Prediction of genetic alterations from gastric cancer histopathology images using a fully automated deep learning approach

Figure 7 The classifiers trained with both The Cancer Genome Atlas and SSMH data were used to predict the mutation of CDH1 (A and B), ERBB2 (C and D), and KRAS (E and F) genes. Representative binary heatmaps of the whole slide images (WSIs) correctly classified as mutation, correctly classified as wild-type, falsely classified as wild-type, and falsely classified as mutation were presented. Receiver operating characteristic curves for the folds with the lowest and highest area under the curve and the concatenated ten folds were also presented for each gene. CDH1-M: CDH1 mutated, CDH1-W: CDH1 wild-type, ERBB2-M: ERBB2 mutated, ERBB2-W: ERBB2 wild-type, KRAS-M: KRAS mutated, KRAS-W: KRAS wild-type.