You are on page 1of 1306


Basic Principles, Clinical Applications, and Related Fields

Basic Principles, Clinical Applications, and Related Fields


Donald L. Schomer, MA, MD

Professor of Neurology
Harvard Medical School
Director, Laboratory for Clinical Neurophysiology
Chief, Comprehensive Epilepsy Program
Beth Israel Deaconess Medical Center
Boston, Massachusetts

Fernando H. Lopes da Silva, MD, PhD

Emeritus Professor
Swammerdam Institute for Life Sciences
Center of Neurosciences
Faculty of Science
University of Amsterdam
Amsterdam, The Netherlands
Acquisitions Editor: Frances Destefano
Product Manager: Tom Gibbons
Vendor Manager: Bridgett Dougherty
Senior Manufacturing Manager: Benjamin Rivera
Marketing Manager: Brian Freiland
Design Coordinator: Stephen Druding
Production Service: MPS, a Macmillan Company


Two Commerce Square

2001 Market Street
Philadelphia, PA 19103 USA

Printed in China

All rights reserved. This book is protected by copyright. No part of this book may be reproduced in
any form 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.

Library of Congress Cataloging-in-Publication Data

Niedermeyer’s electroencephalography: basic principles, clinical applications, and related fields/

[edited by] Donald L. Schomer, Fernando H. Lopes da Silva.—6th ed.
p. ; cm.
Rev. ed. of: Electroencephalography/[edited by] Ernst Niedermeyer, Fernando H. Lopes da Silva.
5th ed. c2005.
Includes bibliographical references and index.
ISBN-13: 978-0-7817-8942-4 (hardback: alk. paper)
ISBN-10: 0-7817-8942-7 (hardback: alk. paper) 1.
Electroencephalography. I. Niedermeyer, Ernst, 1920– II. Schomer, Donald L. III. Lopes da Silva,
F. H., 1935– IV. Title: Electroencephalography.
[DNLM: 1. Electroencephalography. 2. Central Nervous System Diseases—diagnosis. WL 150]
RC386.6.E43N54 2011

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 the 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 the 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.
To purchase additional copies of this book, call our customer service department at (800)
638-3030 or fax orders to (301) 223-2320. International customers should call (301) 223-2300.
Visit Lippincott Williams & Wilkins on the Internet: at Lippincott Williams &
Wilkins customer service representatives are available from 8:30 am to 6 pm, EST.
10 9 8 7 6 5 4 3 2 1
Dedicated to Ernst Niedermeyer, M.D.
for all of the help and inspiration offered to us and
to countless clinical neurophysiologists worldwide
over his truly remarkable career.
Preface to the First Edition

The history of clinical electroencephalography (EEG) has just aspects of lesions and tissue changes, neurology would be
passed the 50-year mark. The age of the pioneers was followed shallow and barren without awareness of the constant fluctua-
by the stage of expansion. What began in a few prestigious cen- tion of functional states in the CNS. Another important sti-
ters gradually became a tool of all academic medical institu- mulus has been the establishment of standards of quality for
tions, and eventually of all major hospitals. In more recent electroencephalographers and EEG laboratories.
years, EEG even invaded the private offices of practicing neurol- In these times of challenge, a review of the state of affairs in
ogists and other specialists interested in central nervous system EEG seems to be appropriate. Such thoughts have prompted the
(CNS) disease. genesis of this large one-volume textbook, which, by its mere
From this perspective, the history of clinical EEG looks like size, sets itself apart from the group of smaller introductory
the “via triumphalis” of a buoyantly dynamic new subspecialty. textbooks and from the huge multivolume handbook. The one-
The elucidation of the electrophysiologic process underlying volume character of the book symbolizes the spirit of unity that
epileptic seizure disorders and a variety of other CNS dysfunc- should reign among clinical electroencephalographers, basic
tions was indeed a unique achievement made possible by the science researchers, and the workers in the field of compu -
new method. The original intention of the founder of clinical terized data analysis.
EEG, Hans Berger, had been the exploration of mental and psy- One author can hardly undertake such a task alone. For this
chological processes, and even in this domain the yield has been reason, we have reached out across the ocean for collaboration.
substantial. Moreover, electroencephalographers have not con- It became clear, however, that a two-man effort would not
fined themselves to the spontaneous wave patterns of the brain; suffice to cover the entire field in the relatively short working
forms of EEG data analysis with the aid of computers were period of 2 years. We solicited for assistance and found a
introduced in order to demonstrate evoked and event-related wonderful group of contributors in various special areas. Above
potentials and to investigate the wealth of frequencies that con- all, coverage of the fields of neurophysiology and neurophar-
stitute the EEG. In the search for the sources of EEG generation, macology has depended most heavily on the assistance of
the brain tissue became the target of exploration. prestigious specialists.
Depth electrodes became one of the most important tools of Attempts at synthesis have not been the goal of this book.
experimental neurophysiologists, who also investigated the sin- Instead the reader will find a more individualistic approach
gle neuron using microelectrodes. The implantation of depth from which the personal basic philosophy of each author can be
electrodes in the human brain has aided in the evaluation of derived. No effort has been made to achieve strict standardiza-
chronic epileptics considered candidates for seizure surgery. tion of symbols and terminology; as an example, frequencies
This impressive progress, however, has been counterbal- are described in various terms (10 cps, 10/sec, 10 Hz, etc.).
anced by signs of pessimism, fatigue, and resignation. A certain There is also some overlap between certain chapters; we feel
malaise has inched its way into the hearts of thousands of elec- that the reader will benefit from the presentation of a given
troencephalographers who have started to feel the grip of stag- topic as seen from two somewhat different viewpoints.
nation. Generation of EEG potentials has proved to be A piece of technical information might be worthwhile.
extremely complex and difficult to understand; the feeling of Unless the filter setting is specifically indicated in the illus-
doing pragmatically useful work with an ill-understood trations of EEG tracings, a time constant between 0.15 and
method has been depressing to many workers in the field. The 0.5 seconds was used (above 0.4 seconds when slow frequencies
pragmatists have further suffered from the limitations of EEG played a major role). The use of “muscle filters” was avoided.
as a method of localization of cerebral lesions. These feelings We have tried to combine didactic and academic elements in
have been nourished by the phenomenal achievements of non- this book. Hence, trainees as well as seasoned professionals in
invasive radionuclear and radiologic scanning methods; this the field will, we hope, find what they are searching for. This
progress of new methods in the field of structural diagnosis has dual approach does mean that some parts of the book require
been a matter of concern for many electroencephalographers. greater sophistication of the reader than do others.
A more real danger, perhaps, is presented by the poorly trained Acknowledgments for invaluable help in this undertaking
colleagues who are tarnishing the image of EEG. come from the depth of our hearts. Mr. Braxton Dallam
In reality, however, these challenges present a stimulus for Mitchell, President of Urban & Schwarzenberg in Baltimore,
the electrophysiologic field. The function-oriented aspects of Maryland, deserves the honor of having been the initiator of
neurologic sciences will always be of paramount significance. this book. His encouragement has been very much appreciated.
The loss of function-oriented neurology would foreshadow the Detlev Moos has coordinated the production of this book with
death of neurology. With all due respect for the structural care and efficiency; Suzanne Lohmeyer has copyedited and

Preface vii

indexed it well; and Nan Tyler, Carola Sautter, and Victoria physiology, Dr. A Earl Walker (presently of Albuquerque, New
Doherty at the publisher’s Baltimore office assisted with their Mexico), deserve special gratitude.
experience. In the Johns Hopkins Hospital in Baltimore, Gratitude is expressed (by F.L.S.) to professor Dr. W. Storm
Maryland, the technical staff of the laboratory deserves great von Leeuwen, Dr. G. Wieneke, and Dr. K. Van Hulten (Utrecht
praise: Mrs. Judy Nastalski, R. EEG T. and chief technologist, University Hospital), Mr. N.J.I. Mars (Twente University of
Mr. Eric DeShields, Mrs. Debbie Reichenbach, R. EEG T., Technology), and Mr. A. Van Rotterdam (Institute of Medical
Miss Sharon Vaughan, Mrs. Kathleen Daniecki, R. EEG T., Physics, Utrecht) for their advice and encouragement. The high
Mrs. Cindy Haywood, and Miss Kim Rimel. How deeply the professional competence of Mrs. Ada Van Schaik and Mr. Nico
electroencephalographer depends on the quality of the record- Haagen (Institute of Medical Physics, Utrecht) in their fields of
ings and the dedication of the technologists! Truly invaluable secretarial work and artwork was of invaluable help.
was the secretarial assistance of Mrs. Catherine Bonolis. The Heartfelt thanks are also expressed to the contributing authors
operation of the laboratory was further aided by the experience of the book; they have naturally become a part of this undertak-
of Mrs. Marie Simpson. Important contributions came from ing. The response of these splendid coworkers was exemplary.
Mr. Joe Dieter, who is responsible for the pictorial artwork, Joseph J. Tecce and Lynn Cattanach, the authors of the article on
Mr. Ron Garret (lettering of tracings), and Mr. Zuhair Kareem contingent negative variation, substituted for a colleague who
and Mrs. Lillian Reich, the staff of Medical Photography. had to step down from his obligations at the last minute. They
Assistance and advice in the clinical EEG field was given made possible the almost impossible when they declared their
most freely by Dr. Gisela Freund, visiting assistant professor at willingness to join the team of coworkers. To them, and to all the
the John Hopkins Hospital EEG Laboratory (1980/81), of the contributing authors of this volume, our deepest thanks.
Department of Clinical Neurophysiology, Free University
Berlin (Klinikum Westend). E.N.’s principal teacher in the field Ernst Niedermeyer
of EEG, Dr. John R. Knott (presently of Boston, Massachusetts), Fernando H. Lopes da Silva
and the great master of neurosurgery, epileptology, and neuro- Spring 1981
Preface to the Sixth Edition

We would like to express our most sincere thanks to Dr. Ernst this classic volume continue for the years to come and evolve
Niedermeyer for his leadership and direction over the past several with the field as we have seen Dr. Niedermeyer do so success-
years as we have moved forth to produce this sixth edition of the fully over the last 25 years as its senior editor.
textbook that he had first introduced back in 1982. Additionally, This edition has several new features, reflective of the changes
we would like to honor him as he is so richly deserving by renam- that have occurred in our field over the past 5 years since the fifth
ing this book Niedermeyer’s Electroencephalography: Basic edition. More and more, the field of digital recording has
Principles, Clinical Applications, and Related Fields. expanded; however, in order to understand some of the short-
Dr. Niedermeyer was born in Vienna, Austria, in 1920. He comings and pitfalls of digital EEG, people need to still address
attended medical school in Austria but, during his third year in the issues of basic analog recording principles. With an increased
medical school, he was witness to the rise of Nazism in use of digital recording, laboratories have collected new and dif-
Germany and Austria and was inducted into the German army ferent “technical artifacts.” We present here an attempt to start a
just prior to the onset of World War II. He was stationed in database for such artifacts in a hope that future editions will
Berlin and was then literally drafted into the medical service continue to expand upon this and offer a fairly complete library
even though he had not yet graduated from medical school. He for beginners interested in our field. As noted in the fifth edition,
said that he was basically treating wounded soldiers coming epilepsy monitoring units (EMUs) have continued to mush-
back from the eastern front during the first several years of the room. Similar growth has occurred in the use of EEG monitor-
German invasion of the Soviet Union. He remained there until ing in newborn, cardiac, trauma, and postoperative intensive
Nazi officials learned that his maternal grandmother was Jewish care units. With the significant advances in wireless communica-
which, according to German law, made him an “undesirable.” tion and easy access to the Internet, such recordings can also be
However, because of his medical background, he was not viewed and transmitted locally virtually instantaneously and can
interned but rather was sent to Normandy. He mentioned that allow well-trained clinical neurophysiologists to see and opine
he was, as were many of the soldiers in Normandy, well aware of about patients’ conditions on a very time-relevant basis.
the coming invasion by the Allied forces and he looked upon Hopefully, as future generations may show, this ability will sig-
the day of the invasion as one of his “happiest.” He said he was nificantly influence our patients’ outcomes. Similarly, the field of
wounded during the invasion and was taken as a POW by the intraoperative clinical neurophysiology for spinal cord function,
American forces taking part in the invasion. He was sent back as cranial nerve function, and cranial vascular therapies has con-
a POW to Kansas where he remained and served out the rest of tinued to evolve along with the wireless and Internet communi-
the war as the prison camp doctor. He returned to Austria fol- cations. This has allowed for close monitoring of neurologic
lowing cessation of fighting and finished his medical school and function during critical periods of operations, again with a time
stayed on in Austria, interested in and learning neurology/psy- course that allows for corrective actions to be taken on a mean-
chiatry. He was introduced into clinical neurophysiology ingful time frame.
through some of the historical works in neurophysiology that We have reorganized the text regarding normal EEG and
had been conducted in Europe prior to World War II, including epilepsy to more closely follow the normal aging patterns. We
the work of Hans Berger. He became a clinical neurophysiology have continued to present some of the more classical chapters
addict and was offered a training position at the University of that have evolved over years on evoked potentials and routine
Iowa with Dr. John Knott. He took that position and stayed electroencephalography, including EEG in common neuro-
with Dr. Knott for a number of years but was then offered a logic, metabolic, and heredito-degenerative diseases. We have
position to work with Dr. A. Earl Walker at the Johns Hopkins added and updated chapters and text regarding automated and
Hospital, Baltimore. He continued to do his clinical and basic specialized mathematics-based analysis techniques to try to
science research at Hopkins, where his career blossomed and he keep up with this rapidly expanding field. We will continue to
became well known in the field. He has been an inspiration to update these techniques in future editions. We have included a
many trainees and colleagues, both at Johns Hopkins and chapter on linking clinical neurophysiology to other investiga-
worldwide. Many of our colleagues have come to know him tive techniques such as functional MRI. We have updated the
through his kindness and his willingness to personally take an chapter on magnetoencephalography to, again, reflect the sig-
interest in their research and to make recommendations and nificant changes that have taken place, both from a technical
suggestions about directions for further study. His knowledge and a clinical perspective, in this field. The last several chapters
of the field is legendary and his personality is truly “old world” of this textbook attempt to present an overview of clinical neu-
in the best sense of the term. It is our most sincere desire that rophysiology research that intersects with many other aspects of

Preface ix

the realm of brain sciences such as consciousness and cognitive have offered tremendous help and support in the technical
processing. aspects of getting the chapters organized and together and for
We would like to encourage the readers of these words and Tom Gibbons from Lippincott and all those who have helped
this text to please let us know if they are aware of other areas of again in getting the textbook prepared for publication. Again, in
interest that could or should be involved in future editions of the words of Dr. Niedermeyer, “many, many thanks.”
this textbook that reflect either oversights on our part or are
harbingers of future development. Finally, we thank all of the Donald L. Schomer
authors and coauthors of the chapters of this book for their Fernando H. Lopes da Silva
tremendous effort in keeping this textbook relevant to our field.
Special thanks go to Fran Destefano and Franny Murphy who


Florin Amzica, Ph.D. Sudhansu Chokroverty, M.D., F.R.C.P.

Associate Professor of Stomatology Professor of Neuroscience
Montreal University Seton Hall University
Montreal, Canada South Orange, NJ
Clinical Professor of Neurology
Mary Repole Andriola, M.D.
Robert Wood Johnson Medical School
Professor of Neurology and Pediatrics
New Brunswick, NJ
Director, Divisions of Child Neurology, Clinical Neurophysiology
Professor and Co-chair of Neurology
and Pediatric Epilepsy
Program Director, Sleep Medicine and Clinical Neurophysiology
SUNY at Stony Brook
New Jersey Neuroscience Institute at JFK Medical Center
Stony Brook, NY
Edison, NJ
Helen Barkan, M.D., Ph.D.
Nathan E. Crone, M.D.
Assistant Professor of Neurology
Associate Professor of Neurology
Upstate Medical University Hospital
Johns Hopkins University
SUNY Upstate Medical University
Syracuse, NY
Johns Hopkins Hospital
Gerhard Bauer, M.D. Baltimore, MD
Professor of Neurology
Edgar dePeralta, M.D.
Medical University Innsbruck
Fellow in Neurology and Clinical Neurophysiology
Innsbruck, Austria
State University of New York Upstate
Richard Bauer, M.D., M.Sc. Syracuse, NY
Oberarzt, Department of Neurosurgery
Frank W. Drislane, M.D.
Medical University Innsbruck
Professor of Neurology
Innsbruck, Austria
Harvard Medical School
Robert L. Beach, M.D., Ph.D. Neurologist
Professor of Neurology Comprehensive Epilepsy Center
Upstate Medical University Beth Israel Deaconess Medical Center
Director, Epilepsy Program and EEG Laboratories Boston, MA
Upstate Medical University Hospital
Barbara Ann Dworetzky, M.D.
Syracuse, NY
Assistant Professor of Neurology
Steve Bild, R. EEG/ EP T., C.N.I.M. Harvard Medical School
Manager of Clinical Services, EEG/Evoked Potentials Lab Chief, Division of Epilepsy, EEG, and Sleep Neurology
Rush University Medical Center Brigham and Women’s Hospital
Chicago, IL Boston, MA
Gastone G. Celesia, M.D. Günter Edlinger, M.Sc., Ph.D.
Professor of Neurology (Retired) CEO
Loyola University of Chicago g.tec Medical Engineering GmbH
Chicago, IL Schiedlberg, Austria
Bernard S. Chang, M.D., M.M.Sc. Jonathan Charles Edwards, M.D.
Assistant Professor of Neurology Associate Professor of Neurosciences
Harvard Medical School Director, Comprehensive Epilepsy Center and Clinical
Comprehensive Epilepsy Center Neurophysiology Laboratories
Beth Israel Deaconess Medical Center Medical University of South Carolina
Boston, MA Charleston, SC
Keith H. Chiappa, M.D. Christian E. Elger, M.D., F.R.C.P.
Associate Professor of Neurology Professor of Epileptology
Harvard Medical School University of Bonn
Director, Electroencephalography Laboratory Head, Department of Epileptology
Massachusetts General Hospital University of Bonn Medical Centre
Boston, MA Bonn, Germany

Contributors xi

Ronald G. Emerson, M.D. Richard A. Hrachovy, M.D.

Professor of Clinical Neurology Professor of Neurology
Columbia University College of Physicians and Surgeons Baylor College of Medicine
Attending Neurologist Medical Director of Neurophysiology
New York Presbyterian Hospital St. Luke’s Episcopal Hospital
Columbia University Medical Center Houston, TX
New York, NY
Aatif M. Husain, M.D.
Charles M. Epstein, M.D. Department of Neurology
Professor of Neurology Duke University Medical Center and Neurodiagnostic Center
Emory University School of Medicine Veterans Affairs Medical Center
Emory Healthcare Durham, NC
Atlanta, GA
Kai Kaila, Ph.D.
Bruce J. Fisch, M.D. Professor of Biological and Environmental Sciences
Professor of Neurology Neuroscience Center
University of New Mexico University of Helsinki
Director, Neurodiagnostic Laboratories and MEG Center Helsinki, Finland
University of New Mexico Hospital
Stiliyan Kalitzin, M.D.
Albuquerque, NM
Image Sciences Institute
John N. Gaitanis, M.D. University Medical Center Utrecht
Assistant Professor of Neurology and Pediatrics, Clinical Utrecht, The Netherlands
Warren Alpert School of Medicine at Brown University
Anton Kamp
Director of Pediatric Epilepsy
Biological Centre
Rhode Island Hospital/Hasbro Children’s Hospital
University of Amsterdam
Providence, RI
Amsterdam, The Netherlands
Ali Gorji, M.D.
Andres M. Kanner, M.D.
Institute for Physiology
Professor of Neurological Sciences and Psychiatry
Münster University
Rush Medical College at Rush University
Münster, Germany
Director, Laboratories of EEG and Video-EEG-Telemetry
Jean Gotman, Ph.D. Associate Director, Epilepsy Section
Professor Rush University Medical Center
Montreal Neurological Institute Chicago, IL
McGill University
Peter W. Kaplan, M.B., F.R.C.P.
Montreal, Québec, Canada
Department of Neurology
Riitta Hari, M.D., Ph.D. Johns Hopkins Bayview Medical Center
Academy Professor Baltimore, MD
Brain Research Unit, Low Temperature Laboratory
Andrew D. Krystal, M.D., M.S.
Aalto School of Science and Technology
Professor of Psychiatry and Behavioral Sciences
Espoo, Finland
Director, Quantitative EEG Laboratory
Director, Insomnia and Sleep Research Laboratory
Department of Clinical Neurophysiology
Duke University School of Medicine
HUSLAB, Helsinki University Central Hospital
Durham, NC
Helsinki, Finland
Ekrem Kutluay, M.D.
Adam L. Hartman, M.D.
Associate Professor of Neurology
Assistant Professor of Neurology and Pediatrics
Medical College of Wisconsin
Johns Hopkins University School of Medicine
Milwaukee, WI
Attending Physician
Johns Hopkins Hospital Emma Laureta, M.D.
Baltimore, MD Assistant Professor of Neurology
Albert Einstein College of Medicine
Bin He
Department of Biomedical Engineering and Center for
Montefiore Medical Center
New York, NY
University of Minnesota
Minneapolis, MN Alan D. Legatt, M.D., Ph.D.
Professor of Neurology
Susan T. Herman, M.D.
Albert Einstein College of Medicine
Assistant Professor of Neurology
Director of Intraoperative Neurophysiology
Harvard Medical School
Montefiore Medical Center
Beth Israel Deaconess Medical Center
New York, NY
Boston, MA
xii Contributors

Ronald P. Lesser, M.D. Ernst Niedermeyer, M.D.

Professor of Neurology and Neurosurgery Professor Emeritus of Neurology and Neurological Surgery
Johns Hopkins University Johns Hopkins University School of Medicine
Johns Hopkins Medical Institutions Baltimore, MD
Baltimore, MD
Douglas R. Nordli, Jr., M.D.
Fernando H. Lopes da Silva, M.D., Ph.D. Professor of Pediatrics
Emeritus Professor Northwestern University Feinberg School of Medicine
Swammerdam Institute for Life Sciences, Center of Neurosciences Lorna S. and James P. Langdon Chair of Pediatric Epilepsy
Faculty of Science Children’s Memorial Hospital
University of Amsterdam Chicago, IL
Amsterdam, The Netherlands
Marc R. Nuwer, M.D., Ph.D.
François Mauguière, M.D. Professor of Neurology
Head, Department of Functional Neurology and Epileptology UCLA School of Medicine
Lyon University and Lyon Federative Institute of Neuroscience Department Head
Neurological Hospital Clinical Neurophysiology
Lyon, France Ronald Reagan UCLA Medical Center
Los Angeles, CA
Douglas Maus, M.D., Ph.D.
Assistant Professor of Neurology Trudy D. Pang, M.D., M.M.Sc.
SUNY Downstate Medical Center Instructor in Neurology
State University of New York at Brooklyn Harvard University
New York, NY Staff Neurologist
Beth Israel Deaconess Medical Center
Christoph M. Michel, M.D.
Boston, MA
Functional Brain Mapping Laboratory
Neurology Clinic and Department of Basic Neuroscience Alvaro Pascual-Leone, M.D., Ph.D.
University Hospital Professor of Neurology
University of Geneva Harvard Medical School
Geneva, Switzerland Director, Department of Neurology
Berenson-Allen Center for Noninvasive Brain Stimulation
Eli M. Mizrahi, M.D.
Beth Israel Deaconess Medical Center
Chair, Department of Neurology
Boston, MA
Professor of Neurology and Pediatrics
Peter Kellaway Section of Neurophysiology, Department of Neal S. Peachey, M.D., Ph.D.
Neurology Professor of Ophthalmology
Section of Pediatric Neurology, Department of Pediatrics Cleveland Clinic Lerner College of Medicine
Baylor College of Medicine Associate Chief of Staff for Research
Houston, TX Cleveland VA Medical Center
Cleveland, OH
Solomon L. Moshé, M.D.
Professor of Neurology Pediatrics, and Neuroscience Gert Pfurtscheller, Ph.D.
Albert Einstein College of Medicine Emeritus Professor
Director, Pediatric Neurology and Clinical Neurophysiology Laboratory of Brain–Computer Interfaces (BCI-Lab)
Montefiore Medical Center Institute for Knowledge Discovery
New York, NY Graz University of Technology
Graz, Austria
Janet M. Mullington, Ph.D.
Associate Professor of Neurology Rodney A. Radtke, M.D.
Harvard University Professor of Neurology
Beth Israel Deaconess Medical Center Duke University
Boston, MA Medical Director
Duke Hospital Sleep Disorder Center
Christa Neuper, Ph.D.
Duke University Hospital
Durham, NC
Institute of Psychology
University of Graz James J. Riviello, Jr., M.D.
Head of Institute George Peterkin Endowed Chair in Pediatrics
Laboratory of Brain–Computer Interfaces (BCI-Lab) Professor of Pediatrics and Neurology
Institute for Knowledge Discovery Baylor College of Medicine
Graz University of Technology Director, Epilepsy and Neurophysiology Program
Graz, Austria Director, Neurocritical Service, Section of Neurology
Chief of Neurophysiology
Texas Children’s Hospital
Houston, TX
Contributors xiii

Alexander Rotenberg, M.D., Ph.D. William O. Tatum IV, D.O.

Assistant Professor of Neurology Professor of Neurology
Harvard Medical School Mayo School of Medicine
Attending Staff Physician, Neurology Director, Epilepsy Monitoring Unit
Children’s Hospital Mayo Clinic and Hospital
Boston, MA Jacksonville, FL
Erik Rumpl, M.D. Barbara Tettenborn, M.D., Ph.D.
Professor of Neurology Chair, Department of Neurology
Landeskrankenhaus Kantonsspital St. Gallen
Klagenfurt, Austria St. Gallen, Switzerland
Johannes Gutenberg University
Donald L. Schomer, M.A., M.D.
Mainz, Germany
Professor of Neurology
Harvard Medical School Ab van Rotterdam, Ph.D.
Director, Laboratory for Clinical Neurophysiology Senior Researcher (Retired)
Chief, Comprehensive Epilepsy Program Radiobiological Laboratory
Beth Israel Deaconess Medical Center Radiobiological Institute TNO
Boston, MA Rijswijk, The Netherlands
Margitta Seeck, M.D. Sampsa Vanhatalo, M.D., Ph.D.
Professor Department of Clinical Neurophysiology
Director of the EEG and Epilepsy Unit Children’s Castle
University Hospital of Geneva Helsinki University Central Hospital
Geneva, Switzerland Helsinki, Finland
Megan Selvitelli, M.D. Juha Voipio, Ph.D.
Maine Medical Partners Neurology Professor of Biological and Environmental Sciences
Scarborough, Maine University of Helsinki
Helsinki, Finland
Erw in-Josef Speckmann, M.D.
Professor Emeritus Thoru Yamada, M.D.
Institute of Physiology (Neurophysiology) Professor of Neurology
University of Münster Roy J. and Lucille A. Carver College of Medicine
Münster, Germany University of Iowa
Director, Division of Clinical Electrophysiology
Cornelis Jan Stam, M.D.
University of Iowa Hospitals and Clinics
Department of Clinical Neurophysiology
Iowa City, IA
VU University Medical Center
Amsterdam, The Netherlands Malcolm Yeh, M.D.
Associate Clinical Professor of Neurology
Travis R. Stoub, Ph.D.
Division of Clinical Electrophysiology
Assistant Professor of Neurological Sciences
Roy J. and Lucille A. Carver College of Medicine
Affiliated Scientist
University of Iowa
Rush University Medical Center
Staff Neurologist
Chicago, IL
University of Iowa Hospitals and Clinics
Takeo Takahashi, M.D. Iowa City, IA
Yaotome Clinic
Sendai, Japan

Preface to the First Edition vi 19 Cerebrovascular Diseases and EEG . . . . . . . . . . . . . . . . . . . . . . 351

Barbara Tettenborn, Ernst Niedermeyer, and Donald L. Schomer
Preface to the Sixth Edition viii
Contributors x 20 Dementia and EEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
Cornelis Jan Stam

Part I Basic Principles 21 Metabolic Disorders and EEG . . . . . . . . . . . . . . . . . . . . . . . . . . 395

Trudy Pang, Megan Selvitelli, Donald L. Schomer, and Ernst Niedermeyer
1 Historical Aspects of EEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 22 EEG and Craniocerebral Trauma . . . . . . . . . . . . . . . . . . . . . . . . 411
Ernst Niedermeyer and Donald L. Schomer Erik Rumpl
2 Neurophysiologic Basis of EEG and DC Potentials . . . . . . . . . . 17 23 Anoxia, Coma, and Brain Death . . . . . . . . . . . . . . . . . . . . . . . . 435
Erwin-Josef Speckmann, Christian E. Elger, and Ali Gorji Peter W. Kaplan and Gerhard Bauer
3 Cellular Substrates of Brain Rhythms . . . . . . . . . . . . . . . . . . . . . 33 24 The EEG in Patients with Migraine and Other
Florin Amzica and Fernando H. Lopes da Silva
Forms of Headache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
4 Dynamics of EEGs as Signals of Neuronal Populations: Ernst Niedermeyer and Donald L. Schomer
Models and Theoretical Considerations . . . . . . . . . . . . . . . . . . . 65
Fernando H. Lopes da Silva
Part IV Clinical EEG in Epilepsy
5 Biophysical Aspects of EEG and Magnetoencephalogram and Related Disorders
Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Fernando H. Lopes da Silva With an Appendix by Ab Van Rotterdam 25 Seizures and Epilepsies in the Preterm and Term Neonate ... 465
Emma Laureta, Eli M. Mizrahi, and Solomon L. Moshé
6 Analog Signal Recording Principles . . . . . . . . . . . . . . . . . . . . . . 111
Charles M. Epstein 26 Seizures and Epilepsy in Infants to Adolescents . . . . . . . . . . . . 479
Douglas R. Nordli, Jr., James J. Riviello, Jr., and Ernst Niedermeyer
7 Digital EEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Douglas Maus, Charles M. Epstein, and Susan T. Herman 27 Epilepsy in Adults and the Elderly . . . . . . . . . . . . . . . . . . . . . . . 541
Bernard S. Chang, Donald L. Schomer, and Ernst Niedermeyer
8 Polarity and Field Determinations . . . . . . . . . . . . . . . . . . . . . . . 143
Bruce J. Fisch 28 Convulsive Status Epilepticus . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
Frank W. Drislane, Susan T. Herman, and Peter W. Kaplan

Part II Normal EEG 29 Nonconvulsive Status Epilepticus . . . . . . . . . . . . . . . . . . . . . . . 595

Frank W. Drislane, Peter W. Kaplan, and Susan T. Herman
9 Normal EEG and Sleep: Preterm and Term Neonates . . . . . . . 153 30 Anticipating Seizures Based on EEG . . . . . . . . . . . . . . . . . . . . . 645
Eli M. Mizrahi, Solomon L. Moshé, and Richard A. Hrachovy
Fernando H. Lopes da Silva and Stiliyan Kalitzin
10 Normal EEG and Sleep: Infants to Adolescents . . . . . . . . . . . . 163 31 Nonepileptic Attacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659
James J. Riviello, Jr., Douglas R. Nordli, Jr., and Ernst Niedermeyer
Megan Selvitelli, Trudy Pang, Donald L. Schomer, and Ernst Niedermeyer
11 Normal EEG and Sleep: Adults and Elderly . . . . . . . . . . . . . . . 183
Bernard S. Chang, Donald L. Schomer, and Ernst Niedermeyer
Part V Complementary and Special Techniques
12 Activation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Takeo Takahashi and Keith H. Chiappa 32 Nasopharyngeal, Anterotemporal,
13 Artifacts of Recording . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239 and Sphenoidal Electrodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 669
Andres M. Kanner, Travis Stoub, and Steve Bild
Barbara Dworetzky, Susan Herman, and William O. Tatum IV
33 Intracranial Monitoring: Depth, Subdural,
Part III Clinical EEG: General and Foramen Ovale Electrodes . . . . . . . . . . . . . . . . . . . . . . . . . . 677
Margitta Seeck, Donald L. Schomer, and Ernst Niedermeyer
14 Patterns of Unclear Significance . . . . . . . . . . . . . . . . . . . . . . . . . 267 34 Electrocorticography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 715
Jonathan Charles Edwards and Ekrem Kutluay Marc R. Nuwer
15 EEG of Degenerative Disorders of the Central 35 Principles and Techniques for Long-Term EEG Recording
Nervous System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 (EMU, ICU, Ambulatory) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725
John Gaitanis Jean Gotman, Marc Nuwer, and Ronald G. Emerson
16 The EEG in Congenital Malformations of Cortical 36 Infraslow EEG Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 741
Development, Neurocutaneous Disorders, Cerebral Palsy, Sampsa Vanhatalo, Juha Voipio, and Kai Kaila
Autism/Mental Retardation, and ADHD/Learning 37 High-Frequency EEG Activity . . . . . . . . . . . . . . . . . . . . . . . . . . 749
Disabilities of Childhood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299 Jean Gotman and Nathan E. Crone
Mary Repole Andriola
38 Intraoperative Evoked Potential Monitoring . . . . . . . . . . . . . . 767
17 Brain Tumors and Other Space-Occupying Lesions . . . . . . . . 321 Alan D. Legatt
Adam L. Hartman and Ronald P. Lesser
39 Monitoring EEG during Carotid Surgery . . . . . . . . . . . . . . . . . 787
18 The EEG in Inflammatory CNS Conditions . . . . . . . . . . . . . . . 331 Thoru Yamada and Malcolm Yeh
Robert L. Beach, Helen Barkan, and Edgar DePeralta

Contents xv

40 Polygraphy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 809 50 Neurocognitive Processes and the EEG/MEG . . . . . . . . . . . . . 1083

Anton Kamp, Gert Pfurtscheller, Günter Edlinger, Fernando H. Lopes da Silva
and Fernando H. Lopes da Silva

41 Polysomnography: Technical and Clinical Aspects . . . . . . . . . 817 Part VIII EEG and Neurocognitive Functions
Sudhansu Chokroverty, Rodney Radtke, and Janet Mullington
51 Psychiatric Disorders and EEG . . . . . . . . . . . . . . . . . . . . . . . . . 1113
42 Magnetoencephalography: Methods and Applications . . . . . . 865 Andrew D. Krystal
Riitta Hari
52 Technical Aspects of Transcranial Magnetic
Part VI EEG: Neuropharmacology and Anesthesia and Electrical Stimulation of the Brain . . . . . . . . . . . . . . . . . . 1129
Alan D. Legatt, Alvaro Pascual-Leone, and Alexander Rotenberg
43 EEG, Drug Effects, and Central Nervous System Poisoning . . 901 53 Transcranial Magnetic Stimulation (TMS):
Gerhard Bauer and Richard Bauer
Clinical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1139
Alexander Rotenberg and Alvaro Pascual-Leone
Part VII Evoked Potentials and Event-Related EEG
Phenomena Part IX Computer-Assisted EEG Analysis
44 Event-Related Potentials: General Aspects of 54 EEG Analysis: Theory and Practice . . . . . . . . . . . . . . . . . . . . . 1147
Methodology and Quantification . . . . . . . . . . . . . . . . . . . . . . . 923 Fernando H. Lopes da Silva
Fernando H. Lopes da Silva
55 EEG Mapping and Source Imaging . . . . . . . . . . . . . . . . . . . . . 1179
45 EEG Event-Related Desynchronization (ERD) Christoph M. Michel and Bin He
and Event-Related Synchronization (ERS) . . . . . . . . . . . . . . . . 935 56 Computer-Assisted EEG Pattern Recognition
Gert Pfurtscheller and Fernando H. Lopes da Silva
and Diagnostic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1203
46 Visual Evoked Potentials and Electroretinograms . . . . . . . . . . 949 Fernando H. Lopes da Silva
Gastone G. Celesia and Neal S. Peachey
57 EEG-Based Brain–Computer Interfaces . . . . . . . . . . . . . . . . . 1227
47 Brainstem Auditory Evoked Potentials (BAEPs) Gert Pfurtscheller and Christa Neuper
and Other Auditory Evoked Potentials . . . . . . . . . . . . . . . . . . . 975 58 Multimodal Monitoring of EEG and Evoked Potentials . . . . 1237
Gastone G. Celesia Gert Pfurtscheller
48 Somatosensory-Evoked Potentials: Normal Responses,
Abnormal Waveforms, and Clinical Applications
in Neurologic Diseases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1003 Index 1243
François Mauguière

49 Evoked Potentials in Children and Infants . . . . . . . . . . . . . . . 1057

Aatif M. Husain
Part I Basic Principles

Historical Aspects of EEG

DISCOVERY OF ELECTRICAL PHENOMENA electricity) and placed the emphasis on physics—on his “pile,”
the first battery (around 1800). This bimetallic pile was a gen-
Thales from Miletos has been credited with the discovery of erator capable of producing a steady flow of electricity. Volta’s
static electricity produced by friction (rubbing fur or glass with view more or less prevailed in this hotly debated argument. The
silk). He was one of the pre-Socratic “natural philosophers” of laws governing flowing electricity were soon discovered by
Greece (around 620–550 BC) and considered water the origin Georg Ohm in 1827.
of all things. Thus, friction was recognized as the generator of a Nevertheless, Galvani’s belief in “animal electricity” was not
phenomenon that derived its name from the Greek work “elec- lost with other discarded false ideas. There still remained the
tron,” which stands for amber. This discovery fell into a dor- nagging question of an active electrical contribution of animal
mant stage for more than two millennia. muscle tissue.
Around 1600, William Gilbert began to study the electrical
properties of various substances, and Otto von Guericke
(1602–1686) invented the friction machine to create electrical BEGINNINGS OF ELECTROPHYSIOLOGY
fields. This machine eventually found its way into doctors’
offices and even university hospitals. Its electrical field would The introduction of the galvanometer has been associated
make a patient’s hair stand up, creating a strong impression on chiefly with the name of Nobili in Florence; this instrument was
a psychologically gullible patient. These friction machines now refined in 1858 by William Thompson (Lord Kelvin) in
ornament high school laboratories and technical museums. In England (O’Leary and Goldring, 1976). These galvanometers
the 17th and 18th centuries, the friction machine taught invalu- would faithfully demonstrate continuous electrical currents
able lessons on attraction and repulsion of charged bodies, on and their variations in intensity but failed in the detection of
conductors and nonconductors, and on the rather questionable instantaneous electrical phenomena.
dualism of positive and negative electricity. Carlo Matteucci (1811–1868) in Bologna and Emil Du Bois-
A new and very important piece of electrical equipment Reymond (1818–1896) in Berlin became the major proponents of
entered the scene in 1746 when the Leyden jar was introduced an electrophysiologically based physiology of the nervous system.
by Pieter van Musschenbroek (following the earlier work of (The French name of Du Bois-Reymond indicates the Huguenot
Ewald von Kleist). This invention resulted in the storage of elec- origin of this Prussian investigator.) Du Bois-Reymond coined the
tricity, and its upshot, the condenser or capacitor, turned into term negative variation for a phenomenon occurring during mus-
an indispensable part of modern electronics. Benjamin cle contraction when the galvanometer indicated an unexpected
Franklin’s bold experiment caught electrical discharges of a decrease in current intensity (O’Leary and Goldring, 1976). This
thunderstorm in a Leyden jar. term was later resurrected in earliest electroencephalogram (EEG)
What the friction machine could generate, the Leyden jar research (Caton, 1875) and with the discovery of the “contingent
could store. Its sudden discharge was used in many experiments negative variation” (Walter, 1964).
(O’Leary and Goldring, 1976). Hermann von Helmholtz (1821–1894) accurately measured
The role of static electricity in medicine appeared to be for- the velocity of nerve conduction, which had been vastly overes-
gotten for about 150 years and became resurrected with the timated up to that time. The electrodes used in physiologic
introduction of the defibrillating cardioversion by William B. research were improved and made nonpolarizable (Du Bois-
Kouwenhoven and his coworkers in the 1950s and 1960s; this Reymond). The concept of “action current” was introduced by
approach may hold promises for cerebral applications L. Hermann (1834–1919) and thus clarified Du Bois-Reymond’s
(Niedermeyer, 2003a). negative variations found during muscle contraction. Julius
A serious scientific controversy developed in Italy between Bernstein (1839–1917) proposed a membrane theory of nerve
Luigi Galvani (1737–1798), a professor at the University of tissue, which ultimately was elucidated as late as 1939 and the
Bologna, and Alessandro Volta (1745–1832) in the wake of following years by A. L. Hodgkin and A. F. Huxley in England.
Galvani’s discovery of frog leg contractions within an electrical Against this background of strongly evolving electrophysiology
circuit and especially in the presence of a thunderstorm (1780). of the nervous system, the first observation of EEG-like electri-
Volta doubted the biologic nature of the contraction (animal cal brain activity took place.

2 Part I ■ Basic Principles

CATON: THE FIRST ATTEMPT AT THE Caton also described a few more interesting observations.
ELECTRICAL ACTIVITY OF THE BRAIN He noted that the external surface of the gray matter was posi-
tive in relation to deep structures of the cerebrum. He also
Richard Caton (1842–1926) (Fig. 1.1) was a physician practic- noted that the electric currents of the cerebrum appeared to
ing in Liverpool who became deeply interested in electrophysi- have a relation to underlying function: “When any part of the
ologic phenomena and eventually received a grant from the grey matter is in a state of functional activity, its electric current
British Medical Association to explore electrical phenomena of usually exhibits negative variation.” Thus, Caton has also been
the exposed cerebral hemispheres of rabbits and monkeys. credited with pioneer work on evoked potential. Furthermore,
According to Brazier (1961), Caton presented his findings to the the difference in polarity found between cortical surface and
association on August 24, 1875, and a very short report of 20 deeper areas could be interpreted as the discovery of the “steady
lines subsequently appeared in the British Medical Journal. A potential” (“DC potential”), but it might be wise to refrain from
more detailed report was presented in the same journal in 1877 such statements that cannot be fully supported by the evidence.
on experiments of more than 40 rabbits, cats, and monkeys, the With regard to the fluctuations, Geddes (1987) pointed out that
rabbit having been principally employed. Caton’s galvanometer had a very limited frequency response
Caton used a galvanometer. A beam of light was thrown on range from 0 to 6 Hz.
the mirror of the galvanometer and reflected on a large scale Caton found some measure of success and recognition with
placed on the wall. With this type of visualization, Caton found this work and held the chair of physiology at the University
that “feeble currents of varying direction pass through the mul- College of Liverpool from 1884 to 1891, when he resigned from
tiplier when the electrodes are placed on two points of the this post. Later he became dean of the medical faculty and, in
external surface, or one electrode on the grey matter, and one 1907, mayor (Lord Mayor) of Liverpool. The electrical activity of
on the surface of the skull.” This sentence is regarded as indicat- the brain did not occupy a predominant position in his further
ing the birth of the electrophysiologram because one can endeavors. Even though Caton became an EEG research dropout,
assume that EEG phenomena made the needle move from one his bold work will always remain a milestone in the history of the
direction to the other. (The suffix “gram” naturally is out of electrical activity of the brain. (More information on Caton’s life
place since “graphein” means “to write” and there was no writ- and work is found in Mary Brazier’s (1961) fine account.)
ten recording.) Even though artifacts could have played a major
role, Caton deserves credit for the discovery of the fluctuating EASTERN EUROPEAN STUDIES OF
potentials that constitute the EEG. ELECTRICAL BRAIN ACTIVITY
The time was ripe for further studies of electrical phenomena of
the cerebrum. Concurrent with Caton’s epochal work of 1875,
physiologists of Eastern Europe began to demonstrate their
independent observations and discoveries concerning the brain
and its electrical activity. Another discovery of the 1870s had
an incomparably greater impact on the neuroscientific world
than Caton’s demonstration of electrical activity of the brain.
The capability of the human cerebral cortex to be electrically
stimulated was discovered by G. Fritsch (1838–1927) and Julius
Eduard Hitzig (1838–1907) in a joint study in 1870. According
to O’Leary and Goldring (1976), an unusual observation had
prompted Fritsch in his work: he had observed contralateral
muscle contractions during dressing of an open brain wound in
the Prussian–Danish War of 1864. The work of Fritsch and
Hitzig was furthered by D. Ferrier and G. F. Yeo in 1880, who
performed electrical stimulations of the cerebrum in apes and
also in a patient who was operated on for a brain tumor. The
repercussions of the stimulation studies were considerable since
many investigators of that time held the view that the entire
cerebrum is a homogeneous organ that harbors mental functions.
The response of the cortex to electrical stimulation probably
was a special incentive for the study of its spontaneous electri-
cal phenomena. This incentive was particularly strong in
Eastern Europe, that is, in laboratories of Russian and Polish
Figure 1.1 Richard Caton at the time of his work on the electrical universities. (In spite of the important historical ethnic and
activity of the brain. (From Brazier MAB. A History of the Electrical national differences, the fact cannot be ignored that most of
Activity of the Brain. The First Half-Century. London: Pitman; 1961, with Poland was part of the Czarist Russian Empire throughout the
permission from Macmillan.) 19th century.)
Chapter 1 ■ Historical Aspects of EEG 3

Vasili Yakovlevich Danilevsky (1852–1939) was only 25 years the most important advances was Willem Einthoven’s introduc-
old when he finished his thesis entitled “Investigations into the tion of the string galvanometer in 1903, a very sensitive instru-
Physiology of the Brain” (Danilevsky, 1877), written at the ment that required photographic recording and became the
University of Kharkov. This work was based on electrical stim- standard instrumentation for electrocardiography at the turn of
ulation as well as on spontaneous electrical activity in the the century.
brains of animals. Thus, Danilevsky walked in Caton’s foot- Kaufman’s work and life are portrayed in Brazier’s (1961)
steps; in 1891, he gave full credit to Caton’s priority. Mary historical account. Kaufman expressed the view that an epilep-
Brazier (1961) comments on the disappointment of Danilevsky tic attack would have to be associated with abnormal electrical
who saw his high hopes unfulfilled as far as the spontaneous discharges, and he studied the effects of cortical electrical stim-
electrical activity of the brain was concerned; he had expected ulation. With World War I, he took the name of Rostoutsev and
better correlation with psychic and emotional processes. He worked mainly at the University of Baku.
remained deeply involved in brain physiology and published an Pravdich-Neminsky began recording electrical brain activity
extensive textbook of human physiology in 1915. He was not of animals in 1912 with the string galvanometer. As Brazier
the only EEG researcher with shattered hopes in the field of (1961) has pointed out, his recordings, published in 1912, were
psychophysiology. the first pictorial demonstration of EEG and appeared 2 years
The life and work of Adolf Beck (1863–1939) have been earlier than Cybulski’s tracings. Pravdich-Neminsky recorded
described in great detail by Brazier (1961). Beck worked in the EEG from the brain, the dura, or the intact skull of the dog.
Kraków as well as in Lwow (the Polish province of Galicia, at He described a 12 to 14/sec rhythm under normal conditions
that time a part of the Austrian–Hungarian monarchy). With and marked slowing under asphyxia. Furthermore, he coined
nonpolarizable electrodes, Beck investigated the spontaneous the term electrocerebrogram (Fig. 1.2).
electrical activity of the brain in rabbits and dogs. He observed The achievements of the Eastern European neuroscientists
the disappearance of rhythmical oscillations when the eyes were during those 50 years preceding the outbreak of World War I fill
stimulated with light and thus became a forebear of Berger’s us with awe and clearly demonstrate their special talent for elec-
discovery of alpha blocking. His work became widely known trophysiologic neurophysiology. Limiting discussion to EEG
due to its publication in the Centralblatt. history can show only the tip of the iceberg. To assess the true
To present a chronologic account of the events, let us leave strength of their neuroscientific institutions, one must mention
Eastern Europe for a moment, but not Vienna. In 1883, Ernst investigators in somewhat related electrophysiologic areas. Ivan
Fleischl von Marxow (1846–1891) deposited a sealed letter at Michailovich Sechenov (1829–1905) appears to be founder of
the Imperial Academy of Sciences in Vienna that contained this powerful school of eminent neurophysiologists. He studied
observations on cerebral electrical activity recorded over the the electrical activity of the spinal cord and oblongata in the
visual cortex in various species of animals. He did not observe frog and was a predecessor of Pavlovian thought. Nikolai
oscillatory activity. He claimed priority when Beck in 1890 pub- Yevgenevich Wedensky followed Sechenov as the chair and pro-
lished his data but was not aware of earlier work done by Caton fessor of physiology at St. Petersburg (known for the concept of
and Danilevsky. The oddity of this episode is underscored by Wedensky inhibition). Vladimir Efimovich Larionov, also
more recent historical accounts (Brazier, 1961; O’Leary and working in St. Petersburg, conducted beautiful studies of the
Goldring, 1976) that indicate that Fleischl von Marxow’s work auditory cortex in the dog.
was not of first-rate quality. This does not detract from the ren- The greatest Russian neuroscientist was also the most eminent
aissance-man versatility of this Austrian physiologist, who was clinical neurologist of his country: Vladimir Mikhailovich
well versed in linguistics (even Sanskrit), swimming, hunting, Bechterev (also “Bekhterev”) (1857–1927). He occupied the chair
and mountain climbing. of psychiatry in St. Petersburg, which included the field of clini-
Exciting new studies were being conducted in Eastern cal neurology. He was a disciple of Du Bois-Reymond, Paul Emil
European universities. Napoleon Cybulski (1854–1919), who Flechsig, and Wilhelm Wundt, and also worked at Charcot’s clinic
was Beck’s teacher in Kraków and an internationally renowned in Paris. The influence of Wundt prompted Bechterev’s associa-
leader in general physiology, presented experimental electroen- tive reflexology (I treasure his work “Allgemeine Grundlagen der
cephalographic studies in graphical form by using a gal- Reflexologie des Menschen,” Deuticke, Leipzig, 1926, even
vanometer with a photographic attachment. He provided EEG though I can hardly agree with his “objective study of personal-
evidence of an epileptic seizure in the dog caused by electrical ity”). He combined his clinical work and private practice with
stimulation. tireless psychophysiologic methods. A photograph with one of
Two Russian physiologists made further studies along these his two assistants shows Bechterev wearing a thick winter coat in
lines: Pavel Yurevich Kaufman (1877–1951) and Vladimir his institute in Leningrad, giving testimony to the icy cold in the
Vladimirovich Pravdich-Neminsky (1879–1952). Prior to the unheated laboratory. “Functional anatomy of the brain, experi-
discussion of their work, a few words must be said about tech- mental psychology and clinical neurology were the three fields in
nological developments. The d’Arsonval galvanometer featured which Bekhterev carved out a place for himself” (Yakovlev, 1953).
a mirror mounted on a movable coil; light focused on the mir- The Soviet regime had to choose between Bechterev and
ror was deflected when a current passed the coil. The capillary Ivan Petrovich Pavlov (1849–1936), a physiologist who had won
electrometer was introduced by G. Lippmann and H. J. Marey the Nobel Prize in 1904 for his early work on conditioned
(for further details, see O’Leary and Goldring (1976)). One of reflexes. Pavlov was the choice, and the magic of conditioned
4 Part I ■ Basic Principles

Figure 1.2 The first photographs to be published of electroencephalograms. In the upper record Neminsky shows (in the
third trace) the brain potentials of a curarized dog with the pulsations from an artery in the brain recorded above them.
In the lower record the sciatic nerve is being stimulated from time to time, and the decrease in activity noted by Neminsky
can be seen. The record reads from right to left, line I being a time marker in one fifth of a second, line III the galvanometer
string, and line V the signal for stimulation. (From Pravdich-Neminsky VV. Ein Versuch der Registrierung der elektrischen
Gehirnerscheinungen. Zbl. Physiol. 1913;27:951–960, with permission from Dr. Mary Brazier and Macmillan.)

reflex overshadowed all Soviet neurophysiology by highest The Western neurophysiologists followed attentively the
decree (even though Pavlov himself was highly critical of the work of their neuroanatomical confreres and the great contro-
regime). The Pavlovian concept was closer to the ideology of versy between network theories (the nerve cells forming a felt-
dialectic materialism, and this maxim with all its intolerant like net) and the neuron theory (the neuron representing a
dogmatism outlasted Pavlov’s death by two decades. This unit). The net theory was supported by Joseph von Gerlach and
ideopolitically governed form of neuroscience stifled all by Camillo Golgi, the discoverer of the silver chromate staining
progress of customary neurophysiology. The world leadership method, but the neuron theory with its great principal propo-
in EEG and related fields quickly crumbled, accompanied by a nent Santiago Ramón y Cajal (1852–1934) proved to be victo-
terrifying decline of a dogmatically governed neurophysiology. rious in this struggle despite further attempts of new
generations of “reticularist” believers in a continuous network
AND CENTRAL EUROPE In a similar manner, cerebral localizationists struggled
against antilocalizationists. In Germany, Friedrich Leopold
Electroencephalographic research was in a dormant state in Goltz (1834–1902) removed the cerebral hemispheres in the
Western and Central Europe while it was flourishing in Eastern “dog without cerebrum” living in a state of extreme lethargy
European countries. This is quite amazing because neurophys- and mental inertia (Goltz, 1888). H. Rothmann (1923) per-
iology in general was healthy and well outside Russia and her formed similar investigations. This type of research was aimed
neighbors, but the ancestral lineage from Galvani to Du Bois- at the working of the cerebral hemispheres as a whole and de-
Reymond and Caton broke off and the neurophysiological emphasized the aspects of cortical localization. As it was
field was watered by rivulets of different orientation. Thus, the pointed out previously, the cerebral stimulation studies of
work of Fleischl von Marxow lies like an erratic patch in the Fritsch, Hitzig, Ferrier, and Yeo initiated a new era of interest in
vast field. cortical localization.
Chapter 1 ■ Historical Aspects of EEG 5

In this period—the last third of the 19th and the dawning of

the 20th century—the work of Charles Scott Sherrington
(1857–1952), performed in Liverpool and Oxford, became
most influential in the development of a modern Western type
of reflexology. The Integrative Action of the Nervous System
(Sherrington, 1906) was based on a series of lectures held at Yale
University in 1904. The scope of this work reaches from reflex-
ology to decerebrate rigidity, from motor cortex to sensory
function, while the issue of mental function is being skirted
with the modesty of a truly great neuroscientist. Inhibition is
one of the great Sherringtonian discoveries. Even Ramón y
Cajal’s net of independent synaptically connected neurons
stood solely in the service of neural excitation. (Incidentally, the
term synapse was introduced by Sherrington.)
It is unfortunate that this greatest master of neurophysiology
stood miles away from electrophysiologic thought. His work
was based chiefly on ablation techniques. He may hardly ever
have given a thought to EEG methods even though he lived a
full active life. His disciples—to name only Edward Liddell and
Derek Ernest Denny-Brown—held similar views, while in
Cambridge electrically oriented neurophysiology found its
greatest proponent in Edgar Douglas Adrian (1889–1977),
whom we discuss later in his relationship to Hans Berger’s

HANS BERGER AND THE HUMAN Figure 1.3 Hans Berger. (From Kolle K. Hans Berger. In: Kolle K, ed.
Grosse NervenŠrzte. Vol. 1. Stuttgart: Thieme; 1956: 1–6.)
Hans Berger (1873–1941) (Fig. 1.3), the discoverer of the
human EEG, was a neuropsychiatrist. What neuropsychiatry capillary electrometer after Lippmann, but the results were dis-
really meant in those years is poorly understood nowadays. appointing. Naturally, Berger was aware of the scanty pertinent
Neurology and psychiatry formed one specialty, one discipline, literature from Caton to Cybulski and Pravdieh-Neminsky. His
in Germany, Austria, and a considerable number of other coun- studies of the human EEG started in 1920; the introduction to
tries. Neuropsychiatric departments at university hospitals and Gloor’s authoritative translation of Berger’s work contains a
other institutions consisted of neurologic and psychiatric plethora of interesting details (Gloor, 1969).
floors; medical specialty training meant rotation from one dis- Every electroencephalographer should be familiar with
cipline to the other, and a the head of one department was sup- Berger’s work, an undertaking that has been greatly facilitated
posed to be a master of both disciplines. Pure neurology was by Gloor’s English translation. It is true that the original
just beginning to emerge as a special discipline in the German- German text is cumbersome and does not make easy reading,
speaking countries (with the work of Wilhelm Erb which is probably a reason for the very slow acceptance of
(1840–1921), Max Nonne (1861–1959), and the incomparable Berger’s work. Those 14 reports bear the same title: “On the
Otfrid Foerster who, like Hans Berger, lived from 1873 to 1941). Electroencephalogram of Man.” Surely, a more attractive title
Berger was not a leader, neither in neurology nor in psychi- would have helped somewhat. Berger’s humanistic educational
atry. Without his pioneering EEG work, his name would have background becomes obvious in the rejection of the term elec-
been forgotten. Biographic sketches (especially Kolle, 1956) trocerebrogram of Pravdich-Neminsky for strictly linguistic
portray Berger as an extremely meticulous and conscientious reasons: the “ugly” mixture of Greek (“electro,” “gram”) and
person, somewhat aloof in his contact with his patients, a very Latin (“cerebro”) fragments. What Berger proposed in German
strict and authoritarian department head, and an “anima can- was the term Elektrenkephalogram (sic) since the root
dida” (a pure soul), a hard-working professor without any enkephalo from the Greek is linguistically more correct than
interest in faculty schemes and diatribes. He hardly ever encephalo.
attended the annual meetings of the German neuropsychiatric Before Berger’s work as such is brought into focus, we must
society. His electroencephalographic work was carried out in a discuss his electrophysiologic instrumentation. He used a string
small and very primitive laboratory. His first scientific interest galvanometer starting in 1910—first with the Einthoven type,
aimed at the cerebral circulation; plethysmographic methods later with the smaller Edelmann model, and after 1924 with
were used in patients with skull defects. From 1902 to 1910, he the larger Edelmann model. In 1926, Berger started to use the
studied the electrical activity of the cerebrum in the dog with a more powerful Siemens double-coil galvanometer (attaining a
6 Part I ■ Basic Principles

sensitivity of 130 V/cm; Grass, 1984). With this instrument depression, which remained undiagnosed. He ended his life by
and the use of nonpolarizable pad electrodes, Berger recorded suicide on June 1, 1941 at the age of 68. External factors may
the human EEG tracings shown in his first report of 1929. have contributed to his depression: in addition to his forced
The records were made on photographic paper with record- retirement at 65 (a few additional “years of honor” were granted
ings from 1 to 3 minutes’ duration. Berger used a bipolar record- to most retiring directors of university institutions), he also felt
ing technique with fronto-occipital leads for his one-channel challenged by a group of independent EEG workers at the
EEG tracings along with simultaneous electrocardiogram Institute of Brain Research at Berlin-Buch. This group was led
(ECG) recording and a time marker. In 1932, he received an by A. E. Kornmüller and produced excellent experimental EEG
oscillograph from Siemens but was unable to obtain further work, which is discussed later. Kornmüller might have had the
amplifiers with oscillographs in order to obtain multichannel better connections to the government institutions in Berlin,
recordings. and the highly sensitive and often insecure Berger was afraid
Studies of the human EEG began in 1924. Appointments that his discovery was being taken away from him by his more
were made for several patients with large skull bone defects aggressive colleagues in Berlin-Buch.
(there was no scarcity of such patients in post-World War I Berger was a very complex person and investigator. It was
Germany). On July 6, 1924, the small Edelmann string gal- pointed out previously that he did not excel clinically, neither
vanometer showed oscillations presumably coming from the as psychiatrist nor as neurologist (even though he was very
underlying brain. In 1925, Berger recognized that skull defects interested in cerebral localization and particularly in the local-
were not necessarily advantageous to obtaining a recording ization of brain tumors). Berger also developed very unscien-
because of thickening of dura, postoperative adhesions, etc., tific ideas about the nature of the EEG, even though he was a
and he found that recordings could be made just as well (or meticulous scientist in his EEG work. The driving force in all
even better) through the intact skull and scalp. Between 1926 his research work was the quest for the nature of the all-pow-
and 1929 Berger obtained good records with alpha waves; the erful force of mental energy (“psychische Energie”). An early
double-coil galvanometer was crucial for these observations. personal experience convinced him that such a mental
The data were often uncertain and, in 1928, Berger was beset energy—even capable of transmitting thoughts and emotions
with doubts concerning the authenticity of his observations from person to person—does exist. According to Berger’s con-
(according to his diary entries, very impressively demonstrated cept, influenced by the Danish physiologist Alfred Lehmann in
by Jung, 1963). 1901, mental energy is thought to be a partial product of meta-
The first report of 1929 features the alpha rhythm and the bolic energies (warmth and electricity being the other two
alpha blocking response (naturally along with a description of products). This concept gives the EEG waves the eerie charac-
the smaller beta waves). Chlorinated silver needle electrodes, ter of messengers within the mental activities, even as messen-
platinum wires, and zinc-plated steel needles were used in those gers from person to person.
years (Fig. 1.4). The bold first report of 1929 produced no Friedrich Rueckert (1788–1866), a great German poet of the
“waves” until a confirming report came from Adrian in Romantic period, was Berger’s maternal grandfather. Behind
Cambridge (Adrian and Matthews, 1934). Throughout the the strict directorial facade of Berger was the gentle soul of a
1930s, Berger’s reports on the human EEG contained veritable highly vulnerable man. It was the psychophysiologist Berger
gems: studies of fluctuation of consciousness, first EEG record- who searched for the correlate of mental energy and, on this
ings of sleep (the first recording of spindles), the effect of voyage, he found the human EEG. It was one of those
hypoxia on the human brain, a variety of diffuse and localized “Columbus syndromes”: that a discovery is made as a by-
brain disorders, and even an inkling of epileptic discharges. product of a search for a different goal. Jung (1963) noted that
Eventually, Berger was invited to an international congress of Berger pursued his goal with the extremely powerful energy of
psychologists in Paris in 1937 and to Bologna where the bicen- a dilettante who had found a concept. Specialists, like the excel-
tennial of Galvani’s birthday was celebrated (also 1937). His lent physiologist (and electrophysiologist) Wilhelm
relationship to the Nazi regime was not good, and Berger was Biedermann, in Jena were convinced that Berger’s dilettantism
most unceremoniously made a professor emeritus at earliest would lead nowhere, but it was the dilettante and not the sea-
convenience, in 1938. This was indeed a hard blow to his plans soned specialist who emerged victorious. Even though the EEG
for further electroencephalographic studies and, in the wake of is not exactly what Berger assumed, his contribution was the
a flu-like disease, he evidently developed a severe endogenous greatest in the history of electroencephalography.

Figure 1.4 The first recorded electroencephalogram of a human.

The lower line is a 10 cycles/ sec sine wave for use as a time marker.
The upper line is the recording from Berger’s young son made in
1925. (From Berger H. Über das Elektrenkephalogramm des
Menschen. 1st report. Arch Psychiat Nervenkr. 1929;87:527–570,
with permission from Dr. Mary Brazier and Macmillan.)
Chapter 1 ■ Historical Aspects of EEG 7

BERGER’S CONTEMPORARIES of personal gratitude. Decades earlier, Vogt successfully treated

an ailing Krupp—with hypnosis. Indeed, Vogt also used to be
The Berliner Group a master of this method!) Hugo Spatz replaced Vogt and con-
The Institute of Brain Research (Hirnforschungs-Institut) in siderably changed the goals of research. Toennies stayed with
Berlin-Bush harbored a group of ambitious and energetic H. S. Gasser at the Rockefeller Institute in New York, where he
investigators in various neurosciences. Oskar Vogt constructed the first cathode follower to record from high-
(1870–1959), one of the great neuroanatomists and neu- resistance electrodes. This was the birth of microelectrode
ropathologists of his time and a remarkably independent recording, which developed into an enormously powerful sci-
thinker with a wide intellectual horizon, was the director of the entific tool in the 1950s and 1960s.
institute. In 1936 he lost his “directorship for lifetime” when the Kornmüller’s work declined after World War II, when he
Nazi government became aware of Vogt’s activities at a similar appeared to be obsessed with a totally unproven theory of glia as
institute in Moscow and his reluctance to get rid of Jewish the generator of slow brain potentials. Jung, however, developed
coworkers (Hassler, 1959). into one of the greatest electroneurophysiologists of his time.
The Berliner Institute (a section of the Kaiser-Wilhelm
Institutes) was composed of a variety of departments. In this Developments in Great Britain
context, the Department of Physiology under M. H. Fischer and Edgar Douglas Adrian (Baron of Cambridge and, as such, Lord
the Department of Electrophysiology under A. E. Kornmüller Adrian) (1889–1977) was not only one of the greatest electrophys-
must be singled out. These departments enjoyed the collabora- iologic neurophysiologists of the 20th century; his name is also
tion of an outstanding physicist and electronic engineer, J. F. intimately associated with the discovery of the EEG because of his
Toennies (1902–1970), a personal friend of Oskar Vogt (both confirmation of Berger’s observations (Adrian and Matthews,
coming from the town of Husum in Holstein). 1934). He showed his colleagues his own beautiful alpha rhythm
Toennies built the first ink-writing biologic amplifier for the and the blocking effect due to eye opening, but, alas, his great
recording of brain potentials. While in New York as a fellow of electronic wizard, Brian Matthews, apparently had a low-voltage
the Rockefeller Foundation in 1932, he designed the differential EEG with no alpha rhythm. Due to some strange quirk, Adrian’s
amplifier—the still all-important principle of EEG amplifica- recording from the head ganglion of a water beetle happened to
tion—but he shares this achievement with Brian Matthews, be indistinguishable from Adrian’s alpha rhythm and was blocked
Adrian’s ingenious coworker whose work is discussed in the in the same manner, namely by light falling on the beetle’s eyes.
next section. The Adrian-water beetle similarity and the Adrian-Matthews EEG
The collaboration with Toennies gave the Berlin group a dissimilarity must have been terribly confusing to the onlookers
much better tool for EEG research in comparison with Berger’s (also see Adrian, 1936).
instrumentation. Kornmüller quickly recognized the impor- Adrian was already a neurophysiologist of great prestige
tance of recordings from a greater number of electrodes. His when he confirmed Berger’s data. He had been credited with
EEG studies in the humans placed particular emphasis on the the demonstration of single sensory nerve fiber potential and
differences between given regions of the cerebrum the analysis of unit activity, which resulted in the Adrian–Bronk
(“Hirnrindenfelder”) (Kornmüller, 1932, 1933, 1935, 1937). His law (Adrian and Bronk, 1929). Incidentally, his collaborator
studies of the clinical significance of EEG (Kornmüller, 1944) Detlef Bronk became president of Johns Hopkins University in
appear to be somewhat pale when compared with the impor- later years. Prior to Matthews, Keith Lucas had been Adrian’s
tance of his earliest experimental EEG work carried out with brilliant electronic engineer and experimental coworker.
Fischer and also with H. Löwenbach, who later came to the W. Grey Walter became the pioneer of clinical electroen-
United States, where he became one of the earliest EEG pioneers. cephalography in England, and his discovery of foci of slow
In those early studies, the EEG was obtained from the cortex of activity (delta waves, named by Walter) generated enormous
animals following poisoning with convulsive substances. This is clinical interest in the new method. Grey Walter, however, was a
the first EEG work focusing on epileptic manifestations and the Ph.D. basic scientist. The fact that he was principally an aca-
first demonstration of epileptiform spikes (Fischer, 1933; demic scholar, not a clinician, could have laid the foundation for
Fischer and Löwenbach, 1934a,b; Kornmüller, 1935). the aversion of England’s great neurologists toward the method
Oskar Vogt (his wife Cécile also being known as a great neu- of EEG, which in the following years was either ignored or left to
ropathologist) developed a concept of strict cerebral compart- Ph.D. electroencephalographers in the laboratory. We will find
mentalization in sharply separated areas. He showed indeed Grey Walter again in later decades; let it be said that he was one
extremely impressive boundaries between healthy and dis- of the most brilliant minds in all neurosciences—an independ-
eased areas in the hippocampus (Vogt and Vogt, 1937) and ent thinker, a powerful writer, and quite often a man nearly con-
conceived the cortex as divided into about 200 regions with sumed by the flame of his own brilliance. He founded a small
precise demarcation from field to field. This concept became a but very effective school in Bristol at the Burden Institute.
powerful leitmotiv for Vogt’s coworkers, and Kornmüller’s EEG
work clearly shows the marks of his authoritarian boss. Developments in France and Belgium
Richard Jung joined this group in 1937 when Vogt was Thus far, France seems to be unduly neglected in this historical
already fired. (The tycoon Alfred Krupp provided the Vogts overview. It had its own proud neurophysiologic schools in the
with a privately built institute in the Black Forest in his expression 19th century, and the names of François Magendie (1783–1855)
8 Part I ■ Basic Principles

and, above all, Claude Bernard (1813–1878) belong in the pan- America Enters the Scene
theon of neuroscience. Around 1935, the center of gravity in the still modest bulk of
A fine school of early electroencephalographers developed in EEG work started to shift from Europe to North America.
Paris in the 1930s. A. Fessard at the Collège de France must be Fascinating new reports came from the United States. European
singled out as the most towering figure. Together with G. investigators started to travel across the Atlantic, and even Hans
Durup, he also confirmed the results of Berger. Durup and Berger was about to accept an invitation to the United States in
Fessard even used EEG in the study of conditioned reflexes. 1939, when the beginning of World War II thwarted his plans.
Clinical electroencephalography started in France under the In the pre-Berger development of experimental EEG studies,
aegis of A. Baudouin and G. Fischgold. Fischgold had come America had not played any role. Schwab (1951) reports that, in
from Rumania, developed into a leading clinical electroen- 1918, a medical student of Harvard Medical School, Donald
cephalographer, and—what an unusual combination—became McPherson, worked under the eminent physiologist Alexander
a leader in neuroradiology. Baudouin was the key figure in Forbes. When McPherson placed two electrodes on the exposed
the invitation of Berger to Paris in 1937. brain of a cat and ran the output into a string galvanometer, he
Neighboring Belgium was the home of a giant in electro- saw rhythmical 10/sec EEG activity. This finding was rejected as
physiologic neurophysiology: Frederic Bremer (1892–1982) from an artifact by Forbes. Was Forbes completely unaware of the
the Université Libre of Brussels. Bremer quickly recognized the work from Caton to Pravdich-Neminsky?
usefulness of EEG methods in the experimental investigation of The rise of American EEG work to international fame is cus-
the brain. He recognized the influence of afferent signals on the tomarily associated with the work of Hallowell Davis, Frederic
state of vigilance and compared his feline preparation called A. Gibbs, and Erna Gibbs at Harvard and also with Herbert
“cerveau isolé” (with midbrain transection) with the “encéphale Jasper’s work at Brown University in Providence, Rhode Island.
isolé” resulting from transection at the boundary between the According to O’Leary and Goldring (1976), A. J. Derbyshire, a
medulla oblongata and cervical cord. The former preparation graduate student of Hallowell Davis, brought Berger’s paper of
would produce permanent coma; the latter would cause a vari- 1929 to Davis’s attention. Derbyshire, Pauline Davis, and H. N.
able state of vigilance, with waking and sleeping demonstrated Simpson then tried in vain to demonstrate their own alpha
on the EEG recording. In other words, trigeminal-sensory, audi- rhythms. There were finally shouts of joy when Hallowell Davis
tory, visual, and probably also olfactory influences would help himself was found to have a good alpha rhythm. Otherwise, the
to keep the (artificially ventilated) encéphale isolé preparation in first human EEG study in America would have been a negative
a waking–sleeping rhythm (Bremer, 1935). one. This work was done in 1934, just before human EEG stud-
The greatness of this investigator must be reemphasized, ies started to mushroom in the United States.
especially in today’s era of short memory. Whoever reads his The EEG, however, had been used for animal experiments
study entitled “Cerebral and Cerebellar Potentials” (Bremer, for some years in the United States, starting with Bartley and
1958) will roughly understand the dimensions of this Newman (1930, 1931) and Bartley (1932), who produced EEG
neurophysiologist. tracings in the dog. Howard Bartley did his work at Washington
University in St. Louis, a place that had already developed into
Developments in Other European Countries
a hotbed of neurophysiology due to the magnificent work of
Italy was one of the first countries where the EEG found fertile Herbert S. Gasser, Joseph Erlanger, and George Bishop—a
soil. Mario Gozzano, for many years professor of clinical neu- group that made excellent use of Braun’s cathode ray oscillo-
rology in Bologna (later in Rome), published his experiences scope (oscillograph) in the study of peripheral nerve potentials.
with the new method as early as 1935. Gozzano personifies the This outstanding group was joined later by James L. O’Leary, a
(not too common) example of a leading neurologist assuming prominent neurophysiologist, electroencephalographer, and
leadership in clinical electroencephalography (Mazza et al., neurologic clinician. Early experimental EEG work was done by
2002). All too often, eminent clinical neurologists spurned the Davis and Saul (1931), Travis and Dorsey (1932), Travis and
new method. Herren (1931), Bishop and Bartley (1932), Bartley (1932), and
A. Gemelli came from the diametrically opposite area of Gerard et al. (1933) (after Grass, 1984). The work of Ralph W.
neurosciences. This great scholar was a monk, psychologist, Gerard (1900–1974) is linked with the introduction of a con-
philosopher, polyhistor, and president of the Catholic centric needle electrode for the stereotaxic exploration of the
University in Milan. In 1937, he reported his first studies of the brain in experimental animals. Gerard joined forces (in 1934)
human EEG. Gemelli hence represents the psychological wing with Franklin Offner, one of the leading electronic engineers in
of EEG research, which subsequently spawned a number of the development of EEG and related equipment.
outstanding Ph.D. electroencephalographers. (Others would American EEG work in the human started, as it was pointed
come from the ranks of experimental neurophysiologists.) out before, at Harvard in Boston (Hallowell and Pauline Davis,
The Austrian psychologist Hubert Rohracher falls into the Frederic and Erna Gibbs, and William G. Lennox), at Brown in
Gemelli category. He held the chair of psychology for many Providence (Herbert H. Jasper), but also at the University of
years at the University of Vienna, but, in his early academic Iowa in Iowa City where Lee Travis worked, an experimental
work, he fell under the spell of the alpha rhythm and even made psychologist who became the founder of a powerful school
a “pilgrimage” to Hans Berger in Jena (in the 1930s). His early (Herbert Jasper, Donald Lindsley, John R. Knott, and Charles
EEG studies can be dated back to 1938. Henry).
Chapter 1 ■ Historical Aspects of EEG 9

