Use of artificial intelligence for the detection of Helicobacter pylori infection from upper gastrointestinal endoscopy images: an updated systematic review and meta-analysis

B ackground Helicobacter pylori ( H. pylori ) infection is associated with various gastrointestinal diseases and may lead to gastric cancer. Currently, endoscopy is the gold standard modality used for diagnosing H. pylori infection, but it lacks objective indicators and requires expert interpretation. In the past few years, the use of artificial intelligence (AI) for diagnosing gastrointestinal pathologies has increased tremendously and may improve the diagnostic accuracy of endoscopy for H. pylori infection. This study aimed to evaluate the diagnostic accuracy of AI algorithms for detecting H. pylori infection using endoscopic images. Methods Three investigators searched the PubMed, CINHAL and Cochrane databases for studies that compared AI algorithms with endoscopic histopathology for diagnosing H. pylori infection using endoscopic images. We assessed the methodological quality of studies using the QUADAS-2 tool and performed a meta-analysis to estimate the pooled sensitivity, specificity, and accuracy of AI for detecting H. pylori infection. Results A total of 11 studies were identified that met our inclusion criteria. All were conducted in different countries based in Asia. Our meta-analysis showed that AI had high sensitivity (0.93, 95% confidence interval [CI] 0.90-0.95), specificity (0.92, 95%CI 0.89-0.94), and accuracy (0.92, 95%CI 0.90-0.94) for detecting H. pylori infection using endoscopic images. However, there was also high heterogeneity among the studies (Tau 2 =0.87, I 2 =76.10% for generalized effect size; Tau 2 =1.53, I 2 =80.72% for sensitivity; Tau 2 =0.57, I 2 =70.86% for specificity). Conclusion This systematic review and meta-analysis showed that AI had high diagnostic accuracy for detecting H. pylori infection using endoscopic images.