Simulating sensorineural pathologies to help refine their diagnosis

Sensorineural hearing loss (SNHL) can express itself in many ways, but underlying pathologies cannot currently be finely diagnosed. In order to establish the psychophysical signature(s) of a given pathology, the performance of young normally-hearing listeners is measured as they attend to a SNHL simulator based on a version of the physiological Model of the Auditory Periphery (MAP, Meddis et al. 1986~2018) impaired accordingly. The acoustic stimulus is passed through MAP, then reconstructed from the predicted auditory-nerve (AN) firing response. Identifying psychophysical tests that can help discriminate between pathologies is essential to refining the diagnosis of SNHL, such that targeted treatment may become conceivable.

Early work (SpIN 2019) demonstrated that medial olivo-cochlear (MOCR) and acoustic (AR) efferent reflexes are essential to the faithful coding of the temporal modulations that carry speech information. Efferent reflexes play a key role in the AN dynamic-range adaptation to context level that prevents information loss (e.g. through AN saturation). The simulator was also validated for its normal-hearing version by obtaining speech reception thresholds (SRT) only 1 dB higher than those obtained with unprocessed stimuli.

A first study measured the impact of simulated pathologies on speech intelligibility in noise. While deactivating 70% of ANs or halving the endocochlear potential did not lead to any appreciable SRT inflation, total loss of outer haircells led to a significantly smaller SRT inflation than the 3-4 dB found earlier, when both MOCR and AR were knocked out. Making use of different maskers (broad-band speech-modulated noise, or speech, instead of steady speech-shaped noise) may enable improved pathology discrimination.

A second study aimed to measure the amount of simulated deafferentiation that leads to significant SRT elevation. 90-95% deafferentiation was found to be required to significantly elevate SRTs in steady noise. This outcome is consistent with the effect of stochastic under-sampling predicted by Lopez-Poveda and Barrios (2013, Front. Neurosci. 7:124).

A third study established changes in binaural masking level difference (BMLD) for the pathologies simulated in the first study. This was done by comparing diotic (N0S0) to dichotic (N0Sπ) thresholds of audibility of a 250-Hz tone in white noise, for which a 13-dB BMLD is found with normal hearing (Hirsh and Burgeat, 1958, JASA 30:827). Only a modest BMLD drop was found with simulated pathologies, despite individual thresholds being inflated.