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.
23
,
No
.
3
,
Sep
tem
b
er
2
0
2
1
,
p
p
.
1
6
1
1
~
1
6
1
9
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
23
.i
3
.
pp
1
6
1
1
-
1
6
1
9
1611
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Big
trans
fer
l
ea
rn
ing
f
o
r auto
ma
ted
skin cancer
clas
si
ficatio
n
Z
ina
h
M
o
hs
i
n Ar
k
a
h
1
,
Da
ly
a
S.
Al
-
Dula
im
i
2
,
Ahla
m
R.
K
hek
a
n
3
1
De
p
a
rtme
n
t
o
f
Bi
o
in
f
o
rm
a
ti
c
s,
Un
iv
e
rsit
y
o
f
In
f
o
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
a
n
d
C
o
m
m
u
n
ica
ti
o
n
s,
Ira
q
2,
3
De
p
a
rtme
n
t
o
f
I
n
telli
g
e
n
t
M
e
d
i
c
a
l
S
y
ste
m
,
Un
iv
e
rsit
y
o
f
In
f
o
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
a
n
d
Co
m
m
u
n
ica
ti
o
n
s,
Ira
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
M
a
y
4
,
2
0
2
1
R
ev
is
ed
Ju
l
1
8
,
2
0
2
1
Acc
ep
ted
J
u
l
26
,
2
0
2
1
S
k
in
c
a
n
c
e
r
is
a
n
e
x
a
m
p
le
o
f
th
e
m
o
st
d
a
n
g
e
ro
u
s
d
ise
a
se
.
Early
d
i
a
g
n
o
sis
o
f
sk
in
c
a
n
c
e
r
c
a
n
sa
v
e
m
a
n
y
p
e
o
p
l
e
’s
li
v
e
s.
M
a
n
u
a
l
c
las
sifica
ti
o
n
m
e
th
o
d
s
a
re
ti
m
e
-
c
o
n
su
m
in
g
a
n
d
c
o
stl
y
.
D
e
e
p
lea
rn
in
g
h
a
s
b
e
e
n
p
r
o
p
o
s
e
d
fo
r
t
h
e
a
u
to
m
a
ted
c
las
sifica
ti
o
n
o
f
sk
i
n
c
a
n
c
e
r.
Alth
o
u
g
h
d
e
e
p
lea
rn
in
g
sh
o
we
d
imp
re
ss
iv
e
p
e
rfo
rm
a
n
c
e
in
se
v
e
r
a
l
m
e
d
ica
l
ima
g
i
n
g
tas
k
s,
it
re
q
u
ires
a
b
ig
n
u
m
b
e
r
o
f
ima
g
e
s
to
a
c
h
iev
e
a
g
o
o
d
p
e
rfo
rm
a
n
c
e
.
Th
e
sk
in
c
a
n
c
e
r
c
las
sifica
ti
o
n
tas
k
su
ffe
rs
fro
m
p
ro
v
i
d
in
g
d
e
e
p
lea
rn
i
n
g
wit
h
su
ffi
c
ien
t
d
a
ta
d
u
e
t
o
t
h
e
e
x
p
e
n
siv
e
a
n
n
o
tati
o
n
p
ro
c
e
ss
a
n
d
re
q
u
ired
e
x
p
e
rts.
On
e
o
f
t
h
e
m
o
st
u
se
d
so
l
u
ti
o
n
s
is
tran
sfe
r
lea
rn
in
g
o
f
p
re
-
train
e
d
m
o
d
e
ls
o
f
th
e
Im
a
g
e
Ne
t
d
a
tas
e
t.
Ho
we
v
e
r,
t
h
e
lea
rn
e
d
fe
a
tu
re
s
o
f
p
re
-
train
e
d
m
o
d
e
ls
a
re
d
iffere
n
t
fro
m
sk
i
n
c
a
n
c
e
r
ima
g
e
fe
a
tu
re
s.
T
o
e
n
d
th
is,
we
in
tr
o
d
u
c
e
a
n
o
v
e
l
a
p
p
ro
a
c
h
o
f
tran
sfe
r
lea
rn
in
g
b
y
train
i
n
g
t
h
e
p
re
-
train
e
d
m
o
d
e
ls
o
f
th
e
Im
a
g
e
Ne
t
(V
GG
,
G
o
o
g
leN
e
t,
a
n
d
Re
sN
e
t5
0
)
o
n
a
larg
e
n
u
m
b
e
r
o
f
u
n
lab
e
ll
e
d
sk
in
c
a
n
c
e
r
ima
g
e
s,
first.
We
th
e
n
train
th
e
m
o
n
a
sm
a
ll
n
u
m
b
e
r
o
f
lab
e
led
sk
in
ima
g
e
s.
Ou
r
e
x
p
e
rime
n
tal
r
e
su
lt
s
p
ro
v
e
d
t
h
a
t
th
e
p
ro
p
o
se
d
m
e
th
o
d
is
e
fficie
n
t
b
y
a
c
h
iev
i
n
g
a
n
a
c
c
u
ra
c
y
o
f
8
4
%
with
Re
sN
e
t5
0
wh
e
n
d
irec
tl
y
train
e
d
wi
th
a
sm
a
ll
n
u
m
b
e
r
o
f
lab
e
led
s
k
in
a
n
d
9
3
.
7
%
w
h
e
n
train
e
d
wit
h
t
h
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
.
K
ey
w
o
r
d
s
:
C
NN
Dee
p
lear
n
in
g
Pre
-
tr
ain
ed
m
o
d
els
Sk
in
ca
n
ce
r
T
r
an
s
f
er
lear
n
i
n
g
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
:
Z
in
ah
Mo
h
s
in
Ar
k
a
Dep
ar
tm
en
t o
f
B
io
in
f
o
r
m
atics
Co
lleg
e
o
f
B
io
m
ed
ical
I
n
f
o
r
m
atics
Un
iv
er
s
ity
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
an
d
C
o
m
m
u
n
icatio
n
s
B
ag
h
d
ad
,
I
r
aq
E
m
ail: z
in
ah
2
0
1
8
@
u
o
itc.
ed
u
.
i
q
1.
I
NT
RO
D
UCT
I
O
N
No
wad
ay
s
,
o
n
e
o
f
th
e
d
ea
d
lie
s
t
an
d
f
astes
t
-
s
p
r
ea
d
in
g
ca
n
ce
r
s
in
th
e
wo
r
ld
is
s
k
in
ca
n
ce
r
.
Sev
en
ty
-
f
iv
e
p
er
ce
n
t
o
f
th
e
s
k
in
ca
n
ce
r
p
atien
ts
ar
e
d
y
in
g
ev
e
r
y
y
ea
r
[
1
]
-
[
3
]
,
wh
ile
in
th
e
USA,
th
er
e
is
a
d
an
g
er
o
f
s
k
in
ca
n
ce
r
in
e
v
er
y
f
iv
e
p
er
s
o
n
s
,
m
o
s
tly
liv
i
n
g
i
n
an
ex
tr
e
m
ely
h
ig
h
s
u
n
s
h
in
e
ar
ea
an
d
h
av
in
g
p
ale
s
k
in
t
y
p
e
[
1
]
.
Als
o
f
o
r
2
0
1
7
,
m
o
r
e
th
a
n
8
7
,
0
0
0
e
m
er
g
in
g
ca
s
es
o
f
m
el
an
o
m
a
a
r
e
e
x
p
ec
ted
to
d
iag
n
o
s
e
in
th
e
USA
[
2
]
.
I
n
Au
s
tr
alia,
1
5
2
0
p
e
o
p
le
d
ie
d
f
r
o
m
m
elan
o
m
a
an
d
6
4
2
f
r
o
m
n
o
n
-
m
elan
o
m
a
in
2
0
1
5
.
At
ea
r
ly
s
tag
es
o
f
an
y
d
is
ea
s
e,
in
clu
d
in
g
s
k
in
ca
n
ce
r
,
ca
n
b
e
d
iag
n
o
s
ed
,
tr
ea
ted
,
a
n
d
cu
r
e
d
ea
s
ily
lik
e
o
t
h
er
d
is
ea
s
es,
ea
r
ly
d
iag
n
o
s
is
is
s
o
s
ig
n
if
ican
t
f
o
r
h
an
d
lin
g
an
d
ca
n
q
u
ick
l
y
at
in
itial
d
eg
r
ee
s
[
1
]
-
[
4
]
.
Fo
r
h
elp
in
g
d
e
r
m
a
to
lo
g
is
ts
in
clin
ical
test
in
g
,
a
s
k
in
ca
n
ce
r
d
ia
g
n
o
s
is
h
as
b
ee
n
im
p
r
o
v
ed
b
y
in
tr
o
d
u
cin
g
a
d
er
m
o
s
co
p
y
tec
h
n
iq
u
e.
I
t
is
a
n
o
n
-
in
v
asiv
e
im
ag
in
g
tec
h
n
iq
u
e
w
h
ich
p
r
o
v
i
d
es
a
h
ig
h
-
q
u
ality
v
is
u
al
ap
p
ea
r
an
ce
o
f
s
k
in
lesi
o
n
.
De
r
m
o
s
co
p
y
is
lo
wer
s
cr
ee
n
in
g
e
r
r
o
r
s
,
m
o
r
e
s
atis
f
ac
to
r
y
d
ee
p
la
y
er
s
d
etails,
an
d
f
ewe
r
s
u
r
f
ac
e
r
ef
lectio
n
s
,
as
co
m
p
ar
e
d
with
co
n
v
en
tio
n
al
m
icr
o
s
co
p
y
.
Hen
ce
,
d
er
m
o
s
co
p
y
im
a
g
es o
f
f
er
g
r
ea
tly
b
etter
ac
c
u
r
ac
y
a
n
d
v
is
ib
ilit
y
[
3
]
-
[
5
]
.
R
ec
en
tly
,
th
e
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
tech
n
i
q
u
e,
as
an
ex
am
p
le
o
f
th
e
d
ee
p
lear
n
i
n
g
ap
p
r
o
ac
h
,
h
as
d
ev
elo
p
ed
an
o
v
er
p
o
wer
in
g
ex
is
ten
ce
in
im
a
g
e
class
if
icatio
n
task
s
[
6
]
-
[
10
].
T
h
e
ex
citin
g
an
d
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.
23
,
No
.
3
,
Sep
tem
b
er
2
0
2
1
:
1
6
1
1
-
1
6
1
9
1612
b
r
illi
an
t
ca
p
ab
ilit
y
o
f
v
is
u
al
r
ep
r
esen
tatio
n
f
o
r
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
task
s
,
d
ep
en
d
in
g
o
n
th
e
s
p
ec
if
ied
tr
ain
in
g
d
ata
g
r
o
u
p
,
is
th
e
m
o
s
t
im
p
o
r
tan
t
ad
v
a
n
tag
e
o
f
C
NN
[
11
]
.
I
n
g
en
er
al,
s
ev
e
r
al
r
esear
ch
wo
r
k
s
ac
h
iev
ed
u
p
-
to
-
d
ate
p
e
r
f
o
r
m
a
n
ce
in
th
e
task
o
f
s
k
in
ca
n
ce
r
class
if
icatio
n
[
12]
,
[
1
3
]
.
T
h
e
C
NN
m
o
d
els,
p
r
e
-
tr
ain
ed
o
n
b
ig
d
atasets
o
f
I
m
a
g
eNe
t
[
14
]
,
p
r
o
v
id
e
d
ass
u
r
in
g
r
esu
lts
f
o
r
im
a
g
e
d
ia
g
n
o
s
is
task
s
,
s
till
ex
clu
s
iv
e
o
f
r
etr
ai
n
in
g
.
Aim
ed
at
th
is
ex
p
lan
atio
n
,
th
ese
tr
an
s
f
er
r
ed
f
e
atu
r
es
ar
e
u
tili
ze
d
in
t
h
e
an
al
y
s
is
o
f
d
er
m
o
s
co
p
ic
im
ag
es,
as
well,
in
th
e
later
y
ea
r
s
[
15]
,
[
1
6
]
.
No
te
th
at
th
ese
f
ea
tu
r
es
ar
e
o
b
tain
ed
,
b
y
d
ef
au
lt,
f
r
o
m
th
e
en
tire
ly
co
n
s
is
ten
t la
y
er
s
o
f
th
e
C
NN
f
o
r
m
atio
n
.
T
h
e
m
ain
d
r
awb
ac
k
s
o
f
th
ese
d
ee
p
f
ea
tu
r
es a
r
e
s
u
s
ce
p
tib
ilit
y
o
r
s
en
s
itiv
ity
to
th
e
g
eo
m
etr
i
c
v
ar
ian
ts
an
d
th
e
s
ca
r
city
o
f
v
ar
ieties
o
f
lo
ca
l
p
atter
n
s
[
17
]
.
