Memoire Online - Contrainte Psycho-Physiques et Electrophysiologiques sur le codage de la stimulation électrique chez les sujets porteurs d'un implant cochléaire

L'objectif de cet article est de faire une revue étendue des différents types de traitement numérique des PEAP et d'essayer de comprendre leur principe.

L'article décrit ensuite, à partir de modèles de conduction nerveuse, un nouveau type de filtrage où la fonction de transfert varie avec la latence. En effet, le spectre des PEAEP et / ou PEAP devient de plus en plus grave en fonction de la latence.

Cette nouvelle technique de filtrage tout d'abord simulée sur des modèles de PEAEP bruités puis sur des PEAEP physiologiques a montré une très bonne robustesse vis à vis de différent types de bruits qui viennent se superposer aux tracés:

- Bruit biologique spécifique correspondant à des réponses non-auditives dues à la stimulation électrique (vestibule, muscle, nerf facial).

Pour finir l'article montre que les caractéristiques des PEAEP après traitement numérique ne sont pas différentes de celles publiées par d'autres auteurs sans ce type de filtre.

DIGITAL TIME-VARIANT FILTERS ADAPTED FOR RECORDING OF
ELECTRICALLY AUDITORY BRAINSTEM RESPONSES (E-ABR).

ABSTRACT : The electrically evoked auditory brainstem response (EABR) is useful in objectively measuring the auditory system's response to stimulation via cochlear implant. However, the observed waveform typically is distorted by the electric artifact produced by the implant and other nonauditory signais (i. e. myogenic and vestibular). This paper describes a digital, time-variant filter designed specifically for surface-recorded electrical activity of the hearing system's afferent pathways, specifically under conditions of electrical stimulation. It thus was our objective to improve the quality of EABRs. The transfer function varies along the time continuum, in deference to nuances of neural propagation along a chain of neurons (i.e. the afferent pathway). Simulation showed that this function can extract EABRs in a noisy environment, that is with signal-to-noise ratio (SNRs) of less than -36 dB and facilitate measurement of wave latencies under such noisy conditions. lndeed, we demonstrate detection component wave without significant temporal distortion (i.e. latency shifts) after filtering of noisy EABRs of SNRs down to --24 dB. Such filtering also can reduce substantially both electrical artifact and non-auditory waves of the EABR. Use of this filter under real recording conditions permitted us to measure reliably latencies and interpeak intervals for waves II, III, and V, and observed values compared favourably with data of other authors. We were able to record and measure EABRs even with the most basal electrodes. No significant differences were found according to electrode number.

The auditory brainstem response (ABR) (Sohmer and Feinmesser, 1967 ; Jewett and Williston, 1971) is well-known and routine in clinical assessments of the auditory system. It is used to assess quickly the functioning state of the afferent auditory system, namely from the cochlea through the pontine brain-stem pathways. ABR via acoustical stimulation has been known for at least 25 years (Jewett and Williston, 1971). However, interest in electrically-stimulated auditory responses also is long- standing and was fuelled further by the development of the cochlear implant (House et al., 1976; Michelson, 1971; Simmons, 1966; Starr and Brackman, 1979). Use of the EABR to facilitate adjustment of cochlear implant, namely by measurement of the functional status of the auditory system (characterization of EABR I/O functions) and estimation of the threshold level (TL) (measure of apearance of EABR), is of interest. In general, it permits objective measures to be used in cases where subjective responses would be doubtful. However, recording of electrically evoked responses tend to be plagued with electrical stimulus artifact and with responses from nonauditory systems stimulated by the spread of current from the implanted device.

Because the ABR is minute of signal (sub-microvolt range), it is, in general, vulnerable to interference from a variety of signais and/or noise, both physical and physiological. Much work has been dedicated to the processing of acoustically stimulated ABR to minimize the effects of such interference. For example stimulation using alternating polarity clicks helps the suppression of microphonic potentiel and stimulation artifact. Analog filtering following the differential preamplifiers stage of the bioelectric amplifier decreases noise in the recording. Digital filters can improve greatly the signal-to-noise ratio of the ABR (Fridman et al., 1982; Urbach and Pratt, 1986; Pratt et al., 1989; Grônfors et al., 1992,1993) with the advantage of zero phase-shift (Boston and Ainslie,1980). These or comparable approaches to the processing of the EABR, however, have been much less successful.

That interference from electrical stimulus artifact is a considerable problem and is readily appreciated from the fact that it can be over a million-fold greater than the ABR itself. Thus, great care must be taken to avoid saturation of amplifier input and ringing of the analog filter at the preamplification stage. The stimulus artifact also can

last a relatively long time (i.e. up to several milliseconds). Most cochlear implants utilize inductive coupling and the transmission of information from the external device to the implanted package, namely via, a pulse-modulated carrier wave.

