The Responsive Brain: The Proceedings of the Third International Congress on Event-Related Slow Potentials of the Brain

J.R.K.

Participants

Abraham, Lt-Col P. Southampton, U.K.

Becker, Dip-Ing W. Ulm, German Federal Republic

Blowers, Mr G. Hong Kong

Cohen, Dr J. Chicago, U.S.A.

Cooper, Dr R. Bristol, U.K.

Crow, Dr H. J. Bristol, U.K.

Deecke, Dr L. Ulm, German Federal Republic

Donald, Dr M. Kingston, Canada

Donchin, Professor E. Urbana, U.S.A.

Dubrovsky, Dr B. Montreal, Canada

Fenelon, Mr B. Newcastle, Australia

Fruhstorfer, Dr H. Marburg, German Federal Republic

Ganglberger, Dr J. Vienna, Austria

Grözinger, Dr B. Ulm, German Federal Republic

Hazemann, Dr P. Paris, France

Hillyard, Dr S. La Jolla, California, U.S.A.

Järvilehto, Dr T. Helsinki, Finland

Kamp, Mr T. Utrecht, Netherlands

Karrer, Dr R. Chicago, U.S.A.

Knott, Professor J. R. Iowa City, U.S.A

Kohn, Dr H. Piscataway, U.S.A.

Leifer, Dr L. Cambridge, U.S.A.

Lelord, Professor G. Tours, France

Little, Mrs B. Keele, U.K.

Loveless, Dr N. Dundee, Scotland

Low, Dr M. Vancouver, Canada

Marsh, Dr G. Durham, U.S.A.

McAdam, Dr D. Rochester, U.S.A.

McCallum, Dr W. C. Bristol, U.K.

Näätänen, Dr R. Helsinki, Finland

Naitoh, Dr P. San Diego, U.S.A.

Otto, Dr D. Chapel Hill, U.S.A.

Papakostopoulos, Dr D. Bristol, U.K.

Peters, Mr J. Iowa City, U.S.A.

Rebert, Dr C. Stanford, U.S.A.

Ritter, Dr W. New York, U.S.A.

Storm van Leeuwen, Professor W. Utrecht, Netherlands

Tecce, Dr J. Boston, U.S.A.

Timsit-Berthier, Dr M. Liège, Belgium

Tueting, Dr P. New York, U.S.A.

Walter, Dr W. Grey, Bristol, U.K.

Weinberg, Dr H. Vancouver, Canada

Wilkinson, Dr R. Cambridge, U.K.

Winter, Mr A. L. Bristol, U.K.

Zappoli, Professor R. Florence, Italy

Abbreviations

Editorial note: The following standard abbreviations have been used in this volume. Other abbreviations are defined in the text when first used. A more extensive list of abbreviations used in event related potential research can be found in the Proceedings of the Second Congress (McCallum and Knott, 1973).

BP Bereitschaftspotential (synonymous with Readiness Potential, RP).

CNV Contingent Negative Variation.

DC Direct current or, when used in relation to amplifiers, ’directly coupled’.

EDR Electro-dermal response.

EKG Electrocardiogram.

EMG Electromyogram.

EOG Electro-oculogram.

EP Evoked potential.

ERP Event related potential.

GSR Galvanic skin response.

Hz Hertz or cycles per second.

i.s.i. Inter stimulus interval.

i.t.i. Inter trial interval.

MP Motor potential.

msec Millisecond.

P3 A term used synonymously with P300. The latter is generally to be preferred.

P300 Positive wave with peak latency approximately 300 msec after stimulus.

R Response.

RP Readiness potential.

RT Reaction time.

S Stimulus.

S1, S2, etc. Stimulus one, stimulus two, etc.

S Subject.

sec Second.

TC Time constant.

President’s Introduction

JOHN R. KNOTT

University of Iowa (Professor Emeritus), Iowa City, U.S.A.

Presently, Professor (Neurology), Boston University Medical Center, Boston, U.S.A.

