Fig. 2: The architecture of the OMG-Net.

The two-step architecture includes mask generation and mitotic figures (MF) classification. First, the post-process cell masks from patched WSIs are generated by Segment Anything (SAM) using an evenly sampled point grid as a prompt. Second, the RGB image of the segmented cell and the binary mask (presented in green) are used to classify MFs by employing an adapted ResNet18.