- Email: [email protected]
Application of finite impulse response digital filters to auditory brain-stem evoked potentials
Electroencephalography and clinical Neurophysiology , 1986, 64:269-273 Elsevier Scientific Publishers Ireland, Ltd.
269
Technical Section APPLICATION O F FINITE I M P U L S E R E S P O N S E DIGITAL FILTERS T O AUDITORY BRAIN-STEM EVOKED P O T E N T I A L S D. URBACH and H. PRATT Microshev Ltd., Alon Shvut, and Evoked Potentials Laboratory, Israel Institute of Technology, Technion City, Haifa 32000 (Israel) (Accepted for publication: January 26, 1986)
Summary Finite impulse response digital filtering is a linear non-distorting operation and hence can be used to differentiate ABEP components. Since the power spectrum of the ABEP indicates that it contains more than two frequency bands, they can be distinguished by applying digital filters with special characteristics: up to 200 Hz (slow components), 200-400 Hz (medium components) and above 500 Hz (fast components). The results, using these filters, indicate a different nature of components II and IV as compared to I, III and V. While I1 and IV are not detected under 500 Hz and thus are called 'fast components,' I, III and V are the 'middle components' and appear above 200 Hz. The slow filter results in a 'pedestal' whose peak coincides with peak V. The results of this study show that the ABEP wave form is obtained by the superposition of at least 3 major waves having distinct frequency bands.
Keywords: finite impulse response - digital filter - A B E P
The increased use of digital filters in improving the recording of brain-stem auditory evoked potentials (ABEPs) (Boston and Ainslie 1980; Moiler 1980, 1983; Doyle and Hyde 1981; Fridman et al. 1982; John et al. 1982; Boston 1983; Takagi et al. 1983) has indicated their superiority over analog filters with respect to latency distortions. However, even when digital filters are used, the latencies of waves I - V depend on the type and the characteristics of the filter (Moiler 1983). Since latencies of peaks I, III and V are most important for clinical applications (Chiappa and Yiannikas 1983), it is necessary to keep phase linear while applying the filter. For example, use of a linear phase filter causes an input signal, which falls entirely in the passband, to be accurately copied to the output with a constant delay. In order to have an exactly linear phase, the filter should be a symmetric causal finite impulse response (FIR) system (Oppenheim and Schafer 1975). Spectral analysis of averaged ABEPs has shown that there is very little energy beyond 2 kHz (Elberling 1979; Kevanishvili and Aphonchenko 1979; Boston and Ainslie 1980; Doyle and Hyde 1981). However, there is no consensus about the
lower cut-off frequency, i.e., the high pass of the filter. Boston and Ainslie (1980) have suggested that 200 Hz may be reasonable if a zero phase shift FIR filter is applied, while Doyle and Hyde (1981) prefer a high-pass cut-off of 100 Hz. The power spectrum of the ABEP in the lower frequency portion (up to 1 kHz) is not flat and contains a few major notches, as can be seen in Doyle and Hyde (1981) and Fridman et al. (1982). In our study we examined the effect of specific band filtering, trying to isolate components from the total ABEP. We divided the spectrum of the ABEP into 3 major bands: (a) slow - - up to 200 Hz, (b) medium - - 200-400 Hz, and (c) fast - over 500 Hz (up to 3 kHz).
Methods Fourteen adults (8 males and 6 females), ranging in age from 19 to 31 years, with normal audiometric function and no neurological complaints, served as our subjects. Potentials were recorded from silver disc electrodes (9 mm diameter) in the conventional vertex-mastoid mode ipsi-
0013-4649/86/$03.50 © 1986 Elsevier Scientific Publishers Ireland, Ltd.
