Figure 3.

Example of the true positive score of the convolutional neural network (CNN) for the capillary collapse (CC). (Left) One of the periodic acid−Schiff (PAS)−stained images from the concordance data. The CNN diagnosed this image as “capillary collapse is positive,” and all 5 clinicians scored it as “capillary collapse is positive.” (Center) The blue lines depict the collapsed capillary, and the yellow lines show the fibrous crescents. (Right) Visualization of the CNN’s focusing points using the Gradient-weighted Class Activation Mapping when the CNN diagnosed the capillary collapse as positive or negative. Red lesions are the CNN’s focusing points. This shows that the CNN was focused correctly on the capillary area.