Machine-Learning Based Analysis of Usefulness of Wideband Tympanometry in Various Middle Ear Disorders

Abstract

Background and Objectives Wideband tympanometry (WBT) provides information additional to what can be provided by the relms of traditional tympanometry, such as absorbance, resonance frequency, and peak pressure. We investigated the characteristics of WBT in various middle ear disorders, especially for chronic otitis media (COM), otosclerosis, and patulous eustachian-tube disorder (ETD).Subjects and Method We recruited 165 patients who presented 179 normal ears and 151 abnormal ears due to COM (113 ears), otosclerosis (14 ears) and ETD (24 ears). We analyzed peak pressure and resonance frequency data using the Mann-Whitney test and absorbance data using the machine learning modeling.Results The only significant difference in peak pressure and resonance frequency was observed in COM ears in contrast to normal ears. For absorbance data from WBT, we made 3 models for machine learning to compare normal ears agaist COM, ETD, and all middle ear disorders. Models for otosclerosis ears and normal ears were impossible to analyze due to the small numbers of otosclerosis patients. Of the 3 models, the model comparing COM and normal ears had only meaningful area under receiver operating characteristic results from least absolute shrinkage and selection operator analysis and Elastic Net analysis.Conclusion WBT could provide useful information for the diagnosis of various middle ear disorders, especially for chronic otitis media.