Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
,
No.
6
,
D
ece
m
ber
201
8,
pp. 506
1~50
70
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp5061
-
50
70
5061
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Optimi
zation
Based Li
ver Cont
ou
r Extra
ctio
n
of
Abdomi
nal
CT Im
ages
Jayanthi
Muthuswam
y
1
, B
. Kan
ma
ni
2
1
Depa
rtment of
El
e
ct
roni
cs
and
Com
m
unic
at
ion Engi
ne
eri
ng,
B
MS
Coll
ege of E
ngine
er
ing,
Indi
a
2
Depa
rtment of
Te
l
ec
om
m
unic
ation,
BMS
Col
le
g
e
of Engin
ee
ring
,
Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
r 5
, 2
01
8
Re
vised
Ju
l
15
,
201
8
Accepte
d
J
ul
22
, 2
01
8
Thi
s
pap
er
in
tro
duce
s
computer
ai
ded
anal
y
s
is
sy
stem
for
dia
gn
osis
of
li
v
er
abnor
m
al
ity
in
a
bdom
ina
l
CT
i
m
age
s.
Segm
enting
the
li
ver
an
d
visual
i
zi
ng
the
reg
ion
of
i
nte
rest
is
a
m
ost
cha
llenging
t
ask
in
the
f
ie
ld
of
ca
nc
er
imaging,
due
to
sm
al
l
observa
b
le
ch
ange
s
bet
w
ee
n
he
al
th
y
and
unhea
l
th
y
li
ver
.
In
thi
s
pap
er,
h
y
brid
appr
o
ac
h
for
au
tomatic
ext
r
ac
t
ion
of
liver
cont
ou
r
is
proposed.
To
obta
in
opt
imal
thre
shold,
the
proposed
work
int
egr
at
es
segm
ent
at
ion
m
et
hod
with
optim
iz
at
ion
techniq
ue
in
ord
er
to
pr
ovide
b
et
t
er
ac
cur
acy
.
Thi
s
m
et
hod
uses
bil
at
er
al
filter
for
pre
proc
essing
a
nd
Fuzz
y
C
m
ea
ns
cl
uster
in
g
(FCM
)
for
segm
ent
at
ion
.
Me
an
Gre
y
W
olf
Optimiza
ti
o
n
te
chn
ique
(m
GW
O)
has
bee
n
used
to
get
the
opti
m
al
thres
hold.
Thi
s
thre
shold
is
used
for
segm
en
ti
ng
the
reg
ion
of
int
ere
st
.
From
th
e
segm
ent
ed
output
,
la
rg
est
connect
ed
reg
ion
is
id
entified
using
L
abe
l
Connecte
d
Com
ponent
(L
CC)
al
gori
thm.
The
eff
e
ct
iv
en
ess
of
proposed
m
et
hod
is
quant
itati
v
ely
e
val
ua
te
d
b
y
co
m
par
ing
with
ground
trut
h
obt
ai
ned
from
ra
diol
ogists
.
Th
e
per
form
ance
cr
i
te
ri
a
l
ike
d
ic
e
co
eff
icient,
true
po
siti
ve
err
or
and
m
iscl
assifi
c
at
ion
rate are
ta
k
en
for
evalua
t
ion
.
Ke
yw
or
d:
Bil
atera
l
fi
lter
Fuzz
y
C
m
e
ans
La
be
l
Conne
cted
Com
ponent
Optimiza
ti
o
n
Preproc
essing
Copyright
©
201
8
Instit
ute of
Ad
v
ance
d
Engi
ne
eri
ng
and
Sc
ie
n
ce
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Jay
anthi Mut
huswam
y,
Re
search
Sc
hola
r,
Dep
a
rtm
ent o
f El
ect
ro
nics
and C
omm
un
ic
ation
En
gin
ee
rin
g,
BM
S Co
ll
ege
of
En
gin
ee
rin
g,
Ba
ngal
ore
,
I
ndia
.
Em
a
il
:
j
ay
anthi.sathish
kum
ar@live.c
om
1.
INTROD
U
CTION
Liver
is
the
la
r
gest
a
nd
m
os
t
i
m
po
rtant
orga
n
for
s
urvival
.
It
is
one
an
d
a
half
kg
orga
n
and
locat
e
d
in
the
uppe
r
righ
t
quad
ra
nt
of
the
a
bdom
inal
cavit
y
[1
]
.
T
he
li
ver
perf
or
m
s
i
m
po
rta
nt
f
unct
ions
li
ke
filt
er
th
e
blood,
process
the
fats
and
to
m
et
abo
li
ze
and
store
carbo
hydr
at
es.
It
is
al
so
pro
ne
to
sev
er
al
disti
nct
typ
es
of
li
ver
ai
l
m
ents
.
Liver
disease
s
hav
e
dif
fer
e
nt
colo
rs
s
uch
as
blu
e
in
dic
at
e
cy
st,
ye
ll
o
w
in
dicat
e
fatt
y
li
ver
,
brown
is
fib
r
os
is
et
c.
D
if
f
eren
t
im
aging
te
chn
iq
ues
li
ke
ultras
ound,
com
pu
te
d
Tom
og
ra
ph
y
I
m
aging,
Ma
gn
et
ic
res
onance
im
agin
g,
posit
rons
e
m
issi
on
tom
og
ra
phy
et
c
ar
e
avail
abl
e
f
or
dia
gnos
is
of
li
ver
diseases
[2]
.
Am
on
g
these
,
CT
scan
is
a
w
el
l
kn
ow
n
non
-
inv
asi
ve
i
m
aging
m
od
al
it
ie
s
and
it
is
m
or
e
pr
e
ferred
by
diag
nosti
ci
ans
since
they
ha
ve
high
sig
nal
t
o
noise
rati
o,
good
s
patia
l
reso
luti
on,
patie
nt
fr
ie
nd
ly
pro
tocols,
le
ss
cost
an
d
le
ss
exam
inatio
n
ti
m
e.
It
also
pr
ov
i
des
m
or
e
accurate
anatom
ic
al
i
nfor
m
at
ion
ab
ou
t
the
visu
al
iz
ed
str
uc
ture.
Ma
nu
a
l
segm
entat
ion
of
li
ve
r
is
te
dio
us
a
nd
tim
e
c
on
s
um
ing
ta
sk
fo
r
rad
i
ologis
ts
[3
]
.
Au
t
om
ati
c
segm
entat
ion
is
m
or
e
trou
bles
om
e
becau
se
of
lo
w
co
ntras
t,
m
ulti
ple
s
lices
of
CT
i
m
age
an
d
si
m
il
ar
intens
ity
of
both
the
l
iver
an
d
tum
or.
Th
us
,
the
re
is
a
need
f
or
C
om
pu
te
r
ai
ded
analy
sis
syst
e
m
to
diag
nose the
li
ver ab
norm
al
ity from
the
huge
am
ou
nt of m
edical
d
at
a.
The
m
ai
n
goal
o
f
com
pu
te
r
ai
ded
analy
sis
sy
stem
is
to
pro
vi
de
com
pu
te
r
outp
ut,
as
a
sec
ond
opi
ni
on
to
assist
physi
ci
an
in
the
det
ect
ion
of
a
bnorm
aliti
es
and
to
im
pr
ov
e
the
segm
entat
ion
a
ccur
acy
.
This
syst
e
m
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5061
-
5070
5062
al
so
re
du
ce
s
the
im
age
readi
ng
ti
m
e.
This
syst
e
m
has
var
io
us
phases:
Data
acq
uisit
ion
,
prep
ro
ce
s
sing,
com
bin
ed
se
gm
entat
ion
a
nd
opti
m
iz
at
ion
al
gorithm
and
post
processi
ng.
Cl
us
te
rin
g
is
one
of
the
m
os
t
popula
r
se
gm
e
ntati
on
te
ch
niques
us
e
d
to
org
anize
the
obje
c
ts
into
gro
up
s
.
The
F
uzzy
C
-
Me
ans
al
gorith
m
is
a
cl
us
te
rin
g
al
go
rithm
wh
ere
ea
ch
it
em
m
ay
belong
to
m
or
e
than
one g
r
oup (h
e
nce
t
he
w
or
d
`
f
u
zzy
'
),
w
he
re
the
degree
of m
e
m
ber
s
hip f
or eac
h
it
em
is g
iven
b
y
a
pro
ba
bili
t
y dist
rib
ution o
f
the
clusters
.
Howe
ver,
it
suffe
rs
f
r
om
the
po
s
sibil
it
y
of
f
al
li
ng
into
loc
al
m
ini
m
a.
