I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
39
,
No
.
1
,
J
u
l
y
2
0
2
5
,
p
p
.
644
~
6
5
5
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
39
.i
1
.
pp
64
4
-
6
5
5
644
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Adv
a
nced cer
v
ica
l cancer c
la
ss
ificatio
n:
en
ha
ncing
p
a
p smea
r
ima
g
es with
hy
bri
d P
M
D
filt
er
-
CL
AH
E
Ach
K
ho
za
im
i
1,
2
,
I
s
na
ni Da
rt
i
1
,
Sy
a
if
ul Ana
m
1
,
Wury
a
ns
a
ri
M
uh
a
rini K
u
s
um
a
wina
h
y
u
1
1
D
e
p
a
r
t
me
n
t
o
f
M
a
t
h
e
m
a
t
i
c
s
,
F
a
c
u
l
t
y
o
f
M
a
t
h
e
mat
i
c
s
a
n
d
N
a
t
u
r
a
l
S
c
i
e
n
c
e
s,
B
r
a
w
i
j
a
y
a
U
n
i
v
e
r
si
t
y
,
M
a
l
a
n
g
,
I
n
d
o
n
e
si
a
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
,
F
a
c
u
l
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
a
s
Tr
u
n
o
j
o
y
o
M
a
d
u
r
a
,
M
a
d
u
r
a
,
I
n
d
o
n
e
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
16
,
2
0
2
4
R
ev
is
ed
Mar
12
,
2
0
2
5
Acc
ep
ted
Mar
26
,
2
0
2
5
Ce
rv
ica
l
c
a
n
c
e
r
re
m
a
in
s
a
sig
n
ifi
c
a
n
t
h
e
a
lt
h
p
r
o
b
lem
,
e
sp
e
c
ially
in
d
e
v
e
lo
p
in
g
c
o
u
n
tr
ies
.
Early
d
e
t
e
c
ti
o
n
is
c
rit
ica
l
f
o
r
e
ffe
c
ti
v
e
trea
tme
n
t.
Co
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
(
CNN
)
h
a
v
e
sh
o
w
n
p
ro
m
ise
in
a
u
to
m
a
ted
c
e
r
v
i
c
a
l
c
a
n
c
e
r
sc
r
e
e
n
i
n
g
,
b
u
t
t
h
e
i
r
p
e
r
f
o
r
m
a
n
c
e
d
e
p
e
n
d
s
o
n
p
a
p
s
m
e
a
r
i
m
a
g
e
q
u
a
l
i
t
y
.
T
h
i
s
s
t
u
d
y
i
n
v
e
s
t
i
g
a
t
e
s
th
e
i
m
p
a
c
t
o
f
v
a
r
i
o
u
s
i
m
a
g
e
p
r
e
p
r
o
c
e
s
s
i
n
g
t
e
c
h
n
i
q
u
e
s
o
n
C
N
N
p
e
r
f
o
r
m
a
n
c
e
f
o
r
c
e
r
v
i
c
a
l
c
a
n
c
e
r
c
la
s
s
if
i
c
a
t
i
o
n
u
s
i
n
g
t
h
e
S
I
P
a
K
M
e
D
d
a
ta
s
e
t
.
T
h
r
e
e
p
r
e
p
r
o
c
e
ss
i
n
g
t
e
c
h
n
i
q
u
e
s
w
e
r
e
e
v
a
l
u
a
t
e
d
:
P
e
r
o
n
a
-
M
a
l
ik
d
i
f
f
u
s
i
o
n
(
P
M
D
)
f
i
l
t
e
r
f
o
r
n
o
i
s
e
r
e
d
u
c
t
i
o
n
,
c
o
n
t
r
a
s
t
-
l
i
m
i
t
e
d
a
d
a
p
t
i
v
e
h
i
s
t
o
g
r
a
m
e
q
u
a
l
i
z
a
t
i
o
n
(
C
L
A
H
E
)
f
o
r
i
m
a
g
e
c
o
n
t
r
a
s
t
e
n
h
a
n
c
e
m
e
n
t,
a
n
d
t
h
e
p
r
o
p
o
s
e
d
h
y
b
r
i
d
P
M
D
f
i
l
t
e
r
-
C
L
A
H
E
a
p
p
r
o
a
c
h
.
T
h
e
e
n
h
a
n
c
e
d
i
m
a
g
e
d
a
t
a
se
ts
w
e
r
e
e
v
a
l
u
a
t
e
d
o
n
p
r
e
t
r
a
i
n
e
d
m
o
d
e
l
s
,
s
u
c
h
a
s
R
e
s
Ne
t
-
3
4
,
R
e
s
Ne
t
-
5
0
,
S
q
u
e
e
z
e
Ne
t
-
1
.
0
,
M
o
b
i
l
e
N
e
t
-
V
2
,
E
f
f
i
c
i
e
n
t
Ne
t
-
B
0
,
E
f
f
ic
i
e
n
t
N
e
t
-
B
1
,
D
e
n
se
N
et
-
1
2
1
,
a
n
d
D
e
n
s
e
N
e
t
-
2
0
1
.
T
h
e
r
e
su
l
t
s
s
h
o
w
t
h
a
t
h
y
b
r
i
d
p
r
e
p
r
o
c
e
s
si
n
g
P
M
D
f
i
l
t
e
r
-
C
L
AH
E
c
a
n
i
m
p
r
o
v
e
t
h
e
p
a
p
s
m
e
a
r
im
a
g
e
q
u
a
l
i
t
y
a
n
d
C
N
N
a
r
c
h
i
t
e
c
t
u
r
e
p
e
r
f
o
r
m
a
n
c
e
c
o
m
p
a
re
d
t
o
t
h
e
o
r
i
g
i
n
a
l
i
m
a
g
e
s
.
T
h
e
m
a
x
i
m
u
m
m
e
t
r
ic
i
m
p
r
o
v
e
m
e
n
t
s
a
re
1
3
.
6
2
%
f
o
r
a
c
c
u
r
a
c
y
,
1
0
.
0
4
%
f
o
r
p
r
e
c
i
s
i
o
n
,
1
3
.
0
8
%
f
o
r
r
e
c
a
l
l
,
a
n
d
1
4
.
3
4
%
f
o
r
F
1
-
s
c
o
r
e
.
T
h
e
p
r
o
p
o
s
e
d
h
y
b
r
i
d
P
M
D
f
i
l
t
e
r
-
C
L
A
H
E
t
e
c
h
n
i
q
u
e
o
f
f
e
r
s
a
n
e
w
p
e
rs
p
e
c
t
i
v
e
i
n
i
m
p
r
o
v
i
n
g
c
e
r
v
i
c
a
l
c
a
n
c
e
r
c
las
s
i
f
i
c
a
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
u
s
i
n
g
C
N
N
a
rc
h
i
t
e
c
tu
r
e
s
.
K
ey
w
o
r
d
s
:
C
er
v
ical
ca
n
ce
r
C
L
AHE
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Pap
s
m
ea
r
Per
o
n
a
-
m
alik
d
if
f
u
s
io
n
f
ilter
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
:
Sy
aif
u
l A
n
am
Dep
ar
tm
en
t o
f
Ma
th
em
atics,
Facu
lty
o
f
Ma
th
e
m
atics a
n
d
Na
tu
r
al
Scien
ce
s
,
B
r
awijay
a
Un
iv
er
s
ity
Ma
lan
g
,
6
5
1
4
5
,
I
n
d
o
n
esia
E
m
ail: sy
aif
u
l@
u
b
.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
C
er
v
ical
ca
n
ce
r
is
th
e
f
o
u
r
th
m
o
s
t
co
m
m
o
n
ca
n
ce
r
in
wo
m
en
wo
r
ld
wid
e
[
1
]
.
Acc
o
r
d
in
g
t
o
th
e
g
lo
b
al
ca
n
ce
r
o
b
s
er
v
at
o
r
y
(
GC
O)
,
c
er
v
ical
ca
n
ce
r
was
r
esp
o
n
s
ib
l
e
f
o
r
an
esti
m
ated
5
7
0
,
0
0
0
n
ew
d
iag
n
o
s
es
an
d
3
1
1
,
0
0
0
f
atalities
wo
r
ld
wid
e
in
2
0
1
8
[
2
]
.
I
n
d
ev
elo
p
in
g
n
a
tio
n
s
,
n
ea
r
ly
9
0
%
o
f
th
e
d
e
ath
s
ar
e
r
elate
d
to
ce
r
v
ical
ca
n
ce
r
[
3
]
.
A
c
o
m
m
o
n
ly
em
p
lo
y
ed
tech
n
iq
u
e
f
o
r
e
ar
ly
ca
n
ce
r
d
etec
tio
n
is
t
h
e
p
a
p
s
m
ea
r
test
,
wh
ich
in
v
o
lv
es
ex
tr
ac
tin
g
ce
lls
f
r
o
m
th
e
ce
r
v
ix
a
n
d
a
n
aly
zin
g
t
h
e
m
to
id
en
tify
p
r
ec
an
ce
r
o
u
s
o
r
ca
n
ce
r
o
u
s
ch
a
n
g
es
[
4
]
.
Desp
ite
its
ef
f
ec
tiv
en
ess
,
tr
ad
itio
n
al
p
ap
s
m
ea
r
an
aly
s
is
is
lab
o
r
-
in
ten
s
iv
e
an
d
s
u
b
je
ct
to
h
u
m
an
er
r
o
r
,
n
ec
ess
itatin
g
th
e
d
ev
elo
p
m
en
t
o
f
a
u
to
m
ated
an
d
r
eliab
le
d
i
ag
n
o
s
tic
to
o
ls
[
5
]
.
W
ith
th
e
a
d
v
en
t
o
f
ad
v
an
ce
d
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es,
p
ar
tic
u
lar
ly
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NN)
,
th
er
e
i
s
an
o
p
p
o
r
t
u
n
ity
t
o
en
h
an
ce
th
e
ac
cu
r
ac
y
an
d
ef
f
icien
cy
o
f
ce
r
v
ical
ca
n
ce
r
d
etec
tio
n
u
s
in
g
a
u
to
m
ated
im
ag
e
an
aly
s
is
[
6
]
.
C
NN
h
as
r
ev
o
l
u
tio
n
ized
m
ed
ical
i
m
ag
e
class
if
icatio
n
b
ec
a
u
s
e
it
ca
n
a
u
to
m
atica
lly
e
x
t
r
ac
t
a
n
d
lear
n
f
ea
tu
r
es
f
r
o
m
co
m
p
lex
d
atasets
[
7
]
.
H
o
wev
e
r
,
th
e
p
er
f
o
r
m
an
ce
o
f
C
NN
m
o
d
els
is
h
ea
v
ily
i
n
f
lu
en
ce
d
b
y
im
ag
e
q
u
ality
[
8
]
.
Mo
s
t
p
ap
s
m
ea
r
im
ag
es
ex
h
ib
it
lo
w
co
n
tr
ast
an
d
s
ig
n
i
f
ican
t
n
o
is
e
[
9
]
.
No
is
e
r
ed
u
ctio
n
an
d
c
o
n
tr
ast
en
h
an
ce
m
e
n
t
h
a
v
e
b
ee
n
p
r
o
p
o
s
ed
to
im
p
r
o
v
e
C
NN
p
er
f
o
r
m
an
ce
[
1
0
]
.
I
m
ag
e
p
r
ep
r
o
ce
s
s
in
g
is
a
cr
u
cia
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
d
va
n
ce
d
ce
r
vica
l c
a
n
ce
r
cla
s
s
ifica
tio
n
:
en
h
a
n
cin
g
p
a
p
s
me
a
r
ima
g
es w
ith
h
yb
r
id
P
MD
…
(
A
ch
K
h
o
z
a
imi
)
645
s
tep
th
at
ca
n
s
ig
n
if
ican
tly
af
f
e
ct
th
e
o
u
tco
m
es
o
f
d
ee
p
lea
r
n
i
n
g
m
o
d
els.
I
m
ag
e
c
o
n
tr
ast
is
r
eq
u
ir
ed
to
p
r
o
d
u
ce
an
ac
cu
r
ate
m
o
d
el
b
ec
a
u
s
e
class
if
icat
io
n
f
ea
tu
r
es
ar
e
u
s
u
ally
b
ased
o
n
n
u
clea
r
an
d
cy
to
p
lasmic
ch
ar
ac
ter
is
tics
[
1
1
]
.
Sev
er
al
s
tu
d
ies
h
av
e
d
em
o
n
s
t
r
ated
th
at
r
ed
u
cin
g
im
ag
e
n
o
is
e
s
ig
n
if
ican
tly
en
h
an
ce
s
class
if
icatio
n
p
er
f
o
r
m
an
ce
[
1
2
]
.
T
h
e
Per
o
n
a
-
Ma
lik
d
if
f
u
s
io
n
(
PMD
)
f
ilter
r
em
o
v
es n
o
is
e,
s
m
o
o
t
h
en
s
im
ag
es,
an
d
m
ain
tain
s
cr
u
ci
al
ed
g
e
d
etails
[
1
3
]
.
I
t
ap
p
lies
a
m
o
d
if
ied
Gau
s
s
ian
f
u
n
ctio
n
,
ass
ig
n
in
g
g
r
ea
ter
weig
h
ts
to
ce
n
tr
al
p
ix
els
an
d
lo
wer
weig
h
ts
to
th
o
s
e
at
th
e
ed
g
es
[
1
4
]
.
R
esear
ch
in
d
icate
s
th
at
PMD
f
il
ter
s
p
lay
a
k
ey
r
o
le
in
d
etec
tin
g
an
d
ex
tr
ac
ti
n
g
m
alig
n
an
t
tu
m
o
r
s
in
m
e
d
ical
im
ag
es
[
1
5
]
a
n
d
h
a
v
e
b
ee
n
s
h
o
w
n
to
e
n
h
a
n
ce
d
ee
p
lear
n
in
g
p
er
f
o
r
m
an
ce
i
n
ce
r
v
ical
ca
n
ce
r
class
if
icatio
n
[
1
6
]
.
I
n
a
d
d
iti
o
n
,
n
o
is
e
r
em
o
v
al
a
n
d
c
o
n
tr
a
s
t
en
h
an
ce
m
en
t
ar
e
ess
en
tial f
o
r
im
p
r
o
v
in
g
p
ap
s
m
ea
r
im
ag
e
q
u
ality
.
C
o
n
tr
ast
-
l
im
ited
ad
ap
tiv
e
h
is
to
g
r
am
e
q
u
aliza
tio
n
(
C
L
AHE
)
,
an
im
p
r
o
v
e
d
v
er
s
io
n
o
f
ad
a
p
t
iv
e
h
is
to
g
r
am
eq
u
aliza
tio
n
(
A
HE
)
,
r
estricts
ex
ce
s
s
iv
e
co
n
tr
ast
en
h
an
ce
m
en
t
to
p
r
ev
en
t
a
r
tifa
cts
[
1
7
]
.
I
t
h
as
s
ig
n
if
ican
tly
en
h
an
ce
d
p
a
p
s
m
ea
r
im
ag
es
an
d
im
p
r
o
v
ed
th
e
class
if
icatio
n
ac
cu
r
ac
y
o
f
d
ee
p
lear
n
i
n
g
m
o
d
els s
u
ch
as
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
-
1
6
(
VGG1
6
)
,
I
n
ce
p
tio
n
V
3
,
an
d
E
f
f
icien
tNet
in
ce
r
v
ical
ca
n
ce
r
d
iag
n
o
s
is
[
1
8
]
.
