Predicting synkinesis caused by Bell's palsy or Ramsay Hunt syndrome using machine learning-based logistic regression

Area of Science:

• Neurology
• Medical Informatics
• Biostatistics

Background:

• Facial synkinesis is a common complication of Bell's palsy and Ramsay Hunt syndrome.
• Predicting synkinesis aids in patient management and treatment planning.
• Current prediction methods may not fully leverage early post-onset data.

Purpose of the Study:

• To evaluate machine learning (ML) algorithms for predicting facial synkinesis.
• To compare the predictive accuracy of ML models against conventional statistics-based logistic regression (LR).
• To utilize early post-onset parameters for enhanced prediction.

Main Methods:

• Retrospective study of 362 patients with facial palsy.
• Evaluation of electrophysiological examinations, including electroneurography (ENoG) and nerve excitability test (NET).
• Application of ML algorithms (LR, Random Forest, k-NN, GBDT) and statistics-based LR using early parameters.

Main Results:

• Synkinesis rates were 20.2% for Bell's palsy and 40.0% for Ramsay Hunt syndrome.
• ML-based LR using ENoG achieved an Area Under the Curve (AUC) of 0.910.
• ML models demonstrated comparable or superior predictive performance to statistics-based LR.

Conclusions:

• ML-based LR shows potential for predicting facial synkinesis probability.
• The ML approach offers reliable prediction for synkinesis in Bell's palsy and Ramsay Hunt syndrome.
• Early post-onset parameters combined with ML can enhance synkinesis prediction.