Differentiating Laparoscopic Skills of Trainees with Computer Vision Based Metrics

Abstract

Context dependent gaze metrics, derived from eye movements explicitly associated with how a task is being performed, are particularly useful for formative assessment that includes feedback on specific behavioral adjustments for skill acquisitions. In laparoscopic surgery, context dependent gaze metrics are under investigated and commonly derived by either qualitatively inspecting the videos frame by frame or mapping the fixations onto a static surgical task field. This study collected eye-tracking and video data from 13 trainees practicing the peg transfer task. Machine learning algorithms in computer vision were employed to derive metrics of tool speed, fixation rate on (moving or stationary) target objects, and fixation rate on tool-object combination. Preliminary results from a clustering analysis on the measurements from 499 practice trials indicated that the metrics were able to differentiate three skill levels amongst the trainees, suggesting high sensitivity and potential of context dependent gaze metrics for surgical assessment.