Principal component analysis of pain-related cerebral potentials to mechanical and electrical stimulation in man

94

Electroencephalography and Clinical Neurophysiology, 1982, 53 : 94--103 Elsevier/North-Holland Scientific Publishers, Ltd.

PRINCIPAL COMPONENT ANALYSIS OF PAIN-RELATED CEREBRAL POTENTIALS TO MECHANICAL AND ELECTRICAL STIMULATION IN MAN I B. BROMM AND E. SCHAREIN

Institute of Physiology, Department of Neurophysiology, Universitd'ts-Kranhenhaus Eppendorf, Hamburg (G.F.R.) (Accepted for publication: September 3, 1981)

Analyses of event-related cerebral potentials (ERPs) due to noxious stimuli in man are of special interest in attempts to quantify pain experience. Such investigations have been performed with electrical dental (Chatrian et al. 1975, Sano 1977; Harkins and Chapman 1978; Chen et al. 1979) or skin stimuli (Stowell 1972; Sitram et al. 1977; Buchsbaum and Davis 1979}, mechanical skin stimuli (Stowell 1975a,b), and recently also with thermal (laser) radiation of the skin (Carmon et al. 1978; Kenton et al. 1980). Relationships have been found between amplitudes or amplitude differences of several components in averaged ERP epochs, and strength of noxious stimuli or magnitude estimation of pain perception. There is some evidence that especially the late ERP amplitudes, above 100 msec after stimulus onset, seem to have a higher correlation to pain experience than to stimulus intensity (cf., Chapman et al. 1979). However, the relationships between single ERP components, pain experience, stimulus quality and stimulus quantity, have n o t yet been examined in detail. Experimentally induced pain is presumably always accompanied by specific sensations, elicited by the kind of the noxious stimulus applied. Thus the question arises which component amplitudes of ERP can actually be related to pain experience, in contrast to information processing due to the specific kind of stimulation used. In the present paper I Supported by the Deutsche Forschungsgemeinschaft (SFB 115).

experiments are described in which cerebral ERP components were correlated with pain experience in man, where in the same session two different kinds of noxious skin stimuli were applied: electrical and mechanical pulses of the same duration and at the same sites of the body. Cerebral potentials were averaged for equal categories of sensation. Wave form similarities and differences were quantified by principal c o m p o n e n t analysis, decomposing the ERP wave form into a small number of statistically uncorrelated components. After eliminating the effects of stimulus intensity, two independent ERP components were extracted discriminating between nonpain and pain perception.

Method and Material

Subjects The investigations were performed on 15 male volunteers, paid medical students. Ages ranged from 21 to 29 years. All subjects were right handed, free of medication and with no history of neurological diseases. No psychological abnormalities were detected using the 'Freiburg Personality Inventory' (Fahrenberg and Selg 1970). None of the subjects had previously participated in such a laboratory experiment. Stimuli and apparatus Two different kinds of stimuli were used within the same session. Mechanical force controlled step indentations of 20 msec

0013-4649/82/0000---0000/$02.75 © 1982 Elsevier/North-Holland Scientific Publishers, Ltd.

PCA OF PAIN-RELATEDCEREBRAL POTENTIALS

95

duration were applied to the nail bed of the left middle finger via an electromechanical stimulator and a plexiglass probe (1.2 mm contact diameter), starting from a constant preset level of 0.15 N. The rise time of the step was less than 2 msec. Electrical constant current stimuli, consisting of 20 msec trains of 200 Hz rectangular monophasic waves, were applied ground-free to the tip of the same finger (electrode diameter 1.2 mm, for detail see Bromm and Treede 1980). Stimulus intensities were referred to the corresponding mean threshold intensities as measured in preliminary experiments (15 subjects). As such, mechanical stimulus intensities of 0.8, 1.6, 2.4, 3.2 N (mean pain threshold at 1.9 + 0.6 N) and electrical stimuli with intensities of 2, 4, 6, 8 mA (mean pain threshold at 4.8 +0.9 mA) were applied.

ment, temperature 22--24°C, comfortably supported subjects, closed eyes, measurements around 10.00a.m.). Each of the 2 kinds of stimuli was presented in 4 different intensities. For each session lasting 90 rain the number of stimuli per stimulus intensity and quality was 40. The stimulus intensities were delivered by a microcomputer in quasi-randomized order, with interstimulus intervals between 10 and 20 sec, in order to minimize effects of habituation (for detail see Bromm and Scharein 1982). In two preliminary sessions the subjects were familiarized with the experimental procedure.

