- Email: [email protected]
Linear and nonlinear transient evoked otoacoustic emissions in humans exposed to noise
Hearing Research 174 (2002) 290^295 www.elsevier.com/locate/heares
Linear and nonlinear transient evoked otoacoustic emissions in humans exposed to noise A. Moleti a b c
a;
, R. Sisto b , M. Lucertini
c
Dipartimento di Fisica, Universita' di Roma ‘Tor Vergata’, Via della Ricerca Scienti¢ca, 1, 00133 Rome, Italy Dipartimento Igiene del Lavoro, ISPESL, Via Fontana Candida, 1, 00040 Monte Porzio Catone (Rome), Italy Italian Air Force ^ CSV Aerospace Medicine Department, Pratica di Mare AFB, 00040 Pomezia (Rome), Italy Received 31 January 2002; accepted 18 September 2002
Abstract Transient evoked otoacoustic emissions (TEOAEs) have been analyzed in a population of 134 ears, divided into three classes: (1) nonexposed ears in bilaterally normal hearing subjects, (2) audiometrically normal ears of subjects exposed to noise and affected by unilateral high-frequency (f s 3 kHz) hearing loss in the contralateral ear, and (3) the contralateral impaired ears of the exposed subjects. The statistical distributions of global and spectral signal-to-noise ratio (SNR) were analyzed. TEOAEs were recorded both in the linear and nonlinear acquisition mode to evaluate the effectiveness of two standard averaging techniques with respect to their sensitivity to the early effects of noise exposure. Good discrimination between nonexposed and exposed ears was obtained using either the linear or the nonlinear mode. Despite its intrinsically higher SNR, the linear mode is not more sensitive than the nonlinear mode for this purpose because it is not possible to find a window for effectively cancelling the linear artifact while keeping a suitable sensitivity to the short-latency high-frequency aspect of the response. Moreover, with respect to another measurable parameter, the TEOAE latency, good discrimination is obtained only by using the nonlinear mode because, again, the linear artifact masks the high-frequency TEOAE response. 5 2002 Elsevier Science B.V. All rights reserved. Key words: Transient evoked otoacoustic emissions; Hearing loss; Exposure to noise
1. Introduction Otoacoustic emissions (OAEs) provide a wide set of experimental techniques that are e¡ectively used to investigate cochlear function. Important studies have established a correlation between subjective hearing sensitivity expressed by audiometric threshold and objective OAE presence and level. For example, it has been shown that the separation between normal-hearing and hearing-impaired ears can be e¡ectively performed above 1 kHz by analyzing the transient evoked OAE (TEOAE) response, the sig* Corresponding author. Tel.: +39 (06) 7259 4288; Fax: +39 (06) 2023 507. E-mail address: [email protected] (A. Moleti). Abbreviations: SNR, signal-to-noise ratio; OAE, otoacoustic emission; TEOAE, transient evoked otoacoustic emission; SOAE, spontaneous otoacoustic emission
nal-to-noise ratio (SNR) and reproducibility (Prieve et al., 1993; Hussain et al. 1998). A further possible clinical application of OAEs is the early detection of sub-clinical noise-induced cochlear damage. The resonant nature of the cochlear response and the well-known correspondence between minima of the audiometric threshold ¢ne structure and spontaneous OAE (SOAE) frequencies (Furst et al., 1992) suggest that the most sensitive cochlear regions, corresponding to the frequencies of the largest OAEs, also are the most sensitive to cochlear damage. TEOAE signals are dominated by the contribution of a few resonant lines (Sisto et al., 2001) whose amplitude is sensitive to even small variations in local cochlear properties. Thus the TEOAE response is expected to be sensitive to exposure also within a sub-clinical hearing loss range. Recently, Lucertini et al. (2002), using TEOAE parameters such as SNR, response level, reproducibility
0378-5955 / 02 / $ ^ see front matter 5 2002 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 8 - 5 9 5 5 ( 0 2 ) 0 0 7 0 3 - 7
HEARES 4021 30-10-02
A. Moleti et al. / Hearing Research 174 (2002) 290^295
and latency, uncovered statistically signi¢cant di¡erences, not only between normal and impaired ears but also between audiometrically normal exposed and nonexposed ears (ear classes 1 and 2 of the present work). The ¢rst result was expected, due to the well-known correlation between audiometric hearing loss and reduced OAE level (e.g., Hall and Lutman, 1999). The second result, con¢rming the ¢ndings of Attias et al. (1995) and Kowalska and Sulkowski (1997), was an indication that the OAE technique is particularly sensitive in the range of mild ( 6 20 dB) hearing losses, which correspond to normal hearing according to a standard dichotomous audiometric clinical criterion. It should be stressed that the above results only demonstrate the possibility of showing signi¢cant statistical di¡erences between the TEOAE average parameters measured in exposed and nonexposed populations in a sub-clinical hearing loss range. This is just a prerequisite for investigating if a reliable prediction of mild hearing impairment may actually be obtained for an individual from a single TEOAE measurement. This is a much more di⁄cult task, due to the large variability of the TEOAE response within a population of normal subjects. Another, and perhaps more promising, application of TEOAEs could be the monitoring of exposed populations of workers, by repeated TEOAE measurements on the same subjects. In the present work, the same analysis of Lucertini et al. (2002) was performed using TEOAE data from the same subjects, recorded with the ILO-96, V.5 system (Otodynamics), both in the linear and nonlinear modes of acquisition, to compare the e¡ectiveness of the two standard TEOAE-acquisition paradigms. The performances of the linear and nonlinear averaging paradigms were compared elsewhere (Tognola et al., 2001) with respect to SNR and reproducibility that can be obtained and the rejection of noncochlear artifacts. It was found that, in the saturated regime using standard protocols, the SNR of the nonlinear technique was smaller by a factor of two (6 dB) due to the di¡erential averaging procedure. This is an obvious consequence of the fact that the nonlinear paradigm yields a quantity that is half the nonlinear response due to the subtraction algorithm used for cancelling the linear artifact. On the other hand, the linear waveform is a¡ected by the linear ringing artifact during the ¢rst 5^6 ms. The standard analysis window using the commercial ILO system starts 2.5 ms after the click, and is not su⁄cient to cancel it. A window starting 5^6 ms after the stimulus could solve this problem (Tognola et al., 2001), with the obvious drawback of cancelling also the high-frequency cochlear response, whose latency is shorter than 5 ms (Sisto and Moleti, 2002). In the present work, the comparison between the linear and nonlinear data of the same subjects is made to
291
compare the e¡ectiveness of the two techniques concerning their capability of discriminating between normal and impaired (and also between exposed and nonexposed) ears. The relevant parameter is not in this case the absolute SNR that can be obtained by a given acquisition paradigm, but the di¡erence between the average SNR values obtained in the two ear classes to be compared. More precisely, the relevant quantity is the ratio between this di¡erence and the square root of the weighted mean of the variances of the two SNR distributions, i.e., the e¡ect size (Cohen, 1992), which is directly related to the statistical signi¢cance of the di¡erence, quantitatively expressed by the Student’s t-test probability. Standard TEOAE parameters, such as global and band reproducibility, response level and SNR were used in the present work. As these three parameters are strictly correlated to each other, very similar results are obtained using any of them, and thus only results for the SNR are presented. Another parameter was also used: the TEOAE spectral latency, i.e., the time delay between the click stimulus and the time of maximum response in a given frequency band, which has also recently been shown to be correlated to audiometrically determined hearing loss (Sisto and Moleti, 2002).