The present assembly to view the responsive brain through its slow potential processes is the third in some 6 years, and marks only the ninth anniversary of the original publication of the existence of these phenomena in man, which was brought about through the serendipity of a key group of workers at the Burden Neurological Institute, led by W. Grey Walter, and including Ray Cooper, Vivian Aldridge, Cheyne McCallum and Arthur Winter. While to some it may appear that holding this Congress in Bristol is akin to carrying coals to Newcastle, none of us would regard it as such, since the phrase ‘slow potentials—or more properly, ‘event-related slow potentials’—covers a far greater semantic field than does the term ‘coals’. Further, it has been our experience in the past that when we carry our data to some common meeting point we all emerge with more than that which we brought, and we find that we each have had generated within us new programs of research, based upon the admixture of the ‘coals’ we all have carried and the various patterns of flames into which they have been fanned by our colleagues.

When one compares the Second (Vancouver) Congress with this one (and both with the original informal meeting in 1967 in Liège), there are three types of ‘progress’ apparent. First, slow potential techniques are being applied to a wider range of problems. Second, there is increasing technical elegance in experimentation. To a noticeable degree, experimental design is emerging from the primitive fixed foreperiod reaction time experiment into more complex behavioral situations. Third, and perhaps most important, is the fact that we, as a collection of scientists with some common interests in sub-electroencephalo-graphic frequencies, have stopped being ‘a CNV group’ and are passing into a stage of being ‘an event-related slow potential group’. As such the polarity and timing of the potentials seem less crucial, and the events have come to have a more critical implication, the goals being the obtaining of information about the responsiveness of the human brain. It is a sign of health that these are being considered collectively as indices of the complex processes of cerebration and not as competing items in the scientific market-place, one of greater value than another. We are not learning more and more about less and less, but we are learning more about more.

The unique feature of ‘live’ demonstrations of certain technical aspects of slow potential recording, utilizing closed-circuit television, will enable Dr Grey Walter to realize a wish he has expressed since 1967 (first in Liège, then in Vancouver), that we design and carry out a ‘critical experiment’ in order to satisfy ourselves regarding certain points involved in the generation of event-related cerebral slow potentials. To what extent ‘experiment-by-committee’ is viable will be seen—but the very execution of this attempt may itself be an important experiment.

We are, at the opening of this Congress, coming face to face with new sets of data, new ways by which these data may have been generated, better—or at least different—methods of analysing them and broader views of what may be regarded as the data themselves. The data are not, however, the end-point. The final goal is an understanding of the functional structuring of the human brain characteristics and its moment to moment activities. We should anticipate that if we keep our own brains responsive we shall be changed, ourselves, by the experiences we shall share with one another.

METHODOLOGY OF SLOW POTENTIAL CHANGES

Chairman’s Opening Remarks on Present State of Methodology

R. COOPER,     Burden Neurological Institute, Bristol, U.K.

Publisher Summary

This chapter discusses the methodology in slow potential work, which can be categorized into characteristics of the signal, the electrodes, the amplifiers and display system, and the signal analysis. The electrodes themselves do not change very much, but the potentials at the skin change as the conducting jelly soaks in. DC amplifiers are of such high stability that few problems remain. When using RC-coupled amplifiers, the time constant reduces the amplitude of the Contingent Negative Variation (CNV) and other slow phenomena. For intracerebral recordings, particularly in man, the polarization of the electrodes causes a major problem. Acceptable metals such as gold or stainless steel act like a capacitor and present very high impedance to DC and low frequency activity. The low frequency activity is reduced 20-fold by input impedance of 1 megohm if the electrode impedance is 19 megohm. Although the impedance might be reduced by changing the surface layer of the electrode by etching or chemical reaction, EEG amplifiers with input impedance of 100 megohm can be used. Using such amplifiers, the CNV and other slow phenomena can be recorded from cortical and subcortical electrodes without too much loss.

Methodology in slow potential work can be separated into 4 areas concerned with: (1) characteristics of the signal; (2) characteristics of the electrodes; (3) characteristics of the amplifiers and display system; (4) signal analysis.

1 CHARACTERISTICS OF THE SIGNAL

These are well known—amplitude less than 100 μV, and bandwidth DC to 100 Hz. What is not yet known although it must be of considerable importance is the output impedance of the cortical generators and the distribution of current flow that enables us to record the potentials on the scalp.

2 CHARACTERISTICS OF THE ELECTRODES

The problems that still exist in scalp recordings are primarily due to slow drift of potential which is troublesome during DC recordings. It seems likely that the electrodes themselves do not change very much but that the potentials at the skin change as the conducting jelly soaks in. In addition, fluid may collect at the site of any abrasion that has been made to improve electrode contact and change the electrical potentials. Drift rates of more than 200 μV per minute are common even with well prepared and applied electrodes.

