<s>
An	O
MRI	B-Algorithm
artifact	I-Algorithm
is	O
a	O
visual	B-Algorithm
artifact	I-Algorithm
(	O
an	O
anomaly	O
seen	O
during	O
visual	O
representation	O
)	O
in	O
magnetic	O
resonance	O
imaging	O
(	O
MRI	O
)	O
.	O
</s>
<s>
Artifacts	B-Algorithm
can	O
be	O
classified	O
as	O
patient-related	O
,	O
signal	O
processing-dependent	O
and	O
hardware	O
(	O
machine	O
)	O
-related	O
.	O
</s>
<s>
A	O
motion	O
artifact	B-Algorithm
is	O
one	O
of	O
the	O
most	O
common	O
artifacts	B-Algorithm
in	O
MR	O
imaging	O
.	O
</s>
<s>
Phase-encoded	O
sampling	O
takes	O
several	O
seconds	O
,	O
or	O
even	O
minutes	O
,	O
owing	O
to	O
the	O
collection	O
of	O
all	O
the	O
k-space	B-Algorithm
lines	O
to	O
enable	O
Fourier	O
analysis	O
.	O
</s>
<s>
Several	O
methods	O
can	O
be	O
used	O
to	O
reduce	O
motion	O
artifacts	B-Algorithm
,	O
including	O
patient	O
immobilisation	O
,	O
cardiac	O
and	O
respiratory	O
gating	O
,	O
signal	O
suppression	O
of	O
the	O
tissue	O
causing	O
the	O
artifact	B-Algorithm
,	O
choosing	O
the	O
shorter	O
dimension	O
of	O
the	O
matrix	O
as	O
the	O
phase-encoding	O
direction	O
,	O
view-ordering	O
or	O
phase-reordering	O
methods	O
and	O
swapping	O
phase	O
and	O
frequency-encoding	O
directions	O
to	O
move	O
the	O
artifact	B-Algorithm
out	O
of	O
the	O
field	O
of	O
interest	O
.	O
</s>
<s>
Flow	O
can	O
manifest	O
as	O
either	O
an	O
altered	O
intravascular	O
signal	O
(	O
flow	O
enhancement	O
or	O
flow-related	O
signal	O
loss	O
)	O
,	O
or	O
as	O
flow-related	O
artifacts	B-Algorithm
(	O
ghost	O
images	O
or	O
spatial	O
misregistration	O
)	O
.	O
</s>
<s>
Flow	O
enhancement	O
,	O
also	O
known	O
as	O
inflow	O
effect	O
,	O
is	O
caused	O
by	O
fully	O
magnetised	O
protons	B-Application
entering	O
the	O
imaged	O
slice	O
while	O
the	O
stationary	O
protons	B-Application
have	O
not	O
fully	O
regained	O
their	O
magnetization	O
.	O
</s>
<s>
The	O
fully	O
magnetized	O
protons	B-Application
yield	O
a	O
high	O
signal	O
in	O
comparison	O
with	O
the	O
rest	O
of	O
the	O
surroundings	O
.	O
</s>
<s>
High	O
velocity	O
flow	O
causes	O
the	O
protons	B-Application
entering	O
the	O
image	O
to	O
be	O
removed	O
from	O
it	O
by	O
the	O
time	O
the	O
180-degree	O
pulse	O
is	O
administered	O
.	O
</s>
<s>
The	O
effect	O
is	O
that	O
these	O
protons	B-Application
do	O
not	O
contribute	O
to	O
the	O
echo	O
and	O
are	O
registered	O
as	O
a	O
signal	O
void	O
or	O
flow-related	O
signal	O
loss	O
(	O
Fig	O
.	O
</s>
<s>
Spatial	O
misregistration	O
manifests	O
as	O
displacement	O
of	O
an	O
intravascular	O
signal	O
owing	O
to	O
position	O
encoding	O
of	O
a	O
voxel	O
in	O
the	O
phase	O
direction	O
preceding	O
frequency	O
encoding	O
by	O
time	O
