Available online at www.sciencedirect.

com

Magnetic Resonance Imaging xx (2011) xxx – xxx

Joint reconstruction of multiecho MR images using correlated sparsity

Angshul Majumdar ⁎, Rabab K. Ward
Department of Electrical and Computer Engineering, University of British Columbia, Kaiser 2010, Vancouver, BC, Canada V6T1Z4
Received 14 September 2010; revised 20 March 2011; accepted 23 March 2011

Abstract

This works addresses the problem of reconstructing multiple T1- or T2-weighted images of the same anatomical cross section from
partially sampled K-space data. Previous studies in reconstructing magnetic resonance (MR) images from partial samples of the K-space used
compressed sensing (CS) techniques to exploit the spatial correlation of the images (leading to sparsity in wavelet domain). Such techniques
can be employed to reconstruct the individual T1- or T2-weighted images. However, in the current context, the different images are not really
independent; they are images of the same cross section and, hence, are highly correlated. We exploit the correlation between the images,
along with the spatial correlation within the images to achieve better reconstruction results than exploiting spatial correlation only.
For individual MR images, CS-based techniques lead to a sparsity-promoting optimization problem in the wavelet domain. In this article,
we show that the same framework can be extended to incorporate correlation between images leading to group/row sparsity-promoting
optimization. Algorithms for solving such optimization problems have already been developed in the CS literature. We show that significant
improvement in reconstruction accuracy can be achieved by considering the correlation between different T1- and T2-weighted images. If the
reconstruction accuracy is considered to be constant, our proposed group sparse formulation can yield the same result with 33% less K-space
samples compared with simple sparsity-promoting reconstruction. Moreover, the reconstruction time by our proposed method is about two to
four times less than the previous method.
© 2011 Elsevier Inc. All rights reserved.

Keywords: T1- and T2-weighted MR imaging; quantitative MRI; optimization; fast imaging; compressed sensing; group-sparsity; MMV

