I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
11
,
No
.
5
,
Octo
b
er
2
0
2
1
,
p
p
.
4
5
3
1
~
4
5
4
1
I
SS
N:
2
0
8
8
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/
ijece
.
v
1
1
i
5
.
p
p
4
5
3
1
-
4
5
4
1
4531
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Deep
l
ea
rni
ng
f
o
r
COVI
D
-
19
d
ia
g
n
o
sis
b
a
sed o
n
c
hes
t
X
-
r
ay
i
ma
g
es
Na
s
ha
t
Alre
f
a
i
1
,
O
t
hm
a
n Ib
ra
him
2
1,
2
S
c
h
o
o
l
o
f
C
o
m
p
u
t
in
g
,
F
a
c
u
l
ty
o
f
En
g
i
n
e
e
rin
g
,
Un
i
v
e
rsiti
Tek
n
o
l
o
g
i
M
a
lay
sia
,
8
1
3
1
0
UTM
J
o
h
o
r
Ba
h
ru
,
M
a
lay
sia
2
Az
m
a
n
Ha
sh
im
In
tern
a
ti
o
n
a
l
Bu
sin
e
ss
S
c
h
o
o
l,
U
n
iv
e
rsiti
Tek
n
o
lo
g
i
M
a
lay
sia
(UT
M
),
S
k
u
d
a
i,
J
o
h
o
r
8
1
3
1
0
,
M
a
lay
sia
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
2
0
,
2
0
2
0
R
ev
is
ed
Mar
16
,
2
0
2
1
Acc
ep
ted
Ma
r
2
9
,
2
0
2
1
Co
ro
n
a
v
iru
s
d
ise
a
se
2
0
1
9
(COV
I
D
-
1
9
)
is
a
re
c
e
n
t
g
l
o
b
a
l
p
a
n
d
e
m
ic
th
a
t
h
a
s
a
ffe
c
ted
m
a
n
y
c
o
u
n
tri
e
s
a
ro
u
n
d
t
h
e
wo
rld
,
c
a
u
sin
g
se
rio
u
s
h
e
a
lt
h
p
ro
b
lem
s,
e
sp
e
c
ially
in
th
e
l
u
n
g
s.
Alt
h
o
u
g
h
tem
p
e
ra
tu
re
tes
ti
n
g
is
su
g
g
e
ste
d
a
s
a
first
-
li
n
e
tes
t
fo
r
COV
ID
-
1
9
,
it
wa
s
n
o
t
re
li
a
b
le
b
e
c
a
u
se
m
a
n
y
d
ise
a
se
s
h
a
v
e
th
e
sa
m
e
sy
m
p
to
m
s.
T
h
u
s,
we
p
r
o
p
o
se
a
d
e
e
p
lea
rn
in
g
m
e
th
o
d
b
a
se
d
o
n
X
-
ra
y
ima
g
e
s
th
a
t
u
se
d
a
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
(CNN
)
a
n
d
tran
sf
e
r
lea
rn
in
g
(TL
)
fo
r
COV
ID
-
1
9
d
iag
n
o
sis,
a
n
d
u
sin
g
g
ra
d
ien
t
-
we
i
g
h
te
d
c
las
s
a
c
ti
v
a
ti
o
n
m
a
p
p
in
g
(G
ra
d
-
CAM)
tec
h
n
i
q
u
e
fo
r
p
ro
d
u
c
i
n
g
v
isu
a
l
e
x
p
lan
a
ti
o
n
s
fo
r
th
e
COV
ID
-
1
9
i
n
fe
c
ti
o
n
a
re
a
in
t
h
e
lu
n
g
.
Th
e
lo
w
sa
m
p
le
siz
e
o
f
c
o
ro
n
a
v
iru
s
sa
m
p
les
wa
s
c
o
n
sid
e
re
d
a
c
h
a
ll
e
n
g
e
,
th
u
s,
th
is
issu
e
wa
s
o
v
e
rrid
d
e
n
u
sin
g
d
a
ta
a
u
g
m
e
n
tatio
n
tec
h
n
iq
u
e
s.
Th
e
stu
d
y
f
o
u
n
d
th
a
t
t
h
e
p
ro
p
o
se
d
(
CNN
)
a
n
d
th
e
m
o
d
ifi
e
d
p
re
-
train
e
d
n
e
two
rk
s
VG
G
1
6
a
n
d
In
c
e
p
ti
o
n
V3
a
c
h
iev
e
d
a
p
ro
m
isin
g
re
su
lt
fo
r
COV
ID
-
1
9
d
iag
n
o
sis
b
y
u
si
n
g
c
h
e
s
t
X
-
ra
y
i
m
a
g
e
s.
Th
e
p
ro
p
o
se
d
CNN
wa
s
a
b
le
t
o
d
iff
e
re
n
ti
a
te
2
8
4
p
a
ti
e
n
ts
wit
h
CO
VID
-
1
9
o
r
n
o
rm
a
l
wit
h
9
8
.
2
p
e
rc
e
n
t
f
o
r
tr
a
in
in
g
a
c
c
u
ra
c
y
a
n
d
9
6
.
6
6
p
e
rc
e
n
t
f
o
r
tes
t
a
c
c
u
ra
c
y
a
n
d
1
0
0
.
0
p
e
rc
e
n
t
se
n
s
it
iv
it
y
.
Th
e
m
o
d
ifi
e
d
VG
G
1
6
a
c
h
iev
e
d
th
e
b
e
st
c
las
sifica
ti
o
n
r
e
su
lt
b
e
twe
e
n
a
ll
with
1
0
0
.
0
p
e
rc
e
n
t
fo
r
train
in
g
a
c
c
u
ra
c
y
a
n
d
9
8
.
3
3
p
e
rc
e
n
t
f
o
r
tes
t
a
c
c
u
ra
c
y
a
n
d
1
0
0
.
0
p
e
rc
e
n
t
se
n
siti
v
it
y
,
b
u
t
th
e
p
ro
p
o
se
d
CNN
o
v
e
rc
a
m
e
th
e
o
th
e
rs
in
t
h
e
sid
e
o
f
re
d
u
c
in
g
t
h
e
c
o
m
p
u
tatio
n
a
l
c
o
m
p
lex
it
y
a
n
d
train
i
n
g
ti
m
e
sig
n
ifi
c
a
n
tl
y
.
Ke
ywo
r
d
s
:
C
h
est X
-
r
ay
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
C
OVI
D
-
19
Dee
p
lear
n
in
g
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
:
Nash
at
Alr
ef
ai
Sch
o
o
l o
f
C
o
m
p
u
tin
g
,
Facu
lty
o
f
E
n
g
in
ee
r
in
g
Un
iv
er
s
iti T
ek
n
o
lo
g
i M
ala
y
s
ia
8
1
3
1
0
UT
M
J
o
h
o
r
B
ah
r
u
,
J
o
h
o
r
,
Ma
lay
s
ia
E
m
ail:
am
n
ash
at@
g
r
ad
u
ate.
u
t
m
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
n
o
v
el
C
o
r
o
n
av
ir
u
s
2
0
1
9
o
r
C
OVI
D
-
1
9
is
a
p
an
d
em
ic
th
at
was
f
ir
s
t
an
n
o
u
n
ce
d
in
W
u
h
an
,
C
h
in
a
in
Dec
em
b
er
2
0
1
9
[
1
]
.
T
h
e
cli
n
ical
s
y
m
p
to
m
s
o
f
C
OVI
D
-
1
9
in
clu
d
e
r
esp
ir
ato
r
y
s
y
m
p
t
o
m
s
,
h
ig
h
tem
p
er
atu
r
e,
co
u
g
h
,
d
y
s
p
n
ea
,
an
d
v
ir
al
p
n
eu
m
o
n
ia
[
2
]
,
[
3
]
.
Do
cto
r
s
also
d
iag
n
o
s
ed
p
n
eu
m
o
n
ia,
l
u
n
g
in
f
lam
m
atio
n
,
ab
s
ce
s
s
es
an
d
en
lar
g
ed
ly
m
p
h
n
o
d
es
b
y
X
-
r
ay
s
an
d
C
T
s
ca
n
s
.
T
h
e
s
p
r
ea
d
o
f
C
OVI
D
-
1
9
h
as
in
c
r
ea
s
ed
in
s
ev
er
al
co
u
n
tr
ies
ar
o
u
n
d
th
e
wo
r
ld
,
th
u
s
,
n
ew
au
to
m
ate
d
an
d
ef
f
ec
tiv
e
m
eth
o
d
s
o
f
ac
c
u
r
ate
d
etec
tio
n
ar
e
n
ec
ess
ar
y
in
o
r
d
er
to
r
ed
u
ce
t
h
e
o
u
tb
r
ea
k
o
f
th
e
p
a
n
d
em
ic.
Dete
ctio
n
o
f
C
OVI
D
-
1
9
b
y
u
s
in
g
X
-
r
ay
im
a
g
es
is
s
ig
n
if
ican
t
f
o
r
d
ia
g
n
o
s
is
,
ev
alu
atio
n
,
an
d
tr
ea
tm
en
t.
B
ec
au
s
e
C
OVI
D
-
1
9
tar
g
ets
t
h
e
e
p
ith
elial
ce
lls
lin
in
g
o
u
r
r
esp
ir
ato
r
y
tr
ac
t,
t
h
e
h
ea
lth
o
f
a
p
atien
t'
s
lu
n
g
s
ca
n
b
e
ex
am
in
e
d
b
y
X
-
r
ay
s
[
4
]
.
All
h
o
s
p
itals
ar
e
eq
u
ip
p
ed
with
X
-
r
ay
im
ag
in
g
d
e
v
ices;
X
-
r
ay
s
co
u
l
d
b
e
u
s
ed
to
tes
t
C
OVI
D
-
1
9
.
T
h
e
o
b
s
tacle
is
th
at
X
-
r
ay
an
al
y
s
es
n
ee
d
a
s
p
ec
ialis
t
in
r
ad
i
o
lo
g
y
a
n
d
c
o
n
s
u
m
e
co
n
s
id
er
ab
le
tim
e,
wh
ich
is
a
v
alu
ab
le
co
m
m
o
d
ity
esp
ec
ially
wh
en
th
er
e
ar
e
m
an
y
p
atie
n
ts
in
th
e
h
o
s
p
ital
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t
J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
5
3
1
-
4
5
4
1
4532
d
u
r
in
g
th
e
p
an
d
e
m
ic.
T
h
u
s
,
it
is
n
ec
ess
ar
y
to
d
e
v
elo
p
a
co
m
p
u
ter
ized
d
etec
tio
n
a
n
d
d
iag
n
o
s
is
m
eth
o
d
to
s
av
e
v
alu
ab
le
tim
e
f
o
r
ex
p
er
ts
in
th
e
m
ed
ical
f
ield
.
New
f
in
d
in
g
s
s
u
g
g
est
th
at
C
T
im
ag
es
ar
e
m
o
r
e
h
elp
f
u
l
th
a
n
X
-
r
ay
s
f
o
r
C
OVI
D
-
1
9
d
iag
n
o
s
is
,
b
u
t
it
is
co
s
tly
.
Mo
r
eo
v
er
,
it
is
ea
s
y
to
h
av
e
an
X
-
r
ay
im
ag
e,
b
u
t
it
i
s
d
if
f
icu
lt
to
d
iag
n
o
s
e
f
r
o
m
it.
T
h
er
ef
o
r
e
,
ex
p
er
tis
e
is
n
ee
d
ed
.
Ou
r
p
r
o
p
o
s
ed
m
eth
o
d
h
as
th
e
ab
ilit
y
t
o
d
iag
n
o
s
e
an
y
n
ew
X
-
r
ay
im
ag
e
with
p
r
o
m
is
in
g
r
esu
lts
an
d
h
ig
h
p
er
f
o
r
m
an
ce
co
m
p
ar
ed
to
th
e
tr
a
d
itio
n
al
d
ia
g
n
o
s
is
p
r
o
ce
d
u
r
e.
T
h
is
s
tu
d
y
f
ir
s
t
ex
p
lo
r
es
th
e
s
tate
-
of
-
th
e
-
ar
t
s
o
lu
tio
n
s
to
f
i
g
h
t
C
OVI
D
-
1
9
.
T
h
en
,
we
co
n
s
tr
u
cted
a
p
r
ep
r
o
ce
s
s
ed
an
d
in
clu
s
iv
e
d
ataset
o
n
m
u
lti
-
s
o
u
r
ce
X
-
r
ay
s
im
ag
es
an
d
d
eliv
er
a
p
r
ec
is
e
C
OVI
D
-
1
9
d
etec
tio
n
tech
n
iq
u
e
u
s
in
g
d
ee
p
lear
n
in
g
an
d
tr
a
n
s
f
er
lear
n
in
g
tech
n
iq
u
es.
T
r
an
s
f
er
lear
n
in
g
is
u
s
ed
wid
ely
in
d
ee
p
lear
n
in
g
if
th
er
e
is
a
s
m
all
d
ataset,
wh
er
e
a
m
o
d
el
is
r
eu
s
ed
f
o
r
a
p
ar
ticu
lar
task
to
an
o
th
er
task
.
T
h
e
tr
an
s
f
er
lear
n
in
g
ca
n
b
e
d
o
n
e
in
two
c
ateg
o
r
ies:
T
h
e
“o
f
-
t
h
e
-
s
elf
”
f
e
atu
r
e
ex
tr
ac
tio
n
,
an
d
f
in
e
-
tu
n
in
g
[
5
]
.
