Fig. 1

Architecture of our V-net segmentation network. Each block represents a residual convolution block (see Eq. 5); strided convolutions (downsampling) are blue and transposed convolutions (upsampling) are purple. The output resolution at each level is displayed and the number of filters/channels is displayed with the ‘f’ suffix. Green spheres represent concatenation. The last block performs 1x1x1 convolution with a sigmoid activation function