Gastric cancer remains a major cause of cancer-related mortality worldwide. Endoscopy is the most reliable modality for detecting early gastric cancer (EGC), yet the identification of subtle early lesions is often challenging. Artificial intelligence (AI) has emerged as a promising tool to support endoscopic diagnosis. This meta-analysis aimed to compare the diagnostic performance of AI systems with that of human endoscopists in detecting EGC

PubMed, Embase, Google Scholar, Cochrane, and Web of Science were searched for articles. Studies were eligible if they allowed extraction of TP, FP, TN, FN. Histopathology was the gold standard. Endoscopists were classified as experts or non-experts. Pooled sensitivity and specificity were estimated using random-effects models; an SROC curve was constructed, and AUC calculated

Eight studies from Asia were included. A total of 3409 patients were analyzed, comprising 1444 with EGC and 1965 with non-neoplastic conditions. AI achieved a sensitivity of 92% (95% CI, 89–94%) and a specificity of 87% (95% CI, 84–90%). Expert endoscopists reached a sensitivity of 90% (95% CI, 86–93%) and specificity of 90% (95% CI, 80–95%). Non-expert endoscopists showed sensitivity of 78% (95% CI, 72–83%) and specificity of 78% (95% CI, 69–85%). AI performance was comparable to that of experts and significantly superior to non-experts (sensitivity p ≤ 0.001; specificity p ≤ 0.01)

AI-assisted endoscopy demonstrates high diagnostic accuracy for early gastric cancer. Its performance closely parallels that of expert endoscopists and substantially exceeds that of non-experts