Current Practice of Clinical Electroencephalography - John S. Ebersole

661 Pages • 294,548 Words • PDF • 37.9 MB
Uploaded at 2021-09-24 14:42

This document was submitted by our user and they confirm that they have the consent to share it. Assuming that you are writer or own the copyright of this document, report to us by using this DMCA report button.


Current Practice of Clinical Electroencephalography FOURTH EDITION

Current Practice of Clinical Electroencephalography FOURTH EDITION

EDI TO R S Editor

John S. Ebersole, md Professor of Neurology and Director Adult Epilepsy Center and Clinical Neurophysiology Laboratories Department of Neurology The University of Chicago Chicago, Illinois 

Associate Editor

Associate Editor

Aatif M. Husain, md

Douglas R. Nordli Jr., md

Professor Department of Neurology Duke University Medical Center Director, Neurodiagnostic Center Veterans Affairs Medical Center Durham, North Carolina

Professor of Pediatrics and Neurology Northwestern University Medical School; Lorna S. and James P. Langdon Chair of Pediatric Epilepsy Children’s Memorial Hospital Chicago, Illinois

Acquisitions Editor: Julie Goolsby Senior Product Development Editor: Kristina Oberle Production Project Manager: Alicia Jackson Senior Manufacturing Manager: Beth Welsh Marketing Coordinator: Stephanie Manzo Production Service: S4Carlisle Publishing Services © 2014 by Wolters Kluwer Health 2001 Market Street Philadelphia, PA 19103 USA LWW.com All rights reserved. This book is protected by copyright. No part of this book may be reproduced in any form or by any means, including photocopying, or utilized by any information storage and retrieval system without written permission from the copyright owner, except for brief quotations embodied in critical articles and reviews. Materials appearing in this book prepared by individuals as part of their official duties as U.S. government employees are not covered by the above-mentioned copyright. Printed in China Library of Congress Cataloging-in-Publication Data Current practice of clinical electroencephalography / editors, John S. Ebersole, Douglas R. Nordli Jr., Aatif M. Husain—Fourth edition.   p.; cm.   Includes bibliographical references.   ISBN 978-1-4511-3195-6 (hardback)   I. Ebersole, John S., editor of compilation. II. Nordli, Douglas R., Jr., editor of compilation. III. Husain, Aatif M., editor of compilation.   [DNLM: 1. Electroencephalography—methods. WL 150]  RC386.6.E43  616.8’047547—dc23 2014001992 Care has been taken to confirm the accuracy of the information presented and to describe generally accepted practices. However, the authors, editors, and publisher are not responsible for errors or omissions or for any consequences from application of the information in this book and make no warranty, expressed or implied, with respect to the currency, completeness, or accuracy of the ­contents of the publication. Application of this information in a particular situation remains the professional responsibility of the practitioner. The authors, editors, and publisher have exerted every effort to ensure that drug selection and dosage set forth in this text are in accordance with current recommendations and practice at the time of publication. However, in view of ongoing research, changes in government regulations, and the constant flow of information relating to drug therapy and drug reactions, the reader is urged to check the package insert for each drug for any change in indications and dosage and for added warnings and precautions. This is particularly important when the recommended agent is a new or infrequently employed drug. Some drugs and medical devices presented in this publication have Food and Drug Administration (FDA) clearance for limited use in restricted research settings. It is the responsibility of the health care provider to ascertain the FDA status of each drug or device planned for use in their clinical practice. 10 9 8 7 6 5 4

To our mentors, who taught us and gave us the opportunity to discover new things on our own; to our colleagues, who supported us during this endeavor; and to our wives, who lovingly tolerated the long process of completing this volume.

CONTRIBUTORS

A.G. Christina Bergqvist, MD

François Dubeau, MD

Mohamad Z. Koubeissi, MD

Associate Professor Department of Neurology and Pediatrics Perelman School of Medicine at the University of Pennsylvania; Director, Dietary Treatment Program of Epilepsy Division of Neurology The Children’s Hospital of Philadelphia Philadelphia, Pennsylvania

Associate Professor Department of Neurology and Neurosurgery McGill University; Head, EEG Laboratory and Epilepsy Monitoring Unit Montreal Neurological Hospital and Institute Montréal, Québec, Canada

Associate Professor Director, Epilepsy Center Department of Neurology The George Washington School of Medicine Washington, DC

John S. Ebersole, MD Professor of Neurology and Director Adult Epilepsy Center and Clinical Neurophysiology Laboratories Department of Neurology The University of Chicago Chicago, Illinois 

Professor Department of Neurology and Bioengineering University of Pennsylvania; Director Penn Epilepsy Center Hospital of the University of Pennsylvania Philadelphia, Pennsylvania

Lawrence J. Hirsch, MD

Douglas Maus, MD, PhD

Professor of Neurology Chief, Division of Epilepsy and EEG; Co-Director, Yale Comprehensive Epilepsy Center Yale School of Medicine New Haven, Connecticut

Assistant Professor Departments of Neurology and Bioengineering University of Pennsylvania Epilepsy Division Hospital of the University of Pennsylvania Philadelphia, Pennsylvania

Robert R. Clancy, MD Professor of Neurology and Pediatrics Perelman School of Medicine University of Pennsylvania; Founder and Former director, Pediatric Regional Epilepsy Program The Children’s Hospital of Philadelphia Philadelphia, Pennsylvania

Darryl C. De Vivo, MD Sidney Carter Professor of Neurology Professor of Pediatrics Department of Neurology Attending Neurologist Attending Pediatrician New York Presbyterian Hospital Columbia University New York, New York

Dennis J. Dlugos, MD, MSCE Associate Professor Department of Neurology and Pediatrics Perelman School of Medicine at the University of Pennsylvania; Director, Pediatric Regional Epilepsy Program Attending Neurologist Division of Child Neurology The Children’s Hospital of Philadelphia Philadelphia, Pennsylvania

Aatif M. Husain, MD Professor Department of Neurology Duke University Medical Center Director, Neurodiagnostic Center Veterans Affairs Medical Center Durham, North Carolina

Philippe Kahane, MD, PhD

Brian Litt, MD

Douglas R. Nordli Jr., MD Professor of Pediatrics and Neurology Northwestern University Medical School; Lorna S. and James P. Langdon Chair of Pediatric Epilepsy Children’s Memorial Hospital Chicago, Illinois

Faculty of Medicine, Joseph Fourier University Head, Epilepsy Unit, Neurology Department University Hospital of Grenoble Grenoble, France

vii

viii

CONTRIBUTORS

Rodney A. Radtke, MD

Joseph I. Sirven, MD

William O. Tatum IV, DO

Professor of Neurology Chief, Division of Epilepsy and Sleep Department of Neurology Duke University School of Medicine Medical Director, Duke Hospital Neurodiagnostic Laboratory Medical Director, Duke Hospital Sleep Laboratory Duke University Medical Center Durham, North Carolina

Professor of Neurology Professor and Chairman Department of Neurology Mayo Clinic Arizona Phoenix, Arizona

Professor of Neurology Mayo Clinic College of Medicine Mayo Clinic Florida Jacksonville, Florida

Catherine A. Schevon, MD, PhD Assistant Professor Department of Neurology College of Physicians and Surgeons Columbia University New York, New York

Saurabh R. Sinha, MD, PhD Associate Professor of Neurology Vice-Chair for Education, Neurology Duke University Medical Center Director, Epilepsy Monitoring Unit Duke University Hospital Durham, North Carolina

Elson L. So, MD Director, Section of Electroencephalography Department of Neurology Mayo Clinic Rochester, Minnesota

Andrew Trevelyan, MD, DPhil Senior Lecturer in Network Neuroscience Institute of Neuroscience Newcastle University Medical School Newcastle upon Tyne, United Kingdom

James Tao, MD, PhD

Elizabeth Waterhouse, MD

Associate Professor Director of Electroencephalography Laboratory Department of Neurology The University of Chicago Chicago, Illinois

Professor Department of Neurology Virginia Commonwealth University School of Medicine Richmond, Virginia

PREFACE

This volume represents the fourth iteration of Current Practice of Clinical EEG. As such, we hope it reflects the progressive changes and improvements in EEG and evoked potential recording and interpretation that have occurred since the publishing of the third edition 10 years ago. The fourth edition features two new associate editors, with expertise complementary to mine, and 12 new chapter authors, who are expert in their own right. Our goal was to assemble a group of nationally recognized authors who would produce a substantial, yet practical, compendium of EEG ­know-how to serve as a reference for students, physicians-in-training, researchers, and practicing electroencephalographers in the 21st century. In addition to updating areas of clinical EEG that are well established, we wanted to emphasize its neurophysiologic bases in order to promote a deeper understanding of EEG, rather than simply reemphasize a recognition of its patterns. We also expanded the discussion of rapidly evolving areas in clinical neurophysiology, including intraoperative monitoring, ICU EEG, and advanced digital methods of EEG and EP analysis. It is our hope that EEG interpretation will be appreciated again as a science and not simply as a clinical art. As a field of endeavor, EEG is not stagnant, nor has it reached the end of its evolution; rather, there is much remaining to learn and much to be done to exploit to the fullest these electrical signals for the benefit of our patients. John S. Ebersole, MD ix

ACKNOWLEDGMENTS

A number of individuals contributed to this volume both directly and indirectly through the ­software that they developed, which we used to create figures. These include Patrick Berg (Dipole ­Simulator), Michael Scherg (BESA), Manfred Fuchs and Michael Wagner (Curry). We sincerely thank them.

xi

CONTENTS

Contributors  vii Preface  ix Acknowledgments  xi

Chapter 1

The Cellular Basis of EEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

Catherine A. Schevon and Andrew J. Trevelyan

Chapter 2

Cortical Generators and EEG Voltage Fields . . . . . . . . . . . . . . . . . . 28 John S. Ebersole

Chapter 3

Engineering Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Douglas Maus and Brian Litt

Chapter 4

Recording Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 Saurabh R. Sinha

Chapter 5

Normal Adult EEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 William O. Tatum IV

Chapter 6

Normal Pediatric EEG: Neonates and Children . . . . . . . . . . . . . . . . 125 Robert R. Clancy, A.G. Christina Bergqvist, Dennis J. Dlugos and Douglas R. Nordli Jr.

Chapter 7

Generalized Encephalopathy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 Elizabeth Waterhouse

Chapter 8

EEG in Focal Encephalopathies: Cerebrovascular Disease, Neoplasms, and Infections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 Joseph I. Sirven

Chapter 9

Progressive Childhood Encephalopathy . . . . . . . . . . . . . . . . . . . . . 258 Douglas R. Nordli Jr. and Darryl C. De Vivo

xiii

xiv

CONTENTS 

Chapter 10 Pediatric Epilepsy Syndromes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 Douglas R. Nordli Jr.

Chapter 11 EEG in Adult Epilepsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Mohamad Z. Koubeissi and Elson L. So

Chapter 12 EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338 John S. Ebersole

Chapter 13 Subdural Electrode Corticography . . . . . . . . . . . . . . . . . . . . . . . . . 367 James Tao and John S. Ebersole

Chapter 14 Intracerebral Depth Electrode Electroencephalography (Stereoencephalography) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393 Philippe Kahane and François Dubeau

Chapter 15 Evoked Potentials Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442 Aatif M. Husain

Chapter 16 Neurophysiologic Intraoperative Monitoring . . . . . . . . . . . . . . . . . 488 Aatif M. Husain

Chapter 17 Continuous EEG Monitoring in the Intensive Care Unit . . . . . . . . . 543 Saurabh R. Sinha and Lawrence J. Hirsch

Chapter 18 Sleep Disorders: Laboratory Evaluation . . . . . . . . . . . . . . . . . . . . . 599 Rodney A. Radtke Index  631

1

The Cellular Basis of EEG CATHERINE A. SCHEVON • ANDREW J. TREVELYAN

Introduction Electrical Flow in the Brain Action Potentials Synaptic Currents Active Conductances Gap-Junction Coupling Nonneuronal Currents The Anatomical Organization of Cortical Currents Hippocampal Anatomy Basket Cells Neocortical Anatomy and Thalamic Connections Oscillations The Structure of EEG

Introduction EEG remains, as it has been since Berger made his first recordings in the 1920s (1), a pivotal diagnostic clinical tool for assessing brain activity. Traditionally, EEG is interpreted by visual inspection of the signal traces, using a set of qualitative rules developed through collective clinical experience, to define features of the EEG that are associated with particular brain states.

The Relationship between Oscillations and Cellular Activity Hierarchical Phase-Amplitude Coupling Cellular Basis of Epileptiform Activity Neural Activity during Epileptiform Discharges Lessons from Microelectrode Arrays: Ictal Discharges and the “Ictal Penumbra” Surround Inhibition EEG Markers of Ictal Territories: High-frequency Oscillations Conclusion Acknowledgments References

These qualitative properties of the signal include the structure and symmetry of prominent spontaneous oscillations such as the posterior dominant rhythm and sleep spindles, the relative mixture of frequencies and their spatial organization, and the presence of paroxysmal waveforms such as epileptiform discharges. This empirical approach to EEG interpretation, in which certain features of the EEG signal have become associated with particular brain states, has

1

2

The Cellular Basis of EEG

served generations of neurologists and neurophysiologists well. Historically, the reason this approach has been predominant is obvious: Berger and his early followers quickly realized the significance of certain characteristic features of EEG long before we had tools that could make sense of the signals. Over the ensuing decades, clinicians have learned to use these EEG features as powerful indicators of focal or widespread cortical abnormalities. These diagnostic associations have proved robust; they are standing the test of time well, and will be covered extensively in later chapters. Is it enough then simply to understand EEG as a “black box,” and go no further? We believe not. There are powerful arguments that we should strive for deeper understanding, asking why these associations exist. One reason is that technology is advancing rapidly, allowing us to record and manipulate brain signals in ways that will shed new light on the meaning of old data sets. Another motivation is that in many ways, the EEG is an impoverished signal, which prevents us from differentiating between various possible underlying activities. This leads to these different activities being pooled together, which then weakens what associations we can draw. Another important reason to explore the neural basis of these signals is that there are interpretative pitfalls. For instance, some uses of EEG, such as the localization of activity, are fraught with the problem of circular logic, if independent measures of activity patterns are lacking. It is important to be able to recognize these cases. There is a tendency to rest on the simple intuition that the EEG signal is merely a weighted average of everything that goes on beneath the electrode. In some ways, this is a truism, but the danger comes from confusing the different levels of understanding the brain. The EEG derives from currents flowing in and out of cells, but the activity that most neuroscientists and neurologists are ultimately interested in is neuronal firing behavior, because that is how information is thought to be processed. One of our aims in this chapter is to show that the ionic level and the firing levels do not always relate in a simple manner. Most importantly, we want to stress that these issues are not merely of academic interest, but may seriously impact on clinical practice. In this chapter, we will describe findings in basic neuroscience that link these EEG signal features to the sources of electrical activity in the brain. Our goal is to provide electroencephalographers with an understanding of the cellular activity that contributes to the electrical signals measured by EEG, the sources and structure of cortical oscillations, and the network behaviors contributing to normal and pathologic EEG activity. We are at an exciting time in neuroscience, when new ways of recording, analyzing, and even manipulating neuronal activity are being developed at

great pace. Recent neuroengineering advances have resulted in clinical sensors capable of high spatiotemporal resolution recordings that augment standard EEG with information about neuronal firing. This makes available a broad array of neurophysiological data types, ranging from oscillations synchronized over large brain areas to unit firing in a single cortical macrocolumn, which electroencephalographers should develop familiarity with, as the new technologies find increased use in the clinical setting. There is also increasing use of sensors recording from below the dura or within the brain parenchyma, which not only afford ever greater diagnostic possibilities but create new opportunities for the investigation of human cognitive function, a venture that requires close cooperation between neuroscientists and clinicians.

Electrical flow in the brain Changes in the extracellular field potential arise because ions flow in and out of cells at very focal sites, thereby creating ionic flow also in the extracellular space (Fig. 1.1). The driving force for the initial movement is invariably the electrochemical gradients that exist across the cell membrane. By far, the largest currents flow through ion channels; indeed a simple lipid bilayer without channels is essentially impermeant to ions. There are also small currents associated with electrogenic ion pumps—proteins that generally are working to restore ion balances and thus are working against the electrochemical gradient. Examples of these are the various ATP-dependent pumps that restore Na+, K+, Ca2+, and H+ across various membranes. These can ultimately shift the membrane potential by up to about 10 mV, but in terms of rates of flow of ions, these currents are far smaller than channel currents. It is possible that these pumps may contribute to extremely slow frequency changes in the extracellular field, including so-called DC shifts, small step changes in the electric field that have been seen in certain brain states including epileptic seizures—this remains an open question. It is quite clear, however, that any rhythms higher than about 1 Hz are too fast to be attributed to pumps, and that all such faster rhythms must therefore arise from flow through ion channels. For the purpose of this chapter, therefore, when referring to cell membrane currents, we will use the term conductance as a synonym for ion channel. Patch clamp studies can derive the unitary conductance of a single ion channel, and so the total conductance is simply the product of this unitary conductance and the number of channels that are open. Since the total conductance is a nice intuitive term, reflecting the total number of channels open, it is used in preference to its reciprocal, the resistance, when talking about membrane currents.

The Cellular Basis of EEG

Figure 1.1: Local circuits induced by transmembrane currents. A: Changes in intracellular membrane potential (upper panel), and in the membrane conductance of Na+ and K+ during an action potential. B: As the action potential propagates, the highly focal changes in transmembrane conductance create large movements of Na into, and K out of the cell. Currents moving into cells are referred to as current sinks (i.e., relative to the extracellular space), and those moving out of cells and into the extracellular space are called current sources. Charge redistribution away from these sinks and sources can be measured by electrodes as fluxes in the electric field. C: Schematic showing how the electric field drops with distance down a cable. A steady state holding potential drops off with a space constant, l = (d·Rm/4·Ri)0.5. The space constants for oscillating currents are shorter (the drop off is faster), because the capacitance absorbs current proportional to the rate of change of the potential. The dissipation of electric field in the extracellular space is likely to be similar, although the complexities of the circuit are far greater than for intracellular currents, and so we do not have analytical solutions for the extracellular space. D: Local electrical circuits may also be created through synaptic conductances. E: Linear IV relationships follow Ohm’s law, where the gradient is the conductance (reciprocal of the resistance), and (F) the intersection of the abscissa is called the reversal potential. Persistent currents are also termed “passive conductances”—synaptic currents that are Ohmic are not strictly speaking “passive” because they are gated currents, by neurotransmitter binding. In contrast, other currents show large voltage-dependent changes in conductance. These are referred to as “active conductances.” All gated currents, including both neurotransmitter-gated and voltage-gated currents, can be very rapidly activated, thereby generating large currents and rapid fluxes in the extracellular field.

3

4

The Cellular Basis of EEG

Action Potentials The elucidation of the currents flowing during an action potential, by Alan Hodgkin and Andrew Huxley, was a seminal moment in the history of neuroscience (2–5) (Fig. 1.1). They described how, as a neuron is depolarized from rest, there is a point at which voltage-dependent Na+ channels are suddenly opened. There is then a surge in the inward current, creating a positive feedback cycle, which is only terminated by the equally sudden inactivation of the Na+ channels. Inactivation occurs very quickly, after about a m ­ illisecond, and coincides roughly with the slightly slower opening of voltage-dependent K+ channels. This combination of closing of Na+ channels and opening of K+ channels rapidly brings the membrane potential back down to resting membrane potential (or even overshoots briefly). ­Restoration of the resting membrane potential results in shutting of the voltage-dependent K+ channels. The density and kinetics of the ion channels (particularly the K+ channels) vary in different neuronal classes, allowing some cells to fire at far higher rates than others. Firing patterns are also influenced by other slower currents, which shape the tail end of action potentials, leading to after-­ hyperpolarizations and after-depolarizations. The major currents underlying an action potential, though, last in the region of just 1 to 3 milliseconds. Consequently, action potentials are very high-frequency signals, and this affects how far they travel through electrical fields with significant capacitance (Fig. 1.1). Electrophysiologists refer to the “transfer function” (6), which is the drop in electrical field going from point A to point B. A steady holding potential at point A drops off rapidly when recorded at increasing distances. Most theoretical work on transfer functions has considered the flow of membrane potential changes intracellularly; for instance, examining how a synaptic potential drops from the dendritic spine to where it is usually recorded at the soma. Extracellular transfer functions are considered to be analogous, except that the mathematical model is far more complex, with multiple parallel routes through the network and a less clearly defined capacitive element, and an analytical solution, even if it were possible, has not been worked out. Empirically though, it is clear that high-frequency voltages such as action potentials dissipate rapidly in the extracellular space. In contrast, slower electrical fields, such as those associated with synaptic currents, influence the electric field over a larger volume of the brain. The very transient nature of action potential currents is also an important issue when there are multiple neurons firing. The spatial and temporal

juxtaposition of the firing neurons, and their positions relative to the recording electrodes, influence the recorded signal. Consider, for instance, the case when two closely apposed neurons are firing repeatedly at high frequency (>250 Hz) but 1 to 2 milliseconds out of phase with each other. Locally, this induces small currents reverberating in the extracellular space between the cells, as Na+ ions go in and K+ ions come out of the cells. These reverberating extracellular currents may be recorded, but they are small and very local, and thus their visibility is very dependent on the exact positioning of the recording and reference electrode, and the impedance of the electrodes. Now consider a larger cluster of neurons, say 10 cells firing at more typical rates for pyramidal cells, at 50 Hz, but again with some millisecond jitter. As before, if the recording electrode is located within the cluster, then one may be able to discern the direction and amplitude of small, local extracellular currents relating to particular cells (relative to the distant location of the reference electrode), and this is the principle behind spike-sorting algorithms. When the recording electrode is outside the cluster of neurons, the jitter in firing results in some destructive interference, and there may be little current flow at the actual electrode. If, however, firing becomes more synchronous, then the currents become far more visible at that electrode. The visibility of the currents, therefore, requires synchrony at the timescale of the transmembrane currents. We will return to this point shortly when considering the much longer duration synaptic currents. A second factor determining the visibility of currents is their amplitude. Hodgkin and Huxley based their model on recordings of the squid giant axon, the largest axon in the natural world, but much subsequent work has shown that the essential features of the Hodgkin-Huxley model are also replicated in the far smaller neurons in mammalian brains. Regarding how these currents appear in the EEG, however, in this instance, size matters. The essential details are exactly analogous to changes in concentration. A drop of ink added to a thimbleful of water will cause a big change in color, but the color change might be imperceptible when a drop is added to a bucketful. With neuronal membrane potential, we are dealing with charging up membranes; so the structure of interest is a surface area, not a volume, but the principle is exactly the same. For large structures, like a cell body, or the squid giant axon, a large charge flow is needed to substantially change the membrane potential. Conversely, a large membrane potential change implies a large transmembrane current. But for tiny structures, like most mammalian axons, the currents involved are, relatively, very small. Furthermore, many mammalian axons are myelinated (although many locally connecting grey matter axons are not), and so still smaller currents

The Cellular Basis of EEG

are involved per unit length of axon, because the current only flows at particular hotspots, the nodes of Ranvier. Consequently, as it propagates through the axonal tree, an action potential does not constitute a large “sink” of current away from the extracellular space. Moreover, since action potentials propagate rapidly down the axon, the current sink is distributed over a spatially extensive area very quickly. In contrast, a rather more ­visible current in the EEG arises from the postsynaptic consequence of axonal firing.

Synaptic Currents Neurotransmitters released from axonal terminals and varicosities cause postsynaptic receptors to open on dendritic spines, on the dendrites themselves, and also on the soma and axon hillock. The same size argument as outlined earlier, for axons, also applies for dendritic spines. In contrast, the dendritic trunks may be relatively large (>5 μm diameter, compared to approximately 1 μm or smaller for most axons and dendritic spines). Theoretical calculations show that these size considerations impact on the currents flowing through receptors at these different locations. The currents flowing are dictated by Ohm’s law

V = IR (1) Turning this equation round, we get I =

V R

(2)

This may be rewritten more appropriately for transmembrane currents, in terms of the membrane driving forces and the conductance, g, which is the reciprocal of resistance (g = 1/R). Thus, the current (note that the convention is to plot inward currents as negative) is expressed as

I = g(Em − Erev) (3)

where Em is the membrane potential, and Erev is the reversal potential for the ion in question, which is set by the concentrations inside and outside the cell, as given by the Nernst equation Erev α In

[ion]out [ion]in

(4)

5

According to these equations, the driving force for any transmembrane current is the difference between the reversal potential and the membrane potential. When a receptor opens on a dendritic spine, small ion movements cause big changes in membrane potential (the ink and thimble analogy mentioned earlier), and so the membrane potential will shift rapidly toward the reversal potential of the receptor. As a result, the driving force drops sharply in these small structures, and in turn, the current also drops quickly. In contrast, receptors on larger structures, such as the dendritic trunk or soma, do not shift the membrane potential toward the reversal potential as easily; so the driving force persists and the currents also remain high. A second important feature of synaptic receptor currents is that, relative to action potential currents, they are long lasting; most last 5 to 20 milliseconds, while others last hundreds of milliseconds. The best characterized of these long-lasting synaptic currents are N-methyl-d-aspartate (NMDA) plateau potentials (7). NMDA receptors require glutamate binding to open, but like Na+ channels are also voltage-gated, and can thus also produce regenerative currents. This is a kind of action potential, but unlike our previous description of the Hodgkin-Huxley model, NMDA receptors are inactivated far more slowly than Na channels and thus induce a selfsustaining depolarization. This “plateau potential” may also be boosted by other ­voltage-dependent Ca2+ currents, because these too are inactivated slowly, and hence also contribute to the self-sustaining depolarizing current over tens of milliseconds. A further important feature of NMDA receptors is their modulation by other extracellular ions and molecules. Mg2+ ions are critical for the voltage-­dependency (8); in addition various amino acids, including ­glycine and serine, bind at extracellular sites and modulate the NMDA opening properties  (9). Thus, changes in the ionic and molecular constitution of cerebrospinal fluid may influence the likelihood of such plateau ­potentials occurring and also their duration. The protracted nature of these postsynaptic currents means that they will summate strongly in the extracellular field, even when the initial, causative, presynaptic action potentials are not absolutely synchronous. The time scales of these currents are simply far longer than those of action potentials. Furthermore, there is one obvious, and commonly occurring, situation when the postsynaptic currents are highly synchronized: when they arise from an action potential in a single axon, releasing vesicles at multiple varicosities. We shall return to this point later, in our discussion on basket cells. To summarize the section so far, postsynaptic currents are larger and longer lasting and have lower-frequency components than those underlying action potentials, and consequently are more visible in extracellular recordings.

6

The Cellular Basis of EEG

Active Conductances In addition to numerous synaptic receptors, the dendrites and somata have many other conductances. The most relevant for EEG are so-called active, or non-Ohmic, conductances. These change their conductivity with changes in membrane voltage, unlike Ohmic conductances, which have a linear I-V relationship. Hodgkin-Huxley channels and NMDA receptors are all examples of active conductances, but there are many others besides. Of particular importance are the voltage-dependent Ca2+ channels, which in addition to their role in gating neurotransmitter release at synaptic terminals, are also found in the dendritic tree and soma where they underlie bursting behavior in many neuronal classes, including certain classes of pyramidal cells in cortex and hippocampus. In these pyramidal cells, there are great concentrations of voltage-dependent Ca2+ channels at the tuft of the apical dendrite, which may play a role in certain rhythmic activity and in distorting, and especially boosting, synaptic inputs. Since these are concentrated within the upper cortical layers, they are examples of highly localized current sinks. An important feature of these active conductances is that they can give rise to sudden and large changes in local transmembrane currents near their threshold voltage. However, synchronization of these currents in populations of neurons, which is likely to be a prerequisite for such currents being visible in EEG recordings, requires interactions between neurons. The most obvious way this happens is through chemical or electrical synapses, although recent work has provided experimental evidence also for so-called “­ephaptic” interactions. Ephaptic interactions refer to how very small fluctuations in the local field potential caused by one cell’s firing, passively entrain the a­ ctivity of other neurons (10–13) (Fig. 1.2). Several groups have now shown that electrical fields, of comparable amplitude to those actually recorded in the brain (2 to 4 mV/mm), can modulate firing patterns (13). Interestingly, the effect appears to be proportional to the rate of change of the field, so higher frequency oscillations entrain more powerfully than lower frequencies, although as we discussed earlier, lower frequencies propagate further in the brain. Thus, even though these local ephaptic interactions are tiny, the entrainment may still grow to synchronize many neurons, in the same way that fireflies may coordinate their flashes, or theater audiences come to clap in synchrony.

Gap-Junction Coupling Electrical synapses, through gap junctions, have been described in a number of cortical populations. The best evidence for such connections is between

Figure 1.2: Entrainment of neuronal bursting induced by small fluctuations in the extracellular field (adapted from Fröhlich F, McCormick DA. Endogenous electric fields may guide neocortical network activity. Neuron 2010;67:129–143). Spontaneous bursting activity in a brain slice was modulated by application of small extracellular voltages of comparable size to those recorded in vivo. The switch between up and down states is rapidly entrained to whatever extracellular rhythm is being imposed, as shown by the close correlation between the period of imposed field and the period of the entrained activity. Demonstrations like this provide a proof of principle that ephaptic effects are real, albeit very small. Indeed the intracellular effect is likely to be no larger than the summed activation of less than about three to five excitatory synapses.

members of the same inhibitory neuronal classes. Thus, parvalbumin-­ expressing basket cells are connected through gap-junction syncytia, as are somatostatin positive neurons, but it appears that the two classes of neuron do not connect with each other, and that the syncytia are thus entirely separate. There are theoretical arguments and some experimental data also suggesting that pyramidal cells may be interconnected through gap junctions located between axons, creating what is referred to as an axonal plexus. It has been suggested that, in certain conditions, activity may spread and escalate through this plexus, building up eventually to a full ictal event. This model of ictal initiation derives from a rather artificial, in vitro epilepsy model in which all synaptic activity is prevented by removing Ca2+ ions from the extracellular medium. In contrast to synaptic

The Cellular Basis of EEG

function, gap-junction coupling appears enhanced in zero Ca2+ medium, and sharp electrode recordings of neurons commonly show “spikelets,” which are presumed to be retrogradely propagating action potentials that invade the soma, but which fail to induce a full action potential there because of the impedance mismatch going from the small axon to the large soma. The trigger for the back-propagating action potential has been suggested to be a gap-junction connection with another active axon. An important point to make is that gap-junction-mediated currents between cells are likely to be virtually invisible to extracellular recordings. This is because gap-junction currents flow directly from one cell to the next, without creating a local sink or source of current in the extracellular compartment. They may be visible indirectly, however, by synchronizing firing between cells (14,15). As such, gap-junction coupling may lead to summation of transmembrane currents in such synchronized cells. Two other points are worth making. First, gap junctions may extend the electrotonic length of neurons: Quite simply, the cellular “unit” is extended into the adjacent cell, creating a syncytium, and so charge entry into one cell is then dissipated and eventually returned to the extracellular space over a larger area. In this way, the effective circuit is increased in size, although in doing so, the “return” part of the circuit, away from the initial current, carries a lower density current because it is distributed over a larger surface area. The second is that while the conventional arrangement is for hemichannels to be juxtaposed with hemichannels of adjacent cells, thus making a full gap junction, there are cases when the hemichannel appears to exist in isolation. It is possible that such hemichannels may be induced by epileptic activity (16), and cause further trauma to the participating neurons, while also allowing currents that may be visible in EEG recordings.

Nonneuronal Currents Finally, we come to currents flowing through glial networks. Glia are also highly connected through gap junctions, and this facilitates the spread of waves of Ca2+ transients from cell to cell. The cytosolic Ca2+ surges come primarily from intracellular stores, making them unlike those in neurons, which primarily reflect transmembrane currents. As such, glial Ca2+ transients are not likely to have a direct impact on EEG signals. They may do so indirectly though, by releasing neuroactive substances, which cause local neuronal discharges. There is good evidence from many laboratories supporting this view, although much of this derives from in vitro work, and

7

there is continued debate regarding to what extent these data merely show the potential of glia, rather than what really happens in vivo. Our understanding of these phenomena continues to evolve, and will soon clarify what glia do in both physiological and pathologic brain states. A more established function for glial syncytial networks is the redistribution of local inhomogeneities of extracellular ions. After intense bursts of neuronal activity, extracellular K+ rises and Ca2+ drops markedly. Following an epileptic burst, extracellular K+ may easily exceed 10 mM. Steven Kuffler, in the 1960s, suggested that glial networks may redistribute K+ ions in these conditions. Recently, strong support for this has been provided by the demonstration that glucose imbalances in parts of cortical networks cause increased coupling of glia, and thereby also enhanced the spread of a fluorescent glucose analog (17). If the same process occurs with charged ionic species such as K+ ions, then focal epileptic activity would trigger potentially a large current sink, with characteristically slow kinetics, and the current circuits would match the dimensions of the extent of changes in extracellular K+. This needs further research, but the theoretical argument suggests that such low-frequency currents, reflecting their slow kinetics, appear in the EEG as a change in very low-frequency (or DC) power, and may thus provide a useful marker of the location of intense epileptiform discharges.

The anatomical organization of cortical currents Hippocampal Anatomy The hippocampus in humans is tucked away from accessible sites for placement of EEG electrodes, or even subdural electrodes. As such, activity in the hippocampus is largely invisible in EEG or ECoG recordings, although it can be recorded readily using depth electrodes. Hippocampal recordings both in vivo and in vitro have contributed hugely to our understanding of synaptic physiology and network oscillations. The usefulness of the hippocampus for these researches is that its anatomical organization is remarkably laminar. Pyramidal somata are packed densely in the pyramidal layer, with dendritic trees that project almost perpendicular to that layer on both sides. The excitatory drive on to pyramidal cells is arranged perpendicular to the dendritic trees, and parallel with the pyramidal cell layer. In short, one could not design a structure more suited to the study of associative learning, allowing the simple independent stimulation of parallel

8

The Cellular Basis of EEG

inputs onto a single cell by placing the stimulating electrodes at different depths along the somato-dendritic axis. Such stimuli, and indeed any barrage of presynaptic inputs, induce a focal synaptic current with respect to this same axis, and conversely, one may record current sinks along that axis and associate them with particular presynaptic inputs. Further, one may progressively move a recording electrode along the somato-dendritic axis, and record an inversion of the sign of the recorded field, and thus ascertain the location on the dendrites of the transmembrane current. The direction of the sign inversion may then indicate whether the transmembrane current represents a current sink (positive charge entering the cell, and thus “leaving” the extracellular space) or a source (negative charge entering the cell [e.g., a hyperpolarizing GABAergic current] or positive charge leaving the cell [late action potential K+ currents]). This approach to dissecting out the location of synaptic currents and neuronal firing is referred to as “current source density analysis” (18). The arrangement of inhibitory circuitry in the hippocampus is similarly ordered (19,20). Thus, anatomists have been able to classify many different types of hippocampal interneuron, characterizing them according to their stereotyped axonal and dendritic morphology and thus their drivers and their outputs, the location of their somata, the expression of particular proteins (parvalbumin, somatostatin, VIP, calretinin, etc.), their firing patterns, and many other cellular attributes. A major research goal in recent years has been to characterize how these different interneuronal groups participate in different network oscillations, or even sculpt these rhythms.

Basket Cells Special mention needs to be made of one particular inhibitory cell class, that which targets the somata of pyramidal cells. The interneurons wrap their axons around the pyramidal somata, which thus appear like baskets cupping the somata, hence the derivation of Cajal’s term, the basket cell (his Golgi stains typically only labelled small number of cells, and so he saw many examples when the presynaptic axonal basket was seen without the contents of the basket, the pyramidal soma, being visible—it looks even more basket-like when the basket appears empty!). These cells also go by other names, most commonly fast-spiking interneurons, and parvalbumin positive neurons, but these are not, strictly speaking, synonyms. It is clear, as the classification of cortical interneurons becomes more sophisticated, that there are at least two populations of basket cells, only one of which fires

at very high frequencies (>200 Hz). Furthermore, while current thinking is that all fast-spiking basket cells express parvalbumin, it is also the case that other cell classes express parvalbumin too, and even that parvalbumin expression may be altered by activity levels. Thus, all fast-spiking basket cells may be parvalbumin positive, but not all parvalbumin positive neurons are fast-spiking basket cells. We make these points mainly to emphasize the fact that interneuronal classification is a minefield for the unwary, and that many specialists, quite rightly, get agitated when others are sloppy with their classification. As ever, the devil is in the details, and we would encourage readers to consult more specialist texts on this topic, of which there are many (19,20). We will limit ourselves here to some general, but important, points that are particularly relevant to epilepsy and to the nature of EEG signals. There are many notable features of these remarkable cells; the most important though regard their output. Basket cells project extremely densely on to the local pyramidal population. Indeed, each basket cell appears to synapse on to every single pyramidal cell locally, and furthermore, makes on average at least 10 synapses per connection. These synapses are all clustered densely around the proximal dendrites and the somata of the pyramidal cells. The synapses show a high transmitter release probability, and the postsynaptic receptors have a high conductance. Thus a single action potential in a basket cell may trigger a very large summed conductance in a very focal region of cortex. We make a very important distinction here between conductance and current, because as Ohm’s law states, the current depends on both the driving force and the resistance. Thus there may be a large conductance, but if the membrane potential is close to the GABAergic reversal potential, the driving force is small. At rest, EGABA (typically about −60 mV) lies close to the membrane potential, and under these conditions, the large basket cell-induced, postsynaptic current is small. In one very important situation, however, the driving force is large, and therefore so is the GABAergic current. This occurs when there is concurrent glutamatergic drive and basket cell firing. The glutamatergic input drives the pyramidal cell to a more depolarized state, thus shifting the membrane potential away from the GABAergic reversal potential and creating a large driving force. The inhibitory postsynaptic currents (IPSCs) are thus large, and so constitute a powerful hyperpolarizing drive. They also have a second inhibitory action, which is referred to as shunting inhibition: They make the cell far more “leaky” and thus more current is required to depolarize them. This is particularly relevant for pyramidal cells, which have a very striking arrangement of synapses on their dendritic trees: All their excitatory inputs are located on distal dendrites, with nothing on the most proximal dendrites

The Cellular Basis of EEG

and soma, where instead, a huge density of inhibitory synapses is located. This distribution of synaptic input means that these proximal inhibitory synapses can veto virtually any level of excitation. This is clearly predicted on theoretical grounds, and has been shown experimentally for a comparable arrangement of synapses in crayfish (21), and more recently, in pyramidal cells themselves, in a simple model of epileptiform propagation (22). In this model, ictal discharges are induced by removing Mg2+ ions from the bathing medium, thereby increasing excitation while preserving (at least initially) inhibition. The cortical territories immediately ahead of the ictal wavefront are bombarded by excitatory barrages arising from the wavefront itself, but this is not necessarily translated into postsynaptic firing because of the power of the inhibitory restraint on these neurons. Thus, there arises a kind of “ictal penumbra” around territories that have been recruited to a seizure, in which there may be an extreme discrepancy between the level of synaptic currents (both excitatory and inhibitory) and the level of neuronal firing. This is a very important scenario in human seizures as well, which we will return to later. An important factor in the coordination of this restraining inhibitory burst is that gap junctions interconnect these interneurons. Gap-junction coupling has the effect of synchronizing firing in connected cells (14,15). Consequently, the high-frequency discharges of basket cells during these bursts of activity may be synchronized with submillisecond precision across an extended syncytium of cells. The pattern of inhibition during this restraint is of high frequency (100 to 300 Hz) and large amplitude IPSCs are experienced by all pyramidal cells in a local area. Studies of these types of discharges in the magnesium washout model described earlier showed that these currents are highly visible in the local field potential. Although results from an in vitro model should be regarded only as proof of principle rather than conclusive evidence of cause, this may be one of the mechanisms that can give rise to high-frequency oscillations in clinical recordings, a ­potentially important epileptic biomarker. We will address this topic in more detail in a later section. In addition to a possible role in high-frequency oscillations, basket cells have a central role in gamma rhythm activity. Again, fundamental to this rhythm is the power of basket cells in inhibiting pyramidal firing, which allows them to entrain pyramidal activity extremely well. This entrainment has been shown compellingly by recording the initial desynchronized firing of pyramidal cells from a baseline depolarizing state, and then imposing a gamma frequency basket cell inhibition on the pyramidal cells (23). Each volley of IPSCs shut the pyramids down transiently, but this is followed by a narrow window between IPSCs when the pyramid can fire. Since all local

9

pyramidal cells are receiving essentially the same IPSCs, they are all inhibited synchronously, but then all “bounce back” to fire in synchrony too. Essentially, the same mechanism explains how a depolarizing agent such as kainate, applied continuously to a cortical slice, sets up a gamma rhythm. In short, it creates a powerful depolarizing drive to most of the network, which is then overridden by firing of basket cells, but then the depolarizing influence kicks in again to create a pacemaker current. The exact rhythm is essentially dictated by the rhythm of the basket cell firing, which is also coordinated by gap-junction coupling as described earlier. There is ongoing debate about whether this mechanism is the same for all gamma rhythms. Gamma activity is considered to extend from about 30 to 150 Hz, but this may be further subdivided into “low” and “high” gamma. It is also relevant that in neocortical brain slices, certain pharmacological manipulations may induce concurrent, but separate, gamma frequency rhythms in the supragranular and infragranular layers (24). It remains to be resolved what distinguishes the two rhythms in this model. It is notable, though, that the predominant laminar frequencies are the same as the firing rates of basket cells in those same layers, suggesting that the two frequencies arise simply because the different laminar basket cells vary in their sensitivity to the pharmacological agent, kainate. What this work does show, however, is that a single cortical column need not follow the same rhythm throughout the cortical depth, and that the separation of rhythms may reflect the external drive. Gamma activity in vivo may show increased variations in the instantaneous peak frequency, which is thought to reflect the fluctuations in the amplitude of the population IPSC: a large IPSC, perhaps unsurprisingly, delays the rebound firing, and thus extends the period of the oscillation.

Neocortical Anatomy and Thalamic Connections Most of the same cell classes have been identified in neocortex, as have been described in hippocampus. One difference is that in neocortex, pyramidal somata are distributed across almost the entire cortical depth, from layers 2 to 6. It is possible that this arrangement reduces the potential for ephaptic influences in neocortex, but the truth is we do not know why this evolutionarily more recent cortical structure (“neo”-cortex) is arranged differently from archeocortex (hippocampus). A consequence is that the neocortical anatomy is not as crystalline as in hippocampus, but it seems reasonable to presume that the same basic principles elucidated using hippocampal preparations will also apply in neocortex. Because the somato-dendritic axes of the dominant cell type (pyramidal cells constitute about 80% to 85% of the

10

The Cellular Basis of EEG

total neuronal population) are not all aligned, though, neocortical current sinks and sources may not be so cleanly recorded as in hippocampal tissue. An important feature of neocortical circuits is the pattern of recurrent connections with thalamus. Thalamic nuclei not only provide the major external drive to cortical circuits, but also receive back from cortex a large descending input, primarily from layer 6 pyramidal cells. This corticofugal pathway sends collateral branches also to the reticular nucleus, a group of inhibitory neurons that then project on to the same set of thalamic neurons as the direct cortical pathway targets. Cortical discharges therefore not only create a direct excitation of thalamic neurons, but also cause a disynaptic inhibition mediated through the reticular nucleus. A further key feature of this thalamocortical circuit is that both thalamic and reticular neurons have a peculiar arrangement of low-threshold activated Ca2+ channels. The two nuclei express slightly different forms of voltage-gated Ca2+ channel: Both nuclei express the Cav33.3 isoform, but the reticular nucleus also expresses Cav2.3, which is activated at higher threshold than those found in thalamocortical neurons, and this may be relevant to the subtle distinction between pathologic and physiologic thalamocortical bursting (25), which we will discuss shortly. The presence of these channels means that thalamic neurons discharge in two different ways. If they are activated from a relatively hyperpolarized state (below about −65 mV), then the depolarization opens both Na+ and Ca2+ channels. The latter inactivate far slower than Na+ channels, and consequently, cause secondary firing; in other words, the combined activation of both Na+ and Ca2+ channels causes a burst of action potentials. This is the dominant pattern in sleep and also underlies the spike and wave pattern of idiopathic generalized epilepsies. In the waking state, thalamic neurons at rest are in a more depolarized state, at about −60 mV, and in this state, the Ca2+ channels are already inactivated. A further, transient depolarization then only activates Na+ currents, and the cells fire just a single action potential. This slightly paradoxical state of affairs, whereby neurons burst more intensely if driven from a more hyperpolarized state, is thought to underlie two different patterns of information transfer in the thalamus. Bursts of activity may provide some kind of wakeup call, but the normal waking mode, in which thalamic neurons only fire single action potentials may more faithfully transfer sensory information to the cortex. There are three other notable membrane currents in thalamic neurons, which may influence the likelihood of rhythmic bursting (26,27): Ih, GABAB and tonic GABAA currents. Ih is the hyperpolarization-activated, nonspecific cation current (Erev = −30 mV), encoded by HCN1 and HCN2 genes (the

acronym, HCN stands for “hyperpolarization-activated, cyclic-­nucleotide modulated”, which already tells us much about these conductances) (27). These channels carry maximal current at relatively hyperpolarized levels (about −80 mV), dropping down to almost nothing when cells are depolarized to about −30 mV. The key effect of this activation curve is that the currents act to resist changes from a moderately hyperpolarized state. Take the situation where a cell is stable at rest at −65 mV. At this membrane potential, there will be a subpopulation of these channels opened, which are providing a tonic depolarizing current that will be contributing to the steady state at that membrane potential. If the cell experiences a hyperpolarizing drive though, some of these HCN channels will close, thus reducing the depolarizing drive, and the membrane potential heads back toward −65 mV. ­Remarkably, exactly the same effect happens if a hyperpolarizing drive is given, because this will cause HCN channels to open, thus providing a stronger depolarizing current, and again causing the membrane potential to head back toward −65 mV. The currents have quite slow kinetics however, opening or closing over a time course of about 10 to 15 m ­ illiseconds. Thus, if  these cells are given a sharp inhibitory synaptic drive, arising for instance in the reticular nucleus and mediated by both GABAA and ­GABAB receptors, the HCN channels will then create a pacemaker current from this inhibitory trough. If the inhibition is sufficient to de-inactivate the low-­threshold ­voltage-gated Ca2+ channels (i.e., membrane repolarization following closure and inactivation of the Ca2+ channel changes its conformation to a closed, but activatable, state), the Ih pacemaker current can then reactivate these channels, and the neurons fire a burst of action potentials. This ­reactivates the reticular nucleus, which reinhibits the thalamus, and we have the potential for a self-perpetuating rhythm. This pattern of bursting provoked by voltage-gated Ca2+ channel activation, the latency of synaptic delays between the reticular and thalamic nuclei, and the slope of the Ih-driven pacemaker drive a rhythmic 7 to 14 Hz oscillation in the cortex, referred to as sleep spindles (28). These occur every 3 to 10 seconds in the early stages of sleep, and are most prominent in the frontal and midline regions. A related mechanism is responsible for the spike and wave discharges (SWDs) that are a hallmark of idiopathic generalized epilepsy syndromes. These differ from spindles in several respects. They are slower (3 to 6 Hz), larger in amplitude, have a prominent spike component preceding the slow wave, are recorded throughout the cortical mantle, but may be either frontally or posteriorly predominant. The most notable difference, though, is that unlike sleep spindles, SWDs may also occur during wakefulness, when they may be clinically manifest as absence seizures.

The Cellular Basis of EEG

Sleep spindles and SWDs also appear to differ in the relative involvement of the reticular and thalamic neurons. In vivo recordings in cats (29) and the GAERS rats (“Generalized Absence Epilepsy Rats of Strasbourg”) (30) suggest that the dominant bursting in SWDs occurs in the reticular nucleus neurons, possibly secondary to some initial event in cortex (31). The burst of reticular firing creates a flurry of inhibitory synaptic potentials in thalamic neurons, which consequently fire only single spikes or are silent, unlike their behavior during sleep spindles (29,32). The failure of the burst of inhibitory postsynaptic potentials to deinactivate the Ca2+ channels in the thalamic neurons in these recordings has been attributed to a pathologically high level of tonic inhibitory currents (these “tonic” currents pass through GABAA channels that are persistently open—this may happen because there remain low levels of GABA in the CSF, and certain receptor subtypes, notably those that include δ-subunits, do not inactivate), which act to clamp the membrane potential above the level needed for deinactivation of the voltage-gated Ca2+ channels (33). Intense cortical pyramidal and interneuron firing is also observed during the spike component of the pathologic discharges, which is not present in spindles (34). The two rhythms may be linked by the phenomenon of augmenting potentials in thalamocortical circuits; stimulation anywhere in the circuit at 10 Hz leads to incremental potentials, and ultimately to spike and wave discharges very similar to those in absence seizures (35). Thus, both these rhythms involve burst firing in the thalamic and reticular nucleus, but differ in the relative involvement of the two nuclei and cortex. The rhythms may be influenced by neuromodulation (36), such as cholinergic projections from the brainstem and basal forebrain, which act by modulating the various active conductances, and in particular Ih (26). In both cases however, it is easy to see how the intensity of these rhythms, cycling between the thalamus and the reticular nucleus, may disrupt the flow of sensory information through this relay, during both sleep and absence seizures. We have provided just a brief outline of these fascinating thalamic behaviors, which is covered in far greater detail in more specialist monographs (35,37).

Oscillations One of the most important tasks performed by electroencephalographers interpreting clinical EEG studies is the identification of visually prominent rhythmic patterns, or oscillations. These oscillations often provide clinically useful markers of both normal and pathologic activity. The first oscillation noted in human EEG was the 8 to 12 Hz posterior dominant (alpha) rhythm. The presence of this rhythm during wakefulness and its attenuation

11

(“blocking”) with eye opening were described by Berger in 1929 (1). Similarly, the mu rhythm, with components in the alpha and beta frequency bands, is related to sensorimotor processing in the frontal and parietal lobes (38). As we have discussed, spindles and rhythmic slow waves are distinctive features of the sleep states, whereas during wakefulness, polymorphic and rhythmic, local or diffuse slow waves are commonly associated with cerebral dysfunction. There are also many highly characteristic patterns associated with particular seizure phenotypes, and we could list many other examples. While these long-recognized clinical associations are reason enough to study oscillations, it is increasingly clear that the very nature of brain function is tied to oscillations (39,40). We will not understand the brain until we understand how and why groups of neurons behave this way. Interest in brain oscillations has grown immensely in recent years, motivated largely by two pieces of research that suggested particular functional significance to oscillations. The first was a series of recordings made by John O’Keefe and Colleagues using extracellular electrodes implanted into the entorhinal cortex of rats that were trained to run along simple mazes and tracks (41). They could isolate the activity of single cells (“unit” recordings, detected and sorted from the high-pass filtered signal as we described earlier), and also the local field potential (low-frequency bandpass recordings), all done with the same electrode. They found that as the rat moved along a simple track, cells consistently fired when the rat was at certain points along the track. That is to say, individual cells represented particular places along the track, and these cells were duly termed “place cells.” The other notable finding was that the firing showed a very special relationship with the local field potential, which had a predominant theta rhythm (5 to 8 Hz): As the rat moved through the place field for a particular neuron, the cell fired at a fractionally earlier time point during each oscillation of the theta rhythm. Another way of putting this is that the cell’s firing was at a very slightly higher frequency than the field oscillation. Its firing appeared to “progress” forward as the animal moved forward, and this gave rise to the term “phase progression” for this phenomenon. Thus the position of the rat was represented not only by activity of particular neurons, but also by the rhythm of their firing. The second study was of recordings in cat visual cortex made by Singer and Colleagues, of the responses of groups of neurons to presentations of bars of light (42). What Singer and his team did was identify neurons whose receptive fields were aligned, such that they could be stimulated by flashes of light, either independently with two short bars or together, by a single extended long bar of light. They found that both approaches increased the firing of the pair of neurons, but when they were activated together by a single

12

The Cellular Basis of EEG

B Comp 1.5 Hz 7 Hz 40 Hz 0

1

2

3

time (s)

Figure 1.3: Structure and interaction of brain oscillations. A: Amplitude spectrum of EEG recorded from a lateral frontal subdural grid electrode during wakefulness, with frequency ranges shown as color-coded bars. Note the near linear decrease in amplitude with frequency shown on a natural logarithm scale. B: Schematic of hierarchical, phase-amplitude coupling in EEG. The top trace illustrates the typical observation: Oscillations recorded in the brain are normally complex mixtures of components at different frequencies. The traces below illustrate the individual oscillatory components in the delta, theta and gamma band that comprise the composite waveform. Gamma oscillatory amplitude varies with the phase of the underlying theta oscillation, and theta oscillatory amplitude varies with the phase of the underlying delta oscillation. C: Hierarchical coupling in human microelectrode data recorded from lateral temporal neocortex during an auditory task (Besle and Schevon, unpublished). A modulation index (56) was computed for phases between 0.1 and 20 Hz, and for both amplitudes between 1 and 250 Hz and the amplitude of multiunit activity (MUA, above), filtered 500 to 5 kHz and rectified. Significant phase-amplitude coupling was found between delta phase and MUA (A), gamma (B), and theta (C) amplitudes, and between ­alpha phase and both MUA (D) and high gamma amplitude (E) color version is available online.

long bar of light, that is to say they were responding to the same object, their firing became synchronized at gamma frequencies. The implication was that gamma synchronization was how the brain represents “unity”—that activity in these disparate neurons were reflecting one and the same stimulus. This has proved to be a very influential observation, triggering many subsequent studies, but is limited in that it is just a correlation; what we need now is to be able to modulate these rhythms, either to create or destroy the binding oscillation, so that we can really demonstrate causality between gamma rhythms and the so-called “binding problem.” With the development of

optogenetic techniques, we may now have the tools to test Singer’s remarkable hypothesis.

The Structure of EEG Cerebral signals recorded as EEG contain superimposed oscillations in a wide range of frequencies (Fig. 1.3). In humans, recorded oscillations range from 0.05 to 600 Hz, divided into fixed frequency bands. At the high end of the spectrum, above approximately 150 to 200 Hz, oscillations begin to

The Cellular Basis of EEG

overlap with multiunit activity, or the extracellular signatures of action potentials, rather than synaptic currents. Signal amplitude is much higher in the lowest frequency bands, and decreases with frequency in a steep “1/f” curve. This inverse relationship of signal amplitude to frequency is characteristic of many biological signals. A similar inverse relationship also exists between the spatial extent of oscillations and frequency; as alluded to earlier, higher frequency oscillations are restricted to small cortical volumes, while low frequencies are more broadly distributed (39). EEG recorded from the scalp is typically limited to frequencies under about 30 Hz, due to attenuation of the recorded signals by the skull and intervening layers of tissue (43), as well as interference from extracerebral electrical sources such as muscle activity. Higher frequencies are preferentially screened out due to the smaller amplitude of the high-frequency signals, and due to their limited spatial extent, which causes signal to be lost due to the effects of averaging across the listening sphere of scalp electrodes (44). For example, interictal epileptiform discharges are rarely detected in scalp EEG if they encompass less than 6 cm2 of cortex (45). It is not surprising, therefore, that EEG recorded from electrodes implanted onto the cortical surface (e.g., subdurally) is far better than scalp electrodes at detecting gamma and high gamma oscillations (up to 150 to 200 Hz). Frequencies above 200 Hz are dominated by action potentials, or the extracellular signatures of firing neurons. To detect action potentials, which are most prominent in neocortical layers 3 to 5 about 1 mm below the cortical surface (46), we need to be yet more invasive. Action potentials can be recorded by tiny, high-impedance electrodes inserted into the brain parenchyma, such as microwires added to the end of a clinical depth array (47,48), or the microelectrodes built into the “Utah” array (9–1). Magnetoencephalography, which records from sensors on the scalp but uses magnetic signals that are less prone to attenuation by skull and scalp tissue (52), is therefore more sensitive to high frequencies than is scalp EEG, but is subject to a similar spatial averaging effect.

The Relationship between Oscillations and Cellular Activity In the 1930s, Bishop proposed that EEG is the direct result of rhythmic fluctuations in neuronal activity (53). Specifically, postsynaptic potentials, generated from neuronal ensembles firing in synchrony, induce extracellular potential fluctuations that summate to create waves that are the substrate of EEG signals. These waves may appear at sites far removed from the location of the presynaptic neurons, due to rapid and wide distribution of synaptic potentials down axonal pathways. This is a fundamental property that is important to keep in mind when interpreting clinical studies: EEG is an

13

excellent tool for detecting brain rhythms with fine temporal resolution, but its usefulness for locating the source of these rhythms is often limited. We will revisit this point later on, in the context of seizure recordings. In previous sections, we described the electrical currents induced by firing neurons, and gave examples of EEG rhythms that are generated by bursting cells in the thalamus. We now shift our focus to the EEG oscillations themselves. Earlier, we discussed several mechanisms by which neuronal firing can be synchronized and regulated to generate rhythmic extracellular field potentials, such as shunting inhibition from basket cells onto perisomatic pyramidal neurons. Conversely, slow oscillations themselves may influence the firing probability of a neuron through their effect on endogenous electrical fields, as described previously in the discussion of ephaptic interactions. The best described of these slow oscillations at the cellular and network levels are transitions from a relatively hyperpolarized membrane potential generally below −70 mV, when there is little firing activity, to a more depolarized state around −60 mV and far higher rates of firing. These two states are conventionally referred to as “down” and “up” states, respectively. Such activity is seen in many anesthetized states, and also during sleep, and because many neurons switch in unison, the transitions between up and down states are readily seen in EEG recordings as a delta rhythm (roughly 0.5 to 2 Hz). The synchrony appears far more global in the anesthetized state, whereas during sleep, evidence suggests that the transitions propagate across the cortex as a wave, and are only synchronized locally, over a few square centimeters. The transition from down to up state can be triggered by thalamic inputs. It is also likely to occur spontaneously within cortical networks, arising from activity reverberating within the extensive recurrent connections. The greatly increased firing during the up state is modulated broadly in the gamma frequencies (20 to 80 Hz) (54), which is evident both in the statistics of population firing and also in the pattern of synaptic currents recorded in individual cells. This is an excellent example of higher frequency rhythms being nested within lower-frequency oscillations. Again, this rhythm is likely to arise simply from the pattern of recurrent connections with their characteristic synaptic latencies and kinetics. What terminates these up states is less clear, although obvious possibilities include a progressive synaptic depression, the kinetics of NMDA receptors and the consequent time course of NMDA plateau potentials, increasing levels of Ca2+, which in turn raise K+ conductance by opening Ca2+-gated K+ channels, and changes in other intrinsic neuronal conductances. The clearest examples of spontaneous EEG oscillations with up and down states are the physiological rhythms that occur during sleep, as we discussed

14

The Cellular Basis of EEG

previously, and also pathologic slow waves. The delta rhythms of pathologic focal or diffuse slow activity, as opposed to the slow waves that characterize stage 3 sleep, appear to be generated from the cerebral cortex, and their presence in the waking EEG is usually ascribed to underlying white matter abnormality or cortical deafferentation. Recordings from large deafferented cortical slabs taken from cats revealed spontaneously synchronized slow rhythms, in which neurons were hyperpolarized during positive peaks of the field potential, and depolarized during negative peaks, at which time bursts of action potentials were seen. The EPSPs (excitatory postsynaptic potential) propagated at speeds of 10 to 100 mm/second, to recruit nearby neurons (55). Thus, rhythmic oscillations with up and down states may have either a thalamic or cortical origin.

