Artificial Intelligence Applications in Endoscopic Sleeve Gastroplasty: A Systematic Review of Preliminary Evidence

Endoscopic Sleeve Gastroplasty (ESG) is an established minimally invasive bariatric intervention, while artificial intelligence (AI) has been increasingly explored across clinical and procedural domains. However, evidence integrating AI into ESG remains limited and fragmented.

To systematically identify and critically interpret the clinical, technical, and educational applications of AI in ESG.

This systematic review followed PRISMA guidelines and was prospectively registered in PROSPERO. Searches were conducted in PubMed, Embase, Scopus, Web of Science, and the Cochrane Library. Original studies applying AI to ESG in clinical practice, simulation, or education were included. Given marked heterogeneity, results were synthesized qualitatively with emphasis on clinical interpretability.

Five studies met inclusion criteria. A preoperative model predicting 30-day reintervention, 3.3% in 3,583 patients, showed moderate discrimination, AUC 0.74, supporting risk stratification and targeted early surveillance rather than treatment modification. A multicenter study predicting 12-month weight-loss success demonstrated limited preoperative performance, ROC-AUC ~ 0.60, but improved with early follow-up data, reaching ROC-AUC 0.79-0.88, supporting dynamic monitoring rather than baseline selection. Two simulation-based studies reported high accuracy, up to 1.00, for skill classification and up to 100% for procedural recognition, but were based on small or experimental datasets without clinical validation. One study evaluating AI-generated educational content showed comparable performance to standard materials, with accuracy 3.81 vs. 3.78 and readability 4.11 vs. 4.02. No study demonstrated prospective improvement in patient outcomes or integration into routine clinical workflows.

Current evidence suggests that AI in ESG has selective, task-specific signals of potential utility, particularly in risk stratification, longitudinal follow-up, training, and education, but remains at an early stage of clinical translation. At present, AI should be considered an adjunct to clinical judgment rather than a tool for autonomous decision-making, and further prospective validation is required before routine implementation.