Often in medical imaging- a very small fraction of the image is responsible for the findings, but GRADCAM or its later versions show a big red patch, including much larger areas.
Is GRADCAM a useful tool, in context to medical imaging? Is GRADCAM effective to pinpoint such small features of image, that are responsible for their class labelling(especially in medical imaging.)
It works nicely on images where class object covers larger portions of the image, but Is GRADCAM a useful tool, in context when these objects don’t cover a large portion of the image?