Artificial intelligence-based model for identifying anatomical and surgical structures in the videos of laparoscopic left hemicolectomy.

BACKGROUND: Precise surgical procedures are critical to improving the survival outcomes for colon cancer patients. Currently, there is no dedicated intraoperative navigation system available for performing precise surgery. OBJECTIVE: We aimed to innovate an artificial intelligence-based model for identifying anatomical structures and surgical instruments during laparoscopic left hemicolectomy, to assist surgeons in performing the surgery with greater precision. DESIGN AND SETTINGS: A total of 7474 images, extracted from the operation videos of 56 patients, were included. Anatomical structures, including the fascia, avascular dissection plane, inferior mesenteric vessels and their branches, veins, ureter, reproductive blood vessels, pancreas, and nerves around the inferior mesenteric artery, were analyzed in each image. Surgical instruments, including clips, energy devices, and tissue forceps, were also analyzed. The Mask2Former model was trained using the annotated images. PATIENTS: This was a multicenter study involving 56 randomly selected patients who had undergone laparoscopic or robotic left hemicolectomy between September 26, 2022, and June 25, 2025, at Sun Yat-Sen Memorial Hospital, Qilu Hospital of Shandong University and Guangdong Second Provincial General Hospital. MAIN OUTCOME MEASURES: Intersection over Union (IoU), precision, recall and F1 score of the model. RESULTS: A total of 7474 images were included; 6727 images were selected as a training set and 747 images as validation set. The mean precision of this model was 0.8643, the mean IoU was 0.7768, and the mean recall was 0.8756. LIMITATIONS: First, the quality of the videos was inconsistent due to the different recording systems. Second, we should add images of suboptimal dissections to make this technology a more effective navigation aid during difficult dissections. Third, this model should be tested in real-time surgery navigation. CONCLUSIONS: The artificial intelligence-based model can accurately identify anatomical structures and surgical instruments in laparoscopic left hemicolectomy.