Combined Similarity to Reference Image with Joint Sparsifying Transform for Longitudinal Compressive Sensing MRI

Ruirui Kang,Long Yan,Gangrong Qu,Bin Cao

doi:10.1155/2016/4162194

Abstract

It is challenging to save acquisition time and reconstruct a medical magnetic resonance (MR) image with important details and features from its compressive measurements. In this paper, a novel method is proposed for longitudinal compressive sensing (LCS) MR imaging (MRI), where the similarity between reference and acquired image is combined with joint sparsifying transform. Furthermore, the joint sparsifying transform with the wavelet and the Contourlet can efficiently represent both isotropic and anisotropic features and the objective function is solved by extended smooth-based monotone version of the fast iterative shrinkage thresholding algorithm (SFISTA). The experiment results demonstrate that the existing regularization model obtains better performance with less acquisition time and recovers both edges and fine details of MR images, much better than the existing regularization model based on the similarity and the wavelet transform for LCS-MRI.

Highlights

Magnetic resonance imaging (MRI) is a noninvasive and nonionizing imaging modality that offers a variety of contrast mechanisms and enables excellent visualization of anatomical structures and physiological functions
To verify the performance of the proposed algorithm for LCSMRI, the experiments are designed in two different clinical MR imaging (MRI) scenarios
The performance of joint sparsifying transform is compared to the performance of single transform for longitudinal compressive sensing (LCS)-MRI

Summary

Introduction

Magnetic resonance imaging (MRI) is a noninvasive and nonionizing imaging modality that offers a variety of contrast mechanisms and enables excellent visualization of anatomical structures and physiological functions. For LCS-MRI, Weizman et al [17, 18] proposed a general sampling and reconstruction scheme, where the degree of similarity to the reference image is considered and the sparsity is exploited in both image and transform domain via the similarity and the wavelet, respectively Their method still suffers from point-like artifacts due to the inherent shortcoming of wavelet transform. It is natural to combine wavelet and Contourlet as the joint sparsifying transform to produce more physically plausible solutions instead of using either of them individually [26] Inspired by this idea, in this paper, a novel reconstruction method is proposed for the LCS-MRI, where the similarity between the reference image and the acquired image is employed to save acquisition time and improve peak signal to noise ratio (PSNR).

Proposed Model and Algorithm

Experiment Results

Conclusion