Predicting Acoustic Hearing Preservation Following Cochlear Implant Surgery Using Machine Learning.

Area of Science:

• Otolaryngology and Biomedical Engineering
• Artificial Intelligence in Healthcare
• Predictive Analytics in Medicine

Background:

• Cochlear implantation (CI) is a common treatment for severe to profound hearing loss.
• Predicting acoustic hearing preservation post-CI is crucial for patient counseling and surgical planning.
• Current methods for predicting hearing outcomes are limited.

Purpose of the Study:

• To develop and validate supervised machine-learning classifiers for predicting acoustic hearing preservation after CI.
• To identify preoperative clinical factors associated with hearing preservation.
• To enhance shared clinical decision-making for CI patients.

Main Methods:

• Retrospective analysis of a prospectively collected CI dataset (n=175 patients).
• Supervised machine learning classifiers were trained and tested using preoperative clinical data.
• Audiometric testing (standard pure tone average, SPTA; low-frequency PTA, LFPTA) was performed preoperatively and one month post-surgery.
• Model performance was evaluated using metrics like AUC and MCC.

Main Results:

• Machine learning models successfully predicted acoustic hearing preservation.
• Factors like meningitis history, preoperative hearing levels, sudden hearing loss, noise exposure, and anatomy were associated with hearing preservation.
• Random forest models showed the highest classification performance.

Conclusions:

• Machine learning is a valuable tool for predicting residual acoustic hearing in CI patients.
• Identified associations aid in interpreting model predictions and understanding factors influencing hearing preservation.
• These models and findings can improve clinical decision-making and patient outcomes.