Fig. 5: Tile-level cross-classifications as a function of test set. | Nature Communications

Fig. 5: Tile-level cross-classifications as a function of test set.

From: Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images

Fig. 5

Correlations of predicted tumor/normal status (i.e., logit of tumor probability) between pairs of classifiers, specified on the x and y axis. Correlations are first calculated using the tile values for all slides of a given test tissue. a Average correlation across tissues, using both tumor and normal slides in the tissue test sets. b Correlation for the tissue set with the maximal correlation, using both tumor and normal slides in the tissue test sets. c Average correlation across tissues, using only tumor slides in the tissue test sets. d Correlation for the tissue set with the maximal correlation, using only tumor slides in the tissue test sets.

Back to article page