Artificial Intelligence for Endoscopic Diagnosis of Gastric Premalignant Lesions

Gastric cancer is a leading cause of cancer-related mortality worldwide. The Correa cascade posits that intestinal-type gastric adenocarcinoma develops through a stepwise progression from chronic gastritis to atrophic gastritis, intestinal metaplasia, and dysplasia, highlighting the importance of early detection of premalignant lesions. The use of artificial intelligence (AI), particularly convolutional neural network-based deep-learning models, in upper gastrointestinal endoscopy has been expanding rapidly. This review examines the current role of such models in the endoscopic diagnosis of chronic atrophic gastritis, intestinal metaplasia, and gastric dysplasia. Data from meta-analyses indicate pooled sensitivities and specificities exceeding 90%, with performance comparable to that of expert endoscopists. The results of randomized trials suggest that AI-assisted endoscopy may improve the detection of intraepithelial neoplasia, particularly for small and subtle lesions. However, these trials are few and were conducted exclusively in China. Moreover, they largely evaluated systems of a single lineage. The one trial that prespecified centralized pathologic review found no improvement in neoplasm detection. Thus, multicenter prospective validation and real-world data are necessary before AI can be reliably used to inform routine clinical decision-making. Moreover, further integration of AI models into practice will depend on the availability of appropriate human-AI interfaces and unified multitask platforms and the resolution of hitherto unresolved legal and ethical concerns.