The great international breakthrough in clinical electroen- few examples of paroxysmal EEG discharges in a case of pre-
cephalography came in 1934 with studies of epileptic patients. sumed petit mal attacks and also during a focal motor seizure
Frederic Gibbs had come from Johns Hopkins University in in a patient with general paresis. These observations were just
Baltimore to join the Harvard group. He sought out William G. mentioned in passing and the opportunity of a major break-
Lennox, who had already become a widely known epileptologist. through was missed.
It might be interesting to point out that Lennox had started As to the other great pioneers of electroencephalography in
studies of the cerebral circulation by measuring the O2 and CO2 North America, Hallowell and Pauline Davis produced fine
content of the jugular veins (Lennox, 1930, 1931; Lennox and work on the normal EEG and its variants. They were also
Gibbs, 1932). E. L. Gibbs was Erna L. Gibbs, originally the tech- among the earliest investigators of the human sleep EEG. In the
nical coworker of Lennox but who became the wife of Frederic domain of sleep, A. L. Loomis and his coworkers E. N. Harvey
Gibbs and one of the world’s first EEG technicians and the coau- and G. A. Hobart were the first who methodically studied the
thor of numerous papers. She had come to Boston as an immi- human sleep EEG patterns and the stages of sleep. This research
grant from Germany. The pre-EEG work of Lennox and Gibbs was done off the academic track in Tuxedo Park, New Jersey
on the cerebral blood flow was a milestone in this field. (One of (Loomis et al., 1935, their first study). The Davises eventually
the great present-day masters of cerebral blood flow, Louis turned to audiology and moved to St. Louis. At Brown
Sokoloff from the National Institutes of Health, expressed to me University in Providence, Rhode Island, Jasper studied the EEG
in a personal communication his profound admiration for this of behavior disorders in children before he found his niche in
pioneering work.) EEG simply exerted a greater degree of fasci- basic and clinical epileptology at McGill University in Montreal
nation to W. G. Lennox than did cerebral blood flow. in his epochal collaboration with Wilder Penfield (discussed
Twelve children with petit mal epilepsy were the clinical sub- later). Lee Travis gradually disappeared from the scene but his
jects for the petit mal epilepsy study of Gibbs and Davis (1935) foremost disciples, John R. Knott and Charles E. Henry (Ph.D.
and Gibbs et al. (1935, 1937). This work remains evergreen in electroencephalographers with strong clinical inclinations)
the entire EEG literature; hardly any EEG finding has left such were bound to assume a very important role in America’s EEG
an indelible impression as the association of petit mal absences work. Their ultimate skill and supreme dedication turned them
and 3/sec spike-wave complexes. (Of course, it was found out into the “conscience of EEG,” steering developments into the
later that spike waves could occur without petit mal.) While right direction and correcting the course when there was dan-
Berger was gripped by the rhythms, Frederic Gibbs came under ger of going astray. D. Lindsley became one of the pioneers in
the fascination of paroxysmal patterns such as spike waves. the investigation of maturational EEG aspects; at the University
Shortly afterward, the EEG patterns of grand mal and psy- of California at Los Angeles, he directed excellent neurophysio-
chomotor seizures were reported by the same team (Gibbs, logic EEG research.
Lennox, and Gibbs), but the stretches of fast spikes (in grand This was the first wave of American EEG pioneers and their
mal) and the rhythmical activity in 4 or 6/sec frequency (in psy- immediate disciples and followers. It is impossible for the his-
chomotor seizures) were no match in popularity for the 3/sec torian to do justice to the second wave, which started before the
spike waves of petit mal. decade of the 1930s was over. There was Robert Schwab at
The technical quality of the EEG tracings shown in these Harvard and at the Massachusetts General Hospital in Boston,
studies left much to be desired. Dr. and Mrs. Gibbs traveled to in whom the mastery of EEG was combined with great clinical
Germany in the summer of 1935, paid a visit to Hans Berger, neurologic talents (especially in the field of myasthenia gravis,
spent some time at the Berlin-Buch Institute, and studied the parkinsonism, and epilepsy). Across the Charles River, at the
“polyneurograph” instrument of Jan F. Toennies; they also saw Massachusetts Institute of Technology, there was Warren
the instrumentation of Matthews in England. Frederic Gibbs McCulloch, a fiery genius like Grey Walter in Bristol and a pro-
then contracted Albert Grass (then at the Massachusetts found thinker. His scope would range light years beyond the
Institute of Technology) to build a three-channel preamplifier. limits of EEG and neurophysiology. (One must read his
In 1935, the Grass Model I went into use; it had three channels Embodiments of Mind to fathom his greatness, even though one
and an ink writer that recorded on rolls of paper (the folded may be inclined to disagree in many points.) He and Grey
paper not yet being in use). Walter lived in the world of brain machines, but there was
The Gibbs–Gibbs–Lennox era of the 1930s proved to be per- still a niche for a psyche (when one tries to read between the
haps the most exciting period in the history of EEG. In those lines). Earlier at Northwestern University in Evanston, Illinois,
years, EEG found the domain of greatest effectiveness: the outside Chicago, McCulloch had been involved with Dusser de
realm of the epileptic seizure disorders. Epileptology can Barenne in “neuronographic” work, an import from Utrecht,
be divided historically into two periods: before and after the Netherlands; this work was based on topical strychnine poison-
advent of EEG. Insights into the nature of the epileptic mecha- ing of the cortex and exploration of transmitted spiking to other
nisms deepened, not in a subtle manner but with a huge leap. regions.
What Fischer had started in 1931 with his experimental studies Clinical EEG research already started to conquer certain
on picrotoxin and its effect on the cortical EEG in animals, the fields outside epileptology. Grey Walter’s discovery of the delta
Gibbses and Lennox applied to human epileptology, and a wide focus (Walter, 1936) located over hemispheric brain tumors
door was flung open for the work of future decades. It is true had opened the search for further relationships between brain
that Berger in his seventh report (Berger, 1933) had shown a lesions and focal EEG correlates; metabolic disturbances and
10 Part I ■ Basic Principles

especially hypoglycemia were explored with EEG. (The work of (or Canadian) EEG laboratories started to become more
H. Hoagland and his coworkers dates back to 1937.) sophisticated than that of their European counterparts because
With the end of the 1930s, North America found itself in a of the inclusion of sleep.
leading position in the domain of EEG, whereas progress in Frederic Gibbs enjoyed enormous international prestige at
Europe was quite limited. that time as the world’s leader in clinical electroencephalogra-
phy. Nevertheless, his position at Harvard was much less presti-
World War II and the 1940s gious; he held the academic rank of an instructor (below the
During World War II, from 1939 to 1945, research and clinical professorial ranks), even though a visit to his laboratory was the
EEG activities were not flourishing, particularly in Europe. goal of European colleagues (who could afford the trip). This
There were some neurologic units where the EEG was used in disproportion drastically shows the negative attitude toward
the localization of traumatic brain lesions and epileptogenic EEG in neurologic departments (not universally, of course).
foci. After World War II, the gap between North America and Robert Schwab did not fare much better at Harvard in spite of
Europe was bigger than ever before, and European EEG his fine clinical neurologic talents.
research found itself at a low point. Toward the end of the 1940s, Herbert H. Jasper turned into
After the war, new activities started in England and France, a strong competitor of Frederic Gibbs. Jasper had moved to the
while the situation in Germany looked desperate. Neurologic Institute of McGill University in Montreal, joining
W. Grey Walter with his associates V. J. Dovey and H. Shipton forces with Wilder Penfield, a neurosurgeon with a profound
(a brilliant electronic engineer who later moved to Iowa City neuroscientific background. We discuss the rise of the Montreal
and then St. Louis) at the Burden Institute in Bristol discovered group in section Developments in the 1950s.
the paroxysmal response to flickering light at critical frequen- Two new developments started in the late 1940s. The EEG
cies between 10 and 20/sec. Further work on epileptic photo- technique started to become invasive and, with the use of spe-
sensitivity immediately shifted from Bristol to Marseille, cial depth electrodes, the exploration of deep intracerebral
France, where a young and incredibly talented Henri Gastaut regions began. This is discussed in section Developments in the
used this method, in combination with intravenous dosages of 1950s. Automatic frequency analysis also started in the 1940s,
pentylenetetrazol, to determine the “seuil épileptique,” that is, but this development reached loftier heights in the 1960s.
the individual threshold for paroxysmal responses (Gastaut, A discussion of the 1940s would be incomplete without a
1949; Gastaut et al., 1948). brief glance at the work of the neurophysiologists. A large seg-
In 1947, the American EEG Society was founded and the ment of neurophysiologic work was dominated by the use of
First International EEG Congress was held in London; a second EEG. One of the most fascinating results of these researchers
one followed in 1949 in Paris (in association with clinical neu- was the demonstration of thalamocortical relationships, thus
rology and other neurologic disciplines). EEG activities in far explored solely with anatomical methods (e.g., the study of
Germany were still minimal; Japan, however, gained attention the thalamus by A. Earl Walker in 1938, which propelled this
by the work of K. Motokawa, a researcher of EEG rhythms. young neuroscientist to great fame for decades to come). The
Switzerland started to develop its own profile; the neurophysi- work of Morison and Dempsey (1942) on the recruiting
ologist Marcel Monnier was instrumental in this regard. W. R. response had great impact on the neuroscientific world with the
Hess, however, a Nobelist, had gained great prestige by the func- demonstration of cortical responses to relatively slow stimula-
tional mapping of thalamus and hypothalamus with regard to tion of the intralaminar structures of the thalamus in the cat.
autonomic responses to electrical stimulation. This work emphasized the role of the thalamus in the cortical
The American scene was bustling with activities. Frederic A. electrogenesis and broke the ground for the concept of a “cen-
Gibbs with his coworkers Erna Gibbs and B. Fuster from trencephalic epilepsy,” a concept promoted by Penfield and
Uruguay produced another epochal study on the interictal Jasper in Montreal (somewhat naively understood as a concept
anterior temporal spike or sharp wave discharge in the inter- of the thalamic origin of primary generalized epilepsy).
seizure interval in patients with psychomotor seizures. This was Even greater was the impact of the work of Horace W.
an important step in the elucidation of temporal lobe epilepsy, Magoun, who had studied the effects of descending and mostly
a work with far-reaching consequences for the entire develop- inhibitory influences of the brainstem reticular formation dur-
ment of EEG laboratories and their routine work. It was found ing his work at Northwestern University. Together with G.
(Gibbs et al., 1948) that the anterior temporal discharges were Moruzzi (a fine neurophysiologist from Pisa, Italy, and investi-
often limited to the state of sleep. This observation meant that gator of basic epileptic mechanisms), Magoun subsequently
a tracing without a sleep portion could be insufficient, uninfor- studied the ascending system of the brainstem reticular forma-
mative, and even misleading. Thus, EEG laboratories would tion (chiefly in the midbrain level) and the effect of high-
include sleep in most (if not all) of their EEG evaluations. This frequency electrical stimulation, consisting of EEG
required pasted electrodes (rather than rubber bands or caps), desynchronization and behavioral arousal, on cortical function.
a much longer recording time, and a much smaller numerical Magoun, who had moved to the University of California at Los
output of recordings per technician (incidentally, an evolving Angeles, subsequently investigated the effects of acute lesions
profession, which is discussed later). made in the midbrain-level reticular formation in cats. These
Transatlantic communication was poor at that time, and cats remained in a comatose state with EEG slowing in spite
it was at this point when the routine work in American of electrical brainstem stimulation because of the destruction
Chapter 1 ■ Historical Aspects of EEG 11

of the all-important ascending portion of the brainstem reticu- effects. At that time, local anesthesia was still widely used in
lar formation (Lindsley et al., 1949). It is no exaggeration when neurosurgery. Jasper was chiefly a neuroscientist and not
one describes the effect of these studies on the world of neuro- merely an electroencephalographer. The book entitled Epilepsy
science as a “bombshell.” For the ensuing 10 to 15 years, the and the Functional Anatomy of the Human Brain (Penfield and
association of consciousness with reticular formation and Jasper, 1954) was a result of this fruitful collaboration.
Magoun’s name was so strong that it even had considerable Very controversial, however, was a concept of the primary
influence on the Pavlovian dogmatism of the Eastern Bloc generalized form of epilepsy characterized by generalized syn-
countries. Nowadays, however, even talented young neuroscien- chronous paroxysmal EEG discharges and exemplified by the
tists react to Magoun’s name with a blank expression—sic transit 3/sec spike waves of petit mal absences. Penfield and Jasper
gloria mundi! listed these epilepsies as “centrencephalic” with the concept of
The reason for discussing this experimental work in a histor- “center of the encephalon” (i.e., “thalamic midline structures”)
ical overview is to demonstrate the incredibly powerful role of serving as the starting point of the bilateral discharges. Henri
EEG in the neurophysiology of the 1940s. This was a high Gastaut from Marseille would follow the lead and so did many
watermark. Subsequently, experimental EEG work started to others, but Frederic Gibbs and a host of other electroen-
concentrate on single neurons while the “macro-EEG” gradu- cephalographers and neuroscientists became detractors of the
ally declined. centrencephalic concept. It wasn’t until the late 1960s that it
became clear that the centrencephalic concept stood on a very
Developments in the 1950s shaky ground and was ripe for being dismantled. Montreal’s
This is the last decade presented for historical analysis. Our story own Pierre Gloor helped to do this in a cautious and diplomatic
is gradually approaching the present, and a historical outline manner; others buried the centrencephalic concept more
must shy away from events that occurred over the last 40+ years. bluntly.
It does not behoove the historian to place living and active col- Frederic Gibbs had moved to the University of Illinois
leagues into the focus of discussion (with few exceptions). School of Medicine in Chicago (where full professorship was
The 1950s was the decade when EEG became a household given to him instantly after Harvard had denied him any pro-
word. During the early stretch of the decade, almost every uni- motion for more than a decade). Chicago—especially the
versity (teaching) hospital had at least one EEG machine. At the University of Chicago but also Northwestern University and the
end of the decade, EEG apparatuses had found their way into a University of Illinois—had become a world leader in neurologic
large number of other hospitals and even into private practice. sciences over the past 20 years. Percival Bailey, Paul Bucy, Roy
At university hospitals, central as well as departmental EEG lab- Grinker, A. Earl Walker, Gerhardt Von Bonin, C. J. Herrick,
oratories emerged. The latter were usually limited to children or Frederic Gibbs, and many others give testimony to the glory of
adults, and pediatric EEG units evolved (while specialized neona- neurologic science in Chicago at the middle of that century. In
tologic EEG units followed suit about 10 years later). Some psy- the field of EEG, the Chicago group under the Gibbses and the
chiatric departments took particular pride in their clinical and Montreal group under Jasper and Gloor were strong rivals
research-oriented EEG work. It is absolutely true that psychiatry throughout the 1950s, especially with respect to leadership in
was always “nice” to the electroencephalographer. Psychiatry’s epileptologic electroencephalography. One of the greatest mas-
domain was in need of organic or neurophysiologic substrata of ters from the Chicago school, A. Earl Walker, came to Johns
disorders and dysfunctions of psychiatric–psychological nature. Hopkins in Baltimore in 1947, introducing depth EEG, electro-
What could the electroencephalographer give in return? It was corticography, epilepsy surgery, and a scientifically oriented
very little, but the psychiatrists did not seem to mind. On the epileptology to his new place. Walker, who in 1972 had moved
other hand, there was so much to give to neurology. At that to the University of New Mexico and died in 1995, will always
time, it had become clear that the majority of diseases affecting be remembered as one of the great scholars of neurosurgery
the central nervous system (CNS) had more or less impressive and epileptology.
EEG correlates, but the majority of neurologists remained The 1950s saw a strong comeback of the Europeans. Henri
either reserved or hostile to EEG. Neurosurgeons were inter- Gastaut’s intellectual brilliancy was hard to match. In Marseille,
ested as long as EEG could contribute to the determination of disciples of great stature flocked around him, especially Robert
focal cerebral lesions (and before EEG became overpowered by Naquet, Joseph Roger, and Annette Beaumanoir, to mention
noninvasive neuroimaging techniques). Some epilepsy-ori- only the earliest nucleus of this group. At the great world cen-
ented neurosurgeons like W. Penfield or A. Earl Walker ters of neurology, Salpêtrière Hospital in Paris and National
remained interested in EEG and its use in the depth of the cere- Hospital, Queen Square, London, Antoine Remond and
brum or on the cortex. William Cobb, respectively, represented the EEG, but unfortu-
The epileptologic EEG work of Herbert Jasper in collabora- nately too much in the shadow of the leading neurologists.
tion with Wilder Penfield reached new heights, and Montreal Remond later turned into a protagonist of computerization of
reigned supreme as the place for neurosurgical treatment of EEG data.
focal epilepsies. Penfield was far more than a neurosurgeon. His The star neurologists of both Queen Square and Salpêtrière
operations for the removal of epileptogenic foci and, in a later lived in the world of classical neurology, which gave them so
phase, large portions of affected lobes were associated with elec- much satisfaction and happiness that one could hardly expect
trical stimulation and a systematic study of the behavioral their openness for the world of brain potentials. The Queen
12 Part I ■ Basic Principles

Square guard appeared to be more detached from EEG than Signals in a Large Irregular Background,” George D. Dawson
their Parisian confrères, perhaps due to the fact that Gastaut, from the National Hospital, Queen Square, in London demon-
the man from Marseille, came from the neurologic ranks to strated evoked potentials to electrical stimulation of the ulnar
achieve instant stardom with his EEG achievements. Probably nerve (Dawson, 1951). This required advanced analog technol-
no other famous neurologist has expressed his opinion about ogy. Thus, Dawson became the father of evoked potential stud-
EEG more scathingly than Francis M. R. Walshe has done. He is ies, which developed into a major outcropping of
the brilliant Queen Square star who apparently knew every- electroencephalography, eventually constituting a field of its
thing about neurology except EEG. To Sir Francis, the electrical own. The ingenious superimposition method of Dawson was
activity of the brain was, “a bloodless dance of action potentials eventually superseded by the advent of computerized averaging
… hurrying to and fro of its molecules” (after Critchley, 1990). methods in the 1960s.
Let us assume, for everyone’s benefit, that Sir Francis had meant Computational techniques of wave analysis started early in
it to be a joke. the history of EEG. First attempts were made by Hans Berger
Fine schools of EEG developed in the Netherlands with O. (1932); he was assisted by the physicist Dietsch (1932), who
Magnus (son of a Nobel Prize winning physiologist) in applied Fourier analysis to short EEG sections. Further work in
Wassenaar and Storm van Leeuwen in Utrecht. In Switzerland, this field was produced by Grass and Gibbs (1938) and Knott
Rudolf Hess (also son of a Nobelist physiologist) created an and Gibbs (1939). At the Massachusetts Institute of Technology
important school of electroencephalographers at the Zurich near Boston, Guillemin applied Fourier analysis to communica-
University Hospital. Giuseppe Pampiglione, from Italy, initiated tion theory, and one of his students was Albert Grass who
pediatric electroencephalography at the Hospital for Sick “could not wait to get the Gibbs interested” (Grass, 1984). The
Children in London, concurrently with William Lennox’s work 1950s saw the early generation of automatic frequency analyz-
at the Boston Children’s Hospital. ers approaching and eventually saw the end of these magnifi-
These European centers of EEG activities had rather an cent but mostly unused machines.
international flavor with strong North American influence. Eventually, the EEG branched out into the world of single
This cannot be said about the evolving field of clinical EEG in neurons, and the microelectrode technique was introduced in
West Germany, which was dominated by its prestigious leader the early 1950s. Microelectrodes can be made of metal such as
Richard Jung in Freiburg. Jung’s greatness pertained to experi- tungsten with tips of 1- to 3- m diameter; glass electrodes
mental neurophysiology; he also had great interest in the clini- filled with electrolytes such as KCI have tips of 0.5 m or even
cal fields and even in philosophy. This renaissance man smaller. Because of their characteristics, microelectrodes reach
designed the outline for EEG training and the routine of the very high impedance values (1–60 M ), which render conven-
EEG laboratory—unfortunately not without shortcomings, tional EEG recording techniques unsuitable. The introduction
which hamstrung the further development of clinical EEG in of the cathode follower by Toennies created the technical pre-
West Germany. requisite for single-cell recordings.
The 1950s also saw EEG sprouting into related fields. Extracellular microelectrode recording was used on a larger
Depth electroencephalography with implanted intracerebral scale in the early 1950s (Jung et al., 1952; Li et al., 1952;
electrodes was used in the humans for the first time by Moruzzi, 1952). About 10 years later, extracellular microelec-
Meyers and Hayne (1948) and Knott et al. (1950) at the trode studies were even done intraoperatively in humans. Far
University of Iowa, Iowa City, and also by Hayne et al. more revolutionary was the introduction of the extremely labo-
(1949a) in Chicago. These short recordings served the study rious intracellular microelectrode technology (Brock et al.,
of EEG activity in the human basal ganglia and thalamus 1952, in the spinal cord; Phillips, 1961, in the cortex). This tech-
with regard to basal ganglia dyskinesias and epilepsy. In the nique opened the gates to a new world of biochemical
following years, deep structures were also explored in processes. These insights taught lessons in humility to the elec-
patients with psychiatric disorders until doubt was cast upon trophysiologic neurophysiologist. There was no doubt that the
the ethical basis of this invasive approach (see Chapter 33, chemical changes were of primary significance, while the elec-
“Depth Electroencephalography”). In the 1960s, depth EEG trical phenomena were more or less by-products.
would find its true field in epileptic patients considered can- We cannot leave the 1950s without mentioning epochal
didates for epilepsy surgery. developments in the field of sleep research. At the University of
The origin of intraoperative electrocorticography dates back Chicago, N. Kleitman stood out as one of the world’s leading
to Foerster and Altenburger (1935). How was it possible that investigators of the organization of sleep. This institution pro-
Otfried Foerster, perhaps the greatest clinical neurologist ever duced the first study of rapid eye movement (REM) sleep
and an amazing self-taught master of neurosurgery, failed to (Aserinsky and Kleitman, 1953), but it must be pointed out that
recognize the future potential of EEG? Did his mind work Blake and Gerard (1937) described a “null stage” in the EEG of
mainly in the world of Sherringtonian concepts? Most of the nocturnal sleep, thus indicating the desynchronization of EEG
work in electrocorticography remains associated with the in REM sleep but without observation of the accompanying
Montreal Neurologic Institute and the names of Penfield and ocular, muscular, and other autonomic changes. William C.
Jasper (also see Chapter 34, “Electrocorticography”). Dement continued the work of Kleitman and, following his
The related fields of EEG started to bloom in the 1950s. In a move from Chicago to Stanford, became a world leader in the
study entitled “A Summation Technique for Detecting Small study of nocturnal sleep.
Chapter 1 ■ Historical Aspects of EEG 13

Sleep research gradually became based on polygraphic somatosensory evoked potentials is associated with the names
recording, and its share in the overall EEG research declined. of Roger and Joan Cracco.
This development led to a constantly widening gap between The 1960s and 1970s witnessed a regrettable alienation of
EEG and nocturnal sleep research (in the 1960s and the follow- EEG and epileptology, which had existed before in an almost
ing decades). perfect marriage. A sizable number of epileptologists lost inter-
est in EEG. Was it early enthusiasm about the introduction of
The Rest of the Story antiepileptic serum levels? Was the path to mastery of EEG
The last 30 years of the history of EEG and related fields can be becoming too laborious? This situation changed in the 1980s
gleaned directly from this book. The events of the 1960s, 1970s, due to the rapidly increasing emphasis on EEG and related
and 1980s are just too close for us to see with the eyes of the his- techniques in the presurgical workup of patients considered
torian. Nevertheless, modern trends are briefly discussed in this candidates for seizure surgery.
final section. The 1970s and 1980s saw brilliant structural neuroimaging
The development of clinical and experimental EEG work techniques emerging: computed tomography and magnetic res-
reached a high point around 1960 after 30 years of steady onance imaging. This seemed to knock out electroencephalog-
progress. There is no doubt that the 1960s slowed down the raphy from the contributors to focal CNS diagnosis. Such a
smooth progress. The interest of electroencephalographers in knockout blow, however, was also more apparent than real. The
academic institutions tended to shift from the tracing, with all EEG, by its very nature, never was a structure-oriented test.
its waves and patterns, to automatic data analysis. Whether the patient has a hemispheric brain tumor, a vascular
Computerization was the direction—tendencies reaching back lesion, or a traumatic contusion, EEG can always demonstrate
to Berger’s coworker Dietsch (1932), but flourishing in the the degree of dysfunctional changes around the lesion (or sec-
1960s and 1970s. This development had many positive aspects. ondary diffuse cerebral dysfunction). The sad story is that the
The names of Barlow, Brazier, Remond, Lopes da Silva, neurologist of our day seems to lose interest in the realm of
Bickford, Saltzberg, Dumermuth, Matousek, D. O. Walter, function and dysfunction. This development must be halted
Cooper, Künkel, Lehmann, Gasser, Burch, Hjorth, Schenk, (and eventually will be).
Matejcek, and Low should be mentioned in this context. In par- Topical EEG diagnosis, however, has made a comeback of its
ticular, Cooley and Tukey (1965) have been credited with the own in the form of computerized brain mapping. This fascinat-
introduction of the fast Fourier transforms as the basis of ing recent development is associated chiefly with the name of
power spectral analysis. Frank Duffy. Again, it is the old EEG in new clothes: it can be
This work led us into a “brave new world” of EEG comput- understood only by an expert of the conventional EEG.
erization and, as early as in 1967, we were told that custom- Starting in the late 1960s, a completely new development
ary EEG reading would soon be a thing of the past, replaced took place in the EEG exploration of the full-term and the pre-
by a fully automatic EEG interpretation. The fears of many mature newborn. The historical aspects of this development are
clinical electroencephalographers were unfounded; nobody found in Chapters 11 and 49.
became jobless, because such an automatization of EEG read- This historical overview is not complete without a few words
ing was fictional. It was found that EEG is far too complex for about the technicians (technologists) doing the EEG laboratory
such automation. Its interpretation requires the wonderful routine work. They had to place electrodes with greatest accu-
computer that is located between the ears—the educated and racy and to obtain a readable tracing, even under the most
experienced brain. One simply must consider that the meth- adverse conditions. They were considered the electroencephalo-
ods of those years—frequency or time domain—were limited grapher’s attendants for a long time, even though their work
to an analysis of frequencies. Automatic spike detection had required considerable sophistication.
barely reached its earliest stage. John R. Knott and Charles E. Henry invested incredible energy
The electroencephalographer needed to be aware that all into the founding of the American Society of EEG (later
types of data computerization were nothing but the EEG in Electroneurodiagnostic) Technologists, which came into being in
disguise—“an analog of an analog.” Computerized frequency 1962. Soon afterward, the first group of technicians underwent a
analysis was here to stay and to prove to be of enormous value stiff examination that made them registered EEG technologists.
not only in psychophysiologic research but also in the assess- There have been similar developments in many other countries.
ment of neuropharmacologic effects. This evolution has been helpful in giving EEG technologists the
In the 1970s, the evoked potential technique progressed dignity they deserve, but this process is far from being completed.
greatly. The introduction of the pattern changer in the visual With every record we read, we must be thankful for the work of
evoked potential technique made this method highly reliable; our technical staff and invest some of our energies in their con-
the names of H. Speckrejse and R. Spehlmann ought to be men- tinuous education and training, for the sake of better technical
tioned in this context. In the field of auditory evoked potentials, EEG quality and thus for the sake of our patients.
the location of primary cortical discharge was elusive for many This historical overview remains a fragmented account
years, and the late vertex potential of limited clinical value. The because much remains untold in this story. However, it is hoped
introduction of the far-field technique for the demonstration of that the historical perspective this chapter brings to the reader
the brainstem auditory evoked potentials (Jewell), however, will foster in our ranks insights that may help us avoid the mis-
proved to be extremely valuable. Analogous work in the field of takes of the past.
14 Part I ■ Basic Principles


PRESENT AND FUTURE Major Historical Works
Clinical Electroencephalogram Brazier MAB. A History of the Electrical Activity of the Brain. The First
Half-Century. London: Pitman; 1961.
The role of clinical electroencephalography has been diminish- Critchley M. The Ventricle of Memory. New York: Raven Press; 1990.
ing throughout the past 30 years except for epileptology, which Gloor P. Hans Berger on the Electroencephalogram of Man. Amsterdam:
has been using the tool of EEG video monitoring to its full Elsevier; 1969.
extent. This produces an enormous accumulation of data and Grass AM. The electroencephalographic heritage. Am J EEG Technol.
needs thorough analysis and interpretation—a highly time- 1984;24:133–173.
consuming task that is often “farmed out” or subcontracted to Hassler R. Cécile und Oskar Vogt. In: Kolle K, ed. Grosse NervenŠrzte.
outside readers of sometimes unacceptable EEG expertize. Vol. 2. Stuttgart: Thieme; 1959:45–64.
Presence, number or absence of spikes, and spike-related dis- Jung R. Hans Berger und die Entdeckung des EEG nach seinen
Tagebüchern und Protokollen. In: Werner R, ed. Jenenser EEG-
charges are not the only criteria; full justice must be done to all
Symposium: 30 Jahre Elektroenzephalographie. Berlin: VEB Verlag
EEG abnormalities of nonparoxysmal character, and this task is Volk und Gesundheit; 1963:20–53.
a lot more difficult. Such expensive long-term recordings should Kolle K. Hans Berger. In: Kolle K, ed. Grosse NervenŠrzte. Vol. 1.
be either artfully interpreted or not done at all. Stuttgart: Thieme; 1956:1–6.
Acutely requested EEG on an emergency basis nowadays Mazza S, Pavone A, Niedermeyer E. Mario Gozzano: the work of an
deals mainly with the question of “rule out status epilepticus” EEG pioneer. Clin Electroencephalogr. 2002;33:155–159.
(Varelas et al., 2003). With presently widespread confusions con- Niedermeyer E. Benjamin Franklin and static electricity.
cerning the limitations of nonconvulsive status epilepticus (see Considerations of past, present, and future. Am J End Technol.
Chapters 28 and 29), the answer needs profound neurologic– 2003a;43:26–29.
epileptologic understanding. Cases of acute cerebral anoxia O’Leary JL, Goldring S. Science and Epilepsy. New York: Raven Press;
(postcardiopulmonary arrest) may cause both convulsive as 1976.
Schwab RS. Electroencephalography. Philadelphia: WB Saunders; 1951.
well as nonconvulsive pictures with massive paroxysmal EEG
Upton M. Electronics for Everyone. New York: New American Library of
abnormalities but do not represent true status epilepticus World Literature (Signet Key Books); 1960.
(see Chapter 23). Werner R. Hans Berger zum Gedächtnis. In: Werner R, ed. Jenenser
Outside the confines of epileptic conditions, the EEG is full EEG-Symposium: 30 Jahre Elektroenzephalographie. Berlin: VEB
of information for the vast majority of neurologic diseases. It is Verlag Volk und Gesundheit; 1963:13–19.
most regrettable that this rich source of information has been Work of Specially Mentioned EEG Pioneers
badly underused in most neurologic teaching institutions and Adrian ED. The Berger rhythm in the monkey’s brain. J Physiol.
hospitals in general (Niedermeyer, 2003b). 1936;87:83P–84P.
Adrian ED, Bronk DW. The frequency of discharge in reflex and volun-
The Basis of Electroencephalogram tary contractions. J Physiol. 1929;67:119–151.
The cellular basis of EEG activity has been the topic of intensive Adrian ED, Matthews BHC. The interpretation of potential waves in
studies of extracellular current flow and voltage-dependent the cortex. J. Physiol. 1934;81:440–471.
Aserinsky W, Kleitman N. Regularly occurring episodes of eye motility
intrinsic oscillations (see Buzsaki et al., 2003). Ongoing work
and concomitant phenomena during sleep. Science. 1953;118:
on cortical and subcortical generators is found in Chapter 3
(also see Ebersole, 2003). This shows that the genesis of EEG is Bartley SH. Analysis of cortical response to stimulation of the optic
still a widely discussed issue. nerve. Am J Physiol. 1932;101:4P.
Bartley SH, Newman EB. Recording cerebral action currents. Science.
Electroencephalogram and Neurocognition
This area has become the perhaps most fascinating aspect of Bartley SH, Newman EB. Studies on the dog’s cortex. Am J Physiol.
modern EEG interest (see Chapter 51). It is true that most of 1931;99:1–8.
this type of research has been done with other tools such as Berger H. Über das Elektrenkephalogramm des Menschen. 1st report.
functional magnetic resonance imaging, positron emission Arch Psychiat Nervenkr. 1929;87:527–570.
tomography (PET) scanning, and single-photon emission com- Berger H. Über das Elektrenkephalogramm des Menschen. 4th report.
puted tomography (SPECT) (methods demonstrating regional Arch Psychiat Nervenkr. 1932;97:6–26.
blood flow and metabolic needs). With the extension of EEG Berger H. Über das Elektrenkephalogramm des Menschen. 7th report.
Arch Psychiat Nervenkr. 1933;100:301–320.
into the ultrafast frequency ranges (Chapter 37), a powerful
Bishop GH, Bartley SH. Electrical study of the cerebral cortex as com-
upswing of EEG-oriented neurocognitive research in animals pared to the action potential of excised nerve. Proc Soc Exp Biol
and humans can be expected. Neurocognition has assumed the (New York). 1932;29:698–699.
position of the “Holy Grail” of all neuroscience, and EEG stands Blake K, Gerard RW. Brain potentials during sleep. Am J Physiol.
a great chance to become a principal contributor. After years of 1937;119:692–703.
frantic efforts, it will be found that neuroscience can illuminate Bremer F. Cerveau isolé et physiologie du sommeil. C R Soc Biol
the “brain-mind barrier” only to a certain degree. For the time (Paris). 1935;118:1235–1241.
being, however, let us not disturb the glowing enthusiasm with Bremer F. Cerebral and cerebellar potentials. Physiol Rev. 1958;38:
Cassandra calls. 357–388.
Chapter 1 ■ Historical Aspects of EEG 15

Brock LG, Coombs JS, Eccles JC. The recordings of potentials from Hayne R, Meyers R, Knott JR. Characteristics of electrical activity of
motor neurons with an intracellular electrode. J Physiol. 1952;117: human corpus striatum and neighboring structures. J Neurophysiol.
431–460. 1949b;12:185–195.
Buzsaki G, Traub R, Pedley TA. The cellular basis of EEG activity. In: Hoagland H, Rubin MA, Cameron DF. The electroencephalogram of
Ebersole JS, Pedley TA, eds. Current Practice of Clinical schizophrenics during insulin hypoglycemia and recovery. Am J
Electroencephalography. 3rd ed. Philadelphia: Lippincott Williams Physiol. 1937;120:559–570.
and Wilkins; 2003:1–11. Jung R, Baumgarten RV, Baumgartner G. Mikroableitungen von
Caton R. The electric currents of the brain. Br Med J. 1875;2:278. einzelnen Nervzellen im optischen Cortex der Katze. Die lich tak-
Cooley JW, Tukey JW. An algorithm for the machine calculation of tiviertcn B-Neurone. Arch Psychiat Z Ges Neurol. 1952;189:
complex Fourier series. Math Comp. 1965;19:267–301. 521–539.
Danilevsky VD. Investigations into the Physiology of the Brain. Doctoral Knott JR, Gibbs FA. A Fourier transform of the electroencephalogram
thesis. University Charkov (cited in Brazier, 1961), 1877. from one to eighteen years. Psychol Bull. 1939;36:512–513.
Davis H, Saul LV. Action currents in the auditory tracts of the midbrain Knott JR, Gibbs FA, Henry CE. Fourier transforms of electroen-
of the cat. Science. 1931;86:448–450. cephalogram during sleep. J Exp Psychol. 1942;31:465–477.
Dawson GD. A summation technique for the detection of small signals in Knott JR, Hayne RA, Meyers HR. Physiology of sleepwave characteris-
a large irregular background. J Physiol (London). 1951;115:2P. tics and temporal relations of human electroencephalograms
Dietsch G. Fourier-Analyse von Elektrenkephalogrammen des recorded from the thalamus, the corpus striatum and the surface of
Menschen. Pflugers Arch Ges Physiol. 1932;230:106–112. the scalp. Arch Neurol Psychiatry (Chicago). 1950;63:526–527.
Ebersole JS. Cortical generators and EEG voltage fields. In: Ebersole JS, Kornmüller AE. Architektonische Lokalisation bioelektriseher
Pedley TA, eds. Current Practice of Clinical Electroencephalography. Erscheinungen auf der Grosshirnrinde. 1. Mitteilung:
3rd ed. Philadelphia: Lippincott Williams and Wilkins; 2003:12–31. Untersuchungen am Kaninchen bei Augenbelichtung. J Psychol
Fischer MH. Elektrobiologische Auswirkungen von Krampfgiften am Neurol. 1932;44:447–459.
Zentralnervensystem. Med Klin. 1933;29:15–19. Kornmüller AE. Die Ableitung bioelektischer Effekte architektonischer
Fischer MH, Löwenbach H. Aktionsströme des Zentralnervensystems Rindenfelder vom uneröffneten Schadel. J Psychol Neurol.
unter der Einwirkung von Krampfgiften. 1. Mitteilung Strychnin und 1933;45:172–184.
Pikrotoxin. Arch Exp Pathol Pharmakol. 1934a;174:357–382. Kornmüller AE. Der Mechanismus des epileptischen Anfalles auf
Fischer MH, Löwenbach H. Aktionsströme des Zentralnervensystems Grund bioelektrischer Untersuchungen am Zentralnervensystem.
unter der Einwirkung von Krampfgiften. 2. Mitteilung: Cardiazol, Fortschr Neurol Psychiatry. 1935;7:391–400, 414–432.
Coffein und andere. Arch Exp Pathol Pharmakol. 1934b;174: Kornmüller AE. Die Bioelektrischer Erseheinungen der Hirnrindenfelder.
502–516. Leipzig: Thieme; 1937.
Foerster O, Altenburger H. Elektrobiologische Vorgänge an der Kornmüller AE. Klinische Elektrenkephalographie. Munich: Lehmann;
mensehlichen Hirnrinde. Dtsch Z Nervenheilk. 1935;135:277–288. 1944.
Gastaut H. Effets des stimulations physiques sur l’E.E.G. de l’homme. Kornmüller AE. Die Elemente der Nervšsen TŠtigkeit. Stuttgart:
Electroencephalogr Clin Neurophysiol Suppl. 1949;2:69–82. Thieme; 1947.
Gastaut H, Roger J, Corriol JH, et al. Les formes expérimentales Lennox WG. The oxygen and carbon dioxide content of blood from
d l’épilepsie humaine. L’épilepsie induite par la stimulation the internal jugular and other veins. Arch Intern Med.
lumineuse intermittente ou épilepsie photogénique. Rev Neurol 1930;46:630–636.
(Paris). 1948;80:161–183. Lennox WG. The cerebral circulation. Arch Neurol Psychiatry
Geddes LA. What did Caton see? Electroencephalogr Clin Neurophysiol. (Chicago). 1931;26:719–724.
1987;67:2–6. Lennox WG, Gibbs EL. The blood flow in the brain and the leg of man,
Gerard RW, Marshall WH, Saul LJ. Cerebral action potentials. Proc Soc and the changes induced by alteration of blood gases. J Clin Invest.
Exp Biol (New York). 1933;30:1123–1125. 1932;1:1155–1177.
Gibbs EL, Gibbs FA. Diagnostic and localizing value of electroen- Li CL, Jasper HH, McLennan H. Décharge d’unités cellulaires en rela-
cephalographic studies in sleep. Publ Assoc Res Nerv Ment Dis. tion avec les oscillations électriques de l’écorce cérébrale. Rev
1947;26:366–376. Neurol (Paris). 1952;87:149–151.
Gibbs EL, Fuster B, Gibbs FA. Peculiar low temporal localization of Liberson WT. Recherches sur les électroencéphalogramme tran-
sleep-induced seizure discharges of psychomotor epilepsy. Arch scraniens de l’homme. Travail Hum. 1937;5:431–463.
Neurol Psychiatry (Chicago). 1948;60:95–97. Lindsley DB, Bowden JW, Magoun HW. The effect of subcortical
Gibbs FA, Davis H. Changes in the human electroencephalogram asso- lesions upon the electroencephalogram. Am Psychol. 1949;4:
ciated with loss of consciousness. Am J Physiol. 1935;113:49–50. 233–234.
Gibbs FA, Davis H, Lennox WG. The electroencephalogram in epilepsy Loomis AL, Harvey EN, Hobart GA. Potential rhythms of the cerebral
and in conditions of impaired consciousness. Arch Neurol cortex during sleep. Science. 1935;82:198–200.
Psychiatry (Chicago). 1935;34:1133–1148. Meyers HR, Hayne R. Electrical potentials of the corpus striatum and
Gibbs FA, Gibbs EL, Lennox WG. Epilepsy paroxysmal cerebral dys- cortex in Parkinsonism and hemiballism. Trans Am Neurol Assoc.
rhythmia. Brain. 1937;60:377–388. 1948;73:10–14.
Goltz FL. Ueber die Verrichtungen des Groszhirns. Pflueger’s Arch Ges Morison RS, Dempsey EW. A study of thalamo-cortical relations. Am J
Physiol. 1888;42:419–467. Physiol. 1942;135:281–292.
Grass AM, Gibbs FA. A Fourier transform of the electroencephalo- Moruzzi G. L’attività dei neuroni corticali durante il sonne e durante la
gram. J Neurophysiol. 1938;1:521–526. reazione elettroencefalografica di risveglio. Ricerca Sci. 1952;22:
Hayne R, Belinson L, Gibbs FA. Electrical activity of subcortical 1165–1173.
areas in epilepsy. Electroencephalogr Clin Neurophysiol. Moruzzi G, Magoun HW. Brain stem reticular format and activation of
1949a;1:437–445. the EEG. Electroencephalogr Clin Neurophysiol. 1949;1:455–473.
16 Part I ■ Basic Principles

Niedermeyer E. The clinical relevance of EEG interpretation. Clin Varelas PN, Spanaki MV, Hacein-Bey L, et al. Emergent EEG.
Electroencephalogr. 2003b;34:93–98. Neurology. 2003;61:702–704.
Penfield W, Jasper HH. Epilepsy and the Functional Anatomy of the Vogt C, Vogt O. Sitz und Wesen der Krankheiten im Lichte der topistis-
Human Brain. Boston: Little, Brown; 1954. chen Hirnforschung und des Variierens der Tiere. Leipzig: Barth;
Phillips CG. Some properties of pyramidal neurones of the motor cor- 1937.
tex. In: Wolstenholme GEW, O’Conner M, eds. The Nature of Sleep. Walter WG. The location of brain tumors by electroencephalogram.
Boston: Little, Brown; 1961:4–24. Proc R Soc Med. 1936;30:579–598.
Rothmann H. Zusammenfassender Bericht über den Rothmann’schen Walter WG. Slow potential waves in the human brain associated with
groszhirnlosen Hund nach klinischer und anatomischer expectancy, attention and decision. Arch Psychiat Nervenkr.
Untersuchung. Z Ges Neurol Psychiat. 1923;87:247–313. 1964;206:309–322.
Sherrington CS. Integrative Action of the Nervous System. New Haven, Walter WG, Dovey VJ, Shipton H. Analysis of electrical responses
CT: Yale University Press; 1906. of the human cortex to photic stimulation. Nature. 1946;
Travis LE, Dorsey JM. Action current studies of simultaneously active 158:540–541.
disparate fields of the central nervous system of the rat. Arch Neurol Walter WG, Cooper R, Aldridge VJ, et al. Contingent negative varia-
Psychiat. 1932;28:331–338. tion. An electric sign of sensorimotor association and expectancy in
Travis LE, Herren RY. The relation of electrical changes in the brain to the human brain. Nature. 1964;203:380–384.
reflex activity. J Comp Psychol. 1931;12:23–29. Yakovlev PI. Vladimir Mikhailovich Bekhterev (1857–1927). In:
Travis LE, Knott JR. Brain potential studies. I. Perseveration time to Haymaker W, ed. The Founders of Neurology. Springfield, IL:
light. J Psychol. 1936;3:97–100. Charles C Thomas; 1953:244–247.
Neurophysiologic Basis of
EEG and DC Potentials
he clinical electroencephalographer correlates central numbers. Thus, nerve cells are usually covered with several
nervous system (CNS) functions as well as dysfunctions thousand synapses (2). The glia cells are imbedded between
and diseases with certain patterns of the electroen- nerve cell somata, dendrites, and axons. They usually have sev-
cephalograhy (EEG) on an empirical basis. Obviously, this eral processes that make contact with somata and processes of
method has been found valuable in clinical practice. Therefore, nerve cells; they may also make contact with vessels. This histo-
why should the clinical electroencephalographer study the basic logic arrangement results in a cerebral extracellular space con-
elementary processes underlying the EEG? There is little doubt sisting of very narrow intercellular clefts (3).
that the range of EEG interpretations can be much widened and
misinterpretations avoided when the underlying elementary NEURONAL MEMBRANE POTENTIALS:
processes are also considered. This is true especially for convul-
sive disorders and cerebral metabolic disturbances. For exam-
ple, an isoelectric EEG can be caused by selective pCO2 increase Essential potentials that can be demonstrated with intracellular
while the brain is sufficiently supplied with O2. On the other recordings are characterized briefly. When the membrane of the
hand, in the presence of practically normal pCO2 levels, cere- nerve cell body is penetrated by a microelectrode, a potential of
bral hypoxia may be the cause. It will be pointed out below that about 60 to 70 mV with negative polarity in the intracellular
the prognosis may be quite different in these two cases. space can be recorded. This membrane potential is subject to
various fluctuations that are elicited chiefly by synaptic activi-
POTENTIAL GENERATION AT NEURONAL ties. Their mechanisms are shown in greater detail in Figure 2.2.
AND GLIAL ELEMENTS: MEMBRANE As can be derived from this schematic illustration, the neuron
The basic mechanisms that give rise to potentials recorded out-
side the CNS elements will be described. Such extracellular
potentials are generally known as field potentials (1).
In the course of this presentation, the morphology of gen-
erator structures is discussed briefly. Then, the electrical activ-
ity demonstrable with intracellular recordings from neurons
and glia cells is described. On the basis of this information,
the principles of the generation of extracellular field poten-
tials are outlined and the various types of field potentials are

The CNS essentially consists of nerve cells and glia cells. The
arrangement of neurons usually shows a specific type of lami-
nar character. Glia cells are located between neurons. As shown
in Figure 2.1, several processes emerge from the nucleus-con-
taining cellular soma (body) of the nerve cell. These processes
can be divided into two types according to their function. Most
of the processes are dendrites that branch off into numerous
small ramifications. Every cell also has an axon that may split
up into multiple collaterals. Such an axon provides contact with
other nerve cells or with other target organs. In the case of
interneuronal connections, the contact consists of synapses that Figure 2.1 Schematic drawing of morphology and histology of neu-
cover the dendrites, the soma, and the axon hillock in large ronal and glial elements.

18 Part I ■ Basic Principles

contribute primarily to the generation of the extracellular field

potentials in question (1,5,7,8). For this reason, the ionic
mechanisms of these potentials are discussed in greater detail.
The individual events of this process are presented with a mag-
nified time base (Fig. 2.3). With the elicitation of an EPSP, a net
inflow of cations occurs across the subsynaptic membrane. As
shown in Figure 2.2B, a potential gradient develops along the
neuronal membrane in the intra- and extracellular spaces.
Because of this potential gradient, cations move along the
nerve cell membrane through the extracellular space in the
direction of the subsynaptic region. An inversely directed flow
takes place in the intracellular space. With the generation of an
IPSP, there is an outflow of cations from the nerve cell and/or
an inflow of anions into the nerve cell. These changes first
increase the membrane potential at the subsynaptic membrane
in comparison with the surrounding segments of the mem-
brane. For this reason, a potential gradient develops along the
nerve cell membrane, as in the case of the EPSP genesis. This
potential gradient causes, in the extracellular space, a flow of
cations from the subsynaptic region to the surrounding por-
tions of the membrane. An inverse process develops in the
intracellular space (5).
The ion fluxes in the extracellular space are of paramount
significance in the generation of field potentials. Therefore,
these processes are further discussed in the following chapters.


Figure 2.2 Membrane potential (MP) changes and current flows dur-
In addition to the neurons, glial cells may also play a role in the
ing synaptic activation. A: The MP of the postsynaptic neuron and the
generation of extracellular field potentials (9,10). Therefore, the
MP of the presynaptic fibers are recorded by means of intracellular
bioelectric properties of glial cells are summarized.
microelectrodes. Action potentials in the excitatory and inhibitory
If a glia cell is penetrated with a microelectrode, a membrane
presynaptic fiber lead to excitatory postsynaptic potential (EPSP) and
potential can be recorded with a polarity similar to that of the
inhibitory postsynaptic potential (IPSP), respectively, in the postsynap-
nerve cells. The size of this membrane potential approximates
tic neuron. Two EPSPs sum up to a superthreshold potential, triggering
the potassium equilibrium potential and hence somewhat
an action potential in the postsynaptic neuron. B: During EPSP and
exceeds the membrane potential of nerve cells. In contrast to
IPSP, ionic current flows occur through as well as along the neuronal
neurons, glial cells fail to show any action potentials, and there
membrane, as shown by arrows. The density of and signs indicate
are also no postsynaptic potentials. Thus, in contrast to neu-
the polarization of the subsynaptic (dark area) as well as that of the
rons, glial cells do not show characteristic potentials that distin-
postsynaptic membrane during synaptic activation.
guish them unmistakably from other cells. The glial membrane
potential, however, is also not constant. An augmentation of the
extracellular potassium concentration (potassium activity)
from which the soma membrane potential is recorded has causes depolarization of glial cells (Fig. 2.4A). Concentration
synaptic connections. The corresponding presynaptic struc- changes of other ions cause only negligible alterations of the
tures are also explored with microelectrodes. If an action poten- glial cell membrane potential. The glial cell is hence compara-
tial travels along the fiber, which ends in an excitatory synapse, ble to a potassium electrode (9,11).
an excitatory postsynaptic potential (EPSP) occurs in the The dependency of the glial membrane potential on the
following neuron (Fig. 2.2A). If two action potentials travel extracellular potassium concentration is the reason for a
along the same fiber with a short interval, there will be a sum- functional linkage with adjacent neuronal structures.
mation of EPSP triggering an action potential on the postsy- Neuronal activity is associated with outflow of potassium
naptic neuron after reaching the membrane threshold. If an ions. As shown schematically in Figure 2.4B, repetitive firing
action potential travels along a fiber ending in an inhibitory of neurons gives rise to increased extracellular potassium
synapse, then hyperpolarization will occur, representing an concentration and hence to glial cell depolarization (12,13).
inhibitory postsynaptic potential (IPSP) (4–6). If the potassium concentration does not affect the entire glial
Because of the time course of the various membrane poten- cell membrane and remains increased only locally, then
tial fluctuations, the postsynaptic potentials are thought to potential gradients build up along the glial cell, giving rise to
Chapter 2 ■ Neurophysiologic Basis of EEG and DC Potentials 19

Figure 2.3 Basic mechanisms underlying

generation of potentials (electroencephalogra-
phy [EEG]) and of magnetic fields (magnetoen-
cephalography [MEG]) in the extracellular space
of central nervous system. The description is
based on the assumption that an extended neu-
ronal process, for example, a dendrite, is locally
depolarized by activation of an excitatory synapse.

intra- and extracellular current flows similar to the ones extension may develop on the basis of the aforementioned
described in reference to neuronal synaptic transmissions mechanisms (1,10,14,15). In view of the above-described
(Fig. 2.2). Glial cells frequently have widespread processes functional interconnections, it is quite likely that in the gen-
and furthermore may have close connections with each other. esis of extracellular field potentials an amplifying effect can
For this reason, potential fields of considerable spatial be attributed to the glial cells.

Figure 2.4 Membrane potential (MP) changes of glia cells induced by an increase in the extracellular K concentration
(arrows in the schematic drawings). A: Potassium is applied extracellularly to the glia cell. B: The potassium concentra-
tion is increased due to an activation of a neighboring neuron. (From original tracings from Kuffler SW, Nicholls JG, Orkand
RK. Physiologic properties of glial cells in the central nervous system of amphibia. J Neurophysiol. 1966;29:768–787.)
20 Part I ■ Basic Principles

It has been shown in the preceding section that primary trans-
membranous currents generate secondary ionic currents along
the cell membranes in the intra- and extracellular spaces. The
portion of these currents that flows through the extracellular
space is directly responsible for the generation of field poten-
tials (Fig. 2.3). Particular significance must be ascribed to the
synaptic processes as causing events for the field potentials in
question, especially for their time course. In accordance with
these statements, the generation of extracellular field potentials
will be discussed as exemplified by extracellular fields accompa-
nying synaptic activity (5,8,15,16). The discussion of these
events will again make use of a very protracted time axis
(Fig. 2.3). The explanation of the events is given in reference to
the schematic view in Figure 2.5. This figure shows a widely
stretched neuronal element, with one end segment lying close
to the surface of a central nervous structure. At both ends of
this neuronal unit, the microelectrodes ME1 and ME2 are
inserted. At the same time, the extracellular electrodes E1 and E2
are located at the surface and at the deeper end of the neuronal
element. The potentials picked up from the intra- and extracel-
lular electrodes are shown in the vicinity of each electrode. The
potential recorded from the surface of the nervous structure is
accentuated by thicker lines. Figure 2.5 shows active excitatory
and inhibitory synapses, either close to the surface or located in
the depth. As described elsewhere, the activation of an excita-
tory synapse leads to a net inward flow of cations. If this state-
ment is applied to Figure 2.5A1, then it becomes evident that Figure 2.5 Membrane potential (MP) changes and field potentials
the upper end of the neuronal element will be depolarized in (FPs) elicited by the activation of excitatory and inhibitory synapses
comparison with other segments of the same cell. Accordingly, in the central nervous system. The elementary processes are explained
the synaptic current flow causes an EPSP at the microelectrode by means of a neuronal element (hatched area), the one end of which
ME1. This local depolarization then gives rise to further intra- contracts the surface of a structure in the central nervous system. The
and extracellular ionic currents along the nerve cell membrane. MP of the neuron element is recorded at both ends by the microelec-
Because of the intracellular movements of positive charges, trodes ME1 and ME2. The extracellular field is picked up at the surface
depolarization in the area of microelectrode ME2 also takes of the neuronal structure by the electrode E1, as well as in the vicin-
place. This depolarization, however, is less steep and of smaller ity of ME2 by the electrode E2. Active excitatory and inhibitory
amplitude. At the superficially located extracellular electrode synapses are marked by open triangles and black triangles (S), respec-
E2, the inflow of positive charges into the neuronal element tively. A1: The inward current at S generates an EPSP that appears in
causes a negative field potential. The extracellular electrode E2 the region of ME1, as well as in that of ME2. Because S is located
is, metaphorically speaking, approached by positive charges so superficially, the FP generated, due to the direction of the extracellu-
that a positive field potential will develop in this area. The point lar current flow (arrows), is of negative polarity at the surface (E1)
of reversal of the field potentials is localized between electrodes and of positive polarity in the deeper recording (E2). A2: The activa-
E1 and E2. The exact position of the point of reversal depends tion of a deep excitatory synapse elicits a current flow with inverse
on the distribution of extracellular impedances. direction as compared with A1. Therefore, the extracellular FP con-
Current flows of reversed direction (in reference to the sists of a positive deflection at the surface and a negative one at the
recording electrodes) will occur if the active excitatory synapse depth. B1: The outward current at S generates an IPSP in the region
is located at the deeper end of the neuronal element (Fig. of ME2, as well as in that of ME1. Due to the direction of the extracel-
2.5A2). In this case, positive charges approach the superficially lular current flow, the FP generated consists of a positive fluctuation
located electrode (E1) (again speaking metaphorically) and in the depth (E2) and a negative one in the surface recording (E1).
remove themselves from the deeply located electrode (E2). This B2: The current flow during the activation of a superficial inhibitory
arrangement of the active synaptic structures causes a positive synapse is inverse as compared with B1. Therefore, the FP recorded
field potential at the surface and a negative one at the deep elec- from the surface consists of a positive fluctuation. Differences in the
trode. The current flows accompanying the activation of time course of the various potentials are caused by the electrical prop-
inhibitory synapses located in deeper and in more superficial erties of the tissue.
areas, respectively, are shown in Figure 2.5B. As can be derived
from this illustration, the activation of a deep inhibitory
Chapter 2 ■ Neurophysiologic Basis of EEG and DC Potentials 21

synapse (Fig. 2.5B1) produces a current flow that is largely sim-

ilar to the one generated by the activation of a superficial exci-
tatory synapse (Fig. 2.5A1). In the same manner, there are also
similar current flows in the extracellular space when a superfi-
cial inhibitory synapse (Fig. 2.5B2) or a deeply located excita-
tory synapse (Fig. 2.5A2) is activated. Accordingly, a negative
field potential will develop at the surface of a central nervous
structure (in the schematic view of Fig. 2.5) whenever a super-
ficial excitatory or a more deeply located inhibitory synapse is
activated. The corresponding principle applies to generation of
the superficial field potentials of positive polarity.
Due to their higher densities especially the currents flowing
through the intracellular spaces give rise to magnetic fields sur-
rounding the elements. These magnetic fields can be picked up
from the scalp as magnetoencephalography (MEG; Fig. 2.3).

Types of Field Potentials

The field potentials, whose generation has been described, can
be subdivided into different types. If field potentials are
recorded against an inactive reference point with an upper fre-
quency limit of about 100 Hz, then two types of field potentials
can be distinguished, depending on the time constant of the
amplifying recording device. In the case of a time constant of
1 second or less, the extracellular field potentials correspond
with that which is commonly known as the EEG. If the record-
ing is carried out with an infinite time constant, that is, with
direct current (DC) amplifier, then slower potentials can also be
picked up. Potentials recorded with this technique are generally
known as DC potentials (1,8,15,17). Thus, DC potentials com-
prise slow as well as fast field potentials. The fast components
correspond with the potential fluctuations of the EEG. Due to
different time constants, however, the faster potential compo-
nents may differ from each other as far as their time course is
concerned when recordings are done either with conventional
EEG amplifiers or with DC amplifiers.
Thus far, technical problems have made it difficult to carry
out DC recordings from the scalp. Except for special areas of
application, DC recordings are usually performed in animal
experiments. DC potentials directly reflect the state of activity
of central nervous cells and therefore contribute to the explana- Figure 2.6 Recording of neuronal membrane potentials using voltage-
tion of the mechanisms of genesis of cerebral field potentials sensitive dyes, and principles and schematic example of application.
(14,18). For this reason, DC potentials will be discussed jointly A: (1) A dye is incorporated into the double lipid membrane of nerve
with EEG waves. cells and illuminated by light with dye-specific wave length; simultane-
For the sake of comparison, Figure 2.6 shows the EEG and ously the membrane potential is recorded with an intracellular micro-
the DC potentials during convulsive activity, hypercapnia, and electrode against a reference electrode in the extracellular space.
asphyxia. As shown in this illustration, a tonic–clonic convul- (2) Changes of the membrane potential (MP) starting from the resting
sion is associated with a negative DC shift (8,13–15,19–21). level passing a decrease (depolarization) with action potentials superim-
Furthermore, it can be seen that the hypercapnia-induced dis- posed and a subsequent increase (hyperpolarization) and eventually
appearance of the EEG is associated with a monophasic positive returning to resting level. (3) In correspondence to the different MP levels
DC shift. In the case of EEG extinction due to primary fluorescence and absorption of the dye change. With fluorescent dyes a
asphyxia, however, there are characteristic patterns of DC fluc- depolarization is associated with a decrease and a hyperpolarization with
tuation. Hence, similar findings in the conventional EEG may an increase of fluorescence (symbols). B: (1) Two neurons in a popula-
be associated with different DC shifts. tion; one stays in the resting state and the other changes its MP as in
A2. (2) By the aid of a microscope and a connected array of diodes
Original Remarks Made by Ernst Niedermeyer (squares) the different MP changes of both neurons can be detected via
What does the term DC shift mean? What is DC? Speaking from the different optical behaviors. (Speckmann E-J. Das Gehirn meiner Kunst.
experience, many electroencephalographers have no clear Kreativität und das selbstbewußteGehirn. Münster: Daedalus; 2008.)
22 Part I ■ Basic Principles

concept regarding DC potentials or DC shifts. One cannot

blame them because, for strange reasons, “DC” has two mean-
ings in this context:
1. DC means direct current (and this is, of course, common-
place): a current without oscillations; a current derived from
a battery source; a current maintained in one direction
through a circuit. A more imperfect DC is produced by a rec-
tifier, used to change alternating current (AC) into DC. For
multilingual readers, DC is courant continu in French,
Gleichstrom in German, and corrente continuo in Italian.
Electroneurophysiologically, DC shifts are ultraslow poten-
tials, about as slow as 0.1 to 0.2/sec. This, however, is not true
DC. Such slow activity is just a bit more “DC-like” since it
does not show the faster “AC-like” activity. One simply has to
live with this kind of misnomer.
2. DC also means direct coupling (and this is much less
known). What coupling? The coupling between the stages of
EEG amplification. Conventional EEG machines have stages
coupled by capacitors. Now one has to remember that capac-
itors (a) reject DC and (b) determine the time constant.
Even a very long time constant (several seconds duration)
may not suffice for the recording of DC potentials. Direct
coupling is a capacitor-free coupling between the stages of Figure 2.7 Simultaneous detection of membrane potentials (MP) of all
amplification and provides the optimal condition for DC neurons in a population (voltage-sensitive dye) and conventional
recording. This is technically quite difficult in clinical con- recording of the local field potential (FP) at the same time. Living brain
ventional EEG recording but easier under experimental neu- tissue (0.5 mm thick) from the temporal lobe of a patient who under-
rophysiologic conditions in animals.
went epilepsy surgery. A: Recording of the local FP (“local EEG”).
Hence, be aware of the dual significance of the term DC. (1) Resting state; (2 to 4) epileptic discharges of different intensities.
Epileptic discharges appeared spontaneously, that is, they were not
Validation of Field Potentials induced experimentally. B: MP changes indicated by the intensity of flu-
As to Their Functional Significance orescence of the dye (black, decrease of the MP, depolarization).
Correlating field potentials with other signals accompanying Resting state (1); similar epileptic potentials in the FP (2 and 3) are
neuronal activity may contribute to a validation of them. There associated with different extents of neuronal depolarizations and simi-
is a variety of methods other than electroencephalography for lar extents of neuronal depolarizations with different epileptic poten-
detecting brain activity. Thus, single photon emission tomogra- tials in the FP (3 and 4). (Speckmann E-J. Das Gehirn meiner Kunst.
phy (SPECT), positron emission tomography (PET), functional Kreativität und das selbstbewußteGehirn. Münster: Daedalus; 2008.)
magnetic resonance imaging (fMRI), and intrinsic optical imag-
ing (IOI) are based on metabolic changes associated with
increases of local neuronal activity. The latter methods are “very Figure 2.7. The living brain slices are stained with fluorescence
indirect” (22). EEG including evoked potentials (EP) and MEG (or absorption) dyes (Fig. 2.7A1). With depolarization and
represent “more direct ones” since they measure the field effects hyperpolarization the fluorescence is decreased and increased,
of the proper neuronal activity and therewith of the information respectively (Fig. 2.7A2 and A3). The changes in fluorescence are
processing brain activity. For the analysis of neuronal network measured via a microscope by an array of detectors and there-
functions, the immediate and simultaneous recording of mem- with the actual membrane potentials of the neurons are
brane potentials of all neurons in a population by application of observed (Fig. 2.7B1 and B2).
voltage-sensitive dyes is the “only direct” method available yet The method of simultaneous measurement of neuronal
(23–28). All these methods have advantages and disadvantages. membrane potentials of all neurons in a population is success-
Thus, the functional imaging using voltage-sensitive dyes cannot fully applied in living human brain slices (0.5 mm thick) in
be applied in patients for several reasons, for example, prerequi- vitro obtained from neurosurgical interventions (tumor and
site of direct access to the brain structure to be investigated, pho- epilepsy surgery) (25–27,29).
totoxicity, and pharmacologic side effects of the dyes. But, this A comparison of the field potentials, that is, the local EEG,
method is helpful to analyze the principles of functional mean- and of the neuronal membrane potentials detected by the aid
ing of field potentials in living human brain slices in vitro, espe- of voltage-sensitive dyes is given in Figure 2.8. The tissue is a slice
cially with spontaneously occurring epileptic discharges. preparation from the temporal neocortex resected from a patient
A principle and schematic example of recording neuronal suffering from pharmacoresistant complex partial seizures. Most
membrane potentials using voltage-sensitive dyes is displayed in of these living human brain slices show spontaneous epileptic
Chapter 2 ■ Neurophysiologic Basis of EEG and DC Potentials 23

Figure 2.9 Principles of wave generation. The excitatory synapses of

two afferent fibers contact the superficial dendritic arborization of two
longitudinal neuronal elements. The afferent fiber activity is recorded by
means of the intracellular electrodes E1 and E2, and the membrane
potentials (MPs) of the dendritic elements are recorded by the electrodes
Figure 2.8 EEG (time constant: 1 second; upper frequency limit: E3 and E4. The field potential at the surface of the neuronal structure
100 Hz) and DC/ EEG recordings (DC recording: upper frequency limit, (cortex) is led by the electrode E5. Synchronized groups of action poten-
100 Hz) during a generalized seizure induced by pentylenetetrazol tials in the afferent fibers (E1, E2) generate wavelike EPSPs in the den-
(A), during hypercapnia (B), and during asphyxia (C). Original record- dritic areas (E3, E4) and corresponding field potentials in the EEG and
ings were obtained from cats and rats. Note the different time scales. DC/ EEG recording (E5). Tonic activity in the afferent fibers results in a
long-lasting EPSP with small fluctuations. During this period the EEG
(5b) shows only a reduction in amplitude, whereas the DC/ EEG record-
EEG potentials, that is, epileptic discharges not induced by exper- ing (5a) reflects the depolarization of the neuronal elements as well.
imental procedures. One can derive the following from the
microelectrodes E1 and E2 are located in the ascending fibers and
1. During epileptic discharges only a certain portion of the neu-
the microelectrodes E3 and E4 are in the superficial dendrites of
rons in the population is active simultaneously, that is, a com-
the postsynaptic neurons. To pick up the extracellular field
plete synchronization is missing (Fig. 2.7, numbers 2 to 4).
potentials, the electrode E5 lies on the surface of the central
2. Similar epileptic potentials in the EEG (Fig. 2.7, numbers
nervous structure.
2 and 3) are associated with different extents of neuronal
As shown in tracings 1 and 2, action potentials occur synchro-
depolarizations and similar extents of neuronal depolar-
nously in the afferent fibers. There are grouped discharges that
izations with different epileptic potentials in the EEG
are temporarily supplanted by tonic activity. The ascending
(numbers 3 and 4).
action potentials elicit individual EPSP in the upper dendrites of
the neurons; these EPSPs are subsequently summated into major
WAVE GENERATION depolarizations in accordance with the discharge frequency. As
shown in tracings 3 and 4, amplitude and duration of the depo-
In the preceding sections, the generation of single field poten- larizations depend on the discharge pattern of the afferent fibers.
tials was described. In this section, the principles of the genera- The synaptic activity at the superficial structures gives rise to
tion of wavelike potential fluctuations are outlined. This is extracellular current flows resulting in superficial field potentials.
followed by the discussion of the laminar distribution of such With the use of DC recording techniques, the superficial field
potentials in the cerebral cortex. potentials reflect the potential fluctuations of the dendritic mem-
To present the generation of wavelike potential fluctuations brane. If, however, the superficial field potentials are recorded
on the surface of a central nervous structure, a simple model as with a time constant of 1 second or less, then only the fast fluctu-
shown in Figure 2.9 is used. This model consists of two extended ations of the superficial field potentials are demonstrable.
pyramidal neurons of vertical orientation. Terminals of afferent Thus far, the principles of genesis of EEG and DC waves
fibers make contact with the superficial dendrites of both neu- have been shown in the schematic view of Figure 2.9.
rons via excitatory synapses. The bioelectrical activity of these Accordingly, the generation of physiologic EEG waves may be
structures is recorded with intracellular microelectrodes. The explained as follows. If a grouped and synchronous influx
24 Part I ■ Basic Principles

takes place in afferent fiber systems toward the superficial

generator structures, then EEG waves evolve that are of high
amplitude and distinctly separated from each other. In case
of a periodic sequence of the afferent bursts, the recording of
the field potentials shows sinusoidal potential fluctuations.
This mechanism has been presumed by several groups of
investigators as the principle of the generation of the alpha
rhythm and slower periodic EEG waves. According to these
workers, thalamocortical feedback loops are believed to play
a significant role in the generation of the alpha rhythm
(see Chapters 3 and 4 for detailed descriptions of the genera-
tion of rhythmic activities) (1,30).
If the afferent influx of impulses occurs at a high frequency
for a longer period and/or synchronously, then negative field
potentials with small fluctuations will result from the extracel-
lular current flows. Accordingly, the EEG recording will pick
up only waves of smaller amplitude and mostly higher fre-
quency. In the DC recording, however, the prolonged depolar-
ization of the superficial structures caused by the afferent
high-frequency influx will express itself by a negative DC
potential shift (31,32). There is a close correlation between the Figure 2.10 Surface (1) and laminar recordings (2 to 6) of EEG waves
amplitude of the negative DC shift and average discharge fre- of the cortex. The schematic drawing symbolizes conical neuronal ele-
quency in the afferent fiber systems. This mechanism may ments densely packed with synapses. (Drawings from original tracings
apply principally to the generation of beta activity and other obtained in experiments in the rat’s motor cortex during pentobarbital
EEG waves of higher frequencies (see also cortical sources of anesthesia.)
beta and gamma rhythms in Chapter 3 and Fig. 3.16). A
decrease of the amplitudes of the EEG waves can also occur
when the afferent activity is diminished. In this case, however, CORTICAL FIELD POTENTIALS
the depression of EEG waves is accompanied by a positive DC DURING EPILEPTIFORM ACTIVITY
shift (see Fig. 2.17) (33).
The principles of generation of individual and wavelike In the following subsections, the generation of cortical field
field potentials at the surface of central nervous structures potentials during convulsive activity is discussed. The first sub-
such as the cerebral cortex have been described. If the wave- section deals with focal activity, and the second discusses gener-
like potential fluctuations are recorded not only from the alized tonic–clonic convulsive activity. For methodical reasons,
cortical surface, but also from different cortical layers, then it we refer to data derived from experimental work in animals.
can be shown that potential fluctuations in the latter record-
ings may differ considerably from those at the surface. Focal Activity
These differences imply polarity, frequency, and amplitude If a convulsive substance such as penicillin is applied to the sur-
(1,34,35). Such a recording from the cortex of the rat is shown face of the cerebral cortex, steep negative potentials of high
in Figure 2.10. According to this illustration, field potentials amplitude can be picked up from the area of application after a
reverse their polarity between electrode 1 (on the surface) and short latency period. These discharges repeat themselves in
electrode 2 (located 300 m beneath the cortical surface). Two stereotyped form and periodicity (Fig. 2.11A) (13,36,37). If the
and sometimes more of such phase reversals may be observed membrane potential of a cortical neuron is simultaneously
in deeper recording sites depending on the experimental con- recorded with a microelectrode while a second microelectrode
ditions (phase reversals of alpha oscillations are shown in picks up the corresponding field potentials, then potential fluc-
Chapter 3, Figs. 3.9 and 3.10). The vertical distribution type of tuations occur as shown in Figure 2.11B. It can be derived from
field potential will be discussed in greater detail in connection this illustration that the monotonously recurrent negative field
with the generation of cortical field potentials during convul- potentials are associated with equally stereotyped membrane
sive activity. potential fluctuations. These oscillations of the membrane
In the course of the discussion of cerebral field potentials, it commence with a steep depolarization that, having exceeded
was pointed out that particular significance must be attributed to the membrane threshold, triggers a series of action potentials.
synaptic activity. A view of the laminar distribution of neurons in This is followed by a plateau that, after 80 to 100 msec, changes
the cortex and the dense coverage of these unitary structures with into a steep repolarization, and frequently also into a hyperpo-
synapses makes it clear that different patterns of potentials must larization. These membrane potential fluctuations have proved
necessarily occur in different layers when populations of synapses to be characteristic in the epileptiform activity of individual
are activated in a different manner. This should be clarified by the neurons. They are generally known as paroxysmal depolariza-
schematic drawing in Figure 2.10. tion shifts (PDSs) (13,38).
Chapter 2 ■ Neurophysiologic Basis of EEG and DC Potentials 25

There are, however, field potentials with predominantly posi-

tive components in the deeper contacts 4 to 6. If penicillin is
applied to the surface without penicillinase, then negative
field potential will also develop in deeper cortical layers. If it
is assumed that the negative field potentials mirror the direct
epileptiform activity of neuronal structures (Fig. 2.11), then it
must also be assumed that deeper cortical elements are
involved in convulsive activity shown in of Figure 2.12B in
contrast with Figure 2.12A. This is further supported by the
observation that neuronal activity descending to the spinal
cord and producing characteristic spinal field potentials
occurs only under the experimental conditions shown in
Figure 2.12B. If one compares the recordings in Figure 2.12A
and B, it becomes clear that, with a monotonous epileptiform
Figure 2.11 EEG (A) and membrane potential (MP) A changes of a potential at the cortical surface, the intracortical potential dis-
pyramidal tract neuron and extracellular field potential (FP) recorded in tribution and the occurrence of descending activity may differ
the vicinity of the impaled neuron (B) during focal interictal activity considerably (34,39–44).
elicited by application of penicillin to the cortical surface (hatched area If penicillin is applied to deeper cortical laminae
in A). Drawings of original tracings from experiments in the rat. The (Fig. 2.12C), then negative field potentials will be confined to
sweep speed in B is five times that in A. The recording sites are shown that region. These potentials are consistently accompanied by
in the schematic drawings. descending activity to the spinal cord. Under these conditions,
there is frequently nothing but a positive potential fluctuation
of minor amplitude at the cortical surface (34,40).
Investigation of potential distribution within the cerebral In summary, it can be derived from the described experi-
cortex after the local application of penicillin yields a variety mental models that, in focal convulsive activity limited to the
of findings. An appropriate model is shown in Figure 2.12. In cortex, the surface potential does not necessarily reflect the bio-
this experiment, recordings of interictal field potentials were electrical events in deeper cortical layers.
carried out from the cortical surface, from inside the cortex,
and from the spinal cord. The spinal field potentials permit Generalized Tonic–Clonic Activity
the observation of electrical activity descending from the cor- Here, possible mechanisms involved in the generation of cortical
tex to the spinal cord. In Figure 2.12A, negative field poten- field potentials during tonic–clonic convulsive activity are
tials are recorded from the cortical surface and from the two described. Again, data are based on experimental observations in
upper intracortical contacts after the application of penicillin animals. Tonic–clonic convulsive activity was triggered by
together with penicillin-metabolizing enzyme penicillinase. repeated injections of pentylenetetrazol (also see Refs. 13 and 37).

