Development and application of explainable artificial intelligence using machine learning classification for...

Area of Science:

• Neurosurgery
• Machine Learning in Medicine
• Cerebellopontine Angle Tumors

Background:

• Vestibular schwannomas (VSs) are common cerebellopontine angle tumors.
• Preserving facial nerve (FN) function during VS surgery is a significant clinical challenge.

Purpose of the Study:

• To develop and validate a machine learning model for predicting long-term facial nerve (FN) outcomes after vestibular schwannoma (VS) surgery.
• Classify long-term FN function as good (House-Brackmann grades 1-2) or bad (grades 3-6).

Main Methods:

• Retrospective analysis of 256 patients undergoing VS surgery.
• Utilized the Extreme Gradient Boosting (XGBoost) machine learning classifier for binary classification.
• Employed explainable artificial intelligence (SHapley Additive exPlanations - SHAP) for model interpretability.

Main Results:

• The XGBoost model achieved an average accuracy of 0.83, ROC AUC of 0.91, and MCC of 0.62.
• Short-term FN function was the most influential predictor of long-term outcomes (tau=0.6).
• Large tumor volume and absent preoperative auditory brainstem responses correlated with poor outcomes.

Conclusions:

• An effective machine learning model can classify long-term FN outcomes post-VS surgery.
• Short-term FN function is a critical determinant of long-term recovery.
• The model aids in patient evaluation and guiding recommendations for FN dysfunction management.