Fig. 1: The workflow of EBVNet for predicting EBV status with hematoxylin and eosin-stained WSIs. | Nature Communications

Fig. 1: The workflow of EBVNet for predicting EBV status with hematoxylin and eosin-stained WSIs.

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

Fig. 1

Each WSI was preprocessed and tessellated into non-overlapped tiles of ×10 magnification. After color normalization, tiles were resized to 224 × 224 pixels and then input to the tumor detector. Only tiles from regions recognized as tumor were fed to EBVNet to get tile-level probabilities for EBV status. The five well-trained individual classifiers were ensembled to form the EBVNet at the output layer of individual classifiers. The average probability outputs of the five individual classifiers were used as the prediction of the ensembled model EBVNet. Tile-level probabilities were averaged to generate a slide-level probability of EBV status. EBV Epstein-Barr Virus, WSI whole slide image.

Back to article page