Structure based non-local mean method for diffusion weighted image denoising

To improve the signal to noise ratio of the diffusion-weighted image, a non-local mean based denoising method was proposed. The method constructs the weighted coefficients according to the dissimilarity of intravoxel fiber structure between two voxels and then smooths the diffusion-weighted images. The experiments on phantom data demonstrated that the proposed method was superior to the conventional non-local mean method according to the regional homogeneity. Furthermore, the intravoxel fiber structure estimating results on the data denoised by the proposed method was more consistent with the real fiber structure. The experiment results demonstrated the proposed method was an available diffusion-weighted image denoising method.