To
overc
om
e
this
dr
a
w
bac
k,
the
FCM
al
go
r
it
h
m
is
co
m
bin
ed
with
an
ot
he
r
so
ft
c
om
pu
ti
ng
al
go
rithm
.
Gr
ey
w
o
lf
opti
m
iz
at
ion
te
chni
qu
e
is
a
new
m
et
a
-
heu
risti
c
opti
m
i
zat
ion
al
gorit
hm
based
on
th
e
so
ci
al
hiera
r
chy
an
d
hunting
be
ha
vio
r
of
grey
wo
l
ves
a
nd
it
is
us
e
d
t
o
sea
rc
h
a
nd
hu
nt
a
prey
(s
olu
ti
on)
[4
]
.
T
his
te
ch
ni
qu
e
is
us
ed
to
op
ti
m
iz
e
the
cl
us
te
r
center
i
n
orde
r
to
fin
d
t
he
optim
a
l
thres
ho
l
d.
T
his
te
ch
nique
is
us
e
d
i
n
t
he
pro
po
s
ed
w
ork.
Th
e
r
est
of
thi
s
pap
e
r
is
organ
i
zed
as
fo
ll
ows:
Sect
ion
2
giv
e
s
a
detai
l
descr
ipti
on
o
f
the
propose
d
m
e
thod
.
Sect
ion
3
pre
sents
exp
e
r
im
ental
r
esults a
nd
pe
rfor
m
ance
analy
sis. Secti
on
4
pro
vid
es
the c
on
cl
us
io
ns
a
nd fu
ture
sc
ope.
Liver
segm
entat
ion
sti
ll
rem
ai
ns
a
n
op
e
n
ch
al
le
ng
e
pro
ble
m
fo
r
re
searc
he
rs.
Ma
ny
rese
arch
e
rs
ha
ve
dev
el
op
e
d
dif
f
eren
t
m
et
ho
ds
and
te
c
hn
i
ques
to
extract
the
li
ver
an
d
tum
or
from
t
he
abd
om
inal
CT
im
ages
ov
e
r
the
rece
nt
ye
ars.
And
num
ber
of
re
sear
ches
has
been
carried
out
on
so
ft
c
om
pu
ti
ng
base
d
opti
m
izati
on
m
et
ho
ds. T
his
li
te
ratur
e s
urve
y has
been d
on
e on th
ree
ph
a
s
es:
p
re
processi
ng, s
e
gm
entat
i
on and
optim
izati
on
.
The
e
xisti
ng m
et
ho
ds
us
e
d for
the e
xtracti
on
of li
ver
c
onto
ur are
d
isc
us
se
d i
n
this sect
i
on.
In
R
us
s
o
a
nd
Sonali
auth
or
pr
ese
nted
li
ne
ar
filt
erin
g
m
et
hod
to
rem
ove
Ga
us
sia
n
no
ise
s
in
a
CT
i
m
age
s
.
The
m
ai
n
dr
aw
bac
k
in
their
w
or
k
is
blurrin
g
pro
blem
[5
,
6]
.
To
overc
om
e
t
his,
Cha
ux
et
al
hav
e
dev
el
op
e
d
nonl
inear
filt
er
[7
]
.
This
filt
er
no
t
on
ly
rem
ov
es
the
blurri
ng
eff
ect
an
d
al
so
pr
ese
rv
e
s
the
edge
inf
or
m
at
ion
;
thereb
y
im
pr
ov
e
the
eff
ect
ive
ne
ss
of
non
-
li
ne
ar
filt
er.
Propo
sed
bilat
eral
filt
er
to
pr
ese
rve
the
edg
e
s
an
d
im
a
ge
sm
oo
the
n
in
g
us
i
ng
nonlin
ear
com
bin
at
io
n
of
im
age
pix
el
s
[
8
]
.
T
her
e
f
or
e
,
it
can
be
s
een
that
no
isy
CT
i
m
age
will
deg
ra
de
the
qu
al
it
y
of
a
n
i
m
age
so
pre
processi
ng
is
e
ssentia
l
in
com
pu
te
r
ai
ded
a
na
ly
si
s
syst
e
m
.
In
S.
Gunasundari
and
M.
Sugan
y
a
auth
or
ha
s
rev
ie
we
d
c
om
par
at
ive
stud
y
of
li
ver
segm
entat
i
on
m
et
ho
ds
to
e
xtract
the
li
ver
con
t
our
an
d
al
so
disc
us
sed
th
e
lim
it
ation
of
var
io
us
m
et
ho
ds
[
9]
.
K
Ma
la
et
al
hav
e
propose
d
adap
ti
ve
th
res
hold
base
d
m
or
phologica
l
pro
cessi
ng
f
or
li
ver
segm
en
ta
ti
on
[10
]
.
M
Jay
anthi
et
al
ha
ve
pro
pos
ed
an
a
ppr
oac
h
f
or
e
xtracti
ng
the
li
ver
a
nd
tum
or
fr
om
abdom
inal
CT
i
m
ages
an
d
us
e
d
f
or
com
pu
te
r
ai
de
d
diag
nosis
[
11
]
.
T
he
a
uthor
sh
a
ve
us
ed
see
ded
re
gion
gro
wing
m
et
ho
d
a
nd
f
oc
us
ed
on
lim
it
ed
a
m
ou
nt
of
sa
m
ple
i
m
age
a
nd
perf
or
m
ance
m
easur
es
no
t
evaluated
in
the
ir
w
ork.
S
S
Ku
m
ar
et
al
have
dev
el
op
e
d
a
C
AD syst
em
f
or
segm
enting
the
li
ver
a
nd tum
or
e
xtracti
on
[
12
]
.
Ah
m
ed
F
ou
a
d
Ali
et
al
ha
ve
presente
d
a
novel
a
ppr
oac
h
base
d
on
n
a
ture
i
ns
pi
red
optim
iz
at
ion
al
gorithm
s
and
hi
gh
li
ghte
d
the
pro
blem
pr
esent
in
t
he
CT
li
ver
s
eg
m
entat
ion
[
1
3
]
.
The
aut
hor
s
have
exp
la
ine
d
how
the
nat
ur
e
in
sp
ire
d
al
gorit
hm
s
can
be
ap
plied
to
so
l
ve
the
se
gm
enta
ti
on
prob
le
m
.
Geh
e
d
Ism
ail
et
al
ha
ve
pr
ese
nte
d
a
Com
pu
te
r
ai
de
d
diag
nosis
s
yst
e
m
fo
r
a
bd
om
inal
CT
i
m
ages
[
14
]
.
The
auth
ors
hav
e
pro
posed
a
hybr
i
d
m
eth
od
to
reduce
the
false
posi
ti
ve
error
rate.
In
Mit
ta
l
N
et
al
the
aut
hor
s
hav
e
exten
ded
a
m
od
ifie
d
m
ean
grey
wo
l
f
o
ptim
i
zat
ion
a
ppr
oac
h
f
or
bio
m
edical
p
r
oble
m
s
and
the
pe
rfo
rm
a
nce
was
c
om
par
ed wit
h othe
r
m
eta
-
he
ur
ist
ic
op
ti
m
iz
at
ion
algorit
hm
s
[15]
.
Ther
e
f
or
e,
it
can
be
see
n
tha
t
there
are
dif
f
eren
t
m
et
ho
ds
to
extract
the
li
ver
co
ntour
.
S
om
e
of
the
m
et
ho
ds
discu
ssed
in
the
li
te
ratur
e
s
urve
y
us
ed
pri
or
know
le
dge
of
reg
io
n
of
in
te
rest
and
a
sing
le
segm
entat
ion
m
et
ho
d
f
or
se
gm
enting
the
li
ver
c
on
t
our
from
abdom
in
al
CT
im
ages.
S
om
et
i
m
es,
these
m
et
ho
ds
m
igh
t
create
fak
e
se
gm
entat
ion
er
r
or.
I
n
Edy
Fr
a
din
at
a
et
al
a
nd
Ra
tna
Nit
i
n
Pati
l
et
al
the
auth
or
s
hav
e
e
xpla
ine
d
the
i
m
po
rta
nc
e
of
op
ti
m
iz
ati
on
a
nd
eval
uation
of
cl
assifi
c
at
ion
al
gorith
m
s
[1
6,
17]
.
T
he
ne
xt
sect
ion
disc
us
s
es
about
the
lim
it
a
ti
on
of
e
xisti
ng
m
et
ho
ds
and
al
so
e
xp
la
i
ns
how
the p
r
opose
d
m
et
ho
d
is
us
e
d
for
li
ve
r
se
gm
e
ntati
on
i
n
c
ompu
te
r
ai
de
d
a
n
a
ly
sis sy
stem
.