C
L
AHE
h
as
also
b
ee
n
s
u
cc
ess
f
u
l
in
im
p
r
o
v
in
g
im
ag
e
q
u
ality
an
d
b
o
o
s
tin
g
th
e
p
er
f
o
r
m
a
n
ce
o
f
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
s
u
ch
as
ar
tific
ial
n
eu
r
a
l
n
etwo
r
k
s
(
ANN)
an
d
K
-
n
ea
r
est
n
eig
h
b
o
r
s
(
KNN)
in
ce
r
v
ical
ca
n
ce
r
class
if
icatio
n
[
1
9
]
.
Ad
d
itio
n
ally
,
r
esear
c
h
h
as
s
h
o
wn
th
at
C
L
AHE
en
h
an
ce
s
th
e
ab
ilit
y
o
f
th
e
y
o
u
o
n
ly
lo
o
k
o
n
ce
(
YOL
O
)
alg
o
r
ith
m
to
r
ec
o
g
n
ize
r
o
ad
m
ar
k
in
g
s
at
n
ig
h
t
[
2
0
]
,
im
p
r
o
v
es
C
NN
-
b
ased
s
eg
m
en
tatio
n
o
f
lu
n
g
ca
n
ce
r
in
co
m
p
u
ted
to
m
o
g
r
ap
h
y
(
CT
)
s
ca
n
im
ag
es
[
2
1
]
,
a
n
d
en
h
an
ce
s
wate
r
-
im
ag
e
class
if
icatio
n
[
2
2
]
.
Giv
en
th
eir
s
u
cc
es
s
,
it is
e
s
s
en
tial
to
in
v
esti
g
ate
wh
eth
er
co
m
b
i
n
in
g
b
o
th
tech
n
iq
u
es c
an
im
p
r
o
v
e
c
lass
if
icatio
n
p
er
f
o
r
m
an
ce
.
T
h
is
s
tu
d
y
ex
p
lo
r
es
th
e
ef
f
ec
t
s
o
f
a
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
o
n
ce
r
v
ical
ce
ll
im
ag
e
q
u
a
lity
.
W
h
il
e
PMD
f
ilter
s
ar
e
k
n
o
wn
f
o
r
n
o
is
e
r
ed
u
ctio
n
an
d
ed
g
e
p
r
ese
r
v
atio
n
,
th
eir
im
p
ac
t
o
n
co
n
tr
ast
en
h
an
ce
m
en
t
is
less
s
tu
d
ied
.
Similar
ly
,
C
L
AHE
im
p
r
o
v
es
co
n
t
r
ast,
b
u
t
its
r
o
le
in
n
o
is
e
s
u
p
p
r
es
s
io
n
an
d
s
tr
u
ctu
r
al
co
n
s
er
v
atio
n
in
m
ed
ical
im
a
g
in
g
is
u
n
clea
r
.
T
h
is
s
tu
d
y
ex
am
in
es
th
eir
co
m
b
in
ed
ef
f
e
cts
to
ad
d
r
ess
th
es
e
g
ap
s
.
T
h
e
PMD
f
ilter
r
ed
u
ce
s
n
o
is
e
wh
ile
p
r
eser
v
in
g
e
d
g
e
d
etails
an
d
im
p
o
r
tan
t
im
ag
e
s
tr
u
ctu
r
es
[
2
3
]
.
B
y
co
n
tr
ast,
C
L
AHE
en
h
an
ce
s
im
ag
e
co
n
tr
ast
an
d
im
p
r
o
v
es
th
e
v
is
ib
ilit
y
o
f
cr
itical
f
ea
tu
r
es
f
o
r
C
NN
-
b
ased
class
if
icatio
n
[
2
4
]
.
Stu
d
ies
h
av
e
s
h
o
wn
th
at
C
L
AHE
-
b
ase
d
en
tr
o
p
y
an
al
y
s
is
is
m
o
r
e
ef
f
ec
tiv
e
th
an
o
th
er
m
eth
o
d
s
in
en
h
a
n
cin
g
m
ed
ical
im
ag
es
[
2
5
]
.
T
o
ass
ess
th
e
ef
f
e
ctiv
en
ess
o
f
th
ese
p
r
ep
r
o
ce
s
s
in
g
tech
n
i
q
u
es,
we
co
n
d
u
cte
d
ex
p
e
r
im
en
ts
u
s
in
g
th
e
SIPaK
Me
D
d
ataset,
a
wid
ely
u
s
ed
d
ataset
f
o
r
ce
r
v
ical
c
an
ce
r
class
if
icatio
n
[
2
6
]
.
T
h
e
s
tu
d
y
was
ev
alu
ated
u
s
in
g
C
NN
ar
ch
itectu
r
es,
in
clu
d
in
g
R
esNet3
4
,
R
esNet5
0
,
Sq
u
ee
ze
Net
-
1
.
0
,
Mo
b
ileNet
-
V2
,
E
f
f
icien
tNet
-
B
0
,
E
f
f
icien
tNet
-
B
1
,
Den
s
eNe
t1
2
1
,
an
d
Den
s
eNe
t2
0
1
,
to
m
ea
s
u
r
e
th
ei
r
class
if
icatio
n
p
er
f
o
r
m
an
ce
u
s
in
g
d
if
f
er
e
n
t
p
r
ep
r
o
ce
s
s
in
g
m
et
h
o
d
s
.
T
h
e
k
ey
ev
alu
atio
n
m
etr
ics
in
clu
d
ed
im
ag
e
q
u
ality
ass
ess
m
en
t,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
tr
ain
in
g
tim
e.
I
m
ag
e
q
u
ality
was
ass
e
s
s
ed
u
s
in
g
th
e
co
n
tr
ast en
h
a
n
ce
m
en
t
-
b
ased
im
ag
e
q
u
ality
(
C
E
I
Q)
m
etr
ic
[
2
7
]
.
C
NN
ar
ch
itectu
r
es
wer
e
s
elec
ted
b
ased
o
n
th
eir
u
n
iq
u
e
s
tr
en
g
th
s
in
d
ee
p
lear
n
in
g
a
p
p
licatio
n
s
.
R
esNet
in
tr
o
d
u
ce
s
r
esid
u
al
lear
n
in
g
wi
th
s
k
ip
c
o
n
n
ec
tio
n
s
,
wh
ich
ef
f
ec
tiv
ely
m
itig
ates
th
e
v
an
is
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
a
n
d
e
n
ab
les
th
e
tr
ain
in
g
o
f
v
er
y
d
ee
p
n
etwo
r
k
s
[
2
8
]
.
Sq
u
ee
ze
Net,
d
esig
n
ed
f
o
r
ef
f
icien
cy
,
u
s
es
f
ir
e
m
o
d
u
les
to
r
ed
u
ce
th
e
n
u
m
b
er
o
f
p
ar
am
ete
r
s
,
m
ak
in
g
it
a
n
id
ea
l
c
h
o
ice
f
o
r
ta
s
k
s
th
at
r
eq
u
ir
e
lig
h
tweig
h
t
m
o
d
els
with
m
in
im
al
m
em
o
r
y
u
s
ag
e
wh
ile
m
ain
tain
in
g
co
m
p
etitiv
e
ac
cu
r
ac
y
[
2
9
]
.
Mo
b
ile
Net,
o
p
tim
ized
f
o
r
m
o
b
ile
an
d
em
b
ed
d
ed
v
is
io
n
a
p
p
licatio
n
s
,
em
p
lo
y
s
d
ep
th
-
wis
e
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
s
,
s
ig
n
if
ican
tly
r
ed
u
cin
g
th
e
m
o
d
el
s
ize
an
d
co
m
p
u
tati
o
n
al
r
eq
u
ir
em
e
n
ts
f
o
r
ef
f
icien
t
p
r
o
ce
s
s
in
g
[
3
0
]
.
E
f
f
icie
n
tNet
b
alan
ce
s
n
etwo
r
k
d
ep
th
,
wid
th
,
a
n
d
r
eso
lu
tio
n
,
ac
h
iev
in
g
s
tate
-
of
-
t
h
e
-
ar
t
ac
cu
r
ac
y
with
f
ewe
r
p
ar
am
eter
s
an
d
f
lo
atin
g
-
p
o
in
t
o
p
er
atio
n
s
p
er
s
ec
o
n
d
(
FLO
Ps
)
th
an
o
th
er
m
o
d
els
[
3
1
]
.
Fin
al
ly
,
Den
s
eNe
t
e
s
tab
lis
h
es
d
en
s
e
co
n
n
ec
tio
n
s
b
etwe
en
lay
er
s
,
m
ax
im
izin
g
in
f
o
r
m
atio
n
f
lo
w
an
d
g
r
a
d
ien
t
p
r
o
p
a
g
atio
n
,
wh
ic
h
en
h
a
n
ce
s
f
ea
tu
r
e
r
eu
s
e
an
d
m
itig
ates th
e
v
an
is
h
in
g
g
r
ad
ie
n
t p
r
o
b
lem
[
3
2
].
T
h
e
ex
p
er
im
en
tal
r
esu
lts
s
h
o
w
th
at
ea
ch
p
r
e
p
r
o
ce
s
s
in
g
tec
h
n
iq
u
e
o
f
f
er
s
u
n
iq
u
e
ad
v
a
n
tag
es,
with
th
e
m
o
s
t
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t
o
b
s
er
v
ed
u
s
in
g
th
e
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
.
T
h
ese
f
in
d
in
g
s
em
p
h
asize
th
e
im
p
o
r
tan
ce
o
f
s
elec
tin
g
an
d
in
teg
r
atin
g
p
r
ep
r
o
ce
s
s
in
g
tec
h
n
iq
u
es
to
en
h
a
n
ce
m
ed
ical
im
ag
e
q
u
ality
an
d
im
p
r
o
v
e
d
iag
n
o
s
t
ic
ac
cu
r
ac
y
.
PMD
f
ilter
in
g
is
p
ar
ticu
la
r
ly
ef
f
ec
tiv
e
f
o
r
lig
h
tweig
h
t
ar
ch
itectu
r
es,
a
n
d
C
L
AHE
b
en
ef
its
d
ee
p
e
r
C
N
N
ar
ch
itectu
r
es.
Mo
s
t
im
p
o
r
t
an
tly
,
th
e
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
ap
p
r
o
ac
h
co
n
s
is
ten
tly
im
p
r
o
v
es
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
o
s
t
C
NN
ar
c
h
itectu
r
es,
b
alan
ci
n
g
n
o
is
e
r
e
d
u
ctio
n
an
d
co
n
tr
ast
en
h
an
ce
m
e
n
t
to
o
p
tim
ize
f
e
atu
r
e
ex
tr
ac
tio
n
.
T
h
is
s
tu
d
y
p
r
o
v
id
es
v
alu
a
b
le
in
s
ig
h
ts
f
o
r
r
ef
in
in
g
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
s
tr
ateg
ies.
T
h
is
co
n
tr
ib
u
tes
to
m
o
r
e
p
r
ec
is
e
an
d
r
eliab
le
ce
r
v
ical
ca
n
ce
r
d
etec
tio
n
u
s
in
g
d
ee
p
lear
n
in
g
m
o
d
el
s
.
W
e
f
o
u
n
d
t
h
at
im
ag
e
co
n
tr
ast
en
h
a
n
ce
m
en
t
an
d
n
o
is
e
r
ed
u
ctio
n
c
o
r
r
elate
with
f
ea
tu
r
e
ex
tr
ac
tio
n
ef
f
icie
n
cy
,
as
h
ig
h
er
co
n
tr
ast
lead
s
to
b
etter
-
d
ef
i
n
ed
ce
ll
s
tr
u
ctu
r
es.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
in
th
is
s
tu
d
y
ten
d
e
d
to
h
av
e
an
in
o
r
d
i
n
ately
h
ig
h
er
p
r
o
p
o
r
tio
n
o
f
we
ll
-
en
h
an
ce
d
im
ag
e
r
eg
io
n
s
as si
g
n
if
ican
t
f
ea
tu
r
es
in
C
NN
ac
tiv
atio
n
s
,
im
p
r
o
v
in
g
class
if
icatio
n
p
er
f
o
r
m
an
ce
a
cr
o
s
s
d
if
f
er
en
t a
r
c
h
itectu
r
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
39
,
No
.
1
,
J
u
ly
20
25
:
644
-
6
5
5
646
2.
M
E
T
H
O
D
2
.
1
.
M
a
t
er
ia
l,
m
et
ho
ds
,
a
nd
s
im
ula
t
io
n
Fig
u
r
e
1
illu
s
tr
ates
th
e
s
eq
u
en
tial
s
tep
s
o
f
p
r
ep
r
o
ce
s
s
in
g
p
ap
s
m
ea
r
im
ag
es
f
o
r
ce
r
v
i
ca
l
ca
n
ce
r
class
if
icatio
n
u
s
in
g
d
ee
p
lear
n
in
g
.
T
h
e
p
r
o
ce
s
s
b
eg
i
n
s
with
in
p
u
t
d
ata
u
s
in
g
th
e
SIPaK
Me
D
d
ataset.
Nex
t,
th
e
i
m
a
g
e
p
r
e
p
r
o
c
e
s
s
i
n
g
i
n
cl
u
d
e
d
a
P
MD
f
i
lt
e
r
a
n
d
C
L
AH
E
.
A
f
te
r
p
r
e
p
r
o
c
e
s
s
i
n
g
,
t
h
e
d
a
ta
s
et
wa
s
s
p
l
i
t
i
n
t
o
t
r
a
i
n
i
n
g
(
8
0
%
)
,
v
a
l
i
d
a
t
i
o
n
(
1
0
%
)
,
a
n
d
t
e
s
t
(
1
0
%
)
d
a
t
a
t
o
e
n
s
u
r
e
p
r
o
p
e
r
m
o
d
e
l
t
r
a
i
n
i
n
g
,
f
i
n
e
-
t
u
n
i
n
g
,
a
n
d
e
v
a
l
u
a
t
i
o
n
.
T
h
e
e
n
h
a
n
c
e
d
i
m
a
g
e
s
we
r
e
e
v
a
l
u
a
te
d
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
r
c
h
i
tec
t
u
r
e
s
,
s
u
c
h
a
s
R
es
N
et
,
E
f
f
i
c
ie
n
t
N
e
t
,
M
o
b
il
e
N
et
,
a
n
d
D
e
n
s
e
N
et
.
T
h
e
t
r
a
i
n
i
n
g
s
et
w
a
s
e
m
p
l
o
y
e
d
t
o
d
e
v
e
l
o
p
t
h
e
m
o
d
e
l
,
w
h
e
r
e
a
s
t
h
e
v
al
i
d
a
ti
o
n
s
e
t
w
as
u
t
il
i
z
e
d
to
o
p
t
i
m
i
z
e
h
y
p
e
r
p
a
r
a
m
e
t
e
r
s
a
n
d
m
i
t
i
g
a
te
o
v
e
r
f
i
t
ti
n
g
i
s
s
u
es
.
Af
t
e
r
t
r
a
i
n
i
n
g
,
t
h
e
m
o
d
e
l
w
as
ev
a
l
u
a
t
e
d
u
s
i
n
g
t
e
s
t
d
a
t
a
.
P
e
r
f
o
r
m
a
n
c
e
m
e
t
r
i
c
s
s
u
c
h
a
s
a
c
c
u
r
a
c
y
,
r
e
ca
l
l
,
F
1
-
s
c
o
r
e
,
an
d
p
r
e
c
i
s
i
o
n
w
e
r
e
ca
l
c
u
l
at
e
d
.
T
h
e
SIPaK
Me
D
d
ataset
is
a
p
u
b
licly
av
ailab
le
co
llectio
n
d
e
s
ig
n
ed
to
s
u
p
p
o
r
t
r
esear
ch
o
n
au
to
m
ated
ce
r
v
ical
ca
n
ce
r
class
if
icatio
n
[
2
6
]
.
