AI tools to improve endoscopic training in early-stage GI cancer management

Researchers from the Chinese University of Hong Kong (CUHK) have demonstrated the potential application of two artificial intelligence (AI) tools, ENDO-AID and AI-Endo, for training endoscopists in early-stage gastrointestinal (GI) cancer management.

ENDO-AID for adenoma detection

ENDO-AID is a real-time automatic computer-aided polyp detection system trained on about 12 million images and videos from Japan and other countries. To assess adenoma detection rate (ADR) among endoscopists in training using ENDO-AID, the CUHK researchers conducted a single-blind, parallel-group, superiority study involving 22 endoscopists with <3 years of training and experience of <500 procedures. A total of 766 adults were randomized to receive ENDO-AID colonoscopy (n=386) or standard colonoscopy (n=380) between April 2021 and July 2022. [Clin Gastroenterol Hepatol 2023;doi:10.1016/j.cgh.2023.10.019]

Results showed a 41 percent increase in ADR with ENDO-AID vs standard colonoscopy (57.5 percent vs 44.5 percent; adjusted relative risk [aRR], 1.41; 95 percent confidence interval, 1.17–1.72; p<0.001) among the endoscopists in training. Notably, novice endoscopists (<200 procedures) achieved a greater relative increment in ADR with ENDO-AID vs standard colonoscopy (aRR, 1.58; p=0.015) than intermediate endoscopists (200–500 procedures; aRR, 1.36; p=0.009).

“AI technology [ie, ENDO-AID] can enhance adenoma detection by providing real-time guidance and systematic colonoscopy training to endoscopists,” highlighted Dr Louis Lau of the Department of Medicine and Therapeutics, CUHK.

AI-Endo for endoscopic submucosal dissection

AI-Endo is a deep learning–based platform for intelligent real-time recognition of four surgical phases (ie, marking, injection, dissection, idle) in endoscopic submucosal dissection (ESD) workflows. Developed by the CUHK researchers, the platform was built upon a dataset of >2 million labelled frames extracted from 47 high-quality ESD videos that covered operations involving different lesion sizes, locations, and surgical instruments. [Nat Commun 2023;14:6676]

In the developmental validation dataset, AI-Endo demonstrated promising performance in both overall and each surgical phase of ESD. Its high performance was reaffirmed with external validation datasets involving surgeons with diverse ESD experience, using various operation methods and endoscopy systems, and across centres in China and Germany.

In preclinical studies, AI-Endo seamlessly integrated with endoscopic system. During surgical training sessions, junior surgeons successfully perform ESD with the AI-Endo–integrated system on porcine models in both ex vivo and in vivo settings. Additionally, AI-Endo provided data derived from surgical phase recognition results, facilitating postoperative assessment and review.

The researchers believe that application of AI can assist junior surgeons in improving ESD technique. “AI-Endo can serve as a reliable tool for reducing the duration of [endoscopic] training,” commented Dr Hon-Chi Yip of the Department of Surgery, CUHK.