Fig. 3: Receiver operating characteristic (ROC) curves and area under the curves (AUCs) for HRD classification across CT energy levels.

a ROC curves on the training set comparing the performance of the handcrafted radiomics (HCR), supervised deep learning (sDL), and foundation model (FM) for HRD classification across 40 kVp CT, 80 kVp CT, and combined CT datasets. b Same as (a), but on the test set. c AUC values as a function of training set size for each model. Lines represent mean AUCs; shaded areas (a, b) and error bars (c) indicate 95% confidence intervals (CIs). * p < 0.05, ** p < 0.01.