1. Introduction The traditional method to reconstruct T1/T2-weighted

This work aims at reconstructing multiple T1- or T2-
weighted magnetic resonance (MR) images from their partial
K-space samples (Fourier coefficients). T1- and T2-weighted
imaging acquire scans of the same anatomical slices at
different repetition and echo times. These images correspond
to the same anatomy, but the tissue contrast in each image
varies from one image to another. For a particular T1/T2
weight, the contrast between tissue A and tissue B may be
high in a certain image, but for another T1/T2 weight, the
contrast between these two tissues may be low. Detailed
discussion on this matter is beyond the scope of this work;
the interested reader is encouraged to read the work of
Haacke et al. [1]. When compared with photography, such
images correspond to multiple shots of the same scene taken
at varying exposures.
⁎ Corresponding author.
E-mail addresses: angshulm@ece.ubc.ca (A. Majumdar),
rababw@ece.ubc.ca (R.K. Ward).
0730-725X/$ – see front matter © 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.mri.2011.03.008
images is to acquire dense K-space scans on a Cartesian grid
and apply an inverse Fourier transform to obtain the
corresponding T1/T2-weighted images. Such dense sampling
of the K-space, however, results in high signal acquisition
time. The state-of-the-art magnetic resonance imaging (MRI)
reconstruction methods aim at sparsely sampling the K-space
to reduce the data acquisition time and applying smart
reconstruction algorithms to obtain the image. The concept of
reducing the scan time by reducing the number of acquired
K-space samples is called “acceleration.”
Acceleration can be either hardware based or software
based. The hardware-based acceleration uses multiple
receiver coils to scan the K-space, but they scan the K-
space sparsely. Depending on their field of view, each coil
has a sensitivity profile. There are different techniques to
obtain the final image by combining the partial K-space
scans with the information about the sensitivity profiles.
However, hardware-based acceleration is not the topic of this
article and will not be discussed further. In this work, we are
interested in the rapidly developing field of software-based
2 A. Majumdar, R.K. Ward / Magnetic Resonance Imaging xx (2011) xxx–xxx
acceleration. Software-based acceleration is based on the
theory of compressed sensing (CS) [2,3]. Compressed
sensing-based techniques exploit the spatial correlation in
an MR image. They obtain only partial K-space samples
[4,5] of an image and then reconstruct it from these partial
samples by exploiting the correlation present in the image.
In this article, we propose to further reduce the number of
required partial K-space samples (thereby increasing accel-
eration) than required by conventional CS technique for
reconstructing T1/T2-weighted images by exploiting the
interimage correlation present among the different images as
well as the spatial correlation within the image.
Compressed sensing exploits only the spatial correlation
of an MR image. In the current problem, one should
remember that the T1/T2-weighted images correspond to the
same cross section. Thus, there is correlation among different
images of the T1/T2-weighted images. We propose to
exploit both the spatial correlation in individual images as
well as the correlation among the images to further reduce
the K-space required for reconstruction.
The problem of jointly reconstructing multiple T1- or T2-
weighted MR images has not be identified or attempted
before. In this work, we will show how this problem can be
formulated into either a group sparse optimization problem
or a row sparse multiple measurement vector (MMV)
