I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14
,
No.
4
,
Augus
t
2025
,
pp.
3099
~
3108
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
4
.
pp
30
99
-
3108
3099
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
E
f
f
ic
ie
n
t
lu
n
g
d
is
e
as
e
c
la
ssi
f
ic
at
io
n
t
h
r
o
u
gh
lu
m
in
e
s
c
e
n
t
f
e
a
t
u
r
e
se
le
c
t
io
n
u
si
n
g
f
ir
e
f
ly
al
gor
it
h
m
Anj
u
gam
S
h
an
m
u
gave
lu
,
Ar
u
l
L
e
e
n
a
Ros
e
P
e
t
e
r
Jos
e
p
h
D
e
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ie
nc
e
,
F
a
c
ul
ty
of
S
c
ie
nc
e
a
nd
H
um
a
ni
ti
e
s
,
S
R
M
I
ns
ti
tu
te
of
S
c
ie
nc
e
a
nd
T
e
c
hnol
ogy,
C
he
nga
lp
a
tt
u
,
I
ndi
a
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
M
a
y
5,
2024
R
e
vis
e
d
F
e
b
24,
2025
Ac
c
e
pted
M
a
r
15,
2025
O
v
er
t
h
e
p
as
t
co
u
p
l
e
of
d
ecad
e
s
,
t
h
ere
h
a
s
b
een
a
s
u
b
s
t
an
t
i
al
i
n
creas
e
in
the
p
rev
a
l
en
ce
of
l
u
n
g
ai
l
men
t
s
,
res
u
l
t
i
n
g
in
3
.
5
mi
l
l
i
o
n
fat
a
l
i
t
i
e
s
each
y
ear.
T
h
i
s
n
eces
s
i
t
at
e
s
the
ad
o
p
t
i
o
n
of
a
l
u
n
g
d
i
s
ea
s
e
d
et
ec
t
i
o
n
t
ech
n
o
l
o
g
y
t
h
a
t
is
effect
i
v
e,
t
r
u
s
t
w
o
rt
h
y
,
a
n
d
c
o
s
t
-
effect
i
v
e.
In
t
h
i
s
s
t
u
d
y
,
we
p
r
o
p
o
s
e
an
o
p
t
i
mi
zed
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
ral
n
et
w
o
r
k
(CN
N
)
mo
d
el
,
u
s
e
d
fo
r
m
u
l
t
i
c
l
as
s
cat
eg
o
ri
za
t
i
o
n
of
l
u
n
g
ai
l
me
n
t
s
b
as
e
d
on
fro
n
t
al
ch
es
t
X
-
ra
y
s
.
T
h
e
cl
as
s
i
f
i
cat
i
o
n
i
n
cl
u
d
e
s
fo
u
r
cat
e
g
o
r
i
es
:
CO
V
ID
-
19,
v
i
ral
p
n
eu
m
o
n
i
a,
l
u
n
g
o
p
ac
i
t
y
,
an
d
n
o
n
-
i
n
fect
i
o
u
s
n
o
rmal
g
r
o
u
p
.
We
i
mp
l
emen
t
e
d
the
fi
ref
l
y
al
g
o
ri
t
h
m
to
o
p
t
i
m
i
ze
t
h
e
g
l
o
b
a
l
effi
ci
e
n
cy
of
feat
u
re
s
el
ec
t
i
o
n
of
t
h
e
l
u
n
g
ab
n
o
rmal
i
t
y
in
the
X
-
ray
i
ma
g
es
of
l
u
n
g
d
i
s
eas
e
an
d
C
O
V
I
D
-
19
to
c
l
as
s
i
f
y
the
i
n
p
u
t
acco
r
d
i
n
g
to
t
h
e
t
arg
e
t
cl
a
s
s
.
T
h
e
p
ro
p
o
s
ed
al
g
o
ri
t
h
m
w
a
s
t
es
t
ed
f
o
r
acc
u
racy
,
p
rec
i
s
i
o
n
,
recal
l
,
an
d
F1
-
s
co
re.
T
h
e
fi
n
d
i
n
g
s
w
ere
v
al
i
d
a
t
ed
u
s
i
n
g
the
t
ra
n
s
fer
l
e
arn
i
n
g
mo
d
e
l
V
G
G
-
16;
the
al
g
o
r
i
t
h
m
ach
i
e
v
ed
a
s
u
p
er
i
o
r
accu
racy
of
9
9
.
3
%
co
mp
are
d
to
t
h
a
t
of
o
t
h
er
cu
t
t
i
n
g
-
e
d
g
e
mo
d
el
s
s
u
c
h
as
In
cep
t
i
o
n
v
3
an
d
Res
N
et
5
0
.
K
e
y
w
o
r
d
s
:
C
onvolut
ional
ne
ur
a
l
ne
twor
k
F
ir
e
f
ly
a
lgo
r
it
hm
Hybr
id
f
e
a
tur
e
s
e
lec
ti
on
L
ung
dis
e
a
s
e
c
la
s
s
if
ica
ti
on
VGG
-
16
Th
i
s
is
an
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
the
CC
BY
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
Ar
ul
L
e
e
na
R
os
e
P
e
ter
J
os
e
ph
De
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ienc
e
,
F
a
c
ult
y
of
S
c
ien
c
e
a
nd
Huma
nit
ies
S
R
M
I
ns
ti
tut
e
of
S
c
ienc
e
a
nd
T
e
c
hnology
Ka
tt
a
nkulathur
-
603203,
C
he
nga
lpattu,
T
a
mi
l
Na
du
,
I
ndia
E
mail
:
lee
na
.
r
os
e
527@gmail.
c
om
1.
I
NT
RODU
C
T
I
ON
T
he
lung
is
a
c
r
uc
ial
or
ga
n
in
the
human
body,
a
nd
lung
dis
e
a
s
e
s
can
c
a
us
e
s
e
r
ious
he
a
lt
h
is
s
ue
s
,
including
im
pa
ir
e
d
lung
f
unc
ti
on,
dif
f
iculty
b
r
e
a
thi
ng,
a
nd
e
ve
n
de
a
th
if
not
tr
e
a
ted
or
diagnos
e
d
pr
ompt
ly.
C
ons
e
que
ntl
y,
r
e
s
pir
a
tor
y
a
il
ments
a
r
e
the
th
ir
d
-
lea
ding
c
a
us
e
of
mor
talit
y
globally
,
r
e
s
ult
ing
in
a
r
o
und
f
ive
mi
ll
ion
f
a
talit
ies
pe
r
ye
a
r
.
T
his
c
ould
be
a
tt
r
ibu
t
e
d
to
s
e
ve
r
a
l
va
r
iable
s
includ
ing
c
ontaminants
in
the
a
ir
,
mi
c
r
obial
inf
il
t
r
a
ti
on,
c
he
mi
c
a
l
c
ons
umpt
ion,
a
nd
phys
ica
l
a
il
ments
.
F
ur
ther
mor
e
,
lung
dis
e
a
s
e
s
,
pa
r
ti
c
ular
ly
pne
umoni
a
,
C
OV
I
D
-
19,
tuber
c
ulos
is
,
a
nd
pne
umot
hor
a
x
a
r
e
indi
c
a
ted
as
the
major
c
a
us
e
s
of
de
a
th
wor
ldwide
by
the
W
or
ld
He
a
lt
h
Or
ga
niza
ti
on
(
W
H
O)
[
1]
.
P
n
e
u
mo
ni
a
is
a
l
un
g
i
n
f
e
c
ti
on
c
a
us
e
d
by
ba
c
te
r
ia
,
vi
r
us
e
s
,
or
f
u
ng
i
.
It
r
e
s
ul
ts
in
in
f
la
mm
a
t
io
n
a
nd
f
lu
id
b
u
il
du
p
in
the
l
un
g's
a
i
r
s
a
c
s
,
l
e
a
di
ng
to
b
r
e
a
th
i
ng
d
i
f
f
i
c
u
lt
ies
a
n
d
s
y
m
pt
oms
l
ike
c
o
ug
h
,
c
he
s
t
pa
i
n
,
a
n
d
f
e
ve
r
[
2
]
.
S
i
m
il
a
r
l
y
,
C
OV
I
D
-
19
,
or
t
he
n
ov
e
l
c
o
r
o
na
vi
r
us
d
is
e
a
s
e
,
is
c
a
us
e
d
by
th
e
v
i
r
us
S
AR
S
-
C
oV
-
2.
It
a
pp
e
a
r
e
d
in
l
a
te
20
19
a
n
d
r
a
pi
dl
y
be
c
a
m
e
a
g
lo
ba
l
p
a
n
de
mi
c
.
T
h
e
dis
e
a
s
e
ma
i
nl
y
im
pa
c
ts
th
e
r
e
s
pi
r
a
t
or
y
s
ys
t
e
m
,
le
a
d
in
g
to
s
ym
pt
oms
l
i
ke
f
e
ve
r
,
c
ou
gh
,
s
h
or
t
ne
s
s
of
b
r
e
a
th
,
a
nd
f
a
ti
gu
e
.
In
s
e
ve
r
e
c
a
s
e
s
,
it
c
a
n
c
a
us
e
p
ne
u
m
on
ia
a
n
d
a
c
ute
r
e
s
pi
r
a
t
or
y
dis
t
r
e
s
s
s
yn
d
r
o
me
(
AR
D
S
)
.
As
a
r
e
s
ul
t
of
d
e
la
ye
d
r
e
s
ul
ts
r
e
po
r
ti
ng
,
l
im
i
ted
t
e
s
t
in
g
c
a
pa
c
i
t
y
,
a
nd
i
na
de
qu
a
te
di
a
g
nos
is
,
t
he
C
O
V
I
D
-
19
p
a
n
de
mi
c
h
a
s
r
a
is
e
d
mo
r
tal
i
ty
r
a
tes
.
T
h
e
r
e
f
o
r
e
,
im
p
r
o
ve
d
tes
t
in
g
i
n
f
r
a
s
t
r
uc
t
ur
e
,
un
iv
e
r
s
a
l
a
c
c
e
s
s
,
a
n
d
e
f
f
e
c
ti
ve
d
iag
n
os
t
ic
te
c
h
ni
qu
e
s
a
r
e
e
s
s
e
n
t
ial
f
o
r
pa
nd
e
m
ic
m
it
i
ga
ti
on
.
A
s
e
r
i
e
s
of
tes
ts
,
in
c
l
ud
in
g
a
nt
ig
e
n
tes
t
in
g
,
r
e
a
l
-
t
im
e
po
ly
me
r
a
s
e
c
ha
i
n
r
e
a
c
ti
on
,
th
e
M
a
nt
ou
x
tu
be
r
c
ul
i
n
s
k
in
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
309
9
-
3108
3100
t
e
s
t
,
a
nd
a
c
o
mp
le
te
b
lo
od
c
o
un
t
(
C
B
C
)
lev
e
l
a
r
e
r
e
c
om
me
nd
e
d
f
o
r
di
a
g
nos
is
.
T
he
s
e
m
e
t
ho
ds
a
r
e
ti
me
-
i
n
te
ns
i
ve
a
nd
h
a
v
e
li
m
it
a
t
io
ns
s
uc
h
as
a
20
%
r
a
te
of
i
na
c
c
u
r
a
c
ies
a
nd
an
80
%
l
e
v
e
l
of
a
da
p
ta
bi
l
it
y
[
3
]
.
E
a
r
l
y
id
e
nt
i
f
ica
t
io
n
of
t
he
s
e
d
is
e
a
s
e
s
g
r
e
a
t
ly
e
n
ha
nc
e
s
t
he
c
ha
nc
e
s
of
s
ur
v
iva
l
a
nd
r
e
d
uc
e
s
f
a
t
a
l
it
ies
[
4
]
.
A
c
c
u
r
a
t
e
d
ia
gno
s
is
a
n
d
c
las
s
i
f
i
c
a
ti
on
of
c
he
s
t
d
is
e
a
s
e
s
a
r
e
e
s
s
e
nt
ia
l
f
o
r
e
f
f
e
c
t
iv
e
t
r
e
a
t
men
t
a
nd
m
a
n
a
ge
me
nt
.
R
e
c
e
nt
ly
,
tec
hno
l
og
ica
l
a
dv
a
nc
e
m
e
n
ts
,
i
nc
lu
di
ng
de
e
p
lea
r
ni
ng
a
n
d
a
r
t
if
i
c
i
a
l
in
te
l
li
ge
n
c
e
a
l
go
r
i
t
hms
,
ha
ve
p
r
ov
e
n
be
n
e
f
i
c
i
a
l
in
d
ia
gn
os
in
g
a
nd
c
las
s
i
f
y
in
g
c
he
s
t
d
is
e
a
s
e
s
th
r
oug
h
r
a
d
io
g
r
a
p
h
ic
i
mag
i
ng
l
ik
e
c
he
s
t
X
-
r
a
ys
or
c
om
pu
te
d
t
o
mo
g
r
a
p
hy
(
CT
)
s
c
a
ns
.
