I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
,
p
p
.
4
0
8
0
~
4
0
9
4
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
13
.i
4
.
p
p
4
0
8
0
-
4
0
9
4
4080
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
H
y
brid
im
pro
v
ed f
uzzy
C
-
mea
ns
a
nd watersh
ed seg
menta
tion
to cla
ss
ify
Alzhei
mer’s
using
deep
lea
rning
E
s
ra
a
H
.
Ali
1,
2
,
Sa
wsa
n Sa
dek
1
,
Z
a
id F
.
M
a
kk
i
3
1
D
o
c
t
o
r
a
l
S
c
h
o
o
l
o
f
S
c
i
e
n
c
e
s
a
n
d
Te
c
h
n
o
l
o
g
y
,
L
e
b
a
n
e
se
U
n
i
v
e
r
si
t
y
,
B
e
i
r
u
t
,
Le
b
a
n
o
n
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
C
o
l
l
e
g
e
o
f
S
c
i
e
n
c
e
,
A
l
-
N
a
h
r
a
i
n
U
n
i
v
e
r
si
t
y
,
B
a
g
h
d
a
d
,
I
r
a
q
3
Al
-
N
a
h
r
a
i
n
C
e
n
t
e
r
f
o
r
S
t
r
a
t
e
g
i
c
S
t
u
d
i
e
s
,
B
a
g
h
d
a
d
,
I
r
a
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
1
9
,
2
0
2
4
R
ev
is
ed
Ma
r
2
4
,
2
0
2
4
Acc
ep
ted
J
u
n
1
,
2
0
2
4
Bra
in
d
a
m
a
g
e
a
n
d
d
e
ficits
in
in
te
ra
c
ti
o
n
s
a
m
o
n
g
b
ra
i
n
c
e
ll
s
a
re
th
e
p
rima
ry
c
a
u
se
s o
f
d
e
m
e
n
ti
a
a
n
d
Alz
h
e
ime
r
’
s d
ise
a
se
(AD
)
.
De
sp
it
e
o
n
g
o
i
n
g
re
se
a
rc
h
,
n
o
e
ffe
c
ti
v
e
m
e
d
ica
ti
o
n
s
h
a
v
e
y
e
t
b
e
e
n
d
e
v
e
l
o
p
e
d
f
o
r
t
h
e
se
c
o
n
d
it
io
n
s
.
Th
e
re
fo
re
,
e
a
rly
d
e
tec
ti
o
n
is
c
ru
c
ial
fo
r
m
a
n
a
g
in
g
t
h
e
p
ro
g
re
ss
io
n
o
f
th
e
se
d
iso
rd
e
rs.
In
t
h
is
st
u
d
y
,
we
i
n
tro
d
u
c
e
a
n
o
v
e
l
t
o
o
l
f
o
r
d
e
tec
ti
n
g
AD
u
sin
g
n
o
n
-
in
v
a
siv
e
m
e
d
ica
l
tes
ts,
su
c
h
a
s
m
a
g
n
e
ti
c
re
so
n
a
n
c
e
ima
g
i
n
g
(M
RI).
Ou
r
m
e
th
o
d
e
m
p
lo
y
s
fu
z
z
y
C
-
m
e
a
n
s
c
lu
ste
rin
g
to
id
e
n
ti
f
y
fe
a
tu
re
s
t
h
a
t
e
n
h
a
n
c
e
ima
g
e
a
c
c
u
ra
c
y
.
Th
e
sta
n
d
a
rd
fu
z
z
y
C
-
m
e
a
n
s
a
lg
o
rit
h
m
h
a
s
b
e
e
n
a
u
g
m
e
n
te
d
with
f
u
z
z
y
c
o
m
p
o
n
e
n
ts
to
imp
ro
v
e
c
lu
ste
ri
n
g
p
e
rfo
rm
a
n
c
e
.
Th
is
e
n
h
a
n
c
e
d
a
p
p
ro
a
c
h
o
p
ti
m
ize
s
se
g
m
e
n
tatio
n
b
y
e
x
trac
ti
n
g
ima
g
e
in
f
o
rm
a
ti
o
n
a
n
d
u
ti
li
z
i
n
g
a
sli
d
in
g
win
d
o
w
t
o
c
a
lcu
late
c
e
n
ter
c
o
o
rd
i
n
a
tes
a
n
d
e
sta
b
li
sh
a
sta
b
le
g
ro
u
p
m
a
tri
x
.
Th
e
se
c
rit
ica
l
fe
a
tu
re
s
a
re
su
b
se
q
u
e
n
tl
y
in
te
g
ra
ted
with
a
two
-
p
h
a
se
wa
ters
h
e
d
se
g
m
e
n
tatio
n
p
ro
c
e
ss
.
Th
e
re
su
lt
i
n
g
se
g
m
e
n
t
e
d
ima
g
e
s
a
re
th
e
n
u
se
d
to
train
a
n
o
p
ti
m
a
l
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
(CNN
)
fo
r
AD
c
las
sifica
ti
o
n
.
O
u
r
m
e
th
o
d
o
lo
g
y
d
e
m
o
n
stra
ted
a
9
8
.
2
0
%
a
c
c
u
ra
c
y
ra
te
in
th
e
d
e
tec
ti
o
n
a
n
d
c
las
sifica
ti
o
n
o
f
se
g
m
e
n
ted
M
RI
b
ra
i
n
ima
g
e
s,
h
i
g
h
li
g
h
t
in
g
it
s
e
ffica
c
y
in
id
e
n
ti
fy
in
g
d
ise
a
se
ty
p
e
s.
K
ey
w
o
r
d
s
:
Alzh
eim
er
’
s
d
is
ea
s
e
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
I
m
p
r
o
v
ed
f
u
zz
y
C
-
m
ea
n
s
Ma
g
n
etic
r
eso
n
an
ce
im
a
g
in
g
W
ater
s
h
ed
s
eg
m
en
tatio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
E
s
r
aa
H.
Ali
Do
cto
r
al
Sch
o
o
l
o
f
Scien
ce
s
a
n
d
T
ec
h
n
o
lo
g
y
,
L
e
b
an
ese
Un
i
v
er
s
ity
Al
-
Had
ath
Dis
tr
ict,
B
eir
u
t,
L
eb
an
o
n
E
m
ail:
esra
a.
ali@
u
l.e
d
u
.
ib
1.
I
NT
RO
D
UCT
I
O
N
Alzh
eim
er
’
s
d
is
ea
s
e
(
AD)
is
th
e
lead
in
g
ca
u
s
e
o
f
d
em
en
tia
am
o
n
g
o
ld
er
ad
u
lts
,
ch
a
r
ac
te
r
ized
as
a
m
en
tal
h
ea
lth
d
is
o
r
d
er
th
at
r
esu
lts
in
b
r
ain
d
a
m
ag
e
a
n
d
im
p
a
ir
s
th
e
ab
ilit
y
to
p
er
f
o
r
m
d
aily
ac
tiv
ities
[
1
]
.
I
t
is
a
ch
r
o
n
ic
n
e
u
r
o
d
eg
en
e
r
ativ
e
c
o
n
d
itio
n
with
an
i
n
s
id
io
u
s
o
n
s
et
an
d
g
r
ad
u
ally
wo
r
s
en
in
g
s
y
m
p
to
m
s
o
v
er
tim
e.
T
h
e
etio
lo
g
y
o
f
AD
r
em
ai
n
s
u
n
clea
r
,
a
n
d
tr
ea
t
m
en
ts
ar
e
o
f
ten
ex
p
e
n
s
iv
e.
I
n
r
ec
en
t
y
ea
r
s
,
th
er
e
h
as
b
ee
n
a
s
ig
n
if
ican
t
f
o
cu
s
o
n
ea
r
ly
d
A
D
is
t
h
e
lead
in
g
ca
u
s
e
o
f
d
em
e
n
tia
am
o
n
g
o
ld
er
ad
u
lts
,
ch
ar
a
cter
ized
as
a
m
en
tal
h
ea
lth
d
is
o
r
d
er
th
at
r
esu
lts
in
b
r
ain
d
am
a
g
e
an
d
im
p
air
s
th
e
ab
ilit
y
to
p
er
f
o
r
m
d
aily
ac
tiv
i
ties
.
I
t
is
a
ch
r
o
n
ic
n
eu
r
o
d
eg
en
e
r
ativ
e
co
n
d
itio
n
with
an
in
s
id
io
u
s
o
n
s
et
an
d
g
r
ad
u
ally
w
o
r
s
en
in
g
s
y
m
p
t
o
m
s
o
v
er
tim
e.
T
h
e
etio
lo
g
y
o
f
AD
r
em
ain
s
u
n
clea
r
,
an
d
tr
ea
tm
en
ts
a
r
e
o
f
ten
e
x
p
en
s
iv
e.
I
n
r
ec
en
t
y
ea
r
s
,
t
h
er
e
h
as
b
ee
n
a
s
ig
n
if
ican
t
f
o
cu
s
o
n
ea
r
ly
d
etec
tio
n
o
f
th
is
f
o
r
m
o
f
d
em
en
tia
b
y
ac
ad
em
ics
an
d
r
esear
ch
er
s
.
T
h
e
cu
r
r
en
t
g
lo
b
al
d
em
o
g
r
a
p
h
i
c
o
f
in
d
iv
id
u
als
with
d
em
en
tia
is
esti
m
ated
to
b
e
4
7
.
5
m
illi
o
n
,
p
r
o
jecte
d
to
in
cr
ea
s
e
to
7
5
m
illi
o
n
b
y
2
0
3
0
[
2
]
,
[
3
]
.
T
h
e
ad
v
an
ce
m
en
t
o
f
d
ig
ital
n
eu
r
o
im
a
g
in
g
tech
n
iq
u
es
h
as
en
h
an
ce
d
th
e
an
aly
s
is
o
f
clin
ical
im
ag
in
g
d
ata
f
o
r
d
iag
n
o
s
in
g
b
r
ain
d
is
o
r
d
er
s
.
T
ec
h
n
iq
u
es
s
u
ch
as
m
ag
n
etic
r
eso
n
an
c
e
im
ag
in
g
(
MRI)
,
ce
r
eb
r
o
s
p
in
al
f
lu
id
(
C
SF
)
an
aly
s
is
,
s
in
g
le
p
h
o
to
n
em
is
s
io
n
co
m
p
u
ted
t
o
m
o
g
r
ap
h
y
(
SP
E
C
T
)
,
an
d
f
lu
o
r
o
d
eo
x
y
g
lu
c
o
s
e
p
o
s
itro
n
em
is
s
io
n
to
m
o
g
r
a
p
h
y
(
FDG
-
PET
)
ar
e
in
s
tr
u
m
e
n
tal
in
id
en
tify
in
g
s
tr
u
ct
u
r
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
imp
r
o
ve
d
fu
z
z
y
C
-
mea
n
s
a
n
d
w
a
ters
h
ed
s
eg
men
ta
tio
n
to
cla
s
s
ify
A
lz
h
eime
r
’
s
…
(
E
s
r
a
a
H.
A
li
)
4081
ch
an
g
es
in
th
e
b
r
ain
.
Sin
ce
th
e
ea
r
ly
1
9
8
0
s
,
th
e
m
ed
ical
co
m
m
u
n
ity
h
as
b
eg
u
n
u
tili
zin
g
th
e
s
e
ad
v
an
ce
d
m
ed
ical
im
ag
in
g
tech
n
i
q
u
es
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
h
ea
lth
ca
r
e
im
ag
er
y
[
4
]
,
[
5
]
.
T
h
e
ev
o
lu
tio
n
o
f
tr
ad
itio
n
al
im
ag
e
p
r
o
ce
s
s
in
g
m
eth
o
d
s
alo
n
g
s
id
e
m
ac
h
in
e
lear
n
in
g
(
ML
)
a
n
d
d
ee
p
lear
n
in
g
(
DL
)
h
as
led
to
s
ig
n
if
ican
t
ad
v
an
ce
m
e
n
ts
in
m
e
d
ical
im
ag
e
an
aly
s
is
.
