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.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
,
p
p
.
1
4
6
9
~
1
4
8
0
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
24
.i
3
.
pp
1
4
6
9
-
1
4
8
0
1469
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Pneumo
nia
de
tectio
n bas
ed on tran
sfer le
a
rning
and
a
co
mbina
tion o
f
V
G
G
1
9
a
nd a
CNN
built
from scra
t
c
h
O
us
s
a
m
a
Da
hm
a
ne,
M
us
t
a
ph
a
K
helifi,
M
o
ha
m
m
ed
B
ela
dg
ha
m
,
I
bra
him
K
a
dri
Lab
o
ra
to
r
y
o
f
TIT
,
De
p
a
rtme
n
t
o
f
El
e
c
tri
c
a
l
En
g
in
e
e
rin
g
,
Tah
ri
M
o
h
a
m
m
e
d
Un
iv
e
rsity
,
Be
c
h
a
r,
Alg
e
ria
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
13
,
2
0
2
1
R
ev
is
ed
Oct
25
,
2
0
2
1
Acc
ep
ted
Oct
27
,
2
0
2
1
In
th
is
p
a
p
e
r,
to
c
a
teg
o
rize
a
n
d
d
e
tec
t
p
n
e
u
m
o
n
ia
fro
m
a
c
o
l
lec
ti
o
n
o
f
c
h
e
st
X
-
ra
y
p
ictu
re
sa
m
p
les
,
we
p
ro
p
o
s
e
a
d
e
e
p
lea
rn
in
g
tec
h
n
i
q
u
e
b
a
se
d
o
n
o
b
jec
t
d
e
tec
ti
o
n
,
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s,
a
n
d
tran
sfe
r
lea
rn
i
n
g
.
Th
e
p
ro
p
o
se
d
m
o
d
e
l
is
a
c
o
m
b
in
a
ti
o
n
o
f
th
e
p
r
e
-
train
e
d
m
o
d
e
l
(VG
G
1
9
)
a
n
d
o
u
r
d
e
sig
n
e
d
a
rc
h
it
e
c
tu
re
.
Th
e
G
u
a
n
g
z
h
o
u
Wo
m
e
n
a
n
d
Ch
i
ld
re
n
'
s
M
e
d
ica
l
Ce
n
ter
in
G
u
a
n
g
z
h
o
u
,
C
h
in
a
p
ro
v
id
e
d
th
e
c
h
e
st
X
-
ra
y
d
a
tas
e
t
u
se
d
in
th
is
st
u
d
y
.
T
h
e
re
a
re
5
,
0
0
0
sa
m
p
les
i
n
th
e
d
a
ta
s
e
t,
with
1
,
5
8
3
h
e
a
lt
h
y
sa
m
p
les
a
n
d
4
,
2
7
3
p
n
e
u
m
o
n
ia
sa
m
p
les
.
P
re
p
ro
c
e
ss
in
g
tec
h
n
i
q
u
e
s
s
u
c
h
a
s
c
o
n
tra
st
li
m
it
e
d
a
d
a
p
ti
v
e
h
isto
g
ra
m
e
q
u
a
li
z
a
ti
o
n
(CLAHE)
a
n
d
b
rig
h
t
n
e
ss
p
re
se
rv
in
g
b
i
-
h
ist
o
g
ra
m
e
q
u
a
li
z
a
ti
o
n
wa
s
a
l
so
u
se
d
(BBHE)
t
o
imp
ro
v
e
a
c
c
u
ra
c
y
.
D
u
e
to
th
e
imb
a
lan
c
e
o
f
t
h
e
d
a
ta
se
t,
we
a
d
o
p
te
d
so
m
e
train
i
n
g
tec
h
n
i
q
u
e
s
to
imp
ro
v
e
th
e
lea
rn
i
n
g
p
r
o
c
e
ss
o
f
th
e
sa
m
p
les
.
T
h
is
n
e
two
rk
a
c
h
i
e
v
e
d
o
v
e
r
9
9
%
a
c
c
u
ra
c
y
d
u
e
to
th
e
p
r
o
p
o
se
d
a
rc
h
it
e
c
tu
re
th
a
t
is
b
a
se
d
o
n
a
c
o
m
b
in
a
ti
o
n
o
f
two
m
o
d
e
ls.
Th
e
p
re
-
train
e
d
VG
G
1
9
a
s
fe
a
tu
re
e
x
trac
to
r
a
n
d
o
u
r
d
e
sig
n
e
d
co
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
(
CNN
)
.
K
ey
w
o
r
d
s
:
B
B
HE
C
L
AHE
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Dee
p
lear
n
in
g
Ob
ject
d
etec
tio
n
T
r
an
s
f
er
lear
n
i
n
g
VGG1
9
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
:
Ou
s
s
am
a
Dah
m
an
e
L
ab
o
r
ato
r
y
o
f
T
I
T
,
De
p
ar
t
m
en
t o
f
E
lectr
ical
E
n
g
i
n
ee
r
in
g
T
ah
r
i M
o
h
am
m
ed
Un
iv
e
r
s
ity
B
ec
h
ar
,
Alg
er
ia
E
m
ail:
d
ah
m
an
e.
o
u
s
s
am
a@
u
n
iv
-
b
ec
h
ar
.
d
z
1.
I
NT
RO
D
UCT
I
O
N
E
p
id
em
ics
an
d
ch
r
o
n
ic
d
is
ea
s
es
h
av
e
tak
en
m
an
y
liv
es
an
d
ca
u
s
ed
m
ass
iv
e
cr
is
es
ar
o
u
n
d
th
e
wo
r
ld
o
v
er
th
e
d
ec
ad
es.
Pn
eu
m
o
n
ia,
a
co
m
m
o
n
lu
n
g
in
f
ec
tio
n
ca
u
s
ed
b
y
b
ac
ter
ia,
v
ir
al,
o
r
f
u
n
g
i,
is
o
n
e
o
f
th
ese
d
is
o
r
d
er
s
.
I
t
is
th
e
wo
r
l
d
'
s
to
p
ca
u
s
e
o
f
d
ea
t
h
.
T
h
e
r
e
ar
e
tw
o
v
ar
ieties
o
f
p
n
eu
m
o
n
ia:
v
ir
a
l
an
d
b
ac
ter
ial.
T
h
e
d
if
f
er
en
ce
b
etwe
en
th
e
two
is
th
e
tr
ea
tm
en
t
tech
n
iq
u
e;
b
ac
ter
ial
p
n
eu
m
o
n
ia
is
u
s
u
ally
tr
ea
ted
with
an
tib
io
tics
,
wh
ils
t
v
ir
al
p
n
eu
m
o
n
ia
im
p
r
o
v
es
o
n
its
o
wn
[
1
]
.
I
t
is
a
p
r
ev
alen
t
d
is
ea
s
e
th
at
h
as
m
an
y
ca
u
s
es,
o
n
e
o
f
wh
ic
h
is
th
e
h
ig
h
am
o
u
n
t o
f
p
o
llu
tio
n
[
2
]
.
I
n
2
0
1
6
,
p
n
eu
m
o
n
ia
r
an
k
ed
ei
g
h
th
o
n
th
e
lis
t
o
f
th
e
to
p
te
n
ca
u
s
es
o
f
d
ea
th
in
th
e
Un
ited
States
[
3
]
.
I
n
2
0
1
7
,
th
is
r
esp
ir
ato
r
y
d
is
ea
s
e
k
illed
g
lo
b
ally
o
v
er
a
m
illi
o
n
liv
es
o
f
b
o
t
h
ch
ild
r
e
n
u
n
d
er
th
e
ag
e
o
f
5
an
d
th
e
eld
er
ly
o
v
er
5
0
,
an
d
it
co
n
tin
u
es
to
b
e
a
d
ea
d
ly
d
is
ea
s
e
n
o
wad
ay
s
if
n
o
t
d
iag
n
o
s
ed
ea
r
lier
[
4
]
.
P
n
e
u
m
o
n
i
a
c
a
n
b
e
i
d
e
n
t
i
f
i
e
d
t
h
r
o
u
g
h
t
h
e
e
x
a
m
i
n
a
t
i
o
n
o
f
t
h
e
l
u
n
g
r
a
d
i
o
g
r
a
p
h
i
c
a
l
l
y
,
b
y
C
T
-
s
c
a
n
,
o
r
e
v
e
n
b
y
m
o
l
e
c
u
l
a
r
d
i
a
g
n
o
s
t
i
c
a
s
s
a
y
s
(
s
u
c
h
a
s
PC
R
)
[
5
]
,
b
u
t
t
h
e
m
o
s
t
p
o
p
u
l
a
r
d
i
a
g
n
o
s
i
s
m
e
t
h
o
d
is
b
y
u
s
i
n
g
r
a
d
i
o
g
r
a
p
h
y
(
c
h
e
s
t
X
-
r
a
y
e
x
a
m
i
n
a
t
i
o
n
)
,
u
s
i
n
g
a
v
er
y
s
m
a
l
l
d
o
s
e
o
f
i
o
n
i
z
i
n
g
r
a
d
i
at
i
o
n
t
o
p
r
o
d
u
c
e
p
i
c
t
u
r
es
o
f
t
h
e
i
n
s
i
d
e
o
f
t
h
e
c
h
es
t
.
T
h
e
e
x
p
e
r
t
u
s
e
s
t
h
o
s
e
p
ic
t
u
r
e
s
to
c
o
n
f
i
r
m
o
r
v
a
l
i
d
a
t
e
i
f
t
h
e
p
e
r
s
o
n
i
s
i
n
f
e
c
t
e
d
b
y
p
n
e
u
m
o
n
i
a
o
r
n
o
t
.
On
th
e
o
t
h
er
h
an
d
,
m
an
y
r
es
ea
r
ch
er
s
r
ely
o
n
a
v
a
r
iety
o
f
ar
tific
ial
in
tellig
en
ce
an
d
d
ee
p
lear
n
in
g
-
b
ased
p
n
e
u
m
o
n
ia
d
etec
tio
n
s
tr
ateg
ies,
p
ar
ticu
lar
ly
i
n
s
e
g
m
en
tatio
n
an
d
class
if
icatio
n
.
C
o
m
p
u
ter
-
aid
ed
d
iag
n
o
s
is
(
C
AD)
,
wh
ich
u
s
es
ar
tific
ial
in
tellig
en
ce
-
b
ased
s
o
lu
tio
n
s
,
is
a
ter
m
u
s
ed
in
th
e
m
ed
ical
in
d
u
s
tr
y
to
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.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
4
6
9
-
1
4
8
0
1470
d
escr
ib
e
th
ese
au
to
m
atic
d
ete
ctio
n
p
r
o
ce
d
u
r
es
[
6
]
,
[
7
]
.
I
n
p
o
o
r
r
e
g
io
n
s
a
n
d
n
atio
n
s
s
u
ch
as
So
u
th
Asi
a
an
d
Su
b
-
Sah
ar
an
Af
r
ica,
w
h
er
e
it
is
d
if
f
icu
lt
to
o
b
tain
ex
p
e
r
ien
ce
d
h
ea
lth
ca
r
e
p
e
r
s
o
n
n
el
an
d
r
ad
io
lo
g
is
ts
wh
o
s
e
p
r
ed
ictio
n
s
o
n
s
u
ch
d
is
ea
s
es a
r
e
cr
u
cial
[
8
]
,
[
9
]
,
th
e
C
AD
tech
n
iq
u
e
ca
n
b
e
q
u
ite
b
e
n
ef
icia
l.