T
h
e
s
e
tec
hn
o
lo
gi
e
s
h
e
l
p
h
e
a
l
thc
a
r
e
p
r
o
f
e
s
s
io
na
ls
q
ui
c
k
ly
a
n
d
a
c
c
u
r
a
t
e
l
y
i
d
e
n
ti
f
y
v
a
r
i
ous
c
he
s
t
c
on
di
t
io
ns
,
r
e
s
u
l
ti
ng
in
t
im
e
l
y
i
nt
e
r
ve
n
t
io
ns
a
nd
be
t
te
r
pa
ti
e
n
t
o
ut
c
o
mes
.
M
or
e
ove
r
,
f
e
a
tur
e
s
e
lec
ti
on
is
a
n
e
s
s
e
nti
a
l
tas
k
in
f
e
e
ding
knowle
dge
;
it
is
us
ua
ll
y
c
a
r
r
ied
ou
t
du
r
ing
the
da
ta
pr
e
pa
r
a
ti
on
s
tage
[
5
]
.
T
he
objec
ti
ve
is
to
identif
y
the
mos
t
s
uit
a
ble
c
oll
e
c
ti
on
of
f
e
a
t
ur
e
s
that
e
nc
o
mpas
s
im
por
tant
f
e
a
tur
e
s
of
the
da
tas
e
t
whil
e
e
li
mi
na
ti
ng
unne
c
e
s
s
a
r
y
or
r
e
dunda
nt
f
e
a
tu
r
e
s
t
ha
t
may
ha
ve
ne
ga
ti
ve
im
pa
c
ts
on
the
a
c
c
ur
a
c
y
of
the
c
las
s
if
ica
ti
on
a
nd
the
e
f
f
ica
c
y
of
a
c
las
s
if
ier
.
T
his
is
c
r
uc
ial
f
or
va
li
da
ti
ng
the
a
c
c
ur
a
c
y
of
the
pr
opos
e
d
a
lgor
it
hm
s
[
6]
.
T
he
s
ugge
s
ted
tec
hnique
uti
li
z
e
s
f
ir
e
f
l
ies
f
il
ter
ing
to
opti
mi
z
e
f
e
a
tur
e
s
e
lec
ti
on
in
mac
hine
lea
r
ning.
T
his
f
il
ter
ing
s
tr
a
tegy
f
a
c
il
it
a
tes
the
de
tec
ti
on
of
a
p
pr
opr
iate
f
e
a
tur
e
s
while
mi
nim
izing
the
pr
oba
b
il
it
y
of
f
a
ls
e
pos
it
ives
[
7]
.
Nume
r
ous
meta
he
ur
is
ti
c
a
lgor
it
hms
,
pa
r
ti
c
ular
ly
thos
e
ins
pir
e
d
by
na
tur
e
,
s
uc
h
a
s
in
t
e
ll
igenc
e
f
r
om
s
wa
r
ms
a
nd
e
volut
ionar
y
a
lgo
r
it
h
ms
,
ha
ve
s
hown
im
pr
e
s
s
ive
e
f
f
e
c
ti
ve
ne
s
s
in
a
ddr
e
s
s
ing
th
e
dif
f
icult
ies
in
f
e
a
tur
e
s
e
lec
ti
on
[
8]
.
How
e
v
e
r
,
ther
e
is
potential
f
or
f
u
r
ther
a
dva
nc
e
ments
in
c
las
s
if
ying
t
he
dis
e
a
s
e
,
e
ns
ur
ing
pr
e
c
is
ion,
im
pr
oving
tr
a
ini
ng
dur
a
ti
on,
a
nd
incr
e
a
s
ing
the
number
of
pa
r
a
mete
r
s
.
S
e
ve
r
a
l
e
s
tablis
he
d
meta
he
ur
is
ti
c
methods
e
n
c
ounter
s
t
a
gna
ti
on
withi
n
s
ub
-
opti
mal
domains
[
9]
,
[
10]
.
Additi
on
a
ll
y,
it
is
c
ha
ll
e
nging
to
s
c
r
e
e
n
f
or
s
ys
temic
pulm
ona
r
y
a
il
ments
owing
to
a
lac
k
of
tes
t
p
r
oc
e
dur
e
s
a
nd
li
mi
ted
hos
pit
a
l
r
e
s
our
c
e
s
.
C
ons
e
que
ntl
y,
a
utom
a
ti
on
in
the
f
ield
of
medic
a
l
im
a
ging
in
lung
dis
e
a
s
e
c
las
s
if
ica
t
ion
c
a
n
e
nha
nc
e
e
a
r
ly
diagnos
is
.
F
ur
ther
mor
e
,
the
us
e
of
de
e
p
lea
r
ning
tec
hniques
li
ke
c
onvolut
ional
ne
ur
a
l
ne
two
r
ks
(
C
NN
s
)
ha
s
pr
ove
n
s
uc
c
e
s
s
f
ul
in
e
xt
r
a
c
ti
ng
f
e
a
tur
e
s
f
r
om
c
he
s
t
X
-
r
a
y
im
a
ge
s
,
a
idi
ng
in
the
quick
a
nd
a
c
c
ur
a
te
d
e
tec
ti
on
of
lung
dis
e
a
s
e
s
.
How
e
ve
r
,
ther
e
is
a
r
e
s
e
a
r
c
h
ga
p
in
e
xplor
ing
mor
e
a
dva
nc
e
d
f
e
a
tur
e
s
e
lec
ti
on
m
e
thods
to
f
ur
ther
i
mpr
ove
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
a
nd
opti
mi
z
e
the
pe
r
f
or
manc
e
of
de
e
p
lea
r
ning
models
in
dia
gnos
ing
lung
dis
e
a
s
e
s
us
ing
c
he
s
t
X
-
r
a
y
im
a
ge
s
.
T
his
s
tudy
inves
ti
ga
ted
the
e
f
f
e
c
ts
of
us
ing
meta
he
ur
is
ti
c
a
l
gor
it
hms
to
opti
mi
z
e
f
e
a
tur
e
s
e
lec
ti
on
in
de
e
p
lea
r
ning
m
ode
ls
,
s
pe
c
if
ica
ll
y
f
oc
us
ing
on
the
f
ir
e
f
ly
na
tur
e
-
ins
pir
e
d
opti
mi
z
a
ti
on
a
lgor
it
hm
.
W
hil
e
e
a
r
li
e
r
s
tudi
e
s
h
a
ve
e
xplor
e
d
the
im
pa
c
t
of
t
r
a
dit
ional
f
e
a
tu
r
e
s
e
lec
ti
on
methods
,
they
ha
ve
not
e
xpli
c
it
ly
a
ddr
e
s
s
e
d
the
ir
inf
luenc
e
on
the
pe
r
f
or
manc
e
a
nd
e
f
f
icie
nc
y
of
de
e
p
lea
r
ning
models
whe
n
opti
m
ize
d
us
ing
meta
he
ur
is
ti
c
a
lgor
it
h
ms
.
T
he
f
ir
e
f
ly
a
lgor
it
hm
,
known
f
o
r
its
s
im
pli
c
it
y
a
nd
a
bil
it
y
to
a
void
loca
l
opti
ma
,
of
f
e
r
s
a
dva
ntage
s
s
uc
h
as
e
nha
n
c
e
d
e
xplor
a
ti
on
of
th
e
s
e
a
r
c
h
s
pa
c
e
,
e
f
f
icie
nt
ha
ndli
ng
of
high
-
dim
e
ns
ional
da
ta,
a
nd
r
obus
t
pe
r
f
or
manc
e
in
nois
y
e
nvir
onments
.
T
his
s
tudy
int
r
oduc
e
s
an
opti
mi
z
e
d
C
NN
model
f
or
c
las
s
if
ying
lung
dis
e
a
s
e
s
f
r
om
c
he
s
t
X
-
r
a
ys
int
o
f
our
c
a
tegor
ies
:
C
OV
I
D
-
19,
vir
a
l
pne
umoni
a
,
lung
opa
c
it
y,
a
nd
non
-
inf
e
c
ti
ous
nor
mal.
F
ur
the
r
mor
e
,
the
f
ir
e
f
ly
a
lgor
it
hm
wa
s
us
e
d
to
im
pr
ove
f
e
a
tur
e
s
e
lec
ti
on
e
f
f
icie
nc
y
f
o
r
lung
a
bno
r
malit
ies
.
2.
L
I
T
E
RA
T
UR
E
RE
VI
E
W
T
he
r
e
s
e
a
r
c
h
pr
e
s
e
nts
a
method
f
o
r
diagnos
ing
pne
umoni
a
us
ing
X
-
r
a
y
im
a
ge
s
.
T
his
method
c
ombi
ne
s
C
NN
a
nd
tr
a
ns
f
e
r
lea
r
ning
tec
hniques
.
S
pe
c
if
ica
ll
y,
it
a
dopts
a
pr
e
vious
ly
tr
a
ined
M
obil
e
Ne
t
model
a
nd
incor
por
a
tes
a
ddit
ional
laye
r
s
f
or
f
ine
-
tuni
ng.
Nota
bly,
the
pr
opos
e
d
s
tr
a
tegy
e
xhibi
ted
s
upe
r
ior
pe
r
f
or
manc
e
in
c
ompar
is
on
to
other
C
NN
mod
e
ls
uti
li
z
ing
c
onve
nti
ona
l
mac
hine
lea
r
ning
tec
hniques
,
ther
e
by
de
mons
tr
a
ti
ng
it
s
e
f
f
e
c
ti
ve
ne
s
s
in
diagnos
ing
pne
umoni
a
[
11]
.
Additi
ona
ll
y
,
de
e
p
lea
r
ning
c
a
n
c
r
e
a
te
models
that
c
a
n
e
f
f
e
c
ti
ve
ly
f
or
e
c
a
s
t
a
nd
de
tec
t
c
e
r
tain
dis
e
a
s
e
s
,
including
C
OV
I
D
-
19,
us
ing
im
a
ge
s
[
12]
.
T
his
tec
hnology
is
pa
r
ti
c
ular
ly
e
f
f
e
c
ti
ve
in
a
c
hie
ving
pr
e
c
is
e
r
e
s
ult
s
in
medic
a
l
diagnos
is
.
F
or
in
s
tanc
e
,
a
c
a
tegor
iza
ti
on
s
ys
tem
f
or
C
OV
I
D
-
19,
dif
f
e
r
e
nt
f
o
r
ms
of
pne
umoni
a
,
tuber
c
ulos
is
,
a
nd
nor
mal
X
-
r
a
y
im
a
ge
s
ha
s
be
e
n
s
ugge
s
ted
[
13]
.
M
or
e
ove
r
,
Kha
s
a
wne
h
e
t
al.
[
14
]
p
r
opos
e
d
the
us
e
of
a
utom
a
ti
on
f
or
id
e
nti
f
ying
C
OV
I
D
-
19
dis
e
a
s
e
f
r
om
c
he
s
t
X
-
r
a
ys
us
ing
de
e
p
l
e
a
r
ning
a
lgor
it
hms
.
T
his
s
tudy
e
xa
mi
ne
s
a
pplyi
ng
de
e
p
lea
r
ning
a
ppr
oa
c
he
s
to
de
tec
t
pne
umoni
a
in
c
he
s
t
X
-
r
a
y
im
a
ge
s
.
T
he
a
uthor
d
is
c
us
s
e
d
the
c
ha
ll
e
nge
s
a
nd
li
mi
tations
of
a
nnotate
d
da
tas
e
ts
,
inequa
li
ti
e
s
in
c
las
s
e
s
,
c
ompr
e
he
ns
ion
of
model
p
r
e
dictions
,
ge
ne
r
a
li
z
a
ti
on,
a
nd
ins
e
r
ti
on
int
o
c
li
nica
l
p
r
oc
e
dur
e
s
.
F
ur
ther
mor
e
,
the
li
ter
a
tur
e
r
e
view
e
xa
mi
ne
s
C
NN
-
ba
s
e
d
f
r
a
mew
or
ks
,
t
r
a
ns
f
e
r
lea
r
ning,
da
tas
e
t
de
ve
lopm
e
nt
,
a
nd
da
ta
pr
e
pr
oc
e
s
s
ing
methods
,
s
uc
h
a
s
im
a
ge
s
c
a
li
ng,
a
n
d
nor
maliza
ti
on,
to
im
pr
ove
model
pe
r
f
o
r
manc
e
[
15]
.
F
o
r
e
xa
mpl
e
,
Oz
yur
t
e
t
al.
[
16
]
im
pleme
nted
VGG
-
16,
De
ns
e
Ne
t
-
169,
a
nd
De
ns
e
Ne
t
-
201
to
diagnos
e
the
c
he
s
t
X
-
r
a
y
im
a
ge
s
.
