Patch Wise Ornstein-Uhlenbeck Bridge Diffusion for Low Dose CT Denoising

Abstract

Despite uncertainty about the nature of low-dose CT degradation, denoising them using deep learning methods gives good results. In this work, Patch wise Ornstein-Uhlenbeck Bridge (POUB) diffusion is used for LDCT denoising. The results are compared with existing state of-the-art methods in Mayo 2016 dataset. The quantitative performance reveals that the proposed work outperforms the third-best compared method by +0.9741 dB PSNR, +0.47% SSIM. This method outperforms the best compared method in terms of SSIM and FSIM, proving that anatomical information is preserved even after accomplishing the complex denoising task. The high contrast between the lesion and the back ground in the resulted denoised CT proves that the method can be helpful for further abnormality diagnosis using radiologists or computer assisted means. The success of Patch wise Ornstein Uhlenbeck diffusion model for LDCT denoising shows that the bridge diffusion models have huge potential for medical image restoration.