Signaal Leakage(L--factor) as a measure
m
for pa
arallel imaging
g performancee among simu
ultaneously mu
ulti-Slice (SM
MS) excited and
d
acq
quired signals
1
Steen
S
Moeller1, Jun
nqian Xu1, Edward
d J Auerbach1, Esssa Yacoub1, and K
Kamil Ugurbil1
Center for Magnetic
M
Resonan
nce Research, Univversity of Minnesoota, Minneapolis, M
Minnesota, Unitedd States
Introd
duction Parallell imaging metho
ods have significcantly improved
d the temporal effficiency of manny applications iin MRI. The geoometry (g) facto
or
has beeen widely utilizzed to quantify noise
n
amplificatiion and subsequ
uent SNR losses in PI methods aarising from the limited ability oof coil sensitivity
profilees to un-alias sig
gnals. Howeverr, the g-factor do
oes not account for
f potential artiifacts due to resiidual aliasing inn un-aliased imagges. At 3T, using
typicaal resolutions, a limited amount of in-plane undeer-sampling, if any,
a
is used in E
EPI-based acquissitions since traddeoffs in reducinng readout timess,
TR, eetc., do not warrrant much accelleration, especiaally given modeern high perform
ming gradients. This, coupled w
with high coil ccounts, has mad
de
residuual aliasing less concerning.
c
How
wever, recently we
w demonstrated
d the possibilityy and utility of eextremely high aaccelerations for EPI-based fMR
RI
and diiffusion imaging
g based on simu
ultaneous slice ex
xcitations [1-2]. By exciting muultiple slices sim
multaneously, wee are able to conncurrently reduce
the voolume TR in a 2D
2 acquisition, and as many feewer excitations and encodings are needed. Thhis TR reductionn increased fMR
RI sensitivity and
permittted significant reductions in diiffusion scan tim
mes [2] as multtiple slices sharee diffusion encooding gradients. The benefits ffrom volume TR
R
reducttions become ev
ven more significcant when higherr spatial resolutiions (i.e. more sllices) are used. A
As such, even hiigher slice accelerations could be
benefiicial, making eff
ffects due to resiidual aliasing a concern. These effects could reesult in spatiallyy mis-mapped acctivations (or i.ee. spurious fiberrs
througghout the brain) due to signal leeakage in simultaaneously acquireed slices. To adddress this conceern, we describe here a new mettric, the (leakage
e)
L-facttor, to quantify th
hese effects and propose that both l-factors and conventional g-ffactors should bee considered in hhighly accelerateed acquisitions.
Methoods Multi-slicce GRE-EPI datta (FOV: 192 x192 mm2, 2mm
m isotropic) weere acquired on a 3 Tesla Siem
mens Skyra (Sieemens, Erlangen
n,
Germaany) scanner equ
uipped with SC7
72 gradients, a body
b
transmitter and a 32-channeel head receiverr coil. Multibandd separation was performed using
a coil--by-coil data driiven (CCDD) allgorithm [5]. It consists of a series of linear m
matrix multiplicattions projecting the aliased signnal (MB(k)) onto
distincct signals related
d to separate slicces (Sn) in a coiil-by-coil processs. G-factor noisse amplification is calculated usiing the Monte-C
Carlo [6] method
d,
wheree the change in noise-distribution
n
n is used as a meeasure of noise-d
degeneration duee to slice-separaation. L-factor quuantification (described below) is
i
calcullated using un-acccelerated (singlle band) EPI dataa. Subsequently,, simulated multtiband data are ggenerated offlinee by adding signaals from multiple
slices using a variety of different pseeudo (simultaneo
ous slice) acceleerations and conntrolled aliasingg schemes [3-4]. Prior to simulaating a multiband
acquissition and subseq
quent slice unaliiasing, a unique slice specific tem
mporal modulattion was globallyy imposed on eaach slice (simulaating i.e. an fMR
RI
signall modulation). Frrom a single ban
nd multi-slice acq
quisition, a multtiband time seriees with M frames can be formedd such that:
∑
∙ ,
0⋯
,
,
where,
,
1 0.1 ∗ cos 4 ∙
1 . D
Designate by S
,
th
he
reconsstruction of
, into distinct slices. Thee frequency response of
,
is uniqueely defined for each slice; deenoted with fslice,
slice=
={1,..,N}. For thee formulated MB
B time series, sig
gnal leaking from
m a given slice too all others is deetermined from S
,
forr
(i.e
e.
