Fig. 4: Successful cases predicted by EBVNet. | Nature Communications

Fig. 4: Successful cases predicted by EBVNet.

From: A deep learning model and human-machine fusion for prediction of EBV-associated gastric cancer from histopathology

Fig. 4

a–c Histological image (left column) of patients with EBVaGC in a–c were from Internal-STAD, MultiCenter-STAD, and TCGA-STAD, respectively. The heatmaps overlapped on these three WSIs (middle column) showed that tumor tiles were mainly predicted as EBVaGC with a high score (reddish color). Tiles with a high score were mainly localized in areas of medullary histology, poor differentiation, and tumor with vacuolar nucleus or recognizable nucleolus (right column, tiles at ×10 magnification). d–f Histological image (left column) of patients with EBVnGC in d–f were from Internal-STAD, MultiCenter-STAD, and TCGA-STAD, respectively. The heatmaps overlapped on these three WSIs (middle column) showed that tumor tiles were mainly predicted as EBVnGC with a low EBV score (bluish color). All results could be reproduced stably by EBVNet. Tiles with a low score were more likely localized in areas of adenoid differentiation, mucinous differentiation, and signet-ring cell differentiation (right column, tiles at ×10 magnification). EBV Epstein-Barr Virus, EBVaGC Epstein-Barr Virus-associated gastric cancer, EBVnGC Epstein-Barr Virus negative gastric cancer.

Back to article page