optimization problem. These problems can be solved using
existing algorithms.
The rest of the article is organized into several sections.
The following section discusses the mathematical formulation
behind the T1/T2 joint reconstruction problem. In Section 3,
the experimental results are provided. Finally in Section 4, the
conclusions of this work and future directions are discussed.
2. Reconstruction of T1/T2-weighted images
As mentioned earlier, most traditional MR scanners
collects dense K-space samples on a uniform Cartesian grid.
Reconstructing the image from such samples is trivial—
applying an inverse Fourier transform. However, such dense
sampling is time consuming, and one of the main challenges in
MRI lies in reducing the MR acquisition time by reducing the
number of K-space samples collected. First, we will discuss
the reconstruction of single MR image from partially sampled
K-space data. Later, we will describe how further reduction in
the number of samples can be achieved by reconstructing all
the T1 (or T2)-weighted images simultaneously.
2.1. Reconstructing single MR images
The mathematical representation expressing the partially
sampled K-space data acquisition model for the underlying
image can be expressed as
ym
1 = Fm
N 2 xN 2
1 + gm
1 ; mVN 2 a
or y = Fx + g
where y is the K-space samples, x is the underlying image
2×1
expressed in vectorized form, F: ℝN →ℂm×1 is the Fourier
mapping from the real image space to the complex frequency
domain and η is white Gaussian noise.
The underdetermined inverse problem (1) does not
have a unique solution. To reconstruct the underlying
image, some prior information is necessary. In CS-
based MR reconstruction techniques, it is assumed that
the image to be reconstructed is approximately sparse
in the wavelet domain [4,5]. This turns out to be a
realistic assumption and yields extremely good recon-
struction results when partially sampled K-space data
are used.
The wavelet analysis and synthesis equations are:
Analysis : a = Wx ð2aÞ
Synthesis : x = W T a ð2bÞ
The wavelet coefficient vector α is approximately s-
sparse, i.e., only s coefficients are nonzeroes, while the rest
are zeroes or very close to zero. Wavelet transform is
orthogonal, in which case, the dimensionality of the original
image and the transform coefficients is the same, or the
wavelet transform is a tight-frame, in which case, the
dimensionality of the wavelet coefficients is larger. In this
work, we will be using orthogonal wavelets.
Incorporating the wavelet coefficients into Eq. (1),
y = FW T a + g ð3Þ
Instead of solving for the image x directly, CS solves for
the wavelet coefficient of the image (α) (Eq. (3)). CS aims at
recovering only the s high-valued wavelet coefficients. The
coefficients that are close to zero are indistinguishable from
noise and cannot be recovered.
In the pixel domain, an N×N image under consideration
has N2 unknowns. However, the pixel values are highly
correlated locally. The wavelet transform effectively
“whitens” the image by removing the spatial correla-
tions [6]. This leaves s nearly independent high-valued
wavelet coefficients.
The inverse problem (Eq. (3)) has effectively only 2s
unknowns—s positions and s values. Intuitively, solving for
these 2s unknowns will require far fewer equations than
directly solving for the image, which has N2 unknowns.
Compressed sensing-based recovery algorithms solves for
the sparse wavelet transform coefficients by solving the
following optimization problem [4],
argmin jjajj1 subject tojjy − AW T ajj2 Vr ð4Þ
ð1Þ where σ is proportional to the S.D. of noise.
 A. Majumdar, R.K. Ward / Magnetic Resonance Imaging xx (2011) xxx–xxx
To get a reasonably good estimate of the signal via l1
minimization, the number of K-space samples needed is [3],


