Model-based prediction of otoacoustic emission level, noise level, and signal-to-noise ratio during time-synchronous averaging

Abstract

Although averaging is effective in reducing noise, its efficiency rapidly decreases beyond several hundred averages. Depending on environmental and patient noise levels, several hundred averages may be insufficient for informed clinical decision making. The predictable nature of the otoacoustic emission (OAE) and noise during time-synchronous averaging implicates the use of predictive modeling as an alternative to increased averaging when noise is high. Click-evoked OAEs were measured in 98, normal-hearing subjects. Average OAE and noise levels were calculated for subsets of the total number of averages and then fit using variants of a power function. The accuracy of the models was quantified as the difference between the measured value and model output. Models were used to predict the OAE signal-to-noise ratio (SNR) for a criterion noise level. Based on predictions, the OAE was categorized as present or absent. Model-based decisions were compared to decisions from direct measurements. Model accuracy improved as the number of averages (and SNR in the case of OAEs) from which the model was derived increased. Model-based classifications permitted correct categorization of the OAE status from fewer averages than measurement-based classifications. Furthermore, model-based predictions resulted in fewer false positives (i.e., absent OAE despite normal hearing).