機械学習を用いた内視鏡下副鼻腔手術の技量評価

抄録

Endoscopic sinus surgery (ESS) is one of the standard minimally invasive surgical procedures for diseases of a nasal cavity and paranasal sinuses. This paper describes the development and details of a surgeon skill classification system using machine learning. A learning model was generated based on the features calculated from the dynamic measurement data of surgical instruments in the ESS and the enlarged sinus volume as an indicator for surgical efficacy. By using three machine learning algorithms, Support Vector Machine (SVM), PCA (Principal Component Analysis based)-SVM and Gradient Boosting Decision Tree (GBDT), three-group discrimination (expert vs. intermediate vs. novice) and two-group discrimination (expert vs. intermediate/novice) were performed. A comparison of the accuracy in the methods revealed that SVM is the most accurate method for three-group discrimination, while SVM and GBDT are the most accurate for two-group discrimination.