Skip to main content
. 2020 May 6;7:177. doi: 10.3389/fmed.2020.00177

Figure 3.

Figure 3

On the left side, the 4 test images are shown for which the outputs of the CNN trained with dermatological labels and those trained with biopsy-verified labels differ the most. In the upper part of each image the majority decision of the dermatologists (MD) and the result of the biopsy (BIO) is given. On the right side, the boxplots of the corresponding outputs of the two CNNs over the 10 conducted simulation runs are presented.