Interference from nonauditory signais may arise from several possible sources. Aside from the common source (e.q. electroencephalogram and electrocardiogram), the current may spread to the facial nerve and cause strong interference by myogenic potentiels. There also is the possibility of stimulating the vestibular apparatus and evoking a response from this system.

Techniques used for conventional ABR measurement are less effective with the EABR. The typical analog filtering (100-3000 Hz; -3dB, RC/passive) cannot be used as the artifact actually is prolonged, particularly if the high-frequency cut-off frequency is too low or the low-frequency cut-off frequency is too high. With the EABR, the input analog fiiter ideally should have a wide bandpass (Van den Honert and Stypulkowski, 1986). Digital filtering used for the acoustical ABR also cannot be used, due to the shift in the EABR, namely toward shorter latencies by virtue of the elimination of sound- and hydromechanical-wave propagation and synaptic delays in the auditory periphery. This can cause dramatic distortions, especially during the first milliseconds of the analysis window.

To avert these problems, several techniques can be used. It is possible to put an analog fiiter in the front end of the differential amplifiers (Clarke et al., 1990) to attenuate the electromagnetic interference coming from the transmission of information through the skin, from the external device. If the signal is by-passed during the first milliseconds of a recording (so-called 'blanking') it also is possible to avoid the saturation of the head differential amplifiers. Of course, the waves occurring simuitaneously with the artifact are lost. Techniques based on the subtraction of a model of the artifact, or taking into account the shift have been used (Durrant and Krieger, 1996). However, this signal processing and/or editing can introduce flaws, proportional to the magnitude of the artifact of stimulation. Several cochlear implants also allow the stimulation pulse to be biphasic in order to decrease the influence of the artifact on the recording. Nevertheless, the success of EABR recording still is strongly affected by the capabilities of the recording amplifier (not equal among

systems/manufacturers). The techniques just described are by no means optimal, and not all of these techniques are possible via available ABR test systems and/or within the expertise of the examiner.

There thus has been little work to establish signal processing methods specifically for the EABR. The technique described herein was developed in consideration of the properties of the EABR and was tested both by simulation and in actual implanted subjects. Although the clinical application of interest here is the post-implant evaluation via EABR measurement, it also is conceivable to use this technique for pre-operative testing in the implant candidate, i.e. via promontory stimulation using a transtympanic electrode. This technique also should be adapted ultimately to the study of electrically evoked middle- and long- latency auditory evoked potentials and to the EABR obtained with a brainstem implant.

The cochlear implant employed in this study was the DIGISONIC, a 15-electrode transcutaneous device (Beliaeff et al., 1994). It is manufactured by the firm MXM (Antibes, France). It is composed of two distinct parts; an external and an internai component (figure 1). The external device performs an acoustical signal analysis leading to the extraction of relevant speech information. This information is passed to the internai (implanted) device. The signal processing is based on the Fast Fourier Transformer. The external device computes, in real time, the spectrum of the acoustical signal for 64 bands of 122 Hz in the range of 0 to 7.8 kHz. The acoustical signal is sampled at 15.6 kHz. The analysis window is 128 points. Frequency bands are grouped before being distributed to the electrodes. The transcutaneous interface is composed of two antennas, one emitting and the other receiving. Generally, active electrodes are located from 5 to 20 mm from the base. The internai part performs the decoding of the information received by the internai antenna and stimulates the programmed electrodes.

Figure 1: Block diagram of the DX10 DIGISONIC cochlear implant. Information is transmitted from the speech processor to the internai part on a carrier, wave using amplitude modulation. The carrier frequency was 4 MHz.

Stimulation was accomplished via the Common Ground mode, wherein the addressed electrode is activated and ail the others are connected to ground. The stimulus is produced by a pulse generator; the pulse is of a constant amplitude, but variable in length. As illustrated in figure 2, the equivalent circuit of each electrode/channel of stimulation is a series capacitor and resistor, fed by the pulse generator. The stimulus delivered is biphasic and asymmetric. Adjustment of the series capacitor permits very precise equalization of the positive and negative charges transferred which, in turn, is extremely important for preservation of physiologie integrity of the tissues. The pulse duration ranges from 5 to 310 ps. The pulse generator can source 3 mA over a 2 kohm load. For recording EABRs, the external package is replaced by a specialized device manufactured by MXM, the DIGISTIM. The DIGISTIM is battery-powered and controiled by a persona! computer through an opto-isolated serial port. This system allows the generation of a pulse with several adjustable parameters, i.e. choice of electrode, magnitude, pulse duration, and stimulation frequency. Synchronization of the evoked potential test system is made possible by an external trigger pulse produced by DIGISTIM.

C=150 nF

R=1 kohm

Figure 2: Simplified description of the DIGISONIC cochlear implant stimulation in the labyrinth. Impedance can be represented by a pure resistance (about 2 kohm in vivo). Mean amplitude is zero volt. Positive and negative parts of the pulse are not symmetrical but charges are equal. Phase-Iocking of the fibers occurs mostly on the positive part.

