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.
15
,
No
.
5
,
Octo
b
er
20
25
,
p
p
.
4
7
5
1
~
4
7
6
1
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
5
.
pp
4
7
5
1
-
4
7
6
1
4751
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Efficien
t
ma
s
k
r
e
g
io
n
-
ba
sed co
nv
o
lutiona
l neural n
et
wo
rk
-
ba
sed a
rchitec
tur
e
for C
O
VI
D
-
19
detec
tion
from
co
mputed
tom
o
g
ra
phy
da
ta
Na
der
M
a
hm
o
ud
1,
2
,
Ash
ra
f
B
.
E
l
-
Si
s
i
1
1
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
D
e
p
a
r
t
me
n
t
,
F
a
c
u
l
t
y
o
f
C
o
m
p
u
t
e
r
s
a
n
d
I
n
f
o
r
mat
i
o
n
,
M
e
n
o
u
f
i
a
U
n
i
v
e
r
si
t
y
,
S
h
e
b
i
n
El
-
K
o
m,
E
g
y
p
t
2
C
y
b
e
r
s
e
c
u
r
i
t
y
D
e
p
a
r
t
m
e
n
t
,
En
g
i
n
e
e
r
i
n
g
a
n
d
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
C
o
l
l
e
g
e
,
B
u
r
a
y
d
a
h
P
r
i
v
a
t
e
C
o
l
l
e
g
e
s,
B
u
r
a
y
d
a
h
,
K
i
n
g
d
o
m
o
f
S
a
u
d
i
A
r
a
b
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
1
4
,
2
0
2
5
R
ev
is
ed
J
u
n
5
,
2
0
2
5
Acc
ep
ted
J
u
n
3
0
,
2
0
2
5
Th
e
wo
rl
d
wid
e
e
ffe
c
t
o
f
th
e
c
o
r
o
n
a
v
iru
s
d
ise
a
se
(COV
ID
-
19)
p
a
n
d
e
m
ic
h
a
s
b
e
e
n
c
a
tas
tro
p
h
ic,
lea
d
in
g
to
a
si
g
n
ifi
c
a
n
t
n
u
m
b
e
r
o
f
fa
talit
ies
w
o
r
ld
wid
e
.
In
re
sp
o
n
se
to
th
e
o
u
t
b
re
a
k
,
h
e
a
lt
h
c
a
re
in
stit
u
ti
o
n
s
h
a
v
e
p
ro
p
o
se
d
t
h
e
u
se
o
f
c
h
e
st
c
o
m
p
u
ted
to
m
o
g
ra
p
h
y
(CT
)
a
s
a
n
imp
o
rtan
t
d
iag
n
o
sis
t
o
o
l
fo
r
ra
p
i
d
d
iag
n
o
sis,
lev
e
ra
g
i
n
g
d
e
e
p
lea
rn
i
n
g
a
p
p
r
o
a
c
h
e
s
f
o
r
d
ise
a
se
d
e
tec
t
io
n
.
T
h
is
p
a
p
e
r
a
ims
to
p
r
o
g
re
ss
a
r
o
b
u
st
m
e
th
o
d
o
lo
g
y
to
wa
rd
s
a
c
c
u
ra
te
d
i
a
g
n
o
sis
o
f
COV
ID
-
1
9
b
a
se
d
o
n
d
e
e
p
lea
rn
i
n
g
a
p
p
ro
a
c
h
e
s
with
c
h
e
st
CT
ima
g
e
s.
We
p
ro
p
o
se
a
m
a
sk
re
g
io
n
-
b
a
se
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
(
M
a
sk
R
-
CNN
)
m
o
d
e
l
a
rc
h
it
e
c
tu
re
th
a
t
is
we
ll
-
t
ra
in
e
d
a
n
d
u
se
d
to
d
isc
rimin
a
te
b
e
twe
e
n
COV
ID
-
19
-
in
fe
c
ted
a
n
d
u
n
in
fe
c
ted
c
a
se
s.
In
o
r
d
e
r
to
imp
ro
v
e
fe
a
tu
re
e
x
trac
ti
o
n
,
t
h
e
p
r
o
p
o
se
d
m
o
d
e
l
in
c
o
r
p
o
ra
tes
a
fu
z
z
y
c
o
l
o
r
e
n
h
a
n
c
e
m
e
n
t
p
re
p
ro
c
e
ss
in
g
tec
h
n
iq
u
e
th
a
t
re
d
u
c
e
s
ima
g
e
fu
z
z
in
e
ss
a
n
d
i
n
c
re
a
se
s
c
o
n
tras
t
.
A
p
u
b
li
c
l
y
a
v
a
il
a
b
le
c
h
e
st
C
T
d
a
tas
e
t
is
c
o
n
sid
e
re
d
fo
r
q
u
a
n
ti
tati
v
e
e
v
a
lu
a
ti
o
n
o
f
t
h
e
p
ro
p
o
se
d
a
rc
h
it
e
c
tu
re
m
o
d
e
l,
w
h
ich
in
c
l
u
d
e
s
v
a
rio
u
s
fro
n
tal
ima
g
e
v
iew
s
o
f
COV
ID
-
1
9
a
n
d
n
o
n
-
COV
ID
-
1
9
c
a
se
s.
Th
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
y
iel
d
e
d
a
n
a
c
c
u
ra
c
y
o
f
9
8
.
8
%
with
9
8
.
4
%
p
re
c
isio
n
a
n
d
9
8
.
5
%
re
c
a
ll
.
Ad
d
it
io
n
a
ll
y
,
t
h
e
p
r
o
p
o
se
d
m
o
d
e
l
a
rc
h
it
e
c
tu
re
h
a
s
b
e
e
n
q
u
a
n
ti
tati
v
e
ly
e
v
a
lu
a
ted
i
n
c
o
m
p
a
riso
n
wit
h
b
e
n
c
h
m
a
rk
a
p
p
ro
a
c
h
e
s,
y
ield
i
n
g
su
p
e
rio
r
p
e
rfo
rm
a
n
c
e
in
term
s o
f
c
o
n
v
e
n
ti
o
n
a
l
e
v
a
l
u
a
ti
o
n
m
e
tri
c
s
.
K
ey
w
o
r
d
s
:
C
h
est co
m
p
u
ted
to
m
o
g
r
ap
h
y
C
OVI
D
-
19
Dee
p
lear
n
in
g
Fu
zz
y
co
lo
r
tec
h
n
iq
u
e
Ma
s
k
r
eg
io
n
-
b
ased
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
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
:
Nad
er
Ma
h
m
o
u
d
C
o
m
p
u
ter
Scien
ce
Dep
ar
tm
e
n
t,
Facu
lty
o
f
C
o
m
p
u
ter
s
an
d
I
n
f
o
r
m
atio
n
,
Me
n
o
u
f
ia
Un
iv
er
s
i
ty
Sh
eb
in
E
l
-
Ko
m
,
E
g
y
p
t
E
m
ail: n
ad
er
.
m
a
h
m
o
u
d
@
ci.
m
en
o
f
ia.
ed
u
.
eg
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
co
r
o
n
av
ir
u
s
d
is
ea
s
e
(
C
O
VI
D
-
1
9
)
r
is
in
g
in
W
u
h
an
,
C
h
in
a,
2
0
1
9
,
h
as
q
u
ick
l
y
ev
o
l
v
ed
in
to
a
g
lo
b
al
h
ea
lth
c
r
is
is
.
T
h
e
v
ir
u
s
s
p
r
ea
d
s
p
r
im
a
r
ily
t
h
r
o
u
g
h
r
esp
ir
ato
r
y
d
r
o
p
lets
,
th
u
s
ac
c
u
r
ate
an
d
e
f
f
icien
t
d
iag
n
o
s
is
is
v
ital
in
p
r
ev
en
tin
g
tr
an
s
m
is
s
io
n
an
d
m
an
ag
in
g
p
atien
t
ca
r
e.
T
h
e
r
ev
er
s
e
tr
an
s
cr
ip
tio
n
p
o
ly
m
e
r
ase
ch
ain
r
ea
ctio
n
(
R
T
-
PC
R
)
was
th
e
p
r
in
cip
le
d
iag
n
o
s
tic
tech
n
i
q
u
e
th
at
h
as
b
ee
n
u
s
ed
g
lo
b
ally
in
o
r
d
er
t
o
d
etec
t
v
ir
al
r
ib
o
n
u
cleic
ac
i
d
(
R
NA
)
f
r
o
m
n
aso
p
h
ar
y
n
g
ea
l
s
wab
s
[
1
]
.
W
h
ile
R
T
-
PC
R
is
s
p
ec
if
ic
an
d
wid
ely
ad
o
p
ted
,
it
s
u
f
f
er
s
f
r
o
m
a
n
u
m
b
er
o
f
lim
itatio
n
s
,
s
u
ch
as
l
o
n
g
p
r
o
ce
s
s
in
g
tim
es,
h
ig
h
d
e
p
en
d
e
n
cy
o
n
s
p
ec
ialized
r
ea
g
en
ts
an
d
e
q
u
ip
m
e
n
t,
an
d
m
o
d
er
ate
s
en
s
itiv
ity
r
ates,
with
s
tu
d
ies
r
ep
o
r
tin
g
s
en
s
itiv
ities
as
lo
w
as
7
1
%
[
2
]
.
C
o
m
p
u
ted
to
m
o
g
r
a
p
h
y
(
C
T
)
i
m
ag
in
g
is
a
well
-
k
n
o
wn
m
ed
i
ca
l
im
ag
in
g
m
o
d
ality
th
at
all
o
ws
f
o
r
n
o
n
-
in
v
asiv
e
v
is
u
aliza
tio
n
o
f
in
ter
n
al
b
o
d
y
s
tr
u
ctu
r
es
an
d
is
wid
ely
u
tili
z
ed
in
a
v
ar
iety
o
f
ap
p
licatio
n
s
[
3
]
.
C
T
s
ca
n
s
h
av
e
d
em
o
n
s
tr
ated
s
ig
n
if
ican
t
s
e
n
s
itiv
ity
wh
en
co
n
s
id
er
ed
f
o
r
d
etec
tin
g
C
OVI
D
-
19
-
r
elate
d
p
u
lm
o
n
ar
y
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
5
1
-
4
7
6
1
4752
ab
n
o
r
m
alities
—
r
ep
o
r
ted
to
b
e
as
h
ig
h
as
9
8
%
[
4
]
.
C
T
s
ca
n
s
ar
e
ca
p
ab
le
o
f
ex
p
lo
itin
g
ch
ar
ac
ter
is
tic
f
ea
tu
r
es
s
u
ch
as
g
r
o
u
n
d
-
g
lass
o
p
ac
ities
an
d
o
th
er
lu
n
g
ir
r
eg
u
lar
ities
,
ev
en
with
p
atien
ts
wh
o
ar
e
asy
m
p
to
m
atic
o
r
h
av
e
n
eg
ativ
e
R
T
-
PC
R
.
As
a
r
esu
lt,
s
ev
er
al
h
ea
lth
ca
r
e
in
s
titu
tio
n
s
h
av
e
a
d
o
p
ted
C
T
im
ag
in
g
a
s
a
co
m
p
lem
e
n
tar
y
o
r
alter
n
ati
v
e
d
ia
g
n
o
s
tic
to
o
l,
p
ar
ticu
lar
ly
d
u
r
i
n
g
ea
r
ly
-
s
tag
e
in
f
ec
tio
n
s
o
r
in
s
ce
n
ar
io
s
r
e
q
u
ir
in
g
r
ap
id
p
atien
t
tr
iag
e
[
5
]
.
Desp
ite
its
ad
v
an
tag
es,
ac
cu
r
ate
in
ter
p
r
etatio
n
o
f
ch
est
C
T
im
ag
es
is
s
till
v
er
y
d
if
f
icu
lt,
wh
er
e
d
iag
n
o
s
tic
in
d
icato
r
s
ca
n
v
ar
y
wid
ely
d
ep
e
n
d
in
g
o
n
th
e
s
tag
e
o
f
in
f
ec
tio
n
a
n
d
w
h
eth
e
r
co
m
o
r
b
id
ities
ar
e
p
r
esen
t.
Mo
r
eo
v
er
,
m
a
n
u
al
i
n
ter
p
r
etatio
n
is
tim
e
-
co
n
s
u
m
i
n
g
an
d
also
s
u
b
ject
to
i
n
ter
-
o
b
s
er
v
er
v
ar
iab
ilit
y
t
h
at
ty
p
ically
in
cr
ea
s
es
an
d
b
ec
o
m
es
an
is
s
u
e
in
ca
s
e
o
f
h
ig
h
p
at
ien
t
v
o
lu
m
es
d
u
r
in
g
th
e
p
an
d
em
ic.
Du
e
t
o
th
ese
lim
itatio
n
s
,
th
er
e
is
in
cr
ea
s
in
g
in
ter
est
in
u
s
in
g
th
e
ap
p
licati
o
n
o
f
ar
tific
ial
in
tellig
en
ce
a
n
d
d
ee
p
lear
n
in
g
to
m
ed
ical
im
ag
in
g
,
in
o
r
d
er
t
o
p
r
o
v
i
d
e
f
ast,
co
n
s
is
ten
t,
an
d
s
ca
lab
le
d
iag
n
o
s
tic
ass
is
tan
c
e
[
6
]
.
Sev
e
r
al
d
ee
p
lear
n
in
g
m
o
d
els
h
av
e
b
ee
n
r
e
s
ea
r
ch
ed
r
ec
en
tly
f
o
r
C
OVI
D
-
1
9
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
task
s
f
r
o
m
ch
est
X
-
r
ay
an
d
C
T
m
ed
ical
im
ag
e
s
.
