Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
12
,
No.
3
,
Decem
ber
201
8
, p
p.
974
~
983
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
2
.i
3
.pp
974
-
983
974
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
A Re
view on R
egist
ration
of
Med
ical Ima
ges Usin
g Graph
Theoreti
c A
pp
ro
ac
h
es
Ak
sh
aya R
, He
ma
P
M
e
non
Depa
rtment
o
f
C
om
pute
r
Scie
n
ce a
nd
Engi
n
ee
rin
g,
Am
rit
a
Schoo
l
of Engin
ee
ring
,
Coim
bat
ore
Am
rit
a
Vishw
a
Vid
y
ap
eetha
m
,
I
ndia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
N
ov
22, 201
7
Re
vised Jan
9
, 201
8
Accepte
d Oct
18, 201
8
Im
age
reg
istr
at
i
on
is
ver
y
esse
nti
al
for
image
ana
l
y
sis
espe
cially
in
t
h
e
m
edi
ca
l
field.
T
he
reg
iste
red
image
h
el
ps
us
to
find
a
lot
of
things
li
ke
th
e
pre
senc
e
of
a
t
um
or,
ane
ur
y
sm
,
and
m
an
y
m
ore
.
Im
ag
e
reg
ist
rat
ion
is
a
proc
ess of
al
igning two
images
tha
t
were
c
apt
ur
e
d
during
diffe
ren
t
condi
t
ions
tha
t
m
ake
s a
nalyzing
of
imag
es
poss
ibl
e.
Th
ere
ar
e
var
ious
m
et
ho
ds
in
whic
h
images
ca
n
be
r
egi
ster
ed.
Thi
s
pape
r
m
ai
n
l
y
di
scuss
es
the
m
ethods
tha
t
us
e
the
gra
ph
appr
o
a
ch
to
reg
iste
r
th
e
m
edi
ca
l
images.
The
m
ai
n
foc
us
is
on
rigi
d
and
n
on
-
rig
id
r
egi
stra
ti
on
te
ch
nique
s
and
gives
a
bri
ef
des
cription
of
th
e
m
et
hods pre
sent
.
Ke
yw
or
ds:
Gr
a
ph the
or
y
Im
age r
egistrat
ion
Me
dical
i
m
ages
Non
-
rigi
d regis
trat
ion
Ri
gid
r
e
gistrati
on
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed.
Corres
pond
in
g
Aut
h
or
:
Ak
s
haya R,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce a
nd E
ng
i
ne
erin
g
,
Am
rita
Sch
ool
of Enginee
rin
g,
Coim
bator
e
A
m
ri
ta
V
ishw
a
Vidyapeet
ham
, Ind
ia
.
Em
a
il
:
cb.
en.p
2cv
i
16001@c
b.st
udents.am
rita
.edu
,
p_hem
a@cb
.am
rita
.ed
u
1.
INTROD
U
CTION
In
pr
es
ent
day’
s
world
,
with
va
st
dev
el
op
m
ent
in
the
im
age
captu
rin
g
te
ch
niques,
im
ages
are
us
e
d
in
alm
os
t
al
l
area
s
from
m
edical
industry
to
spa
ce
te
chnolo
gy.
The
im
ages
t
hat
are
capt
ur
e
d
he
re
are
no
t
on
e
or
two
but
hu
ge
in
num
ber
s,
if
at
al
l
we
wan
t
our
com
pu
te
r
s
to
detect
any
changes,
ide
nt
ify
so
m
e
thing
fr
om
them
,
then,
al
l
m
us
t
be
wei
ghed
on
t
he
sa
m
e
scal
e.
But
the
im
ages
that
are
ca
ptured
are
diff
e
ren
t
due
to
var
i
ou
s
reas
ons
on
t
he
pret
e
xt
of
e
xtracti
ng
m
or
e
inf
orm
at
ion
from
t
he
im
ages
of
the
obj
ect
or
scene.
To
an
al
yz
e
these
i
m
ages,
firs
t
they
hav
e
to
be
bro
ught
to
t
he
sam
e
scal
e,
that
is,
they
ha
ve
to
be
al
igne
d
t
o
each
oth
e
r.
Thi
s is do
ne by t
he
pro
ces
s
of
Im
age
reg
ist
rati
on.
1.1.
Back
gr
ound
Durin
g
Im
age
reg
ist
rati
on,
i
m
ages
are
m
a
pp
e
d
to
a
com
m
on
co
-
ordi
na
te
syst
e
m
wh
ic
h
helps
to
do
analy
sis
on
the
giv
en
set
of
im
ages
thereb
y
beco
m
ing
an
i
m
po
rtant
proce
ss
in
Im
age
pr
ocessin
g
an
d
a
naly
sis.
The
im
ages
c
aptu
red
are
di
ff
e
ren
t
from
on
e
a
no
t
her
be
cause
of
the
f
ollow
i
ng
rea
sons
(i)
ca
pture
d
from
diff
e
re
nt
Vie
wpoints
an
d
Dep
t
h
(ii)
ca
ptured
at
dif
f
eren
t
ti
m
ing
s
(iii
)
us
i
ng
dif
fer
e
nt
m
od
es
[1
]
,
[2]
.
The
im
ages
are
captu
red
fro
m
diff
ere
nt
Vi
ewpoint
in
cas
e
of
a
Sate
ll
it
e
,
the
im
ages
that
are
sent
are
by
the
sat
el
li
te
are
fr
om
diff
eren
t
View
po
i
nt
as
they
are
m
ov
in
g
co
ntin
uous
ly
;
If
we
are
m
on
it
ori
ng
a
pa
rtic
ular
obj
ect
or
a
sce
ne
the
n
the
im
ages
that
are
got
are
at
dif
fere
nt
tim
ing
s;
In
m
edical
i
m
ag
es,
there
a
re
diff
e
rent
m
od
es in whic
h
the
im
ages can
be
ca
ptured
,
f
or e
xam
ple the b
rai
n
im
ages can
be
go
t
fro
m
CT scan
or
MR
I
to
analy
ze
the
con
diti
on.
T
he
ab
ov
e
ar
e
the
scenari
os
w
her
e
we
w
ou
l
d
requ
ire
Im
age
reg
ist
rati
on
to
be
an
al
yz
ed
befor
e
w
e
m
ake
certai
n
co
ncl
us
io
ns
.
T
he
m
ajor
a
reas
w
he
re
Im
age
reg
is
trat
ion
is
use
d
are
rem
ote
sensing
,
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A Revi
ew
on R
egistrati
on
of
Medical
Ima
ge
s U
si
ng G
r
aph
Th
e
or
et
ic
Appr
oa
c
hes
(
Aksh
ay
a
R
)
975
weathe
r
m
on
it
or
i
ng,
cha
nge
detect
ion,
En
vi
ronm
ent
m
on
it
or
i
ng,
Me
dical
app
li
cat
io
ns
e
tc
.,
I
n
this
paper
our
m
ajo
r
conce
ntr
at
ion
would be
on th
e
Medica
l im
age r
egistr
at
ion
[3
]
,
[
4].
Adva
ncem
ents
in
m
edical
i
maging
m
akes
it
i
m
po
rtant
not
on
ly
in
diag
no
sis
of
diseases
bu
t
al
s
o
in
areas
of
plan
nin
g,
exec
utin
g
and
a
ssessin
g
a
su
r
ge
ry.
The
re
are
wide
a
ppli
cat
ion
s
of
I
m
age
analy
sis
ta
sk
s
in
the
fiel
d
of
m
e
dicines
li
ke
Re
ti
nal
fun
du
s
de
te
ct
ion
,
T
um
or
detect
ion,
et
c.
,
T
he
var
io
us
m
od
es
that
an
i
m
age
can
be
obta
ine
d
is
Com
pu
te
d
Tom
og
ra
phy,
Ma
gn
et
ic
Re
sonance
Im
aging,
Posi
tro
n
Em
i
ssion
Tom
ogra
ph
y
,
Sing
le
-
P
ho
t
on
Em
issi
on
Com
pu
te
d
T
om
og
r
aph
y,
Functi
on
al
Ma
gn
et
ic
Re
sonance
Im
aging
,
IVUS
im
a
ges
or
a
com
bin
at
ion
o
f
t
hese
s
ource
s
w
hich
is
cal
l
ed
a
m
ultim
od
al
i
m
ages
et
c.,
All
these
im
ages
are
us
e
d
to c
aptu
re
var
i
ou
s
par
ts
of
bo
dy
li
ke
the
Brai
n,
Re
ti
na,
L
ungs
,
Bl
a
dd
e
r,
Breast
,
Liver,
S
pin
e
e
tc
.,
Ther
e
a
re
a
lot
of
researc
h
t
hat is
h
a
pp
e
ni
ng i
n
t
he Im
age r
egis
trat
ion
[5]
-
[
8].
2.
RESEA
R
CH MET
HO
D
The
proc
ess
of
al
ign
in
g
tw
o
or
m
or
e
i
m
a
ges
is
cal
le
d
I
m
age
reg
ist
rat
ion
.
On
e
of
t
he
i
m
age
is
consi
der
e
d
a
s
t
he
ref
e
ren
ce
im
age
an
d
al
l
the
oth
e
r
ca
ptu
r
ed
im
ages
are
al
ign
ed
with
re
sp
ect
to
it
.
W
e
deci
de
on
a
c
omm
on
co
-
ordinate
sys
tem
and
al
l
the
i
m
ages
are
arra
ng
e
d
in
t
hat
sp
ace.
T
her
e
a
r
e
nu
m
ero
us
wa
ys
we
can
cl
as
sify
the
Re
gistrati
on
proce
ss.
T
he
y
are
(i)
bas
ed
on
dim
ension
s
(ii)
base
d
on
the
natu
re
of
trans
form
ation
(iii
)
base
d
on
the
opti
m
iz
at
i
on
proce
dures
(iv)
base
d
on
t
he
m
od
es
(
v)
base
d
on
the
s
ubj
ect
(v
i)
ba
sed
on
t
he
obj
ect
a
nd
m
any
m
or
e.
B
as
ed
on
dim
en
sion
s
,
they
c
an
be
cl
assi
fied
a
s
2D
-
2D
Re
gistrati
on
,
2D
-
3D
Re
gistr
at
ion
or
3D
-
3D
Re
gistrati
on
al
l
these
are
done
in
the
spa
ti
al
do
m
ai
n
wh
ereas
im
ages
can
be
reg
ist
ere
d
i
n
t
i
m
e
series
al
so
.
Ba
se
d
on
t
he
natu
re
of
trans
form
ation
we
c
an
cl
ass
ify
th
em
as
Ri
gid
,
Non
-
Ri
gid
,
Proj
ect
iv
e,
Affin
e,
Curve
d.
Sim
il
arly
,
in
case
of
opti
m
izati
on
te
ch
niqu
es
there
a
re
va
rio
us
m
et
ho
ds
that
c
an
be
broa
dly
pu
t
un
der
cat
e
gories
cal
le
d
the
Param
et
er
com
pu
te
d
or
th
e
Param
et
er
Se
arch
e
d.
Mod
e
of
the
im
age
cou
ld
be
ei
ther
a
sing
le
m
od
e
or
m
ultim
od
e
wh
ic
h
m
os
tl
y
co
m
es
under
m
edical
i
mages
wh
e
re
we
us
e
i
m
ages
from
diff
ere
nt
m
od
es
to
predict
or
fin
d
so
m
et
hin
g.
Objects
ca
n
be
va
stl
y
cl
as
sifie
d,
it
is
ac
tuall
y
t
he
obj
e
ct
that
is
being
ca
ptured
,
as
we
are
con
ce
ntrati
ng
on
m
edic
al
i
m
age
reg
ist
rati
on
the
obj
ect
s
t
hat
co
m
e
under
this
are
Head
(
br
ai
n,
reti
na,
de
nta
l,
et
c.,),
T
hora
x
(breast,
ca
r
di
ac,
et
c.,),
Abd
om
en
(liver,
ki
dn
ey
, etc.,)
, S
pi
ne
a
nd
Lim
bs
.
