TDNet Discover

Background: Pancreatic cancer is a leading cause of cancer deaths, with late-stage diagnosis limiting treatment options. CT scans are widely used but have limited sensitivity for early disease. Artificial intelligence (AI), including machine learning (ML), may improve diagnostic accuracy. Objective: To evaluate whether AI can improve the sensitivity of CT scans in detecting pancreatic cancer at an earlier stage compared with standard radiologist interpretation. Methods: PubMed and Medline were searched for studies from the past 10 years on AI, pancreatic cancer, and CT imaging. Inclusion criteria required peer-reviewed studies in English reporting diagnostic accuracy. Data on sensitivity, specificity, PPV, and AUC were extracted. Quality was assessed using NHLBI tools. Results: Of 588 records, four studies met inclusion criteria: two retrospective cohorts and two systematic reviews. Sample sizes ranged from <100 to >20,000 patients. AI models achieved area under the curve (AUC) values of 0.84–0.98 and sensitivities up to 94.7%, often outperforming radiologists, particularly for tumors <2 cm. However, study heterogeneity, retrospective design, and lack of standardized validation limited generalizability. Conclusion: AI shows promise as a diagnostic aid in pancreatic cancer imaging, especially for early detection. Prospective trials, standardized metrics, and external validation are needed before clinical integration.