AI Revolution in Upper GI: Endoscopic Diagnosis of Gastric Cancers and Premalignant Neoplasms

Background: Discrepancies between endoscopic and pathological diagnoses commonly occur. From pre-malignant neoplasms to early gastric cancers (EGC), this inconsistency often leads to significant alteration of management. AI (artificial intelligence) assisted diagnostic tools have the potential to mitigate the differences. From optimizing endoscopic examination through standardizing quality and risk stratification by identifying at-risk populations to enhancing EGC detection and characterization guiding therapeutic interventions, AI-assisted endoscopy has shown potential to overcome current limitations in standard care. Methods: In this state-of-the-art narrative review, we synthesize the current evidence and prevailing paradigm, discuss existing limitations, and outline the future directions of AI-assisted endoscopy in EGC. Results: AI models demonstrate potential in enhancing all steps of the EGC management pathway, from optimizing endoscopy quality, identification of at-risk populations, and lesion detection, lesion characterization, to endoscopic resection guidance and lymph node metastasis prediction. Yet, existing evidence largely remains retrospective, with interventional data demonstrating heterogeneity. The conflicting data largely stems from inherent fundamental limitations of AI research in EGC, such as the lack of standardized objective performance metrics, inconsistent methodologies, and reporting standards. These issues fundamentally hinder generalizability and implementation in routine clinical practice. Conclusions: AI-enhanced endoscopy demonstrates potential to overcome current limitations in EGC management. Despite this promising future, there are obstacles to validation, implementation, and mitigating disparities owing to inherent fundamental issues.