Dur
ing
the
f
e
a
tur
e
e
xtr
a
c
ti
on
s
tag
e
,
they
identif
ied
the
im
po
r
tant
c
ha
r
a
c
ter
is
ti
c
s
us
ing
the
s
c
a
le
-
invar
iant
method,
a
nd
the
r
e
s
ult
s
we
r
e
e
n
ha
nc
e
d
by
us
ing
the
binar
y
-
r
obus
t
invar
iant
s
c
a
lable
ke
y
point
s
.
T
he
y
de
ve
loped
a
dyna
mi
c
-
s
ize
d
pyr
a
mi
d
f
us
e
d
with
the
ba
s
e
model
f
o
r
f
e
a
tur
e
s
e
lec
ti
on
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
ff
icie
nt
lung
dis
e
as
e
c
las
s
if
ication
thr
ough
lumines
c
e
nt
featur
e
s
e
lec
ti
on
…
(
A
njugam
Shanmugav
e
lu)
3101
c
las
s
if
ica
ti
on
of
c
he
s
t
X
-
r
a
y
im
a
ge
s
a
c
c
or
ding
to
the
tar
ge
t
dis
e
a
s
e
,
a
c
hieving
a
n
a
c
c
ur
a
c
y
o
f
95.
84%
us
ing
thi
s
f
e
a
tur
e
ge
ne
r
a
ti
on
ne
two
r
k
[
17
]
.
M
a
c
h
in
e
l
e
a
r
n
in
g
e
nc
ou
nt
e
r
s
n
ot
a
b
le
di
f
f
ic
ul
ti
e
s
in
f
e
a
t
ur
e
s
e
le
c
t
io
n;
t
his
i
nc
lu
de
s
c
ho
os
in
g
a
s
ubs
e
t
of
c
om
p
li
me
nta
r
y
a
tt
r
ib
ut
e
s
f
r
o
m
c
ol
lec
t
io
n
of
c
ha
r
a
c
te
r
is
t
ics
to
e
n
ha
nc
e
c
l
a
s
s
i
f
ica
t
io
n
a
c
c
u
r
a
c
y
.
T
h
e
t
r
a
i
t
'
c
o
mp
le
me
nt
a
r
y
'
is
e
s
s
e
nt
ia
l
d
ue
to
t
he
po
te
nt
ia
l
f
o
r
s
e
ve
r
a
l
in
te
r
a
c
t
io
ns
be
twe
e
n
f
e
a
t
u
r
e
s
,
wh
ic
h
b
e
c
om
e
s
mo
r
e
c
ha
ll
e
n
gi
ng
as
n
u
mbe
r
of
f
e
a
t
ur
e
s
i
nc
r
e
a
s
e
s
.
F
ur
t
he
r
m
o
r
e
,
the
di
f
f
i
c
u
l
ty
in
i
mp
le
me
nta
t
io
n
c
o
ul
d
be
a
t
tr
ib
u
ted
to
t
he
ne
e
d
f
o
r
a
c
h
ie
vi
ng
a
b
a
l
a
nc
e
d
c
las
s
i
f
ica
ti
on
a
c
c
u
r
a
c
y
a
nd
m
in
im
iz
in
g
f
a
ls
e
pos
it
iv
e
s
,
to
a
vo
id
i
nc
o
r
r
e
c
t
t
r
e
a
tm
e
n
t
r
e
c
om
me
nda
t
io
ns
.
F
e
a
tu
r
e
s
e
l
e
c
t
i
on
mi
n
i
mi
z
e
s
d
i
men
s
i
ona
l
it
y
of
th
e
i
np
ut
da
ta
by
e
li
m
i
na
ti
ng
e
xt
r
a
ne
ous
qu
a
l
it
ies
a
nd
no
is
e
,
th
us
ob
ta
in
i
ng
an
op
t
im
a
l
s
ubs
e
t
of
f
e
a
t
ur
e
s
.
T
he
r
e
a
r
e
t
hr
e
e
p
r
im
a
r
y
t
y
pe
s
of
f
e
a
t
ur
e
s
e
lec
t
io
n
a
p
p
r
oa
c
h
e
s
:
w
r
a
p
pe
r
-
b
a
s
e
d
te
c
h
ni
qu
e
,
f
i
lt
e
r
-
ba
s
e
d
me
th
od
,
a
nd
e
m
be
dd
in
g
m
e
t
ho
ds
[
18
]
.
M
e
tahe
ur
is
ti
c
s
a
lgor
it
hms
,
pa
r
ti
c
ular
ly
na
tu
r
e
-
ins
pir
e
d
tec
hniques
s
uc
h
a
s
e
volut
ionar
y
ge
ne
r
a
ted
a
lgor
it
hms
a
nd
s
wa
r
m
opti
mi
z
a
ti
on
,
a
r
e
ve
r
y
e
f
f
icie
nt
f
or
f
e
a
tur
e
s
e
lec
ti
on
be
c
a
us
e
they
c
a
n
e
f
f
e
c
ti
ve
ly
e
xplor
e
the
va
s
t
s
e
a
r
c
h
a
r
e
a
.
F
o
r
e
xa
mpl
e
,
the
ge
ne
ti
c
a
lgor
it
hm
(
GA
)
is
f
r
e
que
ntl
y
e
mpl
oye
d
a
s
a
wr
a
ppe
r
tec
hnique
in
thes
e
methodologi
e
s
[
19]
.
S
w
a
r
m
in
tel
l
ig
e
nc
e
is
a
r
o
bus
t
op
t
im
iz
a
t
io
n
me
th
od
th
a
t
is
e
f
f
e
c
t
ive
f
o
r
s
ol
vi
ng
a
c
t
ua
l
pr
ob
le
ms
t
ha
t
a
r
e
di
f
f
icu
l
t
to
s
o
l
ve
us
i
n
g
t
r
a
d
i
ti
on
a
l
a
lg
o
r
i
th
ms
[
20
]
.
T
h
is
is
a
c
hi
e
ve
d
b
y
im
i
ta
ti
ng
n
a
t
u
r
a
l
s
ys
te
ms
a
n
d
us
in
g
o
pe
r
a
ti
ons
th
a
t
inc
l
ude
bo
th
e
xp
l
oi
ti
ng
a
nd
e
x
pl
o
r
i
ng
the
is
s
u
e
s
pa
c
e
.
P
o
pu
la
r
i
n
te
ll
ig
e
n
t
s
ys
te
ms
ha
ve
be
e
n
d
e
ve
lo
pe
d
ba
s
e
d
o
n
a
nt
c
ol
on
y
-
i
ns
p
i
r
e
d
op
t
im
iza
t
io
n
te
c
h
ni
que
s
,
in
he
r
it
a
n
c
e
o
f
w
o
r
k
in
g
p
a
t
te
r
ns
o
f
b
e
e
c
ol
on
ies
,
ba
t
a
lg
o
r
i
th
m
,
a
nd
t
he
f
i
r
e
f
l
ies
a
l
go
r
it
h
m
(
F
A
)
[
17
]
,
[
1
8
]
,
[
2
1]
–
[
2
4
]
.
Gupta
e
t
al.
[
25]
pr
opos
e
d
the
F
A
meta
he
ur
i
s
ti
c
s
,
ins
pir
e
d
by
the
gr
e
ga
r
ious
a
nd
f
las
hing
c
ha
r
a
c
ter
is
ti
c
s
of
f
ir
e
f
li
e
s
.
F
A
e
mpl
oys
s
e
ve
r
a
l
a
p
pr
oxim
a
ti
on
r
ules
to
a
c
c
ur
a
tely
r
e
pr
e
s
e
nt
the
int
r
i
c
a
te
a
nd
a
dva
nc
e
d
na
tur
e
of
the
pr
a
c
ti
c
a
l
a
ppli
c
a
ti
ons
of
n
a
tur
a
l
s
ys
tems
.
T
he
f
it
ne
s
s
va
lue
is
obtaine
d
ba
s
e
d
on
the
lum
inos
it
y
of
the
br
ight
f
ly
a
nd
the
a
tt
r
a
c
ti
on
o
f
f
i
r
e
f
li
e
s
with
the
br
ight
e
r
f
ly;
the
a
tt
r
a
c
ti
on
in
mos
t
c
omm
on
FA
im
pleme
ntations
is
c
onti
nge
nt
upon
the
lum
in
os
it
y,
whos
e
va
lue
is
de
ter
mi
ne
d
by
the
obj
e
c
ti
ve
f
unc
ti
on.
T
he
a
ppr
oa
c
h
in
thi
s
s
tudy
is
s
uit
a
ble
f
or
mi
nim
i
z
a
ti
on
s
c
e
na
r
ios
.
Additi
ona
ll
y,
a
pplyi
ng
pr
e
-
tr
a
ined
ne
ur
a
l
ne
twor
ks
li
ke
VG
G
-
16,
C
a
ps
Ne
t
,
M
obil
e
Ne
t,
I
nc
e
pti
onV3,
a
nd
De
ns
e
Ne
t
f
or
a
na
lyzing
lung
im
a
ge
s
,
e
s
pe
c
ially
in
c
a
s
e
s
s
uc
h
a
s
lung
in
f
e
c
ti
ons
a
nd
C
OV
I
D
-
19,
is
pr
omi
s
ing
[
26]
.
P
ne
umoni
a
is
one
o
f
the
ke
y
s
ympt
oms
of
C
OV
I
D
-
19
dis
e
a
s
e
.
T
r
a
ns
f
e
r
lea
r
n
in
g
f
a
c
il
it
a
tes
the
identif
ica
ti
on
of
a
s
ha
r
e
d
c
a
us
a
ti
ve
a
ge
nt
f
or
both
pne
umoni
a
a
nd
C
OV
I
D
-
19
dis
e
a
s
e
a
n
d
it
s
va
r
iants
.
T
his
s
tudy
pr
e
s
e
nts
e
s
s
e
nti
a
l
knowle
dge
obtaine
d
us
ing
a
tr
a
ined
model
in
dis
ti
nguis
hing
be
twe
e
n
vir
a
l
pne
umoni
a
,
tuber
c
ulos
is
,
a
nd
C
OV
I
D
-
19.
3.
P
ROP
OS
E
D
M
E
T
HO
DOL
OG
Y
3.
1
.
Dat
as
e
t
d
e
s
c
r
ip
t
ion
T
he
C
OV
I
D
-
19
r
a
diogr
a
phy
da
taba
s
e
is
a
c
omp
r
e
he
ns
ive
r
e
pos
it
or
y
of
c
he
s
t
X
-
r
a
y
im
a
ge
s
that
c
a
ptur
e
ins
tanc
e
s
of
C
OV
I
D
-
19,
nor
mal
c
ondit
ions
,
lung
opa
c
it
y,
a
nd
vir
a
l
pne
umoni
a
.
T
he
da
taba
s
e
e
xpa
nde
d
s
igni
f
ica
ntl
y
s
tar
ti
ng
with
an
ini
ti
a
l
s
e
t
of
219
im
a
ge
s
.
A
major
upda
te
a
dde
d
1,
200
C
OV
I
D
-
19
X
-
r
a
y
im
a
ge
s
.
T
he
s
e
c
ond
e
dit
ion,
r
e
lea
s
e
d
in
2021,
include
s
im
a
ge
s
of
3
,
616
C
OV
I
D
-
19
c
a
s
e
s
,
10,
192
nor
mal
c
a
s
e
s
,
6,
012
lung
opa
c
it
y
c
a
s
e
s
,
a
nd
1,
345
vi
r
a
l
pne
umoni
a
c
a
s
e
s
.
T
he
da
taba
s
e
is
e
x
pe
c
ted
to
be
c
onti
nuous
ly
upda
ted
with
ne
w
X
-
r
a
y
im
a
ge
s
,
pa
r
ti
c
ular
ly
thos
e
of
C
OV
I
D
-
19
pa
ti
e
nts
with
ne
w
va
r
iants
a
nd
pne
umoni
a
[2
7]
,
[
28]
.
T
he
X
-
r
a
y
im
a
ge
s
a
nd
mas
ks
us
e
d
in
thi
s
s
tudy
a
r
e
s
hown
in
F
igur
e
s
1
(
a
)
to
1(
d)
.
(
a
)
(
b)
(
c
)
(
d)
F
igur
e
1.
L
ung
dis
e
a
s
e
types
of
(a
)
nor
mal
,
(
b
)
C
O
VI
D
-
19,
(
c
)
pne
umon
ia,
a
nd
(
d
)
lung
opa
c
it
y
with
their
r
e
leva
nt
mas
k
T
he
methodology
s
e
c
ti
on
outl
ines
the
de
e
p
lea
r
ni
ng
a
ppr
oa
c
h
e
mpl
oye
d
f
or
e
f
f
icie
ntl
y
c
a
tegor
izing
c
he
s
t
X
-
r
a
y
im
a
ge
s
us
ing
pr
e
-
tr
a
ined
C
NN
s
.
