Quantitative assessment of human cochlear function by evoked otoacoustic emissions

99

Hearing Research, 52 (1991) 99-112 0 1991 Elsevier Science Publishers B.V. 0378-5955/91/$03.50

HEARES

01515

Quantitative

assessment of human cochlear function by evoked otoacoustic emissions

P. Avan ‘, P. Bonfils 2,3, D. Loth I, Ph. Narcy 3 and J. Trotoux 2 ’ Laboratory of Audition, Central Service of Biophysics and Nuclear Medicine, Fact&y of Medicine Lariboisi&eSt-Lmis, UniuersifyParis VII; ’ ENT Service, Boucicaut Hospital, Fact&y of Medicine Necker-En~ants-Molades, University Paris V; ’ ENT Service, R. DebrL; Hospital, Faculty of Medicine Bichat, University Paris VII, France (Received

16 June 1990; accepted 22 September 1990)

The amplitudes of evoked otoacoustic emissions (EOE) and their detection threshold were measured in 44 normal young adults and 118 patients with two categories of cochlear dysfunction, acoustic trauma and presbycusis. A different method was used for each category: detection of click EOE or of stimulus frequency emissions. A partial correlation and multivariate analysis was performed for both groups of results to investigate the relations between EOE threshold and pure tone audiometric thresholds (250 to 8000 Hz). Only one significant correlation was found, linearly relating EOE threshold and hearing threshold at 2 kHz (P < O.OOl), independently of the origin of co&ear dysfunction. It suggests that EOE threshold is not frequency-specific since the frequency of EOE at threshold was nearly always close to 1 kHz. A simple model is proposed, based on the assumption that EOE amplitudes and threshold are proportional to the total number of residual active sites in the organ of Corti. i.e. to the total length of active basilar membrane. It is shown that this model accounts for the results disclosed by the statistical analysis and fits the experimental data. It can be used for quantitatively predicting the residual cochlear activity of a patient. However, the EOE threshold is only sensitive to already important cochlear alterations and this parameter does not seem to allow a follow-up of early stages of cochlear dysfunction. Ciick evoked Evaluation

otoacoustic

emissions;

Stimulus

frequency

emissions;

Evoked otoacoustic emissions, which Kemp was the first to propose as a clinical test in 1978, are now becoming one of the routine audiological methods to detect any cochlear pathology associated with an alteration of active mechanisms (e.g. Probst et al., 1987, Bonfils et al., 1988a,b; Kemp et al., 1990). In neonates and children, this method could provide a rapid, objective, reliable and noninvasive test for screening peripheral auditory dysfunction. It seems now clear that the presence of EOE (as well as distortion products or spontaneous otoacoustic emissions) is associated with

Correspondence to: P. Avan, Laboratory of Biophysics, Faculty of Medicine Lariboisiere St-Louis, 10 Avenue de Verdun, 75010 Paris, France.

Active

mechanisms;

EOE threshold;

Pure tone audiometry;

active mechanisms arising from the normal function of outer hair cells in the cochlea (Kim, 1986). Both the selectivity and the sensitivity of the healthy inner ear depend on these vulnerable active mechanisms. Any cochlear pathology with even minimal auditory threshold shifts should thus give rise to early changes in EOE. Until now, clinical studies have indeed shown the sensitivity of EOE tests in screening: EOE are present in nearly every healthy inner ear but tend to disappear as soon as an endocochlear hearing loss of 30 dB or more is present (e.g. Kemp et al., 1986, 1990; Bonfils et al., 1988b; Johnsen et al,, 1988). However, quantitative relationships between the hearing loss and EOE parameters (threshold, amplitude, frequency spectrum) remain to be found (if any), and EOE tests cannot replace audiometric data. From a more fundamental point of view, the properties of evoked acoustic emissions originate

100

in as yet unclear mechanisms: they can be described by two main categories of models, local impedance discontinuities giving rise to isolated place-fixed EOE (Wit and Ritsma, 1980; Ruggero et al., 1983) and more extensively distributed activity with some kind of interference and unbalance between contributions of various regions of the cochlea (Kemp, 1978; Wilson, 1980b; Sutton and Wilson, 1983). Moreover, several types of click EOE can be identified in most normal subjects, corresponding to sharp isolated peaks or to a broad continuum (Wit and Ritsma, 1980; Kemp and Chum, 1980; Wilson 1980b; Manley, 1983, Zwicker and Schloth, 1984). The situation has not much progressed during these last years: click EOE can be readily recorded in human ears since available devices are now reliable (Kemp et al., 1990). Unfortunately, it is not the case in animals (Avan et al., 1990) so that the usual techniques of experimental pathology cannot help to understand what happens to EOE when cochlear mechanics is altered. On the other hand, distortion-product emissions are easy to detect in animals and even in man, but are useless for understanding some EOE properties which are quite different (in particular their unique frequency pattern in a given ear). The aim of this study was to determine the parameters influencing the amplitudes and detection threshold of EOE around 1 kHz by looking for statistical correlations between EOE threshold and various audiometric frequencies (from 250 to 8000 Hz). A multivariable analysis was performed on a population of 44 ears from normal subjects, 78 ears with presbycusis and 38 ears with acoustic trauma. Materials and Methods Two completely separate sets of experiments were carried out, using different devices and methods, with two categories of subjects corresponding to different endocochlear pathologies. In the following sections, they will be described separately and referred to as experiment El and E2. Experiment

El

Recording of evoked otoacoustic emissions Since the acoustic probe, stimulus generation, signal processing and EOE identification have

been described in detail previously (Bonfils et al., 1988a,b), only a brief account will be presented here. EOE recordings were obtained using a small acoustic probe with built-in miniature microphone and loudspeaker. The probe was sealed into the external ear canal using an impedance probe protector. All recordings were done in a doublewalled, sound-treated chamber. Acoustic stimuli were rarefaction clicks generated by 0.1 ms rectangular pulses delivered to the earphone at a repetition rate of 19/s. The response was amplified (x5000), high-pass filtered (250 Hz, 16 dB per octave), sampled within a time window of 30 ms starting 5 ms after the onset of the stimulus, and averaged in the time domain over 1000 repetitions. The averaged signal was fed into a frequency analyser (HP3561 A) for Fast Fourier Transform (FFT) analysis. EOE were recorded for decreasing stimulus intensities ranging from 40 dB HL to the detection threshold in 5 dB steps.