The ABR represents electrical fields sampled on the scalp. These fields are produced by sources remote to the electrodes; the recorded signais thus are characterized as being far-field potentials. They are the consequence of nerve-impulse propagation along the auditory pathways from the inner ear to the inferior colliculus at the upper margin of the pontine portion of the brain stem (for review, see Durrant and Lovrinic, 1995 and Durrant and Wolf, 1991). The stimulation of choice for these characteristically short-latency potentials (i.e. with principle waves falling within 10 ms.) is a brief transient, such as the acoustic click, which excites a relatively wide population of nerve fibers (due to spectral splatter) with excellent synchronisation (due to temporal precision of the stimulus). The ABR primarily reflects, therefore, afferent activity of auditory nerve fibers in the Vllith nerve and the brainstem (Moore, 1987b). Each wave cornes from a virtual source which can be represented as a dipole with an amplitude and an angle (phase and space) (Williston et al., 1981; Pratt et al., 1983; Scherg and Cramon, 1985). Consequently, a given pair of recording electrodes, that is sampling between any two points on the head, registers an overall waveform that is the resultant of the combined phasor and vector summation of the overlapping time-series of compound potentials excited at the various levels of the

peripheral and brain-stem auditory pathways. Waveform morphology thus is influenced not only by stimulus spectral and temporal characteristics, but also orientation of the recording `dipole' with respect to those of constituent sources dipoles.

Clinical methods of recording employ disc electrodes (e.g. silver or gold) with an electrolyte paste affixed to the skin. At moderate intense stimulus levels, the typical ABR appears as a series of 5-7 waves occuring within a time frame of 10 milliseconds. The EABR, however, typically shows only three waves in a 6ms time interval (waves II, III & V, in reference to the acoustically evoked response) (Van den Honert and Stypulkowski, 1986; Abbas and Brown, 1988; Shallop et al., 1990). Nevertheless, the generation of EABRs waves is fundamentally the same as for the ABR, but with the input stage of the system (the organ of Corti) being effectively`bypassed (Van den Honert and Stypulkowski, 1986; Waring, 1995). This eliminates the initial propagation and synaptic delays, moving the ensemble earlier into the measurement time window. Wave I, from distal Vlllth nerve usually is obscured by the stimulus artifact, regardless of method of analysis. Wave II derives from the transition between the VIllth nerve and the passing into the brainstem (WIler et al., 1988). Wave III is generated when the nervous signal leaves the cochlear nucleus. The IV-V complex originates somewhere along the contrelateral lemniscus to the controlateral inferior colliculus (Moore, 1987b; Millier and Jannetta, 1982; Legatt et al., 1988; and Durrant et al., 1994).

Figure 3: Aspect of the compound nerve impulse related to the distance from the stimulation.

Most of the afferent auditory nerve fibers (90% or more) are myelinated (Spoêndlin and Schrott, 1988, 1989; Moore, 1987a,b). The fiber diameter is characterized by a
normal distribution centred between 2 and 44.im .(Spoëndlin and Schrott, 1988, 1989),

as is conduction velocity. Furthermore, the farther from the origin of stimulation, the Iess synchronous are the neural potentials. Consequently, the density of discharge of the fibers also is typically represented by a normal distribution which becomes increasingly broader the farther the potentials are sampled from the source. Namely, the width increases Iinearly with the distance (Figure 3). Normative studies of the ABR support this theory (John et al., 1982). This fact however, is somewhat obscured in the ABR waveform, because of the diverse origins of wave I to V (Zapala et al., 1992), further complicated by temporal dispersion due to the frequency-place encoding performed by the cochlea (Don and Eggermont, 1978; Gorga et al., 1988). This resuits in the superposition of wave components, depending upon stimulus intensity (Zapala et al., 1992). Electrical stimulation, however, simplifies matters in that the responses originate from essentially the same region of the VIllth nerve (i.e., again, cochlear propagation delays are eliminated).

The Pathfinder II (Nicolet Biomedical) was used to record the EABRs. The ipsilateral montage commonly employed clinically for recording the ABR was used here: the noninverting input of the preamplifier was connected to a gold-cup electrode at hairline, at midiine, and the inverting input was connected to an electrode placed on the ipsilateral earlobe. The ground electrode was affixed to the contralateral earlobe. The recording parameters employed, however, were slightly different of those used in ABR protocol. Full-scale sensitivity was +/-50pV. A wide bandpass (0.2 Hz-8000 Hz) was used. The sampling rate (50kHz) was well above the highest frequency of interest in the EABR; such oversampling is desirable in order to minimize the influence of noise (Grônfors and Juhola, 1995). Each response recording derived from an average of 512 stimulus repetitions. This is a relatively low N by clinicat standards, but it should be born in mind that the subjects were under general anesthesia. Consequently, time was of the essence, but some sources of noise (i.e. myogenic from movement) were naturally reduced. A test and two retests of EABR were performed for each stimulus condition for purposes of determining reproducibility of the responses.

