Figure 1.

A schematic architecture of residual 3D U-Net. The number above each block corresponds to the number of features for each layer. The number in the left of each block corresponds to the matrix size at every level. Each convolutional layer has a 3 × 3 × 3 kernel and a rectified linear unit, except for the last layer that produced the residual update. Between different layers, a 2 × 2 × 2 max pooling size was used.