The
m
ajo
r
c
oncent
r
at
ion
o
f
th
e
pa
per
is Gr
a
ph
-
ba
sed
m
et
ho
ds
t
o
s
olv
e
i
m
age
reg
ist
ra
ti
on
an
d
the
pa
per
m
os
tl
y
d
eal
s
with
Ri
gid
an
d
N
on
-
ri
gi
d
reg
ist
rati
on.
Re
gistrati
on
is
an
i
m
po
rtant
ste
p
no
t o
nly
in
m
e
dical
i
m
age
an
al
ysi
s
bu
t
al
so i
n
f
us
io
n,
i
n
s
uper r
es
olu
ti
on, s
ha
pe
analy
sis,
point
cl
oud
re
gistrati
on,
bi
om
et
rics
and
i
n
3D
rec
onst
r
uction
[
9]
-
[
14
]
.
Gr
a
ph
m
e
thod
is
prefe
rred
in
case
of
i
m
age
reg
ist
rati
on
be
cause
of
the
f
act
that
e
ven
if
the
im
age
is
r
otate
d
or
tra
ns
la
te
d
t
he
no
de
i
n
the
gr
a
ph
of
a
n
i
m
age
are
c
onnected
i
n
t
he
s
a
m
e
way.
Lez
or
ay
et
al
.
has
exp
la
ine
d
t
he
con
ce
pts
of
gr
ap
hs
a
nd
ho
w
gr
a
phs
us
e
d
in
im
ages
[15
]
.
2.1.
Rigid
Re
gistr
at
i
on
Ri
gid
i
m
age
R
egistrat
ion
is
ve
ry
si
m
ple
as
t
he
cha
ng
es
in
the
i
m
ages
ob
ta
ined
is
on
ly
translat
io
n,
ro
ta
ti
on,
s
hea
r,
zo
om
et
c.
Its
only
m
app
ing
of
tw
o
or
m
or
e
im
ages.
Its
assum
ed
that
on
e
im
a
ge
rem
ai
ns
sta
ti
on
ary
w
he
reas
th
e
oth
e
r
i
m
ages
are
sp
at
ia
ll
y
transf
orm
ed.
A
m
app
ing
has
to
be
de
fi
ned
to
tran
sf
orm
the
ta
rg
et
i
m
age
to
m
at
ch
the
ref
e
ren
ce
im
age.
The
pa
ram
et
ers
of
the
tra
nsfo
r
m
are
est
i
m
a
ted
an
d
the
im
ages
are
trans
form
ed
usi
ng
these.
I
n
t
his
sect
io
n,
the
resea
rc
h
work
wit
h
re
s
pect
to
rigi
d
reg
ist
rati
on
ha
s
bee
n
discusse
d.
On
e
of
the
old
est
m
et
ho
d
use
d
f
or
im
age
reg
ist
rati
on
w
as
pro
posed
by
Jasiobe
dzk
i
Piotr
wh
e
r
e
al
ign
m
ent
of
im
ages
us
in
g
t
he
ada
ptive
a
dj
ace
ncy
gra
phs
is
do
ne.
T
he
te
ch
niqu
e
us
e
d
in
this
pa
per
is
descr
i
bing
the
reti
nal
vasc
ular
netw
ork
that
is
detect
ed
fro
m
the
ref
eren
c
e
i
m
age
to
al
i
gn
it
with
the
oth
e
rs.
So
m
e
of
t
he
c
on
t
ro
l
points
use
d
a
re
t
he
Bi
furcati
ons,
a
rte
rio
veno
us
c
ros
sing
s
an
d
ves
s
el
s.
T
he
se
gme
nted
ref
e
ren
ce
im
a
ge
is
the
de
note
d
as
an
A
da
ptive
a
dj
ace
nc
y
gr
a
ph
or
th
e
AAG.
The
AAG
c
on
ta
in
s
act
ive
con
t
ours
t
hat
a
re
c
onnected
in
the
net
wor
k.
Re
gistrati
on
of
i
m
ages
is
do
ne
by
placi
ng
th
e
ad
j
ace
ncy
gr
aph
of
the
ref
e
ren
c
e
im
age
on
the
ot
her
an
d
al
lo
w
ing
the
data
to
adap
t
t
o
the
pr
e
vious
on
e
[
16
]
.
Lat
er
Sabunc
u
Me
rt
R.,
and
P
et
er
Ram
adg
e
pro
vid
e
d
a
detai
le
d
analy
sis
on
the
us
e
of
m
ini
m
u
m
sp
ann
in
g
trees
as
a
too
l
f
or
al
ign
m
ent
of
i
m
ages
go
t
fro
m
m
ul
tim
od
al
so
urces.
It
use
s
a
gr
ap
h
the
or
et
ic
m
e
tho
d
that
si
m
ul
ta
neo
usl
y
est
i
m
at
es
the
al
ign
m
ent
and
t
he
desce
nt
dir
ect
ion
with
th
e
help
of
tra
nsfo
rm
s.
The
al
ign
m
ent
is
m
ea
su
r
e
d
us
in
g
the
m
ini
m
u
m
entro
py
gr
a
phs
an
d
a
de
scent
base
d
optim
iz
at
ion
.
The
al
ign
m
ent
m
easur
e
is
co
m
par
ed
with
the
plugi
n
ent
ropy
est
im
at
or
s
w
hich
giv
es
us
in
f
orm
at
ion
on
the
how
the
data
i
s
co
ns
ide
red
le
adin
g
to
the
re
gistrati
on
of
the
im
age.
The
al
gorith
m
is
com
par
ed
with
t
he
oth
e
r
by
sim
ulati
ng
it
into
a
3D
m
od
el
.
In this m
et
ho
d, pri
or in
form
ation
is
co
m
bin
e
d
to
avoid
pr
oblem
s that o
ccu
r
at the init
ia
l st
ages
[17].
Weibel
T
hom
a
s
et
al
.
s
uggest
ed
a
m
et
ho
d
that
can
re
duce
the
c
om
pu
ta
tio
nal
ti
m
e
that
a
grap
h
-
c
ut
requires
.
T
he
node
s
f
ro
m
the
blad
der
im
age,
that
are
in
t
he
gr
a
ph
are
bro
ught
dow
n
by
com
pu
ti
ng
t
he
s
patia
l
pro
per
ti
es
of
a
n
im
age,
wh
ic
h
dec
reases
t
he
am
ou
nt
of
in
form
ation
that
is
lost.
The
re
du
ct
io
n
of
t
hes
e
node
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
974
–
983
976
do
e
s
no
t
a
ff
ect
the
qual
it
y
of
the
im
age
reg
i
strat
ion
an
d
t
he
bla
dd
e
r
wall
s
are
c
om
pu
te
d
us
in
g
t
he
pano
ram
ic
m
aps.
T
he
e
dg
e
val
ue
of
the
gr
a
ph
is
fou
nd
with
the
help
of
a
waters
hed
trans
f
or
m
.
Th
e
eval
uation
of
t
he
al
gorithm
is
do
ne
us
i
ng
t
he
R
egistrat
ion
e
rro
r,
E
uclidean
pi
xel
de
viati
on
(
m
ean
dev
ia
ti
on
bet
ween
t
he
so
urce
and
ta
r
get
im
age)
[
18]
.
F
urt
her
,
De
ng
Ke
xin
et
al
.
i
n
t
heir
pa
per
s
uggested
an
al
gorithm
cal
le
d
Gr
a
ph
m
at
ching
-
Itera
ti
ve
cl
os
est
point
to
al
ig
n
the
i
m
ages
of
reti
na
pair
wi
se.
The
pro
ble
m
of
reti
na
l
i
m
age
reg
ist
rati
on
is
handled
her
e
us
in
g
G
raph
Ma
tc
hin
g.
In
t
his
m
e
tho
d,
th
e
featur
es
e
xtr
act
ed
are
the
reti
nal
vasc
ular
feat
ures.
A
n
undirec
te
d
grap
h
is
c
onstr
ucted
a
nd
t
he
norm
al
iz
ed
path
is
cal
c
ulate
d
f
or
m
at
ching
th
e
edg
e
s
in
t
he
gr
a
ph.
It
inte
gr
at
es
t
he
G
M
and
the
IC
P
al
gorithm
to
fi
nd
out
a
trans
form
ation
m
od
el
.
The
S
TRUCT
SA
C
al
go
rith
m
is
app
li
ed
to
it
to
rem
ov
e
the
false
m
a
tc
hin
g
i
n
the
res
ult
obta
ined
.
The
e
valuati
on
of
the
m
et
ho
d
is
ba
sed
on
how
sal
ie
nt
th
e
vasc
ular
gr
a
ph
is
an
d
the
ov
erall
w
or
king
of
the
al
gorithm
.
Th
e
vasc
ular
gra
ph
is
sal
ie
nt
base
d
on
the
fo
ll
owin
g
c
rite
rio
ns
li
ke
the
aver
a
ge
of
c
orrectl
y
m
at
ched
no
des
,
recall
rate
a
nd
the
su
cces
s
of
the
reg
ist
rati
on.
T
he
par
am
et
ers
us
e
d
for
evaluati
ng
the
m
et
ho
d
are
the
norm
al
iz
ed
Co
rr
el
at
io
n
C
oeffici
ent
(
NCC),
norm
al
i
zed
M
utu
al
Inf
or
m
at
ion
(
NM
I)
,
ve
ssel
Ce
nt
erli
ne
Error
Me
as
ur
e
(CEM)
[
19
]
.
Lom
baer
t,
H.,
&
Cheriet
,
F
s
how
that
the
grap
h
-
c
uts,
wh
i
ch
m
os
tl
y
has
bee
n
us
e
d
to
ta
ckle
de
-
noisi
ng
a
nd
fin
ding
co
rr
es
pondence
s
se
pa
ratel
y,
can
be
us
e
d
s
im
ultan
eousl
y.
T
his
m
et
ho
d
can
be
a
dopted
w
hen
only
on
e
cl
ear
im
age
is
present
an
d
t
he
oth
e
rs
a
re
a
ll
degrad
e
d.
Th
e
pe
rfor
m
ance
of
t
he
m
et
ho
d
is
m
e
asur
e
d
us
i
ng
t
he
RM
SE
value
an
d
wh
e
n
it
is
do
ne
to
ge
ther
giv
i
ng
a
bette
r
per
f
or
m
ance.
The
m
et
ho
d i
s
adv
a
ntag
e
ous
wh
e
n we
get a
seq
uen
ce
of c
orr
up
te
d
im
ages [20
]
.
Lu
paşcu
Ca
rm
en
Alina
et
al
.
proposes
a
se
m
i
-
autom
atic
m
et
ho
d
f
or
re
gistrati
on
of
re
ti
nal
i
m
ages
with
e
xtracti
on
of
li
ne
featu
r
es
co
nnect
ing
the
point
li
ke
bif
ur
cat
io
n
points,
br
a
nc
hing’
s,
cr
os
s
ove
r
poi
nts,
endp
oin
ts.
A
li
ne
on
the
gr
a
ph
in
dicat
es
that
it
is
a
con
nec
ti
on
in
the
vas
cular
se
gm
ent
in
the
reti
nal
im
age.
The
la
ndm
ark
s
are
extracte
d
m
anu
al
ly
to
av
oid
se
gm
entat
i
on
e
rrors
that
can
caus
e
loss
of
data.
A
str
ai
gh
t
-
li
ne
m
od
el
is
com
pu
te
d
to
m
easur
e
the
s
i
m
i
la
rity
in
the
m
at
ched
li
ne
s.