An
aly
tical
im
a
g
e
p
r
o
ce
s
s
in
g
tech
n
i
q
u
es
ar
e
ca
teg
o
r
ized
in
to
r
eg
is
tr
atio
n
,
class
if
icatio
n
,
d
etec
tio
n
,
s
eg
m
en
tatio
n
,
a
n
d
lo
ca
lizatio
n
,
with
s
eg
m
en
tatio
n
b
e
in
g
a
cr
u
cial
s
tep
i
n
is
o
latin
g
th
e
d
esi
r
ed
tis
s
u
e
o
r
r
eg
io
n
o
f
in
ter
est
(
R
o
I
)
f
r
o
m
t
h
e
co
llected
im
ag
es
[
6
]
,
[
7
]
.
T
h
e
m
o
d
el
a
r
c
h
i
tec
tu
r
e
f
o
r
d
ia
g
n
o
s
in
g
AD
e
n
c
o
m
p
ass
es
f
i
v
e
k
e
y
s
ta
g
es:
d
a
ta
a
cq
u
is
i
ti
o
n
,
s
eg
m
e
n
t
ati
o
n
,
r
e
g
is
tr
ati
o
n
,
m
o
r
p
h
o
m
e
tr
y
,
a
n
d
cl
ass
if
ic
ati
o
n
.
V
ar
i
o
u
s
s
ta
n
d
a
r
d
d
at
asets
,
s
u
c
h
Alz
h
ei
m
e
r
'
s
d
is
ea
s
e
n
e
u
r
o
im
ag
in
g
in
iti
ati
v
e
(
A
D
NI
)
,
in
te
r
n
ati
o
n
a
l
c
o
n
s
o
r
t
i
u
m
f
o
r
b
r
a
in
m
ap
p
i
n
g
(
I
C
B
M)
,
m
in
im
al
in
te
r
v
al
r
es
o
n
a
n
c
e
im
ag
in
g
i
n
Alz
h
eim
er
'
s
d
is
ea
s
e
(
M
I
R
I
A
D)
,
Ka
g
g
l
e
,
o
p
en
a
cc
ess
s
er
ies
o
f
i
m
a
g
i
n
g
s
t
u
d
i
es
(
OAS
I
S
)
,
Har
v
a
r
d
Me
d
ic
al
Sc
h
o
o
l
,
a
n
d
o
t
h
e
r
s
,
a
r
e
u
tili
ze
d
to
g
at
h
e
r
e
x
te
n
s
i
v
e
d
at
a
f
r
o
m
m
o
r
p
h
o
lo
g
ical
a
n
d
a
n
at
o
m
i
c
a
l
im
a
g
es
.
T
h
es
e
im
a
g
es
a
r
e
ess
e
n
ti
al
f
o
r
i
d
e
n
ti
f
y
i
n
g
ab
n
o
r
m
a
li
ties
i
n
th
e
a
f
f
ec
t
e
d
b
r
ai
n
.
E
f
f
e
cti
v
e
s
e
g
m
e
n
ta
ti
o
n
an
d
class
if
ica
ti
o
n
,
p
ar
t
ic
u
l
ar
l
y
i
n
MRI
s
t
u
d
ies
,
n
e
ce
s
s
it
at
e
a
r
o
b
u
s
t
im
a
g
e
p
r
e
-
p
r
o
c
ess
i
n
g
a
p
p
r
o
ac
h
.
I
n
th
e
d
e
v
e
lo
p
m
e
n
t
o
f
A
lz
h
e
im
er
'
s
d
ete
cti
o
n
s
y
s
te
m
s
,
p
r
o
c
ess
es
s
u
ch
as
n
o
is
e
r
e
d
u
cti
o
n
,
s
m
o
o
t
h
i
n
g
,
s
k
u
ll
s
t
r
i
p
p
in
g
,
cr
o
p
p
in
g
,
a
n
d
n
o
r
m
a
liz
ati
o
n
a
r
e
i
n
d
is
p
e
n
s
a
b
le
.
T
h
r
o
u
g
h
th
e
r
e
g
is
t
r
a
ti
o
n
p
r
o
c
ess
,
im
ag
es
ar
e
a
lig
n
e
d
to
a
s
ta
n
d
a
r
d
r
e
f
e
r
e
n
c
e
a
r
e
a,
f
ac
ilit
ati
n
g
i
n
t
r
a
-
i
m
a
g
e
an
d
i
n
t
er
-
i
m
a
g
e
m
at
c
h
i
n
g
c
r
u
cial
f
o
r
t
r
a
c
k
i
n
g
d
is
ea
s
e
p
r
o
g
r
ess
i
o
n
a
n
d
id
e
n
t
if
y
i
n
g
a
f
f
ec
te
d
i
n
d
iv
id
u
a
l
s
[
8
]
.
C
l
ass
i
f
ic
ati
o
n
,
t
h
e
f
i
n
al
s
tag
e,
i
n
v
o
l
v
es
c
at
eg
o
r
iz
in
g
p
atie
n
ts
as
n
o
r
m
al
o
r
ex
h
i
b
it
in
g
ab
n
o
r
m
a
liti
es.
A
r
t
if
icia
l
i
n
tel
li
g
e
n
c
e
(
AI
)
,
in
co
n
j
u
n
ct
io
n
w
it
h
MRI
,
e
m
e
r
g
es
as
a
p
r
o
m
is
i
n
g
m
et
h
o
d
f
o
r
d
is
ea
s
e
cl
ass
i
f
ic
ati
o
n
.
T
h
e
d
e
v
el
o
p
m
e
n
t
a
n
d
a
p
p
li
ca
ti
o
n
o
f
ML
a
n
d
DL
a
r
e
p
iv
o
t
al
i
n
c
r
a
f
t
in
g
AI
-
b
ase
d
class
i
f
i
ca
t
io
n
al
g
o
r
it
h
m
s
t
h
a
t
e
n
h
a
n
c
e
o
u
tc
o
m
es
,
q
u
al
it
y
,
a
n
d
ef
f
ic
ie
n
c
y
[
9
]
.
I
n
v
ar
io
u
s
s
tu
d
ies,
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN
)
-
b
ased
lear
n
in
g
h
as
b
ee
n
f
o
u
n
d
to
lac
k
r
o
b
u
s
tn
ess
,
p
r
o
m
p
tin
g
th
e
ex
p
lo
r
atio
n
o
f
alter
n
ativ
e
m
eth
o
d
s
to
en
h
an
ce
p
er
f
o
r
m
an
ce
.
Fo
r
MRI
b
r
ain
im
ag
es,
a
h
y
b
r
id
ap
p
r
o
ac
h
co
m
b
in
in
g
en
h
an
ce
d
f
u
zz
y
C
-
m
ea
n
s
clu
s
t
er
in
g
with
wate
r
s
h
ed
s
eg
m
en
t
atio
n
(
W
s
)
h
as
b
ee
n
u
tili
ze
d
as
a
f
ea
tu
r
e
d
etec
tio
n
an
d
ex
tr
ac
tio
n
m
ec
h
a
n
is
m
to
d
elin
ea
te
g
r
ay
m
atter
(
GM
)
,
wh
ite
m
atter
(
W
M)
,
an
d
C
SF
r
eg
io
n
s
o
f
th
e
b
r
ain
.
Ou
r
liter
atu
r
e
r
ev
iew
r
e
v
ea
led
th
at
lim
ited
r
esear
ch
h
as
b
ee
n
co
n
d
u
cted
o
n
d
ev
elo
p
in
g
s
p
ec
ialized
C
NN
ar
ch
itectu
r
es f
o
r
m
o
r
e
e
f
f
ec
tiv
e
AD.
T
h
e
tech
n
iq
u
e
p
r
o
p
o
s
ed
in
[
1
0
]
in
itiates
with
a
g
en
etic
alg
o
r
ith
m
(
GA)
f
o
r
f
ea
tu
r
e
s
elec
tio
n
,
id
en
tify
in
g
th
e
m
o
s
t
in
f
o
r
m
ati
v
e
s
u
b
s
et
o
f
f
ea
tu
r
es.
Fu
zz
y
C
-
m
ea
n
s
(
FC
M)
clu
s
ter
in
g
is
ap
p
lied
to
th
is
s
elec
ted
s
u
b
s
et.
T
h
is
ap
p
r
o
ac
h
r
e
d
u
ce
s
th
e
d
im
e
n
s
io
n
ality
o
f
th
e
f
e
atu
r
e
s
p
ac
e,
th
er
eb
y
r
en
d
e
r
in
g
th
e
class
if
icatio
n
p
r
o
ce
s
s
v
ia
s
u
p
p
o
r
t
v
ec
to
r
m
a
ch
in
e
(
SVM
)
b
o
t
h
m
o
r
e
ef
f
ici
en
t
an
d
u
n
d
er
s
tan
d
ab
le.
I
t
n
o
t
ab
ly
e
n
h
an
ce
s
t
h
e
ac
cu
r
ac
y
o
f
ea
r
ly
AD
d
etec
tio
n
b
y
ac
ce
n
tu
atin
g
th
e
d
if
f
er
en
tiatio
n
b
etwe
en
AD
a
n
d
n
o
n
-
AD
clu
s
ter
s
.
T
h
e
r
esu
lts
u
n
d
er
s
co
r
e
t
h
e
ef
f
icac
y
o
f
th
is
tech
n
iq
u
e
in
p
r
ec
is
el
y
id
en
tify
in
g
in
d
iv
id
u
als
at
r
ick
o
f
Alzh
eim
er
'
s
a
t
an
ea
r
ly
s
tag
e.
T
o
tr
ac
k
AD
p
r
o
g
r
ess
io
n
,
Sap
p
ag
h
et
a
l.
[
1
1
]
in
tr
o
d
u
ce
d
a
m
u
lti
-
m
o
d
al
en
s
e
m
b
le
DL
tech
n
iq
u
e
th
at
ex
tr
ac
ted
b
o
th
l
o
ca
l a
n
d
l
o
n
g
itu
d
i
n
al
in
f
o
r
m
ati
o
n
f
r
o
m
ea
ch
m
o
d
ality
.
Ad
d
itio
n
ally
,
p
r
io
r
k
n
o
wled
g
e
was
u
tili
ze
d
to
d
er
iv
e
lo
ca
l
f
ea
t
u
r
es
f
o
r
m
MRI,
p
o
s
itro
n
e
m
is
s
io
n
to
m
o
g
r
ap
h
y
(
PET
)
,
co
g
n
itiv
e
s
co
r
es,
n
eu
r
o
p
ath
o
lo
g
y
,
an
d
ADNI
ass
es
s
m
en
ts
.
E
m
p
lo
y
in
g
a
c
o
m
b
in
atio
n
o
f
lay
e
r
ed
C
NN
-
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
iLST
M
)
,
all
g
ath
er
e
d
f
ea
tu
r
es
wer
e
in
teg
r
ated
f
o
r
r
eg
r
ess
io
n
an
d
cla
s
s
if
icatio
n
task
s
.
A
m
u
lti
-
m
o
d
al
ap
p
r
o
ac
h
f
o
r
au
to
m
ated
h
ip
p
o
ca
m
p
u
s
s
eg
m
en
ta
tio
n
u
s
in
g
3
D
p
atch
es
was
d
etailed
[
1
2
]
.
Utilizin
g
s
MRI
(
T
1
)
im
ag
es
f
r
o
m
th
e
ADNI
d
ataset,
a
h
y
b
r
id
m
u
lti
-
task
d
ee
p
C
NN
an
d
3
D
Den
s
eNe
t+so
f
tm
ax
wer
e
em
p
lo
y
ed
f
o
r
AD
class
if
icatio
n
.
T
h
e
m
o
d
el
ac
h
iev
ed
an
ac
c
u
r
ac
y
o
f
8
8
%,
s
en
s
itiv
ity
8
6
%
,
an
d
an
ar
ea
u
n
d
e
r
th
e
cu
r
v
e
(
AUC)
o
f
9
2
%
[
1
3
]
.
T
h
is
s
tu
d
y
p
r
esen
ts
a
m
eth
o
d
f
o
r
ea
r
l
y
d
etec
tio
n
o
f
Alzh
e
im
er
'
s
u
s
in
g
SVM
s
tr
ain
ed
o
n
v
ar
io
u
s
tex
tu
r
e
d
e
s
cr
ip
to
r
s
,
wh
ich
aid
in
d
im
en
s
io
n
ality
r
ed
u
ctio
n
d
er
iv
e
d
f
r
o
m
MRI
alo
n
g
s
id
e
SVMs
tr
ain
ed
o
n
m
ar
k
e
r
s
o
b
t
ain
ed
f
r
o
m
ADNI
.
