Disentangled generative adversarial network for low-dose CT

Abstract Generative adversarial network (GAN) has been applied for low-dose CT images to predict normal-dose images. However, the undesired artifacts and details bring uncertainty clinical diagnosis. In order improve visual quality while suppressing noise, in this paper, we mainly studied two key components of deep learning based (LDCT) restoration models—network architecture loss, proposed a disentangled noise suppression method on GAN ( DNSGAN ) LDCT. Specifically, generator network, which contains structure recovery modules, is proposed. Furthermore, multi-scaled relativistic loss introduced preserve finer structures generated Experiments simulated real LDCT datasets show that can effectively remove recovering provide better perception than other state-of-the-art methods.