TE/2.The	O
intensity	O
of	O
the	O
artifact	B-Algorithm
is	O
dependent	O
on	O
the	O
signal	O
intensity	O
from	O
the	O
vessel	O
,	O
and	O
is	O
less	O
apparent	O
with	O
increased	O
TE	O
.	O
</s>
<s>
Metal	O
artifacts	B-Algorithm
occur	O
at	O
interfaces	O
of	O
tissues	O
with	O
different	O
magnetic	O
susceptibilities	O
,	O
which	O
cause	O
local	O
magnetic	O
fields	O
to	O
distort	O
the	O
external	O
magnetic	O
field	O
.	O
</s>
<s>
Metal	O
artifacts	B-Algorithm
are	O
caused	O
by	O
external	O
ferromagnetics	O
such	O
as	O
cobalt	O
containing	O
make-up	O
,	O
internal	O
ferromagnetics	O
such	O
as	O
surgical	O
clips	O
,	O
spinal	O
hardware	O
and	O
other	O
orthopaedic	O
devices	O
,	O
and	O
in	O
some	O
cases	O
,	O
metallic	O
objects	O
swallowed	O
by	O
people	O
with	O
pica	O
.	O
</s>
<s>
Manifestation	O
of	O
these	O
artifacts	B-Algorithm
is	O
variable	O
,	O
including	O
total	O
signal	O
loss	O
,	O
peripheral	O
high	O
signal	O
and	O
image	O
distortion	O
(	O
Figs	O
3	O
and	O
4	O
)	O
.	O
</s>
<s>
Reduction	O
of	O
these	O
artifacts	B-Algorithm
can	O
be	O
attempted	O
by	O
orientating	O
the	O
long	O
axis	O
of	O
an	O
implant	O
or	O
device	O
parallel	O
to	O
the	O
long	O
axis	O
of	O
the	O
external	O
magnetic	O
field	O
,	O
possible	O
with	O
mobile	O
extremity	O
imaging	O
and	O
an	O
open	O
magnet	O
.	O
</s>
<s>
Further	O
methods	O
used	O
are	O
choosing	O
the	O
appropriate	O
frequency	O
encoding	O
direction	O
,	O
since	O
metal	O
artifacts	B-Algorithm
are	O
most	O
pronounced	O
in	O
this	O
direction	O
,	O
using	O
smaller	O
voxel	O
sizes	O
,	O
fast	O
imaging	O
sequences	O
,	O
increased	O
readout	O
bandwidth	O
and	O
avoiding	O
gradient-echo	O
imaging	O
when	O
metal	O
is	O
present	O
.	O
</s>
<s>
A	O
technique	O
called	O
MARS	O
(	O
metal	O
artifact	B-Algorithm
reduction	O
sequence	O
)	O
applies	O
an	O
additional	O
gradient	O
,	O
along	O
the	O
slice	O
select	O
gradient	O
at	O
the	O
time	O
the	O
frequency	O
encoding	O
gradient	O
is	O
applied	O
.	O
</s>
<s>
==	O
Signal	O
processing	O
dependent	O
artifacts	B-Algorithm
==	O
The	O
ways	O
in	O
which	O
the	O
data	O
are	O
sampled	O
,	O
processed	O
and	O
mapped	O
out	O
on	O
the	O
image	O
matrix	O
manifest	O
these	O
artifacts	B-Algorithm
.	O
</s>
<s>
Chemical	O
shift	O
artifact	B-Algorithm
occurs	O
at	O
the	O
fat/water	O
interface	O
in	O
the	O
frequency	O
encoding	O
direction	O
(	O
Fig	O
.	O
</s>
<s>
These	O
artifacts	B-Algorithm
arise	O
due	O
to	O
the	O
difference	O
in	O
resonance	O
of	O
protons	B-Application
as	O
a	O
result	O
of	O
their	O
micromagnetic	O
environment	O
.	O
</s>
<s>
The	O
protons	B-Application
of	O
fat	O
resonate	O
at	O
a	O
slightly	O
lower	O
frequency	O