2. Materials and methods In this work audiometric and TEOAE linear and nonlinear data of a population of 134 ears from 67 young (age = 18^25 yr) male subjects were studied. Except for two subjects for which the linear TEOAE data are missing, the population is the same as that analyzed in a previous study (Lucertini et al., 2002), in which only the nonlinear data were reported. More details about the examined population and the methodology can be found there. Pure tone audiograms were recorded for all subjects in an acoustically shielded room, using a clinical audiometer (Amplaid A-460) equipped with standard headphones (TDM-39), according to the procedure described by Yantis and Katz (1994). The audiometric test frequencies were 0.25, 0.5, 1, 2, 3, 4, 6, and 8 kHz. As in Lucertini et al. (2002), a standard dichotomous clinical de¢nition of hearing impairment was used. Speci¢cally, the examined ear was de¢ned as ‘normal’ if no absolute threshold level greater than 20 dB was measured over the whole frequency range. If a threshold greater than 20 dB was observed in any audiometric range f v 3 kHz, the ear was de¢ned as ‘high-frequency impaired’. Subjects were also divided, on the basis of a preliminary interview, into two subsets: non-noise-exposed (n = 46) and noise-exposed (n = 21). Nonexposed sub-
HEARES 4021 30-10-02
292
A. Moleti et al. / Hearing Research 174 (2002) 290^295
jects stated they had never been exposed to either recreational noise or to other ototoxic agents. As all subjects live in urban areas of Italy, this negative statement means they had actually been exposed to the standard urban noise level, which is an unavoidable exposure background for all subjects considered in the present study. However, the audiometric test showed that none of them had hearing loss. Exposed subjects had been exposed for 1 yr without ear protection to ¢rearm noise with variable exposure durations. They were a¡ected by unilateral high-frequency hearing loss, which is compatible with the exposure to impulsive noise. None of them had pre-existing hearing loss, and none of them reported exposure to other ototoxic agents, thus hearing loss was presumably related to the training with ¢rearms. According to the results of the audiometric test, three classes of ears were identi¢ed: (1) ears of nonexposed bilaterally normal subjects (92 ears from 46 subjects), (2) audiometrically normal ears of exposed subjects exhibiting unilateral high-frequency hearing loss in the contralateral ear (21 ears from 21 subjects), and (3) the impaired ears of the same subjects (21 ears from 21 subjects). TEOAE recordings were obtained using the ILO-96 system in the nonlinear mode of acquisition, with a stimulus level of 80 N 3 dB peSPL. The standard ILO window, which is a trapezoidal window from 2.5 to 20 ms with 2.5 ms rise time, was applied to the nonlinear data. Synchronized SOAEs (SSOAEs) from the same subjects were recorded in the linear mode, using the same stimulus level (within N 1 dB). SSOAE recordings yield information about the long-lasting components of the TEOAE response (SOAEs) that was not used in the present work. Only the ¢rst 20 ms of the 80-ms SSOAE linear recordings were used, to allow comparison between linear and nonlinear responses. Three di¡erent windows were applied to the linear recordings: the standard ILO window, an ILO window shifted forward in time by 2.5 ms (i.e., with onset and £at-top at 5 ms and 7.5 ms after the click, respectively), and an ILO window shifted forward in time by 5 ms (onset at 7.5 ms, £attop at 10 ms). These three di¡erent temporal windows
were applied o¡-line to the linear data, to analyze the e¡ect of windowing on data quality, which is a critical issue when using linear responses due to the presence of the linear ringing artifact. Both the TEOAE and SSOAE recordings were processed by using customized software developed in LabVIEW 5.0 (National Instruments). Global and thirdoctave band reproducibility, response and SNR were evaluated using the de¢nitions and the method described in Lucertini et al. (2002). Spectral latency was measured by means of a wavelet transform technique based on the iterative application of ¢lter banks to the TEOAE waveform, as described in Sisto and Moleti (2002). The statistical distributions of the examined TEOAE parameters were analyzed, and the statistical signi¢cance of the di¡erences between the three classes were evaluated according to the Student’s t-test. A 95% con¢dence level (corresponding to a probability p 6 0.05) was arbitrarily chosen as the criterion for statistical signi¢cance.
