Preoperative prediction of early recurrence in resectable pancreatic cancer integrating clinical, radiologic, and CT radiomics features

Jeong Hyun Lee et al. developed a predictive model for early recurrence in resectable pancreatic cancer by integrating clinical, radiologic, and CT radiomics features. Retrospective data from 190 patients were used to create three models: radiomics-only, clinical-radiologic (CR), and clinical-radiologic-radiomics (CRR). Using the Aview software, an experienced radiologist manually segmented tumor volumes from CT images to create Volumes of Interest (VOIs), which were analyzed with PyRadiomics to extract 572 quantitative features. Random forest algorithms optimized model performance. Among the models, the CRR model demonstrated the highest accuracy (AUC = 0.83) during external validation, with balanced sensitivity (65%) and specificity (87%). Key predictive factors included elevated CA19-9 levels and specific radiomics markers. This approach improves preoperative risk stratification and guides personalized treatment, offering alternatives like neoadjuvant chemotherapy for high-risk patients. Future work should aim for automated segmentation, larger sample sizes, and prospective validation. This study advances tailored cancer management.