Computed tomography-based delta-radiomics of tumor core_edge combination for systemic treatment response evaluation in pancreatic cancer

Abstract

Background: As a systemic disease, pancreatic cancer (PC) can be treated systemically to raise the R0 resection rate and enhance patient prognosis. The best ways to assess the treatment response to systemic treatment of patients with PC are still lacking. Methods: A total of 122 PC patients were enrolled; 25 of these patients were used as an independent testing set. According to the pathologic response, PC patients were classified into the responder and non-responder groups. The whole tumor, core, edge, and peritumoral were segmented from the enhanced CT images. Machine learning models were created by extracting the variations in radionics features before and after therapy (delta radiomics features). Finally, we compared the performance of models based on radiomics features, changes in tumor markers, and radiologic evaluation. Results: The model based on the core (Area under Curve, AUC=0.864) and edge features (AUC=0.853) showed better performance than that based on the whole tumor (AUC=0.847) or peritumoral area (AUC=0.846). Moreover, the tumor core_edge combination model (AUC=0.899) could better increase confidence in treatment response than using either of them alone. The accuracies of models based on changes in tumor markers and radiologic evaluation were relatively poorer than of the radiomics model. Moreover, Patients predicted to respond to therapy using the radiomics model showed a relatively longer overall survival (43 months vs 27 months), although there were no significant differences (p=0.063). Conclusions: The tumor core_edge combination delta radiomics model is an effective approach to evaluate pathologic response in PC patients with systemic treatment.