3. Results As pointed out in a previous work (Lucertini et al., 2002), reproducibility, response and SNR are strictly correlated quantities because the noise level is quite independent of the response level and reproducibility is just a di¡erent de¢nition of the SNR. Thus very similar results are obtained using any of these parameters, and in the following, for simplicity, only the results for the SNR will be shown. As mentioned in the Introduction, the main result of Lucertini et al. (2002) was the observation of a statistically signi¢cant di¡erence not only between normal and impaired ears, but also between nonexposed and exposed (either normal or impaired) ears. The present work shows that nonlinear and linear data are both capable of showing this result, with di¡erent e¡ectiveness. In Table 1, the average global SNR of the ears of the three classes are shown, as obtained from the nonlinear analysis, compared to those obtained by applying
Table 1 Average values of the SNR for the three classes of ears (1 = nonexposed, normal hearing, 2 = exposed, normal hearing, 3 = exposed, hearing impaired), average SNR di¡erence (vi3j ), Cohen’s e¡ect size (ESi3j ), and probability associated to the Student’s t-test computed between pairs of the classes (pi3j ) Acquisition mode/window
SNR1
SNR2
SNR3
v132
v133
ES132
ES133
p132
p133
Nonlinear/ILO Linear/ILO Linear/ILO+2.5 ms Linear/ILO+5 ms
12.7 18.1 9.3 6.2
9.0 16.4 6.8 4.0
8.1 15.7 5.2 2.5
3.7 1.7 2.5 2.2
4.6 2.4 4.1 3.7
0.75 0.69 0.43 0.34
0.98 0.97 0.75 0.58
0.018 0.014 0.12 (ns) 0.20 (ns)
0.0005 0.002 0.004 0.017
The linear data were windowed with three di¡erent windows, the standard ILO (a trapezoidal window from 2.5 to 20 ms with a 2.5 ms rise time), and two equally shaped windows, increasingly shifted forward in time, by steps of 2.5 ms.
HEARES 4021 30-10-02
A. Moleti et al. / Hearing Research 174 (2002) 290^295
Fig. 1. Di¡erence of the average SNR in third-octave frequency bands between nonexposed normal hearing ears (ne) and exposed normal hearing ears (e) (classes 1 and 2), obtained using the nonlinear data (a), and statistical signi¢cance of the di¡erence, as expressed by the probability of the Student’s t-test parameter (b). The 5% probability level assumed as the criterion for signi¢cance is shown by the horizontal solid line.
three di¡erent temporal windows to the linear data. Pushing forward in time the window caused a progressive degradation of the data due to the drastic elimination of the short-latency high-frequency response. In the following, the emphasis is put on the di¡erence between classes 1 and 2, because the early detection of exposure e¡ects in audiometrically normal subjects could be the most promising application of this study. As mentioned above, statistically signi¢cant di¡erences between classes 1 and 2 can be evidenced by using global TEOAE parameters, but more signi¢cant di¡erences can be found using the corresponding spectral band parameters. In Fig. 1a the di¡erence of the average SNR in third-octave frequency bands between classes 1 and 2 obtained using the nonlinear data is shown. The horizontal line at 3.7 dB is the di¡erence between the average global SNR of the two classes. It is clear that most of this di¡erence comes from the highfrequency bands (f s 2 kHz). In Fig. 1b the statistical signi¢cance of the di¡erence is expressed by the probability of the Student’s t-test parameter. The low-frequency part of the signal adds a non-signi¢cantly di¡erent contribution to the overall signal, which tends to lower the signi¢cance of the global SNR di¡erence between classes 1 and 2. With respect to the di¡erent
293
discrimination capability of global and band parameters, very similar results are obtained with the linear data. In Fig. 2a the di¡erence of the average SNR in thirdoctave frequency bands between classes 1 and 2 is shown for the nonlinear and the linear data, using three di¡erent delay times for the window applied to the linear data. The result shown in Table 1 for the global SNR is found again and clari¢ed by the spectral analysis. Most of the SNR di¡erence comes from the highfrequency bands, also for the linear data, and this difference tends to decrease with increasing start time of the window due to the elimination of part of the shortlatency high-frequency response. This result may be interpreted in terms of a loss of discrimination capability between exposed and nonexposed ears, as shown in Fig. 2b, where the statistical signi¢cance of the di¡erence is expressed by the probability of the Student’s t-test parameter. As shown in Table 1, the di¡erence between nonexposed normal ears and exposed impaired ears (classes 1 and 3) was always statistically signi¢cant, considering either global or spectral band SNR (above 1 kHz), for both acquisition modes and for all the windows applied to the linear data. Also in this case, the statistical sig-
Fig. 2. Di¡erence of the average SNR in third-octave frequency bands between nonexposed normal hearing ears (ne) and exposed normal hearing ears (e) (classes 1 and 2), as obtained using the nonlinear and the linear data, for three di¡erent choices of the window applied to the linear data (a) and statistical signi¢cance of the difference (b).
HEARES 4021 30-10-02
294
A. Moleti et al. / Hearing Research 174 (2002) 290^295
ni¢cance of the di¡erence decreased, for the linear data, with increasing time delay of the window. The di¡erence between classes 2 (exposed normal ears) and 3 (exposed impaired ears) was never statistically signi¢cant, also due to the smaller size. It should be noted here that e¡ect sizes of 0.2, 0.5 and 0.8 are interpreted by Cohen (1992) as small, medium and large, respectively. With respect to spectral latency, a statistically signi¢cant di¡erence between classes 1 and 2 had been found in a previous work (Lucertini et al., 2002) using nonlinear data only. In the present work it was found that the linear data cannot be used for measuring the latency with the proposed algorithm because no reasonable compromise can be found between the con£icting necessities of e¡ectively cancelling the linear artifact contribution while maintaining a su⁄cient sensitivity to the short-latency high-frequency ( s 3 kHz) response.
