<s>
Iterative	B-Algorithm
reconstruction	I-Algorithm
refers	O
to	O
iterative	B-Algorithm
algorithms	O
used	O
to	O
reconstruct	O
2D	O
and	O
3D	B-Algorithm
images	I-Algorithm
in	O
certain	O
imaging	B-Algorithm
techniques	O
.	O
</s>
<s>
For	O
example	O
,	O
in	O
computed	O
tomography	B-Algorithm
an	O
image	O
must	O
be	O
reconstructed	O
from	O
projections	O
of	O
an	O
object	O
.	O
</s>
<s>
In	O
recent	O
research	O
works	O
,	O
scientists	O
have	O
shown	O
that	O
extremely	O
fast	O
computations	O
and	O
massive	O
parallelism	O
is	O
possible	O
for	O
iterative	B-Algorithm
reconstruction	I-Algorithm
,	O
which	O
makes	O
iterative	B-Algorithm
reconstruction	I-Algorithm
practical	O
for	O
commercialization	O
.	O
</s>
<s>
In	O
this	O
case	O
,	O
a	O
direct	O
algorithm	O
has	O
to	O
approximate	O
the	O
solution	O
,	O
which	O
might	O
cause	O
visible	O
reconstruction	O
artifacts	B-Algorithm
in	O
the	O
image	O
.	O
</s>
<s>
Iterative	B-Algorithm
algorithms	O
approach	O
the	O
correct	O
solution	O
using	O
multiple	O
iteration	B-Algorithm
steps	O
,	O
which	O
allows	O
to	O
obtain	O
a	O
better	O
reconstruction	O
at	O
the	O
cost	O
of	O
a	O
higher	O
computation	O
time	O
.	O
</s>
<s>
The	O
Algebraic	B-Algorithm
Reconstruction	I-Algorithm
Technique	I-Algorithm
(	O
ART	O
)	O
was	O
the	O
first	O
iterative	B-Algorithm
reconstruction	I-Algorithm
technique	O
used	O
for	O
computed	O
tomography	B-Algorithm
by	O
Hounsfield	O
.	O
</s>
<s>
The	O
iterative	B-Algorithm
Sparse	I-Algorithm
Asymptotic	I-Algorithm
Minimum	I-Algorithm
Variance	I-Algorithm
algorithm	O
is	O
an	O
iterative	B-Algorithm
,	O
parameter-free	O
superresolution	B-Algorithm
tomographic	B-Algorithm
reconstruction	I-Algorithm
method	O
inspired	O
by	O
compressed	O
sensing	O
,	O
with	O
applications	O
in	O
synthetic-aperture	B-Application
radar	I-Application
,	O
computed	O
tomography	B-Algorithm
scan	O
,	O
and	O
magnetic	O
resonance	O
imaging	B-Algorithm
(	O
MRI	O
)	O
.	O
</s>
<s>
There	O
are	O
typically	O
five	O
components	O
to	O
statistical	O
iterative	B-Algorithm
image	B-Algorithm
reconstruction	I-Algorithm
algorithms	O
,	O
e.g.	O
</s>
<s>
An	O
algorithm	O
,	O
usually	O
iterative	B-Algorithm
,	O
for	O
minimizing	O
the	O
cost	O
function	O
,	O
including	O
some	O
initial	O
estimate	O
of	O
the	O
image	O
and	O
some	O
stopping	O
criterion	O
for	O
terminating	O
the	O
iterations	B-Algorithm
.	O
</s>
<s>
In	O
learned	O
iterative	B-Algorithm
reconstruction	I-Algorithm
,	O
the	O
updating	O
algorithm	O
is	O
learned	O
from	O
training	O
data	O
using	O
techniques	O
from	O
machine	O
learning	O
such	O
as	O
convolutional	B-Architecture
neural	I-Architecture
networks	I-Architecture
,	O
while	O
still	O
incorporating	O
the	O
image	O
formation	O
model	O
.	O
</s>
<s>
The	O
advantages	O
of	O
the	O
iterative	B-Algorithm
approach	O
include	O
improved	O
insensitivity	O
to	O
noise	O
and	O
capability	O
of	O
reconstructing	O
an	O
optimal	O
image	O
in	O
the	O
case	O
of	O
incomplete	O
data	O
.	O
</s>
<s>
The	O
method	O
has	O
been	O
applied	O
in	O
emission	O
tomography	B-Algorithm
modalities	O
like	O
SPECT	O
and	O
PET	B-Application
,	O
where	O
there	O
is	O
significant	O
attenuation	O
along	O
ray	O
paths	O
and	O
noise	O
statistics	O
are	O
relatively	O
poor	O
.	O
</s>
<s>
Such	O
algorithms	O
compute	O
estimates	O
of	O
the	O
likely	O
distribution	O
of	O
annihilation	O
events	O
that	O
led	O
to	O
the	O
measured	O
data	O
,	O
based	O
on	O
statistical	O
principle	O
,	O
often	O
providing	O
better	O
noise	O
profiles	O
and	O
resistance	O
to	O
the	O
streak	O
artifacts	B-Algorithm
common	O
with	O
FBP	O
.	O
</s>
<s>
Good	O
's	O
roughness	O
method	O
may	O
yield	O
superior	O
performance	O
to	O
expectation-maximization-based	O
methods	O
which	O
involve	O
a	O
Poisson	O
likelihood	O
function	O
only	O
.	O
</s>
<s>
In	O
Magnetic	O
Resonance	O
Imaging	B-Algorithm
it	O
can	O
be	O
used	O
to	O
reconstruct	O
images	O
from	O
data	O
acquired	O
with	O
multiple	O
receive	O
coils	O
and	O
with	O
sampling	O
patterns	O
different	O
from	O
the	O
conventional	O
Cartesian	O
grid	O
and	O
allows	O
the	O
use	O
of	O
improved	O
regularization	O
techniques	O
(	O
e.g.	O
</s>
<s>
reconstruct	O
images	O
from	O
data	O
acquired	O
in	O
a	O
very	O
short	O
time	O
as	O
required	O
for	O
real-time	B-Algorithm
MRI	I-Algorithm
(	O
rt-MRI	O
)	O
.	O
</s>
<s>
In	O
Cryo	O
Electron	O
Tomography	B-Algorithm
,	O
where	O
the	O
limited	O
number	O
of	O
projections	O
are	O
acquired	O
due	O
to	O
the	O
hardware	O
limitations	O
and	O
to	O
avoid	O
the	O
biological	O
specimen	O
damage	O
,	O
it	O
can	O
be	O
used	O
along	O
with	O
compressive	O
sensing	O
techniques	O
or	O
regularization	O
functions	O
(	O
e.g.	O
</s>
<s>
Here	O
is	O
an	O
example	O
that	O
illustrates	O
the	O
benefits	O
of	O
iterative	B-Algorithm
image	B-Algorithm
reconstruction	I-Algorithm
for	O
cardiac	O
MRI	O
.	O
</s>
