IDDF2021-ABS-0176 Influence of artificial intelligence on colonoscopy in different examination periods

Background Influenced by various factors, endoscopists have shown differences in the detection rate of adenomas (ADR) during different periods of endoscopy every day. The purpose of this study is to observe whether AI-assisted endoscopy can bring an improvement in ADR and whether the increase is consistent in different periods. Methods We established a real-time AI-assisted colonoscopy system. The core of the system is a deep neural network for object detection, YOLOV3. The processing speed of this model was 60FPS. When polyps appear in the video of colonoscopy, the system can give a real-time warning on the monitor of the endoscope. We randomly divided the enrolled patients into four groups, which were arranged to receive colonoscopy in the first hour of the morning and the last hour of the morning, respectively. Meanwhile, we differentiated the AI-assisted group and the non-AI-assisted group. In order to remove the bias caused by the experience of the endoscopist, the trial was performed by two colonoscopists with comparable experience. Finally, the ADRs among the groups were compared. Results With or without AI, endoscopy scheduled in the first hour of the morning had higher ADRs than endoscopy scheduled in the last hour of the morning (23% Vs 18%). AI assistance had a positive effect on the increase in ADR, but it had a greater effect on endoscopy scheduled in the latter period. (0.5% Vs 4.2%). Conclusions The change of ADR detection rate in different time periods was mainly due to the change of attention of endoscopists, which had little correlation with the level of endoscopic operation. The use of artificial intelligence assistance can help endoscopic doctors improve their attention and ensure the stability of the ADR in different periods.