Evaluation in real-time use of artificial intelligence during colonoscopy to predict relapse of ulcerative colitis: a prospective study

Background and Aims: The use of artificial intelligence (AI) during colonoscopy is attracting attention as an endoscopist-independent tool to predict histologic disease activity of ulcerative colitis (UC). However, no study has evaluated the real-time use of AI to directly predict clinical relapse of UC. Hence, it is unclear whether the realtime use of AI during colonoscopy helps clinicians make real-time decisions regarding treatment interventions for patients with UC. This study aimed to establish the role of real-time AI in stratifying the relapse risk of patients with UC in clinical remission. Methods: This open-label, prospective, cohort study was conducted in a referral center. The cohort comprised 145 consecutive patients with UC in clinical remission who underwent AI-assisted colonoscopy with a contact microscopy function. We classified patients into either the Healing group or Active group based on the AI outputs during colonoscopy. The primary outcome measure was clinical relapse of UC (defined as a partial Mayo score 2) during 12 months of follow-up after colonoscopy. Results: Overall, 135 patients completed the 12-month follow-up after AI-assisted colonoscopy. AI-assisted colonoscopy classified 61 patients as the Healing group and 74 as the Active group. The relapse rate was significantly higher in the AI-Active group (28.4% [21/74]; 95% confidence interval, 18.5%-40.1%) than in the AI-Healing group (4.9% [3/61]; 95% confidence interval, 1.0%-13.7%; P < .001). Conclusions: Real-time use of AI predicts the risk of clinical relapse in patients with UC in clinical remission, which helps clinicians make real-time decisions regarding treatment interventions.

Automated summary

This study evaluated the use of real-time artificial intelligence during colonoscopy to predict the risk of clinical relapse in patients with ulcerative colitis (UC) in clinical remission. The results showed that real-time AI can help clinicians make treatment interventions decisions.