Denoising DTI Images Based on Regularized Filter and Fiber Tracking

The Rician noise introduced into the diffusion weighted (DW) images can bring serious impacts on tensor calculation and fiber tracking. To decrease the effects of the Rician noise, many anisotropic diffusion denoising methods have been presented. Among all these methods, the P&M (Perona &Malik)filtering methods are most popular. Although efficient in alleviating the effects of the noise, the P&M filter has shortcomings such as instability and ill‐posedness. In this paper, we analyzed the effects of the noise on the calculated tensors and tracked fibers and presented a regularized P&M filter to denoise the DW images. The presented filtering strategy is convolving the gradient used for calculating the diffusion coefficient with a non‐Gaussian smoothing kernel. To evaluate the efficiency of the regularized filter in accounting for the Rician noise, the PSNR and SMSE metrics were used. The experiment results acquired from the synthetic and real data prove the better performance of the regularized filter.