By using artificial intelligence (AI) in a colonoscopy, colorectal neoplasms were missed about two times less often than during a standard colonoscopy. Especially for small, subtle lesions, AI can reduce perception errors, Wallace et al. write in Gastroenterology.
The research group wanted to find out whether AI, consisting of ‘deep learning computer aided diagnosis endoscopy’, can help to reduce the number of polyps missed and thus contribute to the prevention of colorectal cancer. In 8 centers in Italy, the United Kingdom and the United States, participants in colorectal cancer screening programs were randomized 1:1. One group first had a colonoscopy with AI and then a colonoscopy without AI (n = 116) and in the other group the order was reversed (n = 114). The proportion of missed adenomas was calculated as the number of lesions detected during the second colonoscopy divided by the total number of lesions found during both colonoscopies. In addition, the proportion of false negative patients (no lesion during the first colonoscopy and at least 1 during the second) was calculated.
The proportion of missed adenomas was 15.5% (38/246) in the first AI colonoscopy group and 32.4% (80/247) in the standard colonoscopy group (odds ratio corrected for endoscopist, age, gender and indication for colonoscopy: 0.38; 95% CI 0.23-0.62). The mean (±SD) number of adenomas found during the second colonoscopy was lower in the AI-first group (0.33 ± 0.63) than in the standard-first group (0.70 ± 0.97; p < 0.001). The proportion of false negatives was 6.8% in the AI-first group and 29.6% in the standard-first group (adjusted odds ratio 0.17; 95% CI 0.05-0.67). The number of adverse events did not differ between the two groups.
Bron:
Wallace MB, Sharma P, Bhandari P, et al. Impact of Artificial Intelligence on Miss Rate of Colorectal Neoplasia. Gastroenterology. 2022;163:295-304. e5.