On he impac o a i ac s induced by misma ches be ween au o calib a ion
signal and accele a ed 3D GRE da a a 11.7T
Joseph Ob io 1, F anck Maucondui 1, Vincen G as1, Chai hya Giliya Radhak ishna1,2, Maxime
Be ai 1,2, Philipp Ehses4, Rüdige S i nbe g4, Ca oline Le S e 1, and Nicolas Boulan 1
1Uni e si é Pa is-Saclay, CEA, CNRS, BAOBAB, Neu ospin, Gi -su -Y e e, F ance
2MIND, In ia, Palaiseau, F ance
4Ge man Cen e o Neu odegene a i e Diseases (DZNE), Bonn, Ge many
Oc obe 6, 2025
Abs ac
1 Pu pose
The s udy aims a in es iga ing B0 ield inhomogenei y a -
i ac s a ising om emo e loca ions in he FOV and en-
coun e ed in accele a ed 3D g adien - ecalled echo (GRE)
sequences a ul a-high ield and a p o iding mi iga ion
s a egies.
2 Me hods
Measu emen s we e conduc ed a 11.7T using a head-
shaped phan om and an accele a ed 3D GRE sequence wi h
ei he in eg a ed o ex e nal au o-calib a ion signal (ACS)
lines. Simula ions we e pe o med o ep oduce he a i-
ac s. The e ec s o a ying GRAPPA econs uc ion pa-
ame e s (ke nel size and egula iza ion) we e also exam-
ined.
3 Resul s
B0 ield inhomogenei ies loca ed ou side he B0shimmed
egion o in e es (i.e. he b ain) we e obse ed o e u n
ipple-like a i ac s wi hin his egion, pa icula ly a long
echo imes. The simula ion esul s suppo hese indings,
and he idea ha he obse ed a i ac o igina es om mis-
ma ch be ween ACS and accele a ed da a due o in a- oxel
dephasing a di e en esolu ions (ACS lines ha ing an in-
insically lowe esolu ion). The sho echo ime enabled
by ex e nal (i.e., p eacqui ed) ACS lines educed a i ac s
compa ed o in eg a ed ones. Va ying GRAPPA ke nel sizes
and inc easing he numbe o ACS lines can imp o e image
quali y ye wi hou ull compensa ion.
4 Conclusion
This s udy highligh s ipple-like a i ac s ampli ied wi h
ield s eng h and a ising om lack o cohe ence be ween
he ACS and imaging (3D GRE) signal caused by B0in a-
oxel dephasing. To minimize hese a i ac s, ca e should
be aken in o de o p ese e he ele an in o ma ion in he
ACS da a o p ope ly compu e he GRAPPA ke nels.
Keywo ds
a i ac s, au o-calib a ion signal, ul a-high ield
1
1 In oduc ion
The ubiqui ous use o MRI in neu oscience has d i en he
pu sui o inc easingly highe main magne ic ields (B0) o
imp o e he image spa io empo al esolu ion by ading o
signal- o-noise a io (SNR) gain. Ul a-high ields o 7T a e
now common in esea ch, while unique MRI scanne s going
as high as 10.5T and 11.7T a e ope a ional and ha e al eady
shown success in p oducing images in i o1,2 and in i o3
.
As he main ield inc eases, howe e , challenges eme ge,
in pa icula he p onounced inhomogenei y o he B+
1and
B0 ields. Speci ic abso p ion a e (SAR) is also a majo
conce n o mos sequences.
Much p og ess has been made o mi iga e a i ac s in-
duced by B+
1 ield inhomogenei y, such as wi h pa -
allel ansmission me hods —calib a ion- ee o subjec -
ailo ed— yielding homogeneous exci a ion p o iles4
.B0
ield o se s (∆B0, mos ly o igina ing om impe ec ions
in he main magne ic ield and suscep ibili y di e ences be-
ween ai and issue) a e mi iga ed by shimming wi hin a
olume o in e es (usually se o he b ain in neu oimag-
ing). Inside he b ain he s anda d de ia ion o ∆B0is
a ound 80 Hz a 11.7T1bu no able la ge excu sions oc-
cu close o ai ca i ies, eaching se e al hund eds o Hz.
Func ional MR images, ha ypically equi e T∗
2-weigh ed
g adien - ecalled echo plana imaging (EPI), a e hus no o-
iously difficul o acqui e in b ain egions nea by sinuses
and ea canals due o he loss o signal induced by B0g a-
dien s (∇B0), i.e. in a- oxel dephasing. Bu also, s uc-
u al T∗
2-weigh ed imaging o T∗
2mapping, suscep ibili y-
weigh ed imaging o quan i a i e suscep ibili y mapping us-
ing g adien - ecalled echo (GRE) acquisi ions, can su e
om such signal loss.
In pa icula high- esolu ion s uc u al and quan i a i e
GRE imaging is p e e ably pe o med using 3D sequences,
whe e phase encoding is pe o med along wo di ec ions.
This is p ima ily o maximize SNR5–8
, bu u he easons
include acili a ing e y high slice esolu ion and efficien
use o pa allel imaging9along wo phase encode (PE) di ec-
ions. While educing oxel size is a e y e ec i e means o
coun e ac in a- oxel dephasing, e en high- esolu ion GRE
imaging can su e om ∇B0-induced a i ac s1
.