4. Discussion TEOAE data, either recorded in the linear or in the nonlinear mode, have shown good discrimination capability between normal and impaired (and also between normal exposed and nonexposed) ears. Thus the results of a previous study (Lucertini et al., 2002) are extended to the case of TEOAE data acquired using the linear paradigm. The presence of the linear ringing artifact, which typically a¡ects the linear recordings over the ¢rst 5^6 ms, requires proper windowing of the linear data. On the other hand, pushing the onset of the analysis window forward in time caused a decrease in the discrimination capability of the TEOAE due to the fact that most of the high-frequency signal, whose latency is shorter than 5 ms, is canceled by windows starting at times longer than 5 ms. This fact is particularly important because most of the di¡erence between normal and impaired (and between exposed and nonexposed) subjects comes indeed from the high-frequency bands. It is clear that the linear acquisition mode provides a SNR that, for the same values of all other parameters, is 6 dB higher than that given by the nonlinear mode (Tognola et al., 2001). This factor of two is due to the coherent sum of all the acquired signals in the linear mode, while in the nonlinear mode only two out of four acquired signals actually contribute to the output. Despite this well-known advantage of the linear technique, it should be stressed that the inclusion of a large linear artifact contribution has the e¡ect of lowering the difference between di¡erent classes of ears, particularly when using global TEOAE parameters. This fact may be observed in the global SNR data of Table 1 : the di¡erence between classes 1 and 2 as well as that be-
tween classes 1 and 3 is much smaller using the linear data and the standard ILO window. In fact, adding to the signal of all subjects a large spurious and deterministic component has the e¡ect of lowering the value (expressed in dB) of the SNR (or response level) di¡erence. Therefore, the comparison between the e¡ectiveness of the two acquisition modes must be done keeping into account not only the SNR level that can be obtained in given experimental conditions but also the degree of contamination by noncochlear (thus uncorrelated to cochlear functionality) contributions. Signi¢cant cochlear signal variations become less evident if accompanied by a spurious noncochlear signal. It is interesting how an added noncochlear signal and added noise di¡erently a¡ect the statistical signi¢cance of the di¡erence between the OAE response of two populations expected to have di¡erent cochlear functionality. An added spurious deterministic signal has the e¡ect of lowering the di¡erences of cochlear origin in the overall signal. Added spurious noise has instead the e¡ect of increasing the variance of the data, thus lowering the statistical signi¢cance of a given di¡erence.
5. Conclusions The results presented in this work con¢rm the idea that TEOAEs parameters, measured from either nonlinear and linear data, can be used to successfully discriminate a population of normal ears from one of impaired ears and also a population of exposed ears from one of nonexposed ears. The comparison between the e¡ectiveness of the nonlinear and linear averaging paradigms has shown that both acquisition paradigms are e¡ective if the window applied to the linear data to eliminate the artifact does not start later than 2.5 ms after the click. Longer window start times eliminate too much of the short-latency high-frequency response, from which most of the di¡erence between impaired and normal or between exposed and nonexposed ears arises. On the other hand, the addition of a large noncochlear artifact has been shown to decrease the response di¡erence between classes and its statistical signi¢cance. Thus the nonlinear mode seems to be preferable in the general case, also because it makes it possible to e¡ectively use another independent indicator of cochlear function, i.e., the TEOAE spectral latency. References Attias, J., Furst, M., Furman, V., Reshef, I., Horowitz, G., Breslo¡, I., 1995. Noise-induced otoacoustic emission loss with or without hearing loss. Ear Hear. 16, 612^618. Cohen, J., 1992. A power primer. Psychol. Bull. 112, 155^159.
HEARES 4021 30-10-02
A. Moleti et al. / Hearing Research 174 (2002) 290^295 Furst, M., Reshef, I., Attias, J., 1992. Manifestations of intense noise stimulation on spontaneous otoacoustic emission and threshold microstructure: experiment and model. J. Acoust. Soc. Am. 91, 1003^1014. Hall, A.J., Lutman, M.E., 1999. Methods for early identi¢cation of noise-induced hearing loss. Audiology 38, 277^280. Hussain, D.M., Gorga, M.P., Neely, S.T., Keefe, D.H., Peters, J., 1998. Transient evoked otoacoustic emissions in patients with normal hearing and in patients with hearing loss. Ear Hear. 19, 434^ 449. Kowalska, S., Sulkowski, W., 1997. Measurements of click-evoked otoacoustic emissions in industrial workers with noise-induced hearing loss. Int. J. Occup. Med. Environ. Health 10, 441^459. Lucertini, M., Moleti, A., Sisto, R., 2002. On the detection of early cochlear damage by otoacoustic emission analysis. J. Acoust. Soc. Am. 111, 972^978. Prieve, B.A., Gorga, M.P., Schmidt, A., Neely, S., Peters, J., Schultes,
295
L., Jesteadt, W., 1993. Analysis of transient-evoked otoacoustic emissions in normal-hearing and hearing-impaired ears. J. Acoust. Soc. Am. 93, 3308^3319. Sisto, R., Moleti, A., 2002. On the frequency dependence of the otoacoustic emission latency in hypoacoustic and normal ears. J. Acoust. Soc. Am. 111, 297^308. Sisto, R., Moleti, A., Lucertini, M., 2001. Spontaneous otoacoustic emissions and relaxation dynamics of long decay time OAEs in audiometrically normal and impaired subjects. J. Acoust. Soc. Am. 109, 638^647. Tognola, G., Ravazzani, P., Molini, E., Ricci, G., Alunni, N., Parazzini, M., Grandori, F., 2001. ‘Linear’ and ‘derived’ otoacoustic emissions in newborns: a comparative study. Ear Hear. 22, 182^ 190. Yantis, P.A., Katz, J., 1994. Puretone air-conduction threshold testing. In: Handbook of Clinical Audiology, 4th edn. Williamson and Wilkins, Baltimore, pp. 97^108.
