Skip to main content
. 2021 Apr 12;11:7995. doi: 10.1038/s41598-021-86995-5

Figure 5.

Figure 5

Separation of binary classes increases with increasing number of training images. tSNE reduces the 32 dimensions of the last hidden layer of networks trained with 10, 50, and 100 images to two dimensions. Coloring the last hidden layer for the respective 215, 175 and 125 lymphoblast and non-lymphoblast images shows an increasing cluster separation of the two classes with increasing training set size even though the ROC-AUCs for 50 or 100 training images are similar (see Fig. 2A).