Fig. 3

Receiver operating curves (ROC) for a radiomics (i.e., R) and b radiomics with molecular subtype (i.e., R+MS) classifier models in the training and testing sets. Repeated five-fold, nested cross-validation was performed in the training set wherein the accuracy values for the classifier were produced by evaluating the classifier only on the data not used in the model building in each fold of the validation. An independent hold-out set that was never seen by the model during training was treated as the test set