Denoising of 3D magnetic resonance images via edge-enhanced low-rank tensor decomposition

Magnetic Resonance images (MRI) denoising is to obtain high quality image from infectant version. Recently, low-rank tensor (LRT) methods have been developed and attained resounding success in MRI denoising. However, these pure LRT models are incapable of utilizing the comprehensive inherent information of clean MRI. To overcome these drawbacks, we design a novel edge-enhanced low-rank tensor approximation (EELRTA) framework for Rician noise removal. The tensor gradient L0 norm regularization with describing the local structure information is incorporated into the weighted core tensor rank model for improving texture edge preservation. The application of weights can further preserve the potentially useful information distributed on the different core tensor coefficients with different physical meanings. What's more, non-local self-similarity tactic is employed for low-rank sparsity-encourage and enhancing anti-noise capability of EELRTA model. The proposed EELRTA method is tackled by an efficient alternating direction method of multipliers (ADMM). The Experiment results on simulation and multiple sclerosis lesion (MSL) data illustrate that the proposed method can effectively remove noise while reasonably retaining pathological structure information.