Ho
wev
e
r
,
in
t
h
e
ca
s
e
o
f
im
ag
es
h
av
in
g
im
p
r
ess
iv
e
d
if
f
er
en
ce
s
in
r
eso
lu
tio
n
an
d
p
er
s
p
ec
tiv
e,
it
co
u
ld
b
e
a
v
ast
d
if
f
icu
lty
to
d
o
an
aly
s
is
im
m
ed
iately
u
ti
lizin
g
C
NN
p
iece
s
.
T
h
e
g
en
er
ally
u
s
ed
s
o
lu
tio
n
s
ar
e
d
ata
au
g
m
en
tatio
n
an
d
r
e
-
s
ca
lin
g
,
s
u
ch
as
r
o
tate,
f
lip
,
o
r
cr
o
p
[
18
]
.
T
h
e
p
er
f
o
r
m
a
n
ce
m
ay
d
ec
r
ea
s
e
d
u
e
to
s
o
m
e
tr
an
s
f
o
r
m
atio
n
s
.
Fo
r
ex
am
p
le,
th
e
o
b
ject
m
ay
n
o
t w
ith
in
th
e
b
ac
k
g
r
o
u
n
d
a
r
ea
,
o
r
th
e
r
an
d
o
m
cr
o
p
im
a
g
es
m
ig
h
t
s
im
p
ly
p
ick
u
p
an
in
d
if
f
er
en
t
p
o
r
tio
n
o
f
th
e
g
ad
g
et
in
th
e
p
r
im
ar
y
p
ict
u
r
e.
T
h
er
ef
o
r
e
,
th
e
p
er
f
o
r
m
an
ce
e
n
h
an
ce
m
en
t
is
v
er
y
r
estricte
d
an
d
m
ay
b
e
wo
r
s
e
wh
en
em
p
l
o
y
in
g
C
NN
f
o
r
m
ed
i
ca
l p
u
r
p
o
s
es.
Alter
n
ativ
ely
,
n
u
m
er
o
u
s
r
esea
r
ch
es
ap
p
lied
a
c
o
m
b
in
atio
n
o
f
lo
ca
l
d
escr
ip
tio
n
en
co
d
in
g
with
d
ee
p
f
ea
tu
r
es
tech
n
iq
u
es
f
o
r
im
p
r
o
v
in
g
th
e
d
is
tin
ctio
n
ca
p
ab
ilit
y
o
f
s
u
ch
r
ep
r
esen
tatio
n
s
,
an
d
t
o
av
o
id
em
p
lo
y
in
g
d
ir
ec
t
C
NN
f
ea
tu
r
es
as
a
c
o
m
m
o
n
im
a
g
e
r
e
p
r
esen
tatio
n
[
1
9
]
,
[
2
0
]
.
T
h
ese
r
esear
ch
es
h
av
e
v
er
y
h
ig
h
i
n
ten
s
iv
e
co
m
p
u
tatio
n
b
ec
au
s
e
o
f
e
n
d
-
to
-
en
d
en
co
d
in
g
la
y
er
in
te
g
r
ate
d
with
tr
ain
in
g
a
C
NN
s
tr
u
ctu
r
e,
th
e
em
p
lo
y
m
e
n
t
o
f
p
o
o
lin
g
s
tr
ateg
y
b
ased
m
u
lt
ip
le
-
s
ca
le
p
y
r
am
id
to
co
m
p
o
s
e
FV r
ep
r
esen
tatio
n
s
[
21
]
,
o
r
th
e
s
lid
in
g
-
win
d
o
ws
ad
o
p
tio
n
t
o
cr
ea
te
wid
e
d
esc
r
ip
to
r
s
o
f
d
ef
in
ed
r
eg
io
n
s
in
th
e
p
r
im
ar
y
im
ag
es
[
17
]
.
I
n
a
later
s
tu
d
y
[
22
]
,
co
n
f
in
ed
p
iece
s
ar
e
cr
ea
ted
ar
b
itra
r
ily
o
f
a
d
er
m
o
s
co
p
ic
p
ic
tu
r
e
an
d
u
tili
ze
d
th
e
p
atch
F
V
ag
g
r
eg
ated
C
NN
f
ea
tu
r
es f
o
r
au
to
m
ated
m
ela
n
o
m
a
class
if
icatio
n
.
I
n
th
e
p
r
ev
io
u
s
liter
atu
r
e
[
2
3
]
,
ea
ch
ac
tiv
a
ted
C
NN
in
s
id
e
th
e
f
ea
tu
r
e
m
ap
,
b
e
ab
le
t
o
m
ap
o
u
t
b
ac
k
to
a
n
in
ter
esti
n
g
f
ield
(
a
s
p
ec
if
ied
ar
ea
)
o
f
th
e
in
p
u
t
im
ag
e
an
d
m
i
r
r
o
r
th
e
ch
ar
ac
ter
is
tics
o
f
th
e
s
p
ec
if
ied
ar
ea
.
As
a
r
esu
lt,
an
ex
tr
a
ef
f
ec
tiv
e
an
d
th
e
co
m
p
ac
ted
s
o
lu
tio
n
is
in
tr
o
d
u
ce
d
b
y
[
24
]
.
I
t
is
f
o
u
n
d
ed
o
n
c
o
m
p
ac
tly
c
o
llectin
g
lo
ca
l
d
escr
ip
to
r
s
f
r
o
m
a
c
o
n
v
o
lu
ti
o
n
al
C
NN.
Alth
o
u
g
h
s
ev
e
r
al
s
o
lu
tio
n
s
(
s
u
ch
as
d
ata
au
g
m
en
tatio
n
,
p
r
e
-
tr
ain
ed
m
o
d
els
o
f
im
a
g
eNe
t)
wer
e
p
r
o
p
o
s
ed
t
o
ad
d
r
ess
th
e
s
h
o
r
tag
e
o
f
tr
ain
i
n
g
d
ata
in
th
e
task
o
f
s
k
in
c
an
ce
r
class
if
icatio
n
,
s
till
th
es
e
s
o
lu
tio
n
s
ar
e
n
o
t
ef
f
ec
tiv
e.
T
h
er
e
f
o
r
e,
we
in
tr
o
d
u
ce
a
n
ew
s
tr
ateg
y
th
at
is
d
ep
en
d
en
t
o
n
en
h
a
n
cin
g
th
e
le
ar
n
ed
f
ea
tu
r
e
o
f
th
e
p
r
e
-
tr
ain
ed
m
o
d
els
b
y
tr
ai
n
in
g
th
em
o
n
a
lar
g
e
n
u
m
b
er
o
f
u
n
lab
elled
s
k
in
ca
n
ce
r
im
ag
es
t
h
en
a
s
m
all
n
u
m
b
er
o
f
lab
eled
s
k
in
ca
n
ce
r
im
ag
e
s
as
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
g
u
ar
a
n
te
es
th
at
th
e
m
o
d
els
lear
n
in
g
th
e
r
elev
an
t
f
ea
tu
r
es
an
d
r
ed
u
ce
th
e
an
n
o
tatio
n
p
r
o
ce
s
s
in
g
tim
e.
Mo
r
e
o
v
er
,
it
ca
n
b
e
u
s
ed
with
a
n
y
m
ed
ical
im
ag
in
g
task
.
Fig
u
r
e
1
.
T
h
e
o
v
er
all
wo
r
k
f
lo
w
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
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
B
ig
tr
a
n
s
fer lea
r
n
in
g
fo
r
a
u
to
ma
ted
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
(
Zin
a
h
Mo
h
s
in
A
r
ka
h
)
1613
2.
RE
L
AT
E
D
WO
RK
T
h
e
d
er
m
o
s
co
p
y
im
a
g
e
r
ec
o
g
n
itio
n
tech
n
iq
u
es a
r
e
ca
teg
o
r
iz
ed
in
to
two
m
ai
n
ca
teg
o
r
ies.
2
.
1
.
T
ec
hn
iqu
es ba
s
e
d o
n t
he
ha
nd
cr
a
f
t
ed
f
ea
t
ure
T
h
e
s
tan
d
ar
d
tech
n
iq
u
e
f
o
r
d
iag
n
o
s
in
g
co
l
o
r
ed
s
k
in
lesi
o
n
s
in
to
m
elan
o
m
a
an
d
b
en
ig
n
is
th
e
"ABC
D"
r
u
le
[
5
]
.
Sev
e
r
al
a
u
to
m
ated
d
ia
g
n
o
s
is
tech
n
iq
u
es
f
o
u
n
d
e
d
o
n
th
is
r
u
le
ar
e
p
r
o
p
o
s
ed
[
2
5
]
.
Fo
r
ex
am
p
le,
a
c
o
m
b
in
atio
n
o
f
KNN,
f
ea
tu
r
e
o
p
tim
izatio
n
f
r
am
ewo
r
k
,
an
d
h
an
d
-
d
esig
n
ed
f
ea
tu
r
es
(
co
l
o
r
d
escr
ip
to
r
s
,
b
o
r
d
er
-
g
r
a
d
ien
t,
a
n
d
s
h
ap
e
)
f
o
r
d
if
f
er
e
n
tiatio
n
b
etwe
en
b
en
ig
n
m
ela
n
o
m
a
lesi
o
n
s
is
ad
o
p
ted
b
y
,
Gan
s
ter
et
a
l.
[
2
6
]
.
A
s
im
ila
r
ap
p
r
o
ac
h
h
as
b
ee
n
in
tr
o
d
u
ce
d
b
y
C
eleb
i
et
a
l.
[
2
]
b
ased
o
n
ex
tr
ac
tin
g
a
s
eq
u
en
ce
o
f
ch
ar
ac
ter
is
tics
f
r
o
m
th
e
d
er
m
o
s
co
p
ic
p
ictu
r
e,
wh
ich
in
clu
d
es;
tex
tu
r
e
r
elate
d
d
escr
ip
to
r
s
,
co
lo
r
,
an
d
s
h
a
p
e
f
ea
t
u
r
es,
a
n
d
co
m
b
i
n
in
g
t
h
em
with
s
ev
er
al
alg
o
r
it
h
m
s
o
f
f
ea
tu
r
e
s
elec
tio
n
to
estab
lis
h
a
n
o
n
-
lin
e
ar
SVM
cla
s
s
if
ier
.
Oth
er
s
im
ilar
r
esear
ch
p
r
esen
ted
b
y
C
ap
d
e
h
o
u
r
at
et
a
l.
[
2
7
]
is
b
ased
o
n
ch
ar
ac
ter
izin
g
ea
ch
ap
p
lican
t
lesi
o
n
ar
ea
v
ia
d
es
cr
ip
to
r
s
et,
wh
ich
in
clu
d
es
t
ex
tu
r
e,
c
o
lo
r
,
a
n
d
s
h
a
p
e
in
f
o
r
m
atio
n
,
an
d
t
h
en
em
p
lo
y
in
g
th
is
in
f
o
r
m
atio
n
f
o
r
tr
ain
in
g
th
e
Ad
aBo
o
s
t
class
if
ier
.
Mo
r
e
o
v
er
,
Xie
et
a
l.
[
2
8
]
in
tr
o
d
u
ce
d
a
s
elf
-
g
en
er
ated
n
eu
r
al
m
o
d
el
f
o
r
g
en
er
atin
g
lesi
o
n
ar
ea
s
an
d
ass
em
b
ly
n
eu
r
al
n
etwo
r
k
p
r
o
to
ty
p
e
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
(
b
o
r
d
er
,
tex
tu
r
e
,
an
d
tu
m
o
r
c
o
lo
r
)
,
an
d
th
e
n
,
f
o
r
m
elan
o
m
a
r
ec
o
g
n
itio
n
.
B
i
et
a
l.
[
2
9
]
,
p
r
esen
ted
an
au
to
m
ated
m
elan
o
m
a
r
ec
o
g
n
itio
n
tech
n
iq
u
e,
b
y
u
tili
zin
g
jo
in
t
r
ev
er
s
e
class
if
icatio
n
an
d
m
u
ltip
le
s
ca
le
r
ep
r
esen
tatio
n
s
.
C
o
n
s
id
er
i
n
g
ad
d
itio
n
al
tec
h
n
iq
u
es
b
ased
o
n
o
b
tain
in
g
lo
ca
l
ch
ar
ac
te
r
is
tics
(
tex
tu
r
e
an
d
co
lo
r
.
)
f
r
o
m
p
etite
1
6
x
1
6
p
at
ch
es,
an
d
n
ex
t
co
llected
th
ese
p
atch
es
in
to
last
r
ep
r
esen
tat
io
n
s
u
s
in
g
th
e
B
o
F
m
o
d
el.
I
n
a
s
im
ilar
ap
p
r
o
ac
h
,
B
ar
ata
et
a
l.