Dif
f
er
e
n
t
f
ea
tu
r
e
s
elec
tio
n
m
et
h
o
d
s
,
ea
c
h
tr
ain
in
g
a
d
is
tin
ct
SVM,
wer
e
ap
p
lied
to
r
ed
u
ce
t
h
e
d
im
en
s
io
n
ality
o
f
v
o
x
el
-
b
as
ed
f
ea
tu
r
es.
Gee
th
a
an
d
Pu
g
az
h
en
th
i
[
1
4
]
s
u
g
g
ests
a
n
o
v
el
ap
p
r
o
ac
h
f
o
r
Alzh
eim
er
'
s
cla
s
s
if
icat
io
n
f
r
o
m
MRIs
u
s
in
g
a
f
u
zz
y
n
eu
r
al
n
etwo
r
k
(
FNN)
.
T
h
e
wav
elet
tr
an
s
f
o
r
m
atio
n
(
W
T
)
is
em
p
lo
y
ed
f
o
r
im
ag
e
d
ec
o
m
p
o
s
itio
n
,
with
th
e
d
is
cr
ete
wav
elet
tr
an
s
f
o
r
m
ca
lc
u
latin
g
th
e
o
u
tp
u
t
co
e
f
f
icien
t
v
ec
to
r
s
.
T
h
e
s
e
g
en
er
ated
f
ea
tu
r
es
ar
e
th
en
u
s
ed
to
tr
ain
FNN.
A
C
NN
m
o
d
el
is
p
r
o
p
o
s
ed
f
o
r
AD
clas
s
if
icatio
n
u
s
in
g
MR
i
m
ag
es
with
h
ip
p
o
ca
m
p
u
s
d
esi
g
n
ated
as
th
e
R
o
I
[
1
5
]
.
T
h
e
R
o
I
is
ex
tr
ac
ted
th
r
o
u
g
h
an
au
to
m
ated
p
atch
-
b
ased
s
ep
ar
atio
n
m
eth
o
d
th
at
u
tili
ze
s
g
eo
m
etr
ic
v
alu
es
f
r
o
m
th
e
i
n
ter
n
atio
n
al
co
n
s
o
r
tiu
m
f
o
r
b
r
ai
n
m
ap
p
i
n
g
(
I
C
B
M)
s
tan
d
ar
d
.
C
NN
was
ap
p
lied
f
o
r
d
ataset
class
if
icatio
n
,
d
em
o
n
s
tr
atin
g
n
o
ta
b
le
p
er
f
o
r
m
an
ce
s
.
A
n
o
v
el
m
eth
o
d
o
lo
g
y
c
o
m
b
in
in
g
ex
tr
em
e
le
ar
n
in
g
an
d
d
ee
p
lear
n
in
g
f
o
r
AD
clas
s
if
icatio
n
is
in
tr
o
d
u
ce
d
[
1
6
]
.
T
h
is
ap
p
r
o
ac
h
ev
alu
ates
two
d
ee
p
lear
n
in
g
m
o
d
els
f
o
r
f
u
n
ctio
n
al
b
r
ain
-
n
e
two
r
k
class
if
icatio
n
,
alo
n
g
s
id
e
an
ex
tr
em
e
lear
n
i
n
g
m
ac
h
in
e
(
E
L
M)
en
h
a
n
ce
d
f
r
am
ewo
r
k
f
o
r
lear
n
in
g
d
ee
p
r
e
g
io
n
al
-
co
n
n
ec
tiv
ity
an
d
d
ee
p
ad
jace
n
t
p
o
s
itio
n
al
f
ea
tu
r
es.
T
h
e
co
n
s
tr
u
ctio
n
o
f
t
h
e
b
r
ain
n
etwo
r
k
u
tili
ze
s
th
e
Pear
s
o
n
co
r
r
elatio
n
co
ef
f
icien
t.
I
n
s
u
m
m
ar
y
,
o
u
r
r
ev
iew
h
ig
h
lig
h
ts
th
r
ee
k
ey
f
in
d
in
g
s
:
−
T
h
e
m
ajo
r
it
y
o
f
th
e
liter
atu
r
e
wer
e
r
ev
iewe
d
r
ep
o
r
ted
e
v
alu
atio
n
s
co
r
es
b
el
o
w
9
2
%,
wh
e
r
ea
s
o
u
r
s
tu
d
y
ac
h
iev
ed
an
ex
ce
p
tio
n
al
p
er
f
o
r
m
an
ce
o
f
9
8
%.
T
h
is
s
ig
n
if
i
ca
n
t
ad
v
an
ce
m
en
t
is
attr
ib
u
te
d
to
o
u
r
n
o
v
el
s
eg
m
en
tatio
n
an
d
f
ea
tu
r
e
ex
tr
a
ctio
n
m
o
d
el,
wh
ich
ef
f
ec
t
iv
ely
r
ed
u
ce
s
v
ar
iab
le
p
ar
a
m
eter
s
wh
ile
en
h
an
cin
g
co
m
p
u
tatio
n
al
s
p
ee
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
0
8
0
-
4
0
9
4
4082
−
Ou
r
r
esear
ch
in
tr
o
d
u
ce
s
a
p
o
t
en
t
class
if
icatio
n
ar
ch
itectu
r
e
th
at
u
tili
ze
d
if
f
er
en
t
k
er
n
el
s
izes
to
ex
tr
ac
t
ess
en
tial
f
ea
tu
r
es.
B
y
in
co
r
p
o
r
atin
g
two
s
m
aller
k
er
n
els
(
3
×
3
an
d
5
×
5
)
,
o
u
r
m
o
d
el
ac
h
iev
es
r
o
b
u
s
t
tr
ain
in
g
an
d
test
in
g
p
r
o
ce
s
s
es,
th
er
eb
y
im
p
r
o
v
in
g
its
p
er
f
o
r
m
an
ce
a
n
d
r
eliab
ilit
y
.
−
Fu
r
th
er
m
o
r
e
,
o
u
r
s
tu
d
y
lev
er
a
g
es
a
s
u
b
s
tan
tial
d
ataset
co
m
p
r
is
in
g
6
4
0
0
b
r
ain
im
ag
es,
ca
teg
o
r
ized
in
to
f
o
u
r
d
is
tin
ct
class
e
s
:
m
ild
,
n
o
r
m
al,
m
o
d
er
ate,
an
d
v
er
y
m
ild
.
T
h
is
ex
ten
s
iv
e
d
ataset
o
f
f
er
s
a
c
o
m
p
r
eh
e
n
s
iv
e
v
iew
ac
r
o
s
s
v
ar
io
u
s
s
tag
es o
f
AD
,
th
er
eb
y
e
n
r
ich
in
g
th
e
d
e
p
th
an
d
d
iv
er
s
ity
o
f
o
u
r
an
aly
s
is
.
2.
M
E
T
H
O
D
I
n
th
is
p
ap
er
,
we
u
tili
ze
th
e
im
p
r
o
v
e
d
f
u
zz
y
C
-
m
ea
n
s
clu
s
ter
in
g
(
I
m
FC
m
)
f
o
r
s
eg
m
en
tin
g
b
r
ain
tis
s
u
e,
o
win
g
to
its
ef
f
icien
cy
in
s
eg
m
en
tin
g
h
o
m
o
g
en
e
o
u
s
in
ten
s
ity
r
eg
io
n
s
o
f
MRI
im
ag
es.
W
e
in
tr
o
d
u
ce
a
h
y
b
r
id
ap
p
r
o
ac
h
c
o
m
b
in
in
g
I
m
FC
m
with
Ws
,
ac
h
iev
in
g
m
o
r
e
ef
f
e
ctiv
e
r
esu
lts
in
ac
cu
r
ately
p
ar
t
itio
n
in
g
im
a
g
es
an
d
en
h
an
cin
g
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
T
h
e
o
u
tco
m
es
o
f
th
ese
two
m
eth
o
d
s
ar
e
th
en
in
te
g
r
ate
d
in
to
an
o
p
tim
ized
C
NN
ar
ch
itectu
r
e,
aim
in
g
t
o
im
p
r
o
v
e
th
e
ac
cu
r
ac
y
a
n
d
r
o
b
u
s
tn
ess
o
f
t
h
e
AD
d
etec
tio
n
s
y
s
tem
.
T
h
e
ADNI
d
ataset
was
em
p
lo
y
ed
to
v
ali
d
ate
th
e
f
in
d
in
g
s
,
with
a
p
p
r
o
x
im
ately
6
4
0
0
MRI
b
r
ain
im
ag
es
an
aly
ze
d
.
T
h
ese
im
ag
es
ar
e
an
n
o
tated
in
to
f
o
u
r
ca
teg
o
r
ies:
m
ild
,
m
o
d
er
ate,
v
e
r
y
m
ild
,
an
d
n
o
r
m
al
.
Fig
u
r
e
1
i
llu
s
tr
ates
th
e
b
lo
ck
d
iag
r
am
o
f
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
.
E
ac
h
s
ec
tio
n
o
f
o
u
r
m
eth
o
d
is
ex
p
lai
n
ed
i
n
d
etail
i
n
th
e
s
u
b
s
eq
u
en
t
s
ec
tio
n
s
.
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
2
.
1
.
P
re
-
pro
ce
s
s
ing
Pre
p
r
o
ce
s
s
in
g
is
a
tech
n
iq
u
e
o
f
im
ag
e
en
h
a
n
ce
m
en
t
th
at
f
o
cu
s
o
n
b
o
t
h
th
e
d
ata
s
tr
u
ctu
r
e
an
d
p
r
o
ce
s
s
in
g
co
n
s
tr
ain
ts
.
I
t
en
c
o
m
p
ass
es
th
e
r
em
o
v
al
en
h
a
n
ce
m
en
t
o
f
th
e
im
ag
e
to
im
p
r
o
v
e
s
y
s
tem
p
er
f
o
r
m
a
n
ce
.
C
r
o
p
p
in
g
is
u
s
ed
to
elim
in
ate
u
n
n
ec
ess
ar
y
c
o
m
p
o
n
en
ts
f
r
o
m
an
im
a
g
e.
Ad
d
itio
n
ally
,
co
n
v
er
tin
g
th
e
im
ag
es to
g
r
ay
s
ca
le,
an
ess
en
tial
s
tep
is
p
er
f
o
r
m
ed
.
T
h
e
c
o
n
tr
ast
ad
a
p
tiv
e
h
is
to
g
r
am
eq
u
aliza
tio
n
(
A
HE
)
f
ilter
an
d
B
ay
es
wav
elet
tr
an
s
f
o
r
m
(
W
T
)
ar
e
u
tili
ze
d
to
r
ed
u
ce
n
o
is
e,
en
h
an
ce
b
r
ig
h
tn
ess
an
d
co
n
tr
ast,
an
d
n
o
r
m
a
lize
th
e
im
ag
e
.
Fig
u
r
e
1
d
e
m
o
n
s
tr
ates
th
e
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
.
T
h
is
p
r
o
ce
s
s
aim
s
to
r
em
o
v
e
n
o
is
e
f
r
o
m
M
R
I
im
ag
es.
T
h
e
DB
3
wav
elet
is
u
s
ed
f
o
r
d
ec
o
m
p
o
s
in
g
th
e
im
a
g
e,
a
n
d
th
e
n
o
is
e
s
tan
d
ar
d
d
ev
iatio
n
is
co
n
s
id
er
ed
wh
en
estab
lis
h
in
g
wav
elet
d
etail
co
ef
f
icien
t
th
r
esh
o
ld
.
T
h
e
ty
p
e
o
f
wav
elet
ap
p
lied
is
d
eter
m
in
ed
p
y
wt.
wav
elis
t
f
u
n
ctio
n
,
with
b
io
r
6
.
8
s
elec
ted
as
th
e
wav
el
et
ch
o
ice.