T
h
e
a
d
v
a
n
t
a
g
e
o
f
u
s
i
n
g
c
o
m
p
u
t
e
r
-
a
i
d
e
d
d
i
a
g
n
o
s
i
s
is
t
h
a
t
i
t
c
an
b
e
e
x
p
e
d
i
e
n
c
y
f
o
r
a
l
a
r
g
e
p
o
p
u
l
a
t
i
o
n
a
t
a
m
i
n
i
m
a
l
c
o
s
t
.
I
n
a
d
d
i
t
i
o
n
,
an
o
t
h
e
r
i
s
s
u
e
wi
t
h
p
n
e
u
m
o
n
i
a
d
e
t
e
c
t
i
o
n
is
t
h
a
t
i
n
s
o
m
e
c
a
s
es
t
h
e
f
e
a
t
u
r
es
t
h
a
t
d
e
s
c
r
i
b
e
t
h
is
d
i
s
e
as
e
c
a
n
b
e
m
i
x
e
d
w
i
t
h
f
e
a
t
u
r
e
s
o
f
o
t
h
e
r
d
i
s
e
a
s
es
l
i
k
e
m
i
d
d
l
e
E
as
t
r
es
p
i
r
a
t
o
r
y
s
y
n
d
r
o
m
e
(
M
E
R
S
)
a
n
d
C
O
V
I
D
-
1
9
.
D
ee
p
l
e
a
r
n
i
n
g
a
n
d
i
t
s
a
p
p
l
i
c
at
io
n
s
s
o
l
v
e
t
h
is
i
s
s
u
e
wi
t
h
h
ig
h
e
r
a
c
c
u
r
a
c
y
a
n
d
s
o
m
e
t
i
m
es
e
v
e
n
b
e
tt
e
r
t
h
a
n
e
x
p
e
r
t
r
a
d
i
o
l
o
g
i
s
t
s
d
o
[
1
0
]
.
I
n
th
e
m
ed
ical
im
ag
in
g
f
ield
,
th
e
q
u
ality
o
f
th
e
X
-
r
a
y
im
ag
es
to
d
iag
n
o
s
e
p
n
e
u
m
o
n
i
a
th
r
o
u
g
h
a
d
ee
p
lear
n
in
g
m
o
d
el
m
ak
e
s
a
d
if
f
er
en
ce
in
th
e
ac
cu
r
ac
y
an
d
th
e
p
r
ec
is
io
n
o
f
th
e
m
o
d
el
[
1
1
]
.
T
h
er
ef
o
r
e
,
in
t
h
is
p
ap
er
,
we
p
r
e
s
en
t
a
co
s
tu
m
e
tr
an
s
f
er
lear
n
i
n
g
ar
c
h
itectu
r
e
th
at
co
m
b
in
es
a
p
r
e
-
tr
ain
e
d
m
o
d
e
l
(
VGG1
9
)
an
d
o
u
r
class
if
icatio
n
m
o
d
el
with
a
wh
o
le
im
ag
e
p
r
e
p
r
o
ce
s
s
in
g
s
ec
tio
n
to
im
p
r
o
v
e
th
e
q
u
alit
y
o
f
ch
est
X
-
r
a
y
im
a
g
es
b
ef
o
r
e
f
ee
d
i
n
g
it
to
th
e
d
ee
p
lear
n
in
g
p
r
o
ce
s
s
e
s
.
T
h
is
p
r
ep
r
o
ce
s
s
in
g
ac
tio
n
h
elp
s
to
i
n
cr
ea
s
e
th
e
ac
cu
r
ac
y
o
f
p
n
e
u
m
o
n
ia
d
etec
tio
n
b
y
X
-
r
ay
f
ilm
s
.
2.
RE
L
AT
E
D
WO
RK
S
T
h
e
b
ig
g
est
ch
allen
g
e
f
o
r
d
o
cto
r
s
is
to
r
ed
u
ce
t
h
e
p
atien
ts
s
u
f
f
er
in
g
a
n
d
to
tr
ea
t
th
em
,
t
h
is
ch
allen
g
e
ca
n
b
e
o
v
er
c
o
m
e
o
n
l
y
if
th
ey
ca
n
m
ak
e
a
g
o
o
d
d
iag
n
o
s
is
an
d
in
ter
v
en
tio
n
,
f
r
o
m
th
is
p
o
in
t
th
e
in
teg
r
atio
n
o
f
th
e
au
to
m
ated
d
etec
tio
n
s
y
s
tem
s
an
d
co
m
p
u
ter
-
aid
e
d
d
iag
n
o
s
is
(
C
AD)
s
tar
ted
to
b
e
v
er
y
im
p
o
r
tan
t
a
n
d
v
e
r
y
u
s
ef
u
l
[
1
2
]
.
I
n
th
e
m
e
d
ical
im
ag
in
g
f
ield
,
th
e
im
p
lem
e
n
tatio
n
an
d
th
e
u
s
e
o
f
d
i
f
f
er
en
t
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
an
d
m
o
d
els
h
av
e
s
h
o
w
n
an
in
ter
esti
n
g
an
d
en
c
o
u
r
ag
in
g
r
esu
lts
.
So
m
e
o
f
th
e
p
o
wer
f
u
l
an
d
m
o
s
t
u
s
ed
d
ee
p
co
n
v
o
lu
ti
o
n
al
n
etwo
r
k
s
lik
e
r
esid
u
al
n
etwo
r
k
R
esNet
[
1
3
]
,
Xce
p
tio
n
[
1
4
]
,
I
n
ce
p
tio
n
[
1
5
]
,
VGG
[
1
6
]
,
De
n
s
eNe
t
[
1
7
]
r
e
ac
h
ed
o
v
er
9
5
%
ac
cu
r
ac
y
in
d
if
f
er
en
t
d
is
ea
s
es
d
etec
tio
n
s
u
ch
as
s
k
in
ca
n
ce
r
class
if
icatio
n
[
1
8
]
,
d
iab
etic
r
etin
o
p
at
h
y
d
etec
tio
n
[
1
9
]
,
ar
r
h
y
th
m
ia
d
etec
tio
n
[
2
0
]
,
an
d
h
em
o
r
r
h
ag
e
id
en
tific
atio
n
[
2
1
]
.
I
n
o
u
r
wo
r
k
,
we
f
o
cu
s
ed
o
n
d
e
v
elo
p
in
g
an
au
to
m
ated
d
etec
tio
n
s
y
s
tem
th
at
d
etec
t
s
th
e
p
n
eu
m
o
n
ia
d
i
s
ea
s
e
th
r
o
u
g
h
lu
n
g
X
-
r
a
y
f
ilm
s
s
in
ce
Pn
eu
m
o
n
ia
is
in
cr
ea
s
in
g
ly
b
ec
o
m
in
g
o
n
e
o
f
th
e
r
esear
ch
h
o
ts
p
o
ts
in
r
ec
en
t
y
ea
r
s
.
I
n
th
i
s
s
ec
tio
n
,
we
p
r
esen
t r
elate
d
w
o
r
k
s
th
at
ar
e
in
th
e
s
am
e
v
ein
:
I
n
2
0
1
8
,
Ok
e
k
e
et
a
l.
[
2
2
]
r
el
ea
s
ed
a
v
er
y
in
ter
esti
n
g
p
a
p
er
,
wh
er
e
th
ey
b
u
ilt
a
co
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
m
o
d
el
f
r
o
m
s
cr
atch
t
o
ex
tr
ac
t
f
ea
tu
r
es
f
r
o
m
a
g
iv
e
n
ch
est
X
-
r
ay
im
a
g
e.
T
h
eir
tech
n
iq
u
e
was
ab
le
t
o
m
itig
ate
th
e
d
ep
en
d
ab
ilit
y
an
d
in
ter
p
r
etab
ilit
y
is
s
u
es
th
at
ar
e
f
r
eq
u
en
tly
en
c
o
u
n
te
r
ed
wh
en
wo
r
k
in
g
with
m
ed
ical
im
ag
es.
I
n
2
0
2
0
,
Hash
m
i
et
a
l
.
[
2
3
]
p
u
b
lis
h
ed
an
e
f
f
icien
t
m
o
d
el
th
at
d
etec
t
s
p
n
e
u
m
o
n
ia
b
ased
o
n
a
weig
h
ted
class
if
ier
,
wh
ich
co
m
b
in
es
th
e
weig
h
ted
p
r
ed
ic
tio
n
s
f
r
o
m
d
if
f
e
r
en
t
p
r
e
-
tr
ain
ed
m
o
d
els
s
u
ch
as
R
esNet
,
Xc
ep
tio
n
an
d
Den
s
e
Net1
2
1
.
T
h
e
wo
r
k
p
r
esen
te
d
i
n
2
0
2
0
b
y
L
u
ján
-
Gar
cía
et
a
l
.
[
2
4
]
d
escr
ib
ed
th
e
u
s
e
o
f
th
e
tr
a
n
s
f
er
lear
n
in
g
a
n
d
th
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
e
x
ce
p
tio
n
m
o
d
el
t
o
d
etec
t
th
e
a
b
n
o
r
m
ality
in
ch
est
X
-
r
ay
im
ag
es.
L
i
v
i
e
r
is
e
t
a
l
.
p
r
e
s
e
n
te
d
i
n
2
0
1
9
[
2
5
]
a
s
e
m
i
-
s
u
p
e
r
v
is
e
d
l
ea
r
n
i
n
g
a
l
g
o
r
i
t
h
m
b
a
s
e
d
o
n
a
n
ew
w
e
i
g
h
t
e
d
v
o
t
i
n
g
s
c
h
e
m
e
.
T
h
e
y
p
r
o
v
e
d
t
h
a
t
t
h
e
w
e
i
g
h
t
s
o
f
m
o
d
e
l
s
m
a
k
e
a
d
i
f
f
e
r
e
n
ce
i
n
t
h
e
p
r
e
c
i
s
i
o
n
o
f
t
h
e
c
l
a
s
s
i
f
i
c
a
ti
o
n
p
r
o
c
e
s
s
.
I
n
2
0
2
0
,
Asn
ao
u
i
et
a
l
.
[
2
6
]
p
r
o
p
o
s
ed
an
ap
p
r
o
ac
h
b
ased
o
n
m
u
ltip
le
m
o
d
els
s
u
ch
as
VGG1
9
to
p
r
ed
ict
th
e
p
n
eu
m
o
n
ia
ex
is
ten
ce
in
th
e
lu
n
g
X
-
r
ay
f
ilm
s
.
T
h
ey
r
ea
ch
ed
an
ac
cu
r
ac
y
o
f
m
o
r
e
th
an
9
8
%.
Als
o
,
in
2
0
2
0
Ap
o
s
to
lo
p
o
u
lo
s
an
d
Mp
esian
a
[
2
7
]
p
r
o
v
ed
in
t
h
ei
r
p
a
p
er
t
h
at
th
e
u
s
e
o
f
a
p
r
e
-
t
r
ain
ed
m
o
d
els
as
a
f
ea
tu
r
e
ex
tr
ac
to
r
r
ev
ea
ls
a
g
o
o
d
p
er
f
o
r
m
an
ce
an
d
a
less
lo
s
e
ac
cu
r
ac
y
o
n
th
e
m
o
d
el
ev
alu
at
io
n
.
J
aiswal
et
a
l
.
