|
| |||
|
Long-term Reproducibility of the TEOAE time-frequency (TF) distributions
INTRODUCTIONThe cochlea emits sounds (otoacoustic emissions) of a very low intensity, which can be recorded and separated from the acoustic stimulus and acoustic background of the surrounding environment by using a very sensitive microphone. Characteristics of the otoacoustic emission, which help to identify them from other sounds are: non-linearity of the relationship between the stimulation and the receptor’s response, and in case of the transient emissions (TEOAEs) the estimated delay of the response associated with the course of travelling wave. Despite the fact that the recording of otoacoustic emissions does not reproduce an audiogram, it helps to assess very precisely changes in the cochlea, which otherwise cannot be observed in the audiogram. This characteristic has been used as a sensitive way of monitoring changes in the inner ear caused by exposure to noise or administration of ototoxic drugs. However, it should be kept in mind that otoemissions reflect functions of peripheral part of the hearing organ and generally are independent from functioning of the central part of the hearing tract. The analysis of the transiently evoked otoacoustic emission (TEOAE) signal is usually performed with the Fourier transform (FT). Because the TEOAE signal is non-stationary and its spectrum is changing over time i.e. the amplitude as well as the number of oscillations per time unit are changing. This method can be less useful, mainly due to a considerable short time of the TEOAE signal (about 20 ms), which cannot be divided into quasi-stationary segments. The objective of the traditional analysis with FT is to examine the spectrum of whole TEOAE course, while the range of spectrum frequency is being divided into bands. In the time-frequency analysis the time range can be divided into segments and the spectrum can be examined in shorter parts – it is the so called short term Fourier transform (STFT). In a short segment of the analysed signal by means of STFT, the resolution in time is better, but the resolution in frequency is worse, and conversely, by lengthening the analysed segment the resolution in time is deteriorating. This phenomenon is called time and frequency broadening[1]. To partially resolve the time AND frequency resolution conflict, several signal processing approaches are available. It is possible to use the Wavelet transform (WT) which can adjust the resolution in frequency of the analysed frequency band [2,3]. This technique allows us to achieve a time-frequency distribution of constant but relative frequency broadening. For low frequencies, the lower time-resolution and the more precise frequency-resolution are used and the inverse is applied to higher frequencies (ie higher time resolution and lower frequency resolution). The reader might comprehend better the latter statement by considering that a shorter time of observation is needed to evaluate higher frequencies, as the signal oscillations are faster. Another technique to adjust the frequency resolution in the frequency band of interest, is the Wigner-Ville time-frequency distribution with regional smoothing [4]. For multi-component signals (as TEOAEs) the representation on the time-frequency plane is highly disturbed by the so-called cross-terms. The cross-terms are by-products of the mutual interplay of single TEOAE components and appear as false, highly fluctuating artefacts overlaid on the time-frequency representation (spectrogram) . This causes the spectrogram to have local negative values which increase the difficulties of a coherent TF interpretation. The cross-terms of the Wigner’s transform have a number of characteristic properties, which can be used to identify and suppress these components. The first feature is that the distribution takes negative values in the places where cross-terms appear. The other characteristic is the high variability of cross-terms, both in time and frequency domain. The method of regional smoothing of discrete Wigner’s distribution, proposed in this work, consists in applying a special technique of filtering in the TF distribution. The proposed filter has a low-pass characteristic, and it is applied only in the places where the cross-terms are to be suppressed. One can locate the cross-terms knowing that they are confined to the places, where negative values of power appear in the distribution. In these areas, and in their vicinity, one applies smoothing. Due to this procedure, extreme negative values are decreased, and the sum of “negative powers” in the spectrum image is reduced at the expense of positive values located at the outskirts of the positive-value areas of the spectrum. This way, one obtains a TF representation whose physical interpretation is more coherent. On the other hand, the effect of “blurring” of spectrum structure is minimal due to the fact that smoothing is applied locally, only in the places where it is necessary. The objective of this study was to analyse via the TF approach TEOAE responses and compare the corresponding spectrograms taken in the time interval of over one year.
MATERIAL AND METHODS
A total of 152 young men (304 ears) aged 18-19 years (mean 18.5 years) without otolaryngologic problems participated in the study. Ear examinations were performed twice in the time interval of one year, using audiological examination, impedance audiometry, and click evoked otoacoustic emissions (TEOAE). TEOAEs were recorded with a Otodynamics ILO 292 Echoport, 5.0 version. The intensity of sound stimulus ranged from 75 to 82 dB SPL at 50 reps / s. Responses were averaged following 260 repetitions and the time of the analysis ranged from 2.5 to 20 ms. The level of artefacts was set to 4.6 mPa .i.e. 47.3 dB SPL. The level of TEOAEs was measured in the range of 0.5 kHz from 0.5 to 5 kHz. A signal to noise ratio (S/N) of at least 3 dB was considered as an indication of a TEOAE component present at the tested band. The evaluation of the TEOAE signal magnitude was based on the S/N for frequency bands of 1, 2, 3, 4 and 5 kHz, separately for right and left ears. For the time –frequency analysis (TF), a grey colour scale on TF spectrograms has been standardised in such a way that the maximum of the distribution corresponds to maximum black colour.