P
r
e
-
t
r
a
ined
models
tr
a
ined
on
s
ubs
tantial
da
tas
e
ts
a
r
e
a
va
il
a
ble
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
309
9
-
3108
3102
f
or
r
e
late
d
tas
ks
a
nd
ha
ve
a
lr
e
a
dy
lea
r
ne
d
d
iver
s
e
f
e
a
tur
e
s
a
nd
pa
tt
e
r
ns
.
T
his
s
tudy
im
p
leme
nted
a
VGG
-
16
pr
e
-
tr
a
ined
model.
T
he
e
xpa
nde
d
da
tas
e
t
include
s
lung
im
a
ge
s
a
nd
thei
r
c
or
r
e
s
ponding
mas
ks
to
e
nha
nc
e
the
pe
r
f
or
manc
e
of
the
model
.
T
he
FA
wa
s
uti
l
ize
d
to
s
e
lec
t
a
li
mi
ted
s
e
t
of
f
e
a
tur
e
s
,
f
oll
owe
d
by
the
de
p
loym
e
nt
of
a
t
r
a
ns
f
e
r
lea
r
ning
model.
T
h
is
methodology
e
ns
ur
e
s
a
higher
a
c
c
ur
a
c
y
in
c
las
s
if
ying
lung
dis
e
a
s
e
s
,
pa
r
ti
c
ular
ly
in
the
c
ontext
of
C
OV
I
D
-
19
diagnos
is
.
3.
2
.
F
ire
f
li
e
s
algorit
h
m
T
he
f
ir
e
f
ly
a
lgor
it
hm
is
a
f
ly
-
ins
pir
e
d
na
tur
e
a
lg
or
it
hm
that
r
e
pli
c
a
tes
the
be
ha
vior
of
f
ir
e
f
li
e
s
at
night
,
s
pe
c
if
ica
ll
y
thei
r
f
las
hing
pa
tt
e
r
n,
to
de
ter
mi
ne
the
a
tt
r
a
c
ti
ve
ne
s
s
of
the
br
ight
e
r
f
i
r
e
f
ly.
T
he
thr
e
e
r
ules
a
r
e
de
r
ived
f
r
om
the
be
ha
vior
a
l
pa
tt
e
r
ns
e
xhibi
ted
by
f
ir
e
f
li
e
s
[
29]
–
[
31
]
.
T
he
r
ules
e
t
f
o
r
the
f
e
a
tur
e
s
e
lec
ti
on
with
FA
is
r
e
pr
e
s
e
nted
in
F
igur
e
2.
T
he
im
pleme
n
ted
r
e
s
u
lts
a
r
e
s
hown
in
F
igu
r
e
3
ha
s
an
it
e
r
a
ti
on
of
50
f
or
va
lue
n=
4
a
nd
F
igur
e
4
r
e
p
r
e
s
e
nts
the
r
e
c
ur
s
ive
s
tr
ide
to
c
a
ptur
e
the
f
e
a
tur
e
f
or
one
it
e
r
a
ti
on
.
T
h
e
logi
c
a
l
s
teps
of
the
FA
a
r
e
as
f
oll
ows
:
‒
R
ule
1
s
tate
s
that
∀
f
ir
e
f
li
e
s
,
r
e
ga
r
dles
s
of
ge
nde
r
,
a
r
e
a
tt
r
a
c
ted
to
each
other
ba
s
e
d
on
their
br
ight
ne
s
s
,
whic
h
is
ti
e
d
to
an
objec
ti
ve
f
unc
t
ion
B
=
f
(
O
).
‒
W
he
n
two
f
ir
e
f
li
e
s
(
F1
a
nd
F2
)
a
r
e
p
r
e
s
e
nt,
the
moveme
nt
of
F1
towa
r
ds
F2
is
de
ter
mi
ne
d
by
th
e
ir
br
ight
ne
s
s
dif
f
e
r
e
nc
e
,
m
ove
ment
F1
→
F2
=
α
(
B
F2
−
B
F1
),
whe
r
e
α
is
a
c
ons
tant
f
a
c
tor
.
‒
F
ir
e
f
li
e
s
with
lowe
r
lum
inos
it
y
move
towa
r
ds
f
ir
e
f
li
e
s
wi
th
highe
r
lum
inos
it
y
,
M
ove
ment
low
→
high
=
β
⋅
(
B
h
i
g
h
−
B
l
o
w
),
whe
r
e
β
is
a
nother
c
ons
tant
f
a
c
tor
.
‒
T
he
leve
l
of
a
tt
r
a
c
ti
ve
ne
s
s
(
A)
is
dir
e
c
tl
y
c
o
r
r
e
late
d
to
the
br
ight
ne
s
s
of
the
f
i
r
e
f
ly
(
B
)
:
A=
B
.
‒
Obs
e
r
va
ti
ona
ll
y,
if
f
i
r
e
f
ly
F1
is
les
s
br
ight
than
F2
(B
F1
<B
F
2
),
both
f
ir
e
f
li
e
s
be
c
ome
les
s
br
ight
as
t
he
ir
dis
tanc
e
(
D)
incr
e
a
s
e
s
f
r
om
the
obs
e
r
ve
r
,
B
F1
=B
F1
⋅
e−α
⋅
D,
B
F2
=B
F2
⋅
e−α
⋅
D
.
‒
If
no
f
i
r
e
f
li
e
s
e
mi
t
s
tr
onge
r
li
gh
t,
a
f
ir
e
f
ly
mov
e
s
r
a
ndoml
y
,
M
ove
ment
=
γ
⋅
r
a
ndom(
)
,
whe
r
e
γ
is
a
c
ons
tant
f
a
c
tor
.
‒
R
ule
2
s
tate
s
that
the
f
ir
e
f
ly
br
ight
ne
s
s
is
inf
luen
c
e
d
by
the
e
nc
ode
d
objec
ti
ve
f
unc
ti
on,
guidi
ng
t
he
ir
int
e
r
a
c
ti
ons
.
‒
R
ule
3
s
tate
s
that
a
tt
r
a
c
ti
ve
ne
s
s
is
dir
e
c
tl
y
li
nke
d
to
the
br
ight
ne
s
s
.
F
o
r
ins
tanc
e
,
as
s
hown
in
F
igu
r
e
2,
f
ir
e
f
ly
F1
is
les
s
br
ight
than
F
2
,
a
nd
both
de
c
r
e
a
s
e
in
br
ight
ne
s
s
with
dis
tanc
e
f
r
om
the
obs
e
r
ve
r
.
If
no
br
ight
e
r
f
ir
e
f
li
e
s
a
r
e
ne
a
r
by,
r
a
ndom
moveme
nt
e
n
s
ue
s
.
‒
T
he
a
lgor
it
hm
pr
opos
e
s
a
method
f
o
r
f
e
a
tur
e
s
e
lec
ti
on
ba
s
e
d
on
the
int
e
ns
it
y
of
li
ght
I
(
r
)
a
dhe
r
e
s
to
the
inver
s
e
s
qua
r
e
r
ule.
I
0
de
notes
the
int
e
ns
it
y
of
l
ight
at
the
s
our
c
e
as
s
hown
in
(
1
)
.
(
)
=
0
2
(
1)
‒
T
he
inver
s
e
s
qua
r
e
law
can
be
tr
a
ns
f
or
med
int
o
Ga
us
s
ian
f
or
m
f
o
r
e
s
ti
mating
a
bs
or
pti
on
:
(
)
=
0
−
2
(
2)
w
he
r
e
γ
is
a
c
ons
tant.
‒
T
he
va
r
iation
in
the
f
or
c
e
of
a
tt
r
a
c
ti
on
may
be
c
h
a
r
a
c
ter
ize
d
as
the
inver
s
e
s
qua
r
e
r
e
lations
hip
be
tw
e
e
n
the
int
e
ns
it
y
of
li
ght
I
a
nd
the
dis
tanc
e
r
(
r
a
ndom
mot
ion)
.
(
)
=
0
−
2
(
3)
‒
I
n
(
4
)
upda
tes
the
a
tt
r
a
c
ti
ve
ne
s
s
pos
it
ion
X
i
ba
s
e
d
on
the
a
tt
r
a
c
ti
ve
ne
s
s
of
ne
ighbor
ing
f
i
r
e
f
li
e
s
a
nd
r
a
ndom
moveme
nt
,
+
1
=
+
0
−
2
(
−
)
+
+
∈
(
4)
T
he
a
lgor
it
hm
wa
s
im
pleme
nted
s
tepw
is
e
in
the
da
tas
e
t
a
f
ter
pr
e
-
pr
oc
e
s
s
ing
the
number
of
c
las
s
e
s
f
or
c
las
s
if
ica
ti
on,
with
the
s
e
lec
ti
on
of
f
li
e
s
ini
ti
a
t
e
d
a
t
n=
4
.
T
he
nu
mber
of
f
ir
e
f
li
e
s
wa
s
f
ixed
a
t
50
a
nd
the
maximum
it
e
r
a
ti
on
wa
s
f
ixed
a
t
100.
T
he
gr
a
p
h
s
r
e
pr
e
s
e
nt
the
maximum
f
e
a
tur
e
s
of
50
it
e
r
a
ti
ons
,
a
nd
f
r
e
que
nc
y
is
a
pos
it
ive
f
e
a
tur
e
o
f
the
model
that
pr
oduc
e
s
a
c
c
ur
a
c
y.
T
he
f
e
a
tur
e
s
ubs
e
t
a
r
r
a
y
wa
s
pa
s
s
e
d
int
o
the
C
NN
to
c
las
s
if
y
the
dis
e
a
s
e
s
.
T
he
F
A
a
lgor
it
hm,
pr
opos
e
d
in
[
32]
,
is
ins
pir
e
d
by
the
lum
in
ous
a
nd
gr
e
ga
r
ious
be
ha
vior
o
f
f
ir
e
f
li
e
s
.
F
A
e
mpl
oys
many
a
ppr
oxim
a
ti
on
pr
inciples
to
s
im
ulate
c
o
mpl
e
x
a
nd
s
ophis
ti
c
a
ted
r
e
a
l
-
wor
ld
biol
ogica
l
s
ys
tems
.
F
ir
e
f
l
y
br
ight
ne
s
s
a
nd
a
tt
r
a
c
ti
ve
ne
s
s
a
r
e
us
e
d
to
r
e
pr
e
s
e
n
t
f
it
ne
s
s
f
unc
ti
ons
in
a
wa
y
that,
in
mos
t
c
omm
on
F
A
i
mpl
e
menta
ti
ons
,
a
tt
r
a
c
ti
ve
ne
s
s
is
de
p
e
nde
nt
on
b
r
ight
ne
s
s
,
whic
h
is
de
ter
mi
ne
d
by
the
va
lue
of
the
f
unc
ti
on
with
the
objec
ti
ve
.
F
o
r
m
ini
mi
z
a
ti
on
is
s
ue
s
,
the
f
or
mul
a
is
e
xpr
e
s
s
e
d
a
s
de
s
c
r
ibed
by
[
32]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
ff
icie
nt
lung
dis
e
as
e
c
las
s
if
ication
thr
ough
lumines
c
e
nt
featur
e
s
e
lec
ti
on
…
(
A
njugam
Shanmugav
e
lu)
3103
F
igur
e
2.
Dia
gr
a
m
il
lus
tr
a
ti
ng
wo
r
kf
low
of
the
FA
F
igur
e
3
.
T
he
pr
opos
e
d
f
i
r
e
f
ly
a
lgo
r
it
hm
a
nd
its
s
e
l
e
c
ted
f
e
a
tur
e
s
with
50
it
e
r
a
ti
ons
F
igur
e
4.
T
he
f
r
e
que
nc
y
de
ns
it
y
g
r
a
ph
of
the
f
ir
e
f
l
y
a
lgor
it
hm
s
hows
pos
it
ive
f
e
a
tur
e
a
tt
a
inm
e
nt
3.
3
.
P
r
op
os
e
d
c
las
s
if
icat
io
n
m
o
d
e
l
F
or
c
las
s
if
ica
ti
on,
the
p
r
opos
e
d
model
e
mpl
oye
d
a
c
ombi
na
ti
on
of
a
pr
e
-
tr
a
ined
VGG
-
16
model
a
nd
a
C
NN
with
f
ull
y
int
e
r
c
onne
c
ted
laye
r
s
.
T
he
X
-
r
a
y
im
a
ge
s
us
e
d
to
de
ter
mi
ne
the
pr
e
s
e
nc
e
of
s
pe
c
if
ic
types
of
c
he
s
t
dis
e
a
s
e
s
we
r
e
a
na
lyze
d
us
ing
the
VGG
-
16
model
[
20]
,
f
oll
owe
d
by
the
C
NN
model
.