Figure 2.12 Cortical field potentials recorded at the

surface (1) and from within the cortex (2 to 6) and
spinal field potentials (7) during interictal activity.
The interictal activity was elicited by penicillin.
A,B: Potential distribution after surface application of
the drug. In A, the spread of penicillin is limited by the
use of penicillinase. C: Potential distribution after intra-
cortical application of penicillin at recording point 4.
The areas directly involved in the epileptiform activity
as indicated by negative field potentials are marked by
hatching in the schematic drawings. Spinal field poten-
tials are linked to the occurrence of negative field
potentials in lamina V (B and C, 4). Distance between
the intracortical electrodes, 300 m. (From original trac-
ings from Elger CE, Speckmann EJ, Caspers H, et al.
Focal interictal epileptiform discharges in the cortex of
the rat: laminar restriction and its consequences for
activity descending to the spinal cord. In: Klee MR, Lux
HD, Speckmann EJ, eds. Physiology and Pharmacology
of Epileptogenic Phenomena. New York, NY: Raven
Press; 1981.)
26 Part I ■ Basic Principles

Figure 2.13 Simultaneous recordings of EEG and DC/ EEG (A) and of
DC/ EEG and membrane potential (MP) of a pyramidal tract neuron (B)
during generalized tonic–clonic seizures elicited by pentylenetetrazol.
(Drawings after original tracings from experiments in the cat’s motor
cortex. The sweep speed in B is 10 times that in A.) Figure 2.14 Single potential fluctuations at the cortical surface
(DC/ EEG) and concomitant membrane potential (MP) of a pyramidal
tract cell (PTC) and field potentials (FP) in the PTC layer during gener-
Figure 2.13A shows a tonic–clonic convulsion recorded with alized tonic–clonic seizures. The seizure activity was induced by
a conventional EEG amplifier, as well as with a DC amplifier. pentylenetetrazol. A: The negative potential (1), the positive–negative
There is a negative DC shift from the baseline during a convul- fluctuation (2), and the positive potential (3) in the DC/ EEG recording
sive seizure. This negative DC shift gradually recedes during the coincide with monophasic negative FP and stereotyped paroxysmal
termination of the convulsions and frequently changes into a depolarization shift in the neuron. The negative DC shift occurring dur-
transient positive after shift (13–15,18,19,21,41). ing the seizure is indicated by a dashed line in the upper row.
When the membrane potential of a pyramidal tract neuron Monophasic negative potentials in the DC/ EEG recording occur with
of lamina V is recorded during a convulsive seizure, it can be small and monophasic positive fluctuations along with a marked DC
shown that under these conditions typical PDSs become mani- displacement. B: The relations between DC/ EEG potentials and MP of
fest (Fig. 2.13B). If these PDSs are correlated with the potential PTC as described for A1 and A3 also hold true for trains of potentials
fluctuations in the DC recording, it can be noticed that the (1 and 2). (From original tracings from Speckmann EJ, Caspers H,
PDSs in pyramidal tract neurons are coupled at the beginning Jansen RWC. Laminar distribution of cortical field potentials in relation
of the convulsive seizure with superficial negative potential to neuronal field activities during seizure discharges. In: Brazier MAB,
fluctuations and at the end of the convulsive seizure with sur- Petsche H. eds. Architectonics of the Cerebral Cortex, IBRO Monograph
face-positive potential fluctuations (Fig. 2.13B) (1,18,45). Series. Vol 3. New York, NY: Raven Press; 1978:191–209.)
In addition to the field potentials of the cortical surface and
the membrane potentials of the pyramidal tract cells, field
potentials were also recorded in the fifth lamina. Under these when the negative DC shift at the cortical surface reaches and
conditions, it can be shown that every PDS is associated with a exceeds a critical value (1,18,45,46).
negative monophasic field potential in the depth (Fig. 2.14A). These data are interpreted with flow charts in Figure 2.15.
These stereotyped potential fluctuations in deep cortical layers The amplitude of the negative DC shift at the cortical surface
correspond with field potentials at the cortical surface with depends greatly on the amount of the afferent influx of
either monophasic negative or positive (Fig. 2.14A1 and A3) or impulses to the generator structures in the superficial cortical
with polyphasic (Fig. 2.14A2) configurations. This statement laminae. This predominantly asynchronous afferent influx is
does not merely apply to individual ictal potentials but is also symbolized by the width of hatched arrows in Figure 2.15.
true for prolonged trains of potentials during the convulsion. Accordingly, the afferent influx in Figure 2.15A is smaller than
As Figure 2.14B shows, paroxysmal depolarizations of pyrami- that in Figure 2.15B. Therefore, there is a smaller DC shift in
dal tract cells may be accompanied by a sequence of either neg- Figure 2.15A and a prominent one in Figure 2.15B. In the case
ative or positive potentials on the cortical surface. If one of Figure 2.15A, a synchronized inflow of impulses from sub-
correlates these various field potentials on the cortical surface cortical structures is assumed to reach the cortex (widened
with the slow DC shifts occurring during the convulsion (also afferent fiber in schematic view). As a consequence, pyramidal
see Fig. 2.14A), then it can be demonstrated that the surface- tract cells will be stimulated to generate a PDS, and structures
negative field potentials are associated primarily with a slight close to the surface will be depolarized through the mediation
DC shift and that surface-positive field potentials will appear of interneurons. Accordingly, in such a constellation of
Chapter 2 ■ Neurophysiologic Basis of EEG and DC Potentials 27

Figure 2.15 Flow charts of neuronal processes possibly responsible for the generation of DC/ EEG waves of opposite polar-
ity during a generalized tonic–clonic seizure. (Hatched arrows) Symbols for continuous asynchronous input to the cortex;
(heavy lines) symbols for phasic volleys giving rise to single convulsive discharges; (PTC) pyramidal tract cell; (IN)
interneuron; (MP) membrane potential; (UA) extracellularly recorded unit activity. A: During a moderate asynchronous
input to the cortex (small hatched arrow), a burst of UA triggers a paroxysmal depolarization shift in a PTC.
Simultaneously, it leads to a depolarization of superficial neuronal structures and therewith to a negative fluctuation in the
DC/ EEG recording at the cortical surface. B: With an increased asynchronous input to the cortex (wide hatched arrow),
the DC potential shifts to a more negative level than in A (1). When in these conditions a phasic volley reaches the cor-
tex, paroxysmal depolarization shifts are also triggered in PTC, whereas the enhanced asynchronous UA is interrupted
mainly due to inactivation. The latter process results in a disfacilitation of the upper neuronal structures and therewith to
a positive fluctuation of the superficial DC/ EEG potential (2). (From original tracings from Speckmann EJ, Caspers H,
Jansen RWC. Laminar distribution of cortical fluid potentials in relation to neuronal field activities during seizure dis-
charges. In: Brazier MAB, Petsche H, eds. Architectonics of the Cerebral Cortex. IBRO Monograph Series. Vol 3. New York,
NY: Raven Press; 1978:191–209.)

excitatory processes, the paroxysmal depolarization in the characteristic physiologic feature of SD is a negative DC poten-
depth will be coupled with a surface-negative field potential. tial shift with maximal amplitude of 5 to 30 mV and duration
With augmentation of the already existing afferent inflow of of 30 to 90 seconds (Fig. 2.16A) (47). SD recordings by extracel-
impulses, the interneurons involved will necessarily exhibit a lular microelectrodes inserted into the gray matter of the neo-
heightened level of excitation (Fig. 2.15B). If an additional cortex demonstrate that SD can be biphasic or triphasic with
highly synchronized afferent influx of impulses takes place the main negative component. SD begins with a first small pos-
under these conditions, then further PDSs will be triggered in itive component that is sometimes not observed. The main fea-
the pyramidal tract cells, but, in the interneurons, the previ- ture of the SD is a negative wave of high amplitude, which is
ously existing high-frequency activity will be temporarily inter- then followed by a positive-going wave of smaller amplitude,
rupted, chiefly due to inactivation. This causes a decline of the but longer duration. The white matter beneath the neocortical
excitatory inflow of impulses to the superficial cortical struc- gray substance becomes positive, while the gray matter itself
tures. This disfacilitation gives rise to a positive field potential undergoes the negative wave. Apical dendrites were preferen-
at the cortical surface. In this manner, a massive afferent inflow tially involved in the generation of the negative DC deflection.
of impulses provides the basis for a correlation of positive When recordings were made in hippocampal CA1 region,
epicortical field potentials with stereotyped paroxysmal depo- SD invariably started earlier and ended later in the layer of the
larizations and monophasic negative field potentials in the dendritic trees than among the cell body. SD flows in the oppo-
depth (18,45). site direction compared with the current underlying
tonic–clonic seizure discharges, which is inward in the neuron
SPREADING DEPRESSION soma layers (48,49).
A brief period of excitation heralds SD, which is immediately
In this section, the generation and propagation of cortical field followed by prolonged neuronal activity depression. This is
potentials during spreading depression (SD) is discussed. SD is replaced by a sustained increase in the neuronal activity; SD
a strong and rapid depolarization of neuronal tissues and prop- also produces transient suppression of synaptic transmission
agates slowly (3 to 5 mm/min) as a wave in brain tissue (47). that is replaced by a sustained increase in the efficacy of synaptic
The SD phenomenon is exclusive to the CNS and appears to transmission. SD significantly enhanced the amplitude of field
influence both the neuronal and the glial cells. The most EPSP following a transient suppressive period and increased the
28 Part I ■ Basic Principles

from a depolarization of both presynaptic and postsynaptic

processes. In addition, presynaptic depolarization evokes
Ca2 -dependent synaptic transmission and glutamate release.
The latter in turn enhances depolarization of postsynaptic ele-
ments and causes further increases in [K ]o. In addition to potas-
sium and proton release from the cells, sodium, calcium, and
chloride enter together with water causing cells to swell and the
volume of the extracellular compartment to be decreased. SD is
accompanied by an increase of glucose utilization and O2 con-
sumption. Recovery of SD depends on energy metabolism (51,52).
SD can be initiated by different stimuli. The most common
model of SD initiation is KCl application to the neuronal
tissues. Other methods of SD induction include mechanical,
electrical, and thermal stimulations, application of the excita-
tory amino acids such as glutamate and aspartate, the Na –K
ATPase inhibitor ouabain, and metabolic inhibitors such as
NaCN (53). The ionic processes responsible for the generation
of SD are complex. Several ionic currents are involved; both
high [K ] o and low[Ca2 ] o can support either SD or epilepti-
form discharges depending on the state of the various mem-
brane ion channels. Insight into the interplay of these different
factors can best be obtained by the study of computer models
solidly based on biophysical data. Thus, the model of Kager
et al. (2008) revealed how the ignition of SD can result from the
shift in the balance between slowly inactivating (persistent)
inward and outward currents in apical dendrites of pyramidal
Figure 2.16 Propagation of negative DC potential shift in the neocor- cells in favor of the former. The effects of hypoxia and anoxia,
tex and its effect on neuronal excitability. A: Recording of slow spread through changes in Na/K pump and glial buffering, can also be
of negative DC potential wave after injection of KCl in human neocorti- put in evidence with these models (54,55).
cal tissue. Voltage variations were recorded simultaneously by four Processes similar to SD in animal and human neocortex
electrodes (DC1 to DC4), which are set apart by 1 mm. KCl was injected observed in experimental studies are thought to take place in a
in layer VI. Injecting and recording electrodes arranged as shown. number of pathologic conditions in humans. SD belongs to the
B: Suppression of evoked potentials (EP) during spreading of negative domain of the pathophysiology of the brain, and there are rea-
DC potential fluctuation. The DC potential was recorded from layer III sons to believe that it is involved in different clinical disorders,
of a human neocortical slice. EP was elicited by a single electrical including migraine, cerebrovascular diseases, head injury,
stimu lation (ST) applied to afferent fibers. Note the transient suppres- spinal cord disorders, epilepsy, and transient global amnesia
sion of EP during propagation of spreading depression. (Modified from (51). SD can alter the neurochemistry of cortical and subcorti-
Gorji A. Spreading depression: a review of the clinical relevance. Brain cal structures, modulating oxygen distribution, cell survival in
Res Brain Res Rev. 2001;38:33.) these structures, and behavior. Direct alterations of electrical
activity of cortical neurons by the locally spreading wave can
lead to clinical symptoms (e.g., the aura phase of migraine).
The most common symptoms in the aura phase are visual, aris-
induction of long-term potentiation in the neocortex (50). The ing from dysfunction of occipital lobe neurons. The positive
prodromal excitation as well as late hyperexcitability is mani- (excitatory) neurologic symptoms, for example, flashing lights,
fested in appearance of a burst of multiple action potentials and are usually followed by negative (suppressive) ones, for exam-
intense synaptic noise representing synchronized firing of neu- ple, scotoma or hemianopia, in this phase. Subdural recordings
rons. The widely spread synchronization seems best explained in patients demonstrated that SD is critically involved in vari-
by electrical continuity that could be provided by gap junctions. ous disorders associated with acute neuronal injury including
Patent gap junctions may provide a pathway not only for electric traumatic and spontaneous intracerebral hemorrhage (56).
currents and ions, but also for intracellular “second” messengers
and other active ingredients in cytosol (48). CORTICAL FIELD POTENTIALS DURING
SD is associated with transient, but major, localized redistrib- GAS TENSION CHANGES IN TISSUE
utions of ions between the extra- and intracellular spaces. The
[K ] o rapidly rises and can reach levels as high as 40 to This section deals with the alterations of epicortical field poten-
80 mmol/L. This causes neuronal excitation followed by depolar- tials and concomitant changes of the membrane potentials
ization and a period of electrical silence during which the DC caused by deviations of the gas tension in brain tissue. Such
potential at the brain surface falls. Increases in [K ]o could result changes of the gas tension may occur when, for instance, the
Chapter 2 ■ Neurophysiologic Basis of EEG and DC Potentials 29

pulmonary and circulatory function is disturbed or when the With isolated increment of the CO 2 tension in the cerebral
local cerebral blood flow is inadequate. tissue by means of the apnea technique, the amplitude of the
First, the alterations of epicortical field potentials during conventional EEG decreases progressively. This amplitude
selective hypercapnia are discussed; then, those associated with reduction affects first the waves of higher frequency and then
primary asphyxia are considered. It is shown that EEG changes those of lower frequency. Prior to the extinction of normal
may be similar under both conditions. The cortical DC poten- EEG activity, there is once again a phase characterized by
tial, however, shows typical shifts that permit inferences con- high-frequency EEG activity in the range of 50 to 70 Hz
cerning the cause of the accompanying EEG changes. The (1,57). The extinction of the EEG is associated with a shift of
discussion of the effects of gas tension alterations on the bio- the DC potential in a positive direction. If the CO 2 tension is
electrical activity of the CNS is based, again, on data derived then lowered again by reventilation, the EEG waves return in
from experimental work in animals. the original spectral composition after a short latency. At the
same time, the positive DC shift resolves (Fig. 2.17).
Hypercapnia Experiments in animals have shown that, with reduction of
If the CO2 tension in the brain tissue is increased in a selective the pCO2, the EEG returns to normal activity even though the
manner, typical reactions of the cortical field potentials as well hypercapnia-induced suppression lasted for 1 hour or more.
as of the membrane potential and the postsynaptic potentials of In these cases, a positive DC deflection of monophasic charac-
individual neurons are found. These findings are shown in a ter was found to occur during the whole period of apnea
summarized schematic view in Figure 2.17. (33,57,58).
The animal experiments on which Figure 2.17 is based were Under the aforementioned conditions, the recording of the
carried out with the use of the so-called apnea technique. With membrane potential of a cortical nerve cell shows a hyperpolar-
this technique, interference of the effects of hypercapnia with ization while the CO2 tension is increased. Extensive experi-
simultaneous effects of hypoxia could be avoided. According to mental studies in animals have demonstrated that such a
this technique, the experimental animal is ventilated for at least hyperpolarization is caused primarily by a reduction of the
a half hour with pure oxygen. Thereafter, artificial ventilation is EPSP (Fig. 2.17; also see Ref. 58). Consideration of field poten-
discontinued while the trachea of the animal remains con- tials, membrane potentials, and EPSP shows that epicortical DC
nected with the O2 reservoir. Under these conditions, the CO2 potentials reflect neuronal hyperpolarization. The disappear-
tension progressively rises in the tissue for about 15 minutes ance of the EEG waves is presumed to be caused mainly by the
without a concomitant fall of the oxygen tension below the reduction of postsynaptic activity.
baseline level.
Primary asphyxia exemplified by respiratory arrest after air ven-
tilation is associated with combined CNS effects of hypercapnia
and hypoxia. The effects of gas tension changes on the field
potentials and on the membrane potential of individual neu-
rons are schematically shown in Figure 2.18. In the correspon-
ding animal experiments, the artificial ventilation with air was
either temporarily (Fig. 2.18A) or persistently (Fig. 2.18B)
With such an interruption of artificial ventilation with
air, the conventional EEG waves disappear within less than
1 minute. This process is accompanied by a negative DC poten-
tial shift from the baseline, which has been characterized as ini-
tial negativity (1 in Fig. 2.18). While the EEG shows an
isoelectric line in the further course of asphyxia, additional
potential shifts are detectable with DC recording technique.
The initial negativity is followed by a positive DC shift termed
Figure 2.17 Effects of an isolated hypercapnia on epicortical field intermediate positivity (2 in Fig. 2.18). If reventilation is per-
potentials (EEG, DC/ EEG) and on membrane potential (MP). With formed in this phase of asphyxia, an additional positive DC
increasing pCO2, the EEG disappears even if the pO2 is above normal shift is observed, appropriately termed reactive positivity (3 in
levels. The disappearance of the EEG is associated with a positive DC Fig. 2.18A). According to the analysis of the experimental work,
shift and a hyperpolarization of most of the neurons. Simultaneously, the intermediate and the reactive types of positivity are due to
the amplitudes of stimulus (St) evoked EPSP are markedly reduced. an increase of CO2 tension in the brain tissue. With the resolu-
(From original tracings from Speckmann EJ, Caspers, H. The effect of tion of the reactive positivity, a restitution of the fast field
O2 and CO2 tensions in the nervous tissue on neuronal activity and DC potentials occurs that is also demonstrable with the conven-
potentials. In: Remond A, ed-in-chief. Handbook of Electroence- tional EEG. A comparison of the DC shifts and the alterations
phalography and Clinical Neurophysiology. Vol 2. Part C. Amsterdam: of the membrane potentials shows a parallelism of both events
Elsevier; 1974:71–89.) (1,14,15,33,57,58).
30 Part I ■ Basic Principles

5. Hubbard JI, Llinas R, Quastel DMJ. Electrophysiological Analysis of

Synaptic Transmission/Monographs of the Physiological Society.
London: Edward Arnold; 1969.
6. Shepherd GM. The Synaptic Organization of the Brain. London:
Oxford University Press; 1974.
7. Creutzfeldt O, Houchin J. Neuronal basis of EEG waves. In:
Remond A, ed-in-chief. Handbook of Electroencephalography and
Clinical Neurophysiology. Vol 2. Part C. Amsterdam: Elsevier;
8. Speckmann EJ, Caspers H, Elger CE. Neuronal mechanisms
underlying the generation of field potentials. In: Elbert T,
Rockstroh B, Lutzenberger W, et al., eds. Self-Regulation of the
Brain and Behavior. New York, NY: Springer; 1984:9–25.
9. Kuffler SW, Nicholls JG. The physiology of neuroglial cells. Erg
Physiol. 1966;57:1–90.
10. Somjen GG, Trachtenberg M. Neuroglia as generator of extracellu-
Figure 2.18 Alterations of EEG, DC/ EEG, and neuronal membrane lar current. In: Speckmann EJ, Caspers H, eds. Origin of Cerebral
potential (MP) during primary asphyxia. A: The abolition and the reap- Field Potentials. Stuttgart: Thieme; 1979:21–32.
pearance of EEG during a transient asphyxia goes in parallel with typi- 11. Kuffler SW, Nicholls JG, Orkand RK. Physiological properties of
cal DC shifts: (1) initial negativity, (2) intermediate positivity, and (3) glial cells in the central nervous system of amphibia. J Neuro-
physiol. 1966;29:768–787.
reactive positivity. These DC fluctuations are accompanied by corre-
12. Orkand RK, Nicholls JG, Kuffler SW. Effect of nerve impulses on
sponding reactions of the MP. B: With continuing asphyxia, the EEG
the membrane potential of glial cells in the central nervous system
remains abolished and the intermediate positivity (2) turns over into a of amphibia. J Neurophysiol. 1966;29:788–806.
terminal negativity (4). The latter DC negativity corresponds to a break- 13. Speckmann E-J. Experimentelle Epilepsieforschung. Darmstadt:
down of neuronal membrane potential. (From original tracings from Wissenschaftliche Buchgesellschaft; 1986.
Speckmann EJ, Caspers H. The effect of O2 and CO2 tensions in the 14. Caspers H, Speckmann E-J, Lehmenkühler A. Electrogenesis of
nervous tissue on neuronal activity and DC potentials. In: Remond A, cortical DC potentials. In: Kornhuber HH, Deecke L, eds.
ed-in-chief. Ha ndbook of Electroen cepha logra phy a nd Clinica l Motivation Factor and Sensory Processes of the Brain: Electrical
Neurophysiology. Vol 2. Part C. Amsterdam: Elsevier; 1974:71–89.) Potentials, Behaviour and Clinical Use. Progress in Brain Research.
Vol 54. New York, NY: Elsevier; 1980:3–15.
15. Caspers H, Speckmann E-J, Lehmenkühler A. Electrogenesis of
slow potentials of the brain. In: Elbert T, Rockstroh B,
If the interruption of the artificial ventilation is continued Lutzenberger W, et al., eds. Self-Regulation of the Brain and
for a longer period of time, then the intermediate positivity Behavior. New York, NY: Springer; 1984:26–41.
converts into the so-called terminal negativity (4 in Fig. 2.18B). 16. Rall W. Core conductor theory and cable properties of neurons. In:
This negative DC shift correlates with the breakdown of the Kandel ER, ed. Handbook of Physiology/The Nervous System. Vol 1.
neuronal membrane potential. The terminal effects are due to a Part 1. Bethesda, MD: American Physiological Society;
critical lack of oxygen. The terminal negativity may be 1977:39–97.
reversible for a substantial period of time under certain experi- 17. Caspers H, ed. DC potentials recorded directly from the cortex. In:
mental conditions if the artificial ventilation is resumed and the Remond A, ed-in-chief. Handbook of Electroencephalography
reduction of the cerebral circulation is counteracted with circu- and Clinical Neurophysiology. Vol 10. Part A. Amsterdam: Elsevier;
lation support measures (58).
18. Speckmann E-J, Caspers H. Cortical field potentials in relation to
In summary, a comparison of EEG and DC potentials in
neuronal activities in seizure conditions. In: Speckmann E-J,
selective hypercapnia and primary asphyxia shows that the Caspers H, eds. Origin of Cerebral Field Potentials. Stuttgart:
recording of cortical field potentials with DC amplifiers pro- Thieme; 1979:205–213.
vides a more accurate picture of the actual functional state of 19. Caspers H, Speckmann E-J. DC potential shifts in paroxysmal
nerve cells. states. In: Jasper HH, Ward AA Jr, Pope A, eds. Basic Mechanisms of
the Epilepsies. Boston, MA: Little, Brown & Company; 1969:
REFERENCES 20. Gumnit RJ, Matsumoto H, Vasconetto C. DC activity in the depth
of an experimental epileptic focus. Electroencephalogr Clin
1. Speckmann E-J, Caspers H, eds. Origin of Cerebral Field Potentials. Neurophysiol. 1970;28:333–339.
Stuttgart: Thieme; 1979. 21. Speckmann E-J, Elger CE. The neurophysiological basis of epilep-
2. Palay SL, Chan-Palay V. General morphology of neurons and neu- tic activity: a condensed overview. In: Degen R, Niedermeyer E,
roglia. In: Kandel ER, ed. Handbook of Physiology/The Nervous eds. Epilepsy, Sleep, and Sleep Deprivation. Amsterdam: Elsevier;
System. Vol 1. Part 1. Bethesda, MD: American Physiological 1984:23–34.
Society; 1977:5–37. 22. Speckmann E-J. Das Gehirn meiner Kunst. Kreativität und das
3. De Robertis EDP, Carrea R, eds. Biology of Neuroglia/Progress in selbstbewußteGehirn. Münster: Daedalus; 2008.
Brain Research. Vol 15. New York, NY: Elsevier; 1964. 23. Cohen LB, Salzberg BM. Optical measurement of membrane
4. Eccles JC. The Physiology of Synapses. Berlin: Springer; 1964. potential. Rev Physiol Biochem Pharmacol. 1978;83:35–83.
Chapter 2 ■ Neurophysiologic Basis of EEG and DC Potentials 31

24. Ebner TJ, Chen G. Use of voltage-sensitive dyes and optical record- 42. Petsche H, Pockberger H, Rappelsberger P. Current source density
ings in the central nervous system. Prog Neurobiol. 1985;46: 463–506. studies of epileptic phenomena and the morphology of the rabbit’s
25. Köhling R, Höhling J-M, Straub H, et al. Optical monitoring of neu- striate cortex. In: Klee MR, Lux HD, Speckmann E-J, eds.
ronal activity during spontaneous sharp waves in chronically epilep- Physiology and Pharmacology of Epileptogenic Phenomena. New
tic human neocortical tissue. J Neurophysiol. 2000;84: 2161–2165. York, NY: Raven Press; 1981.
26. Köhling R, Reinel J, Vahrenhold J, et al. Spatio-temporal patterns 43. Speckmann E-J, Elger CE, eds. Epilepsy and Motor System.
of neuronal activity: analysis of optical imaging data using geo- Baltimore: Urban & Schwarzenberg; 1983.
metric shape matching. J Neurosci Methods. 2002;114:17–23. 44. Wieser HG. Electroclinical Features of the Psychomotor Seizure. A
27. Straub H, Kuhnt U, Höhling J-M, et al. Stimulus induced patterns Stereoencephalographic Study of Ictal Symptoms and
of bioelectric activity in human neocortical tissue recorded by a Chronotopographical Seizure Patterns Including Clinical Effects of
voltage sensitive dye. Neuroscience. 2003;121:587–604. Intracerebral Stimulation. Stuttgart: Gustav Fischer; 1983.
28. Grinvald A, Hildesheim R. VSDI: a new era in functional imaging 45. Speckmann E-J, Caspers H, Janzen RWC. Laminar distribution of
of cortical dynamics. Nat Rev Neurosci. 2004;5:874–885. cortical field potentials in relation to neuronal activities during
29. Gorji A, Straub H, Speckmann E-J. Epilepsy surgery: perioperative seizure discharges. In: Brazier MAB, Petsche H, eds. Architectonics
investigations of intractable epilepsy. Anat Embryol (Berlin). of the Cerebral Cortex. IBRO Monograph Series. Vol 3. New York,
2005;210:525–537. NY: Raven Press; 1978:191–209.
30. Andersen P, Andersson SA. Physiological Basis of the Alpha Rhythm. 46. Speckmann E-J, Caspers H, Janzen RWC. Relations between corti-
New York, NY: Meredith; 1968. cal DC shifts and membrane potential changes of cortical neurons
31. Caspers H. Relations of steady potential shifts in the cortex to the associated with seizure activity. In: Petsche H, Brazier MAB, eds.
wakefulness–sleep spectrum. In: Brazier MAB, ed. Brain Function. Synchronization of EEG Activity in Epilepsies. New York, NY:
Berkeley, CA: University of California Press; 1963:177–213. Springer; 1972:93–111.
32. Goldring S. DC shifts released by direct and afferent stimulation. 47. Leao AAP. Spreading depression of activity in the cerebral cortex.
In: Remond A, ed-in-chief. Handbook of Electroencephalography J Neurophysiol. 1944;7:359–390.
and Clinical Neurophysiology. Vol 10. Part A. Amsterdam: Elsevier; 48. Somjen GG. Mechanisms of spreading depression and hypoxic
1974:12–24. spreading depression-like depolarization. Physiol Rev.
33. Caspers H, Speckmann E-J. Cortical DC shifts associated 2001;81:1065–1096.
with changes of gas tensions in blood and tissue. In: Remond A, 49. Gorji A, Scheller D, Straub H, et al. Spreading depression in
ed-in-chief. Handbook of Electroencephalography and Clinical human neocortical slices. Brain Res. 2001;906:74–83.
Neurophysiology. Vol 10. Part A. Amsterdam: Elsevier; 1974:41–65. 50. Berger M, Speckmann EJ, Pape HC, et al. Spreading depression
34. Elger CE, Speckmann E-J. Penicillin-induced epileptic foci in enhances human neocortical excitability in vitro. Cephalalgia.
the motor cortex: vertical inhibition. Electroencephalogr Clin 2008;28:558–562.
Neurophysiol. 1983;56:604–622. 51. Gorji A. Spreading depression: a review of the clinical relevance.
35. Petsche H, Muller-Paschinger IB, Pockberger H, et al. Depth profiles Brain Res Brain Res Rev. 2001;38:33–60.
of electrocortical activities and cortical architectonics. In: Brazier 52. Smith JM, Bradley DP, James MF, et al. Physiological studies of
MAB, Petsche H, eds. Architectonics of the Cerebral Cortex. IBRO cortical spreading depression. Biol Rev Camb Philos Soc.
Monograph Series. Vol 3. New York, NY: Raven Press; 1978:257–280. 2006;81:457–481.
36. Klee MR, Lux HD, Speckmann E-J, eds. Physiology and 53. Bures J, Buresova O, Krivanek J. The Mechanisms and Applications
Pharmacology of Epileptogenic Phenomena. New York, NY: Raven of Leao’s Spreading Depression of Electroencephalographic Activity.
Press; 1982. New York, NY: Academic Press; 1974.
37. Purpura DP, Penry JK, Tower DB, et al., eds. Experimental Models 54. Kager H, Wadman WJ, Somjen GG. Conditions for the triggering
of Epilepsy. New York, NY: Raven Press; 1972. of spreading depression studied with computer simulations.
38. Jasper HH, Ward AA, Pope A, eds. Basic Mechanisms of the J Neurophysiol. 2002;88:2700–2712.
Epilepsies. Boston, MA: Little, Brown & Company; 1969. 55. Somjen GG, Kager H, Wadman WJ. Computer simulations of neu-
39. Elger CE, Speckmann E-J. Focal interictal epileptiform discharges ron–glia interactions mediated by ion flux. J Comput Neurosci.
(FIED) in the epicortical EEG and their relations to spinal field 2008;25(2):349–365.
potentials in the rat. Electroencephalogr Clin Neurophysiol. 56. Strong AJ, Fabricius M, Boutelle MG, et al. Spreading and synchro-
1980;48:447–460. nous depressions of cortical activity in acutely injured human
40. Elger CE, Speckmann E-J, Caspers H, et al. Focal interictal epilep- brain. Stroke. 2002;33:2738–2743.
tiform discharges in the cortex of the rat: laminar restriction and 57. Caspers H, Speckmann E-J, Lehmenkühler A. Effects of CO2 on
its consequences for activity descending to the spinal cord. In: cortical field potentials in relation to neuronal activity. In:
Klee MR, Lux HD, Speckmann E-J, eds. Physiology and Speckmann E-J, Caspers H, eds. Origin of Cerebral Field Potentials.
Pharmacology of Epileptogenic Phenomena. New York, NY: Raven Stuttgart: Thieme; 1979:151–163.
Press; 1981. 58. Speckmann E-J, Caspers H. The effect of O2 and CO2 tensions in the
41. Gumnit R. DC shifts accompanying seizure activity. In: Remond A, nervous tissue on neuronal activity and DC potentials. In: Remond
ed-in-chief. Handbook of Electroencephalography and Clinical A, ed-in-chief. Handbook of Electroencephalography and Clinical
Neurophysiology. Vol 10. Part A. Amsterdam: Elsevier; 1974:66–77. Neurophysiology. Vol 2. Part C. Amsterdam: Elsevier; 1974:71–89.

Cellular Substrates of Brain Rhythms*

uasiperiodically recurring waves, henceforth termed 2. The study of the intrinsic properties of cells is mostly carried
oscillations, are, besides evoked potentials, one of the out in vitro or in cultures, preparations that are as far as one can
essential components of the electroencephalogram imagine from the complex reality of the whole brain, both in
(EEG). The curiosity of understanding the cellular counterparts terms of network linkages and physiologic state. At best, these
of EEG waves started almost simultaneously with the discovery preparations would correspond to a deeply comatose brain.
and description of different EEG patterns during the early 1930s 3. The study of the network interactions, however multisite
and 1940s as an effort to relate the global graphic aspect of the they might be, is still based on recordings from spatially dis-
EEG with its more intimate cellular counterparts. In this chap- crete and limited locations. The assumption of continuity
ter, we will discuss how various patterns of oscillatory activity should therefore bear reserve.
develop in cellular brain networks and are further reflected at 4. Anesthesia is often an unavoidable companion of animal
the EEG level. The EEG is essentially an electric phenomenon; studies, and the elimination of the pharmacologic effect is
therefore, we will focus on the ionic currents active during vari- time and again more complicated than a mere subtraction.
ous oscillating patterns but we will also identify the (minimal)
anatomical structures necessary and sufficient for the genesis of
the aforementioned activities. This approach has unavoidably SLOW DELTA RHYTHMS
oversimplifying virtues, while the EEG signal is a complex one.
Grey Walter (1) was the first to assign the term “delta waves” to
This reality will be recreated through the coalescence of various
particular types of slow waves recorded in the EEG of humans.
oscillations that coexist during a given behavioral context.
Although Walter introduced the term delta waves in correspon-
From its very beginning, the EEG was characterized through
dence to pathologic potentials due to cerebral tumors, with
its oscillatory nature. Thus the dominant frequency of an oscil-
time delta activities became more related to sleep and anesthe-
lation became its main descriptor. This trend was further
sia. The IFSECN (2) defines delta waves as waves with a dura-
enforced with the advent of spectral analysis (e.g., Fourier
tion of more than 0.25 seconds (which implies a frequency
transforms) and is still largely used nowadays. Although this
band between 0 and 4 Hz). Thus the delta term is associated
might enclose certain advantages, it also contains numerous
with frequency bands rather than with phenomena generating
pitfalls that prevent the reaching of correct conclusions. This
specific electrographic patterns. There have been various stud-
situation is generated by the fact that a given spectral compo-
ies aiming at disclosing the relationship between cellular activ-
nent might ambiguously result either from the presence in the
ities and EEG (3) and the sources of delta activities (4,5). In the
EEG of oscillations or from the contribution of particular
following it will be emphasized that the 0- to 4-Hz frequency
waveforms, not necessarily rhythmic, to the EEG spectrum.
range reflects more than one phenomenon and that definitions
In the following we will organize this chapter according to
based exclusively on frequency bands may conceal the underly-
the frequency bands traditionally encountered in the EEG
ing mechanism. Studies in the last two decades have unveiled
praxis, however, emphasizing the possible sources of haziness
the electrophysiologic substrates of several distinct activities in
related to the interpretation of EEG signals. Furthermore, as a
the frequency range below 4 Hz during sleep and anesthesia.
section devoted to the understanding of the cellular mecha-
Their interaction within corticothalamic networks yields a
nisms underlying the genesis of the oscillatory behavior of the
complex pattern whose reflection at the EEG level takes the
EEG, the intracellular data constitutes a very strong, although
shape of polymorphic waves.
not absolute, source of knowledge. The reader should, however,
It should also be stressed from the beginning that delta activ-
keep in mind that, when extrapolating from cellular data to
ities cover two EEG phenomena: waves and oscillations, with no
EEG, one should not overlook essential details, such as
clear separations between the two, as far as spectral analysis is
1. Cellular recordings are almost exclusively performed in ani- concerned. The term oscillation designates a repeated variation
mals, whose philogenetical development, behavioral, and of a parameter (e.g., current, voltage) between two extreme val-
structural peculiarities are often neglected. ues, with an optional requirement of regularity of the variation.

*This chapter was initiated and written in the previous editions by the late Dr. Mircea Steriade (1924–2006). The present authors wish to pay tribute to his outstand-
ing work that has much advanced our understanding of neuronal mechanisms underlying EEG rhythms. They hope to continue and preserve the spirit of his

34 Part I ■ Basic Principles

However, most of the biologic phenomena resulting from large delta waves (as obtained by lesions of the subcortical white mat-
population interactions, such as the EEG, are rarely associated ter, the thalamus, or the mesencephalic reticular [RE] forma-
with clocklike rhythmicity. At the other limit of the definition is tion) have shown a relationship between the firing probability
the wave, as a single variation of a parameter between two and the surface-positive (depth-negative) delta waves, whereas
extreme values. From the above definitions it appears that oscil- the depth-positive waves were associated with a diminution in
lations are made of waves. discharge rates (15). These field–unit relationships led to the
As for now, there are at least two cellular sources of delta assumption that the depth-positive component of delta waves
activities: one originating in the thalamus and the other one in reflects the inhibition of pyramidal-shaped neurons by local-
the cortex. circuit cells, implying maximal firing of inhibitory interneurons
Thalamic delta oscillations have been found in a series of in during the depth-positive delta waves. However, this has not
vitro studies. They revealed that a clocklike oscillation within the been found. It was then suggested that, far from resulting exclu-
delta frequency range is generated by the interplay of two intrin- sively from inhibitory postsynaptic potentials (IPSPs), EEG
sic currents of thalamocortical cells. Whereas most of brain delta waves are rather generated by summation of long-lasting
oscillations are generated by interactions within networks of afterhyperpolarizations produced by a variety of potassium
neurons, but also glial cells, the thalamic delta oscillation is an currents in deeply lying pyramidal neurons (16,17).
intrinsic oscillation depending on two inward currents of thala- The discovery of a novel slow-wave sleep- or anesthesia-related
mocortical cells (7–9). It was shown that thalamocortical neu- oscillation (around, but generally less than 1 Hz) (18), together
rons recorded in vitro display rhythmic bursts of high-frequency with its thorough investigation at the cellular level (see below), set
spikes with an interburst frequency of 1 to 2 Hz. This oscillation the basis for revisiting the cellular basis of delta rhythms. The fre-
results from the interplay between two membrane currents: the quency of the slow oscillation depends on the depth or type of
transient calcium current (It) underlying the low-threshold anesthesia, or the sleep stage: it is mainly between 0.3 and 0.6 Hz
spike (LTS) and a hyperpolarization-activated cation current under urethane anesthesia, between 0.6 and 0.9 Hz under keta-
(Ih). Ih is a noninactivating inward (anomalous) rectifier carried mine-xylazine anesthesia, and between 0.7 and ~1 Hz during nat-
by sodium and potassium. The model proposed for the genesis ural sleep. This oscillation, termed slow oscillation by its authors,
of this clocklike delta oscillation is depicted in Figure 3.1A: when was first described in intracellular recordings from neocortical
the membrane potential hyperpolarizes (more negative than neurons in anesthetized animals and in EEG recordings from
–70 mV), the Ih becomes activated and produces a depolarizing humans (18). It was subsequently found during natural slow-
sag, which in turn will activate the It (deinactivated by the hyper- wave sleep of animals (19) and humans (20–24).
polarized membrane). The ensuing LTS deactivates the Ih but The slow oscillation is generated within the cortex because it
also generates a hyperpolarizing overshoot that will reactivate survives after thalamectomy (25), is absent from the thalamus
the Ih, thus restarting a new oscillatory cycle. of decorticated animals (26), and is present in cortical slices
This oscillation was also found in vivo, in the thalamocorti- (27). At the intracellular level, cortical neurons throughout lay-
cal neurons of the cat, after decortications (10–12). ers II to VI (with physiologically identified thalamic and/or cal-
Thalamocortical neurons from a variety of sensory, motor, asso- losal inputs, and with projections to the thalamus and/or
ciational, and intralaminar thalamic nuclei were able to display homotopic points of the contralateral hemisphere) displayed a
a clocklike delta rhythm either induced by imposed hyperpolar- spontaneous oscillation recurring with periods of 1–1.5 to
izing current pulses (Fig. 3.1B) or spontaneously (Fig. 3.1C). 5 seconds, depending on the anesthetic, and consisting of pro-
Interestingly, the deafferentation achieved through decortica- longed depolarizing and hyperpolarizing components (Fig.
tion, by removing the powerful depolarizing impingement from 3.2A). The long-lasting depolarizations of the slow oscillation
corticothalamic neurons and setting thalamic cells at a more consisted of excitatory postsynaptic potentials (EPSPs), fast
hyperpolarized membrane potential (around –70 mV), was prepotentials (FPPs), and fast IPSPs, reflecting the action of
instrumental in permitting the delta oscillation to emerge. synaptically coupled GABAergic local-circuit cortical cells (18).
However, in intact preparations, with functional corticothalamic Data also demonstrated the contribution of both NMDA-
loops, the regular thalamic delta oscillation was absent or largely mediated synaptic excitatory events and a voltage-dependent
prevented by the ongoing cortical activity (13). At the cellular persistent sodium current, INa(p), to the depolarizing component
level, the occurrence of the Ih was blocked during the depolariz- of the slow oscillation. On the other hand, the long-lasting
ing phases of the slow oscillation ( 1 Hz; see below), due to the hyperpolarization, interrupting the depolarizing events, is asso-
interference of synaptic currents and the ensuing decreased ciated with network disfacilitation in the cortex. This is sup-
membrane resistance with the intrinsically generated Ih. These ported by the fact that the membrane input resistance is highest
more recent findings raised concerns as to the emergence of this during the long-lasting hyperpolarizing component of the slow
intrinsically generated thalamic rhythm at the level of the EEG. oscillation (28). The disfacilitation is probably achieved by a
The existence of cortical delta oscillations was suggested on progressive depletion of the extracellular calcium ions during
the basis that delta waves, at 1 to 2 Hz, survive in the EEG of the depolarizing phase of the slow oscillation (29) (Fig. 3.2B–C).
athalamic cats (14). Whether such procedures generate a phys- All major cellular classes in the cerebral cortex, as identified
iologic or a pathologic pattern remains an open question; by electrophysiologic characteristics and intracellular staining,
nonetheless, they create a deafferentation of the cortex. display the slow oscillation: regular spiking and intrinsically
Extracellular recordings of cortical activity during pathologic bursting cells, as well as local-circuit inhibitory basket cells
Figure 3.1 Delta oscillations in thalamocortical cells result from the interplay between two intrinsic currents, Ih and It. A: Proposed model for interac-
tion between these intrinsic currents. Activation of the low-threshold calcium current, It, depolarizes the membrane toward threshold for a burst of
sodium-dependent fast action potentials. The depolarization inactivates the portion of Ih that was active immediately before the calcium spike.
Repolarization of the membrane due to It inactivation is followed by a hyperpolarizing overshoot, due to the reduced depolarizing effect of Ih. The hyper-
polarization in turn deinactivates It and activates Ih, which depolarizes the membrane toward threshold for another calcium spike. (Modified from
McCormick DA, Pape HC. Properties of a hyperpolarization-activated cation current and its role in rhythmic oscillation in thalamic relay neurones.
J Physiol. 1990;431:291–318.) B: Delta oscillation of cat ventrolateral thalamocortical cell triggered by hyperpolarizing current pulses (0.7 nA in 1, 1 nA
in 2, 1.1 nA in 3, and 1.2 nA in 4). Note increasing number of cycles at a frequency of 1.6 Hz. C: Latero-posterior (LP) thalamocortical cell after decor-
tication of areas projecting to LP nucleus. The cell oscillated spontaneously at 1.7 Hz. A 0.5-nA depolarizing current (between arrows) prevented the
oscillation, and its removal set the cell back in the oscillatory mode. Three cycles after removal of depolarizing current in 1 are expanded in 2 to show
high-frequency bursts crowning the low-threshold calcium spike. In this and following figures, polarity of EEG and field potentials is as for intracellular
recordings (positivity up). (Modified from Steriade M, Curró Dossi R, Nuñez A. Network modulation of a slow intrinsic oscillation of cat thalamocorti-
cal neurons implicated in sleep delta waves: cortically induced synchronization and brainstem cholinergic suppression. J Neurosci. 1991;11:3200–3217.)

36 Part I ■ Basic Principles

Figure 3.2 A: Pyramidal cell from the somatosensory cortex (area 3b) during the slow ( 1 Hz) oscillation. The left panel
shows intracellular recordings and simultaneously recorded EEG in the vicinity (~ 1 mm) of the cell. The EEG was recorded
by means of coaxial electrodes located on the surface and at a depth of approximately 0.6 mm. The cell oscillated at 0.9 Hz
with depolarizing phases corresponding to depth-EEG negative (surface-positive) potentials. The right panel shows the cor-
responding cell stained with Neurobiotin (calibration bar in mm). (Modified from Contreras D, Steriade M. Cellular basis
of EEG slow rhythms: a study of dynamic corticothalamic relationships. J Neurosci. 1995;15:604–622.) B: Fluctuations dur-
ing the slow oscillation. Relationships between intracellular membrane potential, extracellular calcium [Ca] out, and field
potential. Periodic neuronal depolarizations triggering action potentials were interrupted by periods (300–500 msec) of
hyperpolarization and silenced synaptic activity. [Ca]out dropped by about 0.25 mM during the depolarizing phase reach-
ing a minimum just before the onset of the hyperpolarization. Then, [Ca] out rose back until the beginning of the next cycle.
C: Thirty cycles were averaged (spikes from the neuronal signal were clipped) after being extracted around the onset of
the neuronal depolarization. The vertical dotted lines tentatively indicate the boundaries of the two phases of the slow
oscillation (B–C modified from Massimini M, Amzica F. Extracellular calcium fluctuations and intracellular potentials in
the cortex during the slow sleep oscillation. J Neurophysiol. 2001;85:1346–1350.).
Chapter 3 ■ Cellular Substrates of Brain Rhythms 37

Figure 3.3 Effect of lidocaine inactivation on synchrony between intracellular activities and FPs. A: Double intracellular
recording in the anterior and posterior parts of the suprasylvian gyrus (see scheme). Synchrony was present between all
three recording sites. Intracortical injection of lidocaine (40 L, 20% ; bottom) inactivated the electrical activity close to
the cannula in the middle part of the gyms, disrupted the synchrony between pools of neurons, and between individual
neurons. B: Sequential correlation analyses of intracellular and field potential data. Two-minute periods were analyzed
with the sequential correlation method. (Modified from Amzica F, Steriade M. Disconnection of intracortical synaptic link-
ages disrupts synchronization of a slow oscillation. J Neurosci. 1995;15:4658–4677.)

(18,30,31). All these neuronal types exhibit similar relation- Such cortical coherence is expected to equally involve the
ships with the EEG components of the slow oscillation: during thalamus. Indeed, RE thalamic cells exhibit similar and time-
the depth-positive EEG wave cortical neurons are hyperpolar- locked variations of their membrane potential, with prolonged
ized, whereas during the sharp depth-negative EEG deflection depolarizations interrupted by prolonged hyperpolarizations
cortical neurons are depolarized (Fig. 3.2). The spectacular (34) in reflection of the excitatory cortico-RE projections. The
coherence between all types of cortical neurons and EEG wave- depolarizing component in RE neurons is transmitted to thala-
forms raised the question of the underlying synchronizing mocortical neurons, via GABAergic projections, where it trig-
mechanisms. Dual intracellular recordings in vivo revealed that gers rhythmic IPSPs, leading to rebound spike bursts (Fig. 3.4),
intracellular potentials were highly synchronized among neu- which in turn will generate phasic excitations in the cortex,
rons (31,32). This synchronization, displaying time lags as a shaping the steady depolarization of cortical neurons.
function of the distance between cortical areas (33), mainly The better comprehension of the mechanisms determining
relies on the integrity of intracortical synaptic linkages. the pacing of the slow oscillations and the switch between
Although the coherence was impaired after local blockage of depolarizing and hyperpolarizing phases came from experi-
axonal transmission through local injections of lidocaine (Fig. ments considering the possible dialogue between neurons and
3.3), the slow oscillation survived at both locations and con- glial cells. Dual simultaneous intracellular recordings from neu-
served a rudiment of synchronization suggesting that, besides rons and adjacent glial cells were performed during sleeplike
direct intracortical linkages, other projections, possibly cortico- patterns produced by ketamine-xylazine anesthesia, exploring
thalamo-cortical, as well as networks of gap junctions (see the possibility that glia may not only passively reflect, but also
below) might contribute to the synchronization of the slow influence the state of neuronal networks (35–37). The behavior
oscillation. of simultaneously recorded neurons and glia is illustrated in
38 Part I ■ Basic Principles

Figure 3.4 The slow oscillation (~ 0.9 Hz) in dual simultaneous intracellular recordings from a regular-spiking cell
in cortical area 4 and a thalamocortical cell in the ventrolateral nucleus. Cat under ketamine-xylazine anesthesia.
Arrow in A points to a low-threshold spike burst. An expanded cycle is shown in B. Notes: (a) Depth-positive
(upward) EEG waves are associated with hyperpolarization of cortical and thalamic cells, whereas the sharp depth-
negatives are associated with depolarization and action potentials in cortical cell, while the thalamic neuron displays
a rebound spike burst with a delay of 150–200 msec; (b) brief sequence of EEG spindles after depth-negative (surface-
positive) sharp deflection; and (c) fast depolarizing waves (40–50 Hz) in cortical neuron during the sustained depo-
larization. (Unpublished data by M Steriade and D Contreras.)

Figure 3.5. During spontaneously occurring slow oscillations, glial cells might control the pace of the oscillation through
the onset of the glial depolarization did, in the vast majority of changes in the concentration of the extracellular potassium (38).
the cases, follow the onset of the neuronal depolarization with The overall synchronization of the slow oscillation in the
an average time lag of around 90 msec (Fig. 3.5). This time lag cortex is also assisted by glial cells, which are imbedded in a gap
is, however, longer than the one obtained from pairs of neurons junction-based network, through the phenomenon of spatial
(around 10 ms in Ref. 33). The glial depolarization reflects with buffering (41–43). Spatial buffering evens local increases of
virtually no delay the potassium uptake (38). As potassium is extracellular potassium by transferring it from the extracellular
also continuously recaptured by neurons through Na /K -ATP milieu to neighboring glial cells, then through gap junctions,
pumps, the actual onset of the glial depolarization may thus and along the potassium concentration gradient, at more dis-
mark the moment where the increase of extracellular potassium tant sites with lower concentrations of potassium, where it is
concentration exceeds the capacity of the neuronal pumps. This again expelled in the extracellular space. This pathway is pre-
behavior suggests that glial cells are essential for maintaining ferred to the direct one through the extracellular space because
extracellular potassium within controlled values. of the tortuosity of the latter (reviewed in Ref. 44). The rela-
Toward the end of the depolarizing phase, the glial membrane tionship between intracellular glial potentials and extracellular
begins to repolarize before the neuronal membranes (Fig. 3.5) potassium (38) shows little accumulation of potassium in the
(37). Interestingly, this observation may expose the mechanism extracellular space ( 1.1 mM/cycle), and this causes a minimal
ruling the onset of the hyperpolarizing phase of the slow oscilla- imbalance between the extra- and intracellular potassium con-
tion. It was initially proposed that this hyperpolarization relies centrations. Given that normally glial cells deal with low
either on active GABA inhibition or on the activation of slow amounts of potassium but have a high propensity to uptake it
calcium-dependent potassium currents. In the first case, the (43,45,48), it is likely that spatial buffering would occur at a
release of GABA by neurons would either depolarize glial cells reduced spatial scale, thus contributing to an uniform distribu-
through direct action on GABAA receptors (39,40) or indirectly tion of the potassium around the membrane of neurons (38).
through the potassium output resulting from the neuronal acti- The local spatial buffering during slow oscillations might
vation of neuronal GABAB receptors. The alternative implication play two roles: (i) it contributes to the steady depolarization of
of calcium-dependent potassium currents should be discarded neurons during the depolarizing phase of the slow oscillation
because of the fact that the extracellular potassium starts to drop in a Nernst-related manner and (ii) it modulates the neuronal
before the onset of the hyperpolarizing phase. In addition, the excitability. The latter mechanism could favor the synaptic
glial membrane potential returns to control value at the end of interaction within cortical networks at the onset of the depo-
each of the recurring oscillatory cycles, leading to the assumption larizing phase of the slow oscillation, but it could equally
Chapter 3 ■ Cellular Substrates of Brain Rhythms 39

Figure 3.5 Simultaneous recording

of a neuron–glia pair in the cortex of
the cat during slow oscillatory pat-
terns. A: Relationship between neu-
ronal and glial depolarizations on
the one hand, and between neu-
ronal hyperpolarization and glial
repolarizations on the other hand
(note enlargement within the box).
The glial impalement occurred dur-
ing this period of the recording and
is depicted as a brisk drop of the
potential from the extracellular
(0 mV) to the intracellular (–90 mV)
level. B: Histogram of membrane
potential variations in the neuron
(in gray) and glial cell (in black). C:
Typical time relationship between
the onset and offset moments of the
two phases of a slow oscillatory
cycle. Note that the onset of the
depolarizing phase starts in the neu-
ron, but that the end of this period is
anticipated in the glia. (Modified
from Amzica F, Massimini M. Glial
and neuronal interactions during
slow wave and paroxysmal activities
in the neocortex. Cereb Cortex.

induce a gradual disfacilitation as the extracellular potassium decades (see the excellent review in Ref. 51) during which
reaches higher values toward the top of the glial depolariza- researchers struggled with the inherent shape variability intro-
tion. The amplitude of the potassium variations during the duced by various bipolar electrode configurations, reference sys-
slow sleep oscillation ranges between 1 and 2 mM, which, tems, and filter settings. These technical difficulties, together
when added to the physiologic values of resting concentration with the relative vagueness of the definition, often led to discor-
(~3 mM), may assist cortical neurons in oscillating between dant conclusions and to difficult interpretations. The cellular
hyper- and hypoexcitability (47–49). This hypothesis was con- mechanisms of the K-complex were only recently described (52)
firmed after testing the excitability of the neurons during the and disclosed that the K-complex is a recurrent event with the
various phases of the slow oscillation cycle with cortical elec- rhythmicity of the slow oscillation (less than but close to 1 Hz)
tric stimuli (37). The results demonstrate that the neuronal and that, through its shape, it is one of the main contributors to
excitability is maximal toward the beginning of the depolariz- the delta spectrum (Fig. 3.6) (21,35).
ing cycle and almost zero toward the end of the depolarizing The rhythmicity of K-complexes during the onset of sleep is
phase of the slow oscillation. less obvious due to the lesser organization of the slow oscilla-
tion. This corroborates with the fact that bipolar recordings are
THE SLOW OSCILLATION often the only ones allowing the detection of K-complexes in
AND THE K-COMPLEX the initial part of slow-wave sleep. However, with the deepening
of sleep, the slow oscillation becomes more regular and faster
One of the functional correlates of the slow cortical oscillation is (the frequency approaches 1 Hz). During late stages of sleep
the fact that, at the EEG level, each sequence of most of the K-complexes may be confounded with delta waves.
depolarizing–hyperpolarizing episodes within an oscillatory The frequency shift (usually from 0.6 to 0.9 Hz) of the slow
cycle corresponds to a waveform called K-complex. Although oscillation, as well as the shape modification of the K-complex
the phenomenon was first described in 1938 as a landmark of (Fig. 3.7), accompanying the deepening of sleep can be
slow-wave sleep (50), its study remained at the EEG level for explained by taking into account the intrinsic properties of the
40 Part I ■ Basic Principles

Figure 3.6 K-complex in human sleep. Scalp monopolar recordings with respect to the contralateral ear (see figurine).
A: Short episode from a stage 3 period of sleep. The two arrows point to a K-complex (KC) followed by a spindle sequence
( ) and to a KC in isolation. The two K-complexes are embedded into a slow oscillation at about 0.6 Hz. Note the syn-
chrony of K-complexes in all recorded sites and the diminution of their amplitudes in the occipital area. B: Average of
50 K-complexes aligned on the positive peak (in this figure positivity is downwards) of the upper channel
(vertical dotted line). C: Power spectrum of the C3-lead for a period of 80 seconds of stable stage 3 activity
containing the period depicted in A. The three frequency bands (S, , and ; further illustrated in D) are rep-
resented in the power spectrum. Moreover, the slow (S) activity displays a high peak, distinct from the onset of delta
( ) activity (see inset). D: Frequency decomposition of the C3-lead (upper trace) into three frequency bands:
slow (S, 0–1 Hz), delta ( , 1–4 Hz), and sigma ( , 12–15 Hz). It is shown that the K-complex results from a
combination of slow and delta waves. (Modified from Amzica F, Steriade M. The K-complex: its slow ( 1-Hz) rhythmicity
and relation to delta waves. Neurology. 1997;49:952–959.)

neuronal membrane. At sleep onset, neurons in the midbrain Due to the controversial criteria for shape recognition, the
reticular formation and mesopontine cholinergic nuclei dimi- study of K-complexes has relied mostly on evoked K-complexes
nish their firing rate (53,54), thus removing a steady excitatory because stimulation generally produces stereotyped responses.
drive from thalamocortical neurons and allowing the Although of certain experimental value, the evoked K-
membrane potential of cortical neurons to reach more hyper- complexes have set a bias in the understanding of the underly-
polarized levels. The progressive hyperpolarization of thalamo- ing mechanisms because it was forgotten that the K-complex is
cortical neurons is a constant correlate of sleep deepening (55) mainly a spontaneous phenomenon (50). The comparison at
and induces a gradual reduction of the duration of the depolar- the cellular level between spontaneous and evoked K-
izing phase of the slow cortical oscillation together with the complexes shows, in agreement with the original observation of
acceleration of the rhythm. The former action contributes Davis et al. (56), that the latter appears as a secondary discharge.
mostly to the shape of the field potentials, while the second Both spontaneous and evoked K-complexes display similar
determines the oscillatory frequency. depth profiles and current source density (CSD) distributions
Chapter 3 ■ Cellular Substrates of Brain Rhythms 41

Figure 3.7 Rhythmic KCs in human EEG during natural slow-wave sleep. A: Four leads recorded from the
two hemispheres during stage 2 sleep show quasirhythmic (around 0.5 Hz) KCs. The expanded insets display
a simple K-complex (KC, left) and a K-complex followed by a spindle (right). Stage 3–4 EEG (below) is char-
acterized by a more regular oscillation of the K-complexes (0.7 Hz). Asterisks mark the most obvious
K-complexes in order to suggest their rhythmicity. B: Averaged K-complexes (n 200) from stages 2 (left)
and 3–4 (right). The gray surface around the averaged K-complexes covers the standard deviation. The low-
est trace (dotted line) results from the filtering of the average K-complex in the delta band (1–4 Hz). C: Power
spectra of 2-minute epochs containing the ones displayed above. Note a principal peak at 0.5 Hz for the stage
2 episode (left), surrounded by other lower peaks as evidence for distributed rhythmicity. At variance, deep
sleep shows a dominant peak at 0.7 Hz. The two FFT graphs are scaled with the same ordinate. (Modified
from Amzica F, Steriade M. The K-complex: its slow ( 1-Hz) rhythmicity and relation to delta waves.
Neurology. 1997;49:952–959.)
42 Part I ■ Basic Principles

(52), further supporting their common origin and generating lowest end of the theta frequency range that presented larger
mechanisms. power spectrum during rapid eye movement (REM) sleep than
The complex electrographic pattern of slow-wave sleep during other states and suggested that this oscillation may be
results from the coalescence of the slow oscillation with sleep considered the counterpart of the hippocampal theta of mam-
spindles (see below) and, possibly, thalamic delta oscillations malian REM sleep. Tesche and Karhu (67) recorded MEG sig-
(57). The weight of each of these major components is dynam- nals in human subjects and reconstructed from these signals the
ically modulated during sleep by synaptic coupling, local- corresponding hippocampal sources of theta rhythmic activity.
circuit configurations, and the general behavioral state of the They reported that during the presentation of a memory task
network. The discussion of the functional significance of the the duration of theta bursts increased with memory load and
K-complex is determined by the role of the slow sleep oscilla- suggested a relation between “stimulus-locked hippocampal
tion of which the K-complex is the electrographic reflection. theta” and the processing of working memory. Using a whole-
Through its rhythmic occurrence at a frequency that is lower head 275- channel MEG system in normal subjects, Cornwell et
than the one of other sleep rhythms (spindles and delta), and al. (68) reported hippocampal and parahippocampal theta
due to its wide synchronization at the cortical level, the sharp oscillations during a spatial navigation task (virtual reality
onset of the K-complex provides a synchronous input to thala- Morris water maze). They carried out source analyses by means
mic neurons, thus triggering and/or grouping sequences of of which they were able to show larger theta activity in the left
spindles or delta oscillations. anterior hippocampus and parahippocampal cortices during
goal-directed navigation relative to aimless movements. Thus,
THETA RHYTHMS similar to other species (see below), the human theta may rep-
resent a dynamic state emerging from hippocampal networks
Theta waves are usually defined as occurring within the 4- to engaged in spatial navigation and in memory processes.
7-Hz frequency range. The normal theta activity should not be
confused with pathologic theta waves, described as a slowing Limbic Theta Rhythms in Animals
down of alpha activity, expressed during cerebral blood flow Stewart and Fox (69), working on urethane-anesthetized pri-
reduction (58) or metabolic encephalopathies (59). Rhythmic mates, recorded hippocampal theta activity similar to that of rats
activities in the theta frequency range are conspicuous in limbic but at a slightly higher frequency and appearing as relatively
areas in various animal species as extensively reviewed by short bursts. Arnolds et al. (63) made a comparative study of
Buzsáki (60). hippocampal theta rhythms in cat, dog, and human and showed
that all three displayed the same type of qualitative changes in
Limbic Theta Rhythms in Humans spectral parameters (peak frequency, peak amplitude, and
The presence of theta rhythms was first denied in humans rhythmicity) in relation to behavioral tasks involving motor
(61,62), but they were later recorded in the hippocampus of activities. Particularly in cat, remarkable relationships between
epileptic patients with indwelling electrodes (63–65) or by way theta spectral properties and vestibular stimulation (e.g., body
of mesio-temporal corticography with foramen ovale elec- acceleration) and/or eye movements were put in evidence. Theta
trodes (66), and also using magnetoencephalography (MEG) in rhythms appear conspicuously in rodents such as rabbits (first
normal subjects (67,68). In all these studies human hippocam- description of the rhythm was made by Green and Arduini (70))
pal theta rhythm was studied under specific behavioral condi- and rats in relation with sensory processing and different types
tions. In the study of Arnolds et al. (69) the hippocampal of movements (71) and have been recorded in several structures
activity was recorded while the subject was freely moving and of the cortical limbic system (72,73). Depending on species and
performing a cognitive task; the frequency and rhythmicity of conditions, these rhythmic may extend from 3–4 to 10 Hz,
the hippocampal theta component while the subject was writ- which is somehow larger than the conventional range of the
ing was consistently higher than while sitting or walking. In a EEG theta rhythm (4 to 7 Hz). This is the reason why these lim-
word association task the amplitude, frequency, and rhythmic- bic theta oscillations are also named “rhythmic slow activity
ity showed a significant rise during the period of silent mental (RSA)” in animal neurophysiology to avoid confusion with the
activity immediately following a verbal cue to which the subject classic human EEG theta frequency band. The two designations
was requested to give a verbal answer. Ekstrom et al. (65) found are commonly used in the literature.
evidence for movement-related theta oscillations in human In the rat, theta oscillations are most regular in frequency
hippocampus and suggested that both cortical and hippocam- and present the largest amplitude in the stratum lacunosum-
pal oscillations play a role in attention and sensorimotor inte- moleculare of the hippocampal CA1 region but are also found
gration. Kahana et al. (64) investigated hippocampal EEG in the dentate gyrus and the CA3 region. In addition to the hip-
activity using subdural recordings from epileptic patients dur- pocampal formation, theta oscillations have been observed in
ing spatial navigation in computer-generated mazes and found the subiculum, entorhinal cortex, perirhinal cortex, amygdala,
that theta rhythmic activity occurred more frequently when the and cingulate cortex (74). Although these areas are capable of
subject was confronted with solving complex mazes. In addi- displaying theta oscillations that can be recorded extracellularly,
tion, theta oscillations were more evident during recall trials they are not able to generate theta activity on their own, that is,
than during learning trials. Bódizs et al. (66) recorded in the isolated from the rest of the brain, unless manipulated by intra-
human hippocampus a rhythmic activity with frequency at the cellular current injection or by the addition of pharmacologic
Chapter 3 ■ Cellular Substrates of Brain Rhythms 43

substances. In this respect it is interesting to mention that neu- frequency, can thus cause rhythmic IPSPs on their target pyrami-
rons of several limbic structures have intrinsic membrane prop- dal cells. Consequently somatic outward currents at the level of
erties that might facilitate their entrainment in theta the CA1 pyramidal layer can contribute to the theta extracellular
oscillations as shown in vitro. This was demonstrated for some field (81,90). Further cholinergic agonists applied to hippocampal
types of CA1 interneurons depolarized near spike threshold by slices cause rhythmic depolarizations of lacunosum-moleculare
current injection that generated rhythmic firing at about 7 Hz interneurons at the theta frequency, which are blocked by atropine
(75) and also for neurons of the perirhinal cortex that can dis- (91). This indicates that muscarinic induction of theta-frequency
play oscillations around 8 to 9 Hz when depolarized at thresh- membrane potential oscillations in lacunosum moleculare
old levels (76). These intrinsic properties could facilitate the interneurons may contribute to the generation of rhythmic inhi-
entrainment of these neurons in theta frequency oscillations. bition that paces intrinsically generated theta activity in CA1
pyramidal cells. In addition, the CA3 can also contribute to the
Theta Rhythm Generation generation of theta field potentials, since the recurrent network of
Since the early years of this research line several studies were CA3 pyramidal cells, and possibly hilar mossy cells, may form an
carried out with the aim of finding a system that may act as intrahippocampal oscillator that may contribute to the atropine-
theta “pacemaker.” The search for a theta pacemaker has led to sensitive component (73).
controversial results: first it was claimed that the medial A classic view (92) is that hippocampal CA1 extracellular
septo/diagonal band-hippocampal cholinergic system, driven field during theta can be accounted for by a two-dipole model.
by the brainstem reticular core, is the “pacemaker” of theta According to this view the CA1 pyramidal cells receive rhyth-
(77,78). The medial septum provides cholinergic and mic inputs in the theta frequency range from other sources that
GABAergic input to the hippocampus. While cholinergic neu- consist of atropine-sensitive and atropine-resistant inputs. The
rons innervate both principal cells and interneurons, former would drive the somata and the latter the distal den-
GABAergic projection neurons from the medial septum make drites. The atropine-sensitive theta rhythm would be mainly
synapses on hippocampal interneurons, including parvalbu- caused by Cl--mediated IPSPs on pyramidal cells (93,94).
min-positive basket cells (79). Further intrinsic voltage-dependent membrane potential oscil-
More recent investigations revealed that hippocampal lations may modulate the response to a theta frequency driving,
interneurons with long-range projections also innervate the although hippocampal pyramidal neurons do not oscillate in
cells of origin in the medial septum. Further, the supramammil- isolation (60). They show, however, in vitro, resonant properties
lary nucleus of the hypothalamus strongly connected to the in the theta frequency range that are determined by the inter-
medial septum, and to extended networks of the brainstem and play between two main ionic currents with different kinetics,
diencephalon, has also been proposed to play a role in pace- the Ih and Im currents. The former is a hyperpolarization-
making and modulating hippocampal theta (80). activated cation current and the latter is a muscarinic K cur-
The fact that theta rhythms survive after the blockade of rent. In contrast, neurons of the entorhinal cortex layer II show
cholinergic muscarinic inputs from the septal area using subthreshold oscillations at the theta frequency that appear to
atropine brought to the fore the existence of an additional non- result from the interplay of the Ih and the depolarization-
cholinergic source of theta activity in the entorhinal cortex (81). activated persistent Na current (60).
Indeed, a dipole of theta activity was found also in the entorhi- More recent studies (88), however, show that hippocampal
nal cortex with two amplitude maxima, one superficial in layers theta is associated with multiple current sinks and sources with
I–II and the other in layer III (82–85). We may conclude that a wide range of phase relationships between the firing of
the classic view of a septal/diagonal band pacemaker impinging pyramidal cells and of interneurons to the theta recorded
rhythmic activity on the pyramidal cells of the hippocampus, extracellularly from different hippocampal layers.
within the theta frequency band, is too simplistic.
Many investigations analyzed the characteristics of the two Intrahippocampal Inhibitory System
types of hippocampal theta: one atropine-sensitive and the other and Theta Generation
atropine-resistant (86). Removal of the entorhinal cortex makes The phase relationship between spike firing and the polarity of
hippocampal theta atropine-sensitive, so that it may be con- the extracellular field depends on the relative strengths of the
cluded that the atropine-resistant theta originates from the dendritic depolarization and somatic inhibition. The insightful
entorhinal inputs to the hippocampus (81,87). These inputs are review of Somogyi and Klausberger (95) summarizes the com-
glutamatergic and it was shown that NMDA synapses on the dis- plex interrelationship between the firing of different popula-
tal apical dendrites of CA1 pyramidal neurons are important tions of inhibitory interneurons and pyramidal cells during
sources of theta-generating inputs (73). Which pathways are theta oscillations (Fig. 3.8). During theta oscillations several
responsible for the atropine-sensitive theta component is still types of interneurons exhibited considerable variability in the
unclear despite many investigations using sophisticated methods relation between firing probability and the mean theta phase
such as high-density 96-site silicon probes enabling recordings recorded extracellularly. Figure 3.8 shows the relationships of
simultaneously from the dentate gyrus, CA3 and CA1 areas (88). the firing of different kinds of interneurons not only with theta
Cholinergic inputs can cause excitation of interneurons that waves, but also with high-frequency ripples and oscillations in
may be responsible for theta rhythmic discharges of hippocampal the gamma frequency range. Here we limit ourselves to a
interneurons (79,89). These interneurons, discharging at the theta description of the former. Axo-axonic cells displayed the highest
44 Part I ■ Basic Principles

Figure 3.8 Spatiotemporal interaction between pyramidal cells and several classes of interneurons during network oscillations,
shown as a schematic summary of the main synaptic connections of pyramidal cells (P), parvalbumine (PV)-positive-expressing
basket, axo-axonic, bistratified, oriens-lacunosum moleculare (O-LM), and three classes of cholecystokinin (CCK)-expressing
interneurons. The firing probability histograms show that interneurons innervating different domains of pyramidal cells fire with
distinct temporal patterns during theta and ripple oscillations, and their spike timing is coupled to field gamma oscillations to dif-
fering degrees. The same somatic and dendritic domains receive differentially timed input from several types of GABAergic
interneuron. ACh, acetylcholine. (Modified from Klausberger T, Somogyi P. Neuronal diversity and temporal dynamics: the unity
of hippocampal circuit operations. Science. 2008;321:53–57.)

firing rate around the positive peak of extracellular theta field constitute an important mechanism for the occurrence of
recorded at the level of the pyramidal layer, that is, when the epileptogenic activity in the hippocampus (97). Fuentealba et al.
average pyramidal cells are most hyperpolarized. Parvalbumin (98) also described more recently a subpopulation of
immunopositive basket cells fired during the descending phase GABAergic interneurons expressing enkephalin that innervate
of the theta oscillations were recorded at the same level. parvalbumin-positive interneurons and thus may contribute to
Bistratified cells innervating strati radiatum and oriens were the organization of rhythmic activities in CA1 area.
fired during the trough of the extracellular theta in the pyram- The peak of the intracellular theta recorded in the dendrites
idal cell layer. The oriens-lacunosum moleculare cells terminat- of pyramidal cells slightly lags the peak of the extracellular theta
ing on the most distal dendrites in conjunction with the in the pyramidal cell layer (99).
entorhinal input were also fired during the trough of the theta An important new finding (100) that further emphasizes the
field potential. role of interneurons in the generation of hippocampal theta
In addition it should be noted that some inhibitory interneu- was obtained using transgenic mice where the fast synaptic
rons make synapses with other inhibitory interneurons such as GABAergic inhibition was removed from parvalbumin-positive
described by Banks et al. (96) who demonstrated an interaction interneurons. In CA1, these interneurons comprise bistratified
between two groups of CA1 interneurons defined according to and possibly oriens-lacunosum moleculare cells that target
the kinetics of the corresponding IPSCs: GABAA generating dendrites, and basket cells that target somata and axo-axonic
IPSPs with slow kinetics and GABAA with fast kinetics that may cells (101). Each of the perisomatic inhibitory parvalbumin-
contribute to theta and gamma rhythms, respectively. Most positive cells innervates more than 1000 pyramidal cells (102)
interesting is the fact that incoming stimuli in stratum lacuno- and causes IPSPs in the pyramidal cell layer. Through synaptic
sum moleculare do not only inhibit pyramidal cells, by activat- GABAA receptors, parvalbumin-positive basket cells recipro-
ing “GABA-slow” IPSCs in these cells, but simultaneously inhibit cally inhibit each other and receive inhibition from the medial
“GABA-fast” interneurons. The authors hypothesize that this septum and other local interneurons (103). Wulff et al. (100)
inhibitory–inhibitory interaction contributes to nested found that this interneuron-specific ablation of the inhibitory
theta/gamma rhythms in the hippocampus. In Chapter 4, we septohippocampal projection can severely impair theta
describe that an alteration of this inhibitory interaction may rhythm in the hippocampus. This study implies that the
Chapter 3 ■ Cellular Substrates of Brain Rhythms 45

intrahippocampal systems of inhibition of parvalbumin- Alpha Waves: Cortical Dipolar Sources

positive cells contribute to theta generation. and Thalamic Influences
From all these data it emerges that to understand theta gen- In a series of experimental studies performed on dogs, Lopes da
eration in the hippocampus and associated areas it is necessary Silva and colleagues established that alpha rhythms are the
to integrate single neuronal data with knowledge of local net- undertaking of the cerebral cortex, mostly in the visual areas,
works, their local synaptic wiring, and the corresponding although they can also be recorded in the visual thalamus (lat-
dynamical properties. In other words, it is necessary to make eral geniculate and pulvinar nuclei) (113). In the visual cortex,
explicit computational models (104) including networks of dif- alpha waves are generated by an equivalent dipole layer cen-
ferent types of interneurons as discussed in Chapter 4, as well as tered at the level of the somata and basal dendrites of pyrami-
pyramidal neurons and granule cells; the latter should be simu- dal neurons in layers IV and V (Fig. 3.9A) (114). Furthermore,
lated using compartmental models to account for the spatial the coherence of alpha waves within the visual cortex was only
distribution of specific inputs to dendrites and somata. partially dependent on thalamic sources measured in the same
animal (Fig. 3.9B) (114,115) leading to the conclusion that hor-
ALPHA RHYTHMS izontal intracortical linkages are essential to the spread of alpha
activities, with only moderate implication of the thalamus.
This section summarizes the few elements presently known The cortical generator of alpha rhythms was confirmed and
about the mechanism(s) and origin(s) of alpha rhythm occur- extended by more recent studies performed in awake monkeys
ring in the frequency range of 8 to 13 Hz, mainly around 10 Hz. using fine microelectrode arrays implanted across the visual
Its initial description (105) also marked the beginning of EEG. cortical areas V2, V4, and the inferior temporal cortex by
A more recent and exhaustive review of different EEG patterns Bollimunta et al. (116) (Fig. 3.10). CSD analysis was combined
of alpha rhythms in the waking adult can be found in with CSD–multiple unit activity coherence to identify intracor-
Niedermeyer (106). Besides the classic alpha rhythm of the tical alpha current generators. In V2 and V4, alpha current gen-
visual cortex, there are rhythmic activities in the same fre- erators were found in all layers, with the infragranular neurons
quency range that can be recorded from the somatosensory cor- (layer V) acting as primary local generator. In contrast, in the
tex (called the mu rhythm) and the temporal cortex (called the inferior temporal cortex, alpha current generators were found
tau rhythm). Occipital alpha waves usually occur during only in supragranular and infragranular layers, with the supra-
reduced visual attention, while the mu rhythms of the somato- granular generator acting as primary local generator. Granger
motor cortex occur as the subject is in a state of muscular relax- causality analysis revealed that the influence of the infragranu-
ation. In this way, the presence of dominant activity within the lar on the supragranular cells has both a direct component and
alpha frequency range has been interpreted, since Adrian and a component mediated by the granular layer cells. Most inter-
Matthews (107), as indicating an “idling state” of the brain esting, the analysis of the coherence between CSD sinks and
although paradoxical findings with alpha enhancement by local multiunit activity in infragranular and granular layers
attention tasks (108,109) were reported. These observations are showed that these cells are depolarized and tend to generate
now well understood since a number of observations, both of action potentials during the local current sinks; however, at the
visual alpha and somatomotor mu rhythms, have shown that supragranular layer the coherence between the CSD sinks and
enhancement and attenuation of these rhythmic activities can the multiunit activity was poor, suggesting that the cells of this
be recorded simultaneously at different sites and successively layer are inhibited due to inhibitory inputs from layer V
depending on a given task. This can be explained taking into interneurons that innervate superficial pyramidal cells.
consideration the interaction between different thalamocortical
modules as explained in Chapter 45 (event-related alpha desyn-
chronization [ERD]/event-related alpha synchronization [ERS] Alpha Waves: Synaptic Processes Studied In Vitro
phenomena). In this respect, the experimental findings of Rihs Furthermore the nature of the synaptic processes underlying
et al. (110,111) showing enhancement of occipital alpha associ- alpha frequency oscillations in the cortex was studied in in vitro
ated with active suppression of unattended positions during a preparations, although it is always a matter of discussion
visual spatial orienting task provides further evidence for the whether in vitro preparations submitted to pharmacologic
facilitative role of alpha-power decreases (ERD) versus the manipulations and displaying this kind of oscillations provide
inhibitory role of alpha-power enhancement (ERS) of atten- reliable models of the situation in the intact brain, where the
tional processes. With few exceptions, the study of the cellular neurons are in different physiologic conditions. Nonetheless,
behavior underlying alpha oscillations was prevented mainly by Flint and Connors (117) were able to obtain oscillations in the
the fact that alpha rhythms appear only during wakefulness and alpha frequency range when slices of rat somatosensory cortex
no valid anesthesia model was established. The fact that the were activated by low extracellular magnesium in vitro. These
alpha band overlaps with another EEG phenomenon, the sleep oscillations were NMDA-dependent since they were reversibly
spindles (or sigma rhythms, see next section), led to a legitimate abolished by the NMDA receptor antagonist D-2-amino-
although failed attempt to study it under barbiturate anesthesia 5-phosphonovaleric acid (AP-5) but were relatively unaffected
(112). Thus, most of our present knowledge is based on scalp by the non-NMDA receptor antagonist 6,7-dinitroquinoxaline-
recordings in humans and laminar profiles of cortical field 2,3-dione (DNQX). The duration of oscillatory events was
potentials in animals. increased by blocking GABAA receptors with bicuculline or by
46 Part I ■ Basic Principles

Figure 3.9 Alpha rhythm in dog. A: (Left) Photomicrograph of a section of the marginal gyrus (visual cortex). The elec-
trode bundle consisted of seven wires with a bare tip of 0.15 mm. (Middle) An epoch of alpha activity recorded simulta-
neously from six sites corresponding to those at left, against a common frontal cranial reference (negativity upward). Note
that there is a polarity change between the most superficial sites (E25, 24) and the deepest sites (E22, 21, 19). (Right) Phase
shift between the most superficial site (E25) and deeper lying sites computed by using spectral analysis based on the aver-
age of a number of epochs within the frequency band 11.2–13.1 Hz. The mean values of the phase shift and the correspon-
ding standard deviations are indicated. (Modified from Lopes da Silva FH, Storm Van Leeuwen W. The cortical source of
the alpha rhythm. Neurosci Lett. 1977;6:237–241.) B: Schematic view of the relationship between thalamic and cortical
alpha activity. The lines indicate the amount of influence that a thalamic signal has on the coherence between a pair of
cortical signals (shaded area) recorded during alpha activity. This was measured by partializing the intercortical coherence
on the thalamic signals from the lateral geniculate nucleus (LGN) and pulvinar (PUL) or other cortical signals. Despite the
fact that PUL has an influence on corticocortical domains of alpha activity, the intercortical factors play a significant role
in the establishment of cortical domains of alpha activity. (Modified from Lopes da Silva FH, Vos JE. Mooibroek J, et al.
Relative contributions of intracortical and thalamocortical processes in the generation of alpha rhythms, revealed by par-
tial coherence analysis. Electroencephalogr Clin Neurophysiol. 1980;50:449–456.)
Chapter 3 ■ Cellular Substrates of Brain Rhythms 47

