Abstract
BACKGROUND AND OBJECTIVE: Barrett's esophagus (BE) is the principal precursor lesion for esophageal adenocarcinoma (EAC), a malignancy with rising incidence and poor survival when diagnosed at advanced stages. Current screening and surveillance strategies rely on endoscopy and random biopsies, which are invasive, resource-intensive, and prone to sampling error. Artificial intelligence (AI) has emerged as a promising tool to enhance early detection, risk stratification, and surveillance efficiency in BE. This narrative review summarizes contemporary AI applications in BE management, evaluates their diagnostic and predictive performance, and discusses barriers to clinical adoption. METHODS: A narrative literature review was conducted using PubMed, Embase, Scopus, Web of Science, and Cochrane. ClinicalTrials.gov and World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) were checked for ongoing trials and Good Scholar was used for citation chasing to identify peer-reviewed studies up to June 2025. Eligible studies evaluated AI-based approaches for BE screening, dysplasia detection, risk prediction, or surveillance optimization using vision-based or non-vision-based models. KEY CONTENT AND FINDINGS: Vision-aided AI systems, particularly convolutional neural networks applied to high-definition white-light endoscopy and image-enhanced endoscopy, demonstrate sensitivities approaching 90% for dysplasia and early EAC detection. Non-vision-based models leveraging electronic health records, biomarkers, and histopathology achieve area under the receiver operating characteristic curve (AUROC) values up to 0.84 for predicting BE or EAC risk. Multimodal approaches integrating clinical, endoscopic, and molecular data show promise for personalized surveillance strategies. However, challenges remain, including limited external validation, algorithm transparency, data bias, workflow integration, and cost considerations. CONCLUSIONS: AI has the potential to transform BE care by improving early detection and enabling risk-adapted surveillance. Multicenter validation, explainable models, and cost-effectiveness analyses are essential before widespread clinical implementation.