Utility of artificial intelligence in colorectal cancer screening colonoscopies for the detection (CADe) and histological prediction (CADx) of polyps: a randomized clinical trial

Le résumé fourni par la source

Aims To evaluate the utility of artificial intelligence (AI) device for the detection (CADe) and characterization (CADx) of polyps during screening colonoscopy, given that the increase in the lesion detection rate has been associated with a reduction in the risk of interval cancer. Methods Randomized clinical trial conducted in a Spanish hospital with patients between 50-74 years of age undergoing colonoscopy after a positive fecal occult blood test on colorectal cancer (CRC) screening between October 2023 and October 2024. The intervention group underwent colonoscopy assisted with AI device while the control group underwent conventional colonoscopy. Advanced neoplasia is defined as one of the following: advanced adenoma (≥ 10 mm, villous and/or with high-grade dysplasia), advanced serrated (≥ 10 mm and/or with dysplasia) or CRC. Quantitative variables were compared with t test; categorical variables with chi-square. Significance level α=0.05. Analysis was performed using SSPS software [ 1 ] [ 2 ] [ 3 ] [ 4 ] [ 5 ]. Results Out of the 301 patients analyzed (156 patients in the intervention group and 145 in the control group), 51.8% were men with a mean age of 62.6±7 years. A total of 546 polyps were resected in 211 patients (70.1%) and 15 CRCs were diagnosed (4.9% of the total number of patients recruited). No significant differences were found in the detection rate of polyps, adenomas (60.3% with AI and 69.7% without AI, p =0.088), serrated lesions (21.8% with AI and 22.8% without AI, p =0.841) or advanced neoplasia (37.8% with AI and 40% without AI, p =0.698). Moreover, no significant differences were found in the mean number of lesions detected per colonoscopy, in the subanalysis according to location, size and morphology of the polyps or in the mean withdrawal time between groups. In both groups, an accuracy in histological prediction of the polyp greater than 80% was observed. With AI, specificity (66.7% with AI and 52.4% without AI) and positive predictive value (PPV) were gained at the cost of losing sensitivity (88.5% with AI and 92.5% without AI). Conclusions CADe did not improve the identification of any of the lesions studied either by location, size or morphology. One of the possible reasons which could explain the non-superiority of AI in adenoma detection is the higher rate (> 60%) found in the control group than in other CRC screening studies. CADx presented greater adenoma specificity and PPV than the endoscopist. The main limitation of the study is the sample size; it is still in the recruitment phase. The results show the need to improve this technology. Publication History Article published online: 27 March 2025 © 2025. European Society of Gastrointestinal Endoscopy. All rights reserved. Georg Thieme Verlag KG Oswald-Hesse-Straße 50, 70469 Stuttgart, Germany

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.