Hierarchical Phase-Amplitude Coupling Oscillations recorded by EEG can be arranged in a nested structure, in which the amplitude envelope of a higher frequency rhythm modulates (or is modulated by) the phase of a lower-frequency rhythm, as described earlier for gamma activity nested within the delta-waves (Fig. 1.3). An example of such cross-frequency coupling that is often seen in clinical recordings is the so-called “sinusoidal alpha rhythm,” or a spontaneous modulation of the posterior dominant (alpha) rhythm amplitude that a­ ppears to vary in a very slow (80 Hz), provides a reliable index of colocated population firing (62,63) (Fig. 1.3). In addition to the work already described, there is evidence from related fields to support this view. Increases in induced gamma power (>40 Hz) during motor and language tasks in patients implanted with subdural grid electrodes has been found at sites positive for the same functions identified by electrocortical stimulation mapping (64,65). Increased high gamma power is also positively correlated with high values of fMRI (BOLD) signals (66), which are presumed to reflect task-related cortical activation. This tight relationship has proven to be advantageous for applications such as creating control signals from EEG for a brain-­computer interface (67), and it will also be relevant in the discussion of high-frequency oscillations in the next section.

Cellular basis of epileptiform activity Neural Activity during Epileptiform Discharges The primary pathologic EEG findings in epilepsy are transient, high-­ amplitude deflections in the local field potential, typically on the order of hundreds of microvolts and seen over several square centimeters of cortex. At the peak of the deflection, there is a burst of intense, hypersynchronous multiunit activity, which is why epileptiform discharges appear “sharp.” Following this sharp peak, there is an after-going slow wave or amplitude attenuation, with little or no multiunit activity. These may accompany a clinical seizure, or occur during the quiescent interictal period between seizures. Interictal discharges most often occur as isolated events, while discharges that are part of an ongoing seizure generally appear in rhythmic trains. The intracellular signature of both ictal and interictal discharges is thought to be the paroxysmal depolarizing shift (PDS), first described in the early 1960s in an in vitro penicillin (68) and in vivo freeze-induced lesion (68) seizure models. Since these early descriptions of the PDS, it has been demonstrated in many animal models of both acute and chronic seizures (for detailed review, see de Curtis and Avanzini, 2001 [69]). Indeed, the PDS appears to be a defining feature of seizures in these models. The most notable feature of the PDS is that the cell receives such a strong excitation that following each action potential, the cell remains in a relatively

15

The Cellular Basis of EEG A

C

Surf. – EEG

Intra - cell area 4 – 63 mV

20 mV

0.2 s

20 ms

Intra - cell VL –78 mV

0.5 s

20 mV

B

0.2 s

0.5 mV

20 mV

Depth – EEG area 4

Figure 1.4: Burst firing correlated with field oscillations in normal and pathologic conditions (panels A and B adapted from Timofeev I, Steriade M. Neocortical seizures: initiation, development and cessation. Neuroscience 2004;123:299–336). A: Slow wave sleep in anesthetized cat, with simultaneous intracellular recordings from a neuron in cortical area 4 and thalamocortical (TC) neuron in the ventrolateral (VL) nucleus, together with surface and depth EEG from area 4. The depth-positive (upward) EEG waves are associated with hyperpolarization of cortical and thalamic cells, whereas the depth-negativities are associated with cortical depolarization and action potentials, followed by a rebound spikeburst (expanded view, arrow) in the TC neuron. B: Paroxysmal depolarizing shifts during a spontaneous neocortical seizure in anesthetized cat. EEG is shown in the top trace, with fast ripples (80 to 300 Hz) in the second trace. The fast ripple correlates with intracellularly recorded spike bursts. C: Human in vivo microelectrode recording of interictal epileptiform discharges (Schevon, Columbia Univ. unpublished). Fast activity can be seen over the negative EEG peaks (top), correlating with multiunit activity (300 to 3 kHz, middle), and detected action potentials (bottom raster). Note the disorganized appearance of the firing bursts accompanying the discharge peaks, compared to the spontaneous firing at other times. The change in action potential shape, combined with likely distortion and/or contamination of the signal from the filtering process, interferes with unit detection and greatly reduces its sensitivity.

depolarized state, and consequently, there is incomplete deinactivation of the population of Na+ channels. With many Na+ channels not being “activatable,” subsequent action potentials are smaller, and also broader, reflecting a lower K+ conductance too (Figs. 1.4 and 1.5). As a result, the burst

of firing, superimposed on the negative peak of the large depolarizing potential, is characterized by a progressive change in the shape of the action potentials, and ultimately, somatic recordings often show a depolarizing blockade of firing. Theoretic studies show that even when the soma is in a

16

The Cellular Basis of EEG

FIGURE 1.5: Gamma rhythms are dictated by fast-spiking interneurons. A: Cellular firing patterns during gamma rhythms recorded in two different brain slice preparations. Panel Ai (adapted from Shu Y, Hasenstaub A, McCormick DA. Turning on and off recurrent balanced cortical activity. Nature 2003;423:288–293) shows an Up state in a ferret brain slice, induced and terminated by white matter stimulation. Panel Aii (adapted from Ainsworth M, Lee S, Cunningham MO, et al. Dual gamma rhythm generators control interlaminar synchrony in auditory cortex. J Neurosci 2011;31:17040–17051) shows persistent gamma rhythms induced in auditory cortex by bathing the brain slice in kainate. Note the strong depolarizing drive reflected in the “pacemaker” voltages, which follow each IPSC. B: Firing patterns recorded during gamma rhythms in vivo, in ferret prefrontal cortex (adapted from Hasenstaub A, Shu Y, Haider B, et al. Inhibitory postsynaptic potentials carry synchronized frequency information in active cortical networks. Neuron 2005;47:423–435). The fast-spiking interneurons fire most cycles, in contrast to regular spiking cells (presumptive pyramidal cells), which fire only intermittently. There is a marked dip in firing probability of the pyramidal cells shortly after the firing of the fast-spiking interneurons, consistent with the model of gamma that has been derived primarily from brain slice models, and which is schematized in (C). The schematic shows a pattern of large rhythmic IPSCs onto pyramidal cells, which then have a narrow window of opportunity to fire between these IPSCs. If the frequency of IPSCs increases, however, this window progressively shuts, and the network is suppressed locally. In this case, an extrinsic excitatory drive, such as arising from an adjacent cortical territory, may still provide a strong depolarization, thereby shifting Em away from EGABA and creating a driving force for current through these large oscillatory inhibitory conductances. We predict that this “relatively pure” inhibitory oscillation may have a very distinctive spatial distribution reflecting an inhibitory surround mechanism, although this remains to be tested experimentally. Bi. The fast-spiking interneurons. Bii. There is a marked dip.

The Cellular Basis of EEG

depolarizing block, there may still be firing down the axon, away from the excitatory drive that is located on the dendrites. Thus an apparent somatic depolarizing block should not be taken to mean that there is no axonal output from these cells. This change in action potential shape distinguishes the pathologic PDS from the physiological “up” state, in which likewise there is a sustained, and often rhythmically repeated, depolarization, with bursts of action potentials superimposed, but the individual action potentials all have the same shape. The primary difference, however, is probably simply the intensity of the excitatory synaptic drive. The PDS then terminates with a period of hyperpolarization and suppressed neural firing, resulting in attenuation of local field potentials. This after-hyperpolarization may contribute to the periodicity of epileptiform discharges during seizures and other situations, such as periodic discharges in acutely an injured brain. How PDSs are initiated remains unclear. At least part of the difficulty discerning the mechanism is that there are many possible ways this might happen. It is appropriate to distinguish PDSs that occur during a full ictal event, or the afterdischarges during the clonic phase of a seizure or following electrical stimulation, from those that are interictal or at the very start of a seizure. The full ictal PDSs and afterdischarges are likely to arise simply from the coordinated synaptic drive from other neurons. Computer modeling work has suggested that the transition from sustained, tonic firing to the clonic bursting behavior may arise from the interaction between firing rates and synchrony, combined with slowed firing under strong synaptic drive (70). Once the seizure is established, there are likely to be multiple “re-­entrant” loops (similar to the origin of certain pathologic heartbeats). Newly recruited territories, which are still at the tonic phase of firing, may also act as a pacemaker for territories that have already entered the clonic phase. This is the case in brain slices, when propagation is essentially onedimensional, thereby allowing the visualization of activity propagating both forward and backward from the moving ictal wavefront (71). Regarding PDSs that arise de novo (interictal events, or those that initiate a seizure), early theories focused on the concept of “epileptic neurons.” This concept developed from studies showing intrinsically generated synchronized bursting in multiple types of neuronal populations in neocortical layers IV–V and the CA3 layer of hippocampus (72,73), and that such bursting can entrain the occurrence of PDSs, albeit in a disinhibited brain slice, which generated PDSs spontaneously (74). The intensity of the depolarizing drive, which underlies the PDS, however, suggests that its development requires synchronized neural activity in the local network, hence its association with large population bursts and the characteristically high-amplitude,

17

field potential deflection. Several different synchronization mechanisms have been proposed, including the release of neuroactive substances from glia (5–7), short-range recurrent excitation, release of feedforward inhibition (78), ephaptic interactions resulting from the effect of the large extracellular currents on membrane potential (79,80), electrical coupling between pyramidal cells via gap junctions (81,82), and low-calcium field bursts (10,83). Despite the universal association of the PDS with animal epileptic discharges and its clear extracellular signature, it has rarely been reported in in vivo recordings of seizures from microelectrodes in humans (for obvious reasons, these invariably are extracellular recordings). Recent results have finally begun to shed light on the reasons for this apparent conundrum. The development of microelectrode arrays approved for use in humans has allowed us to link multiunit activity (action potentials from many neurons) recorded over a small patch of cortex to standardly interpreted, clinically relevant EEG events—an unprecedented level of detail. It is now apparent that the extent of cortex that displays hypersynchronous burst firing is ­often smaller than has hitherto been understood from conventional EEG ­recordings. This is such an important point that is worth considering in more detail.

Lessons from Microelectrode Arrays: Ictal Discharges and the “Ictal Penumbra” Epileptiform events that correspond closely to the extracellular signature of the PDS have been detected in a small number of patients during seizures (84) and interictal discharges (49,85) recorded with microelectrode arrays. The low-frequency signals recorded from the microelectrodes correspond very closely to those recorded from the overlying subdural grid electrodes. Critically, though, the multielectrode arrays could also record action potentials, and so for the first time, we have a tool that allows a direct comparison between local firing and the EEG over a spatially extended territory. These multielectrode recordings showed unequivocally that there can be a large discrepancy between the low-frequency (EEG) signal, and the local level of neuronal firing (51), and that this is a routine occurrence at particular locations during the extreme conditions of an electrographic seizure. In short, a large EEG signal does not necessarily equate with local firing. This happens where there are large synaptic currents that are not being translated into postsynaptic firing. The large amplitude, rhythmic oscillations that are traditionally used to define the onset and progression of a seizure were visible in all electrodes in the

18

The Cellular Basis of EEG

multielectrode array, and also simultaneously over an extended region sampled by the overlying subdural grid. Volume conduction, or spread of current in a surrounding field by passive electrical properties of the extracellular medium, has often been advocated as an explanation for the large field of these discharges. However, the fractional delays and directional asymmetry (49,86,87) instead indicate that they reflect the rapid axonal distribution of postsynaptic currents. Similarly rapidly spreading field potential fluxes have also been demonstrated using current source density analysis in laminar microelectrode recordings of animal (88) and human (89,90) epileptiform discharges, and in seizure recordings from cats using a dense subdural grid (91).

A critical observation was that these rapid reverberations of low-­frequency signals ( Rout ,Vmeas  Vs when Rinput

1 ∼ Rout ,Vmeas  Vs 2

In order that the measured voltage Vmeas most accurately reflects Vs, it is easy to see that the input resistance of the voltmeter should be large compared with the equivalent source resistance, Rout. The source resistance for clinical EEG depends on many factors—one significant factor is the electrode resistance, which is typically on the order of 5 kΩ. In practice, the input

impedance of most clinical EEG equipment is >100 MΩ. For equipment with higher electrode impedance, the input impedance of the measuring equipment should be adjusted upward accordingly.

Differential Amplification Measuring the EEG voltages, at some stage, requires amplifying the signals. EEG voltages are measured at each electrode relative to some chosen reference electrode. That is, the measured voltage is the difference in potential between the electrode and the reference. To measure this voltage difference, specific kinds of amplifiers are employed, called differential amplifiers, illustrated in Fig. 3.6. The ideal differential amplifier is insensitive to large offsets of the voltages being compared, but actual differential amplifiers are imperfect. The general formula for the actual output for a differential amplifier follows: 1 Vout  Ad (V − V )  Acm  (V V ) 2 Here Ad is the differential gain, and Acm is the common-mode gain. A good differential amplifier will have a large Ad and as small a Acm as possible. The standard metric for how well a differential amplifier ignores the average (common-mode) voltage is the common-mode rejection ratio (CMRR), typically expressed in the logarithmic decibel scale:  A  CMRR  20 log10  d   Acm 

V+ Vout

Rout Rinput

V−

Vs

Figure 3.5: The model for a real voltage-measuring (voltmeter) device, illustrating the concept of input impedance, Rinput. The measured voltage will be proportional to Rinput divided by the total resistance Rinput + Rout. The measured voltage will be most accurate when Rinput is very large compared to Rout.

(Iso ground)

Figure 3.6: Simple concept of a differential amplifier. Ideally, the output voltage is proportional to the difference between the input voltages. See text for formulas of CMRR.

51

Engineering Principles

(The factor of 20 is because in engineering the convention is to consider the power ratio.) For a differential amplifier that is quoted as having a CMRR of at least 80 dB, then the ratio of the differential gain to the common-mode gain is 4 orders of magnitude, or Ad/|Acm| > 10,000. There are several variations of differential amplifiers. A differential amplifier particularly designed to handle large common-mode voltage is sometimes termed an isolation amplifier. An instrumentation amplifier is one that has input buffers to stabilize the input voltages.

Time-Varying Circuits Next consider the circuit in Fig. 3.7, with one resistor and one capacitor, and a voltage source that is zero, but then is abruptly stepped up to some voltage, V0, at time t = 0. We use Kirchhoff’s voltage law and differential calculus to solve for the voltage across the capacitor and across the resistor as a function of time after the voltage is abruptly stepped.

V0  I R VC  0 (Kirchoff's voltage law) I

I C

+

V0 −

C

The units of resistance (Ohm) and capacitance (Farad) are related such that 1 Ohm·Farad equals 1 second. Using this fact, in equations such as these, the lumped value of the resistance times the capacitance is known as the “time constant,” usually denoted by tau: RC ≡ t The time evolution of these voltages is illustrated graphically in Fig. 3.8. After the voltage is stepped, the capacitor begins to charge—the voltage across the capacitor starts at zero and climbs to eventually approach the same magnitude as the applied voltage. Initially, the voltage across the resistor is large, but then falls. The current also jumps to an initially large value

∂VC (capacitor definition) ∂t Capacitor

∂VC V0 VC (combining equations for current)  R ∂t

∂VC V0 VC  ∂t RC  VC V0  1 e 

VC

Figure 3.7: RC circuit. A resistor and capacitor in series, and examining the voltage across the capacitor.

t  RC

  (solution to differential equation)

VR V0 e I



t RC

t V0 RC e R

Voltage

I C

V0 VC (re-arranged) R

R

0.3679

0.1353

1/e

1/e

Resistor

2

0 Time

2

Figure 3.8: Time evolution of charging RC circuit.

3

4

5

52

Engineering Principles

and then falls, also, eventually approaching zero—this current provides the charge that charges the capacitor. At all times, the sum of the voltage across the resistor and the voltage across the capacitor sum to equal the applied voltage. After 1 time-constant, the voltage across the resistor (in ratio to its starting voltage) falls to e−1 = 1/e ≈ 0.368, after 2 time-constants to 1/e2 ≈ 0.135, and so on. After 5 time-constants, the values are for practical ­purposes near enough to the steady-state values.

resistance, while the imaginary part, which contains frequency information, is called the reactance. The imaginary part of the impedance is a mathematical way of describing any phase difference between the driving oscillation and the output response. The impedance of a pure resistor is just its resistance. ZR = R The impedance of a capacitor is

Impedance and Alternating (Sinusoidal) Circuits Electrical equipment in which the sources alternate (alternating current, AC) typically follow a sinusoidal pattern; see Fig. 3.9. Furthermore, any rhythmic pattern (like square stimulation waves, etc.) can be usefully treated as a combination of sinusoids: the pattern can be broken down into its sinusoidal components at different frequencies, then the effect on the individual frequency components may be considered, and the end result can be reconstructed by combining the modified frequency components. This principle is called superposition. In electrical engineering of alternating sources, a generalization of the resistance can be made, called the impedance. The impedance is technically a complex quantity, with real and imaginary parts. The real part is the pure

1 jwC

(In electrical engineering, the imaginary number is represented by j, since I and i are used for current. Angular frequency, w, is often used for simplicity in place of conventional or temporal frequency, f, and these are related by w = 2p.) The impedance of an inductor is ZL = jwL

Simple Filters Consider a (partial) circuit with just two elements—a resistor and capacitor, arranged to alter an input voltage and generate an output voltage across the capacitor, illustrated in Fig. 3.10. Consider this as a voltage divider, but acting on sinusoidal varying frequencies. At very low frequencies (less than 1/RC), the impedance of the capacitor is very large, much larger than the resistor, so the voltage across the capacitor is nearly identical to the input voltage. At very high frequencies, the impedance of the capacitor falls to nearly zero, so the voltage across it also falls to nearly zero. This circuit will then act to preserve (pass) low frequencies, but eliminate high

1

Amplitude

ZC 

0

−1 Degrees 0 Radians 0 Cycles 0

Figure 3.9: The sine wave.

R

Vin 90 /2 1/4

180 1/2

270 3 /2 3/4

360 2 1

Figure 3.10: RC circuit as a lowpass filter.

C

Vout

53

Engineering Principles

frequencies. The precise formulas for the frequency dependence of the filter are as follows: ZC ZC  ZR 1 Vout jω C  1 Vin R jω C Vout 1 (complex)  Vin 1 jω RC

1.0 −3 dB

0.7071

Vout Vin 

−20 dB

0.1

Vout 1 (magnitude)  Vin 1 (ω RC )2 Vout 1  Vin 1 (2 RC  f )2 Vout  Vin

1  f  1   fcutoff 

2

Here, the cutoff frequency has been defined as fcutoff = 1/(2pRC) = 1/(2p·t). In this example, t = RC, and is known as the “time constant” of the filter. As an example, if the resistor and capacitor are chosen such that 2pRC is 0.01 seconds, which corresponds to a cutoff frequency of 100 Hz (=1/0.01 seconds), then the output voltage as a function of the input voltage will follow the plot in Fig. 3.11. (Note that this graph uses log-log axes, as is the convention for analog electrical filters.) For low frequencies (below the high-frequency cutoff), the output voltage is nearly the same as the input voltage—called the passband. At high frequencies (much above the cutoff), the output voltage is much lower than the input voltage—called the stopband. At the cutoff, where f/fcutoff = 1, the output voltage is 1 / 2 ≈ 0.7071 of the input voltage, and this cutoff is where the output voltage has a “shoulder,” beginning to rapidly fall with increasing frequency. In engineering terminology, this filter is technically known as a lowpass filter (it passes or allows low frequencies). Because it “cuts out” high frequencies, sometimes it is termed a “high-cut” or “high-frequency” filter, which can become confusing. For the remainder of this chapter, we

0.01 1

10

100

1000

−40 dB 10000

Frequency (Hz) Figure 3.11: Plot of the transfer function of an RC filter with a cutoff frequency of 100 Hz, showing the ratio of the output voltage to the input voltage as a function of frequency. This filter passes low frequencies, and attenuates high frequencies. Note the log–log axes, the convention for analog filters.

shall use the standard engineering terminology of referring to the band that is passed or allowed, so these are lowpass filters. More complicated lowpass filters can be constructed that are “sharper”— the attenuation as a function of frequency (or “roll-off ”) is steeper after the cutoff frequency. The complexity of these filters is described in terms of the “order” of the filter, roughly proportional to the number of elements with reactance (capacitors or inductors) that would need to be placed in a circuit to create the particular filter profile. Figure 3.12 compares the transfer function roll-off of first-, second-, and third-order lowpass filters. For higherorder filters, the cutoff frequency is still defined to be where the response falls by −3 dB, and is the same (100 Hz) for each in this figure. The roll-off above the cutoff is seen to be much steeper for the higher-order filters. On log-log graphs as the frequency increases, in theory the roll-off moves closer to a straight vertical line. The slope of this line is often expressed in terms of attenuation per decade (frequency increasing by factor of 10) or per octave (frequency increasing by factor of 2). The first-order filter has a roll-off at 20 dB per decade (or 6 dB per octave), the second order at 40 dB per ­decade (12 dB per octave), and the third order at 60 dB per decade (18 dB per octave).

54

Engineering Principles

1.0 0.7071

−3 dB

Passband

Stopband

−20 dB

Lowpass filter

Stopband

Passband

1st order 0.1

Highpass filter

2nd Stop

3rd Stop

0.01 1

10

100

1,000

Pass

Stop

Pass

Pass

−40 dB 10,000

Frequency (Hz)

Figure 3.12: Plot of transfer functions for lowpass filters of first-, second-, and third-order, all with cutoff frequency of 100 Hz. Note the steeper slope in the stopband for increasing order.

Different designs of higher-order filters include Butterworth (3), ­ hebyshev, and Elliptical. Some of the trade-offs in these designs include C the steepness of the roll-off versus the amount of ripple in the passband or stopband. Also note that while higher-order analog filters may have better stopband attenuation and narrower transition band, they require more circuit elements, greater expense to design, more exacting specifications, and each element can add noise to the overall circuit. More is not always better. Of analog filters, for EEG, the most important type is this lowpass filter, for reasons that will become clear in the section on digitization. Analog filters can also be designed that act to pass high frequencies and attenuate low frequencies (highpass), and the concepts of lowpass and highpass filters can be combined to create bandpass filters, which pass frequency components in a defined band of frequencies bordered by the high and low pass filters, but attenuate very low and very high frequencies. Finally, analog filters can be designed to attenuate (“stop”) a very narrow range of frequencies. This filter is called a notch or a bandstop filter, and analog

Bandpass filter

Notch filter

Frequency

Frequency

Figure 3.13: Diagram of the effects of the four basic types of filters: lowpass, highpass, bandpass, and notch (or bandstop). These all are plotted as log-log graphs, as is the convention for analog filters.

versions of such filters were formerly very important in eliminating the ­60-Hz ­powerline noise. Figure 3.13 shows diagrams of the frequency ­effects of these four basic types of filters. In modern digital EEG acquisition, highpass, bandpass, and notch filtering are typically handled by software digital postprocessing (see later).

Digitization In order to be captured, stored, and processed by computers, the electrical signals of the biologic systems need to be digitized—converted from analog signals to digital representations. For time-varying signals, the digitization acts to quantize along two dimensions: time and magnitude. That is, samples are extracted at discrete and regular time intervals and also converted to discrete values of magnitude (4).

55

Engineering Principles

Time Discretization Digital machines operate with very precise clocks, changing states at very regular intervals. Digitization takes samples at set intervals of time—regardless of whether the magnitude is changing slowly or quickly, the sampling rate is fixed. Since this sampling rate is fixed, care must be taken to judiciously choose an appropriate sample rate. The choice of sampling rate places restrictions on the information available in the digitized series. This constraint is known as the Nyquist limit. In essence, the frequencies that are detectable in a digitized time series are those below half the sampling rate. For concreteness, if the sampling rate is 400 samples per second, then components with frequencies up to 200 Hz are captured. Signals with frequencies above 200 Hz have not been captured in any way that is detectable or recoverable. If signals above the Nyquist limit were simply lost, this would be a simple limitation. However, if signals with frequencies above the Nyquist limit are

present in the signal to be digitized, a nefarious complication can occur, known as aliasing. Figure 3.14 demonstrates the issue. In this example, a sinusoidal signal at 11 Hz (dashed curve) is digitized at 16 samples per second. With this sampling rate, the Nyquist limit is just 8 Hz. The actual digitized sample points are plotted as solid squares, and these are what the computer actually records. It can be seen that the gray curve also fits the sample points. This gray curve is a sinuisoid at 5 Hz. In this way, signals above the Nyquist frequency appear as aliases at frequencies below the Nyquist frequency. The signals above the Nyquist frequency are said to have been “folded back,” like a paper folded over at the Nyquist limit, illustrated in Fig. 3.15. In this way, random noise components with frequencies above the Nyquist limit can worsen the signal-to-noise ratio (SNR) of digitized signals, illustrated in Fig. 3.16, and may introduce spurious features in the EEG, which are not really in the brain signals being recorded. These waveforms are an artifact introduced by improper sampling.

+1

0

0.5

1

−1

seconds Figure 3.14: Example of aliasing. An analog 11-Hz sine wave (represented by the dashed line), digitized at 16 samples per second, yields the sample points represented by solid squares. A sine wave at 5 Hz (the light-grey line) matches the same sample points exactly. Sine waves at frequencies above the Nyquist frequency (here, 8 Hz) can always be equivalently matched by sine waves between 0 and the Nyquist frequency.

0

Nyquist (8 Hz)

Sample rate

Figure 3.15: Illustration of digital sampling aliasing, showing the mapping of frequencies above the Nyquist frequency to apparent frequencies below the Nyquist frequency, as if the paper were “folded” at the Nyquist frequency—the source for the alternative description of aliasing as “folding.”

56

Engineering Principles

Spectrum Original

0 Hz

256

Low−pass filtered Filter

0

0

0

2.0

0

Noise folded

Digitize 128 Hz

2.0 Digitize 128 Hz

0

0

0

seconds

2.0

0

In order to avoid this aliasing of noise above the Nyquist limit, digitizers as a rule have an analog lowpass filter prior to the digitization step. This lowpass filter (which can be similar to that described in the analog electrical filter section) is called an antialiasing filter.

Voltage Discretization (Signal Resolution) Digitization samples the voltage at a specific time and then converts the voltage into a digital value. The precision of this conversion is limited. For example, with an 8-bit digitizer, every sampled voltage is stored in 8 bits. With 8 bits, there are approximately 256 distinct possible values (28 = 256). This would mean that the highest voltage could be stored as +127, while the most negative voltage could be stored as −128 (when using two’s complement signed integer representation). Suppose +127 is chosen as the digital value to correspond to +1,270 m V. Then the digital value 0 would represent 0 m V,

2.0

Figure 3.16: Illustration of the adverse effect of failure to filter out frequency components above the Nyquist frequency before digitizing (data simulated). Top left: an analog tracing with a signal at 5 Hz with “white” noise components up to 256 Hz (visible in the power spectrum—inset top right). Bottom left: if this analog tracing is digitized at a sampling rate of 128 Hz (Nyquist of 64 Hz), the noise above 64 Hz is folded, and adds to the noise below 64 Hz. The SNR is ­actually worsened. Top right: if the analog tracing is lowpass filtered (with an RC filter, for instance), the noise components above the Nyquist are ­attenuated/removed. Bottom right: if the lowpass filtered analog tracing is then digitized at a sampling rate of 128 Hz, the SNR is preserved, and the original signal remains easily visible.

and digital value 1 would represent 10 m V. There is no way to meaningfully store the value 3 m V—this would be approximated to a digital value of 0, while 7 m V would be rounded up and stored as 10 m V = digital 1. This configuration would then have what is known as a resolution of 10 m V, or 10 m V per bit. The number of bits allocated to store each sample value is known as the bit-depth. There is a simple and fixed relationship between the maximum voltage recordable (or the range), the bit-depth, and the resolution. Resolution 

Range 2 bitdepth 1

Figure 3.17 shows an illustrative example if a digitizer has a bit-depth of 5 bits, and the range was chosen to be 310 m V (minimum −160 to maximum +150 m V). For current commercial EEG equipment, a typical dynamic range is from +5 to −5 mV. Some early EEG digitizers had a bit-depth of 8, which

57

Engineering Principles max

2N levels

time, and many are capable of being reconfigured—one can choose to digitize 1 channel at an extremely fast sampling rate, or can choose to digitize 64 channels at a proportionately slower sampling rate. In general, the total throughput (N channels × sampling rate) for a digitizer is limited. This is a measure of the total capacity to digitize data for a given ADC device. For further details on analog-to-digital conversion, the Amateur Radio Relay League (5) handbooks are good guides.

+150 V

32 levels

Range 310 V

10 V

Resolution

Ref+

Flash converter

R − min

−160

V

R

Figure 3.17: Example of digitizer resolution. In this example, the bit-depth is 5, giving 32 different digital values. If the maximum and minimum are chosen to be +150 and −160 m V, then the range is 310 m V. The resolution is then 10 m V.

− R

16

2 21

5

10,000m V 65535

R

Vin

5 0.153m V.

+

2N comparators

Encoder

N bits



Digitizer Electronics There are several different electronic architectures for construction of analog-to-digital converters (ADCs), which we shall cover only briefly ­ here, illustrated in Fig. 3.18. A standard design is the flash converter. This architecture is rather easy to comprehend. A flash ADC uses N comparators simultaneously to bracket the digital representation for a voltage. For a bit-depth of b, the number of comparators required is N = 2b. These flash ADCs are quite fast, but because of the requirement for N high-quality comparators, they are typically rather expensive. Two other common architectures for ADC are the successive approximation and the sigma-delta architecture. These are less expensive than flash ADC, but are also typically somewhat slower. The successive approximation architecture basically uses one comparator, but N clock cycles to successively hone in on the voltage to be measured. In general, an ADC can operate on multiple channels at a

+ −

was nearly inadequate for EEG, giving resolutions of 4 m V. Modern EEG machines have a bit-depth of 16, which gives resolutions on the order of 5,000 2 (25,000)m V

+

R

+ −

R

+ −

R

A

+

Ref−

Figure 3.18: A–C: Three different kinds of ADCs. A = Flash converter B = Successive-Approximation Converter C = Sigma-Delta Converter

(continued)

58

Engineering Principles

Successive-Approximation Converter Converted digits Clock

Register

D2 D1 D0

DN−2

DN−1

DAC

Comparator +

B

Vin

S/H



Sigma-Delta Converter Integrator

Comparator

Vin ∑

C

Digital filter

Bitstream

1-Bit DAC

Figure 3.18: (continued)

EEG-specific acquisition Now armed with the basics of electricity and electrical engineering, we turn to the actual signal path of acquiring EEG recordings and the issues at each step.

Electrodes The charges that actually carry current in an electronic machine are simply electrons, flowing through metal conductors. In biologic systems, there are no free electrons, but currents are carried by ions such as positively charged

sodium cations (Na+) and negatively charged chloride anions (Cl−). At electrodes, these two different methods of carrying current must be converted. At the interface of the biologic system and the electrode, chemical reactions occur, in which electrons are transferred. For example, for flow of negative current from the biologic system to the electronic system, an electron could be transferred from a chloride anion to a platinum electrode. The chloride anion would then become (briefly) a neutral chlorine radical, before participating in other chemical reactions. The ease with which the charges are transferred from electrode to biologic system and vice versa may be asymmetric: It may be much more chemically/ energetically easier for positive current to flow into the electrode than for it to flow out of the electrode to the biologic system. In fact, this asymmetry can be quantified by what is called the electrode potential. When the electrode potential is large compared to the voltages of interest, the electrode interface is said to be polarizable. Electrodes where the asymmetry is small are said to be nonpolarizable or reversible (6). The commonly used silver/ silver-chloride electrodes fall into the category of reversible electrodes. In this example, if the electrode becomes negative in potential, neutral AgCl dissolves into Ag+ and Cl− and the Cl− passes into the biologic system. Silver/silver chloride electrodes may be fabricated by immersing silver wire in a solution of electrolyte containing chloride and placing a positive voltage across the electrode. Chloride ions migrate to the surface of the silver and impart a distinctive gray color. When a chloride-treated silver electrode comes into contact with NaCl solution on the skin, currents of Cl− ions flow freely between the electrode and the solution and prevent the electrode from becoming polarized. Polarization is avoided because the electrode and the solution can communicate with ions (namely, Cl− ions from electrode and electrolyte, respectively) that exhibit identical mobilities in solution. ­Silver chloride electrodes are useful for recording DC and potentials of very low frequency. Conversely, electrodes that are polarizable can be modeled as having a capacitance, and act as highpass filters (attenuate low frequencies), and are thus not well suited for DC or very low-frequency measurements. Figure 3.19 shows this effect from several different metals used as electrodes. Large electrode resistances, greater than 5,000 Ω for skin electrodes or 15,000 Ω for needle electrodes, can result in noise artifacts in the EEG recording. This happens because strong electric fields present around the EEG machine induce currents in the electrodes and cables. The currents are small but relatively fixed in magnitude, and flowing across the large electrode resistance generates large voltages. Because of a small area of contact with the subject, a needle electrode has high impedance and is more susceptible

Engineering Principles

59

by viscous gels, by mechanical restrictions (bands or rubber caps), or by ­collodion. Collodion is a glue formulated from pyroxylin in ether, alcohol, or camphor; it is liquid when applied, but it is able to dry to a strong adhesive within minutes. It is suitable for patients who cannot keep still or who need electrodes in place for more than a few hours. Ether is highly flammable. Flames, excessive heat, or pure oxygen should not be near collodion ­applications, and collodion should be used only in well-ventilated areas. It must be remembered that collodion is not an adequate conductive ­medium: Conductive gel must be injected into the cup electrodes and refreshed ­periodically. A blunt needle is usually employed for this task. C ­ ollodion is removed with acetone scrubs, and its application or removal always requires a very well-ventilated room.

Common Types of Electrodes

Figure 3.19: Ability of various types of commonly used e ­ lectroencephalographic electrodes to reproduce a 10 mV, 10 mA square-wave input. (From ­Cooper  R. Electrodes. Am J EEG Technol 1963;3(4):91–101.)

to line (60 Hz) artifacts. In certain circumstances, the benefits of a needle electrode may outweigh the drawback of high impedance. Electrode impedances are tested after application by ohmmeters, sometimes integrated into clinical EEG stations. These devices pass a small current through the electrode circuit and a remote reference (e.g., on the forehead or ear) attached to the patient. Voltage change is measured, and Ohm’s law is used to calculate impedance. Impedance is frequency dependent, and the ohmmeter should therefore use frequencies in a range relevant to the EEG, such as 10 to 30 Hz. Before placing an electrode, the scalp must be prepared by being rubbed vigorously with alcohol or with a skin preparation agent. This action removes dirt and oil from the electrode site and lowers the impedance of the scalp-electrode junction. Obviously, excessive cleaning can be irritating to the patient. Similarly, some patients are sensitive to paste containing salt solutions or bentonite. Several varieties of electrode attachment media, including sodium chloride pastes or gels, conducting sponges, and other specialized electrolytes, are available to lower impedance at the electrode-tissue interface. Scalp electrodes are usually cupped, with a hole at the peak, to facilitate contact with electrode paste or gel. Electrodes may be held in place

Each of the commonly used electrodes has a simple design: a metal contact surface, flexible and insulated wire, and a connecting pin to mate with the headbox or jackbox of the EEG machine. Wires are usually color-coded for easy tracing during troubleshooting. Traditional scalp cup electrodes (Fig. 3.20A) are suitable for most ­routine recordings and are usually of the reversible type. They are most often made of chloride-treated silver cups 4 to 10 mm in diameter. Electrodes fabricated from platinum, gold, or tin are sometimes used. Properly applied electrodes demonstrate resistances of a few hundred ohms. Resistances smaller than this usually indicate a short circuit in the electrode. According to the ­international standards for the EEG, electrode resistances should be less than 5,000 Ω and greater than 100 Ω. Positioning of scalp electrodes and methods for constructing montages of electrode pairs are considered elsewhere in this book. In recent years, prolonged EEG monitoring of patients in ICU settings has brought new challenges and issues related to electrodes. These critically ill patients may require frequent brain imaging, and so traditional metal cup electrodes are disadvantageous because they cause artifact on X-ray-based CT scans. Equally or more important, they are incompatible with MRI ­because they cause artifact and may pose a hazard to patients from heat and electromagnetic induction effects. Recently, electrodes have been developed that are MRI friendly (7). These electrodes are shaped like the metal cup electrodes but the cup is formed from Teflon-coated plastic, which is then plated with a thin layer of conductive silver. These have been shown to be relatively safe with MRI, and cause less artifact on CTs.

60

Engineering Principles

Figure 3.20: Photograph of commonly used electroencephalographic electrodes. A: Depth (intracranial) cylindrical electrode contacts on right, and guidewire extending from left, with multicontact connector. B: Scalp cup electrode set made of conductive plastic coated with silver/silver-chloride (“CT friendly”). C: Single scalp cup electrode made of gold (with nontouch connector). D: ­Subdermal (scalp) stainless-steel needle electrode. E: Intracranial microelectrode array in plastic button. This particular array has 16 micro-contacts. These arrays can be used alone, or inserted into a larger macro grid (Leuthardt Grid System). F: Intracranial subdural strip 1 × 6. G: Intracranial subdural grid 4 × 8, with four connectors of eight contacts each. (Photo courtesy of Liberty Simmons, R.EEG.T.)

Intracranial electrodes are typically subdivided into subdural and depth electrodes. They are used to detect and localize brain activity not visible with scalp EEG recording. Both typically use platinum or stainless steel, since Chloride-treated silver can irritate brain tissue after direct contact for several days, and stainless steel and platinum are relatively inert and safe. Intracranial electrodes may remain in place for days or weeks. Subdural strips and grid electrodes (see Fig. 3.20) are flat rectangular arrays of electrodes encased in plastic (silastic). These are placed by a neurosurgeon via craniotomy (or craniectomy at some institutions, where the bone flap is stored away from the skull until implantation has concluded)

(8) into the subdural space, ideally lying flat against the cortical surface. The electrode contacts are typically circular disks with exposed surface of roughly 2.3 to 5 mm in diameter. Depth electrodes (see Fig. 3.20) are designed for introduction directly into the substance of the brain by a neurosurgeon (also called “stereo EEG” in some centers). Typically, depth electrodes consist of a plastic tube (diameter approximately 1 mm) with a set of six or eight exposed metal cylinders (the contacts), each about 2.3 mm in length, center spaced 5- or 10-mm apart. These allow spatial sampling along the mesial and neocortical temporal structures. Depth electrodes are usually implanted stereotactically (according to a three-dimensional coordinate reference frame), although some experienced neurosurgeons prefer to place the electrodes freehand or with radiographic guidance, under sterile protocol. Orientation, targets, and methods for implantation differ among institutions. Other chapters in this text review various approaches. The amygdala, hippocampus, entorhinal or orbitofrontal cortex, and supplementary motor areas of the frontal lobe are popular targets for depth electrode placement. Intracranial EEG recordings usually demonstrate excellent SNR, because these electrodes have relatively low impedance, are relatively unaffected by muscle and movement artifact, and bypass the high-resistance skull. Depth EEG clearly increases the ability to detect and localize epileptiform activity in selected patients (9), but it has disadvantages. First, not all brain sites can be studied with this technique, and thus there is a possibility of sampling error. Epileptiform activity originating in intracranial electrode only indicates that the electrode is closer than the others to the epileptic network and not necessarily that it is within the network (or “focus”). Second, the technique is invasive, with risks for hemorrhage, infection, reactive meningitis, edema, and headache. Use of these electrodes should be restricted to experienced centers.

New Electrodes In recent years, a variety of new technologies for electrodes have been developed for research purposes. The Utah Intracortical Electrode A ­ rray (UIEA) (10) is one of the earlier types. This is a high-density array of microelectrodes. As shown in Fig. 3.21, it has a 10 by 10 array of needle electrodes in a square with width 4.2 mm on a side, each electrode etched to project 1.2 mm, further etched to expose about 50 m m of conducting surface at the tips. The needles penetrate the pia mater into the upper layers of the cortex. This electrode array was developed for use in brain-computer

Engineering Principles

Figure 3.21: Utah Intracortical Electrode Array, UIEA. (Photo courtesy of Richard A Normann, University of Utah.)

interface (BCI) systems, with the intent to be able to isolate and measure single units or small groups of multi-unit potentials. However, the penetrating electrode arrays are reported to cause gliosis and may lose function after 6 to 12 months. Recently, another microelectrode array was developed that is nonpenetrating and flexible/foldable (11). Shown in Fig. 3.22, it is constructed on a plastic (polyimide) base, with active electronics (multiplexing and amplification transistors) also on the array itself. It is hoped that these nonpenetrating arrays will avoid the disadvantage of gliosis, as well as possibly permitting even longer functioning. Microelectrode arrays such as these hold great interest for studying epileptic seizure evolution on the scale of multi-unit networks, and while these are being used in BCI research on human subjects (12), use in clinical EEG is not yet mainstream.

Preamplifiers The electrodes are typically next connected to amplifiers. These first amplifiers should be well-designed: to have high input impedance, to be accurate differential amplifiers (high CMRR), and to amplify without adding

61

Figure 3.22: High-resolution, flexible, active electrode array with 360 amplified and multiplexed electrodes. Only 39 wires are needed to sample from all of the 360 electrodes simultaneously. The electrode array is ultrathin and flexible, allowing close contact with the brain and high-resolution recordings of seizures. (Photo courtesy of Travis Ross and Yun Soung Kim, University of Illinois at Urbana-Champaign.)

significant noise. For example, several clinical EEG vendors quote input impedances of more than 100 MΩ, and CMRR of more than 80 dB.

Antialiasing Filter Crucially, just before the digitizer step, an analog antialiasing filter must be inserted. The filter should eliminate the frequency components above the Nyquist. Typically, the cutoff frequency is chosen to be lower than the Nyquist to ensure that adequate attenuation is achieved. The American Clinical Neurophysiology Society (ACNS) 2006 guideline 8 (13) recommends that digitization “should occur at a minimum sampling rate three times the highfrequency filter setting, e.g., 100 samples/s for 35-Hz high filter.” (This corresponds to an antialiasing filter cutoff at 66% of the Nyquist frequency.)

62

Engineering Principles

Digitizer Typical clinical EEG acquisition stations have a bit-depth of at least 16 bits, and input ranges of ±5 mV, thus yielding resolution of 0.153 m V. The ACNS 2006 guideline recommends “at least 11 bits” and a resolution of “0.5 m V.” EEG equipment vendors provide equipment with sampling rates of 512  samples per second on 64 or more channels. Some can reach sample rates of 1,024 samples per second though often with reduced number of channels, such as 32. Intracranial recordings with grids and strips can easily reach 128 electrodes or more. Some research equipment packages are able to sample at rates of up to 32 kHz on 128 channels simultaneously.

Digital Transfer/Storage The carefully acquired digitized samples are then transferred to digital computers and stored in a fixed format. Many vendors have proprietary formats, though the open published standard of EDF/EDF+ (14) is growing in support. File sizes of EEG records can be easily calculated with the knowledge of the bit-depth, number of channels, and sample rates. For example, 40 minutes of 32 channels sampled at 512 Hz with a bit-depth of 16 will be 2 bytes × 512 samples per second × 32 channels × 40 minutes × 60 ­seconds = 78,643,200 bytes, or about 80 MegaBytes (MB). Video recording will typically require much more storage, even with modern video codecs (coding/decoding) such as H.264. With higher-resolution devices and higher sample rates, other data formats employing lossless compression on the fly to a fraction of EDF, such as MEF (approximately 90% compression compared to EDF) may be required (15).

Digital Processing The raw data of each digitized channel can then be recovered exactly, and displayed at any later time. It can also be algorithmically processed—for instance to search for possible epileptiform discharges (“spikes”) (16) or analyzed for trends in alpha power (17). Simple conventional display of EEG tracings on computer screens requires some rudimentary digital ­processing—digital filtering. (See the chapter appendix for a fuller discussion of digital filtering.) The aims of digital filtering are, like analog filtering, to attenuate certain components in the signal based on the frequency—for example, lowpass

filter would pass low frequencies and attenuate high-frequency components. There are digital filtering techniques that are modeled after analog RC filters, but there are many digital techniques that have no such analog counterparts.

Finite Impulse Response Filtering Consider a simple operation on a sequence of digital samples: for each input point x[n], replace it with the average of itself and the just prior sampled value x[n − 1], yielding the new output y[n]. This is called the “two-point moving average”: 1 1 y[ n ]  x[ n ]  x[ n − 1] 2 2 If the sampled sequence is constant—unchanging—then the averaged ­sequence is identical to the original. If, however, the original sampled ­sequence is a sinusoid at the Nyquist frequency (e.g., +100, −100, +100, …), then the new averaged sequence is uniformly zero. It should not be a surprise that this averaging operation acts as to smooth the sequence, attenuating high frequencies, and is thus a lowpass filter. The two-point moving average is one of the simplest finite impulse response (FIR) filters. The general feature is that an FIR filter acts only on current and prior samples in the original sequence. Incorporating a greater number of past samples (higher order) can be used to create a lowpass filter that has a sharper cutoff (see chapter appendix).

Infinite Impulse Response Filtering Next consider a filter in which each point is replaced by half of the current input plus half of the last output. 1 1 y[ n ]  x[ n ]  y[ n − 1] 2 2

This filter will also act to generate a smoother version of the input sequence. However, this filter differs from the two-point moving average in the way that it responds to sharp transients. For a single sample point that is an extreme outlier, the two-point moving average will only show the effect of this outlier on two points in the output—the output point y[n] and the next output point y[n + 1]. However, this filter will continue to have its output reflect the transient since it has feedback —the last output is added into the current output. The transient (impulse) will form part of the response for infinite future points, and infinite impulse response (IIR).

63

Engineering Principles

FIR versus IIR Both FIR and IIR filters can be designed for lowpass, highpass, bandpass, and notch (bandstop) filters. There is no simple reason to claim that FIR or IIR is better. In general, FIR filters require much higher order (more computations) to achieve similar “sharpness,” but have the advantage of linear phase alteration and consequently constant group delay (waveforms may be delayed, but will have similar shape). IIR filters that are “sharp” can be constructed with lower order (and faster computation time), but have nonlinear phase distortion and frequency-dependent group delay (i.e., they may alter the phase and the shape of waveforms, with distortion roughly commensurate with the order of the filter). Despite many claims that FIR filters with linear phase are “better,” it can be pointed out that many experts routinely use IIR filters. For instance, in the widely used Matlab suite, the important digital signal processing function decimate() requires a lowpass filter, and currently the default filter type is an IIR Chebyshev filter, and in the past was an IIR Butterworth filter. If the entire sequence to be filtered is available; phase distortion and group delay can be eliminated from both FIR and IIR filters—through the technique of forward-backward filtering. In this method, forward (conventional) filtering with an IIR or FIR results in the expected phase alteration and group delay. These effects are then neatly reversed by then running the same filter backward over the signal produced by forward filtering, yielding zero-phase distortion and no group delay. If it is desired to emulate a traditional analog filter, it is possible to construct an IIR filter that fairly closely mimics the result (using the matchedZ-­transform or impulse invariance techniques). See the chapter appendix for details.

Filtering Examples Most EEG reading software includes digital filters for highpass (the most typical use is to flatten a wavering baseline), lowpass (to attenuate muscle artifact), and notch (tuned to eliminate 60 Hz or 50 Hz, depending on local powerline frequency). As one example, consider the digital highpass filter. The cutoff frequency can usually be selected by the software user. The effect of a highpass filter on components as a function of frequency is shown in Fig. 3.23, for several different cutoff frequencies: 0.1, 1, and 10 Hz. As can be seen, all three eliminate true zero frequency components (DC). However, these filters are

0.1 Hz

1

1 Hz

10 Hz 0.707

0

0

10

128 Frequency (Hz)

Figure 3.23: Digital highpass filter transfer functions for different cutoff frequencies, with a sample rate of 256 Hz (Nyquist of 128 Hz). These are plotted as linear-linear graphs, since these are for digital time-domain filters. Note there is some attenuation above the cutoff frequencies, and incomplete attenuation below the cutoffs.

revealed to not be ideal. The amplitude of components at the cutoff frequency is shown to be about 70% (30% reduced). That is, at the cutoff, there is still a significant amount of amplitude allowed, there is a still a slight reduction at frequencies even above the cutoff, and frequencies below the cutoff still have nonnegligible amplitude contribution. Figure 3.24 shows a practical example of the effect of highpass filters on the traditional calibration signal. Note the connection between the specified cutoff frequency, the time constant, and the length of time for the exponentials to decay toward the baseline. In particular, with the cutoff of 1 Hz, in three time-constants (0.16 second × 3 approximately 0.5 second), the square wave nearly decays to the baseline, nicely fitting with the exponential decay curve in the analog section of this chapter. Finally, Fig. 3.25 shows the effect of highpass filters of different cutoffs on an actual EEG. With an inadequate highpass filter (0.2 Hz), the baseline is seen to waver. With the cutoff set at 1 Hz, the baseline is improved and the focal slowing of the left temporal region is quite apparent. However, at too high of a cutoff, the focal delta slowing is

64

Engineering Principles

calib (50 µV, 0.5 s)

0.01 Hz / 16 s

0.1 Hz / 1.6 s

0.3 Hz / 0.53 s

1 Hz / 0.16 s

3 Hz / 0.05 s

Figure 3.24: Effect of digital highpass filters of different cutoffs on the traditional calibration signal. The cutoff frequency and the time constant for each filter are indicated. Note closely: for the filter with time-constant 0.16 seconds, after three time-constants (0.5 sec), the signal has nearly decayed to baseline.

10 Hz / 0.016 s

misleadingly reduced. The wise electroencephalographer should always be careful about the settings of the digital filters.

Display Considerations A final engineering consideration that impacts the clinical interpretation of the EEG centers on the specifications and limitations of the final display. Traditionally, electroencephalographers have expected to be presented with 10 seconds of EEG in one “page” (30 cm). However, with modern digital sampling rates, there are more sample points than can be presented with current pixel resolutions. Consider: with a sampling rate of 512 Hz, 10 seconds of EEG contains over 5,120 individually sampled points. Current mainstream LCD displays have between 1,280 and 1,920 horizontal pixels. That is, there are at least 2 time points and up to 4 or 5 points per horizontal pixels. This leads to inability to present all the sample points at one time. Low-frequency activity is not much affected, but the problem becomes very apparent for higher frequencies, as shown in Fig. 3.26.

Electrical Safety EEG is an extremely safe procedure, but a small possibility of injury does exist. It is imperative that the technologist and electroencephalographer understand how to minimize this risk. Current is the most important predictor of electrical injury. It can cause pain and burns if applied to the skin (Table 3.2). Seizures can result from certain types of current applied directly to the brain or to the scalp. Current can electroplate irritating metals from intracranial electrodes into brain tissue. Current can even kill, by inducing ventricular fibrillation. Injury risk can be discussed in terms of three groups with different types of relationships to EEG equipment. The safest group comprises persons who are simply near and possibly touching an electrical device but not intentionally connected to it. The second group comprises people with electrodes attached to skin, in the absence of other medical instrumentation. The third group contains patients at higher risk, such as neonates and patients with intravascular catheters or other medical instrumentation.

Engineering Principles

A

Fp1 - F7

Fp1 - F7

F7 - T3

F7 - T3

T3 - T5

T3 - T5

T5 - O1

T5 - O1

Fp2 - F8

Fp2 - F8

F8 - T4

F8 - T4

T4 - T6

T4 - T6

T6 - O2

T6 - O2

Fp1 - F3

Fp1 - F3

F3 - C3

F3 - C3

C3 - P3

C3 - P3

P3 - O1

P3 - O1

Fp2 - F4

Fp2 - F4

F4 - C4

F4 - C4

C4 - P4

C4 - P4

P4 - O2

P4 - O2

Fz - Cz

Fz - Cz

Cz - Pz

Cz - Pz

LOC

LOC

ROC

ROC

EKG

EKG

65

Figure 3.25: Effect of digital highpass filters at different cutoffs on actual EEG with focal slowing. A = highpass at 0.2 Hz, B = highpass at 1 Hz, C = highpass at 5 Hz.

There are several potential sources of dangerous currents that may flow through patients connected to EEG machines and cause them harm. These sources are described as follows.

Improper Grounding Improper grounding can result from a disruption of the ground circuit inside the EEG machine or from the use of a two-prong socket. The cylindrical contact (green wire) on the three-prong plug is the ground contact. Should a short circuit occur in the machine and a current-bearing element make contact with the chassis of the machine, this current should immediately be shunted to the ground contact, because this is the path of lowest resistance. This would quickly blow a fuse or circuit breaker in the EEG machine, which would sense the abnormally high current flow through the now very low resistance of the short circuit. This would not happen immediately,

and some current might flow through the patient even if the proper safety mechanisms were intact during a short circuit. If the machine ground contact is not intact, substantial current (possibly life-threatening) may pass through the patient. EEG machines should never be powered by an inadequately grounded circuit: Three-prong to two-prong adapters must not be used. Machines must always be protected with regulation fuses. Fuses must not be defeated: There is always a reason when a fuse stops working, and it is important to discover that reason rather than subject the patient to an electrical hazard. Hospital-grade power outlets should be used whenever possible for EEG machines (or for any other machines that are to be connected to patients). These outlets are labeled with a green dot and indicate a higher standard of safety and quality of construction than do other outlets (18). A schedule of preventive maintenance on the EEG machine and outlets should be enacted.

B

66

Engineering Principles

Fp1 - F7 F7 - T3 T3 - T5 T5 - O1 Fp2 - F8 F8 - T4 T4 - T6 T6 - O2 Fp1 - F3 F3 - C3 C3 - P3

a few microamperes. Nevertheless, if applied directly to the heart, 0.1 mA could cause ventricular fibrillation. Each wire carrying current to and through the EEG machine induces a magnetic field that, in turn, creates currents in other wires, including neutral and ground wires. These currents are usually shunted directly to the ground contact, but, again, they may be conducted through the patient should some ground malfunction (ground fault) occur. Stray inductances generally are of less magnitude than stray capacitances. According to Hill and Dolan, cited in Cooper et al. (19), maximum leakage currents allowed for the three groups defined earlier are 500 m A for those having casual contact with a medical device; 100 m A for those connected to electrical devices; and 10 m A for the group at high risk.

P3 - O1 Fp2 - F4 F4 - C4 C4 - P4 P4 - O2 Fz - Cz Cz - Pz LOC ROC

C

EKG

Figure 3.25:  (continued)

Leakage Currents Leakage currents arise from two main sources: stray capacitance and stray inductance. Stray capacitance usually arises from wires connected to a wall socket or to the EEG machine power supply. Capacitance is a function of the construction of the power cords and of their length. Nearby wires in a power cord are insulated from each other and therefore can function as a capacitor. AC current flows through the “hot” (black) wire in the cord and induces small capacitive currents in the neutral (white) and ground (green) wires as they alternately charge and discharge with the AC current. This leakage current is usually shunted directly to the ground contact; however, if the ground connection is not properly made, this current may flow through the patient. Extension cords should not be used with EEG machines, because they increase the capacitive current to a potentially dangerous amount. Because wires are inefficient capacitors, capacitive currents from an EEG machine are generally far less than 0.1 mA and may be only

Double-Grounding If a patient is connected to an EEG machine and to another electrical ­instrument, there is probably more than one ground connection. This creates a situation referred to as double-grounding or a ground loop. Because no two ground connections are at identical potential, current may flow from one ground connection to the other through the patient. Figure 3.27 shows a simplified circuit diagram illustrating the problem. There are several potential sources of ground-loop currents. Short circuits in the machine or other circuit faults can deliver massive current to a ground loop. Less dangerous but more common are currents in the ground circuit as a result of stray capacitance and stray inductance. Additional currents may be induced in the ground wires by nearby magnetic fields. In this case, the induced potentials in question are small, but the resistances of the ground paths are also small. Large currents could flow from one ground circuit to another through the patient. Double-grounding is of particular concern in areas where patients are connected to multiple devices, such as intensive care units and operating rooms. It is not unusual to observe patients connected to EEG machines, electrocardiographic monitors, temperature monitors, electric blood pressure cuffs, ventilators, pulse oximeters, warming or cooling blankets, electric beds, arterial and venous catheters, intracranial pressure monitors, and a variety of other hardware. In such circumstances, the presence of a ground loop is virtually guaranteed. The solution is to connect all devices attached to the patient to a common ground connection plugged into the same wall outlet. If necessary, a grounding bar can be used to “gang” together the various ground connections. This provides only one low-resistance ground path (not through the patient) for stray currents.

Engineering Principles

Figure 3.26: Diagram illustrating the issues related to digital sampling rates in excess of visual display resolution. The upper half shows sine waves at escalating frequencies. The lower half shows how these sine waves may be displayed on a computer monitor with the indicated pixel sizes, where there are two digital time steps per horizontal pixel. Note that the higher-frequency activity is blurred, and one is unable to determine the actual frequency of activity.