The
t
ran
s
f
or
m
at
ion
func
ti
on
is
cal
culat
ed
f
r
om
the
res
ults
that
we
re
obta
ined
du
rin
g
m
a
tc
hin
g.
Pe
rfo
r
m
ance
of
the
r
egistrat
ion
pro
cess
is
com
pu
te
d
fro
m
the
values
of
cum
ulati
ve
inv
e
rse
co
ns
ist
ency
error
(C
I
CE)
[21].
Par
i
so
t
et
al
pr
opos
es
a
sp
ars
e m
et
ho
d t
o
segm
ent an
d register t
he
br
ai
n
tum
or
b
y
m
od
el
li
ng
it
as a p
ai
r
wise d
is
crete M
arko
v R
ando
m
Fiel
d
m
od
el
m
ai
nly
fo
r
ide
ntific
at
ion
of
Gliom
as,
a
ty
pe
of
tum
or
.
A
gr
a
ph
ic
al
m
od
el
i
s
us
ed
to
overl
ay
the
sp
ars
e
gri
d
on
to
t
he
im
age
do
m
ai
n
wh
e
r
e
it
s
cl
assifi
e
d
an
d
dis
plac
ed
base
d
on
t
he
im
age
si
m
i
la
rity
.
Con
st
raints
are
add
e
d
to
c
hec
k
the
sm
oo
th
ne
ss
of
de
form
a
ti
on
fiel
d.
Th
e
m
ai
n
dr
a
wb
ac
k
of
disc
rete
m
et
ho
d
is
the
tra
deoff
betwee
n
t
he
pr
eci
sio
n
a
nd
the
com
plexity
wh
ic
h
is
ha
n
dle
d
by
ad
di
ng
unce
rtai
nty
in
thi
s
m
et
ho
d.
The
re
finem
ent
of
th
e
sp
ars
e
gri
d
re
du
ce
s
the
c
omplexit
y
and
re
m
ov
al
of
inact
ive
points
helps
in
the
reducti
on
of
r
untim
e.
The
re
gi
strat
ion
of
the
se
im
ages
was
evalu
at
ed
bas
ed
on
the
Dice
sco
re,
false
posit
ive
and
the t
ru
e
po
sit
ive r
at
es,
M
AD
was
calc
ul
at
ed
bet
wee
n
t
he
s
ource a
nd t
he float
in
g
im
a
ge [2
2].
Pinh
ei
ro
Mi
guel
Et
al
.
pr
es
en
ts
a
te
chn
i
qu
e
wh
e
re
the
cal
culat
ion
of
pat
h
desc
riptors
in
geo
m
et
rical
gr
a
phs
are
done
to
fin
d
the
si
m
il
arity
between
them
.
It
m
a
inly
aim
s
in
creati
ng
a
m
et
ho
d
that
can
be
a
pp
li
e
d
wh
e
re
sta
ndar
d
m
et
ho
ds
ca
nnot
be
a
pp
li
ed
t
hat
is
wh
e
re
th
e
i
m
age
app
ea
r
ance
is
dif
fer
e
nt,
uniq
ue
key
po
i
nts
cannot b
e i
de
ntifie
d
or the
re is
n
ot m
uch
of
te
xture. T
he
m
ain
aim
is
to calc
ulate
the p
at
h descri
pto
rs
to
m
at
ch
two
grap
hs
.
T
he
par
am
et
ers
us
e
d
for
e
val
uating
the
m
et
hod
with
t
he
sta
te
of
a
rt
m
et
hods
,
are
Eu
cl
idian
distance
bet
we
en
the
re
gister
ed
grap
hs
a
nd
their
proces
sin
g
tim
e
[2
3].
Me
non
Hem
a
P.,
Et
al
.
il
lustrate
s
a
m
et
ho
d
f
or
sel
ect
ing
the
poin
ts
fo
r
fin
ding
corres
pondence
s
based
on
the
structu
ral
inf
orm
at
ion
of
the
i
m
ag
e
as
it
re
m
ai
ns
t
he
sam
e
even
wh
e
n
the
il
lu
m
inati
on
in
th
e
i
m
age
chan
ge
s.
The
m
et
ho
d
pro
posed
is
cal
le
d
Ver
te
x
Degree
Neighb
orhoo
d
(V
D
N).
He
re
the
init
ia
l
m
at
c
h
is
fo
un
d
us
in
g
the
de
g
ree
a
nd
the
ne
ig
hbors
of
the
node
.
The
n
it
s
i
m
pr
ove
d
by
feat
ur
e
m
at
ching
usi
ng
featu
res
li
ke
the
entr
op
y,
m
ean
and
st
and
a
r
d
dev
ia
ti
on.
T
he
init
ia
l
po
ints
wer
e
go
t
us
in
g
the
Harris
corner
detect
or
and
gr
a
ph
is
bu
il
t
us
i
ng
De
la
un
ay
tria
ngulati
on.
The
par
am
et
er
of
c
om
par
iso
n
is
Tar
get
Re
gis
trat
ion
E
rro
r
[
24]
.
Z
hong
Yang
-
jun
an
d
La
n
-
Zhe
n
Chen
,
Et
al
.
pr
esents
the
re
gi
strat
ion
of
m
a
m
m
og
ram
i
m
a
ges
us
in
g
SIF
T
an
d
grap
h
tr
ansfo
rm
ation
.
Feat
ure
po
i
nts
are
de
rived
from
the
i
m
age
and
GT
M
is
m
ai
nly
do
ne
to
obta
in
accur
at
e
res
ults
as
it
m
ini
m
i
zes
the
inco
rr
ect
m
at
c
hes.
T
he
re
gistrati
on
is
done
us
in
g
the
Thi
n
-
plate
sp
li
ne
(
TPS)
inte
rpola
ti
on
.
T
he
evalu
at
ion
is
done
base
d
on
the
c
orrelat
ion
coe
ff
ic
ie
nt
(CC),
t
he
s
um
of
sq
ua
re
d
dif
fer
e
nces
(
SSD)
a
nd
the
visu
al
eff
ect
[25].
Cz
ajko
wsk
a
Joa
nn
a
Et
al
.
pr
opose
d
a
m
et
ho
d
f
or
re
gistrati
on
t
hat
use
s p
a
th
sim
il
arit
y
sk
el
et
on
g
ra
ph
m
at
ching
al
on
g
with
weig
ht
cl
iqu
es
to
ac
hieve
m
at
ching
.
This
m
et
ho
d
al
lows
us
t
o
m
a
tc
h
the
voxels
from
diff
e
re
nt
CT
series.
T
he
m
ai
n
ap
plica
ti
on
of
t
his
m
eth
od
ca
n
be
f
ound
in
detec
ti
ng
a
bdom
inal
aor
ti
c
aneurysm
by
visu
al
iz
ing
the
pr
e
-
an
d
post
-
op
e
rati
ve
CT
da
ta
.
The
m
et
ho
d
com
par
es
tw
o
al
gorithm
s
app
li
e
d
for
3
-
D
reg
ist
rati
on
that
does
a
se
gm
enta
ti
on
of
t
he
a
ort
a
an
d
inte
rnal
orga
n
usi
ng
gr
a
ph
ap
pro
aches
.
The
a
naly
sis
was
car
ried
out
on
real
CT
A
data
of
t
he
patie
nts
ha
vi
ng
a
bdom
inal
aor
ti
c
ane
ur
ys
m
[2
6].
C.
Leng
et
al
.
pro
po
ses
a
sp
a
rse
base
d
m
a
trix
facto
rizat
ion
m
et
ho
d
cal
le
d
the
gr
a
ph
regul
arized
no
n
-
negat
ive
m
at
rix
factor
iz
at
ion
.
Using
graph
re
gu
la
riza
t
ion
a
nd
t
otal
va
riat
ion
will
help
us
pr
e
ser
ve
the
intrinsic
de
ta
il
s
of
the
im
age
and
the
sm
al
l
detai
ls
resp
ect
ive
ly
.
The
featu
re
s
are
extracte
d
us
in
g
the
ext
ra
ct
ion
al
gorith
m
s
li
k
e
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A Revi
ew
on R
egistrati
on
of
Medical
Ima
ge
s U
si
ng G
r
aph
Th
e
or
et
ic
Appr
oa
c
hes
(
Aksh
ay
a
R
)
977
the
Harris
co
rner
detect
or.
Th
e
per
f
orm
ance
of
the
al
go
rith
m
is
co
m
par
ed
w
it
h
oth
e
r
m
e
thods
li
ke
the
Zass’s
and
t
he
Ca
ei
ll
’s
m
e
tho
d
w
here
the
com
pu
ta
ti
on
ti
m
e
and
accuracy
of
th
e
al
gorithm
is
b
et
te
r
the
oth
e
r
two.
The
discrim
ina
ti
ng
abili
ty
of
t
he
al
gorithm
is
cal
culat
ed
us
i
ng
t
he
r
oot
m
e
an
squa
re
er
ror
value
[27].
Ch
en
Li
Et
al
.
pr
op
os
e
d
the
m
et
ho
d
wh
e
re
the
us
e
of
netw
ork
str
uctu
re
an
d
ci
rc
uit
si
m
ulati
on
is
done
to
re
gister
the
giv
e
n
im
ages.
Vessel
im
ages
we
re
tra
nsfo
r
m
ed
into
gra
phs
a
nd
th
ei
r
s
egm
entat
ion
is
done
to
re
du
ce
the
com
plexity
of
the
cal
culat
ion.
W
ei
ghte
d
gra
ph
s
we
r
e
co
nv
erted
int
o
ci
rcui
ts
and
re
gistra
ti
on
was
done
base
d
on
t
he
node
volt
ages.
It
ef
fici
ently
handles
the
pr
ob
le
m
of
m
at
ching
la
rg
e
nu
m
ber
of
vasc
ular
i
m
ages
.
First
the
sal
ie
nt
po
i
nts
are
i
de
ntifie
d
a
nd
a
ci
rcu
it
co
nversi
on
m
od
el
is
pro
po
s
ed
w
her
e
t
he
s
p
at
ia
l
diff
e
ren
c
e
is
ref
le
ct
ed
as
the
no
de
vo
lt
a
ge.
Netw
ork
de
com
po
sit
ion
i
s
done
usi
ng
t
he
NS
I
m
et
ho
d
a
nd
the
bran
chin
g
crit
eria
is
us
e
d
to
deal
with
l
arg
e
scal
e
net
works
[
28]
.
Pa
rek
a
r
et
al
.
propose
d
a
m
et
ho
d
f
or
reti
nal
im
age
reg
ist
rati
on
by
m
at
ching
the
bif
ur
cat
io
ns
in
the
im
age.
A
fu
ll
y
co
nn
ect
e
d
va
scular
net
work
is
got
wi
th
th
e
help
of
Di
j
kst
ra’
s
al
gorit
hm
.
The
i
m
ages
are
subj
e
ct
ed
to
prep
ro
c
essin
g
ste
ps
as
fals
e
m
at
ches
are
to
be
rem
ov
ed.
T
he
pro
po
se
d
m
et
h
od
is
in
var
ia
nt
to
translat
ion,
ro
ta
ti
on,
scal
ing
as
on
ly
nor
m
al
iz
ed
feature
s
are
us
e
d.
Order
sta
ti
sti
c
filt
ers
are
use
d
f
or
processin
g
i
niti
al
ly
,
fo
ll
ow
e
d
by
m
ulti
-
scal
e
rid
ge
detect
ion
a
nd
finall
y
Dijkstr
a’s
for
j
oi
ning
the
bro
ken
ed
ges.
Pr
e
-
proce
ssing
ste
ps
co
ns
ist
of
noise
r
e
m
ov
al
,
non
-
m
axim
a
l
su
p
pressi
on
et
c.
The
pa
ram
eter
s
f
or
eval
uation
are
m
ean
a
nd
sta
ndar
d
de
viati
on
of
the
r
egistrat
ion
er
ror
[
29]
.