N2
m = O slog
s T1/T2-weighted images
Once the wavelet coefficients are recovered, the image
is reconstructed by applying the wavelet synthesis
equation (2b).
2.2. Sparse reconstruction of T1/T2-weighted images
The T1- or T2-weighted images are obtained by imaging
the same object at different repetition and echo times.
Suppose, the K-space data for T such weighted images have
been acquired. This is represented by,
yi = Fxi + gi ; i = 1:::T
where yi is the K-space sample for the ith-weighted image xi.
The problem is to reconstruct the T1/T2-weighted images
(xi's) given their K-space samples (yi's).
A straightforward application of CS will apply Eq. (4) to
reconstruct each T1- or T2-weighted individually image
(for each i).
In this work, we aim at the joint reconstruction of all the
T1- or T2-weighted images. First, we show how a simple
sparsity-based joint reconstruction problem can be formu-
lated for these T1/T2-weighted images.
Incorporating the wavelet transform, Eq. (6) can be
expressed concisely as follows,
2 3 2 32 3 2 3
y1 FW T 0 0 a1 g1
4 ::: 5 = 4 0 ::: 0 54 ::: 5 + 4 ::: 5
yT 0 0 FW T aT gT
or y = Ua + g
The wavelet coefficients (αi's) for each T1/T2-weighted
image are sparse. Therefore, the combined vector α̂ is also
sparse. Hence, we can invoke a CS recovery algorithm,
namely the l1 minimization to jointly solve for the
wavelet coefficients,
argmin jjãjj subject to jjỹ− Uã jj Vr̃
ã 1
X 2
where jjajj = jaj j
j
Once the wavelet coefficients are obtained, the individual
images can be obtained by applying the synthesis equation.
The sparsity-based joint reconstruction problem does not
exploit any further information apart from spatial correlation.
As discussed earlier, the wavelet transform gets rid of the
spatial correlation in an image and leads to “almost
independent” set of sparse coefficients. However, as will
be shown below, in the problem of reconstructing T1- or T2-
weighted images, we can exploit the correlation among
3
different images of the same cross section to achieve better
reconstruction accuracy.
ð5Þ 2.3. Group/row sparse reconstruction of
T1- or T2-weighted images are obtained by changing the
repetition time or echo times of the magnetization. Such T1/
T2 weightings lead to differential contrast among the tissues.
However, it must be remembered that all these images are
different representations of the same anatomical cross
section. For a particular T1- or T2-weighted image, the
contrast between two tissues may be high, leading to a sharp
edge, while for some other T1 or T2 weighting, the contrast
may be reduced and the edge may be less pronounced.
However, since the underlying anatomy is the same, the
ð6Þ positions of the tissue boundaries (edges) are preserved.
The wavelet transform effectively encodes the edges in an
image. When the edge is more pronounced, the wavelet
coefficients are high. When the edge is not so sharp, the
wavelet coefficients are low. In smooth areas, the wavelet
coefficients are zero. The following toy example illustrates
the fact.
Consider a small image matrix with sharp boundary for a
particular T1 weighting (Fig. 1A). The 0's correspond to
tissue B and the 1's correspond to tissue W. The Haar
wavelet transform (Fig. 1B) captures the vertical disconti-
nuity between W and B perfectly. The values in the last
column should be ignored, the wavelet transform assumes
that the signal is periodic and computes the boundary
between B and W.
Fig. 1 shows that when the tissue boundary is
pronounced, the Haar wavelet coefficients along the
ð7Þ boundary are high. Now consider a different value of T1
weighting, such that the contrast between W and B is less
pronounced (Fig. 2A). The Haar wavelet transform of this
matrix is shown in Fig. 2B. We can see now that although the
position of the high-valued Haar coefficients is the same,
their values have changed. As the contrast is reduced, the
value of the wavelet coefficients is reduced as well.
This is a toy example, but this corroborates the fact that as
ð8Þ
Fig. 1. T1 weighting: (A) tissue boundary and (B) wavelet coefficients.
4 A. Majumdar, R.K. Ward / Magnetic Resonance Imaging xx (2011) xxx–xxx
Fig. 2. T1 weighting: (A) tissue boundary and (B) wavelet coefficients.
long as the anatomy of the brain slice does not change, the
positions of the high-valued wavelet transform coefficients
will not change for different T1 or T2 weightings.
Mathematically, this means that there should be a high
degree of mutual correlation between any two T1- and T2-
weighted images of the same anatomical cross section.
Fig. 3 shows the scatter plot between the wavelet
coefficients of two randomly chosen T2-weighted images
of rat's spinal cord. The plot shows that the correlation is
almost perfectly linear.
The linear relationship corroborates our physical under-
standing of the fact why wavelet coefficients of two T1- or
T2-weighted images have similar valued coefficients at
similar positions.
2.3.1. Group sparse formulation
Since the wavelet transform we are considering is
orthogonal, the wavelet coefficient vector for each image
(αi) is N2 (same as the number of pixels in the image). The
coefficients of the vector α̃ in Eq. (14) can be grouped
Fig. 3. Scatter plot of wavelet coefficients.
according to their positions as shown in Fig. 4. If we consider
the T coefficient vector α1,r, … αT,r as one group, then all the
coefficients in the vector α̃ form N2 groups, each group
having T coefficients.
We have argued above that the wavelet coefficients
should have similar values at similar positions. It is known
that the wavelet transform coefficient vectors of individual
MR images are sparse. If each αi is approximately s-sparse,
then it follows from the argument that the high-valued
coefficients of the different αi's will correspond to similar
positions; we can say that the vector α̃is approximately
“s-group sparse,” i.e., there are approximately only s groups
that have high-valued wavelet coefficients and the rest
of the groups have zero coefficients or coefficients close
to zero.
Therefore, instead of solving the l1 minimization (8), one
needs to incorporate group sparsity into the optimization
problem. Previous studies [7–9] have proposed the follow-
ing mixed l2,1 minimization for solving the group sparsity-
promoting optimization problem,
argminjjãjj2;1 subject to jjỹ − Uã jj Vr̃ ð9Þ
ã !1 = 22
X
N2 XT
where jjajj2;1 = a2k;r :
r=1 k =1