Stimulation, again was delivered by the Digistim, triggering the Pathfinder II. In our study, for each patient, we tested from 4 to 9 electrodes equally spaced along the array from base to apex. For each electrode, we have recorded a threshold among 16 intensity levels, along a linear scale from comfort level (CL) to 0. The duration of the pulse was chosen a the parameter by which to vary intensity. For each condition, we made three recording. The stimulus rate was 60 Hz, chosen to be above the line frequency used in Europe (50Hz). (This clearly would not be an appropriate choice in countries like the United States and Canada where the line frequency is 60 Hz.). This frequency admittedly is to high for ABRs elicited by acoustic stimulation (i.e. due adaptation), yet it not poses problems, in our experience, with electrical stimulation. This is because adaptation for electrical stimulation is more dependent upon refractory period than pure adaptation (Brown et al., 1990 ; Abbas and Brown, 1991 b). After acquisition, the data were transferred to a personal computer for further processing, where upon the filter algorithm was implemented and the filter function applied. Consequently, the choice of system for data acquisition is not critical to the analysis procedure.

Considering the neuroanatomical and neurophysiological bases of the ABR, the width of the component waves are expected to increase with latency, which in fact occurs overall. Consequently, each wave has not only a characteristic latency, but also a characteristic spectrum. Wave I then is expected to be more easily detected using a high frequency filter; the opposite is expected for wave V, since, for longerlatency waves, the spectrum is more robust in the lower frequencies. However, the filters previously described for ABR/EABR have fixed transfer functions. The choice of bandpass is made according the ABR/EABR's average spectrum. In more recent studies, on the other hand, Pratt (1989) and Gronfors (1993) have described digital filters adapted to each wave, i.e. by latency.

The filter involved a transfer function designed according to certain nuances of neural conduction. As shown in Fig. 3, the compound action potentials which develop along a neural pathway are products of the conduction velocity and axonal length. If the spectra of such signais are measured , there is observed a band-pals whose center

frequency follows approximately a hyperbolic function. For proprietary reasons (re future product development), the details of the filter function cannot be divulged. Nevertheless, certain important characteristics can be described.

The conduction velocity depends upon diameter of the nerve fiber; as the diameters follow a Gaussian distribution (Spoendlin, 1988, 1989), the waveforms perhaps can be modelled by a Gaussian pulse whose width influences the latency function. The formula approximating the wave can be written as follows:

Where tO corresponds to a latency for which the response is at the maximum, b is the time constant of the wave, and A is the wave's amplitude at t=t0.

As the form of the wave presumably follows a naturel law, the convolution of the wave by a Gaussian pulse of the same half-life duration centered on zero will maximize the peak of a wave. The distribution of the theoretical wave is calculated, starting from work in this laboratory and values in the literature (Gallego et al., 1996, 1997; Abbas and Brown, 1988).

Where b is the time constant of the theoretical wave and E(t) is the waveform centered on the point where one wishes to detect a peak.

The filter that we have developed was inspired by the formula above, however, it takes into account only a part of the total EABR wave, corresponding to essentially 95 % of the energy of the theoretical wave (t = [ -1.96-2b, 1.96+2b]).

This filter is designed with the goal of maximizing the peaks of the EABR waves and trying to protect the integrity of the EABR while eliminating those waves not belonging to the EABR proper. We thus endeavour to eliminate all waves that do not correspond to the desired ones, that is to say, those that do not fall within plus-or-

minus three standard déviations. To- conctruct the theoretic EABR, we -have based it

on the wave III. In effect, this wave, contrary to wave II and IV-V complex, is well isolated from the others. There thus is no overlapping of wave III by the others, so it can easily be dissociated from EABR (fig. 5a). The duration of wave III also is the most stable. Utilizing a conventional digital filter (filter pass band of 150-3000 Hz (Gallégo et al., 1996)), we have measured the width of wave III in a sample of 10 implanted subjects with the Digisonic. The width of wave III is 1.05 ms with a standard deviation of 0.138 and a peak latency of 2.06 ms with a standard deviation of 0.186.1f one considers that this width corresponds to approximately 95 % of the energy of the wave (this would be +/- 2b'), one thus can measure the time constant of wave III (b' = 1.05/4 = 0.26 ms). We, therefore, have filtered the signal in the manner that only waves falling within plus-or-minus three standard deviations ( for t=2.06 ms. b' = [0.16,0.37]). We have made the assumption that at the instant of t=0, the neural influx evoked by electric stimulus was perfectly synchronized. We then varied b' linearly as function of the latency such that at t=2.06, b'=0.25 and b' varies linearly as a function of t.

Figure 4 : Transfer function of the numeric filter when the latency is 2,06 ms and b=0,288 ms.

