Extracting Unusual Movements during Robotic Surgical Tasks: A Semi-Supervised Learning Approach

Abstract

Although modern features in surgical robots such as 3D vision, “wrist” instruments, tremor abolition, and motion scaling have greatly enhanced surgical dexterity, technical skill is a major challenge for surgeons and trainees. Surgeons who get constructive and real-time feedback can make more significant improvement in their performance [1]. Recent years, the research in automated surgical skill assessment has made considerable progress, however, the majority of surgical evaluation methods are post- operation analysis. Few studies introduced real-time sur- gical performance evaluations, for example, using Con- volutional Neural Network [2], Codebook and Support Vector Machine [3], and Convolutional Neural Network - Long Short Term Memory [4]. One common limitation of these studies is data leakage during training which results in a higher estimate of model performance. Moreover, these studies cannot depict an intuitive repre- sentation of what actually differentiates expertise levels. In this study, we introduce a method to extract the unusual movements which are rarely seen in Experts and identify the types of the unusual movements. We believe detecting and correcting the unusual movements is an important aspect for surgeons to improve their skills.