Data analysis Identification of EOE was based upon several criteria (e.g. Probst et al., 1987; Bonfils et al., 1988a,b; Kemp et al., 1990): (1) nonlinear saturating responses for higher stimulus levels, (2) discrete frequency-specific maxima in FFT analysis, and (3) reproductibility of EOE properties. EOE threshold was defined using a visual criterion in both time and frequency domains (Fig. 2a). This frequency domain was restricted to the range 0.5-2 kHz. Anyway, the experimental device did not allow to accurately detect the threshold of EOE above 3-4 kHz due to the beginning of the available time tindow (2 5 ms after click onset). In the absence of EOE, an arbitrary value of 40 dB HL was attributed to the threshold.

Subjects 44 right ears from normal young adults without any history of otologic disease or exposure to ototoxic agents or noise, and with normal otologic findings (e.g., normal otoscopy, normal middle ear pressure and normal stapedius reflex thresholds) were tested. Every subject had audiometric pure tone thresholds of 10 dB HL or less at octave frequencies between 0.25 and 8 kHz (Fig. la,A). Ages ranged from 17 to 34 (mean age 24).

101 0.25

A

0.5

I

4

2

6

6

kHz

0.25

0.5

1

1.5

2

3

4

6

8

ktiz

’

+

1 t

T .

75th oercenl~le

Fig. 1. (a) experiment El, presbycusis: Mean pure tone audiograms (with SD) for normal subjects (A) and for patients with (B) and without (C) detectable EOE. (b) experiment E2, acoustic trauma: Mean pure tone audiograms (the 75th percentiIes are also indicated (filled triangles) in addition to standard deviations, ~lustra~g the large ~te~di~du~ dispersion of audiometric alterations usually found for such kinds of hearing losses). Compared to El, the audiometric threshoids at 1.5 and 3 kHz were also measured to ahow a more accurate description of some audiograms exhibiting sharp dips.

80 right ears from patients exhibiting a pure sensorineural hearing loss (based upon audiogram (Fig. la, B and C), acoustic reflex and auditory brainstem responses) were studied. Ages ranged from 54 to 78 (mean age, 62). Every ear exhibited dominant high-tone hearing losses due to presbycusis. Informed consent for the study was obtained from the subjects.

(subject

Experiment E2 Recording of evoked ~t~ac~~tic emissions The acoustic probe was similar to the one used in El. It was also installed in the same way. Stimulus Frequency Emissions (SFE) were evoked by a continuous pure tone delivered by the miniature loudspeaker of the probe at equalized levels

tsubjecl

J.P. right)

limit due to noise level (-23

-ak -6

0 EOE Threshold

6

12

18

click

24

level(dB

30

36

tiCI

42

48

1 -10

-5

0

5

J.P. right)

dB SPL)

IO

15

20

25

stimulus level (dB SPL) at 1130Hz

Fig. 2. (a) experiment El, presbycusis: input/output characteristic functions for click-EOE at two different frequencies. The click level, expressed in dB HL, was decreased by 6 dB steps. A FFT computation from the averaged signal in the analysed window (S-30 ms) gave the EOE level (in dB SPL) at a given frequency. EOE threshold was defined as the lowest stimulus Level eliciting a detectable EOE above the noise floor (approximateIy constant vahte of -25 dB/50 Hz band~d~). EOE thresholds were respectively 0 and 12 dB HL at 1130 and 3900 Hz, and this study was focused on the first EOE i.e. around 1OOUHz. (b) experiment E2, acoustic trauma: Stimulus frequency emissions close to 1 kHz were identified as cusps or loops in the Nyquist impedance diagram of the tested ear. The ratio between SFE peak-to-trough amplitude and the stimulus level was plotted for decreasing stimulus levels and tended to a constant value (corresponding to a linear behaviour of the SFE as a function of stimulus level). Remark: in this particular case, the two diagrams a and b were built for the same ear, at the same frequency (1130 Hz).

102

of 40 to 0 dB SPL presented in decreasing 10 dB steps (for more technical details see Avan et al., 1990). The pure tone was slowly swept in frequency from 700 to 2000 Hz (sweep velocity: 20 Hz/s). The microphone response was amplified ( X 1000) high pass filtered (500 Hz, 24 dB /octave) and analysed in a lock-in detector (EG and G 5206, 300 ms integration time). Its amplitude and phase were plotted as Nyquist vector diagrams for each acoustic level.

cochlear origin. It was interesting to note that most subjects (over 80%) mainly complained of high frequency continuous tinnitus.

Data analysis

Results

SFE were identified as peaks and troughs on the ph~/~p~tude plots, characterized by their frequency specificity and saturating nonlinearity appearing for levels above 20 dB. The amplitude of the SFE which was closest to 1 kHz was measured: in every case, its frequency was between 920 Hz and 1130 Hz. The ratio SFE-amplitude to stimulus level was then plotted as a function of the stimulus level. When this level decreased ( < 20 dB), the ratio tended towards a constant value, characteristic of the behaviour of the chosen SFE in the linear part of its input/output curve (Fig. 2b). To allow comparisons with El, this value was converted into an equivalent click-EOE threshold by multiplying it by a constant coefficient, characteristic of the probe sensitivity and of the mean level of the noise floor obtained in these experiments (around -30 dB SPL/SO Hz bandwidth centered at 1 kHz). A similar measurement was performed for the SFE with a frequency close to 750 Hz (i.e. between 720 and 780 Hz). Subjects

Patients suffering from acoustic trauma volunteered for taking part in the study. The trauma arose from weapon noise in most cases, and could be recent (4 to 7 days) or older (6 months to 10 years). In some cases where it was bilateral, only the results of one ear were taken into account to avoid any statistical bias on the variance of data (see for instance Grenner et al., 1990). 38 ears were kept for this study (audiograms: Fig. lb). Apart from the acoustic trauma, the patients had no history of other otologic disease and their hearing loss (determined from pure tone, BCkesy and vocal audiograms, acoustic reflex studies and auditory brainstem responses) was purely of

Statistical analysis of the data This analysis was performed

cal package sis, BMDP multivariate gression on

Experiment

on BMDP statisti(Dixon et al., 1985): correlation analy8D software; partial correlation and regression, BMDP 6R; stepwise reall possible subsets, BMDP 9R.

