Predicting cochlear dead regions in patients with hearing loss through a machine learning-based approach: A...

Area of Science:

• Otolaryngology
• Audiology
• Biomedical Engineering

Background:

• Cochlear dead regions (DRs) significantly impact hearing loss, affecting speech understanding.
• Accurate prediction of DRs is crucial for effective hearing rehabilitation strategies.
• Various etiologies of sensorineural hearing loss (SNHL) present challenges in DR identification.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based model for predicting cochlear dead regions (DRs) in patients with diverse hearing loss causes.
• To identify key clinical factors predictive of DR presence.

Main Methods:

• Analysis of 3,770 test samples from 380 patients with SNHL.
• Utilized the threshold-equalizing noise (TEN) test for DR detection.
• Employed recursive partitioning and regression, logistic regression, and random forest for ML model development.
• Collected data included hearing loss etiology, pure-tone thresholds, and word recognition scores (WRS).

Main Results:

• Overall prevalence of one or more DRs was 20.36% in test ears.
• Frequency-specific DR prevalence was 6.7%.
• Word recognition scores, pure-tone thresholds, and specific disease types (vestibular schwannoma, Meniere's disease) were significant predictors.
• Patient sex and age were not associated with DR detection.

Conclusions:

• Machine learning models can effectively predict cochlear dead regions.
• Audiometric measures and specific audiological conditions are valuable predictors of DRs.
• Further refinement with additional clinical data may enhance predictive accuracy for DRs.