T
h
e
SIPaK
Me
D
d
ataset
co
n
s
is
ted
o
f
4
,
0
4
9
h
i
g
h
-
r
eso
lu
tio
n
ce
r
v
ical
ce
ll
im
ag
es
ex
tr
ac
ted
f
r
o
m
p
ap
s
m
ea
r
s
lid
es
an
d
m
a
n
u
ally
a
n
n
o
tated
b
y
cy
to
p
ath
o
lo
g
y
e
x
p
er
ts
.
T
h
ese
im
ag
es
wer
e
ac
q
u
ir
ed
u
s
in
g
a
4
0
×
o
b
jectiv
e
len
s
to
en
s
u
r
e
th
at
d
etailed
ce
llu
lar
s
tr
u
ctu
r
es
w
er
e
p
r
eser
v
ed
.
T
h
e
d
ataset
in
clu
d
ed
is
o
lated
s
in
g
l
e
-
ce
ll
im
ag
es,
in
d
icatin
g
th
at
ea
ch
im
ag
e
co
n
tain
ed
a
s
in
g
le
ce
r
v
ical
ce
ll
r
ath
e
r
th
an
clu
s
ter
s
o
f
m
u
ltip
le
ce
lls
.
T
h
e
im
ag
es
s
h
o
wed
s
ize
,
co
n
tr
ast,
illu
m
in
atio
n
,
an
d
s
tain
in
g
in
ten
s
ity
v
ar
iatio
n
s
.
T
h
e
SIPaK
Me
D
d
ataset
was
d
iv
id
ed
in
to
f
iv
e
ca
teg
o
r
ies:
s
u
p
er
f
icial
s
q
u
am
o
u
s
ep
ith
elial
ce
lls
,
in
ter
m
ed
iate
s
q
u
am
o
u
s
ep
ith
e
lial
ce
ll
s
,
h
ig
h
-
g
r
ad
e
s
q
u
am
o
u
s
in
tr
ae
p
ith
elial
lesi
o
n
(
HSI
L
)
ce
lls
,
co
lu
m
n
ar
ep
ith
elial
ce
lls
,
an
d
lo
w
-
g
r
ad
e
s
q
u
am
o
u
s
in
tr
ae
p
ith
elial
lesi
o
n
(
L
SIL
)
ce
lls
.
E
ac
h
ca
teg
o
r
y
co
m
p
r
is
ed
ap
p
r
o
x
im
ately
8
0
0
–
8
5
0
im
a
g
es.
SIPaK
Me
D
is
wid
ely
u
s
ed
in
d
ee
p
lear
n
in
g
-
b
as
ed
ce
r
v
ical
ca
n
ce
r
c
lass
if
icatio
n
an
d
is
a
b
en
ch
m
ar
k
f
o
r
test
in
g
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es,
f
ea
tu
r
e
ex
tr
a
ctio
n
m
eth
o
d
s
,
an
d
C
NN
ar
ch
itectu
r
es.
T
h
is
d
ataset
p
r
o
v
id
es
a
r
ea
lis
tic
ch
allen
g
e
f
o
r
d
ev
el
o
p
in
g
r
o
b
u
s
t
an
d
g
en
er
aliza
b
l
e
ce
r
v
ical
ca
n
ce
r
s
cr
ee
n
i
n
g
m
o
d
els
[
3
3
]
.
Fig
u
r
e
1
.
Me
th
o
d
an
d
s
im
u
latio
n
f
lo
wch
a
r
t
Pap
s
m
ea
r
im
ag
es
o
f
ten
co
n
ta
in
n
o
is
e
with
a
lo
w
co
n
tr
ast
[
9
]
.
Pre
p
r
o
ce
s
s
in
g
is
cr
u
cial
f
o
r
en
h
an
cin
g
th
e
im
ag
e
q
u
ality
f
o
r
ce
r
v
ical
ca
n
ce
r
class
if
icatio
n
.
T
h
is
s
tu
d
y
ex
p
lo
r
ed
th
e
im
p
ac
t
o
f
d
if
f
er
e
n
t
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es
o
n
C
NN
p
er
f
o
r
m
an
ce
,
s
p
ec
if
ical
ly
f
o
cu
s
in
g
o
n
th
e
PMD
f
ilter
,
C
L
AHE
,
an
d
a
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
ap
p
r
o
ac
h
.
Fig
u
r
e
2
s
h
o
ws a
f
lo
w
ch
ar
t d
ep
ictin
g
th
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
ap
p
lied
to
th
e
p
ap
s
m
ea
r
im
a
g
es
f
r
o
m
t
h
e
SIPaK
Me
D
d
ataset
u
s
in
g
a
co
m
b
in
atio
n
o
f
th
e
PMD
f
ilter
an
d
C
L
AHE
tech
n
iq
u
es.
I
n
itially
,
th
e
p
ap
s
m
ea
r
im
ag
es
wer
e
s
p
lit
in
t
o
c
o
lo
r
s
p
ac
es:
r
ed
(
R
)
,
g
r
ee
n
(
G)
,
an
d
b
lu
e
(
B
)
.
A
PMD
f
ilter
an
d
/
o
r
C
L
AHE
w
er
e
ap
p
lied
to
ea
ch
co
l
o
r
s
p
ac
e.
T
h
is
s
eq
u
en
tial
ap
p
licatio
n
aim
s
to
en
h
an
ce
th
e
co
n
tr
ast
an
d
d
en
o
is
e
ea
ch
co
lo
r
co
m
p
o
n
en
t.
T
h
e
R
,
G,
an
d
B
co
m
p
o
n
e
n
ts
ar
e
en
h
an
ce
d
u
s
i
n
g
a
PMD
f
ilter
o
r
C
L
AHE
,
r
esu
ltin
g
in
im
p
r
o
v
ed
an
d
d
en
o
is
ed
R
,
G,
an
d
B
co
m
p
o
n
e
n
ts
.
Af
ter
p
r
o
ce
s
s
in
g
ea
ch
co
lo
r
co
m
p
o
n
en
t,
th
e
co
n
tr
ast
en
h
a
n
ce
m
en
t
an
d
d
en
o
is
ed
R
-
,
G
-
,
an
d
B
-
co
m
p
o
n
en
ts
wer
e
co
m
b
in
ed
to
o
b
tain
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
d
va
n
ce
d
ce
r
vica
l c
a
n
ce
r
cla
s
s
ifica
tio
n
:
en
h
a
n
cin
g
p
a
p
s
me
a
r
ima
g
es w
ith
h
yb
r
id
P
MD
…
(
A
ch
K
h
o
z
a
imi
)
647
f
i
n
a
l
e
n
h
a
n
c
e
d
p
a
p
s
m
e
a
r
i
m
ag
e
s
.
T
h
e
P
M
D
f
i
l
t
e
r
h
el
p
s
r
e
d
u
c
e
n
o
i
s
e
w
h
i
l
e
p
r
es
e
r
v
i
n
g
es
s
e
n
t
i
a
l
ce
l
l
s
t
r
u
ct
u
r
e
s
.
I
t
w
o
r
k
s
b
y
s
m
o
o
t
h
i
n
g
u
n
i
f
o
r
m
a
r
e
a
s
w
h
i
le
m
a
i
n
t
a
i
n
i
n
g
e
d
g
e
s
,
e
n
s
u
r
i
n
g
t
h
a
t
c
r
u
c
ia
l
d
e
t
a
il
s
a
r
e
n
o
t
l
o
s
t
.
P
MD
f
i
l
t
e
r
h
as
t
h
r
e
e
m
a
i
n
p
a
r
a
m
e
te
r
s
:
i
t
e
r
at
i
o
n
s
(
N
)
,
d
i
f
f
u
s
i
o
n
co
e
f
f
i
c
i
e
n
t
(
κ
)
,
a
n
d
t
i
m
e
s
t
e
p
(
λ
)
.
M
o
r
e
i
t
e
r
at
i
o
n
s
e
n
h
a
n
c
e
n
o
i
s
e
r
e
m
o
v
a
l
b
u
t
m
a
y
b
l
u
r
t
h
e
d
e
t
a
i
l
s
.
T
h
e
d
i
f
f
u
s
i
o
n
c
o
e
f
f
i
c
i
e
n
t
c
o
n
t
r
o
l
s
e
d
g
e
p
r
e
s
e
r
v
a
t
i
o
n
;
l
o
w
v
a
l
u
es
k
e
e
p
e
d
g
e
s
s
h
a
r
p
,
w
h
e
r
e
a
s
h
i
g
h
v
a
l
u
e
s
s
m
o
o
t
h
t
h
e
i
m
a
g
e
.
T
h
e
t
i
m
e
s
t
e
p
a
f
f
e
ct
s
s
t
a
b
i
l
i
t
y
an
d
s
h
o
u
l
d
b
e
≤
0
.
2
5
;
s
m
a
l
le
r
v
a
l
u
e
s
f
i
l
t
e
r
g
r
a
d
u
a
ll
y
,
w
h
i
l
e
l
a
r
g
e
r
v
al
u
e
s
s
p
e
e
d
u
p
d
i
f
f
u
s
i
o
n
b
u
t
r
i
s
k
o
v
e
r
-
s
m
o
o
t
h
i
n
g
[
3
4
]
.
I
n
t
h
i
s
s
t
u
d
y
,
w
e
s
e
t
N
=
2
0
,
κ
=
2
0
,
a
n
d
λ
=
0
.
2
5
.
A
d
e
t
ai
l
e
d
e
x
p
l
a
n
ati
o
n
o
f
t
h
e
P
MD
f
i
lt
e
r
is
p
r
o
v
id
e
d
i
n
s
e
ct
i
o
n
2
.
2
.
C
L
AH
E
e
n
h
a
n
c
e
s
c
o
n
t
r
a
s
t
b
y
l
o
c
a
l
l
y
a
d
j
u
s
t
i
n
g
t
h
e
b
r
i
g
h
t
n
ess
i
n
s
m
al
l
i
m
a
g
e
r
e
g
i
o
n
s
,
t
h
e
r
eb
y
p
r
e
v
e
n
t
i
n
g
o
v
e
r
-
e
n
h
a
n
c
e
m
e
n
t
i
n
b
r
i
g
h
t
o
r
d
a
r
k
a
r
e
a
s
.
C
L
A
H
E
h
a
s
t
w
o
m
a
i
n
p
a
r
a
m
e
t
e
r
s
:
c
l
i
p
l
i
m
it
a
n
d
b
l
o
c
k
s
i
z
e
.
T
h
e
c
l
i
p
l
i
m
it
c
o
n
t
r
o
l
s
c
o
n
t
r
as
t
a
n
d
l
o
w
v
a
l
u
e
s
p
r
e
v
e
n
t
o
v
e
r
-
e
n
h
a
n
c
e
m
e
n
t
,
w
h
e
r
e
a
s
h
i
g
h
v
a
l
u
es
b
o
o
s
t
c
o
n
t
r
a
s
t
b
u
t
m
a
y
a
d
d
n
o
i
s
e
.
T
h
e
b
l
o
c
k
s
i
z
e
d
e
f
i
n
es
th
e
l
o
c
a
l
r
e
g
i
o
n
s
.
S
m
a
l
l
b
l
o
c
k
s
e
n
h
a
n
c
e
d
e
t
a
i
ls
,
w
h
il
e
l
a
r
g
e
b
lo
c
k
s
g
i
v
e
s
m
o
o
t
h
e
r
r
e
s
u
l
ts
[
3
4
]
.
I
n
t
h
i
s
s
t
u
d
y
,
w
e
s
e
t
t
h
e
c
l
i
p
l
i
m
i
t
t
o
2
.
0
a
n
d
t
h
e
b
l
o
c
k
s
i
ze
t
o
8
×
8
.
A
d
e
s
c
r
i
p
t
i
o
n
o
f
t
h
e
C
L
AH
E
is
p
r
o
v
i
d
e
d
i
n
s
e
c
t
i
o
n
2
.
3
.
P
r
o
c
es
s
i
n
g
e
a
c
h
R
GB
c
h
a
n
n
e
l
s
e
p
ar
a
t
e
l
y
e
n
s
u
r
es
b
e
t
t
e
r
n
o
is
e
r
em
o
v
a
l
a
n
d
c
o
n
t
r
a
s
t
a
d
j
u
s
t
m
e
n
t
w
h
i
l
e
p
r
es
e
r
v
i
n
g
t
h
e
n
a
t
u
r
a
l
c
o
l
o
r
b
a
l
a
n
c
e
o
f
t
h
e
ce
r
v
i
c
a
l
c
e
ll
i
m
a
g
es
[
3
5
]
.
T
h
is
s
tu
d
y
co
n
d
u
cted
f
iv
e
s
im
u
latio
n
s
to
e
v
alu
ate
th
e
im
p
ac
t
o
f
d
if
f
er
e
n
t
im
ag
e
p
r
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
o
n
en
h
an
cin
g
p
a
p
s
m
ea
r
im
ag
es.
T
h
e
f
ir
s
t
s
im
u
latio
n
test
ed
ten
p
ap
s
m
ea
r
im
ag
es
u
s
in
g
th
r
ee
p
r
ep
r
o
ce
s
s
in
g
te
ch
n
iq
u
es:
a
PMD
f
ilter
,
C
L
AHE
,
an
d
a
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
.
T
h
e
g
o
al
was
to
ass
es
s
h
o
w
th
ese
tech
n
iq
u
es
im
p
r
o
v
e
th
e
im
ag
e
q
u
ality
.
T
h
e
s
ec
o
n
d
ex
p
er
im
en
t
u
s
ed
th
e
o
r
ig
in
a
l
p
ap
s
m
ea
r
im
ag
es
as
th
e
b
aselin
e.
T
h
e
T
h
ir
d
s
im
u
latio
n
ap
p
lied
th
e
PMD
f
ilter
to
as
s
es
s
it
s
ef
f
ec
tiv
en
ess
in
r
ed
u
cin
g
n
o
is
e
in
p
ap
s
m
ea
r
im
ag
es.
T
h
e
f
o
u
r
th
s
im
u
latio
n
u
s
ed
C
L
AHE
to
en
h
an
ce
th
e
c
o
n
tr
ast
o
f
th
e
p
ap
s
m
ea
r
im
ag
es.
T
h
e
f
if
th
s
im
u
latio
n
ap
p
lied
th
e
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
to
t
est
wh
eth
er
th
eir
co
m
b
in
ed
ar
e
m
o
r
e
ef
f
ec
tiv
e
in
im
p
r
o
v
in
g
p
ap
s
m
ea
r
im
a
g
e
q
u
ality
.
T
h
e
f
ir
s
t
s
im
u
latio
n
was
ev
alu
ated
u
s
in
g
C
E
I
Q
f
o
r
im
ag
e
q
u
ality
,
an
d
th
e
o
th
er
s
im
u
latio
n
was
ev
alu
ated
u
s
in
g
C
NN
ar
ch
itectu
r
e
s
.
T
h
ese
ev
alu
atio
n
s
h
elp
e
d
d
eter
m
in
e
th
e
m
o
s
t
ef
f
ec
tiv
e
p
r
e
p
r
o
ce
s
s
in
g
a
p
p
r
o
ac
h
f
o
r
o
p
tim
izin
g
C
N
N
-
b
ased
ce
r
v
ical
ca
n
ce
r
class
if
icatio
n
m
eth
o
d
s
.
T
o
r
ch
v
is
io
n
m
o
d
els
wer
e
u
s
ed
i
n
th
is
s
tu
d
y
,
in
clu
d
in
g
p
r
e
tr
ain
ed
R
esNet3
4
,
R
esn
et5
0
,
Sq
u
ee
ze
Net
-
1
.
