Med. & Biol. Eng. & Comput., 1977, 15, 641-647

Simple online detector of auditory evoked cortical potentials S. M . M a s o n

A.P.

Su

Medical Physics Department, Nottingham General Hospital, Nottingham, England

R.A.

Hayes

AudiologyDepartment,NottinghamGeneralHospital,Nottingham, England

BasildonGeneralHospital,Essex,England A b s t r a c t - - A simple analogue electronic instrument is described which produces an objective yes/no indication of the presence of an auditory evoked cortical potential (a.e.c.p.). It correlates two a.e.c.p.s obtained by averaging alternative responses in a stimulus run at constant intensity and frequency. The 5 and 1% significance levels of the correlation coefficient are derived and used as the Criterion for whether or not a response is present. A modified correlation coefficient is used which permits considerable simplification of the circuitry involved. The results o f 24 hearing-level thresholds determined by the detection instrument show better agreement with conventional pure-tone audiometry than results from subjective assessment of the averaged a.e.c.p.s. K e y w o r d s - - E l e c t r i c response audiometry

1 Introduction

THE basic problem involved in the detection of an a.e.c.p. (vertex response) is coping with the presence of random biological noise and artefacts associated with the signal. The noise and artefacts are particularly troublesome with low-intensity stimuli, since the response is small, and this leads to inaccuracy in the subjective determination of hearing level thresholds by means of a.e.c.p.s. This interference has similar frequency components to the evoked response and cannot be removed by filtering. Averaging, to enhance the response and cancel the random noise, is usually employed; for practical reasons the number of individual responses which can be averaged is limited and the difficulty still remains of deciding whether an averaged response is present in a noise-contaminated a v e r a g e d signal. This decision is often made subjectively, which can lead to inconsistent results. B~AGL~Y (1973) showed that the hearing-level thresholds from 25 selected patients determined by an inexperienced observer in electric response audiometry (e.r.a.) were only slightly less in agreement with conventional audiometry than the judgments made by an experienced observer; although it was suggested that the inexperienced observer may have been misled on two occasions by residual artefacts and as a result gave objective thresholds which were considerably lower than the subjective thresholds. First received25th November, 1976and in final form 3rd February, 1977

Medical & Biological Engineering & Computing

A study by Rose et al. (1971), however, shows that there is considerable variability in assessing the presence or absence of an a.e.c.p. A panel of six judges with various degrees of experience was used to assess 613 averages taken from 50 subjects. The results showed considerable variability in both interjudge and intrajudge reliability; thus emphasising the need for objectivity in the assessment. When computing facilities are unavailable, methods of analysis are limited, although a number of techniques have been described to help in the detection of averaged a.e.c.p.s. F o r evaluating the background electroencephalographic signal (e.e.g.) during a particular collection of responses, SCHIMMEL (1967) described the formation of a (_+) reference and of a prestimulus average. This ( + ) reference is an average computed by alternate addition a n d subtraction instead of by the usual addition; the mean component in this situation then becomes zero. The ( + ) reference and prestimulus average give guidance to the investigator in detecting and measuring the mean component and in determining the size and structure of the averaged background activity. SALOMONet al. (1973) used a template method for evaluation of threshold hearing levels in sedated and anaesthetised children. The templates or mean response curves were drawn up from responses elicited by high-intensity suprathreshold stimuli in not less than three patients. Future sweeps with lower intensity stimuli were then compared with this template. A high percentage of errors however, both type-1 and -I1, were introduced in the anaesthetised November 1977

641

group using this method, which, in some cases, were clinically unacceptable. In a later paper SALO~ON (1974) describes a rank-correlation technique which CH2

CHI 80dB

0 ", _

v

L,.,

.~ - x._/- ~

is performed on the voltage and its first and second derivatives measured at various points along individual sweeps. A low-incidence of type-/ and -II errors was obtained in both children and adults when using the first derivative of the voltage. The technique appears very promising and presents a fully automated procedure for computerised evalution of e.r.a. It is however very complex and requires considerable computing facilities. An analysis technique for judging whether an 2--

"G

L.-~

9

0o Q 9

~g 8~ .~-E N~ 1

J

Simplified correlation coefficient Rxy

70dB

5Fv m__ 100 ms

Fig. I Averaged pairs of a.e.c.p.s showing the position of the zero- voltage baseline; each trace consists of the odd- and even-numbered sweeps, respectively

Fig. 3 Computed results for the simplified correlation coefficient showing that the assumption of zero-mean properties is reasonable

