Enhancing Autonomous Skill Assessment of Robot-Assisted Minimally Invasive Surgery: A Comprehensive Analysis of Global and Gesture-Level Techniques applied on the JIGSAWS Dataset

Abstract

Improved surgical skills play a crucial role in ensuring optimal patient outcomes. Tra- ditional methods for skill assessment include self-rating questionnaires and expert eval- uations, but these approaches are prone to bias and require substantial qualified human resources. The emergence of Surgical Data Science (SDS) offers a promising avenue for automating skill assessment, leveraging data science techniques to capture, organize, an- alyze, and model surgical data. In this paper, kinematic data was employed from the JIGSAWS – which is the only skill-annotated Robot-Assisted Minimally Invasive Surgery (RAMIS) dataset – to classify surgeons into novice and experienced groups, using various classification methods (Decision Tree, k-Nearest Neighbors, Support Vector Machines, Lo- gistic Regression, Dynamic Time Warping, and 1D Convolutional Neural Network). The research encompasses a thorough analysis of parameter tuning and dimensional reduction techniques with the aim of establishing a universal benchmark for data classification. The surgical training tasks of suturing, knot-tying and needle-passing consistently achieved 100 % accuracy. The accuracy attained during surgical gesture analysis often exceeded the overall accuracy of the global assessment of the dataset.