I
n
th
eir
p
ap
er
r
elea
s
ed
in
2
0
1
9
[
2
8
]
em
p
lo
y
e
d
Ma
s
k
-
R
C
NN
in
a
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
to
id
en
tify
an
d
lo
ca
lize
p
n
eu
m
o
n
ia
in
lu
n
g
X
-
r
ay
f
i
lm
s
in
a
r
esear
ch
p
u
b
lis
h
ed
in
2
0
1
9
.
T
h
e
m
o
d
el'
s
ef
f
icac
y
a
n
d
r
esil
ien
ce
wer
e
d
em
o
n
s
tr
ated
b
y
its
g
o
o
d
p
er
f
o
r
m
an
ce
o
n
th
e
c
h
est
r
ad
i
o
g
r
a
p
h
y
d
ataset.
I
n
th
e
d
ee
p
lear
n
i
n
g
f
ield
esp
ec
ially
in
m
ed
ical
im
a
g
in
g
,
th
e
f
i
r
s
t
q
u
esti
o
n
th
at
co
m
es
to
m
in
d
is
:
wh
at
is
th
e
b
est
d
ee
p
lear
n
in
g
ar
ch
itectu
r
e?
A
d
d
itio
n
ally
,
h
o
w
we
u
s
e
it
to
g
et
th
e
b
est
p
er
f
o
r
m
an
ce
an
d
r
esu
lts
?
Fo
llo
win
g
th
e
co
n
te
x
t
o
f
o
b
ject
d
etec
tio
n
an
d
class
if
icatio
n
in
l
u
n
g
X
-
r
ay
f
ilm
s
,
we
p
r
esen
t
a
m
o
d
i
f
ied
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
with
a
co
m
p
le
te
im
ag
e
p
r
o
ce
s
s
in
g
s
ec
tio
n
d
ed
icate
d
to
im
p
r
o
v
e
th
e
q
u
alit
y
o
f
X
-
r
ay
im
ag
es
th
at
co
m
b
in
es a
p
r
e
-
tr
ain
e
d
m
o
d
el
(
VGG1
9
)
a
n
d
a
c
o
s
tu
m
e
m
o
d
el
as f
ea
tu
r
e
e
x
tr
ac
to
r
s
.
3.
DE
E
P
L
E
A
RNING
AP
P
RO
ACH
3
.
1
.
C
o
nv
o
lutio
na
l neura
l net
wo
rk
(
CNN)
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs
)
ar
e
a
ty
p
e
o
f
d
ee
p
f
ee
d
-
fo
r
war
d
ar
tific
ial
n
e
u
r
al
n
etwo
r
k
,
m
o
s
tly
u
s
ed
in
o
b
ject
d
etec
tio
n
an
d
im
a
g
e
class
if
icatio
n
,
th
e
C
NN
was
f
ir
s
t
u
s
ed
in
1
9
8
9
b
y
C
u
n
et
a
l
.
[
2
9
]
f
o
r
h
an
d
wr
itten
zip
co
d
e
r
ec
o
g
n
itio
n
,
th
eir
ap
p
lic
atio
n
was
ab
le
to
d
etec
t
an
d
ex
tr
ac
t
h
an
d
wr
itin
g
f
ea
tu
r
es
with
o
u
t
an
y
in
ter
v
e
n
tio
n
an
d
s
u
p
er
v
is
io
n
f
r
o
m
h
u
m
an
s
.
As
illu
s
tr
ated
in
Fig
u
r
e
1
,
th
e
o
u
tp
u
t
lay
er
o
f
a
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
is
m
ad
e
u
p
o
f
f
u
lly
co
n
n
ec
ted
lay
er
s
(
co
n
v
o
lu
tio
n
a
n
d
p
o
o
lin
g
o
p
e
r
atio
n
s
)
.
T
h
e
o
u
tp
u
t
lay
er
f
o
r
b
in
ar
y
class
i
f
icatio
n
is
a
s
ig
m
o
id
lay
er
,
an
d
m
o
s
t
C
NNs
u
s
e
r
esid
u
al
n
etwo
r
k
s
to
av
o
i
d
g
r
ad
ien
t d
is
ap
p
ea
r
an
ce
[
2
3
]
.
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
P
n
eu
mo
n
ia
d
etec
tio
n
b
a
s
ed
o
n
tr
a
n
s
fer lea
r
n
in
g
a
n
d
a
co
m
b
in
a
tio
n
o
f V
GG1
9
a
n
d
…
(
Ou
s
s
a
ma
Da
h
ma
n
e
)
1471
Fig
u
r
e
1
.
C
o
n
v
o
lu
tio
n
al
n
eu
r
a
l n
etwo
r
k
with
a
s
ig
m
o
id
lay
er
f
o
r
b
i
n
ar
y
p
r
ed
ictio
n
3.
2
.
VG
G
1
9
pre
-
t
r
a
ined m
o
del
VGG
-
1
9
is
a
1
9
-
l
a
y
e
r
d
e
e
p
co
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
et
w
o
r
k
c
r
e
a
t
e
d
b
y
S
i
m
o
n
y
a
n
a
n
d
Z
i
s
s
e
r
m
a
n
in
2015
[
1
6
]
.
A
s
i
l
l
u
s
t
r
a
t
e
d
i
n
F
i
g
u
r
e
2
,
t
h
e
n
u
m
b
e
r
1
9
r
e
f
e
r
s
t
o
th
e
n
u
m
b
e
r
o
f
l
a
y
e
r
s
w
it
h
t
r
a
i
n
ab
l
e
w
e
i
g
h
ts
,
w
h
ic
h
i
n
c
l
u
d
e
s
1
6
C
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
er
s
a
n
d
t
h
r
e
e
f
u
l
l
y
l
i
n
k
e
d
l
a
y
e
r
s
.
V
G
G
a
d
o
p
t
s
t
h
e
s
i
m
p
le
s
t
.
O
n
ly
3
×
3
c
o
n
v
o
l
u
t
i
o
n
a
n
d
2
×
2
p
o
o
l
i
n
g
is
u
s
e
d
t
h
r
o
u
g
h
o
u
t
t
h
e
w
h
o
l
e
n
e
t
w
o
r
k
.
V
G
G
a
l
s
o
s
h
o
ws
t
h
a
t
t
h
e
d
e
p
t
h
o
f
t
h
e
n
e
t
w
o
r
k
p
l
a
y
s
a
n
i
m
p
o
r
t
a
n
t
r
o
l
e
.
D
e
e
p
e
r
n
e
t
w
o
r
k
s
g
i
v
e
b
e
t
t
e
r
r
e
s
u
l
ts
.
O
n
e
d
r
aw
b
a
c
k
o
f
VG
G
is
t
h
at
t
h
i
s
n
e
tw
o
r
k
i
s
u
s
u
al
l
y
b
i
g
.
I
t
c
o
n
t
a
i
n
s
a
r
o
u
n
d
1
6
0
M
p
a
r
a
m
e
t
e
r
s
.
M
o
s
t
o
f
t
h
e
p
a
r
a
m
et
e
r
s
a
r
e
c
o
n
s
u
m
e
d
i
n
t
h
e
FC
la
y
e
r
s
.
Fig
u
r
e
2
.
VGG
-
1
9
n
etwo
r
k
m
o
d
el
ar
ch
itectu
r
e
3.
3
.
T
ra
ns
f
er
lea
rning
T
r
an
s
f
er
lear
n
in
g
is
th
e
u
s
e
o
f
a
m
o
d
el
tr
ai
n
ed
alr
ea
d
y
o
n
a
h
u
g
e
d
ataset
in
s
u
b
s
eq
u
e
n
t
tr
ain
in
g
p
r
o
ce
d
u
r
es
with
a
s
m
all
av
ail
ab
le
d
ataset
wh
ile
p
r
eser
v
in
g
th
e
weig
h
ts
o
b
tain
ed
in
t
h
e
o
r
ig
in
al
tr
ai
n
in
g
is
k
n
o
wn
as
tr
a
n
s
f
er
lear
n
in
g
.
Usi
n
g
a
p
r
e
-
tr
ai
n
ed
m
o
d
el
as
th
e
s
tar
tin
g
p
o
in
t
f
o
r
s
o
m
e
p
ar
ticu
lar
an
d
r
elate
d
task
s
[
3
0
]
,
i
n
an
o
th
er
ex
p
r
ess
io
n
,
h
el
p
s
to
s
av
e
tim
e
an
d
c
o
m
p
u
tin
g
r
eso
u
r
ce
s
.
Sin
ce
th
e
C
NN
lear
n
ed
to
ex
tr
ac
t
f
ea
tu
r
es f
r
o
m
im
ag
es
i
n
th
e
i
n
itial
tr
ain
in
g
p
r
o
ce
s
s
an
d
d
e
p
en
d
i
n
g
o
n
t
h
e
ca
p
a
b
ilit
y
to
e
x
tr
ac
t
th
e
m
o
s
t
s
ig
n
if
ican
t
an
d
im
p
o
r
tan
t
o
n
e
s
.
I
n
th
e
n
ex
t
p
h
ase
an
d
d
u
r
in
g
th
e
n
ew
tr
ain
in
g
th
e
C
NN
a
n
d
ac
co
r
d
in
g
to
its
p
ast
k
n
o
wled
g
e
in
f
ea
tu
r
es
ex
tr
ac
tio
n
wh
ich
was
o
b
tain
e
d
d
u
r
in
g
th
e
o
r
ig
in
al
tr
ain
in
g
,
th
e
r
e
ar
e
two
way
s
to
u
tili
ze
th
e
c
ap
a
b
ilit
ies
o
f
th
e
p
r
e
-
tr
ain
ed
C
NN.
T
h
e
f
ir
s
t
w
ay
is
to
u
s
e
th
e
p
r
e
-
tr
ain
ed
c
o
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
as
a
f
ea
tu
r
e
ex
tr
ac
to
r
,
th
is
tech
n
iq
u
e
is
ca
lled
f
ea
tu
r
e
ex
tr
ac
tio
n
v
ia
t
r
an
s
f
er
le
ar
n
in
g
[
3
0
]
,
wh
at
d
if
f
er
en
tiates
it
is
th
at
th
e
clas
s
if
icatio
n
o
p
er
a
tio
n
u
s
es
weig
h
s
an
d
f
ea
tu
r
es
f
r
o
m
th
e
p
r
ec
e
d
en
t
ex
tr
ac
tio
n
an
d
f
ee
d
it
in
to
a
n
ew
n
etwo
r
k
th
a
t
p
er
f
o
r
m
s
th
e
class
if
icatio
n
task
[
3
1
]
.
T
h
e
s
ec
o
n
d
way
(
u
s
ed
in
th
is
p
ap
e
r
)
an
d
a
m
o
r
e
s
o
p
h
is
ticated
p
r
o
ce
d
u
r
e
is
to
r
etain
s
p
ec
if
ic
k
n
o
wle
d
g
e
m
i
n
ed
f
r
o
m
th
e
p
r
ev
io
u
s
task
an
d
to
f
ee
d
it
in
to
a
m
o
d
if
ied
C
NN
ar
ch
itectu
r
e
with
th
e
tu
n
in
g
o
f
th
e
tr
ai
n
ab
le
p
ar
am
eter
s
[
3
1
]
.
4.