1.0 --

g

0.5

o~

8

0.5

].0

1.5

Simplified correlation coefficient R~y^

Fig. 2 Computed results for the true correlation coefficient R xy andthe simplified correlation coefficient R ,y showing good agreement 642

Medical

& Biological

Engineering

& Computing

November

1977

a.e.c.p, is stable or unstable by comparing two averages is described by L o w v and WEiss (1968). It employs a split-sweep method to compare two separate averages: one built up of odd-numbered sweeps, the other the even-numbered sweeps. This split method is very useful when only subjective assessment of the averaged a.e.c.p.s, is possible since it is considerably easier to compare two averaged traces for simularities, rather than recognise a characteristic waveform in a single trace. An objective method of comparing these two waveforms is the crosscorrelation function, or, alternatively, the correlation coefficient. BEAUCHAMP (1973) described the correlation function and techniques for the analogue determination of it. A simplified form of the correlation coefficient is evaluated here using simple analogue electronic

circuits. The results are then used to assess the presence or absence of an averaged a.e.c.p.

2 Method

The correlation coefficient for two sets of data points xl x2 ... xN and yl y2 ... y~ which represent two averaged a.e.c.p.s is N

Z (x,-g)(y,-y) t=1

Rxr =

~

(1)

This can be calculated using analogue electronic techniques but involves fairly complex circuitry. By assuming that the data of each averaged a.e.c.p, have zero-mean properties, then eqn. 1 becomes N

Z (x,)(y,)

(

; ......

i=l

i+1

~l IYnl '

Fig. 4 The stages involved in the determination of the simplified correlation coefficient Rxv

Trigger i/P

|

(2)

1=1

The assumption of zero-mean properties implies that the analogue data has a mean value of 0 volts; a reasonable assumption as visual inspection of Fig. 1 shows. Experimental confirmation is given later in Fig. 3. This situation is achieved experimentally by using an e.e.g, amplifier with a low-frequency cutoff of 0-5 Hz (_+ 3 dB). However, calculation of eqn. 2

[~

Rxy ,'- ~

Referencevoltage +

- -

Reference voltage ~Full wave ~Gates=, rectifiers 9 Summators9

4

Output voltage

indicators

~

Fig. 5 Detection instrument

Medical & Biological Engineering & Computing

November 1977

643

using analogue electronics is still difficult and a further modification is applied to simplify it still further, as follows:

patient and averaged, the correlation is executed as follows. The instrument is first reset; this establishes 1

N

!

E (x,)(y,)

AUDI OME1,ER

i=1

R=y =

. N

.

.

.

(3)

N

Y, (ix, l ) E (ly,1) i=1

1-RI GGER S I GNAL

i=l

Fig. 2 shows this modification is valid under the conditions of the experiment and Fig. 3 indicates how well the assumption of zero-mean properties holds for this simplified correlation coefficient. The block diagram in Fig. 4 demonstrates that analogue electronic techniques are now an acceptable method of evaluating eqn. 3, since the number of arithmetic functions have been considerably reduced. The schematic circuit in Fig. 5 is used to determine the modified correlation coefficient, where full-wave rectification is employed for the moduli calculations, with long time-constant integrators for the summation stages (National Semiconductor Applications Handbook) and integrated circuit multipliers/ dividers (Analog Devices i.c. type AD533J) for the remaining arithmetic functions. The complete detection instrument is shown in Fig. 6. A schematic diagram of the electric-response audiometry system, including the online responsedetection instrument, is shown in Fig. 7. A modified Kamplex diagnostic audiometer produces stimuli of 200 ms duration with a 10 ms rise and fall time at a repetition frequency of 0.5 Hz. The alternate outputs circuit enables two traces to be averaged by employing a split-sweep method similar to that described by L o w v and WEISS (1968), each trace consisting of 32 individual responses which are summated using a Data Laboratories DL102A signal averager. The sample-and-hold circuit enables the averaged data of both traces to be output simultaneously. After the a.e.c.p.s have been recorded from the

Fig. 6 Complete detection instrument 644

/

S I GNAL AVERAOER T

y

MONITOR C.R.O.