1.1.
Rese
arch
Pr
obl
em
The
si
gn
i
fican
ce o
f
li
ver se
gm
entat
ion
probl
e
m
s
are as foll
ow
s
a.
Au
t
om
atic
segm
entat
ion
of
li
ver
is
a
dif
ficu
lt
ta
sk
du
e
to
t
he
overla
pp
i
ng
of
re
gion
of
li
ver
with
ad
j
ac
ent
orga
ns
a
nd the
intensit
y of l
iv
er is sam
e as
th
at
o
f
o
t
her
orga
ns
.
b.
Existi
ng m
et
ho
ds
us
e
d
pix
el
-
ba
sed
distrib
utio
n for
fin
ding th
e li
ver
c
onto
ur.
c.
Most
of
the
re
searche
r
fo
c
use
d
on
sin
gle
s
egm
entat
ion
m
et
hod,
but
this
m
e
tho
d
inca
pa
ble
to
so
l
ve
t
he
com
plex
pro
bl
e
m
li
ke
li
ver
s
hap
e
v
a
riat
ion
a
m
on
g
the
pati
ents.
d.
Ma
chine
le
a
rn
i
ng b
a
sed
li
ver
segm
entat
ion
m
et
ho
d fin
ds
l
aborio
us
ly
to
c
al
culat
e optim
a
l t
hr
es
ho
l
d.
So
t
he
pro
po
se
d
work
inte
gr
at
es
m
achine
le
arn
i
ng
with
opti
m
iz
at
ion
m
e
tho
d
to
im
pr
ove
the
ef
ficacy
of
segm
entat
ion
resu
lt
s
a
nd
a
lso
prov
i
de
t
he
opti
m
a
l
so
lu
ti
on
of
fi
nd
i
ng
the
cl
us
te
r
ce
nters.
T
her
e
for
e,
th
e
pro
blem
state
m
ent
of
the
pr
opos
e
d
work
c
an
be
sta
te
d
as
“To
devel
op
a
pre
ci
se
seg
m
entatio
n
meth
od
th
at
incor
pora
te
F
CM
wi
th
bi
o
insp
ire
d
op
ti
m
izati
on
met
hod
fo
r
the
ext
r
action
of
li
ver
co
nt
our
fro
m
th
e
abdomi
na
l C
T
images”
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
Op
ti
miz
atio
n
B
as
e
d
Live
r C
on
tou
r
Extracti
on
o
f A
bdomin
al
CT I
mages
(
J
ay
an
t
hi M
uthus
wam
y
)
5063
1.2.
Prop
os
ed
S
olu
tion
The
pro
pose
d
so
luti
on
is
an
extensi
on
of
our
hy
br
i
d
m
e
tho
d
to
wa
rd
s
li
ve
r
diag
nosis
prob
le
m
[1
8]
.
We
highli
ghte
d
the
im
po
rtan
ce
of
hy
br
i
d
te
chn
i
qu
e
s
for
the
extracti
on
of
li
ver
c
onto
ur.
The
refor
e
,
in
this
pro
po
se
d
so
l
ution
,
we
em
ph
asi
ze
bio
insp
ir
e
d
optim
iz
at
ion
m
et
ho
d
in
ord
e
r
to
fin
d
the
op
tim
a
l
thresh
ol
d
that
can
be
use
d
by
rad
i
ologist
s
f
or
the
e
xtracti
on
of
li
ver
c
onto
ur
is
disc
us
se
d.
The
pro
pose
d
syst
e
m
has
va
rio
us
ph
a
ses to
obtai
n
the
r
e
gion
of
inter
est
.
Fi
gure
1
s
how
s the
var
io
us
ph
a
ses i
nvolv
e
d
i
n
th
e
pro
po
se
d
syst
e
m
.
Figure
1
.
Bl
oc
k diag
ram
of
the
pro
posed
sys
tem
2.
PROP
OSE
D
METHO
D
I
M
PLE
MENT
A
TION
The
va
rio
us
st
eps
i
nvolv
e
d
i
n
th
e
e
xtract
io
n
of
li
ver
co
nt
our
a
re
discu
ssed
i
n
t
his
se
ct
ion
.
Th
e
pr
im
ary
ste
p
i
n
com
pu
te
r
ai
de
d
analy
sis
syst
e
m
is
a
colle
c
ti
on
of
i
m
age
database
for
abdom
inal
CT
i
m
ages.
The
n,
pr
e
proce
ssing
filt
er
is
discusse
d
i
n
order
t
o
im
pr
ove
the
qu
al
it
y
of
an
im
age.
F
ollow
e
d
by
this,
FCM
base
d
optim
iz
a
ti
on
an
d
la
bel
connecte
d
al
gorithm
s
are
exp
la
ined
to
get
the
li
ver
co
ntour
of
ab
dom
in
al
CT
i
m
ages.
The
va
rio
us
ph
a
ses in
the
pro
po
se
d m
et
ho
d
a
re cle
arly
explai
ned
in the
ac
com
pan
yi
ng p
a
ssage
.
2.1.
Ab
d
omin
al CT
Ima
ge
Com
pu
te
d
T
om
og
raphy
sca
n
com
bin
es
se
r
ie
s
of
X
-
ray
i
m
ages
that
are
ta
ken
from
diff
e
ren
t
a
ng
le
s
,
and
us
es
c
om
pu
te
r
processin
g
to
pro
du
ce
c
ro
ss
sect
ion
al
i
m
ages
of
t
he
hu
m
an
body.
I
t
prov
i
des
the
detai
l
inf
or
m
at
ion
of
the
intern
al
orga
ns
a
nd
al
s
o
helps
rad
i
ologist
s
to
dia
gnose
the
diseases
.
For
a
patie
nt,
m
or
e
than
15
0
sli
ces
of
CT
i
m
age
are
ob
ta
ine
d.
A
su
it
able
sel
ect
ion
of
CT
Im
age
is
essenti
al
.
Mi
dd
le
sl
ic
e
is
pr
e
fer
a
ble
an
d
giv
es
m
or
e
de
ta
il
ed
inform
ation
.
B
oth
re
al
tim
e
database
and
sim
ulate
d
database
a
re
use
d
in
the
pro
po
se
d
work.
O
ur
database
is
a
colle
ct
ion
of
var
i
ous
i
m
ages
of
CT
abdom
inal
or
ga
ns
wh
ic
h
include
norm
al
, f
at
ty
an
d o
ve
r
e
xten
de
d
li
ve
r wit
h
tu
m
or
.
2.2.
Bi
latera
l
Fil
te
r
In
orde
r
to
im
pro
ve
the
qu
al
i
ty
of
a
CT
im
a
ge,
pr
e
processi
ng
is
ess
entia
l.
Bi
la
te
ral
filt
er
is
us
ed
a
s
pr
e
processi
ng
filt
er.
It
is
non
-
li
nea
r
,
non
-
it
erati
ve
,
e
dg
e
pr
ese
r
ving
sm
oo
t
hing
filt
er
dev
el
op
e
d
by
Tom
asi
[
19
]
.
The
m
ain
noise
in
the
CT
i
m
age
is
Gau
s
sia
n
noise
.
Bi
la
te
ral
filt
e
r
is
ob
ta
ine
d
by
the
com
bin
at
ion
of
w
ei
ghte
d
f
unct
ion
of
two
Ga
us
sia
n
filt
ers:
par
ti
al
do
m
ai
n
and
intensit
y
do
m
ai
n.
The
intensit
y
value
of
ea
c
h
pix
el
is
rep
la
c
ed
by
weig
hte
d
ave
rag
e
.
Let
G
is
an
i
m
age.
The
n
G
p
is
the
value
of
the
i
m
age
G
at
pix
el
po
sit
io
n u. B
f (
G)
is
the
outp
ut
o
f
b
il
at
eral
fil
te
r
ap
plied t
o
a
n Im
age G
Sp
at
ia
l dista
nc
e is cal
culat
ed a
s
(
,
)
=
−
|
|
−
|
|
2
/
2
2
(1)
intensit
y
dif
fere
nce is calc
ulate
d
as
(
,
)
=
−
|
(
)
−
(
)
|
2
/
2
2
(2)
gu
a
ssian
ke
rn
e
l
coeffic
ie
nt
a
re
σ
sd
a
nd
σ
id
wh
ic
h
c
on
t
ro
ls
sp
at
ia
l
dista
nc
e
an
d
i
ntensity
diff
e
re
nce.