Zha
ng
et
al
.
proposes
a
m
et
ho
d
that
m
akes
us
e
of
Delaunay
Tr
ia
ngulati
on.
The
im
ages
us
ed
f
or
exp
e
rim
entat
io
n
a
re
m
ul
ti
sp
ect
ral
an
d
pa
noram
ic
i
m
ages
the
featu
re
points
a
re
e
xtracted
usi
ng
SIFT
al
gorithm
and
the
Delau
nay
Tria
ngulati
on
to
rem
ov
e
th
e
false
m
at
ches.
Co
rresp
onde
nces
a
re
e
sta
blish
e
d
us
in
g
the
Delaun
ay
tria
ngul
at
ion
s.
T
he
A
lgorit
hm
re
m
o
ves
the
false
m
at
ches
and
has
go
od
sp
ee
d
a
nd
accuracy
[30
].
2.2.
N
on
-
Rigi
d Im
ag
e
R
e
gis
tration
Our
body
is
not
al
ways
rigid
,
there
are
pa
rts
in
our
body
th
at
do
not
rem
a
i
n
sta
ti
on
ary
sa
y
ou
r
hea
rt
,
defor
m
at
ion
s
i
n
our
face
et
c.,
Re
gistrati
on
of
su
c
h
im
ages
are
done
usi
ng
non
-
ri
gid
i
m
age
reg
ist
r
at
ion.
The
tra
ns
f
orm
at
ion
f
unct
ions
us
e
d
he
re
are
non
-
li
near
li
ke
the
sp
li
ne
,
el
a
sti
c
m
od
el
et
c.
[31].
T
he
al
go
rithm
s
that
help
us
i
n
reg
ist
erin
g
s
uc
h
im
ages
are
discusse
d.
O
ne
of
the
old
est
al
gorithm
fo
und
in
no
n
-
rigid
i
m
age
reg
ist
rati
on
wa
s
by
Fischler
a
nd
Elsc
hlag
e
r
to
re
gister
the
de
form
ation
in
f
ace
[32].
Furth
er,
Ta
ng
T
omm
y
W
H
et
al
.
pro
po
ses
a
m
et
ho
d
wh
e
re
t
he
non
-
rigi
d
im
age
r
egistrat
ion
is
consi
der
e
d
as
a
discrete
la
be
ll
ing
pro
blem
.
Pixels
in
t
he
s
ource
i
m
age
is
assig
ned
wit
h
a
dis
placem
ent
vector
that
de
no
t
es
it
s
sp
at
ia
l
po
si
ti
on
i
n
the
fl
oating
im
age.
T
he
optim
iz
at
ion
of
t
his
m
et
ho
d
use
s
the
gr
a
ph
-
c
uts
m
et
ho
d
via
al
ph
a
ex
pa
nsi
on
s
.
The
s
harp
cha
ng
e
s
in
the
di
sp
la
cem
ent
la
bels
is
pen
al
iz
ed
us
i
ng
a
fi
rst
der
ivati
ve
sm
oo
t
hn
e
ss
co
ns
t
raint.
Using
of
gr
a
ph
-
c
uts
e
nsure
the
fin
ding
of
the
m
ini
m
a.
Th
e
m
et
ho
d
is
ac
cur
at
e
a
nd
rob
us
t
wh
e
n
c
ompar
e
d
with
the
rest
of
the
sta
te
of
a
rt
m
et
ho
ds.
Th
e
par
am
et
er
that
is
us
ed
to
as
sess
the
pe
rform
ance
is
the
ov
erla
p
m
easur
e
an
d
w
as
fou
nd
t
o
bet
te
r
than
t
he
FF
D
an
d
dem
on
s
al
gorith
m
s
[3
3]
.
Fu
rt
her
,
R
on
al
d
WK
,
a
nd
A
lbert
CS
Chun
g
pro
po
s
ed
a
m
et
hod
to
s
olv
e
the
pro
blem
of
non
-
ri
gid
reg
ist
rati
on
us
i
ng
graph
-
c
uts.
T
his
pap
e
r
handles
a
m
ult
i
-
le
vel
non
-
rigi
d
re
gistrati
on
,
wh
e
re
the
de
f
orm
ation
fiel
d
of
one
le
vel
is
passed
on
to
the
nex
t
.
Using
this
ap
proac
h,
the
nu
m
ber
of
le
vels
in
the
prob
le
m
is
gr
eat
ly
reduce
d
there
by
the
s
peed
of
r
egistr
at
ion
is
increased.
T
he
perf
or
m
ance
is
evaluated
base
d
on
the
c
om
pu
ta
ti
on
al
tim
e
and
is
co
m
par
e
d
with
D
E
MON
S
and
t
he
FF
D
a
lgorit
hm
s
and
the
sing
le
le
ve
l
reg
ist
rati
on
[
34
]
.
R.
W.
K.
So
a
nd
A.
C.
S.
Ch
ung
propose
d
an
i
m
pr
ovem
ent
to
t
he
le
ar
ning
-
base
d
sim
il
arity
m
easur
e
f
or
non
-
rigi
d
im
age
re
gistrati
on
usi
ng
the
KL
D
wh
ic
h
us
es
the
pri
or
know
le
dg
e
of
the
j
oin
t
dist
rib
ution
of
the
intensit
ie
s
of
the
pr
e
-
al
ig
ne
d
i
m
ages.
Th
e
new
form
ulati
on
is
the
use
of
Ma
r
kov
ra
ndom
fiel
d
(MRF
)
with
KL
D
a
nd
gr
aph
-
c
uts.
T
he
valuati
on
is
ba
sed
on
su
m
of
abs
olu
t
e
diff
e
re
nce
a
nd
m
utu
al
inf
orm
ation
.
It
is
com
par
ed
with
Fr
ee
f
or
m
defor
m
at
ion
(F
F
D
)
an
d
DEMO
NS al
gorithm
[
35]
.
Ron
al
d W
K
et
al. p
rop
os
es a g
ra
ph
-
c
uts b
as
ed
m
e
tho
d
f
or
reg
ist
rati
on of n
on
-
rigi
d
i
m
ages w
he
re th
e
reg
ist
rati
on
is
reg
a
rd
e
d
as
a
discrete
la
belli
ng
prob
le
m
.
The
de
form
ation
m
od
el
us
ed
in
this
case
is
flexible
and
non
-
pa
ram
et
ric.
I
n
t
he
re
gistrati
on
proc
ess,
eac
h
pi
xel
is
al
locat
ed
th
ree
la
bels
an
d
this
the
dis
plac
e
m
ent
vecto
r
that
s
pe
ci
fies
the
posit
ion
i
n
the
fl
oa
ti
ng
im
age.
A
first
order
bas
ed
sm
oo
th
ness
base
d
c
onstra
int
is
us
e
d
to
rem
ove
the
sh
a
rp
c
ha
ng
e
s
in
the
a
dj
ace
nt
dis
plac
e
m
ent
la
bels.
The
op
ti
m
iz
a
tio
n
of
t
he
re
gis
trat
ion
process
is
done
us
in
g
gr
a
ph
-
cuts
m
et
ho
d
w
hich
us
es
(alp
ha
)
-
e
xp
a
ns
i
on
t
o
reac
h
a
l
ocal
or
a
global
m
ini
m
u
m
.
The
al
gorithm
is
com
par
ed
w
it
h
the
existi
ng
m
et
ho
ds
li
ke
the
DEM
ONS
base
d
m
e
tho
d,
li
near
pro
gr
am
m
ing
m
et
ho
d,
f
ree
-
f
or
m
defor
m
at
i
on
ba
sed
m
et
h
od.
T
he
m
et
hod
is
e
valuate
d
us
i
ng
pa
ram
e
te
rs
li
ke
t
he
overla
p
m
easur
e,
distri
bu
ti
on
of
abs
ol
ute
intensit
y
error
s
.
T
he
al
gorithm
is
robust
against
al
l
c
halle
ng
e
s
li
ke
huge
defor
m
at
ion
,
ripp
le
disto
rtion
et
c.,
[
36]
.
Ch
owd
hury,
An
a
nd
a
S
Et
al
.
ha
ve
im
pr
oved
the
pe
rfo
rm
ance
of
a
non
-
rigi
d
im
age
re
gistrati
on
with
us
a
ge
of
data
te
rm
.
Thi
s
can
be
helpful
in
com
pu
ta
ti
on
al
ne
uroan
a
tom
y
wh
e
re
we
ca
n
detect
the
c
onditi
on
s
li
ke
sc
hizo
phre
nia,
re
gi
strat
ion
of
re
ti
nal
i
m
ages
et
c.
The
re
a
re
se
ver
al
adv
a
ntage
s
in
t
his
m
et
ho
d
li
ke
the
dis
placem
ent
te
r
m
s
can
b
e
a
ssig
ned
from
the
data
te
rm
.
It
act
s
as
a
stric
t
pen
al
ty
on
m
i
sm
a
tc
hes
and
the
data
te
rm
can
handle
the
dissim
i
la
rit
y
in
the
intensit
y
patte
rn
s
due
to
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
974
–
983
978
il
lu
m
inati
on
ch
ang
e
s
in
the
im
ages.
The
w
ork
is
done
on
t
he
MR
I
i
m
ages
of
the
brai
n
a
nd
the
r
et
inal
im
ages.
The
res
ults
are
com
par
ed
to
that
of
the
dem
on
s
al
gorithm
us
ing
the
abso
lute
inte
ns
it
y
diff
ere
nc
es
for
reg
ist
rati
on
ac
cur
acy
[
37]
.
A.
Szm
ul,
B
Et
al
.
propose
s
a
m
e
tho
d
t
ha
t
us
es
gra
ph
-
cuts
as
an
e
ffi
ci
ent
opt
im
iz
ation
too
l.
Her
e
the
i
m
age
is
represente
d
as
a
gr
ap
h
co
ns
tr
ucted
f
ro
m
the
sp
arse
s
up
e
r
-
voxels
com
bin
ed
with
gr
ap
h
-
cuts.
Us
ing
the
sli
ding
m
ot
ion
m
od
el
l
ing
,
a
relaxe
d
gr
a
ph
is
con
str
ucted
w
hich
he
lps
in
est
i
m
ation
of
de
form
ation
.
Re
gistrati
on
is
ac
hieve
d
us
i
ng
i
m
age
-
guide
d
fi
lt
er
in
the
sp
ar
se
defor
m
at
ion
fiel
d.
The
e
xp
e
rim
ents
are
do
ne
on
CT
lung
dataset
that
is
received
from
a
sing
le
m
od
e
an
d
it
s
intra
-
patie
nt.
It
is
m
ai
nly app
li
ed
in
ra
dio t
he
rapy
p
la
nnin
g,
m
on
it
or
i
ng of trea
t
m
ent, an
d ve
nt
il
at
ion
q
ua
ntif
ic
at
ion
. T
he
m
et
hod
was
e
valuated
base
d
on
th
e
ta
rg
et
re
gis
trat
ion
e
rror
wh
ic
h
is
f
ou
nd
t
o
be
bett
er
tha
n
the
dem
on
s
al
gorithm
[3
8].
Ser
radel
l
et
al
.
pro
po
se
s
a
m
et
ho
d
t
hat
does
no
t
rely
only
on
local
fe
at
ur
e
sim
il
arity.
This
m
et
ho
d
does
no
t
re
qu
i
re
i
ni
ti
al
m
at
ches
and
ca
n
do
s
om
e
par
ti
al
m
a
tc
hin
g.
N
on
-
li
near
de
f
or
m
at
i
on
s
ar
e
handled
with
the
help
of
Ga
ussi
an
P
ro
c
ess.
Eve
n
in
case
t
he
in
f
or
m
at
ion
is
m
issi
ng
the
corres
ponde
nc
es
ar
e
m
at
ched
us
in
g
an
it
erati
ve
process
.