The formula utilized experimentally is only an approximate of the theoretical one. The
analysis window is from 4 b'(t) to infinity, so the equation becomes:

Where b' +1- 3 sd(t) takes the values following the function of latency: for t=0.5 ms b'(t)-3sd(t)=0.04 ms, b'(t)-3sd(t)=0.10 ms ; t=1 ms b'(t)-3sd(t)=0.08 ms, bi(t)- 3sd(t)=0.20 ms ; t=2 ms b'(t)-3sd(t)=0.16 ms, b'(t)-3sd(t)=0.40 ms....

Figure 4 shows the filter transfer function for a 2.06-ms-latency component. ( This filter function is simulated for 1024 points, ranging from 100 Hz to 15 kHz on a logarithmic scale.) This function can be represented by a bandpass filter with a peak at 750 Hz. The shape of the filter's frequency response is maintained for components of other latencies. There is a slight linear shift with latency: the peak is at 3kHz for t=0.51ms, at 1.5kHz for t=1.03ms, at 750Hz for t=2.06ms, at 375 Hz for t=4.12ms

The EABRs were sampled over a 10.24 ms window to yield a time-ensemble of 512 data points. However, only the data points representing latencies from 0.5 to 8 ms were considered in the calculation of the filtered response. Like Durrant and Krieger (1996), we found some truncation of the initial part of recording to be essential. On the other hand, experience simply dictates that period beyond 8ms is of no value in assessing EABRs.

After having filtered the three tracings from the same intensity of stimulation, we calculated a mean. To facilitate visualization of the response, we then blanked the first 0.5 ms of the trace to eliminate (or minimize) stimulus artifact. To permit objective determination of the presence or absence EABRs, we calculated a 3x3 matrix of cross-correlations among the three traces (note above). The window studied extends from 1 to 7 ms (300 data points). The EABR was considered to be presence when the cross correlation was greater or equal to 0.15 for two out of three traces (Pearson correlation : p=0.01).

Figure 5: EABRs before (5a) and after (5b) digital processing. After filtering the three curves, the intercorrelation function leads to a time shift equal to -20ps between the first and the second averages (Pearson coefficient=0,85), -60ps between the second and the third averages (Pearson coefficient=0,93), 20ps between the first and the third averages (Pearson coefficient=0,91). The detection of the waves indicates LII=1,22ms, LIII=1,98ms, LV=3,74ms, A1l=0,52pV, Al11=0,99pV, AV=0,57pV. (It also is possible to demonstrate wave IV, but focus upon waves II, III, and V as they were the most robust in our recording).

Figure 5 shows a EABR before and after using the digital filter. Latencies of the waves are not affected. With a simple peak-finding subroutine, latencies and amplitudes of the waves can be easily measured. The short-latency components are well separated from the stimulus artifact.

Latency of wave II (ms) Latency of wave III (ms) Latency of wave V (ms) wave H-III interval (ms) wave III-V interval (ms)

Table I: Average EABR latencies ( 11 subjects, for 58 tested electrodes) compared to values observed in other studies (Van den Honert and Stypulkowski, 1986 ; Abbas et Brown, 1988 ; Kasper et al., 1992). A comparison of the mean values has been performed using a comparison of mean values statistical test (NS: no significant difference ; ***: significant difference, less than 0,05 ; ?:no tested).

The robustness of the noise reduction of this filter function also was assessed. A model EABR was constructed based upon results of other researchers (table I, right part) and from our own laboratory (Gallégo et al., 1996, 1997) and imbedded in white noise for several signal-to-noise ratios (SNRs). The model EABR comprised component waves having latencies of 1.28ms, 2.06ms and 3.86ms respectively. Corresponding widths are 0.71ms, 1.15ms, and 2.15ms.

Figure 6 shows the effect of the addition to the model EABR a white noise before and after digital filtering (the SNR was varied between +18 to --24 dB in steps of 6 dB). Despite the addition of noise, it is notable that wave II, III and V always can be identified. There latencies are very stable.

Figure 6: Resistance to noise of the filter function. A white noise has been added to the signal (EABRs + white noise) ; signal is represented before (left) and after (right) this signal processing. The signal-to-noise ratio varied from +18 to --24 dB in steps of 6 dB.

Figure 7 compares the evolution of the cross-correlation, between the original mode! EABR and that buried in noise, with and without digital filtering (with SNR varied between +18 and --48 dB in 6 dB steps). Each point corresponds to an average of 100 repetitions. (Also indicated are the standard errors). Without digital filtering, we could detect the EABR only down to SNRs of --24 dB. With digital filtering, we could detect the EABR down to --36 dB SNR. For noisier waveform, there no longer was a significant correlation between the noisy signal and the original signal.

Figure 7: Cross-correlation between a model EABR and model EABR buried in noise (white noise added, bandwidth of 25,000 Hz) as a function of the signal-to-noise ratio. Each point corresponds to a mean of 100 repetitions (mean and standard errors) before (open circles) and after filtering (close circles). The dashed lines (for R = 0.15) demark the limit of validity of the measurement of EABR (p=0.01).

