Artificial Intelligence in Gastroenterology: Current Applications, Challenges, and Future Directions

AI (artificial intelligence) including ML (machine learning), DL (deep learning) and NLP(Natural language processing) is rapidly entering gastroenterology. Colonoscopy Real-Time Polyp Detection (CADe), Colonoscopy Real-Time Polyp Characterization (CADx), AI-assisted IBD (inflammatory bowel disease) scoring, radiomics for liver disease, and Large language models (LLMs) for documentation are moving toward deployment. This review focuses on (1) luminal endoscopy and IBD, where evidence is most developed, and (2) cross-cutting issues of validation, regulation, workflow, economics and medico-legal considerations. A narrative review (Pub Med, Scopus, Web of Science; Jan 2010–Aug 2025) using predefined terms related to AI, ML, DL and gastroenterology was done. Data extracted included algorithm type, clinical setting, performance metrics, and patient-related outcomes. AI demonstrated high diagnostic performance across multiple domains. Deep learning–based polyp detection systems achieved pooled sensitivities of 92–98% and specificities of 85–93% for real-time colonoscopy. Algorithms for Barrett’s esophagus dysplasia detection reported accuracies up to 94%, while hepatocellular carcinoma prediction models integrating imaging and laboratory parameters reached AUCs of 0.86–0.93. NLP tools improved risk stratification accuracy in endoscopy and pathology reporting by 18–24% compared with manual review. Clinical decision-support systems for IBD and chronic liver disease enhanced treatment selection accuracy by 15–20% in prospective pilot studies. AI is transitioning to early clinical use, but most tools remain developmental, biased and poorly validated. Clear governance, external validation, health-economic assessment, and a pragmatic framework for “when to turn AI on” is needed.