Recording EEG was recorded from 3 scalp sites, each referred to the linked earlobes. Subjects were grounded on the forearm. The scalp sites were according to the International 10-20 system location Cz, and sites over the hand areas of the postcentral gyrus of both hemispheres (Shagass and Schwartz 1964); Ag/AgC1 electrodes were affixed with collodion. EOG was recorded bipolarly between sites inferiorlateral and superior-lateral of the left eye. Electrode impedance was less than 5 k~2. EEG and EOG were digitized (200 Hz, system bandpass= 0.16--30 Hz) and stored on disk, together with parameters of stimulation and perception, and event markers. EEG and EOG segments of 10 sec duration were stored, 5 sec before and 5 sec after the onset of the stimulus. In this paper poststimulus intervals of 500 msec were analysed. Stimulus perception was estimated using 4 categories of increasing aversiveness: K1 and K2 denoted faint and marked pre-pain sensation, K3 and K4 faint and marked pain. Procedure The measurements were carried out under uniform conditions (minimum noise environ-

Data analysis After visual artifact control about 5% of EEG records were discarded (226 trials), the remaining 4574 single ERPs (from 15 subjects × 320 trials) were evaluated by computer. Principal component analysis (PCA, Glaser and Ruchkin 1976; Donchin and Heffley 1978) was used to detect differences in the ERP components which varied with the experimental conditions. Since PCA is not able to discriminate between variance due to changes in latency times or in amplitudes of ERP, Glaser and Ruchkin (1976) and Donchin and Heffley (1978) have suggested the application of an adaptive filtering algorithm (Woody 1967) to adjust the latencies of components before conducting the PCA. With stimulus qualities and intensities such as we applied, the most prominent peak in ERP is the negativity at about 140--160 msec (N150, Chatrian et al. 1975; Johnson et al. 1975; Chapman et al. 1979). In our study, latency jittering of this peak was evident within and between subjects, with no correlation between changes in latency time and stimulus intensity or painfulness (see also Chen et al. 1979). Therefore we filtered our data by applying the algorithm of Woody (1967) to remove latency jittering within each subject. The template was formed with the ERP averaged in each session, separately for mechanical and electrical stimuli, the filter was focussed on the N150 peak, and a maxi-

96 m u m latency shift o f +15 msec was allowed for obtaining the best fit for each trial within a window between 100 and 200 msec after stimulus onset. Because of a sampling frequency of 200 Hz this shift in time means, in fact, a realignment of data b y maximal +3 points. On the other hand, short peaks, as appear in the very early components, m a y be distorted by this procedure. Since there is much evidence in literature that pain sensation is only reflected in the late components, which could be best recorded over the vertex (Chatrian et al. 1975) this paper deals mostly with ERPs (Cz) > 1 0 0 msec, and considers all the early peaks as one component. The latency corrected single trial ERPs were subjected to PCA. Each single trial ERP, consisting of 100 digitized voltages (500 msec analysis interval) was decomposed into a small number of basic wave forms (principal components). Following the suggestions of Donchin and Heffley (1978), the covariance matrix between ERP time points was chosen as association matrix. The principal components (PCs; dimension ~ V ) were the varimax rotated (Kaiser 1959) orthonormalized eigenvectors of the covariance matrix multiplied with the square roots of the corresponding eigenvalues. Each measured single trial ERP could be represented by a linear combination o f the extracted PCs. The coefficients of the resulting linear equation system were averaged for each experimental condition within each subject, yielding mean c o m p o n e n t scores. These scores were evaluated b y analysis of variance after standardization (mean value = 0, S.D. = -+1).

