Fig. 6.

Cascaded randomization test. The first column shows the original Grad-CAM map for tumor (PCa) and Prostate (PG), followed by the Grad-CAM maps generated after randomizing the weights of the respective convolutional layers. Here TN is the trained Network, BL is the bottleneck layer, D1, D2, D3, E1, E2, E3 are the corresponding decoder and encoder blocks of the U-net, and RN is the network with random weights only