1 = 2
P
T
The l2-norm a2k;r over the groups of correlated
k −1
k,r
coefficients α k=1…T promotes a dense solution within the
2
group (k), whereas the summation over the l -norm enforces
group sparsity, i.e., the selection of only a few groups.
Although group sparse optimization is being introduced for
the first time in MR image reconstruction, it has been used
(with slightly different measures on group sparsity) previ-
ously in color imaging [10–12].
2.3.2. Row sparse MMV formulation
The fact that wavelet coefficients have similar values at
similar areas of the different T1/T2-weighted images can be
exploited in a slightly different fashion. Instead of stacking
 A. Majumdar, R.K. Ward / Magnetic Resonance Imaging xx (2011) xxx–xxx 5

Fig. 4. Grouping of wavelet coefficients according to their position.

the multiple T1/T2-weighted images as vectors as in Eq. (7), discussed in Section 2.2. It is assumed that each of the T1/
we stack them as columns of a matrix, T2-weighted images is approximately s-sparse. As there are
T such weighted images, the vector of the wavelet
Y = FW T â + ˆ ð10Þ coefficients of all images α̂ is approximately s·T -sparse.
Recovering ans·T -sparse  vector via l1 minimization requires
where Y=[y1…yT], α̂=[α1…αT] and η̂=[η1…ηT] and and FWT 2
about M1 = O Tslog Ns K-space samples [15].
operates on each column of α̂.
However, we have argued in Section 3.3. that α̂ is
In this arrangement, coefficients with the same positions
actually s-group sparse (each group having T coefficients).
are along the rows of the matrix α̂ (see Fig. 5). This
Sparsity-promoting optimization via l1 minimization only
arrangement (10) is called the MMV formulation.
utilizes the information that each image is spatially
As argued earlier, the wavelet coefficients αk,r for T1/T2-
correlated. Spatial correlation leads to sparsity in the wavelet
weighted images of the same cross section will have similar
domain. In this work, we have identified that T1/T2-
values at the same positions. Thus, for each image k, if the
weighted images are correlated among themselves. There-
wavelet coefficient vector αk is s-sparse, then for the matrix α̂
fore, we exploit both the spatial correlation within each
only s-rows will have significantly high-valued coefficients
image and interimage correlation among different T1/T2-
and the rest will be zeroes or close to zeroes. Therefore,
weighted images. This leads to our proposed group sparsity-
according to the MMV formulation, the solution to Eq. (10)
promoting optimization problem.
should be approximately s-row sparse, i.e., s-rows of the
Instead of reconstructing a vector that is s·T-sparse (as
matrix α̂ will have high-valued coefficients, while the rest of
done by l1 minimization), we reconstruct a vector that is
the rows are zeroes or close to zero.
known to be s-group sparse (each group having T co-
To solve the inverse problem (10) with row-sparsity
efficients). We employ l2,1 minimization (9) to reconstruct
constraints, the following optimization has been proposed by
such a group sparse vector. The number of samples required
van den Berg and Friedlander [13],
to reconstruct an s-group sparse vector having  T coefficients
argmin jjâjj2;1 subject to jjY − FW T âjjFrobenius Vâ in each group is M2;1 = O Ts + slog Ns
2
[15]. This is
!1 = 2
ã X
N2 X T significantly less than the number of samples required by
where jjâjj2;1 = ak;r
2
: ð11Þ l1 minimization (M1). The number of measurements required
r=1 k =1 by the row sparse MMV formulation (11) is the same as
group sparse formulation.
The objective function for both the group sparsity
Thus, our proposed methods (l2,1 minimization and MMV)
problem (9) and the row sparsity problem (11) are the
expect to achieve the same level of reconstruction accuracy but
same, only the constraints are different.
with lesser number of K-space samples than required by l1
The MMV formulation has previously been used for
minimization. In other words, with the same number of K-space
seismic imaging problems [14] but is being introduced in
samples, our proposed group/row sparsity-based methods will
this article for the first time in the context of MR
yield better reconstruction results.
image reconstruction.
2.4. Benefit of group/row sparse reconstruction
3. Experimental evaluation
In the problem of joint reconstruction of T1/T2 maps by
sparse optimization (8), l1 minimization is employed as The experimental evaluation was carried out on ex vivo
and in vivo T2-weighted images of a rat's spinal cord. The

Table 1
Reconstruction results (NMSE) for ex vivo data
Reconstruction method 40 Lines 80 Lines 120 Lines
CSMRI 0.14 0.06 0.04
ExtSparseMRI 0.14 0.06 0.04
GroupSparseMRI 0.11 0.03 0.01
RowSparseMRI 0.11 0.03 0.01
Fig. 5. Grouping of wavelet coefficients according to their position.
6 A. Majumdar, R.K. Ward / Magnetic Resonance Imaging xx (2011) xxx–xxx

