USING THE AUDITORY STEADY-STATE RESPONSE TO DIAGNOSE DEAD REGIONS IN THE COCHLEA. A thesis submitted to The University of Manchester for the degree of

Transcription

1 USING THE AUDITORY STEADY-STATE RESPONSE TO DIAGNOSE DEAD REGIONS IN THE COCHLEA A thesis submitted to The University of Manchester for the degree of Doctor of Philosophy Faculty of Medical and Human Sciences 2011 TIMOTHY STEVEN WILDING SCHOOL OF PSYCHOLOGICAL SCIENCES

2 Table of Contents List of Tables... 5 List of Figures... 8 Abstract Declaration Copyright Acknowledgements Abbreviations Description of thesis format Chapter 1. Introduction Literature Review Dead Regions Current Dead Region diagnosis methods Comparison of DR diagnosis methods Perceptual consequences of Dead Regions The clinical importance of DR diagnosis DR diagnosis in infants Consideration of factors affecting ASSR recording Aims and objectives Objectives Experiments Chapter 2. General Methods RAC Stimuli Procedures Data Analysis PTCs Method of recruitment Signal checks Transducer generated combination tones Chapter 3. An investigation of the optimal recording parameters of an auditory steady-state response amplitude curve method Introduction Methods Participants Stimuli Protocol Data Analysis RAC Tip Determination Experiment One: Investigating recording parameters Results Discussion of Experiment-One results Experiment two: investigating the test time, response stability and repeatability of estimated tip frequencies Results Test Time

3 3.4.3 Discussion of Experiment-Two results Conclusions Chapter 4. Using the auditory steady-state response to record response amplitude curves. A possible fast objective method for diagnosing dead regions Introduction Methods Participants Stimuli Procedures Data Analysis Results Discussion Repeatability Tip frequency shift Effect of EEG noise Method advantages Conclusions Chapter 5. Using the auditory steady-state response to record response amplitude curves in hearing-impaired adults Introduction Methods Subjects Procedures PTC and RAC curve fitting Results PTCs Response Amplitude Curves Discussion Psychophysical Tuning Curves RACs Conclusions Chapter 6. Auditory Steady-State Responses in Normally Hearing and Hearing-impaired Adults: An analysis of between-session amplitude and latency repeatability, test time and F-Ratio detection paradigms Introduction Methods Subjects Stimuli Procedures Data Analysis Results False Response Detection Detection time Repeatability of response Discussion Effect of EEG-noise floor on ASSR Response detection criteria and test time Response Amplitude and Latency Repeatability Conclusions

4 Chapter 7. General Discussion and Future Directions Summary and discussion Future direction of the work Chapter 8. Appendices Chapter 3 Appendices CD technical detail document contents Chapter 4 Appendices Chapter 6 Appendices References Word Count:

5 List of Tables Table 2.1 Example of curve fitting parameters of RACs shown in Figure 2.3. Two ROEX fitting methods are displayed for each RAC Table 2.2 Combination tone audibility level calculation. The audibility level represents the signal level for which the combination tone could reach audibility for a normally hearing subject with a 0 db HL hearing threshold at all frequencies. Reference zero shows the db HL to db SPL conversion factor at each frequency (i.e. the 0 db HL level) Table 3.1 RAC recording test parameters for experiment one. Test numbers in parenthesis indicate test is a repetition of the indicated previous test Table 3.2. RAC recording modulation frequencies in experiment two Table 3.3 RAC for subjects 1-5. Test runs where the minimum amplitudes exceeded the SRA are shown in BOLD Table 3.4. Response amplitudes and estimated RAC-tip deviation from 2 khz by test type in experiment one. Average noise corrected amplitudes shown are the mean response amplitudes across the 16-second sweep Table 3.5 Not-masked response amplitudes in experiment two. Amplitudes shown are noise corrected (nv). The highest response amplitude for each subject in each test session is shown in BOLD Table 3.6 RAC-tip frequencies recorded in experiment two, to AMEXP stimuli presented at 50 db SL with a signal-to-masker ratio of 0 db Table 3.7. RAC-tip frequencies recorded in experiment two, to AMEXP stimuli presented at 50 db SL with a signal-to-masker ratio of 10 db Table 4.1 Example of curve fitting parameters of RACs shown in Figure 4.4. Two ROEX fitting methods are displayed for each RAC Table 4.2 EEG-noise levels for swept- and fixed-method RAC recordings for two test runs. Bold values indicate recordings where the recorded EEG noise was below the recommended 10nV level (Stevens et al. 2009) Table 5.1 PTC results for two recordings of each subject in different test sessions (except subject S2 only one recording and subject S6 two recordings same session). * Dynamic range of PTC less than 10 db Table 5.2 RAC recordings with 0 db SMR. ROEX RAC tip frequencies and fitted curve R-squares for run1 (R1) and run2 (R2). RACs shown in Figure 5.5 and Figure 5.6 (* correlation p<0.05 ** correlation p<0.01)