EI

Incidence of evoked acoustic emissions

Click stimuli elicited EOE from all normally hearing ears (N = 44 ears). In the group of patients exhibiting a pure sensorineural hearing loss, EOE were present in 47 ears and absent in 33 ears. Fig. la shows the mean audiometric thresholds for octave frequencies in these three groups (i.e. normally hearing ears, pathological ears with EOE, and without EOE). The mean EOE thresholds in these groups were respectively 5 dB HL, 16.7 dB HL and 40 dB HL (this last value being set arbitrarily). The frequency value of the threshold EOE was an important parameter. Its dist~bution histogram is represented on Fig. 3. In 85% of cases, the EOE threshold was measured at a frequency in the range 780-1500 Hz, with 15% only between 1500

40

1

0.7

I

0.9

I

.a

frequency

1.2

1.4

15

of threshold

1.7

I8

2.0

22

EOE fkHzl

Fig. 3. Distribution histogram of the frequency of the EOE at threshold, for experiment El. (subjects A and B, with detectable EOE: N = 91). Only 15% of threshold EOE had a frequency ~1.5 kHz.

103

and each of the pure-tone audiometric thresholds mean a direct relationship or is it only the reflection of the correlation between pure-tone audiometric thresholds?

and 2000 Hz and no case below 780 or above 2000 Hz. 44% of subjects had a threshold EOE in the central range 940-1150 Hz. General correlations between the EUE threshoId and pure-tone a~iametric threshold In order to determine the relationship between the EOE threshold and pure tone audiometry, eight variables were analysed: the pure tone audiometric thresholds (0.25,0.5, 1, 2, 4, 6, 8 kHz) and the EOE detection threshold. Table Ia indicates the correlation matrix between all the variables. A statisti~~y significant relation&p was present between the EOE threshold and ah puretone audiometric thresholds (P < 0.001). However, a significant correlation was also found between all couples of pure-tone audiometric thresholds. It was thus necessary to perform a partial correlation analysis (e.g. Montgomery and Peck, 1982) in order to meet the following question: does the correlation between EOE threshold

Partial correlation analysis of the data ~stepw~se regression) Firstly, the frequency 2 kHz was specified as an independent variable and a partial correlation was computed between the EOE threshold and pure tone audiometric thresholds for the remaining frequencies 0.25, 0.5, 1, 4, 6, 8 kHz. In this case, no statistically significant relationship was present between the EOE threshold and these pure-tone audiometric frequencies. The frequency 0.25 kHz was then specified as an independant variable and a partial correlation was again computed between EOE threshold and pure tone audiometric thresholds for the remaining frequencies 0.5, 1, 2,4, 6, 8 kHz. In this case, a statistically sig~fic~t relationship remained be-

TABLE I CORRELATION MATRICES (R-SQUARED) FOR AUDIOMETRIC THRESHOLDS (FREQUENCIES RANGING FROM 0.25-8 kHz IN OCTAVE OR HALF-OCTAVE INTERVALS) AND 1 kHz EOE THRESHOLDS MEASURED IN EXPERIMENTS El (A) AND E2 (B). A) EOE-T 0.25 0.5 1 2 4 6 8 B) EOE-T 0.25 0.5 1 1.5 2 3 4 6 8

EOE-T

0.25

0.5

1

2

4

6

8

1 0.59 0.69 0.70 0.75 0.65 0.67 0.63

1 0.90 0.84 0.72 0.61 0.61 0.57

1 0.90 0.78 0.65 0.67 0.66

1 0.87 0.71 0.71 0.67

1 0.83 0.81 0.75

1 0.96 0.87

1 0.96

1

EOE-T

0.25

0.5

1

1.5

2

3

4

6

8

1 0.33 * 0.61 0.71 0.87 0.87 0.82 0.72 0.72 0.64

1 0.82 0.64 0.45 0.45 0.35 0.29 0.40 0.40

1 0.87 0.74 0.73 0.67 0.53 0.60 0.64

1 0.85 0.77 0.67 0.58 0.57 0.62

1 0.93 0.79 0.67 0.65 0.69

1 0.88 0.74 0.70 0.68

1 0.87 0.85 0.78

1 0.92 0.88

1 0.93

1

* * + *

* indicatesa non significantcorrelation(P > 0.05).

104

EOE-T-

0 59

x 2k-T

40

+ 5 5

0

35 30 2s 20 15 10 5 0 -5

2 kHz pure tone threshold Cd51

Zktiz care tone LhreshoidfdB)

Fig. 4. multiple regression models obtained (a) for El (presbycusis) and (b) for E2 (acoustic trauma) relating the I kHz EOE threshold ieve and the pure tone audiometric thresholds: in both cases, only the threshold at 2 kHz appears in the regression, with a highly significant T-statistics. The regression slopes are nearly identical for El and E2. The points with an EOE threshold at 40 dB correspond to ears without any detectable EOE, for which this value was arbitrarily attributed.

tween EOE threshold and these pure-tone audiometric frequencies. Similar results were found with the frequencies 0.5, 1, 4, 6, 8 kHz. Stepwise regression thus indicated that only one step could be entered in a linear regression model using the variable 2 kHz, with the following regression coefficients:

the other variables to the variance was negligible, since the maximum value of R-squared was negligeably higher (0.63 instead of 0.60) when taking into account all the pure-tone audiometric frequencies (table II). The analysis of residues showed that a linear regression model was satisfactory.

EOE_Threshold

Experiment

= 0.59 X (2 kHz_Threshold) + 5.5 (dB)

Incidence of evoked acoustic ~rn~ss~on~ All the subjects had a typical pure-tone audiogram with a notch at 4 or 6 kHz. Its deepness varied from 30 to 90 dB and it extended towards

(1)

with a R-squared of 0.60 and a T-statistics of 6.62 (highly significant) (Fig. 4a). The contribution of

TABLE

E2

II

PROGRESSION

OF STEPWISE

Step number in regression R2 (at this step) (data from El) 250 Hz 500 Hz 1000 Hz 2000 Hz 4OOOH.Z 6OOOHz 8000 Hz The regression the one giving next steps led account in the

REGRESSION 1 0.605

*

FOR El DATA 2 0.620

3 0.630

4 0.635

5 0.637

6 0.637

7 0.637

*

* *

* *

* *

* *

*

*

*

*

*

* *

* * * *

* * * * * * *

method used in this work computed successive linear regressions, taking into account one more variable at each step the best enhancement of coefficient R2-. after the first step (2 kHz), R2 was already close to its maximum value. the to non significant improvements of R2 (*: means that from this step, the corresponding variable was taken into regression).