The e is simila ly a desi e o educe he acquisi ion ime
(TA) o mi iga e a i ac s induced by a ia ions du ing he
acquisi ion, such as subjec ’s mo ion. A common way o
achie e his is o employ pa allel imaging wi h an accele -
a ion echnique such as GRAPPA10
. The la e consis s in
skipping k-space lines along he phase encode di ec ion(s),
he eby leading o a iola ion o Nyquis c i e ion and yield-
ing an aliased image. In addi ion, a Nyquis -sampled k-
space egion, known as he au o calib a ion signal (ACS),
is equi ed. To his end, addi ional lines in he cen e o k-
space a e o en acqui ed. ACS da a can be in e nal—coming
om he same acquisi ion as he aliased image da a—o ex-
e nal, i.e. acqui ed in a p e-scan wi h minimized echo ime
(TE) and epe i ion ime (TR) and he e o e usually in a
sho amoun o ime o minimize scan ime and mo ion
a i ac s. Since he ACS ypically co e s only a small po -
ion o k-space compa ed o he ull esolu ion acquisi ion,
i inhe en ly has a much la ge oxel size, making i mo e
suscep ible o signal loss om in a- oxel dephasing, which
can p opaga e ac oss he image due o accele a ion.
On he pa h o unc ional MRI a 11.7T, es s we e ini-
ially pe o med o quali y con ol a he a ge esolu ion
in e nal ACS
ex e nal ACS
Figu e 1: Illus a ion o he ipple a i ac encoun e ed du -
ing ini ial es s in 3D GRE scans acqui ed wi h iPAT =
2×2.
o 1.2 mm. These es s showed non- i ial a i ac s which
led us o simpli y he acquisi ion wi h a 3D GRE sequence in
o de o oubleshoo he p oblem. Unde sampled 3D GRE
images he eby e ealed likewise ipple a e ac s (Figu e 1)
which in u n led o his in es iga ion and s udy.
The aim o his s udy is hus o in es iga e a i ac s in o-
duced by in a- oxel dephasing, which causes a misma ch
be ween low- esolu ion ACS and high- esolu ion unde sam-
pled da a in accele a ed 3D GRE. Simula ions we e pe -
o med o suppo he heo y. We inally p o ide ano he
scena io in which misma ches be ween ACS and accele a ed
da a a ise om he use o di e en RF pulses in he wo ac-
quisi ions, simila ly esul ing in a i ac s.
2 Me hods
2.1 Obse a ions
Measu emen s we e pe o med on he Iseul 11.7T scanne
wi h a 8Tx-31Rx RF coil11,12 d i en in a ci cula ly pola ized
(CP), slab-selec i e mode and on a ealis ic head-shaped
phan om13
. An axial 3D g adien - ecalled echo (GRE) se-
quence was used wi h TR = 40 ms, TE = 4 o 20 ms, eso-
lu ion = 1.2×1.2×1.2mm3, ma ix size = 144×120×160,
lip angle = 10°, le - igh phase encoding. This sequence
was acqui ed 3 imes: wi h and wi hou accele a ion (2×2
GRAPPA, espec i e TA = 3 min 07 s and 11 min 31 s),
in eg a ed o ex e nal ACS (24 ×24 lines).
The phan om is based on [14] wi h a wo-compa men
design made o a emp ep oducing he human head and
neck geome y.
2.2 Simula ions
Simula ions we e conduc ed o a i icially ep oduce he ip-
ple a i ac s by applying bo h in e and in a- oxel dephas-
ing o a ully-sampled, a i ac - ee acquisi ion pe o med
wi h a TE o 4 ms. A syn he ic ∆B0 ield was applied wi h
a peak se o 500 Hz abo e he sinuses, dec easing spa ially
in a Gaussian manne as shown in Figu e 2 (A), o ans o m
he TE = 4 ms da a in o de o mimic dephasing occu ing
a TE = 20 ms.
The in e - and in a- oxel dephasing along he eadou
di ec ion a e pa i ion o phase encoding a a oxel le el
was calcula ed using a i s o de decomposi ion o he B0
2
A
C
B
gGRAPPA
unde sampled
k-space
ACS
X
oxel∈ o
DFT( oxel)
×dephasing(B0, )
ully sampled
kspace
0
100
200
300
400
500
Hz
Figu e 2: A i icial induc ion o ipple a i ac s due o B0 ield inhomogenei y in accele a ed 3D GRE. Base image is
acqui ed wi h TE = 4 ms, Equa ion 3 is applied wi h a simula ed TE o 20 ms using a syn he ic B0map o e layed on
he magni ude, ully-sampled image, shown in A. The econs uc ion p ocess is shown in B and he esul ing image in C,
accele a ion di ec ions a e shown wi h do ed ed lines (2×2), and ead di ec ion wi h a solid ed line. Ripple a i ac s
esul ing om he emo e B0g adien a ea a e isible a he h ee aliased co ne s, away om he s ong B0o se s.