0
,
Mo
b
ileNet
-
V2
,
E
f
f
icien
tNet
-
B
0
,
E
f
f
icien
tNet
-
B
1
,
Den
s
eN
et1
2
1
,
an
d
Den
s
eNe
t2
0
1
.
Ad
am
o
p
tim
izer
was
u
s
ed
in
l
ea
r
n
in
g
m
o
d
els
with
0
.
0
0
0
1
f
o
r
lear
n
in
g
r
ate.
T
h
is
s
tu
d
y
co
n
d
u
cte
d
tr
ain
i
n
g
o
v
e
r
3
0
e
p
o
ch
s
,
allo
win
g
s
u
f
f
icien
t
iter
atio
n
s
f
o
r
th
e
m
o
d
el
to
co
n
v
er
g
e
wh
ile
m
itig
atin
g
th
e
r
is
k
o
f
o
v
er
f
itti
n
g
.
C
o
m
b
in
in
g
th
e
p
r
etr
ain
e
d
T
o
r
ch
Vis
io
n
m
o
d
els,
th
e
Ad
am
o
p
tim
izer
,
th
e
lear
n
in
g
r
ate,
a
n
d
th
e
ap
p
r
o
p
r
iate
ep
o
c
h
co
u
n
t
en
ab
led
C
NN
to
tr
ain
t
h
e
m
o
d
els
ef
f
icien
tly
.
T
h
e
s
im
u
lati
o
n
s
u
s
ed
a
Py
th
o
n
J
u
p
y
ter
N
o
teb
o
o
k
with
T
o
r
ch
v
is
io
n
m
o
d
els
o
n
a
W
in
d
o
ws
1
1
o
p
er
atin
g
s
y
s
tem
.
T
o
r
c
h
Vis
io
n
m
o
d
els
o
f
f
er
a
wid
e
r
a
n
g
e
o
f
s
tate
-
of
-
th
e
-
ar
t
C
NN
ar
ch
itectu
r
es
[
3
6
]
.
T
h
e
h
ar
d
war
e
co
n
f
ig
u
r
ati
o
n
c
o
n
s
is
ted
o
f
an
AM
D
R
y
ze
n
5
5
5
0
0
p
r
o
ce
s
s
o
r
co
m
p
lem
en
ted
b
y
3
2
GB
o
f
s
y
s
tem
m
em
o
r
y
.
A
NVid
ia
GeFo
r
ce
R
T
X
3
0
6
0
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
it
(
GPU)
with
1
2
GB
o
f
d
e
d
icate
d
VR
AM
was u
s
ed
f
o
r
ac
ce
ler
ated
C
NN
ar
c
h
itectu
r
e
co
m
p
u
tatio
n
s
.
Fig
u
r
e
2
.
Pap
s
m
ea
r
im
ag
e
p
r
e
p
r
o
ce
s
s
in
g
u
s
in
g
PMD
f
ilter
,
C
L
AHE
,
o
r
h
y
b
r
id
PMD
f
ilter
–
C
L
AHE
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
39
,
No
.
1
,
J
u
ly
20
25
:
644
-
6
5
5
648
2
.
2
.
P
M
D
f
ilte
r
T
h
e
PMD
f
ilter
o
p
er
ates
o
n
th
e
p
r
in
cip
le
o
f
an
is
o
tr
o
p
ic
d
if
f
u
s
io
n
,
wh
er
e
th
e
d
if
f
u
s
io
n
r
ate
i
s
ad
ap
ted
b
ased
o
n
th
e
g
r
ad
ien
t
v
ar
iati
o
n
s
o
f
th
e
im
a
g
e.
I
t
f
ac
ilit
ates
d
if
f
u
s
io
n
p
r
im
ar
ily
in
a
r
ea
s
with
lo
w
g
r
ad
ien
ts
(
u
n
if
o
r
m
r
e
g
io
n
s
)
,
wh
ile
r
estr
ictin
g
it
to
h
ig
h
-
g
r
ad
ien
t
r
eg
i
o
n
s
(
ed
g
es).
T
h
is
c
o
n
tr
o
lled
d
if
f
u
s
io
n
p
r
o
ce
s
s
is
d
ictated
b
y
a
p
a
r
tial
d
if
f
er
e
n
tial
eq
u
atio
n
(
PDE)
,
w
h
ich
r
eg
u
lates
th
e
f
lo
w
o
f
in
f
o
r
m
atio
n
with
in
th
e
im
ag
e
[
3
7
]
.
T
h
e
f
o
llo
win
g
n
o
n
lin
ea
r
d
if
f
u
s
io
n
e
q
u
atio
n
d
escr
ib
es th
e
PMD
Fil
ter
:
(
,
,
)
=
⋅
(
(
∥
∥
)
)
,
(
1
)
(
,
,
)
is
th
e
im
ag
e
in
ten
s
ity
at
th
e
p
o
s
itio
n
(
,
)
an
d
tim
e
,
is
th
e
g
r
ad
ien
t
o
f
th
e
im
ag
e,
(
∥
∥
)
is
th
e
d
if
f
u
s
io
n
c
o
ef
f
icien
t,
⋅
d
en
o
tes
th
e
d
i
v
er
g
e
n
ce
o
p
er
ato
r
.
T
h
e
d
if
f
u
s
io
n
co
ef
f
icien
t
r
e
g
u
lates
th
e
r
ate
o
f
d
if
f
u
s
io
n
,
w
h
ich
is
in
f
lu
e
n
ce
d
b
y
th
e
m
ag
n
itu
d
e
o
f
th
e
g
r
ad
ien
t.
∣
∣
.
T
h
e
two
co
m
m
o
n
ly
u
s
ed
d
if
f
u
s
io
n
co
ef
f
icien
t
f
u
n
ctio
n
s
ar
e
i
n
v
er
s
e
q
u
ad
r
atic
an
d
ex
p
o
n
en
tial
f
u
n
ctio
n
s
.
T
h
e
ex
p
o
n
en
tial
wa
s
p
er
f
o
r
m
e
d
in
th
is
s
tu
d
y
.
(
∥
∥
)
=
−
(
|
|
)
2
(
2
)
W
h
er
e
is
a
co
n
tr
ast
th
at
d
e
ter
m
in
es
th
e
s
en
s
itiv
ity
to
e
d
g
es.
Sm
all
v
alu
es
o
f
m
ak
e
th
e
f
ilter
m
o
r
e
s
en
s
itiv
e
to
s
m
aller
ed
g
es,
wh
ile
lar
g
er
v
al
u
es
m
ak
e
it
less
s
en
s
itiv
e.
T
h
e
PMD
f
ilter
is
a
p
o
wer
f
u
l
to
o
l
f
o
r
im
ag
e
en
h
an
ce
m
en
t,
o
f
f
er
in
g
s
ig
n
if
ican
t
ad
v
a
n
tag
es
o
v
er
tr
ad
itio
n
al
lin
ea
r
s
m
o
o
th
i
n
g
tec
h
n
iq
u
es.
I
ts
ab
ilit
y
to
s
elec
tiv
ely
s
m
o
o
th
im
ag
es
wh
ile
p
r
eser
v
in
g
e
d
g
es m
ak
es
it h
ig
h
ly
v
al
u
ab
le
in
v
ar
io
u
s
a
p
p
licatio
n
s
[
3
8
]
.
2
.
3
.
CL
AH
E
C
L
AH
E
is
u
ti
l
i
ze
d
t
o
e
n
h
a
n
ce
i
m
a
g
e
c
o
n
t
r
as
t
[
3
9
]
.
I
t
o
p
er
a
t
e
s
b
y
d
i
v
i
d
i
n
g
t
h
e
i
m
a
g
e
in
t
o
s
m
a
ll
r
e
g
i
o
n
s
.
I
t
a
p
p
l
ie
s
h
is
t
o
g
r
a
m
e
q
u
a
l
i
z
a
ti
o
n
(
H
E
)
t
o
e
a
c
h
r
e
g
i
o
n
u
s
i
n
g
(
3
)
a
n
d
c
o
n
t
r
o
l
l
i
n
g
c
o
n
t
r
a
s
t
e
n
h
a
n
c
e
m
e
n
t
b
y
l
i
m
i
ti
n
g
a
m
p
l
i
f
i
c
a
ti
o
n
u
s
i
n
g
(
4
)
.
T
h
i
s
m
e
t
h
o
d
e
f
f
e
c
t
i
v
e
l
y
p
r
e
v
e
n
t
s
e
x
c
e
s
s
i
v
e
c
o
n
t
r
a
s
t
e
n
h
a
n
c
e
m
e
n
t
i
n
u
n
i
f
o
r
m
a
r
e
a
s
,
a
n
d
is
p
a
r
t
ic
u
l
a
r
l
y
a
d
v
an
t
a
g
e
o
u
s
f
o
r
i
m
a
g
e
p
r
o
c
e
s
s
i
n
g
.
C
L
A
H
E
i
m
p
r
o
v
e
s
f
i
n
e
d
eta
i
l
s
w
h
i
l
e
r
e
d
u
c
i
n
g
a
r
t
i
f
a
c
ts
.
C
L
A
H
E
i
s
e
f
f
e
c
t
i
v
e
in
i
m
p
r
o
v
i
n
g
t
h
e
m
e
d
i
c
a
l
i
m
a
g
es
q
u
a
l
i
t
y
[
2
5
]
.
=
(
.
(
2
−
1
)
.
ℎ
)
(
3
)
I
n
(
4
)
is
u
s
ed
to
o
b
tain
th
e
n
ew
g
r
ey
v
al
u
e
o
f
t
h
e
h
is
to
g
r
am
eq
u
aliza
tio
n
r
esu
lt.
.
is
th
e
cu
m
u
lativ
e
d
i
s
t
r
i
b
u
t
i
o
n
o
f
th
e
g
r
ay
s
ca
le
v
alu
e
o
f
th
e
o
r
ig
in
a
l
im
ag
e,
an
d
ℎ
ar
e
th
e
wid
th
an
d
h
eig
h
t
o
f
t
h
e
im
ag
e,
is
a
n
u
m
b
er
o
f
co
lo
r
v
ar
iatio
n
s
.
T
h
e
C
L
AHE
h
as
t
wo
v
ar
iab
les
co
n
tr
o
llin
g
co
n
tr
ast
im
ag
e
q
u
ality
:
th
e
b
lo
ck
s
ize
an
d
clip
lim
it
[
3
9
]
.
T
h
e
(
4
)
is
u
s
ed
t
o
ca
lcu
late
th
e
clip
lim
it (
).
=
(
1
+
100
(
−
1
)
)
(
4
)
Her
e,
an
d
d
en
o
te
th
e
to
tal
n
u
m
b
er
o
f
g
r
ey
-
lev
el
p
ix
els
with
in
ea
ch
b
lo
ck
wh
ile
r
ep
r
esen
ts
th
e
h
ig
h
est
p
er
m
is
s
ib
le
s
lo
p
e
in
th
e
h
is
to
g
r
am
'
s
cu
m
u
lativ
e
d
is
tr
ib
u
tio
n
f
u
n
ctio
n
(
C
DF)
.
T
h
i
s
co
n
s
tr
ain
t
h
elp
s
to
m
in
im
ize
ar
tifa
cts
b
y
c
o
n
tr
o
lli
n
g
t
h
e
n
o
is
e
am
p
lific
atio
n
.
I
n
ad
d
itio
n
,
is
th
e
clip
f
ac
to
r
,
w
h
ich
r
a
n
g
es
f
r
o
m
0
to
1
0
0
.
2
.
4
.
Co
nv
o
lutio
na
l
neura
l net
wo
rk
A
C
NN
i
s
a
d
ee
p
n
eu
r
al
n
etwo
r
k
th
at
p
r
o
ce
s
s
es
d
ata
s
u
ch
as
im
ag
es.
I
n
s
p
ir
ed
b
y
th
e
v
is
u
al
co
r
tex
o
f
an
im
als,
C
NN
h
as
b
ec
o
m
e
th
e
co
r
n
er
s
to
n
e
o
f
m
o
d
er
n
co
m
p
u
ter
v
is
io
n
a
p
p
licatio
n
s
[
4
0
]
.
T
h
e
ar
ch
itectu
r
e
o
f
a
ty
p
ical
C
NN
co
n
s
is
t
s
o
f
s
ev
er
al
lay
er
s
,
ea
ch
s
er
v
in
g
a
d
is
tin
ct
p
u
r
p
o
s
e
in
th
e
d
ata
p
r
o
ce
s
s
in
g
p
ip
elin
e.
T
h
e
p
r
im
ar
y
la
y
er
s
in
clu
d
e
d
th
e
f
o
llo
win
g
:
−
T
h
e
in
p
u
t la
y
er
ac
ce
p
ts
r
aw
d
a
ta,
wh
ich
ar
e
r
e
p
r
esen
ted
as m
u
ltid
im
en
s
io
n
al
ar
r
ay
s
o
f
p
ix
el
v
alu
es.
−
I
n
th
e
co
n
v
o
l
u
tio
n
al
la
y
er
,
a
s
er
ies
o
f
lea
r
n
ab
le
k
er
n
el
f
ilter
s
p
er
f
o
r
m
co
n
v
o
lu
tio
n
o
p
er
atio
n
s
o
n
th
e
in
p
u
t
d
ata.
T
h
ese
f
ilter
s
m
o
v
e
ac
r
o
s
s
th
e
in
p
u
t
a
n
d
p
er
f
o
r
m
ele
m
en
t
-
wis
e
m
u
ltip
licatio
n
a
n
d
s
u
m
m
atio
n
to
g
en
er
ate
th
e
f
ea
tu
r
e
m
a
p
s
.
T
h
e
p
r
o
ce
s
s
was c
alcu
lated
u
s
in
g
(
5
)
.
(
∗
)
(
,
)
=
∑
∑
(
−
,
−
)
.
(
,
)
(
5
)
W
h
er
e
is
th
e
in
p
u
t im
ag
e
,
is
t
h
e
k
er
n
el
a
n
d
(
,
)
ar
e
th
e
c
o
o
r
d
i
n
ates o
f
th
e
o
u
t
p
u
t f
ea
tu
r
e
m
ap
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
d
va
n
ce
d
ce
r
vica
l c
a
n
ce
r
cla
s
s
ifica
tio
n
:
en
h
a
n
cin
g
p
a
p
s
me
a
r
ima
g
es w
ith
h
yb
r
id
P
MD
…
(
A
ch
K
h
o
z
a
imi
)
649
−
Activ
atio
n
lay
er
:
f
o
llo
win
g
e
ac
h
co
n
v
o
l
u
tio
n
al
lay
er
,
an
ac
tiv
atio
n
f
u
n
ctio
n
is
ap
p
lie
d
to
in
tr
o
d
u
ce
n
o
n
lin
ea
r
ity
in
to
t
h
e
m
o
d
el.
A
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
o
r
L
ea
k
y
R
eL
U
was
u
s
ed
i
n
th
e
ac
tiv
atio
n
lay
er
.
I
n
(
6
)
is
u
s
ed
f
o
r
t
h
e
R
eL
U
ac
tiv
atio
n
lay
er
.
(
)
=
(
0
,
)
(
6
)
−
T
h
e
s
p
atial
d
im
en
s
io
n
s
o
f
th
e
f
ea
tu
r
e
m
ap
s
ar
e
r
ed
u
ce
d
b
y
th
e
p
o
o
lin
g
lay
e
r
,
wh
ich
h
elp
s
lo
wer
th
e
co
m
p
u
tatio
n
al
r
eq
u
ir
em
en
ts
a
n
d
co
m
b
at
o
v
er
f
itti
n
g
.
Ma
x
p
o
o
lin
g
,
illu
s
tr
ated
in
(
7
)
a
n
d
a
v
er
ag
e
p
o
o
lin
g
,
as sh
o
wn
in
(
8
)
,
ar
e
th
e
two
m
o
s
t f
r
eq
u
en
tly
em
p
lo
y
ed
p
o
o
li
n
g
tech
n
i
q
u
es.