67

68 Table 3.2

Engineering Principles Estimated Effects of 60-Hz Current on Skin/Externally

15/20 A

Common fuse or breaker opens circuit

2A

Cardiac standstill and internal organ damage

100 mA

Ventricular fibrillation threshold

20 mA

Paralysis of respiratory muscles

16 mA

Maximum current an average man can grasp and “let go”

1 mA

Barely perceptible

0.1 to 0.03 mA from internal/intracardiac catheters have been reported to cause ventricular fibrillation and death (see, e.g., Weinberg et al. [24].) Adapted from National Institute for Occupational Safety and Health. Worker deaths by electrocution: a summary of NIOSH surveillance and investigative findings. Cincinnati, OH: National Institute for Occupational Safety and Health, May 1998.

The EEG technologist must remember the principle of single-grounding when an accessory ground connection is necessary to eliminate 60-Hz interference. All ground connections should travel to one point. If the patient is already grounded by another device, it is not necessary (and is potentially dangerous) to attach another ground connection to the patient. Avoidance

of multiple ground connections, in addition to being a requirement for patient safety, improves recording quality. In high-risk circumstances, such as when patients have intravenous catheters, special equipment with “isolated grounds” should be employed. These boxes use optical isolation or solid-state variable resistors to separate the patient from any currents generated in the EEG machine (18).

Exacerbating Factors Predicting the consequences of an electric shock is difficult, because several factors influence the biological response. In the clinical setting, the most important factor is instrumentation. A transvenous pacemaker or a central venous pressure catheter provides a low-resistance route for stray currents to travel directly to the heart. Ventricular fibrillation can result from currents that would not even be perceived through intact skin. Skin wounds or excessive abrasion with cleaning paste may increase the risk for injury by a given current at those sites. Good general health may be a factor in resisting effects of electric shock, but many hospitalized patients are ill. Table 3.3 summarizes important safety rules in EEG recording.

Regulations +V

Safe

+V

Unsafe!

Figure 3.27: Diagram of the potential problems from a “ground loop.” On the left, if a ground electrode is applied on a patient, and connected to the “ground” of a device, but by poor design or malfunction, this device ground is inadvertently at an electric potential different from actual earth ground, the patient is still safe since there is no conductive path from patient to earth ground. On the right, if more than one device ground is connected to the patient, but any of these device grounds have an electric potential, there is now the possibility that current through the patient can complete the “loop.” If there are electrodes inside the patient (intracardiac electrodes), even a small potential difference may be enough to generate a current of 50 to 100 m A, which could cause ventricular fibrillation (microshock).

Engineering principles for the construction of medical equipment can be found in other texts (20). Standards for the construction and design of electrical equipment have been published by the International ­Electrotechnical Commission (IEC). In particular, IEC 60601 is a series of technical standards for the safety and effectiveness of medical electrical equipment. B ­ esides the general requirements for all medical equipment, specific standards for EEG equipment are published under IEC ­60601-2-26 (“Particular safety of electroencephalographs equipment”). A short checklist of rules to follow to ensure electrical safety during EEG is shown in Table 3.3. Table 3.3

Safety Rules for Performing Electroencephalography (EEG)

Maintain machinery to avoid faulty circuits Always use a grounded (three-prong) plug to power Properly fuse the EEG machine Use one ground connection to patient or ground to a common point Use machines with isolated-ground in high-risk situations

Engineering Principles

CONCLUSIONS Electrical engineering is a vast field, but having a basic understanding of the principles is important for all clinical electroencephalographers, so that they may understand the factors required for faithful recording of cerebral activity, be able to detect and correct sources of artifact, and importantly, supervise the proper and safe use of these electrical machines to avoid harm to patients, technologists, and themselves. New developments are being made to acquire more detail both spatially (with smaller electrodes trying to pinpoint seizure onset) and in frequency (much higher sampling rates). These are exciting times to expand and explore the technology underlying EEG.

Acknowledgments The authors greatly appreciate Liberty Simmons, R.EEG.T for providing the photographs of common clinical electrodes, and constructive comments on the text from the wisdom of a superlative technologist.

69

The authors are greatly indebted to Robert Fisher, MD, PhD, for his wonderful and witty input to this chapter in prior editions. The thoughts expressed in this chapter reflect past experiences and continued learning with many wonderful electroencephalographers throughout training and practice, such as Alan Krumholz, Peter Kaplan, Tom Henry, Chip Epstein, Jacqueline French, John Ebersole, Bob Webber, and others too many to be named. Finally, a large measure of gratitude is due Ernst Niedermeyer for providing teaching and mentoring over the years about the importance of a meticulously recorded and judiciously interpreted EEG. His memory lives on through his many students and their intellectual offspring who populate this exciting, thriving field.

Appendix

There are many textbook references for digital filtering—one that covers many concepts in great detail is Discrete-Time Signal Processing by Oppenheim et al. (4). In general, the framework for discrete or digital signal processing operates on an input sequence, x, denoted: (x0, x1, x2, …) = x[n] Digital filtering is the mathematical conversion of the input x[n] into another new or output ordered sequence y[n] using some formula. In the traditional digital filtering convention, the formula for conversion uses constant coefficients (the coefficients do not change throughout the sequence). The formula can use the current input point and previous inputs, as well as previous outputs. As the simplest example, consider the formula: y[n] = 2·x[n]. That is, the output is proportional to the input sequence but each value is doubled, akin to amplifying the input. For the next simple example, consider y [ n ]  1 x [ n ]  1 x [ n 1] 2 2

With this formula, the output is the average of the current input and the prior input. This will tend to produce a smoother version of the sequence. In digital filtering terms, the high-frequency components are attenuated, but the low-frequency components are preserved, so it is a lowpass filter. The effect of this filter as a function of the frequencies components of the original input can be plotted as in Fig. 3.28. Next consider a formula where the output depends on the current input and prior outputs: 1 3 y[ n ]  x[ n ]  y[ n 1] 4 4

In this instance, the current output is formed from one-quarter of the current input plus three-quarters of the prior output. In general, any formula that includes any of the last M inputs or prior outputs can be written as M

M

k0

k1

y[ n ]  ∑ bk  x[ n − k ]  ∑ ak  y[ n  k ]

This is, in general, called a linear constant coefficient difference (LCCD) equation. The filter defined by this is said to be of order M. Our two-point moving average would be of order 1, and b0 = b1 = 1/2, and ak are zero (nonexistent). In general, if only coefficients bk exist (no ak coefficients), then the filter is a finite impulse response (FIR) filter. If any coefficients ak are nonzero, it is an infinite impulse response (IIR) filter. (Note that by necessity a0 is always 1, since it is the necessary coefficient of y[n] on the left side of the equation.) The frequency response of such a filter is simple in form. The transfer function shows the magnitude of the ratio of the output relative to the input amplitude as a function of frequency. M

H( f ) 

∑b e k0 M

k

k  j 2p f

∑a e k0

k

k  j 2p f

For example, for our two-point moving average: H( f ) 

1 1  j 2p f  e 2 2

A computer plotting program can verify that this formula gives the graph in Fig. 3.28.

71

72

Engineering Principles

filter. When an instantaneous step in voltage is applied to this filter, the voltage across the resistor will follow:

1 Magnitude

Vr  e

t  τ

We now try to emulate this with a digital process. Consider that, given the voltage at some time t0, then the voltage at a later time t1 = t0 + tk is then

0.5

Ve



t1 τ

e



tk τ

e



t0 τ

In fact, for a digital time-step of tk, this holds no matter what initial time, t0. This could be viewed analogously in the discrete time formalism as

0 0

Frequency

  tk  y[ n ]   e τ   y [n21]  

0.5

This could be modeled as 0

0° −

tk

a1 e τ e− t ⋅2p fcut

Phase

k

For instance, for a sampling rate of 256 samples per second, for a highpass filter with a cutoff frequency of 1 Hz, we would calculate that − /4

−45°

a1  e− t k ⋅2p fcut e− 2p /256 e−0.0245 0.97575 If one uses Matlab or Octave to generate a first-order highpass IIR (Butterworth form) filter (these use the convention of scaling frequency so that the Nyquist frequency is 1.0, so the denominator is 128 instead of 256):

− /2 0

Frequency

−90° 0.5

Figure 3.28: Graphs showing the effect of the two-point moving average digital filter on the frequency components, in the magnitude of the contributions to the output and the effect on the phase in the output, as a function of frequency (frequency of 0.5 corresponds to the Nyquist frequency, for any chosen sampling rate). The two-point moving average preserves low frequencies and attenuates high frequencies, and has a linear alteration of phase.

IIR Filter Theory and Construction For examples of how to construct IIR filters, let us now consider the simple analog electrical filter composed of an RC circuit arranged as a highpass

[b, a] 5 butter (1,1.0 / 128.0,‘high’) b a

5 0.9878820.98788 5 1.0000020.97575

We have thus demonstrated how the software arrives at the coefficient for a1. (The b coefficients are constructed for normalization reasons.) The transfer function for this digital IIR highpass filter is shown in Fig. 3.29. Now let us consider how to construct a lowpass filter from the analog electrical RC model. If we choose a cutoff frequency of 32 Hz, with a sampling rate of 256 samples per second, and use the technique of “impulse ­invariance”—effectively trying to mimic the impulse function that would be seen in the analog electrical model—we obtain

a1et k 2p f cut e0.7854 0.45594

73

Engineering Principles

1

1

0.707

0 0 0

1

Frequency (Hz)

128

Figure 3.29: The transfer function for the first-order digital IIR highpass filter with coefficient a1 = −0.975755.

To normalize so that the transfer function is 1.0 at zero frequency (DC): b0  (1 a1 )  0.54406 The transfer function for this lowpass filter is shown as the solid line, with corresponding analog electrical RC filter transfer function as dashed line in Fig. 3.30. This can be seen to act as a lowpass filter, and while it fairly closely mimics the RC filter, both can be seen to pass significant amounts of power at the Nyquist frequency. To completely attenuate frequencies at the Nyquist, another technique called the bilinear transform (BLT) may be used for the construction of IIR filters. This technique essentially warps the entire analog frequency range so that the digital Nyquist is equivalent to infinite analog frequency, and a lowpass filter can result in complete attenuation at the Nyquist frequency. Details of the differences between the methods of impulse invariance (II), matched-Z-transform (MZT), and BLT may be found in digital signal processing textbooks.

Frequency

0.5

Figure 3.30: Solid line shows the transfer function for a digital first-order IIR lowpass filter with a cutoff frequency of 32 Hz and a sampling rate of 256 samples per second, which was constructed using the impulse invariance method. The dashed line shows the effective transfer function for an analog RC lowpass filter with the same cutoff.

FIR Filter Theory and Construction Now we return to FIR filters. Conceptually, the ideal lowpass filter is illustrated in Fig. 3.31. The ideal filter is sometimes called the “brick-wall” filter. The brick-wall is an idealization in the frequency domain. Digital filters operate in the time domain. It is instructive to consider what the form of the ideal brick-wall filter would be in the time domain. The (Inverse) Fourier Transform of the frequency-domain brick wall becomes, in the time domain, the sinc function: sinc(t ) ≡

sin(t ) t

The sinc function is illustrated in Fig. 3.32. Examining it closely, we see that it extends infinitely in both time directions. This would imply that to create the brick-wall filter, we would have to be able to apply this filter to future points in the sequence. This could not be done if we wished to filter in real-time, that is, apply the filter to the sequence point-by-point as it enters the digitizer buffer.

74

Engineering Principles

This would be known as an acausal filter. We also note that the sinc function does not become absolutely zero at any specific point away from time 0. A true brick-wall filter would require filtering all points of the sequence. In order to implement the brick-wall filter with a finite and causal filter, we truncate the sinc function and shift it back. Importantly, this shift causes a group delay in the filter of half the order of the filter. As the order of the filter grows large, this delay can become quite significant. Typically, the order (M) of the filter is chosen to be even, leading to an odd number of coefficients (M + 1). In Fig. 3.33, we show the effect of truncation, with an

“Pass” (low frequencies)

1

1

“Cut” (high frequencies)

0 0

0.2 Frequency

0.5

Figure 3.31: The ideal lowpass filter, sometimes called a “brick-wall” filter. In the perfect case, this filter perfectly passes all frequency components below the cutoff (here chosen to be 20% of the sample rate or 40% of the Nyquist frequency), and perfectly eliminates all frequency components above the cutoff. 0

–20

–12

−4

1

4

12

20

0.3

0.4

Nyquist

Time

1

0 0 0.1 −16

−12

−8

−4

0

4

8

12

16

Figure 3.32: The sinc function, which is the (Inverse) Fourier Transform of the “brick-wall” filter.

0.2

Frequency

Figure 3.33: A truncated sinc function (FIR with order 32) and its transfer function in the frequency domain. Note the ripple in both the passband and the stopband.

75

Engineering Principles 1

1 Hann

0

−20

−12

−4

4

12

20

0.3

0.4

Nyquist

Time 0 0

N-1

1

Figure 3.34: The Hann windowing function. By formula: 0.5 − 0.5 cos[2pn/(N − 1)]

order M = 32 filter with a cutoff frequency at 40% of the Nyquist. Note the significant amount of ripple in both the passband and the stopband. For the most part, the ripple (in frequency domain) is a consequence of the abrupt truncation. Higher order can be used to reduce the ripple in the frequency domain, but this is a trade-off (A) in computational time and (B) more prolonged ringing in the time domain when transients (impulses) are encountered. One technique for reducing the frequency-domain ripple is to apply a window to the sinc function, where the window is nearly 1 in the middle, and tails to 0 on either side. There have been many windowing functions developed in engineering. One of the simplest windows in the Hann window is shown in Fig. 3.34. Applying the Hann window to our truncated, order 32 filter is shown in Fig. 3.35. As you can see, the Hann window smoothing has resulted in less ripple in both the passband and the stopband, at the cost of a slower roll-off. There are a multitude of more complicated techniques to determine the coefficients for FIR filters. Some are straightforward, while other more complicated methods require algorithms to repeatedly adjust coefficients until some criterion is reached. As one example, the method of Parks-McClellan (21) iteratively adjusts the coefficients until the transfer function meets certain criteria, such as maximum allowed ripple and transition bandwidth. Further details of FIR filter design may be found in digital filtering textbooks.

0 0.1

0.2

Frequency

Figure 3.35: The FIR filter with Hann windowing function. In the top panel, the truncated sinc function of Fig. 3.33 has been multiplied by the Hann window of Fig. 3.34. The corresponding transfer function for this windowed FIR filter is shown in the bottom panel. Note the reduction in the ripple, but also the slower roll-off in the transition band.

Digital Filters in EEG The actual details of the digital filters used in major commercial EEG vendors’ software are not immediately apparent. Most contain highpass, lowpass, and notch filters, and have settings to select the cutoff frequencies.

76

Engineering Principles

However, the cutoffs are typically only allowed to be in a preselected set of frequencies—for example, 0.1, 0.5, 1.0, 1.6 Hz. The type of filter (IIR or FIR) is not typically named or described. With careful inspection of known and well-described signal sequences, some inferences can be made about the most typical types of filters. Figure 3.36 shows screen captures of calibration signals viewed with three different software viewers (A, B, and C). The calibration signal is a square-wave with 1 second per phase, and amplitude of ±25 m V. For all of the viewers, the filters were selected to be the same,

namely, highpass cutoff of 1.6 Hz and lowpass cutoff of 70 Hz, and sensitivity of 7 m V per mm. All show that the square waves decay to baseline due  to  the highpass filters eliminating zero frequency (DC). The decay is monotonically exponential in both Fig. 3.36A and B, and nearly identical. One of these is known to be an IIR filter of first-order, and it can be inferred that the other is as well. (An FIR filter to generate a similar monotonic exponential decay would require an extremely high order, which is impractical.) However, in Fig. 3.36C, the decay has an overshoot (below and beyond the

A

B

C

Figure 3.36: Screen captures of a calibration signal viewed with three different software viewers, with the same ostensible filter settings (highpass cutoff of 1.6 Hz; lowpass cutoff of 70 Hz), illustrating that these viewers use different kinds of digital filters. The calibration signal is a square wave of 1 second per phase, ±25 m V. The tracings show that after filtering, (A) and (B) are essentially identical (with (B) known to use a highpass first-order IIR filter), while in (C) the square waves decay toward baseline similarly, but have overshoot, and thus likely use a second-order IIR filter for the highpass filter.

Engineering Principles

baseline), and from this it can be inferred that the highpass filter in this case is probably IIR of higher order (likely second-order). While most EEG viewing software hides the implementation details and has little documentation of the digital filter details, there is one major commercial vendor that has settings to change the filter types. The permitted types are IIR filters, FIR of Chebyshev type, and the obscurely termed Walraven FIR. This filter is based on the publication of an algorithm (in FORTRAN source code) for an FIR filter in a computer society magazine from the 1980s (22). This is, in turn, an implementation of a filter design technique published in 1974 by J. Kaiser, a Bell Labs researcher (23). In this algorithm, the user essentially selects a cutoff frequency and a maximum permitted ripple, and the algorithm estimates the required order of the FIR filter, and some parameters of the window function. This is an example of software permitting some additional control over the filter configuration, with a few modifiable parameters, but not requiring extensive expertise in the design of digital filters. In summary, digital filters are a powerful tool in the processing and interpretation of EEG. While they may seem complex and mysterious at first, a basic understanding of the principles enables the clinical electroencephalographer to anticipate the effects of digital filters (both for good and bad) in the interpretation of digital EEG.



REFERENCES









1. Purcell E. Electricity and magnetism, 2nd ed. Cambridge, England: Cambridge University Press, 2011. 2. Horowitz P, Hill W. The art of electronics, 2nd ed. Cambridge, England: Cambridge University Press, 1989. 3. Butterworth S. On the theory of filter amplifiers. Exp Wireless Wireless Eng 1930;7:536–541. 4. Oppenheim AV, Schafer RW, Buck JR. Discrete-time signal processing, 2nd ed. Upper Saddle River, NJ: Prentice Hall, 1999. 5. American Radio Relay League. ARRL handbook for radio communications. Newington, CT: American Radio Relay League, 2012.



77

6. Cooper R. Electrodes. Am J EEG Technol 1963;3:91–101. 7. Mirsattari SM, Lee DH, Jones D, et al. MRI compatible EEG electrode system for routine use in the epilepsy monitoring unit and intensive care unit. Clin Neurophysiol 2004;115:2175–2180. 8. Wyler AR, Ojemann GA, Lettich E, et al. Subdural strip electrodes for localizing epileptogenic foci. J Neurosurg 1984;60:1195–1200. 9. Spencer SS, Spencer DD, Williamson PD, et al. The localizing value of depth electroencephalography in 32 patients with refractory epilepsy. Ann Neurol 1982;12:248–253. 10. Maynard EM, Nordhausen CT, Normann RA. The Utah Intracortical Electrode Array: a recording structure for potential brain-computer interfaces. Electroencephalogr Clin Neurophysiol 1997;102:228–239. 11. Viventi JA, Kim DH, Vigeland L, et al. Flexible, foldable, actively multiplexed, high-density electrode array for mapping brain activity in vivo. Nat Neurosci 2011;14(12):1599–1605. 12. Donoghue J. Connecting cortex to machines: recent advances in brain interfaces. Nat Neurosci 2002;5(Suppl):1085–1088. 13. American Clinical Neurophysiology Society Guideline 8: guidelines for recording clinical EEG on digital media. J Clin Neurophysiol 2006;23(2):122–124. 14. Kemp B, Olivan J. European Data Format ‘plus’ (EDF+), an EDF alike standard format for the exchange of physiological data. Clin Neurophysiol 2003;114:1755–1761. 15. Brinkmann BH, Bower MR, Stengel KA, et al. Large-scale electrophysiology: acquisition, compression, encryption, and storage of big data. J Neurosci Methods 2009; 180:185–192. 16. Gotman J. Automatic detection of spikes and seizures. J Clin Neurophysiol 1999;​ 16(2):130–140. 17. Claasen J, Hirsch LJ, Kreiter KT, et al. Quantitative continuous EEG for detecting ­delayed cerebral ischemia in patients with poor-grade subarachnoid hemorrhage. Clin Neurophysiol 2004;115:2699–2710. 18. Tyner F, Knott J, Mayer WJ. Fundamentals of EEG technology. New York, NY: Raven Press, 1983. 19. Cooper R, Osselton JW, Shaw JC. EEG technology. London, England: Butterworth and Co, 1969. 20. Geddes LA, Baker LE. Principles of applied biomedical instrumentation. New York, NY: Wiley, 1989. 21. Parks SW, McClellan JH. Chebyshev approximation for nonrecursive digital filters with linear phase. IEEE Trans Circuit Theory 1972;19(2):189–194. 22. Walraven R. Digital filters. In: Proceedings of the DECUS (Digital Equipment Corporation Users Society); 1984; fall issue. 23. Kaiser JF. Nonrecursive digital filter design using the I0-sinh window function. In: Proceedings of the 1974 IEEE International Symposium on Circuits and Systems; 1974:20–23. 24. Weinberg DI, Artley JL, Whalen RE, et al. Electric shock hazards in cardiac catheterization. Circ Res 1962;11:1004–1009.

4

Recording Techniques SAURABH R. SINHA

Introduction Cerebral Generators of EEG Potentials Electrode Placement Recording Conventions and Sensitivity/Filter Settings Issues Related to Digital Recording Polarity Conventions Calibration and Sensitivity Filters

INTRODUCTION Electroencephalography (EEG) measures electrical fields generated by neuronal activity by recording the amplified potential differences between electrodes placed on the scalp, directly on the cortex (e.g., with subdural electrodes), or within the brain (e.g., with depth electrodes). EEG analysis attempts to draw conclusions about the nature, location, and configuration of the generator of the EEG patterns; this helps to determine if the activity is normal or abnormal and epileptiform or nonepileptiform. When conclusions about the EEG signal are combined with other information (patient symptoms and signs, imaging results, etc.), diagnostic conclusions and management decisions can be made. Scalp electrodes are traditionally placed according to the International 10–20 system. Assessing the spatial distribution of the recorded electrical field requires the orderly arrangement of multiple channels, termed a

78

Montages Unpaired, Paired-Group, and Paired-Channel Montages Display Conventions Referential and Bipolar Montages Conclusion References

montage. Different types of montages and different arrangements of channel within a montage can help to highlight certain types of activity. In this chapter, we will briefly discuss the generators of the EEG signal, how these signals are recorded (electrodes, amplifier characteristics, and filters), and how they are routinely displayed and analyzed (montages).

CEREBRAL GENERATORS OF EEG POTENTIALS EEG signals reflect neuronal activity at a macroscopic scale; they represent the summated electrical activity generated by large groups of neurons (105 or more) (1). Due to the need for summation of the activity of a large population of neurons, the main source of EEG potentials are cortical neurons, which are arranged in layers in roughly the same orientation beneath the cortical surface. They are aligned with their long axis perpendicular to the cortical surface. They are activated by synapses located on their soma and dendrites. The postsynaptic

Recording Techniques

79

potentials produced in the cortical neurons are the major contributors to the EEG signal. Although other types of neuronal and glial activity may contribute to the EEG, neuronal action potentials do not because their very brief duration prevents significant summation across a population of neurons. All generators of scalp-recorded electrical activity behave like a dipole, a generator with a positive and a negative pole. For example, synaptic activity at a point on a pyramidal cell produces an active ionic current as a result of the opening of a pore in a neurotransmitter receptor or an ion channel. All current flows in a closed loop; in the case of synaptic activation of neurons, the current loop is completed by passive flow through the neuronal membrane at a distant, relatively inactive site. Thus, synaptic input at one end of a pyramidal cell causes current flow through the cell membrane whose direction at the surface is opposite to its direction in the depth. This produces polarization shifts (voltage changes) in the extracellular space in opposite directions at the surface and in the depth. Thus, synaptic activity on a pyramidal neuron produces an extracellular, radially oriented dipole. Because both the type of postsynaptic potential (excitatory versus inhibitory) and its location (somatic or proximal dendrites versus distal dendrites) impact the direction of the dipole, it is not possible to determine whether a given polarization change results from excitatory or inhibitory synaptic activity (Fig. 4.1).

ELECTRODE PLACEMENT A variety of systems have been used in the past to locate electrodes on the scalp for EEG recording. The General Assembly of the International Federation of Clinical Neurophysiology (IFCN) recommended a specific system of electrode placement for use in all laboratories (Fig. 4.2A) (2, 3). This system is known as the International 10–20 system. Specific measurements are made based on bony landmarks (nasion, inion, and preauricular point) to determine the placement of electrodes. From these landmarks, specific measurements are made and then 10% and 20% of a specified distance is used as the interelectrode interval. This enables replication consistently over time, between technologists, and among different laboratories. Uniform interelectrode distances and symmetric placement of electrodes over the left and right hemispheres are important for EEG interpretation, especially for detecting asymmetry between the two hemispheres. The American Clinical Neurophysiology Society (ACNS) has recommended using a minimum of 21 electrodes from the International 10–20 system. Each electrode is named by a letter/number combination. The letter designates the approximate anatomical location (“F” for frontal, “T” for temporal,

Figure 4.1: The direction of the dipole generated by the activation of a pyramidal neuron depends on both the type of activation (excitatory vs. inhibitory) and its location on the neuron (distal dendrites vs. proximal dendrites or soma). A: An excitatory input in the distal apical dendrites causes active (solid line) inward current flow there; passive current flow (dashed lines) completes the circuit with outward current closer to the soma. In the extracellular space, this generates a relative negative potential in dendrites and a relative positive potential near the soma. A dipole that is surface negative is generated. B: An inhibitory input in the distal dendrites generates a surface positive dipole. C: An inhibitory input near the soma generates a surface negative dipole (the same as an excitatory input in the distal dendrites). D: An excitatory input near the soma generates a surface positive dipole (the same as an inhibitory input in the distal dendrites).

80

Recording Techniques

Figure 4.2: A: Electrode nomenclature of the  21 most commonly used electrodes from the International 10–20 system. B: Electrode nomenclature of the International 10–20 system. Electrodes in black are those with different names in the 10–10 system compared to the 10–20 system.

“Fp” for frontopolar, “C” for central, “P” for parietal, and “O” for occipital). Odd-­numbered electrodes are placed over the left side of the head and even-­numbered electrodes are placed over the right. Although less consistent, smaller numbers are closer to the midline (with “z,” indicating zero, being in the midline) and larger numbers are more lateral. In 1991, the American Electroencephalographic Society (the forerunner of the ACNS) added nomenclature guidelines for 75 electrode positions along five anterior-posterior planes, lateral to the midline chain of 11 specific sites (Fig. 4.2B). In addition, there are eight coronal chains (four anterior and four posterior to the chain of 13 electrode sites between the earlobe electrodes along the midline through the Cz electrode). Several electrodes are named differently in the International 10–20 system and the extended nomenclature (the 10% system or International 10–10 system). Electrodes T3 and T4 in the 10–20 system are referred to as T7 and T8 in the expanded system, and T5 and T6 are referred to as P7 and P8. For neonates, fewer electrodes are used and the precise number is somewhat variable from laboratory to laboratory. The ACNS has suggested using at minimum Fp1, Fp2, C3, Cz, C4, T3, T4, O1, O2, A1, and A2 (4). In certain situations, additional electrodes can be applied to increase the yield of EEG recordings. These include sphenoidal and T1 and T2 electrodes in patients with suspected temporal lobe epilepsy. In direct comparisons, T1 and T2 electrodes have been found to be as effective as sphenoidal electrodes and superior to nasopharyngeal electrodes and electrodes from

the International 10–20 system (F7, F8, A1, and A2) for detecting interictal discharges (5, 6). T1 and T2 are placed 1 cm above the line connecting the external auditory meatus to the outer canthus at a point one-third of the distance from the former to the latter. Additional electrodes from the 10% system, especially those near the midline, are often useful for assessing activity near the midline, for example a medial frontal focus.

RECORDING CONVENTIONS AND SENSITIVITY/ FILTER SETTINGS Issues Related to Digital Recording Voltage measurement always requires a reference: The voltage at any point can only be defined with respect to the voltage at another point, and there is no absolute value for voltage. In an electronic circuit, the reference point is usually a fixed voltage referred to as the ground. In EEG, the voltages or potentials on the scalp are measured with respect to another point, usually on the scalp or at least on the body. The output of an EEG amplifier is the amplified difference between the voltages at the points connected to its two inputs. For an analog EEG system, the two points are selected and physically connected to the amplifier at the time of the recording and cannot be changed once the data is acquired.

Recording Techniques

For digital EEG recording, all data are measured with respect to a common reference electrode. The signal is then digitized via an analog-to-digital converter (ADC). Digitization refers to the discretization of the analog input signal (which is continuous) into specific voltage and time steps. The sampling rate (in Hertz) of the ADC indicates the number of times per second that data from each channel is sampled and recorded. By the Nyquist theorem, a sampling rate of N cannot faithfully record frequencies higher than N/2. The resolution of the ADC indicates the number of discrete values that it can produce over the allowed range of values of the analog input. For example, an 8-bit ADC has 28 − 1 (=255) voltage intervals. Thus, if the input voltage range is −5 to +5 mV (=10 mV), the ADC voltage resolution is approximately 39 µV (10 mV/255). Fortunately, most ADCs in use today have resolutions of 16 bits or even higher. After acquisition, for display and analysis, the data from two electrodes are mathematically combined into a “derivation.” For example, for two electrodes, E1 and E2, the data will be collected with respect to the common reference electrode (REF, also called the system reference). These are labeled as E1 − REF (the voltage at E1 with respect to the voltage at REF) and E2 − REF. The derivation E1 − E2 can then be obtained by subtraction as the common reference then cancels out:

E1 − E2 = (E1 − REF) − (E2 − REF)

Thus, theoretically, the choice of the common reference electrode is i­ rrelevant as it mathematically cancels out. Practically speaking, REF is usually still an electrode placed on the scalp to minimize large extracerebral potentials that could saturate the amplifiers. A common choice for system reference is an extra electrode placed midway between Cz and Pz. Not only does digital recording permit post hoc recombination of ­electrodes to obtain different derivations, it also allows for postprocessing of EEG signal by digital filtering. Some minimal filter settings are necessary when recording the raw signal. At the lower end, this is necessary to avoid artifacts related to electrode polarization and drift. At the upper end, this is necessary to attenuate frequencies above the sampling capabilities of the ADC.

Polarity Conventions EEG signals are measured using a differential amplifier, which amplifies the difference between the voltages at its two inputs. These inputs are designated input terminal 1 and input terminal 2, and have historically been referred to as “G1” and “G2.” By convention, an upward deflection occurs when input 1 is more negative than input 2 or when input 2 is more positive

81

than input 1. The polarity convention also specifies that the 10–20 electrode names, separated by a dash, designate electrodes connected to the two inputs of a differential amplifier (“electrode 1” − “electrode 2”), with “electrode 1” connected to input 1 and “electrode 2” connected to input 2.

Calibration and Sensitivity The gain of an amplifier specifies how much it amplifies the input voltage; for example, an amplifier with a gain of 100 will produce a 200-mV output when supplied with a 2-mV input. For EEG amplifiers, the gain is usually expressed as sensitivity, which is expressed as microvolts per millimeter. This specifies the input voltage necessary to produce a given vertical deflection on the screen or paper. For digital recordings, the actual gain of the EEG amplifiers is usually not adjusted; instead, the gain is usually preset to a level that amplifies the typical range of input voltages to make maximal use of the input range of the ADC. This serves to optimally use the full resolution of the ADC. If the input voltage is significantly higher than expected, it will be amplified to values outside the range of the ADC and these values will be clipped (i.e., set to the maximum or minimum value that the ADC can handle). Changes to the sensitivity made by the technologist during acquisition will only change how the digitized signal is processed for display; the data being collected is unchanged. This is obviously also the case for adjustments to sensitivity during review of the data. In analog EEG systems, adjustments made to the sensitivity actually modified the gain of the amplifiers. In these systems, the technologist played a much more active role in selecting an appropriate sensitivity to collect and display data optimally. Calibration of the equipment is an integral part of insuring faithful ­reproduction of the recorded signal; this is especially critical in analog systems where the output device (pen on paper) is an integral part of the hardware and can impact calibration. Typically, a small voltage (approximately 50 µV) is used at a sensitivity of 7 to 10 µV per mm; this is usually applied as a train of square waves. The output signal produced in all channels is assessed to make sure that an appropriate and equivalent deflection is produced. An additional biological calibration (usually by connecting Fp1–O2 to all amplifiers) is performed to ensure that all amplifiers respond equally and correctly to a variety of frequencies and not just the square wave used for calibration. A second calibration is also recommended at the end of EEG recording, with all of the sensitivities and filter settings that were used during the recording. In digital systems, calibration is still important, but is much more reliable and stable. Calibration for digital systems may be performed only once during a recording; amplifier gains and direct current

82

Recording Techniques

offsets can be corrected automatically by the software to yield the same response characteristics across all channels. Calibration at different frequencies is a built-in function, and thus manual biological calibration is no longer necessary (7). When EEG is performed to assess electrocerebral inactivity, there is a requirement to calibrate with a 2- to 5-µV calibration signal.

Filters Voltage changes related to neurophysiological activity occur over a broad range of frequencies, from at least 0.01 Hz at the low end to several hundred Hertz at the high end (8). Recordings covering this entire band of frequencies would best reproduce the biological activity. However, there are biological and technical factors that make recording at extremely high and low frequencies difficult. On the technical side, recording at low frequencies are difficult due to electrode polarization and drift in amplifiers and other electronics. Recording of higher frequencies is technically difficult due to the need for higher sampling rates and electronic/random noise; furthermore, for analog pen/paper system, the physical characteristics of the pen limited the maximum frequency that could be recorded. On the biological

side, artifacts related to movement and sweat, among others, are prominent at low frequencies. Muscle artifact and the significantly lower amplitudes of high-frequency activity are limiting. For these reasons, scalp EEG recordings have traditionally been limited to a band from 0.5 to 70 Hz or 100 Hz. Analog and digital filters are used to select which frequencies are recorded and displayed. For all EEG systems, analog filters are used to select the frequency range to be recorded. In the case of digital EEG systems, further filtering can be performed using digital filters (software) applied to the digitized signal. The high-frequency filter, also referred to as the low-pass filter, selectively attenuates higher frequencies. A commonly used high-frequency filter setting is 70 Hz. An ideal 70-Hz high-frequency filter would remove all activity with a frequency higher than 70 Hz with no effect on frequencies below 70 Hz. A real-world filter has minimal impact on frequencies far below 70 Hz and maximally attenuates frequencies well above 70 Hz. The transition between these two zones is referred to as the roll-off: the steeper the roll-off, the more closely it resembles an ideal filter. Using a high-frequency filter setting of 35 Hz or even lower can significantly lower artifact related to muscle activity; however, it can also significantly alter the biological EEG signal and attenuate/blunt epileptiform discharges and make apparent artifacts appear physiological (Fig. 4.3).

Figure 4.3: A: Ictal pattern for a left hemisphere seizure displayed with LFF of 0.5 Hz, HFF of 100 Hz. Although the rhythmic theta activity over the left hemisphere can be seen, it is marred by muscle artifact. B: The same EEG shown with HFF of 50 Hz. The muscle artifact is reduced and the left hemisphere theta is much more apparent. C: The same EEG with HFF of 15 Hz. The muscle artifact is significantly reduced; however, now the right hemisphere activity, mainly due to muscle artifact, could be mistaken for cerebral activity, especially over the temporal regions.

Recording Techniques

Figure 4.3: (continued)

83

84

Recording Techniques

Figure 4.4: A: EEG with prominent L temporal irregular slowing shown with LFF of 0.5 Hz, and HFF of 70 Hz. B: Same EEG shown with LFF of 5 Hz, which largely attenuates the left temporal slowing.

Recording Techniques

The low-frequency filter, also referred to as the high-pass filter, selectively attenuates lower frequencies. Frequencies below the low-frequency filter setting are selectively attenuated and frequencies above are not. Low-frequency filters are sometimes described by their time constant (τ). For a square wave input, the time constant is defined as the time it takes the voltage to decline by 63% from its peak. For a simple filter (consisting only of resistors and capacitors), a time constant of 1 second corresponds to a low-frequency filter setting of 0.16 Hz:

fcutoff 

1 2pt

Judicious use of low-frequency filter may allow better appreciation of ­high-frequency activity in a record dominated by low-frequency physiological activity or artifact; however, it may also lead to losing relevant information, like focal slowing (Fig. 4.4). Another issue related to filtering is that of phase advance and phase delay. This refers to the fact that a filter not only affects the amplitude of a waveform, it also shifts the position in time. For a periodic signal, this corresponds to a change in the phase of the signal. A phase advance indicates that the peak of the signal is moved to an earlier time—this is the impact of applying a low-frequency filter. A phase delay indicates that the peak of the signal is delayed to a later time—this is the impact of applying a highfrequency filter. For typical filter settings, the phase shifts are typical only a few milliseconds. As long as the filters are applied uniformly to all channels, the phase shifts will have little impact on EEG interpretation. However, this can be a more significant issue if different filter settings are used for different channels. It is also a bigger issue for evoked potentials, where a small shift in latency can be significant (Fig. 4.5).

85

combinations. This is equivalent to filtering the signal across space. Just like high-frequency and low-frequency filters that have the potential to accentuate activity of interest but can also distort the signal, montages can highlight certain types of activity while distorting others. The type of montage and the arrangement of electrodes within that montage impact what spatial information is well-preserved and what is lost. For example, a longitudinal montage preserves information about spatial relationships best in the sagittal plane, while a transverse montage preserves information about spatial relationships best in the coronal plane. Montages may be classified as: (1) unpaired, electroanatomical pairedgroup or paired-channel (referring to the relative arrangement of electrodes within a montage), (2) referential, bipolar or laplacian (referring

Montages The EEG signal is generated by a complex, three-dimensional structure, the cerebral cortex. Significant spatial information is lost by the act of ­recording this signal using electrodes placed at a handful of locations on the scalp. While adding more electrodes can increase the spatial information, it cannot fully compensate for the lack of information from deeper points in the brain. Spatial information is further degraded by the need to display how the signal varies over time. Thus, spatial information is typically only conveyed within the orderly arrangement of different electrode derivations, the montage. Voltages recorded from multiple scalp locations simultaneously are displayed by combining the signal from different electrode pairs or

Figure 4.5: Example of phase shifting. The center panel shows a bipolar ­temporal chain of electrodes with a left temporal sharp wave (near isoelectric at F7–T3) with LFF of 0.5 Hz and HFF of 100 Hz. The dashed line marks the time of the peak. Increasing the LFF to 5 Hz (top panel) shifts the peak earlier. On the other hand, decreasing the HFF to 15 Hz (bottom panel) shifts the peak later.

86

Recording Techniques

to derivations used), and (3) longitudinal, transverse or circular (referring to arrangement of channel along sagittal, coronal or axial planes, respectively).

Unpaired, Paired-Group, and Paired-Channel Montages In unpaired montages, channels are arranged in anatomical neighboring sequence: for example, sequentially from front to back or left to right. An example is a transverse, bipolar montage where electrodes are arranged in coronal planes from anterior to posterior, with electrodes going from left to right in each plane. In paired-group montages, electrodes from homologous areas are placed adjacent to each other. For example, left and right temporal are placed together as are left and right parasagittal chains. Paired-group montages are always longitudinal as the brain does not have homologous regions in the coronal direction. A longitudinal bipolar montage is an ­example. In paired-channel montages, channels from homologous brain areas are paired. Typically, pairs of electrodes from the left and right are then subgrouped together in longitudinal lines, with unpaired midline electrodes placed separately. Paired-group and paired-channel montages allow for ­better detection of asymmetry between the two hemispheres.

Display Conventions In montages, electrodes are always arranged in anterior-to-posterior ­sequ­ence. For longitudinal montages, the chains proceed from front to back; for transverse montages, chains are displayed in a front-to-back sequence. In the US and Canada, “left-over-right” convention is typically used. For longitudinal montages (unpaired or paired), the left-sided electrodes are placed before the right-sided electrodes or electrode groups. For transverse montage, each chain proceeds from left to right. A “right-over-left” convention is used routinely in Europe and is recommended by the IFCN (3).

Referential and Bipolar Montages As mentioned previously, the main purpose of arranging EEG data as ­montages is to aid in localization of the recorded activity. Localization ­requires that multiple measurements of the voltage distribution be made. For longitudinal or transverse chains of electrodes, one is trying to identify the electrode that registers the maximum voltage. Precise localization in two dimensions requires identifying the maximally involved electrode in both the sagittal and coronal planes.

Referential Montages In a referential montage, the voltage at each electrode is displayed with ­respect to a common electrode known as the reference. Note that this ­reference is not the same as the system reference used for recording the EEG signal (see  earlier). Ideally, the common electrode is inactive with r­espect to the activity of interest (i.e., does not contain any of this activity). This is not possible in real life, as the reference electrode is always active to some degree. So, the goal is to identify a relatively inactive electrode. Commonly used reference electrodes include A1 and A2 (“ipsilateral ear” and, rarely, “contralateral ear”), A1 + A2 (“linked ear”), Cz, a balanced noncephalic reference such as the neck-chest region, and the average reference. The ­average reference is derived electronically (analog) or mathematically ­(digital) by ­averaging the voltage in all active scalp electrodes and using it as the reference (9). Sometimes electrodes containing a prominent artifact (e.g., frontal electrodes with eye movement artifact) or prominent signal (e.g., electrodes with ­focal ­slowing) are excluded from the calculation of the average ­reference; this avoids contamination of the reference electrode. With appropriate selection of a reference (where the reference is significantly less active than the other electrodes), a referential montage clearly displays the polarity and field distribution of potential being measured. ­Selection of the reference is crucial. If the selected reference has a significant active contribution, it will distort the recorded activity. For example, using an ipsilateral ear reference in a patient with temporal discharges may actually minimize the activity in nearby temporal electrodes. A major disadvantage of referential recordings is that no single reference choice is optimal for all situations; for example, a Cz reference may be ideal for looking at abnormal activity over the temporal regions but would be an extremely poor choice for abnormal activity near the midline or any type of activity during sleep (Fig. 4.6).

Bipolar Montages In bipolar montages, both inputs are connected to active recording electrodes and no single electrode is common to both inputs. Bipolar montages link sequential pairs of electrodes to form chains in the longitudinal ­(sagittal), transverse (coronal), or circular (axial) directions. In these chains, a single electrode becomes common to two adjacent channels; for example, A–B, B–C, C–D, and D–E. In such a chain, the site of maximal voltage within a potential field becomes a phase reversal, that is, simultaneous deflections in two channels sharing a common electrode with the maximal voltage will be in opposite directions. The direction of the phase reversal

Recording Techniques

Figure 4.6: A: A bifrontal discharge, likely a vertex wave, shown on a longitudinal bipolar montage. Phase reversals are seen at F3 and F4. B: The same discharge shown on a transverse bipolar montage, with phase reversals at Fz, Cz. Fp1–Fp2 is nearly isoelectric. In combination with A, this discharge can be localized to the midline at approximately the level of Fz. C:  The same discharge shown on an average reference montage. Maximum amplitude is seen at Fz, consistent with localization from the bipolar montages. However, note the effect of the large discharge contaminating the average reference: a deflection in the opposite direction at relatively inactive electrodes like O1 and O2. D: The same discharge shown on a Cz reference montage. Note how the use of an active electrode as a reference leads to mislocalization: the discharge appears to be a surface positive discharge located near the back of the head (maximal at O1 and O2).

87

88

Recording Techniques

Figure 4.6: (contiuned)

89

Recording Techniques

(deflections coming together for a surface negative peak and diverging for a surface positive peak) assists in determining polarity. If the voltage associated with a potential field is equal in two adjacent electrodes, they are equipotential. A channel involving these two electrodes will show cancellation and no deflection will be seen in this channel. Localization by phase reversal is only possible if the bipolar montage fully encompasses the voltage peak. If the peak occurs at the end of a chain, for example a spike located at O1, no phase reversal will be seen. Adding an additional electrode, for example a suboccipital lead, may reveal a phase reversal and aid in localization. Bipolar montages are ideally suited to highlighting activity that changes rapidly in space, that is, has a fairly localized field. By subtracting the ­potential in adjacent electrode, only those signals that are changing b ­ etween the two locations will be accentuated. Another way of looking at it is that a bipolar montage provides the first spatial derivative of the EEG signal: signals that are changing rapidly in space are best seen. On the other hand, potentials that have a broad field, that is, those that are not changing rapidly in space, will be relatively equipotential between adjacent electrodes and will cancel out in a bipolar montage. A corollary to this is that the amplitude in a channel of a bipolar montage does not reflect the actual amplitude but only how rapidly it is changing in space. Unlike a referential montage, a bipolar montage does not require an inactive electrode or careful selection of a reference. However, the level of activity at individual electrodes cannot be directly compared using a bipolar montage, only inferred indirectly.

CONCLUSION EEG is the most readily available and clinically relevant tool for ascertaining cerebral activity, especially as it relates to seizures. Proper interpretation and utilization requires an understanding of the biological basis of the EEG, including the inherent limitations in such a macroscopic measurement of complex neuronal activity. The technical aspects of recording, displaying, and analyzing the EEG signal have been simplified over time. The use of digital recording systems that allow for reformatting and reprocessing of the data and that are often engineered with capabilities well beyond older systems has allowed us to be less aware of potential technical issues. These issues were often just under the surface of older systems whose capabilities were sometimes just barely adequate for their purpose. However, even with today’s systems, under special circumstances (e.g., recording high-frequency oscillations), these technical issues can still impact the EEG.

REFERENCES

Selection of Montages As discussed earlier, each type of montage has advantages and disadvantages. Optimal EEG recording combines both referential and bipolar montages. Comprehensive assessment of the potential field requires combining these methods intelligently. With analog equipment, where data could not be reformatted after collection, it was essential to record the data using a variety of montages. The ACNS recommended that a standard record should include, at minimum, a longitudinal bipolar montage, a transverse bipolar montage, and a referential montage. The specific montages to be used are not specified; however, each laboratory should have a set of standard montages. The existence of such standards montages does not replace the active role of a trained technologist who may select or set up special montages or electrodes to highlight a particular activity.



1. Gloor P. Neuronal generators and the problem of localization in electroencephalography: application of volume conductor theory to electroencephalography. J Clin Neurophysiol 1985;2(4):327–354. 2. Jasper HH. The ten-twenty electrode system of the International Federation. ­Electroencephalogr Clin Neurophysiol 1958;10:371–373. 3. Jasper HH. The ten-twenty electrode system of the International Federation. In: International Federation of Societies for Electroencephalography and Clinical Neurophysiology: recommendations for the practice of clinical neurophysiology. Amsterdam, the ­Netherlands: Elsevier, 1983:3–10. 4. American Clinical Neurophysiology Society. Guideline two: minimum technical standards for pediatric electroencephalography. 2006 [cited November 11, 2013]. Available from http://www.acns.org/pdf/guidelines/Guideline-2.pdf. 5. Binnie CD, Marston D, Polkey CE, et al. Distribution of temporal spikes in relation to the sphenoidal electrode. Electroencephalogr Clin Neurophysiol 1989;73(5):403–409. 6. Sadler RM, Goodwin J. Multiple electrodes for detecting spikes in partial complex ­seizures. Can J Neurol Sci 1989;16(3):326–329. 7. Wong PKH. Digital EEG in clinical practice. Philadelphia, PA: Lippincott-Raven, 1996. 8. Vanhatalo S, Voipio J, Kaila K. Full-band EEG (FbEEG): an emerging standard in ­electroencephalography. Clin Neurophysiol 2005;116(1):1–8. 9. Goldman D. The clinical use of the “average” reference electrode in monopolar ­recording. Electroencephalogr Clin Neurophysiol 1950;2(2):209–212.

5

Normal Adult EEG WILLIAM O. TATUM IV

Introduction Parameters of Recording Bandwidths in the Normal Adult EEG Alpha Mu Rhythm Beta Theta Delta

Introduction EEG has made a unique and valuable contribution to our understanding of the brain’s electrical function. The utility of EEG lies in its application to a number of neurological conditions associated with altered brain function. To understand what constitutes an abnormal EEG, a solid foundation of what constitutes the normal boundaries is essential. EEG represents a three-dimensional cerebral source graphically represented in a two-­dimensional plane. Most of the human cortex lies deep beneath the scalp surface where EEG has limited capabilities in assessing subcortical abnormalities. The origin of cerebral potentials is based upon the intrinsic electrophysiological properties of the nervous system (1). Scalp EEG is

90

Activating Procedures Sleep Architecture Benign Variants of Uncertain Significance Variations of the Normal EEG Misinterpreted Normal EEG Conclusion References

routinely used in the assessment of normal EEG, though three-dimensional source localization is possible (2). To be believable as a source of electrical activity that emanates from the brain, an appropriate polarity (Table 5.1) and accompanying electrophysiological field must exist (3). The correct interpretation involves synthesizing complex waveforms from different bandwidths generated by a cerebral source to determine whether the combined result is normal. Because the routine interpretation of EEG as normal is most routinely a qualitative skill, the result is based upon one’s knowledge, training, and experience. No single parameter or collective group of features present in the EEG can identify a tracing as normal. It is the overall pattern of waveform organization and orderly progression over time that best represents the “brain waves” as normal. Recognizing common features

91

Normal Adult EEG TABLE 5.1

Polarity with Respect to Electrode Pairs: Input Terminal 1 (aka G1) and Input Terminal 2 (aka G2) versus Surface Waveform Deflection (Arrows) Electrode 1

Electrode 2

Negative





Positive





By convention, when G1 is relatively electronegative, the deflection is up.

of normal EEG, the infrequent variations seen with a normal record, and the benign variants of uncertain significance that are considered normal is an essential task to ensure appropriate interpretation. The changes that normally occur in the EEG with advancing age are also important to recognize. From the time of birth to old age, the appearance of waveforms that are inappropriate for other ages may occur. Understanding the temporal relationship (Table 5.2) that occurs with time is essential for an ageappropriate assessment of a normal EEG. Misinterpreting an EEG usually occurs from the overinterpretation of a normal record as abnormal. This can result in the incorrect treatment of patients and potentially lead to serious consequences (4). TABLE 5.2

Essential Features of the Normal EEG Involving Waveform Measurements and Temporal Characteristics

Waveform Metric (units)

Temporal Characteristics

Frequency 500 Hz ­(traditionally 1–70 Hz) Polarity (positive vs. negative) Amplitude (μV)

Symmetry

Morphology (mono-/polymorphic, regularity, rhythmicity, special features)

Reactivity (eye opening/closing, ­extremity movement, visual scanning, activation techniques)

Duration (milliseconds)

Appearance (rare, intermittent, ­occasional, frequent, continuous)

Distribution (anterior-posterior, hemispheric, regional, focal)

Synchrony

Parameters of Recording Frequency, polarity, amplitude, morphology, duration, and distribution are elements that govern the description and reflect the identity of individual waveforms. Symmetry, reactivity, appearance, and synchrony constitute the temporal characteristics of all the waveforms. The absolute waveform metrics and their temporal relationship is relative to all other components in the EEG. Analyzing the EEG requires a systematic approach. In the routine interpretation of scalp EEG, recordings are usually performed with standardized parameters. A visual display of 30 mm per second is utilized to reflect the “paper speed.” Amplifier sensitivity is initially recorded at 7 μV per mm, though are adjusted based upon requirements of the individual. Lower amplitudes necessitate a higher amplifier gain (i.e., higher sensitivity setting) that is often seen normally in younger patients. Filter settings are routinely set between 1 and 70 Hz. Adjustment of the low-frequency filter (to a higher cutoff frequency) may be required if artifact is problematic such as with sweating. High-frequency filter modification (to a lower cutoff frequency) may be necessary if myogenic artifact is prominent, and a selective “notched” filter at 60 Hz may be required in cases where electrically hostile environments exist. Standard settings are used in the examples throughout the text unless otherwise noted. The state and age of the patient may be known to the electroencephalographer prior to interpretation, though most typically initially interpret the EEG independent of background information. This is to facilitate a final interpretation of the EEG in the proper clinical context of recording. Physiologic functions such as eye opening and closure, visual scanning, and activating procedures are components that are utilized in analyzing normal EEG. In adults, an interpretation should be performed without prior knowledge of the patient’s history to prevent bias during interpretation. Recognizing an adult EEG that has normal background organization with characteristic waveforms makes even the information about the age and state nonessential. The EEG of neonates, infants, and children under 3 years of age, on the other hand, normally has electrocerebral activity that varies substantially based upon age and state. A  purists’ approach to analysis of the adult EEG is to develop a working impression without any information. Indicating the patient’s level of alertness is important. An example would be to misinterpret a normal background rhythm less than 8 Hz in drowsiness as abnormal. Every routine EEG should include at least one montage that uses the longitudinal bipolar, reference, and traverse bipolar montages. Newer modified electrode systems use electrode placement with more closely spaced

92

Normal Adult EEG

electrodes in a 10–10 system. High-density arrays and complex methods of signal analysis, such as independent component analysis and phase congruency, may increase the sensitivity of scalp EEG further to refine the location of the generators in normal and in abnormal EEG (5,6). Changes in montage, sensitivity, and display speeds as well as post hoc filtering, spike and seizure software applications, quantitative EEG analysis, and artifact rejection may clarify an EEG that is not otherwise ­interpretable (7).

Bandwidths in the Normal Adult EEG Normal EEG produces waveforms represented by a spectrum of frequencies (Table 5.3). Routine interpretation of the routine scalp EEG typically involves bandwidths of frequencies that lie between 1 and 30 Hz. Alpha, beta, theta, and delta frequencies comprise the major frequencies for analysis. Most amplitudes range between 10 and 100 μV. Invasive EEG (iEEG) reveals a similar spectrum of frequencies as those recorded at the scalp (8). “Ultra-frequencies” may be observed below 1 Hz (infraslow activity) and above 30 Hz (high-frequency oscillations), depending upon the technical limitations of the recording. Altered morphology, higher amplitudes (5- to 15-fold on iEEG), and shorter durations contrast the bandwidths noted with scalp EEG (9).

Alpha The alpha rhythm is the starting point to assess the background activity when interpreting the EEG. The normal range for the alpha frequency in the occipital region in adults is 8 to 13 Hz. The alpha rhythm is best TABLE 5.3

Frequencies in the EEG and Their Bandwidths

Frequency

Bandwidth (Hz)

Ultraslow Delta Theta Alpha Beta Gamma Ripples Fast ripples Ultra-fast ripples

0–0.3 3.5–13–30 >30–80 >80–250 >250–500 500–1,000

observed during relaxed wakefulness and normally has a side to side difference of less than 1 Hz. The frequency of the alpha rhythm is coupled with cerebral blood flow and falls when blood flow is compromised. The alpha rhythm has been evaluated in relationship to cognitive and mental function (10,11). ­Reactivity and a posterior location distinguish the alpha rhythm from an alpha frequency. In majority of people, the occipital region is the site of maximal activity for the alpha rhythm (Fig. 5.1). While a posterior dominant rhythm first becomes apparent at 3 to 4 months of age, an alpha rhythm of 8 Hz is achieved in the majority by 3 years of age and is 10 Hz by age 10. The frequency may transiently increase after eye closure for an instant to lie in the beta bandwidth. However, this rhythm is seen on eye closure and is reactive, independent of the beta frequency and is known as alpha squeak. The alpha rhythm is best observed during periods of relaxed wakefulness and often during periods of relative physical and mental inactivity. In normal EEG, a dominant alpha rhythm is present over the posterior aspect of the head and lies within the 8- to 13-Hz bandwidth. Measures of the alpha rhythm vary over time and are age-dependent, reflecting the changing physiology of brain function with aging. Only 1% of a healthy young population has an alpha rhythm of 8 Hz. While an 8-Hz alpha rhythm is considered normal for an adult, this broad interpretation should be liberally applied in adults greater than 65 years of age but more restrictive in younger patients. With advancing age, the alpha rhythm frequency steadily declines, though should remain stable with normal frequencies of 8 Hz even into late life. Therefore, even in those greater than 80 years of age, an alpha rhythm of 7.5 Hz should be viewed critically as an abnormality. Reactivity is a characteristic of the alpha rhythm. Attenuation of the alpha activity is seen in response to eye opening. With eye closure, there is a return of higher voltage alpha frequencies in the occipital region denoting the reaction to visual input. This “block” is temporary and dependent upon a stimulus that may eliminate or attenuate the voltage of the alpha rhythm. Reactivity may be seen with other forms of stimulation, including cognition, though the most powerful remains eye opening and closure. Complete unresponsiveness of the alpha rhythm is not expected in normal EEG. In approximately one-fourth of normal adults, the alpha rhythm is poorly visualized or it may be only intermittently visualized. Amplitudes also vary between individuals, over the time of recording and lifetime. Low-amplitude recordings may be present, and in less than 10% of patients, voltages of less than 15 μV are seen. The alpha rhythm is distributed maximally in the

Normal Adult EEG

93

FIGURE 5.1: A 10-Hz alpha rhythm in an awake 45-year-old woman with spells dissipates when the eyes are opened (gray box). A return of ­alpha is noted demonstrating reactivity upon eye closure (black arrow). Note the increase in frequency immediately following eye closure known as alpha squeak.

occipital regions, and shifts anteriorly during drowsiness. Asymmetries of the alpha rhythm are best assessed at two posterior electrode sites (i.e., parietal and occipital), with both bipolar and referential recordings to ensure consistency. Higher amplitudes are usually noted over the right hemisphere between 20 and 60 μV in the majority of individuals from peak to peak. Using the P4-O2 derivation, normal amplitudes of 15 to 45 μV are typically encountered. Higher amplitudes are noted with slower alpha rhythms. Voltage asymmetries of greater than 50% should be regarded as abnormal. When the left side is greater than the right by greater than 35%, it is more likely to be associated with abnormality.