D. URBACH, H. PRATT
270
lateral to the stimulus. A grounding electrode was placed on the forearm. Electrode resistance was maintained below 5 kl2. Potentials were differentially amplified (Grass P511) (×500,000) at a bandpass of 10-3000 Hz ( - 3 dB points, 6 d B / o c tave slopes). The amplified potentials were averaged using 256 addresses and a dwell time of 71 /~sec per address. Stimuli were clicks generated by transducing 100 ~sec square electric pulses in TDH-39 earphones. Potentials following 75 dB nHL, alternating polarity clicks, presented monaurally at 10/sec, were averaged to produce each trace. In order to minimize residual noise contributions to the spectra, each trace was the average of 8000 sweeps. Alternating polarity clicks were used so as to average possible click polarity-related differences in the ABEP. The averaged data were stored on magnetic media for further analysis. Most of the subjects participated in two sessions, one for each ear, and the total number of records from the 14 subjects was thus 23. Power spectra of typical ABEPs. obtained by the periodogram method for power spectrum estimation (Oppenheim and Schafer 1975), are shown in Fig. 1. The average notches in the spectrum were at 240 Hz (with S.D. of 2.7 Hz) and 484 Hz (with S.D. of 7.6 Hz). The filters according to these findings should be up to 240 Hz, 240-483 and over 483 Hz. The shape of the power spectrum was asymmetrical, i.e., the lower frequencies near the notch were greater in power content than
DIGITAL
FILTERS
FREQUENCY
dB s~w
0 o Z
t t | i ! I I t I i
o I-
w ~
~,~.."
tO-
~.
20..'""
~O- "'""
404
ro
t
fast
• "- . . . . . . . . . . . . . . . . . .
2'0
;o
,~o
~o
s~o
.~..-.
,~ ~
~,~
FREQUENCY (Hz)
Fig. 2. Specifications of the 3 filters. The ripple in the passband is less than 0.5 dB.
on the higher frequency side of the notch. Consequently, we designed the filters as in Fig. 2. The filter coefficients were calculated using the ' F I R linear phase filter design programme' (McClellan et al. 1973). The filtering procedure was the direct form realisation of FIR filter (Oppenheim and Schafer 1975) using 64 filter coefficients. Each ABEP was thus filtered by the 3 filters. All peak latencies of the unfiltered ABEPs and the filtered ones were determined by the same person, to apply the same bias to all the results. Average and standard deviations of the peak latencies were calculated. AUDITORY BRAINSTEM DIGITAL
POWER SPECTRUM
RESPONSE
ESTIMATION
EVOKED POTENTIALS FILTERING v
v
LOG POWER SPECTRUM (dB)
OC - 2 0 0 ~t*
-SO
-tOO
FREQUENCY
(KHz)
Fig. l. Power spectrum estimation of ABEPs from 3 different subjects. The arrows point to the location of the notches.
Fig. 3. The effect of the 3 different filters on ABEP records from 3 of the subjects (left, middle and right columns). Amplitudes of the medium and fast filtered records are normalized to peak-to-peak amplitudes of 1.
