MRI reconstruction based on three regularizations: Total variation and two wavelets

This paper proposes an efficient algorithm for magnetic resonance imaging (MRI) reconstruction based on compressed sensing (CS) to further improve the accuracy of the reconstructed image and simultaneously take into account the reconstruction speed. The proposed algorithm makes full use of prior knowledge of the sparsity of the MR image in different transform domains. It uses the total variation (TV), the orthogonal Haar wavelet, and the orthogonal Daubechies20 wavelet as regularizations, on top the split Bregman iteration algorithm. Experimental results show that when the sample ratio is 20%, SNR is 30 dB, the average reconstruction PSNR of the proposed algorithm is 41.09 dB. The proposed method is compared to nonlinear conjugate gradients combined with the backtracking line-search method, split Bregman algorithm based on TV regularization method and split Bregman algorithm based on TV + orthogonal Haar wavelet regularizations method, and the average reconstruction peak signal to noise ratio (PSNR) increases 3.71 dB, 3.44 dB and 1.14 dB, respectively.