Fig. 4.

Fitted binomial receiver operating characteristic (ROC) curves for single-parametric (dashed line) and mpMRI classifiers (solid line) trained on the full set. The three single-parametric classifiers were trained separately on (i) DCE, (ii) T2w, and (iii) DWI features. The mpMRI models (iv) were trained on the ensemble of features extracted from all available sequences, and (v) aggregated the PM from the single-parametric classifiers via soft voting. The legend gives the AUC with the 95% CI for each classifier.