than	O
those	O
of	O
water	O
.	O
</s>
<s>
High	O
field	O
strength	O
magnets	O
are	O
particularly	O
susceptible	O
to	O
this	O
artifact	B-Algorithm
.	O
</s>
<s>
Determination	O
of	O
the	O
artifact	B-Algorithm
can	O
be	O
made	O
by	O
swapping	O
the	O
phase	O
-	O
and	O
frequency-encoding	O
gradients	O
and	O
examining	O
the	O
resultant	O
shift	O
of	O
fat	O
tissue	O
.	O
</s>
<s>
Partial	O
volume	O
artifacts	B-Algorithm
arise	O
from	O
the	O
size	O
of	O
the	O
voxel	O
over	O
which	O
the	O
signal	O
is	O
averaged	O
.	O
</s>
<s>
Reduction	O
of	O
these	O
artifacts	B-Algorithm
is	O
accomplished	O
by	O
using	O
a	O
smaller	O
pixel	O
size	O
and/or	O
a	O
smaller	O
slice	O
thickness	O
.	O
</s>
<s>
A	O
wrap-around	O
artifact	B-Algorithm
also	O
known	O
as	O
an	O
aliasing	B-Error_Name
artifact	I-Error_Name
,	O
is	O
a	O
result	O
of	O
mismapping	O
of	O
anatomy	O
that	O
lies	O
outside	O
the	O
field	O
of	O
view	O
but	O
within	O
the	O
slice	O
volume	O
.	O
</s>
<s>
If	O
not	O
,	O
the	O
Fourier	B-Algorithm
transform	I-Algorithm
will	O
assign	O
very	O
low	O
values	O
to	O
the	O
frequency	O
signals	O
greater	O
than	O
the	O
Nyquist	O
limit	O
.	O
</s>
<s>
In	O
the	O
phase	O
encode	O
direction	O
,	O
artifacts	B-Algorithm
can	O
be	O
reduced	O
by	O
an	O
increasing	O
number	O
of	O
phase	O
encode	O
steps	O
(	O
increased	O
image	O
time	O
)	O
.	O
</s>
<s>
Gibbs	O
artifacts	B-Algorithm
or	O
Gibbs	O
ringing	O
artifacts	B-Algorithm
,	O
also	O
known	O
as	O
truncation	O
artifacts	B-Algorithm
are	O
caused	O
by	O
the	O
under-sampling	O
of	O
high	O
spatial	O
frequencies	O
at	O
sharp	O
boundaries	O
in	O
the	O
image	O
.	O
</s>
<s>
Lack	O
of	O
appropriate	O
high-frequency	O
components	O
leads	O
to	O
an	O
oscillation	O
at	O
a	O
sharp	O
transition	O
known	O
as	O
a	O
ringing	O
artifact	B-Algorithm
.	O
</s>
<s>
Ringing	O
artifacts	B-Algorithm
are	O
more	O
prominent	O
in	O
smaller	O
digital	O
matrix	O
sizes	O
.	O
</s>
<s>
Methods	O
employed	O
to	O
correct	O
Gibbs	O
artifact	B-Algorithm
include	O
filtering	O
the	O
k-space	B-Algorithm
data	O
prior	O
to	O
Fourier	B-Algorithm
transform	I-Algorithm
,	O
increasing	O
the	O
matrix	O
size	O
for	O
a	O
given	O
field	O
of	O
view	O
,	O
the	O
Gegenbauer	O
reconstruction	O
and	O
Bayesian	O
approach	O
.	O
</s>
<s>
==	O
Machine/hardware	O
-related	O
artifacts	B-Algorithm
==	O
This	O
is	O
a	O
wide	O
and	O
still	O
expanding	O
subject	O
.	O
</s>
<s>
Only	O
a	O
few	O
common	O
artifacts	B-Algorithm
are	O
recognised	O
.	O
</s>
<s>
Magnetic	O
field	O
gradients	O
are	O
used	O
to	O
spatially	O
encode	O
the	O
location	O
of	O
signals	O
from	O
excited	O
protons	B-Application
within	O
the	O
volume	O
being	O
imaged	O
