Diagnostic Performance of Artificial Intelligence -based Computed Tomography Techniques in Detecting Pancreatic Cancer: A Systematic Review

Introduction: Pancreatic cancer is among the most lethal malignancies, often diagnosed at an advanced stage due to its subtle clinical presentation. Artificial intelligence (AI) techniques applied to computed tomography (CT) have shown potential in improving diagnostic accuracy across various imaging tasks, including cancer detection and lesion classification. This systematic review and meta-analysis aimed to evaluate the diagnostic performance of AI-based CT models specifically for the detection of pancreatic cancer, while accounting for the clinical task type and algorithmic heterogeneity among studies. Methods A comprehensive literature search was conducted across PubMed/MEDLINE, the Cochrane Library, Web of Science, Embase, Scopus, CINAHL, and Google Scholar to identify relevant studies published up to April 2025. Eligible studies were screened according to predefined criteria, and diagnostic performance data including true positives, false positives, true negatives, and false negatives were extracted or calculated as needed from the selected publications. Results Ten studies comprising a total of 33,174 patients met the inclusion criteria. The pooled diagnostic sensitivity and specificity were 0.92 with a 95 percent confidence interval of 0.92 to 0.93, and 0.98 with a 95 percent confidence interval of 0.98 to 0.98, respectively. The area under the summary receiver operating characteristic curve was 0.959, and the diagnostic odds ratio was 179.57 with a 95 percent confidence interval of 57.98 to 556.16, indicating high diagnostic accuracy for pancreatic cancer detection. Significant heterogeneity was observed among the studies included. However, subgroup analyses did not reveal any statistically significant differences. No evidence of publication bias was detected. Conclusions The findings of this study suggest that artificial intelligence assisted computed tomography may serve as a valuable tool in supporting the diagnosis of pancreatic cancer. However, further validation using larger and more diverse datasets is necessary to confirm its clinical utility and generalizability.