ield:
S( , ky, kz) =
0+∆ /2
y
0−∆ /2
ρ( )ei(k· )ei(∆B0+∇B0·( − 0)) d ,
(1)
whe e is he ime a e he exci a ion pulse, is he po-
si ion ec o (x, y, z)cen e ed inside a oxel a posi ion 0
and size ∆ , wi h i s associa ed ∆B0and ∇B0. The en-
coding s eps ka e de ined as (kx( ), ky, kz)wi h kyand kz
co esponding o he wo phase-encoded spa ial equencies,
and he eadou along xgi en by:
kx( ) = −γ∫
0
Gx(u)du. (2)
A e in eg a ion15
, he signal can be u he exp essed in
e ms o he ideal (wi hou ∆B0o se ) signal S0 o each
oxel as
S( ) = S0( )
×ei∆B0
×∏
d∈[x,y,z]
sinc ((kd+ [∇B0]d )∆ d/2)
sinc(kd∆ d/2) .(3)
The a io o sincs abo e a ises om he ac ha we
isola e he ∇B0impac as a mul iplica i e ac o o he
undis u bed signal (when ∇B0= ∆B0= 0,S( ) = S0( )).
A consequence o ha exp ession is ha he ully-sampled
magni ude image expe iences no only a signal a enua ion
a kx= 0, bu also ce ain high- equency componen s along
he eadou di ec ion can be dis o ed due o he shape o
he sinc unc ion, hus po en ially leading o ipples in he
same di ec ion.
In he simula ion, Equa ion 3 was applied o he TE =
4 ms k-space da a, 24 ×24 cen al lines we e ex ac ed
o se e as he in eg a ed ACS, while a 2×2accele a ed
da ase was simula ed by unde sampling he ull k-space
2×2- old. O line image econs uc ion was pe o med us-
ing gGRAPPA16 and images o he idi idual channels we e
combined using oo -sum-o -squa es.
2.3 Manipula ion o ACS lines
To con i m he e ec s o in a- oxel dephasing causing mis-
ma ches be ween low esolu ion ACS da a and he highe
esolu ion accele a ed acquisi ion, we econs uc ed images
using unde sampled da a acqui ed a TE = 20 ms, and ACS
lines ex ac ed om ully-sampled TE = 4 ms image (24×24
lines). In addi ion, in he ACS image, a po ion o he FOV
(phan om neck) was a i icially educed in magni ude a
a ying le els.
The associa ed econs uc ions we e compa ed o he ully
sampled image acqui ed a TE = 20 ms using he s uc u al
simila i y index measu e17 (SSIM).
2.4 GRAPPA econs uc ion pa ame e s
The accele a ed image acqui ed a TE = 20 ms wi h in e nal
ACS was econs uc ed o line wi h gGRAPPA using di e -
en GRAPPA ke nel sizes and egula iza ion pa ame e (λ)
o assess he possibili y wi hin his econs uc ion ame-
wo k o compensa e o he a i ac s. Recons uc ion was
also pe o med using a ying numbe s o in eg a ed ACS
lines. De aul pa ame e s we e se o: ke nel size = (4, 4,
5) (pa i ion, phase encode, eadou ), λ= 10−4, numbe
o ACS lines = 24 ×24. The magni ude images we e again
compa ed o he ully sampled image using he SSIM.
3
2.5 Misma ch caused by di e ence in RF
pulse exci a ion
RF ield maps we e measu ed using a magne iza ion-
p epa ed u bo lash sequence a 11.7T wi h 5 mm iso opic
esolu ion and TR = 15 s18
. A non-selec i e pa allel ans-
mission (pTx) pulse ( a ge lip angle = 10°) was calcula ed
om he masked phan om b ain using he as g adien
ascen pulse enginee ing ( as GRAPE) me hod19,20
, which
yielded a much imp o ed lip angle homogenei y e sus CP
(no malized oo mean squa e e o , NRMSE = 4.5% e sus
40.7%). The RF pulse was hen inse ed in he 3D GRE se-
quence o acqui e ully sampled images a TE = 3 and 20 ms
(TA = 15 min 22 s). The acquisi ion a TE = 3 ms was
epea ed in he CP mode o exci a ion. In simula ion, e-
cons uc ion o he unde sampled (iPAT = 2×2), TE =
20 ms pTx da a was pe o med sepa a ely wi h pTx ACS
and CP ACS da a ex ac ed om he TE = 3 ms acqui-
si ions, he eby explo ing he impac o using dis inc RF
pulses in he wo sub-acquisi ions.
3 Resul s
Non-accele a ed and accele a ed (iPAT 2×2,24 ×24 ACS
lines) magni ude and phase images acqui ed a TE = 4 and
20 ms a e shown in Figu e 3, along wi h aliased images. As
expec ed, he non-accele a ed images show ew a i ac s due
o ∇B0beside signal loss (wo se a TE = 20 ms) a he in e -
ace be ween ai and he op o he phan om “b ain”, as well
as in he “neck”. Inc eased dephasing a longe echo ime
can also be isualized on he phase images ( he exp(i∆B0 )
e m in Equa ion 3), and he epe cussion on he aliased
image is isible. The complex alues wi h apidly changing
phases old on o a shimmed a ea ( he “b ain”) in oducing
in e e ence pa e ns. The GRAPPA un olding p ocess is
able o co ec mos bu no all o he ipples isible on
he aliased image, depending on he accu acy o he ke nels
e alua ed wi h lowe esolu ion ACS da a.