F
igur
e
5
pr
ovides
a
c
ompr
e
he
ns
ive
de
piction
of
the
model
a
r
c
hit
e
c
tur
e
a
nd
T
a
ble
1
de
tails
the
im
pleme
ntation
of
the
pr
opos
e
d
s
ys
tem.
Dur
ing
the
f
e
a
tur
e
e
xt
r
a
c
ti
on
s
tage
,
d
im
e
ns
ionalit
y
r
e
duc
ti
on
wa
s
a
c
hieve
d
us
ing
th
e
f
ir
e
f
ly
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
309
9
-
3108
3104
a
lgor
it
hm,
a
f
te
r
whic
h
the
VGG
-
16
model
wa
s
int
e
gr
a
ted
with
a
ddit
ional
C
NN
blocks
.
T
he
VGG
-
16
model,
r
e
nowne
d
f
or
its
e
xc
e
pti
ona
l
pr
e
c
is
ion
in
im
a
ge
-
r
e
late
d
tas
ks
,
s
e
r
ve
d
as
the
ba
c
kbone
of
the
a
r
c
hit
e
c
tur
e
.
It
c
ons
is
ted
of
16
c
onvolut
ional
laye
r
s
,
each
c
onf
ig
ur
e
d
with
3
×
3
c
onvolut
ion
f
il
ter
s
a
nd
a
s
tr
ide
of
1.
F
igur
e
5.
A
p
r
opos
e
d
a
ppr
oa
c
h
f
o
r
f
ou
r
-
c
las
s
c
las
s
if
ica
ti
on
methodology
T
a
ble
1.
De
tailed
ove
r
view
of
the
pr
opos
e
d
s
ys
tem
im
pleme
ntation
S
ta
ge
D
e
s
c
r
ip
ti
on
L
a
ngua
ge
s
/F
r
a
me
w
or
ks
D
a
ta
a
c
qui
s
it
io
n
X
-
r
a
y
im
a
ge
s
r
e
s
iz
e
d
to
128×
128
×
3,
s
ta
nd
a
r
di
z
e
d
,
a
nd
nor
ma
li
z
e
d
P
yt
hon,
O
pe
nC
V
F
e
a
tu
r
e
e
xt
r
a
c
ti
on
F
ir
e
f
ly
a
lg
or
it
hm
f
or
di
me
ns
io
na
li
ty
r
e
duc
ti
on
P
yt
hon,
S
c
iP
y
P
r
e
-
tr
a
in
e
d
mode
l
VGG
-
16
w
it
h
16
c
onvolut
io
na
l
la
ye
r
s
(
3×
3
f
il
te
r
s
,
s
tr
id
e
of
1)
T
e
ns
or
F
lo
w
,
K
e
r
a
s
C
N
N
a
r
c
hi
te
c
tu
r
e
A
ddi
ti
ona
l
C
N
N
la
ye
r
s
w
it
h
R
e
L
U
,
ba
tc
h nor
ma
li
z
a
ti
on, ma
x
-
pool
in
g, dr
opout
T
e
ns
or
F
lo
w
,
K
e
r
a
s
C
la
s
s
if
ic
a
ti
on
F
ul
ly
c
onne
c
te
d
la
ye
r
s
w
it
h
S
of
tM
a
x
f
or
f
our
-
c
la
s
s
c
a
te
gor
iz
a
ti
on
T
e
ns
or
F
lo
w
,
K
e
r
a
s
T
r
a
in
in
g
a
nd
E
va
lu
a
ti
on
C
r
os
s
-
e
nt
r
opy
lo
s
s
,
A
da
m
opt
im
iz
e
r
,
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
,
F1
-
s
c
or
e
P
yt
hon,
T
e
ns
or
F
lo
w
D
a
ta
s
e
t
s
pl
it
ti
ng
T
r
a
in
(
70%
)
,
va
li
da
ti
on
(
15%
)
,
te
s
t
(
15%
)
P
yt
hon,
P
a
nda
s
H
a
r
dw
a
r
e
N
V
I
D
I
A
G
P
U
s
f
or
a
c
c
e
le
r
a
te
d
tr
a
in
in
g
-
S
of
twa
r
e
T
e
ns
or
F
lo
w
,
K
e
r
a
s
,
N
umP
y,
O
pe
nC
V
,
M
a
tp
lo
tl
ib
P
yt
hon
3.
4
.
L
aye
r
c
on
f
igu
r
a
t
ion
A
de
e
p
lea
r
ning
f
r
a
mew
or
k
wa
s
e
mpl
oye
d
f
or
the
mul
ti
-
c
las
s
if
ica
ti
on
of
lung
d
is
or
de
r
s
.
T
he
de
s
ign
im
pleme
nted
a
f
lattening
laye
r
in
the
ini
ti
a
l
pha
s
e
of
c
las
s
if
ica
ti
on,
f
oll
owe
d
by
s
ubs
e
que
nt
thi
c
k
laye
r
s
to
a
c
hieve
f
ur
ther
a
bs
tr
a
c
ti
on.
T
he
VGG
-
16
model
in
c
or
por
a
tes
a
f
lattening
laye
r
as
a
pivot
a
l
c
ompone
nt
in
the
ini
ti
a
l
s
tage
of
its
c
las
s
if
ica
ti
on
pr
oc
e
s
s
.
T
his
lay
e
r
is
e
s
s
e
nti
a
l
f
o
r
c
onve
r
ti
ng
the
mu
lt
i
-
dim
e
ns
ional
output
f
r
om
the
pr
e
c
e
ding
c
onvolut
ional
laye
r
s
int
o
a
o
ne
-
dim
e
ns
ional
a
r
r
a
y.
T
his
tr
a
ns
f
or
mation
is
c
r
it
i
c
a
l
as
it
pr
e
pa
r
e
s
the
e
xtr
a
c
ted
f
e
a
tur
e
s
f
or
s
ubs
e
que
nt
pr
oc
e
s
s
ing
in
the
de
ns
e
laye
r
s
.
I
n
the
VGG
-
16
a
r
c
hit
e
c
tur
e
,
the
f
lattening
laye
r
c
onne
c
ts
the
c
onvolut
ional
laye
r
s
r
e
s
pons
ibl
e
f
or
f
e
a
tur
e
e
xt
r
a
c
ti
on
to
the
f
ull
y
c
o
nne
c
ted
laye
r
s
de
dica
ted
to
c
las
s
if
ica
ti
on.
F
lattening
the
ou
tput
,
the
hier
a
r
c
hica
l
f
e
a
tur
e
s
lea
r
ne
d
by
the
c
onvo
lut
ional
laye
r
s
a
r
e
e
f
f
e
c
ti
ve
ly
f
e
d
int
o
the
de
ns
e
laye
r
s
,
e
ns
ur
ing
a
s
moot
h
t
r
a
ns
it
ion
f
or
a
c
c
ur
a
te
c
las
s
if
ica
ti
on.
F
oll
owing
the
f
lattening
laye
r
,
the
model
e
mp
loys
thr
e
e
de
ns
e
laye
r
s
wi
th
va
r
ying
ne
ur
on
c
ounts
of
5
12,
256
,
a
nd
128
ne
ur
ons
,
r
e
s
pe
c
ti
ve
ly.
T
he
s
e
de
ns
e
ly
int
e
r
c
onne
c
ted
laye
r
s
a
r
e
de
s
igned
to
lea
r
n
int
r
ica
te
pa
tt
e
r
ns
a
nd
r
e
lations
hips
in
the
e
xtr
a
c
ted
f
e
a
tur
e
s
.
T
he
hi
gh
c
onc
e
ntr
a
ti
on
of
ne
u
r
ons
in
each
de
ns
e
laye
r
e
nha
nc
e
s
the
model's
a
bil
it
y
to
c
a
ptur
e
both
low
-
leve
l
a
nd
high
-
leve
l
f
e
a
tur
e
s
,
ther
e
by
c
ontr
ibut
ing
to
its
c
a
pa
c
it
y
to
dis
c
e
r
n
c
ompl
e
x
pa
tt
e
r
ns
a
nd
make
p
r
e
c
is
e
c
las
s
if
ica
ti
ons
.
T
he
f
inal
output
laye
r
c
ons
is
ts
of
f
our
ne
ur
ons
,
each
r
e
pr
e
s
e
nti
ng
a
dif
f
e
r
e
nt
c
las
s
:
C
OV
I
D
-
19,
pne
umoni
a
,
lung
opa
c
it
y
a
nd
nor
mal.
T
he
S
of
tM
a
x
a
c
ti
va
ti
on
f
unc
ti
on
is
a
ppli
e
d
to
thi
s
laye
r
to
ge
ne
r
a
te
a
pr
oba
bil
it
y
dis
tr
ibut
ion
a
c
r
os
s
thes
e
c
las
s
e
s
,
f
a
c
il
it
a
ti
ng
a
c
c
ur
a
te
outcome
c
a
tegor
iza
ti
on.
4.
RE
S
UL
T
S
AND
DI
S
CU
S
S
I
ON
A
c
he
s
t
dis
e
a
s
e
c
las
s
if
ica
ti
on
model
wa
s
de
ve
lope
d
us
ing
the
3
.
11.
0
r
e
lea
s
e
of
P
ython
a
nd
the
Ke
r
a
s
f
r
a
mew
or
k.
T
he
s
im
ulation
of
the
model
wa
s
pe
r
f
or
med
on
a
Google
C
olab
P
r
e
mi
u
m
ve
r
s
io
n,
whic
h
pr
ovided
2
TB
of
s
tor
a
ge
,
25
GB
of
R
AM
,
a
nd
a
C
P
U
-
P
100.
T
he
p
r
e
-
pr
oc
e
s
s
ing
s
tage
invol
ve
d
uti
l
izing
the
I
mage
Da
taG
e
ne
r
a
tor
c
las
s
in
Ke
r
a
s
f
or
tas
ks
s
uc
h
as
im
a
ge
e
xpa
ns
ion,
nor
maliza
ti
on,
a
nd
da
ta
c
onve
r
s
ion.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
ff
icie
nt
lung
dis
e
as
e
c
las
s
if
ication
thr
ough
lumines
c
e
nt
featur
e
s
e
lec
ti
on
…
(
A
njugam
Shanmugav
e
lu)
3105
T
he
pr
opos
e
d
de
e
p
lea
r
ning
a
r
c
hit
e
c
tur
e
f
o
r
c
las
s
if
ying
mul
ti
ple
lung
dis
e
a
s
e
s
wa
s
de
ve
loped
us
ing
the
F
A.
T
he
model
wa
s
tr
a
ined
a
nd
va
li
da
ted
us
ing
an
op
ti
mi
z
a
ti
on
a
lgor
it
hm
a
nd
f
i
t
methods
li
mi
ted
to
100
e
poc
hs
.
E
a
c
h
e
poc
h
c
ons
is
ted
of
f
ou
r
it
e
r
a
ti
ons
a
nd
wa
s
u
s
e
d
at
a
ba
tch
s
ize
of
128
.
T
he
opti
mi
z
e
r
Ada
m
wa
s
us
e
d
with
a
lea
r
ning
r
a
te
of
0.
001
.
T
he
pe
r
f
o
r
manc
e
e
va
luation
wa
s
c
a
r
r
ied
out
with
pr
e
c
is
ion,
a
c
c
ur
a
c
y,
r
e
c
a
ll
,
a
nd
F1
-
s
c
or
e
.
T
he
pe
r
f
o
r
manc
e
mea
s
ur
e
s
a
r
e
s
hown
in
T
a
ble
2.
In
c
ompar
is
on,
to
c
ur
r
e
nt
s
tate
-
of
-
the
-
a
r
t
a
ppr
oa
c
he
s
,
whic
h
of
ten
s
e
r
ve
as
the
"
gr
ound
t
r
ut
h,
"
our
model
de
mons
tr
a
tes
s
upe
r
ior
pe
r
f
or
manc
e
.
T
h
is
is
e
videnc
e
d
by
the
r
e
s
ult
s
pr
e
s
e
nted
in
T
a
ble
3,
whe
r
e
our
model
outper
f
o
r
ms
s
e
ve
r
a
l
highl
y
r
obus
t
pr
e
-
tr
a
ined
models
.
We
a
s
s
e
s
s
e
d
the
c
a
tegor
iza
ti
on
a
nd
identif
ica
ti
on
of
lung
dis
or
de
r
s
,
includ
ing
the
diagnos
is
of
lung
dis
e
a
s
e
s
a
nd
C
OV
I
D
-
19
us
ing
the
pr
e
-
tr
a
ined
ne
twor
k
f
oll
owe
d
by
the
pr
opos
e
d
C
NN
models
.
T
a
ble
2.
