Machine and Deep Learning-Based Detection of the Stapedius Reflex for Automatic Cochlear Implant Fitting

Cochlear implants (CIs) effectively restore hearing in individuals with hearing loss. However, the current fitting process, a patient-specific adjustment of a CI, is time-consuming and dependent on user feedback. This poses significant challenges, particularly for children and persons with communication impairments. The acoustic stapedius reflex, an involuntary muscle contraction triggered by loud sounds, can be used as an objective measure of auditory response. This study focusses on the automatic detection of the stapedius reflex via machine and deep learningbased approaches. A dataset comprising of 4,167 measurements was collected, allowing for the investigation of various approaches such as supervised, semi-supervised, unsupervised and transfer learning approaches. By training our models on adults and testing them on minors we demonstrated that the age has no significant effect on the characteristics of the stapedius reflex signal. The logistic regression model emerged as the most effective, achieving a test accuracy of 96.2%, a sensitivity of 97.0%, a specificity of 95.1%, an F1-Score of 96.6%, an ROC curve with coordinates (0.05, 0.97) and an AUC of 0.992. The algorithm has the potential to significantly improve the efficiency and accuracy of CI fitting, thereby reducing hospital visits and improving the quality of life of CI-users.