FIR [)IGITAL F I L T E R I N G OF ABEP
271
TABLE I Average peak latencies (in msec) and their standard deviation (in brackets) over an ensemble of 23 different ABEP records and the 3 filters described in Fig. 2. Peak IV in the unfiltered ABEP (marked with *) includes an ensemble of only 12 ABEPs because it was impossible to define peak IV in the other 11 records. Filter type
1
II
I II
IV
V
Unfiltered 'Slow' 'Medium' 'Fast'
1.42 (0.12)
2.58 (0.09)
3.57 (0.14)
4.77 (0.19) *
1.28 (0.15) 1.48 (0.12)
2.57 (0.10)
3.42 (0.17) 3.60 (0.16)
4.60 (0.27)
5.45 5.38 5.37 5.51
Results Fig. 3 presents an example of the wave forms obtained with the different filters. It can be seen that the 'slow' filter (up to 200 Hz) easily defines peak V and only peak V while the 'medium' filter (200-400 Hz) results in peaks I, III, V and VI. It might be of interest that peaks II and IV do not appear below 400 Hz in any of the 23 ABEPs and thus they are classified, according to our filters, as the ' fast components.' The average peak latencies are presented in Table I. Peaks I and III of the unfiltered data differ from the medium component peaks by about 1 S.D. while peak V differs by only 0.5 S.D. The location of peak V seems to be better using the medium filter while the slow filter accurately defines peak V, but does not give its accurate latency. The fast filter gives longer average latencies than the unfiltered data (for peaks I, III and V) and it might be of interest that the unfiltered data peaks are between their values with the fast and the slow filters. In Table II, the interpeak latency differences are presented. There is no improvement in V - I latency determination using the medium filter but T A B L E II Average values for interpeak latency differences V-I, V-III, and l l l - I . The latencies are in msec, the standard deviation is in brackets and the ensemble includes 23 different ABEPs. Filter type
V-I
V-III
IlI-I
Unfiltered 'Medium' 'Fast'
4.02 (0.18) 4.09 (0.18) 4.03 (0.20)
1.88 (0.12) 1.95 (0.09) 1.91 (0.14)
2.15 (0.14) 2.15 (0.17) 2.11 (0.16)
(0.17) (0.25) (0.16) (0.18)
there is 25% reduction in the standard deviation of V-III, which comes at the expense of the increase in the III-I standard deviation.
Discussion The results of this study suggest the existence of 3 frequency bands in the ABEP wave form. Earlier reports described only 2 frequency bands corresponding to the slow and fast components of the ABEP. The cut-offs of these bands were: below 300 and above 400 Hz (Takagi et al. 1983); below and above 450 Hz (Fridman et al. 1982); or above 280 Hz (Moiler 1983). However, careful examination of the spectra provided in these and other reports reveals 3 frequency bands. Thus, for example, in one of these studies the averaged spectral characteristics of the ABEP show two notches, one at about 280 Hz and the other at 450 Hz (Fig. 2 in Fridman et al. 1982). In Fig. 7 of that report, phase variance of the ABEPs shows a minimum at about 300 Hz in addition to about 550 Hz, indicating increased synchronization in the generation of the ABEP and smaller noise contributions in these frequencies. While all studies, including ours, agree on the frequency content of the ABEP, earlier studies ignored the low frequency notch at about 240 Hz, concluding that there are only 2 frequency bands. The importance of all 3 frequency bands is demonstrated in their selectivity in component detection. Selective filters for each of these bands revealed enhancement or disappearance of components, depending on the band. The slow filter enhanced the 'pedestal' peaking at the latency of component V.
272
The medium filter enhanced components I, III and V, while II and IV vanished. The fast filter enabled identification of all 5 components. The smooth wave forms obtained by the slow and medium filters, coupled with the similar latencies of peak V in both, may be used for automatic peak detection and identification. T h e most striking finding was the absence of components II and IV in the slow and medium bands. One earlier study, on the effect of high-pass filter (Doyle and Hyde 1981), described the monotonic amplitude increase of peak I! between 150 and 500 Hz, and a similar increase of IV between 200 and 500 Hz. Another report (Moiler 1983) states that peaks II and IV were significantly attenuated when using a bandpass of approximately 400-900 Hz, but were clearly identified with a higher bandpass. In our study, the medium filter passband was at lower frequencies, resulting in the total abolition of II and IV. The different nature of components II and IV as compared to I, III and V has also been indicated by a study on the 3-channel Lissajous' trajectory of the ABEP. In that study (Martin et al. 1986), lowering stimulus intensity resulted in the disappearance of the second and fourth components and the persistence of the first, third and fifth. The results of this study show that the ABEP wave form is obtained by the superposition of at least 3 major waves having distinct frequency bands. The physiological significance of these bands can only be ascertained by the effects of physiological manipulations on the ABEP wave form.