Figure 3.10 A: Multielectrode with 14 equally spaced (200 m) contacts. B: A short segment (200 msec) of alpha
rhythm. C: CSD displayed as a color-coded plot, sinks in red and sources in blue. D: Spectra profiles shown for three sites.
E: Coherence between CSD and multiunit activity (MUA) showing high coherence for the granular and infragranular lay-
ers, and in contrast with the superficial layer. (Modified from Bollimunta A, Chen Y, Schroeder CE, et al. Neuronal mech-
anisms of cortical alpha oscillations in awake-behaving macaques. J Neurosci. 2008;28:9976–9988.)(See color insert.)

activating metabotropic glutamate receptors with trans- vitro conditions may differ from those that are responsible for
1-aminocyclopentane-1,3-dicarboxylic acid (trans-ACPD). alpha rhythm generation in the intact brain. A comprehensive
A similar kind of study in slices of the cat lateral geniculate review of the basic findings with respect to the generation and
nucleus maintained in vitro (118) showed that activation of the organization of alpha rhythms, both in vivo and in vitro, can be
metabotropic glutamate receptor, mGluR1a, which is postsy- found in Ref. 119.
naptic to corticothalamic fibers induces synchronized oscilla- Remarkably, interesting additional findings in the last few
tions at alpha and theta frequencies in a subset of years have enriched our understanding of how alpha rhythms
thalamocortical neurons that can be synchronized by gap junc- are generated and have posed some new challenges for further
tions. The frequency of the oscillations, whether alpha (8 to 13 research. Lörincz et al. (120) confirmed that a sparse network of
Hz) or theta (2 to 7 Hz), depended on the concentration of the high threshold bursting thalamocortical cells coupled through
agonist trans-ACPD. These oscillations are driven by a subset gap junctions is capable of generating alpha rhythmic activity
(~25%) of lateral geniculate thalamocortical neurons display- in cat slices of the lateral geniculate nucleus and ventrobasal
ing high threshold burst activity occurring at membrane poten- complex (somatosensory thalamus). They also found that this
tials more depolarized than –55 mV, which depends on activity can be induced in this preparation by administering a
Ni2 -sensitive Ca2 channels. In these in vitro conditions the cholinergic agonist and can be abolished by muscarinic M3
alpha/theta oscillations do not appear to require chemical receptor antagonists. However, this was evident in only about
synapses because they can be sustained by coupling through 26% of the thalamocortical cells.
gap junctions between thalamocortical neurons. These in An important issue is that of the remaining majority of thal-
vitro results may indicate that activation of mGluRs of corti- amocortical cells. These authors suggest that the remaining
cothalamic synapses is an important mechanism for the gener- cells are probably under phasic inhibition from local interneu-
ation of EEG alpha and theta activity but, as stated above, one rons activated by the bursting thalamocortical cells. It still
should keep in mind that the mechanisms underlying oscilla- remains unclear how these different neuronal populations
tions in the alpha frequency range elicited in this kind of in behave in the intact brain where thalamocortical cells receive
48 Part I ■ Basic Principles

several important inputs from the cortex, RE nucleus of the travel along the dorsal thalamic relay pathway en route to the cor-
thalamus, and several brainstem sources that use various neu- tex, which probably is responsible for the genesis of the dipole
rotransmitters, particularly serotonin, noradrenalin, and hista- that allows them to surface in the EEG. Interestingly, no spindles
mine that are mainly released during the waking state but not are recorded in the EEG of kittens during the first 6 to 7 days of
during REM sleep (121). An important input to these thalamic life, although they are already present in thalamic recordings 3 to
nuclei is the cholinergic input that arises from brainstem and 4 hours after birth (129), suggesting that no viable dipole is gen-
basal forebrain nuclear groups. It has been shown that cholin- erated in the thalamus. The brainstem and basal forebrain pro-
ergic inputs are excitatory to thalamocortical cells causing a jections exert activating effects on the above-mentioned circuit
direct depolarization associated with a decrease of a K cur- with the result of inhibiting spindles.
rent; at the same time there is a decrease of the inhibitory input The RE nucleus has been pointed out as the pacemaker for
arising from the thalamic RE neurons, since the latter are spindle oscillations because spindles were abolished in the dor-
hyperpolarized by cholinergic activation of a K conductance, sal thalamus after disconnection from the RE (130) but were
as was shown in vitro in guinea pig (122) and in vivo in cat preserved in the rostral part of the RE nucleus severed from the
(123,124). In this way these cholinergic inputs favor the block- dorsal thalamus (131). Further support is provided by the fact
age of rhythmic spindles, a phenomenon that is generated in that in cat’s anterior nucleus, which does not receive inputs from
the thalamus and shares the same frequency range with alpha the RE nucleus (132,133), spindles are absent (134). Similarly,
oscillations (see below). It is possible that the major effect of spindles are absent in the cingular cortex (74) and habenular
cholinergic inputs on thalamic oscillations in vivo may be to nuclei, other structures that are devoid of RE inputs. Modeling
induce their blockage through muscarinic inhibition of RE studies of isolated RE neurons, with minimal or more realistic
thalamic neurons, although more subtle effects appear to exist ionic models of RE cells (135–137), further strengthened the
in some thalamic subpopulations, but the relevance of these idea that the deafferented RE nucleus is a spindle pacemaker.
effects for alpha oscillatory activity is not yet clear. These recent Spindle oscillations, in order to produce a coherent pattern
findings provide new insights but it is evident that more at the level of the target structures, need to be synchronized at
research in in vivo conditions is necessary in order to achieve a the very site of their genesis. Besides the axonal projections
deeper understanding of the mechanisms responsible for the allowing obvious connectivity between RE neurons, two other
generation of alpha oscillations of different kinds and localization. mechanisms have been proposed. The first relies on the dendro-
dendritic GABAergic contacts between RE neurons (138–140)
SPINDLE (SIGMA) RHYTHMS and the second on the functional interconnection of RE neu-
rons through gap junctions (141).
Classically, spindles have been regarded as one of the first signs A fundamental question concerns the triggering factor of a
of EEG synchronization during the early stage of quiescent spindle. Steriade and colleagues (131) have assumed that any
sleep. This type of oscillation is defined by the presence of two excitatory drive stimulating RE cells would start the process.
distinct rhythms: waxing and waning spindle waves at 7 to The decreased activity of brainstem cholinergic neurons at the
14 Hz within sequences lasting for 1 to 2 seconds, and the peri- onset of sleep (53) contributes to the overall hyperpolarization
odic recurrence of spindle sequences with a slow rhythm, gen- of thalamocortical cells, thus bringing the membrane potential
erally 0.2 to 0.5 Hz (Fig. 3.11A). Spindles are generated within in the range where bursting discharges can occur. Such clusters
the thalamus because they survive in the thalamus after decor- of high-frequency action potentials would excite the dendrites
tications and high brainstem transection (125). More radical of RE neurons and would trigger the dendrodendritic ava-
procedures, including complete removal of the striatum and lanche leading to the final synchronization of the whole RE
rhinencephalon, do not affect thalamic spindles either (14). nucleus. It should be emphasized that thalamic-RE loops may
Sleep spindles have been mostly studied in cats (in vivo) and assist the onset of a spindle; however, its synchronization can-
ferrets (in vitro). Humans and cats have similar sleep cycles, not rely on the dorsal thalamic circuitry because there is scarce,
EEG patterns, and ultrastructural organization of their thala- if any, communication between thalamocortical neurons and
mus. Both intrinsic properties of various thalamic neurons dorsal thalamic nuclei (142,143).
(126) and network linkages (for review, see Ref. 127) contribute Once a spindle has been started through one of the mecha-
to the patterning of spindles. nisms mentioned above, the pacing of the actual oscillations
As indicated above, the thalamic circuits comprise two main within the spindle sequence depends on the RE–thalamic dia-
structures: (i) the dorsal thalamus, made of several nuclei, each of logue. This was investigated in ferret thalamic visual slices con-
them containing both relay (thalamocortical) and local-circuit taining geniculate and perigeniculate (RE) nuclei (144–146) and
(inter-) neurons and (ii) the RE nucleus of the thalamus (see with multisite recordings within the thalamus (147,148). Results
scheme in Fig. 3.11A). The latter is a thin sheet of GABAergic show that the bursting of RE neurons imposes powerful
neurons that covers the rostral, lateral, and ventral surfaces of the GABAergic IPSP in thalamocortical neurons. The end of this inhi-
thalamus (128). It mainly receives inputs from the cerebral cor- bition triggers a rebound LTS, crowned by a high-frequency burst
tex and dorsal thalamus, but also from the rostral sector of the of action potentials, which in turn will again excite the target RE
brainstem and basal forebrain. The RE–thalamic–cortical–RE cells. This is also the moment where intrinsic properties of thala-
loop constitutes a resonating circuit that reinforces the spindle mic, both thalamocortical and RE, neurons play an important
oscillation. Moreover, spindles generated within the RE nucleus role. The amount of hyperpolarization of the thalamocortical cells
Chapter 3 ■ Cellular Substrates of Brain Rhythms 49

Figure 3.11 Spindle oscillations in reticular

thalamic (RE), thalamocortical (Th-Cx, ventro-
lateral nucleus), and cortical (Cx, motor area)
neurons. A: Circuit of three neuronal types and
two rhythms (7–14 Hz and 0.1–0.2 Hz) of spin-
dle oscillations in cortical EEG. B: Intracellular
recordings in cats under barbiturate anesthesia.
(Modified from Steriade M, Deschênes M.
Intrathalamic and brainstem-thalamic networks
involved in resting and alert states. In:
Bentivoglio M, Spreafico R, eds. Cellula r
Thalamic Mechanisms. Amsterdam: Elsevier;

determines the degree of deinactivation of the LTS (149). Another In an intact sleeping brain, the RE–thalamic network under-
intrinsic property of thalamic cells is the voltage-dependent, non- goes periodic, synchronized excitatory inputs from the cortex
inactivating or very slowly inactivating persistent sodium current, such as K-complexes (see above), which then shape and modu-
INa(p) (149), which modulates postinhibitory rebound depolariza- late the duration of spindles (153). Corticothalamic stimuli
tions. The blockage of the INa(p) by intracellular injection of qua- produce shorter and waning spindles that lack the initial wax-
ternary derivatives of local anesthetics transformed spindles of ing component, probably because the volleys entrained from
thalamocortical cells into a single, long-lasting period of hyperpo- the very beginning large populations of thalamic cells.
larization (because unopposed by INa(p)) without any rhythmic Moreover, the synchronization of spindles in the thalamus was
rebounds within the frequency range of spindles (150). Finally, widespread, with virtual zero time lag, in intact preparations,
the calcium-dependent potassium current, IK(Ca) , or other but dropped drastically after decortications (147,148) (Fig. 3.12).
voltage-dependent potassium currents (151,152) may prolong the This result stands in some contrast to a propagating pattern of
long-lasting IPSPs generated in thalamocortical neurons by RE spindles in thalamic slices (154) likely due to the absence of
neurons and, thus, assist in the production of the LTS and in corticothalamic inputs of this preparation and to the overall
inducing rebound bursts in thalamocortical targets. diminished synaptic activity of the slice.
Figure 3.12 Control of spatiotemporal coherence of thalamic spindles by the cerebral cortex in cat. Top panels: Disruption of the spatiotemporal coher-
ence of thalamic oscillation after removal of the neocortex. Spatiotemporal maps of electrical activity across the thalamus were constructed by plotting
time (time runs from top to bottom in each column; arrow indicates 1 second), space [from left to right, the width of each column represents 8 mm in
the anteroposterior axis of the thalamus), and local field potential (LFP) voltage (from blue to yellow, color represents the amplitude of the negative
deflection of thalamic LFPs; the color scale ranged in ten steps from the baseline [blue] to 100 V [yellow]). Time was divided into frames, each rep-
resenting a snapshot of 4 msec of thalamic activity. A total of 40 seconds is represented (9880 frames). Each frame consisted of eight color spots, each
corresponding to the LFP of one electrode from anterior to posterior (left to right in each column). Middle panels (experiments): Decay of correlation
with distance. Cross-correlations were computed for all possible pairs of thalamic sites, and the value at time zero from each correlation was repre-
sented as a function of the intersite distance for six different consecutive epochs of 20 seconds. Spatial correlation was calculated for thalamic record-
ings in the intact brain (left) and after removal of cortex (right). Bottom panel (model): Decay of correlation with distance (in units of sites). Similar
computation of cross-correlations as in the above panel from experiments, in the presence of cortex (left) and after decortication (right). (Modified from
Contreras D, Destexhe A, Sejnowski TJ, et al. Control of spatiotemporal coherence of a thalamic oscillation by corticothalamic feedback. Science.
1996;274:771–774. and unpublished data by A Destexhe, D Contreras, TJ Sejnowski and M Steriade.) (See color insert.)
Chapter 3 ■ Cellular Substrates of Brain Rhythms 51

The main functional correlate of sleep spindles is the block- a conditioned response to a visual stimulus in monkey (163),
age of incoming stimuli on their way to the cortex. This could during tasks requiring fine finger movements and focused atten-
result from the fact that thalamocortical neurons alternate tion in monkey motor cortical cells (164), and during behavioral
between IPSPs and LTS bursts. Both events preclude the relay- immobility associated with an enhanced level of vigilance while
ing of afferent synaptic signals either due to the shunting nature a cat was watching a visible but unseizable mouse (165). Other
of the former or due to the all-or-none feature of the latter. The studies have described stimulus-dependent oscillations at 25 to
functional blockage of sensory transmission at the thalamic 45 Hz of the focal EEG and/or neuronal firing in the olfactory
level promotes the cortical deafferentation, further contribut- system (166) and visual cortex (167–171).
ing to the process of falling asleep. The underlying mechanism responsible for cortical gamma
A final observation concerns the overlapping frequency oscillations was further unraveled by the finding that synchro-
range of spindles and alpha waves. It should be emphasized that nized gamma frequency oscillations are enhanced with arousal
both behavioral context and mechanisms are dissimilar. Alpha and attention (172) or by electrical stimulation of the mesen-
waves generally occur during relaxed wakefulness, and in some cephalic RE formation (173), and are mediated by acetylcholine
studies they are considered to regulate afferent and efferent sig- muscarinic receptors. In addition, the spectral power of local field
nals (155). At difference, spindle oscillations are occurring dur- potentials in the gamma frequency range in the visual cortex of
ing unconsciousness, thus making the idea of an the cat is closely correlated with hemodynamic signals (174).
alpha-to-spindle continuum obsolete. Nonetheless the same The main functional implication of these rhythms in the cor-
basic neural circuitries seem to be involved in both kinds of tex was proposed to rely on the degree of spatial and temporal
oscillatory activities—sleep spindles and alpha rhythms—but synchronization, thus allowing, at a given moment, the aggrega-
working according to different functional modes. tion of various cortical areas with the purpose of creating global
and coherent properties of patterns, a prerequisite for scene seg-
FASTER (BETA AND GAMMA) RHYTHMS mentation and figure-ground distinction (175). The issue of this
“feature binding” has, however, been toned down by the fact that
At the opposite side of the EEG spectrum of slow sleep rhythms such fast rhythms are coherent also within thalamocortical and
( 15 Hz) are fast oscillations (generally 20 to 50 Hz) that are cortical networks during states such as sleep and deep anesthe-
proper to vigilance states associated with wakefulness, but also sia reputed for the absence of any conscious behavior
paradoxical (REM) sleep. The brain substrate of EEG activation (19,176,177). In these studies it was established that thalamocor-
upon arousal has begun to be understood since the pioneering tical cells do spontaneously oscillate within the beta–gamma fre-
work of Moruzzi and Magoun (156). They reported an “activa- quency range and this occurs coherently with field potentials
tion” response to the stimulation of brainstem structures con- recorded from the related cortical areas (Fig. 3.13).
sisting of the suppression of spindles and slower EEG rhythms. Most likely there exist different kinds of beta–gamma
The cellular mechanisms of this suppressing mechanism have rhythms with different properties and behavioral/cognitive
recently been analyzed: the blockage of spindles occurs at the associations. Since the report by Gray and Singer (169) that the
very site of their genesis (the RE nucleus) through the action of firing probability of neurons of the visual cortex of the cat, in
acetylcholine serotonin and norepinephrine. Although acetyl- response to the presentation of appropriate visual stimuli, oscil-
choline hyperpolarizes RE neurons (124,157), while the latter lates with a frequency in the gamma frequency range, along
two depolarize them (158), the combined action of these neu- with the observation that no evidence for similar oscillations
rotransmitters is far from being understood. The intrinsically were found in the thalamus, these gamma oscillations are con-
generated thalamic delta oscillation is blocked through the sidered to be generated intracortically. This observation led to
depolarizing action of both acetylcholine (10) and the assumption that these gamma oscillations may reflect a gen-
monoamines (159) on thalamocortical cells. Finally, slow corti- eral mechanism that is capable of binding together activities of
cal delta activities are prevented by cholinergic actions of the spatially separate cortical areas (179). In general we may state
nucleus basalis neurons (30,160). that synchronization of neuronal networks by way of common
oscillations in the gamma frequency range may enable fast
Beta–Gamma Rhythms: Functional Implications routing and processing of information in the cortex (180).
The disappearance of slow sleep waveforms during the activat-
ing response is also accompanied by the appearance of peculiar Genesis of Beta–Gamma Rhythms
fast rhythms characterizing wakefulness (161). This study The genesis of fast (beta–gamma) activities lies at the crossroads
reported, in addition to the ocular syndrome of arousal, a clear- of intrinsic and network properties. Cortical neurons generate
cut enhancement in amplitude of the spontaneous EEG rhythms upon depolarization intrinsic oscillations in a range around 40
and their regular acceleration to 40 to 45 Hz instead of the till Hz (181,182). This is equally valid for a subclass of rostral
then observed flattening of the cortical EEG. Since then, several intralaminar thalamic neurons (183) (Fig. 3.14), a property that
papers have mentioned the presence of 20 to 40 Hz waves in var- is in line with their implication in fast oscillations during acti-
ious cortical areas, during different conditions of increased vated states associated with neuronal depolarizations, waking,
alertness. For instance, fast rhythms were observed in a canine and REM sleep. Intralaminar nuclei are particularly fit for
subject in the occipital cortex while the dog paid intense atten- wide range synchronization due to their widespread projections
tion to a visual stimulus (162), during accurate performance of to the cerebral cortex (125) including the visual areas (184).
52 Part I ■ Basic Principles

Figure 3.13 Episodes of tonic activation are

associated with coherent fast rhythms (40 Hz)
in cortical EEG and intracellularly recorded
thalamocortical neuron. Cat under ketamine-
xylazine anesthesia. A: Four traces represent
simultaneous recordings of surface and depth
EEG from motor cortical area 4, extracellular
discharges of neuron from the rostrolateral part
of the thalamic reticular (RE) nucleus, and
intracellular activity of thalamocortical neuron
from ventrolateral (VL) nucleus. EEG, RE, and
VL cells displayed a slow oscillation (0.7–
0.8 Hz) during which the sharp depth-negative
(excitatory) EEG waves led to IPSPs in VL cell,
presumably generated by spike bursts in a cor-
tically driven GABAergic RE neuron. Part
marked by horizontal bar, taken from a short-
lasting period of spontaneous EEG activation, is
expanded in B (arrow), with EEG waves and
field potentials from the RE nucleus filtered
between 30 and 50 Hz; part marked by horizon-
tal bar in this panel is further expanded in C to
illustrate relations between action potentials of
VL cell and depth-negative waves in cortical
EEG at a frequency of 40 Hz. D: Cross-
correlations (CROSS) between action potentials
and depth-EEG shows clear-cut relation, with
opposition of phase, between intracellularly
recorded VL neuron and EEG waves. (Modified
from Steriade M, Contreras D, Amzica F, et al.
Synchronization of fast (30–40 Hz) sponta-
neous oscillations in intrathalamic and thalam-
ocortical networks. J Neurosci. 1996;16:

Despite these intrinsic properties that make neurons prone somatosensory slices with slightly reduced inhibition display
to generate fast oscillations, their imbedding in complex cir- activities around 37 Hz generated by networks of intrinsically
cuits, especially in the cortex, is required for the short- and pyramidal-shaped (excitatory) cells.
long-range synchronization, in order to make the rhythm The complexity of the cortical generators of fast
emerge at the EEG level. As examples of different kinds of (beta–gamma) rhythmic activities is illustrated further by the
circuits that may engage fast rhythmic activities we may men- observation that, under anesthesia, they do not show field rever-
tion the following two findings: Freeman (180) postulated that sal at any depth of the cortex (19). Volume conduction was ruled
the generation of 40- to 80-Hz activity in the olfactory bulb out because the negative field potentials of the fast oscillations
depends on feedback inhibitory circuits involving local-circuit were systematically associated, at all depths, with neuronal firing
GABAergic neurons acting on output elements, the mitral cells, and were not observable in the underlying white matter. This
while Chagnac-Amitai and Connors (185) showed that rat fact can be interpreted as being due to the existence of several
Chapter 3 ■ Cellular Substrates of Brain Rhythms 53

Figure 3.14 Fast oscillatory patterns of neuron recorded from the dorsolateral part of the centrolat-
eral (CL) intralaminar thalamic nucleus, characterized by exceedingly high-frequency (800 to
1000 Hz) spike bursts recurring rhythmically at 50 to 60 Hz. Intracellular recording in cat under bar-
biturate anesthesia. A: Activities triggered by depolarizing current pulses ( 1.2 nA, 50 msec) at dif-
ferent levels of membrane potential (as indicated at left). Two examples are illustrated for each
membrane potential level. At bottom, presence of low-threshold spike (LTS) leading to high-
frequency (800 to 1000 Hz) spike burst at –78 mV and its absence at –84 mV. Note fast-recurring (50
to 60 Hz) spike doublets or spike triplets at relatively depolarized levels (–64 and –58 mV), a feature
that is not seen in other types of thalamocortical neurons. B: Oscillatory patterns similar to those
elicited by current injection occurred spontaneously at similar membrane potentials (from top to bot-
tom, –58, –64, –68, and –78 mV). (Modified from Steriade M, Curró Dossi R, Contreras D.
Electrophysiologic properties of intralaminar thalamocortical cells discharging rhythmic (approxi-
mately 40 Hz) spike-bursts at approximately 1000 Hz during waking and rapid eye movement sleep.
Neuroscience. 1993;56:1–9.)

intracortical distributed microsinks and microsources as shown increased levels of alertness, thus closely following the onset of
by current-source-density analyses (176). Thus, the absence of activity in cholinergic aggregates of the brainstem and basal
potential reversal was due to the lower intensity of vertical cur- forebrain. The cholinergic activation induces a twofold increase
rents, compared with that of transmembrane currents. of cortical EEG waves around 40 Hz, a potentiating effect that
Despite their presence during sleep and under anesthesia, is mediated by muscarinic receptors as it is blocked by scopo-
fast (beta–gamma) oscillations are mainly associated with lamine (30,183). Beyond the depolarizing effect exerted by
54 Part I ■ Basic Principles

acetylcholine on thalamocortical and cortical neurons, this a remarkable quantitative indicator of the pharmacologic effect
neurotransmitter hyperpolarizes a subpopulation of cortical of benzodiazepines. In addition to the classic beta and gamma
glial cells maintained in culture (186). When tested in vivo, oscillations, episodes of oscillations at high frequency (100 to
acetylcholine hyperpolarized most of the glial cells (Fig. 3.15), 600 Hz) and short duration (~50 to 100 ms) have been
in parallel with an overall decrease in the extracellular potas- described in the brain of rodents and human, both under nor-
sium concentration (187) (Fig. 3.15; see also next section). This mal and epileptic conditions. These oscillatory bursts are called
effect was also accompanied by an increased cerebral blood ripples (see Fig. 3.8) or fast ripples depending on frequency and
flow, leading to the interpretation that the glial hyperpolariza- are discussed in Chapter 37.
tion is due to a boosted transport of potassium across capillary
membranes. This is further supported by the fact that EEG acti- Fast (Beta–Gamma) Rhythms in Human EEG
vation, when recorded with DC amplifiers, generally displays a While the mechanisms of generation of gamma and beta
persistent positive DC shift (187). In addition, brain activation rhythms in cortical slices, in vitro, present clear different prop-
was accompanied in glial cells by a reduced membrane capaci- erties, the situation in vivo in humans is even more complex.
tance suggesting that the interglial syncytium shuts down dur- On the one hand, beta EEG or MEG activities recorded over the
ing wakefulness, preventing nonspecific synchronization Rolandic cortical area (about 15 and 30 Hz) display a close tem-
through spatial buffering mechanisms (as mentioned previ- poral relationship with peripheral EMG activity during isomet-
ously). This feature may thus leave the synchronization of ric contractions, as shown using MEG recordings (190–192).
beta–gamma activities strictly confined to synaptic networks, The MEG signals in the beta frequency range lead the EMG sig-
achieving a more specific aggregation of the neuronal popula- nals in time, and the time lag increases with increasing
tions involved. brain–muscle distance. This implies that this beta rhythmic
Recent studies in cortical slices revealed particularly interest- activity is associated with rhythmic firing of neurons of the
ing properties with respect to the generation of gamma and motor cortex that generate the commands to drive spinal
beta rhythmic activities. Roopun et al. (188) showed that motoneurons (193) (see also Chapter 42). On the other hand,
gamma (30 to 70 Hz) rhythms could be generated in the super- however, there are other conditions where beta rhythms are not
ficial somatosensory cortical layers II/III, while beta rhythms associated with a state of active neuronal firing in the sense
(20 to 30 Hz) were generated in deep layer V mainly associated described above. This is the case of an increase of beta activity
with the activity of fast-spiking interneurons when kainate was after a finger movement (postmovement beta rebound at 16 to
applied to the slices. It is interesting to note that in layer IV both 21 Hz) or foot movement (19 to 26 Hz) (194) when the mus-
activities may coexist (Fig. 3.16). A remarkable difference in the cles relax. The combination of EEG recording with the applica-
physiologic mechanisms underlying these two types of cortical tion of transcranial magnetic stimulation showed that during
oscillations is that the beta rhythms of the deep layers are this beta rebound the corticospinal excitability was diminished
reduced by carbenoxolone, which blocks gap junction conduc- (195). The dynamics of changes of beta rhythms under differ-
tances, but are not affected by the blockade of synaptic trans- ent behavioral conditions is further detailed in Chapter 45.
mission. In contrast, gamma rhythms in superficial layers are Related to the beta rebound after a movement it was shown that
reduced by the blockade of chemical synaptic transmission. a surge of gamma activity (above 30 Hz) may immediately pre-
The fact that gamma and beta oscillations rely on different cede a finger movement (196), which is probably directly asso-
mechanisms is further emphasized in experiments with trans- ciated with the activation of motor cortical neurons.
versal cuts within the somatosensory cortical slices that com-
pletely separate layers I to II from layers V to VI. The gamma DC EEG POTENTIALS
rhythmic activity of the superficial layers and the beta rhythmic
activity of the deep layers survived the cut, showing that they Although the chapter is devoted to oscillations in the brain, this
can be maintained separately. The two kinds of rhythmic activ- section will briefly raise the issue of less conventional, very slow
ities also respond differently to pharmacologic agents. While EEG signals, which are overwhelmingly left out of the clinical
the blockage of GABAA receptors in these slices abolished focus (see, however, Refs. 197 and 198). Interestingly, the first
gamma rhythms, it enhanced beta rhythms. This means that EEG machines used by Berger for his first EEG recordings were
gamma oscillations depend on the discharge of pyramidal cells true DC amplifiers. This changed with the introduction of
coupled to GABA-mediated local inhibitory circuits. We should increasingly performing filters offering the possibility to limit
emphasize, however, that demonstrations of this kind of prop- the intrusion of artifacts in clinical recordings. Filters also pre-
erties in vitro are not a guarantee that precisely the same mech- cluded the expression of neurophysiologic relevant informa-
anisms are also essential in vivo. Nonetheless in vitro studies as tion, which, in time, limited the knowledge of brain
this one yield a proof-of-principle that such mechanisms can be mechanisms. The consecutive conventional theories in elec-
operational in the cortex. troencephalography considered that most of the electrical
It was reported in the rat in vivo that benzodiazepines, which activity recorded on the scalp, including very slow waves, is gen-
enhance GABAA receptor-mediated synaptic transmission, also erated by neuronal networks with particular emphasis on corti-
increased beta oscillations in the EEG. In this sense, Mandema cal dipoles. Several studies have also demonstrated the
and Danhof (189) reviewed data showing that changes in the involvement of glial cells in generating slow local field poten-
EEG spectral amplitude within the beta frequency band provide tials during spreading depression (199,200), seizures
Chapter 3 ■ Cellular Substrates of Brain Rhythms 55

Figure 3.15 Comparison between electrically elicited and spontaneously occurring activation in a pair of simultane-
ously impaled neuron and glia in a cat under ketamine-xylazine anesthesia. A: Sequence of slow ( 1 Hz) oscillation,
brainstem cholinergic activation, and return to slow oscillation. Depolarizing phases of the slow oscillation are indi-
cated with a black arrow head (up state), while hyperpolarizing phases are marked with an empty arrow head (down
state). After activation, the sustained neuronal depolarization is accompanied by glial hyperpolarization and the DC
field potential assumes rather positive values. B: Histograms of neuronal (panel 1) and glial (panel 2) Vm during the
slow oscillation (in gray) and during activation (black trace). Histograms were calculated over the respective under-
lined epochs in A. They illustrate the incidence of different values of Vm , sampled at 20 kHz, over periods indicated
by bar below traces. Note the bimodal histograms during sleep (up states correspond to the black arrow head, down
states to the empty arrow head) and the unimodal histogram after activation, as well as the opposite evolution of glial
and neuronal Vm (glia hyperpolarizes and neuron depolarizes). C1: Similar pattern of activity in the same double intra-
cellular recording during a period of spontaneous activation of the EEG. Periods within squares are expanded at right
(panels 2 and 3). D: Histograms from equivalent time periods before and after activation. (Modified from Seigneur J,
Kroeger D, Nita DA, et al. Cholinergic action on cortical glial cells in vivo. Cereb Cortex. 2006;16:655–668.)
56 Part I ■ Basic Principles

Figure 3.16 2 rhythms are generated in deep layers of cortex. A: Spectrograms of field rhythms generated
in somatosensory cortex slices by 400 nM kainate. Superficial layers generated gamma frequency (30 to
50 Hz) signals (layer II/ III). Deep layers concurrently generated 2 frequency signals (layer V, 20 to 30 Hz).
Layer IV recordings show both frequency bands coexisting. Below each spectrogram are representative field
potential traces (scale bars: 0.2 mV, 100 ms). B: Surgical separation of deep from superficial layers at the layer
IV/ V border abolished neither rhythm. Cartoon illustrates the electrode position for the power spectra taken
from 60-second epochs of field potential data in the superficial layers (a–c, black lines) and deep layers
(d–e, red lines). (Modified from Roopun AK, Middleton SJ, Cunningham MO, et al. A 2-frequency (20–30 Hz)
oscillation in nonsynaptic networks of somatosensory cortex. Proc Natl Acad Sci USA. 2006;103:
15646–15650.)(See color insert.)

(199–203), and sleep (36). Besides these concepts, pioneering mechanisms generating ventilation-related DC shifts that
studies in the early 1950s to 1970s have proposed that some of emphasized an almost exclusive role of cortical neurons and their
the slow potential shifts are generated at the interface between dendritic tree (201,213). Recent data (212) demonstrate that such
cerebrospinal fluid and blood as a function of the partial pres- ventilation-induced DC shifts occurred in the absence of any
sure of CO2 (PCO2) (204–208). Such potentials were suggested parallel change in the neuronal or glial membrane potential. No
to originate across the blood–brain barrier (209–212). The extracellular, spatial polarity reversal of the DC shift responses
merit, but also the difficulty, of these studies consisted in intro- was seen across or within the neocortex and the underlying white
ducing among the variables that condition brain phenomena matter. This finding provides further support for the view that
extraneuronal parameters. the DC shifts are not caused by cortical current dipoles generated
The basic aspects related to EEG potentials situated at the in response to neuronal and/or glial activity. And finally, the
infraslow limit of the spectrum (DC to 0.01 Hz, although the breakdown of the blood–brain barrier produced itself a DC shift
upper limit is only informative) rely on the electrical scheme as mark of the polarizing potential of the barrier. Once the
proposed by Voipio and colleagues (211) (Fig. 3.17A). It proved blood–brain barrier was disrupted, ventilation-induced DC
to be particularly useful to understand how large DC shifts that shifts were abolished (Fig. 3.18). These examples clearly state
can be recorded on the scalp in response to hypo/hypercapnia that, other than neurons and glia, a series of other parenchymal
(Fig. 3.17C,D) are not generated by networks of neurons and/or elements (blood flow and capillary epithelial cells) participate in
glia. This is in stark contrast with the prevailing view about the the genesis of potentials that translate into EEG signals.
Chapter 3 ■ Cellular Substrates of Brain Rhythms 57

Figure 3.17 DC shift potentials generated in the EEG by variations of the systemic CO2. A left: Schematic
drawing of the human head divided into four compartments: brain, blood, the blood–brain or blood–CSF bar-
rier (black double line), and all other tissues. EBB, the electromotive force of the voltage source across the
brain–blood interface; RBB, its internal resistance. This voltage source generates a volume current (lines with
arrowheads) that flows first through RB (the distributed resistance that couples brain potential to the
surrounding extracortical tissue layers) and RS (the distributed resistance of the layers between brain surface
and skin surface pooled together) and gives rise to the voltage drop VDC that can be measured on scalp.
Current returns back to the brain–blood interface through RT1 (resistance of wider tissue pathways below the
level of cranial fossae), RT2 (access resistance to blood), and RT3 (resistance of blood). A right: Simplified
equivalent circuit of the scheme depicted at left. IBB denotes the current that is driven in the circuit by the
brain–blood potential difference (VBB). (Modified from Voipio J, Tallgren P, Heinonen E, et al. Millivolt-scale
DC shifts in the human scalp EEG: evidence for a nonneuronal generator. J Neurophysiol.
2003;89:2208–2214.) B: Scheme with the position of the scalp electrodes with respect to the brain of the cat.
DC shifts induced by hyperventilation (C) and hypoventilation (D) in a cat under ketamine-xylazine anesthe-
sia. Below the EEG signals, variations of the CO2 concentration in the expired air. Hyperventilation induced
positive DC shifts, while hypoventilation was associated with negative DC shifts. (Modified from Nita DA,
Vanhatalo S, Lafortune FD, et al. Nonneuronal origin of CO2-related DC EEG shifts: an in vivo study in the
cat. J Neurophysiol. 2004;92:1011–1022.)
58 Part I ■ Basic Principles

Figure 3.18 DC shifts induced by the disruption of the blood–brain barrier. Scalp DC EEG recording in a cat
under ketamine-xylazine anesthesia. After a control period, a volume of 3 ml saline (SAL) was slowly injected
into the right carotid artery with no effect. Then an identical volume of sodium dehydrocholate (DHC; 17.5% )
was injected in order to break the blood–brain barrier. This induced a clear positive DC shift, with right lat-
eralization. During the period with open blood–brain barrier, the amplitude of the responses to hypo/ hyper-
ventilation (hv/ HV) maneuvers was drastically diminished. The asterisk during the first hypoventilation
maneuver indicates that initially for a period of about 1 minute the respirator was stopped. (Modified from
Nita DA, Vanhatalo S, Lafortune FD, et al. Nonneuronal origin of CO2-related DC EEG shifts: an in vivo study
in the cat. J Neurophysiol. 2004;92:1011–1022.)

REFERENCES 10. Steriade M, Curró Dossi R, Nuñez A. Network modulation of a

slow intrinsic oscillation of cat thalamocortical neurons impli-
1. Walter G. The location of cerebral tumors by electro- cated in sleep delta waves: cortically induced synchronization
encephalography. Lancet. 1936;8:305–308. and brainstem cholinergic suppression. J Neurosci. 1991;
2. IFSECN. A glossary of terms most commonly used by clinical elec- 11:3200–3217.
troencephalographers. Electroencephalogr Clin Neurophysiol. 11. Curró Dossi R, Nuñez A, Steriade M. Electrophysiology of a slow
1974;37:538–548. (0.5–4 Hz) intrinsic oscillation of cat thalamocortical neurones in
3. Creutzfeldt OD, Watanabe S, Lux HD. Relations between EEG phe- vivo. J Physiol. 1992;447:215–234.
nomena and potentials of single cortical cells. II. Spontaneous and 12. Nuñez A, Curró Dossi R, Contreras D, et al. Intracellular evidence
convulsoid activity. Electroencephalogr. Clin Neurophysiol. for incompatibility between spindle and delta oscillations in thal-
1966;20:19–37. amocortical neurons of cat. Neuroscience. 1992;48:75–85.
4. Kellaway P, Gol A, Proler M. Electrical activity of the isolated cerebral 13. Nita DA, Steriade M, Amzica F. Hyperpolarisation rectification in
hemisphere and isolated thalamus. Exp Neurol. 1966;14:281–304. cat lateral geniculate neurons modulated by intact corticothalamic
5. Rappelsberger P, Pockberger H, Petsche H. The contribution of the projections. J Physiol. 2003;552:325–332.
cortical layers to the generation of the EEG: field potential and 14. Villablanca J. Role of the thalamus in sleep control: Sleep-
current source density analyses in the rabbit’s visual cortex. wakefulness studies in chronic diencephalic and athalamic cats. In:
Electroencephalogr. Clin Neurophysiol. 1982;53:254–269. Petre-Quadens O, Schlag J, eds. Basic Sleep Mechanisms. New York:
6. Leresche N, Jassik-Gerschenfeld D, Haby M, et al. Pacemaker-like Academic; 1974:51–81.
and other types of spontaneous membrane potential oscillations 15. Ball GJ, Gloor P, Schaul N. The cortical electromicrophysiology of
of thalamocortical cells. Neurosci. Lett. 1990;113:72–77. pathological delta waves in the electroencephalogram of cats.
7. McCormick DA, Pape HC. Properties of a hyperpolarization- Electroencephalogr. Clin Neurophysiol. 1977;43:346–361.
activated cation current and its role in rhythmic oscillation in 16. Steriade M, Buzsáki G. Parallel activation of thalamic and cortical
thalamic relay neurones. J Physiol. 1990;431:291–318. neurons by brainstem and basal forebrain cholinergic systems. In:
8. Leresche N, Lightowler S, Soltesz I, et al. Low-frequency oscillatory Steriade M, Biesold D, eds. Brain Cholinergic Systems. New York:
activities intrinsic to rat and cat thalamocortical cells. J Physiol. Oxford University Press; 1990:3–62.
1991;441:155–174. 17. Steriade M, Gloor P, Llinás RR, et al. Report of IFCN committee
9. Soltesz I, Lightowler S, Leresche N, et al. Two inward currents and on basic mechanisms. Basic mechanisms of cerebral rhythmic
the transformation of low-frequency oscillations of rat and cat activities. Electroencephalogr Clin Neurophysiol. 1990;76:
thalamocortical cells. J Physiol. 1991;441:175–197. 481–508.
Chapter 3 ■ Cellular Substrates of Brain Rhythms 59

18. Steriade M, Nuñez A, Amzica F. A novel slow (< 1 Hz) oscillation 39. Bowman CL, Kimelberg HK. Excitatory amino acids directly
of neocortical neurons in vivo: depolarizing and hyperpolarizing depolarize rat brain astrocytes in primary culture. Nature.
components. J Neurosci. 1993;13:3252–3265. 1984;311:656–659.
19. Steriade M, Amzica F, Contreras D. Synchronization of fast (30–40 40. Kettenmann H, Schachner M. Pharmacological properties of
Hz) spontaneous cortical rhythms during brain activation. J gamma-aminobutyric acid-, glutamate-, and aspartate-induced
Neurosci. 1996;16:392–417. depolarizations in cultured astrocytes. J Neurosci. 1985;5:
20. Achermann P, Borbély AA. Low-frequency (<1 Hz) oscillations in the 3295–3301.
human sleep electroencephalogram. Neuroscience. 1997;81:213–222. 41. Orkand RK, Nicholls JG, Kuffler SW. Effect of nerve impulses on
21. Amzica F, Steriade M. The K-complex: its slow (<1-Hz) rhythmic- the membrane potential of glial cells in the central nervous system
ity and relation to delta waves. Neurology. 1997;49:952–959. of amphibia. J Neurophysiol. 1966;29:788–806.
22. Simon NR, Manshanden I, Lopes da Silva FH. A MEG study of 42. Kettenmann H, Ransom BR. Electrical coupling between astro-
sleep. Brain Res. 2000;860:64–76. cytes and between oligodendrocytes studied in mammalian cell
23. Mölle M, Marshall L, Gais S, et al. Grouping of spindle activity cultures. Glia. 1988;1:64–73.
during slow oscillations in human non-rapid eye movement sleep. 43. Walz W. Role of glial cells in the regulation of the brain ion
J Neurosci. 2002;22:10941–10947. microenvironment. Prog Neurobiol. 1989;33:309–333.
24. Massimini M, Rosanova M, Mariotti M. EEG slow (approximately 44. Nicholson C. Extracellular space as the pathway for neuron-glial
1 Hz) waves are associated with nonstationarity of thalamo- cell interaction. In: Kettenmann H, Ransom BR, eds. Neuroglia.
cortical sensory processing in the sleeping human. J Neurophysiol. New York: Oxford University Press; 1995;387–397.
2003;89:1205–1213. 45. Ballanyi K, Grafe P, ten Bruggencate G. Ion activities and potas-
25. Steriade M, Nuñez A, Amzica F. Intracellular analysis of relations sium uptake mechanisms of glial cells in guinea-pig olfactory cor-
between the slow (< 1 Hz) neocortical oscillation and other tex slices. J Physiol. 1987;382:159–174.
sleep rhythms of the electroencephalogram. J Neurosci. 46. Kettenmann H. K+ and Cl- uptake by cultured oligodendrocytes.
1993;13:3266–3283. Can J Physiol Pharmacol. 1987;65:1033–1037.
26. Timofeev I, Steriade M. Low-frequency rhythms in the thalamus 47. Kocsis JD, Malenka RC, Waxman SG. Effects of extracellular potas-
of intact-cortex and decorticated cats. J Neurophysiol. 1996;76: sium concentration on the excitability of the parallel fibres of the
4152–4168. rat cerebellum. J Physiol. 1983;334:225–244.
27. Sanchez-Vives MV, McCormick DA. Cellular and network mecha- 48. Rausche G, Igelmund P, Heinemann U. Effects of changes in extra-
nisms of rhythmic recurrent activity in neocortex. Nat Neurosci. cellular potassium, magnesium and calcium concentration on
2000;3:1027–1034. synaptic transmission in area CA1 and the dentate gyrus of rat
28. Contreras D, Timofeev I, Steriade M. Mechanisms of long-lasting hippocampal slices. Pflugers Arch. 1990;415:588–593.
hyperpolarizations underlying slow sleep oscillations in cat corti- 49. Hille B. Ion Channels of Excitable Membranes. Sunderland:
cothalamic networks. J Physiol. 1996;494(Pt 1):251–264. Sinauer; 2001.
29. Massimini M, Amzica F. Extracellular calcium fluctuations and 50. Loomis AL, Harvey EN, Hobart G. Distribution of disturbance
intracellular potentials in the cortex during the slow sleep oscilla- patterns in the human electroencephalogram, with special refer-
tion. J Neurophysiol. 2001;85:1346–1350. ence to sleep. J Neurophysiol. 1938;1:413–440.
30. Steriade M, Amzica F, Nuñez A. Cholinergic and noradrenergic 51. Colrain IM. The K-complex: a 7-decade history. Sleep.
modulation of the slow (approximately 0.3 Hz) oscillation in neo- 2005;28:255–273.
cortical cells. J Neurophysiol. 1993;70:1385-1400. 52. Amzica F, Steriade M. Cellular substrates and laminar profile of
31. Contreras D, Steriade M. Cellular basis of EEG slow rhythms: a sleep K-complex. Neuroscience. 1998;82:671–686.
study of dynamic corticothalamic relationships. J Neurosci. 53. Steriade M, Oakson G, Ropert N. Firing rates and patterns of mid-
1995;15:604–622. brain reticular neurons during steady and transitional states of the
32. Amzica F, Steriade M. Disconnection of intracortical synaptic sleep-waking cycle. Exp Brain Res. 1982;46:37–51.
linkages disrupts synchronization of a slow oscillation. J Neurosci. 54. Steriade M, Datta S, Paré D, et al. Neuronal activities in brain-stem
1995;15:4658–4677. cholinergic nuclei related to tonic activation processes in thalam-
33. Amzica F, Steriade M. Short- and long-range neuronal synchro- ocortical systems. J Neurosci. 1990;10:2541–2559.
nization of the slow (< 1 Hz) cortical oscillation. J Neurophysiol. 55. Hirsch JC, Fourment A, Marc ME. Sleep-related variations of
1995;73:20–38. membrane potential in the lateral geniculate body relay neurons of
34. Steriade M, Contreras D, Curró Dossi R, et al. The slow (< 1 Hz) the cat. Brain Res. 1983;259:308–312.
oscillation in reticular thalamic and thalamocortical neurons: sce- 56. Davis H, Davis PA, Loomis AL, et al. Electrical reactions of the
nario of sleep rhythm generation in interacting thalamic and neo- human brain to auditory stimulation during sleep. J Neurophysiol.
cortical networks. J Neurosci. 1993;13:3284–3299. 1939;6:500–514.
35. Amzica F, Steriade M. Electrophysiological correlates of sleep delta 57. Steriade M, Amzica F. Coalescence of sleep rhythms and their
waves. Electroencephalogr Clin Neurophysiol. 1998;107:69–83. chronology in corticothalamic networks. Sleep Res Online.
36. Amzica F, Neckelmann D. Membrane capacitance of cortical neu- 1998;1:1–10.
rons and glia during sleep oscillations and spike-wave seizures. J 58. Ingvar DH, Sjölund B, Ardo A. Correlation between dominant
Neurophysiol. 1999;82:2731–2746. EEG frequency, cerebral oxygen uptake and blood flow.
37. Amzica F, Massimini M. Glial and neuronal interactions during Electroencephalogr Clin Neurophysiol. 1976;41:268–276.
slow wave and paroxysmal activities in the neocortex. Cereb 59. Saunders MG, Westmoreland BF. The EEG in evaluation of disor-
Cortex. 2002;12:1101–1113. ders affecting the brain diffusely. In: Current Practice of Clinical
38. Amzica F, Massimini M, Manfridi A. Spatial buffering during slow Electroencephalography, Klass DW, Daly DD, eds. New York: Raven
and paroxysmal sleep oscillations in cortical networks of glial cells Press; 1979;343–379.
in vivo. J Neurosci. 2002;22:1042–1053. 60. Buzsáki G. Rhythms of the Brain. Oxford University Press; 2006.
60 Part I ■ Basic Principles

61. Brazier MAB. Studies of the EEG activity of limbic structures in 83. Alonso A, Garcia-Austt E. Neuronal sources of theta rhythm in
man. Electroencephalogr. Clin Neurophysiol. 1968;25:309–318. the entorhinal cortex of the rat. I. Laminar distribution of theta
62. Halgren E, Babb TL, Crandall PH. Human hippocampal forma- field potentials. Exp Brain Res. 1987;67:493–501.
tion EEG desynchronizes during attentiveness and movement. 84. Alonso A, Garcia-Austt E. Neuronal sources of theta rhythm in
Electroencephalogr Clin Neurophysiol. 1978;44:778–781. the entorhinal cortex of the rat. II. Phase relations between unit
63. Arnolds DE, Lopes da Silva FH, Aitink JW, et al. The spectral prop- discharges and theta field potentials. Exp Brain Res.
erties of hippocampal EEG related to behaviour in man. 1987;67:502–509.
Electroencephalogr. Clin Neurophysiol. 1980;50:324–328. 85. Boeijinga PH, Lopes da Silva FH. Differential distribution of beta
64. Kahana MJ, Sekuler R, Caplan JB, et al. Human theta oscillations and theta EEG activity in the entorhinal cortex of the cat. Brain
exhibit task dependence during virtual maze navigation. Nature. Res. 1988;448:272–286.
1999;399:781–784. 86. Vanderwolf CH, Buzsáki G, Cain DP, et al. Neocortical and hip-
65. Ekstrom AD, Caplan JB, Ho E, et al. Human hippocampal theta pocampal electrical activity following decapitation in the rat.
activity during virtual navigation. Hippocampus. 2005;15: Brain Res. 1988;451:340–344.
881–889. 87. Vanderwolf CH, Leung LS. Hippocampal rhythmical slow activ-
66. Bódizs R, Kántor S, Szabó G, et al. Rhythmic hippocampal slow ity: a brief history and effects entorhinal lesions and phencyci-
oscillation characterizes REM sleep in humans. Hippocampus. dine. In: Seifert W, ed. The Neurobiology of the Hippocampus.
2001;11:747–753. London: Academic Press; 1983:275–302.
67. Tesche CD, Karhu J. Theta oscillations index human hippocampal 88. Montgomery SM, Betancur MI, Buzsáki G. Behavior-dependent
activation during a working memory task. Proc Natl Acad Sci USA. coordination of multiple theta dipoles in the hippocampus. J
2000;97:919–924. Neurosci. 2009;29:1381–1394.
68. Cornwell BR, Johnson LL, Holroyd T, et al. Human hippocampal 89. Stewart M, Fox SE. Do septal neurons pace the hippocampal
and parahippocampal theta during goal-directed spatial naviga- theta rhythm? Trends Neurosci. 1990;13:163–168.
tion predicts performance on a virtual Morris water maze. J 90. Leung LW. Model of gradual phase shift of theta rhythm in the
Neurosci. 2008;28:5983–5990. rat. J Neurophysiol. 1984;52:1051–1065.
69. Stewart M, Fox SE. Hippocampal theta activity in monkeys. Brain 91. Chapman CA, Lacaille JC. Cholinergic induction of theta-fre-
Res. 1991;538:59–63. quency oscillations in hippocampal inhibitory interneurons
70. Green JD, Arduini AA. Hippocampal electrical activity in arousal. and pacing of pyramidal cell firing. J Neurosci. 1999;19:
J Neurophysiol. 1954;17:533–557. 8637–8645.
71. Buzsáki G. The hippocampo-neocortical dialogue. Cereb Cortex. 92. Leung LS. Generation of theta and gamma rhythms in the hip-
1996;6:81–92. pocampus. Neurosci. Biobehav Rev. 1998;22:275–290.
72. Witter MP, Naber PA, van HT, et al. Cortico-hippocampal com- 93. Buzsáki G, Czopf J, Kondákor I, et al. Laminar distribution of
munication by way of parallel parahippocampal-subicular path- hippocampal rhythmic slow activity (RSA) in the behaving rat:
ways. Hippocampus. 2000;10:398–410. current-source density analysis, effects of urethane and atropine.
73. Buzsáki G. Theta oscillations in the hippocampus. Neuron. Brain Res. 1986;365:125–137.
2002;33:325–340. 94. Soltesz I, Deschênes M. Low- and high-frequency membrane
74. Leung LW, Borst JG. Electrical activity of the cingulate cortex. I. potential oscillations during theta activity in CA1 and CA3
Generating mechanisms and relations to behavior. Brain Res. pyramidal neurons of the rat hippocampus under ketamine-
1987;407:68–80. xylazine anesthesia. J Neurophysiol. 1993;70:97–116.
75. Chapman CA, Lacaille JC. Intrinsic theta-frequency membrane 95. Somogyi P, Klausberger T. Defined types of cortical interneurone
potential oscillations in hippocampal CA1 interneurons of stra- structure space and spike timing in the hippocampus. J Physiol.
tum lacunosum-moleculare. J Neurophysiol. 1999;81:1296–1307. 2005;562:9–26.
76. Bilkey DK, Heinemann U. Intrinsic theta-frequency membrane 96. Banks MI, White JA, Pearce RA. Interactions between
potential oscillations in layer III/V perirhinal cortex neurons of distinct GABA(A) circuits in hippocampus. Neuron. 2000;25:
the rat. Hippocampus. 1999;9:510–518. 449–457.
77. Petsche H, Stumpf C, Gogolak G. The significance of the rabbit’s 97. Wendling F, Bartolomei F, Bellanger JJ, et al. Epileptic fast activ-
septum as a relay station between the midbrain and the hippocam- ity can be explained by a model of impaired GABAergic dendritic
pus. The control of hippocampus arousal activity by septum cells. inhibition. Eur J Neurosci. 2002;15:1499–1508.
Electroencephalogr. Clin Neurophysiol. 1962;14:202–211. 98. Fuentealba P, Tomioka R, Dalezios Y, et al. Rhythmically active
78. Petsche H, Gogolak G, van Zwieten PA. Rhythmicity of septal cell dis- enkephalin-expressing GABAergic cells in the CA1 area of the
charges at various levels of reticular excitation. Electroencephalogr. hippocampus project to the subiculum and preferentially inner-
Clin Neurophysiol. 1965;19:25–33. vate interneurons. J Neurosci. 2008;28:10017–10022.
79. Freund TF, Antal M. GABA-containing neurons in the septum 99. Kamondi A, Acsady L, Wang XJ, et al. Theta oscillations in
control inhibitory interneurons in the hippocampus. Nature. somata and dendrites of hippocampal pyramidal cells in vivo:
1988;336:170–173. activity-dependent phase-precession of action potentials.
80. Vertes RP, Kocsis B. Brainstem-diencephalo-septohippocampal Hippocampus. 1998;8:244–261.
systems controlling the theta rhythm of the hippocampus. 100. Wulff P, Ponomarenko AA, Bartos M, et al. Hippocampal theta
Neuroscience. 1997;81:893–926. rhythm and its coupling with gamma oscillations require fast
81. Buzsáki G, Leung LW, Vanderwolf CH. Cellular bases of hip- inhibition onto parvalbumin-positive interneurons. Proc Natl
pocampal EEG in the behaving rat. Brain Res. 1983;287: Acad Sci USA. 2009;106:3561–3566.
139–171. 101. Klausberger T, Somogyi P. Neuronal diversity and temporal
82. Mitchell SJ, Ranck JB Jr. Generation of theta rhythm in medial dynamics: the unity of hippocampal circuit operations. Science.
entorhinal cortex of freely moving rats. Brain Res. 1980;189:49–66. 2008;321:53–57.
Chapter 3 ■ Cellular Substrates of Brain Rhythms 61

102. Sik A, Penttonen M, Ylinen A, et al. Hippocampal CA1 interneu- 122. McCormick DA, Prince DA. Actions of acetylcholine in the
rons: an in vivo intracellular labeling study. J Neurosci. guinea-pig and cat medial and lateral geniculate nuclei, in vitro.
1995;15:6651–6665. J Physiol. 1987;392:147–165.
103. Toth K, Freund TF, Miles R. Disinhibition of rat hippocampal 123. Hu B, Steriade M, Deschênes M. The effects of brainstem peri-
pyramidal cells by GABAergic afferents from the septum. J brachial stimulation on neurons of the lateral geniculate nucleus.
Physiol. 1997500(Pt 2):463–474. Neuroscience. 1989;31:13–24.
104. Rotstein HG, Pervouchine DD, Acker CD, et al. Slow and fast 124. Hu B, Steriade M, Deschênes M. The effects of brainstem peri-
inhibition and an H-current interact to create a theta rhythm in brachial stimulation on perigeniculate neurons: the blockage of
a model of CA1 interneuron network. J Neurophysiol. spindle waves. Neuroscience. 1989;31:1–12.
2005;94:1509–1518. 125. Morison RS, Bassett DL. Electrical activity of the thalamus and
105. Berger H. Über das Elektroenkephalogramm des Menschen. 1st basal ganglia in decorticate cats. J Neurophysiol. 1945;8:09–314.
report. Arch Psychiat Nervenkr. 1929;87:527–570. 126. Steriade M, Llinás RR. The functional states of the thalamus and
106. Niedermeyer E. The normal EEG of the waking adult. In: the associated neuronal interplay. Physiol Rev. 1988;68:649–742.
Niedermeyer E,Lopes da Silva FH, eds. Electroencephalography: 127. Steriade M. Cellular substrates of brain rhythms. In:
Basic Principles, Clinical Applications, and Related Fields. Niedermeyer E, Lopes da Silva FH, eds. Electroencephalography:
Baltimore: Lippincott, Williams & Wilkins; 2005:167–192. Basic Principles, Clinical Applications and Related Fields.
107. Adrian ED, Mathews BHC. The interpretation of potential waves Baltimore: Lippincott Williams & Wilkins; 2004:31–84.
in the cortex. J Physiol. 1934;81:440–471. 128. Jones EG. The Thalamus. New York: Plenum; 1985.
108. Creutzfeld O, Grünvald G, Simonova O, et al. Changes of the 129. Domich L, Oakson G, Deschênes M, et al. Thalamic and cortical
basic rhythms of the EEG during the performance of mental and spindles during early ontogenesis in kittens. Brain Res.
visuomotor tasks. In: Evans CR, Mulholland TB, eds. Attention in 1987;428:140–142.
Neurophysiology. London: Butterworth; 1969;148–168. 130. Steriade M, Deschênes M, Domich L, et al. Abolition of spindle
109. Ray WJ, Cole HW. EEG alpha activity reflects attentional oscillations in thalamic neurons disconnected from nucleus
demands, and beta activity reflects emotional and cognitive reticularis thalami. J Neurophysiol. 1985;54:1473–1497.
processes. Science. 1985;228:750–752. 131. Steriade M, Domich L, Oakson G, et al. The deafferented reticu-
110. Rihs TA, Michel CM, Thut G. Mechanisms of selective inhibition lar thalamic nucleus generates spindle rhythmicity. J
in visual spatial attention are indexed by alpha-band EEG syn- Neurophysiol. 1987;57:260–273.
chronization. Eur J Neurosci. 2007;25:603–610. 132. Steriade M, Parent A, Hada J. Thalamic projections of nucleus
111. Rihs TA, Michel CM, Thut G. A bias for posterior alpha-band reticularis thalami of cat: a study using retrograde transport of
power suppression versus enhancement during shifting versus horseradish peroxidase and fluorescent tracers. J Comp Neurol.
maintenance of spatial attention. Neuroimage. 2009;44:190–199. 1984;229:531–547.
112. Andersen P, Andersson SA. Physiological Basis of the Alpha 133. Velayos JL, Jimenez-Castellanos J Jr, Reinoso-Suarez F.
Rhythm. New York: Appleton-Century-Crofts; 1968. Topographical organization of the projections from the reticular
113. Lopes da Silva FH, Van Lierop THMT, Schrijer CFM, et al. thalamic nucleus to the intralaminar and medial thalamic nuclei
Organization of thalamic and cortical alpha rhythm: spectra and in the cat. J Comp Neurol. 1989;279:457–469.
coherences. Electroencephalogr Clin Neurophysiol. 134. Paré D, Steriade M, Deschênes M, et al. Physiological character-
1973;35:627–639. istics of anterior thalamic nuclei, a group devoid of inputs from
114. Lopes da Silva FH, Storm Van Leeuwen W. The cortical alpha reticular thalamic nucleus. J Neurophysiol. 1987;57:1669–1685.
rhythm in dog: depth and surface profile of phase. In: Brazier 135. Wang XJ, Rinzel J. Spindle rhythmicity in the reticularis thalami
MAB, Petsche H, eds. Architecture of the Cerebral Cortex, IBRO nucleus: synchronization among mutually inhibitory neurons.
monograph series, Vol. 3. New York: Raven Press; 1978;319–333. Neuroscience. 1993;53:899–904.
115. Lopes da Silva FH, Vos JE. Mooibroek J, et al. Relative contribu- 136. Destexhe A, Contreras D, Sejnowski TJ, et al. A model of spindle
tions of intracortical and thalamo-cortical processes in the gen- rhythmicity in the isolated thalamic reticular nucleus. J
eration of alpha rhythms, revealed by partial coherence analysis. Neurophysiol. 1994;72:803–818.
Electroencephalogr Clin Neurophysiol. 1980;50:449–456. 137. Golomb D, Wang XJ, Rinzel J. Synchronization properties of
116. Bollimunta A, Chen Y, Schroeder CE, et al. Neuronal mechanisms spindle oscillations in a thalamic reticular nucleus model. J
of cortical alpha oscillations in awake-behaving macaques. J Neurophysiol. 1994;72:1109–1126.
Neurosci. 2008;28:9976–9988. 138. Deschênes M, Paradis M, Roy JP, et al. Electrophysiology of neu-
117. Flint AC, Connors BW. Two types of network oscillations in neo- rons of lateral thalamic nuclei in cat: resting properties and burst
cortex mediated by distinct glutamate receptor subtypes and discharges. J Neurophysiol. 1984;51:1196–1219.
neuronal populations. J Neurophysiol. 199675:951–957. 139. Yen CT, Conley M, Hendry SH, et al. The morphology of physi-
118. Hughes SW, Lörincz M, Cope DW, et al. Synchronized oscilla- ologically identified GABAergic neurons in the somatic sensory
tions at alpha and theta frequencies in the lateral geniculate part of the thalamic reticular nucleus in the cat. J Neurosci.
nucleus. Neuron. 2004;42:253–268. 1985;5:2254–2268.
119. Hughes SW, Crunelli V. Thalamic mechanisms of EEG alpha 140. Williamson AM, Ohara PT, Ralston DD, et al. Analysis of gamma-
rhythms and their pathological implications. Neuroscientist. aminobutyric acidergic synaptic contacts in the thalamic reticular
2005;11:357–372. nucleus of the monkey. J Comp Neurol. 1994;349:182–192.
120. Lörincz ML, Crunelli V, Hughes SW. Cellular dynamics of 141. Landisman CE, Long MA, Beierlein M, et al. Electrical synapses
cholinergically induced alpha (8-13 Hz) rhythms in sensory thal- in the thalamic reticular nucleus. J Neurosci. 2002;22:1002–1009.
amic nuclei in vitro. J Neurosci. 2008;28:660–671. 142. Yen CT, Jones EG. Intracellular staining of physiologically identi-
121. Steriade M. Neuronal Substrates of Sleep and Epilepsy. Cambridge: fied neurons and axons in the somatosensory thalamus of the cat.
University Press; 2003. Brain Res. 1983;280:148–154.
62 Part I ■ Basic Principles

143. Steriade M, Deschênes M. The thalamus as a neuronal oscillator. 164. Murthy VN, Fetz EE. Coherent 25- to 35-Hz oscillations in the
Brain Res. 1984;320:1–63. sensorimotor cortex of awake behaving monkeys. Proc Natl Acad
144. von Krosigk M, Bal T, McCormick DA. Cellular mechanisms of a Sci USA. 1992;89:5670–5674.
synchronized oscillation in the thalamus. Science. 1993;261: 165. Bouyer JJ, Montaron MF, Vahnee JM, et al. Anatomical localization
361–364. of cortical beta rhythms in cat. Neuroscience. 1987;22:863–869.
145. Bal T, von KM, McCormick DA. Role of the ferret perigeniculate 166. Freeman WJ. Mass Action in the Nervous System. New York:
nucleus in the generation of synchronized oscillations in vitro. Academic Press; 1975.
J Physiol. 1995;483(Pt 3):665–685. 167. Eckhorn R, Bauer R, Jordan W, et al. Coherent oscillations: a mech-
146. Bal T, von KM, McCormick DA. Synaptic and membrane mech- anism of feature linking in the visual cortex? Multiple electrode
anisms underlying synchronized oscillations in the ferret lateral and correlation analyses in the cat. Biol Cybern. 1988;60:121–130.
geniculate nucleus in vitro. J Physiol. 1995;483(Pt 3):641–663. 168. Gray CM, Konig P, Engel AK, et al. Oscillatory responses in cat
147. Contreras D, Destexhe A, Sejnowski TJ, et al. Control of spa- visual cortex exhibit inter-columnar synchronization which
tiotemporal coherence of a thalamic oscillation by corticothala- reflects global stimulus properties. Nature. 1989;338:334–337.
mic feedback. Science. 1996;274:771–774. 169. Gray CM, Singer W. Stimulus-specific neuronal oscillations in
148. Contreras D, Destexhe A, Sejnowski TJ, et al. Spatiotemporal orientation columns of cat visual cortex. Proc Natl Acad Sci USA.
patterns of spindle oscillations in cortex and thalamus. J Neurosci. 1989;86:1698–1702.
1997;17:1179–1196. 170. Engel AK, Konig P, Gray CM, et al. Stimulus-dependent neuronal
149. Jahnsen H, Llinás R. Electrophysiological properties of guinea-pig oscillations in cat visual cortex: inter-columnar interaction as
thalamic neurones: an in vitro study. J Physiol. 1984;349:205–226. determined by cross-correlation analysis. Eur J Neurosci.
150. Mulle C, Steriade M, Deschênes M. The effects of QX314 on thal- 1990;2:588–606.
amic neurons. Brain Res. 1985;333:350–354. 171. Gray CM, Engel AK, Konig P, et al. Stimulus-dependent neuronal
151. McCormick DA. Functional properties of a slowly inactivating oscillations in cat visual cortex: receptive field properties and fea-
potassium current in guinea pig dorsal lateral geniculate relay ture dependence. Eur J Neurosci. 1990;2:607–619.
neurons. J Neurophysiol. 1991;66:1176–1189. 172. Rodriguez R, Kallenbach U, Singer W, et al. Short- and long-term
152. Budde T, Mager R, Pape HC. Different types of potassium out- effects of cholinergic modulation on gamma oscillations and
ward current in relay neurons acutely isolated from the rat lateral response synchronization in the visual cortex. J Neurosci.
geniculate nucleus. Eur J Neurosci. 1992;4:708–722. 2004;24:10369–10378.
153. Contreras D, Steriade M. Spindle oscillation in cats: the role of 173. Munk MH, Roelfsema PR, Konig P, et al. Role of reticular activa-
corticothalamic feedback in a thalamically generated rhythm. tion in the modulation of intracortical synchronization. Science.
J Physiol. 1996;490(Pt 1):159–179. 1996;272:271–274.
154. Kim U, Bal T, McCormick DA. Spindle waves are propagating 174. Niessing J, Ebisch B, Schmidt KE, et al. Hemodynamic signals
synchronized oscillations in the ferret LGNd in vitro. correlate tightly with synchronized gamma oscillations. Science.
J Neurophysiol. 1995;74:1301–1323. 2005;309:948–951.
155. Sanford AJ. A periodic basis for perception and action. In: 175. von der Malsburg C, Schneider W. A neural cocktail-party
Colquhoun WP, ed. Biological Rhythms and Human Perception. processor. Biol Cybern. 1986;54:29–40.
New York: Academic Press; 1971;179–209. 176. Steriade M, Amzica F. Intracortical and corticothalamic
156. Moruzzi G, Magoun HW. Brain stem reticular formation and coherency of fast spontaneous oscillations. Proc Natl Acad Sci
activation of the EEG. Electroenceph Clin Neurophysiol. 1949;1: USA. 1996;93:2533–2538.
445–473. 177. Steriade, M., Contreras, D., Amzica, F., and Timofeev, I.
157. McCormick DA, Prince DA. Acetylcholine induces burst firing in Synchronization of fast (30–40 Hz) spontaneous oscillations in
thalamic reticular neurones by activating a potassium conduc- intrathalamic and thalamocortical networks. J. Neurosci.
tance. Nature. 1986;319:402–405. 1996;16:2788–2808.
158. McCormick DA, Wang Z. Serotonin and noradrenaline excite 178. Roelfsema PR, Engel AK, Konig P, et al. Visuomotor integration
GABAergic neurones of the guinea-pig and cat nucleus reticu- is associated with zero time-lag synchronization among cortical
laris thalami. J Physiol. 1991;442:235–255. areas. Nature. 1997;385:157–161.
159. McCormick DA, Pape HC. Noradrenergic and serotonergic mod- 179. Fries P, Nikolic D, Singer W. The gamma cycle. Trends Neurosci.
ulation of a hyperpolarization-activated cation current in thala- 2007;30:309–316.
mic relay neurones. J Physiol. 1990;431:319–342. 180. Llinás RR, Grace AA, Yarom Y. In vitro neurons in mammalian
160. Buzsáki G, Bickford RG, Ponomareff G, et al. Nucleus basalis and cortical layer 4 exhibit intrinsic oscillatory activity in the 10- to
thalamic control of neocortical activity in the freely moving rat. 50-Hz frequency range. Proc Natl Acad Sci USA. 1991;88:
J Neurosci. 1988;8:4007–4026. 897–901.
161. Bremer F, Stoupel N, Van Reeth PC. Nouvelles recherches sur la 181. Nuñez A, Amzica F, Steriade M. Voltage-dependent fast
facilitation et l’inhibition des potentiels évoqués corticaux dans (20–40 Hz) oscillations in long-axoned neocortical neurons.
l’éveil réticulaire. Arch Ital Biol. 1960;98:229–247. Neuroscience. 1992a;51:7–10.
162. Lopes da Silva FH, van Rotterdam A, Storm van Leeuwen W, et al. 182. Steriade M, Curró Dossi R, Paré D, et al. Fast oscillations (20–40
Dynamic characteristics of visual evoked potentials in the dog. II. Hz) in thalamocortical systems and their potentiation by meso-
Beta frequency selectivity in evoked potentials and background pontine cholinergic nuclei in the cat. Proc Natl Acad Sci USA.
activity. Electroencephalogr Clin Neurophysiol. 1970;29:260–268. 1991;88:4396–4400.
163. Freeman WJ, van Dijk BW. Spatial patterns of visual cortical fast 183. Cunningham ET, LeVay S. Laminar and synaptic organization of
EEG during conditioned reflex in a rhesus monkey. Brain Res. the projection from the thalamic nucleus centralis to primary
1987;422:267–276. visual cortex in the cat. J Comp Neurol. 1986;254:66–77.
Chapter 3 ■ Cellular Substrates of Brain Rhythms 63

184. Chagnac-Amitai Y, Connors BW. Synchronized excitation and 199. Somjen GG, Trachtenberg M. Neuroglial as generator of extracel-
inhibition driven by intrinsically bursting neurons in neocortex. lular current. In: Speckmann EJ, Caspers H, eds. Origin of
J Neurophysiol. 1989;62:1149–1162. Cerebral Field Potentials. Stuttgart: Thieme; 1979:21–32.
185. Hösli L, Hösli E, Della BG, et al. Action of acetylcholine, mus- 200. Caspers H, Speckmann E-J, Lehmenkuler A. DC potentials of the
carine, nicotine and antagonists on the membrane potential of cerebral cortex: Seizure activity and changes in gas pressures. Rev
astrocytes in cultured rat brainstem and spinal cord. Neurosci Physiol Biochem Pharmacol. 1987;106:127–178.
Lett. 1988;92:165–170. 201. Heinemann U, Walz W. Contributions of potassium currents and
186. Seigneur J, Kroeger D, Nita DA, et al. Cholinergic action on cor- glia to slow potential shifts (SPSs). In: Laming PR, Sykova E,
tical glial cells in vivo. Cereb Cortex. 2006;16:655–668. Reichenbach A, et al., eds. Glial Cells: Their Role in Behaviour.
187. Roopun AK, Middleton SJ, Cunningham MO, et al. A beta2- Cambridge: Cambridge University Press; 1998:197–209.
frequency (20–30 Hz) oscillation in nonsynaptic networks of 202. Amzica F, Steriade M. Neuronal and glial membrane potentials
somatosensory cortex. Proc Natl Acad Sci USA. 2006;103: during sleep and paroxysmal oscillations in the neocortex. J
15646–15650. Neurosci. 2000;20:6648–6665.
188. Mandema JW, Danhof M. Electroencephalogram effect measures 203. Tschirgi RD, Taylor JL. Slowly changing bioelectric potentials
and relationships between pharmacokinetics and pharmacody- associated with the blood-brain barrier. Am J Physiol.
namics of centrally acting drugs. Clin Pharmacokinet. 1992;23: 1958;195:7–22.
191–215. 204. Held D, Fencl V, Pappenheimer JR. Electric potential of cere-
189. Salenius S, Salmelin R, Neuper C, et al. Human cortical 40 Hz brospinal fluid. J Neurophysiol. 1964;27:942–959.
rhythm is closely related to EMG rhythmicity. Neurosci Lett. 205. Kjällquist Å, Siesjö BK. Regulation of CSF pH - influence of the
1996;213:75–78. CSF/plasma potential. Scand J Clin Lab Invest Suppl. 1968;102:I:C.
190. Salenius S, Portin K, Kajola M, et al. Cortical control of human 206. Sorensen SC, Severinghaus JW. Effect of cerebral acidosis on the
motoneuron firing during isometric contraction. J Neurophysiol. CSF-blood potential difference. Am J Physiol. 1970;219:68–71.
1997;77:3401–3405. 207. Hornbein TF, Sorensen SC. dc Potential difference between dif-
191. Gross J, Tass PA, Salenius S, et al. Cortico-muscular synchroniza- ferent cerebrospinal fluid sites and blood in dogs. Am J Physiol.
tion during isometric muscle contraction in humans as revealed 1972;223:415–418.
by magnetoencephalography. J Physiol. 2000;527(Pt 3):623–631. 208. Woody CD, Marshall WH, Besson JM, et al. Brain potential shift
192. Hari R. Action-perception connection and the cortical mu with respiratory acidosis in the cat and monkey. Am J Physiol.
rhythm. Prog Brain Res. 2006;159:253–260. 1970;218:275–283.
193. Pfurtscheller G, Neuper C, Andrew C, et al. Foot and hand area 209. Revest PA, Jones HC, Abbott NJ. The transendothelial DC poten-
mu rhythms. Int J Psychophysiol. 1997;26:121–135. tial of rat blood-brain barrier vessels in situ. Adv Exp Med Biol.
194. Chen R, Yaseen Z, Cohen LG, et al. Time course of corticospinal 1993;331:71–74.
excitability in reaction time and self-paced movements. Ann 210. Voipio J, Tallgren P, Heinonen E, et al. Millivolt-scale DC shifts in
Neurol. 1998;44:317–325. the human scalp EEG: evidence for a nonneuronal generator.
195. Pfurtscheller G, Neuper C, Kalcher J. 40-Hz oscillations during J Neurophysiol. 2003;89:2208–2214.
motor behavior in man. Neurosci Lett. 1993;164:179–182. 211. Nita DA, Vanhatalo S, Lafortune FD, et al. Nonneuronal origin of
196. Vanhatalo S, Tallgren P, Andersson S, et al. DC-EEG discloses CO 2-related DC EEG shifts: an in vivo study in the cat.
prominent, very slow activity patterns during sleep in preterm J Neurophysiol. 2004;92:1011–1022.
infants. Clin Neurophysiol. 2002;113:1822–1825. 212. Speckmann E-J, Elger CE. Introduction to the neurophysiologi-
197. Vanhatalo S, Holmes MD, Tallgren P, et al. Very slow EEG cal basis of the EEG and DC potentials. In: Niedermeyer E,Lopes
responses lateralize temporal lobe seizures: an evaluation of non- da Silva R, eds. Electroencephalography: Basic Principles, Clinical
invasive DC-EEG. Neurology. 2003;60:1098–1104. Applications, and Related Fields. Baltimore: Lippincott, Williams
198. Somjen GG. Electrogenesis of sustained potentials. Prog & Wilkins; 2005:17–30.
Neurobiol. 1973;1:201–237.
Dynamics of EEGs as Signals of
Neuronal Populations: Models and
Theoretical Considerations