105

higher and lower frequencies, down to 2 and sometimes 1 kHz depending on the importance of the trauma. Fig. lb shows the mean audiometric thresholds for the following pure-tone frequencies: 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, 8 kHz. SFE were found in 33 ears and absent in 5 ears. When they were present, it was always possible to identify a SFE close to 1 kHz (between 920 and 1130 Hz), and a SFE close to 750 Hz ( f 30 Hz) for the determination of their thresholds (Fig. 2b). General correlations between I kHz-EOE and pure tone audiometric threshok&

threshold

Similar variables were analysed here compared to El, except that ten variables were taken into account instead of eight, the audiometric thresholds at 1.5 and 3 kHz having been measured in addition to others. Table Ib indicates the correlation matrix between all the variables. As in El, a statistically significant relationship was present between the EOE threshold and nearly all puretone audiometric thresholds (P K 0.001) except 0.25 kHz, and a highly significant correlation (P < 0.001) was present between all couples of puretone audiometric thresholds except three of them: (0.25, 4) (0.25, 6) and (0.25, 8) kHz. A partial correlation analysis with research of stepwise regression was thus also necessary. Nearly identical results were obtained for the EOE threshold at 750 Hz. Partial correlation regression)

analysis

of the data (stepwise

It was performed with the same method as for El and gave the following linear regression relationships: 1 kHz EOE_Threshold = 0.50 x (2kHz_Threshold)

+ 6.7 (dB)

(2)

(with a R-squared of 0.75, highly significant) (Fig. 4b). The contribution of the other variables to the variance, evaluated by forcing them in the regression model, one after the other, was negligible and non significant. In parallel: 750 Hz EOE_ Threshold = 0.41 x (l.SkHz_Threshold)

+ 9.5 (dB)

(with a R-squared of 0.65, highly significant)

(3)

Remark: when EOE were absent, an arbitrary value of 40 dB (the maximal measurable threshold value being about 30-35 dB) had to be set for EOE threshold. It was checked that this had no influence on the results for El and E2: Identical conclusions and regression models were obtained when setting this threshold at 50 dB for instance. All these results indicate for both experiments El and E2 that the correlations appearing between the threshold of 1 kHz EOE and all the pure-tone audiometric thresholds except at 2 kHz, exist only because these thresholds are highly correlated between each other and especially with the 2 kHz one. This last parameter actually appears to be the only relevant one to determine the 1 kHz EOE threshold. Moreover, for E2 in which pure tone audiograms were more accurately sampled, a similar high correlation is found between the 750 Hz EOE threshold and the 1.5 kHz audiometric threshold. Discussion

The notion of threshold for EOE has been used in several papers about the clinical interest of EOE measurements (Probst et al., 1987; Bonfils et al., 1988a,b; Collet et al., 1989): this parameter is defined as the lowest stimulus level (in dB HL) eliciting a detectable EOE when using a click stimulus such as in experiment El. This detection is made easier by comparing the record at threshold and its frequency spectrum with any suprathreshold one. For a given ear, many EOE thresholds can be measured, corresponding to different peaks in the EOE spectrum. The threshold of an EOE is specific of its behaviour in the linear part of its input/output characteristics, for low level stimuli (for every patient, the existence of a linear part in this characteristics was checked-see Fig. 2-at the stimulus levels used around EOE threshold). This parameter obviously also depends on the physical characteristics of the probe and on the level of noise background. In both experiments performed in this work, the same probes were used for all the subjects and only the records obtained in the presence of ~~rnurn noise were kept. This noise level was constant (within 5 dB) consequently it did not influence the definition of EOE threshold. Its measurement was done with a precision of *5 dB.

106

In clinics and for the sake of time-saving, only the best threshold is generally determined (i.e corresponding to the last detectable EOE when the stimulus level is decreased step by step) (Probst et al,, 1987, Bonfils et al., 1988a,b). In experiment El, the frequency of this EOE could not be kept constant and varied between 780 et 2000 Hz. However, Fig. 3 shows that the EOE threshold was explored in a narrow frequency range (less than half an octave wide) for most subjects. In experiment E2, the input/output functions of two given SFE, as close as possible to 1 kHz (i.e. 920-1130 Hz) and 750 Hz (i.e. 720-780 Hz), were plotted. It is known that for normal subjects the distribution of SFE peaks is often regular at this frequency and exhibits a periodicity of about 0.4 Bark or 80 Hz around 1 kHz (Wilson, 1980b; Zwicker and Schloth, 1984; Zwicker, 1986), and it was always possible to measure a SFE in the chosen frequency ranges in subjects having residual EOE. Even if the meaning of the so-called ‘EOE threshold’ was thus slightly different at 1 kHz in El and E2, the same range of frequency was actually tested for EOE in most cases It would have been interesting to study the behaviour of EOE thresholds in other frequency ranges. Unfortunately, for higher frequencies (2 2 kHz), too many subjects had no detectable EOE, probably in relation with their high frequency hearing loss. For lower frequencies (e.g. 500 Hz), no reliable click EOE or SFE could be recorded: in all cases where some peaks at this frequency were detectable in the spectrum of the response, their input/output curve did not exhibit a clear saturating behaviour at higher intensities hence their identification as EOE was considered as doubtful. The existence of strong correlations between the various couples of pure tone audiometric frequencies in both experiments were quite expectable from a clinical point of view: the diseases which were studied in El (presbycusis) and E2 (acoustic trauma) basically induct: high frequency losses centered around 4 to 6 kHz and progressively extending down to low frequencies, 2, 1 and even 0.5 kHz during their evolution. In most cases, a slight audiometric defect limited to 4 and 6 kHz is associated with a normal threshold at frequencies below 2 kHz, and on the opposite, a very