Figure 8: Mean of absolute latency shifts (intercorrelation) between a model EABR and model EABR buried in noise (white noise added, bandwidth of 25,000 Hz) as a function of the signal-to-noise ratio. Each point corresponds to a mean of 100 repetitions (mean and standard errors) before (open circles) and after filtering (close circles). The dashed lines (for t = 50-100 ps) demark the limit of validity latency of the EABR (from 2.5 to 5 data points). Below this zone we consider that there is no significant displacement between the two curves. Above this zone, the model EABR in noise with and without filtering are two different.

Figure 8 compares the variations of the maxima of inter-correlation between the noisy EABR-(filtered or not) and the original EABR. This gives an idea of the changes in latency of the waves due to the addition of the noise. Each point corresponds to an absolute value of the average of 100 repetitions. (Also indicated are the standard errors.). The latency shifts are contained within +0.5 ms (with an mean value of 0.25 ms representing chance). The latency shifts can be considered negligible for mean values from 50 to 100 ps (2.5 to 5 data points). The digital filter does not cause latency shifts for SNRs down to --24 dB. For

Figure 9: Demonstration of suppression of interfering signais (myogenic, vestibular, etc.). Input signais, i.e. EABR + interference (Ieft-hand panel) are auto-scaled to full-scale, dictated in some case by the magnitude of the interfering signal. Filtered, output signais (right-hand panel) are plotted on the same scale.

Another test of the filter transfer function was to evaluate its ability to suppress waves
where the spectrum is incompatible in their latency with waves evoked by auditory
stimulation. It was noted above that such waves could derive from myogenic,

vestibular, or other sources. Figure 9 shows several model non-EABR waves. It can be seen in figure 7 that such waves are strongly attenuated.

In this work, EABRs have been recorded and filtered for 11 implantees, for several electrodes (a total of 58), and with different stimulation levels for each electrode (16 levels). For each recording we first determined if there was presence or absence of an EABR (again, cross-correlation of >= 0.15). Then, for the traces which were recognized as EABRs we measured the latencies and interpeak intervals of waves III, and V. We then averaged the values by electrodes, then by across patients. Table I indicates the latencies and interwave latencies for waves II, III, and V measured on 11 subjects. These average values have been compared with results given other authors on others cochlear implant system (Van den Honert and Stypulkowski, 1986; Abbas and Brown, 1988; Kasper et al., 1992 respectively tested on single channel 3M, multi-channel Nucleus and lnaired, multi-channel Inaired systems). Most of the comparisons are not significantly different (10 out of 12). Results obtained after filtering match the usual values given in the literature. EABRs were recorded even when the perception was that the stimulus was very faint.

EABR quality was found to be improved substantially by the use of a time-variant filter function adapted to the auditory system. Each wave undergoes filtering that is latency-specific and can be easily extracted from the overall EABR and background noise. No latency shift is introduced by the processing. Noise and signais commonly interfering with the EABR were reduced dramatically by this filter function, both in stimulations and actual recordings. Consequently, the myogenic wave typically occurring at about 6.0 ms (Fifer and Novak,1990) and the putative vestibular wave occurring at 0.5 ms (Van den Honert and Stypulkowski, 1986) were effectively suppressed. Indeed, most of the problems seen with EABRS are reduced by this filter function; including suppression of the de shift induced by the stimulus artifact. Furthermore, wave II which was not systematically seen in other studies (Van den Honert and Stypulkowski,1986) is always extracted.

Latencies of waves II, III, and V measured using the filtered responses were found to compare favourably with the values reported by other workers with other cochlear implant systems (Van den Honert and Stypulkowski, 1986; Abbas and Brown,1988; Kasper et aI.,1992). This agreement suggests our method to be valid and to yield accurate Iatency measures. The potential to record responses in all cases and with stimulation by all electrodes is particularly significant for intraoperative and clinical applications of EABR analysis. Altogether, 58 electrodes were tested in 11 patients. We thus did not have the sorts of failures reported by others (e.g. Van den Honert and Stypulkowski, 1986; Abbas and Brown, 1991).

Finaily, the time-variant filter function proved to be very robust in coping with noise added to model EABRs. The model EABRs were reliably recorded in broad-band noise to SNRs below -36 dB. Furthermore, this filter does not cause latency shifts for SNRs <= -24 dB. Consequently, EABRs can be recorded using low-level stimulation, as indeed was demonstrated by actual EABR recordings. Such performance is particularly important in working with paediatric patients in the clinical setting, wherein time is of the essence. Using this filter should substantially reduce the analysis time by reducing the number of stimulus repetitions needed for each average, thereby permitting the exploration of more electrodes.

The authors acknowledge those individuals and institutions who made this work possible: the MXM company, the Hospices Civils of Lyon, the Centre Nationale de Recherche Scientifique, the Université Claude Bernard, professors Eric Truy and Alain Morgon, and certainly the eleven implantees.

