Artificial Intelligence Recognition System of Pelvic Autonomic Nerve During Total Mesorectal Excision.

Abstract

Background: The preservation of the pelvic autonomic nervous system in total mesorectal excision remains challenging to date. The application of laparoscopy has enabled visualization of fine anatomical structures; however, the rate of urogenital dysfunction remains high.

Objective: To establish an artificial intelligence neurorecognition system to perform neurorecognition during total mesorectal excision.

Design: This retrospective study.

Setting: The study was conducted at a single hospital.

Patients: Intraoperative images or video screenshots of rectal cancer patients admitted to the Department of Gastrointestinal Surgery, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, between January 2016 and December 2023 were retrospectively collected.

Main outcome measure: Mean intersection over union, precision, recall, and F1 of the model.

Results: A total of 1424 high-quality intraoperative images were included in the training group. The proposed model was obtained after 700 iterations. The mean intersection over union was 0.75, and it slowly increased with an increase in training time. The precision and recall of the nerve category were 0.7494 and 0.6587, respectively, and the F1 was 0.7011. From the video prediction, we can observe that the model achieves a high accuracy rate, which could facilitate effective neurorecognition.

Limitation: This was a single-center study.

Conclusion: The artificial intelligence model for real-time visual neurorecognition in total mesorectal excision was successfully established for the first time in China. Better identification of these autonomic nerves should allow for better preservation of urogenital function, but further research is needed to validate this claim. See Video Abstract.