HEARES 4021 30-10-02
Linear and nonlinear transient evoked otoacoustic emissions in humans exposed to noise A. Moleti a b c
a;
, R. Sisto b , M. Lucertini
c
Dipartimento di Fisica, Universita' di Roma ‘Tor Vergata’, Via della Ricerca Scienti¢ca, 1, 00133 Rome, Italy Dipartimento Igiene del Lavoro, ISPESL, Via Fontana Candida, 1, 00040 Monte Porzio Catone (Rome), Italy Italian Air Force ^ CSV Aerospace Medicine Department, Pratica di Mare AFB, 00040 Pomezia (Rome), Italy Received 31 January 2002; accepted 18 September 2002
Abstract Transient evoked otoacoustic emissions (TEOAEs) have been analyzed in a population of 134 ears, divided into three classes: (1) nonexposed ears in bilaterally normal hearing subjects, (2) audiometrically normal ears of subjects exposed to noise and affected by unilateral high-frequency (f s 3 kHz) hearing loss in the contralateral ear, and (3) the contralateral impaired ears of the exposed subjects. The statistical distributions of global and spectral signal-to-noise ratio (SNR) were analyzed. TEOAEs were recorded both in the linear and nonlinear acquisition mode to evaluate the effectiveness of two standard averaging techniques with respect to their sensitivity to the early effects of noise exposure. Good discrimination between nonexposed and exposed ears was obtained using either the linear or the nonlinear mode. Despite its intrinsically higher SNR, the linear mode is not more sensitive than the nonlinear mode for this purpose because it is not possible to find a window for effectively cancelling the linear artifact while keeping a suitable sensitivity to the short-latency high-frequency aspect of the response. Moreover, with respect to another measurable parameter, the TEOAE latency, good discrimination is obtained only by using the nonlinear mode because, again, the linear artifact masks the high-frequency TEOAE response. 5 2002 Elsevier Science B.V. All rights reserved. Key words: Transient evoked otoacoustic emissions; Hearing loss; Exposure to noise
1. Introduction Otoacoustic emissions (OAEs) provide a wide set of experimental techniques that are e¡ectively used to investigate cochlear function. Important studies have established a correlation between subjective hearing sensitivity expressed by audiometric threshold and objective OAE presence and level. For example, it has been shown that the separation between normal-hearing and hearing-impaired ears can be e¡ectively performed above 1 kHz by analyzing the transient evoked OAE (TEOAE) response, the sig* Corresponding author. Tel.: +39 (06) 7259 4288; Fax: +39 (06) 2023 507. E-mail address: [email protected] (A. Moleti). Abbreviations: SNR, signal-to-noise ratio; OAE, otoacoustic emission; TEOAE, transient evoked otoacoustic emission; SOAE, spontaneous otoacoustic emission
nal-to-noise ratio (SNR) and reproducibility (Prieve et al., 1993; Hussain et al. 1998). A further possible clinical application of OAEs is the early detection of sub-clinical noise-induced cochlear damage. The resonant nature of the cochlear response and the well-known correspondence between minima of the audiometric threshold ¢ne structure and spontaneous OAE (SOAE) frequencies (Furst et al., 1992) suggest that the most sensitive cochlear regions, corresponding to the frequencies of the largest OAEs, also are the most sensitive to cochlear damage. TEOAE signals are dominated by the contribution of a few resonant lines (Sisto et al., 2001) whose amplitude is sensitive to even small variations in local cochlear properties. Thus the TEOAE response is expected to be sensitive to exposure also within a sub-clinical hearing loss range. Recently, Lucertini et al. (2002), using TEOAE parameters such as SNR, response level, reproducibility
0378-5955 / 02 / $ ^ see front matter 5 2002 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 8 - 5 9 5 5 ( 0 2 ) 0 0 7 0 3 - 7
HEARES 4021 30-10-02
A. Moleti et al. / Hearing Research 174 (2002) 290^295
and latency, uncovered statistically signi¢cant di¡erences, not only between normal and impaired ears but also between audiometrically normal exposed and nonexposed ears (ear classes 1 and 2 of the present work). The ¢rst result was expected, due to the well-known correlation between audiometric hearing loss and reduced OAE level (e.g., Hall and Lutman, 1999). The second result, con¢rming the ¢ndings of Attias et al. (1995) and Kowalska and Sulkowski (1997), was an indication that the OAE technique is particularly sensitive in the range of mild ( 6 20 dB) hearing losses, which correspond to normal hearing according to a standard dichotomous audiometric clinical criterion. It should be stressed that the above results only demonstrate the possibility of showing signi¢cant statistical di¡erences between the TEOAE average parameters measured in exposed and nonexposed populations in a sub-clinical hearing loss range. This is just a prerequisite for investigating if a reliable prediction of mild hearing impairment may actually be obtained for an individual from a single TEOAE measurement. This is a much more di⁄cult task, due to the large variability of the TEOAE response within a population of normal subjects. Another, and perhaps more promising, application of TEOAEs could be the monitoring of exposed populations of workers, by repeated TEOAE measurements on the same subjects. In the present work, the same analysis of Lucertini et al. (2002) was performed using TEOAE data from the same subjects, recorded with the ILO-96, V.5 system (Otodynamics), both in the linear and nonlinear modes of acquisition, to compare the e¡ectiveness of the two standard TEOAE-acquisition paradigms. The performances of the linear and nonlinear averaging paradigms were compared elsewhere (Tognola et al., 2001) with respect to SNR and reproducibility that can be obtained and the rejection of noncochlear artifacts. It was found that, in the saturated regime using standard protocols, the SNR of the nonlinear technique was smaller by a factor of two (6 dB) due to the di¡erential averaging procedure. This is an obvious consequence of the fact that the nonlinear paradigm yields a quantity that is half the nonlinear response due to the subtraction algorithm used for cancelling the linear artifact. On the other hand, the linear waveform is a¡ected by the linear ringing artifact during the ¢rst 5^6 ms. The standard analysis window using the commercial ILO system starts 2.5 ms after the click, and is not su⁄cient to cancel it. A window starting 5^6 ms after the stimulus could solve this problem (Tognola et al., 2001), with the obvious drawback of cancelling also the high-frequency cochlear response, whose latency is shorter than 5 ms (Sisto and Moleti, 2002). In the present work, the comparison between the linear and nonlinear data of the same subjects is made to
291
compare the e¡ectiveness of the two techniques concerning their capability of discriminating between normal and impaired (and also between exposed and nonexposed) ears. The relevant parameter is not in this case the absolute SNR that can be obtained by a given acquisition paradigm, but the di¡erence between the average SNR values obtained in the two ear classes to be compared. More precisely, the relevant quantity is the ratio between this di¡erence and the square root of the weighted mean of the variances of the two SNR distributions, i.e., the e¡ect size (Cohen, 1992), which is directly related to the statistical signi¢cance of the di¡erence, quantitatively expressed by the Student’s t-test probability. Standard TEOAE parameters, such as global and band reproducibility, response level and SNR were used in the present work. As these three parameters are strictly correlated to each other, very similar results are obtained using any of them, and thus only results for the SNR are presented. Another parameter was also used: the TEOAE spectral latency, i.e., the time delay between the click stimulus and the time of maximum response in a given frequency band, which has also recently been shown to be correlated to audiometrically determined hearing loss (Sisto and Moleti, 2002).
