FIGURE 3.

Construction and evaluation of the deep learning radiomics model for preoperative hepatocellular carcinoma (HCC) pathological grading. (A) Violin plots of six selected radiomics features, including ‘original_shape2D_Sphericity’, ‘original_firstorder_10Percentile’, ‘original_shape2D_Elongation’,‘original_firstorder_Minimum’,‘original_glcm_Correlation’ and ‘original_glcm_DifferenceEntropy’. (B) 3D scatterplot of predicted probabilities for the clinical model, radiomics signature, deep learning model and fused model. (C) ROCs and AUCs of the deep learning radiomics model in the three cohorts. (D) Calibration curves of the different models. (E) Decision curves of the different models. (F) Nomogram for preoperative prediction of HCC pathologic grade