Abbas PJ, Brown CJ, Electrically evoked brainstem potentials in cochlear implant patients with multi-electrode stimulation, Hear. Res. vol.36, pp.153-162, 1988.

Abbas PJ, Brown CJ, Electrically evoked auditory brainstem response: Growth of the response with current level, Hear. Res. vol.51, pp.123-37, 1991a.

Abbas PJ, Brown CJ, Electrically evoked auditory brainstem response: refractory properties
and strength-duration functions, Hear. Res. voI51, pp.139-48, 1-991-13-.

Beliaeff M, Dubus P, Leveau J-M, Repetto J-C, Vincent P, Sound Signal Processing and Stimulation Coding of the DIGISONIC DX10 15-Channel Cochlear Implant, advances in Cochlear Implant Ed Hochmair-Desoyer IJ & Hochmair ES, pp.198-203, 1994.

Boston JR, Ainslie PJ, Effects of analogue and digital filtering on brainstem auditory evoked potentials, Electroenceph. Clin. Neurophysiol. vol.48, pp.361-364, 1980.

Brown CJ, Abbas PJ, Gantz B, Electrically evoked whole-nerve action potentials: parametric

data from human cochlear implant users, J. Acoust. Soc. Am. vol. 88, pp.2205-2210, 1990. Clarke CJA, Thomson DR, Gibson WPR, Measurement of auditory brainstem responses evoked

by electrical stimulation with a cochlear implant, Brit. J. Audiol. vol.24, pp.145-149, 1990.

Don M., Eggermont J.J., Analyses of click-evoked brainstem potentials in man using high-pass noise masking, J. Acoust. Soc. Am., vo1.63(4), pp.1084-1098, 1978.

Durrant J, Krieger D, Artifact reduction in evoked potential recording via polynomial curve fitting--feasibility study, Abstracts of the Nineteenth Midwinter Research Meeting of the Association for Research in Otolaryngology, pp.77, 1996.

Durrant J.D., Lovrinic J.H., Bases of Hearing science, 3rd ed. Williams & Wilkins, Baltimore, 1995.

Durrant J.D., Martin W.H., Hirsch B., Schwegler J., 3CLT ABR analyses in a human subject with

unilateral extirpation of the inferior colliculus, Hear.Res., vol.72, pp.99-107, 1994.

Durrant J.D., and Wolf K.E., Auditory evoked potentials: basic aspect. In Hearing Assessment, 2nd Ed., W.F. Rintelmann, ed. Pro-Ed: Austin, pp.321-381, 1991.

Fifer RC, Novak MA, Myogenic Influences on the Electrical Auditory Brainstem Response (EABR) in Humans, Laryngoscope, vol.100, pp.1180-1184, 1990.

Fridman J, John ER, Bergelson M, Kaiser JB, Baird HW, Application of filtering and automatic peak detection to brain stem auditory evoked potentials, Electroenceph. Clin. Neurophysiol., vol.53, pp.405-416, 1982.

Gallégo S, Micheyl C, Berger-Vachon C, Truy E, Morgon A, Collet L, Ipsilateral ABR with cochlear implant, Acta Otolaryngol.(Stockh), vol.116, pp.1604-1610, 1996.

Gallégo S, Truy E, Morgon A, Collet L. EABRs and surface potentials with a transcutaneous multielectrode cochlear implant. Acta Otolaryngol (Stockh), vol 117, pp.164-168, 1997.

Gorga MP, Kaminski JR, Beauchaine KA, Jesteadt W. Auditory brainstem responses to tone burst in normally hearing subjects, J. Speech. Hear. Res., vol.31, pp.87-97, 1988.

Grônfors T, Juhola M, Johansson R, Evaluation of some non recursive digital filters for signais of auditory evoked responses, Biol. Cybern., vo1.66, pp.533-536, 1992.

Grônfors T, Peak identification of auditory brainstem responses with multifilters and attributed

automaton, Computer Methods and Programs in Biomed., vol.40, pp.83-87, 1993.

Grônfors T, Juhola M, Effect of sampling frequencies and averaging resolution on medical

parameters of auditory brainstem responses, Comput. Biol. Med., vol.25(5), pp.447-454, 1995.
House WF, Berliner K, Crary W, Graham M, Luckey R, Norton N, Setters W, Tobin H, Urban J,

Wexler M, Cochlear implants, Ann. Otol. Rhinol. Laryngol., vol.85(suppl 27), pp.1-93, 1976.

Jewett DL, Williston JS, Auditory evoked far-fields averaged from the scalp of humans, Brain, vol.94, pp.681-696, 1971.

John ER, Baird H, Fridman J, Bergelson M, Normative values for brain stem auditory evoked potentials obtained by digital filtering and automatic peak detection, Electroenceph. Clin. Neurophysiol., vol.54, pp.153-160, 1982.

