Sparse Representation Based MRI Denoising with Total Variation

Diffusion tensor magnetic resonance imaging is a newly developed imaging technique: however, this technique is noise sensitive. This paper presents a novel method for sparse representation denoising of MR images that propose sparse representation of the corrupted images with the knowledge of the Rician noise model. The proposed model inferring the prior that MR images are composed of several separated regions with uniform intensity, therefore, total variation can be combined to further smooth every, region. Since sparse representation performs well in extracting features from images, coupled with the total variation regularization, the method offers excellent combination of noise removal and edge preservation. The experiment results demonstrate that the proposed method preserves most of the fine structure in cardiac diffusion weighted images.