PhD student, the University of Melbourne
I use recorded brain activity to decide whether an individual is perceiving a sound stimulus.
Millions of infants worldwide have a hearing loss. Infants learn languages quickly by listening in to their environment, so good sound quality is essential for normal development of language and speech. For some, a hearing aid restores sound sufficiently; but for others, the hearing loss is too severe and at the level of the inner ear. An alternative is the cochlear implant, an electronic device used by many adults which provides electrical stimulation directly to the nerves of the inner ear, accomplishing what the hearing aid cannot. The electrical stimulation needs to be customised for each implantee in a process called implant fitting, and is traditionally adjusted through their verbal feedback. This procedure is time consuming and must be repeated roughly yearly, but more importantly becomes much more difficult and inaccurate for infants as they are not capable of providing verbal feedback.
My research aims to circumvent verbal feedback by reading activity from the brain when sounds are heard. I aim to automatically find hearing thresholds, the lowest audible electrical stimulation level, using recorded brain activity. This is an important part of fitting cochlear implants, and automation can eliminate lengthy sessions with an audiologist.
Brain signals has been explored for decades and many fantastic signal processing tools have been developed; however, only basic tools and exploration exist in research on the hearing brain. In addition, threshold seeking is especially challenging because the brain signal associated with hearing a quiet sound are very small fluctuations compared to surrounding brain activity not related to hearing.
I take these tools and modify and apply them to explore the dynamics of the hearing brain. So far, I have found that looking at the brain signals with statistical models can detect these small brain fluctuations to hearing more optimally. These statistical models also show how brain activity differs across everyone, which can be utilized in hearing research. I am currently to translate the findings from these models to an automated hearing threshold estimate. I am also exploring machine learning, a method whereby patterns are learnt algorithmically, to discern these brain signals.
My work is another step in improving the cochlear implant technology. I hope it will enable more infants to get a cochlear implant with good outcomes and also reduce time and effort for adult implantees by automating the fitting procedure.
Abstract: To report on clinical experience using dichotic multiple-stimulus auditory steady-state responses (ASSRs) as an objective technique to estimate frequency-specific hearing thresholds in hearing-impaired infants.A comparison was made between the click-evoked auditory brainstem response (ABR), auditory steady-state responses and behavioral hearing thresholds (BHTs). Both ears of 10 infants between 3 and 14 months of age were tested. ABR and ASSRs were recorded during the same test session. ABR was evoked by 100 micros clicks. ASSRs were evoked by amplitude- and frequency-modulated tones with carrier frequencies of 0.5, 1, 2 and 4 kHz and modulation frequencies ranging from 82 to 110 Hz. Eight signals (four to each ear) were presented simultaneously. ASSR thresholds were derived after separate recordings of approximately 5, 7.5 and 10 min to compare the influence of test duration. BHTs were defined in later test sessions as soon as possible after the ASSR test, dependent on medical and developmental factors.For the subjects tested in this study 60% of ABR thresholds and 95% of ASSR thresholds for 1, 2 and 4 kHz were found at an average age of 7 months. Only 51% of frequency-specific BHTs could be obtained but on average 5 months later. The correlation of ABR thresholds and ASSR thresholds at 2 kHz was 0.77. The correlation of ASSRs and BHTs was 0.92. The mean differences and associated standard deviations were 4 +/- 14, 4 +/- 11, -2 +/- 14 and -1 +/- 13 dB for 0.5, 1, 2 and 4 kHz, respectively. The average test duration was 45 min for ABR (one threshold in both ears) and 58 min for ASSR (four thresholds in both ears). By reducing the duration of the separate recordings of ASSR, the precision of the hearing threshold estimate decreased and the number of outlying and missing values increased. Correlation coefficients were 0.92, 0.89 and 0.83 for recordings of maximum 10, 7.5 and 5 min, respectively. A compromise between test duration and precision has to be sought.Multiple-frequency ASSRs offer the possibility to estimate frequency-specific hearing thresholds in babies in a time-efficient way.
Pub.: 09 Jun '04, Pinned: 31 Aug '17
Abstract: To investigate the clinical usefulness of the LS-chirp auditory brainstem response for estimation of behavioral thresholds in young children with mild to severe hearing losses.68 infants (136 ears) aged 6-12 months (mean age=9.2 months) with bilateral mild to severe hearing losses were studied at Children's Hospital of Fudan University. In all cases, the children were referred for LS-chirp ABR and visual reinforcement audiometric (VRA) measurements. The low-frequency band chirp (LF-chirp) thresholds (frequency band=0.1-0.85kHz) were compared to the average VRA thresholds (frequency band=0.25-0.5kHz), whereas the high-frequency band chirp (HF-chirp) thresholds (frequency band=1-10kHz) were compared to the average VRA thresholds (frequency band=1-4kHz) using statistical correlation coefficient values.The LS-chirp ABR thresholds are very close to behavioral hearing levels. The mean differences between chirp-ABR and VRA thresholds were within 5dBHL for all measurements. The smallest mean threshold difference (<3dBHL) was obtained for the severe hearing loss group. The correlation coefficient values (r) were 0.97 at low-frequency and high-frequency bands. For each carrier frequency, the best correlations between chirp-ABR thresholds and VRA thresholds were obtained at VRA frequency of 0.25kHz/LF-chirp (r=0.98) and VRA frequency of 1kHz/HF-chirp (r=0.98).This study demonstrates the effectiveness using chirp-ABR predicted frequency-specific thresholds, especially of low and middle frequencies. LS-chirp ABR thresholds determined behavioral thresholds in patients with severe hearing losses were better than for mild hearing losses. The use of a chirp-ABR testing ensures higher sensitivity and accuracy than that of auditory stead-state evoked response (ASSR) for measuring frequency-specific thresholds in young children.
Pub.: 19 Mar '14, Pinned: 31 Aug '17
Abstract: The study aims were to determine the incidence of exaggerated hearing thresholds in individuals complaining of noise-induced hearing loss (NIHL) as a result of impulse noise using cortical evoked response audiometry (CERA) and to identify any associated audiometric features.We conducted an office-based study.In this prospective case series, 1154 males complaining of NIHL were assessed with pure tone audiometry; 673 had CERA. Pure tone averages (PTA) and hearing disability were calculated using the Irish and American Medical Association systems. A PTA of >10 dB worse than the CERA average was considered evidence of exaggerated thresholds.The mean PTA was 33 dB. Seventy-two percent had a hearing disability of an average of 26% when assessed by the Irish system. Fifty-four percent had a hearing disability of an average of 30% when assessed by the American Medical Association system. Twenty-six percent of subjects had exaggerated thresholds based on CERA. A binaural hearing threshold of >25 dB at 500 Hz had a sensitivity of 94% and a specificity of 59% for the detection of exaggerated thresholds.Exaggerated hearing thresholds are common. A hearing threshold of >25 dB at 500 Hz should be considered an indication for CERA testing.
Pub.: 26 Feb '03, Pinned: 31 Aug '17
Join Sparrho today to stay on top of science
Discover, organise and share research that matters to you