AI-assisted colonoscopy improves adenoma detection

A study including more than 3,000 participants showed a significantly better adenoma detection rate when colonoscopy was assisted by a self-learning Artificial Intelligence (AI) system. The AI-assisted colonoscopy was superior, independent of the type or location of the adenoma – even in patients with inadequate preparation for the procedure.

Prof. Joseph Sung (Lee Kong Chian School of Medicine, Singapore) pointed out in his presentation that AI-assisted colonoscopy has been shown to improve polyp detection and characterisation in colonoscopy [1]. However, data from large-scale, multicentre, randomised-controlled trials (RCTs) are still missing. This was the rationale for a multicentre, single-blind RCT in China including Hong Kong.

Recruited participants were randomised (1:1) to receive either AI-assisted or conventional coloscopy. Study participants were asymptomatic, 45–75 years old, and were undergoing colorectal carcinoma (CRC) screening either by direct screening colonoscopy or by a faecal immunochemical test (FIT)-based screening programme. IBD or a colonoscopy within 10 years were exclusion criteria. The AI system used was a deep convolution neural network trained and validated by 112,199 still images and a separate dataset of 21 colonoscopy videos. According to an independent validation, the system achieved both sensitivity and specificity of >89% in recognising polyps. “This AI device has an ordinary desktop computer size and can fit in any endoscopy tower,” said Prof. Sung. In the trial, high-definition colonoscopes, endoscopy processors, and monitors were used. However, neither electronic image enhancing function nor mucosal exposure devices were allowed.

The primary study outcome was the overall adenoma detection rate (ADR), the proportion of patients with at least one colorectal adenoma detected among all patients examined by an endoscopist. All adenomas (non-advanced adenomas, advanced adenoma, and CRC) were assessed.

Altogether, 3,059 participants were included in the intention-to-treat analysis. The ADR was significantly better in the AI group compared with the conventional colonoscopy (606 vs 499; P<0.001). “Even those with inadequate preparation benefitted from the AI-assisted colonoscopy,” Prof. Sung emphasised. A significant benefit was seen in both advanced and non-advanced, large and small adenoma, and proximal and distal adenoma. The only disadvantage was that the procedure itself took significantly longer with AI.

“AI should become standard of class in colonoscopy, because it does not only improve adenoma detection in general but also advanced adenoma, and thus can help in cancer prevention and will improve the outcome of patients,” Prof. Sung concluded.

1. Sung JJ, et al. Artificial intelligence-assisted colonoscopy for colorectal cancer screening: a multicenter randomized controlled trial. Lecture 414, Digestive Disease Week 2022, 21–24 May, San Diego, CA, USA.

Copyright ©2022 Medicom Medical Publishers



Posted on 14 June 2022