The morphology of the alpha rhythms is typically sinusoidal, though at times may normally appear with a sharp surface negative component ­especially in younger patients. It may show a waxing and waning “beating” pattern due to varying amplitudes. These alpha “spindles” may be noted in some people, while in others, this pattern is absent. The spindle pattern, however, is a continuous pattern unlike the proper use of the term spindle that is associated with sleep. The cerebral field involved with the alpha rhythm involves the posterior quadrant of the head. When it extends into the central regions, it may be confused with mu. If the temporal region is involved, it may be misinterpreted as an abnormal

94

Normal Adult EEG

epileptiform discharge (4). Some temporal alpha may be detected in the presence of a breach rhythm and is believed to be independent of the posterior alpha and mu rhythms, and has therefore been referred to as the “third rhythm” (12). Unilateral failure of the alpha rhythm to attenuate reflects an ipsilateral abnormality involving the posterior head region (Bancaud’s phenomenon). Paradoxical alpha occurs when alertness results in the presence of alpha and drowsiness does not. Alpha harmonics are normal features of the alpha where frequencies that are exactly two times the alpha (supraharmonics) or one-half the frequency (subharmonics) are seen (Fig. 5.2). No sex-dependent differences in the alpha rhythm is seen in the EEG (13). The relationship of the menstrual cycle relative to EEG has been proposed, with acceleration of the alpha frequency and amount reduced in the premenstrual phase, with slowing of the alpha and amount increased during menses (14). A relationship to body temperature has been shown with acceleration

of the alpha rhythm during rising body temperatures. Cardiac pacemakers have been shown to increase the a­ lpha frequency by more than 1 to 2 Hz, probably due to increased cardiac ­output and ­improved cerebral blood flow (15). Drugs that are used even as prescribed may result in slowing of the alpha rhythm. In cancer patients exposed to hyperthermia of 41°C for treatment purposes, slowing of the alpha rhythm is encountered with depression of the electrocerebral activity (16).

Mu Rhythm The mu (aka motor) rhythm is a normal, centrally located, reactive, alpha frequency. It has characteristics that are similar to the alpha rhythm but it is physiologically distinct. However, it has been referred to as the “somatosensory alpha rhythm” with a restricted central location approximating the pre- and postcentral gyrus to approximate the sensorimotor cortex in relaxed wakefulness. Mu is typically asymmetric and asynchronous with

FIGURE 5.2: The alpha rhythm noted in the occipital region in two-thirds of patients (black arrows) with subharmonics below (A) during drowsiness that is half the frequency of the normal 10-Hz alpha rhythm. In (B), an alpha rhythm is less than 13 Hz occasionally, not due to a supraharmonic in a 24-year-old after cardiac transplantation.

Normal Adult EEG

interhemispheric shifting. It is well localized to the C3 and C4 electrode locations, and is associated with a characteristic physiologic reactivity. The frequency of mu is 10 Hz and frequencies less than 8 Hz are probably abnormal. The mu rhythm has been identified in the first year of life and is believed to reflect the early developmental necessity of evolving motor function during early life. The reactivity of mu may mimic the central spread of the alpha rhythm. Though in contrast to the alpha rhythm that is eliminated by eye opening, the mu rhythm is eliminated by movement of a contralateral limb (17). The reactivity of a mu rhythm highlights the impact of cognition on EEG desynchronization on motor function (18). The reactivity of mu has been shown in amputees attempting to initiate movement of an amputated limb. Furthermore, just the thought of a movement has been enough to create a bilateral blocking response. Light tactile stimulation to create a sensation and enhancement by photic stimulation has been demonstrated to compound the idea of the sensory-motor function of mu rhythm to implicate both sensory and visual integration (14). Quantitative EEG (qEEG) has suggested that the mu is separate from the alpha rhythm and that the rolandic beta is separate from a harmonic of the mu rhythm (19). The 10-Hz frequency of the mu rhythm is most often encountered. It is believed that

FIGURE 5.3: Transverse bipolar montage demonstrating a mu rhythm (gray arrows) as a “fragment” of the occipital alpha rhythm (black arrow). Not the similar frequencies but different localization, reactivity with eye closure, and mu assymmetry.

95

this frequency reflects the sensory cortical function, while 20-Hz reactive beta frequencies producing a notched appearance in the mu rhythm have been suggested to have a greater correlation with motor function (20). During yoga, more than one element of the mu rhythm may coexist in the same individual (14). Beyond the similar frequency that may slow with age, the distribution of mu is typically asymmetric and asynchronous. It may appear in brief bursts and prolonged runs. It can be differentiated from alpha rhythm ­despite the similarities by the asymmetry of the mu rhythm and the symmetry of an ­alpha rhythm. The amplitude of the mu rhythm is usually lower than the amplitude of the alpha rhythm. Morphological differences may be noted (Fig. 5.3). The morphology is arciform with a comb-like appearance. The prevalence of mu has varied depending on the author, though with ­computerized EEG, mu can be identified in most. When a lesion in the central region exists, this may diminish the presence of the corresponding mu rhythm or reduce its reactivity (21). When mu is strictly unilateral, an ipsilateral breach rhythm or rolandic cortical disturbance should be considered. When it is persistent, unreactive, and associated with focal slowing, mu-like ­frequencies are abnormal.

96

Normal Adult EEG

In epilepsy management, the mu rhythm has served as a reference point for treatment using biofeedback (14).

Beta Beta rhythms represent frequencies that are more than 13 Hz. They are also referred to as “fast” frequencies. Beta is a frequency that is found in most normal adult EEG. Beta activity normally increases during drowsiness, light sleep, and with mental activation. Beta is a common frequency and is most often observed within the 18- to 25-Hz bandwidth in normal people with a voltage of less than 25 μV. Most frequencies higher than 80 Hz are beyond the detection limits of scalp EEG, with beta represented by frequencies greater than 13 Hz and less than 30 Hz. A frontal-central field is most often encountered and may even appear as a brief train that exceeds 30 Hz when sleep is achieved. Rarely, a posterior dominant rhythm may appear to be present that lies in the beta frequency bandwidth. With the faster beta frequencies, voltages beyond 25 μV in amplitude are abnormal and amplitudes greater than 30 μV are unusual. Persistently reduced hemispheric voltages

of greater than 50% suggest a cortical gray matter abnormality within the hemisphere having the lower amplitude. Lesser asymmetries may simply reflect normal skull thickness differences. Vertex beta should prompt a search for cortical dysfunction affecting the frontal region. Focal central beta has been related to a mu rhythm demonstrating normal reactivity to motor or tactile stimuli. A skull defect may produce a breach rhythm that produces focal beta activity (Fig. 5.4). Benzodiazepines, barbiturates, and chloral hydrate are potent activators of generalized “fast activity.” Hypnotic agents usually occur within the 18- to 25-Hz bandwidth and may reach greater than 50 μV occupying greater than 50% of the tracing. However, enhanced beta activity alone is without specificity to interpret an EEG record as abnormal (14,22). Not unexpectedly, with the prevalence of sedatives and tranquilizer use (Fig. 5.5) in patients with psychiatric disorders, the beta frequencies may have a greater representation in this population, though it is of no clinical significance (23). Some frontal beta may appear close to 30 Hz, especially during sleep. Even higher frequencies in the gamma range and above may occur with normal physiological functions that are involved in cognition (24,25).

FIGURE 5.4: Left frontal breach rhythm (gray rectangle) with prominent beta activity at F3 electrode derivation and admixed 5- to 6-Hz theta seen at a focal region following a craniectomy for resection of a cavernous vascular malformation.

Normal Adult EEG

97

FIGURE 5.5: AP longitudinal bipolar montage with diffuse 20- to 25-Hz beta activity associated with alprazolam and oxycodone (gray arrow). Note the photoelectric response (black arrow) with the infraorbital electrode/eye movement monitor in association with the photic stimulation.

Theta Theta frequency is composed of 4- to less than 8-Hz activity in the EEG. Approximately one-third of normal young adults may show intermittent 6to 7-Hz theta rhythms during wakefulness (Fig. 5.6A). It appears as rhythmic activity that is maximal in the frontal or frontocentral head regions, typically with amplitudes that are less than 15 μV, though varying amplitude and morphologies may be encountered. The presence of anterior predominant rhythmical theta may persist into to adolescence and young adulthood. The appearance of frontal theta can be facilitated by heightened emotions, concentration, and during mental activities such as problem solving (26). Theta activity is normally enhanced by hyperventilation (HV). It may also occur in bursts during drowsiness and sleep transitioning. It is normally limited in the adult during the waking record. The appearance of theta in the EEG frequently occurs as a normal developmental feature, yet it may persist until 25 years of age (14). It was felt to be an intermediate bandwidth

between alpha and delta frequencies, with continuous focal theta suggesting nonspecific abnormality (Fig. 5.6B). A long-standing recommendation in considering abnormal theta has been to consider it abnormal in awake patients if it is present in persistent focal bursts or runs. While this quantification seems helpful, its presence is more likely related to the state and age of the patient, rather than reflecting a specific abnormality. In the elderly, intermittent bitemporal 4- to 5-Hz activity or even theta with a lateralized predominance (usually left > right) may occur in about one-third of the asymptomatic normal seniors (14,27).

Delta Delta frequencies consist of activity that is less than 4 Hz. Delta may be considered normal in young children less than 10 years of age as well as in normal elderly individuals. In children, posterior slow waves of youth (PSWY) are posterior delta frequencies that are prominent during childhood persisting

98

Normal Adult EEG

FIGURE 5.6: Bipolar montage demonstrating (A) normal 6- to 7-Hz theta prior to eye closure (black arrow) in an awake 22-year-old during mental concentration and (B) right temporal continuous theta (gray arrow) in a 35-year-old with right temporal lobe epilepsy.

through young adulthood (Fig. 5.7A). Between ages 6 and 12 years, arrhythmic delta complexes may be intermixed with the alpha rhythm in the occipital head regions. They typically have a duration of 200 to 400 milliseconds and are of moderate voltage (i.e., 1 mechanism, such as generalized and ­focal features (or epileptic and nonepileptic)

Normal Adult EEG

123

FIGURE 5.35. A: Normal EEG following a breakthrough seizure after amygdalohippocampectomy and subsequent phase 2 and lateral neocorticectomy. Note the normal EEG (10–20 system of electrode placement) despite the extensive resection of the left temporal lobe on the MRI in (B).

interpretation do not confirm a clinical diagnosis of epilepsy. It is only through an unbiased interpretation and firm knowledge of normal EEG that we may ensure optimal treatment for our patients.

REFERENCES

1. Krauss GL, Abdallah A, Lesser R, et al. Clinical and EEG features of patients with EEG wicket rhythms misdiagnosed with epilepsy. Neurology 2005;64(11):1879–1883. 2. Plummer C, Wagner M, Fuchs M, et al. Dipole versus distributed EEG source localization for single versus averaged spikes in focal epilepsy. J Clin Neurophysiol 2010;27(3): 141–162. 3. Tatum WO, Dworetzky BA, Freeman WD, et al. Artifact: recording EEG in special care units. J Clin Neurophysiol 2011;28(3):264–277. 4. Benbadis SR, Tatum WO. Overintepretation of EEGs and misdiagnosis of epilepsy. J Clin Neurophysiol 2003;20(1):42–44. 5. Iriarte J, Urrestarazu E, Artieda J, et al. Independent component analysis in the study of focal seizures. J Clin Neurophysiol 2006;23(6):551–558.



6. Lantz G, Spinelli L, Seeck M, et al. Propagation of interictal epileptiform activity can lead to erroneous source localizations: a 128-channel EEG mapping study. J Clin Neurophysiol 2003;20(5):311–319. 7. Gao J, Yang Y, Sun J, et al. Automatic removal of various artifacts from EEG signals using combined methods. J Clin Neurophysiol 2010;27(5):312–320. 8. Tatum WO, Vale FL, Anthony KU. Epilepsy surgery. In: Husain AM, ed. A practical approach to neurophysiologic intraoperative monitoring. New York, NY: Demos Medical Publishing, LLC, 2008:283–301. 9. Tatum WO. Invasive EEG. In: Greenfield LJ, Geyer JD, Carney PR, eds. Reading EEGs: a practical approach. Philadelphia, PA: Lippincott Williams & Wilkins, 2010:224–246. 10. Nowak SM, Marczynski TJ. Trait anxiety is reflected in EEG alpha response to stress. Electroencephalogr Clin Neurophysiol 1981;52(2):175–191. 11. Shagass C. An attempt to correlate the occipital alpha frequency of the electroencephalogram with performance on a mental ability test. J Exp Psychol 1946;36:88–92. 12. Niedermeyer E. The “third rhythm”: further observations. Clin Electroencephalogr 1991;22(2):83–96. 13. Veldhuizen RJ, Jonkman EJ, Poortvliet DC. Sex differences in age regression parameters of healthy adults—normative data and practical implications. Electroencephalogr Clin Neurophysiol 1993;86(6):377–384.

124

Normal Adult EEG

14. Chang BS, Schomer DL, Niedermeyer E. Normal EEG and sleep: adults and elderly. In: Schomer DL, Niedermeyer E, eds. Niedermeyer’s electroencephalography: basic principles, clinical applications, and related fields, 6th ed. Philadelphia, PA: Lippincott ­Williams & Wilkins, 2011:183–214. 15. Kellaway P. Orderly approach to visual analysis: elements of the normal EEG and their characteristics in children and adults. In: Ebersole JS, Pedley TA, eds. Current practice of clinical electroencephalography, 3rd ed. Philadelphia, PA: Lippincott Williams & Wilkins, 2003:100–159. 16. Dubois M, Sato S, Lees DE, et al. Electroencephalographic changes during whole body hyperthermia in humans. Electroencephalogr Clin Neurophysiol 1980;50(5–6):486–495. 17. Kuhlman WN. Functional topography of the human mu rhythm. Electroencephalogr Clin Neurophysiol 1978;44(1):83–93. 18. Pfurtscheller G, Lopes da Silva FH. Event-related EEG/MEG synchronization and ­desynchronization: basic principles. Clin Neurophysiol 1999;110(11):1842–1857. 19. Pfurtscheller G. Central beta rhythm during sensorimotor activities in man. Electro­ encephalogr Clin Neurophysiol 1981;51(3):253–264. 20. Salmelin R, Hari R. Spatiotemporal characteristics of sensorimotor neuromagnetic rhythms related to thumb movement. Neuroscience 1994;60(2):537–550. 21. Pfurtscheller G. Rolandic mu rhythm and assessment of cerebral functions. Am J EEG Technol 1986;26:19–32. 22. Tatum WO, Husain AM, Benbadis SR, et al. Normal adult EEG and patterns of uncertain significance. J Clin Neurophysiol 2006;23(3):194–207. 23. Giannitrapani D, Kayton L. Schizophrenia and EEG spectral analysis. Electroencephalogr Clin Neurophysiol 1974;36(4):377–386. 24. Curio G. Ain’t no rhythm fast enough: EEG bands beyond beta. J Clin Neurophysiol 2000;17(4):339–340. 25. Draguhn A, Traub RD, Bibbig A, et al. Ripple (approximately 200-Hz) oscillations in temporal structures. J Clin Neurophysiol 2000;17(4):361–376. 26. Takahashi N, Shinomiya S, Mori D, et al. Frontal midline theta rhythm in young healthy adults. Clin Electroencephalogr 1997;28(1):49–54. 27. Silverman AJ, Busse EW, Barnes RH. Studies in the processes of aging: electroencephalographic findings in 400 elderly subjects. Electroencephalogr Clin Neurophysiol 1955;7(1):67–74. 28. Klass DW, Brenner RP. Electroencephalography of the elderly. J Clin Neurophysiol 1995;12(2):116–131. 29. Liedorp M, van der Flier WM, Hoogervorst EL, et al. Associations between patterns of EEG abnormalities and diagnosis in a large memory clinic cohort. Dement Geriatr Cogn Disord 2009;27(1):18–23. 30. Oken BS, Kaye JA. Electrophysiologic function in the healthy, extremely old. Neurology 1992;42(3, Pt 1):519–526. 31. Katz RI, Horowitz GR. Electroencephalogram in the septuagenarian: studies in a normal geriatric population. J Am Geriatr Soc 1982;30(4):273–275. 32. Yamatani M, Konishi T, Murakami M, et al. Hyperventilation activation on EEG recording in childhood. Epilepsia 1994;35(6):1199–1203. 33. Takahashi T, Chiappa K. Activation methods. In: Schomer D, Lopes da Silva F, eds. Electroencephalography: basic principles, clinical applications and related fields, 6th ed. Baltimore, MD: Lippincott Williams & Wilkins, 2010:215–238. 34. Dionisio J, Tatum WO. Triggers and techniques in termination of partial seizures. ­Epilepsy Behav 2010;17(2):210–214. 35. Benbadis SR, Siegrist K, Tatum WO, et al. Short-term outpatient EEG video with induction in the diagnosis of psychogenic seizures. Neurology 2004;63(9):1728–1730.

36. Radtke RA. Sleep disorders: laboratory evaluation. In: Ebersole JS, Pedley TA, eds. Current practice of clinical electroencephalography, 3rd ed. Philadelphia, PA: Lippincott ­Williams & Wilkins, 2003:803–832. 37. Tatum WO. Normal EEG. In: Tatum WO, Husain AM, Benbadis S, et al, eds. Handbook of EEG interpretation. New York, NY: Demos Medical Publishing, LLC, 2008:1–50. 38. Rechtschaffen A, Kales A. A manual of standardized terminology, techniques and scoring system for sleep stages of human subjects. Bethesda, MD: US National Institute of Neurological Diseases and Blindness Neurological Information Network, 1968. 39. Tatum WO, Spector A. Physiologic pseudoseizures: an EEG case report of mistake in identity. J Clin Neurophysiol 2011;28(3):308–310. 40. A glossary of terms most commonly used by clinical electroencephalographers. Electroencephalogr Clin Neurophysiol 1974;37(5):538–548. 41. Klass DW, Westmoreland BF. Nonepileptogenic epileptiform electroencephalographic activity. Ann Neurol 1985;18(6):627–635. 42. Tharp BR. The 6-per-second spike and wave complex: the wave and spike phantom. Arch Neurol 1966;15(5):533–537. 43. Hughes JR. Two forms of the 6/sec spike and wave complex. Electroencephalogr Clin Neurophysiol 1980;48(5):535–550. 44. Marshall C. Some clinical correlates of the wave and spike phantom. Electroencephalogr Clin Neurophysiol 1955;7(4):633–636. 45. Silverman D. Phantom spike-waves and the fourteen and six per second positive spike pattern: a consideration of their relationship. Electroencephalogr Clin Neurophysiol 1967;23(3):207–213. 46. Saito F, Fukushima Y, Kubota S. Small sharp spikes: possible relationship to epilepsy. Clin Electroencephalogr 1987;18(3):114–119. 47. Westmoreland BF, Reiher J, Klass DW. Recording small sharp spikes with depth electroencephalography. Epilepsia 1979;20(6):599–606. 48. Reiher J, Lebel M. Wicket spikes: clinical correlates of a previously undescribed EEG pattern. Can J Neurol Sci 1977;4(1):39–47. 49. Gelisse P, Kuate C, Coubes P, et al. Wicket spikes during rapid eye movement sleep. J Clin Neurophysiol 2003;20(5):345–350. 50. Westmoreland BF, Klass DW. A distinctive rhythmic EEG discharge of adults. Electroencephalogr Clin Neurophysiol 1981;51(2):186–191. 51. Thomas P, Migneco O, Darcourt J, et al. Single photon emission computed tomography study of subclinical rhythmic electrographic discharge in adults. Electroencephalogr Clin Neurophysiol 1992;83(3):223–227. 52. O’Brien TJ, Sharbrough FW, Westmoreland BF, et al. Subclinical rhythmic electrographic discharges of adults (SREDA) revisited: a study using digital EEG analysis. J Clin Neurophysiol 1998;15(6):493–501. 53. Ciganek L. Theta-discharges in the middle-line—EEG symptom of temporal lobe epilepsy. Electroencephalogr Clin Neurophysiol 1961;13:669–673. 54. Okada S, Urakami Y. Midline theta rhythm revisited. Clin Electroencephalogr 1993;24(1):6–12. 55. Edwards J, Kutluay E. Patterns of unclear significance. In: Schomer D, Lopes da Silva F, eds. Niedermeyer’s electroencephalography: basic principles, clinical applications, and related fields, 6th ed. Philadelphia, PA: Lippincott Williams & Wilkins, 2011: 267–280. 56. Tatum WO, Dworetzky BA, Schomer DL. Artifact and recording concepts in EEG. J Clin Neurophysiol 2011;28(3):252–263. 57. Tatum WO. EEG interpretation: common problems. Clin Pract 2012;9(5):527–538.

6

Normal Pediatric EEG: Neonates and Children ROBERT R. CLANCY • A.G. CHRISTINA BERGQVIST • DENNIS J. DLUGOS • DOUGLAS R. NORDLI Jr.

Introduction to the Pediatric Electroencephalogram Visual Analysis State and Continuity Organization and Principal Components Interhemispheric Symmetry and Synchrony Special Features Abnormalities Overview of Electroencephalographic Ontogeny 24 to 29 Weeks of Conceptional Age 30 to 32 Weeks of Conceptional Age 33 to 34 Weeks of Conceptional Age 35 to 36 Weeks of Conceptional Age 37 to 40 Weeks of Conceptional Age 41 to 44 Weeks of Conceptional Age 45 to 46 Weeks of Conceptional Age Early Infancy: 46 Weeks to 1 Year Late Infancy: 1 to 2 Years Childhood: 2 to 10 Years Adolescence Composition of Electroencephalographic Background Activity (Named Patterns) Neonatal Infancy and Childhood

Types and Significance of Abnormal Pediatric Electroencephalographic Backgrounds Excessive Discontinuity Sharp Electroencephalographic Transients Electrographic Neonatal Seizures and Clinical Correlations Sharp Waves and Spikes in the Older Infant and Child Activity of Drowsiness, Arousal, and Sleep Classification of the Neonatal Electroencephalographic Background and Its Implications An Organized Approach to Visual Analysis of the Electroencephalogram Continuity and State Concordance Gradient and Principal Components Symmetry and Synchrony Special Features Abnormalities Final Words References

125

126

Normal Pediatric EEG: Neonates and Children

INTRODUCTION TO THE PEDIATRIC ELECTROENCEPHALOGRAM The previous chapter covered the normal adult EEG and introduced the major necessary concepts to analyze EEG. Pediatric EEG interpretation builds on these essential skills but requires one additional element: a thorough knowledge of EEG ontogeny, or the orderly maturational changes in pediatric EEG. Children are not small adults, and at first glance their EEGs could not appear more distinct, but the same techniques used to systematically analyze the adult EEG can be applied to children once one appreciates the organizational themes of pediatric EEG. Accordingly, this chapter will not recapitulate elements from the prior chapter, but instead will focus on the development of pediatric EEG from the premature to adolescence. To make the discussion as practical as possible, the pediatric EEG will be described in the same order in which it is typically analyzed, noting first the continuity and then, in turn, the organization of the principal components, the interhemispheric symmetry and synchrony of these components, and the presence of various special features that often serve as signposts of maturation. Finally, abnormalities are noted and described. The discussion will begin with the most premature babies and will continue through to adolescence. Before the visual analysis (VA) is discussed, there are a few technical matters that must be addressed. The international 10–20 system of electrode placement is modified for neonatal EEG recording because of neonates’ small head size and the relative lack of EEG activity in the extreme frontopolar regions (Fig. 6.1a,b). The standard neonatal montage includes electrodes Fp3 (halfway between Fp1 and F3), Fp4 (halfway between Fp2 and F4), C3, C4, T3, T4, 01, 02, Fz, Cz, Pz, A1, and A2 (1). If the earlobes are too small, mastoid leads (M1 and M2) may be substituted. Fp3 and Fp4 electrodes are used because electrographic background activity and frontal physiological sharp waves are better visualized there than at the usual frontopolar locations (Fp1 and Fp2) (2). Beyond 6 weeks of age or a head circumference of 40 cm, many laboratories will use the same complement of electrodes found in adult tracings. At the same time, most EEGers in North America will change the display to 30 mm per second, again resembling adult tracings. The one piece of extraneous information that is needed to properly interpret a pediatric EEG is the age of the patient (although in many circumstances the skilled EEGer can develop a fairly accurate estimate). In neonates and young infants, the most relevant is the conceptional age. This is determined by adding the number of weeks since birth to the estimated gestational age (EGA). The EGA is the age of the fetus since conception, calculated from the time of the mother’s last menstrual period to the day of the infant’s birth. Newborns

Figure 6.1a: Intrauterine MRI of fetus 30  weeks of estimated gestational age. Notice the relatively simple convolutional markings of the frontal lobes in comparison with the occipital cortices.

are considered premature if born before the 37th week. Term infants are born between the 37th and 42nd weeks of gestation, and postterm infants are born after 42 weeks of gestation. In an arbitrary but practical manner, we stop counting children’s ages in months sometime beyond the second or third year, and convert to years, or fraction of years (e.g., 2.5-year-old child).

VISUAL ANALYSIS State and Continuity The VA begins with an assessment of the state of the child and the continuity of the background. For all pediatric EEG tracings, there should be a tight concordance between state and EEG background, and the tracing should be continuous in all states. The notable exception is the neonatal period where tracings are normally discontinuous in the premature. Indeed, an accurate

Normal Pediatric EEG: Neonates and Children

Figure 6.1b: Modification of the international 10–20 system commonly used in neonates. A single montage includes anteroposterior, transverse, and midline arrays. Fp3 and Fp4 replace the usual Fp1 and Fp2 electrode locations.

assessment of the quantity and quality of the continuity of the background in the various states is the single most important factor in neonatal EEG. Simple definitions of state suffice: in sleep, the eyes are closed, and in wakefulness, the eyes are open. In a well-developed newborn in active sleep, there are a variety of small and large body movements, sucking, and even crying behaviors that are punctuated by bursts of predominantly horizontal rapid eye movements—the REM phase of active sleep. Brief apneas are relatively common, especially before term. Newborns often enter into active sleep from wakefulness. This pattern of sleep onset continues until about 3 to 4 months postterm, at which time quiet sleep precedes active sleep, just as it will throughout adult life. In a well-developed newborn and infant in quiet sleep, there are few head, trunk, or limb movements. Respirations are regular, deep, and slow. Apnea

127

is uncommon. Occasional startles or arousals may briefly interrupt quiet sleep. Since quiet sleep is most vulnerable to adverse medical or neurological conditions, it is essential to try to capture quiet sleep for a thorough assessment. In newborns and infants with mild encephalopathies, the awake and active sleep recording may appear normal, but quiet sleep recordings reveal previously unrecognized abnormalities. Even in healthy newborns, much of sleep is indeterminate or transitional: That is, even with a good-quality EEG and careful behavioral observation, it is not possible to determine precisely whether the child is in active or quiet sleep. This is clearly the case when the infant transits from one behavioral state to another (transitional sleep), but it also applies when an exact designation of active or quiet sleep cannot be assigned. A large proportion of total sleep time is indeterminate at term and increases in the setting of medical or neurological illness (Fig. 6.2). The earliest vestiges of EEG activity are believed to arise after the 8th week of gestation where the EEG tracing appears as a completely discontinuous recording in which brief periods of electric activity (“bursts”) are interrupted by periods of quiescence (“interburst” intervals or IBIs). With the development of central nervous system (CNS) maturity and the increased influence of the deep grey structures that modulate cortical function, the duration of the burst (burst interval [BI]) increases, whereas the length of the IBI decreases. EEG signals that regularly vary between the high-amplitude “on” periods of the bursts and low-amplitude “off ” periods of the IBI are called discontinuous EEGs. Those that display a relatively steady amplitude are considered continuous. The duration of the IBIs is a semiquantitative measurement of one aspect of the neonatal EEG. A typical, representative portion of the discontinuous portion of the EEG is selected for review, and the duration of the IBIs is measured and counted over a specific period of time: for example, a 10-­minute sample. In that representative portion of time, numerous measurements of the IBI are made, and the mean, median, and longest IBI values can be measured. The prime determinant of measures of the IBI is the infant’s conceptional age. A typical median IBI at the conceptional age of 24 weeks is 10 seconds; this gradually decreases at older conceptional ages to values around 2 to 4 seconds (Fig. 6.3) (2–12). Beyond 2 months of age, the EEG tracing should be continuous in all states. The importance of this fact cannot be overstated. Any older infant with a discontinuous background is demonstrating signs of an encephalopathy, and this is one of the most powerful prognostic features in a wide ­variety of clinical settings, including epilepsy.

128

Normal Pediatric EEG: Neonates and Children

Figure 6.2: The percentage of total sleep time occupied by indeterminate sleep is significantly increased in sick newborns with abnormal electroencephalographic backgrounds. (Adapted from Watanabe K, Miyazaki S, Hara K, et al. Behavioral state cycles, background EEGs and prognosis of newborns with perinatal hypoxia. Electroencephalogr Clin Neurophysiol 1980;49:618–625.)

Organization and Principal Components

Figure 6.3: Median IBI duration decreases with advancing conceptional age in survivors of prematurity. The premature infants who died are characterized by IBIs that are significantly longer in duration than the IBIs of survivors. (Adapted from Clancy R, Rosenberg H, Bernbaum J, et al. Survival outcome prediction in premature infants with IVH by cranial ultrasonography and EEG. Ann Neurol 1994;36:489.)

Newborns do not have a consistent anterior to posterior voltage and ­frequency gradient, but all older infants and children should. This d ­ evelops by 3 months and from this point forward becomes the second most important general feature of the pediatric EEG. In contrast to adults where low voltage undifferentiated backgrounds may be seen as a normal variant, this is virtually unheard of in infants and children. Awake children beyond 3 months should have a clear gradient. This is disrupted in drowsiness and sleep; therefore, if the gradient is not present, one must wonder whether the child is drowsy or encephalopathic. The gradient is composed of three principal rhythmic components located in the posterior, central, and frontal regions, corresponding to the alpha, mu, and beta rhythms, respectively. While these more mature organized rhythms are absent in the newborn, there are other named patterns that are important to recognize. These named patterns are not truly precursors to the more mature components and are not obligatory like the posterior dominant rhythm, but important signposts of maturation, nonetheless. Therefore, these components of neonatal EEG will be briefly mentioned in the overview of the ontogeny and then in detail in the following section on principal components.

Normal Pediatric EEG: Neonates and Children

Interhemispheric Symmetry and Synchrony Normal EEG activity arising from the two hemispheres or homologous brain regions should be essentially symmetrical at any age, including the newborn. It may be difficult for the reader unaccustomed to pediatric EEGs to accurately judge whether the absolute quantity of various frequency elements is normal

129

for age, but it is still possible to determine whether these same elements are symmetric. There are two facets of background symmetry to be judged: amplitude and waveform composition (Figs. 6.4 and 6.5). Amplitude symmetry implies that, in a suitably large sample of cerebral electrical activity, the background voltages between the hemispheres or specific regions are approximately equal. There is no universal agreement as to what amount of amplitude asymmetry

Figure 6.4: Asymmetry secondary to cerebral pathology in an infant 41 weeks of conceptional age with Sturge-Weber syndrome affecting the right hemisphere. The right occipital region is slow and contains sharp waves, in comparison with the normal-appearing left hemisphere.

130

Normal Pediatric EEG: Neonates and Children

Figure 6.5: Asymmetry secondary to scalp edema in an infant 41 weeks of conceptional age with tetralogy of Fallot and seizures. The infant’s head was turned to the right, and marked right scalp edema was present. The amplitude is decreased on the right side, but the background composition is similar to that of the left. The asymmetry disappeared after the scalp edema resolved.

Normal Pediatric EEG: Neonates and Children

constitutes an electrographic abnormality. A useful interpretation guideline is that an abnormality may be suspected if the amplitude difference between two regions exceeds a 2:1 ratio (11,13–15). Likewise, guidelines for the complexity of the waveforms are not well established, and therefore this is usually left to the discretion of the reader to decide whether the frequency components are relatively similar between two regions or hemispheres. Interhemispheric synchrony is another measure reflective of CNS maturation. In the neonates, interhemispheric synchrony is measured during the discontinuous portions of the EEG. Asynchrony is defined as bursts of morphologically similar EEG activity in homologous head regions separated by more than 1.5 to 2.0 seconds (Fig. 6.6). Somewhat paradoxically, neonates at a conceptional age of less than 30 weeks exhibit hypersynchrony, whereby the majority of bursts arising within the two hemispheres appear at the same time (Fig. 6.7). The physiological basis for interhemispheric hypersynchrony is unknown. After the conceptional age of 30 weeks, hypersynchrony gives way to the appearance of asynchronous bursts of cerebral electrical activity between the two hemispheres. About 70% of bursts during quiet sleep are synchronized at the conceptional age of 31 to 32 weeks, increasing to 80% at 33 to 34 weeks, 85% between 35 and 36 weeks, and 100% after 37 weeks (2,11,14,16). In older infants, synchrony is largely judged by sleep architecture. Sleep spindles and vertex waves are normally asynchronous until 18 to 24 months, at which point they are expressed simultaneously across the hemispheres along with K complexes.

Special Features A variety of special features are encountered in the pediatric EEG that are unique to this age or have a clinical significance that is different from that seen in an adult. In neonatal, infant, and childhood EEGs, some of these special features can be helpful signposts of maturation and will be discussed in some more detail. The reader encountering these features for the first time is encouraged not to get lost in the details but to retain the general gestalt. Overtime, it is critical to master the identification of these special features because these are the elements that are most likely to be confused with pathological waveforms and may therefore lead to erroneous conclusions in the interpretation. One example is the exuberance of rhythmic slowing normally seen in state transitions or during hyperventilation (HV) in the young. The authors’ have seen these patterns misinterpreted as “paroxysmal discharges” when, in fact, they are completely normal for age.

131

Abnormalities Many abnormalities such as slowing and attenuation have very similar significance in pediatric EEGs as in adult studies, but the clinical significance of certain abnormalities like sharp waves and spikes varies as a function of age and location. This is most critical in the newborn, and therefore the special significance of sharply-contoured discharges will be reviewed in some detail. Finally, neonatal seizures will be discussed—even though they are never normal—as no review of neonatal EEG would be complete without some discussion on this important topic.

OVERVIEW OF ELECTROENCEPHALOGRAPHIC ONTOGENY The development of the fetal brain undergoes explosive changes with regard to its overall anatomic appearance, synaptic connectivity, timedependent genetic expression of neurotransmitter receptor subunits, and their consequent functional abilities. In parallel with these anatomical and functional changes is an orderly, predictable pattern of neonatal EEG characteristics that emerge simultaneously with advancing maturity of the fetal brain (Fig. 6.8). EEG maturation continues at a rapid pace during infancy, slows during childhood, and nearly plateaus in adolescence, whereupon it settles into a form that remains relatively stable for the next seven to eight decades. For practical reasons, the discussion of EEG ontogeny begins with the infant at 24 weeks EGA, near the current boundary of fetal viability (7). It will proceed through infancy, into childhood, and finally adolescence. The general hope of this chapter is to provide a fluid discussion of pediatric EEG interpretation from the premature to adolescence, with a consistent method of reading that can be used throughout the age spectrum.

24 to 29 Weeks of Conceptional Age EEGs obtained from the very premature infant are, for the most part, discontinuous recordings where low-voltage (100 cm2) were misleading, whereas dipole modeling

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

347

sources between 18 and 36 cm2 were consistent and accurate. Dipoles are thus a reasonable source model for focal spikes, given that Tao et al. (4) showed that epileptic potentials appreciated on scalp EEG commonly had source areas of 10 to 30 cm2. Plummer et al. (31) reviewed the state of affairs regarding the acceptance of dipole and other forms of electrical source imaging in the presurgical evaluations of focal epilepsy as of 2008. Several ways have been used to determine the adequacy of a given dipole solution. The simplest is the goodness of fit of the model to the actual data at each time point or over a time range. For clinical studies, in particular when individual spikes are used, a residual variance of 10% to 15% can be reasonable. Averaged spikes or seizure potentials should result in a better fit. The residual variance or other measures of fit quality should be compared over time to the mean global field power. A good dipole model shows a goodness of fit curve that matches the shape of the field power. Curve mismatches, that is low goodness of fit during a period of high global field power means an inadequate model. Under these circumstances, additional dipoles may be needed to explain the data over time. Recently, Fuchs et al. (32) have developed a means of visualizing the quality of a dipole solution as a 3D ellipsoid confidence volume.

Coregistering Dipole Models with MRI, Realistic Head Models, and Spatial Sampling

Figure 12.6: In order to simulate the broad scalp voltage field generated by a large cortical source (lightened left temporal cortex), an equivalent dipole (black dot) must be deeper in the brain. One method of interpreting this deep dipole is to project its orientation vector out to intersect the overlying cortex. The net orientation of the convoluted cortical source is appropriately orthogonal to that of the dipole modeling it. Note that opposed voltage fields from each side of activated sulci cancel each other, leaving that of gyral crowns to predominate.

Dipole models display in 3D terms, the same information that a person can perceive by visually inspecting the voltage fields as explained earlier. In addition, these equivalent dipoles can be coregistered either with a head schematic or with an actual 3D MRI to identify putative sources within the brain. With a head or brain schematic, it is possible to identify the most likely lobe or perhaps even sublobar area containing the source. It is very tempting to coregister dipole model data with 3D MRIs, as is commonly done with other functional imaging techniques. By digitizing in 3-space (3D), the scalp electrode locations and certain head landmarks, such as the nasion and preauricular points, the topography of scalp EEG fields can be coregistered with and superimposed on a 3D MRI reconstruction of the patient’s head or brain. In similar fashion, calculated dipole models can be coregistered with the same 3D brain image (Fig. 12.8). The problem with doing so is that dipole sources or, worse yet, colored blobs that are placed on a brain image take on a seductive pseudo-realism that, in reality, must be verified. In the case of EEG dipole models, sources

348

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

Front

Right

Front

Right

Top

Back

Top

Back

Left

Front

Top

Back

C

Figure 12.7: EEG traces, voltage topography, and ­single-dipole models of typical temporal lobe spikes. Isopotential lines in speckled areas of voltage maps denote negativity; those in clear areas denote positivity. Top left: Subtemporal negative field maximum, vertex positive maximum, and elevated, oblique dipole orientation of a Type 1 temporal spike. Top right: Lateral temporal negative field maximum, contralateral temporal positive maximum, and horizontal radial dipole of a Type 2 temporal spike. Bottom left: Interior frontotemporal negative field maximum, posterior positive maximum, and horizontal AP tangential dipole of a temporal tip spike.

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

Figure 12.8: Dipole solutions for a left temporal lobe spike are coregistered with a 3D model of the patient’s brain MRI. Recording electrodes and voltage field lines are also displayed in register. The lower dipole solution was calculated with a realistic, boundary element head model, while the upper dipole was ­derived with a three-sphere head model.

in the cortical convexity are usually localized accurately. This is because even simple spherical head models used by source modeling algorithms work well in this most spherical part of the head and brain. However, when sources were near the base of the brain, such as in the temporal lobe, systematic dipole location errors occur even for known sources (33,34). These systematic errors in dipole location, introduced when using a spherical head model, had been previously recognized (33–36). Later, investigators simulated EEG dipole sources throughout the brain and found that spherical head modeling errors were worse for basal brain regions and in the Z or vertical direction (37–39). These results were also confirmed using temporal lobe spike and seizure foci as the source and validating the true location of these foci with intracranial EEG (3,40). This error was typically in the vertical direction, and it could be as great as two or more centimeters in magnitude, when a spherical head model was used (34,35,37).

349

The boundary element model (BEM) is the most commonly used of the realistically shaped models of the head (Fig. 12.9) to compensate for the errors of a spherical head model. It can account for the individual, nonspherical shape of the main intertissue boundaries of the brain, skull, and scalp (38,41,42). Each of these surfaces is discretized into triangular elements to form a mesh (Fig. 12.9). BEM head models have the shortcoming that conductivity is assumed to be piecewise constant and isotropic, whereas cranial tissues are inhomogeneous and show anisotropy. More sophisticated finite-element methods can overcome these problems, but require considerably more computational effort (43). Also, the detailed conductivity information of various brain, skull, and scalp elements needed to construct such a model is not well-established. Accordingly, BEM realistic head models are a reasonable compromise for clinical purposes. Fuchs et al. (44) described the use of a standardized BEM head model derived from the averaged Montreal MRI data set in order to provide simple access to realistically shaped volume conductor models for source reconstruction. Standardized and individualized BEM head models resulted in more accurate and comparable dipole solutions than those derived from a spherical model. They compared dipole modeling accuracy using spherical head models, individually derived realistic head models, and the standardized realistic head model in both source simulations and epileptic spike data. Spherical head models resulted in dipole mislocation errors at the base of the brain, as previously noted. Standardized and individualized BEM head models resulted in more accurate and comparable dipole solutions. By using a standardized head for the BEM setup, an easier and faster access to the benefits of a realistically shaped volume conductor models can be achieved. Figure 12.10 illustrates dipole solutions for a temporal lobe spike calculated with both spherical three-shell and BEM head models. Dipoles using a spherical model were misplaced 2 cm upward from those of a BEM model and their true temporal lobe source. Such an error can give the false impression that the spike/seizure source was of superior temporal or frontal rather than anterior inferior temporal lobe origin. It is apparent that realistic head models should be considered whenever dipoles are coregistered with MRI for interpretation. This is particularly true for basal frontal, temporal, and occipital sources where the brain cortex and skull departs most from a spherical shape. A simple way to determine whether a BEM model would be worthwhile in modeling spikes and seizures is to look at the voltage maps of the field maxima. If the normally dominant negative field maximum is only partially described and it rests near or below the 10–20 system “equator,” in the “southern hemisphere” of the head, a BEM

350

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

Figure 12.9: Left: Shaded circle on a patient’s head depicts the typical shape of a spherical head model. Middle: Realistic BEM head model illustrating inner and outer surfaces of the skull and scalp derived from 3D MRI reconstructions. Right: Tessellation of these surfaces into a triangular mesh used in the source calculations. Note the divergence of the brain and skull base from the simple spherical shape.

model will be necessary to compensate for the nonspherical nature of the skull in this region. Some have proposed on theoretical grounds that high-resolution EEG requires hundreds of electrodes (45); however, others have concluded that the gain beyond 40 to 64 electrodes is negligible for dipole models (37,46). This is particularly true for epileptogenic spike and seizure sources that tend to be large spatially and produce scalp voltage fields of low spatial frequency, namely, dipolar. Lantz et al. (20) recorded spikes from patients with epileptogenic lesions with 128 electrodes and performed source modeling with various down-sampled arrays. An increase in electrode number from 31 to 63 resulted in significantly improved localization, whereas only minimal additional precision was achieved with an increase in electrode number to 128. Given the fact that the position and orientation/magnitude of a dipole model has only six unknowns, even a modest number of scalp electrodes provide an overdetermined solution. A larger number of electrodes (128 to 256) does, however, improve largely underdetermined extended source ­models (47).

Dipole Modeling of Seizures Dipole modeling can also be applied to seizure rhythms with some modifications in the protocol used for spikes (2,3,19,29,48–51). The earliest recognizable seizure potentials should be preferentially modeled because they are more likely to reflect the seizure origin than are later rhythms, which usually evolve only after significant propagation. In most instances, the EEG must be filtered with a narrow bandpass covering those frequencies that represent the cerebral seizure activity and not the accompanying artifact. For partial seizures of the temporal lobes, this is approximately 2 to 10 Hz. For extratemporal seizures that commonly have a beta frequency component, a broader 2- to 20-Hz bandpass is useful. Because ictal-onset rhythms are typically of low amplitude and are commonly confounded with movement and muscle artifact, averaging successive potentials may be necessary to increase the S/N. The key is to average seizure waveforms with similar voltage topographies (Fig. 12.11); only these reflect

Figure 12.10: EEG traces, voltage topography, and singledipole models of the same left temporal spike in both 2D and 3D MRIs. Dipole solutions in the upper panel were calculated with a three-sphere (3S) head model; dipoles in the lower panel were calculated with a realistic BEM head model. Note that the 3S dipoles are approximately 2 cm higher than the BEM dipoles. In this and subsequent similar figures, the 2D MRI images are anatomical, that is, left on left. Also, negative and positive field maxima are noted by “−” and “+” in the voltage maps.

351

352

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

Left

Right

Top

Bottom

Back

Front

Figure 12.11: Top: EEG traces showing a left temporal seizure onset. Individual ictal waveforms are marked by cursors. These ictal waveforms are ­averaged and displayed at the left side of the traces. Voltage topography of the averaged ictal waveform is displayed at the right side. Bottom: Dipole models of the seizure onset are coregistered with a 3D brain MRI of the patient. Dipoles in the top row were calculated using a 3S head model; those in the bottom row were calculated with a BEM head model.

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

the same source configuration. These periods of stable fields may last only a few seconds. Alternatively, spatiotemporal modeling can be performed over this stable epoch. One or more fixed dipoles possessing cyclic variations in the magnitude and polarity of their source potentials can often explain the repeating evolution of an ictal voltage field. Temporal and orientation strategies are applicable in seizure modeling, just as in spike modeling. Modeling sequential epochs during the course of the seizure can provide insight about seizure propagation. The orientation of dipole models of ictal waveforms carries the same significance as those of spikes, and it is most useful in identifying sublobar sources (2,3,19,29,48–51). Temporal lobe seizures modeled by dipoles with dominant horizontal radial, vertical tangential, or horizontal tangential AP orientations are most likely associated with lateral temporal, hippocampal/basal, or temporal tip seizures, respectively (Fig. 12.12). Additionally, many temporal lobe seizures are modeled best by dipoles having an anterior oblique orientation, that is, a combination of all three of the previous orientations. In this case, the ictally active cortical region includes inferior, tip, and anterior lateral temporal cortex. Multiple fixed-dipole models can also be used to identify the contribution of various sublobar cortical areas to seizure potentials. This technique has been used to determine temporal lobe seizure origins and to predict surgical outcome following standard temporal lobectomy (27,28).

Dipole Models for Specific Cortical Regions Temporal Lobe Sources Perhaps the most common source of epileptiform activity in patients with complex partial seizures is the temporal lobe. The anatomy of this lobe affords the possibility of a number of spike and seizure voltage fields and thus several types of dipoles that model them. Typical temporal spike fields have an anterior or mid-temporal negative maximum (Fig. 12.13). The inferior extent of this negative maximum, and more importantly, the location of the positive field maximum will determine if the orientation of the dipole is radial (horizontal), tangential (vertical or horizontal AP), or oblique (in between radial and tangential) (Figs. 12.14 and 12.15). These dipole ­orientation, as well as location, differences confirm different sublobar spike sources. Dipole models can easily distinguish spikes that originate from ­lateral, inferolateral, basal, tip, and superior temporal cortex.

353

Commonly, spikes propagate locally across the adjacent cortex from their origin or even distally over white matter tracts. When this happens, the geometry of the generator cortex changes and the voltage fields produced evolve as well. As noted previously, in instances of a stable source, voltage fields will rise and fall in amplitude but not change in scalp location or shape. Conversely, propagation from source of origin will result in a scalp voltage field whose maxima move in location and change in shape. When propagation is simple and unidirectional, this can be modeled by sequential or so-called moving dipoles (Fig. 12.16).

Extratemporal Sources An increasing number of cases of extratemporal epilepsy are being seen in epilepsy centers. Spikes and seizures from these patients can also be localized better with voltage topography and dipole modeling techniques. Given that much of the frontal and parietal cortex lay on the cortical convexity, radial sources are common (Figs. 12.17 and 12.18). These have the advantage of being more easily interpreted, namely under the negative maximum is the source. However, fissures and deep sulci can produce tangential fields that are confusing to visual inspection of EEG traces, particularly in bipolar montages (Figs. 12.18 to 12.20). False lateralization can occur if one pays attention to only the negative field maximum. Voltage maps confirm the tangential nature of these fields, and dipole models clearly show that the source lies between the negative and positive field maxima. Source propagation can be characterized by a moving dipole model (Fig. 12.21).

Other EEG Source Models Within the past decade, there has been considerable development of new techniques to expand the use of source modeling in the evaluation of epilepsy patients. Many of these involve new mathematical approaches to the decomposition of the data or to the reconstruction of a source model. Because the number of cortical sources is not known when modeling spike or seizure potentials, some investigators feel that decomposing the data by a variety of techniques is worthwhile before attempting to model them. Several have chosen to use singular value decomposition (52) or independent component analysis (53–55). The likely number of component sources, or at minimum the dominant component, can be estimated and their contributions to the voltage field and waveform separated without prior knowledge.

354

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

Front

Right

Front

Right

Top

Back

Top

Back

Left

Front Left

Top

Front

Back Top

Back

Figure 12.12: EEG traces, voltage topography, and single-dipole models of typical temporal lobe seizures. Isopotential lines in speckled areas of voltage maps denote negativity; those in clear areas denote positivity. Top left: Subtemporal negative field maximum, vertex positive maximum, and elevated, oblique dipole orientation of an anterior, inferolateral temporal seizure. Top right: Lateral temporal negative field maximum, contralateral temporal positive maximum, and horizontal radial dipole of a lateral midtemporal seizure. Bottom left: Interior frontal negative field maximum, posterior positive maximum, and horizontal AP tangential dipole of a temporal tip seizure. Bottom right: Subtemporal negative field maximum, contralateral posterior positive maximum, and oblique dipole of an anterior temporal seizure.

Figure 12.13: EEG traces, voltage topography, and dipole models of typical temporal lobe spike sources in 2D and 3D brain MRIs. Top: Anterior subtemporal negative field maximum, contralateral vertex positive maximum, and oblique dipole of an anterior inferolateral temporal source. Projection of dipole vector intersects inferolateral temporal tip cortex. Bottom: Midtemporal negative field maximum, contralateral positive maximum, and radial orientation of a lateral, midtemporal source.

355

Figure 12.14: EEG traces, voltage topography, and dipole models of typical temporal lobe spike sources in 2D and 3D brain MRIs. Top: EEG traces in a longitudinal bipolar montage show no phase reversals, however a phase-away reversal is evident. Localization of the negative field maximum is unclear. Bottom: EEG traces in a common average reference show more clearly the FP2 negative field maximum and T8 positive maximum that is evident in the voltage maps. Horizontal tangential dipole is consistent with an ­antero-mesial temporal tip source.

356

Figure 12.15: EEG traces, voltage topography, and dipole models of typical temporal lobe spike sources in 2D and 3D brain MRIs. Top: Posterior subtemporal negative field maximum, vertex positive maximum, vertical tangential dipole of a temporal base source. Note that a weak negativity can even be seen in the contralateral subtemporal electrodes. Bottom: Posterior temporal negative field maximum, frontal positive maximum, and oblique dipole of a posteriorinferior temporal source.

357

358

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

Figure 12.16: EEG traces, voltage topography, and dipole models of typical temporal lobe spike sources in 2D and 3D brain MRIs. Note progressive change in the location of the voltage field maxima over 40 milliseconds. The negative field maximum moves from the inferior frontal to the midtemporal region, while the positive maximum moves from the contralateral posterior to vertex region. A moving dipole model over the period shows source movement from the left temporal tip to basal temporal cortex.

Others have devised techniques for systematically scanning the entire brain space for possible sources (56,57). Although the dipole is a useful concept from which practical diagnostic information can be obtained, it is anatomically not very realistic. Brain sources of spikes or seizures are extended cortical patches, not isolated points. Clinicians, in particular, are more comfortable seeing regions of cortical activation, such as displayed in PET, SPECT, or fMRI images, regardless of the physiological soundness of these representations. There has been considerable effort recently to go beyond dipole source models to more realistic extended source models. These take a variety of approaches; however, a common form of source reconstruction works on the hypothesis that brain currents are located only at specified locations. Using the cortical surface, for example, is a powerful reconstruction parameter that restricts the search space to a surface, which can be defined further by local orientation information. Algorithms have been devised that can calculate discrete approximations of the current density distribution on a defined surface. This is an ill-posed

problem, however, if many local sources are to be calculated from measurements at relatively few electrode locations. Regularization is necessary to proceed in these circumstances. This represents a compromise between the demands to explain the measured data and to meet certain source or boundary conditions. One such boundary condition is the minimum norm criterion, in which the source constellation of lowest electrical power is calculated. Variations include the L2- or L1-norms (58) or the minimum spatial Laplacian, also called low-resolution tomography (LORETA) (59,60). Additional techniques for estimating multiple sources or single source with extent include LAURA and EPIFOCUS (61,62), and VARETA (63). The usefulness of these techniques needs to be proven with clinical data and further validated by intracranial EEG recording, however. One investigation demonstrated no advantage of current density reconstruction over source modeling with dipoles (64). Another approach to the problem of unrealistic point-like dipoles is the use of a “dipole patch” model. It is composed of many individual dipoles

Figure 12.17: EEG traces, voltage topography, and dipole models of typical extratemporal spike sources in 2D and 3D brain MRIs. Top: Lateral centro-temporal negative field maximum, contralateral frontal positive maximum, and radial dipole of a left frontal operculum source. Bottom: Inferior frontotemporal negative field maximum, ipsilateral vertex positive maximum, and oblique dipole of a left orbitofrontal source.

359

Figure 12.18: EEG traces, voltage topography, and d ­ ipole models of typical extratemporal spike sources in 2D and 3D brain MRIs. Top: Ipsilateral occipital negative field maximum, contralateral frontal positive maximum, and ­ ­radial dipole of a left occipital source. Bottom: Contralateral occipital negative maximum, ipsilateral occipitotemporal positive maximum, and tangential dipole of a left mesial occi­pital source.

360

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

361

Figure 12.19: EEG traces, voltage topography, and dipole models of typical extratemporal spike sources in 2D and 3D brain MRIs. Frontal vertex negative field maximum, ipsilateral frontotemporal positive maximum, and tangential dipole of a mesial frontal source.

that are constrained to follow the surface of the patient’s own cortex, derived from 3D MRI reconstructions, in both position and orientation. This model has a spatial extent that is adjustable to that known to produce scalp EEG potentials (4). The location of dipole patches was accurate at a sublobar level when compared to underlying cerebral sources, as validated by intracranial EEG (65). Other extended source models, such as those employing current source density estimations, are not limited to realistic source areas, nor are they constrained to anatomically contiguous cortex as is this model.

Clinical Studies Using Dipole Models The early 1990s marked the beginning of clinical studies of patients with epilepsy using voltage topography and dipole modeling. It soon became apparent that the voltage field of spikes and seizure potentials could vary considerably even if their origin was the same lobe. Ebersole and Wade (66,67) first made the distinction between temporal lobe spikes with a radial field orientation (type 2) and those with an oblique to vertical tangential

orientation (type 1) (see Figs. 12.3, 12.4, and 12.7). A variety of clinical ­correlations, as well as simultaneous scalp and intracranial EEG investigations, suggested that the type 2 field topography and its corresponding horizontal radial dipole model originated from a lateral temporal cortical source, whereas type 1 topography and its corresponding vertical tangential dipole model was associated with inferolateral and basal temporal sources. However, the type 1 spike field was not thought to reflect hippocampal or amygdalar activity directly. It was shown that spikes confined to these structures did not generate scalp-recordable voltage fields due to the small source area and curved source shape, which favor voltage cancellation. Rather, it was the common and preferred propagation of this epileptiform activity from mesial structures into the entorhinal, fusiform, and other temporal basal cortex that resulted in a generator of sufficient area to produce scalp EEG potentials (2,68,69). Subsequent investigation demonstrated that a rigid categorization of temporal spike topography was overly simplistic. Spike voltage fields commonly evolve over tens of milliseconds, which usually signify a propagating spike source. A type 1 pattern may evolve into a

Figure 12.20: EEG traces, voltage topography, and dipole models of typical extratemporal spike sources in 2D and 3D brain MRIs. Top: EEG traces in a longitudinal bipolar montage show no phase reversals; however, a phase-away reversal is evident. Localization of the negative field maximum is unclear. Bottom: EEG traces in a common average reference show more clearly the Fz-Cz negative field maximum and P4 positive maximum that is evident in the voltage maps. The horizontal tangential dipole is consistent with a lateral central sulcus source.