The in eg a ed ACS econs uc ions a TE = 20 ms show
an inc ease in he equency o he ipples compa ed o TE
= 4 ms, as sugges ed by he in e - oxel dephasing e m,
and an inc ease in he esidual a i ac s a e GRAPPA un-
olding. Meanwhile, he ex e nal ACS econs uc ions show
no such a i ac s. The echo ime o ex e nal ACS lines is
signi ican ly sho e han he TE o he imaging scan, as
hey a e designed no o p ese e con as bu o be ac-
qui ed as quickly as possible wi h he sho es TE, he eby
minimizing in a- oxel dephasing.
Figu e 4 shows he econs uc ion om unde sampled
da a a TE = 20 ms, wi h ACS lines bo owed om he
24 ×24 cen e k-space om a ully-sampled TE = 4 ms
acquisi ion, simila o an ex e nal, sho -TE, ACS p e-
scan. The esul ing econs uc ion in he le -mos column
p esen s minimal a i ac s. A his s age, wi h a TE o
4 ms, he ACS lines s ill con ain signal in he neck. As we
p og essi ely and a i icially a enua e he neck signal om
he ACS images o mimic in a- oxel dephasing, we obse e
he g adual eme gence o ipple a i ac s in he shimmed
b ain egion o igina ing om a emo e a ea wi h a high B0
g adien .
As seen in Figu e 2 (C), adding B0 ield inhomogenei ies
o an ini ially a i ac - ee TE = 4 ms image, hen sim-
ula ing om Equa ion 3 a 20 ms echo ime and inally
econs uc ing he a i icially unde sampled image using
GRAPPA, yields he same kind o a i ac s as in Figu e 3
and Figu e 4, appea ing a he aliased loca ions o he added
B0o se s.
The se e i y o he a i ac s being econs uc ion-
dependen , we show in Figu e 5 how a ying GRAPPA e-
cons uc ion pa ame e s in a TE = 20 ms acquisi ion wi h
in e nal ACS a ec s image quali y by compu ing he SSIM
agains he ully-sampled image. The ke nel size used o
he econs uc ion was a ied in all h ee dimensions. How-
e e , since he esul s exhibi ed nea diagonal symme y in
he wo accele a ion di ec ions, o cla i y we p esen he e-
sul ing SSIM along he diagonal, whe e he maximum alues
occu . The egula iza ion pa ame e λwas a ied loga i h-
mically om 10−7 o 10. The image quali y change caused
by λwas ma ginal below he h eshold λ= 0.01, abo e
which he econs uc ion quickly ails a un olding he im-
age, leading o a sha p d op in SSIM. In con as , inc easing
he numbe o in e nal ACS lines shows a mono onic in-
c ease in image quali y and educ ion in ipples. Howe e ,
he SSIM alues emain lowe han hose achie able wi h
ex e nal ACS (Figu e 4) unless he numbe o in e nal ACS
lines is inc eased o he poin whe e he bene i o accele a-
ion is e ec i ely los , he eby compensa ing o in a- oxel
dephasing wi h highe ACS esolu ion.
As he esolu ion o he ACS ends o each he esolu ion
o he unde sampled da a, he SSIM con e ges o a alue
sligh ly lowe han 1, explained by he ac ha ACS ke nels
a e e alua ed in a leas squa es sense by exploi ing he com-
plemen a i y be ween p o iles ac oss ecei e channel, which
a e hus unable o e ie e he exac o iginal image.
Figu e 6 shows ha his a i ac can also be caused by
misma ches be ween ACS and accele a ed da a acqui ed
wi h di e en RF pulses. While B+
1 ield inhomogenei y
becomes much mo e se e e a ul a high ield, especially a
11.7T, inding an RF shim mode ha comple ely a oids e-
gions wi h ex emely weak B1 alues is ex emely challeng-
ing i no impossible. I hese weak alues a e p esen in he
ACS da a acquisi ion, his can cause e o s in he GRAPPA
ke nel es ima ions when he accele a ed da a acquisi ion is
pe o med wi h pa allel ansmi pulses ha mi iga e B1
ield inhomogenei y.
4 Discussion
In a ypical accele a ed acquisi ion, ACS da a ha e an in-
insically lowe esolu ion han he unde sampled da a in
he PE di ec ions. This makes ACS da a mo e suscep i-
ble o in a- oxel dephasing, inducing a much la ge loss
o signal in he high B0g adien a eas, unless hey ha e a
sho TE like wi h ex e nal ACS. Such di e ences be ween
he ACS lines and he unde sampled da a lead o e o s
in he GRAPPA ke nels (o ecei e sensi i i y p o iles in
SENSE21), p e en ing p ope un olding o he aliased ox-
els due o apid signal a ia ions. This issue is pa icula ly
p onounced a ul a-high ields.
Since o he signal in Equa ion 3 is dependen on echo
ime and B0 ield inhomogenei ies, we expec he same is-
sue o occu a lowe ields when he TE is p opo ionally
inc eased (as shown in Figu e S1).
