Detection of Anatomical Landmarks During Laparoscopic Cholecystectomy Surgery Based on Improved Yolov7 Algorithm

Many adverse events during laparoscopic cholecystectomy surgery occur due to errors in visual perception and judgment by surgeons, leading to misunderstandings of anatomical structures, particularly in surgeries performed by young and inexperienced doctors. This paper proposes a YOLOv7_CA model for detecting anatomical landmarks during laparoscopic cholecystectomy surgery, based on an attention mechanism. The network efficiently detects the gallbladder, Calot's triangle, and common bile duct, which are prone to intraoperative mis-manipulations. We introduce data augmentation and transfer learning methods to address the issue of small datasets. The mAP of YOLOv7_CA is 69.34%, with improvements of 3.22%, 5.59%, and 6.10% in mAP, Precision, and Recall, respectively, compared to YOLOv7. Experimental results demonstrate the network's excellent performance in detecting critical intraoperative anatomical landmarks, thereby achieving real-time assistance for surgeons. It also provides new ideas for the application of deep learning in the surgical field of vision.