AI in gastroenterology and hepatology A narrative review of applications, evidence maturity, and governance implications for Latin America

Artificial intelligence applications in gastroenterology and hepatology have been evaluated in gastrointestinal endoscopy, image interpretation, liver fibrosis risk assessment, hepatocellular carcinoma diagnosis, and initial referral assessment (authorization or request for additional information). This whitepaper summarizes a narrative corpus of 18 selected sources: ten clinical or technical sources, five implementation and governance frameworks, and three institutional sources relevant to regional digital health and health AI governance. The evidence is uneven. Computer-aided colonoscopy is the most mature domain in the selected corpus, but reported endpoints largely concern detection and workflow rather than mortality, cancer incidence, or broad clinical outcomes. Regional evidence from Brazil and Colombia is valuable but task-specific, mostly single-center or retrospective, and insufficient to infer general transferability across Latin America. Prospective multicenter evidence also supports assisted capsule-endoscopy reading for suspected small-bowel bleeding. Evidence for liver fibrosis, hepatocellular carcinoma diagnosis, and referral gatekeeping requires task-specific interpretation and local validation. Colonoscopy guidance differs between professional societies, reflecting uncertainty about patient-important benefits and burdens. For Latin America, responsible implementation should begin with bounded tasks, representative data, explicit clinical actions, privacy and interoperability safeguards, and prospective monitoring of workload, inequity, false positives, false negatives, and institutional capacity to respond.