The same igu e also demons a es ha inc easing he
esolu ion does no mi iga e he p oblem since i widens
he “in a- oxel dephasing” gap be ween he ACS and he
inal unde sampled da a. Impo an ly, in his case, he e -
4
TE = 20 ms
-
π
0
π
ad
ully sampled
magni ude image
ully sampled
phase image
aliased
magni ude image
GRAPPA
econs uc ion
in e nal ACS
GRAPPA
econs uc ion
ex e nal ACS
TE = 4 ms
ead
ead
Figu e 3: Aliasing e ec s o B0 ield inhomogenei ies on GRAPPA econs uc ions: examples o apid dephasing a eas
and hei aliased phan om in accele a ed da a a e delinea ed wi h solid and dashed ed lines espec i ely. Ripple a i ac s
o igina ing om a emo e B0 ield g adien a e s ill isible a e GRAPPA un olding a TE = 20 ms and o a lesse ex en
a TE = 4 ms, wi h a dec eased spa ial equency. Swi ching o ex e nal ACS o GRAPPA calib a ion signi ican ly
educes a i ac s in he long TE acquisi ion.
0.967
0.964
0.959
0.950
0.920
0.850
SSIM
econs uc ion
ACS
image
100%
80%
60%
40%
20%
0%
signal emaining in he neck o ACS images:
Figu e 4: E ec s o manually a enua ing he neck signal in ACS lines o mimick in a- oxel dephasing. The i s ow
shows he ACS images used o he econs uc ion below a highe esolu ion. The ACS a e ex ac ed om a TE =
4 ms ully sampled acquisi ion, while he unde sampled da a is ex ac ed om a TE = 20 ms acquisi ion ( he same as in
Figu e 3). The dashed ed line indica es he slice below which he signal is a i icially a enua ed.
5
2
3
4
5
6
7
8
9
2
3
4
5
6
7
8
9
0.9042
0.5578
0.9
0.85
0.8
0.89
0.88
0.87
# ACS lines
0.2
0.4
0.6
0.8
0.96
1
SSIM
10−7
10−6
10−5
10−4
10−3
10−2
0.1
1
10
λ
ke nel size (accele a ion di ec ions, y, z)
ke nel size ( eadou di ec ion, x)
SSIM
24 ×24
32 ×32
40 ×40
48 ×48
56 ×56
64 ×64
A
B
Figu e 5: In luence o GRAPPA pa ame e s. The e ec s o
ke nel size [A], numbe o ACS lines and λ[B] o accele -
a ion ac o 4 (2×2) a e shown wi h he SSIM and wi h
e e ence he ully-sampled image (TE = 20 ms). The ed
con ou s indica e he maximum SSIM a ke nel size = (2,
2, 5). Ke nel sizes o which econs uc ion ailed a e shown
in black.
GRAPPA using
pTx
accele a ed da a
GRAPPA using
pTx
accele a ed da a
ACS
CP
pTx
econs uc ion
Figu e 6: E ec o a misma ch be ween he ACS and accel-
e a ed da a caused by di e en RF exci a ion pulses. ACS
da a a e composed o 36 ×36 lines o a TE = 3 ms acqui-
si ion, while accele a ed pTx da a (iPAT 2×2) comes om
a TE = 20 ms acquisi ion.
ec can also o igina e om la ge B0o se a eas loca ed
ou side o he shimmed egion o in e es . This can in-
clude he neck (wi h usually a poo B0shim due o he
p esence o he e eb a), he subjec ’s ace (pa icula ly
he mou h o subjec s holding b aces). A misma ch be-
ween he low- esolu ion ACS images and he unde sampled
images impai s he abili y o GRAPPA o p ope ly un old
GRE olumes. As a consequence, e en shimmed a eas ha
a e aliased wi h high B0o se egions a e no comple ely
immune o he ipples in oduced by in e - oxel dephasing.
The as acquisi ion and sho -TE na u e o ex e nal ACS
lines on he o he hand makes hem ela i ely immune o
he loss o signal induced by in a- oxel dephasing. This sig-
nal p ese a ion inc eases he cohe ence be ween he high
esolu ion image and he ACS, allowing GRAPPA o be -
e un old he o me compa ed o using in eg a ed ACS.
As demons a ed in Figu e 4, signal losses in he ACS a e
accep able as long as one is abo e a ce ain h eshold, jus-
i ying he po en ial con as di e ences be ween ACS and
accele a ed da a. Howe e , as shown in Figu e 6, a mis-
ma ch be ween ACS and accele a ed da a can ha e o he
o igins such as he use o di e en RF pulses leading again
o a i ac s. I is cu en ly he case in he de aul pTx-
enabled 3D GRE sequence on ou MRI sys em: ACS wi h
CP, unde sampled da a acquisi ion wi h pTx.
ACS lines acqui ed ex e nally usually ha e a much sho e
TR han he subsequen imaging scans, which can be a se i-
ous limi a ion due o SAR es ic ions. The pTx pulses a e
ypically designed o wo k op imally and wi hin SAR limi s
du ing he longe imaging TRs. To s ay wi hin SAR limi s
du ing sho ACS TRs, using he same pTx pulses would
en o ce e y low ACS lip angles. RF pulses in CP mode
can be mo e bene icial in he ACS in e ms o SAR and
sho e TR bu can lead o his kind o misma ch. While i
is no clea how d ama ic his can be o b ain acquisi ions
a 7T, he se e i y o he B1 ield inhomogenei y p oblem a
11.7T makes i impo an o conside in u u e expe imen s.