.	O
</s>
<s>
In	O
practice	O
this	O
is	O
not	O
always	O
possible	O
because	O
of	O
the	O
necessity	O
of	O
displacing	O
motion	O
artifacts	B-Algorithm
.	O
</s>
<s>
Variation	O
in	O
intensity	O
across	O
the	O
image	O
may	O
be	O
due	O
to	O
the	O
failure	O
of	O
the	O
RF	B-Algorithm
coil	I-Algorithm
,	O
non-uniform	O
B1	O
field	O
,	O
non-uniform	O
sensitivity	O
of	O
the	O
receive	O
only	O
coil	O
(	O
spaces	O
between	O
wire	O
in	O
the	O
coil	O
,	O
uneven	O
distribution	O
of	O
wire	O
)	O
,	O
or	O
presence	O
of	O
non-ferromagnetic	O
material	O
in	O
the	O
imaged	O
object	O
.	O
</s>
<s>
When	O
using	O
a	O
FLASH	B-Algorithm
sequence	O
,	O
tip	O
angle	O
variations	O
due	O
to	O
B1	O
inhomogeneity	O
can	O
affect	O
the	O
contrast	O
of	O
the	O
image	O
.	O
</s>
<s>
This	O
is	O
due	O
to	O
imperfect	O
flip	B-Algorithm
angles	I-Algorithm
throughout	O
the	O
slice	O
,	O
but	O
particularly	O
around	O
the	O
edges	O
of	O
the	O
body	O
,	O
resulting	O
in	O
imperfect	O
magnetization	O
recovery	O
.	O
</s>
<s>
Human	O
body	O
is	O
full	O
of	O
protons	B-Application
and	O
during	O
imaging	O
,	O
the	O
B0	O
field	O
aligns	O
these	O
individual	O
protons	B-Application
to	O
a	O
net	O
magnetization	O
in	O
the	O
direction	O
of	O
the	O
magnetic	O
field	O
.	O
</s>
<s>
This	O
flip	B-Algorithm
angle	I-Algorithm
scales	O
with	O
the	O
B1	O
field	O
amplitude	O
.	O
</s>
<s>
Accurate	O
flip	B-Algorithm
angle	I-Algorithm
is	O
crucial	O
because	O
the	O
measured	O
MR	O
signals	O
depend	O
on	O
the	O
flip	B-Algorithm
angle	I-Algorithm
of	O
the	O
protons	B-Application
.	O
</s>
<s>
In	O
reality	O
,	O
different	O
areas	O
of	O
a	O
slice	O
see	O
different	O
radio	O
frequency	O
fields	O
,	O
leading	O
to	O
different	O
flip	B-Algorithm
angles	I-Algorithm
.	O
</s>
<s>
But	O
as	O
the	O
main	O
magnetic	O
field	O
is	O
increased	O
,	O
these	O
wavelengths	O
become	O
the	O
same	O
or	O
smaller	O
than	O
the	O
regions	O
of	O
the	O
body	O
being	O
imaged	O
,	O
resulting	O
in	O
flip	B-Algorithm
angle	I-Algorithm
inhomogeneity	O
.	O
</s>
<s>
A	O
similar	O
artifact	B-Algorithm
may	O
be	O
caused	O
by	O
non-uniformity	O
in	O
slice	O
thickness	O
.	O
</s>
<s>
Stray	O
RF	O
signals	O
can	O
cause	O
various	O
artifacts	B-Algorithm
.	O
</s>
<s>
Zero	O
line	O
and	O
star	O
artifacts	B-Algorithm
are	O
due	O
to	O
system	O
noise	O
or	O
any	O
cause	O
of	O
RF	O
pollution	O
within	O
the	O
room	O
(	O
Faraday	O
cage	O
)	O
.	O
</s>
<s>
When	O
the	O
chosen	O
T1	O
equals	O
69%	O
of	O
the	O
T1	O
value	O
of	O
a	O
particular	O
tissue	O
,	O
a	O
bounce	O
point	O
artifact	B-Algorithm
occurs	O
.	O
</s>
<s>
To	O
overcome	O
these	O
interference	O
artifacts	B-Algorithm
,	O
the	O
acquisition	O
