A Dual Regression Scheme to Improve GAN in Low-Dose CT Scan Restoration

The application of Super Resolution (SR) technology on recovering low-dose medical image is gaining its importance nowadays. Because it can provide as many contextual details as high-dose CT for physi-cians without harming the patient or spending more on devices. Usage of Deep Learning (DL) on SR tasks has been developed many years, and Generative Adversarial Network (GAN) plays a leading role among them. In this paper, we find that using a special closed-loop scheme, dual network, can strongly improve the performance of GAN model, with about 1.2 point of improvement in PSNR. And the best way to implement this dual network is also evaluated.