Thes
e
coeffie
ie
nts
ar
e
directl
y
pr
op
or
ti
onal
to
im
a
ge
siz
e
an
d
ed
ge
am
plit
ud
e
and
t
hese
are
m
ai
nly
us
ed
to
c
on
t
ro
l
the w
ei
gh
ts
in spati
al
and inte
ns
it
y d
om
ai
n.
A
t pi
xel p
Acquisi
ti
on ph
ase
(Abdom
inal C
T
i
m
age)
Pr
e
processin
g ph
a
se
(Bil
at
eral fil
te
ring)
Segm
entat
ion
and
Op
ti
m
iz
ation
phase
(F
CM
+ m
GW
O
)
Po
stp
r
ocessin
g Phas
e
(LCC+
Mor
phologica
lfil
li
ng
)
Segm
ented
Liv
er
Liver v
olu
m
e
cal
culat
ion
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5061
-
5070
5064
(
)
=
1
∑
(
)
∈
(
)
(
,
)
(
,
)
(3)
wh
e
re
C
is
a
norm
al
iz
a
ti
on
factor
=
∑
(
,
)
(
,
)
an
d
(
)
sh
ows
the
sp
at
i
al
neighb
orh
ood
of
B
if(p). Eac
h p
ixel i
s r
e
placed
by the
w
ei
gh
te
d
a
ver
a
ge of
n
e
arb
y
pix
el
s i
n
s
pat
ia
l neig
hbor
hood.
Show
s
the
sp
at
ia
l neighb
orh
ood o
f
B
f(p).
T
he
p
se
udo co
de of
b
il
at
eral fil
te
r
is s
how
n
i
n Table
1.
Table
1
.
Pseud
o
C
od
e
for B
il
at
eral fil
te
r
Inp
u
t: CT
I
m
ag
e ;
Ou
tp
u
t: Pr
ep
rocess
ed
i
m
ag
e
1
.
DICO
M
i
m
ag
e i
s co
n
v
erted to
JPE
G us
in
g
acc
u
lite
so
f
tware.
2
.
Co
n
v
ert
th
at CT
i
m
ag
e to g
rey scal
e i
m
ag
e
.
3
.
Def
in
e the b
ilate
ral
f
ilter
p
a
ra
m
eter
w,
σsd
and
σid
4
.
Fo
r
all
th
e pix
el,
do
the f
o
llo
win
g
s
tep
s 5
,6,7
5
.
Calcu
late the sp
atial dis
tan
ce us
in
g
the eq
u
atio
n
1
6
.
Calcu
late the in
t
en
sity
d
if
f
erence
u
sin
g
the eq
u
atio
n
2
7
.
Ap
p
ly
the f
ilteri
n
g
valu
es o
n
CT
g
rey
scal
e i
m
ag
e us
i
n
g
the eq
u
atio
n
3.
8.
Get the
resu
ltan
t pre proces
sed
i
m
a
g
e.
2.3.
Fuz
z
y
C
Me
ans
Al
go
ri
th
m
FCM
[
20
]
bas
ed
obj
ect
ive
functi
on
is
a
m
os
t
po
pula
r
it
er
at
ive
cl
us
te
rin
g
al
gorithm
th
at
al
lows
the
m
os
t
pr
eci
se
f
or
m
ulati
on
of
the
cl
us
te
ri
ng
crit
eria.
T
he
ba
sic
co
ncep
t
is
to
fi
nd
opti
m
al
cl
us
te
r
ce
nter
that
m
ini
m
iz
e
obj
e
ct
ive
functi
on
.
T
h
e
ba
s
i
c
con
c
ep
t
i
s
to
p
art
i
t
i
o
n
a
d
a
t
a
se
t
DX
=
{DX
1
,D
X
2
….
DX
M
}
i
n
t
o
‘C’
nu
m
b
e
r
of
c
lus
t
e
r
s.
F
i
r
s
t
s
t
ep
in
F
C
M
i
s
to
ca
l
c
u
l
a
te
th
e
d
egr
e
e
of
m
e
m
b
e
r
s
h
ip
fu
n
cti
o
n.
Fo
r
a
g
ive
n
d
a
t
a
po
i
n
t D
X
i
,
th
e d
egr
e
e
o
f
i
t
s
me
m
b
e
r
sh
i
p
t
o
clu
s
t
er
j
i
s
ca
l
cul
a
t
e
d
a
s
f
o
l
lo
ws
;
=
1
∑
(
‖
−
‖
‖
−
‖
)
2
−
1
=
1
(4
)
wh
er
e
m
i
s
a
fu
z
z
in
e
ss
co
eff
i
c
i
e
n
t
,
v
a
l
ue
o
f
i
r
ang
e
s
fr
o
m
1
to
M
and
j
v
ar
i
e
s
fr
o
m
1
t
o
C.
Th
en
c
a
l
cu
l
at
e
c
l
u
s
t
er
c
en
t
ro
id
b
as
ed
o
n
w
e
igh
t
ed
av
er
ag
e u
s
ing
equ
a
t
ion
=
∑
=
1
∑
=
1
(5
)
n
ex
t
,
c
a
l
cu
l
a
t
e E
u
c
l
id
e
an
d
i
s
tan
c
e b
e
tw
e
en
c
l
u
s
t
er
v
ec
t
or
s t
o
th
e
da
t
a
p
o
in
t
u
s
ing
th
e
fo
r
m
u
l
a
_
=
|
−
|
2
(6
)
t
h
i
s
d
is
t
an
c
e
fi
nd
s
t
h
e
c
lo
sen
e
ss
of
e
a
ch
da
t
a
po
i
n
t
to
t
he
c
l
u
s
t
er
ve
c
tor
C
j
.
Fo
r
e
ve
ry
i
t
e
r
a
t
i
on
of
F
C
M
a
l
go
r
i
th
m
,
t
h
e fo
l
l
ow
in
g
ob
j
ec
t
i
v
e
p
o
in
t
i
s mi
n
i
mi
z
e
d
.
=
∑
∑
|
−
|
2
=
1
=
1
(7
)
i
f
(
O
bJ
m
-
Ob
J
m
-
1
)
<
=
ε
i
s
s
a
t
i
sf
ied
,
f
o
l
l
ow
in
g
ite
r
a
t
i
on
i
s
ter
m
i
n
a
t
ed
.
Th
e v
a
lu
e
of
ε
ra
ng
es
f
ro
m
0
t
o
1
.
2.4.
Grey
Wol
f
O
p
timi
z
at
ion
Mi
rij
al
i
et
al
propose
d
a
new
heurist
ic
op
ti
m
iz
at
ion
al
gorithm
.
This
al
go
rithm
m
i
m
ic
s
so
ci
al
hierar
c
hy
a
nd h
unti
ng b
eha
vi
or
of
grey
w
ol
ves
in
nat
ur
e
[
21
]
. Th
e
ad
va
nt
age
of g
rey w
olf
is
sim
ple,
easi
ly
be
pro
gr
am
m
ed
a
nd
do
es
no
t
ne
ed
sp
eci
fic
in
pu
t
pa
ram
et
ers.
Gen
e
rall
y,
G
rey
wo
lv
es
li
ve
in
a
gr
oup
(
pack),
each
gro
up
ha
s
an
aver
a
ge
m
e
m
ber
of
5
-
10.
All
the
m
e
m
ber
s
in
the
gr
ou
p
ha
ve
a
own
power
of
dom
inant
hierar
c
hy
as
show
n
in
Fig
ure
2
(
a
)
.
F
our
ty
pe
s
of
grey
w
olve
s
al
ph
a
(
α
)
,
be
ta
(
β
),
delta
(
δ
),
an
d
om
ega
(
ω
)
are
worked
fo
r
si
m
ula
ti
ng
the
l
eaders
hip
hier
arch
y.
T
his
is
us
e
d
to
sea
rch
and
hunt
th
e
pr
ey
(
so
l
ution)
and
t
he
le
ader
s
hip
pr
io
rity
le
vel
to
plan
the
hu
nting
are
α
,
β
,
δ
and
ω
resp
ect
i
ve
ly
.
The
wo
l
ve
s
om
ega
ω,
and
a
re
respo
ns
ible
f
or
gu
i
ding
the
se
arch
(hu
nting),
wh
i
le
ot
her
w
olv
es
fo
ll
ow.