I
n
a
d
d
itio
n
,
th
e
p
r
o
p
o
s
ed
C
NN,
m
o
d
if
ied
VGG1
6
,
an
d
I
n
ce
p
tio
n
V3
m
o
d
els
ar
e
im
p
lem
e
n
ted
as
a
p
r
e
-
tr
ain
ed
n
etwo
r
k
o
n
X
-
r
ay
im
ag
es.
Af
ter
e
x
ten
s
iv
e
test
s
o
n
th
e
d
ataset,
th
e
p
r
o
p
o
s
ed
m
o
d
el
r
ev
ea
ls
h
ig
h
ac
cu
r
a
cy
an
d
lo
w
tr
ai
n
in
g
tim
e
f
o
r
C
OVI
D
-
1
9
d
iag
n
o
s
is
.
Ou
r
d
ev
el
o
p
ed
d
ee
p
lea
r
n
in
g
ar
ch
itectu
r
e
ca
n
au
to
m
atica
lly
ex
tr
ac
t
an
d
s
elec
t
im
p
o
r
tan
t
f
ea
t
u
r
es
f
r
o
m
X
-
r
a
y
im
ag
es
f
o
r
co
r
o
n
av
i
r
u
s
d
et
ec
tio
n
an
d
d
iag
n
o
s
is
to
d
is
tin
g
u
is
h
p
atien
ts
with
co
r
o
n
a
v
ir
u
s
f
r
o
m
th
o
s
e
with
o
u
t
th
e
d
is
ea
s
e.
Mo
r
eo
v
er
,
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
h
as
th
e
ca
p
ab
ilit
y
to
d
etec
t
in
f
ec
tio
n
C
OVI
D
-
1
9
ar
ea
in
th
e
lu
n
g
.
T
h
e
p
ap
er
is
o
r
g
an
ized
as th
e
f
o
llo
win
g
: S
ec
tio
n
2
d
escr
ib
es
th
e
ch
est x
-
r
ay
d
ataset
an
d
th
e
p
r
o
p
o
s
ed
m
eth
o
d
s
f
o
r
C
OVI
D
-
19
d
iag
n
o
s
is
.
Sectio
n
3
p
r
esen
t
th
e
p
er
f
o
r
m
a
n
ce
ev
al
u
atio
n
m
etr
ic
s
an
d
e
x
p
er
im
en
tal
r
esu
lts
.
W
h
ile
s
ec
tio
n
4
d
is
cu
s
s
es
th
e
o
b
tain
ed
r
esu
lts
an
d
c
o
m
p
ar
ed
it
with
o
th
er
r
esear
c
h
er
s
.
th
e
co
n
clu
s
io
n
an
d
f
u
t
u
r
e
wo
r
k
ar
e
illu
s
tr
ates in
s
ec
tio
n
5
.
2.
M
AT
E
R
I
AL
S
AND
M
E
T
H
O
D
S
2
.
1
.
X
-
r
a
y
i
m
a
g
e
d
a
t
a
s
et
T
h
e
C
OVI
D
-
1
9
p
a
n
d
em
ic
is
a
n
ew
in
f
ec
tio
n
v
ir
u
s
an
d
th
er
e
is
n
o
s
u
itab
le
d
ataset
ac
ce
s
s
ib
l
e
th
at
h
as
en
o
u
g
h
d
ata
f
o
r
th
is
s
tu
d
y
.
He
n
ce
,
b
y
c
o
llectin
g
ch
est
x
-
r
a
y
im
ag
es
f
r
o
m
two
d
iv
er
s
e
im
a
g
e
r
ep
o
s
ito
r
ies,
we
h
ad
to
cr
ea
te
a
d
ataset.
C
OVI
D
-
1
9
X
-
r
ay
im
ag
es
ar
e
ac
ce
s
s
ib
le
in
t
h
e
Gith
u
b
r
ep
o
s
ito
r
y
[
6
]
.
T
h
e
r
ep
o
s
ito
r
y
co
n
tain
s
o
p
e
n
-
s
o
u
r
ce
d
atasets
o
f
C
OVI
D
-
1
9
p
atien
ts
with
c
h
est
X
-
r
ay
im
ag
es
an
d
is
r
eg
u
lar
ly
u
p
d
ated
.
All
o
f
th
ese
im
ag
es
ar
e
ex
tr
ac
ted
f
r
o
m
4
3
d
iv
er
s
e
p
u
b
licatio
n
s
.
A
r
ef
er
en
ce
f
o
r
ea
ch
im
a
g
e
is
p
r
o
v
id
ed
in
t
h
e
m
etad
ata
f
ile
in
th
e
s
am
e
r
ep
o
s
ito
r
y
.
No
r
m
al
im
ag
es
wer
e
co
llected
f
r
o
m
an
o
t
h
er
c
h
est
X
-
r
ay
im
ag
es
(
p
n
eu
m
o
n
ia)
d
atab
ase
[
7
]
Fig
u
r
e
1
s
h
o
ws X
-
r
a
y
im
ag
es f
o
r
C
OVI
D
-
1
9
an
d
n
o
r
m
al
ca
s
es.
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
1
.
E
x
am
p
les o
f
ch
est x
-
r
ay
im
ag
es
;
(
a)
C
OVI
D
-
1
9
,
(
b
)
C
OVI
D
-
1
9
,
(
c)
n
o
r
m
al,
an
d
(
d
)
n
o
r
m
al
I
n
o
u
r
s
tu
d
y
,
f
o
r
C
OVI
D
-
1
9
p
o
s
itiv
e
ca
s
es,
we
u
s
ed
a
m
etad
ata
E
x
ce
l
f
ile
th
at
co
n
tain
s
th
e
s
o
u
r
ce
o
f
all
X
-
r
ay
im
ag
es,
th
en
f
ilter
ed
th
e
co
lu
m
n
“
f
in
d
in
g
”
to
p
ic
k
u
p
th
e
C
OVI
D
-
1
9
ca
s
es.
Fu
r
t
h
er
m
o
r
e
,
u
n
d
er
th
e
co
lu
m
n
“
v
iew”
,
th
e
Po
s
ter
o
an
ter
io
r
v
iew
was
s
elec
ted
w
h
ich
d
en
o
ted
as
“PA”.
T
h
e
p
o
s
itiv
e
C
OVI
D
-
1
9
ca
s
es
wer
e
d
iag
n
o
s
ed
b
y
e
x
p
er
ts
.
W
e
u
s
ed
to
o
ls
s
u
ch
as
Gr
ad
-
C
AM
an
d
Salien
cy
t
o
en
s
u
r
e
an
d
d
if
f
e
r
en
tiate
b
etwe
en
p
o
s
itiv
e
an
d
n
eg
ativ
e
ca
s
es
b
y
a
n
s
wer
in
g
th
e
f
o
llo
win
g
q
u
esti
o
n
s
:
W
h
y
it
is
p
o
s
itiv
e?
W
h
er
e
is
th
e
in
f
ec
ted
r
eg
io
n
lo
ca
te
d
?
W
h
at
is
th
e
co
n
f
id
en
ce
th
at
it
is
p
o
s
itiv
e?
R
e
s
ea
r
ch
er
s
h
av
e
o
b
s
er
v
ed
th
at
th
e
lu
n
g
s
o
f
p
atie
n
ts
with
C
OVI
D
-
1
9
s
y
m
p
to
m
s
h
av
e
s
o
m
e
v
is
u
al
s
ig
n
s
s
u
ch
as
g
r
o
u
n
d
-
g
lass
o
p
ac
ities
o
f
d
ar
k
en
ed
d
ar
k
s
p
o
ts
th
at
ca
n
d
is
tin
g
u
is
h
b
etwe
en
C
OVI
D
-
1
9
in
f
ec
ted
p
atien
ts
an
d
n
o
n
-
C
OVI
D
-
1
9
p
atien
ts
[
8
]
-
[
1
0
]
.
Su
b
s
eq
u
en
tly
,
we
cr
ea
ted
a
f
o
l
d
er
n
am
ed
C
o
v
id
Data
s
et
wh
ich
co
n
tain
s
two
s
u
b
-
f
o
ld
er
s
n
a
m
ed
T
r
ain
an
d
Val.
E
ac
h
s
u
b
-
f
o
l
d
er
co
n
tai
n
s
two
f
o
ld
er
s
:
o
n
e
f
o
r
C
OVI
D
-
1
9
i
m
ag
es
an
d
th
e
o
th
e
r
o
n
e
f
o
r
n
o
r
m
al
im
ag
es.
I
n
th
e
T
r
ain
f
o
ld
er
,
we
h
av
e
2
2
4
im
ag
es,
wh
er
e
h
alf
o
f
th
em
ar
e
C
OVI
D
-
1
9
ca
s
es
an
d
th
e
o
th
er
s
ar
e
clas
s
if
ied
as
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
Dee
p
lea
r
n
in
g
fo
r
C
OV
I
D
-
1
9
d
ia
g
n
o
s
is
b
a
s
ed
o
n
c
h
est X
-
r
a
y
ima
g
es
(
N
a
s
h
a
t A
lr
efa
i
)
4533
n
o
r
m
al.
I
n
th
e
Val
f
o
l
d
er
,
we
h
av
e
6
0
im
ag
es,
wh
er
e
h
alf
o
f
th
em
ar
e
C
OVI
D
-
1
9
ca
s
es
an
d
th
e
o
t
h
er
s
ar
e
class
if
ied
as n
o
r
m
al.
All
th
e
im
ag
es
u
s
e
th
e
p
o
r
tab
le
n
etwo
r
k
g
r
a
p
h
ics
(
PNG)
f
o
r
m
at
an
d
h
av
e
a
d
if
f
er
en
t
r
eso
l
u
tio
n
.
T
h
e
im
ag
es
wer
e
d
im
en
s
io
n
ally
co
n
v
er
ted
t
o
2
2
4
x
2
2
4
p
ix
els
to
tr
ain
it
ea
s
ily
in
th
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs).
T
o
in
cr
ea
s
e
th
e
n
u
m
b
er
o
f
im
ag
es,
we
u
s
ed
au
g
m
en
tatio
n
tec
h
n
iq
u
es
wh
ich
ca
n
p
r
ev
en
t
C
NN
f
r
o
m
o
v
er
f
itti
n
g
th
at
c
an
en
h
an
ce
th
e
p
er
f
o
r
m
an
ce
o
f
o
u
r
m
o
d
el
an
d
in
cr
ea
s
e
th
e
ac
cu
r
ac
y
.
T
h
e
au
g
m
en
tatio
n
tech
n
iq
u
es
h
er
e
in
clu
d
e
f
lip
,
r
o
tate,
tr
an
s
late,
an
d
s
ca
le
f
o
r
th
e
im
ag
e.
T
h
e
n
u
m
b
er
o
f
en
lar
g
e
d
im
ag
es m
ak
e
th
e
d
ataset
ap
p
r
o
p
r
iate
f
o
r
C
NN.
2
.
2
.
M
o
dels
a
rc
hite
ct
ure
a
nd
dev
elo
pm
ent
T
h
e
m
ain
p
u
r
p
o
s
e
o
f
th
is
s
tu
d
y
is
to
b
u
ild
a
n
o
v
el
C
OVI
D
-
1
9
d
etec
tio
n
m
o
d
el
b
ased
o
n
X
-
r
ay
im
ag
es
an
d
th
e
g
en
e
r
al
f
o
r
m
o
f
o
u
r
p
r
o
p
o
s
ed
n
etwo
r
k
ar
ch
itectu
r
e
is
d
ep
icted
in
Fig
u
r
e
2
.
T
h
is
s
ec
tio
n
d
is
cu
s
s
es
th
r
ee
n
etwo
r
k
s
d
esig
n
ed
f
o
r
C
OVI
D
-
1
9
d
etec
ti
o
n
.
T
h
e
f
ir
s
t
o
n
e
is
b
u
ild
in
g
a
p
r
o
p
o
s
ed
C
NN
ar
ch
itectu
r
e
f
r
o
m
s
cr
atch
as
illu
s
tr
ate
d
in
Fig
u
r
e
3
.
T
h
e
s
ec
o
n
d
o
n
e
is
b
ased
o
n
th
e
m
o
d
i
f
ied
VGG1
6
ar
ch
itectu
r
e
b
y
u
s
in
g
th
e
tr
a
n
s
f
er
lear
n
in
g
co
n
ce
p
t
to
tr
an
s
f
e
r
th
e
k
n
o
wled
g
e
o
r
weig
h
ts
;
in
o
th
er
wo
r
d
s
,
f
r
o
m
th
e
p
r
e
-
tr
ain
e
d
VGG1
6
n
etwo
r
k
to
o
u
r
a
d
ap
tiv
e
d
esig
n
b
y
f
r
ee
zin
g
th
e
f
ir
s
t
lay
er
s
wh
ich
ex
tr
ac
t
th
e
g
e
n
er
al
f
ea
tu
r
es
f
r
o
m
th
e
im
a
g
e
a
n
d
m
o
d
if
y
th
e
last
lay
er
s
to
ex
tr
a
ct
an
d
s
elec
t
th
e
s
p
ec
if
ic
f
ea
tu
r
es
f
r
o
m
th
e
im
ag
e.