So
f
t
th
r
esh
o
ld
in
g
is
im
p
lem
en
ted
to
f
in
d
th
e
o
p
tim
al
m
atch
f
o
r
th
e
o
r
ig
in
al
im
ag
e
with
ad
d
itiv
e
n
o
is
e.
C
o
n
tr
ast
lim
ited
ad
ap
tiv
e
h
is
t
o
g
r
am
eq
u
aliza
tio
n
(
C
L
AHE
)
is
an
ad
v
an
ce
d
v
er
s
io
n
o
f
AHE
d
esig
n
ed
t
o
p
r
ev
en
t
co
n
tr
ast
o
v
er
a
m
p
lifi
ca
tio
n
.
C
L
AHE
o
p
e
r
ates
o
n
s
m
all
s
ec
tio
n
s
o
f
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
imp
r
o
ve
d
fu
z
z
y
C
-
mea
n
s
a
n
d
w
a
ters
h
ed
s
eg
men
ta
tio
n
to
cla
s
s
ify
A
lz
h
eime
r
’
s
…
(
E
s
r
a
a
H.
A
li
)
4083
im
ag
e
r
ath
er
th
an
th
e
en
tire
im
ag
e,
u
s
in
g
a
C
lip
L
im
it
p
ar
am
eter
to
s
et
co
n
tr
ast
th
r
esh
o
ld
.
T
h
e
in
itial
v
alu
e
is
s
et
at
3
,
with
th
e
tile
g
r
id
s
ize
d
eter
m
in
in
g
th
e
n
u
m
b
er
o
f
til
es
p
er
r
o
w
a
n
d
co
lu
m
n
s
et
t
o
8
×
8
.
T
h
is
ap
p
r
o
ac
h
ap
p
lies
a
co
n
tr
ast
f
ilter
b
y
d
iv
id
in
g
th
e
im
ag
e
in
to
s
ec
tio
n
s
.
T
h
e
p
r
ep
r
o
ce
s
s
in
g
s
tag
e
co
n
cl
u
d
es
with
cr
o
p
p
i
n
g
an
d
n
o
r
m
alizin
g
MRI
im
ag
es.
C
r
o
p
p
in
g
is
a
tech
n
iq
u
e
in
c
o
m
p
u
ter
im
ag
in
g
u
s
ed
to
r
em
o
v
e
ir
r
elev
an
t
a
r
ea
s
an
d
s
u
r
r
o
u
n
d
in
g
s
f
r
o
m
im
a
g
e
s
.
No
r
m
aliza
tio
n
is
th
e
p
r
o
ce
s
s
o
f
r
ed
u
cin
g
th
e
in
ten
s
ity
v
ar
iatio
n
am
o
n
g
p
i
x
el
v
alu
es,
m
ar
k
in
g
th
e
f
i
n
al
p
h
as
e
o
f
th
e
p
r
ep
r
o
ce
s
s
in
g
s
tag
e.
2
.
2
.
Seg
m
ent
a
t
io
n pro
ce
s
s
I
m
ag
e
s
eg
m
en
tatio
n
m
eth
o
d
s
e
n
co
m
p
ass
th
r
esh
o
ld
-
b
ased
,
e
d
g
e
-
b
ased
,
r
e
g
io
n
-
b
ased
,
m
atc
h
in
g
-
b
ased
,
clu
s
ter
in
g
-
b
ased
,
f
u
zz
y
in
f
e
r
en
ce
-
b
ased
,
an
d
g
en
e
r
alize
d
p
r
in
cip
al
co
m
p
o
n
e
n
t
an
aly
s
is
tech
n
iq
u
es.
E
ac
h
m
eth
o
d
o
f
f
er
s
ad
v
an
tag
es
an
d
lim
itatio
n
s
.
C
lu
s
ter
in
g
is
a
m
eth
o
d
f
o
r
d
iv
id
in
g
a
co
llect
io
n
o
f
o
b
jects
in
to
d
if
f
er
en
t
g
r
o
u
p
s
,
ea
ch
k
n
o
wn
as
a
clu
s
ter
.
Me
m
b
er
s
with
in
ea
ch
clu
s
ter
ex
h
ib
it
h
ig
h
s
im
ilar
ity
in
ter
m
s
o
f
f
ea
tu
r
es,
b
u
t
th
e
d
eg
r
ee
o
f
s
im
ilar
ity
co
m
p
ar
ed
to
m
em
b
er
s
o
f
o
th
er
clu
s
ter
s
is
m
in
im
al.
W
h
i
le
m
an
y
clu
s
ter
in
g
alg
o
r
ith
m
s
s
h
ar
e
f
o
u
n
d
atio
n
al
co
n
ce
p
ts
,
d
if
f
er
e
n
ce
s
ar
is
e
in
h
o
w
s
im
ilar
ity
o
r
d
is
tan
ce
is
m
ea
s
u
r
ed
an
d
h
o
w
lab
els
ar
e
ass
ig
n
ed
to
ca
teg
o
r
ies
with
in
ea
ch
clu
s
ter
.
Key
s
tr
ateg
ies
in
clu
d
e
f
u
zz
y
clu
s
ter
in
g
,
d
en
s
ity
-
b
ased
clu
s
ter
in
g
,
d
is
cr
im
in
ativ
e
clu
s
ter
in
g
,
m
o
d
el
-
b
ased
clu
s
ter
in
g
,
an
d
h
ier
ar
ch
ical
clu
s
ter
in
g
[
1
7
]
.
I
n
o
u
r
s
tu
d
y
,
we
h
av
e
co
m
b
in
ed
I
m
FC
m
clu
s
ter
in
g
with
W
s
to
en
h
an
ce
b
o
th
th
e
ac
cu
r
ac
y
a
n
d
e
f
f
icien
cy
o
f
im
ag
e
an
aly
s
is
.
2
.
2
.
1
.
I
m
pro
v
ed
f
uzzy
C
-
m
e
a
ns
clus
t
er
ing
(
t
he
pro
po
s
ed
m
et
ho
d)
I
n
f
u
zz
y
clu
s
ter
in
g
,
u
n
lik
e
tr
a
d
itio
n
al
clu
s
ter
in
g
wh
er
e
ea
c
h
s
am
p
le
is
as
s
ig
n
ed
ex
clu
s
iv
ely
to
o
n
e
clu
s
ter
,
a
s
in
g
le
s
am
p
le
ca
n
b
e
ass
o
ciate
d
with
m
u
ltip
le
clu
s
ter
s
.
T
h
e
co
r
e
p
r
in
ci
p
le
b
e
h
in
d
f
u
zz
y
cl
u
s
ter
in
g
is
th
at
ea
ch
elem
en
t
ca
n
b
e
ass
ig
n
ed
to
d
if
f
er
e
n
t
clu
s
ter
s
wit
h
v
a
r
y
in
g
d
e
g
r
ee
s
o
f
m
e
m
b
er
s
h
ip
[
1
8
]
.
FC
M
is
a
wid
ely
r
ec
o
g
n
ize
d
f
u
zz
y
cl
u
s
ter
in
g
ap
p
r
o
ac
h
.
O
u
r
o
b
jectiv
e
is
to
o
p
tim
ize
th
e
f
o
llo
win
g
m
eth
o
d
o
l
o
g
y
[
1
9
]
u
s
in
g
th
e
FC
M
alg
o
r
ith
m
:
=
∑
∑
2
=
1
=
∑
∑
‖
−
‖
2
=
1
=
1
=
1
(
1
)
wh
er
e
m
is
a
p
o
s
itiv
e
in
teg
er
t
h
at
is
g
r
ea
ter
th
an
o
n
e.
Mo
r
eo
v
er
,
u
ik
is
th
e
k
th
d
ata'
s
lev
el
o
f
m
em
b
er
s
h
ip
in
th
e
ith
clu
s
ter
,
d
ik
is
th
e
r
atio
o
f
f
am
iliar
ity
in
th
e
p
r
ec
e
d
in
g
n
s
p
ac
e,
x
k
in
d
icate
s
th
e
k
th
d
a
ta,
an
d
v
i
is
th
e
it
h
clu
s
ter
's
ce
n
ter
.
I
n
o
u
r
s
tu
d
y
,
we
aim
to
d
ev
elo
p
a
n
en
h
an
c
ed
an
d
r
o
b
u
s
t
f
u
zz
y
C
-
m
ea
n
s
(
FC
M)
clu
s
ter
in
g
tech
n
iq
u
e,
with
m
o
d
if
icatio
n
s
im
p
lem
en
ted
in
t
h
e
f
o
llo
win
g
ar
ea
s
,
as d
ep
icted
in
Fig
u
r
e
2
.
Fig
u
r
e
2
.
I
m
p
r
o
v
ed
f
u
zz
y
C
-
m
ea
n
s
clu
s
ter
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
0
8
0
-
4
0
9
4
4084
First
p
ar
t
(
in
itial
p
ar
am
eter
s
s
ettin
g
)
:
t
h
e
p
ar
am
eter
s
ettin
g
s
wer
e
ca
r
ef
u
lly
c
h
o
s
en
as
f
o
llo
ws
;
w
e
in
itially
s
elec
ted
a
clu
s
ter
n
u
m
b
er
o
f
4
,
as
th
is
co
u
n
t
is
co
n
s
id
er
ed
o
p
tim
al
f
o
r
s
eg
m
en
tin
g
MRI
b
r
ain
im
ag
es
in
to
f
o
u
r
d
is
tin
ct
r
eg
io
n
s
,
ef
f
e
ctiv
ely
h
ig
h
lig
h
tin
g
k
ey
tis
s
u
es
with
in
th
e
h
u
m
an
b
r
ain
.
T
h
e
f
u
zz
in
ess
d
eg
r
ee
p
ar
am
eter
was
s
et
to
2
,
p
r
o
v
id
in
g
in
cr
ea
s
ed
f
lex
ib
ilit
y
in
ass
o
ciatin
g
d
ata
with
s
p
ec
if
ic
clu
s
ter
s
,
th
er
eb
y
ac
h
iev
in
g
a
b
ala
n
ce
b
etwe
en
s
en
s
itiv
ity
an
d
r
o
b
u
s
tn
ess
.
T
o
en
s
u
r
e
ad
e
q
u
ate
c
o
n
v
e
r
g
e
n
ce
,
a
lim
it
o
f
1
0
0
iter
atio
n
s
was
ch
o
s
en
,
th
is
d
ec
i
s
io
n
b
ein
g
in
f
o
r
m
ed
b
y
o
b
s
er
v
atio
n
s
f
r
o
m
co
n
v
er
g
e
n
ce
ex
p
e
r
im
en
ts
.
A
n
eig
h
b
o
r
ef
f
ec
t
o
f
4
was
s
elec
ted
to
r
e
f
lect
th
e
s
ize
o
f
th
e
s
lid
in
g
win
d
o
w
u
s
ed
in
im
a
g
e
f
ilter
in
g
,
wh
ich
f
ac
ilit
ates
th
e
co
m
p
u
tatio
n
o
f
lo
ca
l
f
ea
tu
r
es
an
d
ca
p
tu
r
es
s
p
atial
r
elatio
n
s
h
ip
s
.
An
ep
s
ilo
n
th
r
esh
o
l
d
o
f
0
.
0
5
was
estab
lis
h
ed
f
o
r
th
e
co
n
v
er
g
e
n
ce
cr
iter
ia,
i
n
d
icatin
g
a
s
tr
in
g
en
t
c
o
n
v
e
r
g
en
ce
th
r
esh
o
l
d
d
u
e
to
its
lo
wer
v
alu
e.
T
h
e
k
er
n
el
s
ize
was d
eter
m
in
ed
to
b
e
3
,
t
o
aid
in
ca
p
tu
r
in
g
s
p
atial
in
f
o
r
m
atio
n
d
u
r
in
g
th
e
im
ag
e
f
ilter
i
n
g
p
r
o
ce
s
s
.
Seco
n
d
p
a
r
t:
th
e
d
is
tan
ce
win
d
o
w
h
as
b
ee
n
u
tili
ze
d
t
o
f
ilter
t
h
e
im
ag
e.
A
s
lid
in
g
win
d
o
w
te
ch
n
iq
u
e
is
em
p
lo
y
ed
to
tr
av
e
r
s
e
th
e
en
tire
im
ag
e,
aim
in
g
to
id
en
tify
s
tab
le
g
r
o
u
p
s
.