Thr
ee
ste
ps
of
the
huntin
g
be
hav
i
or
are e
ncircli
ng, hunti
ng a
nd att
ackin
g
the
pre
y.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
Op
ti
miz
atio
n
B
as
e
d
Live
r C
on
tou
r
Extracti
on
o
f A
bdomin
al
CT I
mages
(
J
ay
an
t
hi M
uthus
wam
y
)
5065
Enci
rcl
ing t
he
prey
Wh
e
n
the
gre
y
wo
lves
sta
r
t
hu
ntin
g,
the
y
encircle
pr
e
y.
The
encirc
li
ng
be
hav
i
or
is
pr
esented
as
a
m
at
he
m
at
ic
a
l m
od
el
an
d gi
ve
n
a
s
=
|
.
(
)
−
(
)
|
(8)
(
+
1
)
=
(
)
−
.
(9)
=
2
1
−
(10)
=
2
2
(11)
Wh
e
re
t
is
a
c
urren
t
it
erati
on
,
X
pre
is
a
prey
posit
ion
vecto
r
,
X
gp
is
the
gr
ey
wo
lf
vecto
r
posit
ion
a
nd
A,
C
are
the co
e
ff
ic
ie
nt
vect
or. Ra
ndom
v
ect
or
is
r
1 and r
2
,
the
val
ues
i
n
the
r
a
nge of [0,
1].
The
al
ph
a
guid
es
the
huntin
g
process
an
d
be
ta
and
delta
m
i
gh
t
ha
ve
a
pa
rt
in
huntin
g.
Th
e
updatin
g
pos
it
ion
of
grey
wo
l
ves
i
s
s
how
n
in
F
igure
2
(
b
)
.
T
he
locat
io
n
of
prey
posit
io
n
is
ex
pected
to
c
om
e
fr
om
the
al
ph
a
,
beta
, a
nd
delta
wo
l
ves q a
nd
gi
ven
by the
foll
ow
i
ng equati
on
=
|
1
.
(
)
−
(
)
|
(12)
=
|
2
.
(
)
−
(
)
|
(13)
=
|
3
.
(
)
−
(
)
|
(14)
1
=
(
)
−
1
(15)
2
=
(
)
−
2
(16)
3
=
(
)
−
3
(17)
(
+
1
)
=
1
+
2
+
3
3
(18)
Atta
c
king
Wh
e
n
the
prey
stop
s
m
ov
ing
,
th
e
gr
ey
wo
lve
s
en
d
the
huntin
g
process.
T
hi
s
process
is
m
at
he
m
at
ic
a
lly
ex
pr
esse
d
by
decr
ea
sin
g
a
from
2
to
0.
I
f
|
Av
|
<
1,
wo
l
ves
m
ov
e
towa
rd
s
the
pr
ey
for
at
ta
cking.
T
he pict
or
ia
l
repres
entat
ion
of
gr
e
y wo
l
ves u
pd
at
ing
posit
ion i
s
sh
ow
n
in
Fi
gur
e 2
(
b
)
.
Figure
2
(
a
)
.
S
oc
ia
l hierarc
hy
of wolves
[
22]
Figure
2
(
b
)
.
Posi
ti
on
updatin
g
of
gr
ey
wo
l
f
[
22
]
i
n
m
-
G
WO
optim
iz
at
ion
alg
ori
thm
,
m
ean
is
ta
ken
into
consi
der
at
io
n.
This
al
gorithm
is
us
ed
to
fi
nd
th
e
op
ti
m
al
threshold
value
s
in
orde
r
to
im
pr
ove
the
c
lu
ste
rin
g
re
su
lt
s
pro
duced
by
FCM
.
The
P
su
e
do
c
ode
f
or
m
-
GW
O o
pti
m
iz
at
ion
A
l
gorithm
is sh
own
in
Tab
le
2.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5061
-
5070
5066
Table
2.
m
-
G
WO al
gorithm
p
se
udo
c
ode
1.
Read
the en
h
an
ced b
ilateral
f
ilte
r
o
u
tp
u
t.
2.
Get clus
ter
center
Cj wh
ere
j
=1
,2,…
n
u
m
_
o
f
_
clu
ster
u
sin
g
equ
atio
n
4.
3.
Initialis
e
m
-
G
W
O
p
ara
m
et
ers
–
sea
rc
h
agen
ts, di
m
en
sio
n
,
m
ax
i
m
u
m
it
erati
o
n
,
lower and
up
p
er
sea
rch b
o
u
n
d
ary
.
4.
Gen
erate
wo
lv
es p
o
sitio
n
Pi
rand
o
m
l
y
on
size of
the p
a
ck
.
5.
Ass
ig
n
Pj=Cj th
at i
s jth
pack
of
ith wo
lv
es is Cj.
6.
Initial
ize alph
a,
beta ,
d
elta
.
7.
Initialize wo
lv
es p
o
sitio
n
s Pα, Pβ, Pδ
8.
Set l=0(ite
ration
)
9.
W
h
ile l<
m
ax
_
it do
10.
f
o
r
each search
ag
en
t do
11.
Calcu
late the f
itn
ess
of
eac
h
searc
h
a
g
en
t.
12.
Up
ad
ate curr
en
t se
arch agen
t po
sitio
n
bas
ed
on
f
itn
ess
.
13.
End
f
or
14.
Up
d
ate Pα,
Pβ,
Pδ
u
sin
g
equ
atio
n
9
-
1
5
(by inserti
ng
mea
n bef
o
re
X)
15.
Set l=l+1
16.
End
while
17.
Use th
e bes
t so
lu
tio
n
to g
en
erate
the
p
artition
m
atrix
Ui
j .
An
d
gen
erate
th
e f
in
al clus
ter
with
t
h
e partitio
n
m
at
rix.
2.5.
Label
Connec
ted Alg
orith
m
The
al
gorithm
has
t
wo
m
od
ules:
(i)
Labeli
ng
of
c
onnecte
d
c
om
po
ne
nts
(ii)
Sea
rch
f
or
the
la
r
gest
com
po
ne
nt
[
23
]
.
Fo
r
the
outp
ut
of
la
bel
c
on
nected
c
om
po
ne
nt,
a
pp
ly
m
or
phologica
l
ope
ning
filt
er
is
use
d
t
o
fill
the
hole
s.
The
s
uper
posit
ion
of
the
c
ontour
on
the
o
ri
gin
al
im
age
al
l
ow
s
us
to
de
duct
the
reg
i
on
of
t
he
li
ver
. T
o o
btain the
li
ver re
gi
on, live
r
m
ask
is m
ult
ipli
ed
w
it
h
ori
gin
al
im
age
.
3.
RESU
LT
S
AND A
N
ALYSIS
The
pro
pose
d
appr
oach
is
an
al
yz
ed,
app
li
e
d
and
te
ste
d
on
abdom
inal
CT
i
m
age
dataset
[2
4
]
.
I
n
this
work,
20
Im
ages
is
ta
ken
for
analy
sis.
All
i
m
ages
are
axial
i
m
ages.
The
pro
po
se
d
w
ork
s
are
carried
out
on
Pentium
proce
sso
r
an
d
im
pl
em
ented
in
M
A
TLAB
9
.
The
par
am
et
er
set
tin
g
of
bilat
eral
filt
er
an
d
m
-
GW
O
is
sh
ow
n
in
Ta
ble 3
.
Table
3
.
Param
et
er s
et
ti
ng
T
er
m
Descripti
o
n
Valu
es
w
W
in
d
o
w size
5
σsd
Sp
atial dis
tan
ce
3
σid
Inten
sity
d
if
f
erence
0
.1
n
u
m
_
clu
s
Nu
m
b
e
r
o
f
f
u
zzy
c
lu
sters
3
m
Fu
zzin
ess
2
n
o
_
search
Nu
m
b
e
r
o
f
searc
h
ag
en
ts
20
Max_
it
Nu
m
b
e
r
o
f
iter
a
tio
n
10
lb
,ub
Lower a
n
d
up
p
er
b
o
u
n
d
[
0
255]
The
ab
dom
inal
CT
i
m
age
i
s
conve
rted
t
o
gray
scal
e
i
m
age.
To
re
m
ov
e
the
no
i
se
arti
facts,
pr
e
processi
ng
i
s
app
li
ed
on
gr
ay
le
vel
i
m
age
CT
i
m
ages.
Inpu
t
to
the
bilat
eral
filt
er
is
no
rm
alized
with
cl
os
e
d
interval
of
[
0,
1].
Since
im
age
was
norm
al
i
zed,
pa
ram
et
e
r
of
bilat
eral
fil
te
r
was
al
so
no
rm
alized
by
div
idin
g
with m
axi
m
u
m
intensit
y. T
he ou
t
pu
t
of
bilat
eral fil
te
r
is s
ho
wn in t
he
Fi
gur
e 3
.