[3
0
]
,
en
co
d
e
d
c
o
lo
r
an
d
tex
t
u
r
e
-
r
elate
d
f
ea
t
u
r
es
b
y
ap
p
ly
in
g
th
e
B
o
F m
o
d
el
f
o
r
lesi
o
n
r
ec
o
g
n
itio
n
.
2
.
2
.
T
ec
hn
iqu
es ba
s
e
d o
n d
ee
p CN
N
C
NN
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
m
u
lti
-
p
r
o
ce
s
s
lay
er
s
f
o
r
lear
n
in
g
v
ar
io
u
s
le
v
els
o
f
r
ep
r
es
en
tatio
n
s
.
T
h
er
ef
o
r
e,
co
n
n
ec
tin
g
th
ese
f
ea
tu
r
es
m
ain
tain
s
v
er
y
d
is
tin
g
u
is
h
ed
a
n
d
e
f
f
icien
t
d
ee
p
r
e
p
r
esen
tatio
n
s
[
31
]
-
[
33
]
.
I
n
g
en
er
al,
ap
p
ly
i
n
g
C
NN
f
o
r
d
e
r
m
o
s
co
p
y
im
ag
e
c
l
ass
if
icatio
n
ca
n
b
e
ca
teg
o
r
ized
in
two
way
s
.
T
h
e
f
ir
s
t
way
is
th
e
d
ir
ec
t
-
tr
ain
in
g
o
r
f
in
e
-
tu
n
in
g
ex
ten
s
iv
e
m
o
d
el
in
th
e
en
d
-
to
-
en
d
s
ty
le.
Fo
r
in
s
tan
ce
,
Dem
y
an
o
v
et
a
l.
[
34
]
p
r
o
p
o
s
ed
a
5
-
lay
er
C
NN
s
tr
u
ctu
r
e
f
o
r
class
if
y
in
g
s
k
in
lesi
o
n
d
ata
in
to
two
d
if
f
e
r
en
t
ty
p
es.
Mu
ltip
le
s
tag
in
g
s
y
s
tem
s
,
f
o
u
n
d
e
d
o
n
t
h
e
f
in
e
-
tu
n
in
g
ex
tr
em
ely
d
ee
p
r
em
ain
i
n
g
s
y
s
tem
f
o
r
au
to
m
atic
m
elan
o
m
a
class
if
icatio
n
in
d
er
m
o
s
co
p
ic
p
ictu
r
es,
ar
e
d
e
v
elo
p
ed
by
Yu
et
a
l.
[
35
]
.
I
n
v
er
y
r
ec
en
t
tim
es,
E
s
tev
a
et
a
l
.
[
36
]
a
p
p
lied
a
1
-
lay
er
ex
te
n
s
iv
e
m
o
d
el
f
o
r
a
u
t
o
m
ated
s
k
in
ca
n
ce
r
r
ec
o
g
n
itio
n
.
I
t
is
f
o
u
n
d
e
d
o
n
Go
o
g
leNe
t I
n
ce
p
tio
n
V3
ar
ch
i
tectu
r
e
an
d
u
tili
ze
d
1
2
9
4
5
0
p
i
ctu
r
es f
o
r
tr
ain
in
g
.
An
o
th
e
r
s
tu
d
y
b
y
Me
n
eg
o
la
e
t
al
,
[
16
]
ex
am
i
n
ed
th
e
in
f
lu
e
n
ce
o
f
in
f
o
r
m
atio
n
tr
an
s
f
er
e
n
ce
o
f
in
ten
s
e
lear
n
in
g
in
d
er
m
o
s
co
p
y
p
ictu
r
e
class
if
icatio
n
.
T
h
e
r
esear
ch
u
ti
lized
n
u
m
er
o
u
s
d
ata
s
ets,
s
u
ch
as I
m
ag
eNe
t,
R
etin
o
p
ath
y
,
I
S
I
C
,
an
d
Atlas.
T
h
e
s
ec
o
n
d
way
is
th
e
u
tili
za
t
io
n
o
f
d
ee
p
f
ea
tu
r
es e
x
tr
ac
ted
f
r
o
m
p
r
e
-
tr
ain
ed
C
NN
in
m
ed
ical
im
ag
e
r
ec
o
g
n
itio
n
,
r
ath
er
t
h
an
tr
ain
i
n
g
th
e
C
NN
b
y
ex
tr
em
ely
r
e
lian
t
o
n
ca
lcu
latin
g
r
eso
u
r
ce
s
an
d
lar
g
e
tr
ain
i
n
g
d
ata.
I
n
th
e
an
aly
s
is
f
ield
o
f
d
er
m
o
s
co
p
ic
im
ag
es,
C
o
d
ella
et
a
l
,
[
4
]
p
r
o
p
o
s
ed
a
p
r
e
-
tr
ai
n
e
d
C
N
N
(
I
m
ag
eNe
t)
f
o
r
d
is
cr
im
in
atin
g
h
ea
lth
y
an
d
m
elan
o
m
a
im
ag
es
b
y
ex
t
r
ac
tin
g
h
ig
h
-
lev
el
f
ea
t
u
r
e
r
e
p
r
ese
n
tatio
n
s
.
A
f
ea
tu
r
e
ex
tr
ac
to
r
,
b
ased
o
n
p
r
e
-
tr
ain
ed
C
NN
b
y
Kaw
ah
ar
a
et
a
l.
[
15
]
,
is
co
m
b
in
ed
with
f
ea
tu
r
es o
f
s
u
b
-
im
ag
e
p
o
o
lin
g
to
class
if
y
1
0
-
class
lesi
o
n
r
ec
o
g
n
itio
n
.
Oth
e
r
r
esear
ch
b
y
C
o
d
ella
et
a
l.
[
37
]
,
ex
am
in
e
d
th
e
tech
n
iq
u
e
ca
lle
d
"d
ee
p
lear
n
in
g
en
s
em
b
le"
f
o
r
m
elan
o
m
a
class
if
icatio
n
.
Yu
et
a
l
.
[
38
]
in
tr
o
d
u
ce
d
a
n
au
t
o
m
ated
m
elan
o
m
a
r
ec
o
g
n
itio
n
,
v
ia
co
llectin
g
th
e
C
NN
ac
tiv
atio
n
s
o
f
th
e
r
an
d
o
m
ly
s
elec
ted
s
u
b
-
im
ag
es
f
r
o
m
a
d
er
m
o
s
co
p
ic
p
ictu
r
e.
T
h
e
m
ain
is
s
u
e
with
th
ese
m
eth
o
d
s
is
th
e
s
h
o
r
tag
e
o
f
tr
ain
in
g
im
ag
es
[
1
]
,
[
7
]
.
An
o
th
er
is
s
u
e
is
th
e
lo
w
co
m
p
u
tatio
n
al
to
o
ls
.
T
o
ac
h
iev
e
b
etter
r
esu
lts
,
m
o
d
els
s
h
o
u
ld
b
e
tr
ain
ed
o
n
eith
er
GPU
[
3
9
]
o
r
FP
GA
[
4
0
]
,
[
4
1
]
.
I
n
th
is
a
r
ticle,
we
i
n
tr
o
d
u
ce
a
n
o
v
el
s
tr
ateg
y
to
a
d
d
r
ess
th
e
s
h
o
r
tag
e
o
f
tr
ain
in
g
is
s
u
es
an
d
r
ed
u
ce
th
e
an
n
o
tatio
n
p
r
o
ce
s
s
tim
e
an
d
we
tr
ain
o
n
GPU
.
3.
RE
S
E
ARCH
M
E
T
H
O
D
3
.
1
.
D
a
t
a
s
et
I
n
th
is
a
r
ticle,
we
h
a
v
e
u
tili
ze
d
two
d
atasets
.
T
h
e
f
ir
s
t
o
n
e
is
I
SIC
Ar
ch
iv
e
(
s
o
u
r
ce
d
ata
s
et)
wh
ich
co
n
tain
s
2
3
,
9
0
6
d
er
m
o
s
co
p
ic
im
ag
es
[
42
]
.
T
h
is
d
ataset
is
u
s
ed
to
tr
ai
n
th
e
p
r
e
-
tr
ain
e
d
m
o
d
e
ls
in
th
e
m
id
d
le
s
tag
e
wh
er
e
h
u
g
e
p
len
ty
o
f
u
n
lab
eled
s
k
in
ca
n
ce
r
p
ictu
r
es
ar
e
u
s
ed
.
T
h
e
s
ec
o
n
d
d
ataset
is
SII
M
-
I
SIC
2
0
2
0
d
ataset
[
43
]
(
tar
g
et
d
ataset)
.
I
t
co
n
tain
s
3
3
,
0
0
0
s
k
in
lesi
o
n
p
ictu
r
es
d
iv
id
ed
in
to
two
ca
te
g
o
r
ies:
B
en
ig
n
an
d
m
alig
n
an
t.
I
t
co
n
s
is
ts
o
f
o
n
ly
5
8
4
im
ag
es
o
f
t
h
e
m
alig
n
an
t
class
an
d
th
e
r
est
f
o
r
b
en
ig
n
.
T
o
h
av
e
an
e
q
u
al
n
u
m
b
er
o
f
im
a
g
es,
we
to
o
k
o
n
ly
5
8
4
i
m
ag
es f
r
o
m
th
e
b
e
n
ig
n
class
.
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.
23
,
No
.
3
,
Sep
tem
b
er
2
0
2
1
:
1
6
1
1
-
1
6
1
9
1614
3
.
2
.
T
he
s
t
a
t
e
-
of
-
t
he
-
a
rt
a
rc
hite
ct
ures
Nu
m
er
o
u
s
C
NN
ar
c
h
itectu
r
es,
in
th
e
ar
ea
o
f
n
atu
r
al
im
ag
e
r
ec
o
g
n
itio
n
,
ar
e
r
elea
s
ed
in
r
ec
en
t
y
ea
r
s
,
s
u
ch
as
VGGN
et
[
44
]
,
Go
o
g
L
eNe
t
[
45
]
,
a
n
d
R
esNet
[
46
]
.
A
n
u
m
b
er
o
f
th
ese
C
NNs,
lik
e
R
es
Net
an
d
Go
o
g
L
eNe
t,
ar
e
o
f
f
er
ed
as
p
r
e
-
tr
ain
ed
s
tr
u
ctu
r
es.
T
h
ey
ar
e
tr
ain
ed
o
n
a
r
o
u
n
d
1
.
2
8
m
illi
o
n
g
e
n
etic
p
ictu
r
es
f
r
o
m
th
e
I
m
a
g
eNe
t
d
atab
ase
[
47
]
.
Hen
ce
,
t
h
e
ab
ilit
y
to
u
tili
ze
b
iases
an
d
weig
h
ts
f
o
r
th
ese
m
o
d
els,
i.e
.
,
f
in
e
-
tu
n
in
g
all
th
e
m
o
d
el
lay
e
r
s
t
h
r
o
u
g
h
co
n
tin
u
in
g
with
th
e
b
ac
k
p
r
o
p
ag
atio
n
an
d
em
p
lo
y
in
g
th
e
d
ata,
s
o
,
t
h
e
ab
ilit
y
to
ap
p
ly
th
em
to
th
e
s
p
ec
if
ic
r
ec
o
g
n
itio
n
tas
k
,
as
well.
W
h
ile
th
e
VGG
Net
i
s
in
i
tialized
s
o
th
at
th
e
b
iases
an
d
weig
h
ts
ar
e
n
o
t a
f
f
ec
ted
b
y
th
e
v
is
u
al
d
ata
(
m
i
g
h
t v
ar
y
f
r
o
m
s
k
in
i
m
ag
es).
T
h
e
f
o
llo
win
g
is
a
s
h
o
r
t o
u
tlin
e
o
f
th
e
g
en
er
ally
u
s
ed
C
NN
ar
ch
itectu
r
es:
A.
VGGN
et:
A
d
ee
p
C
NN
was
d
ev
elo
p
ed
b
y
K.
S
im
o
n
y
an
,
an
d
Z
is
s
er
m
an
A
,
[
44
]
in
2
0
1
5
.
I
t
co
n
s
is
ts
o
f
ex
tr
em
ely
s
m
all
co
n
v
o
l
u
tio
n
a
l
f
ilter
s
an
d
its
d
ep
th
is
in
th
e
r
an
g
e
o
f
1
6
-
1
9
lay
er
s
.
I
t
u
tili
ze
d
1
3
C
o
n
v
lay
er
s
with
a
3
×3
f
ilter
s
ize.
Fiv
e
m
ax
-
p
o
o
lin
g
lay
er
s
ar
e
em
p
lo
y
ed
to
ca
r
r
y
o
u
t
th
e
s
p
atial
p
o
o
lin
g
.
