Fig. 2

The schematics of the 3D Attention Network. The light-yellow blocks and the dark-yellow blocks indicate the convolutional layers and the ReLU, respectively. The orange blocks indicate the max-pooling operation resulting in a 1/2 image size. The purple blocks indicate the fully connected layer, and the green blocks indicate additional clinical metadata available. The green arrow-line and the red dashed-arrow-line are connecting operations and concatenating vectors, respectively