An unsupervised reconstruction method for low-dose CT using deep generative regularization prior

•Uses deep CNNs as regularizer due to that CNNs can converge to natural images faster than noise.•A randomly initialized deep neural network trains only with one image.•Projection domain measurement loss, image domain SSIM loss, and total variation loss are used as loss terms.•Loss weights, number of parameters and network architecture use as the parameters to tune the reconstruction.