Figure 4.

Same images as Figure 2 were used to generate these attention maps. Normalized images [0-1] are pseudocoloured based on the level of “attention” the CNN gave to the region in the image modality (low represented by blue, to high represented by red).