Abstract: Micro-CT has a wide range of application fields, and has been studied in biomedicine and materials science, and has developed rapidly in recent years. Micro-CT images often suffer from noise due to radiation dose limitations, so it is important to develop an appropriate algorithm to suppress noise in Micro-CT images. The noise level of the Micro-CT image is related to the scanned samples, scanning parameters and other parameters. The noise suppression algorithm should have good performance under different noise levels. In the past, the noise suppression algorithm of Micro-CT images was mainly an iterative reconstruction algorithm, but the iterative reconstruction algorithm was relatively slow. As a popular image processing method in recent years, the deep learning method has better effect and faster processing speed than traditional methods in clinical low-dose CT image processing, and has the potential for further application in low-dose Micro-CT image processing. In addition, generative adversarial networks have better results than convolutional neural networks in maintaining image details. In this paper, ordinary convolutional neural networks and generative adversarial networks are designed to compare their performance differences. Limited to the power of the radioactive source, high-dose Micro-CT images are difficult to acquire. This paper has proposed an innovative scanning method that can effectively obtain low-noise Micro-CT data of mice. Due to high noise of low-dose Micro-CT, combined with the imaging principle of Micro-CT, a progressive low-dose Micro-CT image processing method was proposed, which processed the Micro-CT data before and after analytical reconstruction. Compared with processing only on tomographic images, the progressive method is better for processing high-noise Micro-CT data. From the objective indicators and visual effects, the differences between the progressive method and other deep learning methods or iterative reconstruction algorithms are analyzed and compared. This paper quantitatively analyzes the effect of the processing network on Micro-CT images with different noise levels, and analyzes the advantages and limitations of GAN in progressive Micro-CT image processing. The progressive Micro-CT image processing method generates images with high quality, fast generation speed, high robustness, and high objective index, which can help further advanced applications such as image segmentation and image 3D visualization.