T
H
E
P
RO
P
O
SE
D
B
A
SE
L
I
NE
C
NN
W
h
en
C
u
n
et
a
l
.
f
ir
s
t
p
u
b
lis
h
ed
h
is
p
ap
er
s
[
2
9
]
in
1
9
8
9
wi
th
th
e
f
ir
s
t
u
s
e
o
f
C
NNs,
h
is
wo
r
k
wen
t
lar
g
ely
u
n
n
o
ticed
,
u
n
til
a
g
r
o
u
p
o
f
r
esear
ch
er
s
f
r
o
m
th
e
U
n
iv
er
s
ity
o
f
T
o
r
o
n
to
b
r
o
u
g
h
t
C
NN
to
lig
h
t
1
4
y
ea
r
s
later
d
u
r
in
g
a
co
m
p
u
ter
v
is
io
n
co
m
p
eti
tio
n
,
th
eir
wo
r
k
n
am
ed
Alex
n
et
m
ad
e
an
e
r
r
o
r
o
f
o
n
ly
1
5
.
8
%
in
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.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
4
6
9
-
1
4
8
0
1472
class
if
icatio
n
p
r
o
ce
s
s
es.
I
n
2
0
1
8
,
C
NNs
s
tar
ted
to
b
e
m
o
r
e
d
ev
elo
p
ed
an
d
h
a
d
r
ea
ch
ed
ac
c
u
r
ac
ies
th
at
s
u
r
p
ass
h
u
m
an
-
le
v
el
p
er
f
o
r
m
a
n
ce
[
3
2
]
.
Mo
tiv
ated
b
y
th
ese
in
ter
esti
n
g
r
esu
lts
,
we
s
et
o
u
t
to
e
x
p
lo
r
e
h
o
w
C
NNS
ca
n
b
e
b
u
ilt
f
r
o
m
s
cr
atch
a
n
d
h
o
w
th
ey
g
et
th
e
ab
ilit
y
to
r
ec
o
g
n
ize
p
atter
n
s
p
r
esen
ted
in
i
m
ag
es.
W
e
ca
m
e
to
b
u
ild
a
C
NN
u
s
ed
in
th
e
d
ete
ctio
n
o
f
p
n
eu
m
o
n
ia
with
an
in
ter
esti
n
g
ac
cu
r
ac
y
an
d
p
r
ec
is
i
o
n
.
Fig
u
r
e
3
s
h
o
ws
th
e
ar
ch
itectu
r
e
o
f
th
e
r
ec
o
m
m
en
d
ed
b
aselin
e
C
NN
f
o
r
o
u
r
e
x
p
er
im
en
t
.
Fig
u
r
e
3
.
T
h
e
p
r
o
p
o
s
ed
b
aselin
e
C
NN
f
o
r
o
u
r
ex
p
e
r
im
en
t
T
h
e
co
n
v
o
lu
tio
n
al
lay
er
(
C
o
n
v
L
)
is
th
e
m
ain
b
u
ild
in
g
b
lo
ck
o
f
a
co
n
v
o
lu
tio
n
al
n
etwo
r
k
th
at
d
o
es th
e
b
u
lk
o
f
th
e
h
ea
v
y
liftin
g
[
3
3
]
,
with
th
e
C
o
n
v
L
'
s
in
p
u
t
b
ein
g
th
e
o
u
tp
u
t
o
f
th
e
p
r
ec
ed
i
n
g
C
o
n
v
L
an
d
t
h
e
C
o
n
v
L
'
s
o
u
tp
u
t b
ein
g
t
h
e
o
u
t
p
u
t o
f
th
e
p
r
ec
ed
i
n
g
C
o
n
v
L
.
T
h
e
m
ath
em
atica
l e
x
p
r
ess
io
n
f
o
r
C
o
n
v
L
[
3
4
]
is
,
(
)
=
(
)
+
∑
,
(
)
(
−
1
)
=
1
∗
(
−
1
)
(
1
)
w
h
er
e
B
i
(
l
)
is
a
b
ias
m
atr
ix
an
d
K
i
,
j
(
l
)
is
th
e
f
ilter
o
f
s
ize
2ℎ
1
(
)
+
1
×
2ℎ
2
(
)
+
1
co
n
n
ec
tin
g
th
e
ℎ
f
ea
tu
r
e
m
ap
in
lay
er
(
−1
)
with
ℎ
f
ea
tu
r
e
m
ap
i
n
lay
er
.
W
e
u
tili
ze
d
m
ax
p
o
o
lin
g
la
y
er
s
af
ter
th
e
C
o
n
v
L
b
ec
a
u
s
e
th
e
o
u
tp
u
t
f
ea
tu
r
e
m
ap
s
o
b
t
ain
ed
ar
e
ty
p
ically
n
o
t
s
u
f
f
icien
tly
d
ec
r
ea
s
ed
in
d
im
en
s
io
n
;
th
er
eb
y
,
m
ax
p
o
o
lin
g
lay
er
s
is
th
e
o
n
ly
way
t
o
g
et
a
r
ea
s
o
n
ab
le
o
u
tc
o
m
e
an
d
r
ed
u
ce
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
t
h
at
th
e
n
etwo
r
k
n
ee
d
s
to
lear
n
with
o
u
t
ch
a
n
g
in
g
th
e
n
u
m
b
er
o
f
f
ea
tu
r
e
g
r
a
p
h
s
.
T
h
e
m
ax
im
u
m
v
alu
e
is
ca
p
tu
r
ed
in
th
e
p
o
o
lin
g
lay
e
r
,
with
a
s
am
p
lin
g
s
ize
o
f
2
×
2
[
3
5
]
.
W
e
ad
d
ed
a
f
latten
la
y
er
o
n
t
o
p
o
f
th
at,
f
o
llo
wed
b
y
f
o
u
r
f
u
lly
lin
k
ed
lay
er
s
.
Fu
lly
co
n
n
ec
ted
la
y
er
s
ar
e
u
s
ed
f
o
r
h
i
g
h
-
lev
el
r
ea
s
o
n
in
g
.
A
co
m
p
letely
lin
k
e
d
lay
e
r
'
s
n
eu
r
o
n
s
ar
e
co
n
n
ec
ted
to
a
ll
ac
tiv
atio
n
s
in
th
e
p
r
ev
io
u
s
lay
er
,
f
o
r
m
in
g
a
m
u
lti
-
lay
er
p
er
ce
p
tr
o
n
(
ML
P)
th
at
ac
ts
as
a
clas
s
if
ier
[
3
5
]
.
I
n
ad
d
itio
n
,
d
r
o
p
o
u
ts
wer
e
u
s
ed
t
o
r
e
d
u
ce
o
v
er
-
f
itti
n
g
.
E
x
ce
p
t
f
o
r
th
e
last
lay
e
r
,
wh
ich
was
Sig
m
o
id
as
a
b
i
n
ar
y
class
if
icatio
n
p
r
o
b
lem
,
Ad
am
as th
e
o
p
tim
iz
er
,
an
d
c
r
o
s
s
-
en
tr
o
p
y
as th
e
lo
s
s
,
th
e
ac
tiv
atio
n
f
u
n
ctio
n
was ReL
u.
5.
T
H
E
P
RO
P
O
SE
D
M
E
T
H
O
D
I
n
th
is
s
ec
tio
n
,
we
d
escr
ib
e
th
e
p
r
o
p
o
s
ed
m
eth
o
d
th
at
d
etec
t
s
th
e
ex
is
ten
ce
o
f
p
n
e
u
m
o
n
ia
i
n
ch
est
X
-
r
ay
f
ilm
s
b
ased
o
n
d
ee
p
lear
n
i
n
g
an
d
tr
an
s
f
er
lear
n
i
n
g
.
W
e
d
ed
icate
d
th
e
f
ir
s
t p
ar
t o
f
th
e
o
p
er
atio
n
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
th
e
d
ataset
b
y
th
e
a
p
p
licatio
n
o
f
co
n
tr
ast
lim
ited
ad
ap
tiv
e
h
is
to
g
r
am
e
q
u
aliza
tio
n
(
C
L
AHE
)
[
3
5
]
an
d
to
ad
ju
s
t
th
e
b
r
ig
h
tn
ess
b
y
u
s
in
g
th
e
b
r
ig
h
tn
ess
p
r
eser
v
in
g
b
i
-
h
is
t
o
g
r
am
eq
u
aliza
tio
n
(
B
B
HE
)
[
3
6
]
.
W
e
ch
o
s
e
th
e
VGG1
9
o
n
to
p
o
f
o
u
r
m
o
d
el
b
ec
au
s
e
it
g
en
er
ally
p
r
o
d
u
ce
s
a
g
r
ea
t
p
er
f
o
r
m
an
ce
in
d
etec
tin
g
a
b
n
o
r
m
ality
in
m
ed
ical
im
ag
in
g
.
Af
ter
th
at,
we
u
s
ed
th
e
o
u
tp
u
t
o
f
th
e
VGG1
9
as
an
in
p
u
t
in
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
.
5
.
1
.
D
a
t
a
s
et
T
h
e
o
r
ig
i
n
al
ch
est
X
-
r
ay
d
at
aset
was
f
ir
s
t
r
elea
s
ed
o
n
J
u
n
e
1
s
t,
2
0
1
8
b
y
Ker
m
a
n
y
et
a
l
.
o
f
th
e
Un
iv
er
s
ity
o
f
C
ali
f
o
r
n
ia,
San
Dieg
o
[
3
7
]
.
Gu
a
n
g
zh
o
u
W
o
m
en
an
d
C
h
ild
r
en
'
s
Me
d
ical
C
e
n
ter
in
Gu
a
n
g
zh
o
u
,
C
h
in
a,
p
r
o
v
i
d
ed
th
e
d
ata
f
o
r
th
is
d
ataset.
T
h
e
d
ataset
is
d
iv
id
ed
in
to
t
h
r
ee
s
u
b
f
o
ld
e
r
s
:
tr
ain
,
test
,
an
d
v
alid
atio
n
,
with
p
n
e
u
m
o
n
ia
a
n
d
n
o
r
m
al
as
s
u
b
f
o
l
d
er
s
in
ea
ch
o
f
th
e
s
e
f
o
ld
er
s
.
T
h
e
c
o
llectio
n
co
n
tain
s
5
,
8
5
6
an
ter
io
r
-
p
o
s
ter
io
r
c
h
est
X
-
r
ay
s
ca
n
s
,
in
clu
d
in
g
4
,
2
7
3
im
ag
es
o
f
p
n
e
u
m
o
n
ia
p
atien
ts
a
n
d
1
,
5
8
3
h
ea
lth
y
p
e
o
p
le.
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
P
n
eu
mo
n
ia
d
etec
tio
n
b
a
s
ed
o
n
tr
a
n
s
fer lea
r
n
in
g
a
n
d
a
co
m
b
in
a
tio
n
o
f V
GG1
9
a
n
d
…
(
Ou
s
s
a
ma
Da
h
ma
n
e
)
1473
T
o
co
n
tr
o
l
th
e
tr
ain
s
p
lit
an
d
test
s
p
lit
v
alu
es,
we
m
er
g
ed
th
e
tr
ain
,
test
,
an
d
v
alid
atio
n
d
ir
ec
to
r
ies
in
to
a
s
in
g
le
d
ir
ec
to
r
y
.
W
e
wer
e
ab
l
e
to
s
ep
ar
ate
th
e
d
ataset
in
to
6
0
%
tr
ain
,
3
0
%
test
,
an
d
1
0
%
v
alid
atio
n
p
r
o
ce
s
s
es
u
s
in
g
Py
th
o
n
'
s
s
k
lear
n
p
ac
k
a
g
e.
5
.
2
.
CL
AH
E
a
nd
B
B
H
E
C
o
n
tr
ast
lim
ited
ad
ap
tiv
e
h
is
t
o
g
r
am
eq
u
aliza
tio
n
(
C
L
AHE
)
is
a
tec
h
n
iq
u
e
th
at
u
s
es
lo
ca
l
co
n
tr
ast
en
h
an
ce
m
e
n
t
to
o
v
er
co
m
e
th
e
co
n
s
tr
ain
ts
o
f
g
lo
b
al
ap
p
r
o
ac
h
es.