of	O
two	O
independent	O
sets	O
of	O
gapped	O
multi-slice	O
images	O
need	O
to	O
be	O
included	O
,	O
and	O
subsequently	O
reordered	O
during	O
display	O
of	O
the	O
full	O
image	O
set	O
.	O
</s>
<s>
Gating	O
is	O
best	O
suited	O
for	O
mitigating	O
breathing	O
and	O
cardiac	O
artifacts	B-Algorithm
.	O
</s>
<s>
The	O
Pilot	O
Tone	O
method	O
is	O
great	O
for	O
detecting	O
respiratory	O
motion	O
artifacts	B-Algorithm
.	O
</s>
<s>
The	O
method	O
corrects	O
motion-related	O
artifacts	B-Algorithm
by	O
acquiring	O
a	O
joint	O
estimation	O
of	O
the	O
desired	O
motion-free	O
image	O
and	O
the	O
associated	O
motion	O
trajectory	O
by	O
minimizing	O
the	O
data	O
consistency	O
error	O
of	O
a	O
SENSE	O
forward	O
model	O
that	O
includes	O
rigid-body	O
subject	O
motion	O
.	O
</s>
<s>
Group	O
coils	O
based	O
on	O
artifact	B-Algorithm
properties	O
.	O
</s>
<s>
TAMER	O
is	O
applied	O
to	O
groups	O
of	O
coils	O
with	O
the	O
largest	O
correlation	O
artifacts	B-Algorithm
to	O
attain	O
the	O
motion	O
and	O
image	O
estimation	O
.	O
</s>
<s>
In	O
recent	O
years	O
,	O
neural	B-Architecture
networks	I-Architecture
have	O
generated	O
a	O
great	O
deal	O
of	O
interest	O
by	O
outperforming	O
traditional	O
methods	O
on	O
longstanding	O
problems	O
across	O
many	O
fields	O
.	O
</s>
<s>
Machine	O
learning	O
,	O
and	O
by	O
extension	O
neural	B-Architecture
networks	I-Architecture
,	O
have	O
been	O
used	O
in	O
many	O
facets	O
of	O
MRI	O
—	O
for	O
instance	O
,	O
speeding	O
up	O
image	O
reconstruction	O
,	O
or	O
improving	O
reconstruction	O
quality	O
when	O
working	O
with	O
a	O
lack	O
of	O
data	O
.	O
</s>
<s>
Neural	B-Architecture
networks	I-Architecture
have	O
also	O
been	O
used	O
in	O
motion	O
artifact	B-Algorithm
correction	O
thanks	O
to	O
their	O
ability	O
to	O
learn	O
visual	O
information	O
from	O
data	O
,	O
as	O
well	O
as	O
infer	O
underlying	O
,	O
latent	O
representations	O
in	O
data	O
.	O
</s>
<s>
Network	O
Accelerated	O
Motion	O
Estimation	O
and	O
Reduction	O
(	O
NAMER	O
)	O
is	O
a	O
retrospective	O
motion	O
correction	O
technique	O
that	O
utilizes	O
convolutional	B-Architecture
neural	I-Architecture
networks	I-Architecture
(	O
CNNs	B-Architecture
)	O
,	O
a	O
class	O
of	O
neural	B-Architecture
networks	I-Architecture
designed	O
to	O
process	O
and	O
learn	O
from	O
visual	O
information	O
such	O
as	O
images	O
.	O
</s>
<s>
This	O
is	O
a	O
follow-up	O
from	O
the	O
authors	O
of	O
the	O
TAMER	O
paper	O
titled	O
Network	O
Accelerated	O
Motion	O
Estimation	O
and	O
Reduction	O
(	O
NAMER	O
)	O
:	O
Convolutional	B-Architecture
neural	I-Architecture
network	I-Architecture
guided	O
retrospective	O
motion	O
correction	O
