Flowchart of the particular V-Net architecture we have implemented. The network can take as input multiple channels with dimensions of 144³ (top left green cube) and generates predictions for the central voxels with dimensions 128³ (top right red cube). The flowchart illustrates the encoder and decoder paths, along with other distinctive features of the network. Notably, the hidden layers and skip connections are represented by purple and yellow cubes, with their respective dimensions annotated at their centres. The down-sampling and up-sampling blocks are shown as brown and purple trapezoids, in their centres we indicate the number of filters employed for the convolution (or transposed convolution) operations.