362

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

363

Figure 12.21: EEG traces, voltage topography, and dipole models of typical extratemporal spike sources in 2D and 3D brain MRIs. Note progressive change in the location of the voltage field maxima over 32 milliseconds. The negative field maximum moves from the fronto-central to frontopolar region. A moving dipole model over the period shows source movement from the right fronto-central to frontopolar cortex.

type 2 pattern and vice versa. Spikes that originate in the basal temporal cortex frequently propagate into the temporal tip cortex by the time sufficient cortex is activated to produce a scalp EEG field. Temporal tip cortex has a net anterior-facing orientation. Accordingly, spike sources in this cortex result in a voltage field with a frontotemporal to frontopolar negative maximum and a posterior positive maximum (3,19,29) (see Fig. 12.7). By the mid-1990s, several other investigators began to apply EEG dipole modeling for the evaluation of epilepsy surgical candidates. In general, they confirmed findings from earlier studies. In a series of investigations, Boon et al. (48–51) demonstrated that patients with medial temporal lesions had EEG spikes and early seizure rhythms that were modeled by dipoles having a elevated orientation above the horizontal (a combination of radial and tangential) similar to Ebersole type 1 spike, whereas patients with lateral temporal or extratemporal lesions had EEG spike and early seizure rhythms that were modeled by dipoles having a radial orientations similar to Ebersole type 2 spike. Lantz et al. (70–72), in a series of studies using simultaneous

scalp and intracranial EEG recordings, confirmed that dipole orientation was the major distinguishing factor to distinguish spikes originating from baso-mesial temporal sources from those of lateral temporal origin. Merlet and Gotman (73,74) in a pair of papers validated dipole models of spikes or seizures with a later combination of intracranial and scalp EEG. They concluded that only simple cortical sources could be well-modeled by one dipole and that spike modeling yielded better and more frequent results than seizure modeling. Numerous other publications began appearing in the late 1990s that confirmed a good correlation between spike dipole models and other localization tools such as intracranial EEG (75,76), positron emission tomography (77), single photon emission tomography (77,78), and MRI lesions in both adults and pediatric patients (79). Dipole source models were later shown by the Marseille group, headed by Chauvel, to be accurate not only for temporal, but also for frontal and posterior foci, if the geometry of the focus was not overly complex or multifocal (80–82). Recently, both dipole and extended EEG source models have

364

EEG Voltage Topography and Dipole Source Modeling of Epileptiform Potentials

shown to be particularly useful in patients with normal MRI, where accurate functional localization of the epileptic focus is key (83,84). Ictal EEG dipole models have been correlated with intracranial EEG and surgical outcome (2,3,19,27–29,48–51). The results are similar to those found with spike dipole modeling. Patients with temporal lobe seizures modeled by horizontal radial dipoles, which suggest a lateral cortex origin, do less well following a standard anteromesial resection and should be considered candidates for invasive monitoring and possibly tailored temporal lobe resections. Patients whose seizures are modeled best by dipoles with a vertical, horizontal AP, or anterior oblique orientation have seizures originating probably in mesial structures, basil, entorhinal or temporal tip cortex, respectively. All of these patients do well following surgery because these cortices are customarily removed in the standard temporal lobe resection. Recently, ictal source modeling using both dipoles and several extended source algorithms were compared (85). Dipole modeling provided the closest concordance (90%) with the ictal-onset zone as determined by intracerebral EEG. Starting in 2004, the Geneva-based group, headed by Michel and Seeck, published a series of investigations demonstrating the usefulness of EEG source modeling in localizing the epileptogenic focus, as verified by seizure freedom following resection. In all their studies, a Laura-based extended source model of spikes localized to cortical regions, which when resected led to seizure freedom (62,86). These included temporal and extratemporal foci and pediatric patients. In the largest prospective study to date, the sensitivity and specificity of their EEG source imaging method, using both a low (100), longer duration of intracranial monitoring

(>10 days), left-sided implantation, and elderly patients (62,64). Experience is also an important factor. For example, the Cleveland Clinic group reported a 33% rate for minor and major complications at the beginning of their epilepsy program, which dropped to 19% in more recent years (64). Given that subdural electrode placement can be associated with morbidity and mortality, careful presurgical planning and postimplant patient care is necessary to minimize these. To prevent infection, it is useful to limit the recording period to less than 10 to 14 days. Intraoperatively, the wires of subdural electrodes should be tunneled subcutaneously and brought out at a separate site, where tissues can be tightly sutured. This helps to prevent bacteria from gaining access to the brain and meninges via the exiting wires. In addition, prophylactic use of perioperative antibiotics is a common practice among epilepsy centers. Subdural hematomas are mainly caused by tearing cortical bridging veins during the insertion of strip and grid electrodes. Such bleeding can be

Subdural Electrode Corticography

reduced by assuring that coagulation parameters are appropriate prior to surgery. Cerebral edema can also be a serious complication of invasive EEG recording. It may interfere with accurate epileptic focus and eloquent cortex localization. Moreover, cerebral edema that is refractory to intervention can provoke cerebral herniation and result in death. Administration of dexamethasone following surgery and during intracranial monitoring can be useful in reducing cerebral edema; however, it may also decrease the frequency of habitual seizures and thus prolong the duration of intracranial monitoring (69). Therefore, its use needs to be judiciously considered. In some centers, dexamethasone is only administered during the first 72 postoperative hours in patients experiencing significant headache, nausea, or vomiting.

Clinical indications of subdural electrode EEG recording Currently, there are no universally recognized guidelines for when an intracranial EEG study is indicated. In practice, the following are common clinical indications for invasive EEG studies (70,71): (a) lateralize ictal onset to a specific hemisphere, such as left versus right temporal epilepsy; (b) localize ictal onset to a specific lobe, such as temporal versus frontal lobe epilepsy; (c) localize seizure onset and determine the extent of seizure-onset zone within a lobe, such as mesial versus neocortical or anterior versus posterior TLE; and (d) functional cortical mapping, such as language mapping in patients with dominant hemisphere TLE. The latter is particularly useful when intraoperative functional mapping is not feasible. On the other hand, with the advent of modern neuroimaging studies, many epilepsy patients who have unilateral mesial temporal sclerosis or well-defined structural lesions (i.e., glioma and cavernoma), which are concordant with scalp interictal and ictal EEG localization, are usually eligible for direct surgery without the need for an intracranial EEG study (10). Also, subdural studies are not indicated in patients with evidence of multifocal or generalized seizure onsets, since they are not considered good candidates for epilepsy surgery.

Postoperative care and EEG acquisition Postoperative Care Subdural electrodes are implanted under general anesthesia in the operating room. Afterward, patients are commonly admitted to a neurointensive care unit (NICU) for a brief period of postoperative observation. A head CT and/or brain MRI are usually obtained to rule out acute intracranial

373

hemorrhage and cerebral edema. These neuroimaging studies are also used to localize the implanted subdural electrodes, such as via 3D CT/MRI coregistration (55). Although the choice of antibiotics and duration prophylaxis vary among different centers, antibiotics are commonly administered throughout the course of the intracranial study. Dexamethasone can be also used for patients with signs of increased ICP, such as significant headache and vomiting. Preoperative AEDs are commonly maintained during NICU care or at least until EEG monitoring is begun. Patients are usually transferred from the NICU to the epilepsy monitoring unit (EMU) within 24 hours to begin EEG recording. AEDs are then typically weaned off gradually to facilitate the recording of habitual seizures. EEG recordings are routinely performed 24-hours per day and 7 days per week until sufficient seizures are recorded. This usually necessitates a mean recording duration of 5 to 14 days. The presence of EEG technologists throughout the course of intracranial EEG study is necessary to insure the technical integrity and high quality of the EEG recording. Patient care is provided by a well-trained medical staff, who can provide a neurological evaluation to test awareness, language, memory, and sensory-motor functions during and after a seizure. It is also important to have continuous supervision to prevent the patient from falling or pulling out intracranial electrodes during or after a seizure.

EEG Acquisition Methods for intracranial EEG recording are essentially identical to those used for scalp EEG. Data are routinely digitized at 200 to 250 Hz, which provides an accurate representation of common clinical EEG frequencies, such as delta, theta, alpha, beta, and gamma. For these routine studies, band-pass filters are typically set at 0.3-Hz low-frequency filter and at 70-Hz high-frequency filter. However, modern commercial amplifiers have the ability to sample at much higher rates, ranging from 1,000 to 2,000 Hz. This allows the faithful recording of high-frequency EEG patterns, such as high-frequency oscillations (HFOs) (ripples and fast ripples) from 80 to 500 Hz. Obviously, to record these patterns, the high-frequency filter needs to set much higher, such as at 500 Hz. The number of recording channels required is dependent on the number of implanted subdural electrodes. Although contemporary commercial EEG equipment can simultaneously record 128 to 256 channels, many routine clinical intracranial EEG studies may only require 50 to 100 channels. EEG channels are generally displayed in a logical montage according to anatomic distribution (e.g., anterior to posterior, superior to inferior).

374

Subdural Electrode Corticography

EEG data can be reviewed in either referential or bipolar montages. Each has advantages and disadvantages. Referential montages display the absolute voltage perceived by an e­ lectrode contact, either from a near-field or far-field generator, relative to that of the reference. Because subdural electrodes are placed on the cortex, they record exclusively from gray matter, and near-field signals from underlying cortex predominate. However, in referential recordings from depth electrodes in white matter, there is no local EEG generator, so far-field signals predominate. Referential recordings can be subject to a variety of external artifacts. Therefore, it is important to choose a reference electrode that has minimal signal, less likelihood of artifacts, and is less likely to be involved in seizure activity. Either intracranial or extracranial electrodes can be used as the reference. Intracranial references are less influenced by external movement and muscle artifacts, but unless properly chosen, they may record a significant amount of epileptiform activity, thus potentially confusing the EEG interpretation. Accordingly, intracranial contacts not touching the cortex and distant from likely epileptic sources are typically used for reference. The amplitude of EEG signals recorded at the scalp is attenuated compared to that from the cortex, so a scalp electrode may be a good reference for intracranial EEG. For example, vertex scalp electrodes (such as CPz) are a good choice for reference when evaluating TLE because they record little temporalis muscle artifact and are distant from temporal lobes. For opposite reasons, earlobe electrodes A1 or A2 would not be recommended. Bipolar montages with subdural electrodes display the difference in voltage typically between adjacent electrodes on a strip or grid. Very focal, nearfield activity is accentuated, whereas broad near-field and far-field activity is suppressed. Thus, little or no interictal or ictal potentials may be evident in a channel composed of recordings from two highly, but similarly, active electrodes. Bipolar montages may be more useful in depth electrode recordings where there are few contacts in any given area of cortex, so in-phase cancellation is minimized. This lack of signal on a bipolar depth electrode channel may however be useful in identifying white matter because the far-field signals from the distant cortex will be suppressed. Accordingly, it is important to review intracranial EEG data with both referential and bipolar montages. ECG is typically recorded along with intracranial EEG. Other channels for oxygen monitoring or respiratory monitoring are recorded in some centers. A partial or full set of international 10 to 20 electrodes can be used to record scalp and intracranial EEG simultaneously. In our experience, there is little increased risk of infection, and the overview of brain activity provided by scalp EEG can compensate for the “tunnel vision” often suffered

by intracranial EEG recordings (72). It is also reassuring to know that a given intracranial spike or seizure of interest produced the same scalp discharge that was recorded previously during long-term monitoring.

Intracranial EEG patterns Normal Intracranial Physiological EEG Normal physiological EEG frequencies and patterns commonly observed on the scalp EEG can be recorded with intracranial EEG. These include alpha and mu rhythms; vertex sharp waves, sleep spindles, lambda waves, and positive occipital sharp transients of sleep (POSTs) (70), provided that intracranial electrodes are placed on the pertinent cortical areas. However, there are also significant differences between scalp and cortical EEG recordings (73–75). EEG potentials recorded from the cortex are higher in amplitude than those recorded on the scalp, and morphology of cortical EEG potentials tends to be sharper or spikier. Some cortical EEG patterns such as high-frequency gamma rhythms simply cannot be recorded on the scalp (76). While the difference between scalp and cortical EEG recording can, for the most part, be accounted by signal attenuation due to the increased resistivity of the skull, the dynamic and variable interactions among cortical source area, amplitude, and synchrony also influence whether or not certain intracranial EEG patterns are recordable on the scalp EEG (72).

Intracranial Interictal EEG Similar to scalp interictal epileptiform discharges (IEDs), intracranial IEDs can also be categorized as spikes, polyspikes, sharp waves, spike-and-wave discharges. Given that only patients with partial-onset epilepsy are considered for intracranial EEG studies, generalized IEDs are rarely recorded or cannot be clearly characterized due to the limited intracranial EEG sampling. Cortical interictal spikes are commonly heterogeneous in their source location, area, synchrony, and amplitude. Only a few of those cortical spikes with sufficient cortical area and synchrony are associated with scalp-­recognizable potentials (Fig. 13.4). Cooper et al. (77) proposed in 1965 that 6 cm² of synchronized cortical activity was probably necessary for the recording of a scalp EEG correlate. Their model, however, employed in vitro ­measurements using a piece of fresh cadaver skull, a pulse generator connected to saline-soaked cotton balls placed on the inside of the skull,  an

Subdural Electrode Corticography

artificial dura made from a polyethylene sheet, and EEG recording electrodes on the exterior surface of the skull bone. The 6-cm² estimate of the necessary cortical sources was based on the area of multiple pin holes punched into the polyethylene sheet when the artificial EEG signals were first recorded from the electrodes on the outside of the skull. Using simultaneous scalp and intracranial recording in epilepsy presurgical candidates with TLE, cortical spikes with synchronized 6 cm² of source area in the lateral temporal cortex did not produce recognizable scalp potentials (Fig. 13.5). Cortical spike sources having an area of 10 cm² or more commonly resulted in scalp-recordable EEG spikes (Fig. 13.6). Furthermore, prominent scalp interictal spikes were commonly associated with the activation of 20 to 30 cm² of cortex, which is as extensive as 70% of temporal lobe gyral cortex (78) (Fig. 13.7). Therefore, cortical area of interictal spike generators is considerably larger than commonly thought. Synchronous or at least temporally overlapping activation of 10 to 20 cm² of gyral cortex is commonly required to produce scalp-recognizable interictal spikes. Intracranial IEDs are far easier to record than seizures. The cortical distribution of intracranial IEDs delineates the irritative zone (79,80). IEDs are thought to be closely related to epileptogenesis and ictogenesis. Therefore,

Figure 13.4: Simultaneous intracranial and scalp EEG recording. A: Intracranial EEG recording demonstrates a heterogeneous population of interictal spikes. B: Scalp EEG recording demonstrates that only two of the intracranial spikes (labeled 1 and 2) generate recognizable scalp interictal potentials. LOF, left orbital frontal; LAT, left anterior temporal; LIT, left ­inferior temporal; LMT, left mid temporal.

375

interictal recording likely provides a useful measure to predict the location of seizure-onset zone. Indeed, using visual inspection or automated detection algorithms, several studies have suggested that the cortical areas associated with maximal spiking frequency, highest amplitudes, and shortest spike duration often overlap with the seizure-onset zone. However, their relationship to seizure-onset zone is not consistently strong. Asano and colleagues (81) found that electrodes with the maximal spike frequency identified the seizure-onset zone in 100% (13 of 13) of patients, while Hufnagel and colleagues (82) found that maximal spike frequency of interictal spikes could identify the seizure-onset zone in only 53% (17 of 32) of patients. In another study, electrodes with maximal spike frequency were found to be in seizureonset zone in 58% (11 of 19) of pediatric patients (83). Areas of spiking on subdural electrodes usually extend beyond the ­seizure-onset zone, and in some cases, they appear to originate from multiple independent spike generators (82,84). These independent spike generators may activate independently or synchronously to produce multifocal interictal spikes. For instance, in patients with mesial TLE, independent spikes are commonly identified from mesial and anterior temporal, ipsilateral orbitofrontal, and contralateral temporal cortex (Fig. 13.8). Several

376

Subdural Electrode Corticography

Figure 13.5: Simultaneous intracranial (A) and scalp (B) EEG recording of a left temporal spike (indicated by arrow). No scalp potential is evident from this cortical source. Neither is there an organized scalp voltage field associated with the intracranial spike (C). (D) illustrates the subdural electrodes recording a synchronous negative depolarization (black) during the spike. Note that the spike source area is approximately 6 cm². LOF, left orbital frontal; LFC, left fronto-central; LAT, left anterior temporal; LIT, left inferior temporal; LMT, left midtemporal.

studies have showed that resections incorporating a “significant” portion of regions generating interictal spikes and sharp waves, in addition to the ictalonset zone, improved seizure freedom postoperatively (85–89). These studies strongly suggested that interictal spikes are biomarkers of epileptogenesis. Nevertheless, the role of intracranial interictal spikes in the localization of seizure-onset zone has its intrinsic limitations. IEDs often provide an imperfect representation of seizure-onset zone, and their interpretation needs to be correlated with ictal recordings. Further studies are needed to determine if there is a predictable relationship between the region of IEDs, the seizureonset zone, and surgical outcomes.

Interictal, rhythmic or arrhythmic, regional slowing is commonly observed in patients with temporal and extratemporal lobe epilepsy during intracranial EEG. Traditionally, regional EEG slowing was thought to reflect an underlying structural abnormality of the brain (90–92). However, interictal regional delta activity (IRDA) can be observed in epilepsy patients without clear structural abnormalities on modern neuroimaging studies (93–95). Scalp EEG investigations have demonstrated that IRDA has a similar lateralizing and localizing value as do interictal spikes in patients with temporal epilepsy, and its presence is associated with a favorable surgical outcome (94,96). Simultaneous scalp and intracranial EEG

Subdural Electrode Corticography

377

Figure 13.6: Simultaneous intracranial (A) and scalp (B) EEG recording of a left temporal spike (indicated by arrow). Note the distinct associated scalp potential and temporal scalp voltage field (C). The approximate area of the cortical spike source is 13 cm². Active electrodes: black. LOF, left orbital frontal; LFC, left fronto-central; LAT, left anterior temporal; LIT, left inferior temporal; LMT, left midtemporal.

recordings demonstrated that cortical IRDA is typically a mixture of delta/ theta slowing and cortical spike-slow wave potentials. In patients with TLE, cortical IRDA is predominantly observed from basal and anterolateral temporal cortex. It is infrequently seen from mesial and posterolateral temporal cortex (Fig. 13.9). The cortical distribution of IRDA was shown to be highly correlated with the irritative zone and seizure-onset zone in patients with neocortical TLE (97) (Fig. 13.10). These observations suggested that IRDA might be an intrinsic EEG pattern of the epileptic network. However, its value in localizing the epileptogenic zone remains to be clarified.

Intracranial Ictal EEG Despite the important role of interictal spike analysis, recording seizures has always been the primary goal of an intracranial EEG study. Seizure onsets are typically categorized as focal or regional depending on the number of electrode contacts involved (14,98–102). However, what is “focal” versus “regional” has differed among investigations. In general, focal onset means only one to three electrode contacts involved, whereas regional onset means four or more active electrode contacts. One assumption is that the fewer electrode contacts involved the more likely they are recording from

Figure 13.7: Simultaneous intracranial (A) and scalp (B) EEG recording of a left temporal spike (indicated by arrow). Note the prominent associated scalp potential and temporal scalp voltage field (C). The approximate area of the cortical spike source is 25 to 30 cm². Active electrodes: black. LOF, left orbital frontal; LAT, left anterior temporal; LIT, left inferior temporal; LMT, left midtemporal.

Figure 13.8: Multifocal sources of intracranial interictal spikes in a patient with left TLE: Interictal spikes are observed in the frontal, anterior temporal, midtemporal, and posterior temporal cortices. Spike source activity could be independent or synchronous. LOFS, left orbitofrontal strip; LATS, left anterior temporal strip; LITS, left inferior temporal strip; LMTG, left midtemporal grid; LPTS, left posterior temporal strip.

378

FIgure 13.9: Simultaneous scalp and intracranial EEG recording of interictal temporal delta activity (ITDA). The scalp EEG shows rhythmic 3-Hz delta slowing in the left anterior and inferior temporal regions from electrodes F9, T9, F7, and T7. Intracranial EEG shows synchronous delta slowing, primarily in anterior temporal and basal temporal regions. Lateral temporal and posterior temporal regions are rarely involved. Heterogeneous intracranial spikes are superimposed with delta slowing and are often not recordable at the scalp. LMTG, left midtemporal grid; LATS, left anterior temporal strip; LITS, left inferior temporal strip; LPTS: left posterior temporal strip; LPOS, Left posterior occipital strip.

379

Figure 13.10: Intracranial EEG recording of cortical interictal temporal delta activity (ITDA), interictal spikes, and seizure onset in a patient with left TLE. A similar cortical distribution is observed for cortical ITDA, interictal spikes, (arrow) and ictal-onset discharges, namely, anterior and basal temporal regions. LAT, left anterior temporal; LMTG: left midtemporal grid.

380

Subdural Electrode Corticography

the seizure-onset zone. On the contrary, multiple simultaneous electrode involvement may mean that they are recording propagated activity from a “remote” seizure onset. In TLE, focal onset is commonly observed in mesial disease, whereas regional onset is more common in patients with neocortical epilepsy. Occasionally, focal and regional onsets are recorded independently in the same patient and within the same temporal lobe (Fig. 13.11). Intracranial EEG recording provides a unique window for better understanding of seizure generation and propagation. Seizure-onset rhythms can be of nearly any EEG frequency band, including delta, theta, beta, alpha, and gamma (103). The most common onset pattern is low-voltage beta and gamma (15–40 Hz) activity (104–107). Such low-voltage fast activity is typical of both mesial temporal (Fig. 13.12) and extratemporal seizure onsets (76,99,108,109) (Fig. 13.13). It may or may not be preceded by changes of ongoing IEDs, such as an increment or a reduction in the spike frequency. Experimental findings in animal models of TLE suggest that low-voltage fast activity observed at seizure onset is associated with the reinforcement and synchronization of GABAergic inhibitory networks (107,110). This concept challenges the traditional theory that an increase in excitation and/ or a decrease in inhibition mark the transition to seizures in focal epilepsies. From a practical perspective, low-voltage fast activity at seizure onset has excellent localizing value (102,111,112). Moreover, the removal of the cortical region that generates such activity is associated with favorable surgical outcomes (106,113). High-voltage, low-frequency, delta and theta (1–5 Hz) rhythms are another common intracranial seizure-onset pattern (Fig. 13.14). Such low-frequency ictal onsets are most common in neocortical TLE, although they can sometimes be observed in extratemporal epilepsy. This seizure-onset pattern often appears as a short run of rhythmic, semi-rhythmic, or periodic spikes, sharp waves, or spike waves. High-voltage, low-frequency seizure onsets are commonly “regional.” In patients with TLE, this ictal-onset zone can be as large as 20 to 30 cm². Furthermore, these seizure patterns are recognizable on scalp EEG from their very onset, unlike low-voltage fast patterns that typically take cortical recruitment and a slowing of frequency to be evident (114). Although regional, and thus possibly reflecting seizure propagation from a distant source (115), several studies have suggested that the high-voltage, slow-frequency pattern does identify the onset zone of neocortical seizures (14,108,116). In any given patient, electro-clinical correlation may provide some insights for distinguishing these two possibilities. Commonly, EEG seizure onset tends to precede behavioral onset when intracranial electrodes record directly from the seizure-onset zone, whereas the opposite is true for propagated seizures.

381

High-voltage, low-frequency seizure onsets are generally associated with less-favorable surgical outcomes, compared to low-voltage, fast-frequency seizure onset. This may well be related to the fact that the etiologies of neocortical seizures, such as encephalitis, meningitis, or cortical dysplasias, produce more diffuse cortical damage and/or dysfunction. Other intracranial ictal-onset patterns, such as rhythmic sinusoidal pattern in theta and alpha range (5 to 9 Hz) and periodic low-frequency 1- to 2-Hz spiking, have also been described (105,117). Focal or generalized electrodecremental patterns can also be seen at seizure onset. They are commonly not considered to have a localizing value in identifying the seizure-onset zone, and their physiological underpinnings are uncertain (118,119). Also, potentially confusing and misinterpreted as seizure activity is intracranial EEG slowing at the time of seizure onset, which can be seen distant from the focus. This slowing does not evolve into an ictal pattern over time, but rather subsides. Ipsilateral frontal lobe slowing is not uncommon with temporal lobe seizures (120) (Fig. 13.15). The character of cortical seizure propagation determines seizure semiology and scalp EEG ictal patterns. Accordingly, the latter can provide useful insight into seizure lateralization and localization. Seizures propagate locally by gradually recruiting the cortex surrounding the onset zone. In temporal lobe seizures of mesial origin, focal, low-voltage, fast activity at seizure onset commonly slows to the alpha/theta range and propagates into basal, anterior, and lateral temporal cortex, and eventually into the extratemporal cortex (Fig. 13.15). Usually, sufficient source area and synchrony are achieved in the course of propagation to result in a recognizable scalp EEG rhythm. Scalp ictal onset is typically delayed by up to tens of seconds from the time of intracranial EEG seizure onset (14,121,122). Thus, scalp EEG patterns often reflect seizure propagation. By means of subcortical white matter tracks, seizures can spread to distant cortical areas or even the opposite hemisphere without recognizable local cortical recruitment. This may result in poorly localized or even falsely lateralized scalp EEG ictal rhythms (Fig. 13.16). Rarely, focal seizures can propagate explosively through both hemispheres, producing scalp EEG ictal onsets that cannot be localized or lateralized (Fig. 13.17). Controversy continues regarding the quantity and quality of intracranial EEG data that are needed to recommend surgical resection. No consensus exists as to how many seizures should be recorded. In patients with unifocal interictal spikes and seizures on scalp EEG, recording three similar seizures of similar onset may be sufficient for recommending a localized resection. However, in patients with multifocal interictal spikes, recording of

Figure 13.11: Focal and regional seizure onsets are seen in the same patient with left TLE. Top: Focal, low-voltage, fast-frequency seizure onset from left mesial temporal electrodes, LMTG 9, 17, 25. Bottom: High-voltage, low-frequency, regional neocortical temporal seizure onset. LMTG, left midtemporal grid; LATS, left anterior temporal strip; LITS, left inferior temporal strip; LPTS, left posterior temporal strip.

382

Figure 13.12: Subdural electrode recording of a left mesial temporal seizure. Focal, low-voltage, fast beta and gamma frequency ictal-onset rhythms are noted only from the antero-mesial temporal electrodes (LATS1 and LITS1). The ictal discharge then propagates along mesial, basal, lateral temporal and extratemporal pathways. Seizures were manifested clinically by ipsilateral motor automatisms and contralateral dystonic limb posturing.

Figure 13.13: Subdural electrode recording of a left parietal seizure. Early involvement of electrodes LPTG 36, 42, 52 is noted in this focal, low-voltage, fast beta frequency seizure onset. LPTG, left parietal temporal grid.

383

384

Subdural Electrode Corticography

Figure 13.14: Simultaneous intracranial (left) and scalp (right) EEG recording of a left temporal lobe seizure. The ictal-onset zone is neocortical (approximately 20 cm² in area), involving most of the inferolateral and lateral temporal electrodes. Given that the cortical ictal discharges are widespread and synchronous at onset, a 2-Hz left anterior inferior temporal ictal rhythm appears simultaneously on the scalp (marked by arrow). LMTG, left midtemporal grid; LATS, left anterior temporal strip; LITS, left inferior temporal strip; LPTS.

a minimum of five seizures may be necessary in order to exclude multifocal seizure onsets (71). In practice, however, intracranial recording cannot be continued indefinitely due to increased patient morbidity, when monitoring is extended from days to weeks. Theoretically, complete removal of the epileptogenic zone, the brain region necessary and sufficient to generate seizures, is necessary to achieve longterm, complete seizure freedom (123). Unfortunately, there is no reliable biomarker to define the extent of the epileptogenic zone at the present time. In practice, the seizure-onset zone is commonly used as a surrogate indicator. However, the relationships between seizure-onset zone, epileptogenic zone, and irritative zone remain poorly defined. Therefore, cortex of the seizureonset zone, which is involved in early seizure propagation, and adjacent tissue with active interictal spiking are often targeted for surgical resection.

High-frequency Oscillations Recently, HFOs are being considered as a potential biomarker of the epileptogenic network. With the advent of powerful modern amplifiers that can sample at a rate of 2,000 Hz or more, HFOs known as ripples

(80 to 250 Hz) and fast ripples (250 to 500 Hz) have been recorded from microelectrodes and commercially available intracranial macroelectrodes (subdural and depth) in patients with intractable focal epilepsy (124–126) (Fig. 13.18). HFOs in the range of 80 to 250 Hz (ripples) can be recorded in normal hippocampus and parahippocampus in both animals and humans (126). They are believed to reflect inhibitory field potentials, which facilitate information transfer by synchronizing neuronal activity over longer distances. However, HFOs in the range of 250 to 500 Hz (fast ripples) seem to be linked to epileptogenesis. They have been found in the seizure-onset zone from human ictal and interictal recordings (127). HFOs commonly occur with interictal spikes, but they are also found independently. It has been shown that HFOs have a tighter spatial correlation with seizure-onset zone than do interictal spikes. Accordingly, HFOs may be of greater value in localizing the seizure-onset zone (128). In fact, a significant correlation was found between the removal of cortex under electrode contacts with high rates of HFOs and good surgical outcomes (129,130). These observations suggest that HFOs are a promising biomarker of epileptogenesis. Whether they prove to be clinically reliable in the localization of seizure focus remains to be seen.

Figure 13.15: Simultaneous intracranial (top) and scalp (bottom) EEG recording of a left temporal lobe seizure. The cortical ictal-onset zone is approximately 2 cm² in area and involves only the most mesial electrodes (LATS1, LATS2). Ictal activity propagates gradually over the basal to inferolateral temporal cortex. When sufficient cortex is recruited into ­synchronous involvement, a similar 8-Hz left anterior inferior temporal ictal pattern is seen on scalp EEG (marked by arrow). Latency between intracranial and scalp ictal onsets is 15 seconds. The rhythmic 4.5-Hz frontal burst noted earlier on scalp EEG is a distant effect of the temporal lobe seizure onset. LFCS, left frontal central strip; LATS, left anterior temporal strip; LITS, left inferior temporal strip; LMTG, left midtemporal grid.

385

Figure 13.16: Simultaneous scalp (top) and intracranial (bottom) EEG recording of a right temporal lobe seizure. Scalp EEG ictal rhythms are a confusing mixture of bitemporal activity. The intracranial ictal-onset zone involves only the right mesiobasal temporal electrodes, and the seizure remains focal for approximately 11 seconds. Ictal activity then propagates to both the contralateral and ipsilateral temporal neocortex. Seizures were manifested clinically as mixed bimanual motor automatisms and dystonic posturing.

386

Figure 13.17: Simultaneous scalp (top) and intracranial (­bottom) EEG recording of a left temporal lobe seizure. Scalp EEG ictal (as indicated by arrow) was poorly localized and lateralized. Intracranial ictal-onset zone involves only the mesial temporal electrodes (LMTG1, LMTG9, and LMTG17). The ictal discharge remains focal for approximately 13 seconds and then abruptly spreads to the left lateral and posterior temporal, left frontal and left parietal cortices. Seizures were manifested clinically by hypermotor activity.

387

388

Subdural Electrode Corticography

Figure 13.18: High-frequency oscillations (HFOs, 100 to 300 Hz) recorded from mesial (electrodes LATS1, 2 and LITS1, 2) and basal (electrodes LMTG12, 9, 10, 17, 18, 25, 26) temporal cortex with commercial subdural electrodes in a patient with left TLE. The intracranial EEG was recorded at a sampling rate of 2,000 Hz. HFOs were revealed with input filters of 100 to 500 Hz, a 60 mm per second timescale, and a sensitivity of 10 µV per mm. LOFS, left orbitofrontal strip; LATS, left anterior temporal strip; LITS, left inferior temporal strip; LMTG, left midtemporal grid.

Electrical stimulation and functional localization When placed in large geometrical arrays, subdural electrodes are well-suited for functional mapping of cortex. Electrical stimulation has been a reliable technique for functional localization over the last several decades (37,131). It can be performed intraoperatively under local anesthesia or by means of chronically implanted subdural electrodes (132–134). While more convenient, intraoperative electrical stimulation is subject to significant limitations. It often requires a high level of patient’s cooperation, which may not be possible in certain patient populations, such as children or developmentally delayed patients. Appropriate levels of sustained anesthesia

are difficult to achieve. Patients may have difficulties of staying awake or can be easily fatigued, which will compromise the validity of testing. The duration of testing and the modalities of stimulation are also limited. By comparison, electrical stimulation with chronically implanted subdural electrodes allows an extensive period of testing with more comprehensive protocols. A high level of patient cooperation can be maintained throughout the course of testing over periods of hours or even multiple sessions over several days. Electrical stimulation for functional localization is commonly performed after sufficient habitual seizures have been recorded for localizing the seizure focus. Because electrical stimulation of the cortex can also provoke seizures, AEDs should be resumed prior to mapping. A short

Subdural Electrode Corticography

run of electrical stimulation between two adjacent subdural electrodes is typically used. Biphasic 100 to 300 μs pulses of 2 to 14 mA are delivered in trains of 50 pulses per second for a period of 2 to 5 seconds. Typically, stimulation begins at a low amperage (2 to 4 mA), and it is gradually increased in 2 to 4 mA increments. End points include no effect with the highest current or a sensory or motor response or language disruption at a lesser current. Sensory and motor responses tend to occur at relatively lower stimulus intensities, while language disruption may require a higher current. Effects induced only at high currents are less localizing because a wider area of cortex is activated or inhibited. Cortical afterdischarges, which may also be evoked by stimulation, are a sign of cortical irritability and seizure susceptibility. Afterdischarges appear as rhythmic repetitive high-voltage spikes and spike-wave complexes lasting seconds to minutes. They usually are subclinical or are associated with subtle behavioral changes. Because afterdischarges may propagate to adjacent cortical areas, trials associated with afterdischarges should not be used for functional localization. It is, however, important to recognize the limitations of cortical mapping. Resections guided by negative or positive functional mapping results does not necessarily guarantee the absence of postsurgical functional deficits (135).

Conclusion For decades, intracranial EEG recording with subdural electrodes has been successfully used in localizing seizure foci, when scalp EEG provides insufficient information. Recordings from subdural electrodes provide extensive neocortical coverage and can determine the extent, as well as position, of an epileptic focus. They are especially well-suited for characterizing neocortical epilepsy. In addition, subdural electrodes provide a means for identifying sensory, motor, and language cortex through electrical stimulation techniques. This information allows maximal resection of a seizure focus, while sparing eloquent function. Implantation of subdural electrodes does entail some risk of morbidity. Also, no invasive EEG study, even though considered the “gold standard” for localization, can always provide enough information to assure complete seizure control after surgery. Seizures may continue or recur after seemingly optimal resections. Accordingly, an intracranial EEG study with subdural electrodes should be undertaken only after careful consideration and with a reasonable likelihood that localization of the seizure focus can be achieved.

389

REFERENCES 1. Kwan P, Brodie MJ. Early identification of refractory epilepsy. N Engl J Med 2000;342:314–319. 2. Engel J Jr, Wiebe S, French J, et al. Practice parameter: temporal lobe and localized neocortical resections for epilepsy: report of the Quality Standards Subcommittee of the American Academy of Neurology, in association with the American Epilepsy Society and the American Association of Neurological Surgeons. Neurology 2003;60:538–547. 3. Wiebe S, Blume WT, Girvin JP, et al; Effectiveness and Efficiency of Surgery for Temporal Lobe Epilepsy Study Group. A randomized, controlled trial of surgery for temporallobe epilepsy. N Engl J Med 2001;345:311–318. 4. Ebersole JS. Non-invasive pre-surgical evaluation with EEG/MEG source analysis. Electroencephalogr Clin Neurophysiol Suppl 1999;50:167–174. 5. Mariottini A, Lombroso CT, DeGirolami U, et al. Operative results without invasive monitoring in patients with frontal lobe epileptogenic lesions. Epilepsia 2001;42:1308–1315. 6. Davidson S, Falconer MA. Outcome of surgery in 40 children with temporal-lobe epilepsy. Lancet 1975;1:1260–1263. 7. Polkey CE. Epilepsy surgery: non-invasive versus invasive focus localization. What is needed from the neurosurgical point of view. Acta Neurol Scand Suppl 1994;152:183–186. 8. Blume WT. Current trends in electroencephalography. Curr Opin Neurol 2001;14:193–197. 9. Zumsteg D, Wieser HG. Presurgical evaluation: current role of invasive EEG. Epilepsia 2000;41(suppl 3):S55–S60. 10. Diehl B, Luders HO. Temporal lobe epilepsy: when are invasive recordings needed? Epilepsia 2000;41(suppl 3):S61–S74. 11. Lantz G, Michel CM, Seeck M, et al. Space-oriented segmentation and 3-dimensional source reconstruction of ictal EEG patterns. Clin Neurophysiol 2001;112:688–697. 12. Ebersole JL, Liss WJ, Frissell CA. FORUM: restoration of stream habitats in the Western United States: restoration as reexpression of habitat capacity. Environ Manage 1997;21:1–14. 13. Gloor P. Neuronal generators and the problem of localization in electroencephalography: application of volume conductor theory to electroencephalography. J Clin Neurophysiol 1985;2:327–354. 14. Tao JX, Baldwin M, Ray A, et al. The impact of cerebral source area and synchrony on recording scalp electroencephalography ictal patterns. Epilepsia 2007;48:2167–2176. 15. Michel CM, Murray MM, Lantz G, et al. EEG source imaging. Clin Neurophysiol 2004;115:2195–2222. 16. Lee SK, Yun CH, Oh JB, et al. Intracranial ictal onset zone in nonlesional lateral temporal lobe epilepsy on scalp ictal EEG. Neurology 2003;61:757–764. 17. Sammaritano M, de Lotbiniere A, Andermann F, et al. False lateralization by surface EEG of seizure onset in patients with temporal lobe epilepsy and gross focal cerebral lesions. Ann Neurol 1987;21:361–369. 18. Spencer SS, Williamson PD, Bridgers SL, et al. Reliability and accuracy of localization by scalp ictal EEG. Neurology 1985;35:1567–1575. 19. Feindel W, Penfield W, Jasper H. Localization of epileptic discharge in temporal lobe automatism. Trans Am Neurol Assoc 1952;56:14–17. 20. Penfield W, Baldwin M. Temporal lobe seizures and the technic of subtotal temporal lobectomy. Ann Surg 1952;136:625–634. 21. Almeida AN, Martinez V, Feindel W. The first case of invasive EEG monitoring for the surgical treatment of epilepsy: historical significance and context. Epilepsia 2005;46:1082–1085. 22. Bailey P, Gibbs FA. The surgical treatment of psychomotor epilepsy. J Am Med Assoc 1951;145:365–370.

390

Subdural Electrode Corticography

23. Gloor P. Preoperative electroencephalographic investigation in temporal lobe epilepsy: extracranial and intracranial recordings. Can J Neurol Sci 1991;18:554–558. 24. Hitzig E. Hughlings Jackson and the cortical motor centres in the light of physiological research: being the second Hughlings Jackson Lecture delivered before the Neurological Society of London. Br Med J 1900;2:1564–1565. 25. Jackson JH. A study of convulsions. Trans St Andrews Med Grad Assoc 1870;3:162–204. 26. Jackson JH. On the anatomical investigation of epilepsy and epileptiform convulsions. Br Med J 1873;1:531–533. 27. Horsley V, Schafer EA. Experiments on the character of the muscular contractions which are evoked by excitation of the various parts of the motor tract. J Physiol 1886;7:96–110. 28. Horsley V. Remarks on ten consecutive cases of operations upon the brain and cranial cavity to illustrate the details and safety of the method employed. Br Med J 1887;1:863–865. 29. Feindel W, Leblanc R, de Almeida AN. Epilepsy surgery: historical highlights 1909–2009. Epilepsia 2009;50(suppl 3):131–151. 30. Berger H. Uber das Elektrenkephalogram des Menschen. Arch Psychiatr Nervenkr 1929;87:527–570. 31. Adrian ED, Matthews BHC. The Berger rhythm: potential changes from the occipital lobes in man. Brain 1934;57:355–385. 32. Jasper HH, Carmichael L. Electrical potentials from the intact human brain. Science 1935;81:51–53. 33. Foerster O, Altenburger O. Electrobiologische vorgange an der menschlischen hirnrinde. Dtsch Zeitsch Nervenheilk 1935;135:277–288. 34. Gibbs FA, Gibbs EL, Lennox WG. Epilepsy: a paroxysmal cerebral dysrhythmia. Brain 1937;60:377–388. 35. Gibbs FA, Lennox WG, Gibbs EL. The electro-encephalogram in epilepsy and in conditions of impaired consciousness. Arch Neurol Psychiatry 1936;34:1133–1148. 36. Gibbs FA, Gibbs EL, Lennox WG. Cerebral dysrhythmias of epilepsy: measures for their control. Arch Neurol Psychiatry 1938;39:298–314. 37. Penfield W, Jasper HHJ. Epilepsy and functional anatomy of the human brain. Boston, MA: Little Brown, 1954. 38. Bickford RG, Dodge HW Jr, Sem-Jacobsen CW, et al. Studies on the electrical structure and activation of an epileptogenic focus. Proc Staff Meetings Mayo Clin 1953;28:175–180. 39. Brazier MA. Depth recordings from the amygdaloid region in patients with temporal lobe epilepsy. Electroencephalogr Clin Neurophysiol 1956;8:532–533. 40. Kellaway P. Depth recording in focal epilepsy. Electroencephalogr Clin Neurophysiol 1956;527–528. 41. Bancaud J, Talairach J, Bonis A, et al. La stéréoencéphalographie dans l’épilepsie. Informations neuro-physio-pathologiques apportées par l’investigation fonctionnelle stéréotaxique. Paris, France: Masson, 1965. 42. Talairach J, Bancaud J. Stereotaxic approach to epilepsy: methodology of anatomo-­ functional stereotaxic investigations. Prog Neurol Surg 1973;5:297–354. 43. Wyler AR, Ojemann GA, Lettich E, et al. Subdural strip electrodes for localizing ­epileptogenic foci. J Neurosurg 1984;60:1195–1200. 44. Wyler AR. Subdural strip electrodes in surgery of epilepsy. In: Lüders H, ed. Epilepsy surgery. New York, NY: Raven Press, 1991:395–398. 45. Quesney LF, Gloor P. Localization of epileptic foci. Electroencephalogr Clin N ­ europhysiol Suppl 1985;37:165–200. 46. Arya R, Mangano FT, Horn PS, et al. Adverse events related to extraoperative invasive EEG monitoring with subdural grid electrodes: a systematic review and meta-analysis. Epilepsia 2013;54:828–839. 47. Murphy MA, O’Brien TJ, Morris K, et al. Multimodality image-guided surgery for the treatment of medically refractory epilepsy. J Neurosurg 2004;100:452–462.

48. Hogan RE, Lowe VJ, Bucholz RD. Triple-technique (MR imaging, single-photon ­emission CT, and CT) coregistration for image-guided surgical evaluation of patients with intractable epilepsy. AJNR Am J Neuroradiol 1999;20:1054–1058. 49. Knake S, Triantafyllou C, Wald LL, et al. 3T phased array MRI improves the presurgical evaluation in focal epilepsies: a prospective study. Neurology 2005;65:1026–1031. 50. Kovalev D, Spreer J, Honegger J, et al. Rapid and fully automated visualization of subdural electrodes in the presurgical evaluation of epilepsy patients. AJNR Am J Neuroradiol 2005;26:1078–1083. 51. Schulze-Bonhage AH, Huppertz HJ, Comeau RM, et al. Visualization of subdural strip and grid electrodes using curvilinear reformatting of 3D MR imaging data sets. AJNR Am J Neuroradiol 2002;23:400–403. 52. Mahvash M, Konig R, Wellmer J, et al. Coregistration of digital photography of the human cortex and cranial magnetic resonance imaging for visualization of subdural ­electrodes in epilepsy surgery. Neurosurgery 2007;61:340–344; discussion 344–345. 53. Wellmer J, von Oertzen J, Schaller C, et al. Digital photography and 3D MRI-based multimodal imaging for individualized planning of resective neocortical epilepsy surgery. Epilepsia 2002;43:1543–1550. 54. Hunter JD, Hanan DM, Singer BF, et al. Locating chronically implanted subdural electrodes using surface reconstruction. Clin Neurophysiol 2005;116:1984–1987. 55. Tao JX, Hawes-Ebersole S, Baldwin M, et al. The accuracy and reliability of 3D CT/MRI co-registration in planning epilepsy surgery. Clin Neurophysiol 2009;120:748–753. 56. Hill DL, Batchelor PG, Holden M, et al. Medical image registration. Phys Med Biol 2001;46:R1–R45. 57. Hill DL, Hawkes DJ, Crossman JE, et al. Registration of MR and CT images for skull base surgery using point-like anatomical features. Br J Radiol 1991;64:1030–1035. 58. Dykstra AR, Chan AM, Quinn BT, et al. Individualized localization and cortical surfacebased registration of intracranial electrodes. Neuroimage 2012;59:3563–3570. 59. Yang AI, Wang X, Doyle WK, et al. Localization of dense intracranial electrode arrays using magnetic resonance imaging. Neuroimage 2012;63:157–165. 60. Wyler AR, Walker G, Somes G. The morbidity of long-term seizure monitoring using subdural strip electrodes. J Neurosurg 1991;74:734–737. 61. Van Gompel JJ, Worrell GA, Bell ML, et al. Intracranial electroencephalography with subdural grid electrodes: techniques, complications, and outcomes. Neurosurgery 2008;63:498–505; discussion 505–496. 62. Wellmer J, von der Groeben F, Klarmann U, et al. Risks and benefits of invasive epilepsy surgery workup with implanted subdural and depth electrodes. Epilepsia 2012;53:1322–1332. 63. Tanriverdi T, Ajlan A, Poulin N, et al. Morbidity in epilepsy surgery: an experience based on 2449 epilepsy surgery procedures from a single institution. J Neurosurg 2009;110:1111–1123. 64. Hamer HM, Morris HH, Mascha EJ, et al. Complications of invasive video-EEG monitoring with subdural grid electrodes. Neurology 2002;58:97–103. 65. Cossu M, Cardinale F, Castana L, et al. Stereoelectroencephalography in the presurgical evaluation of focal epilepsy: a retrospective analysis of 215 procedures. Neurosurgery 2005;57:706–718; discussion 706–718. 66. Wong CH, Birkett J, Byth K, et al. Risk factors for complications during intracranial electrode recording in presurgical evaluation of drug resistant partial epilepsy. Acta Neurochir 2009;151:37–50. 67. Swartz BE, Rich JR, Dwan PS, et al. The safety and efficacy of chronically implanted subdural electrodes: a prospective study. Surg Neurol 1996;46:87–93. 68. Ozlen F, Asan Z, Tanriverdi T, et al. Surgical morbidity of invasive monitoring in e­ pilepsy surgery: an experience from a single institution. Turk Neurosurg 2010;20:364–372.

Subdural Electrode Corticography 69. Araki T, Otsubo H, Makino Y, et al. Efficacy of dexamathasone on cerebral swelling and seizures during subdural grid EEG recording in children. Epilepsia 2006;47:176–180. 70. Sperling MR. Intracranial electroencephalography. In: Ebersole JS, Pedley TA, eds. ­Current practice of clinical electrencephalography. Philadelphia, PA: Lippincott Williams & Wilkins, 2003:639–680. 71. Seeck M, Schomer DL, Niedermeyer E. Intracranial monitoring: depth, subdural, and foramen ovale electrodes. In: Schomer D, Lopes da Silvia FH, eds. Niedermeyer’s electroencephalography, 6th ed. Philadelphia, PA: Lippincott Williams & Wilkins, 2011:677–724. 72. Tao JX, Baldwin M, Hawes-Ebersole S, et al. Cortical substrates of scalp EEG epileptiform discharges. J Clin Neurophysiol 2007;24:96–100. 73. Abraham K, Marsan CA. Patterns of cortical discharges and their relation to routine scalp electroencephalography. Electroencephalogr Clin Neurophysiol 1958;10:447–461. 74. Ebersole JS. Defining epileptogenic foci: past, present, future. J Clin Neurophysiol 1997;14:470–483. 75. Williams D, Parsons-Smith G. Cortical rhythms not seen in the electroencephalogram. Brain 1950;73:191–202. 76. Wendling F, Bartolomei F, Bellanger JJ, et al. Epileptic fast intracerebral EEG activity: evidence for spatial decorrelation at seizure onset. Brain 2003;126:1449–1459. 77. Cooper R, Winter AL, Crow HJ, et al. Comparison of subcortical, cortical and scalp activity using chronically indwelling electrodes in man. Electroencephalogr Clin Neurophysiol 1965;18:217–228. 78. Tao JX, Ray A, Hawes-Ebersole S, et al. Intracranial EEG substrates of scalp EEG interictal spikes. Epilepsia 2005;46:669–676. 79. Rosenow F, Luders H. Presurgical evaluation of epilepsy. Brain 2001;124:1683–1700. 80. Palmini A. The irritative zone evaluated with invasive recordings. In: Lüders HO, ed. Textbook of epilepsy surgery. London, England: Informa Healthcare, 2008:521–529. 81. Asano E, Muzik O, Shah A, et al. Quantitative interictal subdural EEG analyses in children with neocortical epilepsy. Epilepsia 2003;44:425–434. 82. Hufnagel A, Dumpelmann M, Zentner J, et al. Clinical relevance of quantified intracranial interictal spike activity in presurgical evaluation of epilepsy. Epilepsia 2000;41:467–478. 83. Marsh ED, Peltzer B, Brown MW III, et al. Interictal EEG spikes identify the region of electrographic seizure onset in some, but not all, pediatric epilepsy patients. Epilepsia 2010;51:592–601. 84. Kobayashi K, Merlet I, Gotman J. Separation of spikes from background by independent component analysis with dipole modeling and comparison to intracranial recording. Clin Neurophysiol 2001;112:405–413. 85. Wyllie E, Luders H, Morris HH III, et al. Clinical outcome after complete or partial cortical resection for intractable epilepsy. Neurology 1987;37:1634–1641. 86. Krsek P, Maton B, Jayakar P, et al. Incomplete resection of focal cortical dysplasia is the main predictor of poor postsurgical outcome. Neurology 2009;72:217–223. 87. Paolicchi JM, Jayakar P, Dean P, et al. Predictors of outcome in pediatric epilepsy ­surgery. Neurology 2000;54:642–647. 88. Asano E, Juhasz C, Shah A, et al. Role of subdural electrocorticography in prediction of long-term seizure outcome in epilepsy surgery. Brain 2009;132:1038–1047. 89. Bautista RE, Cobbs MA, Spencer DD, et al. Prediction of surgical outcome by interictal epileptiform abnormalities during intracranial EEG monitoring in patients with extrahippocampal seizures. Epilepsia 1999;40:880–890. 90. Walter WG. The location of cerebral tumors by electroencephalography. Lancet 1936;2:305–308. 91. Murphy JT, Gloor P, Yamamoto YL, et al. A comparison of electroencephalography and brain scan in supratentorial tumors. N Engl J Med 1967;276:309–313.

391

92. Gloor P, Ball G, Schaul N. Brain lesions that produce delta waves in the EEG. Neurology 1977;27:326–333. 93. Blume WT, Borghesi JL, Lemieux JF. Interictal indices of temporal seizure origin. Ann Neurol 1993;34:703–709. 94. Reiher J, Beaudry M, Leduc CP. Temporal intermittent rhythmic delta activity (TIRDA) in the diagnosis of complex partial epilepsy: sensitivity, specificity and predictive value. Can J Neurol Sci 1989;16:398–401. 95. Koutroumanidis M, Binnie CD, Elwes RD, et al. Interictal regional slow activity in temporal lobe epilepsy correlates with lateral temporal hypometabolism as imaged with 18FDG PET: neurophysiological and metabolic implications. J Neurol Neurosurg ­Psychiatry 1998;65:170–176. 96. Gambardella A, Gotman J, Cendes F, et al. Focal intermittent delta activity in patients with mesiotemporal atrophy: a reliable marker of the epileptogenic focus. Epilepsia 1995;36:122–129. 97. Tao JX, Chen XJ, Baldwin M, et al. Interictal regional delta slowing is an EEG marker of epileptic network in temporal lobe epilepsy. Epilepsia 2011;52:467–476. 98. Pacia SV, Ebersole JS. Intracranial EEG in temporal lobe epilepsy. J Clin Neurophysiol 1999;16:399–407. 99. Lee SA, Spencer DD, Spencer SS. Intracranial EEG seizure-onset patterns in neocortical epilepsy. Epilepsia 2000;41:297–307. 100. So N, Gloor P, Quesney LF, et al. Depth electrode investigations in patients with bitemporal epileptiform abnormalities. Ann Neurol 1989;25:423–431. 101. Jung WY, Pacia SV, Devinsky R. Neocortical temporal lobe epilepsy: intracranial EEG features and surgical outcome. J Clin Neurophysiol 1999;16:419–425. 102. Gotman J, Levtova V, Olivier A. Frequency of the electroencephalographic discharge in seizures of focal and widespread onset in intracerebral recordings. Epilepsia 1995;36:697–703. 103. Stefan H, Hopfengartner R, Kreiselmeyer G, et al. Interictal triple ECoG characteristics of temporal lobe epilepsies: an intraoperative ECoG analysis correlated with surgical outcome. Clin Neurophysiol 2008;119:642–652. 104. McBride MC, Binnie CD, Janota I, et al. Predictive value of intraoperative electrocorticograms in resective epilepsy surgery. Ann Neurol 1991;30:526–532. 105. Spencer SS, Guimaraes P, Katz A, et al. Morphological patterns of seizures recorded intracranially. Epilepsia 1992;33:537–545. 106. Alarcon G, Binnie CD, Elwes RD, et al. Power spectrum and intracranial EEG patterns at seizure onset in partial epilepsy. Electroencephalogr Clin Neurophysiol 1995;94:326–337. 107. de Curtis M, Gnatkovsky V. Reevaluating the mechanisms of focal ictogenesis: the role of low-voltage fast activity. Epilepsia 2009;50:2514–2525. 108. Pacia SV, Ebersole JS. Intracranial EEG substrates of scalp ictal patterns from temporal lobe foci. Epilepsia 1997;38:642–654. 109. Salanova V, Andermann F, Olivier A, et al. Occipital lobe epilepsy: electroclinical manifestations, electrocorticography, cortical stimulation and outcome in 42 patients treated between 1930 and 1991. Surgery of occipital lobe epilepsy. Brain 1992;115(Pt 6):1655–1680. 110. Gnatkovsky V, Librizzi L, Trombin F, et al. Fast activity at seizure onset is mediated by inhibitory circuits in the entorhinal cortex in vitro. Ann Neurol 2008;64:674–686. 111. Fisher RS, Webber WR, Lesser RP, et al. High-frequency EEG activity at the start of seizures. J Clin Neurophysiol 1992;9:441–448. 112. Guggisberg AG, Kirsch HE, Mantle MM, et al. Fast oscillations associated with interictal spikes localize the epileptogenic zone in patients with partial epilepsy. Neuroimage 2008;39:661–668.

392

Subdural Electrode Corticography

113. Wetjen NM, Marsh WR, Meyer FB, et al. Intracranial electroencephalography seizure onset patterns and surgical outcomes in nonlesional extratemporal epilepsy. J Neurosurg 2009;110:1147–1152. 114. Aghakhani Y, Rosati A, Olivier A, et al. The predictive localizing value of tonic limb posturing in supplementary sensorimotor seizures. Neurology 2004;62:2256–2261. 115. Gibbs FA, Amador L, Rich C. Electro-encephalographic findings and therapeutic resluts in surgical treatment of psychomotor epilepsy. In: Baldwin M, Bailey P, eds. Temporal lobe epilepsy. Springfeild, IL: Charles C. Thomas, 1958:358–367. 116. Vossler DG, Kraemer DL, Haltiner AM, et al. Intracranial EEG in temporal lobe ­epilepsy: location of seizure onset relates to degree of hippocampal pathology. Epilepsia 2004;45:497–503. 117. Spencer SS, Kim J, Spencer DD. Ictal spikes: a marker of specific hippocampal cell loss. Electroencephalogr Clin Neurophysiol 1992;83:104–111. 118. Baumgartner C, Flint R, Tuxhorn I, et al. Supplementary motor area seizures: propagation pathways as studied with invasive recordings. Neurology 1996;46:508–514. 119. Arroyo S, Lesser RP, Fisher RS, et al. Clinical and electroencephalographic ­evidence for sites of origin of seizures with diffuse electrodecremental pattern. Epilepsia 1994;35:974–987. 120. Blumenfeld H, Rivera M, McNally KA, et al. Ictal neocortical slowing in temporal lobe epilepsy. Neurology 2004;63:1015–1021. 121. Ebersole JS, Pacia SV. Localization of temporal lobe foci by ictal EEG patterns. Epilepsia 1996;37:386–399. 122. Alarcon G, Guy CN, Binnie CD, et al. Intracerebral propagation of interictal activity in partial epilepsy: implications for source localisation. J Neurol Neurosurg Psychiatry 1994;57:435–449. 123. Engel J, Pedley TA. Epilepsy: a comprehensive textbook. Philadelphia, PA: Lippincott Williams & Wilkins, 2007.

124. Bragin A, Engel J Jr, Wilson CL, et al. High-frequency oscillations in human brain. ­Hippocampus 1999;9:137–142. 125. Jirsch JD, Urrestarazu E, LeVan P, et al. High-frequency oscillations during human focal seizures. Brain 2006;129:1593–1608. 126. Engel J Jr, Bragin A, Staba R, et al. High-frequency oscillations: what is normal and what is not? Epilepsia 2009;50:598–604. 127. Jacobs J, Levan P, Chatillon CE, et al. High frequency oscillations in intracranial EEGs mark epileptogenicity rather than lesion type. Brain 2009;132:1022–1037. 128. Jacobs J, LeVan P, Chander R, et al. Interictal high-frequency oscillations (80–500 Hz) are an indicator of seizure onset areas independent of spikes in the human epileptic brain. Epilepsia 2008;49:1893–1907. 129. Jacobs J, Zijlmans M, Zelmann R, et al. High-frequency electroencephalographic oscillations correlate with outcome of epilepsy surgery. Ann Neurol 2010;67:209–220. 130. Wu JY, Sankar R, Lerner JT, et al. Removing interictal fast ripples on electrocortico­ graphy linked with seizure freedom in children. Neurology 2010;75:1686–1694. 131. Ojemann GA, Fried I, Lettich E. Electrocorticographic (ECoG) correlates of language, I: desynchronization in temporal language cortex during object naming. Electroencephalogr Clin Neurophysiol 1989;73:453–463. 132. Ojemann GA. Individual variability in cortical localization of language. J Neurosurg 1979;50:164–169. 133. Ojemann GA. Neurosurgical management of epilepsy: a personal perspective in 1983. Appl Neurophysiol 1983;46:11–18. 134. Luders H, Lesser RP, Hahn J, et al. Basal temporal language area. Brain 1991;114 (Pt 2):743–754. 135. Seeck M, Pegna AJ, Ortigue S, et al. Speech arrest with stimulation may not reliably ­predict language deficit after epilepsy surgery. Neurology 2006;66:592–594.

