Optimizing Sensor Selection in Laparoscopic Simulators: Lessons Learned in a Robotic Platform

摘要

Laparoscopic simulators provide a safe environment in which surgeons can practice and hone specific skills without risk to patients. However, providing effective performance feedback requires selecting the relevant metrics that most accurately reflect skill levels while remaining actionable for the trainee. This study investigates optimal sensor selection for laparoscopic simulators to enhance training assessment accuracy. Six common sensor types were tested across different combinations to evaluate their impact on recognizing surgical gestures, surgical tasks, and surgeon expertise levels using convolutional neural networks and multidimensional dynamic time-warping classifiers. The results show that linear velocity and gripper angle yield high classification accuracy across all metrics. For gesture recognition, velocity and gripper angle consistently appeared in the top-performing sensor combinations, demonstrating that these two parameters alone are highly indicative of a surgeon’s intent and skill. Surprisingly, adding positional data does not improve accuracy, challenging the traditional emphasis on positional metrics in training systems. With the right sensor selection, surgical simulators can achieve accurate and actionable feedback while reducing complexity and cost without sacrificing performance, which can help make simulators more accessible and effective for training purposes.