T
he
exp
e
rim
ent
is
co
nducted
on
bo
t
h
sy
nth
e
ti
c
i
m
ages
an
d
the
ang
i
ogra
ph
y
i
m
ages.
T
he
num
ber
of
c
orr
esp
onde
nces
i
s
re
du
ce
d
i
n
t
his
m
et
ho
d
wi
thout
the
us
e
of
any
defor
m
at
ion
m
od
el
.
Fi
rst
a
coar
se
al
ign
m
ent
is
do
ne
f
ollo
w
ing
w
hich
a
fi
ner
m
et
ho
d
is
us
e
d
[39].
Che
n
et
al
pro
po
ses
a
m
eth
od
t
hat
i
s
a
c
om
bin
at
ion
of
non
-
rigi
d
im
age
re
gistrati
on
a
nd
sem
i
-
automa
ti
c
segm
entat
i
on
of
li
ver
that
is
use
d
f
or
a
gu
i
de
d
surge
ry.
T
his
m
e
tho
d
pu
ts
k
-
m
eans
cl
us
te
rin
g
a
nd
grap
h
cuts
to
ge
ther
to
dev
el
op
a
rob
us
t
al
gorithm
.
A
f
ree
form
def
orm
ation
is
app
li
ed
to
r
eg
ist
er
these
im
ages.
Ga
us
sia
n
m
ixtur
e
m
od
el
li
ng
is
us
ed
f
or
i
ntensity
distribu
ti
on
of
the
li
ve
r
im
ages.
T
he
sta
ti
sti
cal
par
am
e
ter
s
from
the
i
m
age
is
cal
culat
ed
us
in
g
the
pri
or
knowle
dge
avail
ab
le
.
Thr
es
holdin
g
is
app
li
ed
to
fin
d
out
the
cand
i
date
po
i
nts
fro
m
the
li
ver
an
d
t
he
n
k
-
m
eans
al
gorithm
is
us
e
d
to
fi
nd
out
the
non
-
li
ver
pi
xels.
T
he
see
ds
ar
e
obta
ine
d
f
r
om
the
pr
e
vious
ste
p
and
t
he
segm
entat
ion
is
base
d
on
gr
a
ph
-
c
uts
m
et
ho
d.
The
s
egm
entat
ion
and
t
he
re
gistrat
ion
ar
e
com
bin
ed
to
ge
ther
w
hich
is
us
e
d
dur
i
ng
in
vasive
s
urge
ries.
They
use
bo
t
h
CT
an
d
MR
i
m
ages
fo
r
th
e
process
.
The
pa
ram
et
ers
that
are
us
e
d
f
or
validat
io
n
of
the
al
gorithm
i
s
aver
a
ge
re
gi
strat
ion
e
rror
and
t
he
processi
ng
sp
e
ed
[40].
El
-
Ba
z
et
al
pro
pose
s
a
m
et
ho
d
f
or
reg
ist
rati
on
of
Lu
ng
no
du
le
s.
This
m
et
ho
d
he
lps
us
identify
lung
nodule
s
that
m
ay
la
te
r
le
ad
to
cancer,
thi
s
helps
us
in
early
detect
ion.
W
e
m
on
it
or
the
lung
nodule
s
in a
s
uc
cessi
ve
m
anner
f
ro
m
the
CT
scan of
the
pat
ie
nt.
T
he
proce
ss
sta
rts w
it
h
e
xtracti
on o
f
t
he
lung
reg
i
on
s
by
m
od
el
li
ng
it
s
gray
le
vel
intensit
ie
s,
f
ollow
e
d
by
cl
assifi
cat
ion
of
par
ts
of
the
lung,
a
pp
ly
in
g
rigid
reg
ist
rati
on
a
nd
the
c
om
pen
s
at
ing
f
or
the
he
art
beats
a
nd
br
eat
hing.
T
he
process
us
e
s
the
norm
al
iz
ed
m
utu
al
inf
or
m
at
ion
in t
he
fi
nal steps
. T
o
c
hec
k wh
et
her
the
nodes
have
gro
wn or
not re
-
sli
ci
ng th
e i
m
age [41
]
.
2.3
.
Gr
oup
-
Wise
Im
ag
e
Registr
at
i
on
Group
-
wise
im
age
Re
gistra
ti
on
m
os
tl
y
aim
s
in
reg
ist
eri
ng
a
set
of
im
ages
i
nto
a
co
m
m
on
sp
ac
e
instea
d
of
re
gi
ste
ring
it
to
a
pa
rtic
ular
im
age
sepa
ratel
y.
Re
search
h
a
s
bee
n
done
i
n
t
his
area
a
nd
the
ou
tc
om
e
are
the
f
ollowi
ng
.
S
.
Yi
ng,
G.
Et
al
.
pro
poses
a
m
et
ho
d
fo
r
gro
up
regi
strat
ion
of
th
e
diff
e
ren
t
gro
up
s
of
i
m
ages.
T
he
m
et
hod
us
es
a
t
wo
-
le
vel
hiera
rch
ic
al
gr
a
ph
t
o
m
od
el
the
prob
le
m
by
us
i
ng
int
ra
gr
a
ph
f
or
the
i
m
a
ges
in
a
gr
oup
an
d
inter
gr
a
ph
for
i
m
ages
belonging
to
dif
fer
e
nt
gro
up
s
.
The
e
valu
at
ion
of
the
al
gorithm
is
done
by
cal
culat
ing
t
he
di
ce
rati
o.
T
he
r
at
io
achieve
d
by
pe
rfor
m
ing
this
on
the
im
ages
is
bette
r
than
t
he
existi
ng
m
et
ho
ds
.
The
al
gorit
hm
pr
eser
ves
t
he
to
polo
gy
of
the
im
ages
durin
g
the
proces
s
of
re
gistrati
on
[
42]
.
S.
Yi
ng
Et
al
.
inv
est
igate
s
a
m
et
ho
d
wh
e
re
the
gro
up
im
ages
are
reg
ist
e
red
with
ou
t
th
e
us
e
of
the
re
fer
e
nce
i
m
age
avo
idi
ng
the
bias
in
t
he
re
gistrati
on
process.
T
he
i
m
ages
are
m
o
delle
d
us
i
ng
graphs
w
her
e
th
e
nodes
are
the
i
m
ages
and
the
ed
ges
are
the
geodesi
c
path
betwee
n
the
m
.
Re
gistrati
on
is
achieve
d
us
in
g
the
dy
nam
i
c
sh
ri
nk
i
ng
pr
oblem
un
ti
l
the
nodes
are
cl
ose
to
eac
h
oth
e
r.
T
he
t
opol
ogy
of
the
im
age
is
preser
ve
d
in
th
e
process
.
All
th
e
im
ages
are
warpe
d
sim
ult
aneously
to
wa
rd
s
the
cente
r
el
i
m
inati
ng
t
he
possibil
it
y
of
bias.
The
pa
ram
et
er
fo
r
asse
ssin
g
it
is
Dice
rati
o
[43].
Tan
g
Z
hen
y
u
Et
al
.
pr
esents
a
te
ch
nique
that
accuratel
y
reg
ist
ers
the
si
m
il
ar
i
m
ages
t
o
the
gro
up
center
us
i
ng
the
sh
ort
est
path
.
The
sim
il
arit
y
m
easur
e
is
bu
i
lt
on
sp
ars
e
directed
grap
hs
(
digra
ph),
w
her
e
the
node
s
a
re
c
onnecte
d
by
s
hortest
distance
s.
T
he
reg
ist
ra
ti
on
is
at
ta
ined
by
it
e
rati
ve
m
e
tho
d
wh
e
re
the
la
rge
def
orm
at
ion
fiel
d
is
sp
li
t
i
nto
series
of
s
m
al
le
r
on
e
s.
Re
su
lt
s
su
ggest
that
t
he
m
et
ho
d
is
rob
us
t
an
d
ac
hieves
bette
r
accuracy
wh
e
n
com
par
e
d
w
it
h
existi
ng
m
et
hods
.
The per
f
or
m
ance is ev
al
uated by t
he o
ver
la
p regi
on m
easur
ed usin
g
Jacca
r
d
in
de
x [44].
Gr
a
ph
-
base
d
m
et
ho
ds
are no
t
on
ly
u
sed
f
or m
edical
i
m
ages,
they
a
re
al
so
us
e
d
in
va
rio
us
oth
e
r
a
reas
and
a
fe
w
of
them
are
discuss
e
d
belo
w
.
M.
Iza
di
a
nd
P.
Saeedi,
pr
e
sent
a
n
a
lgorit
hm
wh
er
e
the
corres
pondenc
es
are
m
at
ched
accu
ratel
y.
The
feat
ur
e
po
ints
extracte
d
are
us
e
d
to
fin
d
out
the
geom
et
rical
relat
ion
s
hip
i
n
the
gr
a
ph.
It
st
arts
m
a
tc
hin
g
the
points
in
th
e
i
m
age
and
gr
aph
is
create
d
for
the
point
a
nd
k
-
near
est
ne
ig
hbor
s
.
The
a
ngul
ar
distances
are
us
e
d
to
fi
nd
the
m
at
ching
po
i
nts
in
the
seco
nd
i
m
age.
The
a
dv
a
ntage
of
t
his
m
e
thod
is
it
s
capab
il
it
y
to
han
dle
ou
tl
ie
rs.
T
he
ev
al
uation
par
am
et
ers
are
Accuracy
,
Pr
eci
sio
n,
Re
cal
l,
an
d
S
pecif
ic
it
y
[4
5].
Len
g
Et
al
.
propo
ses
a
c
om
pu
ta
ti
on
te
c
hniq
ue
that
is
ba
sed
on
t
he
pro
j
ect
ion
the
or
em
of
energ
y
con
ser
vatio
n
to
i
m
pr
ov
e
th
e
i
m
age
reg
ist
rati
on
perform
ance.
First
the
in
te
r
-
gr
a
ph
is
c
on
st
ru
ct
e
d
ba
sed
on
the
pro
xim
i
t
y
and
from
that
and
intra
-
gr
aph
us
i
ng
the
node
set
s
.
T
he
n
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A Revi
ew
on R
egistrati
on
of
Medical
Ima
ge
s U
si
ng G
r
aph
Th
e
or
et
ic
Appr
oa
c
hes
(
Aksh
ay
a
R
)
979
pro
j
ect
ion
the
orem
is
app
li
ed
to
pr
oj
ect
that
data
int
o
a
l
ower
dim
ension
s
pace.
A
naly
ti
cal
m
e
tho
ds
are
use
d
to
fi
nd
the
co
r
respo
nd
e
nces
in
th
e
im
ag
e
w
hich
is
ver
y
a
ccur
at
e
e
ve
n
wh
e
n
t
her
e
ar
e
lot
of
no
des
in
it
.
The
acc
uracy
of
this
m
et
ho
d
is
com
par
e
d
with
the
sta
te
of
art
m
et
ho
ds
[
46
]
.
Zh
u
Et
a
l.
presents
a
m
et
hod
t
o
reg
ist
er
the
c
ol
or
im
age
and
dep
th
i
nfor
m
at
ion
.
T
he
co
r
respo
nd
e
nces
betwee
n
the
c
o
lo
r
i
m
age
an
d
de
pth
inf
or
m
at
ion
is
fou
nd
us
in
g
a
con
t
our
base
d
appr
oach.
The
corres
pondenc
es
on
the
c
on
t
ours
will
hav
e
s
i
m
i
la
r
sh
a
pe
descr
i
pto
rs
an
d
t
his
he
lps
us
us
e
M
unkres
al
go
rithm
to
so
lve
the
bip
a
rtit
e
graph.
T
he
sim
ilarity
is
com
pu
te
d
base
d
on
the
c
hi
-
s
quare
m
et
ho
d
a
nd
the
tra
ns
f
or
m
m
od
el
is
esti
m
at
ed
us
ing
the
RAN
SA
C
m
et
ho
d.