T
h
en
,
s
ev
er
al
c
o
n
v
o
lu
tio
n
al
l
ay
er
s
an
d
m
ax
-
p
o
o
lin
g
is
ac
h
iev
ed
o
n
a
m
ask
o
f
s
ize
2
×2
p
ix
els.
T
h
r
ee
f
u
lly
co
n
n
ec
ted
lay
e
r
s
ar
e
f
o
ll
o
win
g
a
s
tack
o
f
co
n
v
o
lu
tio
n
a
l la
y
er
s
.
B.
Go
o
g
L
eNe
t:
T
h
is
ty
p
e
o
f
C
NN
is
p
r
esen
ted
b
y
Szeg
e
d
y
et
a
l.
[
45
]
in
2
0
1
5
.
I
t
co
n
s
is
ts
o
f
twen
ty
-
tw
o
co
n
v
o
l
u
tio
n
al
lay
er
s
in
clu
d
in
g
9
I
n
ce
p
tio
n
b
lo
c
k
s
.
E
v
er
y
I
n
c
ep
tio
n
b
lo
ck
h
as 3
u
n
iq
u
e
f
ilter
d
im
en
s
io
n
s
,
wh
ich
h
o
ld
1
×1
,
3
×3
,
an
d
5
×5
f
o
r
co
n
v
o
lu
tio
n
al,
as
w
ell
as,
3
×3
f
o
r
p
o
o
lin
g
.
Usi
n
g
th
e
g
iv
e
n
p
ar
am
eter
s
with
th
e
R
GB
co
lo
r
s
p
ac
e
,
th
e
r
ec
ep
tiv
e
f
ield
s
iz
e
is
2
2
4
×2
2
4
×3
.
I
n
a
c
o
m
p
ar
a
b
le
way
to
th
e
d
if
f
er
en
t
C
NNs,
th
e
p
r
ep
a
r
atio
n
s
tep
o
f
th
e
co
n
v
o
lu
tio
n
al
k
er
n
el
r
ec
o
r
d
s
is
d
e
p
en
d
e
n
t
o
n
th
e
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
tech
n
iq
u
e
(
SGD)
.
T
h
e
Go
o
g
L
eNe
t
ex
tr
ac
ts
o
f
f
lin
e
th
e
h
ig
h
-
lev
el
f
ea
t
u
r
es
o
f
th
e
d
is
s
im
ilar
class
e
s
th
r
o
u
g
h
th
e
tr
ain
in
g
p
h
ase
at
th
e
tim
e
-
d
em
an
d
in
g
,
w
h
ich
n
ee
d
s
a
v
ast
n
u
m
b
er
o
f
tr
ain
in
g
im
ag
es,
as we
ll.
C.
T
h
e
r
esid
u
al
n
etwo
r
k
(
R
esNet)
:
T
h
e
C
NN
m
o
d
el
with
its
d
ee
p
h
ier
ar
ch
ical
ar
ch
itectu
r
e
h
as
cr
itical
s
ig
n
if
ican
ce
d
u
e
to
it
s
ef
f
ec
tiv
e
lear
n
in
g
a
b
ilit
y
.
He
et
a
l.
[
46
]
,
p
r
esen
ted
a
d
ee
p
r
esid
u
al
n
eu
r
al
n
etwo
r
k
(
R
esNet)
as
a
n
ew
C
NN
g
en
er
atio
n
.
I
n
t
h
e
I
L
SVR
C
ch
all
en
g
e
2
0
1
6
f
o
r
f
ea
tu
r
e
ex
tr
ac
t
io
n
,
wh
ich
is
d
ed
icate
d
to
I
m
ag
eNe
t la
r
g
e
-
s
ca
le
v
is
u
al
r
ec
o
g
n
itio
n
,
its
r
an
k
was n
u
m
b
er
o
n
e
.
T
h
e
k
ey
c
h
ar
ac
ter
is
tic
o
f
R
esNet
i
s
ly
in
g
in
th
e
ad
d
r
es
s
in
g
ca
p
ab
ilit
y
o
f
th
e
d
eg
r
a
d
atio
n
p
r
o
b
lem
wh
ile
tr
ain
in
g
an
ex
tr
em
ely
d
ee
p
n
etwo
r
k
(
i.e
.
,
th
e
ad
ap
tatio
n
o
f
r
esid
u
al
c
o
n
n
ec
tio
n
)
,
as
co
m
p
ar
e
d
with
o
th
er
c
lass
ical
C
NN
ar
ch
itectu
r
es.
As
d
em
o
n
s
tr
a
te
d
ea
r
lier
,
th
e
r
esid
u
al
lin
k
s
ca
n
p
r
eser
v
e
th
e
ac
h
iev
ed
ac
c
u
r
ac
y
g
ain
s
,
as
well
as,
ac
ce
ler
ate
th
e
d
ee
p
n
etwo
r
k
c
o
n
v
er
g
en
ce
b
y
c
o
n
s
id
er
ab
ly
en
lar
g
in
g
t
h
e
n
etw
o
r
k
d
ep
th
.
I
n
g
en
er
al,
w
h
en
l
o
o
k
in
g
in
s
id
e
th
e
d
ee
p
r
esid
u
al
n
etwo
r
k
,
i
t
in
v
o
lv
es
a
g
r
o
u
p
o
f
r
es
id
u
a
l
b
lo
ck
s
,
with
s
ev
er
al
s
tack
ed
c
o
n
v
o
lu
tio
n
al
lay
er
s
in
s
id
e
ea
ch
b
lo
ck
.
No
te
th
at
th
e
ap
p
e
n
d
ag
e
o
f
th
e
co
n
v
o
l
u
tio
n
al
lay
er
is
r
eg
ar
d
ed
as
th
e
b
atc
h
n
o
r
m
aliza
tio
n
lay
er
s
p
lu
s
t
h
e
r
ec
tifie
d
lin
ea
r
u
n
it
lay
er
.
T
h
e
f
o
r
m
u
la
o
f
th
e
r
esid
u
al
b
ase
with
co
n
n
ec
ti
o
n
s
m
ap
p
in
g
ca
n
b
e
co
m
p
o
s
ed
as
in
(
1
)
,
ℎ
+
1
=
(
ℎ
+
(
ℎ
,
)
)
(
1
)
w
h
er
e:
ℎ
+
1
is
th
e
n
-
th
r
esid
u
al
b
a
s
e
o
u
tp
u
t,
R
elu
(
)
is
th
e
r
ec
tifi
ed
lin
ea
r
u
n
it
f
u
n
ctio
n
,
ℎ
is
th
e
n
-
th
r
esid
u
al
b
ase
in
p
u
t,
F
is
th
e
r
esid
u
al
m
ap
p
in
g
f
u
n
ctio
n
,
an
d
is
t
h
e
b
lo
ck
p
ar
am
eter
s
.
also
n
o
te
th
at
a
lin
ea
r
p
r
o
jectio
n
φ
is
n
o
r
m
all
y
u
s
ed
f
o
r
m
atch
in
g
th
e
d
im
e
n
s
io
n
s
wh
en
ℎ
an
d
(
ℎ
,
)
ar
e
u
n
eq
u
al.
Nex
t,
th
e
(
1
)
ca
n
b
e
ex
tr
a
m
o
d
i
f
ied
as in
(
2
)
,
ℎ
+
1
=
(
(
ℎ
)
+
(
ℎ
,
)
)
(
2
)
H
o
wev
er
,
R
esNet
-
1
0
1
an
d
R
esNet
-
5
0
ar
e
two
d
if
f
er
en
t
d
ep
t
h
s
o
f
R
esNet
m
o
d
els.
R
esNet
-
5
0
was
u
tili
ze
d
in
th
is
s
tu
d
y
wh
ich
p
r
e
-
tr
ain
ed
o
n
I
m
ag
eNe
t.
3
.
3
.
T
ra
ini
ng
pro
ce
du
re
a
nd
s
ce
na
rio
s
W
e
h
av
e
tr
ain
ed
th
e
p
r
e
-
tr
ai
n
e
d
n
etwo
r
k
s
in
two
d
if
f
e
r
en
t scen
ar
io
s
:
1.
Scen
ar
io
1
: star
t b
y
a
d
ju
s
tin
g
t
h
e
p
r
e
-
tr
ai
n
ed
n
etwo
r
k
s
th
en
t
r
ain
o
n
t
h
e
d
esti
n
atio
n
d
ataset.
2.
Scen
ar
io
2
:
s
tar
t
b
y
ad
ju
s
tin
g
th
e
p
r
e
-
tr
ain
ed
n
etwo
r
k
s
th
en
tr
ain
o
n
th
e
s
o
u
r
ce
d
ataset
as
a
f
ir
s
t
s
tep
.
Fig
u
r
e
2
illu
s
tr
ates
th
e
tr
ain
i
n
g
p
r
o
ce
s
s
.
T
h
en
ad
ju
s
tin
g
r
e
s
u
lted
f
r
o
m
m
o
d
els
f
r
o
m
th
e
f
ir
s
t
s
tep
an
d
tr
ain
o
n
th
e
d
esti
n
atio
n
d
ataset
.
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
B
ig
tr
a
n
s
fer lea
r
n
in
g
fo
r
a
u
to
ma
ted
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
(
Zin
a
h
Mo
h
s
in
A
r
ka
h
)
1615
Fig
u
r
e
2
.
T
r
ain
in
g
p
r
o
ce
s
s
o
f
t
h
e
f
ir
s
t step
o
f
s
ce
n
ar
io
2
W
e
h
av
e
tr
ain
ed
th
e
m
o
d
el
f
o
r
1
4
0
0
0
iter
atio
n
s
.
So
m
e
well
-
in
f
o
r
m
ed
k
er
n
els
f
r
o
m
th
e
in
itial
co
n
v
o
l
u
tio
n
lay
er
s
o
f
R
esNet
5
0
h
av
e
b
ee
n
s
h
o
w
n
in
Fig
u
r
e
3
.
T
h
e
wo
r
k
f
lo
w
o
f
th
is
s
ce
n
ar
io
is
ex
p
lain
ed
i
n
Fig
u
r
e
4
.
Fig
u
r
e
3
.
So
m
e
lear
n
ed
k
er
n
el
s
f
r
o
m
th
e
in
itial
co
n
v
o
lu
ti
o
n
l
ay
er
s
o
f
R
esNet5
0
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.
23
,
No
.
3
,
Sep
tem
b
er
2
0
2
1
:
1
6
1
1
-
1
6
1
9
1616
Fig
u
r
e
4
.
T
h
e
wo
r
k
f
lo
w
o
f
s
ce
n
ar
io
2
4.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S
I
n
th
is
p
ar
t,
we
h
av
e
e
v
alu
at
ed
th
e
u
s
ed
m
o
d
els
o
n
th
e
t
esti
n
g
s
et
in
s
ev
er
al
ev
alu
atio
n
m
etr
ics
in
clu
d
in
g
ac
c
u
r
ac
y
(
3
)
,
r
ec
all
(
4
)
,
p
r
ec
is
io
n
(
5
)
,
F1
-
s
co
r
e
(
6
)
[
4
8
]
-
[
5
2
]
.
Acc
u
r
ac
y
=
(
T
P+T
N)
/(
T
P+T
N+
FP
+FN)
(
3
)
R
ec
all
=
T
P/(T
P+FN)
(
4
)
Pre
cisi
o
n
=
T
P/(T
P+
FP)
(
5
)
F1
s
c
or
e
=
2
×
(
(
Pre
cisi
o
n
×Reca
ll)/(
Pre
cisi
o
n
+Reca
ll)
(
6
)
W
e
s
tar
ted
to
ev
alu
ate
th
e
p
r
e
-
tr
ain
ed
m
o
d
els
(
VGG1
9
,
Go
o
g
leNe
t,
R
esNet
-
5
0
)
with
s
c
en
ar
io
1
as
r
ep
o
r
ted
in
T
ab
le
1
.
R
esNet
5
0
is
th
e
d
o
m
in
a
n
t
m
o
d
el
am
o
n
g
th
e
o
th
e
r
s
b
y
o
b
tain
i
n
g
t
h
e
h
ig
h
est
ac
c
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
s
co
r
e
o
f
8
4
.
0
%,
8
6
.
6
%,
8
9
.
3
%,
8
7
.
9
%
r
esp
ec
tiv
ely
.
Go
o
g
leNe
t
o
b
tain
ed
s
ec
o
n
d
p
lace
b
y
s
co
r
in
g
8
1
.
2
%
f
o
r
ac
cu
r
ac
y
,
8
4
.
4
%
f
o
r
p
r
ec
is
io
n
,
8
7
.
5
%
f
o
r
r
ec
all,
an
d
8
5
.
9
f
o
r
th
e
F1
s
co
r
e.
L
astl
y
,
th
e
VGG
m
o
d
el
ac
h
i
ev
ed
th
e
lo
west m
ea
s
u
r
em
en
ts
a
m
o
n
g
o
th
e
r
m
o
d
els b
y
o
b
tain
i
n
g
7
9
.