2. Materials and methods In this work audiometric and TEOAE linear and nonlinear data of a population of 134 ears from 67 young (age = 18^25 yr) male subjects were studied. Except for two subjects for which the linear TEOAE data are missing, the population is the same as that analyzed in a previous study (Lucertini et al., 2002), in which only the nonlinear data were reported. More details about the examined population and the methodology can be found there. Pure tone audiograms were recorded for all subjects in an acoustically shielded room, using a clinical audiometer (Amplaid A-460) equipped with standard headphones (TDM-39), according to the procedure described by Yantis and Katz (1994). The audiometric test frequencies were 0.25, 0.5, 1, 2, 3, 4, 6, and 8 kHz. As in Lucertini et al. (2002), a standard dichotomous clinical de¢nition of hearing impairment was used. Speci¢cally, the examined ear was de¢ned as ‘normal’ if no absolute threshold level greater than 20 dB was measured over the whole frequency range. If a threshold greater than 20 dB was observed in any audiometric range f v 3 kHz, the ear was de¢ned as ‘high-frequency impaired’. Subjects were also divided, on the basis of a preliminary interview, into two subsets: non-noise-exposed (n = 46) and noise-exposed (n = 21). Nonexposed sub-
HEARES 4021 30-10-02
292
A. Moleti et al. / Hearing Research 174 (2002) 290^295
jects stated they had never been exposed to either recreational noise or to other ototoxic agents. As all subjects live in urban areas of Italy, this negative statement means they had actually been exposed to the standard urban noise level, which is an unavoidable exposure background for all subjects considered in the present study. However, the audiometric test showed that none of them had hearing loss. Exposed subjects had been exposed for 1 yr without ear protection to ¢rearm noise with variable exposure durations. They were a¡ected by unilateral high-frequency hearing loss, which is compatible with the exposure to impulsive noise. None of them had pre-existing hearing loss, and none of them reported exposure to other ototoxic agents, thus hearing loss was presumably related to the training with ¢rearms. According to the results of the audiometric test, three classes of ears were identi¢ed: (1) ears of nonexposed bilaterally normal subjects (92 ears from 46 subjects), (2) audiometrically normal ears of exposed subjects exhibiting unilateral high-frequency hearing loss in the contralateral ear (21 ears from 21 subjects), and (3) the impaired ears of the same subjects (21 ears from 21 subjects). TEOAE recordings were obtained using the ILO-96 system in the nonlinear mode of acquisition, with a stimulus level of 80 N 3 dB peSPL. The standard ILO window, which is a trapezoidal window from 2.5 to 20 ms with 2.5 ms rise time, was applied to the nonlinear data. Synchronized SOAEs (SSOAEs) from the same subjects were recorded in the linear mode, using the same stimulus level (within N 1 dB). SSOAE recordings yield information about the long-lasting components of the TEOAE response (SOAEs) that was not used in the present work. Only the ¢rst 20 ms of the 80-ms SSOAE linear recordings were used, to allow comparison between linear and nonlinear responses. Three di¡erent windows were applied to the linear recordings: the standard ILO window, an ILO window shifted forward in time by 2.5 ms (i.e., with onset and £at-top at 5 ms and 7.5 ms after the click, respectively), and an ILO window shifted forward in time by 5 ms (onset at 7.5 ms, £attop at 10 ms). These three di¡erent temporal windows
were applied o¡-line to the linear data, to analyze the e¡ect of windowing on data quality, which is a critical issue when using linear responses due to the presence of the linear ringing artifact. Both the TEOAE and SSOAE recordings were processed by using customized software developed in LabVIEW 5.0 (National Instruments). Global and thirdoctave band reproducibility, response and SNR were evaluated using the de¢nitions and the method described in Lucertini et al. (2002). Spectral latency was measured by means of a wavelet transform technique based on the iterative application of ¢lter banks to the TEOAE waveform, as described in Sisto and Moleti (2002). The statistical distributions of the examined TEOAE parameters were analyzed, and the statistical signi¢cance of the di¡erences between the three classes were evaluated according to the Student’s t-test. A 95% con¢dence level (corresponding to a probability p 6 0.05) was arbitrarily chosen as the criterion for statistical signi¢cance.
