Figure 1: GAN architecture of STAN-CT. The generator G is a U-Net with a new latent loss for synthesizing image patches. The discriminator D is an Fully Convolutional Network classifier for determining whether a synthesized image patch is fake or real.