Figure 3
Heatmaps summarizing areas under the ROC curve (AUC) of predictive models developed including (a) clinicopathological predictors alone, (b) T2WI-derived radiomic features alone, (c) ADC-derived radiomic features alone, (d) a combination of clinicopathological predictors, T2WI-, and ADC-derived radiomic features, (e) a combination of clinicopathological predictors and T2WI-derived radiomic features, (f) a combination clinicopathological predictors and ADC-derived radiomic features, and (g) a combination of T2WI- and ADC-derived radiomic features. Each cell presents an AUC for a model developed using a given combination of feature selection and machine learning algorithms. 95% confidence intervals for each model are summarised in Supplementary Tables S1-7, respectively. Blank cells (b,g) denote models shrunk to the intercept only due to the regularisation approach used by GLMnet.