(
)
,
,
=
,
.
+
.
.
+
,
(
7
)
(
)
,
,
=
1
.
∑
.
+
.
.
+
,
,
(
8
)
−
Fu
lly
co
n
n
ec
ted
la
y
er
:
f
o
llo
win
g
m
u
ltip
le
co
n
v
o
lu
tio
n
al
an
d
p
o
o
lin
g
lay
er
s
,
t
h
e
n
etwo
r
k
p
er
f
o
r
m
s
ad
v
an
ce
d
r
ea
s
o
n
in
g
th
r
o
u
g
h
f
u
lly
co
n
n
ec
ted
lay
er
s
.
I
n
th
ese
lay
er
s
,
ea
ch
n
eu
r
o
n
is
li
n
k
ed
t
o
ev
e
r
y
n
eu
r
o
n
in
th
e
p
r
ec
ed
i
n
g
lay
er
,
en
ab
lin
g
s
o
p
h
is
ticated
in
p
u
t d
ata
r
ep
r
esen
tatio
n
.
−
T
h
e
o
u
tp
u
t
lay
er
is
th
e
f
in
al
lay
er
o
f
th
e
C
NN
th
at
o
u
tp
u
ts
th
e
p
r
ed
ictio
n
s
.
T
h
e
So
f
t
Ma
x
ac
tiv
atio
n
f
u
n
ctio
n
p
r
o
d
u
ce
s
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
s
f
o
r
t
h
e
tar
g
et
clas
s
es.
(
)
=
∑
=
1
(
9
)
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
i
s
s
tu
d
y
ex
p
lo
r
ed
t
h
e
im
p
a
ct
o
f
v
ar
io
u
s
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es,
in
clu
d
i
n
g
a
PMD
f
ilter
,
C
L
AHE
,
an
d
h
y
b
r
id
PMD
f
il
ter
-
C
L
AHE
.
T
h
e
ev
alu
atio
n
u
s
ed
m
u
ltip
le
p
er
f
o
r
m
an
c
e
m
et
r
ics,
in
clu
d
in
g
t
h
e
C
E
I
Q
s
co
r
e
f
o
r
im
ag
e
q
u
ality
ass
es
s
m
en
t,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
tr
ain
in
g
tim
e,
to
ev
alu
ate
th
e
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es u
s
in
g
C
NN
m
o
d
els.
3
.
1
.
I
ma
g
es c
o
ntr
a
s
t
a
s
s
ess
m
ent
T
h
e
av
er
a
g
e
C
E
I
Q
m
etr
ic
was
ev
alu
ated
o
n
1
0
p
a
p
s
m
e
ar
im
ag
es
to
ass
ess
th
e
p
er
f
o
r
m
an
ce
o
f
d
if
f
er
en
t
p
r
ep
r
o
ce
s
s
in
g
tech
n
i
q
u
es
in
en
h
an
cin
g
p
ap
s
m
ea
r
im
ag
e
q
u
ality
.
Fig
u
r
e
3
s
h
o
ws
th
e
co
m
p
ar
ativ
e
r
esu
lts
o
f
th
ese
tech
n
iq
u
es
(
th
e
f
ir
s
t
s
im
u
latio
n
)
.
T
h
e
o
r
ig
in
al
im
ag
es
ac
h
ie
v
ed
an
av
er
ag
e
C
E
I
Q
o
f
3
.
5
2
2
8
.
Ho
wev
er
,
ap
p
ly
i
n
g
th
e
PMD
f
ilter
s
lig
h
tly
r
ed
u
ce
d
th
e
s
co
r
e
to
3
.
4
9
5
6
,
s
u
g
g
esti
n
g
th
at
th
e
PMD
f
ilter
alo
n
e
m
ay
n
o
t
s
ig
n
if
ica
n
tly
en
h
an
c
e
im
ag
e
co
n
tr
ast.
H
o
wev
er
,
C
L
AHE
d
em
o
n
s
tr
ated
a
n
o
t
ab
le
im
p
r
o
v
em
e
n
t,
ac
h
ie
v
in
g
a
C
E
I
Q
s
co
r
e
o
f
3
.
7
3
0
1
,
h
ig
h
lig
h
tin
g
its
ef
f
ec
tiv
en
ess
in
im
p
r
o
v
in
g
th
e
c
o
n
tr
ast
o
f
ce
r
v
ical
ce
ll
im
ag
es.
T
h
e
h
y
b
r
id
ap
p
r
o
ac
h
,
wh
ich
co
m
b
i
n
ed
th
e
PMD
f
ilter
with
C
L
AHE
,
r
esu
lted
i
n
a
C
E
I
Q
s
co
r
e
o
f
3
.
7
1
8
8
,
w
h
ich
was
s
lig
h
tly
l
o
wer
th
an
th
at
o
f
C
L
AHE
al
o
n
e
b
u
t
s
till
s
u
b
s
tan
tially
h
ig
h
er
th
an
th
at
o
f
th
e
o
r
ig
in
al
an
d
PMD
f
ilter
im
a
g
es.
T
h
ese
f
in
d
in
g
s
s
u
g
g
est
th
at
C
L
AHE
p
lay
s
a
cr
u
cial
r
o
le
in
co
n
tr
ast
en
h
an
ce
m
e
n
t,
wh
er
ea
s
th
e
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
m
eth
o
d
ef
f
ec
tiv
ely
r
ed
u
ce
s
n
o
i
s
e
wh
ile
im
p
r
o
v
i
n
g
co
n
tr
as
t.
Ho
wev
er
,
ad
d
itio
n
al
p
r
ep
r
o
ce
s
s
in
g
with
th
e
PMD
f
ilter
d
id
n
o
t
s
ig
n
if
ican
tly
en
h
an
ce
th
e
co
n
tr
ast
b
ey
o
n
d
th
at
ac
h
ie
v
ed
u
s
in
g
C
L
AHE
alo
n
e.
Ov
er
all,
C
L
AHE
was
th
e
m
o
s
t
ef
f
ec
tiv
e
tec
h
n
iq
u
e
f
o
r
c
o
n
tr
ast
en
h
an
ce
m
e
n
t
in
th
e
p
ap
s
m
ea
r
im
ag
e
an
aly
s
is
.
I
n
co
n
tr
as
t,
th
e
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
ap
p
r
o
ac
h
o
f
f
er
s
ad
d
itio
n
al
n
o
is
e
r
ed
u
ctio
n
an
d
co
n
tr
ast im
p
r
o
v
em
en
t a
d
v
an
t
ag
es.
3
.
2
.
CNN
a
rc
hite
ct
ures a
s
s
ess
m
ent
3
.
2
.
1
.
P
a
p sm
ea
r
im
a
g
es wit
ho
ut
prepro
ce
s
s
i
ng
T
h
e
s
ec
o
n
d
s
im
u
latio
n
,
s
h
o
wn
in
T
ab
le
1
,
p
r
esen
ts
th
e
ev
alu
ati
o
n
r
esu
lts
o
f
v
ar
i
o
u
s
C
NN
ar
ch
itectu
r
es
o
n
p
ap
s
m
ea
r
im
ag
es
with
o
u
t
p
r
e
p
r
o
ce
s
s
in
g
.
Am
o
n
g
th
e
test
ed
m
o
d
el
s
,
E
f
f
icien
tNet
-
B
0
ac
h
iev
ed
th
e
h
i
g
h
est
ac
cu
r
ac
y
(
9
1
.
5
8
%)
an
d
b
est
F1
-
s
co
r
e
(
9
1
.
1
6
%),
d
em
o
n
s
tr
atin
g
it
s
s
u
p
er
io
r
ab
ilit
y
to
ex
tr
ac
t
r
elev
an
t
f
ea
tu
r
es
f
r
o
m
ce
r
v
ical
ce
ll
im
ag
es.
Ho
wev
er
,
its
tr
ain
in
g
tim
e
(
8
8
0
s
)
is
r
elativ
ely
h
ig
h
co
m
p
ar
ed
to
m
o
d
els
s
u
ch
as
Mo
b
ileNet
-
V2
an
d
Sq
u
ee
ze
N
et
-
1
.
0
.
E
f
f
icien
tNet
-
B
1
clo
s
ely
f
o
llo
wed
,
with
an
ac
cu
r
ac
y
o
f
8
9
.
6
0
%
an
d
an
F1
-
s
co
r
e
o
f
8
9
.
8
4
%;
h
o
we
v
er
,
it
h
ad
th
e
lo
n
g
est
tr
ain
in
g
tim
e
(
1
,
0
4
9
s
)
am
o
n
g
th
e
E
f
f
icien
tNet
m
o
d
els.
Mo
b
ileNet
-
V2
also
d
eliv
er
ed
s
tr
o
n
g
p
e
r
f
o
r
m
an
ce
,
with
8
3
.
9
1
%
ac
cu
r
ac
y
an
d
a
r
elativ
ely
lo
w
tr
ain
i
n
g
tim
e
(
6
8
2
s
)
,
m
ak
in
g
it
a
g
o
o
d
c
h
o
ice
f
o
r
b
alan
cin
g
ac
c
u
r
ac
y
an
d
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
R
esNet3
4
an
d
D
en
s
eNe
t1
2
1
ex
h
ib
ited
m
o
d
er
ate
p
er
f
o
r
m
a
n
ce
,
with
an
ac
cu
r
ac
y
o
f
7
6
.
7
3
%,
wh
er
ea
s
R
esNe
t5
0
p
er
f
o
r
m
e
d
s
lig
h
tly
wo
r
s
e
at
7
1
.
5
3
%
ac
cu
r
ac
y
.
Alth
o
u
g
h
Den
s
eNe
t2
0
1
ac
h
iev
ed
a
s
lig
h
tly
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
39
,
No
.
1
,
J
u
ly
20
25
:
644
-
6
5
5
650
b
etter
ac
cu
r
ac
y
(
7
6
.
9
8
%),
it
h
ad
a
s
ig
n
if
ican
tly
lo
n
g
er
tr
ain
i
n
g
tim
e
(
9
,
8
1
9
s)
.
Den
s
eNe
t2
0
1
is
in
ef
f
icien
t
f
o
r
lar
g
e
-
s
ca
le
im
p
lem
en
tatio
n
s
.
Sq
u
ee
ze
Net
-
1
.
0
,
d
esig
n
e
d
f
o
r
lig
h
tweig
h
t
ap
p
licatio
n
s
,
ac
h
iev
es
7
3
.
7
6
%
ac
cu
r
ac
y
with
th
e
f
astes
t
tr
ain
in
g
tim
e
(
6
3
2
s
)
,
b
u
t
its
lo
wer
F1
-
s
co
r
e
(
7
2
.
5
6
%)
s
u
g
g
ests
a
tr
ad
e
-
o
f
f
b
etwe
en
s
p
ee
d
an
d
p
er
f
o
r
m
an
ce
.
Fig
u
r
e
3
.
Av
e
r
ag
e
C
E
I
Q
s
co
r
e
f
o
r
im
a
g
e
en
h
a
n
ce
m
en
t
T
ab
le
1
.
C
NN
Ar
ch
itectu
r
es’
ass
es
s
m
en
t sco
r
e
f
o
r
p
a
p
s
m
ea
r
im
ag
es with
o
u
t p
r
ep
r
o
ce
s
s
in
g
A
r
c
h
i
t
e
c
t
u
r
e
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
Tr
a
i
n
i
n
g
t
i
me
(
s)
R
e
sN
e
t
3
4
76
.
7
3
%
79
.
1
2
%
76
.
9
0
%
76
.
1
9
%
7
5
5
R
e
sN
e
t
5
0
71
.
5
3
%
75
.
6
6
%
71
.
6
4
%
70
.
4
8
%
8
9
2
S
q
u
e
e
z
e
N
e
t
-
1
.
0
73
.
7
6
%
73
.
9
9
%
74
.
0
7
%
72
.
5
6
%
6
3
2
M
o
b
i
l
e
N
e
t
-
V2
83
.
9
1
%
85
.
7
1
%
84
.
1
0
%
83
.
6
3
%
6
8
2
Ef
f
i
c
i
e
n
t
N
e
t
-
B0
91
.
5
8
%
92
.
3
0
%
91
.
6
8
%
91
.
1
6
%
8
8
0
Ef
f
i
c
i
e
n
t
N
e
t
-
B1
89
.
6
0
%
91
.
5
5
%
89
.
6
4
%
89
.
8
4
%
1
,
0
4
9
D
e
n
seN
e
t
1
2
1
76
.
7
3
%
79
.
8
1
%
76
.
8
5
%
76
.
0
4
%
9
6
4
D
e
n
seN
e
t
2
0
1
76
.
9
8
%
80
.
0
2
%
77
.
1
6
%
75
.
9
3
%
9
,
8
1
9
Ov
er
all,
E
f
f
icien
tNet
-
B
0
em
er
g
ed
as
th
e
b
est
-
p
er
f
o
r
m
in
g
m
o
d
el
r
eg
ar
d
in
g
ac
cu
r
a
cy
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
th
e
F1
-
s
co
r
e.
E
f
f
i
cien
tNet
-
B
0
was
th
e
m
o
s
t
s
u
i
tab
le
ch
o
ice
f
o
r
ce
r
v
ical
ca
n
c
er
class
if
icatio
n
in
th
is
s
im
u
latio
n
.
Ho
wev
er
,
if
co
m
p
u
tatio
n
al
ef
f
icien
cy
is
a
p
r
i
o
r
ity
,
M
o
b
ileNet
-
V2
p
r
o
v
id
es
a
s
tr
o
n
g
alter
n
ativ
e
with
r
elativ
ely
h
ig
h
ac
cu
r
ac
y
an
d
s
ig
n
if
ican
tly
l
o
wer
tr
ain
in
g
tim
e.
T
h
e
r
esu
lt
s
s
u
g
g
est
th
at
d
ee
p
er
ar
ch
itectu
r
es,
s
u
ch
as
E
f
f
icien
tNet
an
d
Mo
b
ileNet
-
V2
,
ar
e
m
o
r
e
ef
f
ec
tiv
e
f
o
r
f
ea
tu
r
e
ex
t
r
ac
tio
n
.
I
n
co
n
t
r
ast,
lig
h
ter
m
o
d
els,
s
u
ch
as
Sq
u
ee
ze
Net
an
d
R
esNet,
r
eq
u
i
r
e
f
u
r
th
er
o
p
tim
izatio
n
to
ac
h
iev
e
co
m
p
etitiv
e
p
er
f
o
r
m
an
ce
.
3
.
2
.
2
.
P
a
p sm
ea
r
im
a
g
es wit
h P
M
D
f
ilte
r
prepro
ce
s
s
ing
T
ab
le
2
p
r
esen
ts
th
e
p
er
f
o
r
m
an
ce
ev
al
u
atio
n
r
esu
lts
o
f
th
e
C
NN
ar
ch
itectu
r
e
o
n
th
e
p
ap
s
m
ea
r
im
ag
es
p
r
ep
r
o
ce
s
s
ed
with
th
e
PMD
f
ilter
tech
n
iq
u
e
in
th
e
th
ir
d
s
im
u
latio
n
.
E
f
f
icien
tNet
-
B
0
r
em
ain
ed
th
e
b
est
-
p
er
f
o
r
m
in
g
m
o
d
el,
ac
h
ie
v
in
g
an
ac
cu
r
ac
y
o
f
8
8
.
8
6
%,
s
lig
h
tly
l
o
wer
th
an
its
p
er
f
o
r
m
an
ce
o
n
th
e
o
r
ig
in
al
im
ag
es
(
9
1
.