14

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography) PHILIPPE KAHANE • FRANÇOIS DUBEAU

Introduction Historical Perspective Technical Considerations Intracranial Electrodes: Overview Intracerebral Electrodes: Characteristics, Insertion, Placement, and Location Advantages and Drawbacks of Intracerebral Depth Electrodes Risks and Complications Recording Procedure Technical Aspects Practical Aspects Indications of Intracerebral Depth Electrode EEG Recording Indications for SEEG in Temporal Lobe Epilepsy

Introduction In spite of the increasing use of noninvasive and nonelectrophysiological tests in the presurgical evaluation of patients with refractory epilepsies (1), invasive EEG methods remain necessary in a number of patients. In the

Indications for SEEG in Extratemporal Lobe Epilepsy Indications for SEEG in Lesional and Nonlesional Focal Epilepsies When Intracerebral EEG Is Not Indicated Intracerebral EEG Monitoring: Interpretation of EEG Data Normal Physiological EEG Activity Artifacts Abnormal Interictal EEG Activity Seizures Onset (Seizure-Onset Zone) and Propagation Electrical Stimulation SEEG-Guided Radiofrequency Thermocoagulation Conclusion Acknowledgments References

1990s and during the onset of the 2000s, the overall proportion of patients evaluated by invasive recordings was probably decreasing. It is our impression, however, that over the more recent years, invasive EEG techniques have gained interest and are more widely and frequently used across c­ enters and countries; notably, a number of cases that were not considered for epilepsy

393

394

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

surgery in the past can nowadays benefit from such investigations (e.g., MRI-negative cases). This apparently paradoxical trend is explained by the perception that invasive EEG methods are still considered the gold standard and the only approach that can provide direct and reliable indication of the epileptic generator. Invasive EEG is perceived as being able to provide specific or accurate localization of the epileptic generator and hence good surgical results with relatively small resections. Moreover, the techniques now appear safer, and the morbidity is acceptable. Intracranial EEG remains, however, an invasive procedure, is expensive and time-consuming, necessitating considerable human and technical resources, and therefore requires proper justifications. Also, as we will see later, the use of invasive EEG often indicates by itself a poor surgical outcome. The first motivation to use invasive EEG in patients with difficult-totreat focal epilepsies is the added value of invasive monitoring to precisely localize the seizure generator and to tailor surgical resections (2) (Table 14.1). ­Invasive EEG also permits to define seizure propagation and the understanding of the neuronal networks involved during epileptic attacks; in some centers, larger resections are performed to include not only the area of seizure generation but also the area(s) containing the first or early propagation of the epileptic discharges (for a review of the ­“Epileptogenic Zone,” see Lüders and Kahane [3]). During acute or chronic invasive EEG, ­cortical electrical stimulation can be used to monitor cortical function (e.g., functional mapping for speech, motor, or sensory activities), to define the cortical areas responsible for the ictal symptoms, the electrical stimulation mimicking the effects of a spontaneous epileptic discharges (4), or to determine low-threshold area(s). By defining the seizure generator and the propagation pathways of the ictal discharge and by providing functional TABLE 14.1

Indications for Invasive Electroencephalography

1. To define seizure generator and tailor surgical resection: • The added value of invasive monitoring to precisely localize the seizure generator. • Definition of seizure propagation. • Electrical stimulation evoked clinical responses mimicking the effect of spontaneous epileptic discharges on cortex. • Comparison of afterdischarge thresholds. 2. Mapping of cortical function. 3. Relationship existing between lesion and seizure focus. 4. For prognosis.

information, intracranial EEG helps in the planning of epilepsy surgery (localization and completeness of resection). Invasive EEG can be indicated to determine the relationship existing between lesions and seizure generator; this relationship may vary with different types of developmental or ­acquired lesions (5–8). I­ntracranial studies can help in prognosis: for instance, in case of bilateral temporal lobe epilepsy (TLE), to determine if a predominance or laterality of epileptogenesis exists (9). Finally, given the privileged access to human brain structures invasive EEG provides, the methods are often used to study and understand the mechanisms underlying normal and abnormal neurophysiology, to measure evoked potentials, or to study neuronal or field potentials, in passive conditions or during ­experimental paradigms (10,11).

Historical Perspective Brain electrical activity was initially measured by direct intracranial recording, and Caton, in 1875 (12), was probably the first to demonstrate the existence of brain electricity by inserting electrodes, through burr holes (openings in the skull), in the gray matter of monkeys. Hans Berger in 1929 (13), for the first time in a human, detected brain electrical potentials through a cranial bone defect and proposed the term “electroencephalography” to describe the method of recording this brain electrical activity. In 1935, Foerster and Altenburger (14) made the first direct recording of epileptic activity during a craniotomy; the electrocorticography (ECoG) was born and the method rapidly expanded due to the work of Gibbs et al. (15) in Boston (1937), Penfield and Jasper (16) in Montreal (1954), and Ajmone Marsan and Baldwin (17) in Bethesda (1958). This recording method became extensively and widely used to detect abnormal brain potentials and define the epileptic focus, and to determine the extension of the surgical resection in patients with refractory epilepsies. The ECoG technique, usually performed under local anesthesia, also allowed determining the functional anatomy of the human brain and, particularly, the mapping of the speech and sensorimotor areas, leading to the famous homunculus (16). Human intracerebral recordings in epileptic patients were introduced in the late 1940s and used to study the pathophysiology of “grand mal” and “petit mal” epileptic seizures; electrodes were implanted in various subcortical areas thought to be involved in the generation of these seizure types ­(18–23). Following these pioneered studies, clinicians began to use chronically implanted “depth” electrodes in patients suffering from focal

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

seizures in order to address the shortcomings of ECoG (24–31): ECoG allowed direct brain recording but required an a priori definition of the origin of seizures, influencing therefore the type of surgical approach and of the explored cortical regions. It appeared then that chronic intracerebral neurophysiological studies of the brain using “depth” electrodes would permit a better access to any cortical and subcortical areas involved in the epileptic process, and for a longer period of time. In these pioneering depth EEG recordings, electrodes were not inserted by stereotactic implantation methods, and therefore could not precisely target the anatomical structures. Also, neurophysiologists, as with ECoG, were looking mainly at the ­interictal spikes. The introduction by Bancaud and Talairach of atlases providing spatial coordinates of the telencephalic structures and the development of neurosurgical techniques that allowed to target and reach such ­coordinates (32,33) permitted more accurate and safer (because they included the visualization of blood vessels) placement of the electrodes. They also demonstrated the importance of using electroclinical information obtained during spontaneous seizures (5,34–38); this approach, referred to as “anatomo-electroclinical correlations,” was based on the assumption that the chronological occurrence of ictal clinical signs reflects the spatiotemporal organization of the epileptic discharge within the brain. By combining anatomo-electroclinical correlations together with the stereotactic placement of intracerebral electrodes, Bancaud and Talairach elaborated a comprehensive methodology, the stereo-­electro-­ encephalography, or SEEG, that allowed studying, in each individual, the origin and spread of ictal discharges. The implantation strategy was thus “custom-tailored” and consisted in the simultaneous investigation, in all planes of the intracranial space, of brain structures that might give rise to ictal clinical signs. Other intracerebral depth electrode EEG procedures and strategies were developed (39–45) with the primary aim to demonstrate, for instance, the side of seizure onset in TLE or to differentiate frontal lobe seizures from temporal lobe seizures. Intracerebral targets and electrode trajectories were in such cases standardized to avoid biasing the exploration strategy in f­ avor of one’s preferred localizing hypothesis (46). The underlying concept of these depth electrode EEG recordings was different from the SEEG method of Bancaud and Talairach, but allowed to perform standardized surgical resections, such as anterior temporal resections. The introduction of prolonged video–EEG monitoring with computer-assisted detection of interictal and ictal epileptiform discharges considerably increased our ability to

395

observe and study seizure semiology with both scalp and intracranial EEG methods (47–49). In parallel, the ECoG methodology evolved with the development of electrodes embedded in strips or grids that could be placed subdurally (or less often epidurally) in chronic conditions, allowing seizure recording and electrical stimulation studies over large portions of the cortical surface (50,51). These recording methods are also now widely used and constitute, together with SEEG, the main approaches to directly record normal and abnormal brain activities in the epileptic patients.

Technical considerations Intracranial Electrodes: Overview Different types of electrodes can be used to record the intracranial EEG activity, for example, intracerebral depth electrodes, subdural grids or strips, and epidural electrodes (Table 14.2). Sphenoidal and foramen ovale electrodes are usually considered as semi-invasive and will not be discussed here. Usually, clinicians develop an expertise with a method, become familiar with it, and use preferentially or only one type of approach. The intracranial method is also selected depending on the question posed by a given epilepsy problem; an approach using intracerebral depth electrodes is ­favored if the epileptic generator is thought to be located deep in the brain, while cortical grids or strips are considered if the generator is thought to be over the brain convexity, basal cortex, or the interhemispheric cortex. Combinations of electrode types, for example, grids and intracerebral electrodes, or subdural strips and intracerebral electrodes, or hybrid electrodes, which contain both a depth and a strip (opercular) component, or epidural electrodes and intracerebral electrodes, or scalp electrodes and intracerebral electrodes may be used to maximize the yield of the intracranial EEG study (52–55). There are very few studies that have addressed the value of each intracranial method in specific epilepsy problems (56–60); each technique has advantages and drawbacks, and selection of one method must consider the clinical hypothesis to be addressed, the technical limitations of each method, and the risk of morbidity or mortality. In our opinion, the choice of intracranial study should be determined by the expertise developed in a center (familiarity with a given method) and, more importantly, by a strict adhesion to a clinical process primarily based on a careful evaluation and correlation of seizure semiology, scalp EEG patterns, and location of a ­lesion, when present.

396 TABLE 14.2

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography) Intracranial Electrode Characteristics

Type

Recording areas

Advantages

Disadvantages

Intracerebral depth electrodes (within cortex)

Deep limbic and paralimbic buried structures: amygdala, hippocampus, entorhinal cortex and parahippocampus, insula; interhemispheric cortical structures; depth of the sulci; hypothalamus, thalamus, and basal ganglia; deep-seated and periventricular lesions

Good sampling of deep structures; findings can be standardized in a common stereotatic space ­allowing intersubject comparisons; well-­tolerated procedure and low morbidity

Limited sampling (especially of ­neocortical structures); not adapted for ­exhaustive cortical functional mapping

Grids (subdural surface)

Cortical convexity, basal and interhemispheric neocortical (gyral) surface

Large number of recording channels and broad coverage of neocortical ­areas; well-suited for mapping of cortical function by electrical stimulation

Large craniotomy; higher ­morbidity; no sampling from deep buried ­structures; needs to be immediately followed by resective surgery

Strips (subdural surface)

Cortical convexity, basal and interhemispheric neocortical surface

Good coverage of neocortical areas; low morbidity

No sampling from deep buried structures

Epidural electrodes (pial surface)

Cortical convexity

Easy to install with low morbidity and satisfactory coverage of neocortical convexity

No sampling of basal and deep ­structures; no electrical stimulation

Intracerebral Electrodes: Characteristics, Insertion, Placement, and Location Characteristics Intracerebral depth electrodes are made as flexible, semi-rigid or rigid wires or catheters that penetrate the brain tissue. They can be homemade (Fig. 14.1A) or commercial (more likely now) (Fig. 14.1B) multicontact electrodes in different metals, including stainless steel, gold–chromium alloy, nickel–chromium composite, or platinum–iridium composite. Electrodes made of nickel–chromium composite are favored because they are nonmagnetic and compatible with magnetic resonance imaging (MRI). Silver and copper electrodes are not used because of their toxic effects (53,61). Each electrode contains several contacts placed at regular intervals (usually 2 to 5 mm, sometimes 10 mm, apart) along the wire. Each contact lies within the brain substance either in the gray or in white matter (surface varies according to the type of electrode: the homemade Montreal Neurological Institute [MNI] electrodes, for instance, have contacts of 0.8 and 0.85 mm2, while

commercial electrodes usually have contacts of 3 to 5 mm2). Electrodes contain between 5 and 18 contacts, 2 mm in length, 0.8 to 1.1 mm in diameter, a surface of 0.8 to 5 mm2, and 1.5 to 5 mm apart. Electrodes can be customized according to the needs of the neurosurgeon and neurophysiologist, and now may contain, in addition to normal-size clinical contacts (macrocontacts), microcontacts placed either at the tip of the electrode or interspaced between the macrocontacts (Ad-Tech Medical Instrument Corporation, Racine, Wisconsin) (Fig. 14.1C). Microcontacts ( R) tonic motor signs. Interictal scalp EEG showed spikes and bursts of fast oscillations over the right fronto-central region (FP2–F4, F4–C4, Fz–Cz), and seizures started from this area. The hypothesis was to demonstrate that seizures arose from the region of the right frontal eye-field. C: Posterior quadrant neocortical epilepsy (a 24-year-old with negative MRI, same patient as in Fig. 14.12). Seizures start without an aura, patient stares with loss of contact, mumbles and sometimes presents head shaking, with postictal amnesia. Interictal EEG showed right parietal or temporoparietal sharp waves and ictal scalp EEG suggested involvement of the right t­ emporo-occipital region. The hypothesis was centered over the posterior temporo-basal neocortex. ­Letters (cases A and B) and numbers (case C) refer to intracerebral ­electrodes. Most electrodes are inserted ­using a lateral ­orthogonal trajectory.

403

404

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

A

All channels (63) (TL: 50 uV/mm, ETL: 20 uV/mm)

B

Selected TL channels (37) (50 uV/mm)

C

Selected Hc channels (16) (50 uV/mm)

Orbito-frontal

Ant. cingulate

Post. cingulate

Temporal pole

Amygdala

Mid-temporal (T3)

Ant. Hippocampus

Post. hippocampus

Figure 14.5: Different displays of SEEG recordings in the same nonlesional temporal lobe patient. A shows a display of all electrodes (67 channels) recording from frontal, temporal, and posterior cingulate regions. B shows the same recording obtained from the temporal lobe, and C from the two hippocampal electrodes. Bipolar recording with seven or eight channels per electrode. Electrodes were inserted through orthogonal or oblique approaches; for each electrode, upper channels refer to mesial structures and lower channels to lateral structures.

reviewed with both bipolar and referential montages because of in-phase cancellation (bipolar montage) and the effects of active reference electrodes (referential montage). Intracererebral EEG activity can be recorded along with a limited number of EEG electrodes placed over the scalp. A reason against the use of ­additional scalp electrodes is the risk of infections. Also, the position of scalp electrodes cannot be standardized due to the presence of intracerebral electrodes. At the MNI, subdermal needle electrodes placed over the frontocentro-parietal parasagittal regions are now used routinely (109,110). They may provide information on the regions unexplored by the intracerebral electrodes, notably in the hemisphere contralateral to the side of investigation, and also help in sleep staging. This later point is of particular interest when looking at events such as HFOs, which preferentially occur during slow-wave sleep (111). These subdermal electrodes are well tolerated by ­patients, remain operational for several weeks (110,112), and are without

side effects. Additionally, the electrocardiography (to capture changes of heart rate during periictal periods), EMG, and EOG (again for artifacts detection and sleep staging) are usually monitored. The risk of SUDEP (sudden unexplained death in epilepsy), although low, makes it reasonable to also monitor breathing and oxygen saturation to look at possible hypoventilation and oxygen desaturation that may occur during and after seizures.

Practical Aspects The intracranial EEGs are reviewed with a high-resolution computer screen. Because the number of channels can easily exceed 100, and therefore beyond the limits of adequate visual resolution, bipolar and ­referential montages are created according to the clinical problem, with a variable number of channels. The different montages can then be r­ eviewed back-and-forth with a large or with a limited number of channels, at

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

­ ifferent sensitivities and timescales and in a logical framework (channels d are grouped according to lobar location and according to the position of the contacts from depth to surface). This approach allows first to get an overall impression of the EEG during an event, and then to focus on more L Amygdala

A 1 2 3 4 5 6 7 8 9 10

L Hippocampus

R Amygdala

R Hippocampus

R Orbito-frontal

R Frontopolar

R Pre-motorant.

Figure 14.6: Examples of seizure recorded with large and limited sampling. A: a 55-year-old patient with adult-onset bilateral temporal lobe epilepsy and right anterior frontal area of encephalomalacia. Patient was suspected to have independent temporal epileptic generators, right more active then left side. Intracerebral evaluation was to demonstrate that main seizure pattern arose from the right mesial temporal regions, and to rule out the contribution of the right frontal lesion. Each electrode contained between 10 and 15 contacts for a total of 100 channels. In this example, selecting only the left amygdala and hippocampal electrodes (inferior panel) shows a clear focal seizure onset in the left hippocampus (recording reference, left parietal). B: A 33-year-old patient with seizures since age 16 and a discrete right mid-frontal convexity focal cortical dysplasia type 2b. Two electrodes (RF2 ant. and post.) were placed along the axis of the abnormal sulcus and gyrus from the depth to the surface. In this example, selecting only these two electrodes (inferior panel) shows the seizure onset involving most of their contacts found in the brain.

R Pre-motorpost.

L Amygdala 1 2 3

4 5 6 7

8 9 10

L Hippocampus 1 2 3

4 5 6

7 8 9 10

405

specific areas of interest (Fig.  14.6A, B). The continuous intracerebral EEG is usually stored on a computer medium, which allows us to go back at any time to review the tracings with a great flexibility. Measures of impedance and routine EEG recordings can be performed at the beginning

406

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

L Ant. cingulate

B

L Mid. cingulate

R Orbito-frontal R Ant. cingulate

R Mid. cingulate R F2 (ant.lesion) R F2 (post.lesion) R Amygdala R Hippocampus

R F2 (ant.lesion) 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10 R F2 (post. lesion) 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10

Seizure onset

Figure 14.6: (continued)

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

and at the end of invasive evaluation to ensure the quality of the recording (see later). EEG recordings are performed over a mean period of approximately 2 weeks (in a recent survey—not published—obtained from six adult or pediatric centers, one Canadian, and five Europeans, the average number of days for evaluations varied from 2.5 to 15 days, with a range from 1 to 6 days in children and 3 to 32 days in adults), with a gradual reduction of patient’s medication, and under different states (waking, after sleep deprivation, and during sleep). The duration of monitoring is tailored according to the patient’s need (frequency of spontaneous seizures), risks of the procedure (e.g., infections) and of drug withdrawal, as well as to expected increase in technical problems over time (electrodes and leads malfunctions). Patients are usually monitored in a dedicated monitoring unit, where they are continuously observed by a member of the epilepsy team. This allows obtaining from seizure onset a precise description of subjective and objective patient experiences, and an accurate neurological evaluation of test awareness, language, memory, muscle tone and activity, and sensory–­motor functions. EEG monitoring usually begins within 24 hours of surgical insertion and after a period of observation in the intensive care unit (ICU) or neurosurgery department. In some centers, recordings are performed 24 hours per day and 7 days per week, while in others, monitoring is restricted to daytime and weekdays. Recently, some centers have designed dedicated intracerebral EEG ICU monitoring suites for their epileptic patients with intracerebral EEG.

Indications of Intracerebral Depth Electrode EEG Recording Invasive EEG recordings are considered when noninvasive studies are discordant or inconclusive. There are no set criteria that determine whether an invasive EEG evaluation should be performed, but factors such as the presence of a lesion or the type of epilepsy (temporal versus extratemporal) influence the decision for the need of invasive recordings. A simple and intuitive scheme to help define the indications, or contraindications, for invasive EEG monitoring in patients with refractory epilepsies (113) is to divide patients in categories determined on the basis that the epileptic generator can be localized and lateralized during adequate comprehensive noninvasive evaluation. Following this scheme, four categories are defined (Table 14.3): (1) localized and lateralized epileptic focus, for example, unilateral mesial temporal epilepsy; (2) localized but nonlateralized focus,

TABLE 14.3

407

Scheme for Intracranial EEG Decision in TL and Extra-TLE

1. Localized and lateralized epileptic generator: e.g., unilateral mesial TLE, unilateral and localized neocortical epilepsy 2. Localized but not lateralized epileptic generator: e.g., side of mesial TL focus not determined (R versus L), bilateral TLE or FLE 3. Lateralized but not localized epileptic generator: e.g., mesial versus neocortical TL focus, temporal “plus” epilepsy, “­pseudo-temporal” lobe epilepsy, frontal and central lobe epilepsy, ­posterior quadrant epileptic generator, insular lobe epilepsy 4. Not lateralized and not localized epileptic generator: e.g., widespread, diffuse or multifocal epileptic generator

for example, TLE without lateralization or bilateral frontal epilepsy; (3) lateralized but not localized focus, for example, right frontal versus right temporal focus; and (4) nonlateralized and nonlocalized focus, for example, multifocal, diffuse, or widespread epileptic generator. With this scheme in mind, invasive EEG would appear primarily indicated in categories 2 and 3. Whether intracerebral electrode, subdural strip or grid recordings are used depends on the specific questions and the brain areas that are suspected to be involved; each technique has advantages in specific situations, and selection must also consider their technical limitations. Also, in children, intracerebral EEG methods are more difficult to perform, for technical reasons (thickness and rigidity of the skull, tolerance to longterm monitoring) and for age-related semiological and neurophysiological peculiarities, which make localizing hypotheses more complex and difficult to elaborate than in adults. ­Nevertheless, intracerebral EEG recording is feasible, effective, and safe, even in children as young as 3 years of age, or less (92,96).

Indications for SEEG in Temporal Lobe Epilepsy Unilateral Mesial Temporal Lobe Seizures (Category 1) In TLE, intracranial EEG is usually not indicated when presurgical clinical, functional, EEG, and imaging features converge, favoring a unilateral temporal lobe focus, and are clear enough to propose a temporal lobe resection with an excellent postoperative long-term prognosis (114,115).

408

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

Nonlateralized or Bilateral Independent Temporal Lobe Seizures (Category 2), and Mesial versus Neocortical Temporal Lobe Seizures (Category 3) Long-term SEEG monitoring is indicated if a doubt persists on the side of the temporal lobe generator, when bilateral TLE is suspected (6), or, when, within the temporal lobe, clinical arguments do not allow to differentiate ­between a mesial and neocortical epileptic generator. Increasing evidence suggests that the distinction between mesial and lateral temporal

lobe seizures is often inaccurate, and that temporal lobe epileptogenic zones may incorporate larger networks that extend beyond these traditional limits (Fig.14.7) (116,117).

Temporal “Plus” Epilepsy (Category 3) This term refers to a specific form of multilobar epilepsy in which seizures involve a complex epileptogenic network that includes the temporal lobe and the closed neighbored structures such as the orbitofrontal cortex, the insula,

1200 uV (* 600 uV) 1s

TP

(0.530-120 Hz band-pass filter)

*

AN * Hc HcG

T3 aT2 mT2 aT1 pT1 FusG TOj AG pCG TPj

Seizure onset

Figure 14.7: Example of a widespread mesial and neocortical temporal seizure onset: a right temporal lobe seizure in a 16-year-old girl with negative MRI. Seizure semiology was characterized by a psychomotor arrest, sometimes inconsistent speech, discrete chewing, and postictal amnesia. During SEEG, seizures involve a large network from onset and start simultaneously in the mesiotemporal lobe structures (AN, amygdaloid nucleus; Hc, hippocampus; HcG, hippocampal gyrus), the temporopolar cortex (TP), basolateral temporal cortex (T3), and anterior part of the second temporal convolution (aT2). It spreads almost immediately to the mid part of T2 (mT2), fusiform gyrus (FusG), lateral temporo-occipital junction (TOj), angularis gyrus (AG), and posterior cingulate gyrus (pCG). The temporoparietal junction (TPj) is much less involved, and the first temporal convolution (aT1, pT1) is spared.

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

the frontal and parietal operculum, and the temporo-parieto-occipital junction. Barba and colleagues (118) compared the clinical and scalp EEG characteristics of patients proven to have TLE (n = 58) with a smaller group of patients with temporal “plus” epilepsy (n = 22, T+). They found that patients with T+ are difficult to differentiate from those with TLE on the basis of general clinical features or MRI data; even the presence of hippocampal sclerosis may not help in distinguishing the two groups. However, ictal clinical symptoms and scalp EEG findings help differentiating TLE from T+ patients. For instance, patients with TLE more frequently present an ability to warn at seizure onset, an abdominal aura, gestural automatisms, and postictal amnesia. On the other hand, patients suffering from T+ epilepsies more frequently have gustatory, vestibular, or auditory aura; they more frequently exhibit contraversive manifestations of the eyes or head, piloerection, and ipsilateral tonic motor signs; and they are more often dysphoric during the postictal phase. Interictal EEGs of T+ patients more frequently exhibit ­bilateral or precentral abnormalities, and ictal EEGs more frequently point over the anterior frontal, temporoparietal, and precentral regions. Hence, a constellation of clinical and scalp EEG findings may be useful for identifying among patients suffering from atypical TL epilepsies, particularly in the nonlesional cases, those who should undergo invasive recordings before surgery (Fig. 14.8). Interictal metabolic 18FDG-PET patterns might also be helpful to differentiate classical mesial TLE from T+ patients (119).

“Pseudo-Temporal” Seizures (Category 3) Invasive EEG is indicated in suspected “pseudo-temporal” epilepsy, a term that encompasses cases with mesial frontal and orbitofrontal seizures, mesial parietal and temporo-occipital seizures, or insular lobe seizures (Fig. 14.9). The epileptic generator localized in these regions can be clinically silent until the discharge propagates to adjacent temporal lobe structures, mimicking temporal lobe seizures both clinically and on scalp EEG (120). In the absence of a structural lesion, this group of focal epilepsies can be difficult to identify and attention must be paid to subtle ictal clinical symptoms that might point to extratemporal areas (121), to atypical EEG ­patterns (122), to posterior abnormal MRI data (123,124), or to unusual ictal SPECT ­patterns (125).

Indications for SEEG in Extratemporal Lobe Epilepsy Frontal Lobe Seizures (Categories 2 and 3) The large volume of the frontal lobe and its complex functional organization and connectivity explain the difficulty clinicians have to precisely define a clear clinical hypothesis in frontal lobe seizures, particularly in nonlesional cases. Frontal lobe seizures have a heterogeneous semiology and are classically compartmentalized in anterior-posterior, superior-inferior, and lateralmesial (126). This clinical classification of frontal lobe seizures has some Seizure onset

Rectus G OrbitoF BasoLat Fcx F pole CG DL preF F operc Insula

Figure 14.8: Left temporal “plus” seizure in a 50-year-old female patient with normal MRI. The discharge is widely extended from the onset, ­ involving not only the mesial and lateral temporal lobe structures (Tpole, temporal pole; HcG, hippocampal gyrus; T2ant, post, anterior/­ posterior part of the second temporal gyrus; T1ant, mid, anterior and mid part of the first temporal ­gyrus), but also the insula and orbitofrontal ­cortex ­(OrbitoF). During the seizure, the patient was motionless, and thereafter amnestic.

T1ant T1mid T1post T2ant T2post T3 Tpole* HcG FusG

Hc 400 uV (* 200 uV) 1s

409

(0.530-120 Hz band-pass filter)

410

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography) Seizure onset

F op C op** P op *

Insula

orbitoF T1 Tpole aHc *** aT2 pHc *** pT2 HcG T3 pCG**** (* 250 uV - ** 400 uV - *** 1200 uV - **** 2000 uV)

500 uV 1s

(0.530 -120 Hz band -pass filter)

Figure 14.9: Right pseudo-temporal seizure in a 41-year-old female patient with a right opercular scar (past history of cavernoma surgery). Seizures were triggered by eating and started with a gustatory hallucination. The SEEG discharge starts in the insular cortex and then involves, 2 seconds later, the temporal lobe, especially the hippocampus (aHc, pHc), giving rise to a “typical” temporal lobe semiology (loss of contact, oroalimentary automatisms, right hand automatisms, and left arm dystonic posturing and postictal confusion).

limitations because of the difficulty to ascribe a specific seizure pattern to a specific frontal region or functional area and the risk of overlap (127). This is explained by (1) the size of the frontal lobe, which represents approximately two-thirds of the brain; (2) the large extent of the mesial, interhemispheric, frontal structures where epileptic generators are inaccessible to scalp EEG; (3) the extended connections within the functional divisions of one frontal lobe and between the two frontal lobes, and the fast spread of ictal discharges within the frontal lobe structures; and (4) the extended connections with cortical and subcortical structures outside the frontal lobe. Typically, frontal lobe seizures fall in categories 2 or 3 of our scheme (Table 14.3), and

good synoptic electroclinical correlations should be able to provide enough information to define indications and types of intracerebral EEG evaluation. The frontal pole, the mesial premotor cortex, the dorsolateral frontal cortex, and the orbito-cingulate region are anatomical areas of predilection for frontal lobe seizures, with distinct features that can be recognized, and should, with their efferents, be adequately sampled. H ­ yperkinetic seizures of frontal origin, for instance (Fig. 14.10), often require investigating the gyrus rectus and the orbital cortex, the lateral fronto-basal cortex, the anterior cingulate gyrus, and the anterior portion of the ipsilateral temporal lobe (128). They may also arise from the mesial surface of the premotor frontal

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography) L orbito-frontal

411

1-2 2-3 3-4

L ant. insula 1-2

2-3 3-4 L ant. cingulate 1-2

2-3 3-4 L mid. cingulate 1-2

2-3 L hippocampus 1-2

4-5 6-7

1s

Figure 14.10: Orbitofrontal hyperkinetic seizure in a 31-year-old patient with left nonlesional frontotemporal epilepsy. Typical seizures were characterized by a sudden onset of intense fear followed by vocal automatisms and complex motor behavior. The seizures were short but ­followed by amnesia. The SEEG seizure starts in the orbitofrontal cortex (first arrow), and possibly at the same time in the anterior insula (rhythmic 2- to 3-Hz spike-and-wave discharge L orbitofrontal 2-3 and 3-4 and L ant insula 2-3 and 3-4). After few seconds, a low-voltage fast discharge is seen in the same electrode contacts (second arrow) and then in the anterior cingulate; it is at this ­ moment that the hyperkinetic manifestations start (third arrow, inferior panel).

cortex and are better evaluated by targeting the rostral and caudal part of the supplementary motor area, the pre-SMA, the cingulate motor area and the cingulate gyrus proper, and the primary motor cortex (129), or even from outside the frontal lobe in the temporal lobe structures or in the insula (130).

Central Cortex (Sensorimotor) Seizures (Category 1) The problem in this type of seizures is not so much the localization of the seizure generator as the sparing of function; the seizure semiology is

usually straightforward, with reliable localizing and lateralizing features. ­Therefore, intracerebral EEG evaluation of the central sensorimotor cortex should be designed to demonstrate that seizures arise from this area and to provide relevant information on what part of this eloquent cortex can be removed and what portion should be spared (Fig. 14.11). In the central region, intracerebral electrodes can sample the depth of the rolandic fissure and the d ­ escending motor and ascending sensory pathways (131). A  ­limitation, however, is that the rolandic fissure in the horizontal plane

412

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography) Am 1-2 2-3

midT (T1) 1-2 2-3 M1 sup. 1-2 2-3 3-4 6-7

Seizure onset

M1 mid. 1-2 2-3 3-4 4-5 S1 sup. 1-2 2-3 S1 inf. 1-2 2-3 Propagation

has an anteroposterior oblique orientation, and one single depth electrode cannot reach, by a lateral orthogonal trajectory, both the mesial and lateral aspects of the motor or sensory cortices. Also, the risk of significant morbidity is important in this region, and hemorrhages caused by electrode insertion may lead to ­important and irreversible motor and sensory deficits.

Posterior Quadrant Seizures (Categories 2 and 3) As for frontal lobe epilepsy, seizures arising from the posterior neocortex (posterior temporal junction and parietal and occipital lobe structures) have a heterogeneous semiology, an heterogeneity usually explained by the frequent simultaneous activations of the occipital, parietal, and posterior temporal lobe structures, and by the multidirectional spread of the epileptic discharge toward the central, frontal, and temporal regions. The clinical and

1s

Figure 14.11: Sensorimotor seizure in a 47-year-old patient with multifocal right frontocentro-parietal nonlesional epilepsy. One seizure pattern consisted of nocturnal seizures characterized by a sudden arousal, grimaces, left arm ictal paresis, and a variable head-eyesbody turn to the right with right arm stiffening. The seizure starts in the mid and superior aspect of the primary motor cortex (M1 sup. and mid.) and then involves, 2 seconds later, the primary sensory cortex (S1 sup. and inf.; an intermediate S1 electrode was dysfunctional).

EEG expressions of posterior seizures are often the result of propagation (132,133). In addition, the semiology of occipital or parietal lobe seizures is subjective, thus often not spontaneously or easily expressed by patients, capable of being masked by other symptoms (for instance, during occipitotemporal propagation) or associated with anterograde amnesia. Seizures originating in the posterior quadrant often fall again in categories 2 and 3 of our scheme (Table 14.3). Typical targets for depth electrode insertions are the calcarine and pericalcarine cortices (in occipital lobe epilepsy), the inferior and superior parietal lobules and the posterior cingulate cortex (in parietal lobe epilepsy), but also with emphasis on the evaluation of “junction territories” such as the lingual lobule and fusiform gyrus, the angular and supramarginal gyri, the precuneus and posterior insular cortex. In ­addition, because posterior seizures often propagate (and rapidly) anteriorly either

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

413

Seizure onset

mesioT

1 2 3 4 5 6 7 8 9 10 11

O-P mes

FusG

T2/T3/O lat

12 13 14 15 16 17* 18 19

T1/P lat

20 21 22 23 24

800 uV (* 400 uV ) 1s

(5.30-120 Hz band-pass filter)

Figure 14.12: Posterior quadrant seizure (same patient as in Fig. 14.4C). The seizure starts in the fusiform gyrus (FusG) and involves, after few seconds, the mesio-temporal lobe structures (mesioT), the mesial ­occipito-parietal cortex (O-P mes), and, to a lesser degree, the lateral temporo-occipital (T2/T3/O lat) and temporoparietal (T1/P lat) junction.

toward the temporal lobe through the inferior longitudinal fasciculus or ­toward the suprasylvian (fronto-central) structures through the superior longitudinal fasciculus, temporal and frontal lobe structures are often other targets for electrode insertions (Fig. 14.12). Finally, posterior seizures tend to spread to the contralateral homologous regions, and therefore, bilateral implantation may be necessary.

Insular Lobe Seizures (Category 3) The insula is known to have an important role in the propagation patterns of the epileptic discharge in a variety of focal seizures originating from different parts of the brain; this is explained by its localization and dense

connectivity with the surrounding frontal, temporal, perisylvian, and posterior cortical structures. Because the insula was considered not to be primarily responsible for seizure generation and because of the surgical risk due to its localization deep in the brain surrounded by a dense vascularization, neurosurgeons did not for some time consider this structure as a significant target in epilepsy surgery. Recently, however, lesional studies (134) and intracranial EEG recordings (75) have revived a new interest in the insula and pointed that the insular cortex can be a primary focus in some forms of focal seizures and plays an important role in the propagation of epileptic discharges arising from different neighboring structures. Moreover, neurosurgeons now feel that insular cortex surgery can be performed more safely

Moves his left hand

Clinical onset of the seizure

orbitoF DLFcx

SSOperc

4

Insula

3

7

1 2 3 4 5 6 7

1

6

TNeocx * **

mesioT

400 uV (4-5 : 800 uV; * 200 uV; ** 100 uV) 1s

4

5

3

(5.30 -120 Hz band -pass filter)

2

7

414

Figure 14.13: Insular lobe seizure (MRI-negative case, same patient as in Figs. 14.16B and 14.25). A typical seizure triggered by repetitive movements of the left hand, which elicit spikes over the posterior part of the insular cortex (contacts 4 and 5). These are followed by a polyspikes discharge, and by a low-voltage fast activity that spread a little bit more anteriorly (contact 3). The patient describes a painful tingling sensation in the left hand. Note the late involvement (arrows) of the dorsolateral frontal cortex (DLFCx) and suprasylvian operculum (SSOperc). orbitoF, orbitofrontal cortex; TNeocx, temporal neocortex; mesioT, mesiotemporal lobe structures.

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

415

and with satisfactory results, even when MRI does not show any epileptogenic lesion (135). Hence, in patients with suspected refractory insular lobe epilepsy, temporo-insular seizures or extratemporal neocortical-insular seizures, intracranial EEG recordings are recommended using intracerebral depth electrodes. Ictal semiology of insular lobe seizures is far from homogeneous and it appears that insular epilepsy can be a great mimicker (136): Although a well-defined “perisylvian” clinical pattern has been individualized, insular seizures may manifest with temporal (75) or frontal lobe semiology (129). Moreover, the insula is a buried structure localized deep in the brain and inaccessible to scalp electrodes and even to intracranial grid or strip electrodes. The SEEG method is better suited to study the insular cortex (Fig 14.13). Electrodes can be inserted using a lateral orthogonal trajectory through the frontoparietal and temporal operculum (75) or, to obtain a larger insular sampling, using oblique trajectories through the frontal or parietal cortices (78,137,138). Combined depth and subdural electrodes or hybrid operculoinsular electrodes can be also used safely to investigate complex insular/­ perisylvian ­epilepsies (54,56).

has nonlesional epilepsy (140): The rates among surgically treated patients range from 3% to 21% (141–143), and results of epilepsy surgery are also usually considered to be less satisfactory (144). Cases of nonlesional focal epilepsy are more challenging and intracerebral recordings remain essential in a significant proportion of them: (a) there is a higher risk for misinterpretation of clinical and scalp EEG findings because different regions of the brain may share the same clinical characteristics (and the presence of a lesion can be used to help in the differential diagnosis); and (b) there is a high percentage of multilobar epilepsies in patients with nonlesional (invisible or occult) focal epilepsy (145). The delineation of extratemporal epileptogenic network is much more difficult than that of temporal lobe seizure origin, and a larger proportion of patients suffering from MRI-negative extratemporal lobe epilepsy are likely to undergo invasive EEG recordings. A larger number of intracerebral electrodes might be necessary, and potentially with a higher risk for complications, with, however, a similar rate of epileptic focus localization compared to lesional cases (146).

Indications for SEEG in Lesional and Nonlesional Focal Epilepsies

There are epileptic conditions (category 4: multifocal, diffuse, or nonlocalizing hypotheses) where invasive EEG methods are unlikely to provide ­additional information and the risks of the procedure outweigh the benefits. In addition, invasive monitoring should be proposed only when it seems reasonable to believe that a resective procedure will lead to a satisfactory surgical outcome, which does not necessarily mean that complete seizure freedom will be achieved. In the Montreal series of patients studied with intracerebral EEG and also in a few other reports from the literature, between 7% and 31% of patients have their electrodes removed and, eventually, no resection (76,77,85,92–94). The reasons usually mentioned include a combination of multifocal, nonlocalizing seizures and the proximity of the eloquent cortex.

Invasive EEG studies are at first not indicated in lesional focal epilepsies. However, invasive EEG is to be considered when contradictions exist between the semiology/EEG presentation and the lesion, suggesting that the lesion is not epileptogenic per se, or solely. Intracerebral EEG in lesional epilepsy may be considered if the extent of the lesion itself is unknown and it is felt that the part seen on MRI represents only a portion of the epileptogenic anomaly (139). Intracerebral EEG may also be considered in multifocal epileptogenic lesions such as in tuberous sclerosis or cases of multiple cavernous angioma where it may be difficult to determine which tuber or cavernoma is responsible for generating seizures. Invasive intracerebral EEG is probably indicated in most of the epileptic patients with refractory ­focal seizures associated with periventricular nodular heterotopia because of their frequent association with “pseudo-temporal” seizures (79,81,124). ­Because of the limitations inerrant to depth electrode methods, intracerebral EEG recording in lesional epilepsy is not primarily considered to ­perform ­electrical functional mapping of the cortex. Despite major advances in neuroimaging, a significant proportion of patients with focal refractory epilepsy and presenting for presurgical evaluation

When Intracerebral EEG Is Not Indicated

Intracerebral EEG Monitoring: interpretation of EEG data Intracerebral electrodes have the advantage of being able to detect very small cortical generators located anywhere in the brain, including those in deep buried regions. They can also sample different lobes simultaneously in one or both hemispheres. As discussed earlier, however, depth electrode contacts sample a very small volume of cortex, which, in a variety of conditions,

416

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

may leave a large portion of the brain unexplored. This limitation has to be considered when studying cortical areas felt, for instance, to be responsible for the generation of seizures (the seizure-onset zone), spikes (the irritative zone), or slow-wave abnormalities (the lesional zone). The same intrinsic limitation allows only a limited sample of the regions responsible to generate normal physiological brain rhythms. Intracerebral depth electrode evaluations primarily aim to record seizures in order to define an epileptic focus and the propagation pattern of the seizures in order to delineate the extent of the cortical epileptogenic network. Interictal epileptiform abnormalities are recorded usually in a much larger amount and with a more widespread pattern than what is normally seen on scalp EEG. Nonepileptiform abnormalities are also present and often difficult to interpret. Each intracerebral investigation is designed for a specific epilepsy problem and, hence, the structures evaluated and the type of implantation may vary considerably. The intracerebral electrodes are still placed in a somewhat uniform approach but never with a standardize frame as for the 10–20 or 10–10 International Recording System used for scalp EEG; the insertion of electrodes is usually asymmetrical or electrodes are not placed in the exact same homologous regions. As already mentioned, brain sampling is limited by the number of electrodes and recording sites used. Finally, depth EEG activity may vary, depending on the parts of the brain that are recorded (different cortical regions have different physiological activities, different thresholds, and different connectivity or networks), and is modulated by the state of vigilance, cognitive and sensorimotor activities, as well as by the antiepileptic medication. For all these reasons, EEG interpretation should be made cautiously, and comparison and interpretation of the normal and abnormal EEG activities between regions and sides can be difficult. To some extent, each patient should be considered individually with his own neurophysiology with which we need to be well accustomed.

Normal Physiological EEG Activity Normal physiological EEG activity can be recorded, as for scalp EEG, with SEEG. At the MNI, a routine SEEG protocol is applied for each patient in order to define the baseline normal brain activity (Table 14.4). Subdermal electrodes are also placed, at least at positions Fz, Cz, Pz and F3, F4, C3, C4, P3, P4, with EOG and EMG electrodes to facilitate sleep staging (110). The routine SEEG recordings are performed twice during evaluation, at the beginning and at the end of investigation. The first study is usually performed 2 to 4 days after electrode insertion, at which time the patient’s

TABLE 14.4

1. 2. 3. 4. 5. 6.

7. 8. 9. 10.

Study Protocol during Routine EEG in Patients with Intracerebral Electrodes (Total Duration of 30 minutes)

Eyes opened 1 min, closed 1 min, opened 1 min, and closed 1 min. Saccades. Look to the right, to the left, upward, downward, 1 min. Clench left fist 5–10 s, right fist 5–10 s, few times. Jaw clenching 5–10 s, few times. Hyperventilation, 3 min, 3 additional minutes for relaxation. Intermittent photic stimulation at 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 30, 40, and 50 Hz. Attention, (a) serial 7s (100 − 7, 93 − 7, 86 − 7, etc.), 1 min; (b) months of the year forward and backward, 1 min. Reading, 1 min. Writing, 1 min. Various stages of sleep, if possible, in epochs of at least 1 min.

brain swelling would have receded and the patient is unlikely to have had multiple seizures. Routine recording should happen at least 12 hours after the last seizure, preferably 24 hours. For each study, the time of the last seizure and the medication are noted. During the first routine study, a verification of all electrodes and contacts is performed, and during the two routine evaluations, electrode impedances are measured (they usually increase from 1–5 to 5–10 kΩ, and sometimes >10 kΩ, between the first and the second studies). The ability to record some of the well-known EEG patterns depends on the position of the intracerebral electrodes; normal alpha, beta, and gamma rhythms and slow-wave activity in the theta and delta ranges are often recorded with these electrodes (Fig. 14.14A, B). The amplitude (usually higher) and morphology (potentials are usually sharper) of these patterns may appear different compared to scalp EEG recordings because of the closest relationship between the recording sites and the neuronal generators. Other normal activities seen during wakefulness and sleep are recorded such as mu rhythm, spindles, POSTS, or lambda waves (Fig. 14.14C, D). It is not possible to perform good sleep staging or to obtain good polysomnography tracings, but, as mentioned, satisfactory sleep staging can still be obtained with additional subdermal electrodes or with EOG and EMG electrodes. The EEG obtained from the white matter shows flat channels or propagated activity from neighborhood cortical generators (Fig. 14.14E).

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

Artifacts As for scalp EEG, artifacts recorded with intracerebral EEG are mostly instrumental (recording equipment, electrodes, and their connections), environmental (external sources, i.e., mains line, other equipment around or connected to patient, electromagnetic, radiofrequency, and electrostatic signals), or biologic artifacts (arising from the patient) (Fig. 14.15). Instrumental artifacts are often due to a loss of signal caused by the amplifier or a dysfunction in the jack-box. Artifacts due to electrodes or their connections can be explained by a breakage of the electrode (often at the point where the electrode is locked onto a bone peg to prevent movement after insertion),

by a loosening of the electrode, by a CSF leak causing a short-circuit, and by the movements of the electrode wires. Environmental artifacts can be frequent, but in the context of invasive EEG, the 60-Hz activity is the most frequent and explained by numerous possible sources, including all electrical equipment in the environment (cellular phones, computers and lap-top, TVs and radios, wall phones, beds, intravenous pumps, etc.). Intracerebral EEG ­recordings are more likely to be free of biologic artifacts. Pulsation artifact, as well as the artifacts of muscle and body movements and eye movements can, however, be encountered. EMG artifacts are commonly seen in the most ­superficial contacts of the electrodes and can be easily confused with fast oscillations. Intracerebral recordings in the temporal pole of implanted

A Figure 14.14: Normal physiological intracerebral EEG patterns. A: Alpha rhythm recorded from depth (RPs 7–13 macrocontacts and RPs17–18 microcontacts) and from surface (subdermal electrodes Fz, Cz, and Pz) electrodes. Left parietal reference. B: Movement-related (patient was asked to close right fist) beta synchronization in the motor cortex (same patient as Figs. 14.14E and 14.18A). Note the prominent beta synchronization over the contacts exploring the anterior bank of the rolandic cortex (contacts 7 to 11, circle). C: Neocortical spindles recorded (bipolar recording) from a parietal electrode and from subdermal electrodes. D: Hippocampal spindles. Each electrode (amygdala, hippocampus, and paraHc, parahippocampus) contains eight channels labeled from 1 to 8, the deepest located in the mesial structures. E: White matter activity. Recording obtained along the axis of a 15-leads depth electrode (L’) exploring the mesial and lateral aspect of the left prefrontal cortex (same patient as Figs. 14.14B and 14.18A). Monopolar (top right) and bipolar (bottom right) montages. Note the low amplitude of the SEEG signal recorded within the white matter (contacts 5 to 9). Spikes are recorded from the lateral surface (contacts 12 and 13).

C

RPs1-ref 15-ref 2-ref 16-ref 3-ref 4-ref 5-ref 6-ref 7-ref 8-ref 9-ref 10-ref 17-ref 11-ref 12-ref 18-ref 13-ref Fz-ref Cz-ref Pz-ref

B

closes right fist

I’

L

600 uV 1s

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

417

D

1s

Amygdala

1

2 3 4 5 6 7 8 Hc

paraHc

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

418

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography) E

5 6 7 8 9 11 13 1234

10 12 14

L’

R

800 uV 1s

(0.530-120 Hz band-pass filter)

1-ref 2-ref 3-ref 4-ref 5-ref 6-ref 7-ref 8-ref 9-ref 10-ref 11-ref 12-ref 13-ref 14-ref 15-ref 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10 10-11 11-12 12-13 13-14 14-15

patients can be contaminated by saccadic eye muscle artifacts, resulting in typical high gamma-band power increases (147): by analyzing data from multiple depth electrodes, Jerbi and colleagues showed that this artifact is confined to the pole of the temporal lobe and explained by its immediate vicinity to extraocular muscles (rectus lateralis).

Abnormal Interictal EEG Activity Interictal nonepileptiform abnormalities include focal, regional or diffuse slow-wave activity, focal loss of fast activities, focal loss of sleep patterns, and attenuation of background activity (Fig. 14.16). In SEEG terminology, the Lesional Zone refers to the brain area occupied by an abnormal slowwave activity, or by a major disturbance of background activity that implies a macroscopic alteration of the neural tissue (148). This area coincides in many instances with the Epileptogenic Lesion of Rosenow and Lüders (149), which is most often revealed on MR images. The location and distribution of the slow-wave activity, however, do not always closely match with a lesion

Figure 14.14: (continued)

as assessed on pathological specimens; the amount and amplitude of this electrical anomaly are far from homogeneous, and the relationship between the areas generating slow waves and those where seizures originate may vary from one case to another. The experience of Talairach and Bancaud in this field was summarized in an important book, EEG et SEEG dans les tumeurs cérébrales et l’épilepsie (5), in which they described the topographic transitions from electrical silence found inside glial tumors to slow-wave patterns, spiking, subclinical paroxysms, and seizure activity. The significance of the slow-wave activity as it is recorded from intracerebral EEG remains unclear, but it probably needs to be carefully weighted when coming to the final surgical decision, even in the absence of a corresponding MRI anatomical lesion. Interictal epileptiform abnormalities consist of spikes, polyspikes, sharp waves, polysharp discharges, and spike-and-wave or polyspike-and-wave complexes. Their extent delineates the Irritative Zone (148–150). They may be very focal and detected over a very short distance by very few contiguous electrode contacts, or they may have a more regional, lobar or m ­ ultilobar and

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

L TNeocx T1

A

B

C

D

1 2 3 4 5

R TNeocx T1 1 2 3 4 5

Figure 14.15: Artifacts. A: Muscle artifact. Fast activity due to muscle artifacts recorded from epidural electrodes placed over the first temporal convolution (left and right TNeocx T1). This activity was not picked up in the other channels that were recording from contacts located in the brain. B: Artifacts caused by “loose” electrode contacts or wires (arrows). C: Artifact due to the reference, which occurred when the examiner was touching the patient, and which is visible only on monopolar montage (upper channels). D: 60-Hz artifact (when the patient uses his ­i-phone), attenuated after applying a notch filter (right panel).

419

420

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

A

Insula

B

500 uV 1s

Eyes opening

Figure 14.16: Examples of SEEG interictal patterns of the lesional zone. A: Electrical silence within a dysembryoplastic neuroepithelial tumor; leads located within the lesion are silent (L’1, L’2, N’1, N’2, K’1, K’2). B: Slow waves within the insular cortex persist during eye closure (same patient as in Figs. 14.13 and 14.25), and spikes are temporarily abolished. C: Extended and continuous slow waves over the left temporal lobe in a patient with bilateral temporal lobe epilepsy after viral encephalitis.

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

421

Right temporal lobe

Left temporal lobe

C

Figure 14.16: (continued)

widespread or even diffuse distribution. There are often multifocal, and they can be generalized and synchronous in one or both hemispheres. Compared to scalp interictal epileptiform potentials, spikes recorded from intracerebral electrodes are less stable, showing variable frequency and ­localization. They are generally more abundant during slow-wave sleep. The morphology of the intracerebral epileptiform potentials, although similar to those recorded from scalp, often shows a higher amplitude, a higher f­ requency, and a more complex and sharper configuration. For instance, very large a­mplitude

10s

and fast spike-and-slow wave or polyphasic potentials with variable polarity (they may be negative or positive alternatively during the same recording epoch) are often recorded from the hippocampus. However, because of the localization and organization of the hippocampus, epileptic activity is usually difficult to detect. Interictal discharges are also frequent in the amygdala, entorhinal cortex, posterior hippocampus or parahippocampus, structures in which the epileptic discharges are synchronous or independent (Fig.  14.17A–D). The mesial temporal structures are located deep in the

422

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

A L Am

LH

B 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 1-2 2-3 3-4 4-5 5-6 6-7 7-8

L Am

C 1-3

L Am

2-3 3-5

3-4

5-6

4-5

5-6

6-7

6-7 7-8

7-8

8-9

LH

1-3

LH

RH

R Hp

8-9 1-2 2-3

3-5

3-4

L Hp 1-2

R Am

1-2

4-5

5-6

2-3 3-4 4-5 5-6 6-7 7-8 8-9 1-2 2-3 3-4 4-5 5-6 6-7 7-8 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9

5-6

6-7

6-7 7-8

7-8

8-9

R Am

1-3

L Hp

8-9 1-2 2-3

3-5

3-4

4-5

5-6

5-6

6-7

6-7

7-8

7-8

8-9

RH

1-3

R Am

8-9 1-2 2-3

3-5

5-6 6-7

3-4 4-5 5-6

6-7 7-8 8-9

7-8 8-9

1s Figure 14.17: Examples of SEEG interictal patterns of the irritative zone. A: Mesiotemporal spikes with positive polarity: one limited at LH1–2 (left hippocampus), no volume conduction, and not seen in the surface contacts; and another recorded simultaneously from right amygdala (RAm) and hippocampus (RH, RHp), with some propagation outside the mesial temporal lobe. B: Right hippocampal polyspikes and spikes with negative polarity. C: Left amygdala sharp discharges. D: Low-amplitude spikes recorded from the left entorhinal cortex (Et’, arrows), and not seen in the amygdala (A’), anterior hippocampus (B’), posterior hippocampus (C’), and parahippocampus (E’). E: Interictal train of spike-and-waves activity in the anterior temporal neocortex (LAm 4–5 to 10–11 and LH 6–7 to 10–11). F: Isolated interictal spike-and-wave in the mid-post temporal ­neocortex (LP 7–8 to 10–11).

D

A’ Et’ 1s

B’ C’ E’

E L Am

LH

F 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10 10-11

RH

1-2

2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10 10-11

Figure 14.17: (continued)

5-6

6-7 7-8 8-9

9-10

LH

1-3 3-5

5-6 6-7 7-8

8-9 9-10

LP

1-3 3-5

1-2

2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10 10-11

1-3 3-5

1-2

2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10 10-11

R Am

L Am

5-6 6-7 7-8

8-9 9-10 10-11

RH

1-3

3-5 5-6 6-7

7-8 8-9 9-10

10-11

1s

423

424

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

brain and realize closed electrical fields with negligible volume conduction. Hence, and in spite of the high frequency of the discharges, the epileptic activity generated is usually not detected by cortical or scalp EEG electrodes (151). In the amygdala, the interictal epileptiform anomalies are often sharp waves with a triangular shape and of lower amplitude, and more difficult to detect than the hippocampal spikes (Fig. 14.17C). This is explained by the intrinsic organization of the amygdala, where neurons are not aligned (as, for instance, the pyramidal cells of the hippocampus) and less likely to generate high-amplitude synchronized postsynaptic potentials (70). In TLE, the mesial temporal potentials (in the hippocampal formation and amygdala) are much more frequent than in the neocortex; they are often associated with polymorphic slow wave anomalies or abnormal background and with a focal reduction or loss of fast beta activity, indicating local neuronal destructive process or dysfunction. They are also often found independently in both mesial temporal structures, and are frequently present even in extratemporal lobe epilepsy. Neocortical temporal discharges are much less abundant even in cases of neocortical TLE, but they represent the electrical correlate of the spikes or sharp discharges seen in the scalp EEG (Fig. 14.17E, F).

Interictal epileptiform activity in extratemporal neocortical epilepsy also consists of potentials of variable morphology and distribution. It is usually present in frontal, central, parietal, occipital, or insular epilepsies and is likely indicative of neocortical epileptogenesis. Again, however, interictal epileptiform discharges often show a distribution and frequency (the Irritative Zone) with a higher variability than in the scalp EEG, and they do not necessarily correlate with the epileptogenic lesion or seizure-onset zone (148,150,152). From a surgical perspective, it would probably be a mistake to ignore the extent of interictal spikes, whether they are focal or not; they can provide additional confirmation of localization, and often may be a marker of the extent of the epileptogenic tissue. This seems to be the case, for instance, in focal cortical dysplasia type II, where a peculiar SEEG pattern (almost total absence of background activity and characteristic repetitive, high-amplitude fast spikes followed by high-amplitude slow waves interspersed with relatively flat periods) is often observed (139) (Fig. 14.18). Similar patterns have been described with subdural and epidural (and sometimes also from the scalp) EEG recordings (153,154). Also, particular attention must be paid to those spikes that do not disappear at

A 400 uV (*250 uv- **1200 uV)

1s (0.530-120 Hz band-pass filter)

* *

Figure 14.18: Examples of interictal patterns in type 2 focal cortical dysplasia (FCD2). A: Continuous or near-continuous interictal spikes (*) well-localized in a rolandic FCD2 (circle). B:  ­Typical subcontinuous paroxystic SEEG pattern of a FCD2 located in the frontal eye-field (*M 5–6; same patient as in Figs. 14.4B, 14.22A, and 14.23).

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

425

B Y1-2 Y8-9 R2-3 R6-7

M

M

*

M 5-6

Figure 14.18: (continued)

G1-2 H1-2 E2-3 F1-2 W1-2 W7-8 Z1-2 Z8-9 X1-2 X8-9 M1-2 M5-6 G14-15 H13-14 E7-8 F7-8 E13-14 F12-13 M11-12 B1-2 J1-2 B13-14 J14-15 S7-8

the seizure onset; this suggests that these spikes do not depend on a common pathophysiological process related to the area(s) from where seizures arise and, therefore, that they might be part of another Potential SeizureOnset Zone (155). Recently, interictal HFOs ranging from 80 to 500 Hz have been identified as a putative marker of the epileptogenic brain tissue (156). HFOs can be recorded in epileptic humans using macroelectrode contacts (Fig. 14.19), preferentially during slow-wave sleep, during which the rate tends to increase. They may occur at the time or independently of interictal spikes, increase during medication reduction but not after the seizures, and can have the same spectral characteristics and spatial extent as fast activities at seizure onset. They can be recorded in mesiotemporal lobe or neocortical epilepsies and do not seem specific to a particular type of epileptogenic lesion. They might therefore represent epileptogenicity per se, in line with the finding that surgical removal of HFOs generating tissue seems to correlate with a good postoperative outcome.

Seizures Onset (Seizure-Onset Zone) and Propagation The Seizure-Onset Zone (SOZ) is a concept that refers to the “area of cortex that initiates clinical seizures” (149): it can be defined as the area that

contains the “minimal” amount of cortex necessary to generate the first clinical symptoms, or the area in which arises the first EEG change prior to or concomitant with the clinical onset of the seizure. The first definition is electroclinical and overlaps with the Symptomatogenic Zone, that is, the area of cortex that when activated produces the initial ictal symptoms or signs (149). The second definition is electrophysiological and encompasses subclinical (also described as asymptomatic or pure EEG seizures) and clinical seizures. Subclinical seizures are ictal discharges without any clinical accompaniment; this definition is subjective and highly dependent on the mode of observation, that is, the presence and type of questioning (Fig. 14.20). Also, the occurrence of symptoms varies according to the localization of the discharge and its electrographic pattern; this is notably true in TLE, in which an ictal discharge can happen and remains confined to the amygdala or to the hippocampus without subjective or objective clinical manifestations (72). Sperling and O’Connor (157) suggested that at least in TLE, those subclinical or asymptomatic seizures recorded with intracerebral electrodes and originating from the same structures and with the same morphology as the habitual electroclinical seizures have a high localizing and prognostic significance. They concluded that their presence indicates a good localization of the SOZ and, also, functional inhibitory mechanisms.