6 Table 5.3 RAC recordings with 0 db SMR (as shown in Table 2). Mean overall noise level for 120 sweeps. Mean and SD amplitudes analysed over 15 sweep slices representing the stability of the overall response during the recording time Table 5.4 RAC recordings at the indicated signal level and signal to masker ratio (SMR). Signal level in db above subject behavioural threshold for the ASSR test stimuli. ROEX RAC tip frequencies and fitted curve R-squares. Mean and SD amplitudes analysed over 15 sweep slices representing the stability of the overall response during the recording time Appendix Table 1 Estimated RAC-tip frequencies and tip difference between test runs for swept-method RACs. Bold values indicate RACs where the ASSR response amplitude was significantly above the EEG noise at every individual masker frequency analysis point across the swept masker frequency range (1 khz to 4 khz). * RMS difference for 1000, 1500, 1800, 2000, 2200, 2500, 3000 and 3500 Hz masker frequency points. ** Correlation coefficient p< Appendix Table 2 Estimated RAC-tip frequencies and tip difference between test runs for fixed-method RACs. Bold values indicate RACs where the ASSR response amplitude was significantly above the EEG noise at all masker frequencies (1000, 1500, 1800, 2000, 2200, 2500, 3000 and 3500 Hz). ** Correlation coefficient p< Appendix Table 3 Swept method curve fitting parameters. Bold values indicate RACs where the ASSR response amplitude was significantly above the EEG noise at every individual masker frequency analysis point across the swept-masker frequency range (1000 Hz to 4000 Hz) Appendix Table 4 Fixed-method curve fitting parameters. Bold values indicate RACs where the ASSR response amplitude was significantly above the EEG noise at all masker frequencies (1000, 1500, 1800, 2000, 2200, 2500, 3000 and 3500 Hz) Appendix Table 5 ASSR recording analysis for normally hearing subjects (N), mild hearing loss subjects (M) and severe hearing loss subjects (S) in session 1 and session 2. F-ratios marked in bold indicate cases where the response amplitude was not significantly above the EEG-noise level (2% level). In subject S2 the 2 nd recording reached significance after 3 sweeps but was not significant after 15 sweeps Appendix Table 6 Pairwise McNemar binomial test p values for variable test time response detection paradigms. p < marked in bold

7 Appendix Table 7 ASSR recording analysis for normally hearing subjects (N), mild hearing loss subjects (M) and severe hearing loss subjects (S) in session 1 and session 2. Four stop when significant detection protocols were applied. Conditions one and two: 1% and 5% indicate the number of sweeps for the recording to reach the required significance level. Condition three 5% > 9 sweeps, indicates a minimum of 9 sweeps. Condition four: 5% level >4 consecutive sweeps indicates significance remains as 4 consecutive sweeps are added to the analysis Appendix Table 8 Mean response detection times and number of false positive response detections for 60 recordings. False response detections analysed at 87 Hz, true responses at 95 Hz

8 List of Figures Figure 1.1 Spectral components of the 95-Hz AM, FM, AMEXP and MM modulated ASSR stimuli. X-axis shows the FFT frequency bin and y-axis the amplitude. All signals were RMS equalized to a RMS pure tone represented by a digital sample RMS value of 10. Y-axis represents unitless digital sample amplitude Figure 2.1 Schematic illustration of swept method averaging process. Upper panel shows the signal (grey line) and masker (black line) centre frequency plotted against test time in each sweep repetition. Lower panel represents the resultant RAC. Dashed arrows show the averaging process of the points of the EEG-waveform obtained for the upward and downward sweep direction Figure 2.2 Illustration of modulation envelope phase difference between the averaged upward and downward sweep direction analysis. Plots represent the 95-Hz modulation envelope. X-axis shows the time position in the sweep relative to the start of the sweep (Left panel) and end of the sweep (Right panel). Vertical lines show the first and second analysis points which were 0 ms and 20 ms on the left panel, 0 and -20 ms on the right panel Figure 2.3 Examples of ROEX curve fitting to RAC data. Left panel, example ROEX curve fitting with r parameters constrained (solid line) and unconstrained (dashed line). Right panel, example ROEX curve fitting with r parameters (r upper and r lower ) separate for frequencies above and below the fitted tip (solid line) and with one single r parameter for all masker frequencies (dashed line). ROEX parameters as shown in Table Figure 2.4 An example of PTC ROEX fitting (Subject M5 reverse masker frequency direction first test run) and Q 10dB calculation. Lighter line represents the turning points of the masked threshold tracking of the automated fast-ptc method. Dashed line represents the ROEX curve fit of the turning points. Darker sold line shows the bandwidth (Q 10dB ) Figure 2.5 Snapshot of spectral content of 95 db SPL ASSR signal at 2 khz and masker at 1.5 khz. Signal recording was performed using a GRAS type 26AC.25 inch microphone connected to Agilent 35670A Dynamic Signal Analyzer in octave analysis mode via a GRAS IEC711 coupler. Square marker represents the level and frequency of the ASSR stimulus, diamond marker the narrow band masker and the combination tone is represented by the cross