The
m
et
ho
d i
s
com
par
ed
a
gai
ns
t exist
in
g
m
et
hods
li
ke
COP
AP
,
A
C
O
a
nd
Hun
gar
ia
n wit
h param
et
ers
li
ke
the
run
ti
m
e and
th
e RM
SE
value
s [47
]
.
Ngo
Et
al
.
pr
opos
es
an
al
gorithm
wh
e
re
the
pa
ram
et
er
sp
ace
of
the
rigid
tra
ns
f
orm
is
local
ly
com
pu
te
d
us
in
g
the
gr
a
dient
decen
t
par
a
digm
s.
The
par
a
m
et
er
sp
ace
ha
s
three
dim
ension
s
a
nd
the
ov
e
rall
sp
ace
is
sub
div
ide
d
as
a
sub
set
of
R3
hav
i
ng
a
po
ly
nom
i
al
com
plexity
.
Her
e
the
gr
a
ph
is
integrated
with
the
desce
nt
sche
m
e
that
will
le
ad
to
the
m
ini
m
u
m
w
hich
s
olv
es
t
he
pro
blem
of
im
age
regi
strat
ion
.
The
opti
m
iz
at
i
on
process
us
e
s
a
m
ult
i
-
scal
e
app
r
oac
h
to
sp
eed
it
up.
Th
e
par
am
et
ers
us
ed
f
or
validat
i
on
are
the
r
un
tim
e
com
plexity
,
distance
bas
ed
on
s
ig
ne
d
f
unct
ion
a
nd
the
a
ve
rag
e
de
gr
ee
of
DRT
gr
a
ph
s
[
48
]
.
Zak
harov
Et
al
.
co
ns
ide
rs
the
pro
blem
of
point
m
at
ching
a
nd
m
ai
n
obj
ect
ive
is
to
el
i
m
inate
the
fals
e
corres
pondenc
es.
Fir
stl
y,
Scot
t
and
Lo
nguet
-
Hi
gg
i
ns
al
go
r
it
h
m
is
us
e
d
a
nd
a
gr
a
ph
is
c
on
st
ru
ct
e
d
bas
ed
th
e
featur
e
s
e
xtrac
te
d
f
ro
m
these
i
m
ages.
Si
ngular
value
dec
om
po
sit
ion
is
done
a
nd
t
he
corres
pondenc
es
are
fou
nd
usi
ng
the
desc
riptor
SU
RF.
T
his
al
gorithm
helps
us
to
find
th
e
sign
ific
ant
f
al
se
cor
res
pondence
s
.
The
s
pee
d
of
the
al
gorith
m
is
al
so
hig
h.
It’s
m
ai
nly
us
e
d
i
n
th
re
e
-
dim
ension
al
rec
on
st
ru
ct
io
n
ta
s
ks
.
The
al
gorithm
is
com
par
ed
w
it
h
RA
NSAC
[
49
]
.
Sa
nrom
à
Et
al
.
proposes
a
m
et
ho
d
f
or
fin
ding
the
po
i
nt
-
set
corres
pondenc
es
us
in
g
the
gr
aph
m
at
ching
m
et
ho
d
wh
ic
h
is
m
od
el
ed
as
a
m
ixtur
e
m
od
el
li
ng
pro
blem
wh
ic
h
accounts
f
or
bo
t
h
struct
ur
a
l
and
the
ge
om
et
rical
par
ts.
Gr
a
ph
m
atch
in
g
is
appr
ox
im
at
ed
us
in
g
the
Ex
pec
ta
ti
on
-
m
axim
iz
at
ion
al
gorithm
.
The
c
orres
pondin
g
c
on
ti
nu
ous
va
ri
ables
are
so
l
ve
d
us
in
g
S
oft
-
a
ssign,
i
m
po
sin
g
a
co
ns
trai
nt
a
nd
al
so
co
ntr
ols
the
le
vel
of
disc
re
ti
zat
ion
.
Ef
fecti
ve
m
echan
is
m
s
are
us
e
d
dete
ct
and
rem
ov
e
outl
ie
r
s.
T
he
e
valuati
on
of
t
he
al
gor
it
h
m
is
done
usi
ng
re
gistrati
on
acc
ur
acy
a
nd
recog
niti
on
abili
ty
[50].
Zh
an
g
Et
al
.
proposes
a
gr
a
ph
sp
ect
r
al
m
et
ho
d
that
use
s
isom
eric
pr
oject
io
ns
.
T
hi
s
is
m
ai
nly
do
ne
to
so
lve
the
pro
bl
e
m
of
m
at
ching
t
he
featur
e
points
us
i
ng
the
struc
tural
inf
orm
ation
that
is
got
f
r
om
the
gr
a
ph.
Using
this
pro
je
ct
ion
,
the
gra
ph
s
a
re
pro
j
ect
ed
to
a
m
at
rix
that
has
the
no
rm
alized
value
s.
The
pro
j
ect
i
on
of
data
is
un
iq
ue
and
th
us
can
be
us
e
d
for
com
pu
ti
ng
the
cor
re
spo
nd
e
nc
es.
A
Laplaci
an
m
a
trix
is
us
ed
to
pr
ese
r
ve
the
i
nfo
rm
at
ion
that
is
relat
ed
to
the
po
sit
io
n
of
a
point.
T
he
grap
h
m
at
ching
pro
blem
is
han
dle
d
us
in
g
the
str
uc
tural
an
d
posit
ion
al
in
form
ati
on.
The
res
ults
go
t
s
how
tha
t
the
al
go
rit
hm
is
accurate
an
d
it
s
rob
us
t
[
51]
.
G
ul
et
al
.
pro
po
ses
a
m
et
ho
d
f
or
the
re
gistrati
o
n
of
br
ai
n
i
m
ages
w
he
re
f
eat
ur
es
are
ext
racted
us
in
g
S
URF
a
nd
the
m
at
ching
is
do
ne
us
ing
Eucli
dean
m
easur
e.
The
false
m
at
ches
that
are
pr
ese
nt
in
th
e
i
m
age
is
rem
ov
e
d
us
i
ng
th
e
R
-
RA
NSAC
al
gorithm
al
on
g
with
the
S
PRT
to
le
sse
n
the
runtim
e.
Final
reg
ist
rati
o
n
of
the
i
m
age
is
do
ne
us
in
g
the
l
east
sq
ua
res
m
et
hod.
T
he
al
gorithm
is
ro
bus
t
and
ef
fici
ent
du
e
t
o
the co
m
bin
at
io
n of
SU
R
F a
nd R
-
RA
NSAC.
The
al
gorithm
is fast as
well
a
s accu
rate w
he
n
c
om
par
ed wi
th the
oth
e
r
m
et
ho
ds
[52].
Z
ha
ng
et
al
.
pro
poses
a
m
et
ho
d
of
re
gistrati
on
that
use
s
wa
velet
tran
sform
.
The
reg
ist
r
at
io
n
ta
ke
s
place
as
tw
o
ste
ps
one
the
c
oar
se
reg
ist
rati
on
a
nd
the
ot
her
is
the
fi
ne
reg
ist
r
at
ion
.
The param
et
ers
of
reg
ist
rati
on are g
ot through
t
he
ant c
olony searc
h
m
et
h
od. T
he
al
gorithm
is ro
bu
st
t
o no
ise
,
has hig
h
acc
uracy
an
d fast i
n com
pu
ta
ti
on
[53].
3.
RESU
LT
S
A
ND AN
ALYSIS
This
sect
io
n
gi
ves
a
su
m
m
a
ry
of
t
he
res
ul
ts
ob
ta
ine
d
by
var
io
us
rese
arch
e
rs
f
or
re
gistrati
on
of
f
m
edical
i
m
age
s
us
in
g
gr
a
ph
base
d
ap
proac
hes.
Ba
se
d
on
the
va
rio
us
pa
ram
et
ers
that
hav
e
bee
n
use
d
f
or
assessi
ng the
quali
ty
o
f
t
he out
pu
t, t
he
m
et
hods
a
nd sam
ple o
ut
pu
ts
are sh
own
in Ta
ble
1.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
974
–
983
980
Table
1.
Su
m
m
ary o
f
Re
gistra
ti
on
on m
edical
i
m
ages
Metho
d
Used
Ou
tp
u
t
Para
m
eters
u
se
d
f
o
r
ev
alu
atio
n
Ref
er
en
ces
Grou
p
-
wise
I
m
ag
e
Reg
istratio
n
Dice ratio
[
4
0
]
Rig
id
I
m
ag
e
Reg
istratio
n
Ro
o
t M
ean Sq
u
are
[
2
0
]
No
n
-
Rig
id
I
m
ag
e
Reg
istratio
n
(
m
ean
±
stan
d
ard
d
ev
iatio
n
.)
o
f
th
e
ab
so
lu
te
in
ten
sity
d
if
f
erences
[
3
5
]
No
n
-
Medical
I
m
ag
e
Reg
istratio
n
Accurac
y
,
P
recisio
n
,
Recall, Specif
icit
y
[
4
4
]
3.1.
O
bser
vation
a
n
d I
nf
er
ences
a)
Gr
a
ph
based m
et
hods
a
re m
ain
ly
b
ase
d o
n pixels
.
b)
Ti
m
e
ta
ken
for
the
process
is
directl
y
proport
ion
al
to
the
num
ber
of
node
s
or
e
dg
es
pr
es
ent
in
the
gra
ph
i.e., as
the
node
s incr
ea
se the
tim
e taken
al
so i
ncr
ease
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A Revi
ew
on R
egistrati
on
of
Medical
Ima
ge
s U
si
ng G
r
aph
Th
e
or
et
ic
Appr
oa
c
hes
(
Aksh
ay
a
R
)
981
c)
The val
ues
of t
he
e
dg
e
s ar
e
b
a
sed o
n
the
inte
ns
it
y or t
he dis
ta
nce b
et
wee
n t
he
pi
xels.
d)
Sele
ct
ion
of po
ints as
nodes
is a v
e
ry cr
ucial
ste
p
in
grap
h b
ased m
et
ho
ds
.
e)
The n
um
ber
of
points i
n bo
t
h of t
he
im
ages should
b
e
the
sa
m
e and
it
s
houl
d be
ho
m
olo
go
us
.
f)
Hen
ce
, c
orres
ponde
nce m
at
ching
betwee
n
t
he
se points
has
al
so
gaine
d
im
portance
.
4.
CONCL
US
I
O
N
A
lot
of
Re
se
arch
ha
s
bee
n
done
in
the
area
of
im
age
reg
ist
rati
on
and
the
it
is
st
il
l
go
ing
on.
It
bec
om
es
very
i
m
po
rtant
t
o
hav
e
a
good
re
gistrati
on
al
gor
it
h
m
as
it
is
th
e
first
ste
p
t
o
a
naly
zi
ng
so
m
eth
in
g.
Ri
gid
Re
gistra
ti
on
has
quit
e
good
am
ount
of
resea
rc
h
as
it
is
easy
to
form
ulate
wh
en
com
par
ed
t
o
t
he
non
-
rigid
re
gistrati
on.
And
m
os
t
work
ta
ke
place
in
a
reas
li
ke
brai
n
an
d
reti
na
as
the
data
is
avail
able,
but
th
e
pro
blem
with
this
is
that
real
tum
or
-
patie
nts
data
is
reall
y
diff
ic
ult
to
get.
Hen
ce
m
os
t
of
the
w
orks
ta
ke
place
us
in
g
te
m
plates
w
hich
is
m
or
e
or
le
ss
pe
rfec
t
and
does
no
t
ha
ve
a
ny
err
or
li
ke
er
r
or
that
occ
ur
duri
ng
captu
rin
g
of
pa
ti
ent’s
data.