Results

Averaged cerebral potentials Nearly half o f all applied stimuli was perceived as painful. All rating categories were used with nearly equal frequency, with considerable overlapping in the assignment between categories and stimulus intensities. In the following the E R P epochs were

B. BROMM, E. SCHAREIN MECHANICALSIIMU[ATION

ELECTRICALSTIMULATION

(pV) n : 40

-i---..---._~

K4

I

0 F i g . 1.

(ms) ERP

500 to

mechanical

0 and

I

I

I

I

(ms)

Zg~n~5L JS

I

500

electrical

stimuli,

Upper traces: EEG record and mean ERP (n = 40 repetitions) at zero stimulus intensity (artifact control). K1, K2: increasing prepain sensation; K3, K4: increasing pain. Electrode site: Cz. Subject: J.S. averaged for equal categories of sensation.

arranged and averaged for equal categories of sensation, independent of the stimulus strength which actually was applied. In Fig. 1 averaged ERP epochs in one subject are shown evoked b y mechanical or electrical stimuli, occurring in a quasi-randomized sequence, within the same session. K1 and K2 denote increasing tactile sensation, a slight pressure, prickling or vibration, K3 and K4 describe increasing pain. Above, the prestimulus EEG of the selected subject is illustrated, showing low amplitudes and an almost complete lack of alpha waves during this session. When the stimulus was initially perceived (K1), typical components appeared within the first 200 msec. When pain was felt, as in K3 and K4, a further c o m p o n e n t became prominent in this subject with peak latencies of a b o u t 330 msec. These findings held true for mechanical as well as for electrical stimuli; differences appeared in the initial c o m p o n e n t amplitudes within the first 100 msec. In all experiments these early components were maximal over the hand area o f the contralateral postcentral gyrus with asymmetries between both hemispheres. In con-

PCA OF PAIN-RELATED CEREBRAL POTENTIALS

97

trast, the late components could best be recorded over vertex (Cz). All amplitudes of evoked brain potentials increased with increasing rating categories. Except for the 330 msec peak, no qualitative changes could be seen when perception altered from tactile sensation to pain. The late 330 msec component could reliably be detected only in subjects having a low amplitude EEG with poor alpha rhythm. The influence of the prestimulus EEG activity on the wave form of the averaged ERPs is documented in Fig. 2. Results for two subjects are given with different alpha contributions in the ongoing EEG segments, as indicated in the insets. The different amounts of alpha waves were quantified by averaging the periodograms of all 5 sec EEG segments preceding the approximately 160 mechanical stimuli, weighing each segment by a split cosine bell to avoid aliasing (Bloomfield 1976). To the right all ERPs to mechanical stimuli are averaged for each subject revealing considerable differences in the ERP wave forms; the lower case suggests the beginning

of alpha synchronization. Furthermore there was an inverse relationship between the amount of alpha waves and the maximum peak-to-peak amplitude differences. Similar results were found even within a single subject when averaged ERPs were related to the power in the alpha band for the corresponding prestimulus EEG segments. In the following, results are given only for those subjects for whom the relative contribution of the alpha band (7.5--12.5 Hz) to the total power of the prestimulus EEG segments did not exceed the arbitrary c u t o f f score of 30% (8 subjects).

MEANERP TOMECHANICALSIIMULATION

MEANPSDOF PRESTIMULUSEEG

i01

1°°I GQ

1oIL L~

0

I

f (Hz)

I

30

I

0

I__1

I

t {ms}

1

I

MJ

500

Fig. 2. Influence o f prestimulus EEG activity on the ERP wave form. Left: mean power spectral densities (PSD) of 160 prestimulus EEG segments (length 5 sec); insets demonstrate typical EEG records. Right: mean ERP to mechanical stimuli. Above: subject G.O. with high frequency low amplitude EEG. Below: subject M.J. with marked alpha activity during the session. Electrode site: Cz.

Principal component analysis Principal component analysis (PCA) was applied to single trial ERPs, of which 2420 epochs (from 8 subjects with low alpha activity in the prestimulus EEG segments) were available; only records from Cz leads were regarded. From the data matrix the variance~ovariance matrix between ERP time points was built from which PCs were extracted. The upper trace in Fig. 3 represents the grand mean of ERPs, averaged for the 8 subjects selected. In principle, the grand mean was similar to the averaged ERPs in the single case study (Fig. 1). Again the early components (<100 msec) were more marked under electrical than under mechanical stimulation, and all ERP component amplitudes increased, when subjective stimulus rating changed over to pain. The lines thereafter in Fig. 3 represent the varimax rotated loadings for the 6 principal components (PCs) which accounted for about 90% of total variance. The number of the extracted components was determined by the magnitudes of the corresponding eigenvalues (Scree test, Cattel 1966). Most of the PCs loaded at practically the same latency times as the amplitude components of the grand mean. Their maxima had well defined locations in time, except the polyphasic PC6. The PCs, therefore, were ordered in sequence of maximum loading latency times. The propor-