residuual energy at im
mposed frequencies of other sllices). To obtaiin comparable ccoefficients, S
,
is noormalized for eeach f subject to
∑ |S
| ≡ 1. Signal leak
|,
,
kage is calculated as the L-facto
or, for each slicee, as
,
′ ∑ |SS
,
∑
but m
masked to only include
i
pixels in
n the brain (deffined with a thrreshold ε) and
,0
. The aveerage L-factor is
i
defineed as
,
, i.e. as the aveerage of all combbinations.
Resullts. Histograms of g-factors for
f a whole brrain acquisition
n for 4 differeent simulated ssliceacceleraations are shown
n in figure 1, whhere a ¼ shift of the
FOV between adjacen
nt slices is impoosed [3-4]. At high
acceleraations, the meean g-factor pplateaus, suggessting
limited information about
a
image ffidelity using this
h
normalizeed L-factor (leakkage)
measuree. On the other hand,
images,, shown in figurre 2, depict signnal leakage in ccases
where g-factor
g
noise iss modest. The M
MB simulation inn the
Fig.1 G-factors for diffferent MB factorss
top row
w, as in [1-2], does
d
not use conntrolled aliasingg [7],
Fig.2 Siignal leakage imagges from 4
usinng controlled aliasiing.
while th
he bottom row does
d
[3-4]. The aaverage L-factorrs for
simultanneous slices with ((bottom) and
differen
nt MB factors (with (PE4) an
and without (PE
E=0)
withoutt (top) controlled aaliasing.
controolled aliasing) are
a listed in tablle 1. In our receent work [2], we
w demonstratedd robust and reliiable
detecttion of resting state
s
networks (i.e. without sig
gnificant detectaable artifacts) uusing
similaar methods. Giveen those results, as an approxim
mate guide, one can
c consider meean L
factorrs ~0.05 or less as
a acceptable. Ap
pparent in both the
t L-factor imaage and mean vaalues,
b
controlled aliasing schemee for EPI [3-4] significantly
s
redduces
the usse of a blipped based
signall leakage permittting much higherr slice acceleratiions.
Concllusions. The L--factor, proposed
d as a new mettric to quantify residual aliasinng in
Table 1. Mean L-factoors for different M
MB factors
acceleeration imaging,, should be con
nsidered in addittion to g-factorss when planninng an
PE=4) and withoout (PE=0) contrrolled aliasing.
with (P
experiiment. For high
h accelerations, recently made possible
p
by imp
proved coil desiigns,
pulse sequence developments (contro
olled aliasing) [3
3-4], and optimiized image recoonstructions, knoowledge of g-factor noise amplification alone is
i
insuffficient. Signal arttifacts due to ressidual aliasing, as
a described by the
t L-factor, cann be quantified foor any arbitrary multi-channel cooil configuration
n,
slice aacceleration, oriientation, and seeparation and caan give critical information aboout image integrrity. Future worrk will aim to ccorrelate L-facto
or
levels with a defined amount
a
of artifacct level (i.e. in a real fMRI experiment) using sim
milar methods.
Refereence [1] Moeller,, et al. Magn. Reso
on. Med. 63:1144–
–53, 2010 [2] Feinb
berg, et al. PLoS O
ONE, 5(12):e1571 0, 2010 [3] Setsom
mpop, et al. Magn. Reson. Med. 2011
[4] Xuu et al ISMRM 20
012 [5] Brau et al,, Magn. Reson. Med.
M
59:382-395, 2008
2
[6] Robson eet al, Magn. Reso n. Med. 60(4):8955-907, 2008 [7] B
Breuer, et al. Magn
n.
Reson. Med. 53:684–91, 2005
unded in part by th
he NIH Human Co
onnectome Project (U54MH091657)), as well as NIH ggrants P41 RR0080079 and P40 NS0557091.
Acknoowledgements Fu
Proc. Intl. Soc. Mag. Reson. Med. 20 (2012)
519