C
L
AHE
is
wid
ely
em
p
lo
y
ed
in
t
h
e
m
e
d
ical
im
ag
in
g
in
d
u
s
tr
y
,
p
ar
ticu
la
r
ly
f
o
r
b
r
ea
s
t
ca
n
ce
r
a
n
d
m
a
m
m
o
g
r
ap
h
y
im
ag
e
e
n
h
an
ce
m
e
n
t
[
3
5
]
,
[
3
6
]
.
I
t
is
u
s
ed
to
im
p
r
o
v
e
p
i
ctu
r
e
co
n
tr
ast
in
a
v
ar
iety
o
f
c
o
m
p
u
ter
v
is
io
n
a
n
d
p
atter
n
r
ec
o
g
n
itio
n
ap
p
lica
tio
n
s
.
T
h
is
m
et
h
o
d
is
k
n
o
wn
as
th
e
clip
lim
it
[
3
5
]
,
an
d
it
is
u
s
ed
to
clip
th
e
h
is
to
g
r
am
at
a
p
r
ed
eter
m
in
ed
v
alu
e
in
o
r
d
e
r
to
r
estrict
co
n
tr
ast am
p
lific
atio
n
b
ef
o
r
e
c
o
m
p
u
tin
g
th
e
C
DF v
alu
e.
B
B
HE
is
an
ex
ten
s
io
n
o
f
h
i
s
to
g
r
am
eq
u
aliza
tio
n
(
HE
)
-
b
ased
co
n
tr
ast
en
h
an
ce
m
e
n
t
t
h
at
av
o
id
s
h
is
to
g
r
am
eq
u
aliza
tio
n
'
s
f
laws,
s
u
ch
as
th
e
b
r
ig
h
tn
ess
lo
s
s
.
T
h
e
p
r
eser
v
in
g
b
i
-
h
is
to
g
r
am
eq
u
aliza
tio
n
tech
n
iq
u
e
d
iv
id
es
th
e
i
n
p
u
t
im
ag
e
h
is
to
g
r
am
in
to
two
s
u
b
-
im
ag
es,
eq
u
alize
s
th
e
h
is
to
g
r
am
s
o
f
th
e
s
u
b
-
im
a
g
es
in
d
iv
id
u
ally
,
an
d
th
e
r
eb
y
p
r
eser
v
es
th
e
im
ag
e'
s
m
ea
n
b
r
ig
h
t
n
ess
[
3
6
]
-
[
3
8
]
.
I
s
t
h
e
m
ea
n
o
f
I
(
in
p
u
t
im
ag
e
)
,
wh
er
e
∊
{
0
,
1
,
…
,
−
1
}
,
b
ased
o
n
th
at,
I
is
d
ec
o
m
p
o
s
ed
as sh
o
wn
in
(
2
)
,
I=
∪
(
2
)
w
h
er
e
= {(
,
)
|
(
,
)
≤
,
∀
(
,
)
∈
}
(
3
)
a
nd
= {
(
,
)
|
(
,
)
>
,
∀
(
,
)
∈
}
(
4
)
5
.
3
.
D
a
t
a
s
et
prepro
ce
s
s
ing
a
nd
a
ug
m
ent
a
t
io
n
T
h
e
s
tr
en
g
th
o
f
d
ee
p
lear
n
in
g
is
th
at
wh
en
ev
er
th
e
d
ataset
is
b
ig
th
e
p
r
ec
is
io
n
o
f
th
e
lear
n
in
g
g
ets
b
etter
.
W
e
em
p
lo
y
ed
a
b
u
n
ch
o
f
d
ata
p
r
ep
r
o
ce
s
s
in
g
an
d
au
g
m
en
tatio
n
tech
n
iq
u
es
to
p
r
o
d
u
ce
a
n
ew
s
im
p
le
f
r
o
m
th
e
av
ailab
le
o
n
es
an
d
i
n
cr
ea
s
e
th
e
q
u
ality
o
f
t
h
e
d
ata
s
et
i
n
o
r
d
er
t
o
av
o
i
d
o
v
er
f
itti
n
g
an
d
u
n
d
e
r
f
itti
n
g
.
I
n
T
ab
le
1
,
we
cite
th
e
s
ettin
g
s
d
ep
lo
y
ed
in
im
a
g
e
au
g
m
e
n
tatio
n
,
an
d
F
ig
u
r
e
4
s
h
o
ws
th
e
r
esu
lt
o
f
th
e
d
ata
au
g
m
en
tatio
n
.
T
ab
le
1
.
T
h
e
im
ag
e
a
u
g
m
en
tat
io
n
s
ettin
g
s
M
e
t
h
o
d
S
e
t
t
i
n
g
s
D
e
scri
p
t
i
o
n
R
e
sc
a
l
e
1
/
2
5
5
I
mag
e
r
e
d
u
c
t
i
o
n
d
u
r
i
n
g
t
h
e
a
u
g
m
e
n
t
a
t
i
o
n
p
r
o
c
e
ss
Zo
o
m ra
n
g
e
0
.
0
5
S
a
mp
l
e
a
s
e
c
t
i
o
n
f
r
o
m
t
h
e
o
r
i
g
i
n
a
l
i
m
a
g
e
.
t
h
e
n
r
e
si
z
e
t
h
i
s
sec
t
i
o
n
t
o
t
h
e
o
r
i
g
i
n
a
l
i
ma
g
e
si
z
e
R
o
t
a
t
i
o
n
r
a
n
g
e
25
R
a
n
d
o
m
l
y
r
o
t
a
t
e
t
h
e
i
ma
g
e
d
u
r
i
n
g
t
r
a
i
n
i
n
g
i
n
2
5
d
e
g
r
e
e
s
W
i
d
t
h
sh
i
f
t
r
a
n
g
e
0
.
0
5
Th
e
h
o
r
i
z
o
n
t
a
l
t
r
a
n
s
l
a
t
i
o
n
o
f
t
h
e
i
ma
g
e
s b
y
0
.
0
5
%
H
e
i
g
h
t
s
h
i
f
t
r
a
n
g
e
0
.
0
5
Th
e
v
e
r
t
i
c
a
l
t
r
a
n
sl
a
t
i
o
n
o
f
t
h
e
i
m
a
g
e
s
b
y
0
.
0
5
%
S
h
e
a
r
r
a
n
g
e
0
.
0
5
C
l
i
p
s
t
h
e
i
ma
g
e
a
n
g
l
e
s i
n
a
c
o
u
n
t
e
r
c
l
o
c
k
w
i
s
e
d
i
r
e
c
t
i
o
n
H
o
r
i
z
o
n
t
a
l
f
l
i
p
Tr
u
e
F
l
i
p
t
h
e
i
m
a
g
e
h
o
r
i
z
o
n
t
a
l
l
y
Fig
u
r
e
4
.
T
h
e
r
esu
lt o
f
d
ata
au
g
m
en
tatio
n
5.
4
.
T
he
s
ug
g
este
d CN
N's
o
v
er
a
ll a
rc
hite
ct
ure
Fig
u
r
e
5
d
ep
icts
th
e
p
r
o
p
o
s
ed
C
NN
m
o
d
el's
o
v
er
all
ar
ch
itectu
r
e,
wh
ich
is
d
iv
id
ed
in
to
t
h
r
ee
p
ar
ts
:
i
)
th
e
im
ag
es
p
r
o
ce
s
s
in
g
p
ar
t
u
s
es
b
i
-
h
is
to
g
r
am
e
q
u
aliza
tio
n
(
B
B
HE
)
an
d
co
n
tr
ast
lim
ited
ad
ap
tiv
e
h
is
to
g
r
a
m
eq
u
aliza
tio
n
(
C
L
AHE
)
alg
o
r
it
h
m
s
,
ii
)
th
e
f
ea
tu
r
e
ex
t
r
ac
to
r
p
ar
t
u
s
es
a
co
m
b
in
atio
n
o
f
a
p
r
e
-
tr
ain
ed
m
o
d
el
(
VGG1
9
)
,
an
d
iii
)
o
u
r
d
esig
n
e
d
m
o
d
el
(
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
)
.
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.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
4
6
9
-
1
4
8
0
1474
Fig
u
r
e
5
.
T
h
e
p
r
o
p
o
s
ed
a
r
ch
it
ec
tu
r
e
6.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
Ou
r
test
s
wer
e
b
ased
o
n
a
d
at
aset
o
f
ch
est
X
-
r
ay
im
a
g
es
p
r
o
p
o
s
ed
i
n
[
3
7
]
.
T
o
cr
ea
te
a
n
d
tr
ain
th
e
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
m
o
d
els,
we
u
s
ed
Ker
as,
an
o
p
en
-
s
o
u
r
ce
d
ee
p
lea
r
n
in
g
f
r
am
ewo
r
k
with
a
T
en
s
o
r
f
lo
w
b
ac
k
e
n
d
[
3
9
]
.
All
ex
p
er
im
en
ts
wer
e
ca
r
r
ied
o
u
t
u
s
in
g
a
wo
r
k
s
tatio
n
PC
eq
u
ip
p
ed
with
a
2
GB
Nv
id
ia
Qu
ad
r
o
k
2
0
0
0
m
GPU
ca
r
d
,
th
e
c
u
DNN
v
9
.
0
lib
r
ar
y
,
th
e
C
UDA
t
o
o
lk
it 1
0
.
1
,
a
n
d
P
y
th
o
n
3
.
7
.
6
.
1
.
T
he
cho
ice
o
f
clip
-
lim
it
(
CL
)
v
a
lue
T
o
d
ef
in
e
d
th
e
b
est
clip
lim
it
v
alu
e
we
d
esig
n
e
d
a
s
im
p
le
C
NN
m
o
d
el
f
r
o
m
s
cr
atch
a
n
d
v
ar
ied
th
e
C
L
f
r
o
m
0
to
1
.
5
an
d
r
u
n
th
e
tr
ain
i
n
g
,
an
d
f
o
r
ea
c
h
v
ar
iatio
n
th
e
ac
cu
r
ac
y
r
esu
lts
ar
e
s
h
o
wn
in
Fig
u
r
e
6
w
e
ch
o
s
e
0
.
5
b
ec
au
s
e
it
lead
s
to
t
h
e
h
ig
h
ac
c
u
r
ac
y
v
al
u
e.
Fig
u
r
e
7
an
d
Fig
u
r
e
8
s
h
o
w
a
c
o
m
p
ar
is
o
n
b
ef
o
r
e
an
d
af
ter
th
e
ap
p
licatio
n
o
f
th
e
C
L
AHE
.
Af
ter
th
e
ap
p
licatio
n
o
f
th
e
C
L
AHE
we
a
p
p
licate
B
B
HE
to
th
e
r
esu
lts
im
ag
es a
s
s
h
o
wn
in
Fig
u
r
e
9
a
n
d
Fig
u
r
e
1
0
.
Fig
u
r
e
6
.
T
h
e
v
ar
iatio
n
o
f
ac
c
u
r
ac
y
f
o
r
ea
ch
C
L
Fig
u
r
e
7
.
T
h
e
ap
p
licatio
n
o
f
C
L
=0
.
5
o
n
n
o
r
m
al
ch
est
X
-
r
ay
Fig
u
r
e
8
.
T
h
e
ap
p
licatio
n
o
f
C
L
=0
.