. TAPE RECORDER

1

X-Y PLOTTER 2

T 1 2

SIGNAL CORRELATOR

Fig. 7 Electric-response-audiometry system including the online response-detection instrument

the initial conditions of the integrators. The averaged analogue data from both channels of the signal averager is then output into the detection instrument, the analysis being performed sequentially during this output period. The final result for the correlation coefficient, which is represented by the output voltage from the circuit of the instrument, is indicated at the end of this period by illumination of two lights on the front panel. One light illuminated shows a significant result at the 5 ~ level (p 0' 05) and both lights a significant result at the 1 level (p = 0.01) as indicated in Figs. 8 and 6. A typical set of averaged a.e.c.p.s, used for a threshold determination was shown in Fig. 1. The bearing-level threshold is then determined from the lowest consecutive pair of stimuli intensities at a 10 dB interval and at the same frequency, which give positive results for the averaged a.e.c.p.s at the 5 ~ significance level. If the lower of these two stimuli gives a result which is only significant at the 5~o level, the bearing-level

Medical & Biological Engineering & Computing

November 1977

threshold is positioned at that stimulus intensity; if it is also significant at the 1~ level, the threshold is put at 5 dB below. Using this criterion, any spurious 3.0 --

2.0

].0

--

5% 1%

1

i, t -0.5

0

I 0.5 TruecorrelationcoefficientRxy

9 1.0

The majority of these thresholds lie within the + 10 dB limits, having a correlation coefficient of 0.94. These results compare very favourably with those given in Fig. 10 where the objective threshold is determined from subjective assessment of the averaged a.e.c.p.s. These thresholds correlate with a value for the correlation coefficient of 0.92. In Fig. 10, there are two type-II errors which are outside the 10 dB limit. By the use of the detection instrument, these two results are brought within this limit. I n Fig. 9 there are no type-lI errors outside the 10 dB limit, but there is an additional type-I error introduced by the detection instrument. This is an acceptable situation since it is clinically preferable to underestimate rather than overestimate the hearing capabilities of patients so that further tests may be conducted. It is also interesting to note the accuracy of the results at the higher hearing-level thresholds. The detection instrument has given considerably more accurate thresholds at these higher levels than those from subjective assessment of the averaged a.e.c.p.s. The results indicate that deviations from the hearing-level thresholds determined from conventional pure-tone audiometry are comparable if not reduced by using the detection instrument in place of our subjective assessment of the averaged a.e.c.p.s. 4 Discussion

-1.0 Fig. 8 Comparison of the output voltage from the detection instrument with computed values of the true correlation coefficient R xy

As evoked-response audiometry evolves from a research tool to a routine clinical test, it becomes important to consider the interpretation of the results in this situation. Whereas a research laboratory will

positive results caused by artefactual disturbances 70will not result in a false threshold since the probability of the results from a stimulus intensity _ 60 of 10 dB above or below this also being a false positive is small, and hence the spurious result will be ignored. 50The hearing-level thresholds of 24 subjects, ranging inage from 16 to 57 years, were determined using the detection instrument. The subjects were selected 2E8 to include a range of levels of hearing loss and were all able to co-operate well in the conventional .~ 30subjective pure-tone audiometric procedure. The // results from the detection instrument were compared 20- , / " with those from pure-tone audiometry and from online subjective assessment of the averaged a.e.c.p.s. The subjective assessment of the response was performed jointly by two of the authors who yL/. between them have had several years experience of I/ , } / -10 / I /,q0 e.r.a.

+ lOdBLimitsJ /o /,/,/" o / ~

/"

/

//

//,

I / / / f /

//

///,,/s. /

,/ / /////

///'///

/" //

FYT," I :

I t 30 40 " / ~ e t e c t i 0 n I

20

.

I I TI 50 60 70 instrumerit

3 Results

A comparison of hearing-level thresholds determined from conventional audiometry and those given by the detection instrument is shown in Fig. 9. M e d i c a l & Biological Engineering & C o m p u t i n g

Fig. 9 Hearing-level threshods given by the detection instrument compared with those from conventional pure-tone audiometry N o v e m b e r 1977

645

holds from a combination of subjective assessment of the response and the detection instrument; both results have to be positive before a response is assessed as being present. This reduces the number of type-I errors recorded and has allowed greater +- 10 dB Limits ,,o// reliability to be placed on our estimation of ' - / thresholds. Although the detection instrument is used here r /ssJ with a Data Laboratories 200 pt. signal averager, it can be easily interfaced to other simple e.r.a, equipment and signal averagers provided that 2-channel averaging and data-output facilities are available. It ," ,/" could therefore be a useful addition to many routine clinics using e.r.a, as a means of threshold determination. It is simple and cheap to build and //" initial results show good reliability in the detection S ~" of responses.

have online computing facilities, it is unusual to find them available in the clinical audiometry service. Unfortunately it is here that some kind of objective

ejsls/

70 --

l/S

60 --

/*

50-

i,0~

/

E

Sp

o 30--

~. ,,/ ./" ~ zo..- ,,"~./-,,"""

/" /

a:l

~"

I0,z- 9 / ~ _ / " ##~* / ., #

V

,/"

,,,"

J

I/

20 30 40 50 60 70 HL threshold from subjective assessmentof the A.E.C.P.