Non
-
rigi
d
im
a
ge
re
gistrati
on
on
t
he
ot
her
hand
is
sti
ll
chall
eng
i
ng
due
to
the
sm
oo
thn
ess r
e
qu
i
rem
ent
and
the
huge
de
gr
e
e
of
f
ree
dom
.
This
pa
pe
r
m
ain
ly
aim
s
at
intro
duci
ng
m
et
hods
that
are
dev
el
o
pe
d
t
o
s
olv
e
im
age r
egistrat
ion
pro
blem
u
sing
gr
a
ph app
ro
ac
hes.
REFERE
NCE
S
[1]
Zi
tova
B,
Flus
s
er
J.
Im
age
reg
i
strat
ion
m
et
hods
:
a
surve
y
.
Ima
ge
and
v
ision
c
omputing
.
2003
Oct
31;
21(11)
:
977
-
1000.
[2]
Brown
LG.
A
surve
y
of
imag
e
r
egi
stra
ti
on
t
ec
hn
ique
s.
ACM
c
om
puti
ng
sur
ve
ys
(
CS
UR).
1992
Dec
1;
24(4):
325
-
76.
[3]
Grane
ro
MA
,
Baroni
MP
,
Costa
ET
,
Rad
eva
PI.
An
alt
ernative
technique
for
Imaging
Re
gistrati
o
n
in
IVUS
image
s.
In
Com
puta
ti
on
al
Sci
enc
e
and
Com
puta
ti
onal
I
nte
lligen
ce
(CS
CI),
2015
Int
ern
at
ion
al
Confer
en
ce
on
2015
De
c
7
(pp.
464
-
469)
IE
EE
.
[4]
W
y
awa
har
e
MV
,
Pat
il
PM
,
Abhy
ank
ar
HK
.
Im
a
ge
reg
istra
t
ion
t
ec
hniqu
es:
an
o
ver
vie
w.
Int
ernati
onal
Journal
o
f
Signal
Proce
ss
in
g,
Image
Proc
essing
and
Pa
tt
ern
Recogni
t
ion
.
20
09
Sep;2(3):
11
-
2
8.
[5]
Hill
DL,
Batchelor
PG
,
Holden
M,
Hawkes
DJ
.
Medic
a
l
image
reg
istration.
Ph
ysic
s
in
medic
in
e
and
biol
ogy
.
200
1
Mar;46(
3):
R1
.
[6]
Ruec
ker
t,
Dani
el,
and
Julia
A.
Schnabe
l
.
Medical
image
reg
istratio
n.
In
Biom
edi
c
al Im
age
Proce
ss
i
ng
,
pp.
131
-
154
.
Springer
Ber
li
n
Heide
lb
erg
,
201
0.
[7]
Z
hang
S,
Zhi
L.
Mu
lt
i
-
modal
i
ty
image
regist
ration
wit
h
gra
die
nt
ori
ent
at
io
n
inf
orm
ati
on
b
ased
on
ent
ropi
c
spanning
graph
.
In
Inform
at
ion
Scie
nc
e
and
Te
c
hnolog
y
(ICIST)
,
2017
Sevent
h
Inte
rna
ti
ona
l
Confer
ence
on
2017
Apr 16
(pp. 470
-
473).
IE
EE.
[8]
Mani
VR.
Surve
y
of
m
edi
c
al
i
m
age
reg
istr
at
io
n.
Journal
o
f
Biom
edi
cal
Engi
n
ee
ring
and
Te
ch
nology
.
2013
Ja
n
23;1(2):
8
-
25.
[9]
Menon
HP
,
Nar
a
y
ana
nku
tty
KA
.
Com
par
ative
p
erf
orm
anc
e
of
di
ffe
ren
t
per
ce
ptu
a
l
contrast
fusion
te
chn
ique
s
using
MLS.
Inte
rnat
io
nal
Journal
of
Biom
edi
cal
Engi
n
e
ering
and
Te
chn
ology
.
2015;18(
1):52
-
71.
[10]
R.
Shw
et
ha
and
Raj
at
h
il
ag
am
B.
Super
resolu
ti
on
of
m
amm
o
gra
m
s
for
bre
ast
ca
n
ce
r
de
tection.
Inte
rnat
iona
l
Journal
of
Appli
ed
Eng
ine
ering
Re
search
.
2015;
10(1):21453
-
214
65.
[11]
Ze
ng
W
,
Yang
YJ
,
Raz
ib
M.
G
raph
-
Constr
aine
d
Surface
R
egi
st
ration
Based
on
Tutte
Embe
ddin
g
.
In
Proce
ed
ing
s
of
the IEEE
Con
fer
ence
on
Com
pute
r
Vision
and
Pattern
R
ec
ogni
ti
on
W
orkshops
2016
(pp. 76
-
83)
.
[12]
Huang
X,
Zha
ng
J,
Fan
L,
W
u
Q,
Yuan
C.
A
S
yste
mati
c
Approach
for
Cross
-
s
ource
Poi
nt
Clo
ud
Re
gistrati
on
by
Prese
rving
Ma
cro
and
Mi
cro
Str
uct
ures
.
I
EEE
Tr
ansa
ctions on
Im
age
Proc
essing.
2017
Apr 19.
[13]
Ong
EP,
Xu
Y,
W
ong
DW
,
Li
u
J.
Ret
in
a
ver
i
fic
a
ti
on
using
a
combined
poin
ts
and
edge
s
ap
proa
ch.
In
Im
age
Proce
ss
ing
(ICIP),
2015
IEEE
In
t
ern
ational Con
fe
ren
ce on
2015
Sep
27
(pp
.
2
720
-
2724).
IE
EE.
[14]
Arath
i.
T.
and
La
th
a
Para
m
es
wara
n.
Im
age
R
ec
onstruc
ti
on
fr
om
2D
stac
k
of
MRI/CT
to
3D
using
Shapel
e
t
s.
Inte
rnational
Jo
urnal
of Engi
ne
e
ring a
nd
Techno
logy
(IJE
T). 201
4;6(1):
2595
–
26
03.
[15]
Lé
zor
a
y
O
,
Grad
y
L
.
Grap
h
the
or
y
conc
ep
ts
a
nd
def
ini
t
ions
used
in
image
proc
essing
and
a
naly
s
is.
Im
age
proc
essing
and
a
naly
s
is wi
th
gra
phs:
the
or
y
and
pra
ctice.
CRC
Press
,
Boca Raton,
FL.
2012
:1
-
24
.
[16]
Jasiobedz
ki
P.
Re
gistrati
on
of
reti
nal
images
using
adapti
ve
adjac
en
cy
graphs
.
In
Com
pute
r
-
Based
Medic
a
l
S
y
stems
,
1993.
Proce
edi
ngs of
S
ixt
h
Annua
l
I
EEE
S
y
m
posium
o
n
1993
Jun 13
(p
p.
40
-
45)
.
I
EE
E
.
[17]
Sabuncu
MR,
R
amadge
P.
Us
in
g
spanning
graphs
for
ef
fici
ent
i
mage
registrati
o
n
.
IEEE
Tr
ansa
ct
ions
on
Im
age
Proce
ss
ing.
200
8
Ma
y
;17(5):
788
-
97.
[18]
W
ei
bel
T
,
Daul
C,
W
olf
D,
Rösch
R,
Ben
-
Ham
ad
ou
A.
Endoscopic
bladde
r
image
registrati
on
using
spar
se
graph
cut
s
.
In
Im
age Proce
ss
ing
(ICIP)
,
2010
17th
I
EEE
Int
ern
ationa
l Confere
nc
e
on
2
010
Sep
26
(pp.
157
-
160).
I
EE
E
.
[19]
Deng
K,
T
ia
n
J,
Zhe
ng
J,
Zh
ang
X,
Dai
X,
Xu
M.
Retinal
fund
us
image
reg
istr
at
ion
v
ia
vasc
ul
ar
struc
ture
gra
p
h
m
at
chi
ng.
Journ
al
of
Bi
omed
ic
a
l
Imaging
.
2010
J
an
1;2010
:14.
[20]
Lombae
rt
H,
Cheri
e
t
F.
Simult
aneous
image
de
-
noising
and
registrati
on
using
gra
ph
cut
s:
Appl
ic
a
ti
on
to
corrupted
medic
al
im
ages
.
In
Inform
at
ion
S
ci
en
ce,
Signa
l
Pr
oce
ss
ing
and
their
Applicat
ions
(
ISS
P
A),
2012
1
1th
Internat
iona
l
Confer
ence
on
2
012
Jul 2 (pp.
26
4
-
268).
I
EE
E
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
974
–
983
982
[21]
Lupa
şcu
CA,
T
egol
o
D,
Bellav
ia
F,
Vale
n
ti
C
.
Semi
-
automatic
registrati
on
o
f
reti
nal
images
based
on
li
ne
m
atc
hing
appro
ach.
In
Com
put
er
-
Based
Med
ical
S
y
st
ems
(CBMS
),
2013
IEEE
26th
Int
ern
a
tional
S
y
m
posiu
m
on
2013
Jun 20
(pp
.
453
-
456).
IEEE
.
[22]
Parisot
S,
W
el
ls
W
,
Chemouny
S,
Duffau
H,
Pa
rag
ios
N.
Concurr
ent
tumor
segm
ent
at
ion
and
r
egi
stra
ti
on
with
unce
rt
ai
nt
y
-
b
as
e
d
sparse
non
-
un
i
form
gra
phs
.
M
e
dic
al
image
anal
ysis
.
2014
Ma
y
31;18(4):
647
-
59.
[23]
Pinhei
ro
MA
,
K
y
bic
J.
Pa
th
desc
riptors
fo
r
geome
tric
gr
aph
matchi
ng
and
registrati
o
n
.
In
Int
ern
a
ti
o
nal
Confer
ence
Im
a
ge
Anal
y
sis
and Rec
ognition 201
4
Oct
22
(pp
.
3
-
11).
Springe
r, Cham.
[24]
Menon
HP
,
Nara
y
ana
nku
tty
KA
,
Indule
kh
a
TS.
Feat
ure
Poin
t
Se
le
c
ti
on
using
Str
uct
ura
l
Graph
Matc
h
ing
for
MLS
base
d
Im
age Registrat
ion
.
In
te
rn
ati
onal Journal of
Computer
Ap
pli
cations
.
2014
Jan
1;100(4)
.
[25]
Zhong
YJ
,
Chen
LZ
.
A
novel
alg
orithm
based
on
SIFT
and
Gr
ap
h
transfor
mation
for
mam
mogr
a
m
registrati
on
.
I
n
Proce
edi
ngs
of
t
he
2012
In
te
rn
ational
Confer
en
c
e
on
C
y
b
ern
e
ti
cs
and
Info
rm
at
ic
s
2014
(pp.
1897
-
1903).
Springer
,
New York,
NY
.
[26]
Cza
jkowska
J,
Feine
n
C,
Grz
e
gorz
ek
M,
Rasp
e
M,
W
ic
kenhö
fer
R.
Skel
et
o
n
Graph
Matc
hin
g
vs.
Maximum
W
ei
ght
Cli
que
s
aor
ta
reg
ist
rat
ion
t
ec
hniqu
es.
Computerized
Me
dic
a
l
I
maging
and
Gr
aphic
s
.
2015
Dec
31;46
:142
-
5
2.
[27]
Le
ng
C,
Chen
Z
,
Cai
G,
Cheng
I,
Xiong
Z,
Ti
an
J,
Basu
A.
Total
variat
ion
constr
aine
d
graph
regularized
NMF
for
med
ic
al
image
r
egi
stration
.
In
Im
age
,
Video,
an
d
Multi
dimensional
Signal
Proc
e
ss
ing
W
orkshop
(IVM
S
P),
2016
IEE
E
12th
2016
Jul 11
(pp
.
1
-
5)
.
IEE
E
.
