highly accelerat ed parameter mapping with joint partial ...tv( ) us and sparsity united states,...
TRANSCRIPT
1Depa
INTRMR pamappiing (Cthis wmappicantly METHConsidimagedata caly proping [3
assum
formed
as P =
coefficthe spdesignspace equivastraint
sU for
where zation VARP RESUThe prin a bscannetimes. ing, anunderspling preconsseparanique. reconsThe Thigher DISCSeveraPCA iand spphysic CONCA newmetho REFE[1] Liapp. 1591973. p
Hig
artment of Electrica
RODUCTION arametric mappinng has been limit
CS) [2] have showwork, we present
ng. The proposedreduced imaging
HODS der T2 mapping s with variable Tan incur severe arposed reconstruc][4]. The same a
mes ( , )ntρ =∑r
d from samples o
s tU V , where tV
cients. Under thearse sampling op
ned to determine data is used as
alent to solving Uts can effectively r the PS model, a
TV(·) representswith continuatio
PRO (variable pro
ULTS roposed techniqu
brain T2 study. Ted at 3T with a sThe imaging par
nd 256x208 matrisampled at an acpattern shown in structed with (1)able model [6]; a Figure 2 compa
structions using (2 map estimated
r spatial resolution
USSION al sparse samplinis based on the bparsity constraintscally acquired trai
CLUSION w image reconstru
d shows superior
ERENCES ang Z-P, IEEE-ISB93-1596. [5] Donepp. 413-432, 1973
ghly Accelerat
al and Computer Ean
ng provides quantted by long scan
wn great potentiala new method to
d method is showg time.
as an example, aT2-weightings. Wrtifacts in the rection methods usi
approach can be
1( ) ( )
Ll l nl
u v t=
r , w
of ( , )ntρ r such t
represents the te
e PS model, the mperator and ξ is Vt. In Fig.1, wetemporal traininsU . However, dirstabilize the ill-c
and the reconstruc
s the spatial total on [3][4]. After ρojection) [8].
ue, named as PS-The brain of a hspin-echo sequencrameters are 2s Tix size. The data
cceleration factor Fig. 1. The down
) compressed senand (3) the propares the T2 map(1) CS, (2) PS, afrom fully samp
n and SNR than t
ng-based fast paraasic PS model ans with Vt being eining data, which
uction method bar performance to e
BI 2007. pp. 988-9eva M, MRM 201.
ed Parameter
Engineering, Univnd Technology, Un
titative informatitimes. Recently, l for acceleratingo accelerate para
wn to achieve muc
and let ( , )ntρ r With sparse samp
onstructed imageng joint PS and seasily adapted fo
where L is the m
that , ( ,m n mρ=P r
emporal subspac
measured data cathe measurement
e illustrate one sang data. With Vt rectly determininconditioned inverction problem can
variation regular( , )ntρ r is obtain
-Sparse, was valihealthy volunteece at 32 different
TR, 17.2 ms echo set was retrospecof 10 using the
n-sampled data sensing [5]; (2) paposed PS-Sparse ps estimated fromand (3) the PS-Spled 32 echoes is the CS and PS co
ametric mappingnd suffers from thestimated from a h can more faithfu
sed on joint partiexisting methods
991. [2] Lustig M,0. pp. 1114-1120.
sU
r Mapping witBo Zhao1,2, Wenm
versity of Illinois aniversity of Illinois
on of intrinsic tissparse sampling
g MR data acquisametric mappingch more accurate
for 1, 2, , n N= L
pling, direct inveres and subsequentsparsity constrainor parametric ma
model order. Co
, )nt . The PS mo
e, and sU repres
an be written as t noise. Multipleampling pattern,
being determineg sU from measu
rse problem. Specn be formulated a
rization. The aboed, the desired p
idated r was t echo spac-
ctively sam-
et was artial-tech-
m the parse. used as the referunterparts with th
g methods have bhe ill-conditioninpre-fixed parame
ully capture under
ial separability anusing a single PS
, MRM 2007. pp. [6] Petzschner F,
Fig. 2. of 10, r
s
arg min ||= − ΩU
d
th Joint Partiamiao Lu2, and Zhi-t Urbana-Champaat Urbana-Champ
ssue properties, wmethods based o
sition. Some of thusing the PS and
e T2 maps than ex
,N represents a sersion of the (k, ttly the T2 map. Wnts for acceleratinapping. Specifica
onsidering the C
del can be equiv
ents the correspo
s s t( ) ξ= Ω +d F U V
e data acquisitionin which fully s
ed, the reconstruured data can suffcifically, we use as,
ve convex optimparameter map is
rence. As can be he CS showing si
been proposed recng issue illustrateetric signal moderlying temporal e
nd sparsity constrS or sparsity cons
1182-1195. [3] Zh, MRM 2011. pp.
a) Estimated T2 mrespectively; b) Er
2s s t 2( ) || + λΩ F U V
al Separability-Pei Liang1,2 aign, Urbana, IL, Upaign, Urbana, IL
which are useful bon the theory of phe methods have d sparsity constrxisting methods u
et of T2-weightet)-space measureWe have previousng dynamic imag
ally, the PS mode
Casorati matrix P
valently expressed
onding the spatia
ξ , where ( )Ω ⋅ isn schemes can besampled central kuction problem isfer from severe ila spatial finite di
mization problem s obtained using
seen in Fig. 2 (aignificant spatial
cently, such as thed in Fig. 2. REPel. The proposed evolution of relax
raints has been prstraint for reconst
hao B, IEEE-EMB706-716. [7] Huan
maps from CS, PS,rror maps of estima
sTV( ) λ U
y and Sparsity
United States, 2BeL, United States
biomarkers. Howpartial separable f
been extended toraints jointly andusing a single PS
ed ed s-g-el
P
d
al
s e k s ll-conditioning isfference to regula
can be efficientlya nonlinear least
a), the proposed Pblurring and the
he k-t PCA methPCOM uses a sim
method uses the xation process.
roposed to acceletructing highly un
BS 2010. pp. 3390ng C, MRM 2011
, and PS-Sparse reated T2 maps usin
Fig. 1. A samSparse, whictemporal trapled imagin
y Constraints
ckman Institute of
wever, the utility functions [1] and o parametric map
d illustrates its pe or sparsity const
sue. Imposing sparize the inverse
y solved by half-t squares fitting p
PS-Sparse producPS low SNR.
hod [6] and REPCmilar formulation
temporal subspa
erate MR paramendersampled data
0-3393. [4] Zhao B. [8] Golub G, SIA
constructions at anng Sparse, PS, and
mpling pattern exach consists of densaining data, and spg data.
(1)
f Advanced Science
of MR parametricompressed sens
pping [5][6][7]. Ierformance in T2traint with signifi
patial sparsity conproblem of fittin
-quadratic minimiprocedure such a
ces a T2 map wit
COM [7]. The k-based on joint P
ace estimated from
etric mapping. Tha.
B, IEEE-ISBI 2011AM J. Numer Ana
n acceleration PS-Sparse
ample for PS-sely sampled arsely sam-
e
ic s-n
2-i-
n-g
i-as
h
-t S m
he
1. l.
2233Proc. Intl. Soc. Mag. Reson. Med. 20 (2012)