This Arkangel AI whitepaper is a narrative review of 18 selected sources on artificial intelligence in gastroenterology and hepatology: ten clinical or technical studies, five implementation and governance frameworks, and three institutional sources. It covers colonoscopy, upper and capsule endoscopy, liver fibrosis, hepatocellular carcinoma and specialist referral review, and classifies each task by evidence maturity, from development to routine implementation.
Computer-aided colonoscopy is the most mature application, but its endpoints are detection and workflow, not mortality or cancer incidence, and the 2025 AGA and ESGE guidelines differ on it. Regional evidence is limited: a Brazilian study of 305 screening colonoscopies found adenoma detection of 48.2% with AI versus 57.0% without, with no significant difference, and a Colombian system reached 85.5% macro sensitivity and 98.8% specificity classifying 13 gastric regions, which is not cancer detection. A Brazilian referral model reached 0.716 accuracy across five specialties. The authors recommend starting with bounded tasks, local validation and human review. It is a whitepaper, not peer reviewed.