I
n
itially
,
p
ad
d
in
g
is
cr
ea
ted
,
eq
u
iv
alen
t
to
h
alf
o
f
th
e
k
er
n
el
s
ize,
to
en
s
u
r
e
th
e
in
clu
s
io
n
o
f
i
m
a
g
e
b
o
r
d
er
s
d
u
r
in
g
th
e
s
lid
in
g
p
r
o
ce
s
s
.
T
h
e
im
ag
e
m
ea
n
is
ca
lcu
lated
u
s
in
g
t
h
is
p
ad
d
in
g
,
a
n
d
t
h
e
cv
2
.
co
p
y
Ma
k
eBo
r
d
er
f
u
n
ctio
n
is
em
p
l
o
y
ed
to
in
co
r
p
o
r
ate
e
d
g
es
d
u
r
in
g
s
lid
in
g
.
Su
b
s
eq
u
en
tly
,
a
s
lid
in
g
win
d
o
w
al
g
o
r
ith
m
is
d
ef
in
ed
,
s
p
ec
if
y
in
g
t
h
e
n
eig
h
b
o
r
ef
f
ec
t
an
d
win
d
o
w
s
ize
as
p
ar
am
eter
s
,
f
o
cu
s
in
g
o
n
lo
ca
tin
g
s
tab
le
g
r
o
u
p
s
o
f
p
ix
els
with
in
clu
s
ter
s
.
T
h
e
f
u
n
ctio
n
f
o
r
lo
ca
tin
g
s
tab
le
g
r
o
u
p
s
o
p
e
r
ates
b
y
id
e
n
tify
in
g
s
tab
le
p
ix
els
t
h
r
o
u
g
h
a
Gau
s
s
ian
f
ilter
,
wh
er
e
th
e
f
ilter
'
s
v
alu
es
ar
e
less
th
a
n
o
r
eq
u
al
to
th
e
s
q
u
ar
e
r
o
o
t
o
f
th
e
n
an
m
ea
n
f
o
r
th
e
p
o
wer
d
if
f
er
en
ce
o
f
Gau
s
s
ian
v
alu
es
an
d
win
d
o
w
s
ize.
T
h
e
r
esu
lt o
f
th
is
p
h
ase
is
a
f
ilter
ed
im
ag
e,
wh
ich
will b
e
f
u
r
th
e
r
u
tili
ze
d
in
th
e
th
ir
d
p
ar
t to
co
m
p
u
te
f
u
zz
i
n
ess
.
T
h
ir
d
p
ar
t:
th
e
h
is
to
g
r
am
o
f
th
e
im
ag
e
is
d
eter
m
in
ed
u
s
in
g
t
h
e
C
L
AHE
f
ilter
.
I
t
is
to
en
h
an
ce
im
ag
e
co
n
tr
ast
an
d
ca
lcu
late
th
e
i
n
ten
s
ity
d
is
tr
ib
u
tio
n
o
f
th
e
en
h
an
ce
d
im
ag
e.
T
h
is
p
h
ase
is
in
s
tr
u
m
en
tal
in
d
eter
m
in
in
g
t
h
e
ce
n
tr
o
id
o
f
th
e
clu
s
ter
.
Fo
u
r
th
p
a
r
t:
th
is
s
eg
m
en
t
en
co
m
p
ass
es
s
ev
er
al
cr
itical
f
u
n
ctio
n
s
.
I
t
b
e
g
in
s
with
th
e
in
itializatio
n
o
f
t
h
e
m
em
b
er
s
h
ip
f
u
n
ctio
n
,
wh
ich
d
eter
m
in
es
th
e
d
eg
r
ee
t
o
wh
ic
h
th
e
p
ix
els
o
f
t
h
e
im
ag
e
b
elo
n
g
to
ea
ch
clu
s
ter
.
T
h
is
is
f
o
llo
wed
b
y
th
e
f
u
n
ctio
n
f
o
r
co
m
p
u
tin
g
th
e
ce
n
tr
o
id
s
o
f
th
e
clu
s
ter
s
,
wh
ich
in
v
o
lv
es
th
e
d
iv
is
io
n
o
f
th
e
n
u
m
er
ato
r
b
y
th
e
d
e
n
o
m
in
at
o
r
.
T
h
e
n
u
m
er
at
o
r
is
th
e
s
u
m
o
f
th
e
p
r
o
d
u
ct
o
f
th
e
d
eg
r
ee
o
f
f
u
zz
in
ess
,
t
h
e
h
is
to
g
r
am
,
an
d
th
e
in
ten
s
ity
,
ea
ch
r
aised
to
th
e
p
o
wer
o
f
m
em
b
er
s
h
ip
.
C
o
n
v
e
r
s
ely
,
th
e
d
en
o
m
in
ato
r
is
th
e
s
u
m
m
atio
n
o
f
t
h
e
h
is
to
g
r
a
m
v
alu
es
r
aised
to
th
e
p
o
wer
o
f
m
em
b
er
s
h
ip
.
T
h
e
f
in
al
s
tep
in
th
is
p
ar
t
is
th
e
co
m
p
u
tatio
n
o
f
weig
h
ts
,
a
p
r
o
ce
s
s
r
elian
t
o
n
th
e
ce
n
tr
o
i
d
co
m
p
u
tatio
n
f
u
n
ctio
n
.
T
h
is
in
v
o
lv
es
d
iv
i
d
in
g
th
e
n
u
m
er
ato
r
b
y
th
e
d
en
o
m
i
n
ato
r
,
wh
er
e
th
e
n
u
m
e
r
ato
r
ca
lcu
lat
es
th
e
ab
s
o
lu
te
d
if
f
er
en
ce
s
b
et
wee
n
th
e
in
ten
s
ity
r
aised
to
a
p
o
wer
an
d
t
h
e
co
m
p
u
ted
clu
s
ter
ce
n
tr
o
i
d
s
.
T
h
e
d
en
o
m
in
ato
r
,
o
n
th
e
o
th
er
h
a
n
d
,
s
u
m
s
th
e
ab
s
o
lu
te
d
if
f
er
en
ce
s
,
ea
c
h
r
aised
to
th
e
p
o
wer
o
f
th
e
f
u
zz
in
ess
d
eg
r
e
e.
A
lg
o
r
ith
m
1
s
h
o
ws
th
e
d
etailed
p
r
o
ce
s
s
.
T
h
e
b
lo
ck
d
ia
g
r
am
in
Fig
u
r
e
2
s
h
o
ws th
e
p
r
o
ce
s
s
o
f
u
tili
zin
g
th
e
im
p
r
o
v
e
d
f
u
zz
y
C
-
m
ea
n
s
clu
s
ter
in
g
.
Alg
o
r
ith
m
1
:
I
m
FC
M
Step 1:
Initialize the following parameters:
-
Number of bits. Number of clusters.
-
Degree of fuzziness.
-
Maximum iteration count.
-
Epsilon threshold for convergence check.
Step 2:
Image Filtering Procedure:
-
Generate
a
padded
image
using
a
sliding
window
with
dimensions
(kerne
l_size/2,
kernel_size/2).
-
Compute the mean based on the padded mask.
-
Pad the resulting mean image to create borders.
-
Utilize a sliding window to account for neighbor effects and kernel size.
-
Determine
center
coordin
ates
using
the
spatial
distance
window
with
Minkowski
distance:
Des_win
=
((abs
(win_si
ze_y
-
center_coordinate
_y))
**
p
+
abs
((win_
size_y
-
center_coordinate_y) ** p)) ** (1/p), where p = 2.
-
Identify the stable group matrix using a Gaussian filter.
-
Obtain the final filtered image using the formula:
Final_image = sum (weighted_coefficients * old_window) / sum (weighted_coefficients)
-
Perform CLAHE.
St
ep
3:
We
ig
ht
In
it
ia
li
z
at
io
n:
In
it
ia
li
ze
a
tw
o
-
dimensional
matrix
based
on
the
number
of
clusters and gray levels to compute weights.
Step 4: Compute Cluster Centroids:
-
Calculate the X and Y values as follows:
-
X = sum (histogram * number of gray levels) * power (weight * number of fuzziness)
-
Y = sum (histogram) * power (weight * number of fuzziness)
-
Z = X / Y
Step 5: Weight Computation Method:
-
Set power =
-
2 / number of fuzziness.
-
Calculate
the
X
value
usi
ng
the
formula:
X
=
(gray
levels
-
centroid
values
)
*
power.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
imp
r
o
ve
d
fu
z
z
y
C
-
mea
n
s
a
n
d
w
a
ters
h
ed
s
eg
men
ta
tio
n
to
cla
s
s
ify
A
lz
h
eime
r
’
s
…
(
E
s
r
a
a
H.
A
li
)
4085
-
Compute
Y
as:
Y
=
sum
(
gray
levels
-
centroid
values)
*
power.
Determi
ne
Z
as:
Z = X / Y.
St
ep
6:
Ch
ec
k
Co
nv
er
ge
nc
e
:
De
te
rm
in
e
wh
et
he
r
th
e
a
bs
ol
ut
e
ma
xi
mu
m
va
lu
e
of
(s
te
p
5
-
st
ep
2)
is less than the epsilon threshold. If so, stop; otherwise, proceed to step 4.
2
.
2
.
2
.
Wa
t
er
s
hed
s
eg
m
ent
a
t
io
n
I
n
Ws
,
an
im
a
g
e
is
c
o
n
ce
p
tu
al
ized
in
t
h
r
ee
d
im
en
s
io
n
s
,
with
th
e
(
x
,
y
)
c
o
o
r
d
in
ates
co
r
r
esp
o
n
d
t
o
th
e
s
p
atial
ax
es
an
d
th
e
in
ten
s
ity
r
ep
r
esen
ted
alo
n
g
th
e
z
-
ax
is
.
T
h
is
ap
p
r
o
ac
h
tr
ea
ts
an
im
ag
e
as
if
it
wer
e
a
to
p
o
g
r
a
p
h
ical
lan
d
s
ca
p
e,
with
th
e
in
ten
s
ity
o
f
ea
ch
p
ix
el
a
n
alo
g
o
u
s
to
elev
atio
n
lev
els.
C
o
n
s
eq
u
en
tly
,
ea
ch
in
ten
s
ity
lev
el
is
ass
o
ciate
d
with
a
d
is
tin
ct
elev
atio
n
p
lan
e
o
n
t
h
is
lan
d
s
ca
p
e.
Utilizin
g
th
is
to
p
o
g
r
ap
h
ical
m
etap
h
o
r
,
p
o
i
n
ts
with
in
th
e
i
m
ag
e
ar
e
ca
te
g
o
r
ized
in
to
th
r
e
e
s
eg
em
ts
: r
eg
io
n
al
m
in
i
m
a,
c
atch
m
en
t b
asin
s
,
an
d
wate
r
s
h
ed
lin
es.
C
atch
m
en
t
b
asin
ar
e
ar
ea
s
wer
e
,
h
y
p
o
th
eti
ca
l,
a
d
r
o
p
let
o
f
wate
r
wo
u
ld
co
v
er
ag
e
to
war
d
s
a
s
in
g
le
r
eg
io
n
al
m
in
im
u
m
.
W
ater
s
h
ed
lin
es,
co
n
v
er
s
ely
,
m
a
r
k
th
e
b
o
u
n
d
ar
ies
wh
er
e
a
d
r
o
p
let
o
f
wate
r
co
u
ld
p
o
ten
tially
b
e
d
r
awn
to
war
d
s
m
u
ltip
le
r
eg
io
n
al
m
in
im
a,
ef
f
ec
tiv
ely
d
elin
ea
tin
g
th
e
d
i
v
is
io
n
b
etwe
en
d
is
tin
ct
ca
tch
m
en
t b
asin
s
[
1
1
]
.
C
o
n
s
id
er
th
e
M1
,
M2
,
…
MR
r
eg
io
n
al
m
in
im
a
o
f
an
im
a
g
e
g
(
x
,
y
)
.
L
et
T
[
n
]
r
ep
r
esen
t
an
ar
r
ay
o
f
p
o
in
ts
b
en
ea
th
th
e
h
o
r
izo
n
tal
a
x
is
with
a
v
alu
e
o
f
n
,
wh
er
e
n
r
an
g
es
f
r
o
m
th
e
im
ag
e'
s
least
to
g
r
ea
test
in
ten
s
ity
.