Figure
3
.
G
ray
scal
e CT im
ag
e an
d ou
t
pu
t
of
b
il
at
eral fil
te
r
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
Op
ti
miz
atio
n
B
as
e
d
Live
r C
on
tou
r
Extracti
on
o
f A
bdomin
al
CT I
mages
(
J
ay
an
t
hi M
uthus
wam
y
)
5067
The
ou
t
pu
t
of
bilat
eral
filt
er
is
ta
ken
f
or
ne
xt
ph
ase
.
W
it
h
t
he
h
el
p
of
m
GW
O
opti
m
iz
at
i
on
m
et
ho
d,
three
cl
us
te
r
c
enters
of
opti
m
iz
ed
values
wer
e
ob
ta
ine
d.
These
val
ues
wer
e
use
d
by
FCM
al
go
rith
m
.Th
e
corres
pond
in
g cl
us
te
r ou
t
pu
t
is sho
wn in F
ig
ur
e
4.
Figure
4
.
FCM
outp
ut of
dif
fe
ren
t cl
us
te
rs
The
best
cl
us
t
er
wa
s
ta
ke
n
f
or
post
proces
sing.
I
n
post
proces
sin
g
phas
e,
la
r
gest
co
nn
ect
ed
re
gion
was
i
den
ti
fie
d and m
or
phol
ogic
al
f
il
li
ng
w
a
s
don
e
. T
his
give
s seg
m
ented
l
iver
c
onto
ur
a
s
shown i
n
Fi
gu
re
5.
Figure
5
.
Live
r
conto
ur outp
ut
and
s
uperim
po
se
d
l
iver
outp
ut w
it
h o
rigi
na
l im
age
Af
te
r
obtai
ni
ng the se
gm
ented
li
ver
,
conto
ur
detect
ion i
s us
ed fo
r
the
v
is
ua
li
zat
ion
.
T
he vo
l
um
e o
f
the li
ver is cal
c
ulate
d usin
g
t
he
for
m
ula g
ive
n belo
w.
F
or s
a
m
ple d
at
abase
,
Vo
l
um
e o
f
li
ve
r
=sl
ic
e
thic
kness* Are
a
Ar
ea=
N
um
ber
o
f
p
i
xels *Pi
xe
l dim
ension
The
e
xperim
e
ntal
res
ults
of
pro
posed
m
eth
od
a
re
c
om
par
ed
with
ot
he
r
se
gm
entat
ion
m
e
tho
ds
that
we
re
discusse
d
in
[
25]
.
The
ou
t
pu
t
of
al
l
m
et
ho
ds
is
pa
inted
in
Figure
6
.
F
rom
the
figure
it
is
in
ferred
tha
t
FCM
base
d
optim
iz
a
ti
on
m
e
tho
d
gi
ves
bette
r
res
ul
ts.
But
the
ey
e
of
the
m
ind
ind
ic
at
es
that
op
t
i
m
iz
ation
res
ults
are
adm
irable
when
c
om
par
ed
to
al
l
oth
e
r
m
et
ho
ds.
T
his
co
nclusi
on
is
purely
based
on
visu
al
iz
at
ion
.
T
o
unde
rstan
d
the
best se
gm
entation
m
et
ho
d,
fo
l
lowing
pe
rform
ance m
easur
es are
done
.
Histo
gr
am
b
as
ed
li
ve
r
segm
entat
ion
Seede
d regi
on
grow
i
ng
base
d
li
ver se
gm
entat
ion
LCC
b
ase
d
li
ve
r
segm
entat
ion
Pr
op
os
e
d
m
et
ho
d
Figure
6.
Re
su
l
ts of va
rio
us
se
gm
entat
ion
m
e
thods
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5061
-
5070
5068
Dice coe
ff
ic
ie
nt
giv
es the
sim
i
la
rity
b
et
ween
autom
at
ic
seg
m
entat
ion
and
gro
und
tr
uth
i
m
age.
A
nd it
def
i
ned as
fo
ll
ow
s
D
ice
=
2
|
A
∩
M
|
|
A
+
M
|
(16)
Tru
e
posit
ive
f
racti
on
al
s
o
cal
le
d
Sensi
ti
vity
and
Re
cal
l,
m
easur
e
s
the
portion
of
posit
iv
e
vowels
i
n
the gr
ound tr
uth
that a
re als
o
i
den
ti
fie
d
as
po
sit
ive b
y t
he
se
gm
entat
ion
b
ei
ng ev
al
uated
.
True
Posi
t
iv
e
fracti
on
=
|
A
∩
M
|
M
(17)
Mi
sc
lassi
ficat
ion
R
a
te
=
1
−
|
A
∩
M
|
M
(18)
A
is
nu
m
ber
of
pix
el
s
of
the
a
uto
m
at
ic
ally
se
gm
ented
li
ver
reg
i
on
s
a
nd
M
is
nu
m
ber
of
pi
xels
of
th
e
m
anu
al
ly
seg
m
ented
li
ver
(
gro
und
tr
uth)
by
the
ex
per
ts
.
A
bove
m
et
ri
cs
are
us
e
d
t
o
fin
d
t
he
sim
i
la
rity
betwee
n
a
utom
at
ic
and
m
anu
al
se
gm
ented
outp
ut
(
GT
).
Table
4
s
hows
the
sta
ti
sti
cal
analy
sis
of
se
gm
ented
ou
t
pu
t.
P
r
opose
d
m
et
ho
d
(PM
)
runs
f
or
D
at
aset
con
ta
ins
20
im
ages
and
this
ta
ble
il
l
us
trat
es
the
de
ta
il
ed
ou
tc
om
e
of
propose
d
w
ork
i
n
te
rm
s
of
the
dice
coeffic
ie
nt
,
true
posit
ive
rate
and
m
isc
l
assifi
cat
ion
rat
e.
Th
e
accuracy i
s m
ore if t
he
m
isc
lassificat
ion
rate
is l
ess.
Table
4
.
Per
for
m
ance
m
et
rics
of
propose
d
m
et
hod
Dataset
PM
GT
TPF
DICE
MCR
ID1
2
0
0
5
4
2
2
1
2
3
0
.90
4
0
.92
1
7
0
.09
6
ID2
2788
3103
0
.89
8
4
0
.90
2
3
0
.10
1
6
ID3
ID4
ID5
6612
1
0
3
1
2
1852
7258
1
1
8
8
6
1942
0
.91
0
9
0
.89
2
0
.95
3
6
0
.93
1
8
0
.91
3
2
0
.92
6
4
0
.08
9
1
0
.10
8
0
.04
6
4
The
pe
rfo
rm
ance
of
propose
d
work
is
c
om
par
ed
with
[
25
]
and
t
he
res
ults
are
ta
bu
la
te
d
in
Table
5
,
it
can
be
obser
ve
d
that
accu
ra
cy
o
f
the
pro
pose
d
work
is
i
m
pr
ov
e
d.
T
hu
s,
FCM
base
d
optim
iz
at
ion
m
et
ho
d
perform
ed
well
fo
r
e
xtracti
ng
the
li
ver
c
on
t
our
a
nd
it
can
be
use
d
i
n
co
m
pu
te
r
ai
ded
a
naly
sis
syst
e
m
to
fi
nd
the abn
or
m
al
ity i
n
the li
ve
r
c
on
t
our.
T
he dic
e coe
ff
ic
ie
nt
pl
ot ar
e
r
e
pr
ese
nt
ed
in
F
ig
ur
e
7
.
Table
5
.
C
om
par
is
on
of
ot
her
segm
entat
ion
techn
i
qu
e
s in
term
s o
f dice
co
e
ff
ic
ie
nt
Metho
d
s
DT2
DT3
DT4
DT5
Histo
g
ra
m
0
.67
0
.65
2
0
.68
9
0
.73
1
2
SR
G
LCC
Prop
o
sed
0
.78
0
.83
0
.90
0
.75
4
0
.84
1
0
.93
1
8
0
.73
4
0
.84
1
0
.91
3
0
.78
6
2
0
.89
1
0
.92
6
4
Figure
7
.
Dice
coeffic
ie
nt
plo
t
of
var
io
us
seg
m
entat
ion
m
eth
ods
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
Op
ti
miz
atio
n
B
as
e
d
Live
r C
on
tou
r
Extracti
on
o
f A
bdomin
al
CT I
mages
(
J
ay
an
t
hi M
uthus
wam
y
)
5069
4.