MODELS OF EEG GENERATION— tions, whether by way of chemical synapses or electrical cou-
INTRODUCTION plings; and (d) the intrinsic ionic and synaptic currents.
There are both inhibitory and excitatory feedback loops. The
In Chapters 2 and 3 the neurophysiological basis of the elec- terms negative feedback and positive feedback are often used in a
troencephalogram (EEG) was discussed with special emphasis rather loose way to characterize the interactions between popula-
on the phenomena at the cellular level. In Chapter 5, the bio- tions of neurons. However, the sign of feedback is defined in an
physical aspects of the generation of EEG phenomena are con- exact way by comparing the overall gain modulus of the system
sidered mainly in terms of volume conduction, but the with feedback (closed loop gain: |Yo(j)|) and without feedback
dynamic properties of EEG patterns are only briefly men- (open loop gain: |Y1(j)|). The feedback is said to be negative if
tioned. These dynamic properties, however, are essential for |Yo(j)| |Y1(j)| and positive in the opposite case (2). The exact
understanding EEG phenomena. In this chapter, EEG signals sign of feedback in the interaction between neuron populations is
are considered as the result of the dynamic behavior of neu- in most cases not known. It therefore seems preferable to use the
ronal populations as revealed by model studies. According to terms excitatory and inhibitory feedback to define the main synap-
this perspective, it is necessary to integrate experimental and tic type of interaction between neuron populations. This will be
theoretical results. The latter must be obtained using models of considered from the reference point of the “main” population,
neuronal networks and corresponding computer simulations. which usually is the population that receives the input from
The mathematical treatment of such models has been avoided another source and/or sends its output to another structure.
here; the interested reader is referred to a number of other In this chapter, models of the generation of characteristic
publications that are cited for their thorough mathematical types of EEGs, including the spatial spread of rhythmic activity
treatment of these areas. over the cortex, and nonlinear dynamics as a new approach for
The fundamental assumption on which the following dis- understanding EEG phenomena are presented.
cussion is based is that EEG signals reflect the dynamics of elec- The processes of generation of EEG signals in a large number
trical activity in populations of neurons. A property of such of neurons forming a neural mass are complex. Therefore, it is
populations that is of utmost importance for the generation of useful to use both analytic and synthetic approaches to under-
EEG signals is the capacity of the neurons to work in synchrony. stand such processes. It is essential to use both approaches in a
This depends on the connectivity between the neurons that dialectic way. The experimental data, obtained through histolog-
form a network. The classic terminology of Freeman (1) pro- ical and physiological analytic methods, must be put together just
vides a useful systematization. Populations of neurons with as pieces of a puzzle are assembled. Very often the relationship
mutual interactions are called KI sets; the interactions may be between the different pieces is still unclear. It is in this context
excitatory (KIe) or inhibitory (KIi). Groups of two interacting that a synthetic approach may be most helpful. The point is then
populations of neurons are called KII sets; they are formed by to put together the available data in such a way as to form the
an interaction of a KIe set with a KIt set. Further, KIII sets most likely pattern. In this way, a model of a neurophysiological
formed by the interaction between two KII sets can be defined. system can be constructed. Such a model can help to advance
A number of such KIII sets constitute a neural mass that may knowledge of the system in two respects. First, it provides the
occupy a few square millimeters of cortical surface or a few possibility for testing the influence of different types of inputs or
cubic millimeters of nuclear volume in the brainstem or spinal of changing some of the properties of the constituting elements
cord. Typically, a neural mass consists of approximately 104 to upon the output of the system. Second, it allows the formulation
107 neurons. An example of a KIII set is given in a symbolic way of hypotheses concerning new elementary properties, relation-
in Figure 4.1. An essential feature of such a basic module is the ships, and overall behavior; it may thus predict new properties of
existence of multiple feedback loops. Such a set of neurons can the system, raise new questions, and suggest new experiments to
produce oscillatory phenomena as revealed in EEG signals. The explore these hypotheses. In this way, the model can constitute a
main parameters that characterize the dynamic behavior of the bridge between experimentation and theory. As Harmon (3)
set are (a) the synaptic time constants; (b) the length constants, stated long ago, it is in suggesting functional relationships
which define the distance of the interactions between different between the activity of single neurons and the behavior of groups
neurons; (c) the gain factors, that is, the strength of interac- of neurons that theoretical neurophysiology may be most potent.
66 Part I ■ Basic Principles

the generation of the main properties of EEG/MEG

(electro/magnetoencephalographic) signals, we pay special atten-
tion to macroscopic models. It should be emphasized that the
brain is a very complex system, and even one cortical column
consists of a large number of elements, neurons, ionic channels,
synapses, neurotransmitter systems, and glial cells that require
enormous amount of parameters and state variables for a proper
description. The dynamics of EEG/MEG signals, however, can be
accounted for within the framework of systems operating within
spaces with much lower dimensions. This is the realm of macro-
scopic models, namely neural mass models (NMMs) that use
mean field concepts, where the properties of subpopulations of
neurons are lumped together based on the fact that many neu-
rons within a given neuronal system share basic properties.
Nonetheless we also consider, albeit in abbreviated form, some
results of studies using microscopic models that are also relevant
for understanding some characteristics of EEG/MEG signals. In
this respect the models developed by Edelman and collaborators
(4) are good examples of models that combine detailed proper-
ties of neurons and synapses within large-scale systems. Although
these models have been introduced mainly to study functional
connectivity at different scales of complexity with respect to neu-
rocognitive processes, their capacity to generate brain rhythms
and EEG-like phenomena should also be stressed. It is to be
expected that the applications of this new generation of large-
scale computer models of complex neuronal networks will grow
fast in the coming years, since these kinds of models can now be
realized by virtue of the current availability of powerful com-
puter systems. An ambitious project along this research line is the
Blue Brain project that has the aim of creating a synthetic mam-
malian brain by reverse-engineering, using IBM’s Blue Gene
supercomputer, in order to investigate the brain’s architectural
and functional principles underlying cognitive processes (5).
Figure 4.1 Lumped circuit diagram of the sets of neurons in the olfac-
Another new trend is the development of Topological Models
tory bulb (OB), the anterior olfactory nucleus (AON), and the prepyri-
of neuronal networks with their patterns of connectivity that
form cortex (PC). PON, primary olfactory nerve; LOT, lateral olfactory
can be studied using the mathematical theoretical approach of
tract; EC, external capsule; R, receptors of the nasal mucosa; gl,
“small-world” graphs. These are mathematical graphs that con-
glomeruli; P, periglomerular neuron; M, mitral neuron; G, granule cell;
sist of nodes that can be reached from every other by a small
E, superficial pyramidal cells of AON; I, inhibitory neuron; A, superficial
number of steps, which have been applied in a wide variety of
pyramidal neuron; B, short local interneuron; C, deep pyramidal neuron;
fields, from gene to social networks. Watts and Strogatz (6)
, excitation; , inhibition; L, latency. (Adapted from Freeman WJ.
classified such graphs according to two independent structural
Analytic techniques used in the search for the physiological basis of the
features, namely the clustering coefficient and average node-to-
EEG. In: Gevins A, Remond A, eds. Handbook of Electroencephalography
node path length. This approach is in essence an analytical tool
and Clinical Neurophysiology, vol. 1: Methods of Analysis of Brain
that can extract interesting parameters to characterize func-
Electrical and Magnetic Signals. Amsterdam: Elsevier; 1986.)
tional connectivity within networks of cortical neurons and
conditions determining neuronal synchrony (7,177) and to ana-
With these general objectives in mind, this chapter considers a lyze fMRI and MEG/EEG data (8), but it has not addressed
few models of the generation of EEG phenomena that have been explicitly the question of the generation of EEG/MEG activities.
elaborated in the past decades. We focus here on models of EEG
rhythmic activities both at the macroscopic and the microscopic
level. The former are simplified models in which detailed mech- MACROSCOPIC MODELS
anisms are kept to a minimum and the main properties of sub- OF EEG GENERATION
populations of neurons are lumped together. These models can,
however, be easily adapted to include more specific neuronal Basic Principles: The Wilson–Cowan Approach
properties and more complex connectivity. Taking into consider- At the macroscopic level we describe NMMs that are based on
ation that in this chapter we are particularly interested in the mean-field approximation approach. Under this heading we
dynamic models of neuronal networks that are responsible for consider two classes of models: (i) models that account for
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 67

time-dependent properties of neural populations and (ii) models average membrane potential. The latter was interpreted as the
that include, in addition, space-dependent properties. In gen- local EEG signal, which in this case consisted of a predominant
eral terms this approach was introduced in a systematic way in alpha rhythm. Later the analysis was extended to take into account
the Neurosciences by Wilson and Cowan (9) who proposed a a first approximation of the nonlinear behavior (11). In the sec-
theoretical framework that emphasizes not the properties of tion of alpha rhythms we describe these models in more detail.
individual cells but rather those of neuronal populations. A More recently these kinds of models were extended to account
basic assumption is that cortical neurons form a dense network for EEG frequency components other than the alpha rhythms.
consisting of, at least, two subpopulations that are intercon- In this context the model study of Jansen and Rit (12) marks an
nected either directly or by way of interneurons. The basic vari- important advance. These authors introduced a lumped-parame-
able is the proportion of neurons of each subpopulation that ter model of a cortical column consisting of a population of
are active at a given time, such that the relevant variable is not pyramidal cells interconnected with populations of excitatory and
the single spike but the average spike frequency. The dynamics inhibitory interneurons, which displays five kinds of simulated
of the neural mass depends on the interaction of excitatory and EEG outputs when fed with random noise. These different EEG
inhibitory neuronal subpopulations. Accordingly the model outputs are the following: low amplitude noise for high values of
requires the use of two variables E(t) and I(t), describing the the inhibitory feedback, slow periodic activity similar to that
activity of the excitatory and inhibitory subpopulations, respec- obtained in comatose patients, typical alpha rhythmic activity,
tively, to characterize the state of the whole population. noisy alpha activity, and low-amplitude high-frequency activity
Based on these assumptions the general Wilson and Cowan similar to beta activity for high values of the excitatory feedback. In
equations (9) governing the dynamics of a localized population order to explore how different cortical columns interact, Jansen
of neurons were derived: and Rit investigated the behavior of two identical single-column
t models and showed that they may produce in-phase oscillations.
E(t r) c1 E(t¿)dt¿d #Se Another case where NMMs of interacting neuronal popula-
tions helped to clarify the origin of a specific EEG/MEG phenom-
t r
enon was studied by Suffczynski et al. (13). This phenomenon is
the so-called focal ERD–surround ERS, described in detail in
e a (t t¿) [c1E(t¿) c2I(t¿) P(t¿) ] dt¿f Chapter 45, where a given event (e.g., finger movement) can cause
q at the same time a decrease (event-related desynchronization or
and ERD) and an increase (event-related synchronization or ERS) of
t EEG/MEG power within the alpha frequency range (mu rhythm)
I(t¿)dt¿d #Si
depending on the site over the scalp. The analysis of coupled
I(t r¿) c1
NMMs representing thalamocortical systems responsible for the
t r¿ hand and for the foot movements, respectively, was able to reveal
how this reciprocal “ERD–ERS” phenomenon takes place.
e a (t t¿) [c3E(t¿) c4I(t¿) Q(t¿) ] dt¿f . Based on the results obtained with NMMs generating rhyth-
q mic activities within the alpha frequency range, several
researchers extended these models to account for different EEG
This formalism describes the activities of the excitatory E(t) frequency bands besides the alpha rhythm. In this respect,
and inhibitory I(t) subpopulations; c1 and c2, and c3 and c4, are David and Friston (14) described a lumped model that gener-
the connectivity constants representing the average number of ates various rhythmic activities ranging from delta to gamma,
excitatory and inhibitory synapses per cell for each subpopula- according to the kinetics of the populations included in the
tion, is the decay time constant, Se(x) and Si (x) are the model. These authors also investigated the dynamics of the
response functions that give the expected proportion of cells in interaction between different cortical areas by coupling two
a subpopulation that would respond to a given level of excita- lumped models. Depending on coupling strength, direction of
tion and are represented by sigmoid functions; P(t) and Q(t) coupling (unidirectional or reciprocal coupling), and propaga-
are the external inputs to the excitatory and the inhibitory sub- tion delays, they showed that the oscillation frequency depends
populations, respectively; r is the refractory period. on the delay and that a reciprocal coupling is characterized by
oscillatory signals in phase or in antiphase. This latter result
Neural Mass Models of EEG Rhythms
reproduces the experimental finding of zero-lag synchroniza-
Based on the theoretical formalism introduced by Wilson and tion among remote cortical areas as found experimentally by
Cowan (9), Lopes da Silva et al. (10) described an NMM with the Roelfsema et al. (15) and Chawla et al. (16).
aim of simulating the conditions under which alpha rhythms With the aim of investigating how different cortical regions
occur. The model consisted of lumped subpopulations of inter- can display complex EEG frequency spectra and several pat-
connected excitatory and inhibitory neurons with properties of terns of functional connectivity, Zavaglia et al. (17,18) also
thalamocortical cells and a special subpopulation of interneurons. developed interconnected lumped models. They showed that
The mathematical analysis was initially carried out for a linearized three populations, consisting of four subpopulations (pyrami-
approximation of the neuronal network in order to express the dal neurons, excitatory, slow, and fast inhibitory interneurons),
transfer function of the network and the power spectrum of the arranged in parallel are sufficient to account for complex EEG
68 Part I ■ Basic Principles

spectra with dominant frequency ranging form theta to alpha Neural Mass Models w ith Spatial-Temporal
and gamma. This model is akin to the comprehensive NMM Features—Continuous Neural Field Models
introduced by Wendling et al. (19) to account for the generation In addition to the lumped NMMs described above, where spatial
of different kinds of neural activities and the corresponding aspects are not explicitly simulated, there are also models that
EEG signals, including epileptic activity as described in more include spatial properties and consist of spatiotemporal chains of
detail in the section dedicated to models of epileptic phenom- different subpopulations of neurons. An example is the model of
ena generated by neural populations. This model illustrates very van Rotterdam et al. (20) that consists of a series of pyramidal
clearly the main features of the currently used NMMs (Fig. 4.2). cells and interneurons interconnected by means of recurrent col-

Figure 4.2 (a): Neural mass model consisting of a subpopulation of principal cells (pyramidal cells) that project to and receive feedback from sub-
populations of local interneurons. Feedback to pyramidal cells is either excitatory (recurrent excitation) or inhibitory: dendritic-projecting interneu-
rons with slow synaptic kinetics—IPSP (slow)—shown in (c)—and somatic-projecting interneurons (gray hexagon) with faster synaptic
kinetics—IPSP (fast)—shown in (c). (b): Correspon ding block diagram representing for each subpopulation, the average pulse density of afferent
action potentials that is transformed into an average inhibitory or excitatory postsynaptic membrane potential using a linear dynamic transfer func-
tion with impulse response h EXC(t), h SDI(t), and h FSI(t) while this potential is converted into an average pulse density of action potentials fired by
the neurons by way of a static nonlinear function (asymmetric sigmoid curve, S(v)). Interactions between main cells and interneurons are repre-
sented by seven connectivity constants (C1 to C7), which account for the average number of synaptic contacts. One of the model outputs represents
the summated average postsynaptic potentials on pyramidal cells. It simulates an EEG signal. The three main parameters of the model respectively
correspond to the average excitatory synaptic gain (EXC), to the average slow inhibitory synaptic gain (SDI), and to the average fast inhibitory synap-
tic gain (FSI). (Adapted from Wendling F, Hernandez A, Bellanger JJ, et al. Interictal to ictal transition in human temporal lobe epilepsy: insights
from a computational model of intracerebral EEG. J Clin Neurophysiol. 2005;22(5):343–356.)
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 69

laterals and inhibitory fibers with connectivity functions Neural Mass Models and Hemodynamic Signals
assumed to be homogenous while the strength of the connectiv- NMMs are also being applied to make a bridge between the
ity was an exponentially decreasing function of distance. This activity of neuronal networks as reflected in EEG signals and
model was constructed to study the factors that determine the hemodynamic signals as measured by way of functional MRI, in
propagation of alpha waves along a cortical sheet. Assuming a order to understand better the working of cortical networks
continuous chain of neurons the model’s output, u(x,t), is the underlying cognitive processes. This constitutes a central feature
space-averaged membrane potential density of the main cells. To of dynamic causal modeling (DCM) (30) that has been applied
study propagation properties in the neuronal chain an expression not only to fMRI signals but also to electrophysiological signals
has been derived as a linear inhomogenous partial differential such as event-related potentials (ERPs) (31).
equation by means of which one can determine group velocity of Valdés-Sosa et al. (32) proposed an approach where the
alpha waves and dispersion properties. Alternatively the neuronal summed postsynaptic potentials generated by an NMM of a cor-
chain may be conceived as a two-dimensional system with a spe- tical area are used in the forward mode to estimate the primary
cific transfer function relating a random input p(x,t) to the out- current density distribution of a cortical area (Fig. 4.3). This is
put u(x,t). The model results were able to emulate experimental then transformed into the signal recorded at EEG sensors by
data, namely the propagation velocity that was found experimen- applying a linear spatial convolution with the EEG lead field, that
tally in the canine cortex to be about 0.3 m/sec (21) based on is, the function that relates any source in the brain to the corre-
parameters taken from physiology and histology, as explained sponding measurements at the EEG sensors. Simultaneously the
further in more detail in this chapter. summed postsynaptic potentials generated by the NMM are trans-
A series of other models, akin to that of van Rotterdam et al. formed into a local Vasomotor feed forward signal that is then fur-
(20), where the cortical neuronal populations are modeled as ther transformed by way of a temporal convolution with the
forming a continuum have been made, including models incor- hemodynamic response function (HRF) into a simulated BOLD
porating stochastic features called continuous neural field models. signal. In this way the information obtained from the EEG signal
In this chapter we focus on those aspects that are specifically rele- and from the corresponding fMRI BOLD signal recorded from the
vant for EEG/MEG generation, while the more general aspects of same region of interest can be fused (Fig. 4.3). We should note that
cortical processing of information with respect to neurocognitive although this approach opens interesting novel perspectives and
functions lie outside the scope of this text. With respect to these
continuous neural field models, in general, the reader may be
referred to comprehensive reviews of Harrison et al. (22), Deco et
al. (23), and Jirsa (24). Examples of continuous neural field models Forward Models for fusion of EEG/MEG & BOLD
signals based on a common Neural Mass Model
are those of Jirsa (24,25) and of the Australian group (26–29). Jirsa
(24) gives a very thorough account of how a set of state equations Neural Mass Model
with parameter values based on physiology can account for the generates an ensemble
of postsynaptic potentials
dynamics of cortical neuronal populations at different spatial
scales: microscopic, mesoscopic, and macroscopic. At microscopic
level, fast synaptic feedbacks appear to be crucial to the stability of
the excited cortex. At mesoscopic scale the model shows how fields Transfer function Transfer function
of synchrony can be established, and at macroscopic scale it shows Estimated cortical Vasomotor feedforward
how traveling waves may be generated in the cortex. The last two primary current density signal
levels are particularly relevant with respect to the dynamics of
EEG/MEG generation. With respect to these models the activity
state of a single neuron can be described by its membrane poten-
tial, V, the capacitive current, I, or the time since the last action
potential, T. Each of these variables can be represented by a EEG/MEG BOLD signal
dimension in the phase space of the neuron’s activity state.
Accordingly this phase space would be three-dimensional and the
state of a neuron would correspond to a point v = {V,I,T} or par- Figure 4.3 The output of a neural mass model in the form of summed
ticle in phase space. Using a general formalism the activity state of postsynaptic potentials can be used to estimate the primary current
a population of neurons at any time, t, can be described in phase density distribution of a cortical area that is transformed into the
space by a density p(v,t). As the state of each neuron evolves the EEG/ MEG signal. Simultaneously the summed postsynaptic potentials
ensemble density, p(v,t) evolves until it reaches some steady state. generated are transformed into a local Vasomotor feed-forward signal
The evolution of the ensemble density dynamics can be described that is then further transformed by way of a temporal convolution with
using the formalism of the Fokker–Planck equation. The latter the hemodynamic response function into a simulated BOLD signal. In
describes in general terms the time evolution of the probability this way EEG/ MEG information can be fused with the corresponding
density function of the position of a particle in a given multidi- fMRI BOLD signal recorded from the same region of interest. (Inspired
mensional space. The interested reader may find a thorough treat- from Valdes-Sosa PA, Sanchez-Bornot JM, Sotero RC, et al. Model
ment of the applications of this formalism in the study of driven EEG/ fMRI fusion of brain oscillations. Hum Brain Mapp.
continuous field models by Deco et al. (23) and Harrison et al. (22). 2009;30(9):2701–2721.)
70 Part I ■ Basic Principles

merits being further elaborated, it is based on concepts that are still the prepyriform cortex (PC). Excitation of the M cells by pri-
insufficiently established. Namely more research is needed to bet- mary olfactory nerve (PON) inputs leads to excitation of the
ter understand how neural activities, generated by neural popula- granule (G) neurons and to feedback inhibition of the M neu-
tions with complex geometrical configurations, are transformed rons. In consequence, the G neurons are disexcited and the M
into cortical current distributions on the one hand, and how they neurons are disinhibited. In this way, the M neurons may
are associated with BOLD signals on the other. become excited again, and the whole process may continue in
an oscillatory mode. The basic oscillation is at 40 Hz in cat and
60 to 80 Hz in rabbit (61). A similar type of inhibitory feedback
MODELS OF SPECIFIC EEG ACTIVITIES loop occurs in the PC, and the EEG of this cortical structure
Here we describe a number of models that account for the gen- resembles that of the olfactory bulb (OB) in frequency content.
eration of specific types of EEG activities, namely the following: In this model and similar ones, the action potentials are con-
verted into graded currents at the somadendritic synapses caus-
• Models describing the generation of the EEG of olfactory ing the postsynaptic potentials. In a first approximation, these
areas of the brain as proposed extensively by Freeman (1) and postsynaptic potentials may be considered to sum at the soma.
later further elaborated (33,34) that may be considered para- At the trigger zone, the axon hillock, these potentials are trans-
digmatic of neural mass modeling approaches; formed into spike trains depending on a threshold.
• Models of alpha rhythms of the thalamus and cortex, since The postsynaptic potentials may be considered as the wave
these are one of the most conspicuous features of the human (W) mode of operation of the neural set; the action potentials
EEG, as proposed by Lopes da Silva et al. (10,11) and elabo- represent the pulse (P) mode. This is, of course, a simplification
rated in more detail by van Rotterdam et al. (20), and of processes taking place at the membrane, but at the level of gen-
extended later by Jansen and Rit (12), Nunez (35), Wright and eralization at which NMMs of the EEG are constructed, these are
Liley (36), Stam et al. (37), Valdes et al. (38), Robinson et al. the main features that must be considered. The sensitivity of the
(39), Suffczynski et al. (13), David and Friston (14), and threshold process in the bulb and cortex is controlled by centrifu-
Zavaglia et al. (17,18); gal influences that originate in the forebrain and brainstem,
• Detailed models of membrane and synaptic properties of among others from the locus coeruleus (62). The conversion
thalamic cells and circuits responsible for the generation of between the postsynaptic potential (W mode) and the pulse rate
7- to 14-Hz spindle rhythmicity that occurs at the onset and (P mode) may be given in terms of the sigmoid curves shown in
in the light stages of sleep, based on the experimental findings Figure 4.4 for a single neuron and for the population or ensem-
of Steriade et al. (40–48); ble. The wave–pulse conversion can be considered as a static sig-
• Models of epileptiform EEG, namely of absence seizures, that moidal nonlinearity at the single neuron level. However, over
display complex nonlinear dynamics, namely bistable behav- small signal ranges this relation may be assumed to be linearized,
ior, where abrupt transitions occur between two states—nor- and therefore it may be represented by a simple coefficient. To
mal and paroxysmal (49,50)—and of epileptic seizures of the simulate the local dynamics of the EEG, delays due to synaptic
limbic cortex (19,51), as well as detailed models of the gener- and conduction processes may be accounted for by a stage of
ation of epileptiform transients by Traub (52), Traub and integration. Along these lines, the dynamics of these structures
Wong (53), Traub et al. (54,55), and Traub and Miles (56); may be represented by a mathematical model consisting of cou-
• Models of beta and gamma rhythms developed by Traub et al. pled differential equations that incorporate static nonlinear func-
(57) with realistic simulations of synapses (55,57–60). We do tions (for a mathematical treatment see Refs. 1, 34, and 63). The
not include here models of hippocampal theta rhythms since main input to the populations of the OB comes from the olfac-
the basic physiology of these rhythmic activities has already tory mucosa and is modulated by the respiration rate. A study of
been described in Chapter 3. the behavior of the KII set described above shows that this set has
multiple stable states. Two are of particular interest because they
Model of Olfactory Area EEG represent the most common situations. The first is a quasiequi-
The main neuronal populations of the olfactory brain struc- librium state that is maintained by a steady input of low intensity;
tures that are of importance for an understanding of the gener- the second is the so-called carrier state (34), characterized by a
ation of the olfactory EEG can be summarized as follows: in the limit cycle with quasisinusoidal EEG activity at 40 Hz (cat) or 60
bulb, the mitral and tufted cell (M) population, which is excita- to 80 Hz (rabbit) that occurs during inspiration. This nonlinear
tory, and the granule cell (G) population, which is inhibitory; in oscillation represents a limit cycle because the feedback gain is
the PC, the surface pyramidal population, which is excitatory displaced to the excitatory side of the resting value with an
(E), and the basket cell population, which is inhibitory (I). The increase in pulse input (either of sensory origin or depending on
interaction between the M and the granule neuron populations arousal level) that may drive the KII set out of the equilibrium
forms an inhibitory feedback loop. The interaction between the state. Certain drugs such as carbachol or picrotoxin can induce a
surface pyramidal and the basket cell populations in the PC also limit cycle state of long duration (34). A later section shall con-
forms an inhibitory feedback loop (Fig. 4.1). sider the relevance of nonlinear dynamics such as the theory of
As can be seen in Figure 4.1, there are also mutual excitatory deterministic chaos for understanding EEG generation. Here it is
interactions between M neurons in the bulb and between sufficient to observe that the EEG signals simulated by the set of
pyramidal A neurons in the PC, as well as mutual inhibitory nonlinear coupled differential equations that describe the KII set
interactions between granule cells in the bulb and basket cells in approximate the experimentally recorded EEGs.
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 71

rhythmic activities is the question of whether such rhythms are

caused by single cells with pacemaker properties. A significant
advance in this discussion was achieved when it was demon-
strated by Jahnsen and Llinás (70,71) that some types of thalamic
neurons display oscillatory behavior in vitro, even after blockage
of synaptic transmission. Most neurons studied in this way tend
to generate oscillations in the frequency range of 6 to 10 Hz.
These intrinsic oscillatory properties of some neurons are most
likely of importance in shaping the rhythmic behavior of the net-
works to which they belong. However, these properties may not
be sufficient to account for the network rhythmic behavior. One
main argument supports this statement: the disconnection of
cortically projecting thalamic cells from their inputs, which arise
in the reticular (RE) nucleus of the thalamus, abolishes spindles
in thalamocortical relay (TCR) neurons (72). However, single
TCR neurons may preserve their intrinsic membrane oscillatory
properties in vitro (70,71). Interestingly, the capacity of both
TCR–RE circuits and the isolated RE nucleus, in vivo, to generate
spindle oscillations contrasts with the fact that the RE nucleus
isolated in vitro does not show spontaneous oscillations (73–78).
Figure 4.4 Conversion operations between the wave mode (W) and This is likely because the input conditions of the RE neurons in
the pulse mode (P) for a neuron and an ensemble of neurons. At the vitro are different from the situation in vivo because the neurons
neuronal level both operations are time-varying (t); the P-W conversion lack neuromodulation influences, including cholinergic, nora-
is nonlinear; the W-P conversion is linear between threshold and catho- drenergic, and serotoninergic inputs. In modeling studies, the
dal block. At the ensemble level, the P-W conversion is restricted to a influence of neuromodulators on RE neurons membrane poten-
small linear range and W-P conversion is static and nonlinear. The tial was simulated (68), namely noradrenaline and serotonin, by
derivative of this curve defines the gain; this may change ( o) depend- blocking leak K currents and thus depolarizing the RE neurons
ing on arousal level. The P-W conversion can vary to represent synap- to the same level as observed in vivo. These studies showed that
tic changes occurring with learning. (Adapted from Freeman WJ. waxing–waning oscillations could occur in the RE network, sim-
Dynamics of image formation by nerve cell assemblies. In: Basar E, ilar to those seen in vivo. This finding provides clear evidence that
Flohr H, Haken H, et al., eds. Synergetics of the Brain. Berlin: Springer- the oscillatory behavior of these neuronal populations of RE neu-
Verlag; 1983:102–121.) rons depends on specific levels of the neuronal membrane poten-
tial that is affected by several inputs, both at short- and
long-range. Among the latter, the neuromodulatory inputs aris-
Alpha Rhythm Models: Spatial ing from the raphe nucleus (serotoninergic), the locus coeruleus
and Temporal Characteristics (noradrenergic), as well as glutamatergic projections form fore-
A current view in neurophysiology is that the origin of certain brain areas are particularly important. Among the former it is
rhythmic electrical activity seen in the EEG, such as that char- interesting to note that gap junctions have been shown to exist
acteristic of some thalamic nuclei and cortical areas after between RE nucleus (79) and thus may also participate in the
administration of a barbiturate (64,65) and possibly also that modulation of their oscillatory properties.
occurring spontaneously, such as sleep spindles and alpha To understand how oscillations may be generated in thala-
rhythms (66,67) is to be found in populations of neurons char- mic nuclei, we must consider the basic structure of these neu-
acterized by the interaction between excitatory and inhibitory ronal networks. In the thalamic relay nuclei where this type of
neuronal populations. Different types of models have been rhythmic activity can be recorded, namely the relay nuclei of
developed to account for the generation of EEG rhythmic activ- the thalamus, three main types of neurons can be distinguished
ities within the alpha frequency range. Some of these models (see also Chapter 3 for a discussion of the physiology of these
operate at the microscopic level of the single neuron or the cellular elements): the TCR neurons, whose axons project to the
small local network (68,69), others at the macroscopic level cortex; the RE nucleus neurons that contribute to the inhibitory
with the aim of accounting for global properties of EEG signals feedback control of the former; and the local, intrinsic, neu-
recorded from the scalp (not limited to alpha rhythms) (35). rons. The RE forms a thin sheet of neurons that surround the
Recently attempts have been made to combine the local and the anterior and lateral side of the thalamus and interact synapti-
global levels within a synthetic theoretical framework, among cally with the TCR neurons. The axons of the RE cells have as
others by Nunez who developed a thoughtful “marriage of local main target the TCR cells. Their dendrites make synaptic con-
and global theories” (Chapter 11 in Ref. 35). A particular inter- tacts with the axons of the TCR cells. Thus the TCR and the RE
esting discussion of this issue is the paper of Wright and Liley neurons are interconnected by means of a feedback loop as
(36) followed by a wide-open peer commentary. schematically shown in Figure 4.5. Both the RE and the local-
At the cellular level, an important issue that has been repea- circuit neurons contain aminobutyric acid (GABA) as trans-
tedly discussed in studies of the mechanisms underlying EEG mitter. This implies that the feedback of the RE on the TCR
72 Part I ■ Basic Principles

due to microphysiological investigations in thalamic slices, in

vitro, and in isolated brain preparations by the groups of Llinás
and collaborators (70,71). For a comprehensive account of
these membrane properties, the reader is referred to the origi-
nal publications and to the excellent reviews by Steriade and
Llinás (80), Steriade et al. (66,67), McCormick (81),
McCormick and Bal (82), Destexhe et al. (69), Destexhe and
Sejnowski (83), (84), Lörincz et al. (85), and to Chapter 3.
A basic question is whether the different types of oscillations
can be found in vitro in the absence of external inputs, since
this would give support to the idea that these thalamic cells may
support pacemaker rhythmic activity. Jahnsen and Llinás (70)
provided an initial answer to this question by showing that the
oscillatory activity at about 10 Hz may be observed in the
absence of external drive in vitro. However, we must also con-
sider what is the contribution of the feedback circuits between
TCR and RE neurons to the generation of EEG oscillations
in vivo.
In this context the most direct experimental evidence per-
tains to the generation of spindle oscillations occurring during
the early stages of sleep. Spindles appear in the EEG of most
mammals as oscillations with a frequency in the range from 7
Figure 4.5 Simplified scheme of a thalamic neural network where the
to 14 Hz, which form bursts lasting for 1.5 to 2 sec that recur
main features are presented. The thalamocortical relay (TCR), the retic-
periodically with an irregular periodicity of about 5 to 10 sec.
ular thalamic (RE), and the local-circuit interneurons (IN) are indicated,
This type of spindle can also be seen in animals under barbitu-
along with projections from a cholinergic population (ACh). The synap-
rate anesthesia. It was shown by Steriade et al. (72) that no thal-
tic triad formed by the dendrites of TCR and IN neurons with the spe-
amic nucleus is capable of generating spindle oscillations after
cific afferent (spec. aff.) fibers is also indicated. Three types of synapse
disconnection from the RE nucleus. Therefore, the existence of
are shown: excitatory, aminobutyric acid (GABA)ergic (inhibitory), and
feedback circuits including the RE nucleus is necessary for spin-
cholinergic. The effect of the latter on RE neurons is assumed to be an
dling to occur under normal conditions in vivo. Nonetheless
increase in potassium conductance (gk), whereas on the TCR neurons
there are other mechanisms that may be responsible for alpha
the effect is the opposite. (From Lopes da Silva FH. Neural mechanisms
rhythmic generation at least in brain slices in vitro but have not
underlying brain waves: from neural membranes to networks.
been incorporated in explicit models as described by Lörincz et
Electroencephalogr Clin Neurophysiol. 1991;79:81–93, with permission.)
al. (85) and in Chapter 3.
In conclusion, under the normal conditions of an intact
brain, both synaptic interactions and input signals play an
neurons is GABAergic and thus inhibitory. The RE neurons essential role in setting the conditions necessary for oscillatory
receive fast GABAA-mediated inhibitory postsynaptic potentials behavior to occur. The frequency of this oscillation depends on
(IPSPs) from neighboring RE neurons; in this way these neu- the intrinsic membrane properties on the membrane potential
rons form a chain of cells interconnected by inhibitory of the individual neurons and on the strength of the synaptic
synapses. In addition they receive fast glutamatergic (AMPA interactions.
type) excitatory postsynaptic potential (EPSPs) from TCR neu-
rons. In turn the TCR neurons receive both fast GABAA and Simplified Distributed Network Model
slow GABAB IPSPs from RE neurons (74,75). In a thalamic A simplified model has been proposed (10,11) that incorpo-
nucleus, a number of feedback circuits between TCR and RE rates the limited physiological and histological data available at
neurons can be distinguished. that time; these assumptions were sufficient to explain the gen-
In the context of this discussion, it is important to note that eration of the alpha rhythm, that is, the EEG stochastic signal
the TCR neurons can work essentially according to two distinct that lies in the frequency range of 8 to 13 Hz. This appears
modes: either (a) as relay cells that simply depolarize and can mainly at eye closure and can be recorded best from the poste-
produce single spikes in response to an adequate input volley or rior regions of the scalp in humans. A similar type of activity
(b) as oscillatory cells producing bursts of high-frequency has also been recorded at the visual cortex and some thalamic
spikes that are repeated in a rhythmic fashion. The resting nuclei in dogs (86) and cats (87,88). The model in its most sim-
membrane potential of a TCR neuron determines whether such ple form consists of a set of simulated neurons, thalamocortical
a neuron will be active in the relay or in the oscillatory mode. relay (TCR) cells, and interneurons (INs). The latter may be
To understand this question, it is necessary to take into account assumed to be the RE cells as shown in Figure 4.5, although the
the intrinsic membrane properties of these cells, namely their IN cells shown in this figure may have a complementary effect.
main ionic conductances and the corresponding dynamics. It For simplicity in the model, the cells providing feedback are
should be noted that these properties became known mainly denoted, in the following, as INs. In contrast with the lumped
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 73

models, some examples of which have been given for the olfac- tory period then followed. A basic characteristic of this model
tory system, this model is a distributed one in the sense that was that the input received by each TCR neuron was in the
each neuron is modeled individually; it occupies a specific posi- form of a series of pulses (action potentials) that had a
tion within a matrix of neurons. Such a distributed model Poisson distribution; the inputs to the individual TCR neu-
offers some special possibilities for theoretical studies that are rons were uncorrelated. The model’s output signal was the
not easily available in lumped models. sum of the membrane potential fluctuations of all TCR neu-
Without entering into great detail about the alpha rhythm rons. This output signal simulated the general spectral prop-
distributed model, it is of interest to point out some of its erties of an alpha rhythm. This result is illustrated in Figures
characteristics. First, the structure of the distributed model 4.6 through 4.8.
initially used (10) consisted of 144 TCR neurons and 36 INs; In the distributed model, it is also easy to investigate the
the 32 TCR neurons that surrounded one IN were responsible result of changing the range of excitatory and inhibitory influ-
for its excitation, while each IN was responsible for the inhi- ence of, respectively, TCRs and INs. In the case of 4.7, there
bition of the 12 TCRs around it. The excitation and inhibition were 12 TCR neurons that were inhibited by one IN. This is the
were represented by time functions that were an approxima- common situation; the effects of reducing this number to four
tion of the waveforms of EPSPs and IPSPs, respectively. The and of increasing it to 21 are shown in Figures 4.7 and 4.8.
matrix of neurons should be viewed as lying at the surface of Reduction of the inhibitory area of influence of an IN results
a torus. Furthermore, each neuron fired whenever the mem- in a decrease of the spectral peak amplitude (alpha frequency)
brane potential exceeded a simple threshold; a short refrac- and an increase in bandwidth; an increase in the inhibitory
area has the opposite effects. These are some of the phenom-
ena that can be studied using a distributed neuronal model of
rhythmic activity. However, this type of model has one main
disadvantage: it is difficult to treat analytically. Therefore, it
was necessary to develop a lumped model that would take into
account the main characteristics of the distributed model. A
lumped model of the same type as those of Freeman, previ-
ously described, permits a more general treatment of the acti-
vity of neuronal populations.

The Neural Mass (Lumped) Alpha Rhythm Model

The development of the lumped model in analytical terms
included the step of translating the properties of the discrete

Figure 4.6 Results obtained with the distributed model of alpha Figure 4.7 Output signals of the distributed alpha model: Vm (t) repre-
rhythm. Upper panel: Input density of the input to all TCR neurons. sents the summed membrane potentials of all TCR neurons and E(t) rep-
Middle panel: Wave mode summed membrane potentials of all TCR resents the impulse density of the same neurons. The power spectrum
neurons. Lower panel: Pulse mode: impulse density of all TCR neurons. of Vm (t) is shown on the left below. In this case, a smaller inhibitory area
Right: The amplitude distributions and the power spectra of the three than in the case of Figure 4.8 has been simulated. Here one interneuron
signals are shown. Note the apparent waxing and waning of the simu- (IN) inhibits only four TCR neurons. (Adapted from Lopes da Silva FH,
lated alpha rhythm and the dominant alpha rhythm with a second har- van Rotterdam A, Barts P, et al. Models of neuronal populations: the
monic. (Adapted from Lopes da Silva FH, Hoeks A, Smits H, et al. basic mechanisms of rhythmicity. In: Corner MA, Swaab DF, eds.
Model of brain rhythmic activity. Kybernetik. 1974;15:27–37.) Perspectives of Brain Research, vol. 45. Prog Brain Res. 1976:281–308.)
74 Part I ■ Basic Principles

Figure 4.8 Similar to Figure 4.7 but for a model with a larger Figure 4.9 Block diagram of the lumped model for rhythmic activity
inhibitory area. Here 1 IN inhibits 21 TCR neurons. Note that the inten- with a possibility of modulation of inhibitory feedback and excitatory
sity of the alpha peak is larger and the bandwidth is smaller in compar- feedback. The cartel consists of four populations. M, main cells with
ison with the model shown in Figure 4.7. (Adapted from Lopes da Silva input and output impulse density, respectively, P(t) and E(t), and mem-
FH, van Rotterdam A, Barts P, et al. Models of neuronal populations: brane potential, Vm (t); spike generation nonlinearity sigmoid function.
the basic mechanisms of rhythmicity. In: Corner MA, Swaab DF, eds. INIF, inhibitory population with membrane potential Vi1(t) and modulat-
Perspectives of Brain Research. vol. 45. Prog Brain Res. 1976;281–308.) ing inputs: excitatory by means of M1(t) and inhibitory by means of
M2(t), spike generation with sigmoid. INFF, excitatory population driven
by the main input P(t) inhibiting the inhibitory population INIF; INEF,
excitatory population feeding back on to the main population, spike gen-
matrix of neurons into a system such as that depicted in eration sigmoid. The interconnectivity constants giving the number of
Figure 4.9. Essentially, the neural mass (here thalamic nucleus), cells of one population projecting to one cell of the other population are
the EEG of which is of interest, forms a KII set, which is consti- C1 and C2 for the inhibitory feedback and C4 and C5 for the excitatory
tuted by the interaction of an excitatory (M) and an inhibitory feedback. C3 represents the number of cells of the INFF population
(INIF) population (KIe and KIi sets). In addition, the scheme of excited by one input fiber (feed-forward). (Adapted from Lopes da Silva
Figure 4.9 shows a second KIe (INEF) set that provides excita- FH, van Rotterdam A, Barts P, et al. Models of neuronal populations: the
tory feedback and an additional KIe set that may receive a direct basic mechanisms of rhythmicity. In: Corner MA, Swaab DF, eds.
input (P(t)) from a source situated outside the neural mass and Perspectives of Brain Research. vol. 45. Prog. Brain Res. 1976:281–308.)
in this way lead to feed-forward inhibition. To obtain an ana-
lytic simplification of the distributed model, it is useful to con-
sider the general formalism of Wilson and Cowan (9) as
described in the introduction to this chapter.
Zetterberg (89) has made use of this formalism to simplify
the distributed model of alpha rhythm. Considering the simpli-
fied block diagram of Figure 4.10, where the EPSPs and IPSPs
are represented by he(t) and hi(t), respectively, and the sigmoid
wave pulse transfer functions are represented by fE(V) and fi (V),
expressions for the transfer function of the simulated neuronal
set can be obtained (for the mathematical treatment, see Ref. 11).
One of the outputs of such a neuronal set represents the EEG
signal of the population, assuming that the latter is determined
by the somadendritic membrane potential of the main cells.
The corresponding spectrum is shown in Figure 4.11 for differ- Figure 4.10 Block diagram of the lumped model for rhythmic activity of
ent values of K, which represents a measure of the gain factor of simplified alpha rhythm model. The thalamocortical relay (TCR) neurons
the neuronal set. The spectrum R(f) indeed shows a peak are represented by two linear systems having impulse responses simulat-
around 10 Hz, as was expected. The bandwidth and the peak ing an EPSP (h c(t)) and an IPSP (h i(t)), respectively, together with the
frequency both depend on the coefficient K. Up to a point, a larger static nonlinearity (fE(V)) representing the spike-generating process. The
K will cause an increase in the frequency of the rhythmic activ- interneurons (IN cells) are represented by one linear system (h e(t)) and a
ity, but too large a K will cause instability. K is also proportional nonlinearity (fI(V)). C1 represents the number of INs to which one TCR
to the derivatives 1/ael and 1/ail of the sigmoid functions in their neuron projects; C2 represents the number of TCRs to which one IN proj-
working point, which in turn is determined by the mean input ects. (From Lopes da Silva FH, Hoeks A, Smits H, et al. Model of brain
pulse density P(t). The peak frequency will increase and the rhythmic activity. Kybernetik. 1974;15:27–37, with permission.)
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 75

Figure 4.12 An epoch of alpha rhythm (A) experimentally recorded

from the visual cortex of the dog (technique as used in Ref. 86). The
Figure 4.11 Power spectra of the output of the lumped alpha rhythm power spectrum (B) and the bicoherence (C) were obtained by ensem-
model shown in Figure 4.10 for different values of the feedback gain ble averaging. Note the peak in the power spectrum at about 10 Hz and
(K), which is mainly determined by the coupling constants representing the smaller peak at the second harmonic frequency, which are related
the synaptic interactions within the neuronal set. As the synaptic inter- to each other as the bicoherence shows. A: Experimentally recorded
actions become stronger and K increases, the spectrum acquires a clear alpha rhythm (arbitrary scale). B: Power spectrum (arbitrary scale). C:
selectivity at the alpha frequency. (Adapted from Lopes da Silva FH, Bicoherence limits 99% at 0.119, 95% at 0.096, five lines from 0.18
Hocks A, Smits H, et al. Model of brain rhythmic activity. Kybernetik. until 0.26. (Adapted from Lopes da Silva FH, van Rotterdam A, Barts P,
1974;15:27–37.) et al. Models of neuronal populations: the basic mechanisms of rhyth-
micity. In: Corner MA, Swaab DF, eds. Perspectives of Brain Research.
vol. 45. Prog. Brain Res. 1976:281–308.)
bandwidth will decrease with an increase of the coupling
constants, C2 and C1, that is, with an increase in synaptic inter-
actions within the neuronal cartel. terms of phase differences in rhythmic components at different
The linear analysis of the preceding section gives only a first locations on the cortical surface.
approximation to the spectral characteristics of the alpha One hypothesis is that the propagation of alpha activity
rhythm, both experimentally measured and simulated. The spec- depends on spatial properties of the cortical neuronal net-
trum of the experimentally measured alpha rhythm also presents works; the strength of inhibition is a function of distance
a clear component at the second harmonic of the peak frequency depending on the length of interneuronal connections. This
(Fig. 4.12). It can be demonstrated by means of bispectral analy- hypothesis has been realized by way of a “spatially distributed
sis that this component is indeed a second harmonic of the peak model” of neurons and interneurons (20), which is an exten-
frequency. This fact implies that static nonlinear properties of the sion of the lumped parameter model described by Lopes da
network have also to be taken into account. Silva et al. (10). In this spatially distributed model, it is assumed
that two populations of neurons (main cells and local interneu-
Continuous Neural Field Model of Alpha Rhythm rons) are arranged in a one-dimensional row and are intercon-
The spatial properties of alpha rhythms are less well under- nected by means of recurrent collaterals and inhibitory fibers.
stood, although it has been found experimentally that, in the Such a neuronal chain is assumed to have the following proper-
visual cortex of the dog, small cortical areas appear to act as ties: (a) the transfer function of input-output spike densities are
“epicenters” from which alpha rhythm activity “spreads” in dif- second-order linear functions with a longer time constant for
ferent directions (21,179). The spreading has been measured in inhibition than for excitation; (b) the interconnectivity properties
76 Part I ■ Basic Principles

Figure 4.13 Transfer properties of the spatially distributed model or

neuronal chain. Left: The time courses of the simulated IPSP and EPSP Figure 4.14 Schematic drawing of a section of the spatially distributed
are shown with the corresponding equations. Right: The spatial connec- model (neuronal chain). For explanation see text. (Adapted with per-
tivity functions are shown. Ce(x) and Di(x) describe the homogeneous mission from van Rotterdam A, Lopes da Silva FH, Van den Ende J, et
interconnectivity functions for the excitatory and the inhibitory connec- al. A model of the spatial-temporal characteristics of the alpha rhythm.
tions, which depend only on distance x. The values of the characteris- Bull Math Biol. 1982;44(2):283–305.)
tic lengths correspond to l/ c and l/ d. The numerical values of c and d
in the figure are only indicative. (Adapted from van Rotterdam A, Lopes
da Silva FH, Van den Ende J, et al. A model of the spatial-temporal char- model may be written as a linear inhomogeneous partial differ-
acteristics of the alpha rhythm. Bull Math Biol. 1982;44(2):283–305.) ential equation, which is a rather complicated type of wave
equation. By assuming a wave packet existing over a restricted
length somewhere in the neuronal chain at some instant of
are homogeneous in space, that is, the strength of inhibition is time, one can try to determine group velocity and dispersion
only a function of distance between the interconnected neu- properties. Alternatively, the neuronal chain may be conceived
rons; and (c) the strength of interconnectivity decreases expo- as a two-dimensional linear system with a specified transfer
nentially as a function of distance, as shown in Figure 4.13. function relating the input p(x,t) to the output u(x,t). The
To simulate alpha activity as a function of both space and response to an arbitrary input can then be obtained in the time
time, the lumped model must be extended in such a way that it domain by means of a two-dimensional convolution or in the
can account for spatial properties. This can be done as follows:
the spatially distributed model consists of a chain of neurons
and interneurons interconnected by means of spatially distrib-
uted excitatory collaterals and inhibitory fibers from the
interneurons. The interconnections are described by Ce(k,l),
denoting the number of excitatory synapses from the kth main
cell projecting on the lth interneuron, and Ci(k,l), denoting the
number of inhibitory synapses from the kth interneuron on the
lth main cell. Figure 4.14 shows a schematic drawing of such a
neuronal chain, where T k is the kth main cell and Ik is the kth
local interneuron. The main cells receive input along afferent
fibers from outside the chain. Figure 4.15 shows the correspon-
ding block diagram. The time-dependent behavior of the mem-
brane potential U(k,t) of the kth cell is determined by the
transfer functions he(t) and hi(t) of the lumped model. Nle and
Nli denote the static nonlinearities that relate membrane poten-
tials to impulse densities of main cells and interneurons,
respectively. These static nonlinearities are assumed to have the
same characteristics as fe(V e) and fi(V i) in the lumped model
and hence can be approximated as linear gain factors in the Figure 4.15 Block diagram of a section of the neural chain model
neuronal chain. Finally, P(k,t) denotes the input impulse den- showing the main excitatory and inhibitory interconnections as
sity on the main cell k. explained in text. (Adapted with permission from van Rotterdam A,
To study the propagation properties in the neuronal chain, Lopes da Silva FH, Van den Ende J, et al. A model of the spatial-tempo-
various approaches can be used. In the first place, the expres- ral characteristics of the alpha rhythm. Bull Ma th Biol.
sion describing the spatial and temporal characteristics of the 1982;44(2):283–305.)
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 77

two-dimensional frequency domain by multiplication of the One of these systems consists of tangential intracortical col-
spectra of the transfer function and the input function (see laterals of the pyramidal cells. These may spread over dis-
mathematical treatment of van Rotterdam et al. (20)). tances of about 3 mm; they form the “pyramid collateral
It is reasonable to assume that, in order to simulate alpha spread module” (95). Another system consists of long sur-
rhythm conditions, the neuronal chain must be excited ran- face parallel terminal axons that may extend over distances
domly in space as well as in time. Due to the linear character of of about 6 to 8 mm; this forms the “surface parallel intracor-
the model, however, a simple local harmonic excitation of the tical system” (95). These systems of overlapping intracortical
chain already reveals the propagation properties of the neu- connections will lead to the establishment of cooperative
ronal chain. action between large populations of cortical neurons and
If the neuronal chain is locally excited at x = 0 with a har- thus to the appearance of dynamic patterns over the cortex
monic component of frequency 0, the model response with (96). In addition to the corticocortical systems described
the same time characteristics has a local component at x = 0 and above, one must also take into consideration the role of local
two components that spread in two directions along the chain. interneurons; Tombol (97) described, among others, basket
The amplitudes of these components decay exponentially pro- cells in the cat cortex with vertical axons branching horizon-
portional to exp[ Re{ki( 0)}|x|] and the phase shifts that tally over 500 to 1,000 m, with terminals surrounding and
reflect the propagation properties of the harmonic component making contact with pyramidal cell somata; these basket
equal exp[ Im{ki( 0)}|x|] radians when measured at a distance cells are probably inhibitory interneurons. The collaterals
x in relation to x = 0. Both phase shift and decay are thus belonging to the pyramid collateral spread module system of
dependent on the frequency of excitation. Szentagothai, which is most likely excitatory, and the axons
The results of the evaluation of the roots of the equation of the inhibitory interneurons can be identified as the exci-
describing the output of the chain as a function of space and tatory collaterals and the inhibitory fibers modeled in the
time (20) show that the rhythmic components existing in the model neuronal chain; because this system of interconnec-
chain in the alpha band are less damped than the components tions ranges over distance in the order of magnitude of hun-
outside the alpha band. Furthermore, it follows that the damp- dreds of micrometers, the estimated characteristic lengths
ing in the alpha band is mainly determined by the interconnec- computed from the experimentally measured alpha rhythm
tivity function with the largest characteristic length. frequency and phase velocity with the help of the neuronal
The model just described schematically has characteristics chain model are in agreement with neuroanatomical data.
that are in qualitative agreement with experimental data on
Finally, note that the neuronal chain can be easily general-
alpha rhythmic activity obtained from the visual cortex of dog
ized to a neuronal network extending in two or three dimen-
(21). This can be examined in relation first to coherence analy-
sions. A more general approach that accounts for the
sis and second to experimental phase measurements:
generation of alpha rhythms and its spread in the cortex was
1. Coherence analysis of EEG signals recorded from chronically proposed more recently by Liley et al. (98–100), who consid-
implanted subdural electrodes lying directly on the marginal ered the cortex as an excitable spatial continuum of recipro-
gyrus revealed a higher correlation between EEG compo- cally connected excitatory and inhibitory neurons interacting
nents in the alpha band than between components outside by way of short-range (intracortical) and long-range (cortico-
that band (90,91). These experimental data support the con- cortical) connections.
clusion from the model study that alpha rhythms are the An essential feature of this general model is that IPSPs are
least damped in the chain. described by third-order differential equations, since according
2. Phase shifts at the alpha frequency (~12 Hz) measured on to these authors lower orders were not able to support wide-
the marginal gyrus in four dogs reflected phase velocities spread activity in the alpha band. This theoretical approach
that were about 0.3 m/sec (21). Using the model, the charac- showed that the strength and the form of synaptic interactions
teristic length corresponding to this phase velocity and fre- between inhibitory interneurons constitute the most relevant
quency was computed. The order of magnitude of the determinants of the frequency and damping of alpha band
characteristic lengths 1/c and 1/d of the connectivity func- oscillations. This feature derives from the assumption that local
tions Ce(x) and Di(x) (Fig. 4.13) was estimated and found to inhibitory–inhibitory loop delays are longer than the corre-
be within the range of 160 to 330 m. The characteristic sponding local and long-range excitatory–excitatory loop
lengths correspond to the distance at which the influence of delays. This theory has been presented in the form of a set of
a neuron on a neighbor is reduced to 37% of the starting nonlinear differential equations that can be solved by numeri-
value and not to the whole length of the interconnecting cal methods, but most important predictions of the theory can
fibers. The figures that were calculated using the model are be obtained by a simplified set of linear equations. In this way
in the same range as the dimensions of the basic cortical the input to the cortex can be assumed to be indistinguishable
space module, which is considered to be a vertical cylinder of from band-limited white noise, while the cortex operates as a
200/300 m diameter (92–95). Different systems of intra- noise filter that passes preferentially frequencies around 8 to 13
cortical fibers, which involve much larger distances than the Hz. The filter can be described by a transfer function, the prop-
diameter of a column (at least distances one order of magni- erties of which are determined by both the physiological state
tude larger), assure the interconnectivity of several columns. and the anatomical structure of the cortex.
78 Part I ■ Basic Principles


In the last decades, important advances were obtained with
respect to the physiology of single cells of the thalamus and to
their patterns of interaction at the level of local networks,
including the corresponding intrinsic membrane and synaptic
properties. Interesting models incorporating these properties
were proposed, among others by Wang (44), Wang et al. (48),
Golomb et al. (42,43), Destexhe et al. (69), and have been
reviewed extensively by Destexhe and Sejnowski (83). These
models are mainly based on the experimental results of Steriade
and colleagues (101) and also of McCormick and collaborators
(74,75). The latter results were obtained in the ferret slice
preparation that facilitates carrying out controlled experi-
ments, although under circumstances that differ from the con-
ditions in the intact brain. Nevertheless, in these slices it was
possible to show (74,75) the occurrence of spindles with main
frequency between 5 and 9 Hz, during which the RE neurons
tend to fire synchronously with almost every cycle of local
rhythmic spindle. In contrast, TCR neurons tend to fire bursts
also phase-locked to the local rhythm, but not at every cycle
since they may skip one, two, or even more cycles. The modula-
tion of this activity by blocking GABAA receptors results in a
decrease of the main frequency to 2 to 4 Hz, whereas the rhyth-
mic pattern is hardly changed by blocking GABAB receptors. In
addition, the spindle bursts appear to propagate in the slice, as
traveling fronts, at a velocity of about 1 mm/sec.
As indicated above, the RE and the TCR neurons can exhibit Figure 4.16 Synaptic architecture of the model of Golomb et al. (43).
oscillations of the membrane potential, which depend strongly A: Reticular neurons (RE) receive thalamocortical (TC) excitation via
on the state of the potassium leak conductance of the cells (gkl). glutamatergic (AMPA) synapses and intra-RE inhibition via GABAA and
In the model of Golomb et al. (43) both RE and TCR neurons GABAB synapses from other RE cells. B: This model has a one-dimen-
are at rest, that is, they do not display oscillatory behavior, for sional architecture. The RE-to-TC coupling strength decays with dis-
too small or too large values of gkl and they oscillate at interme- tance. The typical decay length RT is called the synaptic footprint length.
diary levels. In this model the TCR neurons elicit EPSPs on RE The same applies to the TC-to-RE coupling (TR) and to the RE-to-RE
neurons by way of glutamatergic (AMPA) synapses, whereas RE coupling (RR). C: A typical footprint shape of the synaptic coupling with
neurons elicit both GABAA- and GABAB-mediated IPSPs on distance; the decay length (either TR or RT or RR) is the distance for
TCR neurons. Due to the existence of a low-voltage activated which the synaptic coupling reaches 1/ e of its maximal value. (Adapted
Ca current (T current) the TCR neurons can elicit a rebound from Golomb D, Wang XJ, Rinzel J. Propagation of spindle waves in a
burst after a hyperpolarization due to the GABAergic IPSP, thalamic slice model. J Neurophysiol. 1996;75:750–769.)
whether or not associated with a Ca-dependent K current
(afterhyperpolarization [AHP] current). In the model simula-
tions if the neurons were not in intrinsic oscillatory mode, the
network would be quiescent. However, the introduction of a oscillations, whereas an up-regulation of Ih elicited refractori-
few oscillating TCR neurons was sufficient to recruit the other ness of the neurons, leading to extinction of the oscillations. In
neurons into a rhythmic spindle. The oscillation was critically this model, the responses of the neuronal network to different
dependent on the existence of a cationic current activated by stimulus patterns were not reported, although this is most
hyperpolarization, the Ih or “sag” current, in these neurons (44). important to understand the behavior of the network of the
Using this model it was possible to investigate the patterns of intact brain under physiological conditions.
propagation of the oscillations throughout the network. The
neurons in the network were connected by synaptic couplings, MODELS OF EPILEPTIFORM ACTIVITY
the strength of which decays with spatial distance, as shown in
Figure 4.16. An oscillation started typically at a focal site and Here we consider those models by means of which one attempts
propagated as a wavefront that recruited successively more neu- to explain how EEG phenomena characteristic of epileptiform
rons into the oscillatory mode. The Ih current modulated the seizures may emerge from the normal ongoing activity. Most
occurrence of these oscillations: a reduction of Ih led to sustained theoretical studies have addressed in general terms the question
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 79

of how oscillations may take place in a network of neurons,

assuming that such oscillations are akin to seizure activity. A
general way to investigate this problem using a model of such a
network is to analyze the trajectories in the phase plane whose
coordinates are the potential and its first derivative. A nonlinear
system that goes into stable oscillation even after the input
stimulus has ceased will produce a closed trajectory in the phase
plane; such a closed trajectory is called a limit cycle. Working
from the model initially described by Lopes da Silva et al.
(10,11), Zetterberg et al. (102) have shown that a slightly mod-
ified model may generate limit cycle oscillations that can simu-
late some forms of epileptogenic activity recorded from the
Figure 4.17 Simulation of a bursting neuron showing an epileptiform
cortex. Freeman noted that his KII model was not initially
discharge. Each simulated neuron had a number of compartments and
designed to simulate epileptiform activity. However, by using a
ion conductances, as explained in the text. The intracellular burst
different nonlinear gain curve, Babloyantz and Kaczmarek
model is shown above a recorded burst. The network contains 100 neu-
(103) have generated limit cycle oscillations that are thought to
rons interconnected in a random synaptic network. Vertical calibration,
resemble epileptiform activity. In such models, however, a more
25 mV by 25 msec (model) and 40 msec (experimental). (Adapted from
or less precise fit between simulated and recorded EEG activity
Traub RD, Wong RKS. Cellular mechanism of neuronal synchronization
was not attempted. More recently, models that attempt a precise
in epilepsy. Science. 1982;216:745–747.)
waveform fit have been introduced. Such a model was proposed
by Traub (52) and Traub and Wong (104) and is based on
experimental data at the cellular level concerning the mem-
brane mechanisms responsible for epileptogenesis (53) together experimentally recorded epileptiform transients was obtained,
with the properties of synaptic interactions in hippocampal as shown in Figure 4.17. A second problem analyzed was how
neurons. sustained afterdischarges can be produced in such a model. It
We should note that in recent years the interest for NMMs was found that for the occurrence of repetitive afterdischarges,
with respect to epileptic phenomena has steadily increased, as some form of pacing inhibitory process was needed. This was
illustrated below, most likely due to the fact that epileptic phe- accomplished by introducing two additions in the model: (a) a
nomena involve brain systems with complex dynamics that are slow voltage-dependent K conductance (M current) and (b)
difficult to comprehend exclusively on the basis of knowledge at an axonal refractoriness that causes intermittent conduction
the cellular level. from one to the next depolarized cell, located at the axonal ini-
tial segment, at branch points, or at presynaptic terminals.
Cellular and Network Models Furthermore, a number of loops allowing reentrant paths
of Epileptiform Activity within the neuronal population are of importance for the main-
Observations at the cellular level obtained in the hippocampal tenance of a series of afterdischarges. Using this kind of models,
slice preparation have yielded a number of new facts concern- in a series of studies Traub and collaborators (57,106) investi-
ing the origin of an epileptogenic focus. It has been shown that gated the role of a number of factors in inducing epileptiform
pyramidal cells of the hippocampus (cornus ammonis or CA3) oscillations similar to those observed in brain slices, in vitro,
have active membrane regions, the so-called soma and den- namely synaptic mediated activation of N-methyl- D-aspartate
dritic hot spots that have voltage-dependent ion conductances (NMDA) dendritic synapses and intrinsic voltage-dependent
(g) for Na , K , and Ca2 and a calcium-activated K conduc- conductances. The possible importance of the firing of ectopic
tance; furthermore, a voltage-dependent inactivation of gK and presynaptic action potentials in eliciting the synchronization of
a partial inactivation of gCa by the cytosolic accumulation of neurons during epileptiform seizure-like activity was suggested
Ca2 ions has been assumed (52). Incorporating all these (57). Another factor that may contribute to epileptiform activity
properties, Traub (52) has made a multicompartmental model has been put forward on the basis of models and in vitro physi-
by means of which spontaneous bursts in CA3 cells can be gen- ological studies of Traub’s group (55): gap junctions located
erated, as shown in Figure 4.17. Using these elementary cellular between the axons of principal hippocampal neurons would
components (105), Traub et al. (54) constructed a network of play a role in the generation of very fast (about 200 Hz) oscilla-
CA3 neurons interconnected by excitatory synaptic connec- tions. These model simulations have allowed making the predic-
tions. The generation of epileptiform transients induced by the tion that action potentials can cross from axon to axon via gap
administration of antagonists of the neurotransmitter GABA junctions. Axonal network oscillations would, in turn, induce
was the first problem considered. The conditions for the pro- oscillatory activity in larger neuronal networks, by a variety of
duction of a chain reaction of excitation that spreads over the mechanisms, and could underlie in vivo ripples (to ~200 Hz in
whole network were found using the simulations. The most hippocampal EEG) to drive gamma (30 to 70 Hz) oscillations
important of these conditions was that the excitatory intercon- that appear in the presence of carbachol and to initiate certain
nectivity should be sufficiently dense and the synaptic interac- types of ictal discharges. A comprehensive review of the work of
tions of sufficient strength. A good fit between simulated and Traub and collaborators can be found in Traub et al. (107).
80 Part I ■ Basic Principles

Along the same line high-frequency ripples may be determined further, that the transition between interictal and fast ictal
by high-frequency inhibitory activity, the breakdown of which activity is explained, in the model, by the impairment of den-
may enable seizure activity to propagate from an initial focal dritic inhibition (Fig. 4.18).
area (108,109). Also using NMMs Kramer et al. (119–121) constructed
In addition, model studies have also addressed the possibil- models of human cortical networks that can generate seizure-
ity that changes in extracellular ionic concentrations (increase like activity, as the result of increased connectivity between
of K and/or decrease of Ca2 ) may be important for the gen- excitatory and inhibitory cell populations or of decreased con-
eration of epileptiform seizure activity. Somjen and collabora- nectivity within either excitatory or inhibitory cell populations.
tors (110–112) used computer models to better understand These studies also allowed to test the possibility of controlling
how transmembrane ionic and water movements may affect seizures (120) by means of feedback controllers of the model
neuronal excitability and observed that the model produced dynamics.
epileptiform afterdischarges if during stimulation interstitial In contrast with NMMs where the topology of the network
K concentration was sufficiently enhanced so that the persist- is not taken into account or is much simplified, topological mod-
ent Na (slowly inactivating) current was activated. In order for els there have been developed, that is, graphical representation
firing of a neuron to be maintained after the end of a stimulus, models where topological details of the neuronal networks are
the [K o]/[K i] ratio had to be sufficiently large to maintain explicitly included. These models permit to simulate the effects
the neuronal membrane depolarized so that the inward cur- of histological changes that may occur in an epileptogenic tis-
rents could persist, which activated a series of action potentials. sue, such as the death of certain cellular types and the sprout-
In the course of time the persistent Na current was inactivated ing of new axonal or dendritic connections. Thus Lytton et al.
and the ion pumps, associated with the buffering of K ions by (122) developed a network model of the dentate gyrus of the
glia cells, restored the normal levels of K extracellular concen- hippocampus including granule cells, inhibitory aspiny
tration. These model studies strongly suggest that the elevation interneurons, and excitatory mossy cells. According to this
of extracellular K is not the initial stimulus for a seizure, but it model mossy fibers sprouting and disinhibition were necessary
may cause the transition from interictal to ictal discharges for repetitive granule cell firing, alike to EEG seizure activity, to
(113). The initial stimulus starting the seizure may require occur. More elaborate topological models of similar kind were
synaptic interactions among a population of interconnected developed by Santhakumar et al. (123) and Dyhrfjeld-Johnsen
neurons. et al. (124). The former showed that if sprouting of mossy fibers
In contrast with the models described above, which are exceeds a certain value, seizure activity can occur regardless of
directed to the simulation of epileptiform activity as seen at the the number of hilar cells lost. The latter model study, including
cellular level, especially in in vitro preparations, other models more than one million cells, simulated the topology of the den-
are based on a neural mass approach using mean-field concepts. tate gyrus. This can display globally and locally connected
A paramount example of these is the model study of Wendling (“small world”) architecture. The effect of changes of network
et al. (19) that focuses on high-frequency EEG activities topology during epileptogenesis was modeled on the one hand
(gamma band) that constitute one of the most characteristic by removing long-distance projecting hilar cells which causes a
EEG patterns in focal seizures of human epilepsy, especially reduction in the connectivity and, on the other, by sprouting of
those that have an epileptogenic zone in the hippocampus and granule cell axons which has the opposite effect of increasing
associated limbic cortical areas (114,115). This model starts local connectivity. The model simulations demonstrated that
from the hypothesis that networks of inhibitory interneurons the survival of even a small fraction of hilar cells was enough to
with different properties are essential for the generation of sustain hyperexcitability in the network. This is consistent with
gamma frequency oscillations (116). As already pointed out experimental data since in animal models of hippocampal scle-
above, this lumped model includes different kinds of neuronal rosis the loss of hilar cells is never 100%; in our study (125) it
populations, in particular two kinds of GABAergic neurons: was about 66% to 74% in rats with a progressive form of epilep-
GABAA fast and GABAA slow populations, where the former tic seizures. Also in brain tissue extracted from patients with
cause IPSPs with short time constants on the soma of pyrami- hippocampal sclerosis about 20% of mossy cells have been esti-
dal neurons, while the latter cause IPSPs with slow time con- mated to survive (126), which corresponds closely with the
stants on the dendrites of these neurons. Furthermore as finding in the model of Dyhrfjeld-Johnsen et al. (124) that
demonstrated experimentally by Banks et al. (117), both classes maximal epileptiform activity was seen at a level of about 80%
interact: GABAA slow cells inhibit not only pyramidal cells but sclerosis.
also GABAA fast interneurons. The model is based on the In general we may state that a variety of changes at the cellu-
important experimental observation that a nonuniform alter- lar level may lead to the occurrence of epileptiform seizure
ation of GABAergic inhibition underlies the generation of activity. In fact there is no unique process that can account for
epileptiform activity in hippocampus and associated areas of the generation of epileptiform activity (127). A generalized
the limbic brain, namely a reduced dendritic inhibition and concept is that seizures occur due to a shift from a dynamical
increased somatic inhibition (118). The simulations show that unbalance between excitatory and inhibitory processes with a
strikingly realistic activities are produced by the model when predominance of the former in strongly coupled networks. Van
compared to real EEG signals recorded with intracerebral elec- Drongelen et al. (128) showed, however, in a detailed computa-
trodes in patients with temporal lobe epilepsy. They show, tional model that seizure-like activity may occur when the
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 81

Figure 4.18 A: A real depth-EEG signal

recorded in human hippocampus at the begin-
ning of a partial temporal lobe seizure with
intracerebral electrodes (SEEG). Four phases (a-1
to a-4) are distinguished according to the pseudo-
stationary nature of the activities reflected by the
signal: normal background activity (a-1), dis-
charge of rhythmic spikes (a-2), low-voltage
rapid discharge (a-3), and high-amplitude quasi-
sinusoidal slowing down activity (a-4). The nor-
malized power spectral density (PSD) shows that
part of the activity corresponding to phase a-3
belongs to the gamma band (55 to 60 Hz). B: A
candidate path on the activity map, for a value of
excitation (A = 5) and different values of slow
(B) and fast (G) inhibition, showing the transi-
tions between different types of activity observed
in the real signal (type 1 → type 3 → type 4 →
type 6). C: Slow and fast inhibitory synaptic gain
profiles defined by the candidate path and used
in the model to simulate a timeseries signal with
transitions in dynamics. D: Simulated EEG signal
obtained with the model when the synaptic gains
vary as a function of time and PSDs computed on
the four periods of 10 seconds (b-1 to b-4).
(Adapted from Wendling F, Bartolomei F,
Bellanger JJ, et al. Epileptic fast activity can be
explained by a model of impaired GABAergic
dendritic inhibition. Eur J Neurosci.

