Classification of technical pitfalls in objective universal hearing screening by otoacoustic emissions, using an ARMA model of the stimulus waveform and bootstrap cross-validation.

摘要

Transient-evoked otoacoustic emissions (TEOAE) are widely used for objective hearing screening in neonates. Their main shortcoming is their sensitivity to adverse conditions for sound transmission through the middle-ear, to and from the cochlea. We study here whether a close examination of the stimulus waveform (SW) recorded in the ear canal in the course of a screening test can pinpoint the most frequent middle-ear dysfunctions, thus allowing screeners to avoid misclassifying the corresponding babies as deaf for lack of TEOAE. Three groups of SWs were defined in infants (6-36 months of age) according to middle-ear impairment as assessed by independent testing procedures, and analyzed in the frequency domain where their properties are more readily interpreted than in the time domain. Synthetic SW parameters were extracted with the help of an autoregressive and moving average (ARMA) model, then classified using a maximum likelihood criterion and a bootstrap cross-validation. The best classification performance was 79% with a lower limit (with 90% confidence) of 60%, showing the results' consistency. We therefore suggest that new parameters and methodology based upon a more thorough analysis of SWs can improve the efficiency of TEOAE-based tests by helping the most frequent technical pitfalls to be identified.