5
o
n
in
f
ec
te
d
ch
est
X
-
r
ay
Fig
u
r
e
9
.
T
h
e
ap
p
licatio
n
o
f
B
B
HE
o
n
n
o
r
m
al
c
h
est
X
-
r
ay
Fig
u
r
e
1
0
.
T
h
e
a
p
p
licatio
n
o
f
B
B
HE
o
n
in
f
ec
ted
ch
est
X
-
r
ay
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
P
n
eu
mo
n
ia
d
etec
tio
n
b
a
s
ed
o
n
tr
a
n
s
fer lea
r
n
in
g
a
n
d
a
co
m
b
in
a
tio
n
o
f V
GG1
9
a
n
d
…
(
Ou
s
s
a
ma
Da
h
ma
n
e
)
1475
6
.
2
.
Cla
s
s
if
ica
t
io
n r
esu
lt
s
o
f
t
he
pro
po
s
ed
a
rc
hite
ct
ures
Af
ter
lo
ad
in
g
o
u
r
d
ataset,
we
s
p
lit
it
in
to
3
p
ar
ts
,
6
0
%
f
o
r
tr
ain
in
g
,
3
0
%
f
o
r
test
in
g
an
d
1
0
%
f
o
r
v
alid
atin
g
th
e
r
esu
lts
,
n
ex
t,
we
r
esh
ap
e
all
im
a
g
es to
1
2
5
x
1
2
5
x
3
to
f
it in
to
th
e
n
etwo
r
k
.
Af
ter
th
at,
we
d
ef
in
ed
th
e
n
u
m
b
er
o
f
class
es
(
2
)
,
th
e
B
AT
C
H
s
ize
(
3
2
)
an
d
E
POC
H
(
2
5
)
,
th
o
s
e
two
p
a
r
am
e
ter
s
d
ep
e
n
d
o
n
th
e
wo
r
k
s
tatio
n
ab
ilit
y
,
n
ex
t
we
a
p
p
licate
th
e
d
ata
au
g
m
en
tatio
n
alg
o
r
ith
m
to
p
r
o
d
u
ce
n
ew
tr
ai
n
in
g
s
am
p
les,
a
f
ter
th
at,
we
class
if
y
th
e
attain
ed
d
ata
an
d
ass
ig
n
it
to
a
s
p
ec
if
ic
class
th
r
o
u
g
h
th
r
ee
C
NNs
s
ep
ar
ately
.
W
e
u
s
e
th
e
f
ir
s
t two
ex
p
er
im
en
ts
to
c
o
m
p
ar
e
th
em
with
o
u
r
d
esig
n
e
d
ar
ch
itectu
r
e
(
3
r
d
ex
p
er
im
en
t)
.
Fo
r
ev
alu
atin
g
t
h
e
th
r
ee
C
NN’
s
p
er
f
o
r
m
a
n
ce
we
u
s
ed
t
h
e
f
o
llo
win
g
b
en
c
h
m
ar
k
m
etr
ics,
in
clu
d
in
g
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
an
d
F1
s
co
r
e
[
4
0
]
.
T
h
ese
p
o
p
u
lar
p
ar
a
m
eter
s
ar
e
d
ef
in
ed
as
f
o
llo
ws,
Acc
u
r
ac
y
=
TP
+
TN
TN
+
TP
+
FP
+
FN
(
5
)
Pre
cisi
o
n
=
TP
TP
+
FP
(
6
)
R
ec
all=
TP
TP
+
FN
(
7
)
F1
=
2
×
Recal
l
×
Pr
ecis
i
o
n
Recal
l
+
Pr
ecis
i
o
n
(
8
)
w
h
er
e:
T
P: T
r
u
e
p
o
s
itiv
e
,
FP
: False
p
o
s
itiv
e
,
T
N:
T
r
u
e
n
eg
at
iv
e
,
an
d
FN:
Fals
e
n
eg
ativ
e
.
6
.
3
.
VG
G
1
9
a
s
f
ea
t
ure
e
x
t
r
a
ct
o
r
I
n
th
e
f
ir
s
t
tr
ial,
we
u
s
ed
th
e
p
r
e
-
tr
ain
ed
m
o
d
el
VGG1
9
,
f
r
ee
zin
g
th
e
co
n
v
o
lu
tio
n
b
lo
ck
s
to
u
s
e
it
as
an
im
ag
e
f
ea
t
u
r
e
ex
tr
ac
t
o
r
,
a
n
d
th
en
in
jectin
g
in
o
u
r
o
wn
d
e
n
s
e
lay
er
s
to
ac
co
m
p
lis
h
th
e
c
lass
if
icatio
n
task
at
th
e
en
d
,
th
e
o
b
tain
ed
r
esu
lts
ar
e
p
r
esen
te
d
in
Fig
u
r
e
1
1
.
T
h
e
tr
ain
in
g
cu
r
v
e,
wh
ich
r
e
p
r
esen
ts
h
o
w
ef
f
ec
tiv
el
y
th
e
m
o
d
el
is
lear
n
in
g
,
is
ca
lc
u
lated
f
r
o
m
th
e
tr
ain
in
g
d
ataset,
wh
ile
th
e
v
alid
atio
n
cu
r
v
e,
wh
ich
r
ev
ea
ls
h
o
w
well
th
e
m
o
d
el
is
m
ak
i
n
g
g
e
n
er
aliza
tio
n
s
,
is
ca
lcu
lated
f
r
o
m
a
h
o
l
d
o
u
t
v
alid
a
tio
n
d
ataset.
W
e
ca
n
tell
if
th
e
m
o
d
el
is
o
v
er
f
itted
,
u
n
d
er
f
itte
d
,
o
r
g
o
o
d
f
it b
ased
o
n
th
ese
t
wo
cu
r
v
es.
Acc
o
r
d
in
g
to
th
e
ac
c
u
r
ac
y
c
u
r
v
es,
it
is
clea
r
th
at
th
e
tr
ain
ac
cu
r
ac
y
cu
r
v
e
in
cr
ea
s
es
r
ap
id
ly
f
r
o
m
ep
o
ch
1
to
ep
o
c
h
1
1
a
n
d
s
ta
b
ilizes
ab
o
v
e
9
7
%,
s
im
ilar
ly
to
th
e
v
alid
atio
n
c
u
r
v
e
ex
ce
p
t
s
o
m
e
p
er
t
u
r
b
atio
n
f
r
o
m
ep
o
ch
1
to
ep
o
ch
6
an
d
th
en
it
s
tab
ilizes
ar
o
u
n
d
9
8
%.
Fro
m
ep
o
ch
1
to
e
p
o
ch
1
2
,
wh
er
e
th
e
l
o
s
s
is
ar
o
u
n
d
1
0
%,
th
e
tr
ain
lo
s
s
cu
r
v
e
r
ap
id
ly
d
ec
r
ea
s
es
b
ef
o
r
e
b
ec
o
m
in
g
m
o
r
e
s
tab
le
u
n
til
th
e
last
ep
o
ch
.
Af
ter
a
p
er
tu
r
b
atio
n
f
r
o
m
th
e
f
ir
s
t
to
t
h
e
s
ix
th
ep
o
c
h
,
it
co
n
tin
u
es
to
g
et
m
o
r
e
s
tab
le
u
n
til
th
e
e
n
d
o
f
tr
ain
in
g
an
d
h
as
less
th
an
1
0
%
v
alid
atio
n
lo
s
s
,
s
im
ilar
to
th
e
v
alid
atio
n
lo
s
s
.
T
ab
le
2
lis
ts
th
e
m
o
d
el'
s
p
e
r
f
o
r
m
a
n
ce
m
etr
ics,
wh
ile
Fig
u
r
e
1
2
d
ep
icts
th
e
m
o
d
el'
s
p
r
ed
ictio
n
co
n
f
u
s
io
n
m
atr
ix
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
s
h
o
ws
th
at
th
e
m
o
d
el
p
r
ed
icted
1
,
2
4
0
co
r
r
ec
t c
ases
o
u
t o
f
1
,
2
7
7
ca
s
es
in
f
ec
ted
wit
h
p
n
eu
m
o
n
ia,
an
d
p
r
e
d
icted
4
4
2
co
r
r
ec
t c
ases
o
u
t
o
f
4
8
0
h
ea
lth
y
ca
s
es,
an
d
it c
a
n
b
e
co
n
clu
d
e
d
th
at
th
ese
ar
e
g
o
o
d
r
esu
lts
.
(
a)
(
b
)
F
ig
u
r
e
1
1
.
T
h
e
(
a
)
a
cc
u
r
ac
y
an
d
(
b
)
l
o
s
s
cu
r
v
e
o
f
VGG1
9
as
a
f
ea
tu
r
e
ex
tr
ac
t
o
r
T
ab
le
2
.
Mo
d
el
p
er
f
o
r
m
a
n
ce
m
etr
ics o
f
VGG1
9
as f
ea
tu
r
e
e
x
tr
ac
to
r
p
r
e
c
i
si
o
n
r
e
c
a
l
l
f1
-
sco
r
e
p
n
e
u
mo
n
i
a
0
.
9
7
0
.
9
7
0
.
9
7
h
e
a
l
t
h
y
0
.
9
2
0
.
9
2
0
.
9
2
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.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
4
6
9
-
1
4
8
0
1476
Fig
u
r
e
1
2
.
C
o
n
f
u
s
io
n
m
atr
ix
f
o
r
VGG1
9
as a
f
ea
tu
r
e
ex
tr
ac
t
o
r
6
.
4
.
VG
G
1
9
f
ine t
un
ed
W
e
f
in
e
-
t
u
n
ed
th
e
weig
h
ts
o
f
th
e
lay
er
s
d
is
p
lay
ed
in
t
h
e
las
t
two
b
l
o
ck
s
o
f
o
u
r
p
r
e
-
tr
ain
e
d
VGG
-
1
9
m
o
d
el
in
o
u
r
s
ec
o
n
d
ex
p
e
r
im
en
t.
T
h
e
tr
ain
an
d
v
alid
ati
o
n
cu
r
v
es
ar
e
s
h
o
wn
in
Fig
u
r
e
1
3
d
u
r
in
g
tr
ain
in
g
.
Fro
m
th
e
ac
cu
r
ac
y
cu
r
v
es,
th
e
tr
ain
ac
cu
r
ac
y
s
o
ar
ed
f
r
o
m
e
p
o
ch
1
to
1
1
th
en
s
tar
ted
to
s
t
ab
ilize
u
n
til
ep
o
ch
2
5
,
u
n
lik
e
th
e
v
alid
atio
n
ac
cu
r
ac
y
cu
r
v
e
th
at
was
u
n
s
ettled
d
u
r
in
g
th
e
wh
o
le
tr
ain
in
g
.
On
t
h
e
lo
s
s
cu
r
v
es,
we
o
b
s
er
v
ed
th
at
th
e
r
ain
lo
s
s
d
ec
r
ea
s
ed
r
ap
id
ly
f
r
o
m
e
p
o
c
h
1
to
1
3
a
n
d
th
e
n
s
tar
ted
to
s
tab
ilize,
b
u
t
th
e
v
alid
atio
n
lo
s
s
cu
r
v
e
a
f
ter
s
o
m
e
p
er
tu
r
b
atio
n
d
u
r
in
g
th
e
f
ir
s
t
ten
ep
o
c
h
s
tar
ted
to
in
c
r
e
ase
u
n
til
th
e
e
n
d
o
f
tr
ain
in
g
wh
ic
h
m
ea
n
s
th
at
th
e
m
o
d
el
s
tar
ted
to
o
v
e
r
-
f
it.
