Instrument Detection for the Intracorporeal Suturing Task in the Laparoscopic Box Trainer Using Single-stage object detectors

Intracorporeal suturing is one of the critical tests in the Fundamentals of Laparoscopic Surgery (FLS) training. Performance assessment of the surgical residents is carried out by a supervisor surgeon. It is time-consuming for the surgeon and may lead to a subjective decision. We propose developing an automated system to monitor the suturing process using some recent versions of One-Stage-Object-Detectors such as YOLOv4, Scaled-YOLOv4, YOLOR, and YOLOX. A trade-off between these state-of-arts architectures is presented in this paper by training a dataset of suturing tasks. Since we aim for the most robust detector to use for assessing the residents’ performance, we used mean average precision and inference time metrics for our comparative analysis and evaluations. In the trade-off between accuracy and training speed, the YOLOR-p6 640x model achieved the highest mAP50 with 97.6%.