Pre
lim
in
ar
y
r
esear
ch
wo
r
k
s
h
av
e
ac
h
iev
ed
p
r
o
m
is
in
g
cl
ass
if
icatio
n
r
esu
lts
lev
er
ag
in
g
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
m
ed
ical
im
ag
in
g
d
atasets
[
7
]
,
[
8
]
.
Ho
wev
er
,
th
e
clin
ical
r
elev
an
ce
o
f
th
ese
m
o
d
els
was
lim
ited
d
u
e
to
th
eir
lack
o
f
s
p
atial
awa
r
en
ess
an
d
in
ab
ilit
y
to
lo
ca
te
d
is
ea
s
ed
tis
s
u
es.
Seg
m
en
tatio
n
b
ased
f
r
am
ewo
r
k
s
,
in
cl
u
d
in
g
U
-
Net,
wer
e
p
r
o
p
o
s
ed
to
tack
le
th
is
p
r
o
b
lem
,
wh
ich
allo
ws
f
o
r
p
ix
el
-
lev
el
d
etec
tio
n
an
d
id
e
n
tific
atio
n
o
f
in
f
e
ctio
n
r
eg
io
n
s
.
Desp
ite
th
e
ef
f
ec
tiv
en
ess
o
f
th
ese
m
o
d
els,
th
ey
s
till
s
tr
u
g
g
le
wit
h
g
en
er
alizin
g
to
im
a
g
es with
v
ar
iab
le
q
u
ality
o
r
s
u
b
tle
s
ig
n
s
o
f
d
is
ea
s
e.
Dif
f
er
en
t
ad
v
a
n
ce
d
s
eg
m
en
ta
tio
n
tech
n
iq
u
es
h
a
v
e
b
ee
n
r
e
s
ea
r
ch
ed
;
h
o
wev
er
,
m
ask
r
eg
io
n
-
b
ased
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
Ma
s
k
R
-
C
NN)
[
9
]
h
as
em
er
g
ed
as
an
ef
f
icien
t
an
d
p
o
wer
f
u
l
ap
p
r
o
ac
h
,
wh
ic
h
u
n
iq
u
ely
co
m
b
in
es
o
b
ject
d
ete
ctio
n
an
d
s
em
an
tic
s
eg
m
en
tati
o
n
in
a
u
n
if
ied
f
r
am
ewo
r
k
[
1
0
]
,
[
1
1
]
.
I
t
h
as
b
ee
n
s
tu
d
ied
an
d
s
h
o
wn
ef
f
ec
tiv
e
i
n
a
r
an
g
e
o
f
m
ed
ical
im
ag
i
n
g
task
s
,
in
clu
d
in
g
lu
n
g
n
o
d
u
le
d
etec
tio
n
[
1
2
]
,
liv
er
an
d
m
u
lti
-
o
r
g
an
s
eg
m
en
tatio
n
[
1
3
]
,
b
r
ea
s
t
tu
m
o
r
class
if
icatio
n
[
1
4
]
,
an
d
ea
r
ly
ca
n
ce
r
d
i
ag
n
o
s
is
[
1
5
]
.
Ma
s
k
R
-
C
NN
is
well
-
s
u
ited
f
o
r
id
en
tify
in
g
C
OVI
D
-
19
-
in
f
ec
te
d
r
eg
io
n
s
d
u
e
to
its
ab
ilit
y
t
o
d
eliv
er
ac
cu
r
ate,
in
s
tan
ce
-
lev
el,
p
ix
el
-
wis
e
p
r
e
d
ictio
n
s
.
Ho
wev
er
,
m
a
n
y
o
f
its
cu
r
r
en
t
ap
p
licatio
n
s
o
v
er
lo
o
k
th
e
im
p
o
r
tan
ce
o
f
p
r
ep
r
o
ce
s
s
in
g
,
esp
ec
ially
in
ca
s
es wh
er
e
lo
w
im
ag
e
co
n
tr
ast
m
ak
es in
f
ec
ted
r
e
g
io
n
s
h
ar
d
er
to
id
en
tify
.
T
o
t
a
c
k
l
e
t
h
es
e
d
r
a
w
b
a
c
k
s
,
w
e
p
r
o
p
o
s
e
a
n
o
v
e
l
f
r
a
m
e
w
o
r
k
t
h
a
t
b
r
i
n
g
s
t
o
g
e
t
h
e
r
a
Ma
s
k
R
-
C
N
N
-
b
a
s
ed
s
e
g
m
e
n
t
a
ti
o
n
m
o
d
e
l
w
i
t
h
f
u
zz
y
l
o
g
i
c
-
b
a
s
e
d
c
o
n
t
r
as
t
e
n
h
a
n
c
e
m
e
n
t
.
F
u
z
z
y
l
o
g
ic
is
w
e
ll
-
s
u
i
t
e
d
t
o
h
a
n
d
l
e
t
w
o
p
r
o
m
i
n
e
n
t
c
h
a
l
l
e
n
g
e
s
i
n
m
e
d
i
ca
l
i
m
a
g
e
a
n
a
l
y
s
is
,
w
h
i
c
h
a
r
e
u
n
c
e
r
t
a
i
n
t
y
a
n
d
a
m
b
i
g
u
i
t
y
.
T
h
e
p
r
o
p
o
s
e
d
a
p
p
r
o
a
c
h
t
r
a
n
s
f
o
r
m
s
C
T
i
m
a
g
es
i
n
t
o
a
f
u
z
z
y
d
o
m
a
i
n
u
s
i
n
g
a
d
a
p
t
i
v
e
f
u
z
z
i
f
i
e
r
s
t
o
i
d
e
n
t
i
f
y
c
r
o
s
s
o
v
e
r
p
o
i
n
t
s
,
w
h
i
c
h
i
s
t
h
e
n
f
o
l
l
o
w
e
d
b
y
a
c
o
n
t
r
a
s
t
e
n
h
a
n
c
e
m
e
n
t
o
p
e
r
a
t
o
r
t
h
a
t
a
m
p
l
if
i
e
s
h
i
g
h
-
i
n
t
e
n
s
i
t
y
r
e
g
i
o
n
s
(
e.
g
.
,
l
e
s
i
o
n
s
)
w
h
i
le
s
u
p
p
r
e
s
s
i
n
g
b
a
c
k
g
r
o
u
n
d
n
o
i
s
e
.
T
h
i
s
p
r
e
p
r
o
c
es
s
i
n
g
s
t
e
p
s
i
g
n
i
f
i
c
a
n
t
l
y
i
m
p
r
o
v
e
s
t
h
e
q
u
a
l
i
t
y
o
f
i
n
p
u
t
i
m
a
g
es
a
n
d
e
n
h
a
n
c
e
s
t
h
e
s
e
g
m
e
n
t
a
ti
o
n
a
c
cu
r
a
c
y
o
f
t
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
.
T
h
i
s
p
a
p
e
r
m
a
k
e
s
t
h
e
f
o
l
l
o
wi
n
g
k
e
y
c
o
n
t
r
i
b
u
t
i
o
n
s
:
a.
I
n
tr
o
d
u
ce
a
f
u
zz
y
lo
g
ic
-
b
ased
im
ag
e
co
n
tr
ast
en
h
an
ce
m
en
t
tech
n
iq
u
e
d
esig
n
e
d
to
h
ig
h
lig
h
t
lo
w
-
co
n
tr
ast
C
OVI
D
-
1
9
in
f
ec
tio
n
r
e
g
io
n
s
i
n
C
T
im
ag
es o
f
ch
est.
b.
Dev
elo
p
in
g
a
n
en
h
a
n
ce
d
Ma
s
k
R
-
C
NN
s
eg
m
en
tatio
n
f
r
am
ewo
r
k
th
at
ef
f
ec
tiv
ely
lev
er
a
g
es
th
e
co
n
tr
ast
-
en
h
an
ce
d
in
p
u
t im
a
g
es,
an
d
al
lo
ws f
o
r
p
r
ec
is
e
p
ix
el
-
lev
el
in
f
ec
tio
n
d
etec
tio
n
.
c.
E
x
ten
s
iv
e
ex
p
er
im
e
n
tal
v
alid
atio
n
o
n
a
p
u
b
lic
C
OVI
D
-
1
9
C
T
d
ataset,
s
h
o
win
g
im
p
r
o
v
ed
p
er
f
o
r
m
an
ce
o
v
er
b
aselin
e
a
n
d
s
tate
-
of
-
t
h
e
-
ar
t m
o
d
els in
ter
m
s
o
f
s
tan
d
ar
d
p
er
f
o
r
m
an
ce
m
etr
ics.
T
h
e
r
em
ai
n
in
g
o
f
th
e
p
ap
e
r
is
o
r
g
an
ized
as
f
o
llo
ws:
s
ec
tio
n
2
p
r
o
v
id
es
an
o
v
er
v
iew
o
f
r
el
ated
wo
r
k
in
C
OVI
D
-
1
9
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
f
r
o
m
C
T
s
ca
n
s
,
a
lo
n
g
with
a
cr
itical
an
aly
s
is
o
f
r
ec
en
t
ap
p
r
o
ac
h
es.
Sectio
n
3
d
etails
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
,
in
clu
d
i
n
g
t
h
e
f
u
zz
y
e
n
h
an
ce
m
en
t
p
r
o
ce
s
s
an
d
th
e
Ma
s
k
R
-
C
NN
ar
ch
itectu
r
e.
T
h
e
e
x
p
er
im
e
n
ta
l
s
etu
p
,
p
er
f
o
r
m
an
ce
m
etr
ics,
an
d
q
u
a
n
titativ
e
m
o
d
el
ev
al
u
atio
n
o
n
a
p
u
b
lic
d
ataset
ar
e
p
r
o
v
i
d
ed
in
s
ec
tio
n
4
.
T
h
e
c
o
n
clu
s
io
n
an
d
p
o
ten
ti
al
f
u
tu
r
e
d
ir
ec
tio
n
s
ar
e
d
is
cu
s
s
ed
in
s
ec
tio
n
5.
2.
RE
L
AT
E
D
WO
RK
W
e
p
r
esen
t
a
s
u
r
v
ey
o
f
r
esear
ch
wo
r
k
f
o
c
u
s
ed
o
n
d
ee
p
lear
n
in
g
-
b
ased
d
etec
tio
n
.
Ou
r
em
p
h
asis
is
o
n
ap
p
r
o
ac
h
es
th
at
h
a
v
e
s
ig
n
if
ic
an
tly
co
n
t
r
ib
u
ted
to
d
iag
n
o
s
ti
c
task
s
f
r
o
m
C
T
an
d
X
-
r
ay
i
m
ag
in
g
m
o
d
alities
,
an
d
we
h
ig
h
lig
h
t
th
eir
m
eth
o
d
o
lo
g
ies,
f
i
n
d
in
g
s
,
an
d
th
e
r
em
ain
in
g
c
h
allen
g
es
o
u
r
wo
r
k
aim
s
to
a
d
d
r
ess
.
J
in
et
a
l
.
[
1
6
]
d
ev
elo
p
e
d
a
co
m
p
r
eh
e
n
s
iv
e
d
ee
p
lear
n
in
g
s
y
s
tem
th
at
p
er
f
o
r
m
ed
lu
n
g
s
eg
m
en
tatio
n
an
d
lo
ca
lized
in
f
ec
tio
u
s
s
lices
f
o
r
C
OVI
D
-
1
9
d
iag
n
o
s
is
.
Desp
ite,
th
eir
wo
r
k
h
as
ac
h
iev
ed
en
c
o
u
r
ag
in
g
r
esu
lts
,
it
r
elied
o
n
m
an
u
ally
c
o
n
s
tr
u
cte
d
p
ip
elin
es
f
o
r
s
eg
m
en
tatio
n
,
wh
ich
lim
ited
en
d
-
to
-
e
n
d
au
to
m
atio
n
.
I
n
s
p
ir
ed
b
y
th
e
VGG
ar
ch
itectu
r
e,
Hu
et
a
l
.
[
1
7
]
in
tr
o
d
u
ce
d
a
wea
k
ly
s
u
p
er
v
is
ed
m
u
ltis
ca
le
d
ee
p
lear
n
in
g
f
r
a
m
ewo
r
k
th
at
ef
f
ec
tiv
ely
ass
im
ilates
m
u
lti
-
s
ca
le
lesi
o
n
f
ea
tu
r
es.
Ho
wev
er
,
th
eir
ap
p
r
o
ac
h
d
id
n
o
t
le
v
er
ag
e
p
i
x
el
-
lev
el
s
eg
m
en
tatio
n
ca
p
ab
ilit
ies an
d
ex
h
ib
ited
lim
itatio
n
s
in
h
a
n
d
li
n
g
h
ig
h
in
ter
-
class
s
im
ilar
ities
.
Po
ls
in
elli
et
a
l
.
[
1
8
]
p
r
esen
te
d
a
Sq
u
ee
ze
Net
C
NN
ar
ch
itectu
r
e,
in
o
r
d
er
to
p
r
o
v
id
e
r
a
p
id
in
f
er
en
ce
f
o
r
C
OVI
D
-
1
9
d
iag
n
o
s
is
f
r
o
m
C
T
s
ca
n
s
.
Ho
wev
er
,
th
eir
a
p
p
r
o
ac
h
was
ef
f
icien
t
in
ter
m
s
o
f
p
r
o
ce
s
s
in
g
tim
e,
b
u
t
b
ec
au
s
e
o
f
th
e
s
im
p
lifie
d
ar
ch
itectu
r
e
o
f
th
e
m
o
d
el,
th
e
lev
el
o
f
ac
cu
r
ac
y
s
u
f
f
e
r
ed
.
