Hearing screening based on deep residual shrinkage network

Abstract

Stimulus-frequency otoacoustic emissions (SFOAEs) could be a useful tool for assessing hearing capabilities. Training and testing data were collected from 1084 ears of 725 subjects. Based on SFOAEs, this study applied Deep Residual Shrinkage Network to screen hearing status. Results of a 10-fold cross-validation show that the model performed better than other recent literature in the range of 0.5 to 8 kHz, occupied a larger area under the receiver operating characteristic curve (0.970-0.991) and had a higher accuracy (0.932-0.959). The developed classified model outperformed the previous model by accuracy improvements of 1.28% to 4.43% at 0.5, 2, 4 and 8 kHz.