Journal Article

Development of a Novel Artificial Intelligence System for Laparoscopic Hepatectomy

KOBAYASHI, NAO, SHIBATA, HIDEKI, AOKI, TAKESHI, WATANABE, MAKOTO, MATSUDA, KAZUHIRO, SAITO, KAZUHIKO, KUSANO, TOMOKAZU, YAMAZAKI, TATSUYA, TOMIOKA, KODAI, YAMAZAKI, KIMIYASU, FUJIMORI, AKIRA, HIRAI, TAKAHITO, KUMAZU, YUTA, ENAMI, YUTA, TASHIRO, YOSHIHIKO

2023

Overview

Laparoscopic hepatectomy (LH) requires accurate visualization and appropriate handling of hepatic veins and the Glissonean pedicle that suddenly appear during liver dissection. Failure to recognize these structures can cause injury, resulting in severe bleeding and bile leakage. This study aimed to develop a novel artificial intelligence (AI) system that assists in the visual recognition and color presentation of tubular structures to correct the recognition gap among surgeons.BACKGROUND/AIMLaparoscopic hepatectomy (LH) requires accurate visualization and appropriate handling of hepatic veins and the Glissonean pedicle that suddenly appear during liver dissection. Failure to recognize these structures can cause injury, resulting in severe bleeding and bile leakage. This study aimed to develop a novel artificial intelligence (AI) system that assists in the visual recognition and color presentation of tubular structures to correct the recognition gap among surgeons.Annotations were performed on over 350 video frames capturing LH, after which a deep learning model was developed. The performance of the AI was evaluated quantitatively using intersection over union (IoU) and Dice coefficients, as well as qualitatively using a two-item questionnaire on sensitivity and misrecognition completed by 10 hepatobiliary surgeons. The usefulness of AI in medical education was qualitatively evaluated by 10 medical students and residents.PATIENTS AND METHODSAnnotations were performed on over 350 video frames capturing LH, after which a deep learning model was developed. The performance of the AI was evaluated quantitatively using intersection over union (IoU) and Dice coefficients, as well as qualitatively using a two-item questionnaire on sensitivity and misrecognition completed by 10 hepatobiliary surgeons. The usefulness of AI in medical education was qualitatively evaluated by 10 medical students and residents.The AI model was able to individually recognize and colorize hepatic veins and the Glissonean pedicle in real time. The IoU and Dice coefficients were 0.42 and 0.53, respectively. Surgeons provided a mean sensitivity score of 4.24±0.89 (from 1 to 5; Excellent) and a mean misrecognition score of 0.12±0.33 (from 0 to 4; Fail). Medical students and residents assessed the AI to be very useful (mean usefulness score, 1.86±0.35; from 0 to 2; Excellent).RESULTSThe AI model was able to individually recognize and colorize hepatic veins and the Glissonean pedicle in real time. The IoU and Dice coefficients were 0.42 and 0.53, respectively. Surgeons provided a mean sensitivity score of 4.24±0.89 (from 1 to 5; Excellent) and a mean misrecognition score of 0.12±0.33 (from 0 to 4; Fail). Medical students and residents assessed the AI to be very useful (mean usefulness score, 1.86±0.35; from 0 to 2; Excellent).The novel AI presented was able to assist surgeons in the intraoperative recognition of microstructures and address the recognition gap among surgeons to ensure a safer and more accurate LH.CONCLUSIONThe novel AI presented was able to assist surgeons in the intraoperative recognition of microstructures and address the recognition gap among surgeons to ensure a safer and more accurate LH.