T
h
is
p
r
o
ce
d
u
r
e
ac
h
iev
es
a
r
o
b
u
s
tn
ess
r
esu
lt
with
a
s
m
al
l
d
ataset
an
d
o
v
er
r
o
d
e
th
e
o
v
er
f
itti
n
g
p
r
o
b
lem
.
B
esid
es
VGG1
6
,
we
u
s
ed
I
n
c
ep
tio
n
V3
as
a
p
r
e
-
tr
ain
e
d
m
o
d
el
to
co
m
p
a
r
e
t
h
e
p
er
f
o
r
m
an
ce
b
etwe
en
th
e
t
h
r
ee
ar
ch
itectu
r
e
n
etwo
r
k
s
.
Fig
u
r
e
2
.
T
h
e
g
e
n
er
al
ar
ch
itec
tu
r
e
o
f
t
h
e
p
r
o
p
o
s
ed
C
NN
u
s
ed
in
C
OVI
D
-
1
9
d
ia
g
n
o
s
is
W
e
u
s
ed
a
g
lo
b
al
d
ataset
a
s
a
b
en
ch
m
ar
k
th
at
co
n
tain
s
2
8
4
X
-
r
ay
im
ag
es
(
o
f
wh
ich
1
4
2
o
f
th
em
ar
e
p
o
s
itiv
e
C
OVI
D
-
1
9
p
atien
ts
a
n
d
1
4
2
im
ag
es
ar
e
f
o
r
n
o
r
m
al
p
atien
ts
)
.
T
h
en
,
we
g
en
er
alize
d
o
u
r
r
esu
lt
d
u
r
i
n
g
th
e
d
ata
au
g
m
en
tatio
n
s
tag
e
t
o
b
e
v
alid
f
o
r
an
y
n
ew
p
o
p
u
l
atio
n
.
W
e
u
s
ed
tr
ad
itio
n
al
C
o
n
v
o
lu
tio
n
al
Neu
r
al
Netwo
r
k
(
C
NN)
with
s
o
m
e
ad
ap
tiv
e
h
y
p
er
-
p
ar
am
eter
s
an
d
tr
an
s
f
er
lear
n
i
n
g
f
o
r
C
OVI
D
-
1
9
X
-
r
ay
im
a
g
e
class
if
icatio
n
an
d
d
is
tin
g
u
is
h
ed
th
em
f
r
o
m
n
e
g
ativ
e
ca
s
es
to
b
e
co
m
p
atib
le
with
t
h
e
clin
ic
al
u
n
d
er
s
tan
d
in
g
o
f
Dat
a
Augm
e
n
t
at
ion
-
T
r
a
n
s
f
o
r
m
a
t
i
o
n
-
B
a
t
c
h
i
ng
-
F
l
i
ps
-
Cr
o
ps
T
r
a
i
ni
ng
Da
t
a
s
e
t
(X
-
r
a
y
)
T
e
s
t
i
n
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t
J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
5
3
1
-
4
5
4
1
4534
th
is
d
is
ea
s
e.
I
n
th
is
s
tu
d
y
,
f
o
r
C
OVI
D
-
1
9
d
iag
n
o
s
is
,
a
n
e
w
C
NN
ar
ch
itectu
r
e
an
d
m
o
d
if
ied
VGG1
6
an
d
I
n
ce
p
tio
n
V3
as tr
an
s
f
er
lear
n
i
n
g
m
o
d
els we
r
e
u
tili
ze
d
s
ep
ar
ately
.
Fig
u
r
e
3
.
T
h
e
p
r
o
p
o
s
ed
C
NN
ar
ch
itectu
r
e
f
r
o
m
s
cr
atch
2
.
2
.
1
.
T
he
pro
po
s
ed
c
o
nv
o
lutio
na
l
n
eura
l
n
et
wo
rk
(
CNN
)
m
od
el
a
rc
hite
ct
ure
a
nd
d
ev
elo
pm
ent
T
h
e
C
NN
tak
es
in
p
u
t
ten
s
o
r
s
o
f
s
h
ap
e
(
h
eig
h
t,
wid
th
,
ch
a
n
n
els)
f
o
r
th
e
im
ag
es.
T
h
e
co
n
f
i
g
u
r
atio
n
is
d
esig
n
ed
to
p
r
o
ce
s
s
in
p
u
ts
o
f
s
ize
(
2
2
4
,
2
2
4
,
3
)
.
T
h
is
was
d
o
n
e
b
y
p
ass
in
g
th
e
v
alu
es
to
th
e
in
p
u
t
s
h
ap
e
ar
g
u
m
en
t
i
n
th
e
in
p
u
t
lay
er
.
Ou
r
C
NN
ar
ch
itectu
r
e
h
as
f
o
u
r
co
n
v
o
lu
tio
n
al
lay
er
s
.
W
e
u
s
ed
th
e
s
am
e
k
e
r
n
el
s
ize
d
u
r
in
g
th
e
n
etwo
r
k
(
3
×
3
)
an
d
in
c
r
ea
s
ab
le
lear
n
t
n
u
m
b
er
s
(
3
2
,
6
4
,
an
d
1
2
8
)
.
Mo
r
e
o
v
er
,
we
u
s
ed
th
r
ee
Ma
x
p
o
o
lin
g
in
th
e
n
etwo
r
k
to
r
ed
u
ce
th
e
s
p
atial
d
im
en
s
io
n
s
o
f
th
e
o
u
tp
u
t
v
o
lu
m
e
th
at
lead
to
r
ed
u
cin
g
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
an
d
t
r
ain
in
g
tim
e.
Als
o
,
d
r
o
p
o
u
t
lay
er
s
wer
e
u
s
ed
to
tr
ea
t
o
v
er
f
itti
n
g
,
B
y
d
is
ab
lin
g
th
e
n
eu
r
o
n
s
at
r
an
d
o
m
d
u
r
i
n
g
th
e
lear
n
in
g
p
r
o
ce
s
s
.
I
n
ad
d
itio
n
,
R
eL
U
wa
s
u
s
ed
as
an
ac
tiv
ati
o
n
f
u
n
ctio
n
th
r
o
u
g
h
th
e
h
id
d
e
n
lay
er
s
.
T
h
e
R
eL
U
im
p
r
o
v
es th
e
n
eu
r
al
n
etwo
r
k
b
y
s
p
ee
d
in
g
u
p
th
e
tr
ain
in
g
p
r
o
ce
s
s
.
I
n
th
e
en
d
,
we
u
s
ed
f
latten
t
o
g
et
th
e
im
ag
e
d
im
en
s
io
n
ali
ty
d
o
wn
to
1
D
an
d
a
d
d
2
d
e
n
s
e
f
u
lly
-
co
n
n
ec
ted
la
y
er
s
.
T
h
e
d
en
s
e
l
ay
e
r
s
p
r
o
ce
s
s
ed
1
D
im
a
g
e
v
e
cto
r
s
f
o
r
o
u
r
o
u
tp
u
t.
T
h
e
last
ac
tiv
atio
n
f
u
n
cti
o
n
u
s
ed
was
s
ig
m
o
id
,
wh
ich
is
s
u
itab
le
f
o
r
b
in
ar
y
class
if
icatio
n
.
I
n
o
u
r
s
tu
d
y
,
th
e
f
in
al
d
ec
is
io
n
is
p
o
s
itiv
e
C
OVI
D
-
1
9
o
r
n
o
r
m
al.
Fu
r
t
h
e
r
more
,
we
co
n
f
ig
u
r
ed
a
s
et
o
f
p
ar
a
m
eter
s
s
u
ch
a
s
b
atch
s
i
ze
is
s
et
to
32
,
th
e
n
u
m
b
er
o
f
ep
o
c
h
s
is
s
et
to
3
0
,
lear
n
in
g
r
ate
is
in
itialized
to
3
e
-
4
,
th
e
p
o
o
l
s
ize
f
ac
to
r
s
in
m
ax
p
o
o
lin
g
2
D
ar
e
tu
p
le
o
f
two
i
n
teg
er
s
a
n
d
s
et
t
o
2
f
o
r
b
o
th
,
d
r
o
p
o
u
t
r
ate
i
n
e
ac
h
lay
er
is
0
.
2
5
a
n
d
in
th
e
la
s
t
lay
er
is
0
.
5
,
a
n
d
f
in
ally
th
e
v
alid
atio
n
s
tep
h
as
b
ee
n
s
et
to
2
.
I
n
th
e
co
m
p
ilatio
n
s
tep
,
we
u
s
ed
b
in
ar
y
_
cr
o
s
s
en
tr
o
p
y
as th
e
lo
s
s
f
u
n
ctio
n
an
d
Ad
am
as th
e
o
p
tim
izer
f
o
r
th
is
m
o
d
el.
T
h
e
o
p
tim
izer
is
u
s
ed
f
o
r
m
o
d
if
y
in
g
th
e
we
ig
h
ts
o
f
th
e
n
e
u
r
o
n
s
th
r
o
u
g
h
b
ac
k
p
r
o
p
ag
atio
n
.
I
t
c
o
m
p
u
tes
th
e
d
er
i
v
ativ
e
o
f
t
h
e
lo
s
s
f
u
n
ctio
n
with
r
esp
ec
t
t
o
ea
ch
weig
h
t
an
d
s
u
b
tr
ac
ts
it
f
r
o
m
th
e
weig
h
t.
T
h
at
is
h
o
w
a
n
e
u
r
al
n
etwo
r
k
lear
n
s
.
T
h
e
n
u
m
b
er
o
f
p
ar
a
m
eter
s
u
s
ed
in
th
e
m
o
d
el
is
5
,
6
6
8
,
0
9
7
,
an
d
all
o
f
th
em
ar
e
tr
ain
ab
le
b
ec
a
u
s
e
we
d
id
n
o
t
u
s
ed
a
p
r
e
-
tr
ain
m
o
d
el.
T
h
e
lo
w
s
am
p
le
s
ize
at
th
e
f
ir
s
t
s
tag
e
o
f
co
r
o
n
a
v
ir
u
s
was
co
n
s
id
er
ed
a
ch
allen
g
e.
W
e
wer
e
ab
le
to
o
v
er
r
id
e
th
is
is
s
u
e
b
y
u
s
in
g
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
es
s
u
ch
as
tr
an
s
f
o
r
m
atio
n
,
b
atc
h
in
g
,
f
lip
s
o
r
cr
o
p
s
.
Mo
r
eo
v
er
,
ea
r
ly
an
d
r
ap
i
d
d
etec
ti
o
n
o
f
p
o
s
itiv
e
ca
s
es o
f
co
r
o
n
a
v
ir
u
s
ca
n
e
n
h
a
n
ce
th
e
h
a
n
d
lin
g
o
f
th
e
p
atien
t
s
an
d
d
ec
r
ea
s
e
th
e
s
p
r
ea
d
o
f
th
e
v
ir
u
s
.
2
.
2
.
2
.
T
he
m
o
dified
VG
G
1
6
a
nd
i
ncept
io
nV3
n
et
wo
rk
s
VGG1
6
co
n
s
id
er
s
d
ee
p
c
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
.
I
t
was
p
r
o
p
o
s
ed
in
2
0
1
4
b
y
Si
m
o
n
y
a
n
[
1
1
]
.
T
h
e
n
etwo
r
k
h
as
1
6
c
o
n
v
o
lu
tio
n
al
lay
e
r
s
with
s
m
all
f
ilter
s
ize
(
3
x
3
)
,
1
4
4
m
illi
o
n
p
ar
am
et
er
s
,
5
m
ax
-
p
o
o
lin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
Dee
p
lea
r
n
in
g
fo
r
C
OV
I
D
-
1
9
d
ia
g
n
o
s
is
b
a
s
ed
o
n
c
h
est X
-
r
a
y
ima
g
es
(
N
a
s
h
a
t A
lr
efa
i
)
4535
lay
er
s
(
2
x
2
s
ize)
an
d
3
f
u
lly
-
co
n
n
ec
ted
lay
er
s
an
d
s
o
f
t
-
m
ax
ac
tiv
atio
n
f
u
n
ctio
n
with
t
h
e
f
in
al
lay
er
.
T
h
is
m
o
d
el
p
r
e
-
tr
ain
ed
o
n
I
m
a
g
e
Net
d
ataset
an
d
th
en
th
e
m
o
d
if
ied
weig
h
ts
wer
e
tr
an
s
f
er
r
ed
to
u
p
d
ate
th
e
f
u
lly
co
n
n
ec
ted
lay
e
r
s
o
f
th
e
n
ew
n
etwo
r
k
.
I
n
o
u
r
m
o
d
el,
we
d
ec
r
ea
s
ed
th
e
n
u
m
b
er
o
f
c
o
n
v
o
lu
tio
n
s
an
d
ch
an
g
ed
th
e
s
tr
u
ctu
r
e
o
f
f
u
lly
co
n
n
ec
te
d
lay
er
b
y
u
s
in
g
f
latten
,
d
r
o
p
o
u
t
an
d
two
d
en
s
e
la
y
er
s
,
an
d
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
f
o
r
b
in
ar
y
class
if
ic
atio
n
to
d
is
tin
g
u
is
h
b
etwe
en
C
OVI
D
-
1
9
an
d
n
o
r
m
al
ca
s
es.
