The VISION-AI Trial: protocol for a pragmatic randomized controlled non-inferiority trial comparing artificial intelligence-guided colonoscopy to pancolonic chromoendoscopy for neoplasia detection in adults with colorectal inflammatory bowel disease.

内容摘要

Current guidelines recommend pancolonic chromoendoscopy (pCE) over white light endoscopy (WLE) alone for colorectal neoplasia (CRN) detection in individuals with inflammatory bowel diseases (IBD). However, these techniques are poorly adopted due to technical and logistical limitations. Artificial intelligence-based computer-aided detection (CADe) is a promising new technology integrated into modern endoscopy platforms that has been shown to increase CRN detection in the non-IBD population. We aim to compare CADe-guided colonoscopy to chromoendoscopy in a randomized controlled trial (RCT) in individuals with IBD undergoing surveillance colonoscopy. This is a pragmatic, multicenter, parallel-group, open-label, non-inferiority RCT comparing CADe-guided colonoscopy (intervention) to pCE (virtual or dye spray) (standard of care). Consenting adults (≥ 18 years) with longstanding (≥ 8 years) IBD (or any duration with primary sclerosing cholangitis) involving ≥ 1/3 of the colorectum, in clinical remission and undergoing surveillance colonoscopy, will be included. Those who undergo a high-quality colonoscopy (adequate preparation, minimal inflammation, complete intubation) will be included in the primary analysis. Patients will be centrally randomized by permuted blocks, stratified by endoscopist. The anticipated CRN rate is 15% in both groups. To demonstrate non-inferiority with a 7.5% margin, at a one-sided alpha of 2.5% and 80% power, and allowing for a 5% loss of eligibility, a total of 752 participants are required. The primary outcome will be visible CRN detected at colonoscopy; secondary outcomes include the mean number of CRN, proportion of individuals with high-risk CRN, procedural time, adverse events, and endoscopist acceptance, with per-protocol (primary) and intention-to-treat (secondary) analyses. This protocol follows the SPIRIT 2025 guidelines. AI has shown potential in enhancing CRN detection, but current AI systems are not specifically trained for IBD-related lesions, limiting effectiveness. This multicenter non-inferiority RCT will, for the first time, directly compare CADe-guided colonoscopy to pCE for CRN detection in individuals with IBD. This approach may increase adoption of AI-enhanced endoscopy and improve cancer prevention strategies. Australian New Zealand Clinical Trials Registry (ANZCTR) registration number: ACTRN12626000102370. At the time of initial submission (November 2025), recruitment for this study had not started. Participant enrollment began in April 2026, and as of June 8, 2026, 19 out of the target 752 participants have been recruited. The study is currently in the active recruitment phase and has not yet transitioned to follow-up only.