R6sum~
Application aux potentiels bvoqu~s auditifs du tronc cbrbbral de filtres digitaux h rbponse impulsionnelle bornbe Le filtrage digital h r6ponse impulsionnelle born6e est une op6ration lin6aire sans distorsion et peut donc Etre utilis6 pour diff6rencier les composantes des BAEP. Puisque le spectre de puissance des BAEP montre qu'il existe plus de deux bandes de fr6quence, celles-ci peuvent 8tre dis-
D. URBACH, H. PRATT
tingu6es en utilisant des filtres digitaux aux caract~ristiques sp6ciales: jusqu'~ 200 Hz pour les composantes lentes, de 200 h 400 Hz pour les composantes moyennes et au-dessus de 500 Hz pour les composantes rapides. Les r6sultats obtenus en utilisant ces filtres montrent une nature diff6rente des composantes II et IV de celle des composantes I, lit et V. Alors que II et IV ne sont pas d&ect~es en-dessous de 500 Hz, et sont donc appel6es 'composantes rapides', I, III et V sont [es 'composantes moyennes' et apparaissent au-dessus de 200 Hz. Le filtre bas r6sulte en un 'pi6destal' dont le pic coincide avec le pic V. Les r6sultats de cette 6tude montrent que la forme de l'onde des BAEP est obtenue par superposition d'au moins 3 ondes principales ayant des bandes de fr6quence distinctes.
References Boston, J.R. Effect of digital filtering on the waveform and peak parameters of the auditory brainstem response. J. clin. Engng, 1983, 84: 79-84. Boston, J.R. and Ainslie, P.J. Effects of analog and digital filtering on brain stem auditory evoked potentials. Electroenceph, clin. Neurophysiol., 1980, 48: 361-364. Chiappa, K.H. and Yiannikas, C. Evoked Potentials in Clinical Medicine: Pattern-Shift Visual, Brainstem Auditory and Short-Latency Somatosensory: Technique, Correlations and Interpretations. Raven Press, New York, 1983. Doyle, D.J. and Hyde, M.L, Analog and digital filtering of auditory brainstem responses. Scan& Audiol., 1981, 10: 81-89. Elberling, C. Auditory electrophysiology: spectral analysis of cochlear and brain stem evoked potentials. Scand. Audiol., 1979, 8: 57-64. Fridman, J., John, E.R., Bergelson, M., Kaiser, J.B. and Baird, H.W. Application of digital filtering and automatic peak detection to brain stem auditory evoked potential. Electroenceph, clin. Neurophysiol., 1982, 53: 405-416. John, E.R., Baird, H., Fridman, J. and Bergelson, M. Normative values for brain stem auditory evoked potentials obtained by digital filtering and automatic peak detection. Electroenceph. clin. Neurophysiol., 1982, 54: 153-160. Kevanishvili, Z. and Aphonchenko, V. Frequency composition of brain-stem auditory evoked potentials. Scand. Audiol., 1979, 8: 51-55. Martin, W.H., Pratt, H. and Bleich, N. Three-channel Lissajous' trajectory of human auditory brain-stem evoked potentials. II. Effects of click intensity. Electroenceph. clin. Neurophysiol., 1986, 63: 54-61.
FIR DIGITAL FILTERING OF ABEP McClellan, J.H., Parks, T.W. and Rabiner, L.R. A computer program for designing optimum FIR linear phase digital filters. IEEE Trans. Audio Electroacoust., 1973, AU-21: 506-526. Moiler, A.R. A digital filter for brain stem evoked response. Amer. J. Otolaryng., 1980, 1: 372-377. Moiler, A.R. Improving brain stem auditory evoked potential
273 recordings by digital filtering. Ear Hear., 1983, 4:108 113. Oppenheim, A.V. and Schafer, R.W. Digital Signal Processing. Prentice-Hall, Englewood Cliffs, N J, 1975. Takagi, N., Kobayashi, K. and Suzuki, T. Slow component of auditory brainstem response, intensity and rate function. Audiol. Jap., 1983, 26: 716-721.