E
va
luation
of
the
VGG
-
16
pe
r
f
or
manc
e
w
it
h
lung
dis
e
a
s
e
A
c
c
ur
a
c
y
P
r
e
c
is
io
n
R
e
c
a
ll
F1
-
s
c
or
e
C
O
V
I
D
-
19
0.9916
0.97
0.98
0.97
L
ung
opa
c
it
y
0.9982
0.94
0.91
0.92
N
or
ma
l
0.9964
0.90
0.96
0.95
V
ir
a
l
pne
umoni
a
0.9889
1.00
0.93
0.97
T
a
ble
3.
P
e
r
f
o
r
manc
e
c
ompar
is
on
with
s
tate
-
of
-
the
-
a
r
t
a
ppr
oa
c
he
s
M
ode
ls
A
c
c
ur
a
c
y
(%)
P
r
e
c
is
io
n
(%)
R
e
c
a
ll
(%)
F1
-
s
c
or
e
(%)
D
e
ns
e
N
e
t
-
121
98.67
97.91
98.10
98.0
C
a
ps
N
e
t
96.44
96.74
96.82
96.75
I
nc
e
pt
io
nV
3
97.0
96.73
96.56
96.66
M
obi
le
N
e
t
97.58
97.21
97.42
97.39
E
f
f
ic
ie
nt
N
e
t
96.84
97.64
97.36
97.41
P
r
opos
e
d
mode
l
(
a
ve
r
a
ge
)
99.38
95.25
94.50
95.25
T
he
pr
opos
e
d
model
wa
s
int
r
oduc
e
d
ba
s
e
d
on
a
r
e
f
ined
VG
G
-
16
model.
T
o
identif
y
the
opti
mal
model,
s
e
ve
r
a
l
tr
a
ini
ng/validation
r
a
ti
os
we
r
e
e
mp
loyed.
T
he
pe
r
f
o
r
manc
e
metr
ics
f
or
lung
dis
e
a
s
e
d
e
tec
ti
on
a
r
e
s
hown
in
F
igur
e
6
.
T
he
s
e
metr
ics
we
r
e
c
o
ns
ider
e
d
f
or
a
c
ompr
e
he
ns
ive
a
na
lys
is
of
the
p
r
e
diction,
a
c
c
ur
a
c
y
a
nd
r
obus
tnes
s
of
the
models
.
T
his
mod
e
l
a
c
hieve
d
a
n
a
c
c
ur
a
c
y
of
99.
3%
,
the
highes
t
s
p
e
c
if
icity,
F1
-
s
c
or
e
,
a
nd
r
e
c
a
ll
on
the
im
pleme
nted
X
-
r
a
y
da
tas
e
t.
Our
f
indi
ngs
s
how
a
s
igni
f
ica
nt
im
pr
ov
e
ment
a
s
pr
ovided
by
the
model.
F
igur
e
7
s
how
11
f
e
a
tur
e
s
a
r
e
s
e
lec
ted
a
t
ini
ti
a
l
it
e
r
a
ti
on
1
a
n
d
f
e
a
tur
e
s
a
r
e
r
e
duc
e
d
to
4
f
e
a
tur
e
s
a
t
the
10
th
it
e
r
a
ti
on
tot
a
l
f
e
a
tur
e
r
e
s
ult
f
or
10
it
e
r
a
ti
ons
is
1
,
920.
F
igu
r
e
8
s
hows
the
s
i
gnif
ica
nt
dif
f
e
r
e
nc
e
s
in
the
s
e
lec
ted
f
e
a
tur
e
s
of
f
ir
e
f
li
e
s
'
opti
mi
z
a
ti
on
of
da
ta
a
nd
s
tor
a
ge
o
f
the
s
ubs
e
t
f
o
r
the
p
r
opos
e
d
CN
N.
Ove
r
the
100
th
it
e
r
a
ti
on,
the
f
e
a
tur
e
numbe
r
dr
ops
to
only
13
f
e
a
tur
e
s
a
r
e
pa
s
s
e
d
to
the
c
las
s
if
ica
ti
on
model.
I
n
s
umm
a
r
y,
the
pr
opos
e
d
model
s
tands
out
a
s
the
be
s
t
-
pe
r
f
or
mi
ng
model
a
mong
thos
e
c
ompar
e
d,
of
f
e
r
ing
the
highes
t
a
c
c
ur
a
c
y
a
nd
ba
lanc
e
d
pr
e
c
is
ion
a
nd
r
e
c
a
ll
metr
ics
.
T
his
indi
c
a
tes
that
the
p
r
opos
e
d
model
is
not
only
mor
e
r
e
li
a
ble
in
c
las
s
if
ying
lung
dis
e
a
s
e
s
but
a
ls
o
mi
nim
ize
s
the
c
ha
nc
e
s
of
f
a
ls
e
pos
it
ives
a
nd
f
a
ls
e
ne
ga
ti
ve
s
.
C
ons
e
que
ntl
y,
the
im
pleme
nt
a
ti
on
of
thi
s
model
in
c
li
nica
l
s
e
tt
ings
c
ould
lea
d
to
be
tt
e
r
diagnos
ti
c
outcome
s
,
r
e
duc
e
unne
c
e
s
s
a
r
y
pr
oc
e
dur
e
s
,
a
nd
im
pr
ove
pa
ti
e
nt
c
a
r
e
.
F
igur
e
6.
T
he
lung
dis
e
a
s
e
outcome
with
VG
G
-
16
e
va
luation
r
e
s
ult
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
309
9
-
3108
3106
F
igur
e
7
.
F
e
a
tur
e
s
e
lec
ti
on
of
FA
F
igur
e
8
.
T
he
r
e
duc
e
d
a
tt
r
a
c
ti
on
va
lue
a
r
ound
13
a
f
ter
100
it
e
r
a
ti
ons
of
FA
5.
CONC
L
USI
ON
T
his
pa
pe
r
in
tr
oduc
e
s
an
innovative
a
ppr
oa
c
h
us
ing
the
f
i
r
e
f
ly
a
lgo
r
it
hm
f
or
opti
mal
f
e
a
tur
e
s
e
lec
ti
on
in
the
c
ontext
of
lung
dis
e
a
s
e
de
tec
ti
o
n.
By
e
mpl
oying
r
e
c
ipr
oc
a
l
c
r
os
s
-
e
ntr
opy,
we
e
f
f
e
c
ti
ve
ly
tac
kled
c
ha
ll
e
nge
s
a
s
s
oc
iate
d
with
unde
f
ined
a
nd
z
e
r
o
va
lues
in
the
logar
it
hmi
c
f
unc
ti
ons
.
T
he
a
l
gor
it
hm
leve
r
a
ge
s
a
tt
r
a
c
ti
ve
ne
s
s
a
nd
r
a
ndom
wa
lk
mec
ha
nis
ms
to
identi
f
y
a
nd
s
tor
e
a
bnor
mal
da
ta
as
a
s
ub
s
e
t.
T
his
s
ubs
e
t
is
then
uti
li
z
e
d
f
or
pa
tt
e
r
n
matc
hing
in
c
o
njunction
with
C
NN
,
c
ontr
ibu
ti
ng
to
e
nha
nc
e
d
e
f
f
ica
c
y
in
va
r
ious
lung
dis
e
a
s
e
s
a
nd
C
OV
I
D
-
19
de
tec
ti
on.
T
he
a
ppli
c
a
ti
on
of
the
f
i
r
e
f
ly
a
lgo
r
it
hm
e
xtend
s
to
the
c
a
lcula
ti
on
of
opti
mal
mul
ti
-
thr
e
s
holds
,
f
u
r
ther
r
e
f
ini
ng
pr
e
c
is
ion.
T
he
c
ur
r
e
nt
s
tudy
f
oc
us
e
s
on
a
c
hieving
higher
a
c
c
ur
a
c
y
in
the
identi
f
ica
ti
on
of
C
OV
I
D
-
19
a
nd
other
lung
dis
or
de
r
s
while
r
e
duc
ing
c
omp
utational
r
unti
me.
F
u
r
ther
mor
e
,
the
pa
pe
r
s
pe
c
if
ica
ll
y
c
onc
e
ntr
a
tes
on
the
mul
ti
c
las
s
c
a
t
e
gor
iza
ti
on
of
lung
dis
or
de
r
s
ba
s
e
d
upon
f
r
ontal
X
-
r
a
y
im
a
ge
s
of
c
he
s
t,
a
nd
th
us
,
the
s
c
ope
is
li
mi
ted
to
thi
s
s
pe
c
if
ic
domain.
T
his
wor
k
lays
a
r
obus
t
f
ounda
ti
on
f
or
a
dva
nc
ing
the
s
tate
-
of
-
the
-
a
r
t
methodologi
e
s
in
C
OV
I
D
-
19
a
nd
lung
dis
e
a
s
e
de
tec
ti
on,
a
nd
it
ope
ns
a
ve
nue
s
f
or
f
utur
e
r
e
s
e
a
r
c
h
in
de
ve
lopi
ng
innovative
diagnos
ti
c
tec
hni
que
s
a
nd
tr
e
a
tm
e
nt
s
tr
a
tegie
s
f
or
lung
dis
e
a
s
e
s
us
ing
CT
s
c
a
ns
.
F
UN
DI
NG
I
NF
ORM
AT
I
ON
T
his
r
e
s
e
a
r
c
h
did
not
r
e
c
e
ive
a
ny
s
pe
c
if
ic
gr
a
nt
f
r
om
f
unding
a
ge
nc
ies
in
the
publi
c
,
c
omm
e
r
c
ial,
or
not
-
f
or
-
pr
of
it
s
e
c
tor
s
AU
T
HO
R
CONT
RI
B
U
T
I
ONS
S
T
AT
E
M
E
N
T
T
his
jour
na
l
us
e
s
the
C
ontr
ibut
o
r
R
oles
T
a
xo
nomy
(
C
R
e
diT
)
to
r
e
c
ognize
indi
vidual
a
uthor
c
ontr
ibut
ions
,
r
e
duc
e
a
utho
r
s
hip
dis
putes
,
a
nd
f
a
c
il
it
a
te
c
oll
a
bor
a
ti
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
ff
icie
nt
lung
dis
e
as
e
c
las
s
if
ication
thr
ough
lumines
c
e
nt
featur
e
s
e
lec
ti
on
…
(
A
njugam
Shanmugav
e
lu)
3107
Nam
e
of
Au
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Anjuga
m
S
ha
nmugave
lu
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ar
ul
L
e
e
na
R
os
e
P
e
ter
J
os
e
ph
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
li
z
a
ti
on
M
:
M
e
th
odol
ogy
So
:
So
f
twa
r
e
Va
:
Va
li
da
ti
on
Fo
:
Fo
r
ma
l
a
na
ly
s
is
I
:
I
nve
s
ti
ga
ti
on
R
:
R
e
s
our
c
e
s
D
:
D
a
ta
C
ur
a
ti
on
O
:
W
r
it
in
g
-
O
r
ig
in
a
l
D
r
a
f
t
E
:
W
r
it
in
g
-
R
e
vi
e
w
&
E
di
ti
ng
Vi
:
Vi
s
ua
li
z
a
ti
on
Su
:
Su
pe
r
vi
s
io
n
P
:
P
r
oj
e
c
t
a
dmi
ni
s
tr
a
ti
on
Fu
:
Fu
ndi
ng a
c
qui
s
it
io
n
CONF
L
I
CT
OF
I
NT
E
RE
S
T
S
T
AT
E
M
E
N
T
Author
s
s
tate
no
c
onf
li
c
t
of
int
e
r
e
s
t.
DA
T
A
AV
AI
L
A
B
I
L
I
T
Y
T
he
da
ta
that
s
uppor
t
the
f
ind
ings
o
f
thi
s
s
tudy
a
r
e
ope
nly
a
va
il
a
ble
a
t
htt
ps
:/
/www
.
ka
ggle.
c
om/
da
tas
e
ts
/t
a
ws
if
ur
r
a
hman/c
ovid19
-
r
a
diogr
a
phy
-
da
taba
s
e
RE
F
E
RE
NC
E
S
[
1]
P
.
M
e
ht
a
,
D
.
F
.
M
c
A
ul
e
y,
M
.
B
r
ow
n,
E
.
S
a
nc
he
z
,
R
.
S
.
T
a
tt
e
r
s
a
ll
,
a
nd
J
.
J
.
M
a
n
s
on,
“
C
O
V
I
D
-
19:
c
ons
id
e
r
c
yt
oki
ne
s
t
or
m
s
yndr
ome
s
a
nd i
mm
unos
uppr
e
s
s
io
n,”
T
he
L
anc
e
t
, vol
. 395, pp.
1033
–
1034, 2020, doi:
10.1016
/S
0140
-
6736(
20
)
30628
-
0.
[
2]
S
.
B
.
A
ti
ta
ll
a
h,
M
.