For intracerebral recordings, particularly in man, the polarization of the electrodes causes a major problem. Acceptable metals such as gold or stainless steel act like a capacitor and present a very high impedance to DC and low frequency activity (many megohms). When used with DC amplifiers the low frequency activity is attenuated by a factor determined by the values of the electrode impedance and the amplifier input impedance (see Cooper et al., 1974 or Cooper, 1971 for discussion). The low frequency activity is reduced 20 fold by an input impedance of 1 megohm if the electrode impedance is 19 megohm. Although the impedance might be reduced by changing the surface layer of the electrode by etching or chemical reaction (in the same way as silver chloride changes a silver electrode) we have used EEG amplifiers with input impedance of 100 megohm. Using such amplifiers the CNV and other slow phenomena can be recorded from cortical and subcortical electrodes without too much loss.

3 CHARACTERISTICS OF AMPLIFIERS AND DISPLAY SYSTEM

DC amplifiers are of such high stability that few problems remain. When using RC coupled amplifiers the time constant reduces the amplitude of the CNV and other slow phenomena (Fig. 1). The amount of attenuation depends upon the shape of the CNV, the interstimulus interval and the time constant. As a general rule of thumb the time constant should be at least three times the interstimulus interval. A serious consequence of a short time constant is the deflection below the baseline as the CNV collapses and measurement of such phenomena as P300 may be in serious error. Performance of amplifiers is often quoted as bandwidth rather than time constant although for CNV work in particular the time constant gives a more graphic description. The bandwidth/time constant relationship is illustrated in Fig. 2.

Fig. 1 The effect of time constant on the CNV. Each trace is the average of 50 trials recorded simultaneously but with different time constants. Repetitive clicks (S2) were terminated by a button press. The bottom trace shows the superimposition of all channels. Note the apparent variation in latency of the first peak of the CNV.

Fig. 2 The relationship between frequency and time constant for various percentage losses of sensitivity (p).

The high frequency cut off depends upon the experiment—whether, for example, cerebral evoked responses are being measured; 100 Hz is reasonable and allows most signals to be recorded.

At the present time practically all averaged slow potentials are written out as a voltage/time graph. For spatial distribution some contour display may be necessary, although such a display would imply that the CNV is a unitary phenomenon.

4 SIGNAL ANALYSIS

At present most event related potentials are extracted from the background activity by averaging techniques but the variability or significance of changes of averaged evoked responses are rarely calculated. One reason is that the hardware and software necessary for these calculations is considerable and the calculations make large demands on computer memory. This limitation can be circumvented by storing individual trials on magnetic tape and making the necessary calculations off-line, but this is time consuming. The problem does not end when the trial to trial variability has been calculated since this usually shows that the variability itself varies throughout the epoch of time sampled and it may be that we should study the variability of the variability—‘little fleas have smaller fleas …’.

In cases where there is variability of both latency and amplitude averaging can be misleading. Pfurtscheller and Cooper (1975) describe a method of improving the accuracy of measurement of evoked responses using correlative techniques to estimate latencies and amplitudes. Such methods might lead to the use of a smaller number of trials before a given confidence limit is reached.

The need for reducing the number of trials before a ‘clear’ response is obtained is a fundamental problem of all psychophysiological experimentation. This problem has been approached in a different way by Weinberg and Cooper (1972) who, using correlative techniques, looked for particular patterns of activity in the ongoing EEG. The problem in this approach is to specify the pattern of activity that is being sought. Nevertheless in certain circumstances such methods might be preferable to averaging. The problems involved in the selection of methods for extracting signal from noise can be summed up by saying that in averaging confidence is gained at the expense of repetition whereas in correlative techniques confidence is gained by knowing what one is looking for.

Co-chairman’s Opening Remarks on Measurement in AEP Studies*

EMANUEL DONCHIN,     Department of Psychology, University of Illinois

Publisher Summary

This chapter describes the use of discriminant analysis (DA) as a pattern recognition technique for examining average evoked potential (AEP). The chapter describes an experiment in which synthetic EPs were generated by summing five damped sinusoids, thus producing a waveform with five independent components, each of which could be independently manipulated. Each average evoked response is described in terms of two preselected EPs representing conditions that define a dimension. For example, in an intensity experiment, the stimuli of maximal intensity should yield the largest possible differences between the EP and an average of ongoing EEG. Data obtained with other stimuli can be described in terms of a template that is derived from two curves representing the maximal intensity EP and the ongoing noise. The score for noise alone is quite different from the noise score for signal noise.