excitatory synapses are weakened. The model predictions were respect the single neuron model of a thalamic relay neuron of
tested experimentally and the authors showed that a pharmaco- Wang (44), which shows both 10- and 3-Hz bursting dynamical
logical reduction of excitatory synaptic transmission could lead modes, is particularly instructive. This model demonstrates
to repetitive network bursting. clearly how distinct oscillatory modes depend on different bal-
ances of sets of ion currents. In this model the emergence of
Models of Thalamic Spike-and-Wave Seizures 3-Hz bursts, which resemble the 3-Hz oscillations seen in gen-
Thalamic neurons can display different modes of activity eralized epilepsies, depends on the value of the hyperpolarization-
including rhythmic bursting that are akin to spike-and-wave activated cationic current, Ih the so-called sag current, although
(SW) discharges characteristic of generalized epilepsies. In this the T-type Ca current also plays an important role. According
82 Part I ■ Basic Principles

to this model the 10-Hz oscillatory mode is maintained, at the both kinds of activity without specific adjustments of parame-
level of a single neuron, by the dynamics of the low-voltage- ters being expressly made. Indeed the essence of epilepsy is that
activated Ca current, IT. Under normal conditions the neuronal a patient displays (long) periods of normal EEG activity (i.e.,
membrane does not hyperpolarize sufficiently to activate the Ih nonepileptiform) intermingled with epileptiform paroxysmal
current. activity only occasionally. Thus, it is essential to understand the
Indeed if the neuronal membrane would be deeply hyperpo- mechanisms responsible for transitions from normal activity to
larized, it would stay in such a condition in case only IT would paroxysmal SW discharges. This aspect of the dynamical
be active, because the latter current is not sufficient to overcome process responsible for epilepsy was addressed in a computa-
a strong hyperpolarization. However, the presence of Ih allows tional model of absence epilepsy (50). In this model the behav-
the neuron to escape this strong hyperpolarized state. Thus the ior of populations of interacting neurons integrating neuronal
transition between the 10- and the 3-Hz modes depends on the and network properties was simulated, as a more extended and
level of hyperpolarization. This single neuron model is com- elaborated version of a previous model of the alpha rhythm
pletely deterministic and it may display “strange attractors” typ- (132). This approach facilitated investigating system dynamics
ical of a chaotic dynamic state. at the macroscopic level, that is, at the level where electric brain
A model of a thalamic neuronal network that is closer to the signals such as local field potentials or EEG are recorded. The
reality of epileptic discharges than the single neuron model model may exhibit two qualitatively different types of behavior,
described above was developed to account for SW seizures at such as seen in experimental animals with genetically deter-
the level of a neuronal population. The basic physiological phe- mined absence-like seizure (WAG/Rij rats and GAERS ani-
nomenon simulated was the finding that during SW discharges mals). The output signal may display a waxing and waning
the RE neurons discharge with long spike bursts riding on a “spindle-like” oscillation having a spectrum with a peak at
depolarization, whereas the TCR neurons are either entrained approximately 11 Hz or a high-amplitude “seizure-like” oscilla-
into the seizure oscillation or are quiescent (129). The spatial tion at a frequency around 9 Hz, as seen in these rats during
model of these neuronal populations predicts that in a center of absence seizures (Fig. 4.19). The former behavior corresponds
seizure activity there will be intense RE activity and TCR quies- to the normal ongoing activity, while the latter corresponds to
cence and that this focus will be surrounded by an area where the paroxysmal epileptiform activity. Thus, for a set of neuronal
the TCR neurons will be less hyperpolarized such that low- parameters, the model is in a “bistable regime” where it may
threshold oscillatory spikes will occur. This surrounding area generate both normal and paroxysmal oscillations and sponta-
will be the forefront of a wave of propagating seizure activity. neous transitions between these two types of behavior. This
The model shows that the complex nonlinear dynamics of the model facilitated making a number of important predictions
neuronal network depends critically on the low-voltage-acti- that can be tested experimentally, both in animals and in
vated (LVA) Ca-current IT. In this context it is interesting to
note that in the genetic model of absence epilepsy GAERS
(genetic absence epilepsy rats of Strasbourg), the presence of a
significantly enhanced LVA current in thalamic neurons was
demonstrated by Tsakiridou et al. (130). It is also noteworthy
that drugs, like ethosuximide, that depress IT should suppress
seizure initiation. It should be noted, however, that more recent
studies showed that ethosuximide acts also on the noninactivat-
ing Na current and on a Ca2 -activated K current (131). The
level of hyperpolarization of TCR cells, which appears to be a
crucial determinant of the dynamic behavior of the neuronal
population, depends on synaptic mechanisms. The model of
Destexhe et al. (69) goes further than previous models and
shows that the transition from 9–11 Hz to 3 Hz oscillations
appears to depend on GABAB synapses. Relatively weak RE
activity would mainly elicit GABAA responses, whereas stronger
activity would recruit GABAB synapses and this would cause the
quiescent mode of the TCR neurons. Indeed, GABAB presynap-
tic inhibition is conspicuous between RE neurons, such that an
enhancement of the latter would result in a decrease of intra-RE Figure 4.19 Upper panel: 20 seconds of a simulation with the occur-
inhibition and, thus in an increase of inhibition of TCR neu- rence of a spontaneous paroxysmal episode. Lower panels: Average
rons (quiescent state). Thus GABAergic drugs would aggravate power spectra of signals simulated by the model and recorded from a
the tendency for seizure oscillations to occur. WAG/ Rij rat with absence seizures. Left-hand side: The spectra of nor-
The models discussed above have given insight into some mal activity. Right-hand side: The spectra of paroxysmal activity.
basic neuronal mechanisms of SW discharges but do not (Adapted from Suffczynski P, Kalitzin S, Lopes da Silva FH. Dynamics
address specifically the most essential issue of this type of of non-convulsive epileptic phenomena modeled by a bistable neuronal
epileptic activity—that a given thalamocortical loop can display network. Neuroscience. 2004;126(2):467–484.)
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 83

humans: (a) The transitions between the normal and the parox- nectivity function among chains of cortical neurons; although
ysmal states can emerge spontaneously and are not induced these values are in the same order of magnitude, many factors
necessarily by any parameter changes. (b) Probabilities of tran- may account for the difference between them, such as the dif-
sitions between normal neuronal activity and paroxysmal oscil- ferences in anatomical connectivity between human and
lations depend on a number of model parameters; therefore, canine’s cortex.
these probabilities can be controlled in a nonunique way; in
more general terms these probabilities may be described by MODELS OF EEG GENERATION INVOLVING
gamma distributions, of which the exponential distribution is a CORTICAL AND SUBCORTICAL AREAS
special case. Interestingly, exponential shape of histograms of
durations of ictal events has also been found in other types of The models presented in the previous section describe only a
human epilepsy. (c) Paroxysmal oscillations can be annihilated rather small number of EEG phenomena. There are many EEG
by a well-timed pulse, such that, under given conditions, this patterns that have not yet been modeled in terms of sets of neu-
kind of seizures may be aborted. The model facilitates investi- ronal networks although some NMMs have been shown to gen-
gating these conditions such that the possibility of aborting erate EEG-like oscillations at different frequencies. Moreover,
seizures may be tested experimentally. Another model of most models have been limited to simulations of the behavior
absence seizures gives special attention to the time delays of of local neural networks. This chapter has already considered a
thalamocortical and corticothalamic pathways on the frequency model that involved a KIII set including networks of different
of SW discharges (133). Although these time delays likely con- areas. However, several EEG phenomena have started to be ana-
tribute to the frequency of the seizure oscillations, it is difficult lyzed in these terms. For example, sleep EEG patterns are
to deduce from this model what is the role of these delays in this known in some detail in terms of cellular processes in a num-
process, since this model does not include other important cel- ber of brainstem and cortical regions (136–138; see also
lular elements that may also affect the oscillation frequency, Chapter 3), and a model describing many essential features of
such as GABA synaptic processes with different time constants slow-wave sleep and activated states in the thalamocortical sys-
and calcium T-channels with particular kinetics. Further it does tem as well as the transition between them was described (139).
not account for nonstationary changes in apparent time delays Specific models that address the question of how pathological
between cortex and thalamus that have been shown experimen- slow oscillations (140) may be generated have not yet been
tally to occur during one episode of SW discharge (134). For a developed. It is to be expected that efforts in this direction will
comprehensive review of computer modeling in epilepsy the lead to a better understanding of such EEG phenomena by
reader is referred to Lytton (135). combining physiological data at the cellular level with mathe-
matical modeling.
LARGE-SCALE THALAMOCORTICAL MODELS Another important theoretical aspect of the generation of
EEG phenomena is the question of relating the temporal and
As stated in the introductory remarks brain rhythms and EEG- spatial aspects of such phenomena. Most models are concerned
like phenomena have also been investigated in large-scale mod- with local networks, but not with the question of the coupling
els including detailed neuronal properties by the group of between relatively distant sets of neurons (e.g., between the
Edelman et al. In the context of the present discussion it is par- thalamus and the cortex). However, a model aimed at analyzing
ticularly interesting to note that in a large-scale thalamocortical such a problem concerning the spatial organization of alpha
model rhythmic activities can emerge characterized by frequen- rhythms has been introduced by Lopes da Silva et al. (90,91). A
cies typical of EEG/MEG rhythms, such as delta, alpha, beta, number of experimental studies of alpha rhythm in the dog
and gamma, although these activities were not explicitly built in have revealed that intracortical coherences are in general larger
the neuronal elements (4). than any thalamocortical coherence measured in the same ani-
In this model it was shown that rhythms at specific fre- mal. This led investigators to question the amount of influence
quencies occur in different cortical areas, for example, a beta of thalamic sources on the large intracortical coherences. This
rhythm at about 20 Hz in areas corresponding to motor and amounts to choosing between two models: either thalamocor-
somatomotor cortex, while the microcircuitry of different tical relations are determined by point-to-point projections or
cortical areas was the same. Therefore these authors conclude each thalamic population may project to several cortical areas.
that these differences in frequency of oscillations should be Using partial coherence analysis (91), it has been shown that a
determined by differences in white-matter anatomical con- model based on point-to-point projections cannot be accepted
nectivity (the latter was modeled based on diffusion tensor to explain the organization of thalamocortical alpha rhythms.
imaging data of the human brain) between and within corti- Furthermore, it was found that there are widespread influences
cal areas. Multiple propagating waves were also generated by of populations of the dorsal thalamus on intracortical coher-
the model within limited cortical regions with typical propa- ences. Notwithstanding these influences, the residual intracor-
gation velocities around 0.1 m/sec. We noted above that the tical coherences were so large that other factors must be
propagation speed of alpha waves on the surface of the dog’s assumed to explain this intracortical coherent activity; these
cortex was in the range of 0.3 m/sec also within limited corti- factors may have as an anatomical substrate inter- and intra-
cal areas, and the model of van Rotterdam et al. (20) revealed cortical systems of fibers responsible for the spatial distribution
that this propagation is mainly determined by the intercon- of alpha rhythms within the cortex. Indeed, if such systems did
84 Part I ■ Basic Principles

not exist, investigators would be left with a patchy nonoverlap- Beta rhythms with a different frequency (~28 Hz) have been
ping assembly of cortical modules with little intracortical shown to be rather localized to occipital areas in dogs engaged
coherence; it would be almost impossible in such a case to in a visual discrimination task (146). Particularly in the visual
record an EEG at a relatively large distance from the cortex. cortex, oscillations in the gamma frequency range ( 30 Hz) at
A similar problem has been mathematically elaborated by the single neuronal level (147) have been recorded (reviewed in
Nunez (141), who developed a model, also based on the exis- Ref. 148). In this way, synchronization occurs between neurons
tence of interconnections between different cortical modules, that represent related features of an object that should be inte-
to account for the global wave pattern of the EEG at the scalp. grated for the generation of a coherent visual percept.
In this way, Nunez takes into account that traveling EEG waves Furthermore, it was shown that neurons in other areas of the
may coalesce to form standing waves because, he assumes, the cerebral cortex may also synchronize their activity with those of
cortex forms a closed surface. In his view, there is a fixed rela- the visual cortex, remarkably without time lags; this synchrony
tionship between the nondimensional temporal and spatial fre- was particularly strong between areas subserving related func-
quency of EEG given by the dispersion relation. This relation tions (15).
relates the temporal frequency with the characteristic velocity A large number of studies have demonstrated the existence
of propagation of activity in associative cortical fibers and the of high-frequency rhythmic activities in the human EEG under
so-called wave number. different circumstances, particularly in relation with motor and
Nunez (142) discusses the theory proposed by Lopes da Silva cognitive functions (149–152). A review of clinically relevant
et al. (91) and van Rotterdam et al. (20), which he calls the beta rhythms is given in Chapter 37 and of the influence of
“local theory” of the electrocorticogram, in connection with his drugs on these EEG activities in Chapter 43. Model studies of
“global theory,” which predicts traveling and standing waves these EEG fast rhythms are being developed. The demonstra-
with wavelengths up to the cortical circumference having phase tion of these high-frequency activities in animals, particularly
velocities in the 10 m/sec range. According to the local theory, in the hippocampus and associated areas, has inspired a num-
alpha activity spreads away from multiple epicenters with a pre- ber of model studies that try to account for the generation of
ferred wave number range for alpha activity and with phase these activities, integrating experimental observations and
velocities of about 0.3 m/sec. It is likely that a combined computer simulations. Such a model to account for the occur-
local/global theory is needed. Indeed, as Nunez points out rence of fast oscillations in a neuronal network was proposed by
(35,142), the simultaneous existence of both short- and long- Traub et al. (57) and Jefferys et al. (58,59). The model consists
wavelength phenomena in the same frequency range can occur. of a chain of interconnected excitatory pyramidal and
Some EEG phenomena may be dominated by short-range inhibitory interneurons. According to this model, when the
intracortical interactions, while others are dominated by long- excitatory drive on the interneurons reaches a level sufficient to
range corticocortical interactions. The human scalp EEG results induce pairs of action potentials at short intervals (“spike dou-
probably from both types of activity. In the combined local- blets”) the network generates oscillations in the gamma fre-
global theory of Nunez (35), the input to the cortex consists of quency range on a millisecond time scale from one end of the
the modulating afferent input from subcortical areas and inputs chain to the other. It should be noted, however, that neuronal
from other cortical regions, whether distant or neighboring, oscillations in a similar frequency range have been found in
through corticocortical fibers. In this model, it is assumed that other areas of the brain, mainly in olfactory bulb and cortex (1,
the local networks include inhibitory and excitatory feedback 33,153,154), but also in the hippocampus and other limbic
circuits, whereas the global corticocortical interactions are exci- areas (for review, see Ref. 155). These rhythmic activities have
tatory. Nunez and collaborators (143,144) proposed a theoreti- been modeled by means of cortical neuronal networks with
cal framework supporting experimental measures of dynamic recurrent inhibitory connections, and it was shown that such
properties of human EEG. According to this model both glob- models, with the relevant physiological parameters are capable
ally coherent and locally dominated behavior can occur within of accounting for the 20- to 50-Hz gamma rhythms (see discus-
the alpha band, depending on narrow band frequency, spatial sion on this topic in Refs. 58, 59, 155, and 156).
measurement scale, and brain state. At the same time, alpha and Based on experimental observations in brain slice prepara-
theta phase locking between cortical regions during certain tions of the hippocampus or the neocortex, in vitro, where
behavioral states can occur. oscillations are induced by electrical stimulation and/or by
exogenous drug application, a series of models was created as
MODELS OF CORTICAL BETA/ GAMMA reviewed by Whittington et al. (157). These models demon-
RHYTHMS strate that fast gamma rhythms can be generated by the mech-
anism of mutual inhibition in populations of interconnected
In the cortex fast rhythms have been identified, both in the EEG inhibitory interneurons, which can be studied independently by
and at the level of single neuronal activity, that are usually pharmacological manipulations. Despite the artificial condi-
called beta or gamma rhythms, depending on frequency range. tions of these experiments, these physiological data have led to
It has been found experimentally that fast frequency rhythms interesting theoretical generalizations (60), as for instance the
(33–45 Hz) occurring in the frontoparietal areas may occupy a possible role of electrical coupling via gap junctions in the gen-
restricted cortical area in the awake cat (145). These thalamo- eration of very fast oscillations both under normal and epilep-
cortical beta rhythms are characteristic of attentive immobility. tic conditions (158,159).
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 85

Recently, a generalized model that accounts for changes in freedom can display unpredictable behavior over long time
rhythmic activities with behavioral state, from the slower alpha scales; in other words, they can display what mathematicians
to the faster gamma and beta as arousal increases, was proposed call chaotic behavior (163). Such systems can have multiple sta-
(160). This model is especially interesting because the same set ble states governed by “equilibrium,” “limit cycle,” or “chaotic”
of underlying intrinsic and network mechanisms appears to be (“strange”) attractors. Whether the system will find itself in one
responsible for transitions between the different types of or another stable state depends on the input conditions and on
rhythms. The effects of neuromodulation on ionic conduc- the system’s parameters.
tances were investigated in a cortical circuit model, consisting These mathematical notions can be illustrated using the
of synaptically coupled excitatory and inhibitory neurons that example of the KIII set defined by Freeman (34). Such a set of
support rhythmic activity in the alpha, beta, and gamma neurons (Fig. 4.1) may have a least four equilibrium states.
ranges. It was shown that the effects of neuromodulation on There is a rest state, in which the nonlinear pulse-wave function
ionic conductances are, by themselves, sufficient to induce tran- is set to zero and corresponds to the state during deep anesthe-
sitions between synchronous gamma and beta rhythms and sia. Another is a state in which a near sinusoidal limit cycle
asynchronous alpha rhythms. occurs, such as is produced in the olfactory bulb and cortex by
It should be noted that despite extensive search, no evidence drugs and by hyperthermia (1). Still another state is character-
was found by Menon et al. (161) for globally correlated activity ized by recurrent limit cycle behavior in which the olfactory
in the human EEG, either in narrow (i.e., 35–45 Hz) or in broad bulb depends on the level of the normal respiratory rhythm. If
(i.e., 20–50 Hz) frequency bands. Instead, spatially and tempo- slightly structurally changed, a KIII set can be led to another
rally intermittent synchronization was observed between pairs state in which chaotic activity is generated. In such a case, the
of electrodes in 1-cm 2 regions with high variability. The distri- output consists of long stretches, or bursts, with unpredictable
bution of correlation coefficients differed substantially from amplitudes and phases that resemble the fluctuations of the
background levels at interelectrode distances of 1 and 1.4 cm EEG at rest. A similar behavior can be obtained by giving a low-
but not 2 cm or more. These findings suggest that the surface level random noise input to the KII set. This fact illustrates a
diameters of domains of spatially correlated activity within the general question with which one is faced when modeling the
gamma band in human EEG are limited to less than 2 cm. Thus, EEG: Is the ongoing EEG a manifestation of a chaotic attractor
if such gamma band spatial patterns exist in the human brain, or of an equilibrium state under random perturbation? The
no existing technology would be capable of measuring them at mathematical techniques with which to characterize chaotic
the scalp, and subdural electrode arrays for cortical surface behavior of nonlinear dynamic systems and to differentiate it
recording would have to have spacings under 5 mm. This means from random activity are described in detail in a number of
that on the cortex there are well-characterized spatial domains specialized publications (163–167). Furthermore, the problems
of rhythmic activities that do not spread all over the cortical involved in applying these mathematical tools of nonlinear
surface. dynamical analysis to EEG signals are discussed in a number of
publications (168–174).
RELEVANCE OF NONLINEAR DYNAMIC In general, one can state that it has been difficult to find
SYSTEMS FOR EEG GENERATION unambiguous evidence for the existence of nonlinear determin-
ism in EEG signals except for those recorded during epileptic
The models of neuronal networks presented above possess non- seizures (175). The latter observations have led to the concept
linear wave-pulse functions. In a number of studies of such net- that a given neuronal network may present essentially different
works, a simplification has been introduced; it was assumed modes of activity, that is, it may display different dynamics,
that this nonlinear characteristic could be approximated by a depending on specific conditions. For some values of control
linearized version, at least under those conditions in which the parameters, a neuronal network may change its dynamical
input signals are limited to a small range of amplitude. mode of activity; for example, instead of a random noisy state,
However, such simplification is not necessary, and it may lead a limit cycle or even a chaotic attractor may occur. Under these
one to ignore essential nonlinear properties of the system, such circumstances we say that a bifurcation has taken place. Thus a
as the generation of higher harmonics (Fig. 4.12). Indeed, it is bifurcation represents a qualitative change and depends on a set
possible to account for the behavior of neuronal networks of control parameters that define the operating regimen of the
assuming that these networks can be described by a number of system. We hypothesized (127) that the main difference
coupled nonlinear differential equations as a function of time between a normal and an epileptic brain is that the operating
and space. In general, such a set of nonlinear differential equa- regimen of a neuronal network in the epileptic brain is much
tions may be considered to provide the basis for understanding closer to a bifurcation point than in the normal brain, leading
the generation of all types of EEG activity. to a low-dimensional chaotic mode of behavior. This means
A number of investigators, such as Freeman (34,178) and that compared to the normal brain, where the operating regi-
Babloyantz (162), started to explore the possibilities given by men is situated far from such a bifurcation point, in an epilep-
the modern mathematical field of research of nonlinear tic neuronal network the distance between operating and
dynamic systems in order to obtain a better understanding of bifurcation points is so small that the system may easily switch
EEG phenomena. A relevant aspect for this discussion is that all from a stable equilibrium to a chaotic attractor, even in
nonlinear dynamic systems with more than two degrees of response to a very weak stimulus. The latter may pass undetected.
86 Part I ■ Basic Principles

Accordingly, in the case of epilepsy the basic question is which 10. Lopes da Silva FH, Hoeks A, Smits H, et al. Model of brain rhyth-
factors are responsible for the change in the operating point of mic activity. Kybernetik. 1974;15:27–37.
the involved neuronal networks (176). It should be added that 11. Lopes da Silva FH, van Rotterdam A, Barts P, et al. Models of neu-
this nonlinear dynamical behavior, which we formulate here at ronal populations: the basic mechanisms of rhythmicity. In:
Corner MA, Swaab DF, eds. Perspectives of Brain Research, vol. 45.
the macroscopic level, has also been identified at the neuronal
Prog Brain Res. 1976:281–308.
level, since near the transition from subthreshold to bursting
12. Jansen BH, Rit VG. Electroencephalogram and visual evoked
oscillations in thalamic neurons an apparently chaotic oscilla- potential generation in a mathematical model of coupled cortical
tion was observed. This was made evident in the model studies columns. Biol Cybern. 1995;73:357–366.
of Wang (44,48). 13. Suffczynski P, Kalitizin S, Pfurtscheller G, et al. Computational model
of thalamocortical networks: dynamical control of alpha rhythms in
relation to focal attention. Int J Psychophysiol. 2001;43(1):25–40.
CONCLUSION 14. David O, Friston KJ. A neural mass model for MEG/EEG: coupling
and neuronal dynamics. NeuroImage. 2003;20:1743–1755.
The complexity of EEG phenomena calls for the use of mathe-
15. Roelfsema PR, Engel AK, Koenig P, et al. Visuomotor integration is
matical models and computer simulations in order to under- associated with zero timelag synchronization among cortical areas.
stand the underlying processes of generation. In the last few Nature. 1997;385:157–161.
years, a number of problems in electroencephalography have 16. Chawla D, Friston KJ, Lumer ED. Zero-lag synchronous dynamics
been tackled in this way. Such theoretical studies are justified if in triplets of interconnected cortical areas. Neural Netw.
they are combined with experimental investigations. In this 2001;14:727–735.
way, one may not only obtain new insights about the generation 17. Zavaglia M, Astolfi L, Babiloni F, et al. A neural mass model for the
of EEG patterns, but also formulate hypotheses to be tested simulation of cortical activity estimated from high resolution EEG
under experimental conditions. during cognitive or motor tasks. J Neurosci Methods. 2006;
In the course of the last decades, a shift of attention from 157(2):317–329.
18. Zavaglia M, Astolfi L, Babiloni F, et al. The effect of connectivity on
models describing the behavior of local neuronal networks in
EEG rhythms, power spectral density and coherence among cou-
the temporal domain toward models that take into account the
pled neural populations: analysis with a neural mass model. IEEE
spatial properties of complex networks has occurred. Trans Biomed Eng. 2008;55(1):69–77.
Furthermore, an evolution can be seen in the techniques used. 19. Wendling F, Bartolomei F, Bellanger JJ, et al. Epileptic fast activity
In the early phase, the approach was based mainly on the use of can be explained by a model of impaired GABAergic dendritic
linear systems analysis, but more recently new mathematical inhibition. Eur J Neurosci. 2002;15(9):1499–1508.
tools for the analysis of nonlinear dynamic systems have been 20. van Rotterdam A, Lopes da Silva FH, Van den Ende J, et al. A model
introduced in this domain of theoretical neurophysiology. of the spatial-temporal characteristics of the alpha rhythm. Bull
Moreover, the availability of powerful computer tools opens Math Biol. 1982;44(2):283–305.
new possibilities for modeling both complex membrane phe- 21. Lopes da Silva FH, Storm van Leeuwen W. The cortical alpha
nomena and network properties. The analysis of the dynamics rhythm in dog: the depth and surface profile of phase. In: Brazier
MAB, Petsche H, eds. Architectonics of the Cerebral Cortex. New
of such networks is a challenge for the theoretical neurophysi-
York: Raven Press; 1978:319–333.
ologist interested in understanding EEG phenomena.
22. Harrison LM, David O, Friston KJ. Stochastic models of neuronal
dynamics. Phil Trans Biol Sci. 2005;360(1457):1075–1091.
23. Deco G, Jirsa VK, Robinson PA, et al. The dynamic brain: from
spiking neurons to neural masses and cortical fields. PLoS Comput.
1. Freeman WJ. Mass Action in the Nervous System. New York: Biol. 2008;4(8):e1000092. doi:10.1371/journal.pcbi.1000092.
Academic Press; 1975. 24. Jirsa VK. Neural field dynamics with local and global connectivity
2. Hammond PH. Feedback Theory and Its Applications. London: and time delay. Philos Transact. Series A, Math Phys Eng Sci.
English University Press; 1958. 2009;367(1891):1131–1143.
3. Harmon LD. Problems in neural modeling. In: Reiss RF, ed. Neural 25. Jirsa VK, Haken H. Field theory of electromagnetic brain activity.
Theory and Modeling, Proceedings of the 1962 OGAI Symposium. Phys Rev Lett. 1996;77:960–963.
Stanford, CA: Stanford University Press; 1964. 26. Wright JJ, Robinson PA, Rennie CJ, et al. Toward an integrated
4. Izhikevich EM, Edelman GM. Large-scale model of mammalian continuum model of cerebral dynamics: the cerebral rhythms,
thalamocortical systems. Proc Natl Acad Sci. 2008;105(9): synchronous oscillation and cortical stability. Biosystems.
3593–3598. 2001;63(1–3):71–88.
5. Markram H. The blue brain project. Nat Rev Neurosci. 2006;7(2): 27. Wright JJ, Rennie CJ, Lees GJ, et al. Simulated electrocortical activity
153–160. at microscopic, mesoscopic, and global scales. Neuropsychopharma-
6. Watts DJ, Strogatz SH. Collective dynamics of “small-world” net- cology. 2003;28(suppl 1):S80–SA93.
works. Nature. 1998;393(6684):440–442. 28. Rennie CJ, Robinson PA, Wright JJ. Unified neurophysical model
7. Yu S, Huang D, Singer W, et al. A small world of neuronal syn- of EEG spectra and evoked potentials. Biol Cybern. 2002;86(6):
chrony. Cereb Cortex. 2008;18(12):2891–2901. 457–471.
8. Stam CJ, Reijneveld JC. Graph theoretical analysis of complex net- 29. Robinson PA, Whitehouse RW, Rennie CJ. Nonuniform corti-
works in the brain. Nonlinear Biomed Phys. 2007;1(1):3. cothalamic continuum model of electroencephalographic spectra
9. Wilson HR, Cowan JD. Excitatory and inhibitory interaction in with application to split-alpha peaks. Phys Rev E Stat Nonlin Soft
localized populations of model neurons. Biophys J. 1972;12:1–23. Matter Phys. 2003;68(2 pt 1):021922.
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 87

30. Friston KJ, Harrison L, Penny W. Dynamic causal modeling. 52. Traub RD. Simulation of intrinsic bursting in CA3 hippocampal
NeuroImage. 2003;19:1273–1302. neurons. Neuroscience. 1982;7:1233–1242.
31. Stephan KE, Harrison LM, Kiebel SJ, et al. Dynamic causal models 53. Traub RD, Wong RKS. Synaptic mechanisms underlying interictal
of neural system dynamics: current state and future extensions. spike initiation in a hippocampal network. Neurology. 1983;33:
J Biosci. 2007;32(1):129–144. 258–266.
32. Valdes-Sosa PA, Sanchez-Bornot JM, Sotero RC, et al. Model 54. Traub RD, Knowles WD, Miles R, et al. Synchronized afterdis-
driven EEG/fMRI fusion of brain oscillations. Hum Brain Mapp. charges in the hippocampus: simulation studies of the cellular
2009;30(9):2701–2721. mechanism. Neuroscience. 1984;12:1191–1200.
33. Freeman WJ. Nonlinear dynamics of paleocortex manifested in the 55. Traub RD, Draguhn A, Whittington MA, et al. Axonal gap junc-
olfactory EEG. Biol Cybern. 1979;35:21–34. tions between principal neurons: a novel source of network oscil-
34. Freeman WJ. Analytic techniques used in the search for the physi- lations, and perhaps epileptogenesis. Rev Neurosci. 2002;13
ological basis of the EEG. In: Gevins A, Remond A, eds. Handbook (1):1–30.
of Electroencephalography and Clinical Neurophysiology, vol. 1: 56. Traub RD, Miles R. Neuronal Networks of the Hippocampus.
Methods of Analysis of Brain Electrical and Magnetic Signals. Cambridge: Cambridge University Press; 1991.
Amsterdam: Elsevier; 1986. 57. Traub RD, Whitington MA, Colling SB, et al. Analysis of gamma
35. Nunez PL, ed. Neocortical Dynamics and Human EEG Rhythms. rhythms in the rat hippocampus in vitro and in vivo. J Physiol
New York: Oxford University Press; 1995. (Lond.) 1996;493:471–484.
36. Wright JJ, Liley DTJ. Dynamics of the brain at global and micro- 58. Jefferys JG, Traub RD, Whittington MA. Neuronal networks for
scopic scales: neural networks and the EEG. Behav Brain Sci. induced “40 Hz” rhythms. Trends Neurosci. 1996;19:202–208.
1996;19:285–320. 59. Jefferys JG, Traub RD, Whittington MA. Reply. Trends Neurosci.
37. Stam CJ, Pijn JP, Suffczynski P, et al. Dynamics of the human alpha 1996;19:469–470.
rhythm: evidence for non-linearity? Clin Neurophysiol. 1999; 60. Whittington M, Traub RD. Inhibitory interneurons and network
110(10):1801–1813. oscillations in vitro. Trends Neurosci. 2003;26(12):676–682.
38. Valdes PA, Jimenez JC, Riera J, et al. Nonlinear EEG analysis based 61. Freeman WJ. Linear analysis of the dynamics of neural masses.
on a neural mass model. Biol Cybern. 1999;81(5–6):415–424. Annu Rev Biophys Bioeng. 1972;1:225–256.
39. Robinson PA, Loxley PN, O’Connor SC, et al. Modal analysis of 62. Gray CM, Freeman WJ, Skinner JE. Associative changes in the spa-
corticothalamic dynamics, electroencephalographic spectra, and tial amplitude patterns of rabbit olfactory EEG are norepinephrine
evoked potentials. Phys Rev E Stat Nonlin Soft Matter Phys. dependent. Neurosci Abstr. 1984;10(36,2):121.
2001;63(4 pt 1):041909. 63. Freeman WJ. A proposed name for aperiodic brain activity: sto-
40. Golomb D, Rinzel J. Dynamics of globally coupled inhibitory neu- chastic chaos. Neural Netw. 2000;13(1):11–13.
rons with heterogeneity. Physiol Rev E. 1993;48:4810–4814. 64. Andersen P, Andersson SA. Physiological Basis of Alpha Rhythm.
41. Golomb D, Rinzel J. Clustering in globally coupled inhibitory neu- New York: Appleton-Century-Crofts; 1968.
rons. Physica D. 1994;72:259–282. 65. Andersen P, Andersson SA, Lomo T. Nature of thalamocortical
42. Golomb D, Wang XJ, Rinzel J. Synchronization properties of spin- relations during spontaneous barbiturate spindle activity. J
dle oscillations in a thalamic reticular nucleus model. Physiol. 1967;192:283–307.
J Neurophysiol. 1994;72:1109–1126. 66. Steriade M, Jones EG, Llinás RR. Thalamic Oscillations and
43. Golomb D, Wang XJ, Rinzel J. Propagation of spindle waves in a Signaling. Neuroscience Institute Publications. New York: John
thalamic slice model. J Neurophysiol. 1996;75:750–769. Wiley & Sons; 1990.
44. Wang XJ. Multiple dynamical modes of thalamic relay neurons: 67. Steriade M, Gloor P, Llinás RR, et al. Report of the IFCN commit-
rhythmic bursting and intermittent phase-locking. Neuroscience. tee on basic mechanisms. Electroencephalogr Clin Neurophysiol.
1994;59:21–31. 1990;76:481–509.
45. Wang XJ, Rinzel J. Alternating and synchronous rhythms in recip- 68. Destexhe A, Contreras D, Sejnowski TJ, et al. Modeling the control
rocally inhibitory model neurons. Neural Comp. 1992;4:84–97. of reticular thalamic oscillations by neuromodulators. Neuroreport.
46. Wang XJ, Rinzel J. Spindle rhythmicity in the reticularis thalami 1994;5:2217–2220.
nucleus: synchronization among mutually inhibitory neurons. 69. Destexhe A, Bal T, McCormick DA, et al. Ionic mechanisms under-
Neuroscience. 1993;53:899–904. lying synchronized oscillations and propagating waves in a model
47. Wang XJ, Rinzel J, Rogawski MA. A model of the T-type calcium of ferret thalamic slices. J Neurophysiol. 1996;76:2049–2070.
current and the low threshold spike in thalamic neurons. 70. Jahnsen H, Llinás R. Electrophysiological properties of guinea-pig
J Neurophysiol. 1991;66:839–850. thalamic neurones: an in vitro study. J Physiol (Lond.) 1984;349:
48. Wang XJ, Golomb D, Rinzel J. Emergent spindle oscillations and 205–226.
intermittent burst firing in a thalamic model: specific neuronal 71. Jahnsen H, Llinás R. Ionic basis for the electro-responsiveness and
mechanisms. Proc Natl Acad Sci USA. 1995;92:5577–5581. oscillatory properties of guinea-pig thalamic neurons in vitro.
49. Suffczynski P, Kalitzin S, Pfurtscheller G, et al. Computational J Physiol (Lond.) 1984;349:227–247.
model of thalamo-cortical networks: dynamical control of alpha 72. Steriade M, Deschènes M, Domich L, et al. Abolition of spindle
rhythms in relation to focal attention. Int J Psychophysiol. 2001; oscillations in thalamic neurons disconnected from nucleus retic-
43(1):25–40. ularis thalami. J Neurophysiol. 1985;54:1473–1497.
50. Suffczynski P, Kalitzin S, Lopes da Silva FH. Dynamics of non-con- 73. Avanzini G, de Curtis M, Panzica F, et al. Intrinsic properties of
vulsive epileptic phenomena modeled by a bistable neuronal net- nucleus reticularis thalami neurons of the rat studied in vitro.
work. Neuroscience. 2004;126(2):467–484. J Physiol. 1989;416:111–122.
51. Wendling F, Hernandez A, Bellanger JJ, et al. Interictal to ictal transi- 74. Bal T, von Krosigk M, McCormick DA. Synaptic and membrane
tion in human temporal lobe epilepsy: insights from a computational mechanisms underlying synchronized oscillations in the ferret
model of intracerebral EEG. J Clin Neurophysiol. 2005;22(5):343–56. LGNd in vitro. J Physiol (Lond.) 1995;483:641–663.
88 Part I ■ Basic Principles

75. Bal T, von Krosigk M, McCormick DA. Role of the ferret perigenic- 95. Szentagothai J. The local neuronal apparatus of the cerebral
ulate nucleus in the generation of synchronized oscillations in cortex. In: Buser P, Rougeaul-Buser A, eds. Cerebral Correlates of
vitro. J Physiol (Lond.) 1995;483:665–685. Conscious Experience, Inserm Symposium no. 6. Amsterdam:
76. Huguenard JR, Prince DA. Intrathalamic rhythmicity studied in Elsevier; 1978:131–138.
vitro: nominal T-current modulations causes robust anti-oscillatory 96. Katchalsky AK, Rowland V, Blumenthal R. Dynamic patterns of
effects. J Neurosci. 1994;14:5485–5502. brain cell assemblies. Neurosci Res Prog Bull. 1974;12(1):1–187.
77. Von Krosigk M, Bal T, McCormick DA. Cellular mechanisms of 97. Tombol T. Comparative data on the Golgi architecture of
a synchronized oscillation in the thalamus. Science. 1993;261: interneurons of different conical areas in cat and rabbit. In:
361–364. Brazier MAB, Petsche H, eds. Architectonics of the Cerebral Cortex.
78. Warren RA, Agmon A, Jones EG. Oscillatory synaptic interactions New York: Raven Press; 1978:59–76.
between ventroposterior and reticular neurons in mouse thalamus 98. Liley DTJ. A continuum model of the mammalian alpha rhythm.
in vitro. J Neurophysiol. 1994;72:1993–2003. In: Lopes da Silva F, Cadusch JC, Borges de Almeida L, eds.
79. Landisman CE, Long MA, Beierlein M, et al. Electrical synapses in Spatiotemporal Models in Biological and Artificial Systems.
the thalamic reticular nucleus. J Neurosci. 2002;22:1002–1009. Amsterdam: IOS Press; 1997:89–96.
80. Steriade M, Llinás RR. The functional states of the thalamus and 99. Liley DTJ, Cadusch PC, Wright JJ. A continuum theory of elec-
the associated neuronal interplay. Physiol Rev. 1988;68:649–742. tro-cortical activity. Neurocomputing. 1999;26–27:795–800.
81. McCormick DA. Neurotransmitter actions in the thalamus and 100. Liley DTJ, Cadusch PJ, Dafilis MP. A spatially continuous mean
cerebral cortex and their role in neuromodulation of thalamocor- field theory of electrocortical activity. Network Comp Neural Sys.
tical activity. Prog Neurobiol. 1992;39:337–388. 2002;13:67–113.
82. McCormick DA, Bal T. Sensory gating mechanisms of the thala- 101. Contreras D, Steriade M. Spindle oscillation in cats: the role of
mus. Curr Opin Neurobiol. 1994;4:550–556. corticothalamic feedback in a thalamically generated rhythm. J
83. Destexhe A, Sejnowski TJ. Interactions between membrane con- Physiol (Lond.) 1996;490:159–179.
ductances underlying thalamocortical slow-wave oscillations. 102. Zetterberg LH, Kristiansson L, Mossberg K. Performance of a
Physiol Rev. 2003;83(4):1401–1453. model for a local neuron population. Biol Cybern. 1978;31:15–26.
84. Hughes SW, Crunelli V. Thalamic mechanisms of EEG alpha 103. Babloyantz A, Kaczmarek LK. Self-organization in biological sys-
rhythms and their pathological implications. Neuroscientist. 2005; tems with multiple cellular contacts. Bull Math Biol. 1979;41:
11(4):357–372. 193–201.
85. Lörincz ML, Crunelli V, Hughes SW. Cellular dynamics of cholin- 104. Traub RD, Wong RKS. Synchronized burst discharge in disinhib-
ergically induced alpha (8–13 Hz) rhythms in sensory thalamic ited hippocampal slice. II. Model of cellular mechanism. J
nuclei in vitro. J Neurosci. 2008;28(3):660–671. Neurophysiol. 1983;49:442–458.
86. Lopes da Silva FH, van Lierop THMT, Schrijer CF, et al. Organization 105. Traub RD, Wong RKS. Cellular mechanism of neuronal synchro-
of the thalamic and cortical alpha rhythms: spectra and coherences. nization in epilepsy. Science. 1982;216:745–747.
Electroencephalogr Clin Neurophysiol. 1973;35:627–639. 106. Traub RD, Jefferys JG, Whittington MA. Functionally relevant
87. Chatile M, Milleret C, Buser P, et al. A 10 Hz “alpha-like” rhythm and functionally disruptive (epileptic) synchronized oscillations
in the visual cortex of the waking cat. Electroencephalogr Clin in brain slices. Adv Neurol. 1999;79:709–724.
Neurophysiol. 1992;83:217–230. 107. Traub RD, Bibbig A, LeBeau FE, et al. Cellular mechanisms of
88. Lanoir J. Étude électrocorticographique de la veille et du sommeil neuronal population oscillations in the hippocampus in vitro.
chez le chat, organisation du cycle nycthéméral, rôle du thalamus. Annu Rev Neurosci. 2004;27:247–278.
Thesis Doctor of Science, Centre Régional de recherche et de doc- 108. Trevelyan AJ, Sussillo D, Watson BO, et al. Modular propagation
umentation Pédagogiques, Marseille, France, 1972. of epileptiform activity: evidence for an inhibitory veto in neo-
89. Zetterberg LH. Stochastic activity in a population of neurons—A cortex. J Neurosci. 2006;26(48):12447–12455.
system analysis approach. Rep Inst Med Physics, TNO, Utrecht 109. Trevelyan AJ, Sussillo D, Yuste R. Feedforward inhibition con-
1973;1:53. tributes to the control of epileptiform propagation speed. J
90. Lopes da Silva FH, Vos JE, Mooibroek J, et al. Partial coherence Neurosci. 2007;27(13):3383–3387.
analysis of thalamic and cortical alpha rhythms in dog—a contri- 110. Kager H, Wadman WJ, Somjen GG. Simulated seizures and spread-
bution towards a general model of the cortical organization of ing depression in a neuron model incorporating interstitial space
rhythmic activity. In: Pfurtscheller G, Buser P, Lopes da Silva FH, and ion concentrations. J Neurophysiol. 2000;84(1):495–512.
eds. Rhythmic EEG Activities and Cortical Functioning. Amsterdam: 111. Kager H, Wadman WJ, Somjen GG. Conditions for the triggering
Elsevier; 1980:33–59. of spreading depression studied with computer simulations. J
91. Lopes da Silva FH, Vos JE, Mooibroek J, et al. Relative contribu- Neurophysiol. 2002;88(5):2700–2712.
tions of intracortical and thalamocortical processes in the genera- 112. Kager H, Wadman WJ, Somjen GG. Seizure-like afterdischarges
tion of alpha rhythms, revealed by partial coherence analysis. simulated in a model neuron. J Comput Neurosci. 2007;22(2):
Electroencephalogr Clin Neurophysiol. 1980;50:449–456. 105–128.
92. Braitenberg V. On the Texture of Brains. New York: Springer-Verlag; 113. Borck C, Jefferys JGR. Seizure-like events in disinhibited ventral
1977. slices of adult rat hippocampus. J Neurophysiol. 1999;82:
93. Chow KL, Leimann AL. Aspects of the structural and functional 2130–2142.
organization of the neocortex. Neurosci Res Prog Bull. 1970;8(2): 114. Allen PJ, Fish DR, Smith SJ. Very high-frequency rhythmic activ-
157–183. ity during SEEG suppression in frontal lobe epilepsy.
94. Szentagothai J. Specificity versus (quasi) randomness in cortical Electroencephalogr Clin Neurophysiol. 1992;82(2):155–162.
connectivity. In: Brazier MAB, Petsche H, eds. Architectonics of the 115. Bragin A, Engel J Jr, Wilson CL, et al. Hippocampal and entorhinal
Cortex. New York: Raven Press; 1978:77–97. cortex high-frequency oscillations (100–500 Hz) in human epilep-
Chapter 4 ■ Dynamics of EEGs as Signals of Neuronal Populations: Models and Theoretical Considerations 89

tic brain and in kainic acid-treated rats with chronic seizures. 133. Roberts JA, Robinson PA. Modeling absence seizure dynamics: impli-
Epilepsia. 1999;40(2): 127–137. cations for basic mechanisms and measurement of thalamocortical
116. White JA, Banks MI, Pearce RA, et al. Networks of interneurons and corticothalamic latencies. J Theor Biol. 2008;253:180–201.
with fast and slow gamma-aminobutyric acid type A (GABA-A) 134. Meeren HK, Pijn JP, Van Luijtelaar EL, et al. Cortical focus drives
kinetics provide substrate for mixed gamma-theta rhythm. Proc widespread corticothalamic networks during spontaneous
Natl Acad Sci USA. 2000;97:8128–8133. absence seizures in rats. J Neurosci. 2002;22(4):1480–1495.
117. Banks MI, White JA, Pearce RA. Interactions between distinct 135. Lytton WW. Computer modelling of epilepsy. Nat Rev Neurosci.
GABA(A) circuits in hippocampus. Neuron. 2000;25(2): 2008;9:626–637.
449–457. 136. Steriade M. Mechanisms underlying cortical activation: neuronal
118. Cossart R, Dinocourt C, Hirsch JC, et al. Dendritic but not organization and properties of the midbrain reticular core and
somatic GABAergic inhibition is decreased in experimental intralaminar thalamic nuclei. In: Pompeiano O, Ajmone-Marsan
epilepsy. Nat Neurosci. 2001;4(1):52–62. C, eds. Brain Mechanisms and Perceptual Awareness. New York:
119. Kramer MA, Kirsch HE, Szeri AJ. Pathological pattern formation Raven Press; 1981:327–377.
and cortical propagation of epileptic seizures. J R Soc Interface. 137. Steriade M, Nunez A, Amzica F. A novel slow ( 1 Hz) oscillation
2005;2:113–127. of neocortical neurons in vivo: depolarizing and hyperpolarizing
120. Kramer MA, Lopour BA, Kirsch HE, et al. Bifurcation control of components. J Neurosci. 1993;13:3252–3265.
a seizing human cortex. Phys Rev E Stat Nonlin Soft Matter Phys. 138. Steriade M, Nunez A, Amzica F. Intracellular analysis of relations
2006;73(4 pt 1):041928. between the slow ( 1 Hz) neocortical oscillation and other sleep
121. Kramer MA, Szeri AJ, Sleigh JW, et al. Mechanisms of seizure rhythms of the electroencephalogram. Neuroscience. 1993;13:
propagation in a cortical model. J Comput Neurosci. 2007;22(1): 3266–3283.
63–80. 139. Bazhenov M. Timofeev I, Steriade M, et al. Model of thalamocor-
122. Lytton WW, Hellman KM, Sutula TP. Computer models of hip- tical slow-wave sleep oscillations and transitions to activated
pocampal circuit changes of the kindling model of epilepsy. Artif states. J Neurosci. 2002;22(19):8691–8704.
Intell Med. 1998;13(1–2):81–97. 140. Ball GJ, Gloor R, Schaul N. The cortical electromicrophysiology
123. Santhakumar V, Aradi I, Soltesz I. Role of mossy fiber sprouting of pathological delta waves in the electroencephalogram of cats.
and mossy cell loss in hyperexcitability: a network model of the Electroencephalogr Clin Neurophysiol. 1977;43:346–361.
dentate gyrus incorporating cell types and axonal topography. J 141. Nunez PL. Electric Field of the Brain. New York: Oxford
Neurophysiol. 2005;93(1):437–453. University Press; 1981.
124. Dyhrfjeld-Johnsen J, Santhakumar V, Morgan RJ, et al. 142. Nunez PL. Generation of human EEG by a combination of long
Topological determinants of epileptogenesis in large-scale struc- and short range neocortical interactions. Brain Topogr. 1989;1:
tural and functional models of the dentate gyrus derived from 199–215.
experimental data. J Neurophysiol. 2007;97:1566–1587. 143. Nunez PL. Toward a quantitative description of large-scale neo-
125. Gorter JA, van Vliet EA, Aronica E, et al. Progression of sponta- cortical dynamic function and EEG. Behav Brain Sci. 2000;23(3):
neous seizures after status epilepticus is associated with mossy 371–398; discussion 399–437.
fiber sprouting and extensive bilateral loss of hilar parvalbumin 144. Nunez PL, Wingeier BM, Silberstein RB. Spatial-temporal struc-
and somatostatin-immunoreactive neurons. Eur J Neurosci. tures of human alpha rhythms: theory, microcurrent sources,
2001;13:657–669. multiscale measurements, and global binding of local networks.
126. Blumcke I, Suter B, Behle K, et al. Loss of hilar mossy cells in Hum Brain Mapp. 2001;13:125–164.
Ammon’s horn sclerosis. Epilepsia. 2000;41(suppl. 6):S174–S180. 145. Montaron MF, Bouyer JJ, Rougeul A, et al. Ventral mesencephalic
127. Lopes da Silva F, Blanes W, Kalitzin SN, et al. Dynamical diseases tegmentum (VMT) controls electrocortical beta rhythms and
of brain systems: different routes to epileptic seizures. IEEE Trans associated alternative behavior in the cat. Behav Brain Res.
Biomed Eng. 2003;50:540–549. 1982;6:129–145.
128. Van Drongelen W, Lee HC, Hereld M, et al. Emergent epilepti- 146. Lopes da Silva FH, van Rotterdam A, Storm van Leeuwen W, et
form activity in neural networks with weak excitatory synapses. al. Dynamic characteristics of visual EP’s in the dog. II. Beta fre-
IEEE Trans Neural Syst Rehabil Eng. 2005;13(2):236–241. quency selectivity in evoked potentials and background activity.
129. Lytton WW, Confreras D, Destexhe A, et al. Dynamic interactions Electroencephalogr Clin Neurophysiol. 1970;29: 260–268.
determine partial thalamic quiescence in a computer network 147. Gray C, Konig P, Engel A, et al. Oscillatory responses in cat visual
model of spoke-and-wave seizures. J Neurophysiol. 1997;77: cortex exhibit intercolumnar synchronization which reflects
1676–1696. global stimulus properties. Nature. 1988;338:334–337.
130. Tsakiridou E, Bertollini L, de Curtis M, et al. Selective increase in 148. Singer W, Gray CH. Visual feature integration and the temporal
T-type calcium conductance of reticular thalamic neurons in a correlation hypothesis. Annu Rev Neurosci. 1995;18:555–586.
rat model of absence epilepsy. J Neurosci. 1995;15:3310–3317. 149. Freeman WJ, Burke BC, Holmes MD. Aperiodic phase resetting
131. Leresche N, Parri HR, Erdemli G, et al. On the action of the in scalp EEG of beta-gamma oscillations by state transitions at
antiabsence drug ethosuximide in the rat and cat thalamus. J alpha-theta rates. Hum Brain Mapp. 2003;19(4):248–272.
Neurosci. 1998;18(13):4842–4853. 150. Howard MW, Rizzuto DS, Caplan JB, et al. Gamma oscillations
132. Suffczynski P, Pijn JPM, Pfurtscheller G, et al. Event-related dynam- correlate with working memory load in humans. Cerebral Cortex.
ics of alpha band rhythms: a neuronal network model of focal 2003;13:1369–1374.
ERD/surround ERS. In: Pfurtscheller G, Lopes da Silva F, eds. Event 151. Pfurtscheller G, Graimann B, Huggins JE, et al. Spatiotemporal
Related Desynchronization, Handbook of Electroencephalography and patterns of beta desynchronization and gamma synchronization
Clinical Neurophysiology, Revised Series vol. 6. Amsterdam: Elsevier in corticographic data during self-paced movement. Clin
Science; 1999:67–85. Neurophysiol. 2003;114(7):1226–1236.
90 Part I ■ Basic Principles

152. Sederberg PB, Kahana MJ, Howard MW, et al. Theta and gamma 167. Rapp PE, Albano AM, Schmah TI, et al. Filtered noise can mimic
oscillations during encoding predict subsequent recall. J low-dimensional chaos. Phys. Rev. E. 1993;47:2289–2297.
Neurosci. 2003;23:10809–10814. 168. Basar E. Chaotic dynamics and resonance phenomena in brain
153. Boeijinga PH, Lopes da Silva FH. Differential distribution of beta function: progress, perspectives, and thoughts. In: Basar E, Bullock
and theta EEG activity in the entorhinal cortex of the cat. Brain TH, eds. Chaos in Brain Function. Berlin: Springer; 1990:1–31.
Res. 1988;448:272–286. 169. Casdagli MC, Iasemidis LO, Savit RS, et al. Nonlinearity in inva-
154. Kay LM, Lancaster LR, Freeman WJ. Reafference and attractors in sive EEG recordings from patients with temporal lobe epilepsy.
the olfactory system during odor recognition. Int J Neural Syst. Electroencephalogr Clin Neurophysiol. 1997;102:98–105.
1996;7:489–495. 170. Elbert T, Ray WJ, Kowalik JE, et al. Chaos and physiology: determin-
155. Leung LS. Recurrent inhibition model of hippocampal CA in istic chaos in excitable cell assemblies. Physiol Rev. 1994;74:1–47.
vivo (letter to editor). Trends Neurosci. 1996;19:468–469. 171. Jansen BH. Is it and so what? A critical review of EEG chaos. In:
156. Freeman WJ. Feedback models of gamma rhythms (letter to edi- Duke DW, Pritchard WS, ed. Measuring Chaos in the Human
tor). Trends Neurosci. 1996;19:468. Brain. Singapore: World Scientific; 1991:49–82.
157. Whittington MA, Traub RD, Kopell N, et al. Inhibition-based 172. Pijn JP, Velis DN, Lopes da Silva FH. Chaos or noise in EEG sig-
rhythms: experimental and mathematical observations on net- nals: dependence on state and brain site. Electroencephalogr Clin
work dynamics. Int J Psychophysiol. 2000;38(3):315–336. Neurophysiol. 1991;79:371–381.
158. Perez Velasquez JLP, Carlen P. Gap junctions, synchrony and 173. Theiler J, Rapp PE. Reexamination of the evidence for lowdimen-
seizures. Trends Neurosci. 2000;23:68–74. sional nonlinear structure in the human electroencephalogram.
159. Traub RD, Whittington MA, Buhl EH, et al. A possible role for gap Electroencephalogr Clin Neurophysiol. 1996;99:213–222.
junctions in generation of very fast EEG oscillations preceding the 174. Varela F, Lachaux JP, Rodriguez E, et al. The brainweb: phase syn-
onset of, and perhaps initiating, seizures. Epilepsia. 2001;42:153–170. chronization and large-scale integration. Nat Rev Neurosci.
160. Pinto DJ, Jones SR, Kaper TJ, et al. Analysis of state-dependent 2001;2(4):229–239.
transitions in frequency and long-distance coordination in a 175. Pijn JP, Velis DN, van der Heyden M, et al. Nonlinear dynamics
model oscillatory cortical circuit. J Comput Neurosci. 2003;15(2): of epileptic seizures on basis of intracranial EEG recordings.
283–298. Brain Topogr. 1997;9(4):249.
161. Menon V, Freeman WJ, Cutillo BA, et al. Spatio-temporal corre- 176. Lopes da Silva FH, Pijn JP. Epilepsy: network models of genera-
lations in human gamma band electrocorticograms. Electroen- tion. In: Arbib MA, ed. The Handbook of Brain Theory and Neural
cephalogr Clin Neurophysiol. 1996;98:89–102. Networks. Cambridge, MA: A Bradford Book, MIT Press;
162. Babloyantz A. Strange attractors in the dynamics of brain activ- 1995:367–369.
ity. In: Haken H, ed. Complex Systems Operational Approaches. 177. Breakspear M. “Dynamic” connectivity in neural systems: theoret-
Berlin: Springer-Verlag; 1985:116–123. ical and empirical considerations. Neuroinformatics. 2004;2(2):
163. Schuster HG. Deterministic Chaos. Weinheim: PhysikVerlag; 1984. 205–226
164. Grassberger P, Procaccia I. Measuring the strangeness of strange 178. Lopes da Silva FH. Neural mechanisms underlying brain waves:
attractors. Physica. 1983;9D:189–208. from neural membranes to networks. Electroencephalogr Clin
165. Grassberger P, Schreiber T, Schaffrath C. Chaos. Int J Bifurc. Neurophysiol. 1991;79:81–93.
1991;1:521–547. 179. Freeman WJ. Dynamics of image formation by nerve cell assem-
166. Ott E. Chaos in Dynamical Systems. New York: Cambridge blies. In: Basar E, Flohr H, Haken H, et al., eds. Synergetics of the
University Press; 1993. Brain. Berlin: Springer-Verlag; 1983:102–121.
Biophysical Aspects of EEG and
Magnetoencephalogram Generation
he electrical activity of the brain consists of ionic cur- laminar current along the main axes of the neurons. The net
rents generated by biochemical sources at the cellular membrane current that results from the activation at the level
level. These ionic currents cause electric and magnetic of the synapse can be either a positive or a negative ionic cur-
fields that can be measured in the brain and surrounding tis- rent directed to the inside of the cell. Because there is no accu-
sues. The behavior of these fields can be predicted because they mulation of charge anywhere in the medium (see also
obey physical laws. This chapter is an introduction to the bio- Appendix, expression 5.1), the injected current at the synaptic
physical aspects of the generation of electroencephalogram level is compensated by other currents flowing in the medium,
(EEG) signals. It is convenient to consider the generation of as shown in Figure 5.1. At the level of the synapse and in the case
EEG signals in biophysical terms because this is the exact way to of an excitatory postsynaptic potential (EPSP), the synaptic cur-
determine the potential distribution at the scalp given a set of rent is carried by positive ions; in the case of an inhibitory
intracerebral current sources, that is, the so-called forward postsynaptic potential (IPSP), the corresponding current is
problem of electroencephalography. An understanding of this carried by negative ions. Because the direction of a current is
problem is necessary to discuss the inverse problem that consti- defined by the direction along which the positive charge is
tutes the main concern of clinical electroencephalography,
which is to determine the intracerebral sources given a meas-
ured potential distribution at the scalp. The inverse problem
has no unique solution, as shown long ago by Helmholtz (1);
therefore, it is essential to understand the forward problem so
that constraints of the inverse problem may be well examined.
This chapter treats the biophysical basis of the generation of
EEG potentials nonmathematically so that the general reader
may follow it more easily; in the theoretical appendix, a mathe-
matical treatment of the biophysical aspects is provided.

The electrical potential at a point in the brain in principle can
be computed if the microscopic cellular sources are known. In
expression 5.8 of the Appendix, the field potential is given as a
function of intracerebral current sources; in expression 5.10, it
is given in terms of the membrane potential.
In general terms, the field potential of a population of neu- Figure 5.1 Current flow patterns in and around an idealized neuron
rons equals the sum of the field potentials of the individual owing to synaptic activation. E, current flow caused by activation of a
neurons. To understand EEG phenomena, the activity of popu- synapse at the level of the apical dendrite resulting in depolarization of
lations of neurons must always be considered. EEG phenomena the membrane and flow of a net positive current. This current causes a
can be measured only at a considerable distance from the sink at the site of the synapse. The extracellularly measured excitatory
source if the responsible neurons are regularly arranged and postsynaptic potential (EPSP) is drawn at the left; it has a negative
activated in a more or less synchronous way. A typical regular polarity at the level of the synapse. At the soma there exists a distrib-
arrangement is the palisade, in which the neurons are distrib- uted passive source resulting in an extracellular potential of positive
uted with the main axes of the dendritic trees parallel to each polarity. I, current flow caused by activation of a synapse at the level of
other and perpendicular to the cortical surface. When in such a the soma resulting in hyperpolarization of the membrane and flow of a
population (e.g., the pyramidal neurons of layers IV and V of the net negative current. This results in an active source at the level of the
cortex) the neurons are more or less simultaneously activated soma and passive sinks at the basal and distal dendrites. The extracel-
by way of synapses lying at the proximal dendrites, extracellular lularly measured inhibitory postsynaptic potentials (IPSPs) at the soma
currents will flow; their longitudinal components (i.e., parallel and dendritic level are drawn. Note that an inhibition at the soma gen-
to the main axes of the neurons) will add, whereas their trans- erates about the same extracellular potential field as an excitation at the
verse components will tend to cancel out. The result is a apical or distal dendrites.

92 Part I ■ Basic Principles

transported, the ionic current is directed to the intracellular

medium with an EPSP; with an IPSP, it is directed to the extra-
cellular medium. Therefore, at the level of the synapse, there is
an active sink in the case of a positive ionic current (EPSP) or
an active source in the case of a negative ionic current (IPSP).
The extracellular potential at the sink is negative; at the source,
it is positive, as can be seen from expression 5.8. Along the cell
and at a distance from the synaptic level, there exists a distrib-
uted passive source in the case of an EPSP or a distributed pas-
sive sink in the case of an IPSP.
In addition to postsynaptic potentials, other relatively slow
variations of membrane potential, such as those associated with
depolarizing or hyperpolarizing afterpotentials and dendritic
events as calcium action potentials, may also be the sources of
extracellularly measurable potentials.
At a macroscopic level, it may therefore be stated that the
potential field generated by a synchronously activated palisade
of neurons behaves like that of a dipole layer. This is no more
than a rough model of reality. It may also be considered that an
active sink (for instance, at the level of the cell somata in the
cortex) is flanked by two passive sources, one lying more super-
ficially and the other deeper than the sink; in such a case, the
potential field should behave more like a quadrupole layer. This
is, however, an unnecessary complication because in general the
two sides of the quadrupole are rather asymmetrical; one is
usually dominant. This is to be expected because it can be seen
from histological sections of the cortex that there is a clear Figure 5.2 Examples of closed, open, and open-closed fields for differ-
asymmetry of the pyramidal cells along a direction perpendicu- ent types of neuron pools in the central nervous system. Left: The pop-
lar to the cortex. These cells present a morphological polariza- ulations are drawn. Right: Neurons representing the simultaneously
tion in the vertical direction because they are characterized by a activated pools are shown together with the arrows representing the
rather long vertically directed apical dendrite that ramifies in lines of flow of current at the instant when the impulses have evaded
the most superficial layers of the cortex and by basal dendrites the cell bodies. The zero isopotential lines (0) are also indicated.
distributed around the soma; the axon runs vertically down- A: Oculomotor nucleus represented schematically by only one neuron
ward to the white matter. For the sake of simplicity, it therefore with dendrites oriented radially outward. The isopotential lines are cir-
may be assumed that most potential fields generated in pal- cles; the currents flow entirely within the nucleus, resulting in closed
isades of neurons can be mimicked by simple dipole layers. field (all points outside the nucleus remain at zero potential).
Lorente de No (2) named this type of potential field the “open B: Superior olive represented by neurons each having a single dendrite
field” (Fig. 5.2C) (3), in contrast to those fields generated by oriented radially inward. The currents result in a closed field.
neurons with dendritic arborizations radially distributed about C: Accessory olive represented by only one neuron with a single long
the soma; according to his description, these would generate dendrite. The arrangement of sources and sinks in this structure permits
“closed fields” as shown in Figure 5.2A,B. In the latter case, the the spread of current in the volume of the brain and thus results in an
configuration of the radially oriented neurons is such that each open field. D: Two structures mixed together, generating an open-closed
neuron may be considered as generating a central source (or field. (Modified from Lorente de No R Action potential of the motoneu-
sink, depending on the type of synaptic activity) surrounded by rons of the hypoglossus nucleus. J Cell Comp Physiol. 1947;29: 207–287;
spherically distributed sinks (or sources). The field potential in Hubbard, J I, Llinas, R Quastel D M J Electrophysiological Analysis of
this case is equivalent to that of a distribution of radially ori- Synaptic Transmission. London: Edward Arnold Ltd; 1969.)
ented dipoles at the surface of a sphere. It can be seen intuitively
that, under these circumstances, not only the tangential but also
the radial components of current cancel each other; this can
also be proved by the volume conduction theory (Fig. 5.3) (4). variation caused by an action potential generates a field that is
Postsynaptic potentials in neuronal populations with an equivalent to that of a single dipole perpendicular to the mem-
appropriate spatial organization can be sources of field poten- brane because the piece of membrane that is depolarized at any
tials that can be measured at a distance and therefore are also instant of time is small. In contrast, that of an electrotonically
sources of EEG signals. Can other types of membrane poten- conducted postsynaptic potential extends, at any moment of
tials as action potentials also contribute to those EEG signals? time, over a larger portion of the membrane; thus, it generates
This is not generally the case for two reasons (see also a field that corresponds rather to that of a dipole layer with
Appendix). The first reason is that the membrane potential dipoles perpendicular to the membrane surface. The latter
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 93

Figure 5.3 Equipotential plots and depth profiles for open cortices. Intersection of base lines of the depth profiles is at
the center of the sphere. Tick marks represent the region of cells. Note that the 90-degree profile of the hemisphere and
45-degree profiles of the cap and punctured sphere run along the open edge of the cortex. (Adapted from Klee M, Rall W.
Computed potentials of cortically arranged populations of neurons. J Neurophysiol. 1977;40:647–666.)

attenuates with distance less rapidly than the former, as can be In those cases in which the action potentials occur simultane-
seen by comparing expressions 5.12 and 5.15 of the Appendix; ously in clusters (e.g., when a group of fibers is excited by a short
this is illustrated in Figure 5.10B and has also been discussed by stimulus), the field of these action potentials may also be
Humphrey (5). These properties have been analyzed more rig- recorded at relatively large distances in the form of what is gen-
orously by Pettersen and Einevoll (6) using realistic computer erally called a “compound action potential.”
models of pyramidal and stellate neurons with different mor- Quantitative measurements in vitro and computational neu-
phologies. These authors demonstrated that the neural mor- ronal models used to infer contributions of neuronal populations
phology and passive electrical parameters influence the width to MEG and EEG signals.
and amplitude of extracellular spikes. These modeling studies Studies using in vitro preparations allow the simultaneous
revealed that in all neurons there is a low-pass filtering effect, that measurement of electric and magnetic fields generated by neu-
is, the spike-width increases with increasing distance from soma. ronal activity. A calculation of the magnetic field around an
These authors showed quantitatively that the spike-amplitude axon was performed by Swinney and Wikswo (7) using a prepa-
decays with distance r depending on dendritic morphology, ration where an isolated axon was kept in a bath with saline, in
according to the rule 1/rn with 1 n 2 close to the soma and vitro, and was stimulated electrically. These authors found that
n 2 far away along the dendritic tree. These studies have also the magnetic field resulted mainly from the intracellular cur-
implications for understanding the relation between the extent of rent flowing inside the axon along the axial direction, whereas
fields produced by multiunit activity (MUA) and by synaptic the contribution of the return passive current that flows in the
potentials, which constitute the so-called local field potentials medium was two orders of magnitude smaller, and that of the
(LFPs or local EEG), since they substantiate why MUAs are more transmembrane current was negligible.
confined around the neural sources than the LFPs generated by More recently, Murakami and Okada (8) applied the same
synaptic currents. In the latter case the dipole layer lengths are kind of approach to the neocortex. This study is particularly
much longer than the typical action-potential current dipoles, relevant because it has yielded some results that may help to
and thus the attenuation with distance is weaker than for the interpret EEG and magnetoencephalogram (MEG) recordings
MUAs. The second reason that action potentials do not con- from the scalp. These authors made a computer model, based
tribute to EEG signals is that the latter, owing to their short dura- on that proposed by Mainen and Sejnowski (9), of the four
tion (1 to 2 msec), tend to overlap much less than do postsynaptic main types of cortical neurons taking into account their real-
potentials (EPSP and IPSP), which last longer (10 to 250 msec). istic shapes. Each neuron is described as a three-dimensional
94 Part I ■ Basic Principles

compartmental model, where each compartment has its typical populations (14–18); the interlocking takes place by way of
geometric dimensions, passive electrical properties (membrane recurrent collaterals forming different types of synapses that
capacitance and resistance, intracellular resistance), and five possibly include dendrodendritic synapses and even gap junc-
voltage-dependent ionic conductances. Neuronal activity was tions (19).
obtained by stimulating each neuron with an intracellular cur- These neural masses act as macroscopic sources of EEG sig-
rent injected at the soma. The intracellular current is repre- nals. For example, in the case of the generation of alpha
sented by a vector quantity, the magnitude of which is the rhythms, it has been shown that such a macroscopic source is
current dipole moment and is represented by the scalar quan- located within the cortex and behaves like a dipole layer
tity Q in [pA m]. directed perpendicularly to the cortical surface (20). The
The population Q for a group of pyramidal neurons was cal- dynamic properties of alpha rhythms, such as their frequency,
culated by computing the sum of the dot product of each cur- depend on the parameters of the neuronal populations, such as
rent dipole Qk and the unit vector orthogonal to the cortical feedback gains, strength of connections, time properties of
surface. The same was done for the stellate cells, with a modifi- synaptic potentials, and nonlinear properties (i.e., thresholds
cation due to the fact that these neurons are differently ori- and saturation).
ented. The magnetic field B and electrical potential V at the Does this rhythmic activity propagate along the cortex or
scalp can be expressed as functions of the intracellular currents, form a standing field? It has been shown experimentally (20)
called primary currents, and the return currents that flow in the that, in the visual cortex of the dog, small cortical areas appear
space around the sources. In the model the magnetic field B is to act as “epicenters” from which alpha rhythm activity
assumed to be generated by the tangential component of Q, “spreads” in different directions. The existence of this type of
with respect to the inner skull surface, and the electric field is spread has been concluded on the basis of phase differences in
due to both the tangential and radial components. This implies rhythmic components measured at different locations on the
that B is mainly caused by neuronal activity in sulci, since Q is cortical surface. It has been shown in a model study (21) that a
directed perpendicularly to the pial surface, whereas both radial neuronal chain, where the model neurons are interconnected by
and tangential components contribute to the electric field. This means of recurrent collaterals and interneurons such that the
investigation led to another result of practical interest, namely it connectivity functions are homogeneous and their strength
gave some direct indications about how to interpret the magni- decays as an exponential function of distance, shows propaga-
tude of the magnetic fields recorded in human. According to the tion of electrical activities characterized by frequencies in the
model the overall magnitude of Q for the activity of pyramidal alpha band with a velocity c in the order of magnitude of
neurons of layers V and II/III is in the order of 0.29 to 0.90 pA 0.3 m/sec, which is in the range of experimental results. The
m. Apparently this value deviates appreciably from that esti- dynamics of the activity of neuronal populations are treated in
mated previously for a cell by Hämäläinen et al. (10) that was more detail in Chapter 4.
just 0.02 pA m. That larger value, however, is of the same order The propagation of electric activity in the cortex may result in
of magnitude as those estimated for hippocampal pyramidal the interesting phenomenon of time dilation, that is, an EEG
neurons (0.2 pA m per cell in CA1 and 0.17 pA m per cell in transient recorded at the cortex will seem to be of shorter dura-
CA3, (11,12)). Murakami and Okada (8) point out that assum- tion than the transient recorded simultaneously at the scalp. This
ing a Q of 0.2 pA m per cortical pyramidal neuron, a population has been shown to occur in relation to epileptiform spikes (22).
of 50,000 synchronously active cells would generate a field with The phenomenon of time dilation is due to the fact that a mov-
a magnitude of 10 nA m, which corresponds precisely to the ing dipole seen from a distance can be observed over a longer
smallest value measurable from the human cortex according to time than when seen from nearby. The physical basis for this phe-
Hämäläinen et al. (10). Assuming that a cortical minicolumn nomenon is accounted for in the Appendix (expression 5.18).
with a diameter of 40 m contains 100 cells (13), the cortical
surface that would correspond to 50,000 cells should form a INFLUENCE OF INHOMOGENEITIES
patch with about 0.63 mm 2 area. If this cortical patch would
have a circular form its diameter would be about 0.9 mm. It has been assumed that the brain could be considered an infi-
In order to estimate the number of cortical cells correspon- nite homogeneous medium, but in reality this is not the case.
ding to a cortical column, the synchronous activity of which is The field potentials are influenced not only by the geometry of
responsible for the EEG/MEG signal measured, one has to the neuronal populations and the electrical properties of indi-
know, at least approximately, the density and the spatial distri- vidual neurons but also by the existence of regions with differ-
bution of pyramidal and other cell types within the cortex. ent conductivities in the head, that is, by the presence of
inhomogeneities. For an interpretation of field potentials meas-
TIME-DEPENDENT PROPERTIES ured at the scalp, therefore, it is important to take into consid-
eration the layers lying around the brain: the cerebrospinal fluid
We have emphasized above the importance of synchrony of (CSF), the skull, and the scalp. These layers account, at least in
neuronal activity for the generation of EEG signals. We consider part, for the attenuation of EEG signals measured at the scalp as
here the factors that cause this synchrony. The main factor is of compared to those recorded at the underlying cortical surface.
a structural nature. Neuronal masses are, in general, organized This can be formulated in mathematical terms, as in the
as combinations of interlocked excitatory and inhibitory Appendix (expression 5.19).
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 95

To compute the potential distribution at the surface of the consisting of five concentric shells, solved the forward problem
scalp caused by a dipole placed within the brain, it is necessary in an analytic form for the case in which the skull and the cor-
to compute the potential distribution at the surface of the differ- tex had anisotropic properties. The effects of the anisotropic
ent shells (i.e., to solve the boundary value problem). This can be conductivity of the skull were determined by Bertrand et al.
done by solving the Poisson equation (see equation 5.7) in polar (40) and by Van den Broek (41). The latter showed that the
coordinates as done in the Appendix. Already in the 1970s, the anisotropy of the skull causes the smearing out of the distribu-
influence on the EEG of the specific electrical properties of the tion of the EEG over the scalp, whereas the normal component
tissues surrounding the brain was recognized (23–27). The sys- of the MEG is not affected.
tematic study of Ary et al. (28) led to the proposal of a method In addition, it is relevant to emphasize that the skull does not
that would correct the errors introduced by variations in skull have a homogeneous surface, due to both the existence of
and scalp thickness. Initially, the models of the head volume regions with different thickness and the occipital and the eye
conductor consisted of concentric spheres, usually three to sockets openings (42). The existence of holes in the skull, for
account for the brain, skull, and scalp; in some models a fourth example, due to surgical interventions, influences also the dis-
layer was included representing the CSF. In the classic models, tribution of the EEG over the scalp as demonstrated by
the three main concentric spheres typically had radii of 90, 83, Bertrand et al. (40). In general, the localization of the dipoles is
and 78 mm. The values of the conductivities of the three layers then shifted toward the hole.
commonly used were proposed by Geddes and Baker (29): for
the brain 0.33, for the skull 0.0042, and for the scalp also REALISTIC MODELS OF THE HEAD
0.33 S.m 1; in case the CSF was included, it had a conductivity AND NUMERICAL METHODS
of 1.0 S.m 1. Recently more precise estimates of these conduc-
tivities were obtained by way of measurements in vivo, using an In the past, most solutions of the forward problem assumed
electric impedance tomography (EIT) method combined with a that the head could be reasonably approximated by a spherical
realistic model of the head (30–32). The results of the study of model with three or four shells. However, the geometry of this
Gonçalves et al. (32) showed that the ratio between the conduc- volume conductor is clearly not spherical. Therefore, since the
tivities of the skull and the brain lies between 20 to 50 (for six late 1980s, realistically shaped models were gradually intro-
subjects) rather than the traditionally assumed value of 80. The duced and applied (43–47). This was made possible by the
average values found in this investigation are the following: advances in magnetic resonance imaging (MRI) that allow
brain: 0.33 S.m 1; skull: 0.0081 S.m 1. It is also important to making realistic estimates of the three-dimensional geometry
note that the intersubject variance of the estimates decreased by of the brain and surrounding tissues (48). To solve the forward
half when a realistic model was used, in comparison with a problem in a realistic shaped volume of the head, it is necessary
spherical model. However, a factor of 2.4 was found between the to apply numerical methods. This involves decomposing the
subjects with the extreme values of the ratio of conductivities. volume conductor in a mesh consisting of a relatively large
This implies that these values should be estimated for each indi- number of discrete elements. The accuracy of the method
vidual to increase the reliability of the estimated conductivities. increases with the number of these elements, and, of course, is
In addition, it is known that the conductivity of the various inversely proportional to their size, but the computational
tissues is not homogeneous. The conductivity of the skull varies effort also increases accordingly. The most commonly used
with its thickness and bone structure. Law et al. (33) showed methods to compute the EEG potential distributions are the
that there is an inverse relation between skull resistivity and boundary-element method (49–53), which is appropriate to the
thickness. Cuffin (34) and Eshel et al. (35) investigated the case where the volume conductor is assumed to be isotropic
effects of varying the conductivity of the skull; the former and piecewise homogeneous, and the volume element methods,
determined that the effect of local variations in skull and scalp which are needed in case the assumptions of isotropy and
thickness was slightly larger on the EEG than on the MEG, piecewise homogeneity do not hold. Examples of the latter are
while the latter showed a correlation between interhemispheric the finite-volume (35), the finite-difference, and the finite-
asymmetry in skull thickness and the amplitude of the scalp element methods (for details see reviews by van Uitert,
EEG. The very thorough studies of Haueisen (36,37) on the Weinstein, and Johnson (54)). These methods are powerful, but
influence of various combinations of conductivities on both the they need reliable information about the structure of the brain
EEG and MEG showed that the scalp EEG is most influenced by and surrounding tissues, obtained from MRI scans, and of the
the conductivities of the skull and the scalp, while the MEG was corresponding conductivities, to be applied in a sensible way.
most influenced by the conductivities of the brain tissue and They are particularly suited to calculate the influence of specific
the CSF. We should note that the conductivity of the tissues of inhomogeneities such as the influence of the cerebral ventricles
the head can also present anisotropy. Indeed, it is known that in filled with CSF (41) and anisotropic conductivities (36).
brain tissue, the conductivity measured in a direction parallel to In a number of studies, the advantage of using realistic mod-
a fiber tract can be 10 times larger than in the perpendicular els in comparison with spherical models has been demon-
direction (38). It is, however, difficult to integrate this finding in strated; differences in dipole reconstructions using EEG data
global models of the whole head, since fiber tracts are organized were reported to reach 20 mm (55,56). Relatively smaller differ-
in a most complex way within the anatomical constraints of the ences were obtained with MEG data, in the order of 4 mm (57).
folds of the brain. Nevertheless, De Munck (39), using a model This implies that the necessity of using realistic models,
96 Part I ■ Basic Principles

particularly those accommodating inhomogeneous volumes, is distribution. The number of sensors, however, is much smaller,
stronger for EEG than for MEG recordings. In the latter case, in the order of a maximum of 128 for the EEG and 300 for the
most calculations are based on the boundary-element method. MEG. Thus, the problem of estimating brain sources of EEG or
MEG data is underdetermined. Regularization methods have
THE INVERSE PROBLEM: APPROACHES been proposed to help solve this major difficulty.
TO ESTIMATE SOLUTIONS For solving the inverse problem, the method of the least-
squares source estimation has been applied both to the activity
The introduction stated that the problem of calculating the measured at a single moment in time or within a time interval
intracerebral sources of the potentials measured at the scalp, that (65,66). In this context a convenient approach, proposed by De
is, the so-called inverse problem of electroencephalography, has Munck (39,67), is to split the set of parameters of the ED into
no unique solution. This means that different combinations of linear and nonlinear parameters. This procedure consists
intracerebral sources can result in the same potential distribution schematically of the following steps: first, the time functions
at the scalp. The only way to tackle this problem is to make spe- that describe the change of the source as function of time must
cific assumptions about the intracerebral sources that are be estimated for a given position and orientation of the ED; sec-
assumed to cause a given EEG potential distribution and to make ond, the orientations must be found given the dipole time func-
a model of the conductive media lying between the sources and tions and position. These two steps must be performed
the recording electrodes. This implies that first a particular type alternatively a number of times until the best dipole orientation
of forward problem is solved; thereafter, the theoretical scalp and time functions are found for given dipole positions.
potentials obtained in this way are compared with those recorded Thereafter, the nonlinear position parameters must be updated
experimentally. It should be noted that, on the one hand, a satis- and the process repeated until a best fit is obtained. This proce-
factory comparison does not necessarily imply that the assumed dure starts from the assumption that the position and orienta-
sources are those really responsible for the measured EEG poten- tion of the initial dipole are known, that is, it is based on the
tials. On the other hand, however, a bad comparison must lead to stationary dipole model proposed by Scherg and Von Cramon
the conclusion that the assumptions are not correct with respect (65), although it is not constrained to the initial choice.
to the sources or to the model, or both. If a reasonable assumption on the initial position and orien-
The numerical procedures by means of which the measured tation cannot be made, one may use the alternative approach of
EEG/MEG distribution over the scalp is compared with that estimating EDs for a number of successive time samples of a
computed using the forward approach are not trivial. Usually given potential, or magnetic, distribution (so-called moving
one expresses the difference between the real and the simulated dipole approach) as done by Stok (64) and Stok et al. (68) and
distributions as a summed squared difference. The question is many others. In this way a series of EDs is obtained, as shown in
to find a minimum of this function, which can be accomplished Figure 5.4. Those EDs that have about the same position and ori-
by several algorithms. The mostly used of this is that of entation can be clustered and the clustered dipoles can be aver-
Marquardt (58). In the last decades many solutions of the aged. If desired, these average ED clusters can be used as seed
inverse problem have been proposed, starting from the pioneer points for the procedure of de Munck as described earlier (69).
work of Schneider and Gerin (59), Schneider (25,26), Smith et We should note that, in practice, if the distribution of the
al. (60), Henderson et al. (61), and many others as reviewed by EEG potential or of the MEG field over the scalp is relatively
Scherg (62,63) and Stok (64). simple, a single ED may be an appropriate model. Of course,
According to this approach, the intracerebral source is one needs all available a priori information to speed up the
assumed to be an equivalent current dipole localized within the solution and to give it high reliability. For instance, if it is to be
brain. Here we consider, in general terms, those aspects of the expected that a given activity occurs simultaneously in both
inverse problem that are common to EEG and MEG. The aim hemispheres in areas well defined anatomically, it is advisable to
of the inverse procedure is to obtain an equivalent source, start by assuming two symmetrical dipoles at the appropriate
which in general is a dipole, determined by six parameters. This sites (63). An increase in the number of dipoles can easily lead
solution is called the equivalent dipole (ED), which must be to rather complex and ambiguous interpretations. Validation of
considered a mathematical abstraction. It represents the theo- ED models can be obtained in retrospect by relating the solu-
retical dipole, which generates a potential (or magnetic field) tions obtained to independent physiological or anatomical
distribution at the surface of the outer shell, that is the closest information. An example of an anatomical–physiological vali-
in the least square sense to the measured distribution at the dation is the case of the dipolar sources of alpha, mu rhythms,
scalp. It should be emphasized that the estimation of an ED and sleep spindles (70) that are distributed on brain areas as
located within the brain gives only a rough estimate of the expected on the basis of animal and human physiological
center of gravity of the cortical area active at a given moment. observations. An example of anatomical–pathological valida-
For example, in the case of the dipole layer shown in Figure 5.3 tion is given by the distribution of sources of epileptiform tran-
the corresponding ED would be located much deeper than the sients in patients with cortical lesions visible in MRIs (Fig. 5.5),
surface where the individual dipolar sources are placed. where the main epileptiform sources are located around the
Assuming that the activity of each cortical macrocolumn lesions (71,72).
would be described by an equivalent current dipole, one would A number of alternative methods have been developed to
need several thousands to account for an EEG or MEG scalp circumvent the obvious limitations of the ED approach. One of
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 97