On
th
e
o
th
er
h
an
d
,
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:
2088
-
8
7
0
8
E
fficien
t m
a
s
k
r
eg
io
n
-
b
a
s
ed
c
o
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
-
b
a
s
ed
a
r
ch
itectu
r
e
…
(
N
a
d
er Ma
h
mo
u
d
)
4753
B
is
wa
s
et
a
l
.
[
1
9
]
in
v
esti
g
ated
an
en
s
em
b
le
s
tr
ateg
y
co
m
b
in
in
g
s
u
cc
ess
f
u
l
n
etwo
r
k
ar
c
h
itectu
r
es
in
clu
d
in
g
VGG
-
1
6
,
R
esNet5
0
[
2
0
]
,
an
d
Xce
p
tio
n
[
2
1
]
.
T
h
ei
r
m
o
d
el
ac
h
iev
ed
r
em
ar
k
a
b
le
g
en
er
aliza
t
io
n
ca
p
a
b
ilit
ies
b
y
u
tili
zin
g
tr
an
s
f
er
lear
n
in
g
,
h
o
wev
er
,
th
e
e
n
s
em
b
le’
s
co
m
p
u
tatio
n
al
d
em
an
d
s
wer
e
s
ig
n
if
i
ca
n
tly
h
ig
h
e
r
,
th
u
s
lim
itin
g
its
p
r
ac
ticality
in
r
ea
l
-
wo
r
ld
clin
ical
p
r
o
ce
d
u
r
es.
I
n
co
n
tr
ast,
Z
h
a
o
et
a
l
.
[
2
2
]
h
av
e
r
esear
ch
ed
a
d
if
f
er
en
t
r
o
u
te
t
h
r
o
u
g
h
m
o
d
if
y
in
g
th
e
R
esNet
-
v
2
ar
ch
itect
u
r
e.
I
n
o
r
d
er
t
o
en
h
a
n
ce
tr
ai
n
in
g
r
eliab
ilit
y
a
n
d
o
v
er
all
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
th
ey
im
p
lem
en
te
d
weig
h
t
s
tan
d
ar
d
izatio
n
an
d
r
e
p
lace
d
b
atc
h
n
o
r
m
aliza
tio
n
wit
h
g
r
o
u
p
n
o
r
m
aliza
tio
n
.
Alth
o
u
g
h
th
is
e
n
h
an
ce
d
t
h
e
r
o
b
u
s
tn
ess
o
f
th
eir
p
r
o
p
o
s
ed
s
y
s
tem
,
it
lack
e
d
ex
p
licit
r
eg
io
n
-
le
v
el
s
eg
m
en
tatio
n
,
wh
ich
is
ess
en
tial f
o
r
m
ed
ical
d
ia
g
n
o
s
tic
in
ter
p
r
etab
ilit
y
.
Sig
n
if
ican
t
ad
v
an
ce
m
en
ts
h
a
v
e
b
ee
n
ac
h
ie
v
ed
in
o
b
ject
d
etec
tio
n
f
r
am
ewo
r
k
s
in
a
d
d
itio
n
to
class
if
icatio
n
f
o
cu
s
ed
r
esear
c
h
,
b
e
g
in
n
in
g
with
R
-
C
NN
[
2
3
]
an
d
p
r
o
g
r
ess
in
g
t
h
r
o
u
g
h
Fas
t
R
-
C
N
N
[
2
4
]
a
n
d
Fas
ter
R
-
C
NN
[
2
5
]
.
T
h
ese
m
o
d
els
in
tr
o
d
u
ce
d
r
e
g
io
n
p
r
o
p
o
s
al
m
ec
h
an
is
m
s
an
d
b
o
u
n
d
in
g
b
o
x
p
r
ed
ictio
n
,
wh
ich
s
ig
n
if
ican
tly
im
p
r
o
v
ed
d
etec
tio
n
ac
cu
r
ac
y
.
R
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
s
(
R
PNs
)
wer
e
in
co
r
p
o
r
ated
in
to
Fas
ter
R
-
C
NN,
wh
ich
r
esu
lted
in
s
ig
n
if
ican
t
s
p
ee
d
an
d
ac
c
u
r
ac
y
im
p
r
o
v
e
m
en
ts
.
Du
e
to
r
o
b
u
s
tn
ess
o
f
Fas
ter
R
-
C
NN,
it
s
er
v
ed
as
th
e
f
o
u
n
d
atio
n
f
o
r
n
u
m
er
o
u
s
s
u
b
s
eq
u
e
n
t
en
h
an
ce
m
en
ts
[
2
6
]
a
n
d
r
e
m
ain
s
to
p
p
er
f
o
r
m
e
r
in
s
ev
er
al
b
en
ch
m
ar
k
s
.
R
ec
en
tly
,
Ma
s
k
R
-
C
N
N
[
9
]
h
as
b
ee
n
p
r
esen
ted
,
wh
ich
ex
te
n
d
in
g
th
e
wo
r
k
o
f
Fas
ter
R
-
C
NN
f
o
r
b
o
th
o
b
ject
d
etec
tio
n
an
d
in
s
tan
t seg
m
en
tatio
n
ta
s
k
s
in
co
m
p
u
ter
v
is
io
n
.
T
h
e
m
ain
co
n
tr
ib
u
tio
n
o
f
Ma
s
k
R
-
C
NN
is
in
its
ca
p
ab
ilit
y
to
p
er
f
o
r
m
p
ix
el
-
wis
e
s
eg
m
en
tatio
n
in
a
d
d
itio
n
to
o
b
j
ec
t
d
etec
tio
n
.
T
h
e
ad
d
itio
n
o
f
a
n
ex
tr
a
“m
ask
h
ea
d
”
b
r
an
c
h
im
p
r
o
v
es
p
r
ec
is
e
s
eg
m
en
tatio
n
m
ask
s
f
o
r
in
d
i
v
id
u
al
o
b
jects.
T
h
e
m
o
d
el
also
in
tr
o
d
u
ce
s
two
s
ig
n
if
ican
t
ar
ch
itectu
r
al
wh
ic
h
ar
e
alig
n
m
en
ts
o
f
r
eg
io
n
s
o
f
in
ter
est
(
R
o
I
)
co
r
r
ec
ted
s
p
atial
m
is
alig
n
m
en
t
an
d
f
ea
tu
r
e
p
y
r
a
m
id
n
etwo
r
k
s
.
R
o
I
alig
n
m
en
t
tack
les
lim
it
atio
n
s
o
f
tr
ad
itio
n
al
R
o
I
p
o
o
lin
g
th
r
o
u
g
h
b
ilin
ea
r
in
ter
p
o
latio
n
d
u
r
in
g
t
h
e
p
o
o
li
n
g
p
h
ase.
T
h
is
lead
s
to
im
p
r
o
v
ed
s
eg
m
en
tatio
n
ac
cu
r
ac
y
,
esp
ec
ially
wh
e
n
d
ea
lin
g
with
s
m
all
o
b
ject
in
s
tan
ce
s
.
Desp
ite
th
ese
ad
v
an
ce
m
en
ts
,
m
an
y
ex
is
tin
g
r
esear
ch
wo
r
k
s
eith
er
d
o
n
o
t
s
u
p
p
o
r
t
f
in
e
-
g
r
ain
e
d
s
eg
m
en
tatio
n
o
r
s
tr
u
g
g
le
wit
h
ef
f
ec
tiv
e
C
T
im
ag
e
p
r
e
p
r
o
ce
s
s
in
g
f
o
r
h
an
d
lin
g
lo
w
c
o
n
tr
ast
an
d
im
ag
e
f
u
zz
in
ess
.
Ad
d
itio
n
ally
,
a
n
u
m
b
er
o
f
r
esear
ch
wo
r
k
s
r
ely
o
n
r
elativ
ely
s
m
all
d
atasets
,
wh
ich
lim
its
th
e
g
en
er
aliza
b
ilit
y
o
f
th
eir
r
esu
lts
.
C
o
n
s
eq
u
en
tly
,
th
is
p
ap
er
in
tr
o
d
u
ce
s
an
ef
f
icien
t
M
ask
R
-
C
N
N
–
b
ased
ar
ch
itectu
r
e
tailo
r
ed
f
o
r
C
OVI
D
-
1
9
d
etec
tio
n
in
ch
est
C
T
s
ca
n
s
th
at
ef
f
ec
tiv
ely
lev
e
r
ag
es
th
e
co
n
tr
ast
-
en
h
an
ce
d
i
n
p
u
t
im
a
g
es.
A
k
ey
n
o
v
elty
o
f
th
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
is
th
e
in
tr
o
d
u
ctio
n
o
f
a
f
u
zz
y
c
o
lo
r
en
h
an
ce
m
e
n
t
p
r
e
p
r
o
ce
s
s
in
g
te
ch
n
iq
u
e
th
at
s
ig
n
if
ican
tly
im
p
r
o
v
es
im
ag
e
clar
ity
an
d
co
n
tr
a
s
t.
W
e
v
alid
ate
o
u
r
m
eth
o
d
o
n
a
lar
g
e,
p
u
b
licly
a
v
ailab
le
d
ataset
an
d
b
en
ch
m
ar
k
its
p
er
f
o
r
m
an
ce
ag
ain
s
t
s
ev
er
al
s
tate
-
of
-
th
e
-
ar
t
m
o
d
els,
d
em
o
n
s
tr
atin
g
clea
r
a
d
v
an
tag
es a
cr
o
s
s
s
tan
d
ar
d
ev
a
lu
atio
n
m
etr
ics.
3.
P
RO
P
O
SE
D
M
O
D
E
L
A
RC
H
I
T
E
CT
U
RE
Fig
u
r
e
1
d
e
p
icts
th
e
m
ain
s
tep
s
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
a
r
ch
it
ec
tu
r
e.
Fig
u
r
e
1
.
Pro
p
o
s
ed
m
o
d
el
ar
c
h
itectu
r
e
wo
r
k
f
lo
w
3
.
1
.
I
ma
g
e
prepro
ce
s
s
ing
us
ing
f
uzzy
co
lo
r
enha
ncem
ent
T
o
ad
d
r
ess
th
e
ch
allen
g
es
o
f
C
T
im
ag
es,
s
u
ch
as
lo
w
co
n
tr
ast,
p
ar
ticu
lar
ly
in
ea
r
ly
-
s
tag
e
C
OVI
D
-
19
in
f
e
ct
io
n
s
,
w
e
in
co
r
p
o
r
a
ted
a
f
u
zz
y
lo
g
ic
-
b
a
s
ed
im
ag
e
en
h
a
n
ce
m
en
t
t
ec
h
n
iq
u
e
[
2
7
]
as
a
p
r
ep
r
o
c
e
s
s
in
g
s
t
ep
,
d
eta
i
led
in
A
lg
o
r
i
th
m
1
.
T
h
i
s
t
ec
h
n
iq
u
e
a
llo
w
s
f
o
r
s
ig
n
i
f
ic
an
t
im
p
r
o
v
em
en
t
o
f
b
r
i
g
h
tn
e
s
s
an
d
co
n
tr
a
s
t
ac
r
o
s
s
C
T
im
ag
e
s
,
th
er
ef
o
r
e
en
ab
lin
g
m
o
r
e
a
cc
u
r
at
e
s
eg
m
en
ta
t
io
n
in
th
e
s
u
b
s
eq
u
en
t
M
ask
R
-
C
NN
s
tag
e.
T
h
e
f
u
z
zy
en
h
an
c
em
en
t
p
r
o
c
es
s
b
e
g
in
s
b
y
d
e
co
m
p
o
s
in
g
t
h
e
im
ag
e
i
n
to
o
v
er
l
ap
p
in
g
f
u
zz
y
r
eg
io
n
s
,
ea
ch
r
ep
r
e
s
e
n
t
in
g
a
d
is
t
in
c
t
o
b
j
ec
t
th
a
t
ca
p
tu
r
es
v
ar
y
in
g
le
v
e
ls
o
f
u
n
ce
r
ta
in
ty
in
b
r
i
g
h
tn
e
s
s
an
d
co
n
tr
as
t.
I
n
s
te
ad
o
f
as
s
ig
n
in
g
h
ar
d
b
o
u
n
d
ar
ie
s
,
ea
ch
p
ix
e
l
i
s
g
iv
en
a
m
em
b
er
s
h
ip
v
alu
e
f
o
r
e
ac
h
r
eg
io
n
,
b
as
ed
o
n
h
o
w
f
ar
awa
y
i
t
is
f
r
o
m
th
e
ce
n
te
r
o
f
t
h
e
r
eg
io
n
.
T
h
e
s
e
m
e
m
b
er
s
h
ip
v
alu
e
s
in
f
lu
en
c
e
th
e
p
ix
e
l’
s
c
o
n
tr
i
b
u
t
io
n
d
u
r
in
g
v
ar
i
an
c
e
ca
lcu
la
tio
n
s
,
al
lo
win
g
th
e
a
lg
o
r
ith
m
to
b
e
t
ter
m
o
d
el
u
n
c
er
ta
in
ty
an
d
r
ev
ea
l
s
u
b
t
le
s
tr
u
ct
u
r
a
l
d
e
ta
il
s
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
5
1
-
4
7
6
1
4754
th
a
t
a
r
e
cr
it
ic
al
f
o
r
ea
r
ly
d
i
ag
n
o
s
i
s
.
A
lg
o
r
ith
m
1
d
e
ta
il
s
th
e
en
h
an
c
em
en
t
p
r
o
c
e
s
s
,
wh
i
ch
p
r
o
ce
ed
s
as
f
o
llo
w
s
:
a.
F
o
r
e
a
c
h
p
i
x
el
,
t
h
e
m
e
a
n
a
n
d
v
a
r
i
a
n
c
e
wi
t
h
i
n
i
ts
f
u
z
z
y
n
e
i
g
h
b
o
r
h
o
o
d
a
r
e
c
o
m
p
u
t
e
d
,
w
e
i
g
h
t
e
d
b
y
t
h
e
m
e
m
b
e
r
s
h
i
p
d
e
g
r
e
e
s
.
b.