9 Figure 3.1 Spectral components of the 95-Hz AM, FM, AMEXP and MM modulated ASSR stimuli. X-axis shows the FFT frequency bin and y-axis the amplitude. All signals were RMS equalized to a RMS pure tone represented by a digital sample RMS value of 10. Y-axis represents unitless digital sample amplitude Figure 3.2 RAC recordings for experiment one. X-axis represents the masker centre frequency. Y-axis represents the response amplitudes relative to the SRA. The test number for each RAC is shown in the figure legends Figure 3.3 RAC recordings of experiment two. Axes as for Figure 3.2. Lighter lines represent the RAC recorded in session one, darker lines session two. Left and right panels show RAC recorded at a SMR of 0 db and 10 db respectively Figure 3.4 RAC-tip frequency against test time in subjects 6-9 of experiment two. Two recordings for each SMR condition in each subject are presented. The x-axis represents the number of sweeps in the analysis. The y-axis represents the frequency in khz. Darker lines show the RAC minimum position (tip), lighter line the ROEX-fitted curve minimum position (fitted tip) Figure 3.5 Minimum response amplitude, background-eeg noise and mean ROEX RAC-curve fit residuals against time in subjects 6-9 of experiment two. Two recordings for each SMR condition in each subject are presented. The x- axis represents the number of sweeps in the analysis. The y-axis represents the absolute amplitude in nv. Dashed lines show the SRA, darker line the RMS ROEX residual and lighter line the response amplitude at minimum position (tip) Figure 3.6 The stability of mean response amplitude across the sweep in 15 sweep slices against start position in the sweep during RAC recordings for subjects 6-9 in experiment two. The x-axis shows the start position of each 15 sweep slice, the y-axis the mean response amplitude in nv. The darker line represents the recording in session two, the lighter line session one. The grand mean (and standard deviation) of the individual 15 sweep slice amplitudes are displayed in each figure panel Figure 4.1 Spectral content of 2-kHz 95-Hz exponentially modulated ASSR signal. Y-axis shown in digital sample units, signal generated with overall digital amplitude of 10 units Figure 4.2 Schematic illustration of swept method averaging process. Upper panel shows the signal (grey line) and masker (black line) centre frequency plotted against test time in each sweep repetition. Lower panel represents the resultant RAC. Dashed arrows show the averaging process of the points of the EEG-waveform obtained for the upward and downward sweep direction

10 Figure 4.3 Illustration of modulation envelope phase difference between the averaged upward and downward sweep direction analysis. Plots represent the 95-Hz modulation envelope. X-axis shows the time position in the sweep relative to the start of the sweep (Left panel) and end of the sweep (Right panel). Vertical lines show the first and second analysis points which were 0 ms and 20 ms on the left panel, 0 and -20 ms on the right panel Figure 4.4 Examples of ROEX curve fitting to RAC data. Left panel, example ROEX curve fitting with r parameters constrained (solid line) and unconstrained (dashed line). Right panel, example ROEX curve fitting with r parameters (r upper and r lower ) separate for frequencies above and below the fitted tip (solid line) and with one single r parameter for all masker frequencies (dashed line). ROEX parameters as shown in Table Figure 4.5 Individual RACs recorded for the fixed and swept method for each subject in two sessions. Solid line shows the first test run recorded using the swept masking method. The filled circles represent the first test run recorded using the fixed masking method. The dashed line and unfilled circles are the second test run swept and fixed test method, respectively. Y-axis shows the response amplitude as a percentage of the not-masked amplitude recorded in the same test run as each RAC Figure 4.6 Grand mean RACs recorded in 20 subjects from two recording sessions for the fixed (left panel) and swept (right panel) masking method. Solid line is the mean relative amplitude, dashed line shows the ROEX curve fit of the mean. The mean amplitude ±1 SD is shown as the upper and lower dotted lines in the right panel and as error bars in the left panel Figure 4.7 Example of RACs recorded using the swept masker method. Solid lighter line shows the unmasked response amplitude recorded in the same test session as the RAC. Dashed lighter line shows the amplitude below which the response was not considered significantly higher than the background EEGnoise. Darker solid curve shows the response amplitudes, darker dashed line shows the ROEX curve fit to the response amplitudes Figure 4.8 Example RACs recorded using the fixed masker method. Solid lighter line shows the unmasked response amplitude recorded in the same test session as the RAC. Dashed lighter line with data points marked as X shows the amplitude below which the response was not considered significantly higher than the background EEG-noise. Filled circular data points show the response amplitudes for each masker frequency, darker dashed line shows the ROEX curve fit to the response amplitudes Figure 4.9 Fixed and swept method RAC-tip estimation recording repeatability. Data points show the mean estimated RAC-tip frequency (xaxis) vs. the difference in tip frequency between two test runs (y-axis). The upper and lower dashed lines show the assumed mean difference ± the repeatability coefficient. The error bars show the 95% confidence limits for the repeatability coefficient

