A Hybrid YOLOv8 and Instance Segmentation to Distinguish Sealed Tissue and Detect Tools' Tips in FLS Laparoscopic Box Trainer

Abstract

Intracorporeal suturing is one of the most crucial skills in the Fundamentals of Laparoscopic Surgery. Surgical residents are evaluated by their supervisory surgeon, but surgical assessment demands a substantial amount of time from the surgeons and can result in biased evaluations. In addition, distinguishing sealed tissue among multiple trainees would be a subjective decision. Therefore, we propose an autonomous assessment support system, which can supervise the execution of the suturing task by using YOLOv8 instance segmentation and object detection, tracking the tips of the suturing instruments and monitoring tissue seals and knots. We used mean average precision and inference time metrics to evaluate the performance of the instance segmentation for our proposed suturing assessment system. It was found that the precision of all suturing instruments was 95% and that the mask precision of the tissue was 98.8%. Our proposed autonomous laparoscopic training system saves the supervisor surgeons' time, and the outcomes of the proposed methodology may also be utilized in the development of surgical robots.