Fig. 3

Feature analysis of Radiomics model. (A) Light-colored nodes indicate the best radiomics features for the respective ROIs, while purple nodes represent features shared across multiple models. The edges color-coded by model indicate which features were selected by which models. (B) and (C) illustrate the partial dependence of the model on three features (wavelet-HLL_gldm_LargeDependenceHighGrayLevelEmphasis, original_firstorder_Minimum, wavelet-LHH_glrlm_RunVariance). Each axis represents the range of a feature value, and the color of the points reflects the model’s predictions. The color gradient allows for a visual assessment of the combined effect of these three features on the predictions.