269
Technical Section APPLICATION O F FINITE I M P U L S E R E S P O N S E DIGITAL FILTERS T O AUDITORY BRAIN-STEM EVOKED P O T E N T I A L S D. URBACH and H. PRATT Microshev Ltd., Alon Shvut, and Evoked Potentials Laboratory, Israel Institute of Technology, Technion City, Haifa 32000 (Israel) (Accepted for publication: January 26, 1986)
Summary Finite impulse response digital filtering is a linear non-distorting operation and hence can be used to differentiate ABEP components. Since the power spectrum of the ABEP indicates that it contains more than two frequency bands, they can be distinguished by applying digital filters with special characteristics: up to 200 Hz (slow components), 200-400 Hz (medium components) and above 500 Hz (fast components). The results, using these filters, indicate a different nature of components II and IV as compared to I, III and V. While I1 and IV are not detected under 500 Hz and thus are called 'fast components,' I, III and V are the 'middle components' and appear above 200 Hz. The slow filter results in a 'pedestal' whose peak coincides with peak V. The results of this study show that the ABEP wave form is obtained by the superposition of at least 3 major waves having distinct frequency bands.
Keywords: finite impulse response - digital filter - A B E P
The increased use of digital filters in improving the recording of brain-stem auditory evoked potentials (ABEPs) (Boston and Ainslie 1980; Moiler 1980, 1983; Doyle and Hyde 1981; Fridman et al. 1982; John et al. 1982; Boston 1983; Takagi et al. 1983) has indicated their superiority over analog filters with respect to latency distortions. However, even when digital filters are used, the latencies of waves I - V depend on the type and the characteristics of the filter (Moiler 1983). Since latencies of peaks I, III and V are most important for clinical applications (Chiappa and Yiannikas 1983), it is necessary to keep phase linear while applying the filter. For example, use of a linear phase filter causes an input signal, which falls entirely in the passband, to be accurately copied to the output with a constant delay. In order to have an exactly linear phase, the filter should be a symmetric causal finite impulse response (FIR) system (Oppenheim and Schafer 1975). Spectral analysis of averaged ABEPs has shown that there is very little energy beyond 2 kHz (Elberling 1979; Kevanishvili and Aphonchenko 1979; Boston and Ainslie 1980; Doyle and Hyde 1981). However, there is no consensus about the
lower cut-off frequency, i.e., the high pass of the filter. Boston and Ainslie (1980) have suggested that 200 Hz may be reasonable if a zero phase shift FIR filter is applied, while Doyle and Hyde (1981) prefer a high-pass cut-off of 100 Hz. The power spectrum of the ABEP in the lower frequency portion (up to 1 kHz) is not flat and contains a few major notches, as can be seen in Doyle and Hyde (1981) and Fridman et al. (1982). In our study we examined the effect of specific band filtering, trying to isolate components from the total ABEP. We divided the spectrum of the ABEP into 3 major bands: (a) slow - - up to 200 Hz, (b) medium - - 200-400 Hz, and (c) fast - over 500 Hz (up to 3 kHz).
Methods Fourteen adults (8 males and 6 females), ranging in age from 19 to 31 years, with normal audiometric function and no neurological complaints, served as our subjects. Potentials were recorded from silver disc electrodes (9 mm diameter) in the conventional vertex-mastoid mode ipsi-
0013-4649/86/$03.50 © 1986 Elsevier Scientific Publishers Ireland, Ltd.