Table 2 acquired with 6.728 ms echo time and consecutive echoes

Reconstruction results (NMSE) for in vivo data with the echo spacing of 6.738 ms).
Reconstruction method 40 Lines 80 Lines 120 Lines The aim of this work is to reduce the scanning time for
CSMRI 0.22 0.16 0.13 MRI. Therefore, radial scanning [16,17] of the K-space
ExtSparseMRI 0.22 0.16 0.13 was used. This is because radial sampling trajectories are
GroupSparseMRI 0.17 0.11 0.09 the fastest K-space acquisition techniques. In this work,
RowSparseMRI 0.17 0.11 0.09
for each angular direction of the radial sampling pattern,
128 points were sampled. The mapping between the non-
data were collected with a 7-T MRI scanner. The original Cartesian K-space and the Cartesian image space is
data consisted of a series of 32 echoes acquired with a spin obtained via the nonuniform fast Fourier transform
echo sequence with increasing echo time (first echo was [18,19].

Fig. 6. From top to bottom—reconstructed images of the same slice for 4 different T2 weightings; left column—ground truth; middle column—CSMRI
reconstruction; right column—GroupSparseMRI reconstruction.
 A. Majumdar, R.K. Ward / Magnetic Resonance Imaging xx (2011) xxx–xxx
We have evaluated four strategies to reconstruct the
T2-weighted image from their partial radial K-space
samples. The first one is the simple application of the
CSMRI that employs l1 minimization for reconstructing
the T2-weighted images individually. This serves as our
baseline. In this work, we have proposed three methods
for the joint reconstruction of all the T2-weighted image.
The first one employs l1 minimization on the entire data
set (8) for simultaneous reconstruction of all the T2-
weighted images. This method does not exploit further
information apart from sparsity. It is a straightforward
extension of the CSMRI technique, and we call it
ExtSparseMRI.
The main contribution of this article is to show that in the
case of T1/T2-weighted images, the correlation among
different images can be exploited to further improve the
reconstruction results. Using this correlation information, we
formulated the joint reconstruction problem as a group
sparsity-promoting optimization problem (9) or as a row
sparsity-promoting MMV problem (11). We will call these
reconstruction techniques GroupSparseMRI (16) and Row-
SparseMRI (17), respectively.
Tables 1 and 2 show the reconstruction accuracy in terms
of the normalized mean squared error (NMSE) for the ex
vivo and the in vivo data. The numbers of the radial K-space
lines are 40, 80 and 120. Two-dimensional Haar wavelet is
used as the sparsifying transform.
Tables 1 and 2 show the NMSE up to two significant
digits. As far as reconstruction accuracy is concerned, finer
variations (higher than two significant digits) are not
discernible. We see that the reconstruction accuracy of
both the baseline algorithm CSMRI and its extended version
ExtSparseMRI give the same reconstruction results. This is
to be expected because both of these methods exploit the
same information, i.e., the sparsity of the MR image in the
wavelet domain. The results from both these methods are
worse than our proposed GroupSparseMRI and RowSpar-
seMRI techniques. The results for the ex vivo data are better
than in vivo data; this is expected since in vivo data is
inherently more noisy [20].
When the correlation among the different T2-weighted
images is also exploited along with the spatial correlation
(sparsity in wavelet domain), better reconstruction results are
achieved as can be seen from Tables 1 and 2. Thus, the group
sparsity-promoting optimization and the row sparsity-
promoting MMV problem yield better reconstruction results
than sparse optimization. It is not surprising that both
Table 3
Reconstruction times (seconds) for ex vivo data
Reconstruction method 40 Lines 80 Lines 120 Lines
CSMRI 1908 2347 3592
ExtSparseMRI 2576 4709 4714
GroupSparseMRI 730 1116 1939
RowSparseMRI 710 931 1117
7
Table 4
Reconstruction times (seconds) for in vivo data
Reconstruction method 40 Lines 80 Lines 120 Lines
