Low-dose medical computed tomography (CT) images are associated with noise problems, and it is difficult to obtain relevant paired datasets. To solve these issues, we propose a low-dose CT denoising algorithm, which is based on an improved cycle generative adversarial network. Our algorithm achieves end-to-end mapping from low-dose CT images to standard-dose CT images using unpaired datasets. In addition, to make the generator output image similar to the target image, we creatively put the DenseNet residual learning network model to the generator, wherein feature reusability is beneficial to restore the image details. Research confirms that this algorithm effectively improves the ability of edge keeping and denoising. The quality of the restored image is significantly improved, which is helpful for the detection and analysis of lesions.