D. URBACH, H. PRATT
270
lateral to the stimulus. A grounding electrode was placed on the forearm. Electrode resistance was maintained below 5 kl2. Potentials were differentially amplified (Grass P511) (×500,000) at a bandpass of 10-3000 Hz ( - 3 dB points, 6 d B / o c tave slopes). The amplified potentials were averaged using 256 addresses and a dwell time of 71 /~sec per address. Stimuli were clicks generated by transducing 100 ~sec square electric pulses in TDH-39 earphones. Potentials following 75 dB nHL, alternating polarity clicks, presented monaurally at 10/sec, were averaged to produce each trace. In order to minimize residual noise contributions to the spectra, each trace was the average of 8000 sweeps. Alternating polarity clicks were used so as to average possible click polarity-related differences in the ABEP. The averaged data were stored on magnetic media for further analysis. Most of the subjects participated in two sessions, one for each ear, and the total number of records from the 14 subjects was thus 23. Power spectra of typical ABEPs. obtained by the periodogram method for power spectrum estimation (Oppenheim and Schafer 1975), are shown in Fig. 1. The average notches in the spectrum were at 240 Hz (with S.D. of 2.7 Hz) and 484 Hz (with S.D. of 7.6 Hz). The filters according to these findings should be up to 240 Hz, 240-483 and over 483 Hz. The shape of the power spectrum was asymmetrical, i.e., the lower frequencies near the notch were greater in power content than
DIGITAL
FILTERS
FREQUENCY
dB s~w
0 o Z
t t | i ! I I t I i
o I-
w ~
~,~.."
tO-
~.
20..'""
~O- "'""
404
ro
t
fast
• "- . . . . . . . . . . . . . . . . . .
2'0
;o
,~o
~o
s~o
.~..-.
,~ ~
~,~
FREQUENCY (Hz)
Fig. 2. Specifications of the 3 filters. The ripple in the passband is less than 0.5 dB.
on the higher frequency side of the notch. Consequently, we designed the filters as in Fig. 2. The filter coefficients were calculated using the ' F I R linear phase filter design programme' (McClellan et al. 1973). The filtering procedure was the direct form realisation of FIR filter (Oppenheim and Schafer 1975) using 64 filter coefficients. Each ABEP was thus filtered by the 3 filters. All peak latencies of the unfiltered ABEPs and the filtered ones were determined by the same person, to apply the same bias to all the results. Average and standard deviations of the peak latencies were calculated. AUDITORY BRAINSTEM DIGITAL
POWER SPECTRUM
RESPONSE
ESTIMATION
EVOKED POTENTIALS FILTERING v
v
LOG POWER SPECTRUM (dB)
OC - 2 0 0 ~t*
-SO
-tOO
FREQUENCY
(KHz)
Fig. l. Power spectrum estimation of ABEPs from 3 different subjects. The arrows point to the location of the notches.
Fig. 3. The effect of the 3 different filters on ABEP records from 3 of the subjects (left, middle and right columns). Amplitudes of the medium and fast filtered records are normalized to peak-to-peak amplitudes of 1.
