CT-radiomics and pathological tumor response to systemic therapy: A predictive analysis for colorectal liver metastases. Development and internal validation of a clinical-radiomic model

Introduction The standard treatment of colorectal liver metastases (CRLM) is surgery with perioperative chemotherapy. A tumor response to systemic therapy confirmed at pathology examination is the strongest predictor of survival, but it cannot be adequately predicted in the preoperative setting. This bi-institutional retrospective study investigates whether CT-based radiomics of CRLM and peritumoral tissue provides a reliable non-invasive estimation of the pathological tumor response to chemotherapy. Methods All consecutive patients undergoing liver resection for CRLM at the two institutions were considered. Only patients with a radiological partial response or stable disease at chemotherapy and with a preoperative/post-chemotherapy CT performed <60 days before surgery were included. The pathological response was evaluated according to the tumor regression grade (TRG). The tumor (Tumor-VOI) was manually segmented on the portal phase of the CT and a 5-mm ring of peritumoral tissue was automatically generated (Margin-VOI). The predictive models underwent internal validation. Results Overall, 222 patients were included; 64 had a pathological response (29 %, TRG1-3). Two-third of patients displaying a radiological response (111/170) did not have a pathological one (TRG4-5). For TRG1-3 prediction, the clinical model performed fairly (Accuracy = 0.725, validation-AUC = 0.717 95%CI = 0.652–0.788). Radiomics improved the results: the model combining the clinical data and Tumor-VOI features had Accuracy = 0.743 and validation-AUC = 0.729 (95%CI = 0.665–0.798); the full model (clinical/Tumor-VOI/Margin-VOI) achieved Accuracy = 0.820 and validation-AUC = 0.768 (95%CI = 0.707–0.826). Conclusion CT-based radiomics of CRLM allows an insightful non-invasive assessment of TRG. The combined analysis of the tumor and peritumoral tissue improves the prediction. In association with clinical data, the radiomic indices outperform standard radiological and clinical evaluation.