T
h
e
n
u
m
b
er
o
f
p
ar
am
eter
s
was
d
ec
r
ea
s
ed
to
1
6
,
3
2
0
,
4
4
9
an
d
b
y
u
s
in
g
th
e
tr
a
n
s
f
er
lear
n
i
n
g
tec
h
n
iq
u
e
,
th
e
n
u
m
b
er
o
f
tr
ain
ab
le
p
ar
am
eter
s
d
r
am
atica
lly
d
ec
li
n
ed
to
1
,
6
0
5
,
7
6
1
b
y
f
r
ee
zin
g
th
e
weig
h
ts
in
t
h
e
f
ir
s
t
p
ar
t
o
f
th
e
n
etwo
r
k
a
n
d
tr
an
s
f
er
r
in
g
t
h
e
k
n
o
wled
g
e.
H
en
ce
,
th
e
ch
allen
g
e
o
f
lo
w
s
ize
d
ata
was
o
v
er
co
m
e
an
d
th
e
o
v
er
f
itti
n
g
p
r
o
b
lem
was
s
o
lv
ed
.
At
th
e
s
am
e
tim
e,
th
e
tr
ain
in
g
tim
e
s
h
ar
p
l
y
d
ec
l
in
ed
.
At
th
e
en
d
,
we
u
s
ed
b
in
a
r
y
_
cr
o
s
s
en
tr
o
p
y
as
lo
s
s
f
u
n
ctio
n
an
d
ad
am
as
o
p
ti
m
izer
.
T
h
e
lo
w
s
am
p
le
s
ize
d
u
e
to
th
e
ex
p
en
s
iv
e
p
r
o
ce
s
s
in
ac
q
u
ir
i
n
g
d
ata,
esp
ec
ially
in
th
e
m
e
d
ical
f
ield
,
en
co
u
r
a
g
ed
u
s
to
u
tili
ze
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
es
in
o
u
r
s
tu
d
y
.
Dif
f
er
e
n
t
au
g
m
en
tati
o
n
tech
n
iq
u
es
wer
e
u
s
ed
s
u
ch
as
s
h
if
tin
g
,
z
o
o
m
i
n
g
,
f
lip
p
in
g
,
r
o
tati
n
g
,
s
h
ar
p
e
n
(
lig
h
tn
ess
v
alu
e)
,
Gau
s
s
ian
b
lu
r
(
s
ig
m
a
v
alu
e
)
,
ed
g
es
d
etec
tio
n
(
alp
h
a
v
alu
e)
,
em
b
o
s
s
(
s
tr
en
g
th
v
alu
e
)
,
s
k
ew
(
tilt
)
an
d
s
h
ea
r
(
ax
is
an
d
v
alu
e)
ar
e
u
s
ed
to
in
cr
ea
s
e
th
e
d
ata
v
o
lu
m
e
to
o
b
tain
ac
cu
r
ate
p
er
f
o
r
m
an
ce
[
1
2
]
.
I
n
o
u
r
s
tu
d
y
,
we
u
s
ed
s
o
m
e
o
f
th
e
m
.
Dee
p
lear
n
in
g
r
eq
u
ir
es
a
b
i
g
d
ata
v
o
lu
m
e.
Mo
r
e
o
v
er
,
in
cr
ea
s
in
g
th
e
am
o
u
n
t
o
f
d
ata
ca
n
h
e
lp
u
s
o
v
er
co
m
e
th
e
o
v
er
f
itti
n
g
is
s
u
e
an
d
en
h
a
n
ce
th
e
m
o
d
el
p
er
f
o
r
m
an
ce
.
Fu
r
th
er
m
o
r
e,
tr
an
s
f
er
lear
n
in
g
also
was
u
tili
ze
d
in
th
e
s
ec
o
n
d
m
o
d
el
b
y
f
i
n
e
-
tu
n
in
g
VGG
1
6
.
T
h
e
ea
r
ly
lay
er
s
in
C
NN
ex
tr
ac
t
g
en
er
ic
f
ea
tu
r
es,
b
u
t
th
e
last
lay
er
s
ex
tr
ac
t sp
ec
if
ic
f
ea
tu
r
es
[
1
3
]
.
T
h
er
ef
o
r
e,
s
o
m
e
ea
r
ly
lay
e
r
s
wer
e
f
ix
ed
in
o
u
r
m
o
d
els an
d
th
e
f
ix
ed
lay
er
s
wer
e
ex
clu
d
ed
d
u
r
in
g
t
h
e
tr
ain
in
g
o
f
th
e
m
o
d
els.
I
n
b
o
th
n
eu
r
al
n
etwo
r
k
ar
c
h
itectu
r
es
(
VGG1
6
an
d
I
n
ce
p
tio
n
)
th
e
ea
r
ly
lay
er
s
r
esp
o
n
s
ib
le
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
s
elec
tio
n
ar
e
f
r
o
ze
n
an
d
u
n
f
r
ee
ze
th
e
f
in
al
b
l
o
ck
w
h
ich
co
n
s
tr
u
cted
f
r
o
m
f
latten
,
d
r
o
p
o
u
t,
an
d
two
d
en
s
e
lay
er
s
.
He
n
ce
,
th
e
n
u
m
b
er
o
f
p
a
r
am
eter
s
was
d
ec
r
ea
s
ed
an
d
th
e
co
m
p
u
tatio
n
al
co
m
p
lex
ity
an
d
tr
ain
in
g
tim
e
also
d
ec
lin
e
d
.
T
h
e
th
ir
d
ar
c
h
itectu
r
e
u
s
ed
in
th
is
s
tu
d
y
is
I
n
ce
p
tio
n
V3
,
w
h
ich
was
d
ev
elo
p
ed
f
o
r
th
e
G
o
o
g
L
eNe
t
m
o
d
el
b
y
Go
o
g
le
r
esear
ch
e
r
s
i
n
2
0
1
5
[
1
4
]
.
T
h
e
g
o
al
o
f
th
is
n
etwo
r
k
is
to
ac
t
as
a
m
u
lti
-
lev
el
f
ea
tu
r
e
ex
tr
ac
to
r
with
in
cr
ea
s
ab
le
f
ilter
s
ize
in
ea
ch
co
n
v
o
l
u
tio
n
al
lay
er
.
Fu
r
t
h
er
m
o
r
e
,
th
e
weig
h
ts
f
o
r
I
n
ce
p
tio
n
V3
ar
e
s
m
aller
th
an
VGG1
6
.
3.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S
I
n
o
u
r
ex
p
er
im
en
ts
,
t
h
e
n
e
u
r
al
n
etwo
r
k
s
wer
e
t
r
ain
ed
u
s
in
g
p
y
th
o
n
(
Ker
as
an
d
T
e
n
s
o
r
Flo
w
as
a
b
ac
k
en
d
)
;
th
e
tr
ain
in
g
p
r
o
ce
d
u
r
e
was
d
o
n
e
b
y
u
tili
zin
g
th
e
GPU
f
ea
tu
r
e
wh
ich
is
av
aila
b
le
in
g
o
o
g
le
c
o
lab
.
All
th
e
ex
p
er
im
e
n
ts
ar
e
p
er
f
o
r
m
ed
o
n
an
I
n
tel
i7
2
.
7
GHz
C
PU
with
1
6
GB
o
f
R
AM
.
T
h
e
m
ea
s
u
r
es
o
f
p
er
f
o
r
m
an
ce
u
s
ed
in
th
is
s
tu
d
y
ar
e
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
as sh
o
wn
.
a
c
c
ura
c
y
=
(
TP
+
TN
)
(
TP
+
FN
)
+
(
FP
+
TN
)
(
1
)
s
e
n
s
it
ivity
(
r
e
c
a
l
l
)
=
TP
(
TP
+
FN
)
(
2
)
s
pe
c
ifi
c
it
y
=
TN
(
TN
+
FP
)
(
3
)
pr
e
c
ision
=
Tp
(
Tp
+
FP
)
(
4
)
1
−
=
2
(
×
+
)
(
5
)
T
h
e
ac
cu
r
ac
y
is
th
e
a
b
ilit
y
t
o
d
is
tin
g
u
is
h
b
etwe
en
th
e
cl
ass
es
co
r
r
ec
tly
b
y
t
h
e
class
if
ier
,
wh
ile
s
en
s
itiv
ity
d
en
o
tes
its
ca
p
ab
ilit
y
to
ac
cu
r
ately
id
en
tify
th
e
tr
u
e
p
o
s
itiv
e.
Sp
ec
if
icity
e
v
alu
ates
th
e
ac
tu
al
n
eg
ativ
es
th
at
ar
e
co
r
r
ec
tly
i
d
en
tifie
d
b
y
t
h
e
class
if
ier
.
Als
o
,
tr
u
e
p
o
s
itiv
e
(
T
P)
is
th
e
n
u
m
b
er
o
f
co
r
r
ec
tly
class
if
ied
C
O
VI
D
-
1
9
,
f
alse
p
o
s
itiv
e
(
FP
)
is
th
e
n
u
m
b
er
o
f
n
o
r
m
al
X
-
r
a
y
im
a
g
es
th
at
h
av
e
th
e
wr
o
n
g
class
if
icatio
n
as
C
OVI
D
-
19
,
f
alse
n
eg
ativ
e
(
FN)
is
th
e
n
u
m
b
er
o
f
C
OVI
D
-
1
9
X
-
r
a
y
im
ag
es
wr
o
n
g
ly
lab
ele
d
as n
o
n
-
C
OVI
D
-
1
9
an
d
tr
u
e
n
e
g
ativ
e
(
T
N)
is
th
e
n
u
m
b
er
o
f
t
r
u
ly
id
en
tifie
d
n
o
n
-
C
OVI
D
-
1
9
(
n
o
r
m
al)
ca
s
es.
T
o
ev
al
u
ate
th
e
ef
f
icien
c
y
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
,
we
p
er
f
o
r
m
ed
b
o
t
h
q
u
alitativ
e
an
d
q
u
an
titativ
e
an
aly
s
is
to
g
et
a
b
etter
u
n
d
er
s
tan
d
in
g
o
f
its
d
etec
tio
n
p
er
f
o
r
m
an
ce
an
d
d
ec
is
io
n
-
m
a
k
in
g
b
eh
av
io
r
.
First,
all
im
ag
es
wer
e
r
esized
t
o
2
2
4
x
2
2
4
to
b
e
s
u
itab
le
wit
h
th
e
VGG1
6
,
I
n
ce
p
tio
n
V
3
an
d
t
h
e
p
r
o
p
o
s
ed
C
NN.
Mo
r
eo
v
er
,
all
im
ag
e
p
ix
els
w
er
e
n
o
r
m
alize
d
in
r
an
g
e
[
0
,
1
]
.
T
h
e
s
am
e
p
ar
am
eter
s
to
ea
ch
n
etwo
r
k
wer
e
u
s
ed
,
s
u
ch
as
th
e
n
u
m
b
e
r
o
f
e
p
o
ch
s
wh
ich
was
s
et
to
b
e
3
0
,
th
e
b
atch
s
ize
eq
u
als
to
3
2
,
lea
r
n
in
g
r
ate
s
et
to
b
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t
J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
5
3
1
-
4
5
4
1
4536
0
.
0
0
0
3
an
d
b
i
n
ar
y
cr
o
s
s
en
tr
o
p
y
as
lo
s
s
f
u
n
ctio
n
.
M
o
r
eo
v
e
r
,
we
u
s
ed
d
if
f
er
e
n
t
o
p
tim
izer
s
s
u
ch
as
ad
am
an
d
R
MSp
r
o
p
an
d
th
en
d
ec
id
e
d
wh
ich
o
n
e
is
b
etter
ac
co
r
d
in
g
to
th
e
ac
cu
r
ac
y
an
d
m
o
d
el
p
er
f
o
r
m
a
n
ce
.
I
n
o
u
r
ex
p
er
im
en
ts
,
we
u
s
ed
a
d
am
o
p
tim
izer
in
all
th
e
m
o
d
els
b
ec
au
s
e
it
g
av
e
u
s
b
etter
r
esu
lts
t
h
an
R
MSp
r
o
p
b
ased
o
n
th
e
d
ataset.
GPU
was
u
s
ed
in
o
u
r
ex
p
er
im
en
t
to
ac
ce
ler
ate
th
e
tr
ain
in
g
t
im
e.
T
ab
le
1
illu
s
tr
ates
th
e
tr
a
in
in
g
tim
e
f
o
r
ea
c
h
m
o
d
el.
I
t
is
clea
r
t
h
at
o
u
r
p
r
o
p
o
s
ed
C
NN
h
as
t
h
e
lo
west
tr
ain
in
g
tim
e
an
d
VG
G1
6
h
as
th
e
w
o
r
s
t
tr
ain
in
g
tim
e.
On
th
e
o
th
er
h
a
n
d
,
b
ased
o
n
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
in
th
e
p
r
e
-
tr
ain
e
d
m
o
d
els
I
n
ce
p
tio
n
V
3
an
d
VGG1
6
wer
e
v
er
y
h
ig
h
,
b
u
t
b
y
u
s
in
g
tr
a
n
s
f
er
lear
n
in
g
,
t
h
e
n
u
m
b
e
r
o
f
p
ar
am
eter
s
d
ec
li
n
ed
s
h
ar
p
ly
o
win
g
to
f
r
ee
zin
g
s
o
m
e
o
f
lay
er
s
in
t
h
e
b
eg
i
n
n
in
g
o
f
th
e
n
etwo
r
k
.
