 The proposed method, CGN-CDR, is able to restore low-dose specced synograms with improved quality in both the projection and image domains. It achieves this by using a conditional generative adversarial network with cross-domain regularization, which incorporates a preconditioned alternating projection algorithm with total variation regularization. This allows for more accurate reconstructions of the synogram and image, resulting in improved noise and artifact suppression, contrast enhancement, and structure preservation. Additionally, the proposed method is robust against noise and can better preserve the bone structure of the reconstructed image. This article was authored by Ceeley, Lee Mae Pang, Fing Wan Lee, and others.