impo~ant notch at 4 kHz corresponds to a significant hearing loss at lower frequencies even down to 0.5 kHz or less. Atypical cases of sensorineural hearing loss at 2 or 1 kHz and normal hearing at 4 kHz are seldom or never encountered in presbycusis or acoustic trauma. Standard statistical studies of general correlations are thus insufficient to point out significant correlations between any EOE parameter, for instance their threshold, and particular audiometric frequencies. Rigorous statistical methods with multivariate analysis are thus necessary for this purpose. Both experiments indicate that the 1 kHz EOE threshold is significantly correlated (P < 0.001, highly significant) only with the audiometric threshold at 2 kHz, and E2 shows that the 750 Hz EOE threshold is significantly correlated (P < 0.001, highly significant) only with the audiometric threshold at 1.5 kHz. These results are surprising because they suggest a non local relation between the threshold of EOE (at frequencies around 1 kHz or 750 Hz), and the hearing loss at 2 kHz (resp. 1.5 kHz). Moreover, experiment E2 in which audiometric thresholds were sampled at half-octave intervals between 1 and 8 kHz indicates that the shift between the frequency of EOE at threshold and the correlated audiometric frequency is of the order of 1 octave (and certainly more than l/2 octave). Most studies published up to now and dealing with EOE in pathology have not shown such a relation Correlations have been found between the EOE threshold and three parameters: the subjective threshold to a click (Bonfils et al., 1988a; Collet et al., 1989), the intel~~bility threshold measured through vocal audiometry (Bonfils, 1988), and the mean audiometric threshold in the interval (l-4 kHz) (Bonfils et al., 1988b). The main conclusion of these studies was that EOE disappear when any of these 3 parameters is 2 30 dB. Kemp et al. (1990) also reckoned that EOE tend to disappear in a given octave interval where the ~~rnurn hearing loss measured on the puretone audiogram is greater than 15-20 dB. It must be emphasized that many clinical studies using the IL088 device (Kemp et al., 1990) or any other one using linear artifact rejection are not based on EOE threshold measurements (click level between -5 and 5 dB HL) but rather on the frequency

107

spectrum of saturated click-evoked EOE (click level 2 40 dB HL), which may correspond to different properties of EOE. Before discussing possible models accounting for such a behaviour of EOE, and their consequences for instance for screening of auditory dysfunctions in infants, the question of the validity of results deserves a few comments: - there was no selection of the patients taking part in the protocols but the following one: the hearing losses studied in El and E2 were investigated using traditional otologic methods and were purely of endocochlear origin. They clearly involved the outer hair cells (OHC), and the link between the existence of EOE and the normal function of OHC is widely admitted: presbycusis (El) has been described as a progressive degeneration of OHC in addition to other structures, nerve fibers, stria vascularis and spiral ganglion (Schucknecht, 1974), and OHC seem to be the first altered structures at the beginning of presbycusis (Bonfils et al., 1988a; Bonfils, 1988). Acoustic trauma (E2) gives rise to more or less limited lesions of the stereociliae of the three rows of OHC. For a pure tone overstimulation at a frequency f, these lesions appear mainly basally with respect to the place with characteristic frequency f (with half an octave shift-Davis, 1983; Dancer, 1988). For impulse noise (e.g. weapon noise), the first altered places correspond to a characteristic frequency of 4 or 6 kHz (Dancer, 1988). - the characteristics of EOE (amplitude and spectrum) depend on two parameters besides the inner ear properties themselves: the stimulus frequency spectrum and the transmission through the middle ear. The possible role of these parameters in the apparently prominent relation between 1 kHz EOE threshold and 2 kHz pure tone threshold (and between 750 Hz EOE threshold and 1.5 kHz pure tone threshold) must be carefully discussed to eliminate any bias. In El, the spectrum of the acoustic stimulus in the outer ear canal was generally reasonably flat in the interval (500-4000 Hz), but depending on the individual impedance of some tested ears, it could sometimes exhibit a rather broad peak (about 10 dB enhancement between 2 and 4 kHz). When the stimulus spectrum was not flat enough, the probe was

removed and placed again for another try. In E2, the level of the continuous pure tone stimulus could easily be controlled and set constant in the explored range, especially at 1 kHz and 750 Hz for which the EOE thresholds were measured. - It is difficult to exactly determine the contribution of the middle ear transfer function to the amplitude of the various EOE. However it is clear that at least in E2, it did not play any role since the frequency of the threshold EOE was fixed (between 920 and 1130 Hz, and between 720 and 780 Hz). In El, this frequency was also close to 1 kHz in most cases as previously discussed. The observed correlations between 1 kHz EOE and 2 kHz threshold (and between 750 Hz EOE and 1.5 kHz threshold) thus appear to arise from properties of the cochlea. Most theoretical models for EOE are based on the assumption that, in addition to active mechanisms (Davis, 1983), local cochlear mapping irregularities are necessary to give rise to frequency-specific evoked otoemissions (Wilson, 1980a,b), which would be the consequence of some unbalance in the active mechanisms (it is not the case for their other manifestations such as acoustic distortion products). The role of such irregularities has been proven in some cases for spontaneous emissions (SOE) in chinchilla or dog (Ruggero et al., 1983): the loss of a few OHC at a place with a characteristic frequency f, may be associated with SOE at f. In man, SOE have been found at frequencies corresponding to boundaries between normal and pathological areas. Similar mechanisms have been proposed and investigated in the case of EOE. Mapping irregularities can be non-regular, randomly placed on the basilar membrane (BM), or on the opposite quasi periodical, possibly due to periodical changes in the pattern of OHC. This hypothesis was proposed by Manley (1983) in man to explain the observation that the spacing of EOE frequencies is often found to have a period of 0.4 bark (i.e. 80 Hz around 1 kHz) in normal subjects. In some models (for instance Furst and Lapid, 1988), considering the cochlea as a transmission line with non-linear properties, a discontinuity at a point with characteristic frequency fd gives rise to an EOE at fd. Such models could account for strong isolated peaks often found in the spectrum of EOE (Wit and Ritsma, 1980; Ruggero et al.,

