Machine Learning Models for Predicting Vestibular Function After Cochlear Implantation.

Area of Science:

• Otolaryngology
• Biomedical Engineering
• Data Science

Background:

• Sensorineural hearing loss (SNHL) affects numerous individuals, often necessitating cochlear implantation (CI).
• Vestibular dysfunction is a potential complication following CI, impacting patient quality of life.
• Predicting post-CI vestibular outcomes is crucial for surgical planning and rehabilitation.

Purpose of the Study:

• To evaluate machine learning (ML) for predicting vestibular abnormalities post-CI in SNHL patients.
• To develop a practical ML model for forecasting long-term vestibular function.
• To identify significant risk factors associated with post-CI vestibular dysfunction.

Main Methods:

• Clinical data (imaging, VEMPs, auditory info) collected pre- and post-CI.
• Decision tree algorithm used for data imputation and feature screening.
• Six ML methods applied to predict vestibular dysfunction; feature importance analyzed.

Main Results:

• Logistic regression achieved 80% accuracy for vestibular dysfunction and 78% for abnormal cVEMP.
• Support Vector Machine (SVM) reached 88% accuracy for abnormal oVEMP.
• Key predictors included VEMP latency/amplitude and residual hearing threshold.

Conclusions:

• ML models, including SVM and logistic regression, effectively predict CI's impact on vestibular function.
• These models offer valuable insights for presurgical planning and decision-making.
• Identified risk factors provide a basis for guiding vestibular rehabilitation strategies.