Fig. 2

Architectures of the proposed 3D U-Net like models. The models take as input a real T1nce image of size 128\(\times\)128\(\times\)128 and generate a synthetic T1nce of size 128\(\times\)128\(\times\)128. Res-U-Net: images pass through five descending blocks, each one followed by a residual module, and then through four ascending blocks and one final layer. Att-U-Net: images pass through five descending blocks and then through four ascending blocks and one final layer. One of the inputs of each ascending block is the result of the attention gate. Trans-U-Net: images pass through four descending blocks, four transformer layers and four ascending layers. All the parameters such as kernel size, stride, padding, size of each feature map (N) are reported