T
h
e
s
e
l
o
c
al
i
z
e
d
s
t
a
ti
s
t
i
cs
a
r
e
u
s
e
d
t
o
a
d
j
u
s
t
p
i
x
e
l
i
n
te
n
s
it
y
,
em
p
h
a
s
i
z
i
n
g
r
e
g
i
o
n
s
o
f
i
n
te
r
e
s
t
a
n
d
s
u
p
p
r
e
s
s
i
n
g
b
a
c
k
g
r
o
u
n
d
n
o
i
s
e
.
c.
T
h
e
e
n
h
a
n
c
e
d
i
m
a
g
e
is
r
e
c
o
n
s
t
r
u
c
t
e
d
b
y
a
g
g
r
e
g
a
t
i
n
g
t
h
e
r
e
s
u
l
ts
f
r
o
m
a
l
l
f
u
z
z
y
wi
n
d
o
w
s
,
p
r
o
d
u
c
i
n
g
a
g
l
o
b
a
l
l
y
c
o
n
s
is
t
e
n
t
b
u
t
l
o
c
a
ll
y
a
d
a
p
t
i
v
e
c
o
n
t
r
a
s
t
e
n
h
a
n
c
e
m
e
n
t
.
Alg
o
r
ith
m
1
.
Fu
zz
y
co
l
o
r
-
b
ase
d
im
ag
e
en
h
an
ce
m
e
n
t
Input:
RGB image
I
, Window size
w
, Fuzzifier parameters
α
,
β
Output
: Enhanced image I
enhanced
1.
for
each color channel
c
∈
{
R
,
G
,
B
},
do
:
2.
I
c
← extract the channel
c
from
I
.
3.
I
c
enhanced
← zero matrix of same size as
I
c
.
4.
for
each
pixel
p
in
I
c
do
:
5.
W
p
← extract
w × w
window centered at
p
.
6.
Initialize
total_weight
← 0,
weighted_sum
← 0
7.
for
each
pixel
q
∈
W
p
,
do
:
8.
d
← EuclideanDistance(
p,q
).
9.
μ
←
exp
(
-
α · d
β
)
{Membership degree}
10.
total_weight
←
total_weight + μ
.
11.
weighted_sum
←
weighted_sum +
μ · I
c
(q)
.
12.
End for
13.
mean
←
ℎ
_
_
ℎ
14.
weighted_variance
← 0
15.
for
each
pixel
q
∈
W
p
,
do
:
16.
d
← EuclideanDistance(
p,q
).
17.
μ ← exp
(
-
α · d
β
).
18.
weighted_variance
← weighted_variance
+
μ
· (
I
c
(
q
)
-
mean
)
2
.
19.
End for
20.
variance
=
ℎ
_
_
ℎ
21.
I
c
enhanced
(
p
) = Enhance(
I
c
(
p
),
mean
,
variance
)
22.
End for
23.
Store
I
c
enhanced
in output channels
24.
End for
25.
I
enhanced
= Merge(
I
R
enhanced
,
I
G
enhanced
,
I
B
enhanced
)
Return
I
enhanced
I
n
th
is
p
ap
er
,
th
e
alg
o
r
ith
m
was
ap
p
lied
s
ep
ar
ately
to
ea
ch
o
f
th
e
th
r
ee
co
lo
r
ch
an
n
els
(
R
GB
)
o
f
th
e
C
T
im
ag
e.
T
h
o
u
g
h
C
T
im
ag
e
s
ar
e
o
f
ten
g
r
e
y
s
ca
le,
s
o
m
e
d
atasets
s
to
r
e
th
em
in
th
r
ee
-
ch
an
n
el
f
o
r
m
at;
th
u
s
,
o
u
r
m
eth
o
d
h
a
n
d
les
ea
ch
ch
a
n
n
el
in
p
ar
allel
a
n
d
m
er
g
es
t
h
e
o
u
tp
u
ts
to
r
ec
o
n
s
tr
u
ct
th
e
f
in
al
im
ag
e.
I
n
co
n
t
r
ast
to
co
n
v
e
n
tio
n
al
tech
n
iq
u
es
f
o
r
en
h
an
ci
n
g
im
a
g
e
co
n
t
r
ast,
in
clu
d
in
g
h
is
to
g
r
am
e
q
u
aliza
ti
o
n
an
d
its
ad
ap
tiv
e
v
ar
ian
t,
C
L
AHE
,
f
u
zz
y
en
h
a
n
ce
m
en
t
o
f
f
e
r
s
two
m
ain
ad
v
a
n
tag
es.
Firstl
y
,
it
ef
f
icien
tly
m
o
d
els
u
n
ce
r
tain
t
y
,
wh
ich
is
ess
en
tial
f
o
r
m
ed
ical
im
ag
es,
wh
er
e
p
ath
o
lo
g
ical
ch
ar
ac
ter
is
tics
ar
e
f
r
eq
u
en
tly
s
u
b
tle.
Seco
n
d
ly
,
it
ad
ju
s
ts
lo
ca
lly
,
en
h
an
cin
g
v
is
i
b
ilit
y
with
o
u
t
in
tr
o
d
u
cin
g
an
y
ar
tifa
cts
o
r
n
o
is
e.
Fig
u
r
e
2
(
a
)
s
h
o
win
g
im
p
r
o
v
ed
co
n
tr
ast
in
l
u
n
g
r
eg
i
o
n
s
p
o
ten
tially
af
f
ec
ted
b
y
in
f
ec
tio
n
.
Fig
u
r
e
2
(
b
)
illu
s
tr
ates
th
e
ef
f
ec
t
o
f
f
u
zz
y
co
lo
r
en
h
an
ce
m
e
n
t o
n
a
s
am
p
le
C
T
s
ca
n
,
(
a)
(
b
)
Fig
u
r
e
2
.
Fu
zz
y
C
T
im
ag
e
en
h
an
ce
m
en
t (
a)
o
r
ig
in
al
C
T
im
ag
e
an
d
(
b
)
f
u
zz
y
c
o
lo
r
e
d
en
h
a
n
ce
d
im
ag
e
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:
2088
-
8
7
0
8
E
fficien
t m
a
s
k
r
eg
io
n
-
b
a
s
ed
c
o
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
-
b
a
s
ed
a
r
ch
itectu
r
e
…
(
N
a
d
er Ma
h
mo
u
d
)
4755
3
.
2
.
M
a
s
k
R
-
CNN
-
ba
s
ed
s
eg
m
ent
a
t
io
n a
nd
cla
s
s
if
ica
t
io
n
I
n
th
e
s
ec
o
n
d
s
tep
o
f
o
u
r
p
ip
e
lin
e,
we
ap
p
ly
a
Ma
s
k
R
-
C
NN
f
r
am
ewo
r
k
o
u
tlin
ed
i
n
Alg
o
r
ith
m
2
to
d
etec
t
an
d
s
eg
m
en
t
in
f
ec
ted
r
e
g
io
n
s
f
r
o
m
p
r
ep
r
o
ce
s
s
ed
C
T
im
ag
es,
as
s
h
o
wn
i
n
Fig
u
r
e
1
.
Ma
s
k
R
-
C
NN
ad
d
s
o
n
to
p
o
f
Fas
ter
R
-
C
NN
a
p
ar
allel
b
r
an
ch
f
o
r
p
r
ed
ict
in
g
p
ix
el
-
wis
e
s
eg
m
en
tatio
n
m
ask
s
,
allo
win
g
s
im
u
ltan
eo
u
s
o
b
ject
d
etec
tio
n
an
d
in
s
tan
ce
s
eg
m
en
tatio
n
.
T
h
e
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
th
r
ee
co
r
e
co
m
p
o
n
en
ts
,
as d
escr
ib
ed
in
Alg
o
r
it
h
m
2
:
a.
B
ac
k
b
o
n
e
f
ea
tu
r
e
ex
tr
ac
to
r
(
R
esNet5
0
+
FP
N)
:
W
e
u
til
ize
a
R
esNet5
0
m
o
d
el
as
th
e
b
ac
k
b
o
n
e
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
d
u
e
to
its
a
b
ilit
y
to
b
alan
ce
b
etwe
en
d
ep
th
an
d
c
o
m
p
u
tatio
n
al
ef
f
icien
c
y
.
R
esNet5
0
is
co
m
p
o
s
ed
o
f
4
8
c
o
n
v
o
lu
tio
n
al
lay
er
s
,
alo
n
g
with
a
m
ax
-
p
o
o
lin
g
lay
er
an
d
a
g
lo
b
al
av
e
r
ag
e
p
o
o
lin
g
lay
er
.
Aim
in
g
to
en
h
an
ce
f
ea
tu
r
e
r
ep
r
esen
tatio
n
ac
r
o
s
s
m
u
ltip
le
s
ca
les,
we
in
co
r
p
o
r
ate
a
f
ea
tu
r
e
p
y
r
am
id
n
etwo
r
k
(
FP
N)
[
2
5
]
o
n
t
o
p
o
f
R
esNet5
0
.
FP
N
co
n
s
tr
u
cts
a
f
ea
tu
r
e
p
y
r
am
id
th
at
ca
p
tu
r
es
b
o
th
l
o
w
-
lev
el
an
d
h
ig
h
-
lev
el
f
ea
tu
r
es,
wh
ic
h
en
h
an
ce
s
th
e
ab
ilit
y
to
d
etec
t
lesi
o
n
s
in
C
T
im
ag
es
o
f
d
i
f
f
er
en
t
s
izes
an
d
s
ca
les.
C
O
VI
D
-
1
9
lesi
o
n
s
,
e.
g
.
g
r
o
u
n
d
-
g
lass
o
p
ac
ities
,
ca
n
ap
p
ea
r
at
d
if
f
er
e
n
t
s
izes
an
d
in
ten
s
ities
,
s
o
m
etim
es
s
ca
tter
ed
in
th
e
l
u
n
g
s
.
B
y
co
m
b
in
in
g
h
ig
h
an
d
lo
w
r
eso
lu
tio
n
f
ea
tu
r
es,
FP
N
aid
s
in
th
e
d
etec
tio
n
o
f
b
o
th
la
r
g
er
a
b
n
o
r
m
alities
an
d
tin
y
ea
r
ly
-
s
tag
e
l
esio
n
s
.
T
h
is
is
ess
en
tial
f
o
r
r
eliab
le
d
etec
tio
n
ac
r
o
s
s
d
is
ea
s
e
p
r
o
g
r
ess
io
n
.
b.
R
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
(
R
PN)
:
T
h
e
R
PN
g
en
er
ate
r
eg
io
n
p
r
o
p
o
s
als
th
r
o
u
g
h
ap
p
ly
i
n
g
a
s
m
all
n
etwo
r
k
o
v
er
t
h
e
f
ea
tu
r
e
m
a
p
s
g
en
e
r
ated
b
y
th
e
FP
N.
T
h
e
co
o
r
d
i
n
ates
o
f
ea
ch
an
c
h
o
r
b
o
x
ar
e
f
in
e
-
tu
n
ed
an
d
g
iv
en
an
o
b
jectn
ess
s
co
r
e.
B
y
r
ed
u
ci
n
g
th
e
n
u
m
b
e
r
o
f
p
o
s
s
ib
le
ar
e
as
wh
er
e
in
f
ec
tio
n
m
i
g
h
t
ex
is
t,
th
e
R
PN
h
elp
s
to
im
p
r
o
v
e
th
e
ef
f
icien
cy
o
f
th
e
s
u
b
s
eq
u
en
t c
lass
if
icatio
n
an
d
s
eg
m
en
tatio
n
.
c.
R
eg
io
n
o
f
in
te
r
est
alig
n
m
en
t
(
R
o
I
Alig
n
)
:
T
o
ac
c
u
r
ately
m
a
p
th
e
r
e
g
io
n
p
r
o
p
o
s
als
o
n
to
t
h
e
f
ea
tu
r
e
m
ap
,
we
em
p
lo
y
R
o
I
Alig
n
[
2
8
]
.
T
h
is
m
eth
o
d
o
u
tp
er
f
o
r
m
s
co
n
v
en
tio
n
al
R
o
I
p
o
o
lin
g
b
y
em
p
lo
y
in
g
b
ilin
ea
r
in
ter
p
o
latio
n
in
s
tead
o
f
co
a
r
s
e
q
u
an
tizatio
n
.
T
h
is
en
s
u
r
e
s
s
p
atial
alig
n
m
en
t
is
p
r
eser
v
ed
,
wh
ich
is
esp
ec
ially
cr
itical
f
o
r
m
ed
ic
al
im
ag
in
g
wh
er
e
p
ix
el
-
lev
el
ac
cu
r
ac
y
is
cr
u
cial.
T
h
e
R
o
I
Alig
n
p
r
o
ce
s
s
in
v
o
lv
es
th
e
s
tep
s
d
etailed
Alg
o
r
ith
m
2
.
I
n
p
u
ts
ar
e
f
ea
tu
r
e
m
ap
s
an
d
r
eg
i
o
n
p
r
o
p
o
s
als
g
en
er
ated
b
y
FP
N
an
d
R
PN,
r
esp
ec
tiv
ely
.
R
eg
io
n
p
r
o
p
o
s
als ar
e
s
u
b
d
iv
id
ed
in
t
o
eq
u
al
-
s
ized
g
r
i
d
s
to
ex
tr
ac
t f
ea
tu
r
es f
r
o
m
th
e
m
atch
in
g
r
eg
io
n
s
in
th
e
i
n
p
u
t
f
ea
tu
r
e
m
ap
.