T
ab
le
3
p
r
esen
ts
th
e
m
o
d
el
p
er
f
o
r
m
an
ce
m
etr
ics
a
n
d
Fig
u
r
e
1
4
p
r
esen
ts
th
e
p
r
ed
ic
tio
n
co
n
f
u
s
io
n
m
atr
i
x
,
t
he
co
n
f
u
s
io
n
m
atr
ix
s
h
o
ws
th
at
th
e
m
o
d
el
p
r
ed
icted
co
r
r
ec
tly
1
,
2
4
2
ca
s
es
o
f
p
n
eu
m
o
n
ia
o
u
t
o
f
1
,
2
7
7
ca
s
es
an
d
p
r
ed
icted
4
6
1
ca
s
es
o
u
t
o
f
4
8
0
h
ea
lth
y
ca
s
es
b
u
t
th
e
m
o
d
el
was o
v
e
r
f
itti
n
g
.
(
a)
(
b
)
Fig
u
r
e
1
3
.
T
h
e
(
a
)
a
cc
u
r
ac
y
an
d
(
b
)
l
o
s
s
cu
r
v
e
o
f
VGG1
9
f
in
e
-
tu
n
ed
Fig
u
r
e
1
4
.
C
o
n
f
u
s
io
n
m
atr
ix
f
o
r
VGG1
9
f
in
e
-
tu
n
e
d
T
ab
le
3
.
Mo
d
el
p
er
f
o
r
m
a
n
ce
m
etr
ics o
f
VGG1
9
f
in
e
d
-
tu
n
e
d
p
re
c
isio
n
re
c
a
ll
f1
-
sc
o
re
p
n
e
u
m
o
n
ia
0
.
9
8
0
.
9
7
0
.
9
8
h
e
a
lt
h
y
0
.
9
3
0
.
9
6
0
.
9
4
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
P
n
eu
mo
n
ia
d
etec
tio
n
b
a
s
ed
o
n
tr
a
n
s
fer lea
r
n
in
g
a
n
d
a
co
m
b
in
a
tio
n
o
f V
GG1
9
a
n
d
…
(
Ou
s
s
a
ma
Da
h
ma
n
e
)
1477
6
.
5
.
P
ro
po
s
ed
a
rc
hite
ct
ure
W
e
em
p
lo
y
ed
th
e
C
L
AHE
an
d
B
B
HE
ap
p
r
o
ac
h
es
to
im
p
r
o
v
e
th
e
d
ataset
q
u
ality
b
ef
o
r
e
th
e
tr
ain
in
g
p
r
o
ce
s
s
es,
an
d
th
en
we
m
er
g
e
d
th
e
VGG1
9
p
r
e
-
tr
ain
ed
m
o
d
el
with
o
u
r
cr
ea
ted
m
o
d
el
to
e
x
tr
ac
t
f
ea
tu
r
es
an
d
d
etec
t
p
n
eu
m
o
n
ia
in
th
e
X
-
r
a
y
p
ictu
r
es.
Du
r
in
g
th
e
tr
ai
n
in
g
,
Fig
u
r
e
1
5
d
ep
icts
th
e
tr
ain
an
d
th
e
v
alid
atio
n
cu
r
v
e.
B
o
th
o
f
th
e
ac
cu
r
ac
y
c
u
r
v
es
in
cr
ea
s
ed
r
ap
id
ly
i
n
th
e
f
ir
s
t
1
0
ep
o
ch
s
,
an
d
th
en
s
tab
ilized
u
n
til
th
e
en
d
o
f
th
e
tr
ain
i
n
g
ar
o
u
n
d
9
9
%.
in
th
e
lo
s
s
cu
r
v
es
th
e
tr
ain
lo
s
s
d
ec
r
ea
s
ed
r
ap
id
ly
f
r
o
m
e
p
o
ch
1
to
6
,
th
en
s
tab
ilized
ar
o
u
n
d
0
%,
f
u
r
th
er
m
o
r
e,
th
e
v
alid
atio
n
lo
s
s
cu
r
v
e
in
cr
ea
s
ed
f
r
o
m
1
to
6
e
p
o
ch
,
at
th
at
p
o
i
n
t
th
e
cu
r
v
e
g
o
t
d
o
w
n
to
ar
o
u
n
d
0
%
s
im
ilar
ly
to
th
e
tr
ain
lo
s
s
cu
r
v
e
wh
ich
m
ea
n
s
th
at
we
h
av
e
a
g
o
o
d
f
it.
T
a
b
le
4
p
r
esen
ts
th
e
m
o
d
el
p
er
f
o
r
m
an
ce
m
etr
ics an
d
Fig
u
r
e
1
6
p
r
esen
ts
th
e
p
r
ed
ictio
n
co
n
f
u
s
io
n
m
atr
ix
,
(
a)
(
b
)
F
ig
u
r
e
1
5
.
Acc
u
r
ac
y
an
d
l
o
s
s
cu
r
v
e
o
f
th
e
p
r
o
p
o
s
ed
ar
ch
itec
tu
r
e
T
ab
le
4
.
Mo
d
el
p
er
f
o
r
m
a
n
ce
m
etr
ics o
f
th
e
p
r
o
p
o
s
ed
ar
ch
it
ec
tu
r
e
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
f1
-
sco
r
e
p
n
e
u
mo
n
i
a
0
.
9
9
0
.
9
9
0
.
9
9
h
e
a
l
t
h
y
0
.
9
9
0
.
9
9
0
.
9
9
Fig
u
r
e
1
6
.
C
o
n
f
u
s
io
n
m
atr
ix
f
o
r
VGG1
9
f
in
e
-
tu
n
e
d
T
ab
le
5
.
C
o
m
p
a
r
is
o
n
b
etwe
en
th
e
o
b
tain
ed
r
esu
lts
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
S
c
o
r
e
V
G
G
1
9
a
s fe
a
t
u
r
e
e
x
t
r
a
c
t
o
r
0
.
9
5
7
3
0
.
9
7
0
3
0
.
9
7
1
0
0
.
9
7
0
6
VGG
-
1
9
F
i
n
e
-
t
u
n
e
d
0
.
9
6
9
3
0
.
9
8
4
9
0
.
9
7
2
6
0
.
9
7
8
7
P
r
o
p
o
se
d
M
e
t
h
o
d
o
l
o
g
y
0
.
9
9
8
9
0
.
9
9
8
4
1
.
0
0
.
9
9
9
2
T
h
e
co
n
f
u
s
io
n
m
atr
ix
s
h
o
ws
th
at
th
e
m
o
d
el
p
r
e
d
icted
co
r
r
ec
tly
1
,
2
7
7
ca
s
es
o
f
p
n
e
u
m
o
n
ia
o
u
t
o
f
1
,
2
7
7
ca
s
es
an
d
p
r
ed
icted
4
7
8
ca
s
es
o
u
t
o
f
4
8
0
h
ea
lth
y
ca
s
es,
wh
ich
in
d
icate
s
v
er
y
g
o
o
d
r
esu
lts
.
T
h
e
im
p
r
o
v
em
e
n
t
p
r
o
ce
s
s
es
th
r
o
u
g
h
th
e
a
p
p
licatio
n
o
f
th
e
c
o
n
tr
ast
lim
ited
ad
ap
ti
v
e
h
is
to
g
r
am
eq
u
aliza
tio
n
(
C
L
AHE
)
an
d
th
e
b
r
ig
h
t
n
ess
p
r
eser
v
in
g
b
i
-
h
is
to
g
r
am
e
q
u
aliza
tio
n
(
B
B
HE
)
m
ak
e
a
h
u
g
e
d
if
f
er
en
ce
in
th
e
o
p
er
atio
n
o
f
tr
ain
in
g
,
an
d
m
a
k
e
th
e
ess
en
tial
f
ea
tu
r
es
clea
r
to
ex
tr
ac
t,
wh
ich
lead
s
to
b
etter
p
r
ec
is
io
n
,
as
s
h
o
wn
b
y
th
e
o
b
tain
e
d
cu
r
v
es
an
d
co
n
f
u
s
io
n
m
atr
ices.
T
o
d
e
m
o
n
s
tr
ate
th
e
e
f
f
icien
cy
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
,
T
ab
le
5
s
h
o
ws
th
e
f
in
d
in
g
s
ac
h
iev
ed
a
n
d
c
o
m
p
ar
es
th
em
to
th
e
two
f
ir
s
t
ap
p
r
o
ac
h
es,
wh
il
e
T
ab
le
6
s
h
o
ws
a
co
m
p
ar
is
o
n
o
f
o
u
r
m
et
h
o
d
t
o
o
th
er
h
ig
h
-
ac
cu
r
ac
y
s
y
s
tem
s
alr
ea
d
y
av
ailab
le.
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.
24
,
No
.
3
,
Dec
em
b
er
2
0
2
1
:
1
4
6
9
-
1
4
8
0
1478
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
,
wh
ich
co
n
tain
s
th
e
co
m
b
in
ati
o
n
o
f
o
u
r
C
NN
an
d
th
e
VGG
1
9
,
o
b
tain
e
d
h
ig
h
v
al
u
es
o
f
p
r
ec
is
io
n
(
0
.
9
9
8
4
)
,
r
ec
all
(
1
.
0
)
a
n
d
F1
s
co
r
e
(
0
.
9
9
9
2
)
,
th
u
s
,
we
ca
n
s
ay
th
a
t
th
e
ac
cu
r
ac
y
r
atio
o
f
o
u
r
m
o
d
el
r
ep
r
esen
ted
b
y
(
0
.
9
9
8
9
)
ca
n
b
e
co
n
s
id
er
ed
th
e
h
ig
h
est
co
m
p
ar
ed
to
th
e
d
if
f
er
e
n
t
e
x
is
tin
g
m
eth
o
d
s
.
Mo
r
eo
v
er
,
th
e
u
s
e
o
f
two
C
NNs
a
llo
ws
th
e
n
eu
r
al
n
etwo
r
k
th
e
p
o
s
s
ib
ilit
y
to
ex
tr
ac
t
m
o
r
e
f
ea
tu
r
es
co
m
p
ar
ed
to
th
e
u
s
e
o
f
o
n
e
C
NN.
I
n
a
d
d
itio
n
,
th
e
tr
an
s
f
er
l
ea
r
n
in
g
tech
n
iq
u
e
g
iv
es
th
e
n
etwo
r
k
th
e
ab
ilit
y
t
o
u
s
e
th
e
weig
h
ts
o
b
tain
ed
f
r
o
m
a
b
ig
d
ataset
(
I
m
ag
eNe
t)
to
d
etec
t
p
n
eu
m
o
n
ia
f
r
o
m
X
-
r
ay
im
ag
es.
Ou
r
ar
ch
itectu
r
e
clea
r
ly
o
u
t
p
er
f
o
r
m
s
th
e
o
th
er
s
o
lu
tio
n
s
af
ter
th
e
co
m
p
ar
ativ
e
p
r
o
ce
d
u
r
e.
A
s
a
r
esu
lt,
we
ca
n
co
n
clu
d
e
t
h
at
o
u
r
m
eth
o
d
is
m
o
r
e
r
eliab
le
an
d
ef
f
ec
tiv
e
.
T
ab
le
6
.
C
o
m
p
a
r
is
o
n
o
f
th
e
o
b
tain
ed
r
esu
lts
with
a
cc
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
an
d
F1
s
co
r
e
c
o
r
r
esp
o
n
d
in
g
t
o
d
if
f
er
en
t a
r
ch
itectu
r
es
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
S
c
o
r
e
M
o
b
i
l
e
N
e
t
V
2
9
6
.