11 Figure 4.10 Fixed and swept masking method RAC-tip estimation agreement for the first test run of each method. Data points show the mean estimated RAC tip from the two methods in the first recording (x-axis) vs. the difference in tip frequency between the two recording methods in the first test session. The upper and lower dashed lines show the mean tip difference ± the calculated limits of agreement. The error bars show the 95% confidence limits for the calculated limits of agreement Figure 4.11 Swept method RACs for subject 18 (left) and 20 (right). RAC curve produced from identical EEG data with (solid line) and without (dashed line) variance weighted averaging. The axes are as for Figure Figure 5.1 An example of PTC ROEX fitting (Subject M5 reverse masker frequency direction first test run) and Q 10dB calculation. Lighter line represents the turning points of the masked threshold tracking of the automated fast-ptc method. Dashed line represents the ROEX curve fit of the turning points. Darker sold line shows bandwidth (Q 10dB ) Figure 5.2 Pure tone thresholds and two test runs PTC ROEX fits for all subjects. Pure tone thresholds converted to db SPL shown as circles with missing points indicating no response at the highest possible audiometer output. PTC ROEX fits for ascending masker frequency direction shown as dashed line, downward masker frequency direction as dotted line with thicker lines for first test run and narrower lines for second test run Figure 5.3 A comparison of ROEX fitted PTC tip frequency in two test runs for each subject. X-axis represents the mean PTC tip between two test runs for each subject and Y-axis the difference in tip frequency between the two test runs. The repeatability coefficient markers shown as dashed lines with error bars representing the 95% confidence interval of the coefficient Figure 5.4 Two sessions of each RAC recording with a 0 db SMR for subject and recording number as indicated in each panel. X-axis represents the masker frequency and Y-axis the absolute response amplitude. The recorded amplitudes are represented by the darker solid line, the ROEX RAC curve fit by the darker dashed line. The not-masked response amplitudes are shown as a lighter solid line and the amplitude which the response must exceed to be considered as significantly above the noise at the 1% F-test ratio is shown as the lighter dashed line Figure 5.5 A comparison of first and second test run RACs recorded at 0 db SMR in the mild hearing loss group (see Table 5.2). Lighter solid lines are the response amplitudes in the first test runs, darker solid lines in the second test run. Dashed lighter and darker lines represent the ROEX curve fit of the response amplitudes in the first and second test runs respectively. X-axis represents the masker frequency in khz and Y-axis the response amplitude as a percentage relative to the not-masked response amplitude recorded in each subject and test run preceding the RAC recording Figure 5.6 As for Figure 5.5 except severe hearing loss subjects

12 Figure 5.7 An analysis of the stability of the response amplitude during the RAC recordings in the mild hearing loss subject, corresponding RAC shown in Figure 5.5. Solid lines represent the mean response amplitude for a 15 sweep slice of the 120 sweep recording time at the start sweep as indicated on the X-axis. Dashed lines represent the amplitude above which the response must lie to be considered significant, response amplitudes that fall below this line are not significantly different from background EEG noise. Lighter lines represent analysis for the first RAC recording and darker lines for the second RAC recording Figure 5.8 As for Figure 5.7 except for severe hearing loss subjects with corresponding RAC as shown in Figure Figure 5.9 As for Figure 5.3 except for RAC ROEX fitted tip frequencies for subjects with two recordings at the 0 db SMR condition Figure 5.10 An analysis of the agreement between the ROEX fitted PTC tip frequency and the ROEX fitted RAC tip frequency of the first test runs of subjects with RAC recordings in the 0 db SMR condition. X-axis represents the mean tip frequency between two methods and Y-axis the difference in tip frequency between two methods. Dashed lines represent the upper and lower limits of agreement with error bars representing the 95% confidence interval of the limits Figure 5.11 Comparisons of the additional test runs performed in subject S1 with test parameters as shown in Table 5.4 and as indicated on the figure panels. RAC relative response amplitudes are shown in the upper panels (axes as Figure 5.5). Lighter lines represent the amplitudes (solid line) and ROEX fitted curve (dashed line) in the first condition and darker lines in the second condition. The stability of response amplitudes are shown in the lower panels (axes as Figure 5.7). Solid line represents the mean response amplitude for a 15 sweep slice of the 120 sweep recording time at the start sweep as indicated on the X-axis. Dashed lines represent the amplitude above which the response must lie to be considered significant, response amplitudes that fall below this line are not significantly different from background EEG noise. Lighter lines represent analysis for the first RAC recording and darker lines for the second RAC recording. Relative response amplitudes shown as percentage of notmasked response amplitude recorded in the same session as the RAC at 20 db SL Figure 5.12 Comparisons of RAC test runs performed in subject S5 with test parameters as shown in Table 5.4. Axes as for Figure 5.5. Faint lines represent RAC recording in the first test session, lighter line in the second test session and darker line represents the recording with masker channel disabled to show fluctuations in response amplitude when response is analysed using identical RAC method Figure 5.13 Comparison of RAC test runs performed in subject S6 with test parameters as shown in Table 5.4. Axes as Figure 5.5. Lighter line represents the first condition (20 db SMR) and darker line the second condition (10 db SMR)

13 Figure 6.1 Response amplitude (darker thicker line), significant response amplitude (1% criterion) level (lighter dashed and dotted line), significant response amplitude (5% criterion) level (darker dashed and dotted line) and noise (dashed line) against number of consecutive sweeps in the analysis for the second test session of subject S2. The point at which the response reached the 5% significance level is marked with an arrow head Figure 6.2 F-ratio significance levels shown against test time (number of sweeps) for forty recordings in normally hearing subjects at 50 db SL (two recordings for twenty subjects). Individual shaded rectangles represent the calculated significance level at each position in the recording. Rows represent individual recordings with two recordings for each subject. Columns represent the number of sweeps included in the response calculation. The left panel shows the false responses detected at 87 Hz and the right panel shows true response detection at the stimulus modulation frequency of 95 Hz. White shading indicates response amplitudes not significant (5% level), light grey significant between 2% and 5%, dark grey significant between 2% and 1%, and black significant at <1% Figure 6.3 As for Figure 6.2 except for twenty recordings in hearing-impaired subjects (two recordings per subject) Figure 6.4 Response amplitudes recorded from normally hearing (N1-N20 upper panel) and hearing-impaired subjects (M1-M5 and S1-S5 lower panel) Figure 6.5 Repeatability of ASSR amplitudes in normally hearing subjects (N1-N20) in the left panel and hearing-impaired subjects (M1-M5 and S1-S5) in the right panel. Data points show the mean response amplitude (nv on the x-axis) vs. the difference in response amplitude between two test runs (nv on the y-axis). The upper and lower dashed lines show the assumed mean difference of 0 nv ± the calculated repeatability coefficient. The error bars show the 95% confidence limits of the repeatability coefficients Figure 6.6 As for Figure 6.5 except for mean response latency (x-axis) and difference in response latency between two test runs (y-axis)