FIR [)IGITAL F I L T E R I N G OF ABEP
271
TABLE I Average peak latencies (in msec) and their standard deviation (in brackets) over an ensemble of 23 different ABEP records and the 3 filters described in Fig. 2. Peak IV in the unfiltered ABEP (marked with *) includes an ensemble of only 12 ABEPs because it was impossible to define peak IV in the other 11 records. Filter type
1
II
I II
IV
V
Unfiltered 'Slow' 'Medium' 'Fast'
1.42 (0.12)
2.58 (0.09)
3.57 (0.14)
4.77 (0.19) *
1.28 (0.15) 1.48 (0.12)
2.57 (0.10)
3.42 (0.17) 3.60 (0.16)
4.60 (0.27)
5.45 5.38 5.37 5.51
Results Fig. 3 presents an example of the wave forms obtained with the different filters. It can be seen that the 'slow' filter (up to 200 Hz) easily defines peak V and only peak V while the 'medium' filter (200-400 Hz) results in peaks I, III, V and VI. It might be of interest that peaks II and IV do not appear below 400 Hz in any of the 23 ABEPs and thus they are classified, according to our filters, as the ' fast components.' The average peak latencies are presented in Table I. Peaks I and III of the unfiltered data differ from the medium component peaks by about 1 S.D. while peak V differs by only 0.5 S.D. The location of peak V seems to be better using the medium filter while the slow filter accurately defines peak V, but does not give its accurate latency. The fast filter gives longer average latencies than the unfiltered data (for peaks I, III and V) and it might be of interest that the unfiltered data peaks are between their values with the fast and the slow filters. In Table II, the interpeak latency differences are presented. There is no improvement in V - I latency determination using the medium filter but T A B L E II Average values for interpeak latency differences V-I, V-III, and l l l - I . The latencies are in msec, the standard deviation is in brackets and the ensemble includes 23 different ABEPs. Filter type
V-I
V-III
IlI-I
Unfiltered 'Medium' 'Fast'
4.02 (0.18) 4.09 (0.18) 4.03 (0.20)
1.88 (0.12) 1.95 (0.09) 1.91 (0.14)
2.15 (0.14) 2.15 (0.17) 2.11 (0.16)
(0.17) (0.25) (0.16) (0.18)
there is 25% reduction in the standard deviation of V-III, which comes at the expense of the increase in the III-I standard deviation.
Discussion The results of this study suggest the existence of 3 frequency bands in the ABEP wave form. Earlier reports described only 2 frequency bands corresponding to the slow and fast components of the ABEP. The cut-offs of these bands were: below 300 and above 400 Hz (Takagi et al. 1983); below and above 450 Hz (Fridman et al. 1982); or above 280 Hz (Moiler 1983). However, careful examination of the spectra provided in these and other reports reveals 3 frequency bands. Thus, for example, in one of these studies the averaged spectral characteristics of the ABEP show two notches, one at about 280 Hz and the other at 450 Hz (Fig. 2 in Fridman et al. 1982). In Fig. 7 of that report, phase variance of the ABEPs shows a minimum at about 300 Hz in addition to about 550 Hz, indicating increased synchronization in the generation of the ABEP and smaller noise contributions in these frequencies. While all studies, including ours, agree on the frequency content of the ABEP, earlier studies ignored the low frequency notch at about 240 Hz, concluding that there are only 2 frequency bands. The importance of all 3 frequency bands is demonstrated in their selectivity in component detection. Selective filters for each of these bands revealed enhancement or disappearance of components, depending on the band. The slow filter enhanced the 'pedestal' peaking at the latency of component V.
272
The medium filter enhanced components I, III and V, while II and IV vanished. The fast filter enabled identification of all 5 components. The smooth wave forms obtained by the slow and medium filters, coupled with the similar latencies of peak V in both, may be used for automatic peak detection and identification. T h e most striking finding was the absence of components II and IV in the slow and medium bands. One earlier study, on the effect of high-pass filter (Doyle and Hyde 1981), described the monotonic amplitude increase of peak I! between 150 and 500 Hz, and a similar increase of IV between 200 and 500 Hz. Another report (Moiler 1983) states that peaks II and IV were significantly attenuated when using a bandpass of approximately 400-900 Hz, but were clearly identified with a higher bandpass. In our study, the medium filter passband was at lower frequencies, resulting in the total abolition of II and IV. The different nature of components II and IV as compared to I, III and V has also been indicated by a study on the 3-channel Lissajous' trajectory of the ABEP. In that study (Martin et al. 1986), lowering stimulus intensity resulted in the disappearance of the second and fourth components and the persistence of the first, third and fifth. The results of this study show that the ABEP wave form is obtained by the superposition of at least 3 major waves having distinct frequency bands. The physiological significance of these bands can only be ascertained by the effects of physiological manipulations on the ABEP wave form.