Figure 5.4 Projection of equivalent dipoles (EDs) based on visual evoked potentials (VEPs) (left) and visual evoked fields
(VEFs) (right) during the on-response of subject JM, to the appearance of a checkerboard presented to the left half-visual
field. Each ED is shown in three projection planes, namely occipital (top), horizontal (middle), and sagittal (bottom). An
arrow indicates the direction and length of the projection of the dipole and the starting point indicates the dipole location.
Each arrow is identified by a small number that runs from 1 to 11. A group of arrows represents an interval of 50 msec, in
steps of 5 msec, spanning the time interval of 55 to 160 msec after start of the stimulus. Scale: one division indicates 10 mm
for the position and 6.7 10–9 A m, for the components. Each dipole is plotted with a line width corresponding to the
measure of fit 2: thick for 0.1, medium for 0.1 0.2, thin for 0.2 0.4, very thin for 0.4. (Modified
from Stok CJ, Spekreijse HJ, Peters MJ, et al. A comparative EEG/ MEG equivalent dipole study of the pattern onset visual
response. New trends and advanced techniques in clinical neurophysiology. EEG Suppl. 1990;41:34–50.)

these consists in applying spatial filtering to the data so that sig- issue by creating new algorithms based on the assumption that
nals from a given location may be privileged with respect to the the sources are at fixed locations within the brain, namely at a
rest. This is the so-called linearly constrained minimum vari- given node of the triangular tessellation of the cortical surface
ance beam forming (73) that has been applied in practice with as extracted from the subject’s MRI. In this way the inverse
some interesting results (74). Related methods are the synthetic solution is simplified to finding the linear parameters of
aperture magnetometry (SAM) method introduced by the sources. Nevertheless, this problem is still underdetermined.
Robinson and Vrba (75), and the parametric mapping method The methods used to solve this problem use different forms of
applied by Dale et al. (76). In addition, methods have been pro- regularization parameters (80). In this way particular forms of
posed to obtain estimates of multiple dipoles with little a priori inverse solutions can be obtained: minimum norm (MN)
information, such as the multiple signal classification (MUSIC) (81–84), weighted resolution optimization (WROP) method of
algorithm (77–79). A number of laboratories have pursued this Grave de Peralta Menendez et al. (85), and the low-resolution
98 Part I ■ Basic Principles

Figure 5.5 Top: Channel overlays and

topographic maps of the average magnetic
fields for the five spike clusters in patient
C. The field maps represent the magnetic
field distribution at the time of the marker,
indicated in the channel overlays by the
dashed vertical line. Above each map the
cluster number and, within brackets, the
number of spikes it contains. Bottom:
Position of equivalent dipoles. In the top
axial and sagittal magnetic resonance
imaging (MRI) slices the dipoles fitted to
the average magnetoencephalogram
(MEG) spike of clusters 1 and 2, com-
bined. In the bottom slices are the dipole
locations for cluster 5. For clusters 3 and 4
equivalent dipole sources with less than
10% residual error were not found. The
boundary of the structural lesion in the
right frontal lobe for this patient is marked
by dotted lines. (Adapted from Van’t Ent
D, Manshanden I, Ossenblok P, et al. Spike
cluster analysis in neocortical localization
related epilepsy yields clinically significant
equivalent source localization results in
magnetoencephalogram [MEG]. Clin
Neurophysiol. 2003;114(10):1948–1962.)

brain electromagnetic tomography (LORETA). The latter was compared, in a computer simulation study, the performance of
proposed by Pascual-Marqui et al. (86) and uses the discrete the surface Laplacian and a linear estimation approach with
spatial Laplacian operator for regularization. This method dipoles having free moment orientations, the so-called cortical
yields a very smooth inverse solution (87). potential reconstruction based on linear estimation (Fig. 5.6).
In any case, dipole models provide less ambiguous solutions Compared to the surface Laplacian method, the latter linear esti-
than the more sophisticated methods described above, since mation method provided a more accurate cortical image of the
they are based on simpler assumptions. Nonetheless, they yield sources for a comprehensive review of these methodologies, see
images that are less attractive than the latter methods. In any Ref. 94. Another alternative method consists in reconstructing the
case, dipolar methods are only meaningful if the scalp field has potential distribution on the inner boundary of the skull by the
approximately focal character. method of spatial deconvolution (89).
Alternatively, the scalp EEG may be described using the surface
Laplacian that estimates the local normal component of the cur-
rent through the skull (33) and can be computed by applying SPATIAL DISTRIBUTIONS OF SOURCES
either a local or a global approach. According to the local OF NEUROPHYSIOLOGICAL SIGNALS:
approach (88–90), the surface Laplacian is computed locally using THE CONTRIBUTION OF HEMODYNAMIC
a subset of neighboring sensors for differentiation. According to INFORMATION
the global approach (33,91,92), an interpolation function is first
applied to the measurements obtained over the whole head and Currently a number of groups are attempting to solve the
the differentiation is applied afterward. The relationship between problem of the inherent ambiguity of the EEG/MEG inverse
the surface Laplacian and the cortical potential distribution is only solutions by combining functional MRI (fMRI) data with
simple in case of homogeneous skull and scalp layers. Zanow (93) EEG/MEG data (76), but these approaches are still in
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 99

Figure 5.6 The EEG (top row), the brain

potential reconstructed by means of linear
estimation with minimum norm constraint
(center row), and the surface Laplacian
maps (bottom row) of the deviant tones
responses to a mismatch negativity exper-
iment are shown at two latencies: at 140
msec (left column) and 180 msec (right
column) after stimulus onset. Results were
computed with Advanced Source Analysis
(ASA) software package. [Adapted from
Zanow F. Laplacian maps (bottom row) of
the deviant tones responses to a mismatch
neg-1997. Realistically shaped models of
the head and their application to EEG and
MEG, p. 144. Ph.D. thesis, University of
Twente, Enschedé.]

development. A problem is that the time course of the fMRI mainly of synchronized synaptic potentials, either excitatory or
signal, that is, the blood-oxygen-level dependent (BOLD) sig- inhibitory, with a contribution from membrane oscillations
nal differs from the corresponding local neurophysiological and spike afterpotentials.
signals, since it is delayed with respect to the latter. Devor et al. (97) reported the results of a systematic study in
Nonetheless, Logothetis and Wandell (95) examined the rela- rat somatosensory cortex, to characterize the relationship
tion between LFPs (or local EEG), single- and multiunit activ- between changes in blood-oxygen content and the neural spik-
ity (MUA), and high spatiotemporal fMRI responses recorded ing and synaptic activity, showing that there is a nonlinear rela-
simultaneously in monkey visual cortex, and were able to tionship between electrophysiological measures of neuronal
demonstrate that while both LFP and MUA activity are corre- activity and the hemodynamic response.
lated with BOLD for short stimulation paradigms, this is not The methodologies developed to interpret the results of
the case for sustained oscillations, in the latter case only the combining EEG recordings and fMRI, particularly in epileptic
LFP signal is significantly correlated with the hemodynamic disorders, have been critically reviewed (98). More recently, a
response. The LFPs are commonly thought to be related to the study by Shmuel and Leopold (99) combining fMRI-BOLD and
input of a neural ensemble and dendritic processing within the intracortical neurophysiological signals confirmed the existence
network in contrast with the action potentials or “units” that of correlations between neuronal signals consisting of either the
consist of the output of neurons (96). Thus, LFPs consist firing rate of a small group of neurons, or of relative power
100 Part I ■ Basic Principles

changes in the MUA band, but particularly of LFPs in the In the section on the influence of inhomogeneities we have
gamma frequency band, with BOLD signals at a time lag (peaks seen that, to explain the EEG, we must assume the existence of
at about 6 sec lag) relative to neuronal activity. shells with different conductivities. The existence of these
shells affects the electric potential in two ways: it causes an
attenuation of the potential and it introduces ED layers at the
THE MAGNETOENCEPHALOGRAM (MEG) boundaries of the inhomogeneities as shown in expression
5.19. However, the brain and surrounding tissues behave as a
As discussed above, the ionic currents originating from bio- medium with constant magnetic permeability. Therefore, the
chemical sources at the cellular level in the central nervous sys- magnetic field, in contrast to the electric field, is not influ-
tem generate not only electric fields but also magnetic fields. enced by those layers; it is nevertheless also affected by the
Until Cohen’s work (100), magnetic fields could not be induced dipole layers existing at the boundaries of the inho-
detected owing to the extremely low strength of the MEG and mogeneities in conductivity, as shown in the Appendix
the absence of sufficiently sensitive transducers. The magnetic (expression 5.30).
field of the brain has a strength in the order of magnitude of Other properties of the MEG that should be mentioned are
10 8 gauss (G) or, in meter-kilogram-second (m kg s) units, of the following. Neither electrodes nor a reference point are nec-
10 12 Weber m 2 or tesla (W m 2 or T) (101), whereas the essary for recording the MEG; the transducers for the MEG
earth magnetic field has a strength in the order of magnitude of need not touch the scalp, because the magnetic field does not
0.5 G (102). However, with the introduction of the supercon- disappear where conductivity is zero (free space). From expres-
ducting quantum interference device (SQUID) magnetometer, sion 5.29 it can be seen that the vector potential has the same
based on a superconducting effect at liquid helium tempera- direction as the intracerebral current density. Because the mag-
ture, sensitivities in the order of magnitude of 10 10 G have netic field is perpendicular to the vector potential, we can con-
been obtained (101). clude that current dipoles directed perpendicularly to the
The earth magnetic field and the urban magnetic noise fields surface of the skull result in magnetic fields tangential to the
(10 4 to 10 3 G) can be practically removed while recording surface of the head and, vice versa, tangential current dipoles
the MEG by using the gradiometer technique. This is based on result in magnetic fields perpendicular to the skull.
the fact that the disturbing fields are almost constant over large Research on MEG has advanced considerably in recent years
distances, whereas the MEG falls off rapidly with distance. This (see Chapter 42). Both practical and theoretical aspects of magne-
is done by subtracting the induced potentials in two coils placed toencephalography have been elucidated, as summarized by
close to each other; in this way, the earth and urban noise mag- Cohen and Cuffin (104), Okada (105), Cuffin (106), Modena et al.
netic fields are suppressed. The magnetic field in stationary (107), Hari and Ilmoniemi (108), Hämäläinen (109), Lewine and
conditions obeys Ampere’s laws and can be derived from the Orrison (110), and de Jongh et al. (111). Some theoretical aspects
microscopic sources at the cellular level, as shown in the of general interest will be mentioned here. Although EEG and
Appendix (from expression 5.22 onward). Although the electri- MEG reflect in essence the same elementary phenomena, in prac-
cal potential reflects the strength and localization of the current tice the two types of measurement differ in a number of aspects.
sources, the magnetic field reflects not only these properties but First, the EEG is a relative measurement; it always needs a
also the direction of the current densities. This can be seen in reference electrode. The magnetic field measurement does not
the Appendix by comparing expression 5.29 for the magnetic need a reference point.
quantities with expression 5.8 for the electric potential. Second, the MEG is a measure of the magnetic fields perpen-
Whether the information about the direction of current densi- dicular to the skull, which are caused by tangential current
ties in the brain may be derived from the MEG depends on the dipoles. The radial component of the current dipole does not
practical possibility of recording the magnetic fields in three generate a magnetic field outside a sphere-shaped volume con-
directions. In general we may state that the EEG reflects mainly ductor, and thus does not contribute to the MEG. By contrast, the
the volume or extracellular currents, whereas the MEG is more EEG is a measure of both components. This means that the MEG
sensitive to the primary intracellular currents. There are very reveals tangential current dipoles in a clear way, which in the EEG
few experimental studies where magnetic and electric fields may be obscured by radial sources. The main intracortical dipo-
have been measured along with elementary electrophysiological lar current sources are perpendicular to the cortical surface
properties of simple brain preparations. The recent study of (Fig. 5.1). Therefore, Cohen and Cuffin (104) have pointed out
Murakami et al. (103) validates the assumption indicated that the MEG mainly measures the cortical current dipoles lying
above. This study examined whether evoked magnetic fields in the sulci and not on the convexity of the gyri. Thus, the former
and intra- and extracellular potentials from longitudinal hip- causes mainly tangential dipoles to the skull surface, whereas the
pocampal CA3 slices of guinea pig can be interpreted within a latter produces mainly radial dipoles (Fig. 5.4).
single theoretical framework that incorporates ligand- and volt- Third, the map representing the distribution of the mag-
age-sensitive conductances in the dendrites and soma of the netic field at the surface of the head caused by a tangential
pyramidal cells. The intracellular potentials in this validated current dipole is rotated by 90 degrees to the corresponding
model reveal that the spikes and slow waves of the magnetic EEG potential map as shown in Figure 5.7. This means that to
fields are generated in or near the soma and apical dendrites, localize a dipole-like source in the x and y directions, both
respectively. MEG and EEG should be used (104), but it cannot be stated
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 101

Figure 5.7 Examples of measured and computed maps corresponding to two time samples of a visual evoked potential
(VEP) and magnetic field (VEF). The stimulus was the appearance of a checkerboard pattern as described by Stok (64) and
Stok et al. (68). Top row: Measured VEF and VEP contour plots for two instants of time during the on-response: 100 and
130 msec. Middle row: Simulated VEF and VEP contour plots that correspond to the EDs that were estimated from the
measured VEP distributions. Bottom row: The simulated contour plots that correspond to the VEF-based EDs. The dashed
rectangles indicate the simulation area of 12 14 cm 2. Measures are in centimeters and with respect to the inion. In the
contour plots of measured data, negativity is indicated by dashed lines and in the contour plots of simulated data by thin
lines. The thick lines indicate the zero level. Note that these plots present theoretical and experimental examples of the
three main differences that can be expected between EEG and MEG contour plots. (Modified from Stok CJ, Spekreijse HJ,
Peters MJ, et al. A comparative EEG/ MEG equivalent dipole study of the pattern onset visual response. New trends and
advanced techniques in clinical neurophysiology. EEG Suppl. 1990;41:34–50.)

that one type of measurement is better than the other. significantly in respect to the case of the homogeneous volume
Differences in localization of sources with both methods may conductor under the influence of inhomogeneities in
be determined mainly by the direction along which the meas- conductivity. In certain cases of spherical symmetry, however,
urements were made. the magnetic field distribution due to a current dipole is
Fourth, the EEG represents a sum of the potentials caused by independent of spherically symmetrical variations in resistivity
primary and secondary sources or volume currents. In the of the volume conductor (114). Whether this will hold true for
Appendix, it is shown that the media of the head (brain, CSF, a real head is questionable.
skull, and scalp) have different conductivities; thus, secondary Nevertheless, Cohen and Cuffin (104) assume that the MEG
sources of electric (expression 5.19) and magnetic fields would not be influenced by the volume currents and that this
(expression 5.30) are introduced. It is often stated, however, would explain why the MEG pattern is more focal, or tighter,
that the MEG is not affected by these volume currents than the corresponding EEG potential distribution. It appears
(104,105). It is still debatable how much these secondary that still tighter MEG patterns may be obtained by using trans-
sources contribute to the EEG and MEG under the conditions ducers composed of a number of magnetic field–sensing coils
in which these signals are experimentally measured. Plonsey in an array.
(112) pointed out that both the electric and magnetic fields are Fifth, there are differences between MEG and EEG measure-
affected by secondary sources. Van Rotterdam (113), using a ments regarding the representation of some types of nondipo-
model of a volume conductor represented by a lattice including lar sources, as discussed in detail by Cuffin (106). Different
inhomogeneities, showed that the magnetic field changes types of sources give different results. For side-by-side line
102 Part I ■ Basic Principles

sources forming a dipole layer, the EEG gives a more accurate However, the question of whether the MEG offers real
representation of the actual sources than does the MEG; the practical advantages over the EEG has been raised (64). Stok
reverse is true for in-line sources, that is, a series of dipoles et al. (68) made a systematic theoretical and experimental
aligned along the same direction. However, it should be added comparison of EEG and MEG and concluded that the MEG is
that both EEG and MEG solutions of the inverse problems for not, in general, superior over the EEG, although the MEG
the side-by-side and in-line sources lie deeper in the head and localizes the equivalent sources slightly better than the EEG,
have a larger amplitude than the actual sources. In particular, but it estimates the dipole components slightly worse. Two
the sources that are in the form of lines and are more than 2 cm general conclusions of these studies should be put in evidence:
long give different results when measured using EEG or MEG; (i) when model assumptions are not violated, the MEG and
the MEG solutions are deeper for the side-by-side sources, and EEG lead essentially to the same inferences about the source
EEG solutions are deeper for the in-line sources. These differ- (excluding magnetically silent sources); and (ii) when model
ences may be helpful in identifying the type of such sources in assumptions are violated, the MEG and EEG respond, in gen-
the brain (106). eral, in different ways such that a comparison of the results
Sixth, MEG and EEG appear to differ in their capability of obtained with both methods may give clues regarding the
localizing a current dipole source, given a set of observations kind of model violation present. Therefore, it is important to
corrupted by noise. A single current dipole can be represented use both methods to obtain a higher reliability in functional
by six parameters; three determine its position inside the head localization studies (117).
and three define its components. It has been shown by Stok et This question has been addressed directly by comparing
al. (115) that the EEG gives better estimates of the three com- MEG and EEG maps obtained in patients where dipoles were
ponents of the current dipole than the MEG, based on obser- implanted, using indwelling intracranial electrodes that were
vations to which noise was added. The latter, however, gives placed for the diagnosis of an epileptogenic focus (55,118).
better estimates of the three position parameters. Moreover, These authors concluded, from a limited number of experi-
the inverse procedure, for cases with bad signal-to-noise ratios, mental measurements, that the average errors of localization
fails more often with EEG than with MEG data. The EDs esti- were 10 mm for the EEG and 8 mm for the MEG, and thus that
mated by the inverse procedure using EEG and MEG data sep- the MEG was not superior to the EEG. Nevertheless, the main
arately may be combined in order to obtain an average advantage of the MEG may be the fact that the MEG is not
estimate. Alternatively, the ED may be calculated directly by affected by radial sources and, in this way, the use of both MEG
determining the dipole that provides the least-squares fit to and EEG may complement each other and help to establish a
EEG and MEG simultaneously. It can be demonstrated that the better model of the source, when it cannot be described as
latter procedure provides slightly better results than the com- being a simple dipole.
bination of the two estimates into an average estimate. This In conclusion, more research is needed where the two meth-
reinforces the statement that the two types of measurement, ods are directly compared using the same data. In the tutorial
MEG and EEG, indeed provide complementary information book on “MEG Methods” edited by Hansen et al (119), some
and should be combined whenever precise source localization frequently asked questions about these issues are discussed in
is desired. more detail (120).
Seventh, spherical models of the head perform much bet-
ter with MEG than with EEG data, because the former is less APPENDIX: THE MATHEMATICAL BASIS
sensitive to the influence of volume currents, while these cur-
rents are affected appreciably by deviations from the ideal AB Van Rotterdam
head. This means that inverse solutions based on MEG data The ionic currents in the brain cause electrical and magnetic
may be simpler to obtain since spherical models may be more fields obeying Maxwell’s and Ohm’s laws. These fields have a
readily used in this case than with EEG measurements (116). direction and magnitude; they must, therefore, be represented
Nevertheless, volume currents may still affect the MEG as as vector functions. The electric field is given by E (r ,t) and the
shown by Van Uitert et al. (54). Certainly one of the benefits magnetic field by H (r ,t) , because these vectors depend on
that has been obtained by the introduction of the MEG tech- location r and on time t. (Vector quantities are indicated by an
nique is that a considerable amount of research on brain arrow above the symbol.)
activity by competent multidisciplinary groups has led to a In brain and surrounding tissues, the material constants
renewed interest in the biophysical study of the electric and are approximately the following: conductivity, ≈ 10–1 –1 m –1;
magnetic activity recorded from the scalp. This led to the dielectric constant, ≈ 10–9 F m –1; and magnetic permeability,
development of more sophisticated models of the neuronal ≈ 10–6 H m –1. Do potential changes taking place in the brain
sources and of the volume conductor. These developments at a frequency of, for example, 1 k Hz cause propagated elec-
have been of interest for obtaining a better understanding not tromagnetic waves in the medium that may be detected after
only of the MEG but also of the EEG. Initially the experimen- a certain time at a distant point such as the scalp? It can be
tal results obtained by the new technique have led to the proved (121) that, in a medium such as the brain with the
thought that the MEG is superior to the EEG in locating brain material constants indicated above, the propagation velocity
sources (108). of the electromagnetic waves is of the order of 105 m s–1,
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 103

change of current in the volume of the cube. When no net cur-

rent flows into the cube (i.e., the current that goes into the vol-
ume comes out again), this total change must be zero, as
indicated in expression 5.1. Thus, we may state that the current
is stationary. In the case of a stationary current, the electrical
field E(r ) , or simply E, is related to the electrical potential V(r ) ,
or simply V, by the following expression:

E gradV (5.3)

In this expression the operator gradient (grad) indicates dif-

ferentiation of the scalar function V, of space and time (time is
considered here to be fixed) in the following way:

grad V a , , b (5.4)
x y z
where gradV is a vector quantity called the gradient of V.
Figure 5.8 Cubic volume with sides dx, dy, and dz. Current density In a conductive medium, Ohm’s law is valid; that is:
flowing in: Jz (x — dx, y — dy, z — dz); Jy (,y — dy) and Jz (,,z — dz)
or J(r dr ) . Current densities flowing out: Jx (x,y,z); Jy (,y,) and Jz J Ji sE (5.5)
(,,z) or J(r ) . When divJ 0, no charge is accumulated in volume
where J is the current density injected in the medium (e.g., a
dxdydz. Inset: Decomposition of current density vector J in components transmembrane current density caused by synaptic activity in a
in x, y, and z direction as Jz , Jy, and Jx. neuron).
Taking the divergence of both sides of expression 5.5 and
using expression 5.1, one may write:
which implies that one need not be concerned with the prop-
agation of electromagnetic waves caused by potential changes divJi
within the brain. This means that, in practical terms, the divE (5.6)
effects of those potential changes may be detected simultane-
ously at any point in the brain or surrounding tissues. This
Substituting E by – gradV according to expression 5.3, one
important statement leads to the conclusion that the currents
can write an expression that is called the Poisson equation for
caused by sources in the brain at any moment in time behave
the potential field owing to an injected current:
in a stationary way. This means that no charge is accumulated
at any time in the brain (122). 2 2 2
V V V divJi
Therefore, it can be stated that for the current density J(r ) divgrad V (5.7)
or, for simplicity, J, at any moment of time: x2 y2 z2 s

div J O (5.1) In case the medium where r o is situated is assumed to be

where the operation div J is called divergence of the vector J; infinite, isotropic, and homogeneous, it can be proved that the
J is the current density in A m –2. The operator div must be solution of the Poisson equation is the following:
understood as indicating differentiation of a vector; expression
5.1 can also be written in another way that may help in under- 1 divJi 3
V(r o ) dr (5.8)
standing its meaning. Let us consider the expression for J for 4 s R
the case of an infinitesimally small cube with sides dx, dy, and
dz (Fig. 5.8). It holds that: which gives the values of the potential at a point r o in the
volume conductor from injected current densities Ji, at points
Jx Jy Jz
divJ dxdydz a b dx a b dy a dxdyb dz r , lying at a distance R from r o ( R 0r r o 0); the integral
x dydz y dydz z represents the summation over all current sources within the
(5.2) volume; note that divergence of the vector J gives a scalar
In this expression the factors between brackets, such as Summarizing, it has been shown that the space-varying and
( Jx/ x dydz), represent the rate of change of current through the time-varying parts of the electrical variables can be sepa-
the surfaces dydz, dxdz, and dxdy. To account for the total rated; the time-varying part is the same everywhere in the
change of current, it is necessary that those factors are multi- medium owing to the large propagation velocity, and the space-
plied by the sides of the cube dx, dy, and dz, then these products varying part can be computed for every instant in time accord-
must be summed. Thus expression 5.2 represents the total ing to expression 5.8.
104 Part I ■ Basic Principles

This general expression is not the only way to define V(r o ) . To understand the effect of the solid angle in a more practi-
Two other expressions (5.10 and 5.12) are given below. These cal way, it is useful to derive an explicit expression for d :
are more convenient because they relate the extracellular
potential field directly to membrane processes of the individ- 1 # R cos
ual neurons in the brain. One of these expressions can be d (r r o) grada b dO (r ) #
3 dO (r ) dO
R R R2
derived assuming that the neuronal membrane can be consid-
ered as equivalent to a double layer with an inner (intracellu-
lar) membrane potential Vm and an outer (extracellular) where is the angle between R and dO; it should be noted that
potential V. In these terms, V( r o ) is given by the following R is the vector connecting the point r o where V is measured and
expressions: the point r where the dipole is located (see Fig. 5.8).
It should be noted in expression 5.9 that a change in mem-
1 1 # brane potential such as a depolarization in a small piece of
V(r o ) (s iVm (r ) s e(r ))grada b dO (r ) (5.9)
4 s R membrane potential results in an intracellular potential Vm and
an extracellular potential V. In this case, the membrane behaves
where i is the intracellular and e is the extracellular conduc- like a current dipole consisting of a sink at the outside and a
tivity, and dO is pointing to the extracellular medium; the inner source at the inside. Thus, there is a jump in potential across the
product grad (1>R) #dO is called the solid angle d (r r o )
membrane. A current dipole, or simply a dipole, is defined by its
strength and direction; it is represented by the vector p (r ) . In the
subtended by the surface element dO (r ) on the membrane sur- example described above, the dipole is oriented perpendicularly
face S and seen from the extracellular point r o. It must be to the membrane surface dO; thus, the dipole is represented as
emphasized that expression 5.8 takes the form of expression 5.9 p(r )dO. According to the above, another expression for the
only when the injected current densities Ji originate at the cell potential V(r o ) caused by one dipole can be derived (123):
membrane as derived by Plonsey (121). Expression 5.9 can be
simplified to the following approximate expression, because p(r ) 1 #
V(r o ) grad a b dO (r )
0s eV(r ) 0V 0s iVm (r ) 0: 4 s R

V(r o ) Vm (r )d (r ro) (5.10)
4 se p(r ) cos
V(r o ) dO (5.12)
4 s R2
The solid angle d subtended by a membrane surface dO
seen from a point r o can be interpreted in a geometrical way, as From expression 5.12 it can be concluded that in the case of
shown in Figure 5.9. a single dipole at the membrane, V decreases with the square of
the distant R to the source and it is also proportional to cos ;
thus, it is maximal when R is perpendicular to the surface dO.
In many cases, the membrane depolarization (or hyperpo-
larization) does not remain limited to a small piece of mem-
brane; rather, it spreads over a more or less extended membrane
surface. In this way the equivalent current source cannot be
accounted for in terms of single dipole. One must, instead,
assume that there exists a dipole layer at the membrane.
For a dipole layer located at the surface of the membrane,
one must sum the effects of all individual dipoles; therefore,
expression 5.12 becomes:

1 1 #
V(r o ) p(r )grad a b dO (r )
4 s s R


V(r o ) p(r )d (r ro) (5.13)
4 s s

Figure 5.9 Solid angle subtended by infinitesimal surface dO with

From expression 5.13, it can be noted that, if p is constant
direction dO on membrane cylinder S seen from the extracellular point over the closed membrane surface S, then V(r o ) 0, because
r o lying at a distance R 0ro r 0 from dO; the angle between the the integral of the solid angle d over an external closed surface
direction of the surface dO and r o r is . is zero. Thus, to obtain a value of V(r o ) different from zero,
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 105

Figure 5.10 A: Dipole layer of length 1 and height h in the y-z plane seen from extracellular point r o on the x axis.
B: Attenuation (relative potential, rel. pot.) of extracellular potential as a function of distance (r o) in a log–log plot for the
case of a single dipole and for the case of a dipole layer. Note that, in the latter case, two slopes can be distinguished, the
decay of the potential with distance being more accentuated in the far field (ro 1). The arbitrary values of h = 0.0625
and I = 1 used in this model are indicated on the horizontal axis.

p must vary over the membrane surface; in other words, the where the time behavior of the field is the same as that of the
membrane potential must vary over the surface owing, for sources.
example, to local synaptic activity. In general terms, to allow for time-varying sources, expres-
In this case, the attenuation of the potential in the volume sion 5.10 must be changed as follows:
conductor depends not only on distance R and angle , but also
on the geometry of the dipole layer. For instance, for a dipole 1
layer of constant strength with length 1 and height h located V(r o, t) Vm (r , t)d (r ro ) (5.16)
4 se s
at x = 0 in the y-z plane (Fig. 5.10A) the potential field at a point
r o along the x axis will be given by the following expression: This effect can be illustrated by the following example (Fig.
5.11A). Let us assume a cell that can be simplified to an equiva-
lent cylinder (124) oriented along the z axis where a membrane
1>2 ¢ h>2 ro depolarization front propagates along the cylinder with an arbi-
V(r o ) p 2 2 2 3>2
dyzy (5.14)
y 1>2 z ¢ h>2 (r 0 y z) trary velocity c. In this case, we may write that the resulting
potential V(r o ) is proportional to a time-varying solid angle.
For ro W h and h V 1, this can be estimated as follows: From expressions 5.11 and 5.16, the potential V(r o,t) caused
by a depolarization front along the cylinder is given by:
V(r o ) p¢ h (5.15)
ro5 1 2 1>2
(r o>1) 6 cos ( (t) )
V(r o, t) 2 dO (5.17)
0ctn z r o 0
It can be seen that in the far field, r o W 1, the potential tends
to decay as in the case of the single dipole with r 2o where for where n z is the unit vector in the z direction. Note that the
short distances h ro 1, the potential decays proportion- denominator in expression 5.17 represents the square of the
ally to distance ro. In Figure 5.10B, it can be seen that in the case distance R(t) between r o and the time-varying location of the
of the dipole layer the attenuation has two slopes depending on front.
the relation between r o and 1, whereas in the case of a simple At a point further away in the same direction, r 1 a r o, V
dipole there is only one slope. becomes:
Until now, the discussion has considered only the behavior
of standing sources of EEG signals, that is, sources that do not 1
change in space. However, there is experimental evidence that V(r 1, t) 2 V(r o, t>a ) (5.18)
leads one to assume that some macroscopic sources in the
brain do propagate in space (14,20,92). This propagation of Therefore, the potential at a far location r 1 is equal to the
sources has a particular effect on the behavior of the potential potential at location r o attenuated in amplitude and in a more
field in time. This effect does not appear for standing sources dilated time scale (Fig. 5.11B).
106 Part I ■ Basic Principles

a (s e s k )V(r )d (5.19)
k 1 sk

s e V(r )d

in which S1 represents the membrane surface that encloses the

first neuron with membrane potential Vm (r ) and intracellular
conductivity i; the extracellular conductivity is e except in the
regions bounded by the surfaces Sk, within these regions the
conductivity is k; s (r o ) is the conductivity for the region
where point r o lies: s (r o ) s k if r o is bounded by Sk and it is
e elsewhere, d are the solid angles of surfaces pointing to the
region with conductivity e.
The second term on the right describes the influence that
neurons have as inhomogeneities on the extracellular field; the
third term describes the distorting influences of the M inhomo-
geneities, whereas the fourth term expresses the influence of the
border of the medium.
Using expression 5.19, it is in principle possible to compute
the EEG measured on the scalp and generated by N neurons.
In the following model, the brain and surrounding tissues
are represented as a sphere with a number of shells within
which a single dipole is situated. The potential field in and on
the sphere is a solution of the Poisson equation obeying the
following boundary conditions: the current densities normal
to the surfaces of the shells are continuous and the tangential
components of the electric field at the boundaries are also
Assuming radial symmetry, the solution for a radial dipole
can be written as a Legendre series, in polar coordinates as
Figure 5.11 A: Membrane cylinder oriented along the z axis in which follows (123):
a depolarization front propagates (with velocity c) along the z direc-
tion, r o and r 1 are two extracellular points at which extracellular q
potentials are measured r 1 a r o. Note that the potentials V at r o and V(r, f , ) (n 1)
bn r n 6P0n (cos ) (5.20)
a 5 an r
at r 1 are different, because and R differ when seen from r 1 instead n 0
of r o. The changes in and R as a function of time depend on
the propagation velocity c. B: Extracellular potentials V measured
where P0n ( cos ) is a Legendre polynomial of the nth order. For
simultaneously at r o and r 1 (plot with arbitrary scales) showing time
dilation. a tangential dipole, the solution is:

(n 1)
V(r, , ) a 5 cn r dnr n 6 P1n ( cos ) cos (5.21)
n 0

Let us assume that the field potential V(r o ) generated by the where P1n (cos ) is an associated Legendre polynomial.
activity of N neurons situated in a finite homogeneous medium
The coefficients an, bn , cn , and dn for all n in both cases should
containing M structures with different conductivities and
bounded by surface S (outside the medium the conductivity is be evaluated on basis of the boundary conditions at the different
zero) can be expressed as follows (125): boundaries and the fact that the source is a dipole (23).

4 s (r 0 )V(r 0 ) sia (i)
i i si
In the stationary approximation of Maxwell’s equations for the
sea V(r )d electromagnetic fields generated by microscopic sources at
i i si the cellular level, the behavior of the magnetic field H or the
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 107

induction field B mH is governed by two experimental In this way, it can be seen that the magnetic vector potential
physical laws. There is Ampere’s law in the stationary A reflects the direction of the current densities Ji. If there are
approximation: inhomogeneities in conductivity, the expression becomes (126):

m Ji (r ) 3 V(r )
curlH J (5.22) A( r 0 ) c dr a (s e s k) dO (r )d (5.30)
4p vol R k Sk R
where Sk are surfaces enclosing regions with conductivity k;
divB 0 (5.23) elsewhere is the conductivity e. The surfaces dO point to the
region with conductivity e; is assumed to be constant.
which expresses the experimental fact that magnetic monopoles
do not exist. The term curl H is a differential operation called REFERENCES
curl or rotation on a vector function; it is defined as follows:
1. Helmholtz H. Ueber einige Gesetze der Verteilung elektrischer
0H y 0H z 0H z 0H x 0H x 0H y T Ströme in Körperlichen Leitern mit Anwendung auf die
curlH ca b #a b #a bd thierischelektrischen Versuche. Prog Ann Physik Chemie.
0z 0y 0x 0x 0y 0H y 1855;89:211–233, 353–377.
(5.24) 2. Lorente de No R. Action potential of the motoneurons of the
hypoglossus nucleus. J Cell Comp Physiol. 1947;29:207–287.
For conservative or rotation-free fields, that is, fields deriv- 3. Hubbard JI, Llinas R, Quastel DMJ. Electrophysiological Analysis of
able from vector potentials (e.g., E gradV), this differenti- Synaptic Transmission. London: Edward Arnold Ltd; 1969.
ation results in a zero vector. However, for vector fields directed 4. Klee M, Rall W. Computed potentials of cortically arranged popu-
lations of neurons. J Neurophysiol. 1977;40:647–666.
along a closed path (e.g., closed current density loops), the 5. Humphrey DR. Reanalysis of the antidromic cortical response. II.
operation results in a vector field perpendicular to the original On the contribution of cell discharge and PSPs to the evoked
vector field. It can be proved that expression 5.23 is equivalent potentials. Electroencephalogr Clin Neurophysiol. 1968;25:421–442.
6. Pettersen KH, Einevoll GT. Amplitude variability and extracellular
to the statement that B can be derived from a vector potential A
low-pass filtering of neuronal spikes. Biophys J. 2008;94(3):
as follows: 784–802.
7. Swinney KR, Wikswo JP Jr. A calculation of the magnetic field of a
B curlA (5.25) nerve action potential. Biophys J. 1980;32(2):719–731.
8. Murakami S, Okada YC. Contributions of principal neocortical
By substituting expression 5.25 in 5.22 and using the defini- neurons to magnetoencephalography and electroencephalography
tion of J as a function of impressed current densities (expres- signals. J Physiol. 2006;575(3):925–936.
sion 5.5), we obtain: 9. Mainen ZF, Sejnowski TJ. Influence of dendritic structure on fir-
ing pattern in model neocortical neurons. Nature.
1 1996;382:363–366.
curlcurlA Ji sE (5.26) 10. Hämäläinen MS, Hari R, Ilmoniemi R, et al. Magneto -
m encephalography: theory, instrumentation and applications to the
noninvasive study of human brain function. Rev Med Physics.
It can be shown that curlcurlA divgrad A grad divA, 1993;65:413–497.
so that 5.26 becomes 11. Kyuhou S, Okada YC. Detection of magnetic signals from isolated
transverse CA1 hippocampal slice of the guinea pig.
divgrad A mJi grad divA ms E (5.27) J Neurophysiol. 1993;70:2665–2668.
12. Okada YC, Wu J, Kyuhou S. Genesis of MEG signals in a mam-
One can simplify expression 5.27 by making use of the ambi- malian CNS structure. Electroencephalogr Clin Neurophysiol.
guity in the definition of A (123) and choose A such that 13. Mountcastle VB. The columnar organization of the neocortex.
divA ms V the two last terms of expression 5.27 then can- Brain. 1997;120(pt 4):701–722.
cel out; expression 5.27 becomes the Poisson equation that for 14. Freeman WJ. Mass Action in the Nervous System. New York, NY:
Academic Press; 1975.
each separate component of A can be expressed as follows:
15. Katchalsky AK, Rowland V, Blumenthal R. Dynamic patterns of
brain cell assemblies. Neurosci Res Prog Bull. 1974;12(1):90–94.
divgrad A mJi (5.28) 16. Lopes da Silva FH, Hoeks A, Smits A, et al. Model of brain rhyth-
mic activity. Kybernetik. 1974;15:27–37.
17. Lopes da Silva FH, Van Rotterdam A, Barts P, et al. Models of neu-
In an infinite medium with constant and , the solution of ronal populations. The basic mechanisms of rhythmicity. Prog
equation 5.28 can be proved to be (121): Brain Res. 1976;45:281–308.
18. Wilson HR, Cowan JD. A mathematical theory of the functional
m Ji (r ) 3 dynamics of cortical and thalamic nervous tissue. Kybernetik.
A(r 0 ) dr (5.29)
4 vol R 1973;13:55–80.
108 Part I ■ Basic Principles

19. Schmitt FO, Dev P, Smith BH. Electronic processing of informa- 40. Bertrand O, Thevenet M, Perrin F, et al. Effects of skull holes on.
tion by brain cells. Science. 1976;193:114–120. In: Proceedings of the 14th International Conference of the IEEE
20. Lopes da Silva FH, Storm van Leeuwen W. The cortical alpha Engineering in Medicine and Biology Society, Satellite Symposium
rhythm in dog: the depth and surface profile of phase. In: Brazier on Neuroscience and Technology. Lyon, France; November, 1992.
MA, Petsche H, eds. Architectonics of Cerebral Cortex. New York, pp. 42–45.
NY: Raven Press; 1978. 41. Van den Broek SP. Volume conduction effects in EEG and MEG.
21. Van Rotterdam A, Lopes da Silva FH, van den Ende J, et al. A Ph.D. thesis, University of Twente, Enschedé, 1997.
model of the spatial-temporal characteristics of the alpha rhythm. 42. Yan Y, Nunez PL, Hart RT. Finite element model of the human
Bull Math Biol. 1982;44:283–305. head: scalp potentials due to dipole sources. Med Biol Eng Comput.
22. Lopes da Silva FH, van Hulten K. Analyse quantitative de l’activite 1991;29:475–481.
intercritique en E.E.G. et S.E.E.G. dans l’epilepsie. Rev EEG 43. Hämäläinen MS, Sarvas J. Realistic conductivity geometry model
Neurophysiol. 1978;8:198–204. of the human head for interpretation of neuromagnetic data. IEEE
23. Hosek RS, Sances A, Jodat RW, et al. The contributions of intrac- Trans Biomed Eng. 1989;36(2):165–171.
erebral currents to the EEG and evoked potentials. IEEE Trans 44. He B, Musha T, Okamoto Y, et al. IEEE Trans Biomed Eng.
Biomed Eng. 1978;25:405–415. 1987;34:406–414.
24. Kavanagh RN, Darcey TM, Lehmann D, et al. Evaluation of meth- 45. Meijs JWH, Voorde BJ. ten, Peters MJ, et al. The influence of vari-
ods for three-dimensional localization of electrical sources in the ous head models on EEGs and MEGs. In: Pfurtscheller G, Lopes da
human brain. IEEE Trans Biomed Eng. 1978;25:421–429. Silva FH, eds. Functional Brain Imaging. Toronto: Hans Huber;
25. Schneider MR. A multistage process for computing virtual dipolar 1988:31–45.
sources of EEG discharges from surface information. IEEE Trans 46. Meijs JWH, Peters MJ, Boom HBK, et al. Relative influence of
Biomed Eng. 1972;19:1–12. model assumptions and measurement procedures in the analysis
26. Schneider MR. Effect of inhomogeneities on surface signals com- of the MEG. Med Biol Eng Comput. 1988;26:136–142.
ing from a cerebral current dipole source. IEEE Trans Biomed Eng. 47. Yvert B, Bertrand O, Echallier JF, et al. Improved forward EEG cal-
1974;21:52–54. culations using local mesh refinement of realistic head geometries.
27. Witwer JG, Tresek GJ, Jewett DL. The effect of media inhomo- Electroencephalogr Clin Neurophysiol. 1995;95:381–392.
geneities upon intercranial electric fields. IEEE Trans Biomed Eng. 48. Wieringa HJ, Peters MJ. Processing MRI data for electromagnetic
1972;19:352–362. source imaging. Med Biol Eng Comput. 1993;31:600–606.
28. Ary JP, Klein SA, Fender DH. Locations of sources of evoked scalp 49. Meijs JWH, Weier OW, Peters MJ, et al. On the numerical accuracy
potentials; corrections for skull and scalp thicknesses. IEEE Trans of the boundary element method. IEEE Trans Biomed Eng.
Biomed Eng. 1981;28:447–452. 1989;36:1038–1049.
29. Geddes LA, Baker LE. The specific resistance of biological; materi- 50. De Munck JC, Peters MJ. Mathematical aspects of biomagnetic
als—a compendium of data for the biomedical engineer and phys- modeling. In: Proceedings of the 8th International Conference on
iologist. Med Biol Eng. 1967;5:271–293. Biomagnetism. In: Hoke M, ed. Munster: Excerpta Medica; 1991:
30. Ferree TC, Eriksen KJ, Tucker DM. Regional head tissue conduc- 11–12.
tivity estimation for improved EEG analysis. IEEE Trans Biomed 51. Ferguson AS, Zhang X, Stroink G. A complete linear discretization
Eng. 2000;47:1584–1592. for calculating the magnetic field using the boundary element
31. Gonçalves SI, de Munck JC, Heethaar RM, et al. The application of method. IEEE Trans Biomed Eng. 1994;42:455–459.
electrical impedance tomography to reduce systematic errors in 52. Kybic J, Clerc M, Abboud T, et al. A common formalism for the
the inverse EEG problem—a simulation study. Physiol Meas. integral formulations of the forward EEG problem. IEEE Trans
2000;21:379–393. Med Im. 2005;24:12–28.
32. Gonçalves SI, de Munck JC, Verbunt JPA, et al. In vivo measure- 53. Kybic J, Clerc M, Faugeras O, et al. Generalized head models for
ment of the brain and skull resistivities using an EIT-based MEG/EEG: boundary element method beyond nested volumes.
method and realistic models of the head. IEEE Trans Biomed Eng. Phys Med Biol. 2006;51:1333–1346.
2003;50:754–767. 54. Van Uitert R, Weinstein D, Johnson C. Volume currents in forward
33. Law SK, Nunez PL, Wijesinghe RS. High-resolution EEG using and inverse magnetoencephalographic simulations using realistic
spline generated surface Laplacians on spherical and ellipsoidal head models. Ann Biomed Eng. 2003;31(1):21–31.
surfaces. IEEE Trans Biomed Eng. 1993;40:145–153. 55. Cuffin BN. EEG localization accuracy improvements using realis-
34. Cuffin BN. Effects of local variations in skull and scalp EEG’s and tically shaped head models. IEEE Trans Biomed Eng.
MEG’s. IEEE Trans Biomed Eng. 1993;40:42–48. 1996;43:299–303.
35. Eshel Y, Witman S, Rosenfeld M, et al. Correlation between skull 56. Roth BJ, Balish M, Gorbach A, et al. How well does a threesphere
reflects the direction of the current densities. IEEE Trans Biomed model predict positions of dipoles in a realistically shaped head?
Eng. 1995;42:232–249. Electroencephalogr Clin Neurophysiol. 1993;87:175–184.
36. Haueisen J. Methods of numerical field calculation for neuromag- 57. Menninghaus E, Lütkenhöner B, Gonzalez SL. Localization of a
netic source localization. Ph.D. thesis, Technische Universität bipolar source in a skull phantom: realistic versus spherical model.
Ilmenau, Shaker Verlag, Aachen, 1996. IEEE Trans Biomed Eng. 1994;41:986–989.
37. Haueisen J, Ramon C, Czapski P, et al. On the influence of volume 58. Marquardt DW. An algorithm for least squares estimation of non-
currents and extended sources on neuromagnetic fields: a simula- linear parameters. J Soc Indust Appl Math. 1963;11:431–441.
tion study. Ann Biomed Eng. 1995;23:728–739. 59. Schneider MR, Gerin P. Une methode de localisation des dipoles
38. Nicholson PW. Specific impedance of cerebral white matter. Exp cerebraux. Electroencephalogr Clin Neurophysiol. 1970;28:68–78.
Neurol. 1965;13:386–401. 60. Smith DB, Lell ME, Sideman RD, et al. Nasopharyngeal phase
39. De Munck JC. A mathematical and physical interpretation of the reversal of cerebral evoked potentials and theoretical dipole impli-
electromagnetic field of the brain. Ph.D. thesis, University of cations. Electroencephalogr Clin Neurophysiol. 1973;34:654–658.
Amsterdam, the Netherlands, 1989.
Chapter 5 ■ Biophysical Aspects of EEG and Magnetoencephalogram Generation 109

61. Henderson CJ, Buler SR, Glass A. The localization of equivalent 80. Baillet S, Garnero L. A Baysenian approach to introducing
dipoles of EEG sources by the application of electric field theory. anatamo-functional priors in the EEG/MEG inverse problem.
Electroencephalogr Clin Neurophysiol. 1975;39:117–130. IEEE Trans Biomed Eng. 1997;44:374–385.
62. Scherg M. Fundamentals of dipole source potential analysis. In: 81. Hämäläinen MS, Ilmoniemi RJ. Interpreting measured magnetic
Grandori F, Hoke M, Romani GL, eds. Auditory Evoked Magnetic fields of the brain: estimates of current distributions. Technical
Fields and Potentials. Advances in Audiology. Basal: S Karger; 1990;6: Report TKKFA559. Helsinki: University of Technology; 1984.
40–69. 82. Hämäläinen MS, Ilmoniemi RJ. Interpreting magnetic fields of the
63. Scherg M. Functional imaging and localization of electromagnetic brain: minimum norm estimates. Med Biol Eng Comp. 1994;
activity. Brain Topogr. 1992;5:103–111. 32:35–42.
64. Stok CJ. The inverse problem in EEG and MEG with application to 83. Wang JZ, Williamson SJ, Kaufman L. Magnetic source images deter-
visual evoked responses. Ph.D. thesis, University of Twente, The mined by a lead-field analysis: the unique minimum-norm least-
Netherlands, 1986. squares estimation. IEEE Trans Biomed Eng. 1992;39(7):665–675.
65. Scherg M, Von Cramon D. Two bilateral sources of the late AEP as 84. Wang JZ. Minimum-norm least-squares estimation: magnetic
identified by a spatiotemporal dipole model. Electroencephalogr source for a spherical model head. IEEE Trans Biomed Eng.
Clin Neurophysiol. 1985;62:32–44. 1993;40(4):387–396.
66. Wood CC. Application of dipole localization methods to source 85. Grave de Peralta Menendez R, Hauk O, Gonzalez Andino S, et al.
identification in human evoked potentials. Ann N Y Acad Sci. Linear inverse solutions with optimal resolution kernels applied
1982;388:139–155. to electromagnetic tomography. Hum Brain Map. 1997;5:
67. De Munck JC. The estimation of time varying dipoles on the basis 454–467.
of evoked potentials. Electroencephalogr Clin Neurophysiol. 86. Pascual-Marqui RD, Michel CM, Lehmann D. Low resolution elec-
1990;77:156–160. tromagnetic tomography: a new method for localizing electrical
68. Stok CJ, Spekreijse HJ, Peters MJ, et al. A comparative EEG/MEG activity in the brain. Int J Psychophysiol. 1994;18:49–65.
equivalent dipole study of the pattern onset visual response. New 87. Pascual-Marqui RD, Esslen M, Kochi K, et al. Functional imaging
trends and advanced techniques in clinical neurophysiology. EEG with low-resolution brain electromagnetic tomography
Suppl. 1990;41:34–50. (LORETA): a review. Methods Find Exp Clin Pharmacol. 2002;
69. Lopes da Silva FH, Spekreijse H. Localization of brain sources of 24(suppl C):91–95.
visually evoked responses: using single and multiple dipoles. An 88. Hjorth B. An online transformation of EEG scalp potentials into
overview of different approaches. In: Brumia CHM, Mulder G, orthogonal source derivations. Electroencephalogr Clin
Verbaten MN, eds. Event-Related Brain Research. Amsterdam: Neurophysiol. 1975;39:526–530.
Elsevier; 1991; EEG Suppl. 42:38–46. 89. Le J, Gevins AS. Method to reduce blur distortion from EEGs using
70. Manshanden I, De Munck JC, Simon NR, et al. Source localization a realistic head model. IEEE Trans Biomed Eng. 1993;40:517–528.
of MEG sleep spindles and the relation to sources of alpha band 90. Le J, Menon V, Gevins AS. Local estimate of surface Laplacian der-
rhythms. Clin Neurophysiol. 2002;113(12):1937–1947. ivation on a realistically shaped scalp surface and its performance
71. Schwartz DP, Badier JM, Vignal JP, et al. Non-supervised spatio- on noisy data. Electroencephalogr Clin Neurophysiol. 1994;26:
temporal analysis of interictal magnetic spikes: comparison with 193–199.
intracerebral recordings. Clin Neurophysiol. 114(3):438–449. 91. Babiloni F, Babiloni C, Carducci F, et al. A high resolution EEG
72. Van’t Ent D, Manshanden I, Ossenblok P, et al. Spike cluster method based on the correction of the surface Laplacian estimate
analysis in neocortical localization related epilepsy yields clinically for the subject’s variable scalp thickness. Electroencephalogr Clin
significant equivalent source localization results in magnetoen- Neurophysiol. 1997;103:486–492.
cephalogram (MEG). Clin Neurophysiol. 2003;114(10): 92. Nunez PL. Electrical Fields of the Brain. New York, NY: Oxford
1948–1962. University Press; 1981.
73. Baillet S, Mosher JC, Leahy RM. Electromagnetic brain mapping. 93. Zanow F. Realistically shaped models of the head and their appli-
IEEE Signal Process Mag. 2001;18(6):14–30. cation to EEG and MEG. Ph.D. thesis, University of Twente,
74. Gross J, Kujala J, Hämäläinen M, et al. Dynamic imaging of coher- Enschedé, 1997.
ent sources: studying neural interactions in the human brain. Proc 94. Knösche Th. Solutions of neuroelectromagnetic inverse problems
Natl Acad Sci U S A. 2001;98:694–699. an evaluation study. Ph.D. thesis, University of Twente, Enschedé,
75. Robinson SE, Vrba J. Functional brain imaging by synthetic aper- 1977.
ture magnetometry (SAM). In: Yoshimoto T, Kotani M, Kuriki S, 95. Logothetis NK, Wandell BA. Interpreting the BOLD signal. Annu
et al., eds. Recent Advances in Biomagnetism. Sendai, Japan: Tohoku Rev Physiol. 2004;66:735–769.
University Press; 302–305. 96. Berens P, Keliris GA, Ecker AS, et al. Feature selectivity of the
76. Dale AM, Liu AK, Fischl BR, et al. Dynamic statistical parameter gamma-band of the local field potential in primate primary visual
mapping: combining fMRI and MEG for high resolution imaging cortex. Front Neurosci. 2008;2(2):199–207.
of cortical activity. Neuron. 2000;26:55–67. 97. Devor A, Dunn AK, Andermann ML, et al. Coupling of total
77. Mosher JC, Leahy RM. Source localization using recursively and hemoglobin concentration, oxygenation, and neural activity in rat
projected (RAP) MUSIC. IEEE Trans Signal Process. 1999;47: somatosensory cortex. Neuron. 2003;39(2):353–359.
332–340. 98. Gotman J, Kobayashi E, Bagshaw AP, et al. Combining EEG and
78. Mosher JC, Lewis PS, Leahy RM. Multiple dipole modeling and fMRI: a multimodal tool for epilepsy research. J Magn Reson
localization from spatiotemporal MEG data. IEEE Trans Biomed Imaging. 2006;23(6):906–920.
Eng. 1992;39:541–557. 99. Shmuel A, Leopold DA. Neuronal correlates of spontaneous fluc-
79. Mosher JC, Baillet S, Leahy RM. EEG source localization using tuations in fMRI signals in monkey visual cortex: implications for
multiple signal classification approaches. J Clin Neurophysiol. functional connectivity at rest. Hum Brain Map. 2008;29(7):
1999;16:225–238. 751–761.
110 Part I ■ Basic Principles

100. Cohen, D. Magnetoencephalography. Evidence of magnetic fields 114. Nunez PL. The brain’s magnetic field: some effects of multiple
produced by alpha rhythm current. Science. 1968;161:784–786. sources on localization methods. Electroencephalogr Clin
101. Reite H, Zimmerman JE, Edrich J, et al. The human magnetoen- Neurophysiol. 1986;63:75–85.
cephalogram: some EEG and related correlations. Electro- 115. Stok CJ, Kouijzer WJJ, Peters MJ. Source localization based on
encephalogr Clin Neurophysiol. 1976;40:59–66. EEG’s and MEG’s. In: Proceedings of the 5th Conference on
102. Geselowitz DB. Magnetocardiography: an overview. IEEE Trans Biomagnetics. Vancouver, Canada; 1984.
Biomed Eng. 1979;26:497–504. 116. Leahy RM, Mosher JC, Spencer ME, et al. A study of dipole local-
103. Murakami S, Zhang T, Hirose A, et al. Physiological origins of ization accuracy for MEG and EEG using a human skull phan-
evoked magnetic fields and extracellular field potentials produced tom. Electroencephalogr Clin Neurophysiol. 1998;107:159–173.
by guinea-pig CA3 hippocampal slices. J Physiol. 2002;544(pt 1): 117. Dale AM, Sereno MI. Improving localization of cortical activity
237–251. by combining EEG and MEG with MRI cortical surface recon-
104. Cohen D, Cuffin BN. Demonstration of useful differences struction: a linear approach. J Cogn Neurosci. 1993;5:162–176.
between magnetoencephalogram and electroencephalogram. 118. Cohen D, Cuffin BN, Yunokuchi K, et al. MEG versus EEG local-
Electroencephalogr Clin Neurophysiol. 1983;56:38–51. ization test using implanted sources in the human brain. Ann
105. Okada YC. Inferences concerning anatomy and physiology of the Neurol. 1990;28:811–817.
human brain based on its magnetic field. Il Nuovo Cimento D. 119. Hansen P, Kringelbach M, Salmelin R, eds.MEG: An Introduction
1983;2:379–409. to Methods. New York, NY: Oxford University Press; 2009.
106. Cuffin BN. A comparison of moving dipole inverse solutions using 120. Lopes da Silva FH. Electrophysiological basis of MEG signals. In:
EEG’s and MEG’s. IEEE Trans Biomed Eng. 1985;32: 905–910. Hansen P, Kringelbach M, Salmelin R, eds. MEG: An Introduction
107.Modena I, Ricci GB, Barbanera S, et al. Biomagnetic measurements to Methods. New York, NY: Oxford University Press; 2009.
of spontaneous activity in epileptic patients. Electroencephalogr 121. Plonsey R. Bioelectric Phenomena. New York, NY: McGraw-Hill;
Clin Neurophysiol. 1982;54:622–628. 1969.
108. Hari R, Ilmoniemi R. Cerebral magnetic fields. CRC Crit Rev 122. Rosenfalck R. Intra and extracellular potential fields of active
Biomed Eng. 1986;14:93–126. nerve and muscle fibres. A physicomathematical analysis of dif-
109. Hämäläinen MS. Magnetoencephalography: a tool for functional ferent models. Acta Physiol Scand Suppl. 1969;321:1–168.
brain imaging. Brain Topogr. 1992;5:95–103. 123. Jackson JD. Classical Electrodynamics. New York, NY: Wiley;
110. Lewine JD, Orrison W Jr. Magnetoencephalography and mag- 1962.
netic source imaging. Functional Brain Imaging. St. Louis, MO: 124. Rall W. Electrophysiology of a dendritic neuron model. Biophys J.
Mosby Year Book; 1995:369–417. 1962;2:145–167.
111. de Jongh A, de Munck JC, Gonçalves SI, et al. Differences in 125. Van Rotterdam A. A one-dimensional formalism for the compu-
MEG/EEG epileptic spike yields explained by regional differences tation of extracellular potentials: linear systems analysis applied
in signal to noise ratios. J Clin Neurophysiol. 2005;22(2):153–158. to volume conduction. In: van Eijnsbergen B, Lopes da Silva FH,
112. Plonsey R. The nature of sources of bioelectric and biomagnetic eds. Progress Report No. PR6. Utrecht: Institute of Medical
fields. Biophys J. 39:309–315. Physics TNO; 1978:115–122.
113. Van Rotterdam A. Electric and magnetic fields of the brain: a sys- 126. Plonsey R. Capability and limitations of electrocardiography and
tems analysis approach. Ph.D. thesis, University of Amsterdam, magnetocardiography. IEEE Trans Biomed Eng. 1972;19:239–244.
The Netherlands, 1986.

Analog Signal Recording Principles

ANALOG AND DIGITAL RECORDING appearance in EEG and other neurophysiologic recordings. The
major entry points and barriers to noise consist of the electrode
Digital analysis and storage have transformed modern neuro- interface, the amplifier inputs, and analog filters. The following
physiology in ways that could only be imagined and envied by sections describe from front to back the different components
the founders of EEG. But the “front end” of all recording sys- of the system and the features that contribute to accurate
tems, the portion that interfaces with the patient, remains ana- recording of neurophysiologic signals.
log; and understanding the operation of analog components
can also give insight into aspects of signal processing that are
now performed mostly by digital techniques. Analog refers to
methods of handling data that are physically similar to the orig- EEG and other recording electrodes provide an interface
inal signal. For example, the moving pens and ink traces of the between lead wires, which conduct current as the flow of elec-
original electroencephalographs produced waveforms that mir- trons and human tissues, which transmit it through the move-
rored the electrical potential differences on the scalp, and the ment of ions. Because cutaneous oils and keratin are good
wavy grooves of old-fashioned vinyl records follow the frequen- insulators, placing dry metal electrodes directly on dry skin is
cies and amplitudes of the sound waves produced by stereo sys- usually ineffective. Keratin is often removed by rubbing the
tems. Simple amplification of the microvolt potentials in EEG is skin gently with mild abrasives, and sometimes oils are
an analog process. Analog data “looks like” the original signal, removed separately with alcohol wipes. (Despite such prepara-
which the zeroes and ones of digital representations do not. tions, the epidermis remains the single greatest barrier to cur-
Ideally, each type of data processing would produce a perfect rent flow between the body and the recording system.) In
reflection of the original waveforms. In practice, each is subject addition an ionic solution, or electrolyte, is placed between the
to its own limitations and potential for inaccuracy, which will skin surface and the electrode. As with human extracellular
be discussed in the following chapters. More detailed discussion fluids, the most abundant ions in the electrolyte are sodium
of the material here can be found online in the Guidelines of and chloride. The consistency of the electrolyte ranges from
the American Clinical Neurophysiology Society (1) and in those free-flowing solutions to thick sticky pastes. The latter are used
of the International Federation of Clinical Neurophysiology (2). to help hold cup electrodes mechanically against the scalp. The
former are chosen when the cup is attached by other means
THE HOSTILE WORLD OF CLINICAL and must be filled with electrolyte through a small hole in the
Contact between the electrolyte and the electrode causes
Figure 6.1 shows the approximate amplitudes of electromyo- opposite charges to line up at the metal–ionic boundary. The
graphic activity (EMG), EEG, and evoked potential signals in simplest model for this arrangement is the “electrical double
relation to the many sources of noise that constantly threaten to layer,” first proposed in the 1850s by Helmholtz. The double
overwhelm them. This vast range of voltages requires the use of layer contributes to the stability of the interface and also rep-
a logarithmic scale. So, for example, ECG recorded from the resents a capacitance that augments the apparent current
head can be a thousand times larger than a brainstem evoked flow. At times, especially with new electrodes, the electrical
potential. Within electrified buildings, the 50- or 60-Hz electric double layer undergoes sudden, poorly understood transi-
field on human bodies can be 10,000 times larger than the EEG. tions that produce electrode “pops” and a variety of bizarre
Some physiologic signals, such as EMG and the standing poten- oscillations.
tial on the retina, can also represent noise in relation to smaller The junction between the metal electrode and the ionic elec-
signals that we need to see behind them. trolyte also develops a half-cell potential, which reflects the
At first glance, it may seem impossible that any interpretable electrical force of incipient chemical reactions between them. It
recording could be performed through noise that is orders of also represents half of an electrical battery. The half-cell poten-
magnitude larger than the signal of interest. This chapter dis- tial can be much smaller than that in a flashlight battery and
cusses some of the ubiquitous artifacts in Figure 6.1, the fea- still be enormously larger than the EEG. Ideally this would not
tures of the analog front end that are designed to defeat them be problem, because the half-cell potentials on any pair of
(while allowing faithful recording of the desired signal) and the matching electrodes would be identical, and thus would simply
ways in which they may nonetheless make an unwanted cancel out when those electrodes are connected to an amplifier.
112 Part I ■ Basic Principles

Intracranial strip and depth electrodes are generally made

from stainless steel or platinum. The more favorable electrical
properties of platinum tend to be offset by its greater cost.
With modern amplifiers the high impedance and polarization
of stainless steel represent less of a problem than they did in
the past.


The lead wires, jackbox, and cables are basically passive compo-
nents whose function is transparent. Excessive separation of the
lead wires should be avoided. The capacitance of extremely long
cables can attenuate very high frequencies. These effects can be
ameliorated by placing the amplifiers inside the jackbox or even
directly on the electrodes and by using fiberoptic connections
to replace electrical cables. In exceptional circumstances, such
Figure 6.1 Potential sources of noise in neurophysiologic recording. as inside the bore of a high-field MRI system, resistors should
Sources may represent noise in some contexts and signals in others. be placed in series with every electrode lead to prevent induc-
Note the logarithmic scale. tion of excessive current. When troubleshooting, it is worth
remembering that if things mysteriously go wrong, the com-
monest culprit is a bad lead wire, cable, or connector.
However, half-cell potentials can be altered in a number of
ways, including drying out of the electrolyte, its dilution by ELEMENTARY ELECTRONICS
sweat or other fluids, relative differences in temperature, or
exposure of different metallic surfaces by wear and tear. These Classically the building blocks of the EEG front end are ampli-
potentials are also affected by movement of the electrode fiers, filters, switches, and potentiometers. The top portions of
against the electrolyte and the skin, which disrupts the electri- Figures 6.2 through 6.6 sketch the simplified “transfer functions”
cal double layer. The effect of movement is reduced by using of an amplifier, a voltage divider, a high filter (which, confusingly,
cup-shaped electrodes, which hold a larger reservoir of moist
Chemical interactions at the electrode surface consist of
redox reactions, in which an electron is donated or removed
from an ion. Since the product of such interactions is com-
monly a toxic species, such as hydroxyl ions or chlorine gas, care
must be taken to avoid injury of the underlying tissues. The
safest operating zone for recording and stimulating electrodes is
one where they carry only a limited alternating or balanced
pulse current, small enough to be transmitted entirely through
the double-layer capacitance rather than through redox reac-
tions. Passage of direct current through the double layer will
more readily produce chemical changes at the metal surface,
including electrical polarization, which impedes the free flow of
current in one direction more than another, increases imped-
ance at low frequencies, and may be associated with a substan-
tial change in the half-cell potential.
Scalp and other cutaneous electrodes are usually made
from gold—which is inert and therefore produces smaller
half-cell potentials—or silver coated with a layer of silver Figure 6.2 Amplifier. The top sketch shows the amplifier as just a
chloride. Disposable stick-on electrodes almost always use “black box” that magnifies the size of the signal. At bottom left is the
a silver–silver chloride interface. Combined with a chloride standard circuit symbol for an amplifier, with two inputs on the left and
electrolyte, the latter represents the most stable electrode an output on the right. At bottom right is a graph of the output. It shows
available for neurophysiologic applications, resistant to polar- a straight line representing a gain of 2.0 at the top of the y-axis, because
ization and with the best characteristics for low-frequency and the gain is the same at all frequencies. This simple “single-ended”
direct-current recording. Occasionally the exposure of metallic amplifier has asymmetric inputs: one terminal is “hot” and the other is
silver to the light pulse from photic stimulators produces usually connected to system ground (the bent trident symbol). The
synchronous photoelectric discharges, which may be puzzling inherent asymmetry prevents accurate rejection of noise signals, even
to novice interpreters. when they are equal at both inputs.
Chapter 6 ■ Analog Signal Recording Principles 113

Figure 6.3 Resistance bridge. The circuit symbol for a resistor [R] is a
cylinder. The two resistors here are labeled R1 and R2, and both have
the same value. Current from the AC input signal at the top flows
through the resistors to system ground (the bent trident symbol) at the
Figure 6.5 Sketch of how changing frequencies produce the LF behav-
bottom. The output voltage graph is a straight line with a gain of 0.5,
ior graphed in Figure 6.4. At low frequencies (A) the capacitor imped-
because a pure resistive circuit has the same gain at all frequencies.
ance becomes very high, so the capacitor and signal source practically
drop out of the circuit. At high frequencies (B) the capacitor impedance
becomes very low, and the resistor practically drops out of the circuit.
an engineer would call a low pass filter), and a low filter (LF)
(which an engineer would call a high pass filter). The amplifier
(Fig. 6.2) makes the desired signal larger. The purpose of the LF Amplifiers (Fig. 6.2) have the primary purpose of magnify-
is to block frequencies so low that they carry more noise than ing the signal from a range measured in microvolts up to sev-
signal, and thus are better removed. The high filter blocks fre- eral volts, which can do something useful such as moving
quencies so high that they carry more noise than signal. These mechanical pens or undergoing analog-to-digital conversion to
simple functions remain the basic components of analog a stream of numbers. The ratio of the signal coming out of the
recording. For this discussion the amplifier will remain mostly amplifier to its size going in is called gain. For both analog and
a “black box,” which is the way electrical engineers prefer to use digital systems gain should be in the range of 100,000 or more.
modern integrated circuits anyway. The voltage divider and fil- In neither case, however, is gain easily apparent from the system
ters are represented in the form of their simplest analog com- output. This is because the EEG signal is universally shown in
ponents: resistors and capacitors. terms of sensitivity, the ratio of microvolts to millimeters on a
paper or electronic display.
Figure 6.3 shows a simple voltage divider or potentiometer
in the form of a two-resistor network. A resistor is a circuit ele-
ment with two characteristics: it partially blocks the flow of cur-
rent according to Ohm’s law (voltage = current resistance),
and this relationship is independent of signal frequency.
Potentiometers are useful when for any reason voltage must be
reduced rather than increased (i.e., the gain should be less than
one). The signal enters the two-resistor network as the AC volt-
age source at the top and passes through the resistors to system
ground. According to Ohm’s law, if the resistors share the same
current and have equal values, the voltage output at the middle
of the network must be half the voltage at the top. Since the
function of the resistors is the same at all frequencies, the graph
shows a steady gain of 0.5 all the way across.
The middle and bottom of Figure 6.4 show an LF composed
of one resistor and one capacitor. A capacitor consists basically of
two metal plates that are close together but separated by an insu-
Figure 6.4 Low filter (LF). The low-frequency waveform coming out is lator. Current cannot cross the insulator directly, but positive
smaller than the one going in. The circuit symbol for the capacitor (C) and negative charges attract and repel electrostatically on the
is a pair of parallel lines. Current from the AC input signal at the top plates, producing AC current flow in and out of the capacitor.
flows through the capacitor and the resistor [R] to system ground at the The behavior of the capacitor changes predictably with fre-
bottom. The log–log output voltage graph shows a horizontal line indi- quency, according to the equation Z = 1/(2 FC). Here Z repre-
cating a gain of one on the right side, where frequencies are above the sents the impedance of the capacitor to current flow. Impedance,
cutoff. Below the cutoff frequency, gain drops steadily as frequency which is a generic term for all obstacles to electric current, goes
increases. down as either frequency F or capacitance C goes up.
114 Part I ■ Basic Principles


In addition to gain, neurophysiologic amplifiers require three
other essential properties: linearity and flatness across the entire
frequency range of possible signals, high common-mode signal
rejection, and high input impedance. Any reasonable commer-
cial device should have excellent linearity. The other require-
ments are explained below.
Removing the enormous fields caused by AC wiring—and
by other electrical noise (Fig. 6.1)—is the purpose of the
Faraday cages used with the earliest electroencephalograms and
of shielded rooms still required for modern magnetoen-
cephalograms. As an alternative, patients have occasionally
been connected directly to earth ground. Faraday cages are
unacceptably clumsy and grounding patients is prohibitively
dangerous in medical environments, so more practical alterna-
Figure 6.6 High filter (HF). The circuit is simply an inverted LF. The tives have been found. The key to effective recording of small
high-frequency waveform coming out is smaller than the one going in. neurophysiologic signals is the fact that, large as the noise may
The log–log output voltage graph is a horizontal line for a gain of one be, most sources of electrical noise are of similar amplitude in
on the left side, where frequencies are below the cutoff. Above the cut- nearby regions of the body. Thus the ECG potentials at adjacent
off frequency, gain drops steadily as frequency decreases. EEG electrodes are almost identical—at least, in most normal
individuals. The large AC field produced by capacitive coupling
between protoplasm and AC power systems is similar all over
LF gain changes strikingly with frequency. The right side of the body. Potentials that are similar at different electrode sites
the graph in Figure 6.4 shows a value of 1, meaning that the sig- are called common-mode signals. Common-mode signals are
nal passing through is unchanged. On the left side of the graph removed by very precisely subtracting the whole combination
the signal comes through at lower and lower amplitude as the of potentials at one electrode site from the combination of
frequency drops—essentially, low frequencies are filtered out. potentials at another, leaving behind only those potentials
In the middle of the graph is a break point where the imped- that are different at the two “hot” electrodes—the so-called
ance of the capacitor matches the resistance of the resistor. This differential-mode signal. This process is illustrated in Figure 6.7.
break point is the nominal filter frequency, cutoff frequency, or The simpler amplifiers used in radios, stereos, and other famil-
3-dB point.1 iar electronics (sometimes referred to as single-ended ampli-
The behavior of analog filters far from the cutoff frequency fiers) are incapable of the very precise subtraction needed, so
is sketched for an LF in Figure 6.5. It may be understood most that all neurophysiologic amplifiers have some version of the
easily by remembering the equivalence of the capacitor imped- symmetrical inputs in Figure 6.7. The latter are called differen-
ance and the resistor resistance at the cutoff point in Figure 6.4. tial, balanced, or instrumentation amplifiers to distinguish
At lower frequencies the impedance of the capacitor is so high them from the single-ended variety. The two “hot” inputs in
that the AC signal has less and less influence on the output ter-
minal, and the output approaches a short circuit to system
ground. At higher frequencies, however, the impedance of the
capacitor becomes much lower than the resistance, so that the
system ground has less and less influence on the output
terminal—which is now practically short-circuited to the signal
source. The flat passband with a gain of 1.0 reflects this short
The high filter (HF) circuit (Fig. 6.6) and its transfer func-
tion are simply the reverse of the LF. The elegant symmetry of
the filter curves is seen only in log–log displays, which is one
reason for preferring this type of graph over linear representa-
tions. Merging the HF and LF curves would form a complete
passband of 1.0 to 100 Hz.
Figure 6.7 Differential amplifier with balanced “hot” inputs and sys-
tem ground. The vertical symmetry of this more complicated circuit
1Readers familiar with decibel notation may notice that the output of the RC fil- allows precise rejection of large common-mode signals that are equal at
ter is reduced by only the square root of 2 when the resistive impedance equals
both “hot” inputs. The relatively small differential-mode signal—the
the capacitive impedance, whereas it is attenuated by a factor of 2 in the two-
resistor network. Explaining the difference requires the use of imaginary num- small square at the far left—is the only signal that passes through the
bers, which will not be attempted here. amplifier.
Chapter 6 ■ Analog Signal Recording Principles 115

Figure 6.7 are symmetrical and equally active, despite the uni- would have infinite input impedance. Although perfect ampli-
versal practice of labeling one of them as “active” and the other fiers do not exist, the highest practical impedance is preferred
as “reference” in many EEG recording montages and in for neurophysiologic recording. Figure 6.8 shows why. Imagine
EMG/NCV studies. Their electrical behavior is indistinguish- an amplifier in which the two inputs have an impedance of
able; the only difference is in terminology. The “ground” or 10,000,000 , connected to two “hot” electrodes with imped-
neutral connection in Figure 6.7 is normally present during all ances Z E1 of 1,000 and ZE2 of 10,000 , respectively. Both elec-
neurophysiologic recording (although modern amplifiers may trodes are picking up an equal 60 Hz common-mode noise
appear to function passably without it). signal of 10 mV. Assuming that the input impedances of the
In EEG, a single “ground” or system reference is shared by all amplifier are still much larger than the electrode impedances,
recording channels and is almost universally placed on the fore- the simplified equation in Figure 6.8 allows calculation of the
head. The ground must be distinguished from the recording ref- result. The unequal electrode impedances will produce a differ-
erence, used in constructing referential EEG montages. The ential-mode noise signal at the amplifier inputs of 0.01
“reference” in referential montages is the second hot terminal of (10,000 – 1,000)/10,000,000 = 0.000009 V or 9 V. This noise
the three-terminal differential amplifier, not ordinarily the will be faithfully magnified by the differential amplifier to pro-
“ground.” For digital EEG, the initial recording is most often duce visible widening of the EEG tracings—just as seen in
performed using a recording reference in the vicinity of the Figure 6.9B.
central-parietal region. With modern electronics, it is possible Since it would be impossibly cumbersome to make all the
to use a single digital recording reference (or one of the regular electrode impedances exactly equal, we make sure all electrode
scalp electrodes) rather than a separate recording reference and impedances are low—in which case the difference between
ground. But noise rejection will be better using both, provided them cannot get too large—and make the amplifier input
there is provision for them in the jackbox. impedances as high as practical. Recent guidelines specify that
The classic differential amplifier takes the common-mode the differential input impedance of the amplifier must be at
noise signal as it comes—whatever its size—and then attempts least 100 M (1). The CMRR should be at least 80 dB
to subtract it out. But smaller common-mode voltages would (10,000:1) at the highest sensitivity of the amplifier when the
make the task easier. A more recent technique takes the common- common-mode signal is applied between both inputs and neu-
mode signals seen by all of the amplifiers on the head, averages tral (1). Amplifier input impedances higher than this can be
and inverts them, and injects the result back into the body obtained without much difficulty, but their performance is
through the former “ground” electrode. Since the necessary cur- liable to be degraded by the capacitance between conductors in
rents are measured in microamperes, no risk is involved. This the lead wires and cables.
method, still referred to anachronistically from its original ECG If the only problem caused by unbalanced (i.e., excessively
usage as “driven right leg,” can cancel and reduce excess noise in large) electrode impedance were 50 or 60 Hz noise, it would be
some applications. It also removes the “ground” label and with it easy to deal with using notch filters. But Figure 6.9 illustrates a
the possibility for a misunderstanding to harm the patient (see
section “Electrical Safety and Ground”).
The ability of an amplifier to pass differential-mode signals
and block common-mode signals is described by its common-
mode rejection ratio (CMRR) and is usually expressed in deci-
bels (dB). Thus an amplifier that reduces the common-mode
signal by a factor of 100,000 (105) relative to the differential-
mode signal has a CMRR of 100 dB. Modern integrated circuit
amplifiers can come off the assembly line with CMRR specifi-
cations of 120 dB (1,000,000:1) or higher. However, many
extraneous factors can defeat them and allow noise to get
through despite the nominal superb performance of the differ-
ential amplifier. One mundane consideration is simply that
common-mode noise is not absolutely identical at different
electrode sites, and the difference grows larger as the electrodes
are moved farther apart. Another example is the induction of
AC potentials in the electrode lead wires if they become widely
separated and form an open loop. A third example is the stim-
ulus artifact from the electrical stimulators used in nerve con-
duction studies and evoked potentials. At hundreds of volts, the Figure 6.8 Model of the differential amplifier showing the impedances
latter can produce an enormous difference at the two hot of the electrodes and the two amplifier inputs. Making some reasonable
amplifier inputs. And very commonly, as discussed below, poor approximations, the equation at the upper right represents the way in
electrode application can defeat the differential amplifier. which unequal electrode impedances ZE1 and ZE2 convert a portion of
An ideal amplifier senses electrical potentials without any the common-mode noise voltage VCM into a differential-mode voltage
current actually passing through the inputs. This perfect device VDM, which passes through the amplifier.
116 Part I ■ Basic Principles

Figure 6.9 ICU EEG study with ground recording artifacts in the F4 channels, including apparent slowing and a spike at
the tip of the arrow (A). 60-Hz noise with the notch filter off (B) is an important clue to the existence of a serious prob-
lem. See text for further discussion.

far more insidious consequence. Panel A of this ICU recording lead on the forehead and magnified across the now-unbalanced
appears to show focal delta slowing involving electrode F4, cul- amplifier inputs. Recognizing the presence of ground recording
minating in an apparent spike-wave discharge. Panel B, with the artifacts, and the importance of AC power line artifacts in
60 Hz notch filter turned off, indicates that in fact there is a assisting that recognition, is vital exactly because these artifacts
poor, high-impedance connection at F4 and that 60-Hz com- may not be obvious. In several decades of teaching EEG, none
mon-mode AC noise has entered the involved channels. The of our residents or fellows has ever recognized or deciphered
crucial message of Figure 6.9 is that the apparent focal abnor- ground recording independently. Recording routinely with the
malities also represent common-mode noise. The F4 slowing is AC notch filter on badly skews the odds that many electroen-
actually time-locked to the ECG and reflects an obscure compo- cephalographers ever would.
nent of the t-wave. The apparent spike is due to the small pha- One corollary of the preceding discussion is noteworthy
sic EMG artifact seen in other leads, picked up by the ground for being even more counterintuitive. Poor, high-impedance
Chapter 6 ■ Analog Signal Recording Principles 117

electrode connections do not generally reduce the size of the earth ground at the power station.2 In modern neurophysiology,
EEG signal. Indeed, as in Figure 6.9, channels 9 to 10, they this feature of the power system is the most important concept
commonly make it look bigger. for electrical safety. To operate electrical devices plugged into
the AC wall jack, current flows from the “hot” wire of the power
RANDOM BACKGROUND NOISE line to the “neutral” wire at earth ground. If patients are con-
nected to earth ground, any stray currents (especially those from
In addition to the physiologic and man-made artifacts the hot power line) will happily flow through the patient on the
described above, several sources of random noise can interfere way there. A third wire, which represents a second and more
with recording of very small electrical signals. Of these the direct earth ground connection, is required in hospital equip-
most irreducible is thermal or “Johnson” noise, which repre- ment as well. This additional ground, commonly referred to as
sents Brownian movement of electrons in resistive conduc- “the” ground, increases safety in some situations—but also con-
tors. The magnitude of thermal noise increases with stitutes a risk if the patient is connected to it. Modern safety
temperature, circuit resistance, and recording bandwidth. At standards require that the recording ground on the patient cannot
body temperature, with a high filter of 70 Hz and total elec- form a direct connection to the power line ground(s) or to other
trode resistances of 1000 , thermal noise is about 0.035 V sources of earth ground, including “the” ground wire.
and unlikely to make any difference in recording. But in a Traditionally discussions of electrical safety have focused on
brainstem auditory evoked potential (BAEP) or concentric “ground loops,” which may be formed when the patient is con-
needle EMG study, high electrode resistance can raise the nected to more than one ground electrode through different
minimum noise to the approximate magnitude of the desired pieces of medical equipment. Decades ago, currents flowing
signal. Some examples of possible recording conditions are through the power line ground system could place the different
given in Table 6-1. Note that other noise sources in the ampli- components of that system at somewhat different voltages—in
fier may increase actual noise figures several times above those which case the voltages would attempt to equalize through the
listed. Recent guidelines specify that the noise level of an EEG body of the patient. But the modern requirement that record-
amplifier must not exceed 2 V rms with the inputs con- ing ground on the patient must never be connected to earth
nected to neutral and with a bandpass of 0.1 to 5000 Hz (1). ground means that ground loops can seldom present a safety
risk using commercial systems. However, ground loops may
ELECTRICAL SAFETY AND GROUND contribute substantially to electrical noise.

