Abstarct: It has always been a dream to gain knowledge from the source of influences, and its discoveries have surprised people. The limitations and obstacles that have existed in this direction have posed a hindrance to the advancement of technology. The use of medical imaging technology is an important part of disease diagnosis and screening today. However, its detrimental nature and time-consuming attributes have hampered its accessibility and utilization. By reducing the dosage, noise is generated in the outcomes, rendering their utilization challenging or unfeasible. The computed tomography scan is a medical imaging method with many applications; However, due to its radiation damage, it is limitedly used for long periods. By reducing the dosage, this approach offers the potential for a broader utilization with shorter periods. This is an inverse problem, and one approach to address it is through the utilization of iterative methods, which typically necessitate prior knowledge. By estimating the statistical distribution of normal dose images, we can evaluate the fit of low dose images. This thesis aims to reduce the noise of the final images at the local scale by using infinitesimal continuous normalizing flows as a regularizer in an iterative method. This method provides us with the ability to accurately calculate the probability density function. In model training, we use only 6 images and use SSIM and PSNR criteria to evaluate the proposed model. Our proposed model has a lower number of parameters than other models, and it is trained with 5,184 parameters. On average, the tests performed with the LoDoPaB-Ct dataset show a significant quality improvement in the SSIM, which promises better results in research in this field. In the quantitative evaluation, our model revealed a MSSSIM quality of 0.950, SSIM quality of 0.834, and PSNR quality of 32.02 on an average.