14 Abstract The University of Manchester Timothy Steven Wilding Doctor of Philosophy Using the Auditory Steady-state response to diagnose dead regions in the cochlea 2011 The current behavioural dead region (DR) diagnosis methods such as psychophysical tuning curves and the threshold-equalising noise test require extensive subject co-operation. These present methods cannot be applied to infants. The work presented in the thesis aimed to develop a fast objective DR diagnosis method that could be applied to sleeping hearing-impaired infants. A novel fast objective electrophysiological method of recording response amplitude curves (RACs) which could enable objective DR diagnosis was developed. RACs were derived by recording auditory steady-state response amplitudes using modulated signals in the presence of narrow-band maskers. Two RAC methods were investigated. In the swept method, RACs were recorded in a single test run by recording the response amplitudes across the frequency range of a continuously swept-frequency narrow-band masker. In the fixed method, response amplitudes of eight separate test runs, each in the presence of differing fixedfrequency narrow-band maskers, were recorded. RACs were recorded in normally hearing adult subjects. The results showed that for normally hearing subjects in condition 1 (swept masker), the mean recorded RAC tip for a 2-kHz signal was 2250 Hz and the repeatability coefficient of two repeated recordings in each subject was 389 Hz; in condition 2 (fixed masker), the respective values were 2251 Hz and 342 Hz. These results indicated that the swept masking method is a viable and fast way to record RACs in normally hearing adults. RACs and psychophysical tuning curves (PTCs) were recorded in hearingimpaired adult subjects in order to asses the tip-frequency agreement between the tests. In some cases there were difficulties in using the required signal and masker levels due to maximum sound level limits. The RACs were poorly shaped and had poor repeatability. These findings indicate that the RAC method that was successfully applied to normally hearing subjects requires further development for use with the hearing impaired. The possible causes for the differences in the accuracy of the method between normally hearing and hearing-impaired subjects are discussed. The work presented in this thesis provides the basis upon which further research can be taken forward. It is envisaged that this work, together with further research, will lead to a clinically-effective objective DR diagnosis method. 14

15 Declaration No portion of the work referred to in the thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning. 15

16 Copyright i. The author of this thesis (including any appendices and/or schedules to this thesis) owns certain copyright or related rights in it (the Copyright ) and s/he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes. ii. Copies of this thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made. iii. The ownership of certain Copyright, patents, designs, trade marks and other intellectual property (the Intellectual Property ) and any reproductions of copyright works in the thesis, for example graphs and tables ( Reproductions ), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions. iv. Further information on the conditions under which disclosure, publication and commercialisation of this thesis, the Copyright and any Intellectual Property and/or Reproductions described in it may take place is available in the University IP Policy (see al-property.pdf), in any relevant Thesis restriction declarations deposited in the University Library, The University Library s regulations (see and in The University s policy on presentation of Theses 16

17 Acknowledgements I would like to dedicate this work to my wife, son, daughter, as yet unborn child and my father. I would like to thank: The members of my supervisory team: Karolina Kluk, Colette Mckay and Richard Baker. Terence Picton for providing the initial MATLAB software which was further developed during the course of this research. I am eternally grateful to Karolina for her support and assistance throughout the PhD and for help with proof reading my paper submissions and thesis. I would also like to thank my wife and father-in-law for their assistance with proof reading. This work was supported by a UK Medical Research Council PhD studentship. 17

18 Abbreviations AABR ABR AI AM AMEXP ANSD APD ASSR CF DR EcochG EEG ERB ETC FFT FM HPN MM NBN OAE PTA PTC RAC RETSPL RMS ROEX SEM SMR SRA SSI TEN ULL Automated Auditory Brain Stem Response Auditory Brain Stem Response Articulation Index Amplitude Modulation Exponential Amplitude Modulation Auditory Neuropathy Spectrum Disorder Auditory Processing Disorder Auditory Steady-state Response Characteristic Frequency Dead Region Electrocochleography Electroencephalogram Equivalent Rectangular Bandwidth Electrophysiological Tuning Curve Fast Fourier Transform Frequency Modulation High Pass Noise Mixed Modulation Narrow-band Noise Otoacoustic Emissions Pure Tone Audiogram Psychophysical Tuning Curve Response Amplitude Curve Reference Equivalent Sound Pressure Level Root Mean Squared Rounded Exponential Standard Error of Measurements Signal to Masker Ratio Significant Response Amplitude Speech Intelligibility Index Threshold Equalizing Noise Uncomfortable Loudness Level 18

