Fig. 9
From: Improving lesion detection in mammograms by leveraging a Cycle-GAN-based lesion remover
False positive normal controls that changed to true negative after applying the lesion highlighter. This figure shows two normal controls which were false positive before applying the lesion highlighter but changed to true negative after applying the lesion highlighter. The first and third column show the images with normal tissue with (top row) and without (bottom row) applying the lesion remover. The second and last column show the attention map by the detector. Before applying the lesion highlighter, the detector was falsely focused on the normal tissue as a lesion, but after the lesion highlighter, the detector’s attention was moved away from the previously falsely focused area. Note that the normal breast tissue after applying the lesion highlighter show pixels in green lightly but widely spread in the image, which made the detector correctly identify as normal tissue