/'/0

-IL0 /

,I-10 Fig.

s

//dB

10 Hearing-level thresholds determined from subjective assessment of the same a.e.c.p.s as in Fig. 9. These thresholds are again compared with those from conventional audio-

metry assessment of the response is most necessary for, in contrast to the research laboratory where there will be a small number of experienced observers, the results may be assessed by different less-experienced people at different times. When e.r.a, is used in the paediatric field, assessment of the response often becomes increasingly difficult, particularly for the less experienced observer due to excessive e.e.g. activity and movement artefacts. In our initial routine testing of handicapped children, we have used the detection instrument cautiously and have derived our hearing-level thres-

References ANALOG I~EVICES. Integrated circuit multiplier type AD533J. Application note. BEAGLEY,H. A. (1973) Electrophysiological methods in the diagnosis and management of deafness. Minerva Otorinolar 23, 173-181. BEAUCHAMP, K. G. (1973) Signal processing--using analogue and digital techniques. Allen and Unwin Ltd., London. LowY, K. and W~ISS,B. (1968) Assessing the significance of averaged evoked potentials with an on-line computer: The split sweep method. Electroenceph. Clin. NeurophysioL 25, 177-180. NATIONALSEMICONDUCTOR.(1973)Linear i.e. handbook. (1973) Linear applications handbook. (1974) CMOS i.c.c, handbook. ROSE, D. E., KEATING, L. W., HEDGECOCK, L. D., SCHREURS, K. K. and MILLER, K. E. (1971) Aspects of acoustically evoked response--inter-judgeand i~trajudge reliability. Arch. Otolarying. 94, 347-350. SALOMON, G., BECK, O. and ELBERLINO,C. (1973) The role of sedation in e.r.a, from the vertex. Audiology 12, 150-166. SALOMON,G. (1974) Electric response audiometry (e.r.a.), based on rank correlation. 1bid. 13, 181-194. SCHIMMEL, H. (1967) The (+) reference: accuracy of estimated mean components in average response studies. Science 157, 92-94.

Un appareil simple de ddtection directe des potentiels corticaux suscites par I'appareil autidif Sommaire--Est d~crit un appareil 61ectronique analogique de conception simple qui fournit une indication objective positive ou n6gative sur la pr6sence d'un potentiel cortical suscit6 par l'appareil auditif (a.e.c.p.). I1 6tablit une corr61ation entre deux a.e.c.p, obtenus en faisant la moyenne des r6ponses alternantes avec un stimulus d'intensit6 et de fr6quence constantes. Comme crit6re permettant de d6terminer s'il y a ou non une r6ponse, on utilise les d6riv6es des niveaux significatifs de 5 et de 1 ~ du coefficient de corr61ation. On emploie 6galement un coefficient de corr61ation modifi6 qui permet de simplifier consid6rablement le syst6me de circuits en question. Les r6sultats des 24 seuils de niveau auditif d6termin~s par l'appareil de d6tection concordent davantage avec l'audiom6trie traditiomaelle de tonalit6 pure que les r6sultats obtenus h partir d'une 6valuation subjective du a.e.c.p, moyen. 646

Medical & Biological Engineering & Computing

November 1977

Ein einfacher On-Line-Anzeiger geh6rsbez0glicher Kortikalspannungen Zusammenfassung--Ein einfaches elektronisches Analoginstrument wird beschrieben, das die Gegenwart von geh6rsbeziiglichen Kortikaisparmungen durch eine objektive Ja/Nr162 anzeigt. Es bringt zwei solche Kortikalspannungen, die durch die Mittelbildung yon Alternativreaktionen in einer Reizfolge gleichbleibender Intensit/it und Frequenz erzielt wurden, in Wechselbeziehung. Das 5%und 1%-Giiltigkeitsniveau des Korrelationskoeffizienten wurde abgr und als Kriterium fiir die Gegenwart einer Rcaktion verwendet. Die Verwendung eines modifizierten Korrelationskoeffizienten erm6glicht eine betrfichtliche Vereinfachung der erforderlichen Verdrahtungstechnik. Die Ergebnissr von 24 Gehorschwel|enbestimmungen mit Hilfe des AnzeigegerS,tes stimmen besser mit der herk6mmlichen Tonaudiometrie iiberein als die Ergebnisse subjektiver Beurteilung der Mittelwerte yon geh6rsbeztiglichen Kortikalspanntmgen.

Medical & Biological Engineering & Computing

November 1977

647