C
o
n
s
eq
u
en
tly
,
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
n
ee
d
ed
to
tr
ain
was
d
r
a
m
atica
lly
r
e
d
u
ce
d
f
r
o
m
2
5
,
0
7
9
,
7
1
3
to
3
,
2
7
6
,
9
2
9
an
d
f
r
o
m
1
6
,
3
2
0
,
4
4
9
to
1
,
6
0
5
,
7
6
1
i
n
I
n
ce
p
tio
n
V3
an
d
VGG1
6
r
esp
ec
tiv
ely
.
T
ab
le
1
.
Netwo
r
k
’
s
tr
ain
in
g
ti
m
e
an
d
p
ar
am
eter
s
To
t
a
l
P
a
r
a
m
e
t
e
r
s
Tr
a
i
n
a
b
l
e
P
a
r
a
met
e
r
s
N
o
n
-
t
r
a
i
n
a
b
l
e
p
a
r
a
met
e
r
s
Tr
a
i
n
i
n
g
T
i
me(m
i
n
u
t
e
s)
I
n
c
e
p
t
i
o
n
V
3
2
5
,
0
7
9
,
7
1
3
3
,
2
7
6
,
9
2
9
2
1
,
8
0
2
,
7
8
4
1
5
:
2
0
V
G
G
1
6
1
6
,
3
2
0
,
4
4
9
1
,
6
0
5
,
7
6
1
1
4
,
7
1
4
,
6
8
8
5
2
:
0
2
P
r
o
p
o
se
d
C
N
N
5
,
6
6
8
,
0
9
7
5
,
6
6
8
,
0
9
7
0
0
5
:
4
2
T
r
ain
in
g
an
d
test
in
g
ac
c
u
r
ac
y
ar
e
g
iv
e
n
in
T
ab
le
2
.
I
t
is
clea
r
th
at
th
e
p
r
e
-
t
r
ain
m
o
d
el
VGG1
6
ac
h
iev
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
o
f
1
0
0
%
an
d
9
8
.
3
3
%
in
b
o
th
tr
ain
in
g
a
n
d
test
in
g
r
esp
e
ctiv
ely
.
I
n
co
n
tr
ast,
I
n
ce
p
tio
n
V3
ar
c
h
itectu
r
e
m
o
d
el
h
ad
th
e
wo
r
s
t
ac
cu
r
ac
y
.
On
th
e
o
th
er
h
an
d
,
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
h
ad
th
e
lo
west
lo
s
s
f
u
n
ctio
n
d
u
r
i
n
g
th
e
test
(
0
.
0
0
7
9
)
wh
ich
h
as
co
n
v
er
g
en
ce
to
ze
r
o
an
d
p
r
o
m
is
e
d
test
in
g
ac
cu
r
ac
y
o
f
9
6
.
6
6
%.
T
ab
le
2
.
Per
f
o
r
m
an
ce
o
f
p
r
o
p
o
s
ed
C
NN,
VGG1
6
an
d
I
n
ce
p
tio
n
V3
f
o
r
C
OVI
D
-
1
9
d
iag
n
o
s
is
Tr
a
i
n
i
n
g
A
c
c
u
r
a
c
y
Tr
a
i
n
i
n
g
L
o
ss F
u
n
c
t
i
o
n
Te
st
i
n
g
A
c
c
u
r
a
c
y
Te
st
i
n
g
Lo
ss
F
u
n
c
t
i
o
n
I
n
c
e
p
t
i
o
n
V
3
8
9
.
7
%
0
.
1
0
2
5
8
1
.
6
%
0
.
0
8
1
7
V
G
G
1
6
1
0
0
%
0
.
0
0
6
5
9
8
.
3
3
%
0
.
0
8
5
4
P
r
o
p
o
se
d
C
N
N
9
8
.
2
%
0
.
0
1
6
1
9
6
.
6
6
%
0
.
0
0
7
9
T
ab
le
3
illu
s
tr
ates
th
e
v
alid
atio
n
r
esu
lts
o
f
th
e
th
r
ee
n
e
u
r
al
n
etwo
r
k
s
b
y
ac
c
u
r
ac
y
,
s
p
ec
if
icity
,
s
en
s
itiv
ity
,
p
r
ec
is
io
n
an
d
f
1
_
s
co
r
e
m
etr
ics.
I
t
is
n
o
ticea
b
le
th
at
VGG1
6
h
ad
th
e
b
est
r
esu
lt
b
y
co
m
p
a
r
in
g
with
o
th
er
two
m
o
d
els
with
o
u
t
an
y
co
n
s
id
er
atio
n
to
th
e
co
m
p
u
tatio
n
al
co
m
p
lex
ity
.
Alth
o
u
g
h
o
u
r
p
r
o
p
o
s
ed
C
NN
a
r
ch
itectu
r
e
ac
h
iev
e
d
th
e
s
ec
o
n
d
p
er
f
o
r
m
a
n
ce
,
it h
as si
g
n
if
ic
an
tly
lo
w
tr
ain
in
g
tim
e.
T
ab
le
3
.
Valid
atio
n
r
esu
lts
o
f
t
h
e
th
r
ee
n
e
u
r
al
n
etwo
r
k
s
b
y
ac
cu
r
ac
y
,
s
p
ec
if
icity
,
s
en
s
itiv
ity
,
p
r
ec
is
io
n
an
d
f
1
_
s
co
r
e
m
et
r
ics
A
c
c
u
r
a
c
y
S
e
n
s
i
t
i
v
i
t
y
S
p
e
c
i
f
i
c
i
t
y
P
r
e
c
i
s
i
o
n
F
1
_
S
c
o
r
e
I
n
c
e
p
t
i
o
n
V
3
8
1
.
6
7
70
9
3
.
3
3
9
1
.
3
7
9
.
2
5
V
G
G
1
6
9
8
.
3
3
1
0
0
9
6
.
6
7
9
6
.
7
7
9
8
.
3
6
p
r
o
p
o
se
d
C
N
N
9
6
.
6
6
1
0
0
9
3
.
3
3
9
3
.
7
5
9
6
.
7
7
T
h
e
co
n
f
u
s
io
n
m
atr
ices
f
o
r
t
h
e
th
r
ee
n
eu
r
al
n
etwo
r
k
s
o
f
m
o
d
if
ied
in
ce
p
tio
n
V
3
,
m
o
d
if
i
ed
VGG1
6
an
d
th
e
p
r
o
p
o
s
ed
C
NN
ar
e
p
r
esen
ted
in
Fig
u
r
e
4
I
t
is
clea
r
th
at
VGG1
6
an
d
t
h
e
p
r
o
p
o
s
ed
C
NN
ac
h
iev
ed
1
0
0
%
C
OVI
D
-
1
9
d
iag
n
o
s
is
.
T
h
e
two
m
o
d
els
h
a
v
e
th
e
ab
ilit
y
to
cla
s
s
if
y
all
tr
u
e
p
o
s
itiv
e
ca
s
es
in
th
e
v
alid
atio
n
d
ataset
an
d
h
av
e
p
r
o
m
is
in
g
r
esu
lts
in
class
if
y
in
g
th
e
tr
u
e
n
eg
ativ
e
ca
s
es.
Fu
r
th
er
m
o
r
e,
Fig
u
r
e
5
s
h
o
ws
th
e
p
lo
t
o
f
tr
ain
i
n
g
a
n
d
v
alid
atio
n
ac
cu
r
ac
y
an
d
l
o
s
s
o
n
C
OVI
D
-
1
9
X
-
r
ay
im
ag
es
d
ataset
f
o
r
th
e
p
r
e
-
tr
ain
ed
m
o
d
els
in
ce
p
tio
n
V3
,
VGG1
6
an
d
p
r
o
p
o
s
ed
C
NN.
T
h
e
lin
e
g
r
ap
h
in
th
e
f
i
r
s
t
m
o
d
el
f
lu
ctu
ated
d
esp
ite
u
s
in
g
lo
w
lear
n
i
n
g
r
at
e
th
at
g
av
e
u
s
th
e
in
d
icatio
n
th
at
it
ca
n
n
o
t
d
eter
m
in
e
th
e
g
l
o
b
al
o
p
tim
u
m
an
d
f
ac
ed
d
if
f
ic
u
lties
in
d
ea
lin
g
wi
th
lo
ca
l o
p
tim
u
m
.
I
n
c
o
n
tr
ast,
f
o
r
th
e
o
th
er
two
m
o
d
els,
co
n
v
er
g
en
ce
p
r
o
ce
d
u
r
e
was d
o
n
e
p
er
f
ec
tly
.
I
n
th
e
b
e
g
in
n
in
g
o
f
t
h
is
s
ec
tio
n
,
we
talk
ed
ab
o
u
t
q
u
alitativ
e
an
aly
s
is
to
e
v
alu
ate
t
h
e
ef
f
i
cien
cy
an
d
g
et
a
b
etter
u
n
d
er
s
tan
d
in
g
to
h
o
w
th
e
m
o
d
el
ca
n
class
if
y
p
o
s
itiv
e
C
OVI
D
-
1
9
ca
s
e
s
.
Hen
ce
,
it
ca
n
b
e
b
en
ef
icial
to
v
is
u
alize
wh
at
th
e
C
NN
v
alu
es
ar
e
wh
en
it
d
o
es
a
p
r
ed
ictio
n
,
wh
at
f
ea
t
u
r
e
s
it
f
in
d
s
im
p
o
r
tan
t,
an
d
wh
et
h
er
th
o
s
e
f
ea
tu
r
es
ca
n
d
eter
m
i
n
e
th
e
lo
ca
tio
n
o
f
th
e
in
f
ec
ted
r
eg
io
n
.
T
o
v
is
u
alize
th
e
h
ea
t
m
ap
,
we
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
Dee
p
lea
r
n
in
g
fo
r
C
OV
I
D
-
1
9
d
ia
g
n
o
s
is
b
a
s
ed
o
n
c
h
est X
-
r
a
y
ima
g
es
(
N
a
s
h
a
t A
lr
efa
i
)
4537
u
s
ed
a
tech
n
i
q
u
e
ca
lled
Gr
ad
-
C
AM
(
g
r
ad
ien
t
class
ac
tiv
atio
n
m
ap
)
.
T
h
is
tech
n
iq
u
e
is
u
s
ed
to
d
etec
t
t
h
e
im
p
o
r
tan
ce
o
f
a
p
ar
ticu
lar
cla
s
s
in
o
u
r
m
o
d
el,
we
ju
s
t
tak
e
its
g
r
ad
ien
t
co
n
ce
r
n
in
g
th
e
f
in
al
co
n
v
o
lu
tio
n
al
lay
er
an
d
n
ex
t
weig
h
e
d
it
u
p
o
n
th
e
o
u
tp
u
t
o
f
th
is
lay
er
.
I
t
is
c
lear
in
Fig
u
r
e
6
th
at
th
e
Gr
ad
-
C
AM
h
as
th
e
ab
ilit
y
to
lo
ca
te
th
e
lo
ca
tio
n
o
f
th
e
in
f
ec
ted
r
eg
io
n
in
s
id
e
th
e
ch
est ac
cu
r
ately
.
(
a)
(
b
)
(
c)
Fig
u
r
e
4
.
T
h
e
co
n
f
u
s
io
n
m
atr
ices f
o
r
th
e
th
r
ee
n
eu
r
al
n
etwo
r
k
s
u
s
ed
in
th
e
ex
p
er
im
en
t,
(
a)
I
n
ce
p
tio
n
V3
,
an
d
(
b
)
VGG1
6
,
(
c)
I
n
ce
p
tio
n
V
3
I
n
ce
p
tio
n
V3
VGG1
6
Pro
p
o
s
ed
C
NN
Fig
u
r
e
5
.
T
r
ain
in
g
a
n
d
v
alid
atio
n
ac
cu
r
ac
y
an
d
lo
s
s
o
f
C
OVI
D
-
1
9
x
-
r
ay
im
a
g
es
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t
J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
5
3
1
-
4
5
4
1
4538
Fig
u
r
e
6
.
Gr
a
d
-
C
AM
s
h
o
win
g
h
ea
t m
ap
s
an
d
h
ig
h
lig
h
t th
e
p
n
eu
m
o
n
ia
o
f
th
e
p
o
s
itiv
e
C
OVI
D
-
1
9
p
atie
n
ts
4.
D
I
SCU
SS
I
O
N
I
n
th
is
s
tu
d
y
,
we
p
r
o
p
o
s
ed
n
e
w
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
e
f
r
o
m
s
cr
atch
a
n
d
two
d
ee
p
m
o
d
els
b
ased
o
n
VGG1
6
an
d
I
n
ce
p
tio
n
V3
ar
ch
itectu
r
es
to
d
etec
t
C
OVI
D
-
1
9
f
r
o
m
th
e
ch
est
X
-
r
ay
im
ag
es.
Ou
r
m
o
d
els
wer
e
test
ed
o
n
th
e
v
alid
atio
n
s
et
an
d
ac
h
iev
e
d
an
ac
cu
r
ac
y
o
f
8
1
.
6
7
%,
9
8
.
3
3
%
an
d
9
6
.
6
6
%
f
o
r
I
n
ce
p
tio
n
V3
,
VGG1
6
an
d
th
e
p
r
o
p
o
s
ed
C
NN
r
esp
ec
tiv
ely
.