108

1983). However, there is no proving that this direct relation is general. Other authors proposed that EOE could arise from a global activity of the cochlea, more precisely from imperfect cancellation between responses from various places of the organ of Corti (Wilson, 1980a; Sutton and Wilson, 1983). One hypothesis is that EOE come from stationary waves along the cochlea, with multiple reflections due to mismatch at the cochlear base (Kemp, 1986; Wilson, 1980a,b). Reflections might also occur in the cochlea because of periodic inhomogeneities of BM parameters, due to spatial modulations of active mechanisms. The scattering of waves by such structures was recently described by Strube (1989) as Bragg reflections, giving rise to long-delay EOE with a broad periodic spectrum. Such descriptions partially fit several experimental data, but little attention has been paid to their consequences on the amplitudes of EOE and on their pathological variations. In order to interprete the results of this work, they can be taken into account in the following simplified way: let us assume that the amplitude of EOE in the linear part of their characteristics, and thus EOE threshold, are proportional to the number of ‘scattering sites’, which is itself roughly proportional to the length of active basilar membrane. When active mechanisms are normal all along the BM, the corresponding value of the EOE threshold is of the order of 5 dB HL (mean value in group A, El). Any pathological alteration reducing the length of active BM should increase the EOE threshold in relation with the percentage of residual active BM. This percentage can be evaluated in the following way: Active mechanisms normally exist at least in the basal part of the organ of Corti, in the frequency interval where EOE and distortion products can be detected. Their limit on the low frequency side is not known, but we are dealing with EOE measured at threshold, i.e. evoked for instance at 1 kHz in El and E2. According to BekkCsy (and Davis, 1983), the propagated wave at this frequency extends only basally with respect to the place with characteristic frequency 1 kHz. Let us therefore consider that the contribution of active mechanisms for determining EOE threshold at 1 kHz extends from 1 to 16 kHz in normal subjects. The upper limit of

16 kHz may be inexact but it does not really influence this model, as it will be shown later on. This active interval covers four octaves. It has been shown on various species that the BM repartition of characteristic frequencies is logarithmic (at least in this range), i.e. one octave occupies a constant length on the BM independent of its center frequency (review in Dancer, 1988). The following intervals (l-2), (2-4), (4-8), (8-16) kHz now correspond to the same length and according to our previous hypothesis, to the same amount of active scattering sites: their contribution to EOE amplitude and threshold should therefore be identical. Let us now discuss two examples within these hypotheses: - The first one considers a given subject whose audiogram would reflect the disappearance of active phenomena in the interval (4-16) kHz. This means that only 50% of active sites are available for generating 1 kHz EOE. To make up for that, the stimulus level must be doubled to evoke detectable EOE: their threshold is thus increased by 6 dB. The audiometric changes are already quite significant (pure tone threshold normal at 4 kHz but altered at 6 and 8 kHz), however the change in EOE threshold is hardly detectable: the precision of its measurement for a single given ear is + 5 dB and the dispersion among a population of normal ears is of the same order of magnitude. - the second example assumes that the limit between normal and altered OHC extends down to 2 kHz. 75% of active sites have now disappeared and EOE threshold is increased by 12 dB instead of 6. This change becomes detectable. 2 kHz corresponds to the limit between normal and pathological areas and the pure tone threshold at this frequency would begin to increase if the lesions went on evoluting. It is easy to prove that the exact formula relating the change AT in EOE threshold with the limit frequency f between normal and pathological areas is: AT (dB) = 20 x log(log f in kHz, log base 10

f/log

16) (4)

The first logarithm function arises from the conversion of the percentage of active BM into a

109

threshold value expressed in dB HL (reference normal ear: 100% active BM, AT = 0), the second one (inside brackets) expresses the logarithmic relation between f and position along BM. The absolute value of this function is plotted on Fig. 5. It has a fairly particular shape due to its double log expression: its derivative is proportionai to (f log f )-‘, which is small (I 1) whenever f 2 2.5 kHz and very large (s 1) when f < 2 kHz (tending to infinity when f tends to 1 kHz). As suggested by the qualitative discussion presented above, its behaviour changes very sharply around 2 kHz. That means that, according to this model, the value of audiometric threshold at 2 kHz is strongly correlated with the value of 1 kHz EOE threshold which for instance increases from 12 to 30 dB when the cutoff frequency on the audiogram is shifted down from 2000 to 1125 Hz. Accounting for the shape of most audiograms found in El and E2, this generally corresponds to a variation of 30 to 50 dB of the 2 kHz audiometric threshold, in good agreement with the results obtained from the statistical analysis of the data (Fig. 4a,b, regressions). Obviously, the other audiometric thresholds do not play any role for determining the 1 kHz EOE threshold: - any alteration of the 1 kHz threshold indicates an alteration of the whole length of efficient active BM thus undetectable EOE, - any alteration of 3, 4, 6 and/or 8 kHz thresholds with a normai threshold at 2 and 1 kHz corresponds to a change of EOE threshold of less than 10 dB hardly detectable. Let us now consider the case of the EOE at 750 Hz: the same model can be used and the only difference with 1 kHz EOE is that the active interval on the basilar membrane is now 750 Hz16 kHz (i.e. about 4.5 octaves), so that: AT (dB) = 20 x log(log( f/O.75)/log(16/0.75))

(5) f still being the limit frequency between normal and pathology areas, expressed in kHz. The corresponding function is similar to the one drawn on Fig. 5 except that AT tends to infinity at 750 Hz and that its sharp change of behaviour now occurs at 1500 Hz: AT becomes important (12 dB)

manimum detectable changes minimum

cutoff

frequency

(kHz>

Fig. 5. Theoretical relation between 1 kHz EOE threshold and cutoff frequency on the typical audiogram of a sensorineural high frequency hearing loss, when it is assumed that EOE amplitude is proportional to the length of ‘active’ basilar membrane. (Note that near normal subjects are represented on the right side of the X axis, and that a decrease in abscissa corresponds to an increase of the audiometric notch: around 1 kHz, the cochlear function is deeply altered). An increase in EOE threshold less than 10 dB HL is hardly detectable, and when it is higher than 30-35 dB, EOE have disappeared. The ‘sensitive’ cutoff frequency is thus in between these two boudaries, i.e. 2 kHz in agreement with the statistical analysis summarized on Fig; 4.