6
%,
8
3
.
7
%,
8
5
.
4
%,
8
4
.
6
% f
o
r
th
e
ac
cu
r
a
cy
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
s
co
r
e,
r
esp
ec
tiv
ely
.
W
e
th
en
ev
alu
ated
th
e
p
r
e
-
tr
ain
ed
m
o
d
els
(
VGG1
9
,
G
o
o
g
leNe
t,
R
esNet
-
5
0
)
with
s
ce
n
ar
io
2
as
d
etailed
in
T
ab
le
2
.
T
h
e
r
esu
lts
o
f
all
m
o
d
els
h
av
e
b
ee
n
s
ig
n
if
ican
tly
im
p
r
o
v
ed
.
Ho
we
v
er
,
th
e
R
esNet
5
0
m
o
d
el
is
s
till
th
e
d
o
m
in
an
t
m
o
d
el
am
o
n
g
th
e
o
th
e
r
s
b
y
o
b
t
ain
in
g
t
h
e
h
i
g
h
est
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
s
co
r
e
o
f
9
3
.
7
%,
9
5
.
7
%,
9
4
.
6
%,
9
5
.
1
%
r
esp
ec
tiv
ely
.
G
o
o
g
leNe
t
ea
r
n
ed
s
ec
o
n
d
p
lace
b
y
o
b
tain
in
g
8
8
.
8
%
f
o
r
ac
c
u
r
ac
y
,
8
9
.
8
%
f
o
r
p
r
ec
is
io
n
,
9
2
.
8
%
f
o
r
r
ec
all,
an
d
9
1
.
3
f
o
r
th
e
F1
s
co
r
e.
L
astl
y
,
th
e
VGG
m
o
d
e
l
attain
ed
th
e
lo
west
m
ea
s
u
r
em
en
ts
am
o
n
g
o
th
er
m
o
d
els
b
y
s
co
r
in
g
8
5
.
1
%,
8
7
.
1
%,
8
9
.
7
%,
8
8
.
4
%
f
o
r
th
e
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
s
co
r
e,
r
esp
ec
tiv
el
y
.
O
u
r
o
u
tco
m
es
p
r
o
v
e
d
th
at
t
h
e
aim
ed
s
o
lu
tio
n
in
s
ce
n
ar
io
2
is
v
er
y
p
o
wer
f
u
l in
h
an
d
lin
g
th
e
s
h
o
r
t
a
g
e
o
f
t
r
ain
i
n
g
d
ata
f
o
r
s
k
in
ca
n
ce
r
class
if
icatio
n
task
s
.
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
B
ig
tr
a
n
s
fer lea
r
n
in
g
fo
r
a
u
to
ma
ted
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
(
Zin
a
h
Mo
h
s
in
A
r
ka
h
)
1617
T
ab
le
1
.
T
h
e
ev
a
lu
atio
n
r
esu
lt
s
f
r
o
m
s
ce
n
ar
io
1
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F
1
sc
o
r
e
(
%)
V
G
G
1
9
7
9
.
6
8
3
.
7
8
5
.
4
8
4
.
6
G
o
o
g
l
e
N
e
t
8
1
.
2
8
4
.
4
8
7
.
5
8
5
.
9
R
e
sN
e
t
-
50
8
4
.
0
8
6
.
6
8
9
.
3
8
7
.
9
T
ab
le
2
.
T
h
e
ev
a
lu
atio
n
r
esu
lt
s
f
r
o
m
s
ce
n
ar
io
2
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F
1
sc
o
r
e
(
%)
V
G
G
1
9
8
5
.
1
8
7
.
1
8
9
.
7
8
8
.
4
G
o
o
g
l
e
N
e
t
8
8
.
8
8
9
.
8
9
2
.
8
9
1
.
3
R
e
sN
e
t
-
50
9
3
.
7
9
5
.
7
9
4
.
6
9
5
.
1
5.
CO
NCLU
SI
O
N
I
n
th
is
ar
ticle,
we
p
r
esen
ted
a
n
o
v
el
s
tr
ateg
y
o
f
tr
a
n
s
f
er
lea
r
n
in
g
to
tack
le
th
e
is
s
u
e
o
f
s
h
o
r
tag
e
o
f
tr
ain
in
g
d
ata
in
s
k
in
ca
n
ce
r
cl
ass
if
icatio
n
task
s
b
y
tu
r
n
in
g
t
h
e
lear
n
e
d
f
ea
tu
r
es
o
f
p
r
e
-
tr
ai
n
ed
m
o
d
els
o
f
th
e
I
m
ag
eNe
t.
W
e
tr
ain
ed
th
e
m
o
d
els
o
n
a
la
r
g
e
n
u
m
b
e
r
o
f
u
n
lab
elled
s
k
in
ca
n
ce
r
im
ag
es.
W
e
th
en
tr
ain
th
em
o
n
a
s
m
all
n
u
m
b
er
o
f
la
b
eled
s
k
i
n
.
T
h
is
a
p
p
r
o
ac
h
g
u
ar
a
n
teed
t
h
at
th
e
m
o
d
els
lear
n
e
d
th
e
r
el
ev
an
t
f
ea
tu
r
es
o
f
th
e
s
k
in
ca
n
ce
r
class
if
icatio
n
task
.
T
h
e
r
esu
lts
d
em
o
n
s
tr
ated
th
at
th
e
aim
e
d
m
eth
o
d
is
b
en
ef
icia
l b
y
p
er
f
o
r
m
in
g
a
n
ac
cu
r
ac
y
o
f
8
4
%
with
R
esNe
t5
0
wh
en
d
ir
ec
tly
tr
ai
n
ed
with
a
s
m
all
n
u
m
b
er
o
f
lab
ele
d
s
k
in
an
d
9
3
.
7
%
wh
en
tr
ain
ed
with
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
.
T
h
is
ap
p
r
o
ac
h
is
s
u
itab
le
f
o
r
an
y
m
e
d
ical
im
ag
in
g
task
th
at
h
as
th
e
is
s
u
e
of
p
r
o
v
id
in
g
s
u
f
f
icien
t
lab
ele
d
im
ag
es.
W
e
in
ten
d
to
a
p
p
l
y
th
e
aim
ed
s
tr
ateg
y
f
o
r
o
th
e
r
m
ed
ical
im
ag
in
g
ap
p
licatio
n
s
.
W
e
also
aim
to
u
s
e
n
ew
ty
p
e
o
f
t
r
an
s
f
er
lear
n
in
g
ca
lled
s
am
e
-
d
o
m
ain
tr
an
s
f
er
lear
n
in
g
.
RE
F
E
R
E
NC
E
S
[1
]
L.
Alz
u
b
a
id
i
e
t
a
l.
,
“
No
v
e
l
Tr
a
n
sfe
r
Lea
rn
in
g
Ap
p
ro
a
c
h
fo
r
M
e
d
ica
l
Im
a
g
in
g
wit
h
Li
m
it
e
d
Lab
e
led
Da
ta,”
Ca
n
c
e
rs
,
v
o
l.
1
3
,
n
o
.
7
,
1
5
9
0
,
2
0
2
1
,
d
o
i:
1
0
.
3
3
9
0
/ca
n
c
e
rs1
3
0
7
1
5
9
0
.
[2
]
K.
H.
M
.
Ce
leb
i,
B
.
Ud
d
in
,
H.
I
y
a
to
m
i,
Y.
As
lan
d
o
g
a
n
,
W.
S
t
o
e
c
k
e
r
a
n
d
R.
M
o
ss
,
“
A
m
e
th
o
d
o
lo
g
i
c
a
l
a
p
p
ro
a
c
h
t
o
th
e
c
las
sifica
ti
o
n
o
f
d
e
rm
o
sc
o
p
y
ima
g
e
s,”
Co
mp
u
t.
M
e
d
.
Im
a
g
.
Gr
a
p
.
,
v
o
l.
3
1
,
n
o
.
6
,
p
p
.
3
6
2
-
3
7
3
,
2
0
0
7
,
d
o
i:
1
0
.
1
0
1
6
/
j.
c
o
m
p
m
e
d
ima
g
.
2
0
0
7
.
0
1
.
0
0
3
.
[3
]
A.
R.
A.
Al
i
a
n
d
T.
M
.
De
se
rn
o
,
“
A
sy
ste
m
a
ti
c
re
v
iew
o
f
a
u
t
o
m
a
ted
m
e
lan
o
m
a
d
e
tec
ti
o
n
i
n
d
e
rm
a
t
o
sc
o
p
ic
ima
g
e
s
a
n
d
i
t'
s
g
ro
u
n
d
tru
t
h
d
a
ta,”
Pro
c
.
S
PIE
M
e
d
.
Ima
g
.
,
v
o
l.
8
3
1
8
,
p
p
.
8
3
1
8
1
I
-
1
-
8
3
1
8
1
I
-
1
1
,
2
0
1
2
,
d
o
i:
1
0
.
1
1
1
7
/
1
2
.
9
1
2
3
8
9
.
[4
]
N.
Co
d
e
ll
a
,
J.
Ca
i,
M
.
Ab
e
d
i
n
i,
R
.
G
a
rn
a
v
i
,
A.
Ha
lp
e
rn
a
n
d
J.
R.
S
m
it
h
,
“
De
e
p
Lea
rn
in
g
,
S
p
a
rse
Co
d
in
g
,
a
n
d
S
V
M
fo
r
M
e
lan
o
m
a
Re
c
o
g
n
it
i
o
n
in
D
e
rm
o
sc
o
p
y
Im
a
g
e
s,”
p
re
se
n
ted
a
t
th
e
Pro
c
.
M
e
d
.
Ima
g
.
Co
m
p
u
t
.
Co
mp
u
t
.
Assist.
In
ter
v
.
,
2
0
1
5
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
319
-
2
4
8
8
8
-
2
_
1
5
.
[5
]
D.
G
u
tma
n
e
t
a
l.
,
(2
0
1
6
,
S
k
i
n
Les
io
n
A
n
a
ly
sis
t
o
wa
rd
M
e
lan
o
m
a
De
tec
ti
o
n
:
“
A
Ch
a
ll
e
n
g
e
a
t
th
e
In
ter
n
a
ti
o
n
a
l
S
y
m
p
o
si
u
m
o
n
Bi
o
m
e
d
ica
l
Im
a
g
i
n
g
(I
S
BI)
2
0
1
6
”
,
h
o
ste
d
b
y
th
e
I
n
tern
a
ti
o
n
a
l
S
k
i
n
Im
a
g
i
n
g
Co
ll
a
b
o
ra
ti
o
n
(I
S
IC).
[6
]
L.
Alz
u
b
a
id
i,
O.
Al
-
S
h
a
m
m
a
,
M
.
A.
F
a
d
h
e
l,
L.
F
a
rh
a
n
,
J.
Zh
a
n
g
a
n
d
Y.
Du
a
n
,
“
Op
ti
m
izin
g
t
h
e
p
e
rfo
rm
a
n
c
e
o
f
b
re
a
st
c
a
n
c
e
r
c
las
sifica
ti
o
n
b
y
e
m
p
lo
y
in
g
t
h
e
sa
m
e
d
o
m
a
in
tran
sfe
r
lea
rn
in
g
fro
m
h
y
b
ri
d
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
m
o
d
e
l,
”
E
lec
tro
n
ics
,
v
o
l
.
9
,
n
o
.
3
,
p
.
4
4
5
,
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/ele
c
tro
n
ics
9
0
3
0
4
4
5
.
[7
]
L.
Alz
u
b
a
i
d
i,
M
.
A.
F
a
d
h
e
l
,
O.
Al
-
S
h
a
m
m
a
,
J.
Zh
a
n
g
,
J.
S
a
n
tam
a
ría
,
Y.
Du
a
n
,
a
n
d
S
.
R.
Ole
iwi
,
“
To
wa
rd
s
a
b
e
tt
e
r
u
n
d
e
rsta
n
d
i
n
g
o
f
tran
sfe
r
lea
rn
i
n
g
fo
r
m
e
d
ica
l
ima
g
i
n
g
:
a
c
a
se
st
u
d
y
,
”
Ap
p
li
e
d
S
c
ien
c
e
s
,
v
o
l.
1
0
,
n
o
.
1
3
,
p
.
4
5
2
3
,
2
0
2
0
,
d
o
i:
1
0
.
3
3
9
0
/a
p
p
1
0
1
3
4
5
2
3
.
[8
]
R.
I.
Ha
sa
n
,
S
.
M
.
Yu
su
f
,
a
n
d
L.