F
u
r
th
er
m
o
r
e,
b
o
th
VGG1
6
an
d
th
e
p
r
o
p
o
s
ed
C
NN
ac
h
iev
ed
1
0
0
%
f
o
r
s
en
s
itiv
ity
to
d
ec
r
ea
s
e
th
e
f
alse
n
eg
ativ
e
C
OVI
D
-
1
9
ca
s
es
p
r
ed
ictio
n
a
n
d
in
cr
ea
s
e
t
h
e
tr
u
e
p
o
s
itiv
e
C
OVI
D
-
1
9
ca
s
es p
r
ed
ictio
n
wh
ich
was o
u
r
g
o
al
.
I
n
co
m
p
ar
i
n
g
o
u
r
r
esu
lts
with
o
th
er
s
tu
d
ies
in
liter
atu
r
e
p
r
esen
ted
in
T
ab
le
4
,
o
u
r
s
tu
d
y
u
s
ed
X
-
r
ay
im
ag
es,
b
u
t
s
o
m
e
s
tu
d
ies
u
s
ed
ch
est
C
T
im
ag
es.
T
h
e
ta
b
le
in
clu
d
es
b
o
th
o
f
th
em
.
So
m
e
s
tu
d
ies
wh
ich
ac
h
iev
ed
h
ig
h
ac
c
u
r
ac
y
u
s
ed
t
h
e
tr
ain
in
g
s
et.
I
n
co
n
tr
ast,
o
u
r
s
tu
d
y
u
s
ed
a
v
alid
ati
o
n
s
et
f
o
r
m
o
r
e
r
eliab
ilit
y
.
Go
ze
s
et
a
l.
[
1
5
]
p
r
o
p
o
s
ed
a
u
to
m
ated
d
iag
n
o
s
is
an
d
p
atien
t
m
o
n
ito
r
i
n
g
s
y
s
tem
u
s
in
g
d
ee
p
lear
n
in
g
an
d
C
T
s
ca
n
im
ag
es
in
v
esti
g
atio
n
.
T
h
e
au
th
o
r
u
tili
ze
d
R
esNet
-
5
0
ar
ch
itectu
r
e
to
au
t
o
m
atica
lly
class
if
y
C
OVI
D
-
1
9
as
p
o
s
itiv
e
o
r
n
eg
ativ
e
ca
s
es.
T
h
e
p
r
o
p
o
s
ed
m
e
th
o
d
o
u
tp
er
f
o
r
m
e
d
th
e
p
r
o
f
ess
io
n
als
in
th
is
f
ield
with
ac
cu
r
ac
y
o
f
9
9
.
6
%
a
n
d
9
8
.
2
%
f
o
r
Sen
s
itiv
ity
.
C
h
en
et
a
l.
[
1
6
]
b
u
ilt
a
d
ee
p
lear
n
in
g
m
o
d
el
f
o
r
C
OVI
D
-
1
9
p
n
e
u
m
o
n
ia
d
etec
tio
n
b
ase
d
o
n
h
i
g
h
-
r
eso
lu
ti
o
n
C
T
im
ag
es.
T
h
e
m
o
d
el
d
esig
n
an
d
v
alid
atio
n
r
esu
lted
in
4
6
,
0
9
6
a
n
o
n
y
m
o
u
s
im
ag
es
o
f
1
0
6
p
atien
ts
ad
m
itted
to
th
e
clin
ic,
in
clu
d
in
g
5
1
C
OVI
D
-
1
9
co
n
f
ir
m
e
d
lab
o
r
ato
r
y
p
atien
ts
an
d
5
5
o
th
er
d
is
ea
s
es.
As a
r
esu
lt,
th
ey
ac
h
iev
ed
9
5
.
2
4
% a
cc
u
r
ac
y
.
Nar
in
et
a
l.
[
1
7
]
p
e
r
f
o
r
m
ed
th
e
ex
p
er
im
e
n
t
with
th
r
ee
s
ep
ar
ate
C
NN
m
o
d
els
(
I
n
c
ep
tio
n
V3
,
R
esNet5
0
,
an
d
I
n
ce
p
tio
n
R
esNetV2
)
.
All
o
f
th
em
wer
e
p
r
e
-
tr
ain
ed
o
n
I
m
a
g
eNe
t
d
ataset
b
u
t
f
o
r
2
-
class
class
if
icatio
n
,
R
esNe
t5
0
ac
h
ie
v
ed
th
e
b
est
ac
cu
r
ac
y
o
f
9
8
%
.
T
h
e
e
x
p
er
im
en
t
d
id
n
o
t
h
av
e
p
n
eu
m
o
n
ia
ca
s
es.
Hen
ce
,
th
e
m
o
d
el
m
ay
n
o
t
d
i
s
tin
g
u
is
h
b
etwe
en
C
OVI
D
-
1
9
an
d
o
t
h
er
p
n
eu
m
o
n
ia
d
is
ea
s
es.
L
.
W
an
g
et
a
l
.
[
1
8
]
p
r
o
p
o
s
ed
a
n
ew
ar
c
h
itectu
r
e
n
eu
r
al
n
etwo
r
k
n
a
m
ed
C
OVI
D
-
Net
f
o
r
C
OVI
D1
9
d
etec
ti
o
n
.
T
h
e
m
o
d
el
ac
h
iev
ed
a
9
2
.
4
% a
cc
u
r
ac
y
.
Ap
o
s
to
lo
p
o
u
l
o
s
et
a
l.
[
5
]
d
e
m
o
n
s
tr
ated
th
e
s
u
p
er
io
r
ity
o
f
m
o
b
ile
n
et
th
an
b
u
ild
i
n
g
C
NN
f
r
o
m
s
cr
atch
in
d
ec
r
ea
s
in
g
th
e
f
alse
n
eg
ativ
es
f
o
r
C
OVI
D
-
1
9
d
etec
tio
n
.
Fro
m
th
e
co
m
p
ar
is
o
n
b
etwe
en
t
h
e
two
m
o
d
els,
it
is
clea
r
t
h
at
b
u
i
ld
in
g
th
e
m
o
d
el
f
r
o
m
s
cr
atch
tak
es
a
lo
t
o
f
tim
e
an
d
r
eq
u
ir
es
co
m
p
le
x
co
m
p
u
tatio
n
s
.
O
n
th
e
o
th
er
h
a
n
d
,
th
e
au
th
o
r
u
s
ed
1
0
-
f
o
ld
r
e
p
ea
ted
ly
wh
ich
is
co
n
s
id
er
ed
as
tim
e
co
n
s
u
m
in
g
b
u
t
m
o
r
e
ac
c
u
r
ate.
Mo
r
e
o
v
er
,
p
u
lm
o
n
a
r
y
d
is
ea
s
es
h
av
e
m
an
y
ty
p
es.
T
h
e
au
th
o
r
u
s
ed
s
ix
t
y
p
es
o
f
th
em
in
th
e
d
ataset
an
d
ad
d
ed
C
OVI
D
-
1
9
as
th
e
s
ev
en
th
d
is
ea
s
e.
He
ac
h
iev
ed
9
9
.
1
8
%
ac
cu
r
ac
y
,
9
9
.
4
2
%
s
p
ec
if
icity
,
an
d
9
7
.
3
6
%
s
en
s
itiv
ity
in
th
e
d
ete
ctio
n
o
f
C
OVI
D
-
1
9
,
b
u
t
f
o
r
t
h
e
r
esu
lt
f
o
r
tr
ain
in
g
p
r
o
ce
d
u
r
e,
th
e
m
o
r
e
r
eliab
le
r
esu
lt
s
h
o
u
ld
b
e
o
n
th
e
v
alid
atio
n
s
et.
Ap
o
s
to
lo
p
o
u
lo
s
an
d
Mp
esian
a
[
1
9
]
p
u
b
lis
h
ed
a
n
o
th
er
p
ap
er
a
b
o
u
t
au
to
m
atic
d
etec
tio
n
f
r
o
m
X
-
r
ay
im
ag
es
u
tili
zin
g
tr
an
s
f
er
lear
n
in
g
.
T
h
e
y
u
s
ed
VGG1
9
an
d
Mo
b
ilNetV2
an
d
ac
h
iev
ed
a
9
8
.
7
5
%
an
d
9
7
.
4
%
ac
cu
r
ac
y
r
esp
ec
tiv
ely
f
o
r
t
h
e
2
-
class
p
r
o
b
l
e
m
.
VGG1
9
a
n
d
M
o
b
ilNetV2
ar
e
v
er
y
d
ee
p
an
d
h
av
e
a
lo
t o
f
p
a
r
am
eter
s
th
at
n
ee
d
s
a
lo
n
g
tim
e
f
o
r
r
ai
n
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
Dee
p
lea
r
n
in
g
fo
r
C
OV
I
D
-
1
9
d
ia
g
n
o
s
is
b
a
s
ed
o
n
c
h
est X
-
r
a
y
ima
g
es
(
N
a
s
h
a
t A
lr
efa
i
)
4539
Seth
y
et
a
l.
[
2
0
]
u
s
ed
R
esNet5
0
ar
ch
itectu
r
e
with
SVM
in
th
e
f
u
lly
c
o
n
n
ec
ted
lay
er
f
o
r
C
OVI
D
-
19
d
etec
tio
n
f
r
o
m
c
h
est X
-
r
ay
im
ag
es.
R
esNet5
0
m
o
d
el
wo
r
k
s
as a
f
ea
tu
r
e
ex
tr
ac
to
r
an
d
SVM
ac
ts
as
a
clas
s
if
ier
in
th
e
f
u
lly
co
n
n
ec
ted
lay
er
.
T
h
e
m
o
d
el
was
ap
p
lied
o
n
a
2
-
class
es
d
ataset
an
d
ac
h
iev
ed
a
9
5
.
3
8
%
ac
cu
r
ac
y
.
Hem
d
an
[
2
1
]
u
s
ed
d
if
f
e
r
en
t
p
r
e
-
tr
ain
ed
m
o
d
els
to
d
etec
t
C
OVI
D
-
1
9
ca
s
es
f
r
o
m
ch
est
X
-
r
ay
im
a
g
es
a
n
d
in
tr
o
d
u
ce
d
a
C
OVI
DX
-
Net
m
o
d
el
in
v
o
lv
in
g
s
ev
e
n
C
NN
m
o
d
els.
Oztu
r
k
et
a
l.
[
2
2
]
,
p
r
o
p
o
s
ed
a
n
ew
C
NN
ar
ch
itectu
r
e
n
am
ed
Dar
k
Net
ar
ch
itectu
r
e
f
o
r
C
OVI
D
-
1
9
d
et
ec
tio
n
f
r
o
m
X
-
r
a
y
im
ag
es.
T
h
eir
m
o
d
el
ac
h
iev
ed
9
8
.
0
8
% a
cc
u
r
ac
y
f
o
r
a
2
-
class
p
r
o
b
lem
.
T
ab
le
4
.
C
o
m
p
a
r
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
with
o
th
er
d
ee
p
lear
n
in
g
m
eth
o
d
s
u
s
in
g
C
T
an
d
X
-
r
ay
im
ag
es
S
tudy
T
ype
of
im
a
ge
s
Ar
c
hit
e
c
tu
r
e
Ac
c
ur
a
c
y
(
%
)
S
e
ns
it
ivi
t
y
#of
pa
r
a
mete
r
s
(
in
mi
ll
i
on)
[
1
5
]
C
h
e
st
C
T
R
e
sN
e
t
-
50
9
9
.
6
9
8
.
2
36
[
1
6
]
C
h
e
st
C
T
U
N
e
t
+
+
9
5
.
2
4
1
0
0
-
[
1
7
]
C
h
e
st
X
-
r
a
y
R
e
sN
e
t
-
50
98
96
36
I
n
c
e
p
t
i
o
n
V
3
97
94
26
[
1
8
]
C
h
e
st
X
-
r
a
y
C
O
V
I
D
-
N
e
t
9
2
.
6
8
7
.
1
1
1
6
[
1
9
]
C
h
e
st
X
-
r
a
y
VGG
-
19
9
8
.
7
5
9
2
.
8
5
1
4
3
M
o
b
i
l
e
N
e
t
9
7
.
4
9
9
.
1
0
-
[
5
]
C
h
e
st
X
-
r
a
y
M
o
b
i
l
e
N
e
t
V
2
9
9
.
1
8
9
7
.
3
6
-
[
2
0
]
C
h
e
st
X
-
r
a
y
R
e
sN
e
t
-
5
0
+
S
V
M
9
5
.
3
8
9
5
.
3
3
36
[
2
1
]
C
h
e
st
X
-
r
a
y
VGG
-
19
90
1
0
0
1
4
3
[
2
2
]
C
h
e
st
X
-
r
a
y
D
a
r
k
N
e
t
9
8
.
0
8
8
5
.
3
5
1
.
1
6
[
2
3
]
C
h
e
st
X
-
r
a
y
C
o
r
o
N
e
t
(
X
c
e
p
t
i
o
n
)
99
8
9
.
9
3
3
.
9
[
2
4
]
C
h
e
st
X
-
r
a
y
D
e
Tr
a
C
9
5
.
1
2
9
7
.
9
1
-
[
2
5
]
C
h
e
st
X
-
r
a
y
C
N
N
(
V
G
G
1
6
)
95
96
1
6
.
3
[
2
6
]
C
h
e
st
X
-
r
a
y
4S
-
D
T(
R
e
sN
e
t
)
9
7
.
5
4
9
7
.