7
1
9
6
.
0
8
9
8
.
2
5
9
7
.
1
5
D
e
n
seN
e
t
1
2
1
9
8
.
0
0
9
7
.
5
3
9
9
.
0
0
9
8
.
2
6
I
n
c
e
p
t
i
o
n
V
3
9
7
.
0
0
9
7
.
0
2
9
7
.
7
5
9
7
.
3
9
X
c
e
p
t
i
o
n
9
6
.
5
7
9
5
.
8
5
9
8
.
2
5
9
7
.
0
3
W
e
i
g
h
t
e
d
C
l
a
ssi
f
i
e
r
9
8
.
4
3
9
8
.
2
6
9
9
.
0
0
9
8
.
6
3
P
r
o
p
o
se
d
M
e
t
h
o
d
o
l
o
g
y
0
.
9
9
8
9
0
.
9
9
8
4
1
.
0
0
.
9
9
9
2
7.
CO
NCLU
SI
O
N
W
e
o
f
f
er
ed
a
p
n
eu
m
o
n
ia
d
ete
ctio
n
m
eth
o
d
b
ased
o
n
a
VGG1
9
an
d
a
C
NN
b
u
ilt
f
r
o
m
t
h
e
g
r
o
u
n
d
u
p
to
ca
teg
o
r
ize
ch
est
X
-
r
ay
f
il
m
s
in
to
two
class
es:
n
o
r
m
al
an
d
p
n
eu
m
o
n
ia
in
t
h
is
wo
r
k
.
W
e
u
s
ed
a
n
o
p
en
-
s
o
u
r
ce
d
ataset
with
4
,
2
7
3
im
ag
es
f
r
o
m
p
n
e
u
m
o
n
ia
p
ati
en
ts
an
d
1
,
68
3
im
ag
es
f
r
o
m
h
ea
lth
y
p
eo
p
le;
we
im
p
r
o
v
e
d
th
e
im
a
g
es'
q
u
ality
b
y
r
u
n
n
in
g
th
em
th
r
o
u
g
h
two
alg
o
r
ith
m
s
,
th
e
C
L
AHE
an
d
t
h
e
B
B
HE
,
an
d
th
en
d
iv
id
ed
th
e
d
ataset
in
to
th
r
ee
p
ar
ts
:
6
0
% f
o
r
tr
ain
in
g
,
3
0
% f
o
r
test
in
g
,
an
d
1
0
%
f
o
r
v
alid
at
io
n
.
W
e
u
s
ed
a
d
ata
au
g
m
en
tat
io
n
tech
n
iq
u
e
to
p
r
o
d
u
ce
m
o
r
e
d
ata
f
o
r
th
e
n
etwo
r
k
.
W
e
co
m
p
ar
ed
th
e
r
esu
lts
to
t
h
o
s
e
o
b
tain
ed
u
s
in
g
o
th
e
r
ap
p
r
o
ac
h
es.
T
h
e
p
r
ec
is
io
n
,
r
ec
al
l,
f
1
s
co
r
e,
an
d
ac
cu
r
ac
y
o
f
ea
c
h
m
eth
o
d
wer
e
ca
lcu
lated
,
an
d
we
co
n
clu
d
ed
f
r
o
m
th
ese
r
esu
lts
th
at
co
m
b
in
i
n
g
q
u
ality
i
m
p
r
o
v
em
e
n
t
tech
n
iq
u
es
s
u
ch
as
C
L
AHE
an
d
B
B
HE
wi
th
th
e
co
m
b
in
atio
n
o
f
two
C
NNs
p
r
o
d
u
ce
s
a
r
o
b
u
s
t
an
d
ac
cu
r
ate
d
etec
tio
n
an
d
c
lass
if
icatio
n
s
y
s
tem
wi
th
a
h
ig
h
p
r
ec
is
io
n
.
W
e
e
x
p
e
c
t
t
h
a
t,
i
n
t
h
e
f
u
t
u
r
e
,
o
u
r
t
r
a
i
n
e
d
n
e
t
w
o
r
k
w
i
l
l
b
e
u
s
e
f
u
l
f
o
r
m
e
d
i
c
a
l
d
i
a
g
n
o
s
t
i
cs
a
n
d
p
n
e
u
m
o
n
i
a
d
e
t
e
c
t
i
o
n
,
p
a
r
t
i
c
u
l
a
r
l
y
i
n
d
e
v
e
l
o
p
i
n
g
n
a
t
i
o
n
s
.
W
e
al
s
o
e
x
p
e
c
t
t
h
at
a
d
d
i
t
i
o
n
a
l
d
a
ta
s
et
s
o
f
p
n
e
u
m
o
n
i
a
p
a
t
i
e
n
ts
w
i
ll
b
e
c
o
m
e
a
v
a
il
a
b
l
e
i
n
t
h
e
f
u
t
u
r
e
,
a
l
l
o
w
i
n
g
u
s
t
o
i
m
p
r
o
v
e
t
h
e
a
c
cu
r
a
c
y
o
f
o
u
r
s
u
g
g
e
s
t
e
d
m
o
d
e
l
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
p
r
o
ject
N°
A2
5
N0
1
UN0
8
0
1
2
0
1
9
0
0
0
1
o
f
th
e
Alg
er
i
an
Min
is
tr
y
o
f
Hig
h
er
E
d
u
c
atio
n
an
d
R
esear
ch
(
PR
FU)
i
s
h
elp
in
g
to
f
u
n
d
th
is
r
esear
ch
.
RE
F
E
R
E
NC
E
S
[1
]
S
.
J
o
h
n
so
n
a
n
d
D
.
Wells,
“
Vira
l
P
n
e
u
m
o
n
ia,
S
y
m
p
t
o
m
s,
Ris
k
F
a
c
to
rs,
a
n
d
M
o
re
,
”
2
0
1
9
.
[On
li
n
e
].
Av
a
i
lab
le:
h
tt
p
s:/
/www
.
h
e
a
lt
h
li
n
e
.
c
o
m
/h
e
a
lt
h
/v
iral
-
p
n
e
u
m
o
n
ia
[2
]
C
.
D
M
a
th
e
rs,
T
.
Bo
e
rm
a
,
a
n
d
D
.
M
.
F
a
t
,
“
G
lo
b
a
l
a
n
d
Re
g
io
n
a
l
Ca
u
se
s
o
f
De
a
th
,
”
Pa
t
ter
n
s
a
n
d
T
re
n
d
s
,
v
o
l.
9
2
,
no.
1
,
2
0
1
7
,
d
o
i:
1
0
.
1
0
9
3
/b
m
b
/l
d
p
0
2
8
.
[3
]
He
a
lt
h
c
a
re
,
Un
iv
e
rsity
o
f
Uta
h
,
“
P
n
e
u
m
o
n
ia
M
a
k
e
s
Li
st
fo
r
To
p
1
0
Ca
u
se
s
o
f
De
a
th
,
”
2
0
1
6
.
[O
n
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
s:/
/h
e
a
lt
h
c
a
re
.
u
ta
h
.
e
d
u
/t
h
e
sc
o
p
e
/sh
o
ws
.
p
h
p
?
sh
o
ws
=
0
_
r
iw4
wti
7
[4
]
P
.
Ru
i
a
n
d
K.
Ka
n
g
,
“
Na
ti
o
n
a
l
Am
b
u
lato
r
y
M
e
d
ica
l
Ca
re
S
u
rv
e
y
:
2
0
1
7
Eme
rg
e
n
c
y
De
p
a
rtme
n
t
S
u
m
m
a
ry
Tab
les
,
”
[5
]
W
.
S
.
Li
m
,
D
.
L
.
S
m
it
h
,
M
.
P
.
Wi
se
,
a
n
d
S
.
A
.
Welh
a
m
,
“
Brit
ish
Th
o
ra
c
ic
S
o
c
iet
y
g
u
i
d
e
li
n
e
s
fo
r
th
e
m
a
n
a
g
e
m
e
n
t
o
f
c
o
m
m
u
n
i
ty
-
a
c
q
u
ired
p
n
e
u
m
o
n
ia
in
a
d
u
lt
s,”
BM
J
J
o
u
rn
a
ls
,
v
o
l
.
6
6
,
n
o
.
2
,
p
p
.
1
-
2
3
,
2
0
0
9
,
d
o
i:
1
0
.
1
1
3
6
/t
h
o
ra
x
jn
l
-
2
0
1
1
-
2
0
0
5
9
8
.
[6
]
K
.
Ka
ll
ian
o
s
e
t
a
l
.
,
“
Ho
w
fa
r
h
a
v
e
we
c
o
m
e
?
Artifi
c
ial
i
n
telli
g
e
n
c
e
fo
r
c
h
e
st
ra
d
io
g
ra
p
h
in
terp
re
t
a
ti
o
n
,
”
C
li
n
ic
a
l
ra
d
io
lo
g
y
,
v
o
l.
7
5
,
n
o
.
5
,
p
p
.
3
3
8
-
3
4
5
,
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/
j.
c
ra
d
.
2
0
1
8
.
1
2
.
0
1
5
.
[7
]
N
.
Li
u
,
L
.
Wan
,
Y
.
Z
h
a
n
g
,
T
.
Z
h
o
u
,
H
.
Hu
o
a
n
d
T
.
F
a
n
g
,
“
Ex
p
lo
i
ti
n
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
wit
h
d
e
e
p
ly
lo
c
a
l
d
e
sc
rip
t
io
n
f
o
r
re
m
o
t
e
se
n
sin
g
ima
g
e
c
las
sifica
ti
o
n
,
”
IEE
E
Acc
e
ss
2
0
1
8
,
v
o
l
.
6
,
p
p
.
1
1
2
1
5
-
1
1
2
2
8
,
2
0
1
8
,
d
o
i:
1
0
.
1
1
0
9
/ACCES
S
.
2
0
1
8
.
2
7
9
8
7
9
9
.
[8
]
A.
M
.
Tah
ir
e
t
a
l
.
,
“
A
sy
ste
m
a
ti
c
a
p
p
ro
a
c
h
to
th
e
d
e
sig
n
a
n
d
c
h
a
ra
c
teriz
a
ti
o
n
o
f
a
sm
a
rt
in
s
o
le
fo
r
d
e
t
e
c
ti
n
g
v
e
rti
c
a
l
g
ro
u
n
d
re
a
c
ti
o
n
f
o
rc
e
(v
G
RF
)
in
g
a
it
a
n
a
ly
sis,”
S
e
n
so
rs
2
0
2
0
,
v
o
l.
2
0
,
n
o
.
4
,
p
p
.
1
,
2
0
2
0
,
d
o
i:
1
0
.
3
3
9
0
/s2
0
0
4
0
9
5
7
.
[9
]
M. E
.
H.
C
h
o
w
d
h
u
ry
e
t
a
l
.
,
“
Re
a
l
-
Ti
m
e
S
m
a
rt
-
Dig
it
a
l
S
teth
o
sc
o
p
e
S
y
ste
m
fo
r
He
a
rt
Dise
a
se
s M
o
n
it
o
rin
g
,
”
S
e
n
s
o
rs
2
0
1
9
,
v
o
l.
1
9
,
n
o
.
1
2
,
p
.
2
7
8
1
,
2
0
1
9
,
d
o
i:
1
0
.
3
3
9
0
/s
1
9
1
2
2
7
8
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.