Radiomics analysis of baseline computed tomography to predict oncological outcomes in patients treated for resectable colorectal cancer liver metastasis

Abstract

Abstract Predicting recurrence and survival of patients with upfront resectable colorectal cancer liver metastases (CRLM) is crucial to personalize treatment. The purpose of this work was to determine whether radiomics analysis of baseline computed tomography (CT) images could help predict outcomes of resectable CRLM compared to the clinical risk score (CRS). From a registry of 251 patients treated with systemic chemotherapy and surgery for CRLM, radiomics features extracted from baseline CT images were developed to predict time to recurrence (TTR) and disease-specific survival (DSS) and compared to low- and high-risk groups based on the CRS using Kaplan-Meier estimates and Log-rank test. CRS scores provided significant separation of low- vs. high-risk CRLM patients for TTR (p = 0.002) and DSS (p = 0.002), whereas radiomics signatures improved separation by 4–6 and 6–8 orders of magnitude for TTR and DSS (p < 0.0001), respectively. CRS alone provided median survival times for TTR of 1.67 and 1.05 years for low- and high-risk groups respectively, while radiomics yielded 2.87 and 0.92 years. Radiomics signatures outperformed the CRS for stratifying CRLM patients in low- and high-risk groups. Early prediction of patient outcome may reduce CRLM patient exposure to ineffective yet toxic chemotherapy or high-risk major hepatectomies.