Performance and Reliability Evaluation of an Automated Bone-Conduction Audiometry Using Machine Learning.

Abstract

To date, pure-tone audiometry remains the gold standard for clinical auditory testing. However, pure-tone audiometry is time-consuming and only provides a discrete estimate of hearing acuity. Here, we aim to address these two main drawbacks by developing a machine learning (ML)-based approach for fully automated bone-conduction (BC) audiometry tests with forehead vibrator placement. Study 1 examines the occlusion effects when the headphones are positioned on both ears during BC forehead testing. Study 2 describes the ML-based approach for BC audiometry, with automated contralateral masking rules, compensation for occlusion effects and forehead-mastoid corrections. Next, the performance of ML-audiometry is examined in comparison to manual and conventional BC audiometry with mastoid placement. Finally, Study 3 examines the test-retest reliability of ML-audiometry. Our results show no significant performance difference between automated ML-audiometry and manual conventional audiometry. High test-retest reliability is achieved with the automated ML-audiometry. Together, our findings demonstrate the performance and reliability of the automated ML-based BC audiometry for both normal-hearing and hearing-impaired adult listeners with mild to severe hearing losses.