I in eg a ed ACS canno be a oided, hese a i ac s can
be mi iga ed by, o example, selec ing he accele a ion di-
ec ions so ha a eas wi h high B0g adien s a e no aliased
in o he egion o in e es . Howe e , his app oach is chal-
lenging in p ac ice due o he ana omy o he human head
and he loca ion o ca i ies, pa icula ly he pa anasal si-
nuses and ea canals. I also becomes inc easingly diffi-
cul as he accele a ion ac o inc eases. The mos e ec i e
s a egy is o a oid exci ing a eas wi h high B0g adien s
by employing slab selec ion o ocus on he egion closes
o he shim olume, e.g. o elimina e neck-induced a i-
ac s. As a esul , o in eg a ed ACS, we canno p opose
a gene al solu ion bu only mi iga ion s a egies ha mus
be ine- uned on a case by case basis.
GRAPPA econs uc ion pa ame e s such as ke nel size
and egula iza ion pa ame e can help mi iga e hese a i-
ac s. In Figu e 5, he op imal ke nel size was ound o be
(2, 2, 5) sugges ing ha a smoo he es ima ion o he e-
cei e sensi i i y p o iles in he unde sampled PE di ec ions
ia smalle GRAPPA ke nel size compa ed o he ead-
ou di ec ion may help il e ing ou as signal a ia ions.
Howe e , he SSIMs ob ained wi h hese a ious pa ame e
combina ions could no ma ch hose achie ed wi h ex e nal
ACS. Inc easing he numbe o ACS lines na u ally helps as
well, as i p o ides highe - esolu ion ACS da a and he eby
educing in a- oxel dephasing and, consequen ly, he mis-
ma ch.
6
The a i ac highligh ed in his s udy has implica ion be-
yond 3D GRE acquisi ions, which was conside ed he e as
a ex book case sequence o simplici y. 2D and 3D echo
plana imaging sequences o MRI also ypically ha e long
TE o op imize BOLD sensi i i y. E en hough EPI ACS
is ypically acqui ed ex e nally by de aul , widely used en-
do implemen a ions use he same EPI ain o ACS as o
subsequen accele a ed imaging, only wi h disabled da a ac-
quisi ion o he ou e k-space lines. As his implies he same
long echo ime (and epe i ion ime) bu wi h la ge oxels,
such endo sequences a e pa icula ly suscep ible o he
discussed kind o a i ac s. Thus, beyond educing sensi i -
i y losses due o espi a ion and mo ion ou p oposed mi i-
ga ion s a egies can also be applied o add ess hese issues
in such sequences, bu wi h inc eased complexi y due o he
igh coupling o echo ime and numbe o k-space lines ac-
qui ed pe sho . Fo ins ance, while sho -TE 3D GRE ACS
can be bene icial o EPI, one may ha e o addi ionally con-
side he geome ic misma ch be ween ACS and accele a ed
EPI, especially since a eas a ec ed by such dis o ions a e
also a ec ed by spa ially as a ying phases. This is ele-
an e en when choosing EPI-based ACS. Fo ins ance, in
low PE accele a ion se ings (e.g., in a wide ange o ypical
MRI applica ions), he ACS may be acqui ed wi h a single
sho and hus misma ched dis o ions. In highe PE accel-
e a ion se ings (e.g., high esolu ion EPI) he ACS and EPI
PE bandwid h a e ypically ma ched by swi ching he ACS
o in e lea ed mul i-sho EPI. In his case, he ACS min-
imum achie able echo ime ye au oma ically inc eases as
he ACS oxel size dec eases and he acquisi ion becomes
mo e sensi i e o physiological noise, in pa icula in 2D-
EPI. This can be mi iga ed o a high deg ee by using he
FLEET22 me hod which p oposes, among o he imp o e-
men s, a minimized TE and TR ha makes mul i-sho EPI
ACS also less suscep ible o in a- oxel dephasing.
This s udy was pe o med on a phan om on pu pose
o a oid con ounds om o he phenomena (e.g. mo ion,
b ea hing) and o mo e clea ly demons a e he a i ac s on
simple objec s unde well-con olled condi ions, o be epli-
ca ed in- i o (Figu e 1).
5 Conclusions
We ha e in es iga ed GRAPPA econs uc ion a i ac s in
B0-shimmed accele a ed 3D GRE acquisi ions o he b ain
a ul a-high ield. Inhomogenei ies o he main magne ic
ield cause dephasing leading o ipples in he aliased egions
o he unde sampled image. These ipple a i ac s should
ideally be elimina ed du ing GRAPPA un olding, bu he
misma ch be ween he lowe esolu ion ACS and he accel-
e a ed da a impai s his p ocess. This misma ch is p ima ily
due o he di e ence in signal loss om in a- oxel dephas-
ing, wi h he ACS losing much o he signal in high B0o se
a eas. Coping s a egies include using an ex e nal –sho -
TE– ACS scan, slab-selec i e exci a ions o a oid acqui ing
he MR signal o igina ing om high-∇B0 egions, uning
o he econs uc ion ke nel size, inc easing he numbe o
ACS lines and op imizing he equency encoding di ec ion.
Acknowledgmen s
This p ojec ecei ed unding om AROMA H2020 FET-
Open (g an no. 885876) and Agence Na ionale de la
Reche che (ANR) (F ance 2030 Fu u e In es men P o-
g am, g an no. ANR-21-ESRE-0006, ESR/EquipEx+,
PRESENCE p ojec ). The gGRAPPA package was de el-
oped h ough unding by CEA BlueSky p ojec .
