Artificial intelligence-guided computer-aided intervention system for laparoscopic rectal cancer surgery

Background: Improving surgical quality and safety requires the rapid and stable identification of the safe dissection plane and important anatomical structures. This study aimed to develop an artificial intelligence (AI)-guided computer-aided intervention (CAI) system for laparoscopic rectal cancer surgery. Methods: A total of 8756 images depicting the Holy Plane and 6527 images depicting the pelvic autonomic nerve (PAN) from 386 surgical videos of laparoscopic rectal cancer surgery were manually annotated under the supervision of senior surgeons. The deep-learning models were developed for automatic identification of the Holy Plane and PAN, and the average Dice similarity coefficient (DSC), Recall, and pixel accuracy (PA) were used to evaluated the model performance. Subsequently, the AI-guided CAI system, which utilizes the Holy Plane and PAN as dual identification landmarks, was constructed. Patients who underwent CAI system-assisted laparoscopic rectal cancer surgery were assigned to the CAI-assisted group (CA group), and those without the CAI system were assigned to the control group (CL group). The surgical and functional outcomes of patients were recorded. Results: The U-Net was selected for the automatic identification of the Holy Plane, achieving remarkable performance with DSC, Recall, and PA values as high as 0.898, 0.811, and 0.918, respectively. The ResNet50-U-Net was developed for automatic identification of PAN, demonstrating a satisfactory performance with DSC, Recall, and PA values of 0.815, 0.794, and 0.823, respectively. The AI-guided CAI could achieve dual identification of the Holy Plane and PAN during laparoscopic rectal cancer surgery. Compared with the CL group, patients in the CA group had significantly less surgical blood loss, a lower complication rate and a lower incidence of male sexual dysfunction. Conclusion: It is technically feasible and safe for surgeons to perform AI-guided CAI-assisted laparoscopic rectal cancer surgery, which is expected to reduce variability in surgical quality.