Figure 7.

Performance comparison of DRUNets with contextual loss among different Resnet layers and ordinary U-net for synthetic T1WI and T2WI. The first two row panels A₁/A₂ to D₁/D₂: synthetic T1WI generated by DRUNet-34, 50, 101 and ordinary U-Net. The last two row panels A₃/A₄ to D₃/D₄: synthetic T2WI generated by DRU-Net-34, 50, 101 and ordinary U-net. The last column panels E₁/E₂: real T1WI; E₃/E₄: real T2WI; color bars below each synthetic images: difference maps between synthetic and real images. DRUNet-X, double ResNet-U-Net with double X convolutional layers; T1WI, T1-weighted images; T2WI, T2-weighted images.