Re e ences
[1] Boulan Nicolas, Maucondui F anck, G as Vincen ,
e al. In Vi o Imaging o he Human B ain wi h
he Iseul 11.7-T MRI Scanne . Na u e Me hods.
2024;21(11):2013–2016.
[2] He Xiaoxuan, E ü k M. A can, G an And ea, e al.
Fi s In-Vi o Human Imaging a 10.5T: Imaging he
Body a 447 MHz. Magne ic Resonance in Medicine.
2020;84(1):289–303.
[3] Pi aluga S, Do micchi O, Mau o L, e al. Com-
missioning and Ope a ion o 11.74T Human MRI
Magne o Gachon Uni e si y Gil Medical Cen e ,
Sou h Ko ea. Supe conduc o Science and Technology.
2025;38(3):035004.
[4] G as Vincen , Wu Xiaoping, Boulan Nicolas. Chap e
7 - Pa allel T ansmission: Physics Backg ound, Pulse
Design, and Applica ions in Neu o MRI a Ul a-High
Field. In: Ma ken o h Bloch Ka in, Guye Maxime,
Pose Benedik A., eds. Ad ances in Magne ic Reso-
nance Technology and Applica ions, Ul a-High Field
Neu o MRI, ol. 10: Academic P ess 2023 (pp. 97–
123).
[5] Pose B. A., Koopmans P. J., Wi zel T., Wald L. L.,
Ba h M.. Th ee Dimensional Echo-Plana Imaging a
7 Tesla. Neu oImage. 2010;51(1):261–266.
[6] Van De Zwaag W., Ma ques J. P., Kobe T., Glo e
G., G ue e R., K uege G.. Tempo al SNR Cha ac-
e is ics in Segmen ed 3D-EPI a 7T. Magne ic Reso-
nance in Medicine. 2012;67(2):344–352.
[7] S i nbe g Rüdige , Huijbe s Willem, B enne Daniel,
Pose Benedik A., B e ele Monique, S öcke Tony.
Rapid Whole-B ain Res ing-S a e MRI a 3 T:
Efficiency-op imized Th ee-Dimensional EPI e sus
Repe i ion Time-Ma ched Simul aneous-Mul i-Slice
EPI. Neu oImage. 2017;163:81–92.
[8] Hube Lau en ius, I ano Dimo, Handwe ke
Daniel A., e al. Techniques o Blood Volume
MRI wi h VASO: F om Low-Resolu ion Mapping o-
wa ds Sub-Millime e Laye -Dependen Applica ions.
Neu oImage. 2018;164:131–143.
[9] Deshmane Anagha, Gulani Vikas, G iswold Ma k A.,
Seibe lich Nicole. Pa allel MR Imaging. Jou nal o
Magne ic Resonance Imaging. 2012;36(1):55–72.
[10] G iswold Ma k A., Jakob Pe e M., Heidemann
Robin M., e al. Gene alized Au ocalib a ing Pa ially
Pa allel Acquisi ions (GRAPPA). Magne ic Resonance
in Medicine. 2002;47(6):1202–1210.
[11] Chu Son, Baska an Di ya, Maucondui F anck, G as
Vincen , Boulan Nicolas, Gunamony Shajan. An 8-
Channel T ansmi A ay in Combina ion wi h a 31-
Channel Recei e A ay o Human B ain Imaging a
7
11.7T. In: P oceedings o he 32nd Annual Mee ing o
ISMRM, :1415; Singapo e.
[12] Chu Son, G as Vincen , Maucondui F anck, Massi e
Au élien, Boulan Nicolas, Gunamony Shajan. Elec o-
magne ic and RF Pulse Design Simula ion Based Op-
imiza ion o an Eigh -Channel Loop A ay o 11.7T
B ain Imaging. Magne ic Resonance in Medicine.
2023;90(2):770–783.
[13] Deleba e Thaddée, G as Vincen , Maucondui F anck,
Vignaud Alexand e, Boulan Nicolas, Ciobanu Luisa.
Efficien Op imiza ion o Chemical Exchange Sa u a-
ion T ans e MRI a 7 T Using Uni e sal Pulses and
Vi ual Obse a ion Poin s. Magne ic Resonance in
Medicine. 2023;90(1):51–63.
[14] Jona Ghil, Fu man-Ha an Edna, Schmid Ri a. Re-
alis ic Head-Shaped Phan om wi h B ain-Mimicking
Me aboli es o 7 T Spec oscopy and Spec oscopic
Imaging. NMR in Biomedicine. 2021;34(1):e4421.
[15] Sobol Wlad T., Gaun Da id M.. On he S a iona y
S a es in G adien Echo Imaging. Jou nal o Magne ic
Resonance Imaging. 1996;6(2):384–398.
[16] Be ai Maxime, Giliya Radhak ishna Chai hya, Ciu-
ciu Philippe. gGRAPPA: A Flexible, GPU-accele a ed
Py hon Package o Fas and Efficien Gene alized
GRAPPA Recons uc ion. In: P oceedings o he 33 d
Annual Mee ing o ISMRM, ; 2025; Hawaii, USA.
[17] Wang Zhou, Bo ik A.C., Sheikh H.R., Simoncelli E.P..