426 A

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography) B RPs 1-2 2-3 3-4

4-5 5-6

6-7 7-8 8-9

RLa 1-2 2-3 3-4

4-5 5-6

6-7 7-8 8-9

RLp 1-2 2-3 3-4

4-5 5-6

6-7 7-8

8-9

1s High-pass 80Hz

High-pass 250Hz

0.1s

0.1s

Figure 14.19: HFOs. A: interictal HFOs: ripples outside spikes (RH 1–2, 2–3). B: ripple and fast ripple bands at the onset of a typical seizure (focal cortical dysplasia in the right precuneus, RLa and RLp).

The adequacy of the SEEG assessment of the SOZ depends on an a­ dequate spatial sampling of the cortical structures responsible for the s­ eizure origin and propagation. With this in mind, the first clear ictal electrical changes recorded are likely to be highly localizing, provided that these changes occur prior to the first unequivocal clinical manifestations and that they consist of

a fast synchronizing discharge. The coincidence of these two criteria indicates that the positioning of at least one or some electrodes is correct. Conversely, the absence of one of these two criteria suggests that the clinical assumption is possibly wrong and that the electrodes are not adequately sampling the real SOZ but only the propagated discharge. Another criterion to consider is the

Intracerebral Depth Electrode Electroencephalography (Stereoencephalography)

427

Am

Hc

Figure 14.20: A right mesial temporal SEEG subclinical seizure starts in the hippocampus with periodic high-amplitude spikes (Hc, 1st arrow) and involves, after few seconds, the amygdala (Am, 2d arrow) without further propagation.

1s

extension of the EEG changes at the onset of a seizure; focal changes limited to few contiguous recording contacts are more likely to indicate the exact SOZ than a nonfocal regional or even more diffuse onset (158,159). Intracerebral ictal onsets have classically been described as a lowvoltage fast activity, in the beta and gamma bands (146,159,160), or as recruiting and periodic fast discharge of spikes (161–163). The different seizure-onset morphologies (Table 14.5) likely reflect a variety of ­interictal  to ictal EEG transition mechanisms in different structures (e.g., mesial temporal lobe structures versus neocortex, temporal lobe versus extratemporal lobe) and in different pathological conditions (e.g., normal tissue versus mesial temporal sclerosis versus developmental cortical lesions such as focal cortical dysplasia versus atrophic/gliotic lesions). Hence, the two main factors that determined the ictal intracerebral EEG onset (and probably also the propagation patterns) are the affected lobe or brain structure and the underlying structural abnormality. Other variables that can modulate these patterns are the type and position of the recording electrodes and contacts (already discussed in sections on advantages and drawbacks and on risks and complications

of intracerebral depth electrodes), the host (children versus adults), and probably the type, amount, and withdrawal pattern of antiepileptic medication. Examples of ­ictal intracerebral EEG recordings with typical onset, propagation, and termination patterns for temporal and extratemporal seizures are found in Figs. 14.21 and 14.22. TABLE 14.5

Intracerebral Depth Electrode Ictal-Onset Patterns in Temporal and Extratemporal Lobe Epilepsy

1. Low-voltage fast activity in the beta (13–30 Hz) and gamma (>30 Hz) frequency ranges. 2. Periodic and rhythmic high-amplitude spiking activity (1–5 Hz), followed by low-voltage, fast activity pattern. 3. Spike-and-wave activity (>2 Hz). 4. Rhythmic polyspike activity (>13 Hz). 5. Rhythmic sharp activity (25% Increase in delta frequency 50%

8–12

Severe suppression of all frequencies

58 seconds duration, 88% sensitivity), intermediate

for brief seizures (40%), and poor for slowly evolving seizures (20%) (58) using aEEG. Frequency domain analysis refers to analysis of the contribution of different frequencies to the EEG signal. Filtering of the EEG signal is a simple example. Calculating the Fourier spectrum of the raw EEG is another example: the EEG for a given epoch is represented as the weighted sum of sine waves of different frequencies. For each frequency, there is an amplitude (the magnitude of the contribution made by the sine wave at that frequency) and a phase (the starting point of the sine wave of a given frequency). The plot of the amplitude versus frequency is the Fourier spectrum. Power refers

566

Continuous EEG Monitoring in the Intensive Care Unit

to the area under the Fourier spectrum amplitude curve within a given ­frequency range. For example, the alpha power would be the area under the curve from 8 to 13 Hz. This may be expressed as absolute power or relative power (e.g., normalized by the total power in all frequency bands). Power ratios are the ratio of the power in two different frequency bands. For example, the alpha to delta ratio (ADR) is the ratio of the power in the alpha frequency band to that in the delta frequency band. ADR has been used to monitor for ischemia in both intraoperative monitoring and in the ICU (59). The spectral edge frequency (SEF), defined as the frequency

below which a certain percentage of the total power resides, typically 90%, can potentially be used to monitor for both level of sedation (60) and cerebral ischemia (61). Time-frequency analysis refers to how the contribution of different frequencies to the EEG signal varies over time. Because the EEG is a nonstationary signal (one whose frequency composition is not constant over time), its Fourier spectrum also varies with time. Time-frequency tools are used to describe this variation. CSA and density spectral array (DSA) are tools for displaying the power spectrum over time. For CSA,

A

Figure 17.15: A: Lateralized periodic discharges (LPDs, a.k.a. PLEDs) with superimposed fast activity (LPDs  F) in the right hemisphere, central maximum (boxes). B: An example of intermittent evolution into semirhythmic faster activity (boxes). This shows clear evolution but is not long enough to be considered an electrographic seizure. These are sometimes referred to as BIRDs (brief ictal rhythmic discharges) or similar names. C: Longer event consistent with an electrographic seizure. LFF = 1 Hz, HFF = 70 Hz, notch off.

Continuous EEG Monitoring in the Intensive Care Unit

567

B

Figure 17.15: (continued)

the power spectrum for each epoch is displayed as a line graph; CSAs are rarely used now. Instead, DSAs, where the amplitude of the Fourier spectrum is encoded in different colors and displayed over time, are used. Measures of rhythmicity have also been utilized. For example, “rhythmic run detection and display” is a proprietary QEEG tool based on wavelet analysis (Persyst, Inc., Prescott, Arizona). It highlights the frequency components of the DSA that are the most regular in repetition pattern (i.e., rhythmic or periodic) at a given time. It can help to highlight the rhythmic activity associated with seizures. A diagonal appearance in this

tracing signifies gradual change in the frequency of the rhythmic pattern, fairly specific for an evolving seizure. Examples of QEEG are shown in Figs. 17.19 to 17.23. Some quantitative EEG software packages include automated detectors and alarms. However, most of these tools were developed for seizure detection in patients with relatively normal background EEGs and rapidly evolving seizures seen in epilepsy monitoring units or ambulatory EEG recordings. Their utility for CEEG recordings remains largely unestablished. In one small study (24 patients) comparing a novel seizure detection

568

Continuous EEG Monitoring in the Intensive Care Unit

C

Figure 17.15: (continued)

algorithm with two commercially available, automated seizure detectors, the commercially available software showed limitations (62). The Reveal algorithm (Persyst) had a sensitivity of 12.9% with 1.036 false detections per hour; the IdentEvent algorithm (Optima Neuroscience) had a sensitivity of 10.1% with 0.013 false detections per hour. By comparison, the novel algorithm developed by the authors had a sensitivity of 90.4% with 0.066 false detections per hour. This algorithm and others being developed (63) may be promising tools for CEEG monitoring in the future if their sensitivity and specificity can be confirmed in novel and larger data sets.

With regard to comparison of different QEEG techniques, there are few studies to date. In a retrospective study of 27 CEEG recordings from pediatric ICU patients, DSA and an EEG showed median sensitivity for seizure identification of 83.3% and 81.5%, respectively, with a very low false positive rate (64). However, sensitivity was very different from patient to patient, ranging from 0% to 100%, suggesting that some combination of seizure electrographic characteristics (many missed seizures were of low amplitude), background EEG, and other factors determine the actual sensitivity. More complex algorithms, often based on combinations of

Continuous EEG Monitoring in the Intensive Care Unit

quantitative EEG parameters, are being developed and tested for their ability to correctly analyze and not only detect seizures but also classify the EEG into different categories, such as normal, low voltage, burst suppression, slowing, periodic discharges, and seizures (65). Such algorithms have shown promising results in initial studies, but await larger tests and validation. The potential utility of QEEG tools for detecting delayed cerebral ­ischemia in patients with subarachnoid hemorrhage has been demonstrated (see later) (59).

A

Figure 17.16: A: Bilateral rhythmic delta activity (GRDA in ACNS nomenclature), anterior predominance in a 12-year-old patient who had resection of fourth ventricular mass with persistent change in mental status following surgery. The pattern did not show any evolution and there were no electrographic seizures. Fp2-F4 and F4-C4 have been suppressed owing to missing F4, which was not placed due to the presence of a drain. B: The same page is shown in average reference. This permits Fp2 to be ­interpreted. LFF = 1 Hz, HFF = 70 Hz, notch off.

569

CLINICAL APPLICATIONS There are several common indications for CEEG in the ICU (Table 17.5) (66,67). These include detection of subclinical seizures in patients with fluctuating mental status or unexplained altered mental status. A simple summary of the highest-risk patients is those with acute brain injury or clinical seizure with persistently impaired or fluctuating consciousness. CEEG can also be useful for evaluating spells (i.e., posturing, tremors, twitching, nystagmus, paroxysmal changes in vital signs such as tachycardia, or rises in

570

Continuous EEG Monitoring in the Intensive Care Unit

B

Figure 17.16: (continued)

intracranial pressure) to rule out possible seizures as the underlying cause. CEEG is also necessary for the management of anesthetic coma, especially when being used to treat seizures or elevated intracranial pressures. Lastly, CEEG can be used to detect ischemia or other deteriorations of cerebral functions in patients with limited neurological exams. Certain diagnoses are more commonly associated with seizures. In a series of 570 consecutive patients undergoing CEEG for the detection of subclinical seizures or to investigate the cause of unexplained decrease in level of consciousness, the incidence of seizures was highest in patients with a history of epilepsy (33%) and CNS infection (29%) (68). Most other diagnoses

had an incidence of approximately 20%, including brain tumors, recent neurosurgery, hypoxic-ischemic encephalopathy, subarachnoid hemorrhage, traumatic brain injury, toxic-metabolic encephalopathy, and unexplained decrease in consciousness. Intracerebral hemorrhages (ICHs) and ischemic strokes had a lower incidence of seizures on CEEG, 13% and 11%, respectively. Other factors that have been associated with a higher risk of NCSz include the following: hippus, remote history brain injury (stroke, traumatic brain injury), epileptiform discharges on EEG, burst-suppression pattern on EEG, nystagmus, unexplained eye deviation, coma, younger age, and clinical seizure prior to initiation of EEG.

Continuous EEG Monitoring in the Intensive Care Unit

571

Figure 17.17: Impact of stimulation on generalized rhythmic delta activity (GRDA, reactive). A 20-year-old patient with brief cardiac arrest. EEG showed rhythmic delta activity with a frontal predominance. With stimulation (arrow), this activity subsides as shown on the right side of this page, and is thus a reactive EEG. LFF = 1 Hz, HFF = 70 Hz, notch off.

In neonates, factors that have been associated with increased risk for prolonged subtle or nonconvulsive seizures on EEG include Apgar less than or equal to 5 at 5 minutes, umbilical cord or blood pH less than 7.2 during first hour of life, CNS infection, symptomatic sepsis, congenital cerebral abnormalities, grade III or IV intraventricular hemorrhage, encephalopathy, clinical seizures or suspected seizures, apnea of unknown etiology, and use of paralytic agents (summarized from [34]). Of note, in neonates, the use of EEG is particularly important as the majority of seizures recorded on EEG have no clinical correlate, possibly as high as 80% to 90% (69,70), and even clinical seizures can be difficult to distinguish from non-seizures, even by trained observers (71).

The duration of CEEG recording is highly variable and depends on the clinical circumstances and the initial findings on CEEG. For seizure detection, routine EEGs, even when repeated, clearly miss a substantial portion of seizures in critically ill patients (72), and CEEG significantly increases the detection rate of nonconvulsive seizures and nonconvulsive status epilepticus (73). In a series of 110 patients who had seizures during CEEG (68), factors that predicted time to first seizure included whether the patient was comatose or not and whether or not PEDs were present. Slightly greater than 50% of seizures were detected in the first hour of CEEG (Fig.  17.24). For non-comatose patients, 95% had their first

572 TABLE 17.3 Main Term 1

Main Term 2

Modifiers

Continuous EEG Monitoring in the Intensive Care Unit ACNS Guidelines for Description of Periodic Patterns 1. Generalized (G)—bilateral, bisynchronous, and symmetric. Specify predominance (anterior, posterior, midline, or generalized) 2. Lateralized (L)—includes unilateral, bilateral synchronous but asymmetric, focal, regional, and hemispheric patterns. Specify unilateral vs. bilateral asymmetric. Specify lobes 3. Bilateral independent (BI)—two independent [asynchronous] lateralized patterns, one in each hemisphere. Specify symmetric vs. asymmetric 4. Multifocal (Mf)—three or more independent [asynchronous] lateralized patterns with at least one in each hemisphere

1. Periodic discharges (PD)—periodic refers to a repeating waveform with a quantifiable interdischarge interval and recurrence at nearly regular intervals. Discharge is a waveform with no more than three phases or a waveform lasting 0.5 s regardless of number of phases 2. Rhythmic delta activity (RDA)—repetition of a waveform with relatively uniform morphology and duration, without an interval between consecutive waveforms. For RDA, frequency 4 Hz. 3. Spike-and-wave or sharp-and-wave (SW)—polyspike, spike, or sharp wave with a slow wave occurring in regularly repeating pattern with no interval in between. If there is an interval, it would be classified as a PD.

1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Prevalence Duration Frequency—specify typical and range Number of phases Sharpness Amplitude—specify in standard longitudinal bipolar montage using 10–20 recording Polarity Stimulus-induced Evolving or fluctuating Plus—refers to features that suggest that a periodic pattern may be more ictal. “F” applies to both PDs and RDAs and indicates superimposed fast activity. “R” applies to PDs only and indicates superimposed rhythmic delta activity. “S” applies to RDA only and indicates superimposed sharp waves or spikes or sharp-contoured activity

Adapted from Hirsch LJ, et al. American Clinical Neurophysiology Society’s standardized critical care EEG terminology: 2012 version. J Clin Neurophysiol 2013;30(1):1–27.

TABLE 17.4

Criteria for Nonconvulsive Seizures

Primary Criteria

1. Repetitive generalized or focal epileptiform discharges at a rate of 3 Hz

2. Repetitive generalized or focal epileptiform discharges at a rate of 3 Hz and the secondary criterion

3. Sequential rhythmic, periodic, or quasiperiodic waves at 1 Hz and unequivocal evolution in frequency (gradually increasing or decreasing by at least 1 Hz), morphology, or spatial extent. Excludes evolution in amplitude alone or change in sharpness alone Secondary Criterion

1. Significant improvement in clinical state or appearance of previously absent normal EEG patterns in response to acute administration of a rapid-acting AED, such as a benzodiazepine Definition of Evolution

1. Frequency: at least two consecutive changes in the same direction by at least 0.5/s, e.g., from 2 to 2.5 to 3/s, or from 3 to 2 to 1.5/s

2. Morphology: at least two consecutive changes to a novel morphology 3. Location: sequential spreading into or sequentially out of at least two different standard 10–20 electrode locations In order to qualify as present, a single frequency or location must persist for at least three cycles. The criteria for evolution must be reached without the pattern remaining unchanged in frequency, morphology, or location for 5 min It is important to note that these are criteria for definite NCSzs. If a pattern does not fulfill these criteria, it does not mean that NCSzs have been ruled out; it only means that NCSzs cannot be ruled in definitively. Based on Chong DJ, Hirsch LJ. Which EEG patterns warrant treatment in the critically ill? Reviewing the evidence for treatment of periodic epileptiform discharges and related patterns. J Clin Neurophysiol 2005;22(2):79–91.

seizure within 24 hours of recording, but only 80% of comatose patients did. After 48 hours, these numbers were 98% for non-comatose patients and 87% for comatose patients. Thus, for non-comatose patients, even 2 days of CEEG recording may not capture all seizures, although the clinical significance of such seizures is uncertain. Similar results were found in critically ill children, with 52% of seizures noted in the first hour, 87% noted in the first 24 hours, and 93% within 48 hours (Fig. 17.24) (74). Some recent studies have suggested that the initial background pattern on CEEG

Continuous EEG Monitoring in the Intensive Care Unit

573

Figure 17.18: Example of seizure terminating with lateralized periodic discharges (LPDs) in a 45-year-old man with left parietal hemorrhage. Background EEG showed periodic discharges at P3/T5 and intermittent electrographic seizures over the left hemisphere. This panel shows the termination of a seizure with return of periodic discharges. LFF = 1 Hz, HFF = 70 Hz, notch off.

is highly predictive of whether or not a patient will have seizures. In a retrospective study of 242 consecutive patients undergoing CEEG, only 3% of the 110 patients who did not have epileptiform abnormalities on their initial 30-minute screening EEG later went on to have seizures (75). In another series of 243 consecutive patients undergoing CEEG, no patients with diffuse slowing or triphasic waves on their initial EEG went on to have seizures (76). However, although all patients in both studies were recorded for a minimum of 18 to 24 hours, patients with more benign initial EEGs were likely to be disconnected from the EEG earlier (e.g., median duration of recording of 1 day for those without seizures vs. 4 days for those

with seizures in [76]), possibly leading to an underestimation of seizures in this group. Based on such work, it may be reasonable to terminate CEEG quickly when the baseline study reveals a normal background, diffuse slowing, or pattern suggestive of metabolic encephalopathy, such as triphasic waves, especially with limited resources and other patients who may be at higher risk of nonconvulsive seizures. In situations where the purpose of CEEG is not purely seizure detection, but instead monitoring neurological function to detect changes (e.g., monitoring for delayed ischemia in patients with subarachnoid hemorrhages), clinical circumstances dictate the duration of recording.

574

Continuous EEG Monitoring in the Intensive Care Unit

Figure 17.19: A 78-year-old man with a remote history of a left-sided subdural hemorrhage who presented with persistent change in mental status. The routine EEG is shown on a compressed time scale at the top. The Fourier Transform of the EEG signal is shown as a DSA at the bottom (left hemisphere on top, right hemisphere on bottom). For the DSA, the vertical axis represents frequency (0 to 20 Hz) and the horizontal axis represents time. The colors represent power. Two electrographic seizures are present—arrow indicates the onset of the first one, and the location of the raw EEG sample above. The onset is associated with an increase in power at all frequencies, but especially at low frequencies, less than 5 Hz, mainly on the left but with some spread to the right.

The true utility of CEEG, that is, its impact on outcome in critically ill patients, is somewhat uncertain. As discussed in the coming sections, CEEG is clearly useful for detecting nonconvulsive seizures in critically ill patients, detecting ongoing seizure activity after treatment of convulsive seizures, and potentially for detecting additional cerebral injury in patients with different forms of preexisting neuronal insults (e.g., traumatic brain injury (TBI), stroke, etc.). However, the actual impact of CEEG on outcome in such patients is less certain. The data acquired by CEEG clearly impacts decision-making: in one early study, CEEG in the neurological ICU had a decisive or contributing impact on medical decision-making in 82% of patients (11). In a series of 300 consecutive patients who underwent CEEG monitoring for various indications between 1993 and 1997, Vespa et al. (77) reported an impact on clinical decision-making in greater than 90%. They also argued that clinical efficacy was demonstrated by

the fact that during this period when CEEG was becoming widely used at their institution, there was an improvement in outcome (as measured by Glasgow Outcome Scale), with decreased overall cost of hospitalization and decreased length of stay, although, admittedly, no other factors that may have changed during this time period were accounted for. Furthermore, in their estimation, the cost of CEEG comprised only 1% of the overall cost of the hospitalization. In another retrospective analysis of 300 consecutive CEEG studies, the findings led to a change in antiepileptic drug prescribing in 52% of the cases: 14% had initiation of therapy, 33% had modification of therapy, and 5% had discontinuation of therapy (78). Detection of electrographic seizures led to a change in antiepileptic drug therapy in 28% of the studies. In a prospective series of 100 critically ill children, CEEG led to specific changes in clinical management in 59 (79). In a continuation of the same prospective series with 200 critically

Continuous EEG Monitoring in the Intensive Care Unit

575

Figure 17.20: A 63-year-old man with prior right parieto-occipital stroke with seizures at that time. He presented 6 years later with focal motor seizures involving the left arm and leg. He continued to have episodes of confusion in spite of being on levetiracetam and phenytoin. These seizures are clearly seen on all three QEEG tools shown—DSA at the top, followed by amplitude-integrated EEG (aEEG), followed by the rhythmicity spectrogram. Samples of ictal and interictal EEG are shown later. While these seizures are easy to recognize on the raw EEG, the QEEG tools would permit identification of seizures by a nonexpert and would permit rapid identification of all seizures in a prolonged recording.

ill children, only the 43 patients with electrographic status epilepticus had a significantly increased risk of mortality (odd ratio 5.1) or pediatric cerebral performance category worsening (odds ratio 17.3), whereas the 41 patients with electrographic seizures had no significant increased risk for either (odds ratios of 1.3 and 1.2, respectively) (80). Intuitively, it certainly makes sense that knowledge and potential treatment of seizures that would otherwise go unnoticed or warning of impending cerebral injury should improve outcome. However, actually demonstrating this in a heterogeneous, critically ill population is difficult and has not been done in a controlled fashion.

CLINICAL APPLICATIONS, STATUS EPILEPTICUS Both GCSE and NCSE are fairly common after acute brain injuries from different etiologies, including trauma, intracranial hemorrhage, cerebral ischemia, hypoxia/anoxia, metabolic, and toxic causes. The utility of EEG in managing convulsive status epilepticus is well established; in fact, it is mandatory for management of refractory status epilepticus according to the Neurocritical Care Society and others. When generalized convulsive status epilepticus has been partially treated or has been going on for a prolonged period, the clinical signs may be subtle. Up to 20% of patients continue to

576

Continuous EEG Monitoring in the Intensive Care Unit

Figure 17.21: An 86-year-old woman with a history of multiple ischemic strokes and a prior cerebellar hemorrhage who presented with multiple seizures and was intubated for respiratory distress. QEEG is shown at the top and the raw EEG (at the vertical line) during a subtle seizure from the left temporal lobe is shown in the bottom. Even these subtle seizures have a clear signature on two of the QEEG tools displayed, rhythmicity spectrogram (top) and DSA (middle). This permits the easy identification of five seizures during the epoch displayed. However, note that the aEEG tool does not show these seizures clearly.

have nonconvulsive seizures and NCSE after the clinical GCSE has subsided (6,7). If both nonconvulsive seizures without status epilepticus and NCSE are included, as many as 48% of patients continued to have seizure activity in the 24 hours after cessation of clinical seizure activity (7). Mortality and chance of poor outcome are significantly higher for patients with ongoing ictal activity compared with those where GCSE truly subsided. For the group without ongoing ictal activity, mortality was 13% and poor outcome was 20%; for patients with NCSE, these percentages were 51% and 58%, respectively; for patients with ongoing seizure activity, but not NCSE, the percentages were 32% and 46%, respectively. Owing to the size and heterogeneity of the population in this study, multivariate analysis was difficult but did suggest that age and etiology were important predictors of mortality

and outcome. However, the occurrence of NCSE after GCSE may independently increase the odds of mortality and poor outcome; the occurrence of nonconvulsive seizures did not. Because much of this seizure activity is without obvious clinical correlate, an EEG is necessary in this situation to evaluate for ongoing seizures. Such ongoing seizures and NCSE warrant treatment. NCSE is also being increasingly recognized as a common occurrence in patients with acute neurological insults or unexplained change in mental status. Because of the difficulty in making this diagnosis without EEG, diagnosis is often delayed (81). For example, in patients with subarachnoid hemorrhages, in one series of 233 patients, where only a subset of the 101 patients who had stupor/coma underwent CEEG monitoring, 8 were

Continuous EEG Monitoring in the Intensive Care Unit

577

Figure 17.22: A 65-year-old man with history of right parietal glioblastomamultiforme who had persistent change in mental status after partial resection of the tumor. QEEG is shown at the top along with samples of interictal and ictal EEG during a seizure from the right parietal ­region at the bottom. This seizure is most clearly seen on the aEEG from the right side (middle panel of QEEG) and much less apparent on the DSA or rhythmicity spectrogram.

found to be in nonconvulsive status epilepticus (82), giving a rate of 4% for patients with stupor/coma and subarachnoid hemorrhage (SAH). The diagnosis of NCSE was made an average of 18 days after the SAH (range 5 to 38 days). In another series of 389 consecutive patients with SAH, 3% were found to have nonconvulsive status epilepticus, on average 10 days after the bleed (range 2 to 20 days) (83). However, for both series, this likely ­represents an underestimation, as only patients where there was a clinical suspicion for seizures underwent continuous EEG monitoring. In both series, the patients with NCSE tended to be older and had higher Hunt and Hess or Fisher grades. Furthermore, the NCSE was extremely difficult

to treat/control, and most patients were transitioned to comfort care and expired in the hospital. CEEG is necessary to guide the treatment of refractory status epilepticus of all kinds. As mentioned previously, without CEEG, it is very possible that undertreatment may occur, that is, electrographic seizure activity persists even though clinical seizure activity has subsided. This is especially the case when sedatives or neuromuscular junction blocking agents are being used. Thus, CEEG monitoring is necessary to alert the treating team if the seizure does not stop. In addition, CEEG guides the dosing of anesthetic agents as clinical signs of efficacy are lacking. The continuous or intermittent infusion

578

Continuous EEG Monitoring in the Intensive Care Unit

Figure 17.23: A 59-year-old man with coronary artery disease who was found unresponsive at home. MRI brain showed T2 hyperintensities in the cerebellum and bilateral occipital head regions. He had persistent change in mental status with some fluctuations. EEG on hospital day 3 showed frequent electrographic seizures from the right posterior temporal/occipital region. EEG during one of these seizures is shown at the bottom. QEEG trends are shown at the top. These seizures are apparent on the aEEG and rhythmicity spectrogram from the right hemisphere but not on the DSA.

of agents such as propofol, midazolam, or pentobarbital is adjusted on the basis of CEEG to obtain either a desired level of burst suppression or at least seizure cessation. Overtreatment, with associated cardiovascular or respiratory compromise, is possible without CEEG monitoring. In critically ill patients, the appropriate treatment of NCSE depends on many factors. Deciding on the appropriate treatment and even whether or not treatment should be undertaken requires integrating many clinical factors. These include the underlying etiology and prognosis, the likelihood of response to treatment, the morbidity of the treatment, and others. In specific clinical situations, these factors can combine in multiple ways to dictate the appropriateness of a particular intervention (84).

CLINICAL APPLICATIONS, NONCONVULSIVE SEIZURES Seizures are a common occurrence in the neurological ICU setting (see Fig. 17.2). Early series of continuous EEG monitoring revealed a very high prevalence of seizures, up to 34% (11), with a substantial proportion (76%) exhibiting NCSE. Furthermore, there was recognition that most of these seizures were nonconvulsive, that is, without clear clinical signs other than change in mental status. Thus, without the use of EEG, they would go unrecognized. In the pediatric population, the incidence may be even higher (see Fig. 17.2). In a series of 100 critically ill children with acute encephalopathy, 46% had seizures and/or status epilepticus, 19% had nonconvulsive status

579

Continuous EEG Monitoring in the Intensive Care Unit TABLE 17.5

100

Indications for Continuous EEG Monitoring

a. Fluctuating mental status b. Unexplained alteration of mental status c. Acute supratentorial brain injury with altered mental status d. After convulsive status epilepticus

II. Characterization of spells a. Episodic posturing, other paroxysmal or repetitive movements b. Subtle twitching, nystagmus, eye deviation, chewing c. Paroxysmal autonomic spells, including tachycardia

III. Assessment of level of sedation and following trends

Patients seizures detected (%)

I. Detection of subclinical seizures 75

50 Adults Children 25

IV. Management of burst suppression and anesthetic coma V. Detection of ischemia a. After subarachnoid hemorrhage b. During and after vascular neurosurgical or interventional neuroradiology procedures c. In patients with hemodynamic lesions and borderline flow d. In other patients at risk for in-hospital acute ischemia VI. Prognostication

epilepticus (74). Furthermore, of the 46 children with seizures, 32 had only nonconvulsive seizures. In a prospective series of 100 critically ill children (aged 2 months to 17 years) who had a suppressed level of consciousness (GCS < 8) for any reason, seizures were found in only 7 patients, of whom 6 had a history of epilepsy and had witnessed seizures prior to the CEEG (85). All seizures were recorded within the first 3 hours of the CEEG. The reasons for the low incidence and early detection of seizures in this series are unclear, although it may partially relate to the fact that patients did not necessarily have a neurological diagnosis, just suppressed level of consciousness. Also, CEEG was only performed for a median of 20 hours. CEEG was helpful in ruling out epileptic seizures in 14/18 patients suspected of having seizures by the ICU team. Certain specific clinical situations are commonly associated with nonconvulsive seizures in critically ill patients. After ischemic strokes, the incidence

0 Start of CEEG

25% for >4 hours) 1 to 3 days prior to angiographic confirmation of vasospasm, which was superior to this and other parameters calculated

from scalp EEG recordings (137). Such findings suggest that intracranial multimodal monitoring may have a future role in monitoring the neurological status and function of patients in neurocritical care units; however, the risks, benefits, and potential applications have not been established, and the techniques are not widely available yet (138). In addition, there are several published examples of nonconvulsive seizures recorded with intracranial electrodes that are not apparent on simultaneously recorded scalp EEG (Fig. 17.30).

Continuous EEG Monitoring in the Intensive Care Unit

589

C

Figure 17.27: (continued)

Intracranial recordings have also been used to demonstrate and explore the phenomenon of spreading depolarizations, a form of cortical spreading depression, which requires substantial energy for recovery (repolarization). These depolarizations occur spontaneously and commonly after various forms of brain injury, including ischemia, hemorrhage, and trauma. In permanently damaged tissue, this can reflect a persistent state with electrocorticographic depression. In healthy tissue, this leads to increased perfusion and is transient; however, in damaged tissue, it may lead to an increase in

metabolic demand in tissue already with borderline function; this places the tissue at risk for permanent damage (139). When subdural electrodes have been used to record depolarizations (requires direct current recording—no or minimal low frequency filtering), the presence of prolonged spreading depolarizations (especially when associated with absence of spontaneous activity) has been associated with poor outcome after brain trauma (140,141). Such depolarizations may precede additional neuronal injury in some situations. For example, in patients with major SAHs, these were

590

Continuous EEG Monitoring in the Intensive Care Unit A

Figure 17.28: A: EEG of a 73-year-old patient after cardiac arrest. EEG was performed 14 hours after return of spontaneous circulation. B–D: Consecutive panels showing a nonconvulsive seizure arising from a background of burst suppression. These patterns indicated a poor prognosis for functional recovery; the patient died after withdrawal of care. LFF = 1 Hz, HFF = 70 Hz, notch off.

observed in 13 out of 18 patients and were highly associated with delayed ischemic neurological deficits (positive predictive value of 86% and negative predictive value of 100%) (142). Electrocorticographic seizures have also been associated with cortical spreading depolarization in patients with acute brain injury (143). In a subset of patients, the spreading depolarization was immediately preceded by a prolonged nonconvulsive seizure, lending further support to the idea that such seizures may lead to additional damage in patients with brain injury.

FUTURE DIRECTIONS Even though CEEG has grown rapidly, the field is still evolving and many questions remain unanswered or only partially answered. These include what type of patients should undergo CEEG? Once CEEG is initiated, how long should it be continued, and can the initial findings on EEG be relied

upon to help answer this question? Can CEEG be used reliably and practically to detect changes in neurological function beyond seizures, such as DCI? Can non-neurophysiologists be adequately trained to monitor CEEG or at least QEEG trends? Can QEEG trends be used to speed up/improve review of data? Can other monitoring modalities be integrated with CEEG to provide additional meaningful information regarding neurological and systemic function in critically ill patients? Lastly, what is the true clinical impact of CEEG monitoring—does it truly improve patient outcomes? Will it do so when done in real time with improved automated detection algorithms? If so, will this be cost effective? The use of biomarkers of neuronal injury and individualized, physiology-driven decision-making may play a role in determining the utility of CEEG and interventions based on CEEG. Finally, systematic collection and analysis of data about CEEG, likely in large multicenter cohorts, is needed to answer these questions in a meaningful way.

Continuous EEG Monitoring in the Intensive Care Unit

B

Figure 17.28: (continued)

591

592

Continuous EEG Monitoring in the Intensive Care Unit

C

Figure 17.28: (continued)

Continuous EEG Monitoring in the Intensive Care Unit D

Figure 17.28: (continued)

593

594

Continuous EEG Monitoring in the Intensive Care Unit

Figure 17.29: Relationship of physiological variables with quantitative EEG ­parameters during ictal events. During the repetitive seizure events, partial brain tissue oxygen tension (PbtO2, arbitrary units) consistently decreased followed by a surge in intracranial pressure, brain temperature, and cerebral blood flow (CBF, arbitrary units). In addition, total power on the EEG is well synchronized with the rhythmicity index and the existence of rhythmicity medium- to high-frequency waves in the spectrogram, followed by lower frequency waves, suggesting an evolving ictal rhythm. (Figure courtesy of Dr. Jan Claassen, New York, New York. © Jan Claassen.)

Figure 17.30: Cyclic seizure on intracranial EEG only. EEG sample from an older woman with Hunt and Hess grade III subarachnoid hemorrhage and a left frontal mini-depth electrode. During a 5-hour period, she had cyclic seizures in the depth electrode only, each lasting about one minute and recurring every few minutes. A typical seizure is shown. The bottom six channels from the depth electrode show a clearly evolving seizure; the upper channels of scalp EEG show no hint of a seizure. (With permission from Hirsch LJ, Brenner RP. Atlas of EEG in critical care. West Sussex, UK: John Wiley & Sons, 2010:334.)

Continuous EEG Monitoring in the Intensive Care Unit

References

1. Pohlmann-Eden B, Hoch D, Cochlus JI, et al. Periodic lateralized epileptiform ­discharges—a critical review. J Clin Neurophysiol 1996;13(6):519–530. 2. Kennedy JD, Gerard EE. Continuous EEG monitoring in the intensive care unit. Curr Neurol Neurosci Rep 2012;12(4):419–428. 3. Astrup J, Siesjo BK, Symon L. Thresholds in cerebral ischemia—the ischemic penumbra. Stroke 1981;12(6):723–725. 4. Chiappa KH, Burke SR, Young RR. Results of electroencephalographic monitoring during 367 carotid endarterectomies: use of a dedicated minicomputer. Stroke 1979;10(4):381–388. 5. Salerno TA, Lince DP, White DN, et al. Monitoring of electroencephalogram during open-heart surgery: a prospective analysis of 118 cases. J Thorac Cardiovasc Surg 1978;76(1):97–100. 6. Treiman DM, Meyers PD, Walton NY, et al. A comparison of four treatments for generalized convulsive status epilepticus. Veterans Affairs Status Epilepticus Cooperative Study Group. N Engl J Med 1998;339(12):792–798. 7. DeLorenzo RJ, Waterhouse EJ, Towne AR, et al. Persistent nonconvulsive status epilepticus after the control of convulsive status epilepticus. Epilepsia 1998;39(8):833–840. 8. Jordan KG. Neurophysiologic monitoring in the neuroscience intensive care unit. Neurol Clin 1995;13(3):579–626. 9. Miller JD, Piper IR, Jones PA. Integrated multimodality monitoring in the neurosurgical intensive care unit. Neurosurg Clin N Am 1994;5(4):661–670. 10. Sheinberg M, Kanter MJ, Robertson CS, et al. Continuous monitoring of jugular venous oxygen saturation in ­head-injured patients. J Neurosurg 1992;76(2):212–217. 11. Jordan KG. Continuous EEG and evoked potential monitoring in the neuroscience ­intensive care unit. J Clin Neurophysiol 1993;10(4):445–475. 12. Vespa PM, O’Phelan K, Shah M, et al. Acute seizures after intracerebral hemorrhage: a factor in progressive midline shift and outcome. Neurology 2003;60(9):1441–1446. 13. Maynard D, Prior PF, Scott DF. Device for continuous monitoring of cerebral activity in resuscitated patients. Br Med J 1969;4(5682):545–546. 14. Prior PF, Maynard D, Scott DF. A new device for continuous monitoring of cerebral activity: its use following cerebral anoxia. Electroencephalogr Clin Neurophysiol 1970;28(4):423–424. 15. Trewby PN, Casemore C, Williams R. Continuous bipolar recording of the EEG in patients with fulminant hepatic failure. Electroencephalogr Clin Neurophysiol 1978;45(1):107–110. 16. Eyre JA, Oozeer RC, Wilkinson AR. Diagnosis of neonatal seizure by continuous recording and rapid analysis of the electroencephalogram. Arch Dis Child 1983;58(10):785–790. 17. Eyre JA, Oozeer RC, Wilkinson AR. Continuous electroencephalographic recording to detect seizures in paralysed newborn babies. Br Med J (Clin Res Ed) 1983;286(6370): 1017–1018. 18. Coen RW, McCutchen CB, Wermer D, et al. Continuous monitoring of the electroencephalogram following perinatal asphyxia. J Pediatr 1982;100(4):628–630. 19. Bickford RE, Fleming N, Stewart L, et al. Proceedings: compressed spectral displays in epileptic patients. ­Epilepsia 1972;13(2):347–348. 20. Bricolo A, Turazzi S, Faccioli F, et al. Clinical application of compressed spectral array in long-term EEG monitoring of comatose patients. Electroencephalogr Clin Neurophysiol 1978;45(2):211–225. 21. Cant BR, Shaw NA. Monitoring by compressed spectral array in prolonged coma. ­Neurology 1984;34(1):35–39. 22. American Society of Electroneurodiagnostic Technologists Inc. National competency skill standards for ICU/cEEG monitoring. Am J Electroneurodiagnostic Technol 2008;48(4):258–264.

595

23. Vulliemoz S, Perrig S, Pellise D, et al. Imaging compatible electrodes for continuous electroencephalogram monitoring in the intensive care unit. J Clin Neurophysiol 2009;26(4):236–243. 24. Ives JR. New chronic EEG electrode for critical/intensive care unit monitoring. J Clin Neurophysiol 2005;22(2):119–123. 25. Das RR, Lucey BP, Chou SH, et al. The utility of conductive plastic electrodes in prolonged ICU EEG monitoring. Neurocrit Care 2009;10(3):368–372. 26. Young GB, Ives JR, Chapman MG, et al. A comparison of subdermal wire electrodes with collodion-applied disk electrodes in long-term EEG recordings in ICU. Clin Neurophysiol 2006;117(6):1376–1379. 27. Mirsattari SM, Tapsell LM, Ives JR, et al. MR imaging-compatible electroencephalography electrode system for an epilepsy monitoring unit. AJNR Am J Neuroradiol 2008;29(9):1649–1651. 28. Martz GU, Hucek C, Quigg M. Sixty day continuous use of subdermal wire electrodes for EEG monitoring during treatment of status epilepticus. Neurocrit Care 2009;11(2):223–227. 29. Jarrar R, Buchhalter J, Williams K, et al. Technical tips: electrode safety in pediatric prolonged EEG recordings. Am J Electroneurodiagnostic Technol 2011;51(2):114–117. 30. Kolls BJ, Husain AM. Assessment of hairline EEG as a screening tool for nonconvulsive status epilepticus. Epilepsia 2007;48(5):959–965. 31. Young GB, Sharpe MD, Savard M, et al. Seizure detection with a commercially available bedside EEG monitor and the subhairline montage. Neurocrit Care 2009;11(3):411–416. 32. Karakis I, Montouris GD, Otis JA, et al. A quick and reliable EEG montage for the detection of seizures in the critical care setting. J Clin Neurophysiol 2010;27(2):​100–105. 33. Kolls BJ, Olson DM, Gallentine WB, et al. Electroencephalography leads placed by nontechnologists using a template system produce signals equal in quality to technologistapplied, collodion disk leads. J Clin Neurophysiol 2012;29(1):42–49. 34. Trevathan E, Ellen R. Grass lecture: rapid EEG analysis for intensive care decisions in status epilepticus. Am J Electroneurodiagnostic Technol 2006;46(1):4–17. 35. Hirsch LJ, Laroche SM, Gaspard N, et al. American Clinical Neurophysiology Society’s standardized critical care EEG terminology: 2012 version. 2012. Available from http:// www.acns.org/pdfs/­ACNS2012Nomenclature-2of5_Text_Final-03_Jun.pdf. Accessed July 8, 2012. 36. Hirsch LJ, LaRoche SM, Gaspard N, et al. American Clinical Neurophysiology Society’s standardized critical care EEG terminology: 2012 version. J Clin Neurophysiol 2013;30(1):1–27. 37. Beniczky S, Aurlien H, Brogger JC, et al. Standardized computer-based organized reporting of EEG: SCORE. Epilepsia 2013;54(6):1112–1124. 38. Hirsch LJ, Brenner RP, Drislane FW, et al. The ACNS subcommittee on research terminology for continuous EEG monitoring: proposed standardized terminology for rhythmic and periodic EEG patterns encountered in critically ill patients. J Clin Neurophysiol 2005;22(2):128–135. 39. Gerber PA, Chapman KE, Chung SS, et al. Interobserver agreement in the interpretation of EEG patterns in critically ill adults. J Clin Neurophysiol 2008;25(5):241–249. 40. Mani R, Arif H, Hirsch LJ, et al. Interrater reliability of ICU EEG research terminology. J Clin Neurophysiol 2012;29(3):203–212. 41. Chong DJ, Hirsch LJ. Which EEG patterns warrant treatment in the critically ill? Reviewing the evidence for treatment of periodic epileptiform discharges and related patterns. J Clin Neurophysiol 2005;22(2):79–91. 42. Orta DS, Chiappa KH, Quiroz AZ, et al. Prognostic implications of periodic epileptiform discharges. Arch Neurol 2009;66(8):985–991. 43. de la Paz D, Brenner RP. Bilateral independent periodic lateralized epileptiform ­discharges: clinical significance. Arch Neurol 1981;38(11):713–715. 44. Foreman B, Claassen J, Abou Khaled K, et al. Generalized periodic discharges in the critically ill: a case-control study of 200 patients. Neurology 2012;79(19):1951–1960.

596

Continuous EEG Monitoring in the Intensive Care Unit

45. Husain AM, Mebust KA, Radtke RA. Generalized periodic epileptiform discharges: etiologies, relationship to status epilepticus, and prognosis. J Clin Neurophysiol 1999; 16(1):51–58. 46. Snodgrass SM, Tsuburaya K, Ajmone-Marsan C. Clinical significance of periodic lateralized epileptiform discharges: relationship with status epilepticus. J Clin Neurophysiol 1989;6(2):159–172. 47. Brenner RP, Schaul N. Periodic EEG patterns: classification, clinical correlation, and pathophysiology. J Clin Neurophysiol 1990;7(2):249–267. 48. Ali II, Pirzada NA, Vaughn BV. Periodic lateralized epileptiform discharges after complex partial status epilepticus associated with increased focal cerebral blood flow. J Clin Neurophysiol 2001;18(6):565–569. 49. Handforth A, Cheng JT, Mandelkern JT, et al. Markedly increased mesiotemporal lobe metabolism in a case with PLEDs: further evidence that PLEDs are a manifestation of partial status epilepticus. Epilepsia 1994;35(4):876–881. 50. Ergun EL, Salanci BV, Erbas B, et al. SPECT in periodic lateralized epileptiform discharges (PLEDs): a case report on PLEDs. Ann Nucl Med 2006;20(3):227–231. 51. Ong C, Gilmore E, Claassen J, et al. Impact of prolonged periodic epileptiform discharges on coma prognosis. Neurocrit Care 2012;17(1):39–44. 52. Young GB, Jordan KG, Doig GS. An assessment of nonconvulsive seizures in the intensive care unit using continuous EEG monitoring: an investigation of variables associated with mortality. Neurology 1996;47(1):83–89. 53. Hirsch LJ, Claassen J, Mayer SA, et al. Stimulus-induced rhythmic, periodic, or ictal discharges (SIRPIDs): a common EEG phenomenon in the critically ill. Epilepsia 2004;45(2):109–123. 54. Hirsch LJ, Pang T, Claassen J, et al. Focal motor seizures induced by alerting stimuli in critically ill patients. Epilepsia 2008;49(6):968–973. 55. Ronner HE, Ronner HE, Ponten SC, et al. Inter-observer variability of the EEG diagnosis of seizures in comatose patients. Seizure 2009;18(4):257–263. 56. Scheuer ML, Wilson SB. Data analysis for continuous EEG monitoring in the ICU: ­seeing the forest and the trees. J Clin Neurophysiol 2004;21(5):353–378. 57. Tao JD, Mathur AM. Using amplitude-integrated EEG in neonatal intensive care. J Perinatol 2010;30(Suppl):S73–S81. 58. Abend NS, Dlugos D, Herman S. Neonatal seizure detection using multichannel display of envelope trend. Epilepsia 2008;49(2):349–352. 59. Claassen J, Hirsch LJ, Kreiter KT, et al. Quantitative continuous EEG for detecting delayed cerebral i­schemia in patients with poor-grade subarachnoid hemorrhage. Clin Neurophysiol 2004;115(12):2699–2710. 60. Roustan JP, Valette S, Aubas P, et al. Can electroencephalographic analysis be used to determine s­ edation levels in critically ill patients? Anesth Analg 2005;101(4):1141– 1151, table of contents. 61. Diedler J, Sykora M, Bast T, et al. Quantitative EEG correlates of low cerebral perfusion in severe stroke. Neurocrit Care 2009;11(2):210–216. 62. Sackellares JC, Shiau DS, Halford JJ, et al. Quantitative EEG analysis for automated detection of nonconvulsive seizures in intensive care units. Epilepsy Behav 2011;22 Suppl 1:S69–S73. 63. Temko A, et al. EEG-based neonatal seizure detection with Support Vector Machines. Clin Neurophysiol 2011;122(3):464–473. 64. Stewart CP, Otsubo H, Ochi A, et al. Seizure identification in the ICU using quantitative EEG displays. ­Neurology 2010;75(17):1501–1508. 65. Cloostermans MC, de Vos CC, van Putten MJ. A novel approach for computer assisted EEG monitoring in the adult ICU. Clin Neurophysiol 2011;122(10):2100–2109. 66. Hirsch LJ. Continuous EEG monitoring in the intensive care unit: an overview. J Clin Neurophysiol 2004;21(5):332–340.

67. Claassen J, Taccone FS, Horn P, et al. Recommendations on the use of EEG monitoring in critically ill patients: consensus statement from the neurointensive care section of the ESICM. Intensive Care Med 2013;39(8):1337–1351. 68. Claassen J, Mayer SA, Kowalski RG, et al. Detection of electrographic seizures with continuous EEG monitoring in critically ill patients. Neurology 2004;62(10):1743–1748. 69. Murray DM, Boylan GB, Ali I, et al. Defining the gap between electrographic seizure burden, clinical expression and staff recognition of neonatal seizures. Arch Dis Child Fetal Neonatal Ed 2008;93(3):F187–F191. 70. Clancy RR, Legido A, Lewis D. Occult neonatal seizures. Epilepsia 1988;29(3):256–261. 71. Malone A, Ryan CA, Fitzgerald A, et al. Interobserver agreement in neonatal seizure identification. Epilepsia 2009;50(9):2097–2101. 72. McHugh JC, Downey T, Murphy RP, et al. Analysis of routine EEG usage in a general adult ICU. Ir J Med Sci 2009;178(3):263–266. 73. Sutter R, Fuhr P, Grize L, et al. Continuous video-EEG monitoring increases detection rate of non­convulsive status epilepticus in the ICU. Epilepsia 2011;52(3):453–457. 74. Abend NS, Gutierrez-Colina AM, Topjian AA, et al. Nonconvulsive seizures are common in critically ill children. Neurology 2011;76(12):1071–1077. 75. Shafi MM, Westover MB, Cole AJ, et al. Absence of early epileptiform abnormalities predicts lack of seizures on continuous EEG. Neurology 2012;79(17):1796–1801. 76. Swisher CB, Shah D, Sinha SR, et al. Baseline EEG pattern on continuous ICU EEG monitoring and ­incidence of seizures. 2013. 77. Vespa PM, Nenov V, Nuwer MR. Continuous EEG monitoring in the intensive care unit: early findings and clinical efficacy. J Clin Neurophysiol 1999;16(1):1–13. 78. Kilbride RD, Costello DJ, Chiappa KH. How seizure detection by continuous ­electroencephalographic monitoring affects the prescribing of antiepileptic medications. Arch Neurol 2009;66(6):723–728. 79. Abend NS, Topjian AA, Gutierrez-Colina AM, et al. Impact of continuous EEG monitoring on clinical management in ­critically ill children. Neurocrit Care 2011;​ 15(1):70–75. 80. Topjian AA, Gutierrez-Colina AM, Sanchez SM, et al. Electrographic status epilepticus is associated with mortality and worse short-term outcome in critically ill children. Crit Care Med 2013;41(1):210–218. 81. Kaplan PW. Nonconvulsive status epilepticus in the emergency room. Epilepsia 1996; 37(7):643–650. 82. Dennis LJ, Claassen J, Hirsch LJ, et al. Nonconvulsive status epilepticus after subarachnoid hemorrhage. ­Neurosurgery 2002;51(5):1136–1143; discussion 1144. 83. Little AS, Kerrigan JF, McDougall CG, et al. Nonconvulsive status epilepticus in patients suffering spontaneous ­subarachnoid hemorrhage. J Neurosurg 2007;106(5):805–811. 84. Ferguson M, Bianchi MT, Sutter R, et al. Calculating the risk benefit equation for aggressive treatment of ­non-convulsive status epilepticus. Neurocrit Care 2013;18(2):216–227. 85. Shahwan A, Bailey C, Shekerdemian L, et al. The prevalence of seizures in comatose children in the pediatric ­intensive care unit: a prospective video-EEG study. Epilepsia 2010;51(7):1198–1204. 86. Shinton RA, et al. The frequency, characteristics and prognosis of epileptic seizures at the onset of stroke. J Neurol Neurosurg Psychiatry 1988;51(2):273–276. 87. Davalos A, Cendra E, Genis D, et al. The frequency, characteristics and prognosis of epileptic seizures at the onset of stroke. J Neurol Neurosurg Psychiatry 1988;51(11):1464. 88. Weisberg LA, Shamsnia M, Elliott D. Seizures caused by nontraumatic parenchymal brain hemorrhages. Neurology 1991;41(8):1197–1199. 89. Faught E, Peters D, Bartolucci A, et al. Seizures after primary intracerebral hemorrhage. Neurology 1989;39(8):1089–1093. 90. Arboix A, Garcia-Eroles L, Massons JB, et al. Predictive factors of early seizures after acute cerebrovascular disease. Stroke 1997;28(8):1590–1594.

Continuous EEG Monitoring in the Intensive Care Unit 91. Bladin CF, Alexandrov AV, Bellavance A, et al. Seizures after stroke: a prospective multicenter study. Arch Neurol 2000;57(11):1617–1622. 92. Claassen J, Jette N, Chum F, et al. Electrographic seizures and periodic discharges after intracerebral ­hemorrhage. Neurology 2007;69(13):1356–1365. 93. Ohman J. Hypertension as a risk factor for epilepsy after aneurysmal subarachnoid ­hemorrhage and surgery. Neurosurgery 1990;27(4):578–581. 94. Baker CJ, Prestigiacomo CJ, Solomon RA. Short-term perioperative anticonvulsant ­prophylaxis for the surgical treatment of low-risk patients with intracranial aneurysms. Neurosurgery 1995;37(5):863–870; discussion 870–871. 95. Vespa PM, Nuwer MR, Juhasz C, et al. Early detection of vasospasm after acute subarachnoid ­hemorrhage using continuous EEG ICU monitoring. Electroencephalogr Clin Neurophysiol 1997;103(6):607–615. 96. Labar DR, Fisch BJ, Pedley TA, et al. Quantitative EEG monitoring for patients with subarachnoid ­ hemorrhage. Electroencephalogr Clin Neurophysiol 1991;78(5): 325–332. 97. Butzkueven H, Evans AH, Pitman A, et al. Onset seizures independently predict poor outcome after subarachnoid hemorrhage. Neurology 2000;55(9):1315–1320. 98. Hart RG, Byer JA, Slaughter JR, et al. Occurrence and implications of seizures in subarachnoid hemorrhage due to ruptured intracranial aneurysms. Neurosurgery 1981;8(4):​ 417–421. 99. Annegers JF, Grabow JD, Groover RV, et al. Seizures after head trauma: a population study. Neurology 1980;30(7 Pt 1):683–689. 100. Jennett WB, Lewin W. Traumatic epilepsy after closed head injuries. J Neurol Neurosurg Psychiatry 1960;23:295–301. 101. Kollevold T. Immediate and early cerebral seizures after head injuries, part I. J Oslo City Hosp 1976;26(12):99–114. 102. Lee ST, Lui TN, Wong CW, et al. Early seizures after moderate closed head injury. Acta Neurochir (Wien) 1995;137(3–4):151–154. 103. Vespa PM, Nuwer MR, Nenov V, et al. Increased incidence and impact of nonconvulsive and convulsive seizures after traumatic brain injury as detected by continuous electroencephalographic monitoring. J Neurosurg 1999;91(5):750–760. 104. Ronne-Engstrom E, Winkler T. Continuous EEG monitoring in patients with traumatic brain injury reveals a high incidence of epileptiform activity. Acta Neurol Scand 2006;114(1):47–53. 105. Jette N, Claassen J, Emerson RG, et al. Frequency and predictors of nonconvulsive seizures during continuous electroencephalographic monitoring in critically ill children. Arch Neurol 2006;63(12):1750–1755. 106. Williams K, Jarrar R, Buchhalter J. Continuous video-EEG monitoring in pediatric intensive care units. Epilepsia 2011;52(6):1130–1136. 107. Wusthoff CJ, Dlugos DH, Gutierrez-Colina A, et al. Electrographic seizures during therapeutic hypothermia for neonatal hypoxic-ischemic encephalopathy. J Child Neurol 2011;26(6):724–728. 108. Glass HC, Nash KB, Bonifacio SL, et al. Seizures and magnetic resonance imagingdetected brain injury in ­newborns cooled for hypoxic-ischemic encephalopathy. J Pediatr 2011;159(5):731–735.e1. 109. Abend NS, Topjian A, Ichord R, et al. Electroencephalographic monitoring during hypothermia after pediatric cardiac arrest. Neurology 2009;72(22):1931–1940. 110. McBride MC, Laroia N, Guillet R. Electrographic seizures in neonates correlate with poor neurodevelopmental outcome. Neurology 2000;55(4):506–513. 111. Pisani F, Cerminara C, Fusco C, et al. Neonatal status epilepticus vs recurrent neonatal seizures: clinical findings and outcome. Neurology 2007;69(23):2177–2185. 112. Nevander G, Ingvar M, Auer R, et al. Status epilepticus in well-oxygenated rats causes neuronal necrosis. Ann Neurol 1985;18(3):281–290.

597

113. Nitsch C, Suzuki R, Fujiwara K, et al. Incongruence of regional cerebral blood flow increase and blood-brain barrier opening in rabbits at the onset of seizures induced by bicuculline, methoxypyridoxine, and kainic acid. J Neurol Sci 1985;67(1):67–79. 114. Krsek P, Mikulecka A, Druga R, et al. Long-term behavioral and morphological consequences of nonconvulsive status epilepticus in rats. Epilepsy Behav 2004;5(2):180–191. 115. Williams AJ, Tortella FC, Lu XM, et al. Antiepileptic drug treatment of nonconvulsive seizures induced by experimental focal brain ischemia. J Pharmacol Exp Ther 2004;311(1):220–227. 116. Oddo M, Carrera E, Claassen J, et al. Continuous electroencephalography in the medical intensive care unit. Crit Care Med 2009;37(6):2051–2056. 117. Carrera E, Claassen J, Oddo M, et al. Continuous electroencephalographic monitoring in critically ill patients with central nervous system infections. Arch Neurol 2008;65(12):1612–1618. 118. Bellinger DC, Jonas RA, Rappaport LA, et al. Developmental and neurologic status of children after heart surgery with hypothermic circulatory arrest or low-flow cardiopulmonary bypass. N Engl J Med 1995;332(9):549–555. 119. Vespa PM, Miller C, McArthur D, et al. Nonconvulsive electrographic seizures after traumatic brain injury result in a delayed, prolonged increase in intracranial pressure and metabolic crisis. Crit Care Med 2007;35(12):2830–2836. 120. Vespa PM, McArthur DL, Xu Y, et al. Nonconvulsive seizures after traumatic brain injury are associated with hippocampal atrophy. Neurology 2010;75(9):792–798. 121. DeGiorgio CM, Correale JD, Gott PS, et al. Serum neuron-specific enolase in human status epilepticus. Neurology 1995;45(6):1134–1137. 122. Gross CE, Adams HP Jr, Sokoll MD, et al. Use of anticoagulants, electroencephalographic monitoring, and barbiturate cerebral protection in carotid endarterectomy. Neurosurgery 1981;9(1):1–5. 123. Rathakrishnan R, Gotman J, Dubeau F, et al. Using continuous electroencephalography in the management of delayed cerebral ischemia following subarachnoid hemorrhage. Neurocrit Care 2011;14(2):152–161. 124. Zandbergen EG, Hijdra A, Koelman JH, et al. Prediction of poor outcome within the first 3 days of postanoxic coma. Neurology 2006;66(1):62–68. 125. Legriel S, Bruneel F, Sediri H, et al. Early EEG monitoring for detecting postanoxic status epilepticus during therapeutic hypothermia: a pilot study. Neurocrit Care 2009;11(3):338–344. 126. Laroche S, Mani R, Arif H, et al. In reply. J Clin Neurophysiol 2013;30(2):210. 127. Mani R, Schmitt SE, Mazer M, et al. The frequency and timing of epileptiform activity on continuous electroencephalogram in comatose post-cardiac arrest syndrome patients treated with therapeutic hypothermia. Resuscitation 2012;83(7):840–847. 128. Alvarez V, Oddo M, Rossetti AO. Stimulus-induced rhythmic, periodic or ictal discharges (SIRPIDs) in comatose survivors of cardiac arrest: characteristics and prognostic value. Clin Neurophysiol 2013;124(1):204–208. 129. Kessler SK, Topjian AA, Gutierrez-Colina AM, et al. Short-term outcome prediction by electroencephalographic features in children treated with therapeutic hypothermia after cardiac arrest. Neurocrit Care 2011;14(1):37–43. 130. Rundgren M, Rosen I, Friberg H. Amplitude-integrated EEG (aEEG) predicts outcome after cardiac arrest and induced hypothermia. Intensive Care Med 2006;32(6): 836–842. 131. Rossetti AO, Urbano LA, Delodder F, et al. Prognostic value of continuous EEG monitoring during therapeutic hypothermia after cardiac arrest. Crit Care 2010;14(5):R173. 132. Rossetti AO, Oddo M, Logroscino G, et al. Prognostication after cardiac arrest and hypothermia: a prospective study. Ann Neurol 2010;67(3):301–307. 133. Wartenberg KE, Mayer SA. Multimodal brain monitoring in the neurological intensive care unit: where does continuous EEG fit in? J Clin Neurophysiol 2005;22(2):124–127.