98

B. BROMM, E. SCHAREIN

(pV) MEANERP

.... 17%

PC1 5E (50-80ms) O PC2 (140-160ms)

5F 0L

PC3 5E (200-250ms) 0

~

~

16%

. , ~

19%

PC4 5 (280-360ms) 0 E

15%

PC5 (z.OO-5DO ms)

5 0L

17%

PC6

0

(polyphosi)c

is represented in a single trial ERP. In the covariance analysis, the score for a given PC indicates its amplitude relative to the grand mean (Donchin and Heffley 1978). As such, instead of 100 voltages, n o w 6 scores were taken into account to describe the single ERP epoch on the basis of the 6 extracted PCs. All 2420 c o m p o n e n t scores of each component were normalized {mean value = 0, S.D. = +1). The scores o f each c o m p o n e n t were averaged for each subject and each of the 4 experimental conditions. The so-called mean scores were analysed b y separate 2-way repeated measures analyses of variance: kinds of stimulation (mechanical, electrical) × stimulus perception (non-painful, painful). As shown in Fig. 4, the stimulus perception affected significantly the scores of the first 4 PCs (Fig. 4, left). Even PC5 representing the latest negative wave form in the ERP showed a tendency to discriminate between non-painful and painful stimulation. These

A

r/, I

0

I

I

I

t(ms)

I

-05

0

[

I

B

*0.5 I

-0.5

0

[

I

~0.5 I

l

500

Fig. 3. Mean ERP and principal components. Data matrix contained the ERP records of all subjects under all experimental conditions. Principal components (PCs) are ordered and denoted by latency times of maximum loadings. PC6 is polyphasic. Percentages indicate the percent of total variance accounted for by each PC. Electrode site: Cz.

tion of total variance accounted for by each PC varied b e t w e e n 3 and 19%, as indicated in the figure. In order to separate the influence of the different experimental conditions on the PCs, c o m p o n e n t scores for each single trial ERP were calculated. Principal c o m p o n e n t scores describe the degree to which each PC

Fig. 4. Mean principal component scores. The scores were calculated for the principal components, given in Fig. 3, separately for non-pain (white) and pain (black) stimulus conditions. In the left column (A) the scores are confounded by effects of the stimulus intensity; in B the influence of stimulus intensity was partialed out. p indicates Lhe significance level of discrimination between both stimulus conditions (analysis of variance). Electrode site: Cz.

PCA OF PAIN-RELATED

CEREBRAL

POTENTIALS

effects reflected the fact that Lall ERP amplitudes increased when pain was felt. The kind of stimulation influenced significantly only the scores of PC1 (P < 0.01), reflecting higher negative ERP amplitudes under electrical stimulation in the earlier wave forms ( < 8 0 msec) than under mechanical stimulation. This held true for non-painful and painful stimulus perception. The change of stimulus perception from non-pain to pain sensation is highly correlated with changes in stimulus intensity. As such the question had to be answered whether the influence of stimulus intensity or perception was reflected in the scores of the PCs. Since equal stimulus intensities were often rated b y different categories o f sensation, it was possible to eliminate the variance due to stimulus intensity in all 2420 scores of each PC b y partialing o u t (Cooley and Lohnes 1971), and the remaining variance was re-analysed. As shown in Fig. 4, to the right, the effects of stimulus perception on the component scores were drastically reduced. Only PC2 and PC4 varied systematically with stimulus perception. In other words, only the scores of PC2 (140--160 msec) and PC4 (280--360 mse~) distinguished between nonpain and pain perception, when the variance due to stimulus intensity was removed.