Kasper A, Pelizzone M, Montandon P, Electrically Evoked Auditory Brainstem Responses in Cochlear Implant Patients, O.R.L., vol.54, pp.285-294, 1992.

Legatt A.D., Arezzo J.C., and Vaughan H.G., The anatomic and physiologic bases of brain stem auditory evoked potentials, Neuroligic. Clinics., vol.6, pp.681-704, 1988.

Michelson RP, Electrical stimulation of human cochlear: A preliminary report, Arch. Otolaryngol., vol.93(3), pp.317, 1971.

MOler A.R., and Jannetta P.J., Evoked potentials from the inferior colliculus in man, Electroenceph. Clin. Neurophysiol., vol.53, pp.612-620, 1982.

Miller A.R., Jannetta P.J., Sekhar L.N., Contributions from the auditory nerve to the brainstem auditory evoked potentials (BAEPs): Results of intracranial recording in man, Electroenceph. Clin. Neurophysiol., vol.71, pp.198-221, 1988.

Moore JK, The human auditory brain stem: A comparative view, Hear. Res., vol.29, pp.1-32, 1987a.

Moore JK, The human auditory brain stem as a generator of auditory evoked potentials, Hear. Res., vol.29, pp.33-43, 1987b.

Pratt H., Har'el Z, and Golos E., Geometrical analysis of human three-channel Lissajous' trajectory of auditory brain-stem evoked potentials, Electroenceph. Clin. Neurophysiol., vol.56, pp.682-688, 1983.

Pratt H, Auditory Brainstem Evoked Potentials Peak Identification by Finite Impulse Response Digital Filters, Audiology, vol.28,pp.272-283, 1989.

Scherg M., von Cramon D., A new interpretation of the generators of BAEP wave I-V; Results of spatio-temporal dipole model, Electroenceph. Clin. Neurophysiol., vol.62, pp.290-299, 1985. Shallop JK, Beiter AL, Goin DW, Mischke RE, Electrically Evoked Auditory Brain Stem Responses (EABR) and Middle Latency Responses (EMLR) Obtained from Patients with the Nucleus Miltichannel Cochlear Implant, Ear Hear., vol.11(1), pp.1-15, 1990.

Simmons FB, Electrical stimulation of the auditory nerve in man, Arch. Otolaryngol., vol.84, pp.2-54, 1966.

Sohmer H, Feinmesser M, Cochlear action potentials recorded from the external ear in man, Ann. Otol. Rhinol. Laryngol., vol.76, pp.427-435, 1967.

Spoendlin H, Schrott A, The Spiral Ganglion and the Innervation of the Human Organ of Corti, Acta Otolaryngol.(Stockh), vol.105, pp.403-410, 1988.

Spoendlin H, Schrott A, Analysis of the human auditory nerve, Hear. Res., vol.43, pp.25-38, 1989.

Starr A, Brackman DE, Brainstem potentials evoked by electrical stimulation of the cochlea in human subjectes, Ann. Otol. Rhinol. Laryngol., vol.88, pp.550-560, 1979.

Urbach D, Pratt H, Application of finite impulse response digital filters to auditory brain-stem

evoked potentials, Electroenceph. Clin. Neurophysiol., vol.64, pp.269-273, 1986.

Van den Honert, Stypulkowski PH, Characterization of the electrically evoked auditory

brainstem response (ABR) in cats and humans, Hear Res , vol.21, pp.109-126, 1986.

Waring MD, Auditory brain-stem responses by electrical stimulation of the cochlear nucleus in

human subjects, Electroenceph. clin. Neurophysiol., vol.96, pp. 338-347, 1995.

Williston J.S., Jewett D.L., and Martin W.H., Planar curve analysis of the three-channel auditory

brainstem response: a preliminary report, Brain. Res., vol.223, pp.181-184, 1981.

Zapala DA, Gould HJ, Mendel MI, Place specific influence on the wave I to V interpeak latency

of the auditory brain-stem response, J. Acoust. Soc. Am., vol.92(6), pp.3174-3184, 1992.

Contrainte Psycho-Physiques et Electrophysiologiques sur le codage de la stimulation électrique chez les sujets porteurs d'un implant cochléaire

Article 14 :

DIGITAL TIME-VARIANT FILTERS ADAPTED FOR RECORDING OF ELLECTRICALLY AUDITORY
BRAINSTEM RESPONSES (E-ABR)

Contrainte Psycho-Physiques et Electrophysiologiques sur le codage de la stimulation électrique chez les sujets porteurs d'un implant cochléaire

Article 14 :

DIGITAL TIME-VARIANT FILTERS ADAPTED FOR RECORDING OF ELLECTRICALLY AUDITORYBRAINSTEM RESPONSES (E-ABR)

DIGITAL TIME-VARIANT FILTERS ADAPTED FOR RECORDING OF ELLECTRICALLY AUDITORY
BRAINSTEM RESPONSES (E-ABR)