19 Description of thesis format This thesis presents the author s original work, which aimed to investigate a novel, objective method for diagnosing cochlear dead regions. The thesis is presented in the alternative format. The results chapters are presented in a format required for publication in peer-reviewed journals. In each of these chapters an indication as to the status of the paper (i.e. submitted, in press), the paper title, the journal name and the names of co-authors are given. All of the work presented in this thesis was the original work of the author and was undertaken during the period of the PhD study. The named co-authors were all part of the author s PhD supervisory team with the exception of Terence Picton (The Rotman Research Institute, technical support). Terence Picton provided technical support. He was also the author of the original MATLAB software code that was utilised and further developed by the author of this thesis without collaboration. 19

20 Chapter 1. Introduction Cochlear sensory-neural hearing loss, depending on its severity, can be caused by a mixture of outer and inner hair cell/neuron damage or dysfunction (Moore et al., 2000). The inner hair cells of the cochlea convert the mechanical motion of the basilar membrane into nerve impulses that can be further processed by the auditory pathways of the brain. In normally hearing subjects each position along the basilar membrane is maximally excited by a particular frequency known as its characteristic frequency (CF). It is possible that inner hair cells associated with a particular CF are completely non-functioning. Regions of CFs within the cochlea that are non-functional are known as dead regions (DRs) (Moore, 2004, Moore and Alcantara, 2001, Moore et al., 2000, Moore et al., 1997). Signals at frequencies normally detected within the area of a DR can still generate an auditory response due to the spread of excitation along the basilar membrane. This phenomenon is known as off-frequency listening (Moore and Alcantara, 2001, Moore, 2004). The presence of a DR has implications for providing the best comfort and speech clarity when fitting hearing aids to users with a DR. For example several researchers suggested that there may be little or no benefit from providing amplification at frequencies inside the DR (Baer et al., 2002, Mackersie et al., 2004). It is possible to diagnose DRs by the use of specialised psychoacoustic tests designed to detect the presence of off-frequency listening. However, at present, the methods and procedures used in the diagnosis of DRs cannot be applied to infants and very young children owing to the required level of subject cooperation. This thesis investigates the possibility of diagnosing a DR using a method that could be applied to infants. 20

21 Introduction Chapter Literature Review Dead Regions Moore (2004) described the mechanisms of off-frequency listening and calculated the hypothetical excitation patterns of pure tone stimuli across a range of frequencies and sound pressure levels. Moore (2004) calculated that a 90 db SPL 1.5-kHz pure-tone stimulus generates basilar membrane vibration equivalent to approximately 50 db SPL at the 1-kHz CF point. It is also important to note that the use of the term dead might imply a total lack of function, but the term also includes CFs where the transduction mechanisms are so poorly functioning that the frequencies associated with it are more efficiently detected off-frequency (Moore, 2004). Halpin et al. (1994) provided further psychoacoustic and physiological evidence of off-frequency listening in hearing-impaired subjects. They compared two subjects with nearly identical audiograms revealing normal mid-frequency and moderate low-frequency thresholds. However, post mortem examination revealed that the apical region of the Organ of Corti was absent in one subject, whereas it was present in the other. This comparison supports the hypothesis that behavioural responses occur from test tones within a non-functioning area of the cochlea due to off-frequency hearing. Moore (2001) suggested that it may be possible to predict the presence of a DR using the pure tone audiogram (PTA). He stated that a DR may be present if any one of the following conditions are met: the thresholds are > 90 db HL in the high frequencies, db HL thresholds at low frequencies with near normal at the higher, and >50 db HL at low frequencies better in the high or a rapidly sloping hearing loss of more than 50 db/octave. However, these suggested audiometric configurations are not absolute DR limits and are merely intended to be used as possible DR indicators suggesting the need for further diagnostic testing. Moreover, in a subsequent study Aazh & Moore (2007) found that that the absolute threshold and slope of the PTA could not adequately predict the presence of a DR. Their study focused on high-frequency DRs in a sample of elderly subjects who presented to clinic for hearing assessment. They found that although hearing thresholds at 4 khz were significantly different between groups of subjects with and without a DR, the 21

22 Introduction Chapter 1 sensitivity and specificity of predictive DR diagnosis was poor. They also found that there was no statistically significant difference in the slope of audiogram of subjects with and without a DR. Halpin (2002) presented hypothetical audiograms for differing possible DR configurations primarily based on Zwicker s (1974) work and provided evidence of subjects with confirmed DRs whose audiograms resembled the hypothetical calculations. The hypothetical and actual audiograms shown in this study were of unusual shape with features such as poorer thresholds in mid-frequency sections with normal low- and high-frequency thresholds. Halpin (2002) suggested that although the PTA cannot diagnose a DR, unusual shapes or features can still offer clues as to their presence and indicate when further testing is required Current Dead Region diagnosis methods DR diagnostic tests aim to distinguish when a sound is detected on- or offfrequency. The off-frequency detection of sound is determined by the use of testing in several ipsilateral masking paradigms. Psychophysical tuning curves Moore and Alcantara (2001) described the use of psychophysical tuning curves (PTCs) in DR diagnosis. The PTC test determines the ipsilateral masker level required to prevent the detection of a probe tone (Moore, 1978). The probe tone is fixed whilst the masker frequency is varied. The level of masker required to prevent the subject responding to the probe tone is plotted against multiple discrete masker frequencies. The frequency selectivity at each cochlear CF can be considered to operate as a filter and the effectiveness of the masker is dependent on its power level within the filter bandwidth (Yost, 2000). In normally hearing subjects the lowest effective masker level is for a masker with frequency at, or close to, the probe frequency. The lowest point on the masker intensity versus frequency curve (PTC) is known as the PTC-tip frequency, and represents the CF at which the signal is detected. In subjects with DRs the PTC-tip frequency is shifted away from the probe frequency when the probe is within a DR and occurs at the frequency boundary of the DR (Moore and Alcantara, 2001). 22