Reliability o f the extracted principal components To test the stability of the 6 principal components, separate PCAs were performed for non~verlapping classes of single trial ERPs. For this purpose the 2420 × 100 data matrix was partitioned into 4 submatrices according to the 4 different experimental conditions: mechanical non-painful stimulation, mechanical painful stimulation, electrical non-painful stimulation, electrical painful stimulation. Again the corresponding variance~ovariance matrices were subjected to PCA. Fig. 5 represents the resulting principal c o m p o n e n t s after varimax rotation. In each case 6 PCs were sufficient to account for a b o u t 90% of total variance. As before, the PCs were monophasic,

99

MECHANICALSTIMULATION

ELECTRICALSIIMULATION

(juV)

0

MEAN ERP

0

PC2

5E PC3

0

PC4

PC6 l

0

l

i

[

t (ms)

i

[

500

I

0

I

]

~

t(ms}

I

I

500

Fig. 5. Stability of the extracted principal components. Mean ERPs and principal components (PCs) for all subjects were evaluated selectively for the 4 experimental conditions: mechanical skin stimulation (left), electrical skin stimulation (right), no pain (broken lines), pain (continuous lines). PCs are ordered by latency times of maximum loadings, PC6 is polyphasic. Electrode site: Cz.

except PC6, and again the loading maxima could be classified with respect to distinct ERP time intervals. Strong stimuli caused higher ERP amplitudes and higher covariances between ERP time points; as such the component loadings were generally higher when pain was reported. Position and wave form of comparable PCs, however, were similar in all the non-overlapping classes of ERPs. Moreover, the PCs of the 4 ERP classes were very similar to those extracted from the whole data matrix given in Fig. 3. No specific

100

B. BROMM, E. SCHAREIN

TABLE I Stability o f the extracted principal components. Product-moment correlation coefficients between principal components (PCs) extracted for each experimental condition and the corresponding PC extracted from the whole data matrix. Mech., mechanical skin stimulation; El., electrical skin stimulation. Principal component PC1 PC2 PC3 PC4 PC5 PC6

Non-pain

Pain

Mech.

El.

Mech.

El.

0.89 0.74 0.69 0.67 0.89 0.57

0.76 0.87 0.67 0.84 0.89 0.53

0.92 0.78 0.70 0.72 0.84 0.72

0.91 0.81 0.64 0.87 0.91 0.79

All correlation coefficients are significant at the 0.001 level.

PC emerged or vanished under any of the experimental conditions regarded. The similarity between each PC under each of the 4 experimental conditions and the corresponding PC of the whole data set was quantified by building p r o d u c t - m o m e n t correlation coefficients, as shown in Table I. With a sampling frequency of 200 Hz and an analysis period o f 0.5 sec, 100 data points were available for each PC comparison. With this a m o u n t of data a correlation coefficient o f 0.26 is significant at a level of P = 0.01. The highest stability was found for PC1, describing the early ERP waves. Lowest stability was seen for the polyphasic PC6, which accounted for the smallest a m o u n t of total variance anyway. As such, cerebral potentials evoked by quite different skin stimuli could be decomposed, with high stability, into a unique set of PCs independent from the specific experimental condition regarded.

Discussion

In this study event-related cerebral potentials (ERPs) upon noxious skin stimulation were subjected to principal c o m p o n e n t analy-

sis (PCA) in order to extract specific painrelated components. Recently it was shown that principal components (PCs) extracted by PCA varied with stimulus conditions (Suter 1970; Horst and Donchin 1980; Johnston and Holcomb 1980). Even changes in the meaning of a word in different context seems to influence single PCs (Brown et al. 1979). The cited authors applied PCA to averaged ERPs for equal stimulus conditions, separated for the different subjects or different leads, analysing similarities between the averaged ERP amplitudes at different time points. We applied PCA to single trial ERPs and determined the degree to which each extracted PC was represented in each ERP epoch by building principal c o m p o n e n t scores. This procedure was possible after taking into account the spontaneous activity of the prestimulus EEG. Subjects for w h o m the relative power in the alpha band in the prestimulus EEG exceeded an arbitrary limit of 30% of the total power were discarded (7 of 15 subjects). Furthermore relatively high stimulus intensities were used, of which about 60% were painful, causing an improved signalto-noise ratio. In fact, the stability of the pattern of principal components computed from non-overlapping ERP classes, as well as the results of the analysis of variance of the computed scores, d o c u m e n t e d the practicability of the application of PCA on single trial ERPs. Single trial ERPs to mechanical and electrical, non-painful and painful skin stimuli, applied in the same session, were analysed by PCA for the 8 subjects with low alpha activity. Differences in the ERPs due to different experimental conditions were represented in the c o m p o n e n t scores describing the extent to which each c o m p o n e n t contributed to the ERPs. The c o m p o n e n t scores were tested by analysis of variance. The fact that only one c o m p o n e n t could be extracted within the first 100 msec after stimulus application as well as its relatively small contribution {15%) to the total variance m a y partly be due to Woody filtering the original data, as men-