T
h
e
alig
n
ed
f
ea
tu
r
es
f
r
o
m
th
ese
g
r
i
d
s
r
ep
r
esen
t
th
e
ch
ar
ac
ter
is
tics
o
f
ea
ch
p
r
o
p
o
s
al.
Af
ter
ac
q
u
ir
in
g
th
e
s
p
atially
a
lig
n
ed
f
ea
tu
r
es
,
a
f
u
ll
y
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
[
9
]
is
u
s
ed
to
g
en
er
ate
b
in
ar
y
m
ask
s
f
o
r
ea
c
h
p
r
o
p
o
s
ed
r
eg
io
n
.
I
n
p
ar
allel,
a
class
if
icatio
n
b
r
an
c
h
class
if
ies
ea
ch
r
eg
io
n
in
to
o
n
e
o
f
th
e
p
r
ed
ef
in
e
d
ca
teg
o
r
ies
(
e.
g
.
,
in
f
ec
ted
v
s
.
n
o
n
-
in
f
ec
ted
)
.
T
h
e
class
if
icatio
n
p
ip
elin
e
b
eg
in
s
with
co
n
v
o
l
u
tio
n
al
lay
er
s
f
o
ll
o
wed
b
y
f
u
lly
co
n
n
ec
ted
lay
er
s
w
h
ich
en
c
o
d
e
s
p
atial
an
d
co
n
t
ex
tu
al
in
f
o
r
m
atio
n
f
r
o
m
th
e
R
o
I
-
alig
n
ed
m
ap
s
.
C
o
m
b
in
in
g
th
ese
elem
en
ts
allo
ws
th
e
p
r
o
p
o
s
ed
m
o
d
el
to
ac
cu
r
ately
s
eg
m
en
t
C
OVI
D
-
1
9
lesi
o
n
s
f
r
o
m
C
T
i
m
ag
es
in
an
en
d
-
to
-
en
d
m
an
n
er
,
class
if
y
th
em
,
an
d
en
ab
l
e
lo
ca
lizatio
n
.
T
h
is
ca
p
ab
ilit
y
ad
d
r
ess
es
th
e
lim
ita
tio
n
o
f
ea
r
lier
C
NN
-
b
ased
m
o
d
els,
wh
ich
lack
ed
s
p
atial
lo
ca
lizatio
n
an
d
r
elied
o
n
wh
o
le
-
im
a
g
e
class
if
icatio
n
,
th
er
ef
o
r
e
r
ed
u
cin
g
clin
ical
in
t
er
p
r
etab
ilit
y
.
Alg
o
r
ith
m
2
.
C
OVI
D
-
1
9
lesi
o
n
d
etec
tio
n
u
s
in
g
m
ask
R
-
C
NN
Input:
Preprocessed CT image
I
Output
: Region
-
level classifications and segmentation masks
1.
Extract feature maps from
I
using ResNet50 backbone
2.
Construct multi
-
scale feature maps using Feature Pyramid Network (FPN)
3.
Generate region proposals
ℛ
using RPN on FPN features
for
each
feature map level
P
l
do
:
Slide a small 3
×
3 window over
P
l
for
each
window position:
a. Generate anchor boxes with multiple scales and
aspirations
b. Predict the objectness score for each anchor (foreground/background)
c. Regress bounding box offsets for anchor refinement
End for
End for
4.
For each
region proposal
r
∈
ℛ
do
a.
Apply RoIAlign on
r
to extract a fixed
-
size feature map
b.
Classify the region
r
using a CNN classifier
c.
Generate a binary segmentation mask using a parallel FCN mask branch
End for
5.
Aggregate classification scores and masks for final prediction
Return
Predicted region classes and corresponding segmentation masks
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
is
s
ec
tio
n
,
we
p
r
o
v
id
e
a
th
o
r
o
u
g
h
d
escr
ip
tio
n
o
f
th
e
ex
p
er
im
en
tal
s
etu
p
,
in
clu
d
in
g
d
e
tails
o
f
th
e
d
ataset
u
s
ed
,
th
e
em
p
l
o
y
ed
e
v
alu
atio
n
m
etr
ics,
a
n
d
th
e
m
e
th
o
d
o
lo
g
y
f
o
llo
wed
f
o
r
p
er
f
o
r
m
an
ce
ev
al
u
atio
n
.
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
5
1
-
4
7
6
1
4756
Fu
r
th
er
m
o
r
e
,
we
p
r
esen
t
a
r
ig
o
r
o
u
s
q
u
an
titativ
e
ev
alu
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
ar
ch
ite
ctu
r
e,
co
m
p
ar
in
g
its
r
esu
lts
ag
ain
s
t e
s
tab
lis
h
ed
b
en
ch
m
ar
k
m
o
d
els to
d
e
m
o
n
s
tr
at
e
its
ef
f
ec
tiv
en
ess
an
d
s
u
p
er
io
r
ity
.
4
.
1
.
E
x
perim
ent
a
l
s
et
up
a
nd
da
t
a
s
et
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
a
r
ch
itectu
r
e
was
ca
r
r
ied
o
u
t
an
d
ev
alu
ated
o
n
a
PC
with
an
I
n
tel
C
o
r
e
i5
7
th
g
en
er
atio
n
p
r
o
ce
s
s
o
r
,
8
GB
o
f
R
AM
,
a
GT
X
1
0
8
0
GPU
with
4
GB
o
f
R
AM
,
an
d
r
u
n
n
in
g
a
W
in
d
o
ws
o
p
er
atin
g
s
y
s
tem
.
W
e
u
tili
ze
d
th
e
p
u
b
licly
a
v
ailab
le
C
OVI
Dx
-
C
T
d
ataset
[
2
9
]
,
[
3
0
]
f
o
r
th
e
ev
alu
atio
n
,
wh
ic
h
co
n
s
is
ts
o
f
a
1
0
4
,
0
0
9
ch
est
C
T
s
lice
co
llected
f
r
o
m
1
,
4
8
9
p
atien
t
ca
s
es.
T
h
is
d
ataset
ag
g
r
eg
ates
c
h
est
C
T
ex
am
in
atio
n
s
f
r
o
m
m
u
ltip
le
h
o
s
p
ital
co
h
o
r
ts
ac
r
o
s
s
C
h
in
a
as
p
ar
t
o
f
th
e
C
h
in
a
C
o
n
s
o
r
tiu
m
o
f
C
h
est
C
T
I
m
ag
e
I
n
v
esti
g
atio
n
(
C
C
-
C
C
I
I
)
.
T
h
e
im
ag
in
g
d
ata
in
clu
d
e
ch
est
C
T
v
o
lu
m
es
r
ep
r
esen
tin
g
th
r
ee
ca
teg
o
r
ies:
C
OVI
D
-
1
9
p
n
eu
m
o
n
ia,
c
o
m
m
o
n
p
n
eu
m
o
n
ia
r
esu
ltin
g
f
r
o
m
n
o
n
-
C
OVI
D
-
1
9
in
f
e
ctio
n
s
,
an
d
n
o
r
m
al
(
u
n
in
f
ec
te
d
)
ca
s
es.
Fig
u
r
e
3
s
h
o
ws
im
ag
e
s
am
p
les
f
o
r
ea
ch
o
f
th
e
t
h
r
ee
d
a
taset
ca
teg
o
r
ies:
Fig
u
r
e
3
(
a)
s
h
o
ws
a
C
OVI
D
-
1
9
ca
s
e,
Fig
u
r
e
3
(
b
)
s
h
o
ws
a
n
o
n
-
C
OVI
D
-
1
9
p
n
e
u
m
o
n
ia
ca
s
e,
an
d
Fig
u
r
e
3
(
c)
s
h
o
ws
a
n
o
r
m
al
ca
s
e
f
r
o
m
a
h
ea
lth
y
in
d
iv
i
d
u
al.
T
h
e
C
OVI
Dx
-
C
T
d
ataset
is
ch
o
s
en
d
u
e
to
its
h
ig
h
q
u
ality
,
well
-
o
r
g
a
n
ized
s
tr
u
ctu
r
e,
an
d
r
ea
d
i
n
ess
f
o
r
d
ir
ec
t
u
s
e
in
b
en
c
h
m
ar
k
in
g
ev
alu
atio
n
.
I
ts
s
tan
d
ar
d
ized
f
o
r
m
at
an
d
co
m
p
r
eh
e
n
s
iv
e
co
v
er
ag
e
o
f
in
f
ec
tio
n
ty
p
es
m
ak
e
it
p
ar
t
icu
lar
ly
s
u
itab
le
f
o
r
c
o
n
s
is
ten
t
ev
alu
atio
n
an
d
co
m
p
ar
is
o
n
o
f
d
ee
p
lear
n
i
n
g
m
o
d
el
p
er
f
o
r
m
a
n
ce
.
(
a)
(
b
)
(
c)
Fig
u
r
e
3
.
Sam
p
les o
f
C
OVI
Dx
-
C
T
d
ataset
(
a)
C
OVI
D
-
1
9
an
d
p
n
eu
m
o
n
ia
ca
s
e
,
(
b
)
n
o
n
-
C
OVI
D
-
19
p
n
eu
m
o
n
ia
ca
s
e
,
an
d
(
c)
s
h
o
ws a
n
o
r
m
al
ca
s
e
f
r
o
m
a
h
ea
lth
y
in
d
iv
id
u
al
W
e
f
o
ll
o
w
e
d
t
h
e
d
is
t
r
i
b
u
t
i
o
n
p
r
o
p
o
s
e
d
i
n
[
2
9
]
,
t
h
u
s
,
w
e
a
d
o
p
t
e
d
a
d
a
t
a
s
p
l
it
s
t
r
a
t
e
g
y
t
h
at
d
i
v
i
d
e
s
t
h
e
C
O
V
I
D
x
-
C
T
d
a
t
as
e
t
i
n
t
o
6
0
%
t
r
ai
n
i
n
g
–
2
0
%
te
s
t
–
2
0
%
v
al
i
d
a
t
i
o
n
,
r
es
p
e
ct
i
v
e
l
y
.
T
o
a
v
o
i
d
d
a
t
a
l
e
a
k
a
g
e
a
n
d
m
a
i
n
t
a
i
n
t
h
e
i
n
te
g
r
i
t
y
o
f
t
h
e
e
v
a
l
u
a
t
i
o
n
,
c
a
r
e
w
as
t
a
k
e
n
t
o
e
n
s
u
r
e
t
h
a
t
C
T
s
c
a
n
s
f
r
o
m
a
s
i
n
g
l
e
p
a
t
ie
n
t
w
e
r
e
a
l
l
o
c
a
t
e
d
j
u
s
t
t
o
o
n
e
o
f
t
h
e
s
et
s
.
T
h
e
r
es
u
l
t
i
n
g
d
i
s
t
r
i
b
u
t
i
o
n
is
s
u
m
m
a
r
i
z
e
d
i
n
T
a
b
l
e
1
a
n
d
v
i
s
u
a
l
l
y
r
e
p
r
e
s
e
n
t
e
d
in
F
i
g
u
r
e
4
.
T
h
i
s
s
p
l
i
tt
i
n
g
a
p
p
r
o
a
c
h
c
o
n
s
i
s
t
e
n
t
l
y
p
r
o
d
u
c
e
s
t
h
e
m
o
s
t
c
o
n
s
is
t
e
n
t
a
n
d
a
c
c
u
r
a
t
e
r
e
s
u
l
ts
p
e
r
f
o
r
m
a
n
c
e
a
c
r
o
s
s
al
l
e
v
a
l
u
a
te
d
b
e
n
c
h
m
a
r
k
i
n
g
m
o
d
e
l
s
.
T
o
f
u
r
t
h
e
r
e
n
s
u
r
e
f
a
i
r
n
e
s
s
a
n
d
m
i
n
i
m
i
z
e
s
el
e
c
ti
o
n
b
i
a
s
,
t
h
e
d
a
t
a
s
p
l
i
t
w
a
s
r
a
n
d
o
m
i
z
e
d
f
o
r
e
a
c
h
e
x
p
e
r
i
m
e
n
t
a
l
r
u
n
w
h
i
l
e
m
a
i
n
t
a
i
n
i
n
g
t
h
e
o
v
e
r
a
l
l
d
i
s
t
r
i
b
u
t
i
o
n
p
r
o
p
o
r
t
i
o
n
s
.
T
h
i
s
s
t
r
a
t
e
g
y
c
o
n
t
r
i
b
u
t
e
d
t
o
a
r
o
b
u
s
t
a
n
d
u
n
b
i
a
s
e
d
e
v
al
u
a
t
i
o
n
o
f
t
h
e
p
r
o
p
o
s
ed
a
r
c
h
i
t
e
c
t
u
r
e
p
e
r
f
o
r
m
a
n
c
e
.
T
ab
le
1
.
C
OVI
Dx
-
C
T
d
ataset
d
is
tr
ib
u
tio
n
Ty
p
e
N
o
r
mal
P
n
e
u
mo
n
i
a
C
O
V
I
D
-
19
To
t
a
l
Tr
a
i
n
2
8
2
0
2
2
1
5
6
8
1
2
6
3
5
6
2
4
0
5
V
a
l
i
d
a
t
i
o
n
8
6
5
8
7
2
5
1
4
8
9
3
2
0
2
8
0
2
Te
st
8
6
5
8
7
2
5
1
4
8
9
3
2
0
8
0
2
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:
2088
-
8
7
0
8
E
fficien
t m
a
s
k
r
eg
io
n
-
b
a
s
ed
c
o
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
-
b
a
s
ed
a
r
ch
itectu
r
e
…
(
N
a
d
er Ma
h
mo
u
d
)
4757
Fig
u
r
e
4
.
Data
s
et
d
is
tr
ib
u
tio
n
4
.
2
.
