Prognostic and predictive value of radiomics signatures in stage I-III colon cancer.

Abstract

Accurate identification of patients with poor prognosis after radical surgery is essential for clinical management of colon cancer. Thus, we aimed to develop death and relapse specific radiomics signatures to individually estimate overall survival (OS) and relapse free survival (RFS) of colon cancer patients. In this study, 701 stage I‐III colon cancer patients were identified from Fudan University Shanghai Cancer Center. A total of 647 three‐dimensional features were extracted from computed tomography images. LASSO Cox was used to identify the significantly death‐ and relapse‐associated features and to build death and relapse specific radiomics signatures, respectively. A total of 13 death‐specific and 26 relapse‐specific features were identified from 647 screened radiomics features. The developed signatures can divide patients into two groups with significantly different death (Hazard Ratio (HR): 3.053; 95% CI, 1.78‐5.23; P < .001) or relapse risk (HR: 2.794; 95% CI, 1.87‐4.16; P < .001). Time‐dependent Relative operating characteristic curve showed that the signatures performed better than any other clinicopathological factors in predicting OS (AUC: 0.768; 95% CI, 0.745‐0.791) and RFS (AUC: 0.744; 95% CI, 0.687‐0.801). Further, survival decision curve analyses confirmed the good clinical utility of the two radiomics signatures. In conclusion, we successfully developed death‐ and relapse‐specific radiomics signatures that can accurately predict OS and RFS, which may facilitate personalized treatment.