Accuracy of artificial intelligence–assisted detection of upper GI lesions: a systematic review and meta-analysis

Abstract

Background and aims: Artificial intelligence (AI)-assisted detection is increasingly use in upper endoscopy. We performed a meta-analysis to determine the diagnostic accuracy of AI on detection of gastric and esophageal neoplastic lesions and Helicobacter pylori (HP) status. Method: We searched Embase, PubMed, Medline, Web of Science and Cochrane databases for studies on AI detection of gastric or esophageal neoplastic lesions and HP status. After assessment of study quality by Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool, a bivariate meta-analysis following a random effects model to summarize the data and plotted hierarchical summary receiver-operating characteristic (HSROC) curves. The diagnostic accuracy was determined by the area under the HSROC curve (AUC). Result: A total of 23 studies including 969,318 images were included. The AUC of AI detection of neoplastic lesions in stomach, Barrett's esophagus and squamous esophagus as well as HP status were 0.96 (95% CI, 0.94-0.99), 0.96 (95% CI, 0.93-0.99), 0.88 (95% CI, 0.82-0.96) and 0.92 (95% CI, 0.88-0.97), respectively. AI using narrow-band imaging is superior to white light on detection of neoplastic lesion in squamous esophagus (0.92 vs 0.83 p<0.001). The performance of AI was superior to endoscopists in the detection of neoplastic lesions in stomach (AUC 0.98 vs 0.87, p <0.001), Barrett's esophagus (AUC 0.96 vs 0.82, p<0.001) and HP status (AUC 0.90 vs 0.82, p<0.001). Conclusion: AI is accurate in the detection of upper GI neoplastic lesions and HP infection status. However, most of these studies were based on retrospective review of selected images, which would require further validation in prospective trials.