8
8
-
P
r
o
p
o
se
d
S
t
u
d
i
e
s
C
h
e
st
X
-
r
a
y
VGG
-
16
9
8
.
3
3
1
0
0
1
6
.
3
P
r
o
p
o
se
d
C
N
N
9
6
.
6
6
1
0
0
5
.
6
C
h
ex
Net
[
1
0
]
is
a
n
ew
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
th
at
was
p
r
o
p
o
s
ed
f
o
r
ch
est
X
-
r
ay
p
n
eu
m
o
n
ia
d
etec
tio
n
.
C
h
ex
Net
h
as
o
b
tain
ed
o
u
ts
tan
d
in
g
r
esu
lts
th
at
h
a
s
s
u
r
p
ass
ed
th
e
o
v
er
all
o
u
tp
u
t
o
f
th
e
r
ad
io
lo
g
is
t.
C
h
estNet
i
s
an
o
th
er
s
im
ilar
s
o
lu
tio
n
.
C
h
estNet
is
an
ad
v
an
c
ed
ty
p
e
o
f
C
NN
in
ten
d
ed
to
d
etec
t
ch
est
d
is
ea
s
e
s
o
n
ch
est
r
ad
io
g
r
a
p
h
y
im
ag
es
[
2
7
]
.
Un
ce
r
tai
n
ty
esti
m
atin
g
b
y
u
s
in
g
B
ay
esian
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
B
C
NN)
an
d
R
esn
et5
an
d
O
V2
as
p
r
e
-
tr
ai
n
ed
m
o
d
els
to
t
r
an
s
f
er
th
e
k
n
o
wled
g
e
was
p
r
o
p
o
s
ed
b
y
Gh
o
s
h
al
an
d
T
u
c
k
er
[
2
8
]
t
o
en
h
a
n
ce
t
h
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
d
iag
n
o
s
is
b
y
u
s
in
g
C
OVI
D
-
1
9
ch
est
X
-
r
ay
im
ag
es.
T
h
e
au
th
o
r
u
s
ed
th
e
p
r
e
d
ictiv
e
u
n
ce
r
tain
ty
(
PH)
co
n
ce
p
t
to
a
v
o
id
f
alse
p
r
ed
ictio
n
s
a
n
d
a
u
g
m
en
ted
d
if
f
e
r
en
t
d
atasets
f
r
o
m
d
if
f
er
e
n
t so
u
r
ce
s
to
in
cr
ea
s
e
th
e
am
o
u
n
t o
f
d
at
a.
C
o
r
o
Net
was
a
p
r
o
p
o
s
ed
C
NN
ar
ch
itectu
r
e
b
y
Kh
an
et
a
l.
[
2
3
]
.
T
h
e
au
th
o
r
c
o
m
p
a
r
ed
h
i
s
tech
n
iq
u
e
wh
ich
is
C
o
r
o
Net
with
an
o
t
h
er
tech
n
iq
u
e
wh
ich
is
C
OVI
D
-
n
et
an
d
h
as
ac
h
iev
e
d
b
et
ter
r
esu
lts
f
r
o
m
it.
Fu
r
th
er
m
o
r
e
,
th
e
a
u
th
o
r
co
m
b
in
ed
two
p
u
b
lic
d
atasets
f
r
o
m
d
if
f
er
en
t
s
o
u
r
ce
s
.
He
h
a
d
1
,
3
0
0
im
ag
es
wh
ich
h
e
en
lar
g
ed
u
s
in
g
a
u
g
m
e
n
tatio
n
t
ec
h
n
iq
u
es.
He
n
ce
,
th
e
ac
h
iev
e
d
ac
cu
r
ac
y
was
8
9
.
5
%
f
o
r
4
-
cl
ass
ca
s
es,
9
5
%
f
o
r
3
-
ca
ls
s
es
an
d
9
9
%
f
o
r
b
in
a
r
y
class
.
T
h
e
p
er
f
o
r
m
an
ce
ca
n
b
e
f
u
r
th
e
r
en
h
a
n
ce
d
o
n
ce
ad
d
itio
n
al
tr
ain
in
g
d
ata
ar
e
av
ailab
le.
Fu
r
th
er
m
o
r
e,
C
o
r
o
Net
also
n
ee
d
s
to
u
n
d
er
g
o
clin
ical
tr
ial
s.
Ou
r
p
r
o
p
o
s
ed
m
eth
o
d
will
b
e
tr
ain
ed
to
class
if
y
co
r
o
n
a
v
i
r
u
s
v
er
s
u
s
n
o
n
-
c
o
r
o
n
av
ir
u
s
ca
s
es.
Acc
o
r
d
in
g
to
th
e
r
o
b
u
s
tn
ess
o
f
th
e
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
in
im
ag
e
class
if
icatio
n
,
o
u
r
p
r
o
p
o
s
ed
m
eth
o
d
o
u
tp
e
r
f
o
r
m
s
ex
p
er
t
d
iag
n
o
s
is
esp
ec
ially
wh
en
we
h
a
v
e
a
lar
g
e
am
o
u
n
t
o
f
d
ata.
C
o
m
m
o
n
ly
,
C
NN
g
iv
es
a
h
ig
h
p
e
r
ce
n
t
ag
e
o
f
ac
cu
r
ac
y
in
im
ag
e
class
if
icatio
n
o
f
u
s
u
ally
m
o
r
e
th
a
n
9
0
% a
cc
u
r
ac
y
,
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
.
Ou
r
p
r
o
p
o
s
ed
m
eth
o
d
ca
n
b
e
u
tili
ze
d
in
h
o
s
p
itals
as
a
f
i
r
s
t
-
lin
e
f
o
r
c
o
r
o
n
av
ir
u
s
d
iag
n
o
s
is
.
T
h
e
m
ain
c
o
n
tr
ib
u
tio
n
o
f
o
u
r
r
esear
ch
is
to
en
h
an
ce
C
OVI
D
-
1
9
d
iag
n
o
s
tics
b
y
r
ed
u
cin
g
th
e
c
o
m
p
u
tatio
n
al
co
m
p
lex
ity
wh
ich
lea
d
s
to
a
d
ec
r
ea
s
e
in
tr
ain
in
g
tim
e,
im
p
r
o
v
e
th
e
s
en
s
itiv
ity
b
y
in
cr
ea
s
in
g
th
e
n
u
m
b
er
s
o
f
tr
u
e
p
o
s
itiv
e,
a
n
d
r
e
d
u
ce
th
e
n
u
m
b
er
s
o
f
f
alse n
eg
ativ
e.
I
n
ad
d
itio
n
,
th
e
lear
n
in
g
p
r
o
ce
s
s
was im
p
r
o
v
e
d
b
y
u
s
in
g
d
if
f
er
en
t a
u
g
m
en
tat
io
n
tech
n
iq
u
es.
5.
CO
NCLU
SI
O
N
I
n
th
is
s
tu
d
y
,
we
p
r
o
p
o
s
ed
a
n
ew
C
NN
ar
ch
itectu
r
e
an
d
c
o
m
p
ar
ed
its
p
er
f
o
r
m
a
n
ce
with
o
th
er
p
r
e
-
tr
ain
ed
n
eu
r
al
n
etwo
r
k
s
(
VGG1
6
an
d
in
ce
p
tio
n
V
3
)
f
o
r
C
OVI
D
-
1
9
d
etec
tio
n
b
ased
o
n
X
-
r
a
y
im
ag
es.
C
OVI
D
-
1
9
is
s
till
a
n
ew
p
an
d
em
ic
an
d
th
e
lack
o
r
lo
w
s
am
p
le
s
ize
is
a
ch
allen
g
e,
esp
ec
ially
wh
en
u
s
in
g
d
ee
p
lear
n
in
g
b
ec
au
s
e
it
n
ee
d
s
b
i
g
d
ata
to
b
o
o
s
t
th
e
p
er
f
o
r
m
an
ce
.
Hen
ce
,
d
ata
au
g
m
en
tatio
n
with
d
if
f
er
en
t
te
ch
n
i
q
u
es
was
u
s
ed
.
Mo
r
eo
v
er
,
t
r
an
s
f
er
lear
n
in
g
was
also
u
s
ed
with
f
in
e
-
tu
n
in
g
to
o
v
e
r
r
id
e
th
e
lo
w
s
a
m
p
le
s
ize
ch
allen
g
e.
T
h
e
r
esu
lt
of
th
e
v
alid
atio
n
s
et
s
h
o
wed
th
at
VGG1
6
an
d
o
u
r
p
r
o
p
o
s
ed
C
NN
o
u
tp
er
f
o
r
m
s
I
n
ce
p
tio
n
V
3
.
Alth
o
u
g
h
VGG1
6
is
s
lig
h
tly
b
etter
in
b
o
th
tr
ain
in
g
an
d
test
in
g
ac
c
u
r
ac
y
,
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
o
v
e
r
ca
m
e
o
th
e
r
n
etwo
r
k
s
in
co
m
p
u
tatio
n
al
c
o
m
p
lex
ity
b
y
r
ed
u
cin
g
th
e
n
u
m
b
er
o
f
h
y
p
er
-
p
a
r
am
eter
s
t
h
at
lead
to
r
e
d
u
cin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t
J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
5
3
1
-
4
5
4
1
4540
tr
ain
in
g
tim
e
s
ig
n
if
ican
tly
.
F
o
r
f
u
tu
r
e
wo
r
k
,
a
d
esig
n
tec
h
n
iq
u
e
is
n
ee
d
e
d
th
at
ca
n
a
n
aly
ze
th
e
im
ag
e
to
m
ea
s
u
r
e
th
e
p
er
ce
n
tag
e
o
f
l
u
n
g
v
o
lu
m
e
th
at
is
in
f
ec
ted
b
y
th
e
d
is
ea
s
e.
I
n
a
d
d
itio
n
,
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
was
n
o
t
co
m
p
ar
ed
with
r
ad
io
lo
g
is
ts
.
So
,
f
u
tu
r
e
s
tu
d
ies
co
u
ld
p
lan
t
o
co
m
p
a
r
e
it
with
r
ad
io
lo
g
is
ts
.
RE
F
E
R
E
NC
E
S
[1
]
N.
Zh
u
,
D.
Zh
a
n
g
,
W.
Wan
g
,
X.
Li
,
B.
Ya
n
g
,
J.
S
o
n
g
e
t
a
l.
,
“
A
n
o
v
e
l
c
o
r
o
n
a
v
iru
s
f
ro
m
p
a
ti
e
n
ts
wit
h
p
n
e
u
m
o
n
ia
in
Ch
in
a
,
”
Ne
w
E
n
g
l
a
n
d
j
o
u
rn
a
l
o
f
me
d
icin
e
,
v
o
l.
3
8
2
,
n
o
.
8
,
p
p
.
7
2
7
-
7
3
3
,
2
0
1
9
,
d
o
i:
1
0
.
1
0
5
6
/NEJ
M
o
a
2
0
0
1
0
1
7
.
[2
]
Q.
Li
,
X.
G
u
a
n
,
P
.
W
u
,
X.
Wan
g
,
L
.
Z
h
o
u
,
Y.
To
n
g
e
t
a
l.
,
“
Ear
ly
Tran
sm
issio
n
D
y
n
a
m
ics
in
W
u
h
a
n
,
Ch
i
n
a
,
o
f
No
v
e
l
C
o
ro
n
a
v
ir
u
s
–
In
fe
c
ted
P
n
e
u
m
o
n
ia,
”
Ne
w
En
g
la
n
d
J
o
u
rn
a
l
o
f
M
e
d
icin
e
,
v
o
l
.
3
8
2
,
n
o
.
1
3
,
p
p
.
1
1
9
9
-
1
2
0
7
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
5
6
/NEJ
M
o
a
2
0
0
1
3
1
6
.
[3
]
D.
Wan
g
,
B
.
H
u
,
C
.
Hu
,
F
.
Zh
u
,
X.
Li
u
,
J.
Z
h
a
n
g
e
t
a
l.
,
“
Cli
n
ica
l
Ch
a
ra
c
teristics
o
f
1
3
8
Ho
s
p
it
a
li
z
e
d
P
a
ti
e
n
ts
wit
h
2
0
1
9
No
v
e
l
C
o
ro
n
a
v
iru
s
-
I
n
fe
c
ted
P
n
e
u
m
o
n
ia
in
W
u
h
a
n
,
Ch
in
a
,
”
J
AM
A
T
h
e
J
o
u
rn
a
l
o
f
t
h
e
Am
e
ric
a
n
M
e
d
ic
a
l
Asso
c
ia
ti
o
n
,
v
o
l.
3
2
3
,
n
o
.
1
1
,
p
p
.
1
0
6
1
-
1
0
6
9
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
1
/j
a
m
a
.
2
0
2
0
.
1
5
8
5
.
[4
]
Y.
Z
h
a
n
g
,
X.
G
e
n
g
,
Y.
Tan
,
Q.
Li
,
C.
Xu
,
J.
Xu
e
t
a
l.
,
“
Ne
w
u
n
d
e
rsta
n
d
i
n
g
o
f
t
h
e
d
a
m
a
g
e
o
f
S
ARS
-
C
o
V
-
2
in
fe
c
ti
o
n
o
u
tsi
d
e
t
h
e
re
sp
irato
r
y
sy
ste
m
,
”
Bi
o
me
d
icin
e
a
n
d
Ph
a
rm
a
c
o
th
e
ra
p
y
,
v
o
l.