T
h
is
m
ay
b
e
s
tated
m
ath
e
m
atica
lly
as f
o
llo
ws:
T
[
]
=
{
(
,
)
|
(
,
)
<
}
(
2
)
C
n
(
Mi)
in
d
icate
a
co
llectio
n
o
f
r
eg
io
n
s
in
th
e
ca
tc
h
m
en
t
b
as
in
th
at
ar
e
p
o
u
r
ed
at
p
la
n
e
n
t
h
at
ar
e
r
elate
d
with
th
e
r
eg
io
n
m
in
im
u
m
Mi.
T
h
is
co
u
ld
p
o
s
s
ib
ly
b
e
u
s
ed
to
co
m
p
u
te
it b
y
:
(
)
=
(
)
[
]
(
3
)
C
(
Mi)
is
th
e
s
et
o
f
ca
tch
m
e
n
t
b
asin
p
o
in
ts
lin
k
e
d
with
th
e
r
eg
io
n
al
m
in
im
u
m
Mi.
T
h
e
u
n
io
n
o
f
all
f
lo
o
d
ed
ca
tch
m
en
t b
asin
s
at
a
ce
r
tain
s
tag
e
n
r
ep
r
esen
ted
in
C
[
n
]
:
[
]
=
[
(
)
]
(
4
)
Alg
o
r
ith
m
2
in
tr
o
d
u
ce
s
th
e
s
te
p
s
o
f
th
is
tech
n
i
q
u
e.
Fig
u
r
e
3
illu
s
tr
ate
th
e
wate
r
s
h
ed
m
eth
o
d
in
a
b
lo
c
k
d
iag
r
am
.
Alg
o
r
ith
m
2
: Water
s
h
ed
s
eg
m
en
tatio
n
-
Utilize
OT
SU’
s
b
in
ar
izatio
n
f
ilter
to
esti
m
ate
th
e
o
b
jects p
r
e
s
en
t in
th
e
im
ag
e.
-
Ap
p
ly
m
o
r
p
h
o
lo
g
ical
o
p
en
i
n
g
to
elim
in
ate
an
y
wh
ite
n
o
is
e
p
r
esen
t
in
t
h
e
im
ag
e
,
an
d
p
e
r
f
o
r
m
m
o
r
p
h
o
lo
g
ical
clo
s
in
g
to
ad
d
r
ess
s
m
all
h
o
les with
in
th
e
o
b
jects.
-
E
m
p
lo
y
th
e
d
ilate
m
eth
o
d
to
c
r
ea
te
a
s
ep
ar
atio
n
b
etwe
e
n
th
e
b
ac
k
g
r
o
u
n
d
an
d
th
e
im
ag
e
.
-
Utilize
d
is
tan
ce
tr
an
s
f
o
r
m
an
d
th
r
esh
o
ld
in
g
tech
n
iq
u
es to
is
o
late
th
e
f
o
r
eg
r
o
u
n
d
f
r
o
m
th
e
b
ac
k
g
r
o
u
n
d
.
-
Dete
r
m
in
e
th
e
u
n
k
n
o
wn
ar
ea
s
b
y
s
u
b
tr
ac
tin
g
th
e
f
o
r
eg
r
o
u
n
d
f
r
o
m
th
e
b
ac
k
g
r
o
u
n
d
.
T
h
ese
ar
ea
s
lack
in
g
clar
ity
will b
e
ass
ig
n
ed
ze
r
o
v
alu
es in
th
e
m
ar
k
e
r
s
.
-
L
ab
el
th
e
r
eg
io
n
s
o
f
th
e
f
o
r
eg
r
o
u
n
d
u
s
in
g
th
e
c
o
n
n
ec
te
d
co
m
p
o
n
e
n
ts
m
eth
o
d
as
m
ar
k
er
s
,
an
d
in
cr
em
e
n
t
th
em
b
y
o
n
e
to
en
s
u
r
e
all
b
ac
k
g
r
o
u
n
d
r
e
g
io
n
s
ar
e
m
ar
k
ed
as o
n
es.
-
E
m
p
lo
y
th
e
d
is
tan
ce
v
alu
es
o
b
tain
ed
f
r
o
m
s
tep
5
a
n
d
th
e
m
ar
k
er
s
f
r
o
m
s
tep
6
as in
p
u
t p
ar
a
m
eter
s
f
o
r
th
e
wate
r
s
h
ed
m
eth
o
d
to
g
e
n
er
ate
th
e
f
in
al
s
eg
m
en
tatio
n
m
ap
.
Fig
u
r
e
3
.
W
ater
s
h
ed
s
eg
m
e
n
tatio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
0
8
0
-
4
0
9
4
4086
2
.
3
.
P
o
s
t
-
pro
ce
s
s
ing
Gam
m
a
co
r
r
ec
tio
n
is
a
tech
n
iq
u
e
u
s
ed
f
o
r
d
ata
au
g
m
en
tatio
n
th
at
en
tails
ad
ju
s
tin
g
th
e
g
a
m
m
a
v
alu
e
to
m
o
d
if
y
th
e
im
a
g
e
in
ten
s
ity
.
Gam
m
a
r
ep
r
esen
ts
a
n
o
n
-
lin
e
ar
f
u
n
ctio
n
th
at
ca
n
b
e
ap
p
lied
to
eith
er
en
co
d
e
o
r
d
ec
o
d
e
t
h
e
b
r
ig
h
tn
ess
o
r
in
ten
s
ity
o
f
an
im
ag
e
[
2
0
]
.
Gau
s
s
ian
n
o
is
e,
a
n
o
th
er
f
o
r
m
o
f
d
is
to
r
tio
n
,
is
in
tr
o
d
u
ce
d
in
to
th
e
im
ag
e
th
r
o
u
g
h
r
an
d
o
m
v
alu
es
d
r
awn
f
r
o
m
a
Gau
s
s
ian
d
is
tr
ib
u
tio
n
.
Sin
ce
n
o
is
e
en
co
u
n
ter
e
d
d
u
r
i
n
g
im
ag
e
ac
q
u
is
itio
n
an
d
p
r
ep
r
o
c
ess
in
g
ca
n
escalate
,
ap
p
ly
in
g
Gau
s
s
ian
n
o
is
e
to
a
r
aw
im
ag
e
m
ay
h
elp
th
e
m
o
d
el
b
ec
o
m
e
m
o
r
e
r
esil
ien
t
to
v
ar
ia
tio
n
in
im
ag
e
q
u
ality
[
2
1
]
.
Ou
r
s
tu
d
y
ad
v
o
ca
tes
f
o
r
th
e
u
s
e
o
f
th
ese
two
m
eth
o
d
s
as
a
p
o
s
t
-
p
r
o
ce
s
s
in
g
m
ea
s
u
r
e
f
o
r
d
atasets
th
at
m
ay
s
u
f
f
e
r
f
r
o
m
lo
s
s
in
co
n
tr
ast
an
d
b
r
ig
h
t
n
ess
,
as
well
as
to
in
tr
o
d
u
ce
a
s
lig
h
t
b
lu
r
r
in
g
ef
f
e
ct
to
s
m
o
o
th
tr
a
n
s
itio
n
s
in
p
ix
el
v
alu
e
an
d
s
o
f
ten
th
e
im
ag
e
ed
g
es.
T
h
e
m
ar
k
er
-
co
n
tr
o
lled
Ws
tech
n
iq
u
e
is
ap
p
lied
f
o
llo
win
g
th
e
I
m
FC
m
twice
with
a
m
ar
k
er
v
alu
e
r
an
g
e
o
f
[
1
0
-
9
0
]
to
ca
p
tu
r
e
im
ag
es
h
ig
h
lig
h
tin
g
in
ter
n
al
b
r
ain
f
ea
tu
r
es.
T
h
is
p
r
o
ce
s
s
is
th
en
r
ep
ea
ted
with
a
m
ar
k
er
v
alu
e
r
an
g
e
o
f
[
1
0
-
2
0
0
]
to
o
b
tain
im
a
g
es
s
h
o
wca
s
in
g
ex
ter
n
al
b
r
ain
d
et
ails
.
T
h
e
cu
lm
in
atio
n
o
f
th
is
p
r
o
ce
s
s
in
v
o
lv
es
co
m
b
in
in
g
all
th
r
ee
im
ag
es
to
p
r
o
d
u
ce
th
e
f
in
al
en
h
an
ce
d
im
ag
e.
Po
s
t
-
p
r
o
ce
s
s
in
g
s
tep
s
,
in
clu
d
in
g
g
am
m
a
co
r
r
ec
tio
n
a
n
d
Gau
s
s
ian
b
lu
r
,
ar
e
s
u
b
s
eq
u
en
tly
p
er
f
o
r
m
ed
t
o
f
u
r
th
er
r
ef
in
e
th
e
im
ag
es,
as
d
ep
icted
in
Fig
u
r
e
4.
Fig
u
r
e
4
.
Seg
m
e
n
tatio
n
s
tag
e
2
.
4
.
Cla
s
s
if
ica
t
io
n
C
NN,
a
co
r
n
er
s
to
n
e
o
f
th
e
n
eu
r
al
n
etwo
r
k
f
r
am
ewo
r
k
,
en
co
m
p
ass
es
n
u
m
er
o
u
s
lay
er
s
with
in
its
ar
ch
itectu
r
e
an
d
h
as
g
ain
e
d
s
ig
n
if
ican
t
p
r
o
m
in
en
ce
i
n
v
ar
i
o
u
s
im
ag
e
p
r
o
ce
s
s
in
g
ap
p
licatio
n
s
,
n
o
tab
ly
in
o
b
ject
r
ec
o
g
n
itio
n
[
2
2
]
an
d
im
ag
e
cla
s
s
if
icatio
n
,
wh
er
e
it
h
as
y
ield
ed
p
r
o
m
is
in
g
o
u
tco
m
es.
Pre
v
io
u
s
r
esear
ch
in
d
icate
s
th
e
f
ea
s
ib
ilit
y
o
f
d
ir
ec
tly
in
p
u
ttin
g
im
ag
es
in
to
a
C
NN
n
et
wo
r
k
to
e
x
tr
ac
t
f
ea
tu
r
es
f
o
r
im
ag
e
ca
teg
o
r
izatio
n
.
T
h
e
ar
ch
itectu
r
e
o
f
a
C
NN
co
m
p
r
is
es
s
ev
er
al
f
u
n
d
am
en
t
al
co
m
p
o
n
e
n
ts
,
in
clu
d
in
g
co
n
v
o
lu
tio
n
al
lay
er
s
,
So
f
tMa
x
lay
er
s
,
p
o
o
lin
g
lay
er
s
,
n
o
n
-
lin
ea
r
ac
tiv
atio
n
f
u
n
ctio
n
s
s
u
ch
as
th
e
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
,
an
d
f
u
lly
co
n
n
ec
ted
la
y
er
s
(
FC
)
[
2
3
]
,
[
2
4
]
.
C
NNs
o
p
er
ate
b
ased
o
n
th
e
i
n
ten
s
ities
o
f
im
ag
es,
u
tili
zin
g
d
im
en
s
io
n
s
s
u
ch
as
wid
th
,
h
eig
h
t,
a
n
d
d
e
p
th
to
r
ep
r
esen
t
th
e
in
p
u
t
im
a
g
e
in
ten
s
ities
.
T
h
e
p
r
o
ce
s
s
in
g
b
eg
in
s
f
r
o
m
th
e
t
o
p
lef
t c
o
r
n
er
o
f
t
h
e
im
ag
e
a
n
d
p
r
o
g
r
ess
es to
th
e
r
ig
h
t.
As th
e
f
il
ter
m
o
v
es
f
r
o
m
th
e
to
p
t
o
th
e
b
o
tto
m
o
f
th
e
i
n
p
u
t
v
o
lu
m
e,
it
c
h
an
g
es,
with
e
ac
h
lef
t
-
to
-
r
i
g
h
t
m
o
v
em
en
t
co
n
s
titu
tin
g
a
s
tr
id
e.
