Fig. 2.

Network architecture for extracting computational biomarkers from the WSI and genomic data. We used seven LSTM cells in the network. The calculated pathway scores from the genomic data were forwarded into an MLP that contains three FC layers. The last layer of the MLP was connected with the features extracted from the image patches to serve as the input for the LSTM after an FC layer. On top of the LSTM, we utilized an average pooling layer.