CSMRI 3138 2724 4879
ExtSparseMRI 4191 6730 6768
GroupSparseMRI 851 931 1823
RowSparseMRI 938 1231 1952
GroupSparseMRI and RowSparseMRI yield the same
reconstruction accuracy since both of these methods use
the same information—spatial correlation (in terms of
sparsity in wavelet domain) and the correlation among
different T2-weighted images.
Tables 1 and 2 show that the reconstruction accuracy
achieved by CSMRI and ExtSparseMRI using 120 radial
sampling lines is the same as that of GroupSparseMRI and
RowSparseMRI with 80 radial lines (for both ex vivo and in
vivo data). This means, if we keep the reconstruction
accuracy the same, our proposed method will cut down the
data acquisition time by about 33%.
As this is an imaging experiment, the readers might also
be interested in the visual quality of the results. Unfortu-
nately, it is not possible to show all the 32 T2-weighted
images for all the experimental configurations and for both
ex vivo and in vivo data. We only show results of the in vivo
data since these data are more challenging. We chose four
T2-weighted images to show the reconstruction results from
the baseline method—CSMRI and one of our proposed
methods—GroupSparseMRI (Fig. 6).
The qualitative assessment shows that the reconstructed
images contain some blocky artifacts. The artifacts from
GroupSparseMRI are lower than that from CSMRI. The
“blockiness” of the artifacts arises from the use of Haar
wavelets. Using more sophisticated wavelets such as
complex dual-tree or fractional spline will improve the
visual quality.
To show the speed of the different reconstruction
algorithms, Tables 3 and 4 presents the time required (in
seconds) for reconstructing all the 32 T2-weighted images.
The results show that, depending on the number of lines,
GroupSparseMRI and RowSparseMRI are about two to four
times faster than CSMRI and ExtSparseMRI.
Compressed sensing-based MRI has to solve an
optimization problem for each T2 weighting, whereas
others jointly solve for all the T2-weighted images and
hence solves a single optimization problem. Although
CSMRI solves a smaller optimization problem compared
with others, the overall time taken is longer, as it has ran
the optimization repeatedly. The ExtSparseMRI takes the
longest. This is because it solves a large problem but does
not account for the correct structure (interimage correla-
tion) of the solution. As a result, it takes a long time to
converge. Our proposed formulations (GroupSparseMRI
and RowSparseMRI) solve a large problem, but since they
account for the correct structure (intraimage and inter-
image correlation) in the solution, they converge quickly.
8 A. Majumdar, R.K. Ward / Magnetic Resonance Imaging xx (2011) xxx–xxx
In this work, we experimented on T2-weighted images
since these were the only data available to us. However, the
techniques in this work are applicable to T1-weighted image
reconstruction as well.
4. Conclusion
This work looks at the problem of reconstructing a set
of T1- or T2-weighted MR images from their partially
sampled K-space data. Such a set of images is collected
to construct T1 or T2 maps of the anatomical cross
section under study [21,22].
Compressed sensing-based techniques have previously
been used to successfully reconstruct individual MR images
from their partial Fourier measurements. For the T1/T2
imaging problem, we show that instead of reconstructing the
images individually by only exploiting their sparsity in
wavelet domain (spatial correlation) better results can be
achieved when further information regarding the correlation
amongst the different T2-weighted images is exploited along
with spatial correlation.
We have shown that the correlation among the T1- or T2-
weighted images of the same anatomical cross section can be
used to formulate a group sparsity-promoting optimization
problem or a row sparsity-promoting MMV problem. Our
proposed technique yields significant improvement in the
reconstruction results. By thorough experimentation on ex