T
h
e
c
o
m
p
le
x
ity
o
f
t
h
e
s
tr
id
e
is
au
g
m
en
ted
b
y
th
e
n
u
m
b
er
o
f
s
tep
s
it
en
c
o
m
p
ass
es.
R
eL
U
s
er
v
es
as
an
ef
f
icien
t
ac
ti
v
atio
n
f
u
n
ctio
n
b
y
co
n
v
er
ti
n
g
n
eg
ativ
e
p
ix
el
v
alu
es
to
ze
r
o
[
2
5
]
.
Fo
llo
win
g
t
h
e
co
n
v
o
l
u
tio
n
p
r
o
ce
s
s
,
th
e
s
ize
o
f
th
e
h
id
d
en
lay
e
r
b
ec
o
m
es
s
ig
n
if
ican
tly
lar
g
e,
n
ec
ess
itatin
g
th
e
u
s
e
o
f
a
p
o
o
li
n
g
o
r
s
u
b
-
s
am
p
lin
g
lay
er
to
r
e
d
u
ce
c
o
m
p
u
tatio
n
al
co
m
p
lex
ity
.
Po
o
lin
g
ca
n
b
e
c
ateg
o
r
ized
in
to
two
t
y
p
es:
m
ax
im
u
m
an
d
av
er
ag
e
[
2
6
]
.
W
ith
in
th
e
co
n
te
x
t
o
f
p
o
o
lin
g
,
let
y
=
y
ij
r
ep
r
esen
t t
h
e
m
atr
ix
.
(
)
=
ma
x
(
0
,
)
(
5
)
As n
o
ticed
in
(
6
)
,
m
ax
p
o
o
lin
g
is
th
e
p
r
o
ce
s
s
th
at
s
elec
ts
th
e
m
o
s
t sig
n
if
ican
t c
o
m
p
o
n
en
t in
y
as th
e
o
u
t
p
u
t
=
ma
x
(
)
(
6
)
P
os
t
-
P
r
oc
e
s
s
e
d
I
mage
I
mF
C
m
W
s
with
mar
ke
r
[
10
-
90]
W
s
with
mar
ke
r
[
10
-
200
]
Pre
-
p
r
oc
e
s
s
e
d
I
mage
B
lende
d
I
mage
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
imp
r
o
ve
d
fu
z
z
y
C
-
mea
n
s
a
n
d
w
a
ters
h
ed
s
eg
men
ta
tio
n
to
cla
s
s
ify
A
lz
h
eime
r
’
s
…
(
E
s
r
a
a
H.
A
li
)
4087
Ou
r
s
tu
d
y
an
al
y
ze
d
MRI
b
r
ai
n
im
ag
es
f
r
o
m
AD
p
atien
ts
,
o
r
g
an
izin
g
th
e
d
ataset
i
n
to
f
o
u
r
d
is
tin
ct
ca
teg
o
r
ies:
n
o
r
m
al,
m
ild
,
m
o
d
er
ate,
an
d
v
e
r
y
m
ild
.
Fig
u
r
e
5
illu
s
tr
ates
a
p
r
o
p
o
s
ed
C
NN
a
r
ch
itectu
r
al
m
o
d
el
co
m
p
r
is
in
g
1
3
lay
er
s
.
T
h
is
m
o
d
el
in
clu
d
es
f
o
u
r
co
n
v
o
l
u
tio
n
a
l
lay
er
s
,
with
ea
ch
p
air
o
f
co
n
v
o
lu
ti
o
n
s
f
o
llo
we
d
b
y
a
p
o
o
lin
g
lay
er
an
d
d
r
o
p
o
u
t
lay
er
s
to
e
n
s
u
r
e
r
eg
u
lar
i
za
tio
n
.
T
h
e
ar
c
h
itectu
r
e
c
o
n
c
lu
d
es
with
a
f
u
lly
co
n
n
ec
ted
lay
er
a
n
d
a
class
if
ier
lay
er
.
T
h
e
o
r
ig
i
n
al
im
ag
es,
m
ea
s
u
r
in
g
1
7
6
×
2
0
8
p
ix
els,
ar
e
r
esized
to
2
0
0
×
2
0
0
p
ix
els
b
ef
o
r
e
b
ein
g
in
p
u
t
in
t
o
th
e
C
NN
m
o
d
el.
T
h
e
f
ilter
s
i
ze
is
v
ar
ied
ac
r
o
s
s
th
e
C
NN
lay
er
s
to
ef
f
ec
tiv
ely
id
en
tify
f
ea
tu
r
es.
T
h
e
b
atc
h
s
ize
is
s
et
to
3
2
,
an
d
t
h
e
m
o
d
el
u
n
d
er
g
o
es
tr
ain
in
g
o
v
er
5
0
ep
o
c
h
s
.
Up
o
n
co
m
p
letio
n
o
f
all
ep
o
ch
cy
cles,
t
h
e
C
NN
s
elec
ts
th
e
m
o
d
el
iter
atio
n
with
t
h
e
h
ig
h
est
p
e
r
f
o
r
m
an
ce
m
et
r
ics
f
o
r
class
if
icatio
n
p
u
r
p
o
s
es.
T
h
e
f
in
al
class
if
icatio
n
is
th
en
ap
p
lied
to
th
e
test
s
et
to
d
eter
m
in
e
t
h
e
ac
cu
r
ac
y
r
ate.
Fig
u
r
e
5
.
C
NN
m
o
d
el
lay
e
r
s
T
ab
le
1
o
u
tlin
es
th
e
i
n
ter
n
al
ar
ch
itectu
r
e
o
f
t
h
e
C
NN
m
o
d
el,
d
etailin
g
th
e
s
p
ec
if
ic
lay
er
s
an
d
co
n
f
ig
u
r
atio
n
s
u
s
ed
with
in
th
e
m
o
d
el.
T
ab
le
2
lis
ts
th
e
h
y
p
er
p
ar
am
eter
s
ap
p
lied
d
u
r
in
g
t
h
e
m
o
d
el’
s
tr
ain
in
g
an
d
o
p
tim
izatio
n
p
r
o
ce
s
s
es.
F
o
r
o
p
tim
izatio
n
,
T
h
e
Ad
am
alg
o
r
ith
m
is
u
tili
ze
d
,
with
a
lear
n
in
g
r
ate
o
f
0
.
0
0
1
s
et
fo
r
th
e
e
n
tire
tr
ain
in
g
p
h
ase.
T
ab
le
1
.
C
NN
m
o
d
el
M
o
d
e
l
l
a
y
e
r
s
I
mag
e
v
o
l
u
m
e
F
i
l
t
e
r
s
S
i
z
e
o
f
f
i
l
t
e
r
P
o
o
l
i
n
g
w
i
n
s
i
z
e
v
a
r
.
c
o
n
v
2
d
(
C
o
n
v
2
D
)
(
2
0
0
,
2
0
0
)
32
5
x
5
2
x
2
2
4
3
2
c
o
n
v
2
d
_
1
(
C
o
n
v
2
D
)
(
2
0
0
,
2
0
0
)
32
5
x
5
2
x
2
2
5
6
3
2
M
a
x
P
o
o
l
i
n
g
2
D
(
1
0
0
,
1
0
0
)
32
2
x
2
0
D
r
o
p
o
u
t
(
1
0
0
,
1
0
0
)
32
0
c
o
n
v
2
d
_
2
(
C
o
n
v
2
D
)
(
1
0
0
,
1
0
0
)
64
3
x
3
1
8
4
9
6
c
o
n
v
2
d
_
3
(
C
o
n
v
2
D
)
(
1
0
0
,
1
0
0
)
64
3
x
3
3
6
9
2
8
max
_
p
o
o
l
i
n
g
2
d
_
1
(
5
0
,
5
0
)
64
2
x
2
0
d
r
o
p
o
u
t
_
1
(
5
0
,
5
0
)
64
0
f
l
a
t
t
e
n
(
F
l
a
t
t
e
n
)
(
N
o
n
e
,
1
6
0
0
0
0
)
0
d
e
n
se
(
D
e
n
s
e
)
(
N
o
n
e
,
2
5
6
)
4
0
9
6
0
2
5
6
d
r
o
p
o
u
t
_
2
(
N
o
n
e
,
2
5
6
)
0
d
e
n
se
_
1
(
D
e
n
se)
(
N
o
n
e
,
4
)
1
0
2
8
T
ab
le
2
.
Valu
es o
f
h
y
p
er
-
p
a
r
a
m
eter
s
H
y
p
e
r
-
p
a
r
a
met
e
r
s
V
a
l
u
e
S
p
l
i
t
d
a
t
a
3
8
4
0
t
r
a
i
n
,
1
2
8
1
v
a
l
i
d
a
t
e
D
r
o
p
o
u
t
0
.
3
,
0
.
3
,
0
.
5
B
a
t
c
h
si
z
e
32
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
0
1
N
u
m.
o
f
e
p
o
c
h
s
5
0
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
0
8
0
-
4
0
9
4
4088
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
E
x
perim
ent
a
l
da
t
a
s
et
T
h
e
MRI
im
ag
es
u
tili
ze
d
f
o
r
th
is
r
esear
ch
wer
e
s
o
u
r
ce
d
f
r
o
m
th
e
ADNI
d
atab
ase.
A
to
tal
o
f
6
4
0
0
ADNI
s
am
p
les
wer
e
s
elec
ted
f
o
r
an
aly
s
is
af
ter
ex
cl
u
d
in
g
ce
r
tain
s
am
p
les
with
in
co
r
r
ec
t
in
f
o
r
m
atio
n
.
T
h
e
d
ataset
co
m
p
r
is
es 3
1
4
0
n
o
r
m
a
l sam
p
les,
8
9
6
ea
r
ly
m
ild
co
g
n
itiv
e
im
p
air
m
en
t sam
p
les,
6
4
m
o
d
er
ate
c
o
g
n
itiv
e
im
p
air
m
en
t sam
p
les,
an
d
2
2
4
0
s
ev
er
e
c
o
g
n
itiv
e
im
p
air
m
e
n
t s
am
p
les.
T
h
ese
im
a
g
es a
r
e
in
J
PEG
f
o
r
m
at
with
a
r
eso
lu
tio
n
o
f
1
7
6
×
2
0
8
p
ix
els.
3
.
2
.
E
v
a
lua
t
io
n
m
et
rics
3
.
2
.
1
.
I
m
pro
v
ed
f
uzzy
C
-
m
e
a
ns
clus
t
er
ing
T
h
e
ev
al
u
atio
n
c
r
iter
ia
u
s
ed
to
ass
ess
co
m
p
u
tatio
n
al
c
o
m
p
le
x
ity
f
o
cu
s
o
n
h
o
w
e
f
f
icien
tly
o
u
r
m
et
h
o
d
p
er
f
o
r
m
s
in
ter
m
s
o
f
c
o
m
p
u
tatio
n
al
r
eso
u
r
ce
s
u
s
ed
an
d
th
e
q
u
ality
o
f
r
esu
lts
o
b
tain
e
d
.
I
t
i
n
v
o
lv
es
e
v
alu
atin
g
th
e
tim
e
an
d
s
ca
lab
ilit
y
o
f
o
u
r
m
eth
o
d
wh
e
n
ap
p
lied
to
ex
ten
s
iv
e
d
atasets
.
C
o
n
s
id
er
atio
n
is
g
iv
en
to
tech
n
iq
u
es
o
r
o
p
tim
izatio
n
s
th
at
c
o
u
ld
e
n
h
an
ce
th
e
alg
o
r
ith
m
'
s
ef
f
icien
cy
with
o
u
t
co
m
p
r
o
m
is
in
g
ac
cu
r
ac
y
.
T
h
e
u
s
ed
eq
u
atio
n
to
co
m
p
u
te
th
e
ef
f
icien
cy
o
f
o
u
r
p
r
o
p
o
s
ed
m
et
h
o
d
i
s
d
ep
icted
(
7
)
:
=
∑
∑
|
−
_
|
=
1
=
1
(
7
)
w
h
er
e
d
en
o
tes
th
e
to
tal
ab
s
o
lu
te
d
if
f
er
en
ce
.
s
ig
n
if
ies
th
e
n
u
m
b
er
o
f
clu
s
ter
s
.
r
ep
r
esen
ts
th
e
n
u
m
b
er
o
f
d
ata
p
o
in
ts
.
r
ef
lects
th
e
m
em
b
er
s
h
ip
v
alu
e
o
f
d
ata
p
o
i
n
t
j
in
clu
s
ter
i
d
u
r
in
g
th
e
c
u
r
r
e
n
t
iter
atio
n
.