I shall take as my topic the two-fold problem created by the need to combine the data obtained from different Ss used in an experiment: How do you combine the data so that you know what the results were? How do you communicate your conviction to others? Both are vexing problems in average evoked potential (AEP) research because the measurements made from our Ss are based on very complex waveforms. We must define some specific, consensual methods for reducing them to uniquely defined values which can then be handled by normally available conclusion-deriving techniques.

Our problem is severe because, even though there is some inter-subject consistency, the relationships between components of AEPs vary from one S to the other. For example, most EPs elicited by visual stimuli, of moderate to bright intensities, will have a negative peak with a latency between 75 and 120 msec, a positive peak between 90 and 120 msec and, frequently, another negative peak at about 160 msec. The experimental variables, over Ss, may affect similar components in similar ways. Nevertheless, any one S‘s evoked responses will be sufficiently idiosyncratic to make it difficult to develop specific measures of the characteristics of the components and it is difficult to decide how to combine inter-subject data.

Several approaches to this problem have been employed. Often, one attempts an intuitive conclusion based on visual inspection of each S‘s data. One can superimpose different AEPs in a variety of different combinations and quite often a clearly perceptible pattern appears in the data. However, it is difficult to communicate such conclusions convincingly. Alternatively, some unique characteristic of the AEP, such as the amplitude or latency of one component, is measured. Conclusions are then based solely upon an analysis of variance, without knowing what the original data in fact looked like.

Another approach is to average the data from all Ss, computing a mean AEP for each of the experimental conditions, thus drastically reducing the number of complex waveforms to be digested. However, computing across subjects may lead to inadequate measures of variability.

A measure of variability is preserved in discriminant analysis, which is a pattern recognition approach, similar in its underlying logic to the Recognition Index Technique described by Weinberg et al. (p. 34). All pattern recognition techniques involve the development of a decision rule, so that a score larger than a criterion value leads to one decision, and a score below the criterion value leads to the alternate decision. Pattern recognition techniques differ in the logic underlying the selection of the criterion values and in computation of scores. The advantage of discriminant analysis (DA) as a pattern recognition technique is that it has a fairly well defined manner for determining the decision rules and the criterion values.

To explain the technique I shall describe a simulation study of evoked potential data which we have just completed.

Synthetic EPs were generated by summing five damped sinusoids, thus producing a waveform with five independent components, each of which could be independently manipulated. These were embedded in ‘noise’ generated by a Gaussian random process, which varied in frequency, power, etc. We were thus generating different synthetic ‘experiments’ in which known ‘evoked response’ differences were embedded in defined background noise. We then applied the DA technique to the data and determined if it correctly assayed the differences we had introduced.

An example of the simulation of an ‘intensity’ experiment, generating the two waveforms, is shown in the top row of Fig. 1a. On the left are the results of averaging one hundred synthetic records each containing an ‘evoked response’ mixed with noise. On the right are shown the average of one hundred similar noise processes in which an ‘evoked response’ was not embedded. We repeated this process several times, each time reducing the amplitude of the ‘evoked response’, and the signal waveform gradually disappeared from the ‘average evoked response’.

Fig. 1 a, Simulated evoked potential ‘intensity’ experiment. Each record in the left panel is a synthetic evoked response obtained by adding 5 damped sinusoids. The characteristics of the sinusoids are identical for all records in the left panel, however, the synthetic evoked response was successively multiplied by a fraction, to reduce its overall amplitude. On the right are averages of 100 realizations of a noise process with which the synthetic evoked response was mixed. A discriminant function was developed on the basis of the data obtained for the full amplitude signal and the corresponding noise realizations. b, The discriminant scores for the synthetic AEP shown in Fig. 1a. These are plotted against the relative amplitude of the signal.

To use the DA for analysing these data one starts with known differences, which are used as a baseline in terms of which the rest of the data are described. Thus data are not measured simply in terms of specific physical units; each average evoked response is described in terms of two preselected EPs representing conditions which define a dimension. For example, in an intensity experiment we know that the stimuli of maximal intensity should yield the largest possible differences between the EP and an average of ongoing EEG. Data obtained with other stimuli can be described in terms of a template which is derived from the two curves representing the maximal intensity EP and the ongoing noise.

