Multi-class Detection and Tracking of Intracorporeal Suturing Instruments in an FLS Laparoscopic Box Trainer Using Scaled-YOLOv4.

Intracorporeal suturing is one of the most critical skills in the Fundamentals of Laparoscopic Surgery (FLS) training. Assessment of skills acquisition requires a significant amount of the supervisory surgeons' time, and it can be a very subjective decision. This study uses an object detection algorithm, Scaled-YOLOv4, in conjunction with a centroid tracking algorithm to evaluate the surgeon's skills during advanced intracorporeal suturing. We proposed a system capable of locating and tracking surgical instruments as well as providing an evaluation of the performance of the surgeons. Since the accuracy of the detection is crucial to our proposed tracking system, we evaluated the detection performance using the mean average precision and inference time metrics. An average precision of 85.50% was achieved for the detection of the needle, and 100% was achieved for the work field area.