E
v
a
lua
t
i
o
n
m
et
rics
T
o
ev
alu
ate
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
ar
c
h
i
tectu
r
e,
we
em
p
lo
y
s
tan
d
ar
d
ev
alu
atio
n
m
etr
ics,
in
clu
d
in
g
Pre
cisi
o
n
,
R
ec
all,
F1
-
s
co
r
e,
an
d
Acc
u
r
ac
y
:
a.
Pre
cisi
o
n
:
T
h
e
p
r
o
p
o
r
tio
n
o
f
tr
u
e
p
o
s
itiv
e
C
OVI
D
-
1
9
ca
s
es
p
r
ed
icted
b
y
th
e
m
o
d
el
to
th
e
to
tal
n
u
m
b
er
o
f
m
o
d
el
p
r
ed
ictio
n
s
(
i.e
.
,
tr
u
e
p
o
s
it
iv
es
(
T
P)
an
d
f
alse
p
o
s
itiv
es
(
FP
)
)
.
T
h
is
m
etr
ic
p
r
o
v
id
es
in
s
ig
h
ts
ab
o
u
t
th
e
r
ate
o
f
f
alse p
o
s
itiv
es.
T
h
e
lo
wer
th
e
n
u
m
b
er
o
f
f
alse p
o
s
itiv
es,
th
e
h
ig
h
er
t
h
e
y
ield
.
+
(
1
)
b.
R
ec
all:
T
h
is
is
th
e
s
en
s
itiv
ity
o
f
th
e
m
o
d
el.
I
t
is
th
e
r
atio
o
f
p
r
ed
icted
T
P
to
th
e
to
tal
n
u
m
b
er
o
f
ac
t
u
al
p
o
s
itiv
e
in
s
tan
ce
s
,
wh
ich
in
clu
d
es tr
u
e
p
o
s
itiv
es a
n
d
f
alse n
e
g
ativ
es.
=
+
(
2
)
c.
F1
-
s
co
r
e:
T
ak
es
in
t
o
ac
co
u
n
t
b
o
th
f
alse
p
o
s
itiv
es
an
d
f
alse n
eg
ativ
es
b
y
av
e
r
ag
in
g
th
e
p
r
e
cisi
o
n
an
d
r
ec
all
m
etr
ics.
I
t is im
p
o
r
tan
t in
ca
s
e
s
wh
er
e
th
e
d
is
tr
ib
u
tio
n
o
f
cla
s
s
es i
s
u
n
eq
u
al.
1
−
=
2
∗
∗
+
(
3
)
d.
Acc
u
r
ac
y
:
T
h
is
is
th
e
m
o
s
t
o
f
ten
u
s
ed
an
d
s
tr
aig
h
tf
o
r
war
d
ca
teg
o
r
izatio
n
m
etr
ic.
I
t
is
c
alcu
lated
as
th
e
n
u
m
b
er
o
f
co
r
r
ec
t
p
r
ed
ictio
n
s
d
iv
id
ed
b
y
th
e
n
u
m
b
er
o
f
s
am
p
les.
Alth
o
u
g
h
h
i
g
h
ac
cu
r
ac
y
is
ty
p
ically
d
esira
b
le,
in
s
o
m
e
ca
s
es
wh
er
e
th
e
class
d
is
tr
ib
u
tio
n
is
n
o
t
s
y
m
m
etr
ic,
it
m
ay
n
o
t
b
e
an
in
f
o
r
m
ativ
e
ev
alu
atio
n
.
I
n
th
ese
s
ce
n
a
r
io
s
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
o
f
f
er
a
m
o
r
e
t
h
o
r
o
u
g
h
ass
ess
m
en
t
o
f
m
o
d
el
p
er
f
o
r
m
an
ce
.
A
cc
u
r
a
cy
=
+
(
4
)
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
5
1
-
4
7
6
1
4758
4
.
3
.
P
er
f
o
r
m
a
nce
ev
a
lua
t
io
ns
a
nd
benc
hm
a
rk
ing
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
a
r
ch
itect
u
r
e
h
as
u
n
d
er
g
o
n
e
tr
ain
in
g
f
o
r
1
0
0
ep
o
c
h
s
.
T
h
e
p
er
f
o
r
m
an
ce
m
etr
ics
o
f
th
e
p
r
o
p
o
s
ed
ar
c
h
itectu
r
e
a
r
e
q
u
an
titativ
ely
s
u
m
m
ar
ized
in
T
ab
le
2
with
b
en
ch
m
a
r
k
in
g
m
o
d
els.
W
e
h
a
v
e
co
n
s
id
er
ed
v
ar
io
u
s
b
en
ch
m
ar
k
m
o
d
els
th
at
u
s
e
d
if
f
er
en
t
d
ee
p
lear
n
in
g
m
o
d
els
as
th
eir
b
ac
k
b
o
n
e.
W
e
ar
e
m
ain
ly
in
ter
ested
in
b
en
ch
m
a
r
k
m
o
d
els
th
at
ap
p
ly
t
o
C
T
im
ag
es.
Mo
s
t
o
f
th
e
b
en
ch
m
a
r
k
ap
p
r
o
ac
h
es
h
a
v
e
ev
alu
ated
th
eir
ar
ch
itectu
r
es
o
n
s
m
all
d
atasets
;
in
co
n
tr
ast,
we
h
av
e
p
er
f
o
r
m
ed
th
is
co
m
p
ar
is
o
n
ev
alu
atio
n
o
n
lar
g
er
d
atasets
[
2
9
]
,
[
3
0
]
.
W
e
u
s
ed
t
h
e
s
am
e
d
is
tr
ib
u
tio
n
o
f
d
atasets
in
to
t
r
ain
,
v
alid
atio
n
,
an
d
test
as
d
is
cu
s
s
ed
in
s
ec
tio
n
4
.
2
f
o
r
all
b
en
ch
m
a
r
k
ap
p
r
o
ac
h
es.
T
h
e
r
esu
lts
in
T
ab
le
2
clea
r
ly
s
h
o
w
th
at
th
e
p
r
o
p
o
s
ed
Ma
s
k
R
-
C
NN
m
o
d
el
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
s
all
b
en
ch
m
a
r
k
a
r
ch
itectu
r
es.
T
h
e
y
ield
e
d
g
ain
s
i
n
ac
cu
r
a
cy
,
r
ec
all,
a
n
d
F1
-
s
co
r
e
p
o
i
n
t
to
th
e
p
r
o
p
o
s
ed
m
o
d
el'
s
ad
ap
tab
ilit
y
in
C
OV
I
D
-
1
9
d
etec
tio
n
with
m
in
im
u
m
f
alse
p
o
s
itiv
es
an
d
f
alse
n
eg
ativ
es.
T
h
at
is
p
ar
ticu
lar
ly
v
ital
in
clin
ical
p
r
o
ce
d
u
r
es
wh
er
e
ea
r
ly
a
n
d
ac
c
u
r
ate
d
iag
n
o
s
is
ca
n
s
ig
n
if
ican
tly
af
f
ec
t
tr
ea
tm
en
t
p
lan
n
in
g
a
n
d
is
o
latio
n
p
r
o
to
c
o
ls
.
T
h
e
ab
ilit
y
to
d
is
tin
g
u
is
h
b
etwe
en
C
OVI
D
-
1
9
an
d
o
th
e
r
p
n
eu
m
o
n
ias
f
u
r
th
e
r
en
h
an
ce
s
its
u
tili
ty
in
r
ea
l
-
wo
r
ld
ap
p
licatio
n
s
.
Un
lik
e
p
r
ev
i
o
u
s
m
o
d
els
wh
ich
ar
e
ty
p
icall
y
tr
ain
e
d
o
n
lim
ited
d
ata,
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
s
h
o
ws
co
n
s
is
ten
t
p
er
f
o
r
m
a
n
ce
o
v
er
a
lar
g
e
d
ataset,
r
ef
lectin
g
its
g
en
er
aliza
b
ilit
y
.
T
h
e
s
u
p
er
io
r
m
o
d
el
p
er
f
o
r
m
a
n
ce
in
d
etec
tin
g
p
ath
o
g
en
ic
v
a
r
iatio
n
s
in
C
T
im
ag
es
co
m
es
f
r
o
m
th
e
in
te
g
r
atio
n
o
f
FP
N
an
d
f
u
zz
y
p
r
ep
r
o
ce
s
s
in
g
,
wh
ich
e
n
s
u
r
e
b
etter
s
p
atial
r
ep
r
esen
tatio
n
a
n
d
well
-
co
n
tr
asted
im
ag
e
in
p
u
ts
.
Ad
d
itio
n
ally
,
to
ass
ess
th
e
co
n
tr
ib
u
tio
n
o
f
k
ey
ar
c
h
itectu
r
al
co
m
p
o
n
e
n
ts
,
we
co
n
d
u
cted
a
n
ab
latio
n
s
tu
d
y
,
as
p
r
esen
ted
in
T
a
b
le
3
.
Sp
ec
if
ically
,
we
ev
alu
ated
t
h
e
ef
f
ec
ts
o
f
f
u
zz
y
co
lo
r
en
h
a
n
ce
m
en
t,
FP
N,
an
d
th
e
R
o
I
Alig
n
p
r
o
ce
s
s
.
Fo
r
co
m
p
ar
is
o
n
,
we
u
s
ed
a
b
aselin
e
m
o
d
el
b
ased
o
n
R
esNet
[
2
0
]
,
wh
ich
ex
clu
d
es th
ese
en
h
an
ce
m
e
n
ts
.
T
h
e
b
aselin
e
a
ch
iev
ed
a
p
r
ec
is
io
n
o
f
9
3
.
1
%
an
d
an
F1
-
s
co
r
e
o
f
9
1
.
4
%,
wh
er
ea
s
o
u
r
co
m
p
lete
m
o
d
el
attain
ed
9
8
.
4
%
p
r
ec
is
io
n
an
d
a
9
7
.
4
%
F1
-
s
co
r
e.
T
h
e
g
ain
ed
p
er
f
o
r
m
an
ce
im
p
r
o
v
e
m
en
t
em
p
h
asizes
th
e
im
p
o
r
tan
ce
o
f
ev
er
y
i
n
co
r
p
o
r
ated
co
m
p
o
n
e
n
t.
Hen
ce
,
f
u
zz
y
p
r
ep
r
o
ce
s
s
in
g
en
h
a
n
ce
s
co
n
tr
ast
an
d
s
tr
u
ct
u
r
al
v
is
ib
ilit
y
in
C
T
s
ca
n
s
.
F
PN
e
n
ab
les
ef
f
ec
tiv
e
m
u
lti
-
s
ca
le
f
ea
tu
r
e
ex
tr
ac
tio
n
,
an
d
R
o
I
Alig
n
en
s
u
r
es
ac
cu
r
ate
s
p
atial
alig
n
m
en
t
o
f
f
ea
tu
r
es
d
u
r
in
g
s
eg
m
e
n
tatio
n
.
T
h
ese
d
ev
elo
p
m
en
ts
tak
en
to
g
eth
er
h
elp
to
cr
ea
te
a
m
o
r
e
r
o
b
u
s
t a
n
d
clin
ically
r
elev
a
n
t
C
OVI
D
-
1
9
d
etec
tio
n
s
y
s
tem
.
T
ab
l
e
4
e
v
a
lu
a
te
s
th
e
p
r
o
p
o
s
ed
m
o
d
el
a
r
ch
it
ec
tu
r
e
i
n
th
e
ca
s
e
o
f
u
s
in
g
d
i
f
f
er
en
t
R
e
s
N
et
v
ar
ian
t
s
,
s
u
ch
a
s
R
e
s
N
et4
1
,
R
e
s
Ne
t5
0
,
an
d
R
e
s
N
et1
0
1
a
s
o
u
r
b
ac
k
b
o
n
e
n
et
wo
r
k
s
.
Ac
co
r
d
in
g
t
o
th
e
co
m
p
ar
at
iv
e
ev
al
u
a
tio
n
,
R
e
s
Ne
t5
0
o
f
f
er
s
th
e
b
e
s
t
b
al
an
c
e
b
et
we
en
p
er
f
o
r
m
an
ce
an
d
co
m
p
u
ta
t
io
n
a
l
ef
f
i
ci
en
cy
.
I
t
ac
h
i
ev
e
s
h
ig
h
s
co
r
e
s
ac
r
o
s
s
a
l
l e
v
a
lu
a
tio
n
m
e
tr
ic
s
,
w
i
th
a
p
r
ec
i
s
io
n
o
f
9
8
.
4
3
%,
r
ec
al
l
o
f
9
8
.
5
1
%
,
F1
-
s
co
r
e
o
f
9
7
.
4
6
%
,
an
d
ac
cu
r
ac
y
o
f
9
8
.
8
2
%.
O
n
o
n
e
h
an
d
,
R
e
s
Ne
t4
1
en
ab
le
s
f
a
s
t
er
p
r
o
ce
s
s
in
g
tim
e
s
a
t
th
e
ex
p
en
s
e
o
f
ac
cu
r
ac
y
an
d
r
e
ca
ll
,
wh
er
e
th
e
n
u
m
b
er
o
f
co
n
v
o
lu
t
io
n
al
la
y
er
s
i
s
r
ed
u
ce
d
.
On
th
e
o
th
er
h
an
d
,
R
e
s
Ne
t1
0
1
d
o
es
n
o
t
b
r
in
g
to
o
m
an
y
i
m
p
r
o
v
em
e
n
t
s
in
ter
m
s
o
f
ac
cu
r
ac
y
co
m
p
ar
ed
to
R
e
s
Ne
t5
0
,
d
esp
i
te
i
t
s
n
e
two
r
k
d
ep
th
d
ef
in
ed
b
y
th
e
q
u
an
t
it
y
o
f
co
n
v
o
lu
tio
n
al
l
ay
er
s
an
d
lo
n
g
er
p
r
o
ce
s
s
in
g
t
im
e
.
