Figure 7.

Generated heatmaps (ResNet34-UNet + DS) for sections from the Test part of the data that were wrongly classified: (a,b) false positive, the model wrongly identified a hair-follicle as a tumor (10,656 $\mathsf{\mu }{\mathrm{m}}^2$); (c,d) false negative, the model detected the BCC but not with enough confidence, i.e., the largest connected area above the threshold ($0.60$) was too small ( 544 $\mathsf{\mu }$${\mathrm{m}}^2$).