using	O
a	O
separable	O
motion	O
model	O
.	O
</s>
<s>
Similar	O
to	O
TAMER	O
,	O
the	O
paper	O
aims	O
to	O
correct	O
for	O
motion-related	O
artifacts	B-Algorithm
by	O
way	O
of	O
estimating	O
a	O
desired	O
motion-free	O
image	O
and	O
optimizing	O
parameters	O
for	O
a	O
SENSE	O
forward	O
model	O
describing	O
the	O
relationship	O
between	O
raw	O
k-space	B-Algorithm
data	O
and	O
image	O
space	O
while	O
factoring	O
in	O
rigid	O
motion	O
.	O
</s>
<s>
A	O
SENSE	O
forward	O
model	O
is	O
used	O
to	O
induce	O
synthetic	O
motion	O
artifacts	B-Algorithm
in	O
raw	O
k-space	B-Algorithm
data	O
,	O
allowing	O
us	O
access	O
to	O
both	O
data	O
with	O
motion	O
artifacts	B-Algorithm
,	O
as	O
well	O
as	O
the	O
ground-truth	O
image	O
without	O
motion	O
artifacts	B-Algorithm
.	O
</s>
<s>
This	O
is	O
important	O
to	O
the	O
NAMER	O
technique	O
,	O
because	O
it	O
utilizes	O
a	O
Convolutional	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
CNN	B-Architecture
)	O
to	O
frontload	O
image	O
estimation	O
and	O
guide	O
model	O
parameter	O
estimation	O
.	O
</s>
<s>
Convolutional	B-Architecture
Neural	I-Architecture
Networks	I-Architecture
leverage	O
convolution	O
kernels	O
to	O
analyze	O
visual	O
imagery	O
.	O
</s>
<s>
Here	O
,	O
a	O
27-layer	O
network	O
is	O
used	O
with	O
multiple	O
convolution	O
layers	O
,	O
batch	B-General_Concept
normalization	I-General_Concept
,	O
and	O
ReLU	B-Algorithm
activations	O
.	O
</s>
<s>
The	O
CNN	B-Architecture
attempts	O
to	O
learn	O
the	O
image	B-Algorithm
artifacts	I-Algorithm
from	O
the	O
motion-corrupted	O
input	O
data	O
.	O
</s>
<s>
The	O
estimate	O
for	O
these	O
artifacts	B-Algorithm
,	O
denoted	O
as	O
,	O
are	O
then	O
subtracted	O
from	O
the	O
motion-corrupted	O
input	O
data	O
in	O
order	O
to	O
produce	O
a	O
best	O
estimate	O
for	O
the	O
motion-free	O
image:This	O
serves	O
two	O
purposes	O
:	O
First	O
,	O
it	O
allows	O
the	O
CNN	B-Architecture
to	O
perform	O
backpropagation	B-Algorithm
and	O
update	O
its	O
model	O
weights	O
by	O
using	O
a	O
mean	B-Algorithm
square	I-Algorithm
error	I-Algorithm
loss	I-Algorithm
function	I-Algorithm
comparing	O
the	O
difference	O
between	O
and	O
the	O
known	O
ground-truth	O
motion-free	O
image	O
.	O
</s>
<s>
Using	O
a	O
CNN	B-Architecture
effectively	O
allows	O
us	O
to	O
bypass	O
the	O
second	O
stage	O
of	O
TAMER	O
by	O
skipping	O
the	O
joint	O
parameter	O
search	O
.	O
</s>
<s>
The	O
new	O
model	O
parameters	O
are	O
then	O
used	O
in	O
a	O
standard	O
Least	B-Algorithm
Squares	I-Algorithm
optimization	O
problem	O
to	O
reconstruct	O
an	O
image	O
that	O
minimizes	O
the	O
distance	O
between	O
the	O
k-space	B-Algorithm
data	O
,	O
and	O
the	O
