Surgical Tool Detection in Open Surgery Based on Faster R-CNN, YOLO v5 and YOLOv8

High-precision tool detection can provide rich tool movement information for automated surgical video analysis, so tool detection based on surgical video is of vital significance for understanding surgical scenes. Previous studies on surgical tool detection mainly focused on endoscopic surgery. However, surgical instruments in open surgical video have more difficulties to detect because of the occlusion between instruments. In this paper, we introduce a new simulated open surgery dataset called AEDCSSAD, which includes bowel anastomosis, hepatic rupture repair and closure the abdomen surgeries. Pre-trained Faster R-CNN and different versions of YOLOv5 and YOLOv8 were used and fine-tuned on our dataset respectively, among which YOLOv8x achieved the best results on the dataset, with the mAP up to 94.2%.