5
8
%).
Similar
ly
,
E
f
f
icien
tNet
-
B
1
ex
p
er
ien
ce
d
a
d
r
o
p
in
ac
cu
r
ac
y
f
r
o
m
8
9
.
6
0
%
to
8
5
.
8
9
%,
s
u
g
g
esti
n
g
th
at,
alth
o
u
g
h
th
e
PMD
f
ilter
m
ay
en
h
an
ce
n
o
is
e
r
ed
u
ctio
n
,
it
d
o
es
n
o
t
n
ec
es
s
ar
ily
im
p
r
o
v
e
t
h
e
class
if
ica
tio
n
p
er
f
o
r
m
a
n
ce
o
f
th
ese
ar
ch
itectu
r
es.
I
n
ter
esti
n
g
ly
,
Sq
u
ee
ze
Net
-
1
.
0
an
d
R
esNet3
4
ex
h
ib
ited
no
tab
le
im
p
r
o
v
em
e
n
ts
with
th
e
PMD
f
ilter
.
Sq
u
ee
ze
Net
-
1
.
0
im
p
r
o
v
es
its
ac
cu
r
ac
y
f
r
o
m
7
3
.
7
6
%
to
8
0
.
2
0
%,
d
em
o
n
s
tr
atin
g
th
at
th
e
PMD
f
ilter
en
h
an
ce
s
f
ea
tu
r
e
ex
tr
ac
ti
o
n
f
o
r
lig
h
tweig
h
t
ar
ch
itectu
r
es.
T
h
e
ac
cu
r
ac
y
o
f
R
esNet3
4
also
in
cr
ea
s
ed
f
r
o
m
7
6
.
7
3
%
to
7
9
.
2
1
%,
in
d
ic
atin
g
th
at
n
o
is
e
r
ed
u
ctio
n
c
o
n
tr
ib
u
ted
to
b
etter
class
if
icatio
n
r
esu
lts
in
th
is
ar
ch
itectu
r
e.
T
h
e
ac
c
u
r
ac
y
o
f
R
esNet5
0
im
p
r
o
v
ed
f
r
o
m
7
1
.
5
3
%
to
7
7
.
2
3
%,
alth
o
u
g
h
it r
em
ain
ed
less
co
m
p
etitiv
e
th
an
th
o
s
e
o
f
th
e
o
th
er
m
o
d
els.
On
th
e
o
th
er
h
an
d
,
Mo
b
ileNet
-
V2
,
wh
ich
p
er
f
o
r
m
ed
well
o
n
th
e
o
r
ig
in
al
im
ag
es
(
8
3
.
9
1
%
ac
cu
r
ac
y
)
,
d
ec
r
ea
s
ed
s
lig
h
tly
t
o
8
1
.
9
3
%
with
th
e
PMD
f
ilter
.
A
s
im
ilar
d
ec
lin
e
was
o
b
s
er
v
e
d
in
De
n
s
eNe
t1
2
1
(
7
6
.
7
3
%
to
7
8
.
2
2
%)
an
d
Den
s
eNe
t2
0
1
(
7
6
.
9
8
%
to
7
5
.
5
0
%),
with
Den
s
eNe
t2
0
1
ex
p
er
ien
cin
g
t
h
e
m
o
s
t
s
ig
n
if
ican
t
d
r
o
p
.
T
h
is
s
u
g
g
ests
th
at
d
ee
p
er
ar
ch
itectu
r
es
th
at
r
ely
o
n
h
ig
h
-
d
etail
f
ea
tu
r
e
ex
tr
ac
tio
n
m
ay
n
o
t
b
en
ef
it
as
m
u
ch
f
r
o
m
th
e
PMD
f
ilter
,
p
o
ten
tially
b
ec
au
s
e
th
e
s
m
o
o
th
in
g
ef
f
e
ct
r
ed
u
ce
s
f
i
n
er
d
et
ails
in
ce
r
v
ical
ce
ll
s
tr
u
ctu
r
es.
Fro
m
th
is
co
m
p
a
r
is
o
n
,
lig
h
tweig
h
t a
r
ch
itectu
r
es su
ch
as
Sq
u
ee
ze
Net
-
1
.
0
,
R
esNet3
4
,
an
d
R
esNet5
0
b
en
ef
it
th
e
m
o
s
t
f
r
o
m
th
e
PMD
f
ilter
,
wh
er
ea
s
d
ee
p
er
n
e
two
r
k
s
s
u
ch
as
E
f
f
icien
tNet
-
B
0
,
E
f
f
icien
tNet
-
B
1
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
d
va
n
ce
d
ce
r
vica
l c
a
n
ce
r
cla
s
s
ifica
tio
n
:
en
h
a
n
cin
g
p
a
p
s
me
a
r
ima
g
es w
ith
h
yb
r
id
P
MD
…
(
A
ch
K
h
o
z
a
imi
)
651
an
d
Den
s
eNe
t2
0
1
s
h
o
w
a
d
e
clin
e
in
th
eir
p
er
f
o
r
m
an
ce
.
T
h
is
s
u
g
g
ests
th
at
th
e
PMD
f
i
lter
is
p
ar
ticu
lar
ly
ef
f
ec
tiv
e
f
o
r
m
o
d
els
th
at
s
tr
u
g
g
le
with
n
o
is
e;
h
o
wev
er
,
it
m
ay
r
e
m
o
v
e
f
in
e
-
g
r
ain
e
d
d
etails
o
n
wh
ic
h
d
ee
p
e
r
ar
ch
itectu
r
es r
ely
o
n
cl
ass
if
icatio
n
.
T
ab
le
2
.
C
NN
Ar
ch
itectu
r
es’
ass
es
s
m
en
t sco
r
e
f
o
r
p
a
p
s
m
ea
r
im
ag
es u
s
in
g
PMD
f
ilter
p
r
e
p
r
o
ce
s
s
in
g
A
r
c
h
i
t
e
c
t
u
r
e
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
Tr
a
i
n
i
n
g
t
i
me
(
s)
R
e
sN
e
t
3
4
79
.
2
1
%
80
.
4
9
%
79
.
3
9
%
78
.
7
6
%
7
8
4
R
e
sN
e
t
5
0
77
.
2
3
%
77
.
7
6
%
77
.
5
1
%
76
.
6
0
%
8
9
7
S
q
u
e
e
z
e
N
e
t
-
1
.
0
80
.
2
0
%
79
.
3
3
%
80
.
3
8
%
79
.
2
9
%
6
3
3
M
o
b
i
l
e
N
e
t
-
V2
81
.
9
3
%
84
.
6
3
%
82
.
0
7
%
81
.
4
2
%
6
8
5
Ef
f
i
c
i
e
n
t
N
e
t
-
B0
88
.
8
6
%
90
.
3
2
%
88
.
9
9
%
88
.
1
9
%
8
5
8
Ef
f
i
c
i
e
n
t
N
e
t
-
B1
85
.
8
9
%
88
.
6
6
%
85
.
8
9
%
85
.
5
6
%
1
,
0
4
8
D
e
n
seN
e
t
1
2
1
78
.
2
2
%
82
.
7
3
%
78
.
4
7
%
77
.
4
2
%
1
,
0
4
3
D
e
n
seN
e
t
2
0
1
75
.
5
0
%
77
.
9
6
%
75
.
6
8
%
74
.
1
9
%
11
,
196
3
.
2
.
3
.
P
a
p sm
ea
r
im
a
g
es wit
h CLAH
E
prepro
ce
s
s
ing
T
ab
le
3
p
r
esen
ts
t
h
e
ev
al
u
a
tio
n
r
esu
lts
o
f
th
e
C
NN
ar
ch
itectu
r
es
o
n
th
e
p
ap
s
m
e
ar
im
ag
es
p
r
ep
r
o
ce
s
s
ed
u
s
in
g
th
e
C
L
AH
E
tech
n
iq
u
e
in
th
e
f
o
u
r
th
s
im
u
latio
n
.
T
h
e
ev
alu
atio
n
o
f
C
N
N
ar
ch
itectu
r
es
o
n
C
L
AHE
p
r
ep
r
o
ce
s
s
ed
p
ap
s
m
ea
r
im
ag
es
d
em
o
n
s
tr
at
ed
s
ig
n
if
ican
t
im
p
r
o
v
em
en
t
s
in
class
if
icat
io
n
p
er
f
o
r
m
an
ce
co
m
p
ar
ed
to
t
h
e
o
r
ig
in
al
an
d
PMD
-
f
ilter
ed
im
ag
es.
C
L
AHE
en
h
an
ce
s
co
n
tr
ast,
cr
u
cial
f
o
r
im
p
r
o
v
in
g
f
ea
tu
r
e
ex
tr
ac
tio
n
in
ce
r
v
ical
ce
ll
class
if
icatio
n
.
Am
o
n
g
all
th
e
ar
ch
itectu
r
es,
E
f
f
icien
tNet
-
B
1
ac
h
iev
ed
t
h
e
h
ig
h
est
ac
cu
r
a
cy
(
8
9
.
3
6
%),
f
o
llo
wed
b
y
E
f
f
icien
tNet
-
B
0
(
8
8
.
8
6
%)
a
n
d
Mo
b
ileNet
-
V2
(
8
4
.
6
5
%).
I
n
ter
esti
n
g
ly
,
R
esNet3
4
s
h
o
wed
a
s
u
b
s
tan
tial
im
p
r
o
v
em
e
n
t,
r
ea
ch
in
g
an
a
cc
u
r
ac
y
o
f
8
4
.
1
6
%
co
m
p
ar
ed
with
7
9
.
2
1
%
with
th
e
PMD
f
ilter
an
d
7
6
.
7
3
%
with
th
e
o
r
ig
in
al
im
ag
es.
Similar
ly
,
R
es
Net5
0
im
p
r
o
v
e
d
to
8
2
.
9
2
%,
in
d
ic
atin
g
a
b
etter
a
d
ap
tatio
n
to
co
n
tr
ast
-
en
h
a
n
ce
d
im
a
g
es.
Den
s
eNe
t1
2
1
an
d
Den
s
eNe
t2
0
1
also
ex
h
ib
ited
c
o
n
s
id
er
ab
le
im
p
r
o
v
em
en
ts
,
wi
th
Den
s
eNe
t1
2
1
i
n
cr
ea
s
in
g
f
r
o
m
7
8
.
2
2
%
(
PMD)
to
8
3
.
1
7
%
(
C
L
AHE
)
an
d
Den
s
eNe
t2
0
1
in
cr
ea
s
in
g
f
r
o
m
7
5
.
5
0
%
(
PMD)
to
8
0
.
6
9
%
(
C
L
AHE
)
.
T
h
is
s
u
g
g
est
s
th
at
C
L
AHE
en
h
an
ce
s
th
e
p
er
f
o
r
m
a
n
ce
o
f
d
ee
p
er
a
r
ch
ite
ctu
r
es
b
y
im
p
r
o
v
in
g
co
n
tr
ast
,
allo
win
g
t
h
em
to
ex
tr
ac
t m
o
r
e
m
ea
n
in
g
f
u
l f
ea
tu
r
es.
Ho
wev
er
,
Sq
u
ee
ze
Net
-
1
.
0
,
wh
ich
s
h
o
wed
a
n
o
tab
le
im
p
r
o
v
e
m
en
t
with
th
e
PMD
f
i
lter
(
8
0
.
2
0
%),
d
id
n
o
t
b
en
ef
it
as
m
u
ch
f
r
o
m
C
L
AHE
an
d
ac
h
iev
ed
an
ac
c
u
r
ac
y
o
f
7
4
.
7
5
%.
T
h
is
s
u
g
g
ests
th
at
lig
h
tweig
h
t
m
o
d
els
m
ay
b
e
m
o
r
e
s
en
s
itiv
e
to
n
o
is
e
r
ed
u
ctio
n
(
PMD)
th
an
co
n
tr
ast
en
h
an
ce
m
e
n
t
(
C
L
AHE
)
.
C
L
AHE
en
h
an
ce
d
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
o
s
t
a
r
ch
itectu
r
es,
p
ar
ticu
la
r
ly
d
ee
p
er
C
NNs,
s
u
ch
as
R
esNet3
4
,
R
esNe
t5
0
,
Den
s
eNe
t1
2
1
,
an
d
Den
s
eNe
t2
0
1
.
E
f
f
icien
tNet
-
B
0
an
d
E
f
f
icien
tNet
-
B
1
r
em
ain
ed
s
tr
o
n
g
p
er
f
o
r
m
er
s
;
h
o
wev
e
r
,
t
h
eir
b
est
ac
cu
r
ac
y
was
s
till
o
b
s
er
v
ed
o
n
th
e
o
r
ig
in
al
im
ag
es
r
ath
er
th
an
with
PM
D
o
r
C
L
AHE
p
r
ep
r
o
ce
s
s
in
g
.
Sq
u
ee
ze
Net
-
1
.
0
b
en
ef
its
m
o
r
e
f
r
o
m
t
h
e
PMD
f
ilter
th
a
n
C
L
AHE
,
s
u
g
g
esti
n
g
th
at
n
o
is
e
r
e
d
u
ctio
n
is
m
o
r
e
i
m
p
o
r
tan
t
f
o
r
th
is
lig
h
tweig
h
t
m
o
d
el
th
a
n
c
o
n
tr
ast
en
h
an
ce
m
en
t.
Mo
b
ileNet
-
V2
ac
h
iev
ed
its
h
ig
h
est
ac
cu
r
ac
y
(
8
4
.
6
5
%)
with
C
L
AHE
,
s
lig
h
tly
o
u
tp
er
f
o
r
m
i
n
g
th
e
o
r
ig
in
al
an
d
th
e
PMD
f
ilter
v
er
s
io
n
s
.
T
ab
le
3
.
C
C
N
Ar
ch
itectu
r
es’
ass
es
s
m
en
t sco
r
e
f
o
r
p
a
p
s
m
ea
r
im
ag
es u
s
in
g
C
L
AHE
p
r
ep
r
o
ce
s
s
in
g
A
r
c
h
i
t
e
c
t
u
r
e
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
Tr
a
i
n
i
n
g
t
i
me
(
s)
R
e
sN
e
t
3
4
84
.
1
6
%
86
.
7
5
%
84
.
4
0
%
84
.
0
2
%
7
9
0
R
e
sN
e
t
5
0
82
.
9
2
%
83
.
4
8
%
83
.
0
6
%
82
.
5
7
%
9
0
5
S
q
u
e
e
z
e
N
e
t
-
1
.
0
74
.
7
5
%
76
.
3
3
%
75
.
0
1
%
74
.
4
1
%
6
3
1
M
o
b
i
l
e
N
e
t
-
V2
84
.
6
5
%
85
.
1
8
%
84
.
8
5
%
84
.
1
7
%
6
8
7
Ef
f
i
c
i
e
n
t
N
e
t
-
B0
88
.
8
6
%
89
.
1
8
%
88
.
9
6
%
88
.
1
9
%
8
4
2
Ef
f
i
c
i
e
n
t
N
e
t
-
B1
89
.
3
6
%
90
.
3
5
%
89
.
4
7
%
89
.
1
1
%
1
,
0
5
3
D
e
n
seN
e
t
1
2
1
83
.
1
7
%
84
.
8
5
%
83
.
4
0
%
82
.
5
6
%
9
7
2
D
e
n
seN
e
t
2
0
1
80
.
6
9
%
83
.
6
4
%
80
.
9
2
%
79
.
6
7
%
9
,
8
8
0
3
.
2
.
4
.
