Fig. 2.

Summary of the proposed workflow for organ identification. Blue panels indicate Training and green panels refer to Inference. The workflow involves the generation of tiles at multiple magnifications to be used for our MMO-Net architecture to train binary models (panels 1 and 2). Subsequently, during inference, the multi-magnification tiles pass through all organ models to form the WSI level prediction mask (panel 3). Lastly, the organ metadata is validated comparing it with the organs predicted in the mask (panel 4).