Image Quali y Assessmen : F om E o Visibili y o
S uc u al Simila i y. IEEE T ansac ions on Image
P ocessing. 2004;13(4):600–612.
[18] Amadon Alexis, Cloos Ma ijn An on, Boulan Nicolas,
Hang Ma ie-F ance, Wiggins Ch is ophe John, Fau z
Hans-Pe e . Valida ion o a Ve y Fas B1-mapping Se-
quence o Pa allel T ansmission on a Human B ain
a 7T. In: P oceedings o he 20 h Annual Mee ing o
ISMRM, :3358; 2012; Melbou ne, Aus alia.
[19] Van Damme L., Maucondui F., Chamb ion T.,
Boulan N., G as V.. Uni e sal Nonselec i e Exci a-
ion and Re ocusing Pulses wi h Imp o ed Robus ness
o O -Resonance o Magne ic Resonance Imaging a 7
Tesla wi h Pa allel T ansmission. Magne ic Resonance
in Medicine. 2021;85(2):678–693.
[20] B égea Joseph, Maucondui F anck, Boulan Nico-
las, G as Vincen . Efficien Simul aneous Op imiza-
ion o K-space T ajec o ies and Pa allel T ansmission
RF Wa e o ms Using he Pa ame ic ’ as GRAPE’ Ap-
p oach. In: P oceedings o he 40 h Annual Mee ing o
ESMRMB, :0420; 2024; Ba celona, Spain.
[21] P uessmann Klaas P., Weige Ma kus, Scheidegge
Ma kus B., Boesige Pe e . SENSE: Sensi i i y Encod-
ing o Fas MRI. Magne ic Resonance in Medicine.
1999;42(5):952–962.
[22] Polimeni Jona han R., Bha Himanshu, Wi zel
Thomas, e al. Reducing Sensi i i y Losses Due o Res-
pi a ion and Mo ion in Accele a ed Echo Plana Imag-
ing by Reo de ing he Au ocalib a ion Da a Acquisi-
ion. Magne ic Resonance in Medicine. 2016;75(2):665–
679.
8
Lis o Figu es
1 Illus a ion o he ipple a i ac encoun e ed du ing ini ial es s in 3D GRE scans acqui ed wi h iPAT =
2×2. ........................................................ 2
2 A i icial induc ion o ipple a i ac s due o B0 ield inhomogenei y in accele a ed 3D GRE. Base image
is acqui ed wi h TE = 4 ms, Equa ion 3 is applied wi h a simula ed TE o 20 ms using a syn he ic B0
map o e layed on he magni ude, ully-sampled image, shown in A. The econs uc ion p ocess is shown
in B and he esul ing image in C, accele a ion di ec ions a e shown wi h do ed ed lines (2×2), and ead
di ec ion wi h a solid ed line. Ripple a i ac s esul ing om he emo e B0g adien a ea a e isible a
he h ee aliased co ne s, away om he s ong B0o se s............................ 3
3 Aliasing e ec s o B0 ield inhomogenei ies on GRAPPA econs uc ions: examples o apid dephasing a eas
and hei aliased phan om in accele a ed da a a e delinea ed wi h solid and dashed ed lines espec i ely.
Ripple a i ac s o igina ing om a emo e B0 ield g adien a e s ill isible a e GRAPPA un olding a
TE = 20 ms and o a lesse ex en a TE = 4 ms, wi h a dec eased spa ial equency. Swi ching o ex e nal
ACS o GRAPPA calib a ion signi ican ly educes a i ac s in he long TE acquisi ion. . . . . . . . . . . 5
4 E ec s o manually a enua ing he neck signal in ACS lines o mimick in a- oxel dephasing. The i s
ow shows he ACS images used o he econs uc ion below a highe esolu ion. The ACS a e ex ac ed
om a TE = 4 ms ully sampled acquisi ion, while he unde sampled da a is ex ac ed om a TE = 20 ms
acquisi ion ( he same as in Figu e 3). The dashed ed line indica es he slice below which he signal is
a i iciallya enua ed. ............................................... 5
5 In luence o GRAPPA pa ame e s. The e ec s o ke nel size [A], numbe o ACS lines and λ[B] o
accele a ion ac o 4 (2×2) a e shown wi h he SSIM and wi h e e ence he ully-sampled image (TE =
20 ms). The ed con ou s indica e he maximum SSIM a ke nel size = (2, 2, 5). Ke nel sizes o which
econs uc ion ailed a e shown in black. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
6 E ec o a misma ch be ween he ACS and accele a ed da a caused by di e en RF exci a ion pulses. ACS
da a a e composed o 36 ×36 lines o a TE = 3 ms acquisi ion, while accele a ed pTx da a (iPAT 2×2)
comes omaTE=20msacquisi ion....................................... 6
Lis o supplemen a y igu es
1. The same ipple a i ac s showing in a ious acquisi ion pa ame e s. 11.7T images we e acqui ed wi h a TE o
20 ms and iso opic esolu ions, 7T was acqui ed wi h a TE made o ma ch in a- oxel dephasing o he 11.7T ield
(20×11.7/7≈34 ms). All da a we e econs uc ed wi h 24×24 ACS lines. A 11.7T he acquisi ions we e pe o med
wi h pTx, while he ones a 7T we e pe o med in CP mode.
9