P
a
p sm
ea
r
im
a
g
es wit
h
hy
brid
P
M
D
f
ilte
r
-
CL
AH
E
prepro
ce
s
s
ing
T
ab
le
4
p
r
esen
ts
t
h
e
ev
al
u
a
tio
n
r
esu
lts
o
f
th
e
C
NN
ar
ch
itectu
r
es
o
n
th
e
p
ap
s
m
e
ar
im
ag
es
p
r
ep
r
o
ce
s
s
ed
u
s
in
g
th
e
h
y
b
r
i
d
PMD
f
ilter
-
C
L
AHE
tech
n
i
q
u
e
in
th
e
f
in
al
s
im
u
latio
n
.
T
h
e
h
y
b
r
id
ap
p
r
o
ac
h
co
m
b
in
es
n
o
is
e
r
ed
u
ctio
n
f
r
o
m
th
e
PMD
f
ilter
with
co
n
tr
ast
en
h
an
ce
m
en
t
f
r
o
m
C
L
AHE
,
wh
ich
en
h
an
ce
s
th
e
ab
ilit
y
o
f
th
e
C
NN
m
o
d
el
s
to
ex
tr
ac
t
m
ea
n
in
g
f
u
l
f
e
atu
r
es
f
r
o
m
im
ag
es.
Am
o
n
g
all
ar
ch
itectu
r
es,
E
f
f
icien
tNet
-
B
0
ac
h
iev
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
(
9
1
.
7
4
%),
s
lig
h
tly
s
u
r
p
ass
in
g
its
p
r
ev
io
u
s
p
er
f
o
r
m
a
n
ce
with
th
e
o
r
ig
in
al
im
ag
es
(
9
1
.
5
8
%)
an
d
C
L
AHE
(
8
8
.
8
6
%).
T
h
is
s
u
g
g
ests
th
at
co
m
b
in
in
g
n
o
is
e
r
ed
u
ctio
n
an
d
co
n
tr
ast
en
h
a
n
ce
m
en
t
b
en
e
f
its
th
is
ar
ch
itectu
r
e
m
o
r
e
t
h
an
C
L
AHE
o
r
PMD
alo
n
e.
Similar
ly
,
E
f
f
icie
n
tNet
-
B
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
39
,
No
.
1
,
J
u
ly
20
25
:
644
-
6
5
5
652
m
ain
tain
ed
a
h
ig
h
ac
c
u
r
ac
y
(
8
9
.
7
7
%),
c
o
m
p
ar
a
b
le
to
its
o
r
ig
in
al
im
ag
e
p
er
f
o
r
m
a
n
ce
(
8
9
.
6
0
%)
an
d
s
lig
h
tly
b
etter
th
an
C
L
AHE
(
8
9
.
3
6
%).
Den
s
eNe
t1
2
1
an
d
Den
s
eNe
t2
0
1
s
h
o
wed
n
o
ta
b
le
i
m
p
r
o
v
e
m
en
ts
,
with
Den
s
eNe
t1
2
1
in
cr
ea
s
in
g
to
8
6
.
3
9
%
(
f
r
o
m
8
3
.
1
7
%
with
C
L
AHE
to
7
8
.
2
2
%
wit
h
PMD)
,
wh
er
ea
s
Den
s
eNe
t2
0
1
r
ea
ch
ed
8
3
.
6
6
%
,
s
u
r
p
ass
in
g
C
L
AHE
(
8
0
.
6
9
%)
a
n
d
PMD
(
7
5
.
5
0
%)
p
er
f
o
r
m
an
ce
.
T
h
is
in
d
icate
s
th
at
th
e
Den
s
eNe
t
m
o
d
els
b
en
ef
it
s
ig
n
if
ican
tly
f
r
o
m
n
o
is
e
r
ed
u
ctio
n
a
n
d
co
n
tr
ast
en
h
an
c
em
en
t.
T
h
e
R
esNet
ar
ch
itectu
r
es
also
p
er
f
o
r
m
e
d
well,
with
R
esNet5
0
ac
h
iev
in
g
8
5
.
1
5
%
ac
c
u
r
ac
y
an
d
im
p
r
o
v
in
g
b
o
th
PMD
(
7
7
.
2
3
%)
an
d
C
L
AHE
(
8
2
.
9
2
%).
R
esNet3
4
r
ea
ch
es
8
4
.
4
1
%,
s
h
o
win
g
a
s
im
ilar
tr
en
d
.
T
h
is
co
n
f
ir
m
ed
th
at
co
n
tr
ast
en
h
a
n
ce
m
en
t
with
C
L
AHE
was
ad
eq
u
ate;
h
o
wev
er
,
n
o
is
e
r
ed
u
ctio
n
f
u
r
th
er
r
e
f
in
ed
th
e
ex
tr
ac
ted
f
ea
tu
r
es,
lead
in
g
t
o
b
etter
clas
s
if
icatio
n
ac
cu
r
ac
y
.
Ho
wev
er
,
Sq
u
ee
ze
Net
-
1
.
0
d
id
n
o
t
b
en
ef
it
as
m
u
ch
f
r
o
m
th
e
h
y
b
r
id
a
p
p
r
o
ac
h
,
ac
h
iev
i
n
g
an
ac
cu
r
ac
y
o
f
7
8
.
4
7
%,
wh
ich
is
lo
wer
th
an
its
PMD
-
o
n
ly
r
esu
lt
(
8
0
.
2
0
%),
b
u
t
h
ig
h
er
th
an
its
C
L
AHE
r
esu
lt
(
7
4
.
7
5
%).
T
h
is
s
u
g
g
ests
th
at
lig
h
tweig
h
t
ar
ch
itectu
r
es
lik
e
Sq
u
ee
z
e
Net
m
ay
n
o
t
g
ain
s
ig
n
i
f
ican
t
ad
v
an
tag
es
f
r
o
m
ex
ten
s
iv
e
p
r
ep
r
o
ce
s
s
in
g
an
d
m
ay
b
e
m
o
r
e
o
p
tim
ized
f
o
r
s
tr
aig
h
tf
o
r
war
d
n
o
is
e
r
ed
u
ctio
n
tech
n
i
q
u
es.
T
h
e
h
y
b
r
id
PMD
FIL
T
E
R
-
C
L
AHE
ap
p
r
o
ac
h
ac
h
ie
v
ed
th
e
b
est
o
v
er
all
p
er
f
o
r
m
an
ce
f
o
r
m
o
s
t
C
NN
ar
ch
itectu
r
es,
p
ar
ticu
la
r
ly
f
o
r
d
ee
p
er
n
etwo
r
k
s
s
u
ch
as
E
f
f
icien
tNet
-
B
0
,
E
f
f
icien
tNet
-
B
1
,
Den
s
eNe
t1
2
1
,
an
d
De
n
s
eNe
t2
0
1
.
R
esNet
ar
ch
itectu
r
es
b
en
ef
it
s
ig
n
if
ican
tly
f
r
o
m
th
e
h
y
b
r
id
ap
p
r
o
ac
h
,
with
R
esNet5
0
im
p
r
o
v
i
n
g
th
e
m
o
s
t
(
f
r
o
m
7
1
.
5
3
%
i
n
th
e
o
r
ig
in
al
to
8
5
.
1
5
%
with
h
y
b
r
id
p
r
o
c
ess
in
g
)
.
Sq
u
ee
ze
Net
-
1
.
0
p
er
f
o
r
m
s
b
est
with
th
e
PMD
f
ilter
alo
n
e
(
8
0
.
2
0
%),
i
n
d
icatin
g
t
h
at
lig
h
tweig
h
t
ar
c
h
itectu
r
es
m
ay
b
en
e
f
it
m
o
r
e
f
r
o
m
n
o
is
e
r
e
d
u
ctio
n
th
an
co
n
tr
ast
en
h
an
ce
m
e
n
t.
E
f
f
icien
tNet
-
B
0
ac
h
iev
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
(
9
1
.
7
4
%)
with
h
y
b
r
id
p
r
o
ce
s
s
in
g
,
s
lig
h
tly
o
u
tp
er
f
o
r
m
in
g
th
e
o
r
ig
in
al
im
ag
e
ac
cu
r
ac
y
(
9
1
.
5
8
%).
T
h
e
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
tech
n
iq
u
e
p
r
o
v
id
es
an
o
p
tim
al
b
alan
ce
f
o
r
p
r
ep
r
o
ce
s
s
in
g
p
ap
s
m
ea
r
im
ag
es,
m
ax
im
izin
g
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
u
ltip
le
C
NN
ar
ch
itectu
r
es.
T
h
is
was th
e
m
o
s
t e
f
f
ec
ti
v
e
p
r
ep
r
o
ce
s
s
in
g
m
eth
o
d
u
s
ed
in
th
is
s
tu
d
y
.
T
ab
le
4
.
C
NN
Ar
ch
itectu
r
es’
ass
es
s
m
en
t sco
r
e
f
o
r
p
a
p
s
m
ea
r
im
ag
es u
s
in
g
a
h
y
b
r
i
d
PMD
f
ilter
-
C
L
AHE
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
e
A
r
c
h
i
t
e
c
t
u
r
e
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
Tr
a
i
n
i
n
g
t
i
me
(
s)
R
e
sN
e
t
3
4
84
.
4
1
%
85
.
7
8
%
84
.
5
7
%
83
.
0
9
%
7
4
9
R
e
sN
e
t
5
0
85
.
1
5
%
85
.
7
0
%
85
.
3
2
%
84
.
8
2
%
8
9
4
S
q
u
e
e
z
e
N
e
t
-
1
.
0
78
.
4
7
%
81
.
8
5
%
78
.
7
1
%
77
.
2
4
%
6
3
0
M
o
b
i
l
e
N
e
t
-
V2
83
.
9
1
%
86
.
5
6
%
84
.
1
8
%
83
.
2
5
%
6
8
5
Ef
f
i
c
i
e
n
t
N
e
t
-
B0
91
.
7
4
%
91
.
5
2
%
91
.
2
8
%
91
.
3
3
%
8
6
2
Ef
f
i
c
i
e
n
t
N
e
t
-
B1
89
.
7
7
%
88
.
7
5
%
88
.
4
1
%
89
.
9
4
%
1
,
0
5
2
D
e
n
seN
e
t
1
2
1
86
.
3
9
%
87
.
4
7
%
86
.
5
7
%
86
.
2
1
%
9
5
5
D
e
n
seN
e
t
2
0
1
83
.
6
6
%
85
.
2
9
%
83
.
9
1
%
83
.
1
3
%
10
,
305
4.
CO
NCLU
SI
O
N
C
er
v
ical
ca
n
ce
r
is
a
m
ajo
r
h
e
alth
co
n
ce
r
n
,
esp
ec
ially
in
d
e
v
elo
p
in
g
co
u
n
tr
ies,
wh
er
e
ea
r
ly
d
etec
tio
n
is
cr
u
cial.
Dee
p
lea
r
n
in
g
,
p
a
r
ticu
lar
ly
C
NNs,
h
as
s
h
o
wn
p
r
o
m
is
e
f
o
r
a
u
to
m
ated
ce
r
v
ical
c
an
ce
r
class
if
icatio
n
,
b
u
t
p
a
p
s
m
ea
r
im
ag
e
q
u
ality
s
ig
n
if
ican
tly
af
f
ec
ts
p
er
f
o
r
m
a
n
ce
.
T
h
is
s
tu
d
y
e
v
alu
ated
th
e
ef
f
ec
ts
o
f
th
e
PMD
f
ilter
,
C
L
AHE
,
an
d
th
ei
r
h
y
b
r
id
ap
p
r
o
ac
h
o
n
ce
r
v
ical
ca
n
c
er
class
if
icatio
n
.
R
esu
lts
s
h
o
wed
th
at
th
e
h
y
b
r
id
PMD
f
ilter
-
C
L
AHE
o
u
tp
er
f
o
r
m
ed
in
d
iv
id
u
al
p
r
ep
r
o
ce
s
s
in
g
m
eth
o
d
s
,
with
m
ax
im
u
m
im
p
r
o
v
e
m
en
ts
o
f
1
3
.
6
2
%
in
ac
cu
r
ac
y
,
1
0
.
0
4
%
in
p
r
ec
is
io
n
,
1
3
.
0
8
%
in
r
ec
a
ll,
an
d
1
4
.
3
4
%
in
F1
-
s
co
r
e.
PMD
f
ilter
in
g
wa
s
p
ar
ticu
lar
ly
ef
f
ec
tiv
e
in
lig
h
t
weig
h
t
C
NN
ar
ch
itectu
r
es,
w
h
ile
C
L
AHE
b
en
ef
ited
d
ee
p
e
r
ar
ch
itectu
r
es.
T
h
e
h
y
b
r
id
ap
p
r
o
ac
h
b
alan
ce
d
n
o
is
e
r
ed
u
ctio
n
an
d
co
n
t
r
ast
en
h
an
ce
m
en
t,
im
p
r
o
v
in
g
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
o
s
t
C
NNs
with
o
u
t
s
ig
n
if
ican
tly
i
n
cr
ea
s
in
g
p
r
o
ce
s
s
in
g
tim
e.
Alth
o
u
g
h
th
is
s
tu
d
y
d
em
o
n
s
tr
a
ted
th
e
b
en
ef
its
o
f
h
y
b
r
id
p
r
ep
r
o
ce
s
s
in
g
,
f
u
r
th
e
r
r
esear
ch
is
n
ee
d
ed
to
co
n
f
ir
m
its
ef
f
ec
tiv
en
ess
ac
r
o
s
s
d
if
f
er
en
t
im
ag
in
g
co
n
d
itio
n
s
,
s
tain
in
g
tech
n
iq
u
e
s
,
an
d
d
atasets
.
L
ar
g
er
-
s
ca
le
v
alid
atio
n
an
d
r
ea
l
-
tim
e
im
p
le
m
en
tatio
n
in
clin
ical
wo
r
k
f
lo
ws
co
u
l
d
en
h
a
n
ce
its
p
r
ac
tical
ap
p
licatio
n
.
T
h
ese
f
i
n
d
in
g
s
h
ig
h
lig
h
t
th
e
p
o
ten
tial
o
f
co
m
b
in
in
g
n
o
is
e
r
ed
u
ctio
n
a
n
d
c
o
n
tr
ast
en
h
a
n
ce
m
en
t
to
im
p
r
o
v
e
C
NN
-
b
ased
ce
r
v
ical
ca
n
ce
r
class
if
icati
o
n
.