D
r
is
s
,
W
.
B
oul
il
a
,
A
.
K
ouba
a
,
a
nd
H
.
B
.
G
hé
z
a
la
,
“
F
us
io
n
of
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
ba
s
e
d
on
D
e
mps
te
r
–
S
ha
f
e
r
th
e
or
y
f
or
a
ut
oma
ti
c
pne
umoni
a
de
te
c
ti
on
f
r
om
c
he
s
t
X
-
r
a
y
im
a
ge
s
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
I
m
agi
ng
Sy
s
t
e
m
s
and T
e
c
hnol
ogy
, vol
. 32, no. 2, pp. 658
–
672, 2022, doi:
10.1002/i
ma
.22653.
[
3]
Y
.
F
a
ng
e
t
al
.
,
“
S
e
ns
it
iv
it
y
of
c
he
s
t
C
T
f
or
C
O
V
I
D
-
19:
c
ompa
r
is
on
to
R
T
-
P
C
R
,”
R
adi
ol
ogy
,
vol
.
296,
no.
2,
pp.
E
115
–
E
117,
2020, doi:
10.1148/r
a
di
ol
.2020200432.
[
4]
M
.
H
a
bi
b,
T
.
O
.
T
im
ouda
s
,
Y
.
D
in
g,
N
.
N
or
d,
S
.
C
he
n,
a
n
d
Q
.
W
a
ng,
“
A
hybr
id
ma
c
hi
ne
l
e
a
r
ni
ng
a
ppr
oa
c
h
f
or
th
e
l
oa
d
pr
e
di
c
ti
on
in
th
e
s
us
ta
in
a
bl
e
tr
a
ns
it
io
n
of
di
s
tr
ic
t
he
a
ti
ng
ne
twor
ks
,”
Sus
ta
in
abl
e
C
it
ie
s
and
Soc
ie
t
y
,
vol
.
99,
2023,
doi
:
10.1016/j
.s
c
s
.20
23.104892.
[
5]
A
.
A
.
A
r
da
ka
ni
,
U
.
R
.
A
c
ha
r
ya
,
S
.
H
a
bi
bol
la
hi
,
a
nd
A
.
M
oha
mm
a
di
,
“
C
O
V
I
D
ia
g:
a
c
li
ni
c
a
l
C
A
D
s
ys
t
e
m
to
di
a
gnos
e
C
O
V
I
D
-
19
pne
umoni
a
ba
s
e
d on
C
T
f
in
di
ngs
,”
E
ur
ope
an R
adi
ol
ogy
, vol
. 3
1, no. 1, pp. 121
–
130, 2021, doi:
10.1007/s
00330
-
020
-
07087
-
y.
[
6]
L
.
J
.
S
e
e
la
n,
L
.
P
.
S
ur
e
s
h,
A
.
K
.S
.,
a
nd
V
.
P
.K
.,
“
C
omput
e
r
-
a
id
e
d
de
te
c
ti
on
of
huma
n
lu
ng
nodule
s
on
c
omput
e
r
to
mogr
a
phy
im
a
ge
s
vi
a
nove
l
opt
im
iz
e
d
te
c
hni
que
s
,”
C
ur
r
e
nt
M
e
di
c
al
I
m
agi
ng
R
e
v
ie
w
s
,
vol
.
18,
no.
12,
pp.
1282
–
1290,
2021,
doi
:
10.2174/15734056
17666211126151713.
[
7]
I
. C
houa
t,
A
.
E
c
ht
io
ui
, R
. K
he
ma
khe
m, W
.
Z
ouc
h, M
.
G
hor
be
l
, a
nd A
. B
.
H
a
mi
da
, “
C
O
V
I
D
-
19
de
te
c
ti
on i
n C
T
a
nd C
X
R
i
ma
ge
s
us
in
g de
e
p l
e
a
r
ni
ng mode
ls
,”
B
io
ge
r
ont
ol
ogy
, vol
. 23, no. 1, pp. 65
–
84, 2022, doi:
10.1007/s
10522
-
021
-
09946
-
7.
[
8]
R
.
J
a
in
,
M
.
G
upt
a
,
S
.
T
a
ne
ja
,
a
nd
D
.
J
.
H
e
ma
nt
h,
“
D
e
e
p
le
a
r
ni
ng
ba
s
e
d
de
te
c
ti
on
a
nd
a
na
ly
s
is
of
C
O
V
I
D
-
19
on
c
he
s
t
X
-
r
a
y
im
a
ge
s
,”
A
ppl
ie
d I
nt
e
ll
ig
e
nc
e
, vol
. 51, no. 3, pp. 1690
–
1700, 2
021, doi:
10.1007/s
10489
-
020
-
01902
-
1.
[
9]
B
.
W
a
ng
a
nd
W
.
Z
ha
ng,
“
M
A
R
ne
t:
M
ul
ti
-
s
c
a
le
a
d
a
pt
iv
e
r
e
s
i
dua
l
ne
ur
a
l
ne
twor
k
f
or
c
he
s
t
X
-
r
a
y
im
a
ge
s
r
e
c
ogni
ti
on
of
l
ung
di
s
e
a
s
e
s
,”
M
at
he
m
at
ic
al
B
io
s
c
ie
n
c
e
s
and E
ngi
ne
e
r
in
g
, vol
. 19, no. 1, pp. 331
–
350, 2022, doi:
10.3934/m
be
.2022017.
[
10]
T
. B
e
z
da
n,
M
. Z
iv
kovi
c
, N
. B
a
c
a
ni
n, A
. C
hha
br
a
,
a
nd M
. S
ur
e
s
h, “
F
e
a
tu
r
e
s
e
le
c
ti
on by hybr
id
br
a
in
s
to
r
m opt
i
mi
z
a
ti
on a
lg
or
it
hm
f
or
C
O
V
I
D
-
19 c
la
s
s
if
ic
a
ti
on,”
J
our
nal
of
C
om
put
at
io
nal
B
io
lo
gy
, vol
. 29, no. 6, pp. 515
–
529, 2022, doi:
10.1089/cmb.2021.0256.
[
11]
A
.
M
a
br
ouk,
R
.
P
.
D
.
R
e
dondo,
A
.
D
a
hou,
M
.
A
.
E
la
z
iz
,
a
nd
M
.
K
a
ye
d,
“
P
ne
umoni
a
de
te
c
ti
on
on
c
he
s
t
X
-
r
a
y
im
a
ge
s
u
s
in
g
e
ns
e
mbl
e
of
de
e
p c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
s
,”
A
ppl
ie
d Sc
ie
nc
e
s
, vol
. 12, no. 13, 2022, doi
:
10.3390/a
pp12136448.
[
12]
L
.
T
.
D
uong,
P
.
T
.
N
guye
n,
L
.
I
ovi
no,
a
nd
M
.
F
la
mm
in
i,
“
A
ut
oma
ti
c
de
te
c
ti
on
of
C
O
V
I
D
-
19
f
r
om
c
he
s
t
X
-
r
a
y
a
nd
l
ung
c
omput
e
d
to
mogr
a
phy
im
a
ge
s
u
s
in
g
de
e
p
n
e
ur
a
l
ne
twor
ks
a
n
d
tr
a
ns
f
e
r
le
a
r
ni
ng,”
A
ppl
ie
d
Sof
t
C
om
put
in
g
,
vol
.
132,
2023,
doi
:
10.1016/j
.a
s
oc
.2022
.109851.
[
13]
P
.
C
ha
kr
a
bor
ty
,
“
I
ndi
a
lo
gs
774
ne
w
C
O
V
I
D
-
19
c
a
s
e
s
,
2
de
a
t
hs
,
619
J
N
.1
in
f
e
c
ti
ons
s
o
f
a
r
,”
I
ndi
a
T
oday
.
2024.
A
c
c
e
s
s
e
d
:
F
e
b.
07,
2024
.
[
O
nl
in
e
]
.
A
va
il
a
bl
e
:
ht
tp
s
:/
/ww
w
.i
ndi
a
to
da
y.i
n/
c
or
ona
vi
r
us
-
out
br
e
a
k/
s
to
r
y/
in
di
a
-
c
ovi
d
-
c
a
s
e
s
-
upda
te
-
ke
r
a
la
-
ka
r
na
t
a
ka
-
jn
1
-
in
f
e
c
ti
ons
-
ta
mi
l
-
na
du
-
guj
a
r
a
t
-
w
in
te
r
-
s
e
a
s
on
-
2485047
-
2024
-
01
-
06
[
14]
N
.
K
ha
s
a
w
n
e
h,
M
.
F
r
a
iwa
n,
L
.
F
r
a
iwa
n, B
.
K
ha
s
s
a
w
n
e
h,
a
nd A
.
I
bni
a
n,
“
D
e
te
c
ti
on
of
C
O
V
I
D
-
1
9
f
r
om
c
he
s
t
X
-
r
a
y
im
a
ge
s
u
s
in
g
de
e
p c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
,”
Se
n
s
or
s
, vol
. 21, no. 17,
2021, doi:
10.3390/s
21175940.
[
15]
A
.
W
.
S
a
le
hi
e
t
al
.
,
“
A
s
tu
dy
of
C
N
N
a
nd
tr
a
ns
f
e
r
le
a
r
ni
ng
in
me
di
c
a
l
im
a
gi
ng:
a
dva
nt
a
ge
s
,
c
ha
ll
e
nge
s
,
f
ut
ur
e
s
c
op
e
,”
Sus
ta
in
a
bi
li
ty
, vol
. 15, no. 7, 2023, doi
:
10.3390/s
u15075930.
[
16]
F
.
O
z
yur
t,
T
.
T
unc
e
r
,
a
nd
A
.
S
uba
s
i,
“
A
n
a
ut
oma
te
d
C
O
V
I
D
-
19
de
te
c
ti
on
ba
s
e
d
on
f
us
e
d
dyna
mi
c
e
xe
mpl
a
r
pyr
a
mi
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
nd
hybr
id
f
e
a
tu
r
e
s
e
le
c
ti
on
us
in
g
de
e
p
le
a
r
ni
n
g,”
C
om
put
e
r
s
in
B
io
lo
gy
and
M
e
di
c
in
e
,
vol
.
132,
2021,
doi
:
10.1016/j
.c
ompbi
ome
d.2021.104356.
[
17]
A
.
S
in
gh
e
t
al
.
,
“
S
e
le
c
ti
ve
w
a
ve
le
ngt
h
opt
ic
a
l
f
il
te
r
s
f
r
om
mi
xe
d
pol
ymor
ph
a
nd
bi
na
r
y
in
te
gr
a
ti
on
of
M
oO
3
mul
ti
l
a
ye
r
s
tr
uc
tu
r
e
s
,”
O
pt
ic
al
M
at
e
r
ia
ls
, vol
. 111, 2021, doi:
10.101
6/
j.
o
pt
ma
t.
2020.110709.
[
18]
M
.
A
la
r
oud,
N
.
T
a
ha
t,
S
.
A
l
-
O
ma
r
i,
D
.
L
.
S
ut
ha
r
,
a
nd
S
.
G
ul
ya
z
-
O
z
yur
t,
“
A
n
a
tt
r
a
c
ti
ve
a
ppr
oa
c
h
a
s
s
oc
ia
te
d
w
it
h
tr
a
ns
f
or
m
f
unc
ti
ons
f
or
s
ol
vi
ng
c
e
r
ta
in
f
r
a
c
ti
ona
l
s
w
if
t
-
hohe
nbe
r
g
e
qua
ti
on,”
J
our
nal
of
F
unc
ti
on
Spac
e
s
,
vol
.
2021,
2021,
doi
:
10.1155/2021/
3230272.
[
19]
M
.
S
ha
r
ma
a
nd
P
.
K
a
ur
,
“
A
c
ompr
e
he
n
s
iv
e
a
na
ly
s
is
of
na
tu
r
e
-
in
s
pi
r
e
d
me
ta
-
he
ur
is
ti
c
te
c
hni
que
s
f
or
f
e
a
tu
r
e
s
e
le
c
ti
on
pr
obl
e
m,”
A
r
c
hi
v
e
s
of
C
om
put
at
io
nal
M
e
th
ods
i
n E
ngi
ne
e
r
in
g
, vol
. 28, no
. 3, pp. 1103
–
1127, 2021, doi:
10.1007/s
11831
-
020
-
09412
-
6.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
309
9
-
3108
3108
[
20]
F
.
M
.
J
.
M
.
S
ha
mr
a
t
e
t
al
.
,
“
L
ungNe
t2
2:
a
f
in
e
-
tu
ne
d
mode
l
f
or
mul
ti
c
la
s
s
c
la
s
s
if
ic
a
ti
on
a
nd
pr
e
di
c
ti
on
of
lu
ng
di
s
e
a
s
e
us
in
g
X
-
r
a
y i
ma
ge
s
,”
J
ou
r
nal
of
P
e
r
s
onal
iz
e
d M
e
di
c
in
e
, vol
. 12, no. 5,
2022, doi:
10.3390/j
pm12050680.
