Weighted Two-level Bregman Method With Graph Regularized Sparse Coding For MRI Reconstruction

Magnetic resonance imaging is widely used in the clinical medicine diagnostic, because of its nonionizing, noninvasive high diagnostic significance and it enables excellent and precise visualization. However, there still exists problems: long time of data acquisition and low speed imaging. The proposed compressed sensing theory provides a solution to the above problem, the application of this theory in MRI can reduce the amount of data acquisition, and accelerate the rate of imaging. Nowadays, the magnetic resonance imaging reconstruction based on compressed sensing theory is developing, and the traditional MRI reconstruction methods used the fixed dictionary has undesirability imaging results. Although some proposed dictionary learning algorithms have solved the partly problems that exist in the traditional algorithm, the sparse representation of image and the quality of reconstruction remains to be improved. To solve these problems, we proposed the weighted two-level Bregman method with graph regularized sparse coding for MRI reconstruction algorithm. In this algorithm, the iteratively reweighted1l-norm is used to improve the ability of TBMDU(Two-level Bregman Method with Dictionary Updating)method, then incorporating with the graph regularized sparse coding model. In the process of the iteratively weighted TBMDU, it makes more sparsity of sampling data, and it can complete the higher quality reconstruction under the same sampling condition. Furthermore, the graph regularized sparse coding methods incorporated in WTBMGR can capture image detail information with the local data structure. Experimental results demonstrate that the proposed algorithm can reconstruct images efficiently, and outperforms DLMRI algorithm and TBMDU algorithm in terms of the peak signal-to-noise ratio(PSNR) and high-frequency error norm(HFEN) values.