Learning from EMG: semi-automated grading of facial nerve function

Area of Science:

• Neurosurgery
• Biomedical Engineering
• Machine Learning in Medicine

Background:

• Current facial nerve function grading relies on the subjective House Brackmann (HB) scale.
• High interobserver variability in the HB scale necessitates more objective assessment methods, particularly for research.

Purpose of the Study:

• To develop and evaluate a semi-automated grading system for facial nerve function using surface electromyography (sEMG) data.
• To assess the feasibility of machine learning algorithms in objectively grading facial nerve function post-vestibular schwannoma surgery.

Main Methods:

• Collected sEMG data from 28 patients undergoing vestibular schwannoma surgery preoperatively, postoperatively, and at 3-12 months follow-up.
• Employed machine learning classifiers (Logistic Regression, SVM, KNN) on preprocessed sEMG data for grading.
• Analyzed data across three scenarios: normal vs. slight impairment, normal vs. impaired, and HB grades 1-3 classification.

Main Results:

• The developed system demonstrated good differentiation capabilities across scenarios.
• Median Area Under the Curve (AUC) values were 0.72 (scenario 1), 0.91 (scenario 2), and 0.74 (scenario 3).
• The study confirmed the feasibility of using sEMG and machine learning for objective facial nerve grading.

Conclusions:

• Semi-automated grading using sEMG and machine learning is a viable approach for objective facial nerve assessment.
• This method shows potential as an alternative to the House Brackmann scale, especially for research applications.
• Objective grading can minimize interobserver variability in the evaluation of facial nerve function.