[28]
Chen
L,
Li
an
Y
,
Guo
Y,
W
ang
Y,
Hatsuka
m
i
TS,
Pim
ent
el
K,
Bal
u
N,
Yuan
C.
A
vasc
ula
r
i
m
age
reg
istratio
n
m
et
hod
bas
ed
on
net
work str
uct
u
re
and
c
irc
ui
t
si
m
ula
ti
on.
BMC
bioi
nformatic
s
.
2017
Ma
y
2;18(
1):229.
[29]
Pare
kar
J,
Porw
a
l
P,
Kokare
M
.
Aut
omatic
reti
na
l
image
registrat
ion
using
fully
c
onnec
t
ed
vasc
u
l
ar
tree
.
In
Sign
a
l
and
Inform
at
i
on
Proce
ss
ing
(ICo
nSIP
),
Inte
rn
at
io
nal
Conf
ere
n
ce
on
2016
Oct
6
(p
p.
1
-
5)
.
I
EE
E
.
[30]
Yongm
ei
Zha
ng
,
Li
Ma1,
2
,
Rui
Zha
ng.
A
Quick
Im
age
Regi
str
ation
Algorit
hm
B
ase
d
on
Del
aunay
Tri
angu
lation.
TEL
KOMNIKA
,
2013
Februa
r
y
,
11(2):761~773.
[31]
Crum
W
R,
Hart
kens
T,
Hill
DL.
Non
-
rigi
d
image
reg
istrati
on:
the
o
r
y
and
pra
ct
i
ce.
The
Brit
ish
journal
of
radiology
.
2004
Dec
;77(suppl_2)
:S140
-
53.
[32]
Fis
chl
er
MA
,
El
schla
g
er
RA.
The
rep
res
ent
a
tion
and
m
at
chi
n
g
of
pic
tori
a
l
struct
ure
s
.
I
EE
E
Tr
ansacti
ons
on
compute
rs
.
1973
Jan;100(1):
67
-
9
2.
[33]
Ta
ng
TW,
Chu
ng
AC.
Non
-
rig
id
image
r
egi
stration
using
gra
ph
-
cut
s
.
In
Int
er
nat
ion
al
Conf
er
enc
e
on
Medi
cal
Im
age
Com
puti
n
g
and
Com
puter
-
As
sisted
Inte
rve
nti
on
2007
Oct
2
9
(pp. 916
-
924).
Springer,
B
erlin,
Heidelbe
rg
.
[34]
So
RW
,
Chung
AC.
Mult
i
-
l
evel
non
-
rigid
image
registrati
on
using
graph
-
cut
s.
In
Acoustic
s,
Speec
h
and
Signa
l
Proce
ss
ing,
200
9.
ICAS
SP
2009.
IE
EE Int
e
rna
t
io
nal
Conf
ere
n
ce
on
2009
Apr 19
(pp.
397
-
400)
.
I
EE
E
.
[35]
So
RW
,
Chung
AC.
Non
-
rigid
image
registrati
on
by
using
graph
-
cut
s
wit
h
mutual
inf
orm
ati
on
.
In
Im
age
Proce
ss
ing
(ICIP),
2010
17th
I
EEE
Int
ern
ationa
l Confere
nc
e
on
2
010
Sep
26
(pp.
4429
-
4432).
I
EEE.
[36]
So
RW
,
Ta
ng
TW,
Chung
AC.
Non
-
rigi
d
image
reg
istra
t
ion
of
b
rai
n
m
agne
t
ic
re
sonanc
e
images
using
gra
ph
-
cut
s
.
Pat
te
rn
Recogni
t
ion
.
2011
Nov 3
0;44(10):
2450
-
6
7.
[37]
Chowdhur
y
AS
,
Ro
y
R,
Bose
S
K,
Khali
fa
F,
Elnakib
A,
El
-
Ba
z
A.
Non
-
rigid
biomedi
cal
image
registrati
on
using
graph c
uts wi
th a nov
el
da
ta
t
erm
.
In
Biom
edi
c
a
l
Im
agi
ng
(ISBI),
2012
9th
I
EE
E Inte
rna
ti
ona
l
S
y
m
posium
on
201
2
Ma
y
2
(pp
.
446
-
449).
IE
EE.
[38]
Szm
ul
A,
Papie
z
BW
,
Bat
es
R,
Hall
a
ck
A,
Sch
nabe
l
JA
,
Grau
V
.
Gr
aph
Cuts
-
Based
Re
g
istrati
on
Re
vi
si
te
d:
A
Nove
l
Approach
for
Lung
Image
Re
gistrat
ion
Us
ing
Superv
o
xe
ls
and
Image
-
Gui
ded
Filte
ring
.
In
Proce
ed
ings
of
the
I
EE
E
Confer
enc
e
on
Com
put
er
Vision
and
Pa
tt
ern
Rec
ogn
it
io
n
W
orkshops
2016
(pp. 1
52
-
159)
.
[39]
Serra
dell
E,
Glo
wac
ki
P,
K
y
bic
J
,
Moreno
-
Noguer
F,
Fua
P.
Robust
non
-
rigid
registrati
on
of
2D
a
nd
3D
g
raphs
.
I
n
Com
pute
r
Visio
n
and
Pa
ttern
Re
cogni
ti
on
(CVP
R),
2012
IEEE C
onfe
ren
c
e
on
2
012
Jun 16
(pp
.
996
-
1003).
I
EEE.
[40]
Chen
YW
,
Tsubokawa
K,
Foruza
n
AH
,
Morikaw
S,
Kurum
i
Y.
Image
segment
ation and
registrati
on
te
chni
qu
es
for
MR
-
Guided
Liv
er
Cance
r
Surgery
.
In
Mec
h
at
r
onic
s
and
Embedde
d
S
y
st
ems
and
Applicati
on
s
(MESA
),
201
2
IEE
E
/ASM
E
Int
ern
ational Conference on
2012
Jul 8
(pp
.
105
-
108
)
.
IE
EE.
[41]
El
-
Ba
z
A,
Yukse
l
SE,
El
sha
zly
S,
Fara
g
AA
.
N
on
-
rigid
registrati
on
te
chn
ique
s
for
automati
c
fol
low
-
up
of
lun
g
nodule
s
.
In
Int
er
nat
ion
al
Congr
ess Seri
es
2005
Ma
y
31
(Vol
.
1281
,
pp
.
1115
-
1120)
.
E
lsevier.
[42]
Ying
S,
W
u
G,
Li
ao
S,
Shen
D.
Inte
r
-
group
image
registrati
on
by
h
ie
rar
chi
ca
l
graph
shr
ink
ag
e
.
In
Biom
edica
l
Im
agi
ng
(ISBI), 2013 IE
E
E
10
th Inte
rna
ti
ona
l
S
ym
posium
on
2013
Apr 7
(pp
.
103
0
-
1033).
I
EE
E
.
[43]
Ying
S,
W
u
G,
W
ang
Q,
She
n
D.
Gr
oupwise
registrati
on
v
ia
graph
shr
ink
age
on
th
e
ima
ge
manifol
d
.
In
Proce
edi
ngs of
t
he
IE
E
E
Confer
enc
e
on
Com
put
er
Vision
and
Pa
tt
ern
Rec
ogn
it
io
n
2013
(pp
.
232
3
-
2330).
[44]
Ta
ng
Z
,
Jiang
D,
Fan
Y.
Image
registrati
on
base
d
on
dynamic
d
i
rec
te
d
graphs
wi
th
groupwise
im
age
similarit
y
.
I
n
Biom
edi
cal
Im
a
ging
(ISBI),
201
3
IEEE
10
th
Int
e
rna
ti
on
al
S
y
m
po
sium
on
2013
Apr 7
(pp
.
492
-
49
5).
IE
EE.
[45]
Iza
di
M,
Sa
ee
di
P.
Robust
weigh
te
d
gr
aph
tr
ansf
orm
at
ion
m
at
ch
i
ng
for
rig
id
and
nonrigi
d
imag
e
r
egi
stra
ti
on.
IE
E
E
Tr
ansacti
ons on image
proce
ss
in
g
.
2012
Oct;
21(1
0):4369
-
82.
[46]
Le
ng
C,
Xu
W
,
Li
M,
Ross
ol
N,
He
L,
Li
u
D
.
Image
registrati
on
based
on
the
projec
t
ion
th
eore
m
of
ene
rg
y
conse
rvati
on
in
graphs
.
In
Fu
zzy
S
y
st
ems
and
Know
le
dge
Discove
r
y
(FS
KD
),
2011
Ei
ghth
Inte
rna
ti
ona
l
Confer
ence
on
2
011
Jul 26
(Vol.
3,
pp
.
1976
-
198
0).
IE
EE.
[47]
Zhu
Q,
W
en
T,
Xie
Y,
Gu
J,
W
ang
L.
Contour
-
Ba
sed
Im
age
Regi
strat
ion
usi
ng
Bipa
rti
t
e
Gra
ph
Matc
hing
wi
th
Munkres Algori
t
hm
.
Appl
i
ed
Ma
t
hemati
cs
&
Information
Scienc
e
s
.
2014
Jan
1;8(
1):263.
[48]
Ngo
P,
Kenm
ochi
Y,
Sugim
oto
A,
Talbot
H
,
Pa
ss
at
N.
Discr
et
e
rigi
d
r
egi
stra
ti
on
:
A
lo
ca
l
gra
ph
-
sea
rch
appr
oa
ch
.
Di
screte
Appl
i
ed
Mathe
mat
ic
s
.
2
017
Jan
10;216
:
461
-
81.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A Revi
ew
on R
egistrati
on
of
Medical
Ima
ge
s U
si
ng G
r
aph
Th
e
or
et
ic
Appr
oa
c
hes
(
Aksh
ay
a
R
)
983
[49]
Za
khar
ov
AA
,
Tuz
hil
k
in
AY
,
Zhi
zn
y
a
kov
AL.
Finding
Corre
s
pondenc
es
B
et
w
ee
n
Im
age
s
usin
g
Descri
ptors
an
d
Graphs.
Proc
edia E
ngin
ee
ring
.
2
015
Jan
1;129
:3
91
-
6.
[50]
Sanrom
à
G,
Alq
uéz
ar
R,
Serra
to
sa
F.
A
n
ew
gra
ph
m
at
chi
ng
m
et
hod
for
po
int
-
s
et
cor
responde
n
ce
using
the
E
M
al
gorit
hm
and
S
ofta
ss
ign.
Comp
ute
r v
ision
and
i
mage
understan
ding
.
2012
Feb
2
9;116(2):
292
-
30
4.
[51]
Zha
ng
Z
,
T
ia
n
Z.
Image
reg
i
stration
using
i
som
et
ric
proje
ction
of
graph
.
I
n
Inform
at
ion
Engi
ne
eri
ng
an
d
Com
pute
r
Scie
n
ce
,
2009.
ICIEC
S 2009.
Int
ern
a
tional
Conf
ere
n
ce on
2009
Dec
19
(pp. 1
-
4). IE
E
E.
[52]
Zong
y
un
Gu1
,
Li
Cai,
Yunxia
Y
in,
Yata
o
Din
g,
Hongxing
Ka
n.
Regi
str
at
ion
of
Brai
n
Medi
cal
Im
age
s
Based
on
SU
RF
Algorit
hm
and
R
-
RA
NSAC
Algorit
hm
.
TEL
KOMNIKA
Indone
sian
Journal
of
El
e
ct
rica
l
Engi
nee
ring
,
2014
Marc
h
,
12(3):
22
90
~ 2297.
[53]
Dape
ng
Zha
ng
,
Jia
y
an
Li.
Im
age
Regi
stration
Method
Based
on
W
ave
le
t
Tra
nsfo
rm
and
Ant
Colon
y
Optimiz
at
io
n.
TEL
KOMNIKA,
2015
June,
13(2)
:604
~ 613
.
Evaluation Warning : The document was created with Spire.PDF for Python.