Figure 1.

The architecture of the neural network: 3D U-Net. The input is a sample with the spatial dimension 400 × 400 × 16 of the MR image. The output has the same dimension containing the probability of the three classes (vertebral bodies, intervertebral discs, and spinal canal). The red numbers represent the number of feature maps, whereas the blue numbers define the spatial dimension of the feature maps. All arrows define functions between these feature maps as described in the picture. The Residual Blocks (ResBlock 1, ResBlock 2) are shown in Figure 2.