598

Continuous EEG Monitoring in the Intensive Care Unit

134. Wartenberg KE, Schmidt JM, Mayer SA. Multimodality monitoring in neurocritical care. Crit Care Clin 2007;23(3):507–538. 135. Wright WL. Multimodal monitoring in the ICU: when could it be useful? J Neurol Sci 2007;261(1–2):10–15. 136. Waziri A, Claassen J, Stuart RM, et al. Intracortical electroencephalography in acute brain injury. Ann Neurol 2009;66(3):366–377. 137. Stuart RM, Waziri A, Weintraub D, et al. Intracortical EEG for the detection of vasospasm in patients with poor-grade subarachnoid hemorrhage. Neurocrit Care 2010;13(3):355–358. 138. Stuart RM, Schmidt M, Kurtz P, et al. Intracranial multimodal monitoring for acute brain injury: a single institution review of current practices. Neurocrit Care 2010;12(2):188–198. 139. Dreier JP, Drenckhahn C, Woitzik J, et al. Spreading ischemia after aneurysmal subarachnoid hemorrhage. Acta Neurochir Suppl 2013;115:125–129.

140. Hartings JA, Watanabe T, Bullock MR, et al. Spreading depolarizations have prolonged direct current shifts and are associated with poor outcome in brain trauma. Brain 2011;134(Pt 5):1529–1540. 141. Hartings JA, Bullock MR, Okonkwo DO, et al. Spreading depolarisations and outcome after traumatic brain injury: a prospective observational study. Lancet Neurol 2011;10(12):1058–1064. 142. Dreier JP, Woitzik J, Fabricius M, et al. Delayed ischaemic neurological deficits after subarachnoid haemorrhage are associated with clusters of spreading depolarizations. Brain 2006;129(Pt 12):3224–3237. 143. Fabricius M, Fuhr S, Willumsen L, et al. Association of seizures with cortical spreading depression and peri-infarct depolarisations in the acutely injured human brain. Clin Neurophysiol 2008;119(9):1973–1984.

18

Sleep Disorders: Laboratory Evaluation RODNEY A. RADTKE

Polysomnography: Technical Aspects EEG Sleep Recording Anterior Tibialis EMG Respiratory Monitoring Recording Montage Ambulatory PSG Recording Computerized PSG and Computer-Assisted Scoring Polysomnography: Interpretation EEG-Related Variables Respiratory-Related Variables Movement-Related Variables

The rapid growth of sleep disorders medicine in the United States has led to an increasing role for clinical electroencephalographers (EEGers) in the evaluation of sleep and its disorders. However, most clinical neurophysiology training programs still do not incorporate the neurophysiologic or clinical aspects of sleep disorders into their curriculum. In this chapter, I will provide an introductory overview to the sleep laboratory evaluation of sleep disorders that is aimed at the practitioner or trainee in clinical EEG. The investigation of sleep and its disorders is a comparatively new medical discipline, with the greatest development having occurred in the last three

Multiple Sleep Latency Test MSLT Procedure MSLT Interpretation Maintenance of Wakefulness Test MWT Procedure MWT Interpretation Clinical Evaluation of Sleep Disorders Evaluation of Excessive Daytime Sleepiness Approach to Nocturnal Behavioral Events (Parasomnias) Approach to the Patient with Insomnia References

decades. The use of prolonged EEG recording to investigate sleep was pioneered by Kleitman and culminated in his historic discovery (with Aserinsky) of rapid-eye-movement (REM) sleep in 1953 (1). In the 1960s, early investigations described sleep-onset REM periods in narcolepsy (2,3), and defined sleep apnea (4,5). Important progress in standardizing sleep investigation occurred in 1968 when a committee of sleep researchers published a manual for scoring sleep; this system, known by the editors’ names Rechtschaffen and Kales (R&K), became the accepted standard as the basis for the scoring of sleep stages today (6). The American Sleep Disorders Association

599

600

Sleep Disorders: Laboratory Evaluation

(ASDA) was formed in 1976 to enhance patient care and standardize the practice of sleep disorders medicine. This group of clinical sleep specialists created an extensive classification system for the diagnosis of sleep and arousal disorders, which was initially published in 1979 (7) and has been serially revised (8). The ASDA (which subsequently changed its name to the American Academy of Sleep Medicine, AASM) also spearheaded the development of a free-standing certification board in sleep medicine, known as the American Board of Sleep Medicine. More recently in 2007, sleep medicine became a subspecialty board exam under the direction of the American Committee on Graduate Medical Education (ACGME). Eligibility requires a 1-year training fellowship in sleep in an ACGME-approved fellowship. That same year the AASM published the scoring manual that has markedly improved the standardization of polysomnography and its interpre­ tation (9). An update to that scoring manual has recently been published by that same organization (10). Clinical investigation of sleep disorders relies primarily on two major techniques: the overnight polysomnogram (PSG) and the multiple sleep latency test (MSLT). The overnight PSG allows the evaluation of nocturnal sleep and focuses primarily on identification and characterization of respiratory abnormalities in sleep. PSG is also useful in the evaluation of movement disorders, parasomnias, and nocturnal sleep disruption. The MSLT offers (a) an objective assessment of daytime sleepiness, and (b) a determination of possible early-onset REM sleep. These two study techniques serve as the “gold standard” of sleep evaluation and will be reviewed in detail. The maintenance of wakefulness test (MWT), an adaption of the MSLT, has also begun to be used more, particularly in medicolegal situations, where the documentation of normal alertness is required for employment in a patient with an identified sleep disorder. A brief overview of the initial assessment of patients presenting with complaints of excessive daytime sleepiness, unusual nocturnal behavior, or insomnia will also be presented.

POLYSOMNOGRAPHY: TECHNICAL ASPECTS Overnight sleep studies are usually performed in a facility dedicated to sleep investigations. Most sleep laboratories have one or more bedrooms in which environmental noise, ambient temperature, lighting, and decor are controlled, thus facilitating the patient’s sleep. The technologist and polygraphic equipment are located in a separate room to minimize any disruption to the patient’s sleep. An intercom is used to monitor and communicate with the patient as needed. Most laboratories also videotape the night’s sleep, so

behavioral or respiratory events can be reviewed. The machines used are similar to the commonly used digital EEG machines, but also have the ability to transfer other digital information (e.g., Sao2, CPAP pressure) to the recording. The patient is scheduled to arrive for the nocturnal study 60 to 90 minutes before his usual bedtime. During electrode application, the technologist reviews the patient’s history and offers an explanation of the study procedure. Medical information from the referring physician should be available at the time of the study to assure that the correct procedure is performed, to assist in decision-making during the study, and to complement the subsequent interpretation. Patients are usually asked to complete sleep questionnaires and Epworth Sleepiness Scale (ESS), which are then available to the interpreting physician to assist in assessing the clinical significance of the PSG findings. The usual study duration is 6 to 8 hours, depending on the specific clinical problem. When the patient awakens in the morning, the technologist obtains the patient’s impression of the night’s sleep and how it may have varied from his usual night’s rest. This information may be important in the clinical correlation of the sleep study results. Sleep studies are visually scored by the sleep technologist. The sleep study tracing and scored data are then reviewed by a polysomnographer, who provides the interpretation and clinical correlation. Digital EEG technology has proven to be an extremely useful tool for PSG studies. It has simplified the handling and storage of such large amounts of physiologic data. The ability to expand the number of channels also allows a greater sampling of EEG, which can then be reviewed at “routine EEG paper speed” (30 mm per second) to allow better assessment of a possible epileptic event that can be very difficult to interpret when EEG recordings were available only at PSG paper speed (10 mm per second) (11). The digitalization of the information lends itself to computer analysis. However, at this time, it is still the standard to do visual scoring of sleep stages according to the AASM Scoring Manual, given the unreliability of most sleep staging programs (9,12). Improvement in computer-aided sleep scoring may make sleep study scoring a less laborious task in the future. Computer analysis also has a potential to go beyond the standard (and somewhat arbitrary) epoch-by-epoch scoring of sleep and capture information about the microstructure of sleep that may have important clinical or research implications. Minimal polygraphic requirements to score sleep adequately include two channels of EEG, one channel for the electrooculogram (EOG) and one channel for the submental electromyogram (EMG). Routine PSG utilizes additional channels to assess respiration, leg movements, oxygenation, and

Sleep Disorders: Laboratory Evaluation

601

cardiac rhythm. The following is a brief review of “standard” technology involved in PSG recordings. Other reviews of PSG technique may also be helpful to the reader (13,14).

EEG Sleep Recording As with routine EEG, nonpolarizable silver-silver chloride or gold electrodes are standard and are attached with collodion to maintain adequate contact for the 6 to 10 hours required for an overnight study. The international 10–20 system is used for electrode placement (15), but a much more limited EEG montage is selected as outlined later. Early sleep investigators were limited by channel availability and could commit only one channel to EEG recording (central EEG lead to contralateral ear, e.g., C4-A1). As a result, the scoring manual of Rechtschafen and Kales was based on that single lead derivation. However, the poor sampling of waking alpha activity by this montage led most investigators to add at least one additional channel (occipital lead to more anterior reference, e.g., O1-A2) for better separation of wakefulness from stage 1 (drowsiness). The recent AASM Scoring Manual recommended a minimum of three EEG channels: F4-M1, C4-M1, O2-M1 (9). The use of the F3 or F4 electrode recognizes that sleep spindles and vertex waves are often seen more prominently in a frontal derivation. In addition, many laboratories commit at least two additional channels to EEG. A transverse vertex montage (e.g., T3-Cz, Cz-T4) is often preferred because it is sensitive to the identification of sleep spindles and vertex waves. Bilateral central, occipital, and ear electrodes are routinely placed so that in the event of electrode failure, the homologous contralateral derivation can be used without disturbing the patient. Occasional patients, such as those with possible nocturnal seizures, require a more extensive EEG montage for adequate assessment, usually a minimum of 16 channels (11). Routine “paper speed” or recording display for the PSG is 10 mm per second (6). This is initially disconcerting to the EEGer trained using a display of 30 mm per second. The slower paper speed was chosen to reduce paper use and to improve visualization of events (e.g., apneas) that occur over a relatively long period of time. Displaying the PSG using “slow paper speeds” still allows clear visualization of alpha rhythm and sleep spindles. In fact, the characteristic appearance of sleep spindles, “saw tooth” waves, and eye movements is more easily recognized at slower paper speeds once the interpreter has adjusted to the altered appearance (Fig. 18.1). EEG filter settings are low-frequency filter (LFF) of 0.3 Hz and high-frequency filter (HFF) of 35 Hz. The extended low-frequency

Figure 18.1: EEG activity characteristic for each of the sleep stages. The activity is recorded from the usual sleep study derivation (C4-A1) and displayed at 10 mm per second. Delta sleep is an alternate name for what is now termed stage N3. (From Hari P. The sleep disorders. Kalamazoo, MI: Upjohn Co., 1982:5–62.)

602

Sleep Disorders: Laboratory Evaluation

band is important to assess the prominent slow wave activity seen during deeper stages of NREM sleep. Amplifier sensitivity is usually 5 to 10 μV per mm, with 7 μV per mm used most commonly. Because voltage of the delta range activity is important in determining sleep stage, it is imperative that any change in EEG channel sensitivity be clearly “flagged” to avoid sleep scoring errors.

Electrooculogram The EOG is obtained primarily to identify phasic bursts of REMs that constitute the cardinal sign of REM sleep. In addition, these electrodes allow identification of slow lateral eye movements, which are often the first and most dependable manifestation of drowsiness (16). Gold-plated or silversilver chloride EEG cup electrodes are used to record EOG, but collodion should be avoided because of possible corneal injury. EEG electrolyte paste with additional sticky tape over the electrodes offers adequate attachment and avoids the risk of eye injury. The most common montage to record eye movements (and as recommended by the AASM manual) is a referential recording with the first channel recording from an electrode 1 cm lateral

and 1 cm superior to the outer canthus and referred to the ipsilateral ear. A second channel records activity from a location 1 cm lateral and 1 cm inferior to the contralateral outer canthus referred to the ipsilateral ear. This two-channel derivation shows eye movements as out-of-phase potentials, thus increasing the interpreter’s confidence in correctly identifying REMs (Figs. 18.2 and 18.3). Sensitivity and filter settings for the EOG are similar to those used for EEG recording. REMs are defined as conjugate, irregular, sharply peaked eye movements with an initial deflection of less than 500 milliseconds (9). REMs are characteristic of stage R sleep (stage REM) but are also seen in wakefulness, particularly with eyes open. Most commonly, the sharp slope of a REM is shorter than 300 milliseconds. Slow eye movements (SEMs) are conjugate, somewhat regular, sinusoidal eye movements with an initial deflection lasting greater than 500 milliseconds (9). The sinusoidal SEMs usually have a frequency of 0.25 to 0.5 Hz compared with REMs that are usually very irregular and commonly have a frequency greater than 1 Hz. Obviously, the rate of return to baseline of any EOG waveform is dependent on the alternating current (AC) filter setting and the superimposition of any additional

Figure 18.2: Slow eye movement (SEMs). Note the slow rolling eye movements appearing as out-of-phase activity in EOG channels. SEMs accompany the appearance of mixed-frequency theta activity typical for stage 1 sleep. K complex near end of epoch denotes onset of Stage 2. K complex appears as in-phase activity in EOG channels, reflecting its cerebral origin. (From Rechtschaffen A, Kales A, eds. A manual of standardized terminology, techniques, and scoring system for sleep stages of human subjects. Los Angeles, CA: UCLA Brain Information Service/Brain Research Institute, 1968.)

Sleep Disorders: Laboratory Evaluation

603

Figure 18.3: REM sleep with relatively lowvoltage, mixed-frequency EEG, REMs, and tonic EMG at lowest level during sleep. Note series of typical sawtooth waves at onset of REM activity. (From Rechtschaffen A, Kales A, eds. A manual of standardized terminology, techniques, and sco­ ring system for sleep stages of human ­subjects. Los Angeles, CA: UCLA Brain Information S ­ ervice/ Brain Research Institute, 1968.)

eye movements. Thus, the abrupt slope of the EOG waveform is the most ­reliable differentiating feature.

Axial Electromyogram Submental (chin) EMG activity reflects axial muscle tone and is used as one criterion for identifying REM sleep and movement arousals. Submental EMG is recorded by placing regular EEG electrodes over the mylohyoid muscle. One electrode is placed 1 cm above the tip of the jaw, and the second electrode is placed 2 cm below the inferior edge of the mandible and 2 cm off of the midline. This second electrode should be placed bilaterally, functioning as a backup electrode from which to select the most reliable EMG recording. Tonic EMG activity from axial musculature gradually decreases from wakefulness through stages N1-N3 of sleep and is usually entirely absent during REM sleep. Sensitivity of the submental EEG channel should be adjusted during drowsiness to reflect moderate activity. Subsequent

adjustments should be avoided during the night to permit comparison of EMG activity during different portions of the record. Typical sensitivity of the EMG is 2 μV per mm, with filters set at LFF = 10 and HFF = 100.

Anterior Tibialis EMG Anterior tibialis EMG activity is monitored to detect periodic leg movements (17,18). Regular EEG electrodes are placed over the anterior tibialis muscle bilaterally. The anterior tibialis muscle can be easily identified on the anterior lower leg by having the patient dorsiflex his foot against resistance. Two electrodes 3 cm apart are placed over each anterior tibialis. A bipolar recording from each anterior tibialis is usually obtained. Sensitivity and filter settings are similar to those described for submental EMG recording. A semi-standardized baseline is obtained before the study by asking the patient to dorsiflex a foot gently. Figure 18.4 is an example of periodic leg movements of sleep recorded from the anterior tibialis EMG electrodes.

604

Sleep Disorders: Laboratory Evaluation

Figure 18.4: Periodic limb movements of sleep (PLMS). The anterior tibialis EMG recording demonstrates bursts of periodic EMG activity associated with EEG arousals. (From Hari P. The sleep disorders. Kalamazoo, MI: Upjohn Co., 1982:5–62.)

Respiratory Monitoring Inasmuch as the suspicion of sleep apnea is the most common indication for overnight PSG, respiratory monitoring is arguably the most important technical aspect. Although measurements of airflow and ventilatory effort are technically the most difficult of all polysomnographic variables, they are essential to the adequate assessment of breathing during sleep. Both airflow and a measure of ventilatory effort must be recorded in order to distinguish among central, mixed, and obstructive apneic events (Fig. 18.5). In a central (or nonobstructive) apnea, the absence of respiratory drive causes all mechanical efforts to cease, and no airflow occurs at the nose or mouth. In an obstructive apnea, ventilatory efforts continue but no airflow occurs; this is due to occlusion of the airway. A mixed apnea begins as a central apnea (without ventilatory effort or airflow), but then the picture of obstructive apnea develops (ventilatory effort without accompanying airflow). Apneas are arbitrarily defined as cessation of airflow for 10 seconds or longer (19). In addition, hypopneas (decreased airflow) are also clinically significant. Specific scoring criteria for respiratory events are discussed in the section on polysomnographic interpretation. All measures of airflow or ventilatory effort use a LFF of 0.1 Hz and a HFF of 15 Hz.

Airflow Monitoring Thermistors/Thermocouples Thermistors or thermocouples are the least expensive and most commonly used method of monitoring airflow (13,19). Thermistors are small glass beads or wires whose electrical resistance changes as a function of temperature. When powered by a 1.5-V battery, the voltage drop across the thermistor varies with temperature. Expired air warms the thermistor and produces a signal that can be displayed on a polygraph. A thermocouple consists of two dissimilar metals in electrical contact, which produces a low-voltage signal that varies with the temperature change caused by expired air. Small, lightweight thermistors (or thermocouples) are taped to the face under each nostril and in front of the mouth. It is crucial to monitor both nostrils and mouth because the airflow path frequently changes during the night as a function of patient position. Activity from the thermistors can be summed and displayed in a single channel. Patient movement may disturb the position of the thermistors and result in loss of signal. Thus, careful attention needs to be paid to thermistor position throughout the night. Thermistors and thermocouples are less sensitive to airflow changes than the nasal pressure transducers discussed later. Therefore, hypopneas (reflecting

605

Sleep Disorders: Laboratory Evaluation

a decrease in airflow) are scored from the nasal pressure transducer recording, while apneas are scored off the thermistor/thermocouple channel.

Nasal Pressure Transducer

Glottic Pressure

In the last decade, nasal pressure transducers (NPTs) have become a standard technique in monitoring airflow during polysomnography (9,20,21). Nasal cannula pressure systems generate a respiratory waveform signal by detecting the fluctuations in pressure caused by inspiration and expiration. Unlike the inspiratory and expiratory fluctuations recorded due to temperature changes from a thermocouple or thermistor, these NPT signals are truly proportional to flow and are able to note changes in airway resistance and more modest changes in airflow (Fig. 18.6). Consequently, airflow

Normal Contour

15

20

25

Flow Limitation

No Flow Limitation

Driving Pressure

Figure 18.5: Examples of obstructive, central, and mixed apneas. Compare airflow (measured by CO2 analysis) and ventilatory effort (measured by intraesophageal balloon). (From Hari P. The sleep disorders. Kalamazoo, MI: Upjohn Co., 1982:5–62.)

10

5

Inspiratory Flow

Inspiratory Flow

0

Flattened Contour

Inspiratory Flow

Cannula Flow

Intermediate

Driving Pressure

Driving Pressure

Figure 18.6: Pressure/flow relationships from nasal cannula/pressure transducer during a single respiratory event. The x-axis shows time in seconds. Breaths with normal, intermediate, and flattened flow contours are labeled and a plot of the driving pressure/flow relationship is shown. The flattened flow/time contour shows a nonlinear flow/pressure relationship characteristic of flow limitation. (From Hosselet JJ, Normal RG, Ayappa T, et al. Detection of flow limitations with a nasal cannula/pressure transducer system. Am J Respir Crit Care Med 1998;157:1461–1467.)

606

Sleep Disorders: Laboratory Evaluation

as measured by NPT is far more sensitive and accurate in the capture of hypopneas and “Respiratory Effort Related Arousals” (RERAs) (22). This increased sensitivity is routinely seen on PSG recordings and has led to the AASM recommendation to use NPT in tandem with thermistors or thermocouples for monitoring of airflow during a PSG (9).

CO2 Detectors Measuring the carbon dioxide (CO2) of expired air can serve as an alternative method of monitoring airflow (19,23). It is based on the principle that exhaled air contains 6% to 7% carbon dioxide, whereas inhaled air contains negligible amounts of CO2. In order to sample CO2 content, a cannula is inserted just inside the nostril and is connected to an infrared or mass spectroscopy analyzer. For routine clinical PSG, the small increase in sensitivity for detecting air movement using CO2 detectors does not usually warrant the added expense and technical demands. However, end-tidal Pco2 (Petco2) monitoring is an essential component of assessing sleep-disordered breathing (SDB) in children. The clinical importance of Pco2 monitoring in pediatric sleep apnea has led to its wide application in pediatric sleep laboratories. The Petco2 is prone to providing falsely low values in patients who are mouth breathers or who have a high degree of nasal obstruction. The use of supplemental O2 or CPAP also makes the recording inaccurate. Endtidal CO2 is often supplemented by transcutaneous Pco2 monitoring. However, the transcutaneous method provides only a semiquantitative measure of alveolar ventilation and varies significantly from measured Paco2 (23).

Pneumotachography Pneumotachography is the only method that allows direct quantification of ventilation during sleep. Pneumotachography uses a technology very similar to the NPT devices discussed earlier. However, the technique involves an uncomfortable tight-fitting mask and a flow-to-pressure transducer that offers considerable resistance to respiration. While technically superior to the other methods of monitoring airflow, the sleep disruption and increased respiratory resistance preclude the routine use of pneumotachography in clinical PSG (24).

Ventilatory Effort Monitoring For many years, there was a great deal of variability in the use different techniques to monitor respiratory effort. Early techniques included intercostal EMG and strain gauges. More recently developed techniques included inductive plethysmography (IP) and piezoelectric belts. Esophageal

manometry was available as well, but limited in its use due to its invasive nature and potential impact on sleep. In the 2007 AASM Scoring Manual, the recommendation was for use of calibrated or uncalibrated IP or esophageal manometry for the detection of respiratory effort (9). This has led to IP as the primary technique for monitoring respiratory effort. However, in the most recent update to the scoring manual in 2012, new information led to the piezoelectric belts being listed as an acceptable method of monitoring respiratory effort as well (10,25).

Inductive Plethysmography IP utilizes a conductive wire that is sewn into an elastic band or mesh that encircles the chest or abdomen. With expansion of the body wall, the circle of wire enlarges and the inductance of the loop is changed. When an AC current is applied to this wire, the variable inductance can be displayed on an AC or DC polygraph channel. This technique is much more powerful than the early techniques utilized inasmuch as it provides a quantitative measure of airflow. This method also permits more accurate determination of lung expansion regardless of patient position (26,27). The instrument is carefully calibrated so that the sum of the chest and abdominal signals will be proportional to the volume of airflow. If the airway is obstructed, no air will be exchanged; any change in chest volume will be associated with an opposite change in abdominal volume (paradoxical respiration), and the sum of the two will be zero (representing no airflow) (Fig. 18.7). Theoretically, complete ventilatory monitoring can be accomplished with two bands around the chest. The three-output channels are rib-cage movement, abdominal movement, and total volume. Obstructive apneas are recorded as continued respiratory movement of both thorax and abdomen, but with no significant change in total volume signal (Fig. 18.8). In central apneas, all three signals are suppressed. Mixed apneas, as expected, show a central pattern followed by an obstructive picture. Hypopneas may also be quantified by this technique. Theoretically, an independent measure of airflow (thermistor) is unnecessary. However, because of difficulties in calibration (especially in obese patients) and slippage of coils, this technique usually requires an additional monitor of airflow. Although it increases cost and is more technically demanding, IP offers significant advantages over other methods of monitoring respiratory effort.

Esophageal Pressure Monitors An esophageal pressure monitor (EPM) can be either a catheter-tip or ­balloon transducer that is placed through the nose into the distal esophagus  (28).

Sleep Disorders: Laboratory Evaluation

607

Figure 18.7: Obstructive apnea. Note the development of out-of-phase or paradoxical movement of thoracic and abdominal monitors during period of obstructive apnea. (From Mendelson WB. Human sleep. New York, NY: Plenum Press, 1987:159.)

The output signal of the pressure/voltage transducer shows absolute pressure (with a DC amplifier) or variations in pressure (with an AC amplifier) on the polygraphic record. EPMs measure pleural pressure swings and are the most accurate devices for assessing ventilatory effort. However, the invasive placement and the inability of many patients to tolerate their use make EPMs less than ideal for routine PSG. However, in difficult cases, intraesophageal pressure measurement can accurately resolve ambiguities as to the nature of the apnea.

Piezoelectric Belts This device incorporates a piezo crystal element transducer in an elastic belt. When stretched, the piezo crystal element provides a voltage signal proportional to the stress applied. It requires no power and connects simply to the

electrode board or input cable. It incorporates good sensitivity to ventilatory effort along with simple and reliable technical application (25).

Intercostal EMG Intercostal EMG can be recorded using surface electrodes placed in the intercostal space. The technique identifies thoracic ventilatory movements, but it is frequently inadequate in obese patients. While sometimes useful as a complementary technique, intercostal EMG alone is not an adequate index of ventilatory effort.

Arterial Oxygen Assessment Determination of arterial oxygen desaturation is an important adjunct in assessing the severity of apneic episodes (29–31). Finger pulse oximetry is

608

Sleep Disorders: Laboratory Evaluation

the technique that is most commonly utilized. It measures oxygen saturation (Sao2) by light transmission through the fingernail bed. Sao2 values obtained from pulse oximetry are continuously recorded on the record. The accuracy of pulse oximetry is compromised by the presence of carbon monoxide, hyperbilirubinemia, O2 saturations below 50%, dark skin pigmentation, or significant hypotension. The AASM guideline requires that the maximum acceptable signal averaging time is 3 seconds. It is important to note that the change in Sao2 associated with a respiratory event appears 20 to 40 seconds after the actual event. Incorporated in the delay are both circulation time to the finger and machine measurement delay.

Electrocardiogram

Figure 18.8: Recording demonstrating paradoxical motion of rib cage and abdominal signals along with absence of significant volume signal [VT(RC  ABD)] indicating decreased or absent airflow. Note accompanying oxygen desaturation. (From Cohn M. Respiratory monitoring during sleep: respiratory IP. In: Guilleminault C, ed. Sleeping and waking disorders: indications and techniques. Menlo Park, CA: Addison-Wesley, 1982:213–223.)

The ECG is monitored during sleep studies to detect arrhythmias that may be associated with SDB events. Documentation of an associated cardiac arrhythmia can affect treatment decisions in a patient with sleep apnea. In a PSG, the ECG is usually derived from two electrodes placed over the anterior chest wall. This avoids much of the movement artifact seen with the use of limb leads alone. The AASM guidelines recommend the use of a modified ECG Lead II during PSG recordings. This is achieved by placing one electrode in the second intercostal space between sternum and right shoulder and the second electrode in the sixth intercostal space above the left hip. In morbidly obese patients, movement artifact can sometimes be limited by placing this second electrode in the second intercostal space on the left, again halfway to the shoulder. This one channel recording is ­adequate to monitor heart rate, extrasystoles, and other arrhythmias. However, the differences in recording display, along with the limited derivation, will often not allow adequate assessment of P-wave and QRS abnormalities. If the precise rhythm ­disturbance cannot be determined from the PSG alone, then independent evaluation using an ambulatory ECG monitor or other dedicated ECG recording device is required. Routine filter settings for ECG are LFF = 0.3 and HFF = 70, with a sensitivity of approximately 50 μV per mm.

Recording Montage PSG montages are selected depending on the clinical question. For screening studies of patients with possible sleep apnea, PSG variables recorded should include EEG, EOG, axial EMG, anterior tibialis EMG, ECG,

Sleep Disorders: Laboratory Evaluation TABLE 18.1

Ambulatory PSG Recording

Screening PSG Montage

Channel   1. C4-A1   2. F4-A1   3. O2-A1   4. T3-Cz   5. Cz-T4   6. Left outer canthus—A1   7. Right outer canthus—A1   8. Submental EMG   9. ECG 10. Left anterior tibialis EMG 11. Right anterior tibialis EMG 12. Snore monitor 13. Nasal pressure transducer 14. Oronasal thermistor 15. Thoracic movement 16. Abdominal movement 17. Sao2 (oximetry)

Sensitivity (µV/mm) 7 7 7 7 7 7 7 2 70 3–7 3–7 — — — — — —

609

Filters (LFF/HFF) (Hz) 0.3/35 0.3/35 0.3/35 0.3/35 0.3/35 0.3/35 0.3/35 10/100 0.3/70 10/100 10/100 10/100 DC or 0.03/100 0.1/15 0.1/15 0.1/15 DC/70

LFF, low-frequency filter; HFF, high-frequency filter; EMG, electromyogram; Sao2, ­oxygen saturation.

airflow, ventilatory effort, and Sao2. A representative montage is outlined in Table 18.1. If there is a more specific clinical question, the montage can be adjusted accordingly. For example, if the primary concern is involuntary motor activity during sleep, additional channels recording arm and leg EMG can be added. To evaluate a patient for possible nocturnal seizures, additional EEG channels are needed (11). Present-day digital machines with expanded channel capabilities (e.g., 32 channels) allow greater flexibility in recording all the usual physiologic variables but also allow a choice of which variables to focus on in review. In general, assessment for nocturnal seizures is best performed using dedicated video-EEG monitoring designed for recording seizures. If sleep laboratory equipment is used for such a purpose, additional channels of EEG accompanied by simultaneous time-locked video are required and should be able to be displayed at 30 mm per second.

Historically, ambulatory recording of sleep in the home environment for the evaluation of sleep disorders has not been commonly utilized owing to the inherent limitations of the technique and the lack of third-party reimbursement (32). With the recent CMS (Centers for Medicare and Medicaid Services) approval for coverage and the push to control health-care costs, the role of ambulatory or home sleep tests (HSTs) is being reevaluated and is being used with increasing frequency (33). The various technologies available have been classified into four separate types of sleep studies dependent on what parameters are recorded and whether a technologist is in attendance. In-lab PSG with the recording of seven or more channels and attended by a technologist is the gold standard for the evaluation of sleep disorders and is classified as a type 1 study. A full but unattended PSG (seven or more channels, including the recording of EEG) is classified as a type 2 study. Type 3 studies have four to seven channels and focus on the evaluation of respiratory parameters. Type 3 studies do not record EEG and therefore do evaluate sleep stages, arousals, or sleep duration. Type 3 studies have been the focus in the increasing use of ambulatory HSTs. A type 4 study is a one- or two-channel overnight recording, with at least one of them being oximetry. Type 4 studies have not been demonstrated to be a reliable screen for sleep-related breathing disorders in any patient population (34). The usual routine for HSTs is for the patient to present to the lab the day of the study. The patient is educated regarding appropriate use of the equipment and placement of electrodes and recording devices. That night, after preparing for bed, the patient places the recording sensors and pushes a button as they get into bed. The button is pressed again upon awakening in the morning and a total recording time is determined. This is supplemented by a brief sleep log from the patient outlining any sustained periods of wakefulness or other important information that may impact on the clinical interpretation of the data. Recorded parameters usually include: nasal/ oral airflow, respiratory effort, Sao2, body position, snoring, and heart rate. The studies are scored by automated scoring systems, but the data should be reviewed and edited by a technologist to assure that artifactual information is discarded and that the results represent a reasonable interpretation of the data. From this recording, an estimate of an apnea-hypopnea index (AHI) and oxygen desaturation index is obtained. The recent guidelines from the AASM for unattended sleep studies outline minimal technical expectations for portable monitoring (35). HSTs should

610

Sleep Disorders: Laboratory Evaluation

at a minimum record airflow, respiratory effort, and blood oxygenation, and should use the same sensors that are used in the laboratory studies. Accurate data collection is paramount to the successful use of the techniques, so the device should not be used by technologists or physicians that are inexperienced in sleep disorders and their evaluation. The devices must also allow review of the raw data to allow manual scoring and/or adjustment of the automated scoring algorithms. The advantages of HSTs include lower cost and greater patient accessibility. Patients who cannot access in-lab studies due to lab backlog or physical immobility can be evaluated with a HST. Other patients with irregular or unusual sleep-wake cycles can also have an evaluation that does not have to fit into the usual lab schedule. The primary disadvantage is unreliable or lost data, which may require repeating the test. In addition, the lack of any assessment of sleep stage or quality may lead to a significant underestimation of apnea severity (due to inclusion of prolonged wakefulness) or more importantly, the lack of recognition of hypopneas if significant oxygen desaturations are not present (34). Because of the obvious limitations, portable or ambulatory sleep studies are not appropriate in many clinical settings. These limitations are recognized by the CMS guidelines as well as those of the AASM. Both bodies agree that the HST may be used as an alternative to in-lab PSG for the diagnosis of obstructive sleep apnea (OSA) in patients with a high pretest probability of moderate to severe OSA. If positive, the caregiver can move forward with therapy for the significant respiratory abnormality. Additionally, HSTs can be helpful after CPAP therapy has been initiated to assess adequacy of treatment if there is a concern regarding response to therapy (33–35). However, HST is not indicated for use in patients with significant comorbidity (e.g., chronic obstructive pulmonary disease, congestive heart failure, and neuromuscular disease) as the coexistence of these disorders may lead to a greater degree of inaccuracy. HST has not been evaluated in asymptomatic but high-risk individuals such as bariatric surgery patients or commercial truck drivers.

COMPUTERIZED PSG AND COMPUTER-ASSISTED SCORING Visual sleep scoring has been the gold standard for analyzing sleep, and all subsequent standardizations of sleep scoring have emphasized visual analysis of the PSG (6,9). Any attempt to utilize computerized sleep scoring

in clinical sleep studies has attempted to model itself after the visual sleep analysis. While extremely useful for quantifying easily identified events (e.g., oxygen desaturations), computerized scoring’s inability to adjust to hostile “pathologic states” such as frequent apneas, disrupted sleep, and movement artifacts has limited its clinical application. Although much of the commercially available PSG equipment does come with computerized analysis of sleep stage, the clinical standard remains visual scoring by a trained technologist or, at least, the review and overscoring of the computerized sleep staging by a technologist. Similarly, computerized identification of respiratory or movement events also requires review and editing by a technologist. During the creation of the AASM Scoring Manual, a critical evidencebased look at computerized analysis of PSG was carried out. The consensus of the review suggested that “computer scoring and quantitative analysis is still in the formative stage of development” and have not been proven to be useful in clinical practice (36). The real strength of computerized analysis of sleep is in its ability to look at sleep as a continuum, rather than the arbitrary definitions provided for identifying N1, N2, N3, and stage R sleep. Similarly, information regarding the microstructure of sleep is likely lost in the use of a 30-second epoch. It will be interesting to explore how computerized analysis may contribute to potential new methods for understanding sleep and its disorders (37).

POLYSOMNOGRAPHY: INTERPRETATION The clinical interpretation of polysomnographic sleep studies is based primarily on the analysis of three main variables: EEG-related variables (sleep stage, arousals); respiratory-related variables (apneas or hypopneas); and movement-related variables (periodic leg movements). This discussion will focus on the guidelines for the scoring of these variables and the subsequent assignment of clinical significance. The development of the AASM Scoring Manual (and its recent update) has markedly improved the standardization of PSG recording techniques and interpretation, which improves the ability to compare sleep study results obtained in different laboratories (9,10). Despite the improved standardization, a great deal of uncertainty persists as to the clinical significance of many of the observations noted during PSG recordings. While the International Classification of Sleep Disorders has labeled an AHI greater than 5 as abnormal, a large number of healthy, noncomplaining individuals will have significant elevations of their AHI. Similarly, a movement index of greater than 15 is listed as abnormal, but occurs

611

Sleep Disorders: Laboratory Evaluation

EEG-Related Variables

CHILDREN AWAKE REM SLEEP STAGES

in many normal subjects, and its clinical significance is unclear. Exactly where one should draw the line in attaching clinical significance to various PSG results remains uncertain. I will provide guidelines in this area based on my own standard of practice, but with the understanding that other experienced practitioners may have different opinions and that these standards are continually evolving.

1 2 3 4

Basic Sleep Scoring

1

2

3

4

5

6

7

3

4

5

6

7

3

4

5

6

7

YOUNG ADULTS AWAKE

SLEEP STAGES

REM 1 2 3 4

1

2

ELDERLY AWAKE REM SLEEP STAGES

In 1968, the publication of the Rechtschaffen and Kales sleep scoring atlas (6) represented the consensus agreement between sleep researchers of the time in an effort to standardize the scoring of sleep stages. It is truly remarkable that our most recent sleep scoring criteria (as published in the AASM Scoring Manual) vary only limitedly from this original document (9,10). The nomenclature of sleep stages has changed from those described in R&K Atlas. NREM stage 1 is now called stage N1, and NREM stage 2 is now stage N2. What had previously been NREM stages 3 and 4 (or slow wave sleep) is now unified into a single Stage N3. Stage REM is now labeled as stage R. Normal sleep has a clearly defined architecture that is relatively stable from childhood through senescence (38). Sleep onset begins with a transition from wakefulness to stage N1. Stage N1 is normally brief and is followed by stage N2. Stage N3 (slow wave sleep) follows and is usually sustained, especially in children and young adults. Sleep then briefly lightens to stage N2 before an initial brief REM period. This first period of stage R occurs approximately 90 minutes after sleep onset and completes the first sleep cycle. This complete cycle is then repeated three to five times during the night, but the amount of stage N3 diminishes during ensuing cycles, while the amount of stage R increases. Histograms are useful to visually display the ultradian cycle within a night’s sleep (Fig. 18.9) (39). Predictable changes in sleep architecture occur with age. Beginning in middle age, Stage N3 becomes less prominent, the number of awakenings increase, and sleep efficiency decreases. Published information on normal sleep can serve as an outline for normal values in PSG (39,40) (Table 18.2). Scoring is usually done on an epoch-by-epoch basis, with epoch lengths standardized at 30 seconds. Epochs are scored according to the AASM guidelines, and each epoch is scored as the stage that occupies greater than

1 2 3 4

1

2

HOURS OF SLEEP

Figure 18.9: Histograms representing normal sleep cycles for age. REM sleep (darkened area) occurs cyclically throughout the night at intervals of approximately 90 minutes. REM sleep shows little variation in the different age groups, whereas stage N3 (labeled here as stages 3 and 4 sleep) can be seen to decrease with age. Note also the frequent awakenings and increase in total wake time seen in the elderly. (From Kales A, Kales J. Sleep disorders: recent findings in the diagnosis and treatment of disturbed sleep. N Engl J Med 1974;290:487–499.)

612 TABLE 18.2

Sleep Disorders: Laboratory Evaluation Average Values for Healthy Adults of Different Ages Age (y) 21–30

41–50

61–70

Time in bed (TIB) (min)

Index

480

480

456

Total sleep time (TST) (min)

446

395

350

95 9 54 13 19

85 14 52 7 14

81 16 46 3 15

Sleep efficiency (TST/TIB, %) Stage N1 (%) Stage N2 (%) Stage N3 (%) REM (%) Arousal index

10.8 (±4.6)

16.5 (±5.6)

21.9 (±6.8)

REM, rapid eye movement. From Bonnet MH, Arand DA. EEG arousal norms by age. J Clin Sleep Med 2007;3(3): 271–274.

50% of that epoch (Fig. 18.1). The following is a summary of the rules of sleep scoring: Stage W: Stage W corresponds to the waking stage and is characterized by alpha activity or low-voltage mixed-frequency EEG activity. REMs, eye blinks, and tonic EMG activity are usually present. Stage N1: Stage N1 is scored when greater than 50% of an epoch is lowvoltage 2- to 7-Hz activity. Vertex waves ( 5) (46). This standard is derived from investigation of a small population of middle-aged control subjects free of sleep complaints. Recent studies, however, indicated that a significant percentage (approximately 40%) of normal elderly subjects over the age of 60 have sleep apnea when using that criterion (47–50). Most studies have not been able to identify any associated neuropsychologic or medical complications in the patients identified with mild apnea in these screening studies. Therefore, the rigid application of AHI greater than 5 as diagnostic of sleep apnea may lead to the overdiagnosis of the sleep apnea syndrome, particularly in an elderly population (47). A more reasonable criterion may be an AHI of 10 in middle-aged patients and a AHI of l5 in patients over the age of 60. In our experience, AHIs below this cutoff are uncommonly associated with daytime complaints. However, there are data in the literature, documenting that patients with an AHI greater than 15 to 20 do have a significant increase in subsequent long-term cardiovascular morbidity (51). Multiple factors must be analyzed in determining the severity of sleep apnea. Certainly, the AHI is of great importance, but apnea duration, degree of O2 desaturation, and associated cardiac dysrhythmias must be incorporated into the decision-making process. The positional nature needs to be assessed, as many patients will have severe apnea on their back, but demonstrate little or no apnea on their side. The exact factors that contribute to long-term morbidity or mortality of sleep apnea have not been identified. Surprisingly, the degree of sleepiness is not closely correlated with the AHI

618

Sleep Disorders: Laboratory Evaluation

Figure 18.16: Sleep histogram, CPAP pressure, apneas, hypopneas, and oxygen desaturation displayed for overnight PSG. Note prominent oxygen desaturations associated with recurrent hypopneas. More severe desaturations (arrows) are noted during REM sleep (darkened line). Marked improvement in respiratory events and oxygen desaturations noted with CPAP pressure of 8 cm H2O and above. Slightly higher CPAP pressures were required to control desaturations in REM sleep.

or degree of O2 desaturation. Rather, sleepiness is more closely correlated with the degree of sleep fragmentation, as reflected by an increase in awakenings, arousals, and stage Nl% with an accompanying decrease in stage N3% (52–54). In defining severity of sleep apnea, the AHI is the center of that assessment. The AASM task force recently published their categorization for OSA severity. The AASM defines an AHI of between 5 and 15 as representing mild sleep apnea, an AHI of 15 to 30 as being moderate sleep apnea, and an AHI of less than 30 as severe sleep apnea. The AASM does recognize that there are no current prospective studies to validate these criteria (19). My personal approach is to label an AHI of 5 to 10 as evidence of minimal sleep apnea, an AHI of 10 to 20 as representing mild sleep apnea, an AHI

of 20 to 40 as representing moderate sleep apnea, and an AHI greater than 40 as being severe sleep apnea.

Pediatric Sleep-Disordered Breathing The foregoing discussion focuses on evaluating adult patients. The diagnosis and sleep laboratory evaluation of pediatric patients for OSA is much more challenging. Clinically, the patients usually present with witnessed snoring and ­struggling nocturnal respiration. They are much less likely to be obese or to be sleepy during the day. The abnormalities identified on PSG in pediatric patients are much less profound and require a greater degree of clinical judgment.

Sleep Disorders: Laboratory Evaluation

In a study by Rosen and colleagues (55), the investigators examined ­ olysomnographic data from 20 children who had clinical evidence of p upper-airway obstruction during sleep (loud snoring and labored breathing) and who had accompanying O2 desaturations during sleep. These children were felt to confidently represent a clinical syndrome of OSA in children. Remarkably, the mean apnea index was only 2 in this population. This was despite the fact that they experienced an average of 175 O2 desaturations greater than 5% each night and with an average minimum Sao2 of 66%. The authors concluded from this study that episodes of complete airway obstruction are generally rare in children even in the setting of serious sleep-related upper-airway obstruction and that the adult criteria for OSA failed to identify the majority of children with serious upper-airway obstruction. The presence of any obstructive apneas lasting 10 seconds in a child should be met with suspicion of indicating significant airway compromise. Similarly, greater attention needs to be paid to any episodes of O2 desaturation. The polysomnographic evaluation of normal children also indicates that apneas are much less frequent in children than in adults. In 50 normal children studied by Marcus and colleagues (56), the mean apneic index was 0.1. The mean number of desaturations greater than 4% per hour of sleep was 0.3. Only 9 (18%) children had any obstructive apneas identified and only 2 (4%) children had apneas lasting 10 seconds. Only one child had more than two obstructive apneas in a night’s sleep. Out of these data, the authors recommended a “normal value” for an apnea index in a child is less than 1. In addition, the mean minimum Sao2 was 96% and only one child ever went below 90% O2 saturation. The authors, recommended value for the minimum Sao2 in children is greater than 92%. The recommended normal value for the number of desaturations greater than or equal to 4% per hour of sleep was less than l.4. A subsequent study of normal children yielded remarkably similar results and reinforced these observations (57). This second study by Uliel and colleagues recommended limits for normal values (based on ±2 standard deviations) as follows: OA index 1; CA index 0.9; minimum oxygen desaturation 89%; baseline Sao2 92%. Given the lack of overt apneas and the possible lack of sensitivity of O2 changes, additional markers of SDB have been sought in pediatric patients. The most useful measure identified to date is the end-tidal CO2. Several defined guidelines have been provided, but the most useful appears to be defining the persistence of a Petco2 equal to or greater than 45 mm Hg for greater than 10% of the TST (56,57). Investigators believe this value to reliably distinguish snoring from clinically significant obstructive

619

hypoventilation (55). The AASM Scoring Manual takes a more conservative approach of a Petco2 greater than 50% for greater than 25% of TST. The new Scoring Manual outlines scoring rules for pediatric patients. Most of these apply to patients under the age of 13, as post-pubescent children usually present in a fashion similar to adult patients. Technical specifications are similar to adult recordings except for the use of transcutaneous CO2 and end-tidal CO2. In pediatric patients, an obstructive apnea is scored if the event lasts at least the duration of two breaths in the baseline breathing and is associated with respiratory effort throughout the period. A central apnea is scored if inspiratory effort is absent throughout the period and is accompanied by one of the following: (a) the event lasts at least 20 seconds; (b) the event lasts the duration of at least two breaths and is associated with a 3% oxygen desaturation or an EEG arousal; (c) the event is associated with a decrease in heart rate to less than 50 beats per minute for at least 5 seconds. A mixed event is scored when the event lasts at least the duration of two breaths and has both a period of absent respiratory effort and a period of persistent respiratory effort. The rules for scoring a hypopnea in a pediatric patient are the same as in an adult, except the duration requirement is at least two breaths. Hypoventilation is scored when greater than 25% of TST has a Pco2 greater than 50 mm Hg (as recorded by arterial Pco2, transcutaneous Pco2, or Petco2). The indications for PSG evaluation in a pediatric population have evolved during recent years and were recently summarized in a practice parameter published in 2011 by the AASM (58). The main indication, as in adults, is the evaluation of possible OSA. PSG may be appropriate before adenotonsillectomy (AT) if moderate to severe OSA is suspected owing to a greater degree of perioperative complications in this population. PSG evaluation of residual OSA after AT is indicated if moderate to severe OSA was present preoperatively. PSG is also indicated for evaluation of possible central alveolar hypoventilation syndrome or sleep-related hypoventilation due to neuromuscular disorders or chest wall deformities.

Movement-Related Variables Periodic limb movements in sleep (PLMS) are repetitive, stereotyped movements of the lower extremities that occur during sleep (59–61). They were previously called nocturnal myoclonus, but this term has been replaced by PLMS because the movements are not truly myoclonic and also because there might be confusion with other true myoclonic events that can occur during sleep. PLMS are characterized by tonic extension of the great toe,

620

Sleep Disorders: Laboratory Evaluation

with occasional superimposed clonic activity, variably accompanied by ankle dorsiflexion and knee flexion. Lower-extremity movements are scored as PLMS when anterior tibialis EMG activity lasts 0.5 to –10 seconds and at least four movements occur in a cluster with an intermovement interval of 5 to 90 seconds. The most common intermovement interval is 20 to 40 seconds (see Fig. 18.4). The minimum amplitude for the EMG increase is defined as 8 μV. Movement-associated arousals (also termed PLMS-associated arousals, PLMSAr) are routinely scored with each PLMS if there is no associated respiratory dysrhythmia and if the EEG arousal follows the movement within a few seconds. Oftentimes, EEG arousal will vary in its relation to anterior tibialis EMG activity. It may occur just before, synchronous with, or just after EMG activity (62). The Scoring Manual defines an arousal and a PLMS as associated with each other when there is less than 0.5 second between the end of one event and the onset of the other, regardless of which is first. Obviously, care needs to be taken to be sure that no other cause (e.g., recurrent hypopneas) exists that may underlie these recurrent arousals. While arousals that immediately precede the leg movements may not seem to be “movement-associated,” Lugaresi and colleagues (60) have postulated an internal “pacemaker” that gives rise to both the arousal and the leg movement. As such, the leg movement and EEG arousal often occur in a mildly asynchronous and variable fashion. Scored PLMS events are counted and divided by hours of sleep to yield a PLMS index (PLMSI). The number of PLMS-associated arousals is also divided by the hours of sleep to yield a PLMS arousal index (PLMSArI) (59,63). The basis for labeling an individual as suffering from Periodic Limb Movement Disorder (PLMD) was originally determined to be five movements per hour of sleep (18). However, it soon became apparent that there was a large population of asymptomatic individuals who had more than five movements per hour (59). Bixler et al. (64) demonstrated an 11% incidence of PLMS (using a PLSI > 5) in normal subjects and noted a marked increase in incidence with age. However, none of these subjects had more than five arousals per hour related to these leg movements. Other studies have shown that in patients with hypersomnolence, the degree of sleep disruption is correlated with the severity of excessive daytime sleepiness (46,65), although this has not been a consistent observation. The present definition of PLMD requires a PLMS index of greater than 15 (>5 in children), a complaint of disturbed sleep or daytime fatigue, and the absence of a potential alternative etiology (e.g., OSA). The PLMSArI

does not play a role in the diagnosis. It is important to emphasize that PLMS are present in a broad range of sleep disorders, and their pathophysiology and exact clinical significance remain poorly understood (17,60).

MULTIPLE SLEEP LATENCY TEST The MSLT is a multiple-nap trial designed to quantify the patient’s ­sleepiness and assess the presence of sleep-onset REM periods (SOREM) (66–69). It serves as the “gold standard” for the assessment of excessive daytime ­ sleepiness (EDS), but clearly has significant limitations. Other ­measures of sleepiness (pupillometry [70], ESS [71,72]) are either too technically ­cumbersome or too subjective to warrant widespread clinical use. The MSLT provides an objective quantification of “sleepiness” and is useful in the clinical determination of pathologic hypersomnolence. Opportunities to nap are given at 2-hour intervals across the day, thereby allowing the investigators to obtain a sampling of the diurnal variation of the patient’s sleep tendency.

MSLT Procedure The patient obtains a usual night’s rest prior to the study. Most investigators require that the preceding night’s sleep be documented by PSG to ensure adequate sleep and exclude sleep disruption (e.g., sleep apnea, PLMS) as a contributing etiology to the complaints of EDS. TST of less than 6 hours on the preceding overnight PSG may impact on subsequent mean sleep latency and effect clinical interpretation. Many labs require a sleep log to be kept the week prior to the MSLT to document preceding sleep–wake schedules. Stimulants and REM-suppressing medications should be stopped 2 weeks prior to the study. Urine drug screening the morning of the study may be appropriate to ensure that the sleepiness is not pharmacologically induced. Patients are asked to abstain from caffeine consumption on the day of the study. The patient arrives at the laboratory in time to allow the application of electrodes. At a minimum, electrodes are placed to monitor central and occipital EEG, submental EMG, EOG, and ECG. A sample montage is displayed in Table 18.3. The first nap trial is initiated 1.5 to 3 hours after the patient has awakened from nocturnal sleep. Four or five scheduled nap times are scheduled during the day (i.e., 9:30 a.m., 11:30 a.m., 1:30 p.m., 3:30 p.m., and 5:30 p.m.). At each trial, the patient lies down in his street clothes and assumes a comfortable sleep position. To standardize physical activity, a 15-minute quiet period immediately precedes each nap. The sleep

Sleep Disorders: Laboratory Evaluation TABLE 18.3

MSLT Montage

Channel

Sensitivity (µV/mm)

Filters (LFF/HFF) (Hz)

7 7 7 7 7 7 7 2 75

0.3/35 0.3/35 0.3/35 0.3/35 0.3/35 0.3/35 0.3/35 10/100 0.3/70

1. C4-A1 2. F4-A1 3. O2-A1 4. T3-Cz 5. Cz-T4 6. Left eye-A1 7. Right eye-A1 8. Submental electromyogram 9. Electrocardiogram

621

(awakening) is defined as two consecutive epochs of wakefulness after sleep onset. All scoring is done using the criteria of the AASM Scoring Manual. Each nap is reviewed for evidence of stage R sleep (Fig. 18.17). Stage R is scored as per AASM criteria and is determined by the first epoch scored as REM sleep. REM latency (latency to REM onset after sleep onset) is usually determined, but the most important observation is the presence or absence of REM during each nap trial. Each nap is terminated when: (a) 20 minutes after the trial started if no sleep has occurred, (b) after 15 minutes of continuous sleep as long as sleep-onset criteria are met before the end of 20 minutes, or (c) after 20 minutes if the patient awakens, even if less than 15 minutes of sleep has occurred. The patient is instructed to maintain wakefulness (and is observed if possible) between each nap period.

LFF, low-frequency filter; HFF, high-frequency filter.

MSLT Interpretation

room should be quiet, dark, and free of environmental noise. As the trial is about to begin, the technician instructs the patient to “close your eyes and attempt to sleep,” turns off the lights, exits from the room, and begins recording. The MSLT is routinely scored in 30-second epochs. Sleep onset is defined as the first 30-second epoch scored as stage 1 or deeper sleep. Sleep offset

The interpreter determines the latency to sleep onset and the presence or absence of REM sleep for each nap trial. Sleep onset is usually taken as the first 30-second epoch of stage 1 sleep or deeper sleep stage. A 30-­second epoch is scored as wakefulness if the majority of the 30-second period ­demonstrates a waking pattern. Stage N1 is determined if the majority of the 30-second epoch is stage N1 sleep. Thus, a brief return of alpha activity for 5 to 10 seconds does not result in an awakening, nor does it necessarily

Figure 18.17: Sleep-onset REM (SOREM) period. PSG record demonstrates the appearance of a low-voltage, mixed-frequency EEG, accompanied by the abrupt loss of tonic EMG activity and the appearance of REM activity. (From Hari P. The sleep disorders. Kalamazoo, MI: Upjohn Co., 1982:5–62.)

622

Sleep Disorders: Laboratory Evaluation

prevent the scoring of an epoch as sleep. In subjects who have an identified alpha rhythm, stage N1 is scored when greater than 50% of an epoch has loss of the alpha rhythm, which is replaced by low-amplitude, mixedfrequency activity. In subjects who do not generate an alpha rhythm, stage N1 is scored when any of the following occur: (a) 4 to 7 Hz activity is seen that is at least 1 Hz slower than waking background frequencies; (b) vertex sharp waves; (c) SEMs. These findings need to be present for over 50% of the epoch in question. Individual sleep latencies are then averaged to determine mean sleep latency (MSL). MSLT results in different populations are shown in Table 18.4. An MSL of less than 5 minutes is labeled indicative of pathologic hypersomnolence. It is a direct indication of the individual’s vulnerability to falling asleep in a low-stimulus situation and is associated with performance decrements and unintentional episodes of sleep. An MSL of greater than 10 minutes is “normal” and does not reflect significant sleepiness. An MSL between 5 and 10 minutes is the “gray zone” into which some normal and some hypersomnolent patients fall. There is no consensus as to the clinical significance that should be applied to results in this range. Labeling an MSL of greater than 5 but less than 10 minutes as suggestive but not diagnostic of pathologic hypersomnolence is probably the most reasonable approach. If a single cutoff value is utilized, some authors suggest 8 minutes as a rational dividing point between normal (>8 minutes) and abnormal (
Current Practice of Clinical Electroencephalography - John S. Ebersole

Related documents

2,847 Pages • 898,804 Words • PDF • 95.8 MB

132 Pages • 32,677 Words • PDF • 535.8 KB

7 Pages • 5,913 Words • PDF • 357.3 KB

382 Pages • 80,192 Words • PDF • 1.9 MB

9 Pages • 8,997 Words • PDF • 654.8 KB

1,239 Pages • 914,444 Words • PDF • 35.5 MB

643 Pages • 355,312 Words • PDF • 12 MB

765 Pages • 209,014 Words • PDF • 13.5 MB

192 Pages • 65,178 Words • PDF • 2.9 MB

591 Pages • 457,770 Words • PDF • 14.6 MB