RESULTS
Tympanometric measurements of the subjects revealed tympanograms of type A before and after the one-year period. Stapedius muscle reflex thresholds were recorded at stimulation at the level of 75-85 dB above the hearing threshold. On the basis of the performed TF analysis, a very high individual similarity of the initial and final (after 12 months) spectrograms was observed . The data were classified in three groups: A group of high pre-post spectrogram similarity composed from the responses of 274 ears (90%); A group of low pre-post spectrogram similarity composed from the responses of 24 ears; and A group of zero pre-post spectrogram similarity composed from the responses of 6 ears (2%). Examples of these three categories comparison of TF analysis spectrograms are presented in Fig. 1, 2, and 3 respectively.
Before one year (max.9.91dB) After one year (max.6.85 dB)
Figure. 1: High spectrogram similarity .
In Figure 1, a particular spectrogram similarity has been observed within the interval of 12 ms, where a star-shaped object (in the TF panel under the TEOAE response) can be noticed with a centre for frequency slightly above 3 kHz and a few crossing lines for low frequency with the dominating component in the range of 1.22-1.4 kHz. The time courses (structure of the TEOAE response) shown above the TF distribution panel also present a slight similarity, however an accordance is not so evident. Changes in the morphology of the TEOAEs , which can be observed after one year, are probably associated with the proportions of intensity of particular TEOAE components and not with the configuration of the same components. Thus, only the TF analysis can provide information on the configuration of the TEOAE components in time and frequency co-ordinates even for recordings acquired within long time period.
Before one year (max.12.57dB) After one year (max.9.99dB)
Figure 2: Low spectrogram similarity Figure 2 shows that for the same subject of Figure 1 the spectrogram similarity of the right ear is smaller. This might have caused from a high level of noise during the second examination, especially at the final stage of the recording. Still there is no doubt as to similarity of both spectrograms. It is worth mentioning that spectrograms obtained from the recordings for the left and right ear differ considerably in the same subject.
Before e one year (max.-3.03dB) After one year (max-1.2dB)
Figure 3. Zero Spectrogram similarity Figure 3 shows a typical example of the recorded TF image, where OAE level was minimal and hidden in the noise. The absence of areas and lines with dominating intensities was observed. A similar picture-spectrogram has been achieved in each case of weak otoacoustic emission, therefore, individual identification is not possible.
DISCUSSION
Methods of OAE signal analysis based on the Fourier Transform do not considered the time changeability of the emission signal. The TF analysis according to Wigner-Ville method reveals in a complete way, the time changing spectral structure of the otoacoustic emission signal and the relationships between TEOAE frequency and signal latency. In the present work the basic TF elements we have considered were constant spectrograms components, i.e. oblique and horizontal lines and their location on the time and frequency scales. On the basis of previous comparisons in the authors own investigations on SOAE frequency spectrum with spectrograms of time-frequency distribution [5], signals of constant frequency and changeable amplitude were found to be one of the TEOAE component types visible in TF spectrogram. Due to unchangeability of frequency it is likely that these components are associated with the phenomenon of a spontaneous emission. Comparison of the TF analysis spectrograms was carried out due to some questionable issues in certain cases as regards matching the file and the person or side studied. It appeared that in spite of long time interval that elapsed between two examinations there has been a great similarity of spectrograms, which enable to identify the subject and the examined ear . After establishing the usefulness of this method, spectrograms of all individuals from two examinations performed in the time interval of over one year were compared and their similarity was determined in 90% of cases. It seems that individual similarity confirmed in a long time interval and interpersonal differences of TEOAE signal spectrum visible on spectrograms of the TF analysis may suggest individual distribution and morphology of outer hearing cells, as well as individual anatomical conditions of acoustic signal emitted to the external auditory meatus. In our opinion spectrograms similarity, besides application in medicine, may be used in biometric techniques as a method of individual identification in the form of so called ear-print. According to the data of other studies [6] criteria and characteristics of ideal biometric system facilitating, with a great deal of probability, identification and verification of an individual person should be common for the whole population, and each type of biometric features should be different from the others. We think that the features of the TEOAE time-frequency analysis spectrograms studied meet these criteria.
CONCLUSIONS
REFERENCES [1]
Grzanka A., Hatzopoulos S.: Review of Time-Frequency
Distributions in Applications to Otoacoustic Emissions( in Polish).
Audiofonologia XIII 1998, s. 41-53.
|
 |
 |