Fig. 3.

Network architecture of attention-based classifier (AC). ${\mathbf{f}}_n$ is the top extracted feature map of the FEX, ${\mathbf{f}}_{mask}$ is the lesion-aware feature map learned by LA-Net, ${\mathbf{f}}_{att}$ is the enhanced feature map, MAM is the mask attention module in which $\otimes$ represents element-wise multiplication, and $\oplus$ represents element-wise addition.