_
in
d
icate
s
th
e
m
em
b
er
s
h
ip
v
alu
e
o
f
d
ata
p
o
in
t
j
in
clu
s
ter
i
d
u
r
in
g
th
e
p
r
ev
io
u
s
iter
atio
n
.
T
h
e
eq
u
atio
n
ca
lcu
lates
th
e
s
u
m
o
f
th
e
ab
s
o
lu
te
d
if
f
e
r
en
ce
s
b
etwe
en
co
r
r
esp
o
n
d
in
g
elem
en
ts
o
f
th
e
cu
r
r
e
n
t
an
d
p
r
io
r
m
em
b
er
s
h
ip
m
atr
ices
ac
r
o
s
s
all
clu
s
ter
s
an
d
d
ata
p
o
in
ts
.
T
h
is
m
ea
s
u
r
em
e
n
t
ca
n
ac
t
as
an
in
d
icato
r
o
f
co
n
v
er
g
en
ce
o
r
ch
a
n
g
e
b
etwe
en
s
u
cc
ess
iv
e
iter
atio
n
s
with
in
an
o
p
tim
izatio
n
alg
o
r
ith
m
,
s
u
ch
as
th
e
f
u
zz
y
C
-
m
ea
n
s
clu
s
ter
in
g
alg
o
r
ith
m
.
3
.
2
.
2
.
Co
nv
o
lutio
na
l neura
l
net
wo
rk
T
h
e
an
aly
s
is
o
f
ML
a
n
d
DL
r
ec
o
g
n
itio
n
p
latf
o
r
m
s
,
aim
e
d
a
t
ev
alu
atin
g
th
eir
ca
p
ab
ilit
y
to
ac
cu
r
ately
d
iag
n
o
s
e
AD,
r
elies
o
n
s
ev
e
r
al
p
er
f
o
r
m
a
n
ce
m
etr
ics,
in
clu
d
in
g
ac
cu
r
ac
y
(
Acy
)
,
s
en
s
itiv
ity
(
Sn
y
)
/r
ec
all,
p
r
ec
is
io
n
(
Prn
)
,
an
d
th
e
F1
s
co
r
e.
E
ac
h
o
f
th
ese
p
er
f
o
r
m
a
n
ce
in
d
icato
r
s
o
f
f
er
s
d
if
f
e
r
en
t
in
s
ig
h
ts
in
to
th
e
p
r
o
p
o
s
ed
m
o
d
el'
s
ef
f
ec
tiv
en
e
s
s
.
T
h
e
p
r
im
ar
y
m
ea
s
u
r
e
f
o
r
ev
alu
atin
g
th
e
class
if
icatio
n
s
y
s
tem
is
ac
cu
r
ac
y
,
wh
ich
is
ca
lcu
lated
b
y
d
iv
id
i
n
g
th
e
n
u
m
b
er
o
f
co
r
r
ec
t
p
r
e
d
ictio
n
s
b
y
th
e
to
tal
n
u
m
b
er
o
f
p
r
e
d
ictio
n
s
m
ad
e
.
Ma
th
em
atica
ll
y
,
it c
an
b
e
e
x
p
r
ess
ed
as
(
8
)
:
=
+
+
+
+
(
8
)
w
h
er
e
T
P
an
d
T
N
ar
e
tr
u
e
p
o
s
itiv
e
an
d
tr
u
e
n
e
g
ativ
e
r
esp
ec
ti
v
ely
.
FP
,
FN
ar
e
f
alse
p
o
s
itiv
e
an
d
f
alse
n
eg
ativ
e,
r
esp
ec
tiv
ely
.
Sen
s
itiv
ity
an
d
s
p
ec
if
icity
ar
e
lo
g
ically
s
p
ec
if
i
ed
as
(
9
)
:
=
+
(
9
)
T
h
e
s
en
s
itiv
ity
m
etr
ic
ac
t
s
as
a
n
in
d
icato
r
o
f
th
e
ef
f
ec
tiv
en
es
s
in
d
etec
tin
g
AD
p
atien
t
s
,
r
ef
lectin
g
th
e
m
o
d
el'
s
ab
ilit
y
to
co
r
r
ec
tly
id
en
tify
th
o
s
e
wh
o
ar
e
tr
u
ly
af
f
ec
ted
b
y
th
e
d
is
ea
s
e.
Pre
cisi
o
n
m
ea
s
u
r
es
th
e
r
eliab
ilit
y
o
f
th
e
d
iag
n
o
s
is
,
r
ep
r
esen
tin
g
t
h
e
p
r
o
p
o
r
tio
n
o
f
i
n
d
iv
id
u
als
id
en
tifie
d
b
y
th
e
s
y
s
tem
as
h
av
in
g
th
e
d
is
ea
s
e
wh
o
ar
e
in
d
ee
d
s
er
io
u
s
ly
im
p
ac
ted
b
y
it.
T
h
is
ca
n
b
e
d
escr
ib
ed
as
(
10
)
:
=
+
(
10
)
T
h
e
F1
s
co
r
e
o
f
th
e
s
im
u
latio
n
is
d
escr
ib
ed
as th
e
av
er
a
g
e
o
f
th
e
s
en
s
itiv
ity
an
d
ac
cu
r
ac
y
.
1
=
2
×
(
×
×
)
(1
1
)
3
.
3
.
E
x
perim
ent
a
l
re
s
ults
3
.
3
.
1
.
Co
m
pa
riso
n r
esu
lt
s
t
ra
ditio
na
l
f
uzzy
C
-
m
e
a
ns
a
nd
im
pro
v
ed
f
uzzy
C
-
m
ea
ns
clus
t
er
ing
Up
o
n
co
m
p
letio
n
o
f
th
e
I
m
FC
m
alg
o
r
ith
m
,
c
o
n
v
e
r
g
en
ce
was
ac
h
iev
ed
b
y
t
h
e
3
5
th
iter
atio
n
,
th
er
eb
y
ex
ce
ed
in
g
th
e
p
r
eset
m
ax
im
u
m
iter
atio
n
th
r
esh
o
ld
.
T
h
e
co
s
t
v
alu
e
ex
p
er
ien
ce
d
a
s
ig
n
if
ican
t
r
ed
u
ctio
n
,
d
escen
d
in
g
f
r
o
m
3
8
5
.
0
1
to
0
.
0
4
9
as
th
e
iter
atio
n
s
ad
v
an
ce
d
.
T
h
is
d
ec
lin
e
s
ig
n
if
ies
th
e
alg
o
r
ith
m
’
s
co
n
v
e
r
g
en
ce
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
imp
r
o
ve
d
fu
z
z
y
C
-
mea
n
s
a
n
d
w
a
ters
h
ed
s
eg
men
ta
tio
n
to
cla
s
s
ify
A
lz
h
eime
r
’
s
…
(
E
s
r
a
a
H.
A
li
)
4089
to
war
d
s
an
o
p
tim
al
s
o
lu
tio
n
,
h
ig
h
lig
h
tin
g
t
h
e
ef
f
icien
cy
o
f
o
u
r
m
eth
o
d
in
ac
h
iev
in
g
r
esu
lts
in
a
s
h
o
r
te
r
tim
ef
r
am
e.
I
n
co
m
p
ar
is
o
n
,
th
e
co
n
v
en
tio
n
al
FC
M
alg
o
r
ith
m
a
ttain
ed
co
n
v
er
g
e
n
ce
at
th
e
7
0
t
h
iter
atio
n
,
in
tiatin
g
with
a
co
s
t
v
alu
e
o
f
4
9
0
7
.
9
,
a
n
d
th
u
s
r
eq
u
ir
in
g
m
o
r
e
tim
e
r
el
ativ
e
to
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
.
Fig
u
r
e
6
d
e
p
icts
th
e
co
s
t v
alu
es f
o
r
f
i
v
e
d
if
f
e
r
en
t i
m
ag
es a
lo
n
g
s
id
e
th
e
d
u
r
atio
n
t
ak
en
b
y
b
o
t
h
th
e
FC
M
an
d
I
m
FC
m
alg
o
r
ith
m
s
.
Fig
u
r
e
6
.
C
o
s
t a
n
d
tim
e
f
o
r
F
C
M
an
d
I
m
FC
m
3
.
3
.
2
.
Co
nv
o
lutio
na
l neura
l
net
wo
rk
re
s
ults wit
h t
ra
ditio
na
l
f
uzzy
C
-
m
ea
ns
T
r
ad
itio
n
al
FC
M
clu
s
ter
in
g
w
er
e
em
p
lo
y
ed
f
o
r
s
eg
m
e
n
tatio
n
,
wh
ich
was
th
en
in
p
u
t
in
to
C
NN
f
o
r
AD
class
if
icatio
n
.
Fig
u
r
e
7
illu
s
tr
ates
th
e
tr
ain
in
g
an
d
v
alid
ati
o
n
r
esu
lts
f
r
o
m
th
e
s
eg
m
en
te
d
MRI
b
r
ain
d
ataset.
T
h
e
class
if
icatio
n
r
esu
lts
d
em
o
n
s
tr
ated
a
test
ac
cu
r
ac
y
o
f
9
1
%
ac
h
iev
ed
o
v
er
5
0
ep
o
c
h
s
.
On
e
o
f
th
e
c
u
r
v
es
s
h
o
ws
a
r
ed
lin
e
r
ep
r
esen
tin
g
t
r
ain
in
g
lo
s
s
an
d
a
b
lu
e
lin
e
f
o
r
v
alid
atio
n
lo
s
s
,
wh
ile
an
o
th
er
cu
r
v
e
illu
s
tr
ates
a
r
ed
lin
e
f
o
r
v
alid
atio
n
ac
cu
r
ac
y
an
d
a
b
lu
e
lin
e
f
o
r
ac
c
u
r
ac
y
.
Acc
o
r
d
in
g
to
t
h
is
tech
n
iq
u
e
,
t
h
e
f
ig
u
r
e
in
d
icate
s
th
at
b
o
th
ac
cu
r
ac
y
an
d
v
alid
atio
n
ac
cu
r
ac
y
c
o
n
v
e
r
g
ed
af
ter
1
0
ep
o
ch
s
.
T
h
e
v
alid
ati
o
n
ac
c
u
r
ac
y
r
ea
ch
ed
0
.
8
8
an
d
r
em
ai
n
ed
c
o
n
s
tan
t
f
r
o
m
e
p
o
ch
1
4
t
h
r
o
u
g
h
t
o
ep
o
ch
5
0
.
B
y
ep
o
ch
5
0
,
th
e
lo
s
s
d
ec
r
ea
s
ed
to
0
.
0
4
,
an
d
th
e
v
alid
atio
n
lo
s
s
r
ea
ch
ed
0
.
6
.
0
50
1
0
0
1
5
0
2
0
0
2
5
0
3
0
0
3
5
0
4
0
0
4
5
0
1
4
7
10
13
16
19
22
25
28
31
34
37
C
O
S
T
I
t
e
r
a
t
i
o
n
Im
FC
m
T
i
m
e:
I
m
ag
e1
:
4
.
3
2
2
I
m
a
g
e2
:
4
.
3
4
6
I
m
a
g
e3
:
4
.
6
4
5
I
m
a
g
e4
:
4
.
9
7
3
I
m
a
g
e5
:
4
.
4
2
6
0
1
0
0
0
0
2
0
0
0
0
3
0
0
0
0
4
0
0
0
0
5
0
0
0
0
6
0
0
0
0
1
4
7
10
13
16
19
22
25
28
31
34
37
40
43
46
49
52
55
58
61
64
67
70
C
O
S
T
I
t
e
r
a
t
i
o
n
F
CM
T
i
m
e
:
Im
ag
e
1:
23.723
Im
ag
e
2:
24.4
26
Im
ag
e
3:
25.3
80
Im
ag
e
4:
26.3
12
Im
ag
e
5:
27.2
55
Evaluation Warning : The document was created with Spire.PDF for Python.