1
2
7
,
2
0
2
0
,
Art.
No
.
1
1
0
1
9
5
,
d
o
i:
1
0
.
1
0
1
6
/j
.
b
i
o
p
h
a
.
2
0
2
0
.
1
1
0
1
9
5
.
[5
]
I.
D.
Ap
o
st
o
l
o
p
o
u
l
o
s,
S
.
I.
Az
n
a
o
u
ri
d
is,
a
n
d
M
.
Tza
n
i
,
“
E
x
trac
ti
n
g
p
o
ss
ib
l
y
re
p
re
se
n
tativ
e
COV
ID
-
1
9
Bi
o
m
a
rk
e
rs
fro
m
X
-
Ra
y
ima
g
e
s
with
De
e
p
Lea
rn
in
g
a
p
p
r
o
a
c
h
a
n
d
ima
g
e
d
a
ta
re
late
d
to
P
u
lmo
n
a
ry
Dise
a
se
s,
”
J
o
u
rn
a
l
o
f
M
e
d
ica
l
a
n
d
B
io
l
o
g
ic
a
l
E
n
g
i
n
e
e
rin
g
,
v
o
l.
4
0
,
n
o
.
5
,
p
p
.
4
6
2
-
4
6
9
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
7
/s
4
0
8
4
6
-
0
2
0
-
0
0
5
2
9
-
4.
[6
]
J.
P
.
C
o
h
e
n
,
P
.
M
o
rr
iso
n
,
a
n
d
L.
Da
o
,
“
COV
ID
-
1
9
Im
a
g
e
Da
ta Co
ll
e
c
ti
o
n
,
”
a
rXiv p
re
p
ri
n
t,
2
0
2
0
.
[7
]
P
a
u
l
M
o
o
n
e
y
,
“
C
h
e
st
X
-
Ra
y
Im
a
g
e
s
(P
n
e
u
m
o
n
ia),
”
Ka
g
g
le
,
2
0
2
0
,
[O
n
li
n
e
].
Av
a
ib
le
:
h
tt
p
s:/
/www
.
k
a
g
g
le.co
m
/p
a
u
lt
im
o
th
y
m
o
o
n
e
y
/ch
e
st
-
x
ra
y
-
p
n
e
u
m
o
n
ia
[8
]
X.
Xie
,
Z.
Zh
o
n
g
,
W
.
Zh
a
o
,
C.
Zh
e
n
g
,
F
.
Wan
g
,
a
n
d
J.
Li
u
,
“
Ch
e
st
CT
fo
r
Ty
p
ica
l
2
0
1
9
-
n
Co
V
P
n
e
u
m
o
n
ia:
Re
latio
n
sh
i
p
to
Ne
g
a
ti
v
e
RT
-
P
CR
Tes
ti
n
g
,
”
R
a
d
i
o
l
o
g
y
,
v
o
l.
2
9
6
,
n
o
.
2
,
2
0
2
0
,
Art.
No
.
2
0
0
3
4
3
,
d
o
i:
1
0
.
1
1
4
8
/rad
i
o
l.
2
0
2
0
2
0
0
3
4
3
.
[9
]
T.
Ra
h
m
a
n
,
M
.
E.
Ch
o
wd
h
u
r
y
,
A.
Kh
a
n
d
a
k
a
r,
K.
R.
Isla
m
,
K.
F
.
Isla
m
,
Z.
B.
M
a
h
b
u
b
e
t
a
l.
,
“
Tr
a
n
sfe
r
Lea
rn
in
g
with
De
e
p
Co
n
v
o
lu
t
io
n
a
l
Ne
u
ra
l
Ne
two
rk
(
CNN
)
f
o
r
P
n
e
u
m
o
n
ia
De
tec
ti
o
n
Us
in
g
,
”
Bi
o
M
e
d
i
c
a
l
E
n
in
e
e
rin
g
,
v
o
l.
1
0
,
n
o
.
9
,
2
0
2
0
,
Art.
No
.
3
2
3
3
,
d
o
i:
1
0
.
3
3
9
0
/ap
p
1
0
0
9
3
2
3
3
.
[1
0
]
P
.
Ra
jp
u
r
k
a
r,
Ir
v
in
,
J.,
Z
h
u
,
K.,
Ya
n
g
,
B.
,
M
e
h
ta,
H.,
Du
a
n
,
T
e
t
a
l.
,
“
Ch
e
XN
e
t:
Ra
d
io
l
o
g
ist
-
Lev
e
l
P
n
e
u
m
o
n
i
a
De
tec
ti
o
n
o
n
Ch
e
st X
-
Ra
y
s wi
th
De
e
p
Lea
rn
in
g
,
”
a
rXiv
p
re
p
rin
t
,
2
0
1
7
.
[1
1
]
K.
S
imo
n
y
a
n
a
n
d
A.
Zi
ss
e
rm
a
n
,
“
v
e
ry
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
tw
o
rk
s
fo
r
larg
e
-
sc
a
le
ima
g
e
re
c
o
g
n
it
i
o
n
,
"
a
rXiv
p
re
p
rin
t
,
2
0
1
5
.
[1
2
]
M
.
S
a
jj
a
d
,
S
.
Kh
a
n
,
K.
M
u
h
a
m
m
a
d
,
W.
Wu
,
A.
Ullah
,
a
n
d
S.
W.
Ba
ik
,
“
M
u
lt
i
-
g
ra
d
e
b
ra
in
tu
m
o
r
c
las
sifica
ti
o
n
u
sin
g
d
e
e
p
CNN
wit
h
e
x
ten
si
v
e
d
a
ta
a
u
g
m
e
n
tati
o
n
,
”
J
o
u
rn
a
l
o
f
Co
mp
u
ta
ti
o
n
a
l
S
c
ien
c
e
,
v
o
l.
3
0
,
p
p
.
1
7
4
-
1
8
2
,
d
o
i:
1
0
.
1
0
1
6
/j
.
jo
c
s.2
0
1
8
.
1
2
.
0
0
3
.
[1
3
]
J.
Yo
sin
sk
i,
J.
Cl
u
n
e
,
Y.
Be
n
g
i
o
,
a
n
d
H.
Li
p
so
n
,
“
Ho
w
tran
sfe
ra
b
le
a
re
fe
a
tu
re
s
in
d
e
e
p
n
e
u
ra
l
n
e
two
rk
s?
,
”
a
rXi
v
p
re
p
rin
t
,
p
p
.
3
3
2
0
-
3
3
2
8
,
2
0
1
4
.
[1
4
]
C.
S
z
e
g
e
d
y
,
V.
Va
n
h
o
u
c
k
e
,
S
.
Io
ffe
,
a
n
d
J.
S
h
len
s,
“
Re
th
in
k
in
g
th
e
In
c
e
p
ti
o
n
Arc
h
it
e
c
tu
re
fo
r
Co
m
p
u
ter
Visi
o
n
,
”
in
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
IEE
E
c
o
n
fe
re
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
a
n
d
p
a
t
ter
n
re
c
o
g
n
it
io
n
,
p
p
.
2
8
1
8
-
2
8
2
6
,
2
0
1
5
.
[1
5
]
O.
G
o
z
e
s,
M
.
F
rid
-
A
d
a
r,
H.
G
re
e
n
sp
a
n
,
P
.
D.
Bro
w
n
in
g
,
H.
Zh
a
n
g
,
W.
Ji
e
t
a
l.
,
“
Ra
p
id
AI
De
v
e
lo
p
m
e
n
t
Cy
c
le
f
o
r
th
e
Co
ro
n
a
v
ir
u
s
(COV
ID
-
1
9
)
P
a
n
d
e
m
ic:
In
it
ial
Re
su
lt
s
f
o
r
Au
t
o
m
a
ted
De
tec
ti
o
n
a
n
d
P
a
ti
e
n
t
M
o
n
it
o
rin
g
u
si
n
g
De
e
p
Lea
rn
in
g
CT
Im
a
g
e
An
a
ly
si
s Au
th
o
rs,
”
a
rXiv
p
re
p
rin
t
,
2
0
2
0
.
[1
6
]
J.
Ch
e
n
,
L.
Wu
,
J.
Zh
a
n
g
,
L
.
Zh
a
n
g
,
D.
G
o
n
g
,
Y.
Z
h
a
o
e
t
a
l
.
,
“
De
e
p
lea
rn
in
g
-
b
a
se
d
m
o
d
e
l
fo
r
d
e
tec
ti
n
g
2
0
1
9
n
o
v
e
l
c
o
ro
n
a
v
iru
s
p
n
e
u
m
o
n
ia
o
n
h
i
g
h
-
r
e
so
lu
ti
o
n
c
o
m
p
u
ted
t
o
m
o
g
ra
p
h
y
:
a
p
ro
sp
e
c
ti
v
e
st
u
d
y
,
”
S
c
ien
ti
fi
c
R
e
p
o
rts
,
v
o
l.
1
0
,
n
o
.
1
,
p
p
.
1
-
1
1
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
3
8
/s4
1
5
9
8
-
0
2
0
-
7
6
2
8
2
-
0.
[1
7
]
Z.
P
a
m
u
k
,
A.
Na
ri
n
,
a
n
d
C.
Ka
y
a
,
“
Au
to
m
a
ti
c
De
tec
ti
o
n
o
f
Co
r
o
n
a
v
iru
s
Dise
a
se
(COV
ID
-
1
9
)
Us
in
g
X
-
ra
y
Im
a
g
e
s
a
n
d
De
e
p
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
s,
”
a
rXiv
p
re
p
rin
t
,
2
0
20
.
[1
8
]
L.
Wan
g
,
Z
.
Qiu
Li
n
,
a
n
d
A.
W
o
n
g
,
“
COV
ID
-
Ne
t:
A
Tailo
re
d
De
e
p
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
De
sig
n
fo
r
De
tec
ti
o
n
o
f
COV
ID
-
1
9
Ca
se
s
fro
m
Ch
e
st
X
-
Ra
y
Im
a
g
e
s,
”
S
c
ien
ti
fi
c
Rep
o
rts
,
v
o
l.
1
0
,
n
o
.
1
,
p
p
.
1
-
1
2
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
3
8
/s4
1
5
9
8
-
0
2
0
-
7
6
5
5
0
-
z.
[1
9
]
I
.
D.
Ap
o
sto
l
o
p
o
u
l
o
s
a
n
d
T.
A.
M
p
e
sia
n
a
,
“
Co
v
id
-
1
9
:
a
u
t
o
m
a
ti
c
d
e
tec
ti
o
n
fro
m
X
-
ra
y
ima
g
e
s
u
t
il
izin
g
tran
sfe
r
lea
rn
in
g
wit
h
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s,
”
Ph
y
sic
a
l
a
n
d
En
g
in
e
e
rin
g
S
c
ien
c
e
s
in
M
e
d
ici
n
e
,
v
o
l.
43
,
n
o
.
2
,
p
p
.
6
3
5
-
6
4
0
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
7
/s
1
3
2
4
6
-
0
2
0
-
0
0
8
6
5
-
4
.
[2
0
]
P.
K.
S
e
th
y
,
K.
S
a
n
ti
,
Be
h
e
ra
,
P
.
Ku
m
a
r,
P
.
Biswa
s,
“
De
tec
ti
o
n
o
f
c
o
ro
n
a
v
iru
s
Dise
a
se
(COV
ID
-
1
9
)
b
a
se
d
o
n
De
e
p
F
e
a
tu
re
s
a
n
d
S
u
p
p
o
rt
Ve
c
to
r
M
a
c
h
in
e
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
M
a
th
e
ma
ti
c
a
l,
E
n
g
i
n
e
e
rin
g
a
n
d
M
a
n
a
g
e
me
n
t
S
c
ien
c
e
s,
v
o
l
.
5
,
p
p
.
6
4
3
-
6
5
1
,
2
0
2
0
,
d
o
i:
1
0
.
3
3
8
8
9
/IJ
M
EM
S
.
2
0
2
0
.
5
.
4
.
0
5
2
.
[2
1
]
E.
E
.
He
m
d
a
n
,
M
.
A.
S
h
o
u
m
a
n
,
a
n
d
M
.
E.
Ka
ra
r,
“
COV
IDX
-
Ne
t :
A
F
ra
m
e
wo
rk
o
f
De
e
p
Lea
rn
i
n
g
Clas
sifiers
t
o
Dia
g
n
o
se
COV
ID
-
1
9
in
X
-
Ra
y
I
m
a
g
e
s,
”
a
rXiv p
re
p
rin
t
,
2
0
2
0
.
[2
2
]
T.
Oz
tu
rk
,
M
.
Talo
,
A.
Yild
iri
m
,
U.
B
.
Ba
lo
g
lu
,
O.
Yild
iri
m
,
a
n
d
U.
R
.
Ac
h
a
ry
a
,
“
Au
to
m
a
ted
d
e
tec
ti
o
n
o
f
COV
ID
-
1
9
c
a
se
s
u
sin
g
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
s
with
X
-
ra
y
ima
g
e
s,
”
Co
mp
u
ter
s
in
Bi
o
l
o
g
y
a
n
d
M
e
d
ici
n
e
,
v
o
l.
1
2
1
,
2
0
2
0
,
Art,
No
.
1
0
3
7
9
2
,
d
o
i:
1
0
.
1
0
1
6
/j
.
c
o
m
p
b
i
o
m
e
d
.
2
0
2
0
.
1
0
3
7
9
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.