The program used selects 6 of 80 time points used in developing the discriminant function. Six coefficients provided by the program are applied at these points. The result of these 6 multiplications is a decision score. Fig. 1b presents the results of applying this discriminant function to the simulated data shown in Fig. 1a. Each of the lines begins on the right with the score computed for the template condition. The score for noise alone is quite different from the noise score for signal noise. The rest of the scores were computed for data in which the signal was reduced by the percentages indicated on the abscissa, thus providing a measure of the intensity of the signal, and we have found a way of measuring the characteristics of an evoked response which reflects the aspect of interest in the data. This technique enables the scores for each of the individual trials recorded in an experiment to be computed and, by indicating the distribution of the scores, provides the information required about variability.

Fig. 2a shows that this technique works with real, as well as simulated, data. This is a series of evoked responses elicited by flashes with durations ranging from 12·8 to 0·100 msec and shows that, as the brightness of the flashes is reduced, the AEP amplitude is diminished. The EEG recorded just prior to the presentation of the stimulus has been used to represent ‘noise’ records. We developed a discriminant function using the EP elicited by the 12·8 msec flash and the corresponding pre-stimulus EEG records, and applied the discriminant functions so obtained to the rest of the data. The results are shown in Fig. 2b.

Fig. 2 a, Visual evoked responses elicited by flashes with duration ranging between 12·8 and 0·100 msec. The left hand column shows ongoing EEG activity just preceding stimulus onset. A discriminant function was developed on the basis of the data elicited by the 12·8 msec stimulus, b, Discriminant scores for the data shown in Fig. 2a, plotted against flash duration.

In the stepwise discriminant analysis technique we have a very useful and powerful technique for measuring the characteristics of evoked responses, for summarizing the data obtained in an experiment in a meaningful way and for relating the between-subject and within-subject variability to our conclusions. The use of the technique is supported by a large body of statistical theory and the behaviour of the statistics thus obtained under many different conditions can be derived.

I have been advocating for some time the use of multivariate statistical techniques in the analysis of evoked response data, since the EP is a multivariate observation. Oddly, many evoked response investigators persist in handling the data as if they were recording an assortment of univariate observations. Various strategems are developed for combining basic univariate statistical tests in ways which look intuitively satisfactory. However, multivariate analysis combines univariate tests into one test. It seems to me that rather than develop a variety of palliative measures derived from the textbook statistics we were brought up on, it would behove us to make use of more advanced, more appropriate statistical techniques.

To some extent the reluctance to use multivariate techniques derives from the instrumentation requirements that may be imposed by such techniques, since it is necessary that the data, trial by trial, be in a computer-compatible form. Then the data can be put into statistical packages available at most University computer centres. Such transformation of data is expensive. However, the discriminant analysis described above used the BMD packages developed at UCLA, and this is readily available. It is hoped that more extensive use will be made of this available technique.

*(Editors’ note: A full description of the method of analysis described by Professor Donchin is to be found in Donchin, E. and Herning, R. I. (1975). A simulation study of the efficacy of stepwise discriminant analysis in the detection and relation of event related potentials. Electroenceph. Clin. Neurophysiol. 38, 51–58.)

Topography of Evoked Potential Amplitude Fluctuations

MERLIN W. DONALD,     Queen’s University, Kingston, Ontario, Canada

Publisher Summary

This chapter describes a method that determined the distribution of spontaneous fluctuations in evoked potential (EP) late wave amplitude by subtracting successive topographical displays of the EP obtained during periods of rapid EP amplitude change. Somatosensory-evoked responses were recorded from eight normal adult male Ss, while visual evoked responses were recorded from two males and females each. Individual subjects were observed in detail, some returning for as many as 15 recording sessions. The standard somatic stimulus was a shock to the median nerve of the left wrist of the subjects, with current fixed just below the thumb twitch threshold. The somatic target was a weaker shock, adjusted to allow 80–90 percent identification accuracy. Shock evoked responses were averaged by a PDP-12 computer and were stored on digital tape eight channels at a time. Each S produced standard somatosensory evoked response waveforms with relatively little amplitude variability during the initial recording sessions. These recording conditions did not produce any waves resembling P300 in the EP of any