Extended Data Fig. 2: Validation of semantic segmentation of morphological cell types (“super-classes”).

(a,b) Semantic segmentation workflow. Semantic segmentation of cell nuclei (c), conditioned on five broad annotation groups for spots (b). (c,d) Accuracy of cell type composition by semantic segmentation. Fractions of FG pixels in each H&E image (y axis) assigned to each morphological superclass (x axis) in test and train sets (color code) in adult (c, top) and young (d, top) mice, and percent and number (color scale) of pixels with a ground truth superclass label (rows) classified to each label (columns) in in adult (c, bottom) and young (d, bottom) mice. In Box plots: center black line, median; color-coded box, interquartile range; error bars, 1.5x interquartile range; black dots; outliers. Numbers of images used in train and test sets are denoted by “n_train” and “n_test”, respectively, above each subplot.