R6sum~
Application aux potentiels bvoqu~s auditifs du tronc cbrbbral de filtres digitaux h rbponse impulsionnelle bornbe Le filtrage digital h r6ponse impulsionnelle born6e est une op6ration lin6aire sans distorsion et peut donc Etre utilis6 pour diff6rencier les composantes des BAEP. Puisque le spectre de puissance des BAEP montre qu'il existe plus de deux bandes de fr6quence, celles-ci peuvent 8tre dis-
D. URBACH, H. PRATT
tingu6es en utilisant des filtres digitaux aux caract~ristiques sp6ciales: jusqu'~ 200 Hz pour les composantes lentes, de 200 h 400 Hz pour les composantes moyennes et au-dessus de 500 Hz pour les composantes rapides. Les r6sultats obtenus en utilisant ces filtres montrent une nature diff6rente des composantes II et IV de celle des composantes I, lit et V. Alors que II et IV ne sont pas d&ect~es en-dessous de 500 Hz, et sont donc appel6es 'composantes rapides', I, III et V sont [es 'composantes moyennes' et apparaissent au-dessus de 200 Hz. Le filtre bas r6sulte en un 'pi6destal' dont le pic coincide avec le pic V. Les r6sultats de cette 6tude montrent que la forme de l'onde des BAEP est obtenue par superposition d'au moins 3 ondes principales ayant des bandes de fr6quence distinctes.
References Boston, J.R. Effect of digital filtering on the waveform and peak parameters of the auditory brainstem response. J. clin. Engng, 1983, 84: 79-84. Boston, J.R. and Ainslie, P.J. Effects of analog and digital filtering on brain stem auditory evoked potentials. Electroenceph, clin. Neurophysiol., 1980, 48: 361-364. Chiappa, K.H. and Yiannikas, C. Evoked Potentials in Clinical Medicine: Pattern-Shift Visual, Brainstem Auditory and Short-Latency Somatosensory: Technique, Correlations and Interpretations. Raven Press, New York, 1983. Doyle, D.J. and Hyde, M.L, Analog and digital filtering of auditory brainstem responses. Scan& Audiol., 1981, 10: 81-89. Elberling, C. Auditory electrophysiology: spectral analysis of cochlear and brain stem evoked potentials. Scand. Audiol., 1979, 8: 57-64. Fridman, J., John, E.R., Bergelson, M., Kaiser, J.B. and Baird, H.W. Application of digital filtering and automatic peak detection to brain stem auditory evoked potential. Electroenceph, clin. Neurophysiol., 1982, 53: 405-416. John, E.R., Baird, H., Fridman, J. and Bergelson, M. Normative values for brain stem auditory evoked potentials obtained by digital filtering and automatic peak detection. Electroenceph. clin. Neurophysiol., 1982, 54: 153-160. Kevanishvili, Z. and Aphonchenko, V. Frequency composition of brain-stem auditory evoked potentials. Scand. Audiol., 1979, 8: 51-55. Martin, W.H., Pratt, H. and Bleich, N. Three-channel Lissajous' trajectory of human auditory brain-stem evoked potentials. II. Effects of click intensity. Electroenceph. clin. Neurophysiol., 1986, 63: 54-61.
FIR DIGITAL FILTERING OF ABEP McClellan, J.H., Parks, T.W. and Rabiner, L.R. A computer program for designing optimum FIR linear phase digital filters. IEEE Trans. Audio Electroacoust., 1973, AU-21: 506-526. Moiler, A.R. A digital filter for brain stem evoked response. Amer. J. Otolaryng., 1980, 1: 372-377. Moiler, A.R. Improving brain stem auditory evoked potential
273 recordings by digital filtering. Ear Hear., 1983, 4:108 113. Oppenheim, A.V. and Schafer, R.W. Digital Signal Processing. Prentice-Hall, Englewood Cliffs, N J, 1975. Takagi, N., Kobayashi, K. and Suzuki, T. Slow component of auditory brainstem response, intensity and rate function. Audiol. Jap., 1983, 26: 716-721.