AI-assisted colonoscopy improves polyp detection

Artificial intelligence (AI)–assisted colonoscopy improves adenoma detection in asymptomatic individuals undergoing colorectal cancer (CRC) screening, a multicentre randomized controlled trial (RCT) has shown.

The RCT included 3,059 asymptomatic individuals 45–75 years of age undergoing CRC screening by direct colonoscopy or faecal immunochemical test, who were recruited from six university-affiliated endoscopy centres in Hong Kong, Beijing (two centres), Inner Mongolia, Jilin and Xiamen, China, between November 2019 and August 2021. The participants were randomized to undergo AI-assisted colonoscopy (n=1,519) with an AI polyp detection system that provides real-time notification on the same monitor of the endoscopy system, or conventional colonoscopy (CC; n=1,540). [Clin Gastroenterol Hepatol 2022;S1542-3565(22)00673-5]

The primary outcome of overall adenoma detection rate (ADR) was significantly higher in the AI vs CC group, at 39.9 percent vs 32.4 percent (p<0.001). The mean number of adenoma detected per colonoscopy was 0.59 in the AI group vs 0.45 in the CC group (0.59 vs 0.45; p<0.001).

AI-assisted colonoscopy significantly improved ADR for both expert (42.3 percent vs 32.8 percent; p<0.001) and non-expert (37.5 percent vs 32.1 percent; p=0.023) endoscopists vs CC.

Among patients with inadequate bowel preparation (Boston Bowel Preparation Scale [BBPS] score <6 or any segmental BBPS <2; 18.5 percent vs 16.3 percent in the AI vs CC group), ADR was significantly higher in the AI vs CC group (40.7 percent vs 29.5 percent; p=0.039).

In terms of polyp characteristics, AI-assisted colonoscopy demonstrated significantly higher rates of detection of both non-advanced adenoma (32.3 percent vs 26.7 percent; p=0.001) and advanced adenoma (6.6 percent vs 4.9 percent; p=0.041) compared with CC.

Significantly improved detection rates were also reported in the AI vs CC group for adenoma <5 mm (16.5 percent vs 11.5 percent; p<0.001), adenoma ≥10 mm (6.5 percent vs 4.7 percent; p=0.033), non-pedunculated adenoma (27.6 percent vs 21.8 percent; p<0.001), adenoma in the proximal colon (28.4 percent vs 23.8 percent; p=0.004), and adenoma in the distal colon (10.6 percent vs 7.7 percent; p=0.006).

However, no significant between-group differences were found for rates of detection of CRC (0.9 percent vs 0.8 percent; p=0.819) and sessile serrated lesions (1.1 percent vs 1.3 percent; p=0.65).

In multivariate analysis, AI assistance was independently associated with higher ADR (adjusted odds ratio [aOR], 1.302; 95 percent confidence interval [CI], 1.106–1.532; p=0.001). Other factors independently associated with higher ADR included male gender (aOR, 1.648; 95 percent CI, 1.394–1.948; p<0.001), older age groups (60–64 years: aOR, 1.527; 95 percent CI, 1.156–2.017; p=0.003) (65–69 years: aOR, 2.214; 95 percent CI, 1.653–2.964; p<0.001) (70–75 years: aOR, 2.008; 95 percent CI, 1.393–2.894; p<0.001), alcohol drinking (aOR, 1.399; 95 percent CI, 1.061–1.845; p=0.017), BMI (aOR, 1.026; 95 percent CI, 1.000–1.052; p=0.049), expert endoscopist (aOR, 1.213; 95 percent CI, 1.030–1.428; p=0.020), and endoscope withdrawal time (aOR, 1.179; 95 percent CI, 1.156–1.202; p<0.001).