and clearly detectable when 75% of the active basilar membrane is altered and 25% remains active, thus when the cutoff on the audiogram is at 1.12 octave (i.e. 4.5 octaves/4) above 750 Hz, which corresponds to 1.5 kHz. The exact value of the upper frequency limit of active mechanisms (whether it should be 16 or 8 kHz) does not influence the discussion very much: the important point here is that a slight change in active mechanisms around 2 kHz in the first case, or 1.5 kHz in the second one, corresponds to a small absolute but very large relative change in the percentage of active BM because the remaining percentage of normal OHC is already small when the cutoff at the audiogram is around 2 kHz (resp. 1.5 kHz). Remark: in experiment El, 15% of ears had an EOE frequency in the range 1.5-2 kHz at threshold (Fig. 3). This slight imperfection was due to time saving requirements in the clinical set up. According to the above discussion, a correlation should exist between these EOE thresholds and the audiometric threshold at 3 kHz. Since it occurred only for a small number of ears and since correla-

110

40 ;

35

9

30

" cx ", (u 2

20

25

15 IO

?? -

5 0 -5 IO 0

05

length

IO

15

of actwe Bfl

20

tnumber

25

30

of octaves)

Fig. 6. Variation of 1 kHz EOE threshold as a function of the length of normal basilar membrane (normal active mechanisms in OHC), expressed in octaves between 1 and 8 kHz. The boundary value of pure tone threshold between normal and altered places was fixed at 15 dB (mean value of peaks in the BtSkCsy sweeping-frequency audiogram). Theoretical values (same as Fig. 5): filled triangles and line. Experimental data (E2): open circles.

tions were examined with audiomet~c frequencies 2 and 4 kHz and not for 3 kHz, it had no detectable influence on the final result. This simple model leads to a very good agreement with the results derived from statistical analysis. It clearly appears that the frequencies 2 kHz and 1.5 kHz actually do not play any particular or prominent role in co&ear mechanics but are privileged only because the sensitivity of EOE threshold measurements (in order to separate normal and abnormal cochleas) is best between 10 and 30 dB (so called ‘sensitive area’ on Fig. 5), which is a narrow dynamic range. In particular cases where the variations of auditory threshold are non-monotonous between 1 and 6 kHz, the linear relation generally found between 1 kHz EOE threshold and 2 kHz threshold (or between 750 Hz EOE threshold and 1.5 kHz threshold) is expected to be inadequate. A more straightforward analysis of the results taking into account such particular cases is presented on Fig. 6: experimental data for 1 kHz EOE threshold are plotted in a similar way to Fig. 5, as a function of the length of active operating BM. This parameter was computed from the pure tone (sweeping frequency) automatic BCkCsy audiograms of subjects E2, by measuring the length (in octave) of the frequency intervals with a pure tone threshold (mean value of upper and lower spikes on the

automatic audiogram) better than 15 dB HL between 1 and 8 kHz (three octaves interval). The agreement between experimental data and the expected results (computed from the presented hypothesis) is quite good. EOE threshold values vary between -5 and 15 dB HL for normal subjects and those having a normal audiogram over more than one and a half octave in the interval (l-8) kHz. EOE threshold then increases up to 30-35 dB HL when the length of active basilar membrane is reduced. The interval where EOE threshold is most sensitive to the general state of the cochlea corresponds to half an octave residual activity. Of course, the model is probably too simplified because it considers that every scattering site on the BM plays the same role. It would probably become necessary to introduce some kind of weighting along the BM, particularly if EOE threshold was studied for other frequencies such as 4 kHz for which it can be expected that the most apical parts of BM do not participate. It could be suggested for instance that the shape of the travelling wave envelope at 750 Hz or 1 kHz could be taken into account in the proposed weighting. At first sight, the results presented here and their proposed interpretation do not seem in agreement with those published by some authors (e.g. Probst et al, 1986; Norton and Neely, 1987). Both studies used tone bursts at various frequencies to evoke otoemissions and a close correspondence was found between stimulus and response spectra. This was the main argument to suggest that EOE are generated at sites along the cochlea corresponding to their frequency. Actually, these results were obtained on a small number of subjects (14 in the first paper, 7 in the second one), and all subjects were normal so that such studies could not find any influence of high frequency hearing losses on EOE. Moreover, EOE amplitudes were not analysed. It would thus be unfounded to conclude that our results are in contradiction with these two references, or against the hypothesis that EOE are generated at sites along the cochlea corresponding to their frequency: the data presented here only indicate that all locations (or many locations) along the cochlea contribute to EOE amplitude, and EOE cannot be considered

111

as a purely local phenomenon. From this statement, one could propose two types of hypothesis: - EOE arise from a global cochlear activity (and not from definite places) as stated by some previously discussed models (e.g. Strube, 1989), - EOE are generated at definite sites but they also need a good global cochlear state to propagate back to the oval window and become detectable. The sets of results presented here yield two categories of concluding remarks: - From a fundamental point of view, they suggest that EOE (or at least most of them, maybe those corresponding to a broad frequency spectrum) originate from phenomena (most probably active and fragile ones) widely distributed along the basilar membrane. The behaviour of an EOE at 1 kHz does not describe the cochlear state at this frequency but is related to the shape of the whole audiogram for medium and high frequencies. These data on EOE could help in the modelling of their still unclear mechanisms. - The clinical applications of EOE measurements are now more and more widely used as a routine practice in screening applications (e.g. Probst et al., 1987; Kemp et al., 1990). Their specificity as a OHC probe and their sensitivity have suggested to many users that they might be much useful for early screening (e.g. ototoxicity) or long term mo~to~ng (e.g. noise induced hearing loss). However, the results of this work indicate that some EOE characteristics, in particular their amplitude and threshold, are not frequencyspecific and are altered only when the hearing loss is already important and associated with a change in audiometric threshold at 2 kHz. This means that in most cases of high frequency hearing loss, the possibility of an early diagnosis (based on EOE amplitude measurements, but maybe not on their frequency spectrum) appears highly uncertain: tonal and high frequency audiometry are more sensitive and of course more frequencyspecific (Dreschler et al., 1985). On the other hand, it is difficult to get an objective assessment of the cochlear function in the case of neonates and infants. Since a global evaluation of the active cochlear mechanisms can be obtained by EOE detection, the main interest of this test is that it allows to clearly separate two populations (e.g.