Fu
tu
r
e
r
esear
c
h
s
h
o
u
ld
f
o
c
u
s
o
n
o
p
tim
izin
g
p
r
ep
r
o
ce
s
s
in
g
p
ar
a
m
eter
s
u
s
in
g
ad
v
an
ce
d
alg
o
r
ith
m
s
an
d
i
n
teg
r
atin
g
r
ea
l
-
tim
e
m
eth
o
d
s
to
en
h
an
ce
ac
c
u
r
ac
y
an
d
ef
f
icien
c
y
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
es
ea
r
ch
was
f
u
n
d
ed
b
y
th
e
Min
is
tr
y
o
f
Hig
h
er
E
d
u
ca
tio
n
,
Scien
ce
,
an
d
T
ec
h
n
o
l
o
g
y
o
f
th
e
R
ep
u
b
lic
o
f
I
n
d
o
n
esia
an
d
t
h
e
I
n
d
o
n
esian
E
d
u
ca
tio
n
Fo
u
n
d
atio
n
(
L
PDP)
th
r
o
u
g
h
t
h
e
C
en
ter
f
o
r
Hig
h
er
E
d
u
ca
tio
n
Fu
n
d
in
g
an
d
Ass
ess
m
en
t (
PP
APT)
u
n
d
er
th
e
I
n
d
o
n
esian
E
d
u
ca
tio
n
Sch
o
lar
s
h
i
p
(
B
PI)
p
r
o
g
r
am
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
d
va
n
ce
d
ce
r
vica
l c
a
n
ce
r
cla
s
s
ifica
tio
n
:
en
h
a
n
cin
g
p
a
p
s
me
a
r
ima
g
es w
ith
h
yb
r
id
P
MD
…
(
A
ch
K
h
o
z
a
imi
)
653
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
All a
u
th
o
r
s
co
n
tr
ib
u
ted
as f
o
ll
o
ws an
d
h
av
e
r
ea
d
a
n
d
ap
p
r
o
v
ed
th
e
f
in
al
m
a
n
u
s
cr
ip
t.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Ach
Kh
o
za
im
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
I
s
n
an
i D
ar
ti
✓
✓
✓
✓
✓
✓
✓
Sy
aif
u
l A
n
am
✓
✓
✓
✓
✓
✓
✓
✓
✓
W
u
r
y
an
s
ar
i
Mu
h
ar
in
i
Ku
s
u
m
awin
ah
y
u
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
s
co
n
f
ir
m
th
at
th
ey
h
av
e
n
o
co
n
f
licts
o
f
in
ter
est r
eg
ar
d
in
g
th
e
p
u
b
licatio
n
o
f
th
is
wo
r
k
an
d
n
o
f
in
a
n
cial
o
r
p
er
s
o
n
al
af
f
iliatio
n
s
th
at
co
u
ld
a
f
f
ec
t its
co
n
t
en
t.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
p
u
b
licly
a
v
ailab
le
SIPaK
Me
D
d
ataset
was
u
s
ed
in
th
is
s
tu
d
y
an
d
ca
n
b
e
ac
ce
s
s
ed
at
h
ttp
s
://b
it.ly
/SIPaKM
eD
f
o
r
f
u
r
th
er
r
esear
c
h
in
m
e
d
ical
im
a
g
e
p
r
o
ce
s
s
in
g
a
n
d
ce
r
v
ical
ca
n
ce
r
class
if
icatio
n
.
RE
F
E
R
E
NC
E
S
[
1
]
I
.
P
a
c
a
l
a
n
d
S
.
K
ı
l
ı
c
a
r
sl
a
n
,
“
D
e
e
p
l
e
a
r
n
i
n
g
-
b
a
s
e
d
a
p
p
r
o
a
c
h
e
s
f
o
r
r
o
b
u
s
t
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
c
e
r
v
i
c
a
l
c
a
n
c
e
r
,
”
N
e
u
r
a
l
C
o
m
p
u
t
i
n
g
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
5
,
n
o
.
2
5
,
p
p
.
1
8
8
1
3
–
1
8
8
2
8
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
5
2
1
-
023
-
0
8
7
5
7
-
w.
[
2
]
D
.
S
a
mb
y
a
l
a
n
d
A
.
S
a
r
w
a
r
,
“
R
e
c
e
n
t
d
e
v
e
l
o
p
me
n
t
s
i
n
c
e
r
v
i
c
a
l
c
a
n
c
e
r
d
i
a
g
n
o
s
i
s
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
o
n
w
h
o
l
e
sl
i
d
e
i
ma
g
e
s
:
a
n
O
v
e
r
v
i
e
w
o
f
mo
d
e
l
s,
t
e
c
h
n
i
q
u
e
s
,
c
h
a
l
l
e
n
g
e
s
a
n
d
f
u
t
u
r
e
d
i
r
e
c
t
i
o
n
s
,
”
Mi
c
r
o
n
,
v
o
l
.
1
7
3
,
n
o
.
A
p
r
i
l
,
p
.
1
0
3
5
2
0
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
m
i
c
r
o
n
.
2
0
2
3
.
1
0
3
5
2
0
.
[
3
]
A
.
D
.
J
i
a
,
B
.
Z
.
Li
,
a
n
d
C
.
C
.
Z
h
a
n
g
,
“
D
e
t
e
c
t
i
o
n
o
f
c
e
r
v
i
c
a
l
c
a
n
c
e
r
c
e
l
l
s
b
a
s
e
d
o
n
st
r
o
n
g
f
e
a
t
u
r
e
C
N
N
-
S
V
M
n
e
t
w
o
r
k
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
4
1
1
,
p
p
.
1
1
2
–
1
2
7
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m
.
2
0
2
0
.
0
6
.
0
0
6
.
[
4
]
H
.
J
.
H
o
n
,
P
.
P
.
C
h
o
n
g
,
H
.
L
.
C
h
o
o
,
a
n
d
P
.
P
.
K
h
i
n
e
,
“
A
c
o
m
p
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
f
c
e
r
v
i
c
a
l
c
a
n
c
e
r
s
c
r
e
e
n
i
n
g
d
e
v
i
c
e
s
:
t
h
e
p
r
o
s
a
n
d
t
h
e
co
n
s
,
”
A
s
i
a
n
P
a
c
i
f
i
c
J
o
u
r
n
a
l
o
f
C
a
n
c
e
r
P
r
e
v
e
n
t
i
o
n
,
v
o
l
.
2
4
,
n
o
.
7
,
p
p
.
2
2
0
7
–
2
2
1
5
,
J
u
l
.
2
0
2
3
,
d
o
i
:
1
0
.
3
1
5
5
7
/
A
P
J
C
P
.
2
0
2
3
.
2
4
.
7
.
2
2
0
7
.
[
5
]
Y.
-
M
.
L
e
e
,
B
.
Le
e
,
N
.
-
H
.
C
h
o
,
a
n
d
J.
H
.
P
a
r
k
,
“
B
e
y
o
n
d
t
h
e
mi
c
r
o
sc
o
p
e
:
a
t
e
c
h
n
o
l
o
g
i
c
a
l
o
v
e
r
t
u
r
e
f
o
r
c
e
r
v
i
c
a
l
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
,
”
D
i
a
g
n
o
s
t
i
c
s
,
v
o
l
.
1
3
,
n
o
.
1
9
,
p
.
3
0
7
9
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
st
i
c
s
1
3
1
9
3
0
7
9
.
[
6
]
B
.
C
h
i
t
r
a
a
n
d
S
.
S
.
K
u
m
a
r
,
“
R
e
c
e
n
t
a
d
v
a
n
c
e
m
e
n
t
i
n
c
e
r
v
i
c
a
l
c
a
n
c
e
r
d
i
a
g
n
o
s
i
s
f
o
r
a
u
t
o
m
a
t
e
d
s
c
r
e
e
n
i
n
g
:
a
d
e
t
a
i
l
e
d
r
e
v
i
e
w
,
”
J
o
u
r
n
a
l
o
f
A
m
b
i
e
n
t
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
H
u
m
a
n
i
z
e
d
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
3
,
n
o
.
1
,
p
p
.
2
5
1
–
2
6
9
,
J
a
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
2
6
5
2
-
0
2
1
-
0
2
8
9
9
-
2.
[
7
]
N
.
S
o
m
p
a
w
o
n
g
e
t
a
l
.
,
“
A
u
t
o
m
a
t
e
d
p
a
p
sme
a
r
c
e
r
v
i
c
a
l
c
a
n
c
e
r
scre
e
n
i
n
g
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
A
n
n
u
a
l
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
f
t
h
e
I
EEE
En
g
i
n
e
e
ri
n
g
i
n
M
e
d
i
c
i
n
e
a
n
d
B
i
o
l
o
g
y
S
o
c
i
e
t
y
,
EM
BS
,
p
p
.
7
0
4
4
–
7
0
4
8
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
EM
B
C
.
2
0
1
9
.
8
8
5
6
3
6
9
.
[
8
]
S
.
D
o
d
g
e
a
n
d
L.
K
a
r
a
m
,
“
U
n
d
e
r
st
a
n
d
i
n
g
h
o
w
i
m
a
g
e
q
u
a
l
i
t
y
a
f
f
e
c
t
s
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
i
n
2
0
1
6
E
i
g
h
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Q
u
a
l
i
t
y
o
f
M
u
l
t
i
m
e
d
i
a
Ex
p
e
ri
e
n
c
e
(
Q
o
M
EX)
,
I
EEE,
Ju
n
.
2
0
1
6
,
p
p
.
1
–
6,
d
o
i
:
1
0
.
1
1
0
9
/
Q
o
M
EX
.
2
0
1
6
.
7
4
9
8
9
5
5
.
[
9
]
K
.
P
.
W
i
n
,
Y
.
K
i
t
j
a
i
d
u
r
e
,
K
.
H
a
m
a
m
o
t
o
,
a
n
d
T.
M
.
A
u
n
g
,
“
C
o
m
p
u
t
e
r
-
a
ssi
st
e
d
scr
e
e
n
i
n
g
f
o
r
c
e
r
v
i
c
a
l
c
a
n
c
e
r
u
s
i
n
g
d
i
g
i
t
a
l
i
m
a
g
e
p
r
o
c
e
ss
i
n
g
o
f
p
a
p
sm
e
a
r
i
ma
g
e
s
,
”
Ap
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
0
,
n
o
.
5
,
p
.
1
8
0
0
,
M
a
r
.
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
0
0
5
1
8
0
0
.
[
1
0
]
A
.
K
h
o
z
a
i
m
i
a
n
d
W
.
F
.
M
a
h
m
u
d
y
,
“
N
e
w
i
n
s
i
g
h
t
i
n
c
e
r
v
i
c
a
l
c
a
n
c
e
r
d
i
a
g
n
o
s
i
s
u
s
i
n
g
c
o
n
v
o
l
u
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
a
r
c
h
i
t
e
c
t
u
r
e
,
”
I
A
E
S
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
J
-
A
I
)
,
v
o
l
.
1
3
,
n
o
.
3
,
p
.
3
0
9
2
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
3
.
i
3
.
p
p
3
0
9
2
-
3
1
0
0
.
[
1
1
]
T.
C
h
a
n
k
o
n
g
,
“
A
u
t
o
ma
t
i
c
c
l
a
s
si
f
y
i
n
g
o
f
c
e
r
v
i
c
a
l
c
e
l
l
s
u
si
n
g
f
o
u
r
i
e
r
s
p
e
c
t
r
a
l
f
e
a
t
u
r
e
s
,
”
i
n
2
0
1
8
4
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
G
ree
n
T
e
c
h
n
o
l
o
g
y
a
n
d
S
u
st
a
i
n
a
b
l
e
D
e
v
e
l
o
p
m
e
n
t
(
G
T
S
D
)
,
I
EEE,
N
o
v
.
2
0
1
8
,
p
p
.
7
5
9
–
7
6
2
,
d
o
i
:
1
0
.
1
1
0
9
/
G
TSD
.
2
0
1
8
.
8
5
9
5
6
6
2
.
[
1
2
]
M
.
Z
h
a
o
e
t
a
l
.
,
“
S
E
EN
S
:
N
u
c
l
e
i
s
e
g
me
n
t
a
t
i
o
n
i
n
p
a
p
sme
a
r
i
m
a
g
e
s
w
i
t
h
s
e
l
e
c
t
i
v
e
e
d
g
e
e
n
h
a
n
c
e
me
n
t
,
”
F
u
t
u
r
e
G
e
n
e
r
a
t
i
o
n
C
o
m
p
u
t
e
r
S
y
st
e
m
s
,
v
o
l
.
1
1
4
,
p
p
.
1
8
5
–
1
9
4
,
Ja
n
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
f
u
t
u
r
e
.
2
0
2
0
.
0
7
.
0
4
5
.
[
1
3
]
V
.
K
a
ma
l
a
v
e
n
i
,
R
.
A
.
R
a
j
a
l
a
k
s
h
m
i
,
a
n
d
K
.
A
.
N
a
r
a
y
a
n
a
n
k
u
t
t
y
,
“
I
mag
e
d
e
n
o
i
s
i
n
g
u
si
n
g
v
a
r
i
a
t
i
o
n
s
o
f
P
e
r
o
n
a
-
M
a
l
i
k
m
o
d
e
l
w
i
t
h
d
i
f
f
e
r
e
n
t
e
d
g
e
s
t
o
p
p
i
n
g
f
u
n
c
t
i
o
n
s,
”
Pr
o
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
5
8
,
p
p
.
6
7
3
–
6
8
2
,
2
0
1
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
1
5
.
0
8
.
0
8
7
.
[
1
4
]
S
.
A
n
a
m,
Z
.
F
i
t
r
i
a
h
,
a
n
d
N
.
S
h
o
f
i
a
n
a
h
,
“
H
y
b
r
i
d
o
f
t
h
e
P
M
D
F
i
l
t
e
r
,
t
h
e
K
-
mea
n
s
c
l
u
s
t
e
r
i
n
g
m
e
t
h
o
d
a
n
d
t
h
e
l
e
v
e
l
se
t
m
e
t
h
o
d
f
o
r
e
x
u
d
a
t
e
s
se
g
me
n
t
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
M
a
t
h
e
m
a
t
i
c
s
a
n
d
I
sl
a
m
,
S
C
I
TEPRE
S
S
-
S
c
i
e
n
c
e
a
n
d
Te
c
h
n
o
l
o
g
y
P
u
b
l
i
c
a
t
i
o
n
s
,
Ja
n
.
2
0
1
8
,
p
p
.
1
0
8
–
1
1
6
.
d
o
i
:
1
0
.
5
2
2
0
/
0
0
0
8
5
1
7
9
0
1
0
8
0
1
1
6
.
[
1
5
]
S
.
V
.
Ez
h
i
l
r
a
ma
n
,
S
.
S
r
i
n
i
v
a
sa
n
,
a
n
d
G
.
S
u
s
e
e
n
d
r
a
n
,
“
B
i
l
a
t
e
r
a
l
p
e
r
o
n
a
-
mal
i
k
d
i
f
f
u
s
i
o
n
f
i
l
t
e
r
i
n
g
b
a
s
e
d
t
o
p
o
l
o
g
i
c
a
l
mu
l
t
i
t
u
d
e
f
e
a
t
u
r
e
v
e
c
t
o
r
f
o
r
b
r
e
a
st
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
R
e
se
a
rc
h
o
n
t
h
e
L
e
p
i
d
o
p
t
e
r
a
,
v
o
l
.
5
1
,
n
o
.
1
,
p
p
.
1
1
0
–
1
2
8
,
F
e
b
.
2
0
2
0
,
d
o
i
:
1
0
.
3
6
8
7
2
/
L
EPI
/
V
5
1
I
1
/
3
0
1
0
1
0
.
[
1
6
]
M
.
M
.
R
a
h
a
ma
n
e
t
a
l
.
,
“
A
su
r
v
e
y
f
o
r
c
e
r
v
i
c
a
l
c
y
t
o
p
a
t
h
o
l
o
g
y
i
ma
g
e
a
n
a
l
y
si
s
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
I
EE
E
Ac
c
e
ss
,
v
o
l
.
8
,
p
p
.
6
1
6
8
7
–
6
1
7
1
0
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
0
.
2
9
8
3
1
8
6
.
[
1
7
]
E.
A
.
Tj
o
a
,
I
.
P
.
Y
.
N
.
S
u
p
a
r
t
a
,
R
.
M
a
g
d
a
l
e
n
a
,
a
n
d
N
.
K
u
m
a
l
a
sari
C
P
,
“
T
h
e
u
se
o
f
C
LA
H
E
f
o
r
i
m
p
r
o
v
i
n
g
a
n
a
c
c
u
r
a
c
y
o
f
C
N
N
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
d
e
t
e
c
t
i
n
g
p
n
e
u
mo
n
i
a
,
”
S
H
S
We
b
o
f
C
o
n
f
e
r
e
n
c
e
s
,
v
o
l
.
1
3
9
,
p
.
0
3
0
2
6
,
M
a
y
2
0
2
2
,
d
o
i
:
1
0
.
1
0
5
1
/
sh
sc
o
n
f
/
2
0
2
2
1
3
9
0
3
0
2
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.