Fig. 8

Radiomics algorithm for Glycan-HCCs and Lipid-HCCs distinguishing. a Workflow of radiomics analysis, includes tumor segmentation, features extraction, model development and clinical application. Heatmaps show radiomics features profiles in in-house cohort (b) and TCGA cohort (c). Heatmaps were generated by using package “pheatmap” in R software. d A total of 19 radiomics features were finally included. e LASSO coefficient profiles of the features. f Tuning parameter (λ) selection in the LASSO model used ten-fold cross-validation via minimum criteria. g ROC curves of radiomics models in training (in-house) and test (TCGA cohort)