Johnsen et al., 1988; Bonfils et al., 1990): (i) with normal auditory peripheral function (presence of EOE), (ii) with pathological auditory peripheral function (absence of EOE). Adcnowledgements

A part of this work (E2) was supported by a grant (89-090) of the French Ministry of Environment (the authors are specially indebted to Mr PC. Jacquignon, SRETIE) and another part (El) by grants from ‘Fonds d’Etude du Corps Medical des Hopitaux de Paris’ and from ‘Fondation de France’. The authors would also like to thank Pr. Y. Galifret and Dr. Ph. Gaimerin (University Paris VI) for helpful comments on the statistical part of the work, and Pr. P. Buffe (Val-de-G&e Hospital, ENT Service) for supplying several patients. References Avan, P., Loth D., Menguy C. and Teyssou M. (1990) Evoked otoacoustic emissions in guinea pig: basic characteristics. Hear. Res. 44, 151-160. Bonfils P., Bertrand Y. and Uziei A. (198Sa) Evoked otoacoustic emissions: normative data and presbycusis. Audiology 27, 27-35. Bonfils P., Piron J.P., Uziel A. and Pujol R. (1988b) Correlative study of evoked otoacoustic emissions properties and audiometric thresholds. Arch. Otorhinolaryngol. 245,53-S& Bonfils P. (1988) Les alterations de I’audiometrie vocale dans la presbyacousie: apport des emissions acoustiques cochl&res. J. Otolaryngol. 17: 5, 207-210. Bonfils P., Dumont A., Marie P., Fran&s M. and Narcy P. (1990) Evoked otoacoustic emissions in newborn hearing screening. Laryngoscope 100, 186-189. Collet L., Gartner M., MouIin A., Kauffmann I. and Morgon A. (1989) Evoked otoacoustic emissions and sensorineural hearing loss. Arch. Otolaryngol. Head Neck Surg. 115, 1060-1062. Dancer A. (1988) Biomecanique de l’audition. In: Physiologic de la cochiee, INSERM Eds., Paris, pp. 27-73. Davis H. (1983) An active process in co&ear mechanics. Hear. Res. 9, 79-90. Dixon W.J., Brown M.B., Engelman L., Frane J.W., Jennrich RI. and Toporek J.D. (1985) BMDP statistical software. Berkeley Univ. California Press. DreschIer W.A., v.d. Hutst R.J., Tange R.A. and Urbanus N.A. (1985) The role of high-frequency audiometry in early detection of ototoxicity. Audiology 24, 387-395. Furst M. and Lapid M. (1988) A cochlear model for acoustic emissions. J. Acoust. Sot. Am. 84, 222-229.

112 Grenner J., Nilsson P. and Karbamna B. (1990) Right-left correlation in guinea pig ears after noise exposure. Acta Otolaryngol. 109, 41-48. Johnsen N.J., Bagi P., Parbo J. and Elberling C. (1988) Evoked emissions from the human ear IV: final results on 100 neonates. Stand. Audio]. 17, 27-34. Kemp D.T. (1978) Stimulated acoustic emissions from within the human auditory system. J. Acout. Sot. Am. 64, 13861391. Kemp D.T. and Chum R. (1980) Properties of the generator of stimulated acoustic emissions. Hear. Res. 2, 213-232. Kemp D.T. (1986) Otoacoustic emissions, travelling waves and cochlear mechanics. Hear. Res. 22, 95-104. Kemp D.T., Bray P., Alexander L. and Brown A.M. (1986) Acoustic emission cochteography - practical aspects. Stand. Audiol. suppl 25, 71-95. Kemp D.T., Ryan S. and Bray P. (1990) A guide to the effective use of otoacoustic emissions. Ear Hear. (in press). Kim D.O. (1986) Active and nonlinear co&ear biom~h~cs and the role of outer-hair-cell subsystem in the mammalian auditory system. Hear. Res. 22, 105-114. Manley G. (1983) Frequency spacing of acoustic emissions: a possible explanation. In: W.R. Webster and L.M. Aitkin (Eds.), Mechanisms of Hearing, Monash University Press, Clayton, Australia, pp.36-39. Montgomery D. and Peck E. (1982) Introduction to linear regression analysis. J. Wiley and Sons, New-York. NewYork, pp.275. Norton S.J. and Neely ST. (1987) Tone burst evoked otoacoustic emissions from normal-hearing subjects. J. Acoust. Sot. Am. 81, 1860-1871. Probst R., Coats A.C., Martin G.K. and Lonsbury-Martin B.L. (1986) Spontaneous, click- and tone burst-evoked otoacoustic emissions from normal ears. Hear. Res. 21, 261-276.

Probst R., Lonsbu~-Martin B.L., Martin G.K. and Coats A.C. (1987) Otoacoustic emissions in ears with hearing loss. Am. J. Otolaryngol. 8, 73-81. Ruggero M.A., Rich N.C. and Freyman R. (1983) Spontaneous and impulsively evoked otoacoustic emissions: indications of cochlear pathology? Hear. Res. 10, 283-300. Schucknecht II. (1974) Pathology of the ear, Harvard Univ. Press, Cambridge. Strube H.W. (1989) Evoked otoacoustic emissions as cochlear Bragg refIections. Hear. Res. 38, 35-46. Sutton G.J. and Wilson J.P. (1983) Modelling cochiear echoes: the influence of irregularities in frequency mapping on summed cochlear activity. In: E. de Boer and M.A. Viergever @Is.), Mechanisms of Hearing, Defft Univ. Press. Delft, The Netherlands. pp. 83-90. Wilson J.P. (198Oa) Model of cochlear function and acoustic reemission. In: G. van den Brink and F.A. BiIsen (Eds.), Psychophysical, Physiological and Behavioural Studies in Hearing. Delft Univ. Press, Delft, The Netherlands. pp. 72-73. Wilson J.P. (1980b) Evidence for a cochlear origin for acoustic reemissions. threshold fine structure and tonal tinnitus. Hear. Res. 2, 233-256. Wit H.P. and Ritsma R.J. (1980) Evoked acoustic responses from the human ear: some experimental results. Hear. Res. 2, 253-261. Zwicker E. (1986) Otoacoustic emissions in a nonlinear cochlea hardware model with feedback. J. Acoust. Sot. Am. 80, 154-162. Zwicker E. and Schloth E. (1984) Interrelation of different otoacoustic emissions. J. Acoust. Sot. Am. 75, 1148-1154.