result	O
of	O
applying	O
the	O
SENSE	O
forward	O
model	O
under	O
our	O
new	O
parameter	O
estimate	O
to	O
our	O
best	O
estimate	O
for	O
the	O
motion-free	O
image:This	O
process	O
is	O
repeated	O
until	O
a	O
desired	O
number	O
of	O
time	O
steps	O
,	O
or	O
when	O
the	O
change	O
in	O
reconstructed	O
image	O
is	O
sufficiently	O
low	O
.	O
</s>
<s>
The	O
NAMER	O
technique	O
has	O
shown	O
itself	O
to	O
be	O
very	O
effective	O
in	O
correcting	O
for	O
rigid	O
motion	O
artifacts	B-Algorithm
,	O
and	O
converges	O
much	O
faster	O
than	O
other	O
methods	O
including	O
TAMER	O
.	O
</s>
<s>
Other	O
more	O
advanced	O
techniques	O
take	O
advantage	O
of	O
generative	B-Algorithm
adversarial	I-Algorithm
networks	I-Algorithm
(	O
GANs	O
)	O
which	O
aim	O
to	O
learn	O
the	O
underlying	O
latent	O
representation	O
of	O
data	O
in	O
order	O
to	O
synthesize	O
new	O
examples	O
that	O
are	O
indistinguishable	O
from	O
real	O
data	O
.	O
</s>
<s>
Here	O
,	O
two	O
neural	B-Architecture
networks	I-Architecture
,	O
a	O
Generator	O
Network	O
and	O
a	O
Discriminator	O
Network	O
,	O
are	O
modelled	O
as	O
agents	O
competing	O
in	O
a	O
game	O
.	O
</s>
<s>
Specific	O
to	O
motion	O
artifact	B-Algorithm
correction	O
in	O
MRI	O
,	O
the	O
Generator	O
Network	O
takes	O
in	O
an	O
image	O
with	O
motion	O
artifacts	B-Algorithm
,	O
and	O
outputs	O
an	O
image	O
without	O
motion	O
artifacts	B-Algorithm
.	O
</s>
<s>
Various	O
studies	O
have	O
shown	O
that	O
GANs	O
perform	O
very	O
well	O
in	O
correcting	O
for	O
motion	O
artifacts	B-Algorithm
.	O
</s>
<s>
This	O
correction	O
method	O
was	O
shown	O
to	O
have	O
the	O
greatest	O
effect	O
on	O
sequences	O
that	O
suffer	O
from	O
B1	O
inhomogeneity	O
artifacts	B-Algorithm
but	O
has	O
no	O
effect	O
on	O
those	O
with	O
B0	O
inhomogeneity	O
.	O
</s>
<s>
This	O
method	O
exploits	O
the	O
tradeoff	O
between	O
B1	O
dependence	O
and	O
coil	O
sensitivity	O
dependence	O
in	O
FLASH	B-Algorithm
sequences	O
and	O
allows	O
the	O
user	O
to	O
select	O
an	O
optimized	O
flip	B-Algorithm
angle	I-Algorithm
that	O
will	O
reduce	O
B1	O
dependence	O
.	O
</s>
<s>
For	O
anatomical	O
studies	O
using	O
the	O
FLASH	B-Algorithm
sequence	O
that	O
can	O
be	O
performed	O
with	O
one	O
transmit	O
and	O
receive	O
coil	O
,	O
this	O
method	O
can	O
be	O
used	O
to	O
reduce	O
B1	O
inhomogeneity	O
artifacts	B-Algorithm
.	O
</s>
<s>
However	O
,	O
the	O
method	O
would	O
not	O
be	O
suitable	O
for	O
exams	O
under	O
strict	O
time	O
constraints	O
,	O
since	O
the	O
user	O
first	O
needs	O
to	O
perform	O
flip	B-Algorithm
angle	I-Algorithm
optimization	O
.	O
</s>
<s>
This	O
can	O
be	O
done	O
by	O
changing	O
the	O
way	O
that	O
the	O