Alz
u
b
a
i
d
i
,
“
Re
v
iew
o
f
t
h
e
sta
te
o
f
t
h
e
a
rt
o
f
d
e
e
p
lea
rn
i
n
g
f
o
r
p
l
a
n
t
d
ise
a
se
s:
A
b
ro
a
d
a
n
a
ly
sis a
n
d
d
isc
u
ss
io
n
,
”
P
la
n
ts
,
v
o
l.
9
,
n
o
.
1
0
,
p
.
1
3
0
2
,
2
0
2
0
,
d
o
i:
1
0
.
3
3
9
0
/p
lan
ts9
1
0
1
3
0
2
.
[9
]
L.
Alz
u
b
a
i
d
i,
M
.
A.
F
a
d
h
e
l,
O.
Al
-
S
h
a
m
m
a
,
J.
Zh
a
n
g
a
n
d
Y.
Du
a
n
,
“
De
e
p
lea
rn
in
g
m
o
d
e
ls
fo
r
c
las
s
ifi
c
a
ti
o
n
o
f
re
d
b
lo
o
d
c
e
ll
s
in
m
icro
sc
o
p
y
ima
g
e
s
to
a
id
i
n
sic
k
le
c
e
ll
a
n
e
m
ia
d
iag
n
o
sis
,
”
E
lec
tro
n
ics
,
v
o
l.
9
,
n
o
.
3
,
p
.
4
2
7
,
2
0
2
0
,
d
o
i:
1
0
.
3
3
9
0
/ele
c
tro
n
ics
9
0
3
0
4
2
7
.
[1
0
]
L.
Alz
u
b
a
i
d
i,
M
.
A.
F
a
d
h
e
l,
O.
Al
-
S
h
a
m
m
a
,
J.
Zh
a
n
g
,
J.
S
a
n
tam
a
ría,
a
n
d
Y.
Du
a
n
,
"
R
o
b
u
st ap
p
li
c
a
ti
o
n
o
f
n
e
w d
e
e
p
lea
rn
in
g
t
o
o
ls:
a
n
e
x
p
e
rime
n
tal
st
u
d
y
in
m
e
d
ica
l
ima
g
in
g
,
”
M
u
l
ti
m
e
d
ia
T
o
o
ls
a
n
d
Ap
p
li
c
a
t
io
n
s
,
p
p
.
1
-
2
9
,
2
0
2
1
,
d
o
i:
1
0
.
1
0
0
7
/s
1
1
0
4
2
-
0
2
1
-
1
0
9
4
2
-
9
.
[1
1
]
L.
Alz
u
b
a
id
i
,
e
t
a
l
.
,
“
Re
v
iew
o
f
d
e
e
p
lea
rn
i
n
g
:
c
o
n
c
e
p
ts,
CNN
a
rc
h
it
e
c
tu
re
s,
c
h
a
ll
e
n
g
e
s,
a
p
p
li
c
a
ti
o
n
s,
f
u
t
u
re
d
irec
ti
o
n
s,”
J
o
u
rn
a
l
o
f
B
ig
D
a
ta
,
v
o
l.
8
,
n
o
.
1
,
p
p
.
1
-
7
4
,
2
0
2
1
,
d
o
i:
1
0
.
1
1
8
6
/s
4
0
5
3
7
-
0
2
1
-
0
0
4
4
4
-
8
.
[1
2
]
AG
.
P
a
c
h
e
c
o
a
n
d
Kro
h
li
n
g
R,
“
An
a
tt
e
n
ti
o
n
-
b
a
se
d
m
e
c
h
a
n
ism
t
o
c
o
m
b
in
e
ima
g
e
s
a
n
d
m
e
tad
a
ta
in
d
e
e
p
lea
rn
in
g
m
o
d
e
ls
a
p
p
li
e
d
t
o
s
k
in
c
a
n
c
e
r
c
l
a
ss
ifi
c
a
ti
o
n
,
”
IEE
E
j
o
u
r
n
a
l
o
f
b
i
o
me
d
ica
l
a
n
d
h
e
a
lt
h
i
n
fo
rm
a
ti
c
s
,
F
e
b
2
6
,
2
0
2
1
,
d
o
i:
1
0
.
1
1
0
9
/JBHI.2
0
2
1
.
3
0
6
2
0
0
2
.
[1
3
]
HW.
Hu
a
n
g
,
B.
W
.
Hs
u
,
C.
H.
L
e
e
a
n
d
V.
S
.
Tse
n
g
,
“
De
v
e
lo
p
m
e
n
t
o
f
a
li
g
h
t‐we
ig
h
t
d
e
e
p
lea
rn
in
g
m
o
d
e
l
fo
r
c
l
o
u
d
a
p
p
li
c
a
ti
o
n
s
a
n
d
re
m
o
te
d
ia
g
n
o
si
s
o
f
s
k
in
c
a
n
c
e
rs,”
T
h
e
J
o
u
r
n
a
l
o
f
De
rm
a
t
o
lo
g
y
,
M
a
r,
v
o
l.
4
8
,
n
o
.
3
,
p
p
.
3
1
0
-
6
,
2
0
2
1
,
d
o
i:
1
0
.
1
1
1
1
/
1
3
4
6
-
8
1
3
8
.
1
5
6
8
3
.
[1
4
]
J.
De
n
g
,
W.
D
o
n
g
,
R.
S
o
c
h
e
r,
L.
J.
Li
,
L.
Ka
i,
a
n
d
F
.
-
F
.
Li
,
“
Im
a
g
e
Ne
t:
A
larg
e
-
sc
a
le
h
iera
rc
h
ica
l
i
m
a
g
e
d
a
tab
a
se
,
”
p
re
se
n
ted
a
t
th
e
Pr
o
c
.
IEE
E
Co
n
f
.
Co
mp
u
t.
Vi
s.
P
a
tt
e
rn
Rec
o
g
n
it
.
,
2
0
0
9
,
d
o
i:
1
0
.
1
1
0
9
/CV
P
R.
2
0
0
9
.
5
2
0
6
8
4
8
.
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.
23
,
No
.
3
,
Sep
tem
b
er
2
0
2
1
:
1
6
1
1
-
1
6
1
9
1618
[1
5
]
J.
Ka
wa
h
a
ra
,
A.
Be
n
Taie
b
,
a
n
d
G
.
Ha
m
a
rn
e
h
,
“
De
e
p
fe
a
tu
re
s
to
c
las
sify
sk
in
les
io
n
s,”
p
re
se
n
ted
a
t
th
e
Pro
c
.
IEE
E
1
3
t
h
In
t.
S
y
mp
.
Bi
o
me
d
.
Ima
g
.
,
2
0
1
6
,
d
o
i:
1
0
.
1
1
0
9
/IS
BI.
2
0
1
6
.
7
4
9
3
5
2
8
.
[1
6
]
A.
M
e
n
e
g
o
la,
M
.
F
o
rn
a
c
iali,
R.
P
ires
,
F
.
V.
Bit
ten
c
o
u
rt,
S
.
Av
il
a
,
a
n
d
E.
Va
ll
e
,
"
K
n
o
wle
d
g
e
tran
sfe
r
fo
r
m
e
lan
o
m
a
sc
re
e
n
in
g
with
d
e
e
p
lea
rn
i
n
g
,
”
p
re
se
n
ted
a
t
th
e
Pro
c
.
IEE
E
1
4
th
I
n
t.
S
y
mp
.
Bi
o
me
d
.
Ima
g
.
,
2
0
1
7
,
d
o
i:
1
0
.
1
1
0
9
/I
S
BI.
2
0
1
7
.
7
9
5
0
5
2
3
.
[1
7
]
Y.
G
o
n
g
,
L.
Wan
g
,
R.
G
u
o
,
a
n
d
S
.
Laz
e
b
n
i
k
,
“
M
u
lt
i
-
sc
a
le
Ord
e
rles
s
P
o
o
li
n
g
o
f
De
e
p
Co
n
v
o
lu
ti
o
n
a
l
Ac
ti
v
a
ti
o
n
F
e
a
tu
re
s,"
p
re
se
n
ted
a
t
th
e
Pro
c
.
1
3
t
h
E
u
r.
Co
n
f.
Co
mp
u
t.
Vi
s.
,
C
h
a
m
,
2
0
1
4
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
1
0
5
8
4
-
0
_
2
6
.
[1
8
]
Ho
wa
rd
,
A
n
d
re
w
G
.
“
S
o
m
e
imp
ro
v
e
m
e
n
ts
o
n
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
ra
l
n
e
two
rk
b
a
se
d
ima
g
e
c
las
sifica
ti
o
n
,
”
a
rXiv
p
re
p
ri
n
t
a
rXi
v
:1
3
1
2
.
5
4
0
2
,
2
0
1
3
.
[1
9
]
Z.
Ha
n
g
,
X.
Jia
,
a
n
d
D.
Krist
in
,
“
De
e
p
TE
N:
Tex
tu
re
E
n
c
o
d
in
g
Ne
two
rk
,
”
p
re
se
n
ted
a
t
th
e
Pro
c
.
IEE
E
C
o
n
f
.
Co
mp
u
t
.
Vi
s.
Pa
t
ter
n
Rec
o
g
n
it
.
,
2
0
1
7
,
d
o
i:
1
0
.
1
1
0
9
/cv
p
r.
2
0
1
7
.
3
0
9
.
[2
0
]
Y.
S
o
n
g
,
Q.
Li
,
H.
Hu
a
n
g
,
D.
F
e
n
g
,
M
.
Ch
e
n
,
a
n
d
W.
Ca
i,
“
L
o
w
Dim
e
n
sio
n
a
l
Re
p
re
se
n
tati
o
n
o
f
F
i
sh
e
r
Ve
c
to
rs
fo
r
M
icro
sc
o
p
y
Im
a
g
e
Clas
sifica
ti
o
n
,
”
IEE
E
T
ra
n
s.
M
e
d
.
Im
a
g
.
,
v
o
l.
3
6
,
n
o
.
8
,
p
p
.
1
6
3
6
-
1
6
4
9
,
2
0
1
7
,
d
o
i:
1
0
.
1
1
0
9
/
TM
I.
2
0
1
7
.
2
6
8
7
4
6
6
.
[2
1
]
D.
Yo
o
,
S
.
P
a
rk
,
J.
-
Y.
Lee
,
a
n
d
I
.
S
o
Kw
e
o
n
,
“
M
u
lt
i
-
sc
a
le
p
y
ra
m
id
p
o
o
li
n
g
f
o
r
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
r
e
p
re
se
n
tatio
n
,
”
p
re
se
n
ted
a
t
t
h
e
Pro
c
.
I
E
EE
Co
n
f.
C
o
mp
u
t.
V
is.
P
a
tt
e
rn
Rec
o
g
n
i
t.
W
o
rk
sh
o
p
s
,
2
0
1
5
,
d
o
i:
1
0
.
1
1
0
9
/CVP
R
W.
2
0
1
5
.
7
3
0
1
2
7
4
.
[
2
2
]
Z
.
Y
u
,
X
.
J
i
a
n
g
,
T
.
W
a
n
g
,
a
n
d
B
.
L
e
i
,
“
A
g
g
r
e
g
a
t
i
n
g
D
e
e
p
C
o
n
v
o
l
u
t
i
o
n
a
l
F
e
a
t
u
r
e
s
f
o
r
M
e
l
a
n
o
m
a
R
e
c
o
g
n
i
t
i
o
n
i
n
D
e
r
m
o
sc
o
p
y
I
m
a
g
e
s
,
"
p
r
e
s
e
n
t
e
d
a
t
t
h
e
M
a
c
h
.
L
e
a
r
n
.
M
e
d
.
I
m
a
g
.
,
C
h
a
m
,
2
0
1
7
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
319
-
6
7
3
8
9
-
9
_
2
8
.
[2
3
]
J.
Yu
e
-
He
i
Ng
,
F
.
Ya
n
g
,
a
n
d
L.
S
.
Da
v
is,
“
Ex
p
l
o
it
i
n
g
l
o
c
a
l
fe
a
t
u
re
s
fro
m
d
e
e
p
n
e
two
rk
s
fo
r
im
a
g
e
re
tri
e
v
a
l,
”
p
re
se
n
ted
a
t
th
e
Pro
c
.
I
EE
E
Co
n
f
.
Co
mp
u
t.
Vi
s
.
Pa
tt
e
rn
Rec
o
g
n
it
.
W
o
rk
sh
o
p
,
2
0
1
5
,
d
o
i:
1
0
.
1
1
0
9
/c
v
p
rw.
2
0
1
5
.
7
3
0
1
2
7
2
.
[2
4
]
M
.
Cim
p
o
i
,
S
.
M
a
ji
,
a
n
d
A.
Ve
d
a
ld
i,
"
De
e
p
fil
ter
b
a
n
k
s
fo
r
te
x
tu
re
re
c
o
g
n
it
i
o
n
a
n
d
se
g
m
e
n
tatio
n
,
"
p
re
se
n
ted
a
t
th
e
Pro
c
.