T
h
e
m
ar
g
in
al
g
ain
in
ac
cu
r
ac
y
(
o
n
l
y
0
.
1
2
%
o
v
er
R
e
s
Ne
t5
0
)
d
o
e
s
n
o
t
ju
s
ti
f
y
t
h
e
ad
d
it
io
n
a
l
co
s
t
in
m
o
s
t
p
r
ac
ti
ca
l
s
ce
n
ar
io
s
.
T
h
er
e
f
o
r
e,
R
e
s
N
et
5
0
em
er
g
e
s
as
th
e
m
o
s
t
s
u
i
tab
le
b
ac
k
b
o
n
e
f
o
r
o
u
r
ar
ch
i
te
ctu
r
e,
o
f
f
er
in
g
an
o
p
tim
al
tr
ad
e
-
o
f
f
b
e
tw
ee
n
s
p
ee
d
,
m
o
d
el
s
i
z
e,
an
d
d
e
te
ct
io
n
p
er
f
o
r
m
an
ce
,
m
ak
in
g
it
w
el
l
-
s
u
it
ed
f
o
r
d
ep
lo
y
m
en
t
i
n
r
e
al
-
wo
r
l
d
c
lin
ic
al
s
et
t
in
g
s
.
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
ar
c
h
itectu
r
e
an
d
b
en
ch
m
a
r
k
ar
c
h
itectu
r
e
M
o
d
e
l
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
A
c
c
u
r
a
c
y
V
G
G
1
6
-
b
a
s
e
d
mo
d
e
l
[
7
]
9
3
.
2
0
%
9
4
.
1
0
%
9
3
.
6
0
%
9
4
.
1
0
%
B
a
y
e
s
i
a
n
-
b
a
se
d
m
o
d
e
l
[
3
1
]
9
2
.
4
0
%
9
2
.
8
0
%
9
2
.
1
0
%
9
2
.
5
0
%
D
e
n
sN
e
t
-
b
a
se
d
m
o
d
e
l
[
3
2
]
9
3
.
3
0
%
9
4
.
1
0
%
9
2
.
5
0
%
9
3
.
7
0
%
R
e
sN
e
t
-
b
a
se
d
m
o
d
e
l
[
8
]
9
3
.
8
0
%
9
3
.
4
0
%
9
3
.
6
0
%
9
3
.
9
0
%
Pr
o
p
o
sed
m
o
d
e
l
a
r
c
h
i
t
e
c
t
u
r
e
9
8
.
4
0
%
9
8
.
5
0
%
9
7
.
4
0
%
9
8
.
8
0
%
T
ab
le
3
.
Ab
latio
n
s
tu
d
y
: p
e
r
f
o
r
m
an
ce
co
m
p
ar
is
o
n
with
a
n
d
with
o
u
t k
ey
e
n
h
an
ce
m
en
ts
A
r
c
h
i
t
e
c
t
u
r
e
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
A
c
c
u
r
a
c
y
B
a
se
l
i
n
e
R
e
sN
e
t
[
2
0
]
9
3
.
1
%
9
3
.
6
%
9
1
.
4
%
9
3
.
6
%
P
r
o
p
o
se
d
f
u
l
l
m
o
d
e
l
9
8
.
4
%
9
8
.
5
%
9
7
.
4
%
9
8
.
8
%
T
ab
le
4
.
Pro
p
o
s
ed
m
o
d
el
ar
c
h
itectu
r
e
ev
alu
atio
n
with
r
esp
ec
t
to
d
if
f
er
e
n
t b
ac
k
b
o
n
e
lay
er
s
B
a
c
k
b
o
n
e
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
A
c
c
u
r
a
c
y
R
e
sN
e
t
4
1
9
3
.
7
3
%
9
0
.
1
3
%
9
4
.
6
4
%
9
3
.
7
2
%
R
e
sN
e
t
5
0
9
8
.
4
3
%
9
8
.
5
1
%
9
7
.
4
6
%
9
8
.
8
2
%
R
e
sN
e
t
1
0
1
9
8
.
8
9
%
9
8
.
8
1
%
9
7
.
9
5
%
9
8
.
9
4
%
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:
2088
-
8
7
0
8
E
fficien
t m
a
s
k
r
eg
io
n
-
b
a
s
ed
c
o
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
-
b
a
s
ed
a
r
ch
itectu
r
e
…
(
N
a
d
er Ma
h
mo
u
d
)
4759
Fig
u
r
e
5
d
ep
icts
th
e
co
n
f
u
s
io
n
m
atr
ix
f
o
r
th
e
p
r
o
p
o
s
ed
m
o
d
el
ar
ch
itectu
r
e
.
T
h
e
an
aly
s
is
o
f
m
is
class
if
icatio
n
s
r
ev
ea
ls
th
at
3
3
n
o
n
-
C
OVI
D
-
1
9
p
n
eu
m
o
n
ia
im
ag
es
wer
e
m
i
s
clas
s
if
ied
as
n
o
r
m
al,
an
d
1
3
C
OVI
D
-
1
9
im
ag
es
wer
e
m
is
c
lass
if
ied
as
n
o
r
m
al.
Fu
r
th
e
r
,
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
m
is
class
if
ied
4
2
C
OVI
D
-
1
9
ca
s
es
a
s
n
o
n
-
C
OVI
D
-
1
9
p
n
eu
m
o
n
ia,
wh
ile
5
4
n
o
n
-
C
OVI
D
-
1
9
ca
s
es
wer
e
m
is
cla
s
s
if
ied
a
s
C
O
VI
D
-
1
9
.
T
h
is
in
d
icate
s
d
if
f
icu
lty
in
d
is
tin
g
u
is
h
in
g
b
etwe
en
t
h
ese
two
clin
ically
s
im
ilar
co
n
d
itio
n
s
,
w
h
ich
is
im
p
o
r
tan
t
in
clin
ical
p
r
o
ce
d
u
r
es
wh
er
e
co
r
r
ec
t
tr
ea
tm
en
t
d
ep
en
d
s
o
n
p
r
o
p
er
class
if
icatio
n
.
T
h
e
m
is
clas
s
if
icatio
n
s
o
f
C
OVI
D
-
1
9
as n
o
n
-
C
OVI
D
-
1
9
p
n
eu
m
o
n
ia
(
f
alse n
eg
ativ
es)
ar
e
a
cr
itical
co
n
ce
r
n
,
as th
ey
m
ay
lead
to
d
elay
e
d
o
r
im
p
r
o
p
er
tr
ea
tm
e
n
t
d
ec
is
io
n
s
.
C
o
n
v
er
s
ely
,
th
e
m
is
cl
ass
if
icatio
n
s
o
f
n
o
n
-
C
OVI
D
-
1
9
p
n
eu
m
o
n
ia
as
C
OVI
D
-
1
9
(
f
alse
p
o
s
itiv
es)
co
u
ld
r
esu
lt
in
u
n
n
ec
ess
ar
y
is
o
latio
n
,
th
o
u
g
h
th
e
y
p
o
s
e
a
lo
wer
r
is
k
in
ter
m
s
o
f
p
u
b
lic
h
ea
lth
m
an
a
g
em
en
t.
C
o
n
s
id
er
in
g
th
e
m
o
d
el
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
,
th
e
y
ield
ed
s
en
s
itiv
ity
d
etec
tio
n
f
o
r
C
OVI
D
-
1
9
is
9
8
.
8
7
%,
wh
ile
th
e
s
p
ec
if
icity
is
9
9
.
4
4
%.
So
m
e
m
is
class
if
icatio
n
s
r
esu
lt
f
r
o
m
th
e
r
ad
io
lo
g
ical
s
im
ilar
ities
b
etwe
en
C
OVI
D
-
1
9
an
d
o
th
er
f
o
r
m
s
o
f
p
n
eu
m
o
n
ia,
wh
ich
m
ak
e
d
i
s
tin
ctio
n
d
if
f
icu
lt.
Fig
u
r
e
5
.
C
o
n
f
u
s
io
n
m
atr
i
x
r
e
p
r
esen
tin
g
th
e
class
if
icatio
n
r
e
s
u
lts
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
o
n
th
e
d
ataset
[
2
9
]
5.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
p
r
o
p
o
s
ed
an
ef
f
ec
t
iv
e
Ma
s
k
R
-
C
NN
-
b
ased
ar
ch
itectu
r
e
f
o
r
th
e
d
iag
n
o
s
is
o
f
C
OVI
D
-
19
f
r
o
m
c
h
est
C
T
im
ag
es.
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
in
co
r
p
o
r
ates
a
n
o
v
el
im
a
g
e
e
n
h
an
ce
m
en
t
tech
n
iq
u
e
b
ase
d
o
n
f
u
zz
y
l
o
g
ic,
y
ield
in
g
n
o
ta
b
le
im
p
r
o
v
em
en
ts
in
co
n
tr
ast
an
d
r
ed
u
ce
s
C
T
im
a
g
e
am
b
ig
u
ity
.
Mo
r
eo
v
er
,
ad
d
itio
n
al
c
o
m
p
o
n
en
ts
ar
e
in
teg
r
ated
,
s
u
ch
as
FP
N
a
n
d
R
o
I
Alig
n
,
wh
ich
en
h
an
ce
s
p
atial
f
ea
tu
r
e
r
ep
r
esen
tatio
n
a
n
d
class
if
icatio
n
ac
cu
r
ac
y
.
T
h
e
p
r
esen
ted
ar
ch
itectu
r
e
is
q
u
an
titativ
ely
ass
ess
ed
o
n
a
g
e
n
er
al
d
ataset,
h
en
ce
y
ield
s
a
s
u
p
er
i
o
r
p
er
f
o
r
m
a
n
ce
o
v
er
s
ev
e
r
al
b
en
ch
m
ar
k
m
o
d
els
with
an
ac
cu
r
ac
y
o
f
9
8
.
8
%.
T
h
ese
f
in
d
in
g
s
s
h
o
w
th
e
m
o
d
el’
s
r
eliab
ilit
y
an
d
r
o
b
u
s
tn
e
s
s
,
m
ak
in
g
it
a
p
r
o
m
is
in
g
ca
n
d
id
ate
f
o
r
clin
ical
s
u
p
p
o
r
t
in
th
e
ea
r
ly
d
etec
tio
n
o
f
C
OVI
D
-
1
9
.
Ap
ar
t
f
r
o
m
it
s
im
m
ed
iate
ap
p
licatio
n
to
C
OVI
D
-
1
9
d
iag
n
o
s
is
,
th
e
m
o
d
el'
s
m
o
d
u
lar
d
esig
n
an
d
ad
a
p
tab
ilit
y
im
p
ly
a
g
r
ea
t
p
o
s
s
ib
ilit
y
f
o
r
e
x
ten
s
io
n
to
o
th
er
ch
est
-
r
elate
d
d
is
ea
s
es,
s
u
ch
as
n
o
n
-
C
OVI
D
-
1
9
p
n
eu
m
o
n
ia
o
r
lu
n
g
f
i
b
r
o
s
i
s
.
Fu
r
th
er
m
o
r
e
,
its
p
er
f
o
r
m
an
c
e
o
n
a
lar
g
e
d
ataset
en
h
an
ce
s
its
g
en
er
aliza
b
ilit
y
an
d
p
r
ac
tical
p
r
ep
a
r
atio
n
f
o
r
ap
p
licatio
n
in
m
e
d
ical
im
ag
in
g
s
y
s
tem
s
.
I
n
s
u
m
m
ar
y
,
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
a
r
ch
itectu
r
e
d
e
m
o
n
s
tr
ates
h
o
w
tailo
r
ed
p
r
ep
r
o
ce
s
s
in
g
an
d
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es
ca
n
s
i
g
n
if
ican
tly
im
p
r
o
v
e
d
is
ea
s
e
d
etec
tio
n
ac
c
u
r
ac
y
.
Fu
tu
r
e
wo
r
k
c
o
u
ld
ex
p
l
o
r
e
in
co
r
p
o
r
atin
g
a
d
d
itio
n
al
clin
ic
al
m
etad
ata
—
s
u
ch
as
p
atien
t
s
y
m
p
to
m
s
,
m
ed
ical
h
is
to
r
y
,
o
r
lab
o
r
ato
r
y
r
esu
lts
,
wh
ich
co
u
ld
f
u
r
th
e
r
im
p
r
o
v
e
m
o
d
el’
s
ca
p
ab
ilit
y
to
d
if
f
er
en
tiate
b
etwe
en
v
is
u
ally
s
im
ilar
C
T
s
ca
n
s
an
d
im
p
r
o
v
e
d
iag
n
o
s
tic
p
r
ec
is
io
n
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
au
th
o
r
s
d
ec
lar
e
th
at
n
o
f
u
n
d
in
g
was r
ec
eiv
e
d
f
o
r
th
is
s
tu
d
y
.
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
5
1
-
4
7
6
1
4760
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
i
d
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Nad
er
Ma
h
m
o
u
d
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ash
r
af
B
.
E
l
-
Sis
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
is
av
ailab
le
at
h
ttp
s
:
//g
ith
u
b
.
co
m/h
a
yd
en
g
u
n
r
a
j/C
OV
I
DN
et
-
CT
,
r
ef
er
en
ce
n
u
m
b
e
r
[
3
0
]
.
RE
F
E
R
E
NC
E
S
[
1
]
T.
O
z
t
u
r
k
,
M
.
Ta
l
o
,
E.
A
.
Y
i
l
d
i
r
i
m,
U
.
B
.
B
a
l
o
g
l
u
,
O
.
Y
i
l
d
i
r
i
m,
a
n
d
U
.