RF	B-Algorithm
coil	I-Algorithm
is	O
driven	O
and	O
excited	O
.	O
</s>
<s>
Changing	O
the	O
shape	O
of	O
the	O
coils	O
can	O
be	O
used	O
to	O
reduce	O
B1	O
inhomogeneity	O
artifacts	B-Algorithm
.	O
</s>
<s>
Another	O
method	O
to	O
correct	O
for	O
B1	O
inhomogeneity	O
is	O
to	O
employ	O
the	O
infrastructure	O
in	O
place	O
from	O
a	O
parallel	O
system	O
to	O
generate	O
multiple	O
RF	O
pulses	O
of	O
lower	O
flip	B-Algorithm
angles	I-Algorithm
that	O
,	O
together	O
,	O
can	O
result	O
in	O
the	O
same	O
flip	B-Algorithm
angle	I-Algorithm
as	O
that	O
created	O
using	O
a	O
single	O
transmit	O
coil	O
.	O
</s>
<s>
Overall	O
,	O
when	O
this	O
method	O
is	O
used	O
with	O
the	O
correct	O
selection	O
of	O
RF	O
pulses	O
and	O
optimized	O
for	O
a	O
low	O
power	O
deposition	O
,	O
the	O
artifacts	B-Algorithm
from	O
B1	O
inhomogeneity	O
can	O
be	O
greatly	O
reduced	O
.	O
</s>
<s>
This	O
method	O
is	O
advantageous	O
for	O
reducing	O
artifacts	B-Algorithm
at	O
the	O
source	O
,	O
particularly	O
when	O
accurate	O
flip	B-Algorithm
angle	I-Algorithm
is	O
critical	O
and	O
for	O
increasing	O
signal	O
to	O
noise	O
ratio	O
.	O
</s>
<s>
One	O
way	O
to	O
achieve	O
perfect	O
spin	O
inversion	O
despite	O
B1	O
inhomogeneity	O
is	O
to	O
use	O
adiabatic	B-Algorithm
pulses	I-Algorithm
.	O
</s>
<s>
This	O
correction	O
method	O
works	O
by	O
removing	O
the	O
source	O
of	O
the	O
problem	O
and	O
applying	O
pulses	O
that	O
will	O
not	O
generate	O
flip	B-Algorithm
angle	I-Algorithm
errors	O
.	O
</s>
<s>
Specific	O
sequences	O
that	O
employ	O
adiabatic	B-Algorithm
pulses	I-Algorithm
for	O
increased	O
flip	B-Algorithm
angle	I-Algorithm
uniformity	O
include	O
a	O
slice	O
selective	O
spin-echo	O
pulse	O
,	O
adiabatic	O
180	O
degrees	O
inversion	O
RF	O
pulses	O
,	O
and	O
180	O
degrees	O
refocusing	O
pulses	O
.	O
</s>
<s>
Since	O
the	O
B1	O
field	O
was	O
not	O
homogeneous	O
when	O
the	O
images	O
were	O
acquired	O
,	O
the	O
flip	B-Algorithm
angles	I-Algorithm
and	O
subsequent	O
acquired	O
signals	O
are	O
imprecise	O
.	O
</s>
<s>
To	O
correct	O
RF	O
inhomogeneity	O
artifacts	B-Algorithm
using	O
post-processing	O
corrections	O
,	O
there	O
are	O
a	O
few	O
methods	O
to	O
map	O
the	O
B1	O
field	O
.	O
</s>
<s>
A	O
common	O
and	O
robust	O
method	O
that	O
uses	O
the	O
results	O
from	O
two	O
images	O
acquired	O
at	O
flip	B-Algorithm
angles	I-Algorithm
of	O
and	O
.	O
</s>
<s>
Similar	O
to	O
the	O
double	O
angle	O
method	O
,	O
the	O
phase	O
map	O
method	O
uses	O
two	O
images	O
;	O
however	O
,	O
this	O
method	O
relies	O
on	O
the	O
accrual	O
of	O
phase	O
to	O
determine	O
the	O
real	O
flip	B-Algorithm
angle	I-Algorithm
of	O
each	O
spin	O
.	O
</s>