In electronic circuits “ground” is a reference point that remains

at a virtual zero voltage. Without such a reference point, circuit
operation easily becomes unstable. Many years ago the circuit Prior to the digital era, the vast amount of information con-
ground, and with it the ground electrode on the head, was actu- tained in even a routine EEG could be stored only on paper. The
ally connected to real earth ground through metal plumbing or original signal was amplified up to 1,000,000 times, enough to
other low-resistance conduits. Earth ground is a wonderful power an electromagnetic coil that moved a long mechanical
zero-reference point—a near-infinite reservoir of electrons— pen which carried ink onto fanfold paper. Mechanical pens
which will never be changed by local circuit operation. It is also could move only so fast, setting the upper frequency limit
an unacceptable connection for human bodies, especially in around 70 Hz. They could move only so far, causing the wave-
medical environments. This is because one of the two wires in forms to “square-off” at the limits of the pen range (Fig. 6.10).
alternating current power lines is almost always connected to Special circuitry was used to prevent pens from colliding and
tangling together. The fanfold paper advanced from one tray to
Tabl e 6 . 1 another via a mechanical chart drive with rollers, which could
move at speeds of 15, 30, and 60 mm/sec. The paper was prone
Minimum System Noise Under Representative to ripping or jamming. Ordinarily, EEGs were recorded at
Recording Conditions 30 mm/sec on pages 300 mm wide. The resulting scale of
10 sec/page has been carried over to digital systems. The obso-
Bandwidth Electrode Thermal lete designation of paper speed in mm/sec is so engrained in
Modality (Hz) Resistance Noise older technologists and electroencephalographers that even
( ) ( V) now computer displays are labeled anachronistically in the
same manner.
EEG 35 1,000 0.024
In fully analog EEG systems, constructing different mon-
EEG 70 10,000 0.13 tages requires the use of elaborate mechanical switches. Rows of
BAEP 1,000 1,000 0.13 individual switches are also needed for every channel to set
gain, HF, and LF, and AC notch filters. The available range of
BAEP 3,000 10,000 0.72
EMGa 10,000 100,000 4.1 2Grounding of the power line is necessary because of the astonishingly large
voltage gradients that can develop at different elevations in the atmosphere and
aConcentric needle electrode. over long distances across the earth’s surface.
118 Part I ■ Basic Principles

as compared to about 30 pixels/cm for conventional computer

monitors. To give equivalent resolution, while leaving room for
data labeling, digital monitors would have to be about a meter
wide and incorporate pixel counts that only now are being
approached by high-definition television. Furthermore, paper
can easily be unfolded to provide an overview of several min-
utes’ data while maintaining the same exquisite detail. The
inability of computer displays to show more than 10 seconds
worth of data without severe compromise in frequency resolu-
tion is an underappreciated handicap. It can be overcome only
by knowledgeable adjustment of the display on the part of
Figure 6.10 Detail of two channels from an analog paper writeout. interpreters who understand its limitations for slowly evolving
During a burst of high-voltage frontal theta, the pen in the second chan- activity. Finally, there is no recorded instance of an entire paper
nel reaches the limits of its mechanical range, causing the waveform to EEG tracing vanishing forever on its way to the server, when
“square-off” in the vicinity of the shaded area. Within the shaded area somebody accidentally pushed the “delete” button or when the
is a very narrow spike-like artifact, apparently caused by the collision- operator forgot to hit “record.” And the recent history of digital
avoidance system meant to protect the EEG pens. Note the curvature of recording media suggests that ancient paper records will still be
the high-voltage waveforms and of the artifact, due to the radial nature quite readable when many digital studies can only be viewed in
of the pen movement. a museum.

settings is consequently limited, most notably to the legendary ANALOG RECORDING
HF choices of 15, 35, and 70 Hz. A few generations of “hybrid” Artifacts related to the electrode interface, poor common-
systems are likely to still be in use. In hybrid systems switching mode rejection, and ground recording have been discussed pre-
and the associated notations are done electronically, but the viously. Mechanical pen writers are subject to radial
final writeout remains pen and ink. displacement due to the pens moving in an arc rather than in a
In order to produce a comprehensive overview of cerebral straight line (Fig. 6.10)—at times causing inaccurate alignment
activity, several montages must be used during the course of of high-voltage, sharp-contoured waveforms. The ink may skip
each analog recording. Because the physical settings are not or puddle on the page. The moving paper can transiently slip or
noted automatically on the paper record, analog interpreters jam on the roller, stretching or compressing the EEG wave-
rely on the technologist to write by hand every setting and every forms. Figure 6.10 also shows an unusual artifact apparently
change in parameters at the time of recording. Traditionally, caused by the collision-avoidance system in a hybrid electroen-
every combination of parameters is then replicated during a cephalograph. What appeared at first glance to be spike wave
terminal calibration. Producing a true calibration, which faith- consists only of high-voltage theta slowing plus a spike-like
fully displays the performance of the entire pathway from artifact. When the pen reaches the limit of movement, it para-
inputs to writeout, is a virtue of analog recording that not every doxically moves across its full range in the opposite direction
digital system properly duplicates. (On the other hand, the and quickly back. We have observed this behavior in several dif-
mechanical writeout, the weakest link in the analog recording ferent brands of paper electroencephalograph, where it was
chain, has been abolished from digital systems.) misinterpreted as evidence of epilepsy.
Perhaps the greatest limitation of analog recording is the
dependence on the technologist to make notations accurately,
to recognize abnormalities “online,” and to optimize their dis- REFERENCES
play for the reader by making appropriate changes in montages 1. American Clinical Neurophysiology Society Guidelines.
and recording parameters. This feature does, however, have a
silver lining. The technologist by necessity is far more engaged 2. Ebner A, Sciarretta G, Epstein CM, et al. EEG instrumentation.
in the recording and its eventual interpretation. The International Federation of Clinical Neurophysiology.
Other virtues of paper should not be entirely dismissed. The [Practice Guideline]. Electroencephalogr Clin Neurophysiol Suppl.
spatial resolution of paper EEG is in the range of 100 dots/cm, 1999;52:7–10.

Digital EEG
INTRODUCTION AND HISTORY and epileptologists in particular. Practical digital computation
OF DIGITAL EEG began with the development of ENIAC around 1943 to 1944
(2). As early as the 1950s, the potential advantages of computer
Digital electroencephalography (EEG) refers to the recording, analysis of EEG were foreseen (3). However, the routine use of
storage, and review of the EEG on digital equipment (i.e., computers for EEG acquisition and analysis required several
computers). The process requires conversion of the continuous decades for commercial digital electronics to achieve bench-
analog EEG signal, as described in Chapter 6, into discrete marks that equal or surpass the capabilities of analog EEG
numerical values. Over the past 15 years, digital EEG has grad- recording apparatus. Reliable, power-efficient computation
ually replaced analog systems and is now the standard for clin- began with the semiconductor transistor, invented at Bell Labs
ical EEG. Digital acquisition and processing of EEG allows by William Shockley in 1947. The development of integrated
many advantages, but at the same time imposes limitations that circuits in the 1950s permitted mass production and commer-
should be understood by the clinician to avoid inaccurate inter- cialization. A notable computer algorithm for spectral analysis,
pretation. The practical advantages of digital EEG over analog the Tukey–Cooley version of the discrete fast Fourier transform
EEG are summarized in Table 7.1. In this chapter, we will out- (FFT), was published in 1965 (4). First-generation digital EEG
line the principles of digital signal acquisition and processing systems, with limited digitization technology and screen resolu-
that are important in the design and proper use of clinical EEG. tion, could not match the display properties of paper-based
Clinical digital EEG represents the convergence of two scien- EEG (5), but digital electronics gradually achieved fast sam-
tific fields that evolved in parallel for the last 70 years. In 1929, pling rates and larger data throughput. Finally, in the late 1990s,
Hans Berger first reported his studies on the electroencephalo- digitization rates and computer storage available for reasonably
gram (1). The utility of EEG for clinical diagnosis was quickly priced digital commercial products were sufficient to match
established, and EEG has flourished as a tool of neurologists, routine analog EEG machines.

Tabl e 7 . 1

Advantages of Digital EEG

Efficiency Computer-based display generally permits quick review of many studies

Nondestructive processing Raw data are stored digitally, allowing reconstructed display with different montages, different
filters, and different sensitivity
Precision Can precisely measure frequencies and amplitudes for longitudinal or intergroup comparisons
Archiving Easily make exact copies, retain for easy access on disk drive, or save on optical media
Transmission/comparison Digital copies can be easily and quickly transmitted for second opinion or conferences
Increased frequency range Current and near-future digital sampling rates easily exceed the frequency range accessible
by traditional analog EEG
Reliability Stable and reliable electronics, controlled by microprocessors, decreases likelihood of technical
artifacts compared to analog equipment requiring manual calibration for each channel
Portability Amplifiers can be small enough for patients to carry during ambulatory and long-term studies
Digital signal processing Allows extensive on-line or postacquisition analysis, such as automated detection of spikes and
seizures, topographic maps, and graphical displays of quantitative EEG
Associated software May include patient information databases, report writing programs, and integration with
hospital electronic medical records

120 Part I ■ Basic Principles

DIGITAL EEG COMPONENTS Routine digital EEG has no requirements for specialized elec-
trodes. Current protocols for experimental digital EEG acquisi-
Figure 7.1 shows the basic components of the digital EEG sig- tion at high sampling rate with intracranial implantation may
nal path from the electrodes to the display. Several components derive benefit from specialized microelectrodes that have a
(jackbox, differential amplifier, display) parallel those encoun- smaller contact area and may be able to detect spatially local-
tered in analog EEG machines (Chapter 6). The features of ized multiunit or even single-unit firing trains.
these components that differ from analog EEG will be high-
lighted in the sections below. Components unique to digital Electrode Input Board or Jackbox
EEG systems include (i) the analog-to-digital converter (ADC), Jackboxes for digital EEG typically have a greater number of
which samples the analog EEG signal and converts it into dis- electrode inputs than analog machines. The standard number
crete digital data; (ii) the computer or microprocessor, which of inputs for routine EEG is 32, but jackboxes accepting 64, 128,
stores and manipulates the digital data for optimal review; or even 256 inputs are common for epilepsy monitoring and
(iii) monitors for display of EEG data; and (iv) digital storage intracranial EEG recordings. Additional inputs can accommo-
devices and networking equipment, which allow archiving of date additional EEG electrodes, such as 10-10 modified
EEG data and transmission to remote sites. Some systems International System placements, as well electro-oculogram,
contain video processors and sound cards for simultaneous electromyogram, or other polygraphic inputs (6). Jackbox
acquisition and synchronization of video and audio data. The inputs are labeled with the location of the electrode on the head
hardware and software capabilities of current digital EEG sys- according to the International 10-20 System (6) or numbered.
tems vary significantly between vendors. This review cannot In addition, there are inputs for ground and for a machine ref-
describe all possible digital system specifications but aims to erence (sometimes called recording reference) electrode, which is
provide a summary of the most important features. usually placed in a location not susceptible to large artifacts,
such as the midline central-parietal region.
Electrodes In contrast to analog systems using electrode selectors, most
Along the signal acquisition path, we turn to the first compo- digital amplifiers are “hard-wired,” with the signal from each
nent—the electrodes. These are discussed in detail in Chapter jackbox position going to Input 1 of a specific amplifier. Input
6. One significant difference from analog EEG is the require- 2 for all amplifiers is the machine reference electrode. Some
ment for one additional electrode, the machine reference, digital systems have electronically switched jackboxes to allow
which we shall discuss below in the section on (differential) amplifier input channels to be switched among different elec-
amplification. As with analog EEG, a good recording depends trodes. Because all EEG activity is acquired in relation to the
on low-impedance electrical contact with the scalp. Most machine reference electrode, the data in all channels may be
current commercial EEG acquisition machines have function- distorted if this reference electrode is of high impedance. With
ality for checking electrode impedances. In general, these are a high impedance reference, external noise may overwhelm the
software automated. For each electrode, the impedance is esti- EEG signal, producing a very low amplitude recording with lit-
mated by forming a complete circuit, starting with that elec- tle signal. (This is different from effects of high impedance in
trode, then creating a return path with all the other electrodes any other electrode, which can increase the apparent EEG
joined (so that section of the circuit has an impedance equiva- amplitude through ground reference recording, in addition to
lent to all other electrodes in parallel, and thus small). A small showing the noise.) To prevent this problem, the EEG should be
(microvolt) voltage (typically oscillatory, on the order of 30 reviewed during the recording in a machine reference montage
Hz) is applied, and the current is measured. The impedance of to check for excessive noise in the reference electrode.
that single electrode contact is thereby estimated by the voltage Digital EEG jackboxes may be housed in the same enclosure
(known) divided by the current (measured). For standard as other digital components, such as preamplifiers (see the sec-
scalp electrodes applied with electrolytic paste, impedances tion “Amplifiers”). The short distance between the scalp elec-
should be less than 5 k . This technique relies, of course, on trode and this first amplification step reduces electrical
the assumption that at least some of the electrodes are well interference and artifact in the EEG signal. ADCs may also be
connected. If all electrodes have poor contact, spurious results present in the jackbox itself, as in systems for ambulatory
can be obtained. recording (7). The jackbox also contains or is connected to an

Figure 7.1 The schematic of a digital EEG
machine shows components of the signal Software:
"Jackbox" Differential Anti-alias Analog montage
path from the electrode to the display. digital filter Display
preamp amplifier filter
converter sensitivity
Chapter 7 ■ Digital EEG 121

Paper Pens Low-cut High-cut Montage

Figure 7.2 Schematic of amplifier inputs for analog EEG for
(fixed) (fixed) (fixed)
a longitudinal bipolar montage. One additional electrode
Fp1−F7 + Fp1
Fp1−F7 0.5 Hz 70 Hz input—the ground—is omitted for simplicity.

F7−T3 +
F7−T3 0.5 Hz 70 Hz

T3−T5 +
T3−T5 0.5 Hz 70 Hz

T5−O1 +
T5−O1 0.5 Hz 70 Hz
− O1

impedance meter to allow measurement of electrode imped- (In our figures, we show the electrical symbol for an op-amp,
ance (see the section “Electrodes”) without unplugging the short for operational amplifier. Op-amps do amplify the differ-
electrode wires from the electrode input panel. ence between their inputs, but this is a simplification. Actual
hardware differential amplifiers are more complex circuits, using
Amplifiers a combination of op-amps with other circuit elements, specifi-
The first active components in the signal path are the cally designed and tuned to amplify the difference in voltage
(pre)amplifiers. Prior to digitization, the signals—which are in even when both voltages may be quite large in absolute value.
the microvolt range—must be amplified. This may be done via See more about common-mode rejection below.)
one or more cascaded amplifiers. Modern amplifiers are small, In analog EEG, the montage is fixed at acquisition time by
low-power, single-chip multichannel devices with solid-state the fact that the electrodes are mechanically connected to a set
integrated circuits. Noise (generated by random movement of of differential amplifiers for that montage. Figure 7.2 shows a
electrons within amplifier circuits such as resistors and semi- simple diagram of the arrangement of differential amplifiers for
conductors) should be less than 2 V peak to peak. Amplifiers analog EEG for a longitudinal bipolar montage. Digital EEG
are electrically or optically isolated to prevent conduction of also employs differential amplifiers, but with a significant dis-
current from the EEG machine to the patient through the EEG tinction. Instead of determining the difference between elec-
electrodes. trode inputs at acquisition time, each electrode is acquired
Properly designed, the first amplifier (often called a preampli- relative to one fixed common electrode, the machine reference.
fier) has high input impedance (on the order of mega- to tera- These individual electrode channels are then digitized and
ohms), which allows the source signals to be precisely captured, stored. Later review with tracings displayed in different mon-
without drawing significant current which would create distor- tages can be performed by digitally reconstructing the mon-
tions in the voltage to be measured (8). They also raise the signal tages. Figure 7.3 shows a simple diagram highlighting this
into a range where stray device noise (powerline, television, difference from analog EEG. Note that the vertical dashed line
radio) is negligible and so cannot disturb the signal. Finally, com- separates the fixed, unalterable acquired data on the left, from
mercial ADCs have an expected input range, such as 5 V. To computations on the right showing the data in one of many
take full advantage of the amplitude resolution of the ADC, input possible different montages, which can be easily reconstructed.
signals must be brought up to this voltage range by amplification Also note that the input to the inverting contact for each differ-
(though not over it, since this would “clip” the signal). ential amplifier is the machine reference for all the electrode
For EEG, one of the most important aspects of the amplifi- channels recorded.
cation stage is the use of differential amplifiers. Both analog and There are several other aspects of amplifiers that could
digital EEG employ differential amplifiers. Differential ampli- impact the data acquisition. The common-mode-rejection ratio
fiers have three input connections: (i) inverting input; (ii) non- (CMRR) is a measure of how well the differential amplifier
inverting input; and (iii) ground, typically placed on the reflects the difference between the inputs, especially when both
forehead, and a single output. The output is the difference inputs are large in magnitude. Real differential amplifiers can
between the two inputs multiplied by some constant—the be best modeled as generating an output voltage V0 as
amplification factor, or gain. Gain for clinical EEG machines is 1
usually in the range of 2000 to 20,000, but can be up to V0 Ad (V V ) 2As(V V )
1,000,000. The gain equals output voltage divided by input volt- The coefficient Ad is the differential gain and As is called the
age. Gain is typically expressed as a logarithmic ratio in decibels common-mode gain. The common-mode rejection ratio is
(dB), with gain (dB) = 20(log(Vout/Vin)). For instance, with
inputs of 106 V (noninverting) and 104 V (inverting), and Ad
CMRR 20log10 a b
amplification factor (gain) of 20, the output will be 40 V. As
122 Part I ■ Basic Principles

Figure 7.3 Differential amplifiers Digital

for digital EEG, with each electrode Transmission
input amplified compared to the +
machine reference. Computer soft- ADC …00101010… ‘Fp1’

ware then converts this referential ‘Fp1−F7’
montage to a longitudinal bipolar 10111000
F7 +
montage. See text for details. The ADC …01110010… ‘F7’
− Computer Display
ground electrode is omitted for
simplicity. 00001001 Fp1−F7
T3 +
ADC …01101001… ‘T3’ F7−T3

‘T3−T5’ T3−T5
T5 + T5−O1
ADC …00010011… ‘T5’

O1 +
ADC …01010100… ‘O1’


So, a CMRR of 80 dB, as recommended by American Clinical more components to achieve a faster roll-off in the transition
Neurophysiology Society (ACNS) (9), indicates that the ampli- band, thus a “sharp” cutoff. Such analog filters may be designed
fier will begin to lose accuracy when the absolute magnitude is according to Butterworth or Chebyshev topologies (11).
10,000 (104) times the difference in magnitudes (e.g., 10,001 Other filters can be inserted prior to digitization. DC is not
V – 10,000 V). Most of the common mode signal is 60 Hz commonly used in clinical EEG but may be used for research
noise, and the magnitude of that noise is determined more by purposes. In most commercial systems, filters are inserted to
the electrode impedances than circuit specifications. remove the DC (constant) and very low-frequency components
( 0.5 Hz), which make the record harder to read and require a
Analog Filters larger magnitude resolution for the ADC. These have the addi-
Prior to digitization, analog electrical filtering is applied. The tional benefit of maximally exploiting the ADC resolution, since
fundamentals of analog filter design are discussed in Chapter 6. the average signal is then zero, and the greatest positive and neg-
In contrast to analog machines, digital EEGs are acquired using ative excursions can be designed to fit within the digitizer input
broad bandpass filter settings, typically 0.1 to 70 or 100 Hz. range to take full advantage of the magnitude resolution of the
However, for specialized applications, the bandpass may ADC. Another filter that may appear before digitization is a
include very low frequencies (0.001 Hz) or even direct current power line “notch” filter, which is specifically tuned to eliminate
(DC), as well as high frequencies up to 3000 to 10,000 Hz (10). the frequencies matching that of the power line: 60 Hz in the
The most important filter in this stage is the anti-aliasing fil- United States, 50 Hz in Europe and other countries.
ter as described below. This requires a low-pass filter (passes
low-frequency signals but attenuates high-frequency signals). Analog-to-Digital Conversion
The simplest low-pass analog filter is a resistor (R) and capaci- The stage of conversion of the analog signal to digital represen-
tor (C) in series, with the voltage across the capacitor transmit- tation is the most crucial step in the acquisition process. Design
ted to the next stage, as diagrammed in Figure 7.4. This of an acquisition system requires that the components be
first-order low-pass filter will attenuate high frequencies and matched and appropriate as a whole, so that the final data are
has a cutoff frequency (in Hz) of 1/(2 RC). Commercial anti- not hindered by poor design of any one element in the path.
alias filters may be this simple but are often engineered with Because parameters of the digitization impact acquisition com-
ponents both earlier and later in the signal path, thorough
R understanding of digitizer capabilities and limitations is crucial.
The act of digitization collapses an EEG signal in two independ-
ent axes: time and magnitude.

Vin C Vout Time

Typically, a time-varying signal is digitized at fixed (constant)
discrete intervals. The time intervals are described by sampling
rates. Typical sampling rates are on the order of several hundred
Figure 7.4 A low-pass analog filter is a resistor (R) and capacitor (C) in samples per second (Hz). For 200 Hz, this implies that samples
series, with the voltage across the capacitor transmitted to the next stage. are taken every 5 msec. The time between subsequent samples
Chapter 7 ■ Digital EEG 123

A Figure 7.5 Aliasing. A: A pure sinusoid at 11 Hz

(the solid curve) sampled at 16 samples per second
(Nyquist frequency is then 8 Hz). The sampled data
points can be seen to equivalently match a sinusoid
at 5 Hz (dashed curve). When displayed on a screen
(connected by straight lines), these points would
more closely match the 5-Hz sinusoid than the origi-
nal 11-Hz sinusoid. B: Frequencies above the Nyquist
0 are “folded back” across the Nyquist, as if the paper
0.5 1 were folded with the crease at the Nyquist fre-
quency—best exemplified by the solid curves of the
frequencies between the Nyquist and the sampling
rate. The heavily darkened curve represents the alias-
ing of 11 to 5 Hz from A.


0 Nyquist Sample
(8 Hz) rate

is called the dwell time. The choice of sampling rate impacts (solid curve). Indeed, in practice, when displayed on a screen
how well the obtained data reflect the analog source, with faster (connected by straight lines), these points would more closely
sampling roughly giving more fine detail. A reasonable question match the 5-Hz sinusoid than the original 11-Hz sinusoid. This
is: How fast is fast enough? A well-known theorem from signal phenomenon is known as “aliasing.” Figure 7.5B shows the pat-
processing literature provides a quantitative answer to this tern by which frequencies above the Nyquist are “folded back”
question. The Nyquist theorem states that the highest frequency into the range between 0 and the Nyquist frequency, with the
distinguishable is one half the sampling frequency. The down- example of 11 Hz being folded back to 5 Hz at sampling rate of
side of more rapid sampling is the creation of larger EEG files. 16 samples per second emphasized with the bold arc.
Armed with the Nyquist theorem, one knows the limitations Figure 7.6 shows the deleterious effect of noise above the
of what signal components in the source are capturable, but the Nyquist frequency. Fortunately, methods to eliminate this artifact
question is naturally raised as to the consequence of signals that are quite simple. Electrical filters—specially designed circuits
are above this Nyquist limit. If these were simply eliminated, the composed of elements such as resistors and capacitors—can be
issue would be moot, but sadly high-frequency components used to eliminate undesired frequencies in the analog signal prior
above the Nyquist can actually create nefarious artifacts in the to digitization. In particular, ADC systems incorporate an analog
digitized output. Figure 7.5A shows an example of a pure sinu- low-pass filter prior to digitization, with this filter designed to
soid at 11 Hz (the dashed curve) being sampled at 16 samples pass frequencies below the Nyquist frequency and eliminate
per second (Nyquist frequency of 8 Hz). The sampled data those above it. Such a low-pass filter in this context is termed an
points can be seen to equivalently match a sinusoid at 5 Hz “anti-aliasing” filter.
124 Part I ■ Basic Principles


0 Hz 256
Low−pass filtered

0 0

0 2.0 0 2.0
Noise folded
Digitize Digitize
128 Hz 128 Hz

0 0

0 Seconds 2.0 0 2.0

Figure 7.6 Deleterious effect of noise above the Nyquist frequency. Top left panel shows 2 seconds of “original” (simu-
lated) data composed of a signal at 5 Hz, plus noise with significant components up to 256 Hz. The inset (just above and
to right) shows the power spectrum of this data, showing the white noise and the large signal at 5 Hz. Bottom left panel
shows what would be the result of naïve direct digitization at 128 samples per second, rendering the signal at 5 Hz nearly
unrecognizable within the abundant noise. The signal-to-noise ratio (SNR) has actually been worsened because the noise
above the Nyquist limit (64 Hz = half of 128 samples per second) has been folded back, adding to the noise in the fre-
quencies below the Nyquist, as illustrated in its power spectrum. The right side panels show how anti-aliasing filtering
prior to digitization can prevent this problem. Top-right panel shows the tracing after a low-pass filter (modeled as
Butterworth third order) with cutoff at 50 Hz has been applied. Bottom-right panel shows the result of digitization (again
at 128 samples per second) of this filtered tracing. Frequencies above the Nyquist (64 Hz) have been attenuated by the
low-pass filter and are therefore not folded back, so digitization in this case yields a faithful reproduction of the signal.

The anti-alias filter and ADC in Figure 7.1 are deliberately reflects the fact that two sinusoids with some difference in fre-
joined. The analog hardware anti-alias filter must obviously quency will become out of phase with each other over that time
precede the ADC. The anti-alias filter cutoff must be appropri- frame. For instance, to reliably distinguish a 0.1 Hz difference in
ate for the sampling rate of the joined ADC. The filter must frequency requires a total acquired signal on the order of 1/(0.1
adequately attenuate frequencies above the Nyquist, or else Hz) = 10 seconds. This applies irrespective of the absolute fre-
high-frequency noise will be aliased into the digitized signal. If quency—that is, distinguishing 8.3 from 8.4 Hz or 124.7 from
the cutoff is too low, then signals with frequencies above this 124.8 Hz both require approximately 10 seconds of acquisition.
cutoff and below the Nyquist will be unnecessarily attenuated
and lost, and expense will have been wasted in using an ADC Magnitude or Amplitude
with higher sampling rate than necessary. Digitization occurs along the magnitude axis as well. In this
Another point regarding the time axis regards the “frequency respect, the effect of digitization is typically discussed in terms of
resolution,” a factor important for quantitative analysis of EEG. digitization “depth.” Digitizers presume that each successive
The sampling rate does not impact whether one can distinguish a point is at a known time interval, so this time interval does not
signal as being one of two closely spaced frequencies, for example, need to be stored. Instead, only the value or magnitude at each
8.3 Hz versus 8.4 Hz. Frequency resolution is determined by the successive point is stored. Digital equipment is optimized when
duration of the acquired signal. In general, the duration required storing each successive magnitude in a predefined length. To be
is approximately the inverse of the difference in frequency. This concrete, the output is a stream of 0’s and 1’s. The number of
Chapter 7 ■ Digital EEG 125

Figure 7.7 The practical effect of 4-, 6-, and 8-bit

digitization depth. Note carefully that for the 6-bit
depth, the digital step is most apparent at the peak
and trough, where for each time step the voltage
change is small, but these changes are smaller than
the smallest discrete step and so cannot be repre-

4-bit 6-bit 8-bit

−7 to +7 −31 to +31 −127 to +127

binary digits (bits) set aside for each measurement is the same. the minimum, with 12 bits or more being preferred. The mini-
This storage length can differ between equipment manufacturers. mum amplitude resolution is recommended to be 0.5 V.
Early digitizers used 8, 10, or 12 bits to store each measurement. Regarding the sampling rate, the guideline has no absolute min-
Digitizers now routinely use 16 to 24 bits. The number of bits imum sampling rate but states that “Higher rates are prefer-
determines explicitly the number of distinguishable magnitudes able.” The guideline does suggest that “minimum sampling rate
of the measurements. For example, 8 bits per measurement can (be) three times the high-frequency (anti-aliasing) filter,” which
store up to 28 = 256 different values (0 to 255 unsigned, or –127 is equivalent to saying that for a given sampling rate, the anti-
to 127 signed two’s complement system). Figure 7.7 illustrates aliasing filter cutoff be at 2/3 the Nyquist frequency. Since the
the practical effect of 4-, 6-, and 8-bit digitization depth. ADC rate and anti-aliasing filter are hardware determined,
The three parameters—maximum value (or dynamic range), these are under the control of the manufacturer but should be
minimum distinguishable difference, and digitization depth— investigated and documented clearly when purchasing deci-
are interconnected, and not independent. Choosing any two sions are considered.
specifies the third. With a 12-bit linear digitizer (212 = 4096 dif-
ferent levels), in order to be able to represent magnitudes up to ADC Hardware
1 mV, then the “resolution”—the smallest detectable differ- Some of the basic principles of the hardware involved in AD
ence in magnitude—will be (2000 V/4096) ~0.5 V. The more conversion are worth mentioning. First, no AD converter oper-
bits available, the more accurate each sample. For very low- ates instantaneously. The input voltage is typically measured
amplitude signals (e.g., electrocerebral inactivity recordings), over some finite time (obviously less than the time between
ADC resolution less than 0.5 V is necessary (12). samples). The time over which the signal is actually measured is
called the conversion time. Sampling skew, or loss of time axis
ADC Effects on File Size integrity, can occur if the ADC samples each channel sequen-
Both sampling rate and digitization depth have an impact on tially, since it takes some time to sample one channel and con-
acquired file lengths. For example, for 32 channels with digiti- vert the amplitude to a numerical value before it moves on to
zation of 16 bits (2 bytes) per sample and a sampling rate of 400 the next channel’s sample. Sampling skew is most problematic
samples per second, 1 hour of EEG (3600 seconds) will gener- with high-frequency activity and may cause misalignment of
ate (32 2 400 3600) = 92,160,000 bytes = 90,000 kB = high-frequency events such as spikes between the first and the
87.89 MB. A full day of EEG with these settings would generate last channels sampled. Most ADCs use an input circuit called a
~2 GB. Increasing the sampling rate to 1000 Hz would increase sample-and-hold, employing a capacitor to “hold” the voltage
the file sizes by 2.5, generating ~5 GB per day. during this conversion time. Then, comparator circuits are
applied to bracket the “held” voltage to the desired precision. A
Guidelines second method is to have a series of ADCs, one for each chan-
There are specific guidelines for minimum requirements of dig- nel, all triggered by a single computer command. Some digital
ital EEG acquisition, specifically touching on ADC. The ACNS instruments have combinations of these two methods, with
guideline (9) suggests that 11-bit depth (including sign bit) is ADCs for blocks of four to eight channels using the “sample
126 Part I ■ Basic Principles

and hold” method. Since large ADCs with high throughput are amplifier to be plugged into a network jack near the patient and
expensive, multiple ADCs may be more cost-efficient. to stream data to an acquisition machine anywhere on the net-
There are various ADC hardware structures available. These work. Advanced LTM recorders may have bluetooth connectivity
structures include direct conversion or flash, successive approx- and can wirelessly stream data to nearby storage (15) and often
imation, ramp-compare, delta-encoded, pipeline, and sigma- have battery backup and local storage to allow patients increased
delta. Some of these structures actually involve complex mobility while ensuring uninterrupted data acquisition.
combinations of simpler ADC designs. For example, the
sigma–delta converter first oversamples at a very high rate, dig- Video and Audio Acquisition
itally filters and may use a Flash ADC, then finally downsamples Simultaneous video and audio recording is most commonly
to the desired sampling rate. done in LTM for epilepsy and in the intensive care unit, to allow
The most obvious trade-off of AD converters is between correlation of EEG features with behavioral events (16). Most
sampling rate and number of channels. The product (rate digital EEG systems today utilize digital cameras and store EEG
channels) is basically a fixed constant, with this constant being and video data to computer hard drives, although some older
proportional to the expense of the digitizer. With current hard- systems with analog video and audio recorded to videotape are
ware digitizers, it is often possible to sample very fast (2 kHz) still in operation (17). Early proprietary video capture systems
on a handful (~16) of channels, or at slow rates (256 Hz) on have largely been replaced by standard digital video recording
several dozen channels (64 to 256). Most commercial EEG equipment, markedly decreasing the cost and increasing the
machines aim for a compromise with the ability to monitor 32 efficiency of video acquisition and review. Digital video
to 64 channels at sampling rates of 256 to 1024 Hz, and these recorded directly to hard drives can be accessed randomly,
rates can be user programmed according to the task at hand. allowing the reviewer to quickly jump to any portion of the
record, rather than having to fast-forward or rewind a series of
Calibration videotapes. Video equipment can be customized to optimize
Digital calibration differs significantly from analog calibration. recording for specific types of monitoring. Camera options
The ADC must be calibrated so that it “knows” how large to include analog versus digital, color versus black-and-white, low
represent each voltage point. A 50- or 100- V signal (usually a light versus infrared, fixed versus pan-tilt, and wide-angle ver-
30-Hz sine wave) is passed through each amplifier. The ADC sus remote zoom. Many systems have controls for remote pan-
takes the amplifier outputs and assigns a 50 V value to the tilt and zoom functions. While early systems required custom
peak of each channel. All other amplitude values are deter- cabling for video, modern systems can transmit and control
mined based on this initial calibration. Calibration information digital video over standard TCP/IP cables.
is saved with each EEG record. The digital EEG software should Video and audio are usually encoded into MPEG, MPEG2, or
contain a voltage cursor that allows the user to determine the MPEG4 formats. These are standard formats for digital video,
voltage of individual waveforms. This cursor is then applied to differing in resolution and compression algorithms. All data
the known signal from the signal generator to determine should include time markers so that EEG, video, and audio can
whether the system is properly calibrated. be synchronized precisely (18). Video should be synchronized to
Square wave calibration pulses can sometimes be generated EEG in the millisecond range, or frame-by-frame. Video files
using a digital EEG machine to verify the amplifier and analog contain large amounts of data; 24 hours of video recording can
filter functioning. The calibration signal should be time-locked be 8 to 30 GB, depending on resolution (usually 320 240 pix-
to the ADC clock (13). The square wave calibration should be els or 640 480 pixels), color depth, frame rate, and data com-
examined in the machine reference montage with all digital fil- pression algorithm. EEG machines for video-EEG recording
ters off. Biologic calibration may not be possible with digital should be able to store at least 24 hours of continuous video and
EEG machines. EEG data. Currently available systems, with hard drive capacities
of 200 GB or more, may allow continuous recording for more
Ambulatory and Long-Term EEG Portable Recorders than a week. Because of the size of video files, the entire video is
Systems for acquisition of long-term EEG (hours to days) can not usually archived; rather, video segments of interest (seizures
have a variety of configurations. The number of hours of EEG and behavioral events) are clipped and archived.
that can be stored depends on the number of channels recorded,
recorder battery life, and memory storage capacity. Fully portable Computer and Software
ambulatory recorders, designed for home EEG recording, are The EEG acquisition and digitization devices described above are
small enclosures containing an amplifier, ADC, batteries, and connected to a central computer, which can be either a standard
data storage (usually flash memory card) (7,14). Modern systems desktop personal computer or a laptop computer. The computer
can store up to 96 hours of continuous 16–32-channel EEG data. contains a central processing unit (CPU) that performs calcula-
When recording is complete, data are downloaded to a computer tions, random access memory (RAM) which holds data tem-
via a universal serial bus (USB) or proprietary connection for porarily while they are being manipulated, and a hard disk drive
review. for permanent data storage, as well as a monitor, keyboard,
In systems for long-term monitoring (LTM) of epilepsy, mouse, network connectivity devices, and various connections for
amplifiers/ADCs can be connected directly to a computer via removable media such as CDs, DVDs, and external hard drives.
USB or other cable, or have TCP/IP connectivity that allows the Most EEG vendors currently use standard commercially available
Chapter 7 ■ Digital EEG 127

computers, running on commonly used software platforms such

as Microsoft Windows or Macintosh operating systems.
Most of the variability in modern digital EEG systems comes
from proprietary software applications that allow customization
of EEG acquisition as well as manipulation and display of the EEG
and video data through a graphical user interface (GUI). Features
of review software are discussed in more detail below. In addition,
vendor-provided software applications may include databases for
patient and study information, interfaces to hospital patient infor-
mation systems, report-generation software, security packages,
and a variety of postacquisition signal processing software.
Because the central machine is a standard computer, review sta-
tions can also run other software that may enhance laboratory
productivity such as word processors, voice recognition software,
and hospital electronic medical record systems.

Digital EEG can be displayed in a variety of methods. The
recorded signal can be output to an oscilloscope, printed out on
paper, reconverted to analog signal via digital-to-analog con-
version (DAC) and written by traditional pen-writing systems,
or viewed on a computer monitor. The most practical and ver-
satile method is display on a computer monitor.
Visual display on screen is also a digital process. Whether dis-
played on older CRT (cathode ray tube) or modern digital LCD,
images are drawn as discrete pixels. In fact, current display pixel
resolution may be the limiting factor impeding precise represen-
tation of the signal characteristics to the electroencephalogra-
pher (19). The EEG signal consists of points discretized in time
and voltage but obviously not necessarily in adjacent pixels, so
intermediate pixels are filled to construct lines connecting the
points. The top diagram in Figure 7.8 is an idealized example of
a set of such points and where they might lie on a pixel display. Figure 7.8 Effects of horizontal pixel resolution on EEG display.
Modern software draws lines between these points using line- Supposing two time steps per pixel-width (e.g., 200 Hz digital sampling
drawing routines based on Bresenham’s algorithm (20). rate and 1000 pixel-widths for 10 seconds = 100 pixel-widths per sec-
Screen resolution is described in terms of the number of hor- ond). Top panel shows idealized signals at 12.5, 25, 50, and 100 Hz (the
izontal and vertical points or pixels. In 2010, typical mid-range digitization Nyquist frequency). Bottom panel shows how these fre-
display monitors may have 1280 pixels across with 1024 pixels quencies would be actually drawn with pixels. As can be seen, 50 Hz is
vertically. We examine horizontal resolution first. Display of 10 the highest frequency that can be visualized with these pixel settings
seconds of EEG on a screen implies each second being allotted (and not terribly well) with frequencies above 50 Hz appearing
128 pixels, in the best case. If the sampling rate was 256 samples “smudged.” In analogy with the digitization Nyquist frequency, the
per second, then horizontally there would be two EEG points highest frequency represented by pixels can be said to be half the num-
per pixel. Suppose there were (for simplicity) 100 pixels per sec- ber of pixels per second (100 pixel-widths per second can represent
ond, and an EEG channel sampled at 200 Hz, with signals up to signals up to 50 Hz).
100 Hz (which is at the Nyquist frequency, and representable in
the digitized data stream). Figure 7.8 shows a set of points with
sine waves up to 100 Hz. Unfortunately, by the Bresenham line occur when viewing fast frequency activity, such as 60-Hz arti-
drawing algorithm, these signals will not be able to be traced fact or EMG activity, and can occur even when following rec-
out, instead resulting in blurred uninterpretable lines, as shown ommended guidelines for video resolution. When such video
in Figure 7.8 (bottom). In general, the maximum frequency aliasing artifacts are suspected, decreasing the number of sec-
faithfully representable on a pixel display is one half the number onds displayed per page (e.g., from 10 to 5 seconds) or increas-
of pixels per second. In this case, that would be (100/2) = 50 Hz. ing screen resolution will allow display of all samples in each
Display of EEG on monitors with inadequate resolution (i.e., second and resolve the artifact (Fig. 7.9).
lower than the sampling rate) can result in aliasing, the intro- We now turn to the vertical display resolution. The dimen-
duction of spurious waveforms caused by “undersampling” the sion of pixels on current LCD screens is on the order of 0.25
EEG signal (21). This is analogous to violating the Nyquist the- mm. If we try to replicate the vertical scale of analog EEG,
orem when choosing a sampling rate. Aliasing artifacts typically then 7 V/mm translates into 7 V over 4 pixels. Thus, each
128 Part I ■ Basic Principles

Figure 7.9 Aliasing by inadequate moni- A

tor resolution. A: EEG was viewed on a
hospital computer with monitor resolution
set at 800 600 pixels. An unusual 9.5-Hz
artifact is seen in channel T6 – reference.
B: When the EEG was viewed on a moni-
tor with screen resolution 1600 1200
pixels, this signal is seen to be a sinusoidal
60-Hz artifact.

50 µV
1 sec

50 µV
1 sec

pixel spans about 2 V in vertical height. With current digi- the sensitivity to 2 V/mm. With our 1024 pixels in the ver-
tizers (12- or 16-bit), the amplitude resolution is typically tical dimension, and viewing 20 channels, we see that we have
better than 0.5 V, perhaps as low as 0.03 V. Figure 7.10 approximately 50 pixels per channel. Using our standard 7
shows an idealized example with a low-amplitude signal that V/mm, EEG signals of 50 V can be displayed. Signals
ranges from 1 V to –1 V. This variation is lost with pix- larger than 50 V will begin to spill over into adjacent
els of 0.25 mm at a sensitivity of 7 V/mm, resulting in a flat channel display, causing overlap of tracings in adjacent chan-
line. Conversely, to achieve 0.5 V per pixel requires setting nels and making the EEG more difficult to interpret.
Chapter 7 ■ Digital EEG 129

should be able to print at resolutions of 600 dots per inch

(DPI), or approximately 350 samples per second. Optimally, the
system should be able to print a subset of channels on each page
to maximize vertical resolution.


Digital EEG has several advantages over analog recordings (22). In
routine clinical practice, the ability to reformat montages, change
sensitivities and “paper speed,” and manipulate filter settings dur-
ing EEG review are some of the major advantages. For analog
EEG, what the technologist records is what the electroencephalo-
grapher reviews, with no ability to change or reformat data to bet-
ter localize abnormalities or exclude artifacts. Digital EEG allows
review of the data “as acquired” (as the technologist viewed the
data as they were being recorded) or with a nearly limitless array
of post hoc montage, sensitivity, and filter settings.
Digital reformatting improves EEG interpretation. Levy et al.
(23) demonstrated that interrater agreement for “as-acquired”
digital EEG was the same as for paper EEG, with weighted kappa
scores of approximately 0.65 (good agreement). When elec-
troencephalographers were given the ability to reformat the
digital EEG, however, interpreter agreement improved to kappa
scores of 0.8 (very good agreement). Montage reformatting was
most useful for distinguishing normal variants from abnormal
patterns, identifying artifacts, and classifying abnormalities as
focal or generalized. Correlation of digital EEG with simultane-
ously recorded video also enhances the ability to detect and cor-
Figure 7.10 Effects of vertical display resolution on EEG. Idealized
rectly identify environmental and biologic artifacts.
example with a low-amplitude signal that ranges from 1 to –1 V.
Digital EEG can be reviewed page by page (usually 10 seconds
This variation is lost with pixels of 0.25 mm at a sensitivity of 7
per page) or in continuous “scroll” mode. Scroll mode is typically
V/ mm, resulting in a flat line.
used during acquisition or when reviewing simultaneous video
and EEG, and page mode is used during review. Page mode is
much faster; up to 5 to 10 pages can be presented each second.
Several methods can be used to improve signal display. Only superficial features of the EEG can be determined with this
Screen size for review stations should measure 20 diagonally or rapid paging speed, but it may be useful for screening for seizures
more. Dot pitch (the diagonal distance between display pixels) or determination of sleep states. In page mode, EEG abnormali-
should be as small as possible. Display is best when the moni- ties or comments at the edge of the page may be missed. This can
tor’s “native” screen resolution is used. Large (20 or 24 ) mon- be avoided by overlapping the pages by several seconds.
itors with high resolution (1600 1200 pixels, or widescreen
1920 1200 pixels) are currently reasonably priced, and higher Patient Information
resolution larger monitors are available for those with larger At a minimum, the patient’s name, date of birth, medical record
budgets (Table 7.2). A minimum horizontal resolution of 100 number, laboratory EEG number, and the date the study was
pixels per second is recommended (9). For lower resolution performed should be recorded and stored with the EEG (24).
monitors, decreasing the number of seconds per screen can This information can be stored in the same file as the EEG data
allow display of every data sample point. Similarly, for vertical itself, as a separate text file, or in a database linked to the EEG
resolution, one can reduce the number of channels displayed or data file. Most systems allow recording of more detailed infor-
use a “zoom” feature to hone in on waveforms of interest. Dual mation including patient history, medications, and recording
monitor displays may be a particularly attractive option. Two conditions. More advanced systems may include report-gener-
side-by-side 24 widescreen WUXGA monitors, rotated 90 ation software, allowing reports to be stored with the digital
degrees, give an effective resolution of 2400 1920 pixels. This EEG record.
allows viewing of nearly every data point in 10 seconds of EEG
sampled at 250 samples per second. Data Labeling
Digital EEG signals can also be printed on paper. Printers During recording, the EEG data should be automatically
using fan-fold paper allow printing of a continuous EEG record labeled with time of day, all montage, sensitivity, and filter set-
similar to that acquired with analog systems. Laser jet printers tings used by the technologist during recording, and start and
130 Part I ■ Basic Principles

Tabl e 7 . 2

Commonly Used Computer Monitor Resolutions

Code Name Aspect Ratio Resolution

VGA Video Graphics Array 4:3 640 480
SVGA Super Video Graphics Array 4:3 800 600
XGA eXtended Graphics Array 4:3 1024 768
SXGA Super eXtended Graphics Array 5:4 1280 1024
WXGA Widescreen eXtended Graphics Array 16:9 1366 768
UXGA Ultra eXtended Graphics Array 4:3 1600 1200
WSXGA Widescreen Super eXtended Graphics Array Plus 16:10 1680 1050
WUXGA Wide Ultra eXtended Graphics Array 16:10 1920 1200
WQXGA Wide Quad eXtended Graphics Array 16:10 2560 1600
WQUXGA Wide Quad Ultra eXtended Graphics Array 16:10 3840 2400

end of procedures such as photic stimulation and hyperventila- Montage reformatting allows the electroencephalographer
tion. The system should support entry of comments by the to mentally reconstruct the three-dimensional EEG by combin-
technologist, both as stored “codes” for common events (e.g., ing views of three or more montages, such as longitudinal bipo-
“eyes open,” “awake,” or “seizure”) and as free text. Comments lar, transverse bipolar, and referential montages.
can also be generated by digital analysis programs, such as spike Additional reference montages can be easily constructed by
and seizure detection algorithms. taking the average of data from two or more electrodes (25).
During review, the electroencephalographer should be able to Laplacian montages (source current density montages) are also
view all notations made during acquisition, as well as add com- easy to implement using “nearest neighbor” methods to
ments. A scrollable list of comments can allow the reviewer to approximate the laplacian (26). Montages can include elec-
immediately find an EEG segment of interest. A particularly use- trodes of the modified 10-10 International System, as well as
ful feature is the ability to create a “context-sensitive” comment, other physiologic monitors such as electro-oculogram, electro-
which stores both a text annotation and the montage, sensitivity, cardiogram, oxygen saturation, and electromyogram.
and filter settings at which the EEG was originally viewed. In addition to changing montages, digital EEG software usu-
ally allows the reviewer to optimize vertical resolution by select-
Montage Reformatting ing only certain channels or groups of channels for display or
Since digital EEG data are acquired using a hard-wired referen- “hiding” channels contaminated by artifact. Individual or
tial montage (input 2 for all channels is the machine reference, groups of channels can be moved or reordered to improve com-
or REF), the computer can perform calculations on the stored parisons between homologous brain regions. Extra space or
data to create any desired montage, nearly instantaneously. For “gaps” between channel groups can facilitate review, albeit at the
example, transforming the recorded montage to a bipolar mon- expense of vertical resolution. Using different colors (red vs.
tage requires the following calculations: blue) for left versus right hemisphere channels can help the elec-
troencephalographer quickly localize waveforms but may intro-
Channel Recording Calculations Display duce subtle visual illusions that can mislead the interpreter.
Montage Montage
Sensitivity Changes
1 Fp1-REF Fp1-REF – Fp1-F7
(F7-REF) In analog EEG, sensitivity is the input voltage ( V) divided by
the output pen deflection (mm). For example, a 50- V signal at
2 F7-REF F7-REF – F7-T3 a sensitivity of 7 V/mm would be 7.1 mm high. In digital EEG,
(T3-REF) the same waveform on the display may not actually measure 7.1
3 T3-REF T3-REF – T3-T5 mm. Rather, the amplitude scale is displayed as a vertical icon
(T5-REF) (calibration line) on the monitor, and the amplitude of EEG
4 T5-REF T5-REF – T5-O1 signals is determined relative to this reference amplitude. The
(O1-REF) scale can be changed to provide a variety of sensitivity settings.
Sensitivity changes can be performed in two ways with a
5 O1-REF digital EEG machine. Rarely, the technician can use a software
Chapter 7 ■ Digital EEG 131

sensitivity switch during EEG acquisition to change the actual channel. Others contain a calibration icon that shows the size of a
amplifier sensitivity (e.g., for intracranial vs. surface EEG 50- or 100- V signal. The calibration icon can be moved around
recordings). More commonly, the display sensitivity is changed the screen and compared to various waveforms. Most programs
during acquisition or review. To decrease the amplitude of the also contain a feature that allows a user to draw a line from peak
waveforms on the page, the software program plots each data to trough of any given waveform to get a numerical value for the
point on the monitor as if it were half as large. Sensitivity amplitude in microvolts, or the duration in milliseconds. Finally,
changes can be made for a single channel, groups of channels, or more advanced waveform measurements may include average
the entire page. amplitude or frequency for a selected segment of a channel of
EEG, or even a spectrogram of the component frequencies.
Time Scale Changes
Most EEG software programs display EEG at 10 seconds per Digital Filtering
page, or for larger or widescreen monitors, 25 to 35 mm per sec- One of the major advantages of digital EEG is the ability to
ond. The number of seconds displayed per page can be apply, remove, and reapply a variety of digital filters to enhance
adjusted, however, to optimize interpretation. For example, dis- or minimize activity in certain frequency bands, without alter-
playing 20 seconds or more per page will enhance slow activity ing the original EEG data. This is analogous to inserting electri-
and allow analysis of slow periodic complexes or prolonged cal filters in the acquisition path (see Chapter 6). The field of
events such as seizures. Spreading out the EEG to show only 2 digital time domain filters is vast and complex, but some basic
to 5 seconds per page will spread out faster frequencies, similar principles are important to illustrate (Table 7.3).
to increasing the paper speed on an analog machine. This The three common methods of digital filtering include two
allows more precise analysis of time relationships of signals in time domain methods (finite impulse response [FIR] and
adjacent channels, for example, to analyze the propagation of infinite impulse response [IIR]) and one frequency domain
epileptiform spikes. In addition, time cursors can usually be method using the fast Fourier transform. While most commer-
placed on the screen to measure time relationships exactly. cial systems do not incorporate all the filter choices discussed
Many software programs allow the number of seconds per page below, it is important to understand how different digital filter
to be varied from 1 to 100 seconds. types can be designed to mimic (or improve upon) the analog
Unfortunately, even the largest monitors cannot adequately filters discussed in Chapter 6. Signal processing programs and
display more than 20 to 30 seconds of EEG, limiting simultane- toolboxes used for quantitative EEG analysis (e.g., MatLab) will
ous analysis of long segments of EEG (e.g., evolution of seizures incorporate a larger number of filters and the ability to design
over several minutes). Printing the EEG to fan-fold paper may custom digital filters.
enhance the interpretation of long EEG epochs. Alternatively,
most EEG software programs allow split-screen views to allow Time Domain Filters
side-by-side comparison of noncontiguous segments of EEG. Processing discrete time sequences to accentuate or diminish
frequency bands is made possible by a technique called digital
Waveform Measurements filtering. We begin with an extremely simple example, the
Waveforms can be measured in several ways. Some software dis- two-point moving average. If one takes a sequence and
plays the maximal “peak-to-peak” amplitudes available for each replaces each value by the average of it and the previous value,

Tabl e 7 . 3

Time Domain Digital Filters

Filter Type Pros Cons

FIR Absolutely stable Computationally more expensive
Linear phase, constant group delay Require higher order for equivalent
(peaks and other features of wave performance
shapes will have the same delay Difficult to reproduce effect of analog
in two channels processed by the filters
same filter)
IIR Computationally cheaper Potentially unstable (stable in typical
Can obtain good performance with Nonlinear phase, group delay is
low order frequency dependent
Can simply and exactly reproduce
analog filters
132 Part I ■ Basic Principles

one generates a new sequence. If xk is the input at time t k and the phase delay (in radians and degrees) as a function of (reduced)
yk is the output at t k, then frequency. Note also that the phase shift is a straight line (linear).
1 This simple algorithm (average of 2 points) acts as a filter that
yk 2 (xk xk 1)
attenuates higher frequency inputs and preserves low frequencies
Figure 7.11 shows this concept applied to pure sinusoids of and is thus termed a low-pass filter. A four-point moving average
several different frequencies, up to the Nyquist limit. The squares would act similarly but would more dramatically attenuate high
(connected by solid curve) are the original data points. The cir- frequencies. A simple filter that would attenuate low frequencies
cles (connected by dashed curve) are the points that result from and preserve high frequencies would be a two-point difference fil-
the two-point moving average. We note several features: (i) For ter (computing the difference between the current input point
an input of a pure sinusoid, the amplitude of the output may be and the previous input point).
altered, but the frequency of the output is the same as the input. The graph of the magnitude of the output as a function of the
This is a general feature. (ii) The amplitude of the output dimin- input frequency is known as the transfer function. Several differ-
ishes with higher frequency. This is specific to our example. ent regions or bands in this transfer function have specific names.
Other filters might attenuate low frequencies and preserve high The frequency band in which the output is preserved (ratio of
frequencies. (iii) The output is shifted to the right. In particular, output to input near 1) is called the passband. The frequency
the peaks in the output are all shifted ½ point to the right relative band in which the output is severely attenuated (ratio near 0) is
to the input sinusoid. However, the apparent shift in phase termed the stopband. The frequency band inbetween is known as
increases with higher frequency. (iv) At the Nyquist frequency, the transition band. The slope of the transfer function in the tran-
the output is identically zero, for this example. sition band is called the roll-off. The graph of the phase shift ver-
Figure 7.11B shows graphs of these effects. The top shows the sus frequency is known as the phase plot. There is a delay in the
ratio of the output amplitude over the input amplitude, as a func- output, which is named the group delay and is generally expressed
tion of frequency (in this case, the absolute frequency divided by in number of sample points. The group delay may vary as a func-
the sampling frequency, as this effect does not depend on absolute tion of frequency. In general, the group delay at a given frequency
sampling rate, so 0.5 is the Nyquist frequency). The bottom shows is the (negative) slope of the phase plot at that frequency. The


0 1

0 Seconds 1

−100 0
0 Seconds 1 0 Frequency 0.5

0 0 0°

0 Seconds 1
−π/4 −45°

−100 −π/2 −90°

0 Seconds 1 0 Frequency 0.5

Figure 7.11 Filtering by FIR filter, two-point moving average. See text for explanation.
Chapter 7 ■ Digital EEG 133

order of a digital filter is the number of prior input values that are choices yield a low-pass filter. Note that these coefficients yield
used in the computation of the output. Thus, a two-point aver- a small passband and a cutoff near 0.05 (i.e., low frequencies
age is first order and a four-point moving average is third order. below 0.05 sampling rate are passed, but frequencies above
These are the principal features that are compared and contrasted 0.1 sampling rate are mostly attenuated).
between different digital filters. In this manner, given a set of filter coefficients, one can com-
pute the transfer function and phase plot. There are algorithms for
Finite Impulse Response. The examples described so far are part
generating the coefficients corresponding to traditional analog
of a class known as FIR filters. When the input consists of an
circuits (Butterworth, Chebyshev type 1 and 2, Bessel, etc.) with a
“impulse”—a single nonzero input value in a sequence of zero
given cutoff frequency. The inverse problem—having a desired
inputs—the output sequence (response) will be nonzero for
transfer function, and computing the required order and coeffi-
only a finite number of elements before it necessarily returns to
cients—is beyond the scope of this chapter. For details, see Ref. 27.
zero. This is a direct consequence of the fact that the output yk
In general, for IIR filters, phase shift is nonlinear, so group
depends only on current and prior inputs xk, xk – 1, … This fea-
delay varies with frequency. In fact, particular coefficients for
ture defines the FIR class of filters. FIR filters can be designed to
IIR filters can create a situation in which two finite sinusoidal
be low-pass, high-pass, band-pass, and many other forms. A
signals of different frequencies can appear reversed in order
particular feature of the FIR class of filters is that the phase shift
after being filtered, compared to the input sequence.
is a linear function of frequency, and thus the slope is constant,
Techniques to mitigate group delay for both FIR and IIR filters
so the group delay is the same at all frequencies.
are discussed below.
Infinite Impulse Response. The other class of digital filters is
known as IIR filters. They are also known by other names Undesired Effects
including autoregressive (AR) or recursive. A simple example is Group delay can produce unexpected distortions in digital sig-
1 4 nals, particularly for multichannel comparisons, as with EEG.
yk xk yk 1 Figure 7.13 shows the distorting effect of different filter settings
5 5 when visually comparing channels. Group delay can be prob-
That is, the output (yk) is computed as 1/5 the current input
lematic, and it affects both FIR and IIR filters. As mentioned
(xk) plus 4/5 the prior output.
previously, the group delay generally increases with increasing
Using prior output values is what gives this filter class its
order of the filter. Since FIR filters typically require significantly
important characteristics. For an input sequence that contains an
higher orders to obtain the same transfer function attenuation
impulse (a single nonzero value), this filter will yield an output
as IIR filters, group delay can quickly become very problematic
that continues to have nonzero outputs. Figure 7.12A shows such
for FIR filters. Techniques to eliminate it have been developed.
an IIR (though the values trend toward and approach zero).
Filters that are acausal can eliminate group delay. Acausal filters
This filter applied to another input sequence known as a step
are ones in which future time points as well as past time points
function is shown in Figure 7.12B. This plot will probably
are included in the computation, as opposed to the filters dis-
remind many readers of the graph of voltage charging a capac-
cussed previously that include only past time points, which are
itor. In fact, this filter is the digital analogy of a simple single
called causal. While causal filers can be used to filter live on a
resistor and capacitor circuit arranged as a low-pass filter.
point-by-point basis (but introducing the group-delay artifact),
Digital filters are specified by the coefficients that multiply
acausal filters cannot be implemented live. They can, however,
prior inputs and output to compute the current output. Given
be implemented with a short delay.
a set of coefficients, it is straightforward (though possibly com-
If the entire time series of points is known and immutable,
plicated) to determine the transfer function and phase plot. In
having been acquired and stored, then acausal filters are simple
general, the equation for a filter of order n follows:
to implement and can completely eliminate the group delay. In
yk 5 b0xk 0 b1xk 1 b2xk 2 p bnxk n6 fact, a simple scheme is to filter the entire time series in the for-
5a1yk 1 a2yk 2
p anyk n 6 ward direction, then reverse and use the same filter coefficients
in the backward direction. This eliminates the group delay and
The transfer function (H) and phase are generated by calcu- can even eliminate the phase distortion, achieving zero-phase
lating (or having a graphing program display) distortion. This forward–backward zero-phase filtering works
5b0 b1e i2 x
b2e 2i2 x p 6 for both FIR and IIR filters. A further trick can be employed to
H ` i2 x 2i2 x p 6` minimize transients at the end of the time series, producing the
51 a1e a2e
same result whether filtering forward first or backward first (28).
Frequency Domain Filters
phase tan e f Frequency domain filters use FFT to remove unwanted fre-
Real(H) quencies from the signal (4,29). According to Fourier’s theo-
For our simple example, b0 = 0.2 and a1 = 0.8. With these rem, any time series waveform can be modeled as the sum of a
coefficients, the transfer function (solid) and phase (dashed, in set of sinusoidal waveforms, each with different frequency,
radians) are graphed in Figure 7.12C. It demonstrates that these amplitude, and phase. In frequency filtering, the input signal is
134 Part I ■ Basic Principles

+1 1



0 0
0 10 0 Frequency 0.5

+1 0 0°


−π/4 −45°

0 −π/2 −90°
0 10 0 Frequency 0.5

Figure 7.12 Infinite impulse response (IIR) filter. A: Filter response to an impulse input. The open squares connected by
wide-dash line shows the input, which is zero everywhere but time point 0 where it is 1, and the filled squares connected
by narrow-dash line plots the output, which approaches but never reaches zero, having a persisting, infinite response to
an impulse input, thus the term “infinite impulse response.” B: IIR filter applied to another input sequence known as a
step function. Again, the open squares connected by wide-dash line shows the input, and the filled squares connected by
narrow-dash line plots the output. C: Transfer function and phase plot of IIR low-pass filter. See text for details.

broken into short segments and then transformed into its com- the signal will remain stationary during the entire epoch.
ponent sine wave signals using FFT. To filter the signal, the coef- Breaking the EEG into short epochs, however, may introduce dis-
ficients of the unwanted frequencies above or below a specified continuities between subsequent epochs. “Windowing” proce-
cutoff frequency are set to zero, and the inverse Fourier trans- dures can minimize such discontinuities. Windowing is the
form is then computed to obtain the original EEG signal minus application of a weighting function (e.g., Hamming window) to
the unwanted frequencies (Fig. 7.14). taper the ends of each epoch toward zero. Because windowing
Frequency domain filters have several limitations. Frequency necessarily loses some information at the ends of the epochs,
filtering cannot be performed until the full epoch to be filtered analysis is commonly performed on multiple epochs overlapping
has been acquired, precluding true real-time filtering. To improve each other by 25% to 75%. The length of the analyzed epoch, the
computational speed, certain limitations may be placed on the windowing function utilized, and the degree of overlap all influ-
time series, such as a requirement to contain exactly 2n members ence the computational speed of frequency domain filtering.
(where n is a positive integer). Finally, if the EEG signal is not sta-
tionary (i.e., has variable frequency components) over the epoch, Digital Analysis and Data Reduction
the FFT may introduce artifactual high-frequency noise (30). For Digital EEGs can be very rapidly reviewed by displaying pages
this reason, frequency domain filtering is typically performed at rates of 1 to 10 pages (10 seconds of EEG) per second. The
over short epochs (e.g., 1 second) to improve the likelihood that maximal speed of review is determined by the speed of the
Chapter 7 ■ Digital EEG 135

which can be performed in real time by the computer proces-

sor, while off-line analyses can be more complex. Digital pro-
cessing can highlight specific features of the EEG using
graphical displays, remove or correct artifacts, or detect com-
mon patterns such as seizures or sleep states.

Spike and Seizure Detection

Original signals Algorithms to detect spikes and seizures are commonly used in
long-term video-EEG monitoring, intensive care unit monitor-
ing, and ambulatory EEG. The algorithms are designed to detect
specific features in the digital data, such as the sharpness and
field of an epileptiform discharge or the change in frequencies or
amplitude at the beginning of a seizure (34,35). Other methods
use neural networks that can be trained and updated with a
database of seizures to improve detection accuracy (36,37). The
detection algorithms mark the sections of data which contain
events of interest. The electroencephalographer can then review
this prescreened data rather than the entire EEG recording. The
detection algorithms are not perfect, and may either miss
seizures or spikes, or have high numbers of false detections.
Filtered Topographic Display (Mapping)
Topographic maps display the EEG voltage at various locations on
the head (38). In order to construct smoothed maps, interpolation
Not filtered
techniques such as spline interpretation or nearest neighbor rules
are used (39,40). The maps are most accurate when large numbers
of electrodes are used. Topographic maps can be used to display
the voltage topography of epileptiform activity (41).

Dipole Source Modeling

0 Seconds 1.0
This technique aids in localizing the generators of EEG events,
such as epileptiform discharges (42,43). Multichannel digital
Figure 7.13 Distorting effect of different filter settings when visually EEG is modeled as being generated by a single (or a small num-
comparing channels. The top panel shows (synthesized) traces for two ber) dipole source. Each dipole source produces particular scalp
sequential channels from a bipolar derivation with an idealized sharp- voltage waveforms when placed in a mathematical model of the
and-slow wave complex, but with significant noise on the top channel. head. The locations and orientations of the dipoles are adjusted
It is tempting to try to eliminate the noise in that one channel by selec- to produce the best fit between the mathematical model and the
tively applying a low-pass filter. The bottom panel shows the result of actual EEG. Source localizations can be coregistered with data
an FIR low-pass (cutoff 25 Hz) filter, with order of 50 (a not unreason- from MRI.
able order for an FIR filter). The high-frequency noise is indeed
reduced, but a delay is induced by the filter. The (downgoing) peak of Spectral Analysis/Graphical Displays of Quantitative EEG
the top tracing is delayed by 1/ 8th of a second. All traces are 1-second Spectral analysis involves analyzing the various frequencies that
duration at 200 samples per second. make up the digital EEG signal (44). Typically, the fast Fourier
transform is used to calculate the frequencies present in the EEG.
The spectral analysis of a segment of EEG can then be plotted on
a topographic map or displayed as a graph of power in a partic-
network, computer processor, and graphics card. Even at very ular frequency versus time. This application is increasing in use
rapid review speeds, however, review of an entire 24-hour EEG for long-term EEG monitoring in the intensive care unit (45).
(8640 10-second pages) will take 15 to 60 minutes or more. A
variety of data reduction methods can be used to identify EEG
segments of interest for more rapid interpretation. DATA SHARING, STORAGE, AND
Digital EEG can be analyzed quantitatively to assist interpre- NETWORKING
tation, automatically detect events, and produce graphical dis-
plays (31–33). A full discussion of these techniques is beyond EEG File Formats and Data Sharing
the scope of this chapter, but they are included in later chapters. EEG files acquired with current commercial EEG systems are
Signal processing can be performed either on-line, as EEG is typically stored in proprietary formats that require vendor-
acquired, or off-line. On-line analyses are limited to those specific software to open and view the files. This limitation can
136 Part I ■ Basic Principles

Spectrum − power
Figure 7.14 Filtering in the fre- Original FFT
quency domain. The top left shows
the original signal (this example has
512 samples per second). The fast
(discrete) Fourier transformation 0
(FFT) converts this signal into a fre-
0 Hz 256
quency domain representation (top
right). The spectrum is multiplied by × (multiply by)
a desired filter function (middle
right), in this case preserving all fre- 0 Seconds 1
quencies between 2 and 32 Hz—a
bandpass—and zeroing all other fre-
quency components. The resulting Filtered
spectrum (bottom right) is then 0 Hz 256
inverse Fourier transformed back into
the time domain, yielding the filtered
signal (bottom left). The Fourier 0
domain representation also has FFT
phase information—this has been
omitted for clarity of illustration.

0 Seconds 1
0 Hz 256

cause significant problems in transferring data between differ- EEGs can be archived on a variety of media, including CD,
ent EEG laboratories for review, and even for maintaining read- DVD, tape, hard drives (internal or external), servers, or network-
able archives as file formats are changed over time. There are attached storage devices. These media vary greatly in size
several approaches to overcome this Tower of Babel problem (amount of data which can be stored), cost, convenience (time
(46). Most commercial systems can export EEGs to a compact required to archive and ease of use), and durability. Use of digital
disk (CD) or digital video disk (DVD) with a limited-function media has significantly decreased space requirements and cost of
version of the EEG reader software, so the disk becomes a self- EEG storage. CDs (storing up to 700 MB) and DVDs (4.7 GB,
contained reader station usable on any computer. Others pro- dual layer 9 GB) are most commonly used when EEGs need to be
vide a “light” version of review software that can be installed on transferred to another electroencephalographer for interpreta-
multiple computers at low cost. Some allow EEGs to be con- tion. Archiving to removable media is low cost but inefficient, as
verted to open-source formats (e.g., European Data Format each disk holds only a small amount of data, must be “burned”
[EDF], European Data Format plus [EDF ] (47), Extensible individually, must be placed back in the computer for the data to
Biosignal [EBS] (48), ASCII (49), or the American Clinical be reviewed, and may become unreadable over years. External
Neurophysiology Society’s Technical Standard 1 (50)) that can tape drive storage devices have large capacity and rapid backup,
be read by a variety of freely available reader programs and but data retrieval can be slow because the tape must be trans-
some commercial systems. These open formats may not repli- ferred back to the review station hard drive before review. Tapes
cate all of the features available in the original EEG software, may also deteriorate over time. External hard drives are inexpen-
and most have been adopted by only a few vendors. Finally, sive and provide more rapid access to data than tape drives. Hard
there are several commercial programs that can read EEG files drive costs have decreased significantly over recent years, making
collected by different commercial EEG systems, but these may storage of large amounts of data on servers or network-attached
not read all EEG file types and can be quite expensive. storage devices economically feasible. EEG data stored on a server
can be rapidly retrieved from any computer on the network.
Data Storage
Maintaining digital EEGs archives (e.g., updating to new media
Once EEG is reviewed, video and EEG data are typically clipped or to new file formats) can be costly and time-consuming.
for events of interest and archived to long-term storage. The size of an EEG file depends on the number of channels
Archiving schemes vary; most labs save all routine EEGs, but recorded, sampling rate of the ADC, and the duration of the
only clips of significant events for long-term video-EEG moni- recording. One hour of 32-channel EEG sampled at 200 sam-
toring. Some save all continuous 24-hour EEG data, but contin- ples per second produces an approximately 50-MB file, while 1
uous video files are too large to store in a cost-effective manner. hour of video data can be nearly 1 GB, depending on resolution
A laboratory performing 2000 routine EEGs, 1000 video-EEG and compression. Compression techniques can be used to
studies, and 500 ambulatory EEGs will produce more than a reduce the number of bits necessary to represent information,
terabyte of digital data each year. by removing repeated bits. For example, instead of encoding 10
Chapter 7 ■ Digital EEG 137

Tabl e 7 . 4

Transmission Speed of Digital EEG and Video

EEG Video
Type Data Rate Transfer Time Real Time Transfer Time Real Time
Gigabit Ethernet 1 Gbps 0.5 sec Yes 5.3 sec Yes
T4 (fiberoptic 274.76 Mbps 1.7 sec Yes 19.2 sec Yes
trunk line)
Fast Ethernet 100 Mbps 4.8 sec Yes 52.8 sec Yes
Ethernet 10 Mbps 48 sec Yes 8 min 48 sec Yes
Cable modem 2.5–6 Mbps 1 min 20 sec– Yes 14 min 40 sec– Yes
3 min 12 sec 35 min 20 sec
T1 (trunk line) 1.54 Mbps 5 min 20 sec Yes 57 min 8 sec Yes
DSL 1–3 Mbps 2 min 40 sec–8 min Yes 29 min 20 sec– Maybe
88 min
Modem 56 kbps 2 hr 38 min No 26 hr 20 min No
802.11n 100 Mbps 4.8 sec Yes 52.8 sec Yes
802.11a/g 54 Mbps 8.9 sec Yes 1 min 38 sec Yes
801.11b 11 Mbps 43.6 sec Yes 8 min Yes
4G cellular 6–100 Mbps 4.8 sec–1 min 20 sec Yes 52.8 sec– Yes
14 min 40 sec
3G cellular 144 kbps–2.4 Mbps 3 min 20 sec– Yes 36 min 20 sec– Maybe
55 min 33 sec 10 hrs 10 min

Digital EEG, 32 channels, 16-bit resolution, sampled at 256 Hz, 1 hour = 60 MB = 480 Mbit. MPEG4 color video, 640 pixels 480 pixels resolution, 30 frames per
second, compressed, 1 hour = 660 MB = 5280 Mbit.

binary 1 and 5 binary 0 values, the compression method width 11 to 100 Mbits/sec), although the latter does not incur
encodes that a binary 1 value is repeated 10 times and binary 0 high costs of running cable to all EEG locations and may be
value 5 times (111111111100000 vs. 10-1 5-0). The destination adequate for low-volume laboratories.
computer then decodes the information and expands it back to An efficient and organized network requires careful plan-
the original number of bits. While it is possible to compress dig- ning, usually in conjunction with trained and certified hospital
ital EEG signals slightly, in general the savings are small and network managers. Th