Background and Aims: Peptic ulcer disease (PUD) affects ۵-۱۰% of the global population and can lead to gastric cancer if not detected early. The use of
Artificial Intelligence (AI) in gastroenterology may improve diagnostic accuracy and efficiency, but there is insufficient evidence on its specific performance for diagnosing gastric ulcers. This study aims to evaluate image-based machine learning and deep learning techniques for diagnosing gastric ulcers, with the goal of guiding clinical decision-making and future research. Methods: We performed a systematic search in PubMed, Embase, Scopus, and Web of Science until September ۱۱, ۲۰۲۴. QUADAS-۲ was selected to judge the risk of bias in the included articles. Independent reviewers screened studies and assessed eligibility, certainty of evidence, and risk of bias. A univariate
meta-analysis was conducted to calculate the sensitivity, specificity, and, subsequently, diagnostic odds ratio (DOR), positive likelihood ratio (LR+), and negative likelihood ratio (LR-), along with the bivariate Reitsma method to compute SROC using R version ۴.۴.۲. The study protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO) database [CRD۴۲۰۲۴۶۱۱۵۸۳]. Results: Twenty-three studies were eligible for qualitative analysis, of which sixteen were included in quantitative meta-analysis. The pooled DOR, sensitivity, and specificity of classification of
gastric ulcer against other lesions or normal tissue were ۸۰.۸۰ [۹۵% CI: ۳۷.۸۹ – ۱۷۲.۳۱], ۰.۸۵۴ [۹۵% CI: ۰.۸۲۹ – ۰.۸۷۵], and ۰.۹۲۹ [۹۵% CI: ۰.۹۰۶ – ۰.۹۴۷], respectively. The studies analyzing only CNN showed a DOR of ۷۶.۷۸ [۹۵% CI: ۳۲.۵۳ – ۱۸۱.۲۲], while studies utilizing machine learning and feature fusion, in addition to the convolutional neural network, exhibited a DOR of ۸۷.۹۳ [۹۵% CI: ۶۱.۷۴ – ۱۲۵.۲۳]. The SROC curve analysis reveals a diagnostic test with a high AUC of ۰.۹۳۴ [۹۵% CI], indicating excellent overall accuracy and reliability for clinical application. Conclusion: Machine learning and Deep learning models demonstrate promising potential in detecting gastric ulcers in classification tasks against other pathologies. Further prospective multicenter studies, with standardized
