P0392Narrow Band Imaging versus Artificial Intelligence for colonic surveillance in Inflammatory Bowel Disease: a prospective, randomized, crossover study. Results from the CLEAR-IBD (Colorectal Lesion Evaluation Assisted by Real-time AI in Inflammatory Bowel Disease) trial

Background Colorectal cancer (CRC) is a complication of long-standing ulcerative colitis (UC) and Crohn’s disease (CD). Guidelines recommend different colonoscopy modalities for CRC surveillance, with dye-based chromoendoscopy, narrow-band imaging (NBI), and high definition white light endoscopy showing the highest neoplasia detection rate. Artificial intelligence (AI) systems can assist real-time polyp and adenoma detection, but studies have so far focused only on non-inflammatory bowel disease (IBD) populations. Methods This prospective, single-centre, cross-over study compared the ENDO-AID CADe system (OLYMPUS) with NBI for detecting colonic neoplastic lesions, with the primary outcome being the neoplasia miss rate (NMR). Adenocarcinoma, serrated adenoma, tubular adenoma, and any dysplasia were considered neoplastic. Consecutive UC or colonic CD patients meeting CRC surveillance criteria at the Gastroenterology Outpatient Clinic, Hospital Universitari Vall d’Hebron, Barcelona, were enrolled. Assuming a 1.5% discordance between techniques, 80% power, and α = 0.05, the required sample size was estimated at 60 patients. Each patient was examined by two experienced endoscopists. Endoscopist 1 evaluated the entire colon using either AI or NBI, documented findings, and acted accordingly – resection or biopsy. Endoscopist 2, blinded to the previous findings, re-examined the entire colon with the alternate method, documented additional findings, and treated any lesions missed by Endoscopist 1. The two endoscopists alternated between AI and NBI. Results To date, 48 patients with a median age of 52 years (range 29-84) have been recruited. Four of them were excluded due to inadequate bowel preparation. Thirty-one patients had UC, 16 of them extensive. Seventeen patients were on biologic therapy. No differences were found in terms of baseline characteristics between the patients who underwent AI and NBI as the first examination. Twenty-two dysplastic lesions were found in 11 patients. Fifteen of the lesions were detected at the first examination and 7 at the second examination. In a per-lesion analysis, the NMR was 5/11 (45%) for NBI and 2/11 (18%) for AI (p = 0.17). In a per-patient analysis, the NMR was 4/8 (50%) for NBI and 2/7 (29%) for AI (p = 0.61). The exploration time for NBI and AI was 09:16 and 08:43, respectively (p = 0.805). Histologic inflammatory activity was assessed from biopsies of quiescent mucosa. Activity was detected in 5 patients, with one dysplastic lesion found. Conclusion After recruiting nearly three quarters of the planned sample size, AI appears to be an equally effective alternative to NBI for the detection of IN in patients with IBD. The estimated completion date for recruitment and data analysis is December 16th, 2025. Conflict of interest: Mr. Serra-Ruiz, Xavier: This project was funded by the Research Initiation Grant from the Catalan Society of Gastroenterology (SCD) in 2025. Pigrau, Marc: No conflict of interest Castillo Cejas, María Dolores: No conflict of interest Céspedes Martínez, Elena: No conflict of interest Mayorga Ayala, Luis Fernando: No conflict of interest Herrera De Guise, Claudia Maria: No conflict of interest Robles Alonso, Virginia: No conflict of interest Lastiri, Ernesto: No conflict of interest Dot, Joan: No conflict of interest Borruel, Natalia: No conflict of interest