23 Introduction Chapter 1 Several experimenters reported general difficulties with PTC testing. Stelmachowicz and Jesteadt (1984) reported that it can be difficult to measure tuning curves in hearing-impaired subjects owing to the high level of masking required. Moore and Alcantara (2001) collected data from five subjects who had DRs with varying aetiologies and hearing thresholds. They found variable PTC-tip frequencies were recorded for low signal sensation levels, and experienced difficulty interpreting the results with high signal levels due to broader tuning at high levels. The shapes of the PTCs were distorted at masker frequencies close to the probe tone. The distortion in shape was found to be due to the subject detecting the signal and masker beating. In some cases the detection of the beating led to inaccurate estimation of the PTC-tip frequency. The PTC shape distortions near the tip could have been prevented by adding an additional pair of beating tones to the test stimulus (Moore et al., 1998, Kluk and Moore, 2005). Moore and Alcantara (2001) found that for signal frequencies within a DR there was a tendency for the tip frequency to decrease as the signal level increased. The downward shift was attributed to damaged areas in the cochlea that responded only at the higher signal levels. This factor indicates that the choice of signal level can affect the recorded tip-frequency and thus affect the diagnosed DR boundary frequency. The PTC test is highly time-consuming. PTC testing required for the diagnosis of a DR can take up to two hours to complete (Sek et al., 2005, Kluk and Moore, 2005, Kluk and Moore, 2006b). Stelmachowicz and Jesteadt (1984) suggested testing 10 masker frequencies but this could potentially lead to an inaccurate PTC tip due to poorer curve sampling. Sek et al. (2005) investigated a faster automated method of determining the PTC in normally hearing listeners. In the fast-ptc method a narrow band masking noise was swept across the test frequency range whilst its level was increased or decreased according to the subject s response to the test tone. The probe tone was presented at a supra-threshold level with a low level masker at a frequency far from the probe frequency to ensure the subject could hear the test tone and respond by pressing the response button. The masking noise was then increased until the test signal was inaudible as indicated by the subject releasing the response button. The masker level was then decreased until the response button was depressed to indicate the audibility of the test signal. The process was repeated whilst the masker frequency was adjusted across the masker 23

24 Introduction Chapter 1 frequency test range. In this manner the turning points of the masker level were tracked across the test frequency range. The PTC test runs were repeated with the masker frequency being swept in either an upward or downward direction in each test run. The tip-frequencies were calculated by performing a linear regression fitting method upon sections of the smoothed and averaged masker levels. The fast- PTC method was validated by comparison to conventionally measured PTCs. The fast-ptc method led to the possibility of obtaining PTCs for three probe frequencies in around 30 minutes. Sek et al. (2005) found that the tips were up to 6% above the probe frequency for forward sweeps and 3% below for reverse sweeps owing to the subject s response time. They suggest that it may be necessary to apply a correction factor or perform testing in both masker frequency directions to counteract this effect. The fast-ptc test has also been successfully applied to normally hearing and hearing-impaired children (Malicka et al., 2009, Malicka et al., 2010). In common with conventional PTC testing the fast-ptc method should also be used with caution as the required masker levels can potentially exceed the subject s uncomfortable loudness levels (ULL) (Sek et al., 2005). Threshold equalising noise (TEN) Moore et al. (2000) proposed an alternative fast diagnostic DR test known as the threshold equalising noise (TEN) test. The method involves determining the threshold of pure tones at each audiometric frequency in the presence of a broadband masking noise known as TEN. TEN is spectrally shaped to produce equal-masked thresholds across the audiological test frequency range for normally hearing subjects and for those with cochlear hearing loss without DRs (Moore et al., 2000). The TEN is presented from a commercially available compact disc. The spectral shape of the TEN was designed to produce identical noise power for each equivalent rectangular bandwidth (ERB) filter in the cochlea, thus leading to equal-masked pure tone thresholds. In normally hearing listeners the TEN produces constant masked thresholds across the frequency test range as long as the noise is above absolute threshold and produces sufficient masking effect (Moore et al., 2000). For example, in a normally hearing subject a 50 db SPL/ERBN TEN masker level will produce masked thresholds at 50 db SPL for all frequencies within the TEN frequency range. A DR is detected at any individual frequency when the masked threshold in the 24

