Compressed Sensing MRI Reconstruction with Multiple Sparsity Constraints on Radial Sampling

Jianping Huang,Yuemin Zhu,Lihui Wang

doi:10.1155/2019/3694604

Abstract

Compressed Sensing Magnetic Resonance Imaging (CS-MRI) is a promising technique for accelerating MRI acquisitions by using fewer k-space data. Exploiting more sparsity is an important approach to improving the CS-MRI reconstruction quality. We propose a novel CS-MRI framework based on multiple sparse priors to increase reconstruction accuracy. The wavelet sparsity, wavelet tree structured sparsity, and nonlocal total variation (NLTV) regularizations were integrated in the CS-MRI framework, and the optimization problem was solved using a fast composite splitting algorithm (FCSA). The proposed method was evaluated on different types of MR images with different radial sampling schemes and different sampling ratios and compared with the state-of-the-art CS-MRI reconstruction methods in terms of peak signal-to-noise ratio (PSNR), feature similarity (FSIM), relative l2 norm error (RLNE), and mean structural similarity (MSSIM). The results demonstrated that the proposed method outperforms the traditional CS-MRI algorithms in both visual and quantitative comparisons.

Highlights

Magnetic resonance imaging (MRI) has been widely used in the radiological diagnosis due to its high spatial resolution, noninvasive, and nonionizing radiation merits
We propose a novel Compressed Sensing Magnetic Resonance Imaging (CS-MRI) framework based on multiple sparse priors to increase reconstruction accuracy
The results demonstrated that the proposed method outperforms the traditional compressed sensing (CS)-MRI algorithms in both visual and quantitative comparisons

Summary

Introduction

Magnetic resonance imaging (MRI) has been widely used in the radiological diagnosis due to its high spatial resolution, noninvasive, and nonionizing radiation merits. In the pioneering work of CS-MRI, Lusting et al [7] reconstructed magnetic resonance (MR) images from the Cartesian undersampled k-space data by solving a l1 norm minimization equation with wavelet transform and total variation (TV) sparsity constraints. The wavelet coefficients of MR images are approximately sparse, and have a tree structure The latter provides better image reconstruction quality than standard wavelet sparsity prior [27,28,29] and enables us to exploit the correlation between the parent and child of wavelet coefficients and reduce the required number of k-space data for MR image reconstruction.

Methods

Experimental Results and Discussion

Conclusion