3. Results As pointed out in a previous work (Lucertini et al., 2002), reproducibility, response and SNR are strictly correlated quantities because the noise level is quite independent of the response level and reproducibility is just a di¡erent de¢nition of the SNR. Thus very similar results are obtained using any of these parameters, and in the following, for simplicity, only the results for the SNR will be shown. As mentioned in the Introduction, the main result of Lucertini et al. (2002) was the observation of a statistically signi¢cant di¡erence not only between normal and impaired ears, but also between nonexposed and exposed (either normal or impaired) ears. The present work shows that nonlinear and linear data are both capable of showing this result, with di¡erent e¡ectiveness. In Table 1, the average global SNR of the ears of the three classes are shown, as obtained from the nonlinear analysis, compared to those obtained by applying
Table 1 Average values of the SNR for the three classes of ears (1 = nonexposed, normal hearing, 2 = exposed, normal hearing, 3 = exposed, hearing impaired), average SNR di¡erence (vi3j ), Cohen’s e¡ect size (ESi3j ), and probability associated to the Student’s t-test computed between pairs of the classes (pi3j ) Acquisition mode/window
SNR1
SNR2
SNR3
v132
v133
ES132
ES133
p132
p133
Nonlinear/ILO Linear/ILO Linear/ILO+2.5 ms Linear/ILO+5 ms
12.7 18.1 9.3 6.2
9.0 16.4 6.8 4.0
8.1 15.7 5.2 2.5
3.7 1.7 2.5 2.2
4.6 2.4 4.1 3.7
0.75 0.69 0.43 0.34
0.98 0.97 0.75 0.58
0.018 0.014 0.12 (ns) 0.20 (ns)
0.0005 0.002 0.004 0.017
The linear data were windowed with three di¡erent windows, the standard ILO (a trapezoidal window from 2.5 to 20 ms with a 2.5 ms rise time), and two equally shaped windows, increasingly shifted forward in time, by steps of 2.5 ms.
HEARES 4021 30-10-02
A. Moleti et al. / Hearing Research 174 (2002) 290^295
Fig. 1. Di¡erence of the average SNR in third-octave frequency bands between nonexposed normal hearing ears (ne) and exposed normal hearing ears (e) (classes 1 and 2), obtained using the nonlinear data (a), and statistical signi¢cance of the di¡erence, as expressed by the probability of the Student’s t-test parameter (b). The 5% probability level assumed as the criterion for signi¢cance is shown by the horizontal solid line.
three di¡erent temporal windows to the linear data. Pushing forward in time the window caused a progressive degradation of the data due to the drastic elimination of the short-latency high-frequency response. In the following, the emphasis is put on the di¡erence between classes 1 and 2, because the early detection of exposure e¡ects in audiometrically normal subjects could be the most promising application of this study. As mentioned above, statistically signi¢cant di¡erences between classes 1 and 2 can be evidenced by using global TEOAE parameters, but more signi¢cant di¡erences can be found using the corresponding spectral band parameters. In Fig. 1a the di¡erence of the average SNR in third-octave frequency bands between classes 1 and 2 obtained using the nonlinear data is shown. The horizontal line at 3.7 dB is the di¡erence between the average global SNR of the two classes. It is clear that most of this di¡erence comes from the highfrequency bands (f s 2 kHz). In Fig. 1b the statistical signi¢cance of the di¡erence is expressed by the probability of the Student’s t-test parameter. The low-frequency part of the signal adds a non-signi¢cantly di¡erent contribution to the overall signal, which tends to lower the signi¢cance of the global SNR di¡erence between classes 1 and 2. With respect to the di¡erent
293
discrimination capability of global and band parameters, very similar results are obtained with the linear data. In Fig. 2a the di¡erence of the average SNR in thirdoctave frequency bands between classes 1 and 2 is shown for the nonlinear and the linear data, using three di¡erent delay times for the window applied to the linear data. The result shown in Table 1 for the global SNR is found again and clari¢ed by the spectral analysis. Most of the SNR di¡erence comes from the highfrequency bands, also for the linear data, and this difference tends to decrease with increasing start time of the window due to the elimination of part of the shortlatency high-frequency response. This result may be interpreted in terms of a loss of discrimination capability between exposed and nonexposed ears, as shown in Fig. 2b, where the statistical signi¢cance of the di¡erence is expressed by the probability of the Student’s t-test parameter. As shown in Table 1, the di¡erence between nonexposed normal ears and exposed impaired ears (classes 1 and 3) was always statistically signi¢cant, considering either global or spectral band SNR (above 1 kHz), for both acquisition modes and for all the windows applied to the linear data. Also in this case, the statistical sig-
Fig. 2. Di¡erence of the average SNR in third-octave frequency bands between nonexposed normal hearing ears (ne) and exposed normal hearing ears (e) (classes 1 and 2), as obtained using the nonlinear and the linear data, for three di¡erent choices of the window applied to the linear data (a) and statistical signi¢cance of the difference (b).
HEARES 4021 30-10-02
294
A. Moleti et al. / Hearing Research 174 (2002) 290^295
ni¢cance of the di¡erence decreased, for the linear data, with increasing time delay of the window. The di¡erence between classes 2 (exposed normal ears) and 3 (exposed impaired ears) was never statistically signi¢cant, also due to the smaller size. It should be noted here that e¡ect sizes of 0.2, 0.5 and 0.8 are interpreted by Cohen (1992) as small, medium and large, respectively. With respect to spectral latency, a statistically signi¢cant di¡erence between classes 1 and 2 had been found in a previous work (Lucertini et al., 2002) using nonlinear data only. In the present work it was found that the linear data cannot be used for measuring the latency with the proposed algorithm because no reasonable compromise can be found between the con£icting necessities of e¡ectively cancelling the linear artifact contribution while maintaining a su⁄cient sensitivity to the short-latency high-frequency ( s 3 kHz) response.