25 Introduction Chapter 1 presence of the TEN is at least 10 db higher than the absolute threshold and 10 db higher than the TEN level. The elevated thresholds in the TEN occur in subjects with DRs owing to off-frequency detection of the signal. The excitation on the basilar membrane at the off-frequency place is less than the excitation normally associated with the tone at its CF (on-frequency). Moore et al. (2000) initially validated the TEN in normally hearing subjects and then applied the test criteria to 14 subjects with sensory neural hearing loss of varying aetiologies. Moore et al. concluded that the PTC results diagnosed DR with a higher degree of accuracy, but the TEN test required less test time and the shape of the PTCs can be affected by the detection of the interaction of signal and masker leading to erroneous tip-frequencies (Moore et al., 2000). They reported cases when the tests did not agree. In one case they attributed the test disagreement to poor subject concentration and there were two cases where the TEN produced false positives with unknown cause. Moore et al. (2003) applied the TEN test to a group of teenagers with severe to profound hearing loss. They found that in many cases TEN testing gave inconclusive DR diagnosis as it was not possible to present the TEN at a level high enough to give sufficient masking to fully satisfy the DR criteria. Munro et al. (2005) examined the accuracy and repeatability of the TEN test by re-testing 24 teenaged subjects after an interval of 12 months. Munro found that the repeated testing was largely in agreement with the first test. However, there were two reported cases where the two results disagreed and six that were previously positive became inconclusive at the second time of testing. In one case the initial test suggested a DR and the second test did not. One of the tests had to have been inaccurate since there are no known mechanisms where a DR can recover. Munro et al. (2005) hypothesised that the test error could have been due to the fact that the threshold testing used a 5 db step size. The use of a 5 db step size may incorrectly suggest the presence of a DR in cases where the threshold lies exactly at the 10 db cut-off point and therefore cases which revealed a greater than 10 db difference in the first test run showed a less than 10 db difference in the second. A finer 2 db step size is recommended to more precisely determine the presence of a DR (Moore, 2004). 25

26 Introduction Chapter Comparison of DR diagnosis methods The TEN test is a quick, simple test, and does not depend upon the patient reaction times as, unlike with fast-ptc testing, the response latency is unimportant. The TEN test can be administered by using standard audiological threshold procedures and equipment by presenting the TEN noise via the external input channel of an audiometer, whereas the fast-ptc method requires the use of specialised equipment and software. It is likely that the TEN test can be applied to any patient able to complete a normal masked PTA but its reliability is not sufficient to be sure of the diagnosis and it does not precisely define the DR boundary. This suggests that although the TEN test can be used as a fast DR screening tool further testing is necessary to confirm diagnosis and precisely define the DR boundary. There are other factors that can potentially affect both the TEN and PTC testing. Studies have shown that retro-cochlear pathology and auditory neuropathy spectrum disorder (ANSD) do not affect the tip frequency of PTCs (Papsin et al., 1994, Vinay and Moore, 2007b), whereas ANSD can produce TEN test results that incorrectly suggest the presence of a DR (Vinay and Moore, 2007b). This implies that it is necessary to confirm the presence of a DR using the PTC test unless it is known that the hearing loss is not due to ANSD or of retro-cochlear origin. The TEN test and PTC test can both potentially require high signal levels which could exceed the maximum possible signal output level, causing loudness discomfort or noise damage, making testing difficult or impossible (Moore et al., 2003). The discomfort problems are more likely to occur in the case of a TEN test as it uses a broadband noise whereas the PTC test method uses narrow-band noise (NBN). However, an updated version of TEN calibrated in db HL over a smaller frequency range was made available to partially alleviate the problem of excess loudness (Moore et al., 2004). The fast-ptc test requires approximately four minutes of test time for each test run. Thirty minutes of test time for each ear is required to define the precise DR boundary by the use of multiple fast-ptc test runs (Sek et al., 2005). A set of TEN thresholds can also take approximately twenty minutes for each ear (Moore et al., 2003). If the fast-ptc test frequencies are carefully selected, the location of a DR can be rapidly determined. The current evidence suggests that it is advisable to use 26

27 Introduction Chapter 1 the fast-ptc method where possible, as it offers the best available reliability and requires minimal clinical time. However, since the TEN test is quick and easy to administer using standard test procedures and equipment, it is likely to be suitable for use as a DR screening tool. In a clinical setting the choice of test would therefore be determined by the need to precisely define the boundary, the subject s ability to perform the fast-ptc testing and the availability of the required equipment. It is important to note that all of the current diagnostic tests are behavioural and require extensive co-operation from the subject Perceptual consequences of Dead Regions Halpin et al. (1994) examined speech perception in subjects with poor low frequency hearing by comparing speech scores with predicted values calculated using the speech articulation index (AI). They found that subjects who tested positive for off-frequency detection (evidence of a DR) had worse speech scores than predicted by the speech AI when the AI was calculated including the audibility of the frequencies within the DR. The measured speech scores better matched the predicted speech scores when frequencies within the DR where considered inaudible. Thus they demonstrated that subjects with DRs had a poorer speech performance than indicated by their audiograms. Moreover, Vinay and Moore (2007a) found that amplifying speech components below 0.57 of the boundary of a low-frequency DR had a detrimental effect on speech perception. Vickers et al. (2001), Mackersie et al. (2004), Baer et al. (2002), Vestergaard (2003) and Simpson et al. (2005) analysed the effect of low-pass filtering of speech testing material in adult subjects with and without high-frequency DRs. Vickers et al. (2001) and Baer et al. (2002) found that the speech scores were lower in the subjects with DRs compared to those without, whereas, Simpson et al. (2005) found that in some, but not all, subjects with a DR speech scores increased with increasing speech material bandwidth. Vickers et al. (2001) concluded that increasing the bandwidth of audible speech improved speech scores only in hearing impaired subjects without DRs. However, in a similar study Mackersie et al. (2004) found no difference in speech scores between subjects with and without a DR in all speech bandwidth conditions except when speech was presented in noise, where subjects with a DR had poorer performance. Conversely, Vestergaard (2003) found that 27