PCA OF PAIN-RELATEDCEREBRAL POTENTIALS tioned under Methods. On the other hand, the later components did not discriminate between the stimulus qualities applied; only PC1 separated significantly between ERPs to mechanical or electrical stimuli. This is in agreement with findings of Davis et al. (1972) showing high similarities in the late amplitudes of the average ERPs to acoustical, visual, tactile and electrical stimulation, after adjusting the stimulus intensity to equal magnitudes of sensation. Most of the total variance between the ERP amplitudes at different time points was due to stimulus intensity. The dependence of averaged ERP amplitudes on stimulus intensity under cutaneous electrical stimulation is well documented (for example see Uttal and Cook 1944; Beck and Rosner 1968; Buchsbaum et ai. 1977). In averaged ERPs to mechanical skin stimuli Johnson et al. (1975) found a high correlation especially between the late ERP components (>120msec) and strength of stimulus. PCA, as performed in this study, revealed a nearly equal statistical dependency on stimulus intensity for all 5 PCs (PC1-PC5) describing the main ERP waves between 50 and 500 msec. Only a few investigations have tried to separate effects of stimulus intensity and stimulus perception in cerebral potentials, especially in attempts to quantify pain perception. Carmon et al. (1978) concluded from experiments with thermal (laser) skin stimulation that the painfulness of the stimulus was reflected mainly in the late potentials (ranging between 300 and 500 msec). Chen et al. (1979) studied peak-to-peak amplitude differences of cerebral potentials evoked by electrical dental stimulation; after eliminating the influence of stimulus intensity, they observed significant correlations between painfulness and the 3 amplitude differences: P120-N175, N175-P260, and P260-N340 (P, positive; N, negative). In spite of certain variations in stimulus sites and qualities between the cited and our investigations, we could reproduce these findings with quite another technique of multivariate ERP

101 analysis. After eliminating the effects of stimulus intensity, we found two independent signal components (PC2 and PC4) with latencies in the range of 140--160 msec and 280-360 msec, which varied with stimulus perception changing from non-painful to painful sensations. Therefore it can be suggested that these two independent ERP components reflected parts of the pain processing system, regarding analysis periods up to 500 msec. As such, we propose that these components be used in attempts to quantify pain perception. Of course, the decomposition of ERPs in principal components by PCA is a phenomenological-statistical description, in which no physiological aspects are involved. PCA applied to the 4 disjunctive sets of ERPs built from the 4 different experimental conditions yielded similar sets of PCs with high similarities between corresponding components: none of the stimulus conditions investigated elicited specific principal components within the accuracy of their determination. In other words, no typical ERP components emerged only when the subject felt pain. This illustrates the fact that, on the average, all ERP amplitudes increased with increasing stimulus intensity. In conclusion, in cerebral potentials evoked by noxious and non-noxious skin stimuli, no specific pain-related wave form in the averaged ERP could be observed. Thus from a mere occurrence of amplitudes or amplitude differences in ERPs it cannot be concluded that the subject experienced pain. By means of principal component analysis, however, two components were detected, which distinguished significantly between non-painful and painful sensations, and which therefore may be denoted as pain-related components.

Summary Single trial event-related cerebral potentials (ERPs) in response to skin stimuli ,of

102

various intensities and qualities in man were investigated in respect to their nociceptive information content. Electrical constant current stimuli (20 msec, 2--8 mA) and mechanical force controlled stimuli (20 msec, 0.8-3.2 N) were applied to the tip of the left middle finger. Four intensities of each stimulus quality were given, each intensity appearing 40 times in standardized randomized order. EEG segments (between 5 sec before and 500 msec after stimulus onset) were subjected t o computer analysis. ERP wave form was shown to depend upon the amount of alpha waves in the prestimulus EEG. For analysis, only subjects with low power in the alpha band were selected. Principal component analysis was applied to all single trial ERPs measured using the variance-covariance matrix of association. Six principal components (PCs) were extracted accounting for about 90% of total variance. Five of the extracted PCs had well located loading maxima: PC1 (50--80 msec), PC2 (140--160 msec), PC3 (200--250 msec), PC4 (280--360 msec), PC5 (400--500 msec); PC6 appeared polyphasic. Analysis of variance of the mean PC scores revealed that one PC (PC1) discriminated between quality, and 4 PCs (PC1--PC4) between quantity of stimulation. Eliminating effects of stimulus intensity resulted in two PCs (PC2, PC4) which distinguished exclusively between non-pain and pain. PCA applied to disjunctive subsets of ERPs, corresponding to the different experimental conditions, yielded practically identical sets of PCs, such that no specific ERP component emerged when pain was reported.

