THE DEVELOPMENT OF SKILL EVALUATION SYSTEM FOR LAPAROSCOPIC SURGICAL PROCEDURE

Abstract

This paper describes details of the development of skill evaluation system for laparoscopic surgical procedure. The measurement experiment was conducted for 70 surgeons, and the participants perform 2 tasks: tissue dissection around the aorta and renal parenchymal suturing/knotting using porcine cadaver. In the experiments, the movement of surgical instruments were recorded by motion capture (MoCap) system, and the motion indices were calculated. The participants grouped into three classes (novices, intermediates, and experts) according to their level of experience. Three classification algorithms: support vector machine (SVM), principal component based SVM (PCA-SVM), gradient boosting decision tree (GBDT) were utilized for developing the model of classifier. The accuracy of each model was assessed by nested and repeated k-fold cross validation. Regarding 3-class classification, the GBDT method resulted highest accuracy (the median of the accuracy is A_med = 68.6 %) in the dissection tasks. In the suturing/knotting tasks, PCA-SVM resulted highest accuracy (A_med = 58.4 %). Regarding 2-class classification (experts vs. intermediates/novices), the GBDT method resulted A_med = 72.9 % in the dissection task, and the PCA-SVM method resulted A_med = 69.2 % in the suturing task. This result shows the MoCap based skill evaluation system in wet-lab training could be a practical way to objectively assess trainees' surgical competence.