IEE
E
Co
n
f.
C
o
mp
u
t.
Vi
s
.
P
a
tt
e
rn
Rec
o
g
n
it
.
,
2
0
1
5
,
d
o
i:
1
0
.
1
1
0
9
/CVP
R.
2
0
1
5
.
7
2
9
9
0
0
7
.
[2
5
]
K.
Ko
ro
t
k
o
v
a
n
d
R.
G
a
r
c
ia,
“
Co
m
p
u
teriz
e
d
a
n
a
ly
sis
o
f
p
ig
m
e
n
ted
sk
in
les
io
n
s:
A
re
v
iew
,
"
Arti
f.
I
n
tell.
M
e
d
.
,
v
o
l.
5
6
,
p
p
.
6
9
-
9
0
,
2
0
1
2
,
d
o
i.
o
rg
/1
0
.
1
0
1
6
/j
.
a
rtme
d
.
2
0
1
2
.
0
8
.
0
0
2
.
[2
6
]
H.
G
a
n
ste
r,
P
.
P
i
n
z
,
R.
Ro
h
re
r,
E
.
Wi
l
d
li
n
g
,
M
.
Bin
d
e
r,
a
n
d
H.
Ki
tt
ler,
“
Au
t
o
m
a
ted
m
e
lan
o
m
a
re
c
o
g
n
it
i
o
n
,
”
IEE
E
T
ra
n
s.
M
e
d
.
Ima
g
.
,
v
o
l
.
2
0
,
n
o
.
3
,
p
p
.
2
3
3
-
2
3
9
,
2
0
0
1
,
d
o
i:
1
0
.
1
1
0
9
/
4
2
.
9
1
8
4
7
3
.
[2
7
]
G
.
Ca
p
d
e
h
o
u
ra
t,
A.
C
o
re
z
,
A.
Ba
z
z
a
n
o
,
R.
Alo
n
so
,
a
n
d
P
.
M
u
sé
,
“
To
wa
rd
a
c
o
m
b
i
n
e
d
t
o
o
l
t
o
a
ss
ist
d
e
rm
a
to
lo
g
ists
in
m
e
lan
o
m
a
d
e
tec
ti
o
n
fr
o
m
d
e
r
m
o
sc
o
p
ic
ima
g
e
s
o
f
p
i
g
m
e
n
ted
s
k
in
les
io
n
s,”
P
a
tt
e
rn
Rec
o
g
n
it
.
L
e
tt
.
,
v
o
l
.
3
2
,
n
o
.
1
6
,
p
p
.
2
1
8
7
-
2
1
9
6
,
2
0
1
1
,
d
o
i:
1
0
.
1
0
1
6
/j
.
p
a
trec
.
2
0
1
1
.
0
6
.
0
1
5
.
[2
8
]
F
.
Xie
,
H.
F
a
n
,
L
.
Ya
n
g
,
Z.
Jia
n
g
,
R.
M
e
n
g
,
a
n
d
A.
Bo
v
ik
,
“
M
e
lan
o
m
a
Clas
sifica
ti
o
n
o
n
De
rm
o
sc
o
p
y
Im
a
g
e
s
u
sin
g
a
Ne
u
ra
l
Ne
two
r
k
E
n
se
m
b
le
M
o
d
e
l,
”
IE
EE
T
r
a
n
s.
M
e
d
.
I
ma
g
.
,
v
o
l.
3
6
,
n
o
.
3
,
p
p
.
1
-
1
,
2
0
1
6
,
d
o
i:
1
0
.
1
1
0
9
/
TM
I.
2
0
1
6
.
2
6
3
3
5
5
1
.
[2
9
]
L.
Bi,
J.
Kim
,
E.
A
h
n
,
D.
F
e
n
g
,
a
n
d
M
.
F
u
lh
a
m
,
“
Au
t
o
m
a
ti
c
m
e
lan
o
m
a
d
e
tec
ti
o
n
v
ia
m
u
l
ti
-
sc
a
le
les
io
n
-
b
ias
e
d
re
p
re
se
n
tatio
n
a
n
d
j
o
in
t
re
v
e
rse
c
las
sifica
ti
o
n
,
"
p
re
se
n
ted
a
t
t
h
e
Pr
o
c
.
IEE
E
1
3
t
h
I
n
t.
S
y
mp
.
Bi
o
me
d
.
Ima
g
.
,
2
0
1
6
,
d
o
i:
1
0
.
1
1
0
9
/IS
BI
.
2
0
1
6
.
7
4
9
3
4
4
7
.
[3
0
]
C.
Ba
ra
ta,
M
.
R
u
e
la,
M
.
F
ra
n
c
isc
o
,
T.
M
e
n
d
o
n
ç
a
,
a
n
d
J.
S
.
M
a
rq
u
e
s,
“
Two
S
y
ste
m
s
fo
r
th
e
De
tec
ti
o
n
o
f
M
e
lan
o
m
a
s
in
De
rm
o
sc
o
p
y
Im
a
g
e
s
Us
in
g
Tex
t
u
re
a
n
d
Co
l
o
r
F
e
a
t
u
re
s,”
IEE
E
S
y
ste
ms
J
o
u
rn
a
l
,
v
o
l.
8
,
n
o
.
3
,
p
p
.
965
-
9
7
9
,
2
0
1
4
,
d
o
i:
1
0
.
1
1
0
9
/JS
Y
S
T.
2
0
1
3
.
2
2
7
1
5
4
0
.
[3
1
]
L.
Alz
u
b
a
id
i
,
F
a
d
h
e
l,
M
.
A.,
Ol
e
iwi,
S
.
R.
,
Al
-
S
h
a
m
m
a
,
O.,
a
n
d
Zh
a
n
g
,
J,
“
DFU_
QU
TNe
t
:
d
ia
b
e
ti
c
fo
o
t
u
lce
r
c
las
sifica
ti
o
n
u
si
n
g
n
o
v
e
l
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
,
”
M
u
l
ti
me
d
ia
T
o
o
ls
a
n
d
A
p
p
li
c
a
ti
o
n
s
,
v
o
l.
7
9
,
n
o
.
2
1
,
p
p
.
1
5
6
5
5
-
1
5
6
7
7
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
7
/s1
1
0
4
2
-
0
1
9
-
0
7
8
2
0
-
w
.
[3
2
]
A.
A.
Ab
b
o
o
d
,
Q.
M
.
S
h
a
ll
a
l,
a
n
d
M
.
A.
F
a
d
h
e
l,
“
A
u
to
m
a
ted
b
ra
in
tu
m
o
r
c
las
sifica
ti
o
n
u
sin
g
v
a
rio
u
s
d
e
e
p
lea
rn
i
n
g
m
o
d
e
ls:
a
c
o
m
p
a
ra
ti
v
e
stu
d
y
,
”
In
d
o
n
e
si
a
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
E
n
g
i
n
e
e
rin
g
a
n
d
Co
m
p
u
ter
S
c
ien
c
e
(IJ
EE
CS
)
,
v
o
l
.
2
2
,
n
o
.
1
,
p
p
.
2
5
2
-
2
5
9
,
2
0
2
1
,
d
o
i:
1
0
.
1
1
5
9
1
/i
jee
c
s.v
2
2
.
i
1
.
p
p
2
5
2
-
259
.
[3
3
]
L.
Alz
u
b
a
i
d
i,
R.
I.
Ha
sa
n
,
F
.
H.
Aw
a
d
,
M
.
A.
F
a
d
h
e
l,
O.
Alsh
a
m
m
a
,
a
n
d
J.
Z
h
a
n
g
,
“
M
u
lt
i
-
c
las
s
Bre
a
st
Ca
n
c
e
r
Clas
sifica
ti
o
n
b
y
a
No
v
e
l
Two
-
Bra
n
c
h
De
e
p
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
Arc
h
it
e
c
tu
re
,
”
In
2
0
1
9
1
2
th
In
ter
n
a
t
io
n
a
l
Co
n
fer
e
n
c
e
o
n
De
v
e
lo
p
me
n
ts
i
n
e
S
y
ste
ms
En
g
i
n
e
e
rin
g
(De
S
E)
,
p
p
.
2
6
8
-
2
7
3
.
IEE
E
,
2
0
1
9
,
Oc
to
b
e
r,
d
o
i:
1
0
.
1
1
0
9
/De
S
E.
2
0
1
9
.
0
0
0
5
7
.
[3
4
]
S
.
De
m
y
a
n
o
v
,
R
.
Ch
a
k
ra
v
o
rt
y
,
M
.
Ab
e
d
in
i,
A.
Ha
lp
e
r
n
,
a
n
d
R.
G
a
rn
a
v
i,
"
Clas
sifica
ti
o
n
o
f
d
e
rm
o
sc
o
p
y
p
a
tt
e
r
n
s
u
sin
g
d
e
e
p
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
s,”
p
re
se
n
ted
a
t
th
e
Pro
c
.
IEE
E
1
3
th
I
n
t.
S
y
mp
.
Bi
o
me
d
.
I
ma
g
.
,
2
0
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/I
S
BI.
2
0
1
6
.
7
4
9
3
2
8
4
.
[3
5
]
L.
Yu
,
H.
Ch
e
n
,
Q.
Do
u
,
J.
Qin
,
a
n
d
P
.
A.
He
n
g
,
“
Au
t
o
m
a
ted
M
e
lan
o
m
a
Re
c
o
g
n
it
io
n
in
De
rm
o
sc
o
p
y
Im
a
g
e
s
v
i
a
Ve
ry
De
e
p
Re
sid
u
a
l
Ne
two
r
k
s,”
in
IE
EE
T
ra
n
s
a
c
ti
o
n
s
o
n
M
e
d
ica
l
Ima
g
in
g
,
v
o
l.
3
6
,
n
o
.
4
,
e
d
,
2
0
1
6
,
p
p
.
1
-
1
,
d
o
i:
1
0
.
1
1
0
9
/
TM
I.
2
0
1
6
.
2
6
4
2
8
3
9
.
[3
6
]
A.
Estev
a
e
t
a
l.
,
“
De
rm
a
to
lo
g
ist
-
lev
e
l
c
las
sifica
ti
o
n
o
f
sk
i
n
c
a
n
c
e
r
with
d
e
e
p
n
e
u
ra
l
n
e
two
r
k
s,”
N
a
tu
re
,
v
o
l
.
5
4
2
,
p
p
.
1
1
5
-
1
1
8
,
2
0
1
7
,
d
o
i:
1
0
.
1
0
3
8
/
n
a
tu
re
2
1
0
5
6
.
[3
7
]
N.
Co
d
e
ll
a
,
Q.
-
B.
Ng
u
y
e
n
,
S
.
P
a
n
k
a
n
t
i,
D.
G
u
tma
n
,
B.
He
lb
a
,
A.
Ha
lp
e
rn
,
“
De
e
p
Lea
rn
in
g
En
se
m
b
les
fo
r
M
e
lan
o
m
a
Re
c
o
g
n
it
io
n
i
n
De
rm
o
sc
o
p
y
Im
a
g
e
s,”
v
o
l.
6
1
,
p
p
.
4
,
2
0
1
6
,
d
o
i:
1
0
.
1
1
4
7
/
jrd
.
2
0
1
7
.
2
7
0
8
2
9
9
.
[3
8
]
Z.
Yu
e
t
a
l.
,
“
Hy
b
ri
d
d
e
rm
o
sc
o
p
y
ima
g
e
c
las
sifica
ti
o
n
fra
m
e
wo
rk
b
a
se
d
o
n
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
a
n
d
F
ish
e
r
v
e
c
to
r
,
"
p
re
se
n
ted
a
t
t
h
e
P
r
o
c
.
IEE
E
1
4
th
I
n
t.
S
y
m
p
.
Bi
o
m
e
d
.
Im
a
g
,
2
0
1
7
,
d
o
i:
1
0
.
1
1
0
9
/I
S
BI.
2
0
1
7
.
7
9
5
0
5
2
4
.
[3
9
]
L.
Alz
u
b
a
i
d
i,
O.
Al
-
S
h
a
m
m
a
,
M
.
A.
F
a
d
h
e
l,
L.
F
a
r
h
a
n
,
a
n
d
J.
Zh
a
n
g
,
“
Clas
sifica
ti
o
n
o
f
re
d
b
l
o
o
d
c
e
ll
s in
sic
k
le ce
l
l
a
n
e
m
ia
u
sin
g
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
,
”
In
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
I
n
telli
g
e
n
t
S
y
ste
ms
De
sig
n
a
n
d
Ap
p
li
c
a
ti
o
n
s
,
p
p
.
5
5
0
-
5
5
9
.
