Fig. 2: ROC curves of VGG19 using SDL and WSDL methodologies in patch-level prediction and Grad-CAMs of patches for cancer prognosis prediction.

a Patch-level ROC curves of VGG19 using both SDL and WSDL methodologies. Regions highlighted in the Grad-CAMs represent tumor cells (b) and tumor-infiltrating immune cells (c), which are annotated in the H&E-stained images with yellow dotted lines. ROC receiver operating characteristic curve, VGG Visual Geometry Group, WSDL weakly supervised deep learning, SDL supervised deep learning.