B. BROMM, E. SCHAREIN

tions de la peau d'intensitd et de nature varides chez l'homme ont 6td 6tudi6s en fonction de leur contenu en information nociceptive. Des stimulus en courant constant (20 msec, 2--8 mA) et des stimulus m6caniques de puissance contrSl6e (20 msec, 0,8--3,2 N) furent appliqu6s ~ l'extr6mitd du majeur gauche. Chacun des deux types de stimulus fur appliqud avec 4 intensitds diff6rentes, chacune 40 fois, dans un ordre d6termin6 au hasard. Les segments d'EEG (de 5 sec avant le stimulus ~ 0,5 sec apr6s) furent soumis ~ une analyse informatique. L'onde du PLE s'est prdsentde comme ddpendante de la puissance des ondes alpha dans I'EEG prdcddent le stimulus. Pour l'analyse, seuls les sujets ayant une puissance basse dans la bande alpha furent sdlectionnds. L'analyse de la composante principale a dt6 appliqu6e ~ tousles PLE en utilisant une matrice de variance-covariance d'association. Six composantes principales (CP) furent extraites comptant pour environ 90%~ de la variance totale. Cinq furent bien localisdes par leur maximum: CP1 (50--80 msec), CP2 {140-160 msec), CP3 (200--250 msec), CP4 (280-360 msec), CP5 (400--500 msec); CP6 apparaft polyphasique. L'analyse de variance de la moyenne des scores CP rdv61e qu'une CP (CP1) d6pend de la nature de la stimulation et quatre (CP1 ~ CP4) de son intensitd. En dliminant les effets dus ~ l'intensitd du stimulus pour CP2 et CP4, on s'aperqoit qu'elles ne se distinguent exclusivement que pour la qualitd nociceptive ou non du stimulus. Un PCA appliqu6 ~ des tests disjoints de PLE, correspondant ~ diffdrentes conditions expdrimentales, produisait des complexes CP identiques, tel que des composantes PLE non sp6cifiques dmerg6rent quand une douleur dtait exprim6e.

R~sum~ Analyse de la composante principale des potentiels cdrdbraux lids d la douleur induite par des stimulations dlectriques et mdcaniques chez l'homme Les potentiels cdr~braux lids ~ un seul dvdnement (PLE) en r~ponse ~ des stimula-

References Beck, C. and Rosner, B.S. Magnitude scales and somatic evoked potentials to percutaneous electrical stimulation. Physiol. Behav., 1968, 3: 947---953. Bloomfield, P. Fourier Analysis of Time Series: an Introduction. Wiley, N e w York, 1976. B r o m m , B. and Treede, R.-D. Withdrawal reflex,skin