vivo and in vivo T2-weighted images of a rat's spinal cord,
we have shown that it is possible to reduce the sampling
requirement by around 33% (leading to faster scan times)
compared with standard CS-based methods. Besides, our
proposed reconstruction is also significantly faster (two to
four times).
This study looks at the problem of acquiring fast MRI
scans from a signal processing perspective. Previously
hardware-based acceleration methods were used for acquir-
ing T1- or T2-weighted images [23]. In the future, we will
extend our proposed group/row sparsity-based reconstruc-
tion algorithms to the hardware-based acceleration methods
to achieve further speed-ups in data acquisition.
References
[1] Haacke EM, Brown RR, Thomson MR, Venkatesh R. Magnetic
resonance imaging: physical principals and sequence design. New
York: John Wiley & Sons; 1999.
[2] Donoho D. Compressed sensing. IEEE Trans Info Theory 2006;52(4):
1289–306.
[3] Candès EJ, Romberg J. Tao, Robust uncertainty principles: exact
signal reconstruction from highly incomplete frequency information.
IEEE Trans Info Theory 2006;52(2):489–509.
[4] Lustig M, Donoho DL, Pauly JM. Sparse MRI: the application of
compressed sensing for rapid MR imaging. Magn Reson Med 2007;58
(6):1182–95.
[5] Trzasko J, Manduca A. Highly undersampled magnetic resonance
image reconstruction via homotopic 10-minimization. IEEE Trans
Med Imaging 2009;28(1):106–21.
[6] Strickland RN, Hee HI. The wavelet transform as a multi-
resolution matched filter and zero-crossing detector for detecting
microcalcifications in mammograms. Eng Med Biol Mag 1995:
1047–8.
[7] Eldar YC, Kuppinger P, Bolcskei H. Block-sparse signals: uncertainty
relations and efficient recovery. IEEE Trans Signal Proc 2010;58(6):
3042–54.
[8] Stojnic M, Parvaresh F, Hassibi B. On the reconstruction of block-
sparse signals with an optimal number of measurements, arXiv:
0804.0041v1.
[9] Huang J, Zhang T. The Benefit of Group Sparsity, arXiv:0901.2962v2.
[10] Majumdar A, Ward RK. Compressive color sensing of color images.
Signal Proc 2010;90(12):3122–7.
[11] Majumdar A, Ward RK. Compressive color imaging with group-
sparsity on analysis prior. Int Conf Image Proc 2010 (in press).
[12] Majumdar A. Ward, Compressive non-convex group sparsity:
application to color imaging. Int Conf Acoustics, Speech Signal Proc
2010:469–72.
[13] van den Berg E, Friedlander MP. Theoretical and empirical results for
recovery from multiple measurements. IEEE Trans Info Theory
2010;56(5):2516–27.
[14] Herrmann FJ. Compressive imaging by wavefield inversion
with group sparsity, SEG, 2009, Houston. Technical Report
TR-2009-01.
[15] Baraniuk RG, Cevher V, Duarte MF, Hegde C. Model-based
compressive sensing, arXiv:0808.3572v5.
[16] Xie J, Lai P, Huang F, Li Y, Li D. Cardiac magnetic resonance imaging
using radial k-space sampling and self-calibrated partial parallel
reconstruction. Magn Reson Imaging 2010;28(4):495–506.
[17] Zhang S, Block KT, Frahm J. Magnetic resonance imaging in real time:
advances using radial FLASH. J Magn Reson Imaging 2010;31(1):
101–9.
[18] Matej S, Fessler JA, Kazantsev IG. Iterative tomographic image
reconstruction using Fourier-based forward and back-projectors. IEEE
Trans Med Imaging 2004;23(4):401–12.
[19] O'Connor YZ, Fessler JA. Fourier-based forward and back-projectors
in iterative fan-beam tomographic image reconstruction. IEEE Trans
Med Imaging 2006;25(5):582–9.
[20] Crozier S, Field J, Doddrell DM. Signal-to-noise ratio improvements in
in vivo high resolution micro-volume selected spectroscopy. Magn
Reson Imaging 1990;8(3):277–83.
[21] Laule C, Kolind SH, Bjarnason TA, Boon Li DK, MacKay AL. In vivo
multiecho T2 relaxation measurements using variable TR to decrease
scan time. Magn Reson Imaging 2007;25(6):834–9.
[22] Deichmann R, Adolf H, Nöth U, Morrissey S, Schwarzbauer C. A.
Haase, Fast T2-mapping with SNAPSHOT FLASH imaging. Magn
Reson Imaging 1995;13(4):633–9.
[23] Block KT, Uecker M, Frahm J. Model-based iterative reconstruction
for radial fast spin-echo MRI. IEEE Trans Med Imaging 2009;28(11):
1759–69.