CONCL
US
I
O
N
The
P
rop
os
e
d
al
gorithm
us
es
hybri
d
m
et
hod,
c
om
bin
at
ion
of
FC
M
and
m
ean
Gr
ey
wo
lf
op
ti
m
iz
ation
al
gorithm
with
l
abel
co
nn
ect
e
d
com
po
ne
nt
al
gorithm
fo
r
li
ve
r
segm
entat
ion
.
T
he
pe
rfo
r
m
ance
of
t
he
pro
po
s
ed
m
et
ho
d
w
as
com
par
ed
with
e
xisti
ng
m
et
ho
ds
an
d
t
he
pr
opos
e
d
m
et
ho
d
ac
hie
ve
d
the
accuracy
of
9
1
%
fo
r
li
ve
r
se
gm
entat
ion
.
The
pro
posed
w
ork
is
m
a
inly
us
e
d
to
identif
y
the
su
sp
ic
io
us
pix
el
s
from
the
li
ver
re
gion
a
nd
he
lps
ra
dio
l
og
is
ts
for
diag
nos
ing
li
ve
r
disea
ses.
F
or
the
f
uture
stu
dies,
oth
e
r
op
ti
m
iz
ation
al
gorithm
s w
il
l be co
m
par
ed
w
i
th pr
opos
e
d
m
et
hod for
diff
e
r
ent m
edical
i
m
agin
g pro
blem
s.
REFERE
NCE
S
[1]
Ric
har
d
Car
lt
on
,
Arlen
e
Mcke
nn
a
Adler
,
“
An
ov
erv
ie
w
of
imagi
ng
m
odal
it
ie
s
princ
iples
of
rad
iogra
phic
imaging
:
An a
rt and
A
sci
enc
e
”,
c
engage
l
earning
,
5
th
edi
t
i
on,
2012
.
[2]
Ak
shat
Gotra
,
Loj
an
Sivakuma
ran
,
“
Li
ver
seg
m
ent
at
ion:
indi
c
at
ions,
t
ec
hn
iqu
es
and
future
di
rec
t
ions
”
,
Insigh
t
imaging
,
vo
l
8,
pp
377
–
392,
2017
.
[3]
Ja
y
an
thi,
M.,
Kanm
ani
,
B.
,
“
Ext
ra
ct
ing
the
Li
ver
and
Tum
or
from
Abdo
m
ina
l
CT
Im
a
ges
”
,
I
EE
E
,
F
i
ft
h
Inte
rnational
Co
nfe
renc
e
on
Sign
al
and
Imag
e
Pr
oce
ss
ing
(ICSIP
)
"
,
pp.
122
-
125,
2
014.
[4]
Nari
nder
Singh
and
SB
Singh,
“
A
Modifie
d
Mea
n
Gra
y
W
olf
Optimiza
t
ion
Approac
h
for
Benc
hm
ark
and
Biom
edi
cal
Prob
le
m
s”,
Ev
olu
ti
on
ary
Bi
o
inf
orm
atics
,
Volum
e
13
,
p
p
1
–
28,
2017
.
[5]
Russ
o
F.
“
A
m
et
hod
for
esti
m
at
ion
and
filteri
ng
of
Gauss
ia
n
noise
in
images
”
.
IEEE
Tr
ansacti
ons
o
n
Instrum
ent
ati
on
and
Me
asur
eme
nt
.
52(4)
:
1148
-
5
4,
Aug 2003
.
[6]
Sonali
Pat
il
VR
.
“
Udupi
Prepro
ce
ss
ing
to
b
e
c
onsidere
d
f
or
MR
and
CT
Im
age
s
Containing
Tumors
”
.
IOS
R
Journal
of
Elec
t
rical
and
Elec
tronic
s E
ng
ine
erin
g
.
Vol
1(4):
54
-
7,
2012
.
[7]
C
Chaux L, Duval
A,
Ben
azza
-
B
en
y
ahia and
Pes
quet
JC.
“
A
nonl
ine
ar
Stei
n
-
b
ase
d
esti
m
at
or
for m
ult
ic
hann
el
image
Denoising
”
.
I
EEE
Tr
ansacti
ons on
S
ignal
Proce
ss
ing
,
56
(8
):
3855
-
70,
2008
.
[8]
Nara
in
Ponra
j
D,
Eva
ng
el
in
Jenife
r
M,
Poon
godi
P,
Sam
uel
Manoha
ran
J
.
“
A
Survey
on
t
he
Preproc
essing
Te
chn
ique
s
of
Mam
m
ogra
m
fo
r
the
Dete
ction
of
Brea
st
Canc
e
r
”
.
Journal
of
E
merging
Tr
ends
in
Computing
and
Information
Sc
ience
s
.
D
ec
ember
;
2(12)
;
2011
.
[9]
S.
Gunasundari
and
M.
Sugan
y
a
Ananthi
,
“
Compa
rison
and
Eva
l
uat
ion
of
Metho
ds
for
Li
ver
Tumor
Cla
ss
ifi
ca
tion
from
CT
Dat
ase
t
”
,
In
te
rnat
ional
Journal
of
Computer
App
li
ca
ti
on
s
,
Volum
e
39
–
No.
18,
2012.
[10]
K.
Mala,
V.
Sa
dasiva
m
,
and
S.
Alag
appa
n
,
“
Neura
l
Ne
twork
base
d
T
ext
ur
e
Anal
y
s
is
of
Liver
Tumor
fro
m
Com
pute
d
Tomograph
y
Im
age
s
”
,
In
te
rnationa
l
Journal
of
Bi
o
l
ogic
al
,
B
iomedical
and
Me
di
cal
Scienc
es
,
Vol
.
2
Iss
ue
1,
pp
33
-
3
7,
2007
.
[11]
Ja
y
an
thi,
M.,
Kanm
ani
,
B.
,
“
Ext
ra
ct
ing
the
Li
ver
and
Tum
or
from
Abdo
m
ina
l
CT
Im
a
ges
”
,
I
EE
E
,
F
i
ft
h
Inte
rnational
Co
nfe
renc
e
on
Sign
al
and
Imag
e
Pr
oce
ss
ing
(ICSIP
),
pp
.
122
-
125
,
2
014.
[12]
Kum
ar
SS
,
Moni
RS
,
R
aj
e
esh
,
I
.
“
Autom
at
ic
li
v
er
and
l
esion
se
gm
ent
at
ion:
a
pr
imar
y
step
in
d
i
agnosis
of
l
ive
r
disea
ses Signa
l,
Im
age
and
Vide
o
Proce
ss
ing
”
.
2
011
Marc
h
31.
[13]
A
hm
ed
Fouad
Ali,
Abdalla
Mos
t
afa
,
Geha
d
Ism
ai
l
Sa
y
e
d,
Moha
m
ed
Abd
El
fa
tta
h,
Aboul
El
l
a
H
assanie
n
,
“
Natur
e
Inspi
red
Optimiz
at
ion
Algorit
hm
s for
CT
L
ive
r
se
gm
ent
at
ion
”,
20
16.
[14]
Gehe
d
Ism
ai
l
Say
ed
,
Aboul
El
la
Hass
an
and
Ger
al
d
Scahe
f
er,
“
An
aut
om
at
ed
co
m
pute
r
ai
ded
diagnos
is
sy
stem
for
abdominal
C
T
l
i
ver
images”
,
In
t
ernati
onal
confe
renc
e
on
medic
a
l
imaging
,
under
standin
g
and
an
aly
sis
(
El
sev
e
ir
),
pp
68
-
73,
2016.
[15]
Mitt
al
N,
Singh
U,
Singh
Sohi
B.
“
Modifie
d
gr
e
y
opti
m
izer
for
globa
l
engi
n
ee
r
ing
opti
m
izati
on
”
.
App
l
Comput
Inte
l
Soft Comput
,
1
–
16
,
2016
.
[16]
Ed
y
Frad
ina
t
a,
Sakesun
Suthuma
non,
Suntia
m
or
ntut
,
“
Init
i
al
opt
imal
par
amet
ers
of
art
ificial
neu
ral
net
work
an
d
SVR
”
,
Inte
rnati
onal
Journal
of
El
e
ct
rica
l
and
Computer
Enginee
ring
(
IJE
CE
)
,
Volum
e
8,
Iss
ue
No
5,
2018.
htt
p://doi.
o
rg/10.11591/i
jece
.
v8
i5
.
pp%25p
[17]
Rat
na
Niti
n
Pat
il
,
Dr Sharva
vi
Chandra
shekha
r
T
a
m
ane
,
“
A c
om
pa
rat
iv
e
ana
l
y
sis o
n
the
eva
lu
at
ion of c
la
ss
ifi
c
at
ion
al
gorit
hm
in
the
pre
diction
of
di
abe
t
es”
,
In
te
rnat
ional
Journal
o
f
El
e
ct
rica
l
and
Computer
Engi
n
ee
ring
(
IJE
CE
)
,
Vol
8
,
Iss
ue
5
,
2
01
8.
ht
tp:
/
/doi
.
o
r
g/10.
