DING et al. MR Image Reconstruction Based on Comprehensive Sparse Prior 1
MR IMAGE RECONSTRUCTION BASED ON COMPREHENSIVE
SPARSE PRIOR
1
Ding Xinghao Chen Xianbo Huang Yue Mi Zengyuan
(School of Information Science and Technology, Xiamen University, Xiamen 361005, China)
Abstract In this paper, a novel Magnetic Resonance (MR) reconstruction framework which com-
bines image-wise and patch-wise sparse prior is proposed. For addressing, a truncated beta-Bernoulli
process is firstly employed to enforce sparsity on overlapping image patches emphasizing local struc-
tures. Due to its properties, beta-Bernoulli process can adaptive infer the sparsity (number of non-
zero coefficients) of each patch, an appropriate dictionary, and the noise variance simultaneously,
which are prerequisite for iterative image reconstruction. Secondly, a General Gaussian Distribution
(GGD) prior is introduced to engage image-wise sparsity for wavelet coefficients, which can be then
estimated by a threshold denoising algorithm. Finally, MR image is reconstructed by patch-wise es-
timation, image-wise estimation and under-sampled k-space data with least square data fitting. Ex-
perimental results have demonstrated that proposed approach exhibits excellent reconstruction per-
formance. Moreover, if the image is full of similar low-dimensional-structures, proposed algorithm has
dramatically improved Peak Signal to Noise Ratio (PSNR) 7~9 dB, with comparisons to other
state-of-art compressive sampling methods.
Key words Image-wise sparse prior; Patch-wise sparse prior; Beta-Bernoulli process; Low-
dimensional-structure; Compressive sampling
CLC index TP391
DOI 10.1007/s11767-012-0874-z
I. Introduction
Magnetic Resonance Imaging (MRI) is famous
for its medical diagnosis, because of its non-
invasive manner and excellent depiction of soft
tissue changes. However, the hardware scanner
rate is greatly limiting
Magnetic Resonance (MR)
imaging speed and the under-sampled measure-
ments are also badly reducing the quality of MR
images. For this reason, variety of approximate
methods have been reported to accelerate imaging
speed and at the same time they are required
keeping the image quality.
The recent algorithms based on Compressive
Sensing (CS) theory enable to recover signal using
1
Manuscript received date: April 25, 2012; revised date:
August 17, 2012.
Supported by the National Natural Science Foundation
of China (No. 30900328, 61172179), the Fundamental
Research Funds for the Central Universities (No.
2011121051), the Natural Science Foundation of Fujian
Province of China (No. 2012J05160).
Communication author: Ding Xinghao, born in 1977,
male, Professor. School of Information Science and
Technology, Xiamen University, Xiamen 361005, China.
Email: dxh@xmu.edu.cn.
data with significantly less measurements than
regular, provided the sparsity of the underlying
signal and sophisticated reconstruction procedure.
The research of MR reconstruction based on this
technique have shown promising results
[1–7]
, In
particular, the redundancy of the MR data sam-
pled from frequency domain and implicit sparsity
of MR images have motivated many researchers to
study the application of CS to fast MR Imaging
(CS-MRI). As we know, the sparsity of MR image
are key to accurate CS reconstruction, but the
CS-MRI is limited to use nonadaptive sparsifying
transform (also some set of signal known as dic-
tionary) and Total Variation (TV), that is clearly
ignoring many local structure feature. However,
the application of clinical diagnosis often demands
preserving important diagnostic information (in
particular, sharp edges and local structures). Re-
cent research K-SVD
[8]
has demonstrated the sig-
nificant advantages of learning an over-complete
dictionary matched to the images of interest, as
the K-SVD found that the adaptive dictionary is
advantageous to preserve local feather and match
the original signal. Taking advantage of this