S
p
ri
n
g
e
r,
Ch
a
m
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
1
6
6
5
7
-
1
_
5
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
B
ig
tr
a
n
s
fer lea
r
n
in
g
fo
r
a
u
to
ma
ted
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
(
Zin
a
h
Mo
h
s
in
A
r
ka
h
)
1619
[4
0
]
M
.
A.
F
a
d
h
e
l,
O.
Al
-
S
h
a
m
m
a
,
S
.
R.
Ole
iwi,
B.
H.
Tah
e
r,
a
n
d
L.
Alz
u
b
a
id
i,
“
Re
a
l
-
ti
m
e
P
CG
d
i
a
g
n
o
sis
u
sin
g
F
P
G
A,”
In
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
In
telli
g
e
n
t
S
y
ste
ms
De
sig
n
a
n
d
A
p
p
li
c
a
ti
o
n
s
,
p
p
.
5
1
8
-
5
2
9
.
S
p
rin
g
e
r,
Ch
a
m
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
1
6
6
5
7
-
1
_
4
8
.
[4
1
]
O.
Al
-
S
h
a
m
m
a
,
M
.
A.
F
a
d
h
e
l,
R
.
A.
Ha
m
e
e
d
,
L.
Alz
u
b
a
id
i,
a
n
d
J.
Zh
a
n
g
,
“
Bo
o
stin
g
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
p
e
rfo
rm
a
n
c
e
b
a
se
d
o
n
F
P
G
A
a
c
c
e
ler
a
to
r,
”
In
I
n
ter
n
a
t
io
n
a
l
Co
n
fer
e
n
c
e
o
n
In
tell
ig
e
n
t
S
y
ste
ms
De
sig
n
a
n
d
Ap
p
li
c
a
ti
o
n
s
,
p
p
.
5
0
9
-
5
1
7
.
S
p
ri
n
g
e
r,
Ch
a
m
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
1
6
6
5
7
-
1
_
4
7
.
[4
2
]
Th
e
In
tern
a
ti
o
n
a
l
S
k
i
n
Im
a
g
in
g
Co
ll
a
b
o
ra
ti
o
n
|
Ch
a
ll
e
n
g
e
s.
Ac
c
e
ss
e
d
:
De
c
.
1
,
2
0
2
0
.
[On
li
n
e
].
Av
a
il
a
b
le
:
h
tt
p
s:/
/www
.
isic
-
a
rc
h
iv
e
.
c
o
m
/#
!/
t
o
p
Wi
th
He
a
d
e
r/t
ig
h
tC
o
n
ten
tT
o
p
/c
h
a
ll
e
n
g
e
s.
[4
3
]
S
IIM
-
IS
IC
M
e
lan
o
m
a
Clas
sifica
ti
o
n
.
h
tt
p
s://
ww
w.k
a
g
g
le.co
m
/c/s
ii
m
-
isic
-
m
e
lan
o
m
a
-
c
las
sific
a
ti
o
n
.
Ac
c
e
s
se
d
:
2
6
De
c
e
m
b
e
r
2
0
2
0
.
[4
4
]
K.
S
imo
n
y
a
n
,
a
n
d
A.
Zi
ss
e
rm
a
n
,
"
Ve
ry
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
t
wo
rk
s
f
o
r
lar
g
e
-
sc
a
le
ima
g
e
re
c
o
g
n
it
io
n
,
"
a
rXiv
p
re
p
rin
t
a
rXiv
:
1
4
0
9
.
1
5
5
6
.
2
0
1
4
.
[4
5
]
C.
S
z
e
g
e
d
y
,
W
.
Li
u
,
Y.
Jia
,
P
.
S
e
rm
a
n
e
t,
S
.
Re
e
d
,
D.
An
g
u
e
lo
v
a
n
d
A.
Ra
b
i
n
o
v
ich
,
“
G
o
in
g
d
e
e
p
e
r
wit
h
c
o
n
v
o
lu
ti
o
n
s,”
In
:
Pro
c
e
e
d
i
n
g
s
o
f
th
e
IEE
E
c
o
n
fer
e
n
c
e
o
n
c
o
m
p
u
te
r
v
isio
n
a
n
d
p
a
tt
e
rn
re
c
o
g
n
it
io
n
,
p
p
.
1
-
9
,
2
0
1
5
,
d
o
i:
1
0
.
1
1
0
9
/CVP
R.
2
0
1
5
.
7
2
9
8
5
9
4
.
[4
6
]
K.
He
,
X.
Zh
a
n
g
,
S
.
Re
n
a
n
d
J.
S
u
n
,
“
De
e
p
re
sid
u
a
l
lea
rn
in
g
fo
r
ima
g
e
re
c
o
g
n
it
io
n
,
”
In
Pro
c
e
e
d
i
n
g
s
o
f
th
e
IEE
E
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
a
n
d
p
a
tt
e
rn
re
c
o
g
n
it
io
n
,
p
p
.
7
7
0
-
7
7
8
,
2
0
1
6
,
d
o
i:
1
0
.
1
1
0
9
/CV
P
R.
2
0
1
6
.
9
0
.
[4
7
]
O.
Ru
ss
a
k
o
v
sk
y
,
J.
De
n
g
,
H.
S
u
,
J.
Kra
u
se
,
S
.
S
a
t
h
e
e
sh
,
S
.
M
a
a
n
d
AC.
Be
rg
,
“
Im
a
g
e
Ne
t
larg
e
sc
a
le
v
isu
a
l
re
c
o
g
n
it
i
o
n
c
h
a
ll
e
n
g
e
,
"
I
n
t
J
C
o
mp
u
t
Vi
s
,
v
o
l.
1
1
5
,
n
o
.
3
,
p
p
.
2
1
1
-
2
5
2
,
2
0
1
5
,
d
o
i:
1
0
.
1
0
0
7
/s1
1
2
6
3
-
0
1
5
-
0
8
1
6
-
y
.
[4
8
]
H.
Al
-
Ya
ss
in
,
J.
I.
M
o
u
sa
,
M
.
A.
F
a
d
h
e
l,
O.
Al
-
S
h
a
m
m
a
,
a
n
d
L.
Alz
u
b
a
i
d
i,
“
S
ta
ti
stica
l
a
c
c
u
r
a
c
y
a
n
a
ly
sis
o
f
d
iffere
n
t
d
e
tec
ti
n
g
a
lg
o
rit
h
m
s
fo
r
su
rv
e
il
lan
c
e
sy
ste
m
i
n
sm
a
rt
c
it
y
,
”
In
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
E
n
g
in
e
e
rin
g
a
n
d
Co
m
p
u
ter
S
c
ie
n
c
e
(IJ
EE
CS
)
,
v
o
l.
1
8
,
n
o
.
2
,
p
p
.
9
7
9
-
9
8
6
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
5
9
1
/
ij
e
e
c
s.v
1
8
.
i2
.
p
p
9
7
9
-
9
8
6
.
[4
9
]
A.
Al
-
S
a
b
a
a
wi,
H.
M
.
Ib
ra
h
im,
Z.
M
.
Ark
a
h
,
M
.
Al
-
Am
id
ie,
a
n
d
L.
Alz
u
b
a
i
d
i,
“
Am
e
n
d
e
d
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
with
G
lo
b
a
l
Av
e
ra
g
e
P
o
o
l
in
g
fo
r
Im
a
g
e
Clas
sifica
ti
o
n
,
”
In
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
In
telli
g
e
n
t
S
y
ste
ms
De
sig
n
a
n
d
A
p
p
li
c
a
ti
o
n
s
,
p
p
.
1
7
1
-
1
8
0
,
S
p
ri
n
g
e
r,
C
h
a
m
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
7
1
1
8
7
-
0
_
1
6
.
[5
0
]
L.
Alz
u
b
a
id
i
,
O.
Al
-
S
h
a
m
m
a
,
M
.
A.
F
a
d
h
e
l,
Z.
M
.
Ark
a
h
a
n
d
F
.
H.
Aw
a
d
,
“
A
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
m
o
d
e
l
fo
r
m
u
lt
i
-
c
las
s
fru
it
s
c
l
a
ss
ifi
c
a
ti
o
n
,
”
In
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
In
telli
g
e
n
t
S
y
ste
ms
De
sig
n
a
n
d
Ap
p
li
c
a
ti
o
n
s
,
p
p
.
9
0
-
9
9
.
S
p
ri
n
g
e
r,
Ch
a
m
,
2
0
1
9
,
De
c
e
m
b
e
r,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
4
9
3
4
2
-
4
_
9
.
[5
1
]
L.
Alz
u
b
a
i
d
i
,
A.
A.
Ab
b
o
o
d
,
M
.
A.
F
a
d
h
e
l
,
O.
Al
-
S
h
a
m
m
a
a
n
d
J.
Zh
a
n
g
,
“
Co
m
p
a
ris
o
n
o
f
h
y
b
r
i
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
m
o
d
e
ls
f
o
r
d
iab
e
ti
c
fo
o
t
u
lce
r
c
las
sifica
ti
o
n
,
”
J
o
u
r
n
a
l
o
f
En
g
in
e
e
rin
g
S
c
ien
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l.
1
6
,
n
o
.
3
,
p
p
.
2
0
0
1
-
2
0
1
7
.
2
0
2
1
.
[5
2
]
O.
Alsh
a
m
m
a
,
F
.
H.
Aw
a
d
,
L.
Alz
u
b
a
id
i,
M
.
A.
F
a
d
h
e
l,
Z.
M
.
Ark
a
h
a
n
d
L
.
F
a
rh
a
n
,
“
Emp
l
o
y
m
e
n
t
o
f
M
u
lt
i
-
c
las
sifier
a
n
d
M
u
lt
i
-
d
o
m
a
in
F
e
a
tu
re
s
fo
r
P
CG
Re
c
o
g
n
it
io
n
,
”
In
2
0
1
9
1
2
th
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
De
v
e
lo
p
me
n
ts i
n
e
S
y
ste
ms
En
g
i
n
e
e
rin
g
(De
S
E)
,
p
p
.
3
2
1
-
3
2
5
.
IEE
E
,
2
0
1
9
,
d
o
i:
1
0
.
1
1
0
9
/De
S
E.
2
0
1
9
.
0
0
0
6
6
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Zin
a
h
Mo
h
sin
Ar
k
a
h
M.
S
c
.
in
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
Tec
h
n
o
l
o
g
y
fro
m
S
o
u
t
h
we
st
Jia
o
to
n
g
Un
iv
e
rsity
,
C
h
in
a
-
2
0
1
7
.
Cu
rre
n
tl
y
,
I'
m
wo
rk
in
g
a
s
a
lec
tu
re
a
t
th
e
Un
iv
e
rsity
o
f
i
n
f
o
r
m
a
ti
o
n
tec
h
n
o
l
o
g
y
a
n
d
C
o
m
m
u
n
ica
ti
o
n
.
M
y
re
se
a
rc
h
a
re
a
in
tere
sts
a
re
De
e
p
Lea
rn
in
g
,
M
a
c
h
i
n
e
lea
rn
in
g
,
d
a
ta m
in
in
g
,
b
ig
d
a
ta,
I
m
a
g
e
p
ro
c
e
ss
in
g
a
n
d
Di
g
it
a
l
Im
a
g
e
P
ro
c
e
ss
in
g
.
Da
ly
a
S
.
Al
-
Dul
a
i
m
i
Lec
tu
re
a
t
Un
iv
e
rsiy
u
o
f
In
f
o
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
a
n
d
C
o
m
m
u
n
ica
ti
o
n
M
e
d
ica
l
In
f
o
rm
a
ti
c
s Co
ll
e
g
e
.
B.
S
c
.
d
e
g
re
e
in
Co
m
p
u
ter S
c
ien
c
e
/
B
a
g
h
d
a
d
Un
i
v
e
rsity
(Ira
q
).
M
.
S
c
.
d
e
g
re
e
in
I
n
fo
rm
a
ti
o
n
Tec
h
n
o
l
o
g
y
/
Ça
n
k
a
y
a
Un
i
v
e
rsity
(Tu
r
k
e
y
).
M
y
s
p
e
c
ialize
d
field
i
n
M
e
d
ica
l
In
f
o
rm
a
ti
o
n
Tec
h
n
o
lo
gy.
Ahl
a
m
Ra
shi
d
K
h
e
k
a
n
B.
S
c
.
d
e
g
re
e
in
c
o
m
p
u
ter sc
ien
c
e
/Al
-
M
u
sta
n
siry
a
Un
i
v
e
rsity
(Ira
q
),
M
.
S
c
.
d
e
g
re
e
in
c
o
m
p
u
ter sc
ien
c
e
/Al
-
M
u
sta
n
sir
y
a
Un
i
v
e
rsity
(I
ra
q
).
M
y
sp
e
c
ialize
Artifi
c
ial
In
telli
g
e
n
c
e
Evaluation Warning : The document was created with Spire.PDF for Python.