Surgical skill analysis using explainable AI in endoscopic sinus surgery

Endoscopic sinus surgery (ESS) is a standard procedure performed worldwide for diseases of the nose and sinuses, but it is highly technical and requires effective training. A system for evaluating the skill level of surgeons using machine learning based on measurement data from endoscopic sinus surgeries performed with a 3D sinus model has been developed. In this study, an analysis using SHapley Additive exPlanations (SHAP) values was conducted to investigate behaviors that significantly reflect surgeons’ skill proficiency. Additionally, feature reduction was performed using a variable reduction method, eliminating features in ascending order of their contributions as calculated by SHAP. This approach aimed not only to mitigate the risk of overfitting due to a decrease in explanatory variables but also to improve the accuracy of classifying surgeons based on skill differences. The results demonstrated that reducing features based on SHAP contributions led to an improvement in classification accuracy.