PCA OF PAIN-RELATED CEREBRAL

POTENTIALS

resistance reaction and pain ratings due to electrical stimuli in man. Pain, 1980, 9: 339--354. Brown, W.S., Marsh, J.T. and Smith, J.C. Principal component analysis of E R P differences related to the meaning of an ambiguous word. Electroenceph. clin.Neurophysiol., 1979, 46: 709--714. Buchsbaum, M.S. and Davis, G.C. Application of somatosensory event related potentials to experimental pain and the pharmacology of analgesia. In: D. Lehmann and E. Callaway (Eds.), H u m a n Evoked Potentials: Applications and Problems. Plenum Press,N e w York, 1979: 43--54. Buchsbaum, M.S., Davis, G.C. and Bunney, W.E. Naloxone alterspain perception and somatosensory evoked potentials in normal subjects. Nature (Lond.), 1977, 270: 620--622. Cartoon, A., Dotan, Y. and Sarne, Y. Correlation of subjective pain experience with cerebral evoked responses to noxious thermal stimulations. Exp. Brain Res., 1978, 33: 445--453. Cattel, R.B. The Scree test for the number of factors. Multivar. behav. Res., 1966, 1: 245--276. Chapman, C.R., Chen, A.C.N. and Harkins, S.W. Brain evoked potentials as correlates of laboratory pain: a review and perspective. In: J.J. Bonica, J.C. Liebeskind and D.G. Albe-Fessard (Eds.), Advances in Pain Research and Therapy, Vol. 3. Raven Press, New York, 1979 : 791--803. Chatrian, G.E., Canfield, R.C., Knauss, T.A. and Letrich, E. Cerebral responses to electrical tooth pulp stimulation in man. Neurology (Minneap.), 1975, 25: 745--757. Chen, A.C.N., Chapman, C.R. and Harkins, S.W. Brain evoked potentials are functional correlates of induced pain in man. Pain, 1979, 6: 365--374. Cooley, W.W. and Lohnes, P.R. Multivariate Data Analysis. Wiley, New York, 1971. Davis, M., Osterhammel, P.A., Wier, C.C. and Gjerdingen, D.B. Slow vertex potentials: interactions among auditory, tactile, electric and visual stimuli. Electroenceph. clin. Neurophysiol., 1972, 33: 537--545. Donchin, E. and Heffley, E.F. Multivariate analysis of event-related potential data: a tutorial review. In: D.A. Otto (Ed.), Multidisciplinary Perspectives in Event-Related Brain Potential Research, EPA600/9-77-043. U.S. Government Printing Office, Washington, D.C., 1978: 555--572. Fahrenberg, J. und Selg, H. Freiburger PersSnlichkeitsinventar (FPI). Hogrefe Verlag, GSttingen, 1970. Glaser, E.M. and Ruchkin, D.S. Principles of Neurobiological Signal Analysis. Academic Press, New York, 1976. Harkins, S.W. and Chapman, C.R. Cerebral evoked

103 potentials to noxious dental stimulation: relationship to subjective pain report. Psychophysiology, 1978, 15: 248--252. Horst, R.L. and Donchin, E. Beyond averaging. II. Single-trial classification of exagenous eventrelated potentials using stepwise discriminant analysis. Electroenceph. clin. Neurophysiol., 1980, 48: 113--126. Johnson, D., Jiirgens, R., Kongehl, G. and Kornhubet, H.H. Somatosensory evoked potentials and magnitude of perception. Exp. Brain Res., 1975, 22: 331--334. Johnston, V.S. and Holcomb, P.J. Probability learning and the P3 component of the visual evoked potential in man. Psychophysiology, 1980, 17: 396-4O0. Kaiser, H.F. Computer program for varirnax rotation in factor analysis. Educ. Psychol. Meas., 1959, 19: 413--420. Kenton, B., Coger, R., Crue, B., Pinsky, J., Friedman, Y. and Carmon, A. Peripheral fiber correlates to noxious thermal stimulation in humans. Neurosci. Lett., 1980, 17: 301--306. Sano, H. Influence of intensity-varied electrical stimulation of a tooth and 30% N20 premixed gas inhalation on sornatosensory evoked potentials (SEPs). Jap. J. Dent. Anesthesiol., 1977, 5: 9--21. Shagass, C. and Schwartz, M. Recovery functions of somatosensory peripheral nerve and cerebral evoked responses in man. Electroenceph. clin. Neurophysiol., 1964, 17 : 126--135. Sitram, N., Buchsbaum, M.S. and Gillin, J.C. Physostigmine analgesia and somatosensory evoked responses in man. Europ. J. Pharmacol., 1977, 42: 285--290. Stowell, H. Latency characteristics of primary somatosensory potential evoked by electrical and mechanical stimuli. Int. J. Psychobiol., 1972, 2: 305--320. Stowell, H. Human evoked responses to potentially noxious tactile stimulation, I. Activ. nerv. sup. (Praha), 1975a, 17: 1--7. Stowell, H. Human evoked responses to potentially noxious tactile stimulation, II. Activ. nerv. sup. (Praha), 1975b, 17: 94--100. Suter, C. Principal component analysis of average evoked potentials. Exp. Neurol., 1970, 29: 317-327. Uttal, W.R. and Cook, L. Systematics of the evoked somatosensory cortical potential: a psychophysicalelectrophysiological comparison. Ann. N.Y. Acad. Sci., 1944, 112: 60--80. Woody, C.D. Characterization of an adaptive filter for the analysis of variable latency neuroelectric signals. Med. biol. Engng, 1967, 5: 539--553.