4. Discussion TEOAE data, either recorded in the linear or in the nonlinear mode, have shown good discrimination capability between normal and impaired (and also between normal exposed and nonexposed) ears. Thus the results of a previous study (Lucertini et al., 2002) are extended to the case of TEOAE data acquired using the linear paradigm. The presence of the linear ringing artifact, which typically a¡ects the linear recordings over the ¢rst 5^6 ms, requires proper windowing of the linear data. On the other hand, pushing the onset of the analysis window forward in time caused a decrease in the discrimination capability of the TEOAE due to the fact that most of the high-frequency signal, whose latency is shorter than 5 ms, is canceled by windows starting at times longer than 5 ms. This fact is particularly important because most of the di¡erence between normal and impaired (and between exposed and nonexposed) subjects comes indeed from the high-frequency bands. It is clear that the linear acquisition mode provides a SNR that, for the same values of all other parameters, is 6 dB higher than that given by the nonlinear mode (Tognola et al., 2001). This factor of two is due to the coherent sum of all the acquired signals in the linear mode, while in the nonlinear mode only two out of four acquired signals actually contribute to the output. Despite this well-known advantage of the linear technique, it should be stressed that the inclusion of a large linear artifact contribution has the e¡ect of lowering the difference between di¡erent classes of ears, particularly when using global TEOAE parameters. This fact may be observed in the global SNR data of Table 1 : the di¡erence between classes 1 and 2 as well as that be-
tween classes 1 and 3 is much smaller using the linear data and the standard ILO window. In fact, adding to the signal of all subjects a large spurious and deterministic component has the e¡ect of lowering the value (expressed in dB) of the SNR (or response level) di¡erence. Therefore, the comparison between the e¡ectiveness of the two acquisition modes must be done keeping into account not only the SNR level that can be obtained in given experimental conditions but also the degree of contamination by noncochlear (thus uncorrelated to cochlear functionality) contributions. Signi¢cant cochlear signal variations become less evident if accompanied by a spurious noncochlear signal. It is interesting how an added noncochlear signal and added noise di¡erently a¡ect the statistical signi¢cance of the di¡erence between the OAE response of two populations expected to have di¡erent cochlear functionality. An added spurious deterministic signal has the e¡ect of lowering the di¡erences of cochlear origin in the overall signal. Added spurious noise has instead the e¡ect of increasing the variance of the data, thus lowering the statistical signi¢cance of a given di¡erence.
5. Conclusions The results presented in this work con¢rm the idea that TEOAEs parameters, measured from either nonlinear and linear data, can be used to successfully discriminate a population of normal ears from one of impaired ears and also a population of exposed ears from one of nonexposed ears. The comparison between the e¡ectiveness of the nonlinear and linear averaging paradigms has shown that both acquisition paradigms are e¡ective if the window applied to the linear data to eliminate the artifact does not start later than 2.5 ms after the click. Longer window start times eliminate too much of the short-latency high-frequency response, from which most of the di¡erence between impaired and normal or between exposed and nonexposed ears arises. On the other hand, the addition of a large noncochlear artifact has been shown to decrease the response di¡erence between classes and its statistical signi¢cance. Thus the nonlinear mode seems to be preferable in the general case, also because it makes it possible to e¡ectively use another independent indicator of cochlear function, i.e., the TEOAE spectral latency. References Attias, J., Furst, M., Furman, V., Reshef, I., Horowitz, G., Breslo¡, I., 1995. Noise-induced otoacoustic emission loss with or without hearing loss. Ear Hear. 16, 612^618. Cohen, J., 1992. A power primer. Psychol. Bull. 112, 155^159.
HEARES 4021 30-10-02
A. Moleti et al. / Hearing Research 174 (2002) 290^295 Furst, M., Reshef, I., Attias, J., 1992. Manifestations of intense noise stimulation on spontaneous otoacoustic emission and threshold microstructure: experiment and model. J. Acoust. Soc. Am. 91, 1003^1014. Hall, A.J., Lutman, M.E., 1999. Methods for early identi¢cation of noise-induced hearing loss. Audiology 38, 277^280. Hussain, D.M., Gorga, M.P., Neely, S.T., Keefe, D.H., Peters, J., 1998. Transient evoked otoacoustic emissions in patients with normal hearing and in patients with hearing loss. Ear Hear. 19, 434^ 449. Kowalska, S., Sulkowski, W., 1997. Measurements of click-evoked otoacoustic emissions in industrial workers with noise-induced hearing loss. Int. J. Occup. Med. Environ. Health 10, 441^459. Lucertini, M., Moleti, A., Sisto, R., 2002. On the detection of early cochlear damage by otoacoustic emission analysis. J. Acoust. Soc. Am. 111, 972^978. Prieve, B.A., Gorga, M.P., Schmidt, A., Neely, S., Peters, J., Schultes,
295
L., Jesteadt, W., 1993. Analysis of transient-evoked otoacoustic emissions in normal-hearing and hearing-impaired ears. J. Acoust. Soc. Am. 93, 3308^3319. Sisto, R., Moleti, A., 2002. On the frequency dependence of the otoacoustic emission latency in hypoacoustic and normal ears. J. Acoust. Soc. Am. 111, 297^308. Sisto, R., Moleti, A., Lucertini, M., 2001. Spontaneous otoacoustic emissions and relaxation dynamics of long decay time OAEs in audiometrically normal and impaired subjects. J. Acoust. Soc. Am. 109, 638^647. Tognola, G., Ravazzani, P., Molini, E., Ricci, G., Alunni, N., Parazzini, M., Grandori, F., 2001. ‘Linear’ and ‘derived’ otoacoustic emissions in newborns: a comparative study. Ear Hear. 22, 182^ 190. Yantis, P.A., Katz, J., 1994. Puretone air-conduction threshold testing. In: Handbook of Clinical Audiology, 4th edn. Williamson and Wilkins, Baltimore, pp. 97^108.
HEARES 4021 30-10-02