Fig. 2

Workflow of necessary steps in this study. a ROI was manually delineated on non-contrast CT scans at the level of right portal veins. b Radiomic features including first-order statistics, textural features and wavelet transforms were extracted. c Intra- and interobserver reproducibility and subsequent lasso regression were used for feature selection. d A radiomics signature was constructed with SVM and a radiomics-based nomogram integrates radiomics signature and clinical predictors. e The performance of established models was evaluated by ROC, calibration and DCA curves. ROI, region of interest; LASSO, least absolute shrinkage and selection operator; SVM, support vector machine; ROC, receiver operator characteristic; DCA, decision curve analysis