Dictionary Learning for MRI Denoising based on Modified K-SVD

Abstract

Magnetic resonance imaging (MRI) is one of most powerful medical imaging tools. However, the quality is affected by the noise pollution during the acquisition and transmission. A novel method is presented for adaptively learning the sparse dictionary while simultaneously reconstructing the image from noisy image data. The method is based on a K-singular value decomposition (K-SVD) algorithm for dictionary training on overlapping image patches of the noisy image. A modified dictionary update strategy with an effective control over the self-coherence of the trained dictionary is raised during the dictionary learning. The learned dictionary is employed to achieve effective sparse representation of the corrupted image and used to remove Rician noise, which shows a good performance in both noise suppression and feature preservation. The proposed method was compared with some current MRI denoising methods and the experimental results showed that the modified dictionary learning could obtain substantial benefits in denoising performance.