R
a
j
e
n
d
r
a
A
c
h
a
r
y
a
,
“
A
u
t
o
ma
t
e
d
d
e
t
e
c
t
i
o
n
o
f
C
O
V
I
D
-
1
9
c
a
ses
u
si
n
g
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
w
i
t
h
X
-
r
a
y
i
ma
g
e
s,
”
C
o
m
p
u
t
e
rs
i
n
Bi
o
l
o
g
y
a
n
d
M
e
d
i
c
i
n
e
,
v
o
l
.
1
2
1
,
Ju
n
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
mp
b
i
o
m
e
d
.
2
0
2
0
.
1
0
3
7
9
2
.
[
2
]
F
.
K
h
a
t
a
mi
e
t
a
l
.
,
“
A
m
e
t
a
-
a
n
a
l
y
s
i
s
o
f
a
c
c
u
r
a
c
y
a
n
d
se
n
si
t
i
v
i
t
y
o
f
c
h
e
st
C
T
a
n
d
R
T
-
P
C
R
i
n
C
O
V
I
D
-
1
9
d
i
a
g
n
o
s
i
s,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
r
t
s
,
v
o
l
.
1
0
,
n
o
.
1
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
5
9
8
-
020
-
8
0
0
6
1
-
2.
[
3
]
M
.
A
h
m
a
d
e
t
a
l
.
,
“
D
e
e
p
b
e
l
i
e
f
n
e
t
w
o
r
k
mo
d
e
l
i
n
g
f
o
r
a
u
t
o
mat
i
c
l
i
v
e
r
se
g
me
n
t
a
t
i
o
n
,
”
I
E
EE
Ac
c
e
ss
,
v
o
l
.
7
,
p
p
.
2
0
5
8
5
–
2
0
5
9
5
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
a
c
c
e
ss.
2
0
1
9
.
2
8
9
6
9
6
1
.
[
4
]
T.
A
i
e
t
a
l
.
,
“
C
o
r
r
e
l
a
t
i
o
n
o
f
c
h
e
s
t
C
T
a
n
d
R
T
-
P
C
R
t
e
s
t
i
n
g
f
o
r
c
o
r
o
n
a
v
i
r
u
s
d
i
s
e
a
se
2
0
1
9
(
C
O
V
I
D
-
1
9
)
i
n
C
h
i
n
a
:
A
r
e
p
o
r
t
o
f
1
0
1
4
c
a
ses
,
”
Ra
d
i
o
l
o
g
y
,
v
o
l
.
2
9
6
,
n
o
.
2
,
p
p
.
E3
2
–
E
4
0
,
A
u
g
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
4
8
/
r
a
d
i
o
l
.
2
0
2
0
2
0
0
6
4
2
.
[
5
]
J.
S
h
a
t
r
i
e
t
a
l
.
,
“
Th
e
r
o
l
e
o
f
c
h
e
st
c
o
mp
u
t
e
d
t
o
mo
g
r
a
p
h
y
i
n
a
s
y
m
p
t
o
ma
t
i
c
p
a
t
i
e
n
t
s
o
f
p
o
s
i
t
i
v
e
c
o
r
o
n
a
v
i
r
u
s
d
i
se
a
se
2
0
1
9
:
A
c
a
se
a
n
d
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
,
”
J
o
u
r
n
a
l
o
f
C
l
i
n
i
c
a
l
I
m
a
g
i
n
g
S
c
i
e
n
c
e
,
v
o
l
.
1
0
,
p
.
3
5
,
J
u
n
.
2
0
2
0
,
d
o
i
:
1
0
.
2
5
2
5
9
/
j
c
i
s_
5
8
_
2
0
2
0
.
[
6
]
S
.
S
i
d
d
i
q
u
e
a
n
d
J
.
C
.
L.
C
h
o
w
,
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
i
n
h
e
a
l
t
h
c
a
r
e
c
o
mm
u
n
i
c
a
t
i
o
n
,
”
En
c
y
c
l
o
p
e
d
i
a
,
v
o
l
.
1
,
n
o
.
1
,
p
p
.
2
2
0
–
2
3
9
,
F
e
b
.
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
e
n
c
y
c
l
o
p
e
d
i
a
1
0
1
0
0
2
1
.
[
7
]
K
.
M
.
M
.
U
d
d
i
n
e
t
a
l
.
,
“
F
e
a
t
u
r
e
f
u
si
o
n
b
a
se
d
V
G
G
F
u
s
i
o
n
N
e
t
m
o
d
e
l
t
o
d
e
t
e
c
t
C
O
V
I
D
-
1
9
p
a
t
i
e
n
t
s u
t
i
l
i
z
i
n
g
c
o
m
p
u
t
e
d
t
o
m
o
g
r
a
p
h
y
sca
n
i
m
a
g
e
s,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
r
t
s
,
v
o
l
.
1
2
,
n
o
.
1
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
022
-
2
5
5
3
9
-
x.
[
8
]
Y
.
S
o
n
g
e
t
a
l
.
,
“
D
e
e
p
l
e
a
r
n
i
n
g
e
n
a
b
l
e
s
a
c
c
u
r
a
t
e
d
i
a
g
n
o
si
s
o
f
n
o
v
e
l
c
o
r
o
n
a
v
i
r
u
s
(
C
O
V
I
D
-
1
9
)
w
i
t
h
C
T
i
ma
g
e
s,
”
I
EEE
/
A
C
M
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
Bi
o
l
o
g
y
a
n
d
B
i
o
i
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
1
8
,
n
o
.
6
,
p
p
.
2
7
7
5
–
2
7
8
0
,
N
o
v
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
T
C
B
B
.
2
0
2
1
.
3
0
6
5
3
6
1
.
[
9
]
J.
W
.
J
o
h
n
s
o
n
,
“
A
u
t
o
ma
t
i
c
n
u
c
l
e
u
s
seg
m
e
n
t
a
t
i
o
n
w
i
t
h
M
a
sk
-
R
C
N
N
,
”
i
n
A
d
v
a
n
c
e
s
i
n
C
o
m
p
u
t
e
r
V
i
si
o
n
,
S
p
r
i
n
g
e
r
I
n
t
e
r
n
a
t
i
o
n
a
l
P
u
b
l
i
s
h
i
n
g
,
2
0
1
9
,
p
p
.
3
9
9
–
4
0
7
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
1
7
7
9
8
-
0
_
3
2
.
[
1
0
]
K
.
L
i
n
e
t
a
l
.
,
“
F
a
c
e
d
e
t
e
c
t
i
o
n
a
n
d
s
e
g
me
n
t
a
t
i
o
n
b
a
se
d
o
n
i
mp
r
o
v
e
d
mas
k
R
-
C
N
N
,
”
D
i
s
c
re
t
e
D
y
n
a
m
i
c
s
i
n
N
a
t
u
re
a
n
d
S
o
c
i
e
t
y
,
v
o
l
.
2
0
2
0
,
p
p
.
1
–
1
1
,
M
a
y
2
0
2
0
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
0
/
9
2
4
2
9
1
7
.
[
1
1
]
H
.
A
l
m
u
b
a
r
a
k
,
Y
.
B
a
z
i
,
a
n
d
N
.
A
l
a
j
l
a
n
,
“
Tw
o
-
st
a
g
e
M
a
s
k
-
R
C
N
N
a
p
p
r
o
a
c
h
f
o
r
d
e
t
e
c
t
i
n
g
a
n
d
se
g
me
n
t
i
n
g
t
h
e
o
p
t
i
c
n
e
r
v
e
h
e
a
d
,
o
p
t
i
c
d
i
s
c
,
a
n
d
o
p
t
i
c
c
u
p
i
n
f
u
n
d
u
s
i
m
a
g
e
s,”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
0
,
n
o
.
1
1
,
p
.
3
8
3
3
,
M
a
y
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
0
1
1
3
8
3
3
.
[
1
2
]
S
.
M
u
l
a
y
,
G
.
D
e
e
p
i
k
a
,
S
.
J
e
e
v
a
k
a
l
a
,
K
.
R
a
m,
a
n
d
M
.
S
i
v
a
p
r
a
k
a
sa
m,
“
L
i
v
e
r
se
g
m
e
n
t
a
t
i
o
n
f
r
o
m
mu
l
t
i
m
o
d
a
l
i
m
a
g
e
s
u
s
i
n
g
H
ED
-
M
a
s
k
R
-
C
N
N
,
”
i
n
M
u
l
t
i
s
c
a
l
e
M
u
l
t
i
m
o
d
a
l
Me
d
i
c
a
l
I
m
a
g
i
n
g
,
S
p
r
i
n
g
e
r
I
n
t
e
r
n
a
t
i
o
n
a
l
P
u
b
l
i
s
h
i
n
g
,
2
0
1
9
,
p
p
.
6
8
–
75
,
doi
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
3
7
9
6
9
-
8
_
9
.
[
1
3
]
J.
-
H
.
S
h
u
,
F
.
-
D
.
N
i
a
n
,
M
.
-
H
.
Y
u
,
a
n
d
X
.
Li
,
“
A
n
i
m
p
r
o
v
e
d
M
a
sk
R
-
C
N
N
mo
d
e
l
f
o
r
m
u
l
t
i
o
r
g
a
n
se
g
me
n
t
a
t
i
o
n
,
”
M
a
t
h
e
m
a
t
i
c
a
l
Pro
b
l
e
m
s i
n
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
0
2
0
,
p
p
.
1
–
1
1
,
Ju
l
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
0
/
8
3
5
1
7
2
5
.
[
1
4
]
J.
-
Y
.
C
h
i
a
o
,
K
.
-
Y
.
C
h
e
n
,
K
.
Y
.
-
K
.
Li
a
o
,
P
.
-
H
.
H
s
i
e
h
,
G
.
Z
h
a
n
g
,
a
n
d
T
.
-
C
.
H
u
a
n
g
,
“
D
e
t
e
c
t
i
o
n
a
n
d
c
l
a
ssi
f
i
c
a
t
i
o
n
t
h
e
b
r
e
a
s
t
t
u
m
o
r
s
u
si
n
g
mas
k
R
-
C
N
N
o
n
s
o
n
o
g
r
a
ms,
”
Me
d
i
c
i
n
e
,
v
o
l
.
9
8
,
n
o
.
1
9
,
p
.
e
1
5
2
0
0
,
M
a
y
2
0
1
9
,
d
o
i
:
1
0
.
1
0
9
7
/
m
d
.
0
0
0
0
0
0
0
0
0
0
0
1
5
2
0
0
.
[
1
5
]
T.
S
h
i
b
a
t
a
,
A
.
T
e
r
a
m
o
t
o
,
H
.
Y
a
m
a
d
a
,
N
.
O
h
m
i
y
a
,
K
.
S
a
i
t
o
,
a
n
d
H
.
F
u
j
i
t
a
,
“
A
u
t
o
ma
t
e
d
d
e
t
e
c
t
i
o
n
a
n
d
s
e
g
m
e
n
t
a
t
i
o
n
o
f
e
a
r
l
y
g
a
st
r
i
c
c
a
n
c
e
r
f
r
o
m
e
n
d
o
sc
o
p
i
c
i
m
a
g
e
s
u
si
n
g
M
a
s
k
R
-
C
N
N
,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
0
,
n
o
.
1
1
,
p
.
3
8
4
2
,
M
a
y
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
0
1
1
3
8
4
2
.
[
1
6
]
C
.
J
i
n
e
t
a
l
.
,
“
D
e
v
e
l
o
p
me
n
t
a
n
d
e
v
a
l
u
a
t
i
o
n
o
f
a
n
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
s
y
st
e
m
f
o
r
C
O
V
I
D
-
1
9
d
i
a
g
n
o
si
s,
”
N
a
t
u
re
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
v
o
l
.
1
1
,
n
o
.
1
,
O
c
t
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
4
6
7
-
020
-
1
8
6
8
5
-
1.
[
1
7
]
S
.
H
u
e
t
a
l
.
,
“
W
e
a
k
l
y
su
p
e
r
v
i
se
d
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
C
O
V
I
D
-
1
9
i
n
f
e
c
t
i
o
n
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
ssi
f
i
c
a
t
i
o
n
f
r
o
m
C
T
i
ma
g
e
s,
”
I
EE
E
Ac
c
e
ss
,
v
o
l
.
8
,
p
p
.
1
1
8
8
6
9
–
1
1
8
8
8
3
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
a
c
c
e
ss.
2
0
2
0
.
3
0
0
5
5
1
0
.
[
1
8
]
M
.
P
o
l
si
n
e
l
l
i
,
L.
C
i
n
q
u
e
,
a
n
d
G
.
P
l
a
c
i
d
i
,
“
A
l
i
g
h
t
C
N
N
f
o
r
d
e
t
e
c
t
i
n
g
C
O
V
I
D
-
1
9
f
r
o
m
C
T
s
c
a
n
s
o
f
t
h
e
c
h
e
st
,
”
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
L
e
t
t
e
rs
,
v
o
l
.
1
4
0
,
p
p
.
9
5
–
1
0
0
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
r
e
c
.
2
0
2
0
.
1
0
.
0
0
1
.
[
1
9
]
S
.
B
i
sw
a
s,
S
.
C
h
a
t
t
e
r
j
e
e
,
A
.
M
a
j
e
e
,
S
.
S
e
n
,
F
.
S
c
h
w
e
n
k
e
r
,
a
n
d
R
.
S
a
r
k
a
r
,
“
P
r
e
d
i
c
t
i
o
n
o
f
C
O
V
I
D
-
1
9
f
r
o
m
c
h
e
s
t
C
T
i
m
a
g
e
s
u
si
n
g
a
n
e
n
sem
b
l
e
o
f
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
1
,
n
o
.
1
5
,
p
.
7
0
0
4
,
J
u
l
.
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
1
1
5
7
0
0
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.