Efficacy of artificial intelligence in detecting precancerous lesions among a colorectal cancer screening population undergoing surveillance colonoscopies in Italy: a randomized control trial (GENIAL-CO Follow-Up study)

Background and aims Computer-aided detection (CADe) has demonstrated improved adenoma detection rate (ADR) in screening populations; however, its value in post-polypectomy surveillance colonoscopy remains unclear. This study evaluated CADe efficacy in enhancing detection metrics during surveillance colonoscopy within an organized colorectal cancer (CRC) screening program. Methods In this multicenter randomized controlled trial, patients aged 50–74 scheduled for surveillance colonoscopy were randomized 1:1 to CADe-assisted or standard colonoscopy (SC). The primary endpoint was ADR. Secondary endpoints included advanced ADR, adenomas per colonoscopy (APC), sessile serrated lesion detection rate (SSLDR), and withdrawal time (WT). Results Of 973 enrolled patients, 914 were included after excluding 59 with inadequate bowel preparation (CADe: 455 vs SC: 459). ADR and AADR did not differ significantly between groups (66.6%vs 61.0%, p=0.160; 7.3% vs 6.3%, p=0.263). APC was significantly higher in the CADe group (1.80±2.19 vs 1.32±1.75, p < 0.001), while SSLDR was comparable (17.1%vs 14.2%, p=0.154). Mean WT was significantly longer with CADe (16.35±8.32 vs 14.52±7.24 min, p < 0.001). Conclusions CADe-assisted colonoscopy did not improve ADR or AADR in a CRC organized screening population undergoing surveillance colonoscopy, while significantly increasing WT.