A radiomics-based model can predict recurrence-free survival of hepatocellular carcinoma after curative ablation

Prediction of early recurrence (ER) of HCC after radical treatment is of great significance for follow-up and subsequent treatment, and there is a lot of unmet needs. Here, our goal is to develop and validate a radiomics nomogram that can predict ER after curative ablation. The aim of this study was to evaluate the efficacy and safety of regorafenib after disease progression with sorafenib in Chinese patients with advanced HCC through this retrospective analysis. 149 HCC patients treated between November 2008 and February 2018 were enrolled and randomly divided into training cohort (n = 105) and validation cohort (n = 44). The survival endpoint was recurrence-free survival (RFS). A total of 16908 radiomics features were extracted from the contrast-enhanced MR images of each patient. The minimum redundancy maximum relevance algorithm (mRMR) and random survival forest (RSF) were used for feature selection. Twelve kinds of support vector machine (SVM) models, a Cox regression model (Cox PH), a random survival forest (RSF) model and a gradient boosting model (GBoost) were used to build a radiomics signature. These models were trained after adjusting the model parameters using 5-fold cross-validation. The best models were selected according to the C-index. Using the machine learning (ML) framework, 40 features were identified that demonstrated good prediction of HCC recurrence across all cohorts. The random survival forest (RSF) model showed higher prognostic value, with a C-index of 0.733–0.801 and an integrated Brier score of 0.147–0.165, compared with other SVM models, Cox regression models, etc. (all P