11591/
ij
e
ce.v8i
6.
pp%25p
[18]
Ja
y
an
thi,
M.,
K
anmani,
B.
,
“
E
xtra
c
ti
ng
L
ive
r
and
Tumor
fro
m
Com
pute
r
Tomograph
y
Im
a
ges
Us
ing
Hy
br
id
Te
chn
ique
s
”
,
In
te
rnational
Journal
of
Innov
at
io
ns
&
Adv
ance
ment
in
Compute
r
Sci
ence
,
Volu
m
e:
4(
3
)
:
12
-
17
;
2015
.
[19]
Tomasi
C,
Mand
uchi
R
,
“
Bil
a
te
r
a
l
Filt
eri
ng
Tomasi
C,
Mandu
chi
R.
1998,
Bil
at
er
al
Fil
te
ring
for
G
ra
y
and
Im
age
s
”
.
India
,
Bom
ba
y
:
Proce
ed
ings o
f
t
he
IE
EE
Inte
rna
ti
onal
Con
fe
ren
c
e
on
Comput
er
V
ision
.
p
.
834
-
46
,
1998.
[20]
Kum
ar
SS
,
Moni
RS
,
R
aj
e
esh
,
I
.
“
Autom
at
ic
li
v
er
and
l
esion
se
gm
ent
at
ion:
a
pr
imar
y
step
in
d
i
agnosis
of
l
ive
r
disea
ses Signa
l,
Im
age
and
Vide
o
Proce
ss
ing
”
.
Marc
h
31
,
2011
.
[21]
S.
Mirja
lili
,
S.
Sare
m
i,
S.M.
Mirj
al
ili,
and
L.
D.S
.
Coel
ho,
“
Multi
-
obje
c
ti
ve
gr
e
y
wolf
opti
m
iz
er
:
A
novel
al
gor
it
h
m
for
m
ult
ic
r
it
e
rio
n
opti
m
izati
on
”
,
Ex
pert
Syst
.
App
l.
,
vol
.
47
,
pp
.
10
6
–
119,
Apr 2016
.
[22]
Mirja
lili
,
an
d
A.
Le
wis,
“
Gre
y
w
olf
opt
imize
r
”
,
A
dv.
Eng
.
So
ft
w
.
v
ol.
69
,
pp
.
46
–
61
,
Mar
.
2014
.
[23]
M.
Ja
y
ant
h
i,
“
Segm
ent
ation
of
Li
ver
Abnorm
al
ity
base
d
on
L
abe
l
Connecte
d
Com
ponent
Al
gorit
hm
”,
In
te
r
.
Journal
of
Scien
ti
fic
Engi
ne
ering
and
Techno
logy
.
Volum
e: 6, Iss
ue
:
7
,
PP
:
247
-
2
49
;
2017
.
[24]
The
ca
n
ce
r
Im
agi
ng
Archi
v
e
(
TC
IA),
htt
p
:/
/www
.
ca
nc
eri
m
agi
ng
ar
chi
ve
.
net/
[25]
M.
Ja
y
an
thi,
"Com
par
at
ive
Stud
y
of
Diffe
r
ent
T
ec
hniqu
es
Us
ed
for
m
e
dic
al
im
age
segm
ent
a
ti
o
n
of
Li
ver
from
Abdom
ina
l
CT S
ca
n",
I
EE
E
Wi
SPNE
T 2016
co
nfe
ren
c
e; pp.
14
62
-
1465,
2016
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5061
-
5070
5070
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
M.
Ja
y
ant
hi
is
Senior
As
sistant
Profess
or
in
D
epa
rtment
of
Elec
tron
ic
s
and
C
om
m
unic
at
ion
Engi
ne
eri
ng,
N
ew
Horiz
on
Co
ll
eg
e
Of
Engi
n
ee
ring
,
Beng
al
u
ru
and
Karna
t
a
ka,
Indi
a.
She
rec
e
ive
d
h
er
M.
E
degr
e
e
in
Co
m
m
unic
at
ion
S
y
stems
from
Mepc
o
Schle
nk
Enginee
ring
col
l
e
ge
,
Sivaka
si,
T
amiln
adu
in
the
y
ear 2006.
She
is pur
suing t
he
Ph.D de
gre
e
in
the
Dep
artm
ent
of
ECE
,
BMS
col
le
ge
of
Engi
ne
eri
ng,
B
e
ngal
uru.
Her
res
ea
rch
in
te
r
ests
inc
lude
image
pro
ce
ss
ing,
Digital
Signal
proc
essing
and
Biom
edi
c
al
signa
l
proc
ess
ing.
She
has
1
1
y
e
ars
of
exp
eri
e
nce
in
te
a
chi
ng
.
Email
:ja
y
an
thi
.
s
at
hishkum
ar@l
i
ve.
com
B
K
a
nm
a
n
i
,
Ph
.
D
i
s
P
r
o
f
e
ss
or
a
n
d
D
e
a
n
Ac
a
d
e
m
i
c
s
,
D
e
p
a
r
t
m
e
n
t
o
f
T
e
l
e
c
o
m
m
u
n
i
c
a
t
i
on
E
n
g
i
n
e
e
r
i
n
g
,
B
M
S
c
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
B
e
n
g
a
l
u
r
u
.
S
h
e
r
e
c
e
i
v
e
d
h
e
r
B
a
c
h
e
l
o
r
s
i
n
E
l
e
c
t
r
o
n
i
c
s
a
n
d
C
o
m
m
u
n
i
c
ati
o
n
E
n
g
i
n
e
e
r
i
n
g
f
r
o
m
N
a
g
a
r
u
j
u
na
U
n
i
v
e
r
s
i
t
y
i
n
1
9
8
7
,
M
.
T
e
c
h
.
d
e
g
r
e
e
i
n
D
i
g
i
t
a
l
c
o
m
m
u
n
i
c
a
t
i
o
n
f
r
o
m
I
n
d
i
a
n
In
s
tit
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
K
a
n
p
u
r
i
n
1
9
9
0
,
a
n
d
P
hD
fr
o
m
t
h
e
I
n
d
i
an
I
n
s
t
i
t
u
t
e
o
f
S
c
i
en
c
e
B
a
n
g
a
l
o
r
e
(
I
I
S
c
)
i
n
t
h
e
y
e
a
r
2
0
0
6
.
S
h
e
h
a
s
b
e
e
n
w
i
t
h
BM
S
Co
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
B
a
n
g
a
l
o
r
e
,
s
i
n
c
e
1
9
9
5
,
a
n
d
h
a
s
t
o
h
e
r
c
r
e
d
i
t
2
5
I
n
t
e
r
n
a
t
i
o
n
a
l
p
u
b
l
i
c
a
t
i
o
n
s
,
o
f
w
h
i
c
h
t
h
r
e
e
a
r
e
i
n
J
o
ur
n
a
l
s
.
S
h
e
i
s
a
su
p
e
r
v
i
s
o
r
f
or
t
he
P
h
.
D
s
t
ud
e
n
t
a
t
V
i
s
v
e
s
v
a
r
a
y
a
T
e
c
h
n
o
l
o
g
i
c
a
l
u
n
i
v
e
r
s
i
t
y,
B
e
l
a
g
a
v
i
.
S
h
e
t
e
a
c
h
e
s
u
n
d
e
r
-
g
r
a
d
u
a
t
e
c
o
u
r
s
e
s
o
n
A
n
alo
g
S
i
g
n
a
l
P
r
o
ces
s
i
n
g
,
D
i
g
i
t
a
l
S
i
g
n
a
l
P
r
o
c
e
ss
i
ng
,
A
n
a
l
o
g
C
o
m
m
u
n
i
c
a
t
i
o
n
a
n
d
D
i
g
i
t
a
l
C
o
m
m
un
i
c
a
t
i
o
n
.
S
h
e
i
s
S
e
n
i
o
r
M
e
m
b
e
r
I
E
E
E
,
M
e
m
b
e
r
T
a
t
a
L
i
b
r
a
r
y
a
n
d
L
i
f
e
M
e
m
b
e
r
I
S
T
E
.
Evaluation Warning : The document was created with Spire.PDF for Python.