[
21]
K
.
N
a
im
a
,
B
.
F
a
d
e
la
,
C
.
I
me
ne
,
a
nd
C
.
A
bde
lk
a
de
r
,
“
U
S
E
of
ge
ni
ti
c
a
lg
or
it
hm
a
nd
p
a
r
ti
c
le
s
w
a
r
m
opt
im
is
a
ti
on
me
th
od
s
f
or
th
e
opt
im
a
l
c
ont
r
ol
of
th
e
r
e
a
c
ti
ve
pow
e
r
in
w
e
s
te
r
n
a
lg
e
r
ia
n
po
w
e
r
s
ys
te
m,”
E
ne
r
g
y
P
r
oc
e
di
a
,
vol
.
74,
pp.
265
–
272,
2015,
doi
:
10.1016/j
.e
gypr
o.2015.07.597.
[
22]
A
.
S
a
hoo
a
nd
S
.
C
ha
ndr
a
,
“
I
m
pr
ove
d
c
e
r
vi
x
le
s
io
n
c
l
a
s
s
if
ic
a
ti
on
us
in
g
mul
ti
-
obj
e
c
ti
ve
bi
na
r
y
f
ir
e
f
ly
a
lg
or
it
hm
-
ba
s
e
d
f
e
a
tu
r
e
s
e
le
c
ti
on,”
I
nt
e
r
nat
io
nal
J
ou
r
nal
of
B
io
-
I
ns
pi
r
e
d C
om
put
at
io
n
, vol
. 8, no. 6, pp. 367
–
378, 2016, doi:
10.1504/I
J
B
I
C
.2016.081326.
[
23]
A
.
K
ha
pa
r
de
,
V
.
D
e
s
hmukh,
a
nd
M
.
K
ow
di
ki
,
“
E
nha
nc
e
d
na
tu
r
e
-
in
s
pi
r
e
d
a
lg
or
it
hm
-
ba
s
e
d
hybr
id
de
e
p
le
a
r
ni
ng
f
or
c
ha
r
a
c
te
r
r
e
c
ogni
ti
on i
n s
a
ns
kr
it
l
a
ngua
ge
,”
Se
n
s
in
g and I
m
agi
ng
, vol
. 24, no. 1, 2023, doi
:
10.1007/s
11220
-
023
-
00421
-
w.
[
24]
M
.
N
s
s
ib
i,
G
.
M
a
ni
ta
,
a
nd
O
.
K
or
ba
a
,
“
A
dva
nc
e
s
in
na
tu
r
e
-
in
s
pi
r
e
d
me
ta
he
ur
is
ti
c
opt
im
iz
a
ti
on
f
or
f
e
a
tu
r
e
s
e
le
c
ti
on
pr
obl
e
m:
a
c
ompr
e
he
ns
iv
e
s
ur
ve
y,”
C
om
put
e
r
S
c
ie
nc
e
R
e
v
i
e
w
, vol
. 49, 20
23, doi:
10.1016/j
.c
os
r
e
v.2023.100559.
[
25]
N
. G
upt
a
, D
.
G
upt
a
, A
. K
ha
nna
, P
. P
. R
.
F
il
ho, a
nd V
. H
. C
. de
A
lb
uque
r
qu
e
, “
E
vol
ut
io
na
r
y a
lg
or
it
hms
f
or
a
ut
oma
ti
c
l
ung dis
e
a
s
e
de
te
c
ti
on,”
M
e
as
ur
e
m
e
nt
:
J
our
nal
of
th
e
I
nt
e
r
nat
io
nal
M
e
as
ur
e
m
e
nt
C
onf
e
de
r
at
io
n
,
vol
.
140,
pp.
590
–
608,
2019,
doi
:
10.1016/j
.me
a
s
ur
e
me
nt
.2019.02.042.
[
26]
A
.
K
a
ve
h
a
nd
S
.
M
.
J
a
va
di
,
“
C
ha
os
-
ba
s
e
d
f
ir
e
f
l
y
a
lg
or
it
hms
f
or
opt
im
iz
a
ti
on
of
c
yc
li
c
a
ll
y
la
r
ge
-
s
iz
e
br
a
c
e
d
s
te
e
l
dome
s
w
it
h
mul
ti
pl
e
f
r
e
que
nc
y c
ons
tr
a
in
ts
,”
C
om
put
e
r
s
and Str
u
c
tu
r
e
s
, vol
. 214, pp. 28
–
39, 2019, doi:
10.1016/j
.c
omps
tr
uc
.2019.01.006.
[
27]
T
.
B
e
z
d
a
n,
D
.
C
v
e
tn
ic
,
L
.
G
a
ji
c
,
M
.
Z
iv
kovi
c
,
I
.
S
tr
umbe
r
g
e
r
,
a
nd
N
.
B
a
c
a
ni
n,
“
F
e
a
tu
r
e
s
e
le
c
ti
on
by
f
ir
e
f
ly
a
lg
or
it
hm
w
it
h
im
pr
ove
d i
ni
ti
a
li
z
a
ti
on s
tr
a
te
gy,”
A
C
M
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
P
r
oc
e
e
di
ng Se
r
ie
s
, 2021, doi:
10.1145/3459960.3
459974.
[
28]
T
.
R
a
hma
n
e
t
al
.
,
“
Q
C
ovS
M
L
:
a
r
e
li
a
bl
e
C
O
V
I
D
-
19
de
te
c
ti
o
n
s
ys
te
m
us
in
g
C
B
C
bi
oma
r
ke
r
s
by
a
s
ta
c
ki
ng
ma
c
hi
ne
le
a
r
n
in
g
mode
l,
”
C
om
put
e
r
s
i
n B
io
lo
gy
and M
e
di
c
in
e
, vol
. 143, 2022, doi:
10.1016/j
.c
ompbi
ome
d.2022.105284.
[
29]
M
.
E
.
H
.
C
how
dhur
y
e
t
al
.
,
“
C
a
n
A
I
he
lp
in
s
c
r
e
e
ni
ng
vi
r
a
l
a
nd
C
O
V
I
D
-
19
P
ne
umoni
a
?
,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
8,
pp.
132665
–
132676, 2020, doi:
10.1109/AC
C
E
S
S
.2020.3010287.
[
30]
T
.
R
a
hma
n
e
t
al
.
,
“
E
xpl
or
in
g
th
e
e
f
f
e
c
t
of
im
a
ge
e
nha
nc
e
me
nt
te
c
hni
que
s
on
C
O
V
I
D
-
19
de
te
c
ti
on
us
in
g
c
he
s
t
X
-
r
a
y
im
a
ge
s
,
”
C
om
put
e
r
s
i
n B
io
lo
gy
and M
e
di
c
in
e
, vol
. 132, 2021, doi:
10
.10
16/
j.
c
ompbi
ome
d.2021.104319.
[
31]
N
.
B
a
c
a
ni
n,
K
.
V
e
nk
a
ta
c
ha
la
m,
T
.
B
e
z
d
a
n,
M
.
Z
iv
ko
vi
c
,
a
nd
M
.
A
b
ouh
a
w
w
a
s
h,
“
A
nov
e
l
f
ir
e
f
l
y
a
lg
or
it
hm
a
ppr
o
a
c
h
f
or
e
f
f
i
c
ie
n
t
f
e
a
tu
r
e
s
e
l
e
c
t
io
n
w
it
h
C
O
V
I
D
-
19
da
ta
s
e
t
,”
M
i
c
r
op
r
o
c
e
s
s
or
s
and
M
ic
r
os
y
s
te
m
s
, vol
. 98
, 20
23,
doi
:
10.1
016/
j
.mi
c
pr
o.2
023.
1047
7
8.
[
32]
A
.
S
a
ygı
lı
,
“
A
n
e
w
a
ppr
oa
c
h
f
or
c
omput
e
r
-
a
id
e
d
de
t
e
c
ti
on
of
c
or
ona
vi
r
us
(
C
O
V
I
D
-
19)
f
r
om
C
T
a
nd
X
-
r
a
y
im
a
ge
s
us
in
g
ma
c
hi
ne
le
a
r
ni
ng me
th
ods
,”
A
ppl
ie
d Soft
C
om
put
in
g
, vol
. 105, 2021, doi:
10.1016/j
.a
s
oc
.2021.107323.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
A
njug
a
m
Sha
n
m
ug
a
v
e
l
u
h
o
l
d
s
a
B.
Sc.
in
Co
mp
u
t
e
r
Sci
en
ce,
MCA
a
n
d
M.
T
ec
h
.
in
C
o
mp
u
t
er
Sc
i
en
ce.
S
h
e
is
cu
rre
n
t
l
y
an
as
s
i
s
t
a
n
t
p
ro
f
es
s
o
r
in
t
h
e
D
e
p
art
me
n
t
of
Co
m
p
u
t
er
Sci
en
ce
at
Raj
es
w
ari
V
ed
ac
h
al
am
G
o
v
ern
me
n
t
A
rt
s
Co
l
l
e
g
e,
Ch
en
g
al
p
at
t
u
.
Sh
e
h
as
22
y
ears
of
t
each
i
n
g
ex
p
er
i
en
ce.
Cu
rren
t
l
y
,
she
is
al
s
o
p
u
rs
u
i
n
g
Ph
.
D
.
in
Co
mp
u
t
er
Sci
e
n
ce
at
the
SRM
In
s
t
i
t
u
t
e
of
Sci
en
ce
an
d
T
ec
h
n
o
l
o
g
y
,
Ch
en
n
a
i
.
H
er
res
earch
i
n
t
ere
s
t
i
n
cl
u
d
e
s
d
i
g
i
t
al
i
mag
e
p
ro
ce
s
s
i
n
g
,
mach
i
n
e
l
ear
n
i
n
g
,
an
d
d
ee
p
l
earn
i
n
g
.
Sh
e
h
as
p
res
en
t
ed
h
er
res
earc
h
p
ap
er
s
at
i
n
t
ern
a
t
i
o
n
a
l
co
n
fere
n
ces
t
h
at
are
i
n
d
ex
e
d
in
Sco
p
u
s
.
Sh
e
can
be
co
n
t
act
e
d
at
emai
l
:
s
an
j
u
g
am7
8
@
g
mai
l
.
co
m
.
D
r.
A
rul
Leena
R
o
s
e
P
eter
J
o
s
eph
o
b
t
a
i
n
e
d
a
D
o
ct
o
ra
t
e
in
C
o
mp
u
t
er
Sci
e
n
ce
fro
m
Mo
t
h
er
T
h
eres
a
U
n
i
v
ers
i
t
y
,
l
o
ca
t
ed
in
K
o
d
ai
k
an
a
l
,
In
d
i
a,
in
2
0
1
6
.
Sh
e
h
o
l
d
s
t
h
e
p
o
s
i
t
i
o
n
of
a
Pro
fes
s
o
r
in
t
h
e
D
e
p
art
me
n
t
of
C
o
mp
u
t
er
Sc
i
e
n
ce
at
SRM
In
s
t
i
t
u
t
e
of
Sc
i
en
ce
an
d
T
ech
n
o
l
o
g
y
in
Ch
en
n
ai
.
Sh
e
p
o
s
s
e
s
s
e
s
a
w
eal
t
h
of
k
n
o
w
l
e
d
g
e
w
i
t
h
25
y
ear
s
of
t
each
i
n
g
ex
p
er
i
en
ce
a
n
d
7
y
ear
s
of
re
s
earc
h
ex
p
eri
e
n
ce.
A
d
d
i
t
i
o
n
al
l
y
,
s
h
e
ex
h
i
b
i
t
s
p
ro
f
i
ci
e
n
cy
in
v
ari
o
u
s
t
ech
n
i
cal
d
o
mai
n
s
l
i
k
e
s
o
f
t
w
are
en
g
i
n
eeri
n
g
,
D
BMS,
an
d
art
i
fi
c
i
al
i
n
t
e
l
l
i
g
e
n
ce.
Sh
e
s
p
ec
i
al
i
zes
in
mach
i
n
e
l
ear
n
i
n
g
,
i
mag
e
p
r
o
ces
s
i
n
g
,
an
d
art
i
fi
c
i
al
n
e
u
ral
n
e
t
w
o
rk
s
.
Sh
e
h
as
au
t
h
o
re
d
n
u
mero
u
s
res
earc
h
p
ap
er
s
t
h
a
t
h
av
e
b
ee
n
p
u
b
l
i
s
h
e
d
in
i
n
t
ern
a
t
i
o
n
a
l
l
y
reco
g
n
i
ze
d
j
o
u
rn
a
l
s
an
d
co
n
feren
c
es
i
n
d
e
x
ed
in
Sc
o
p
u
s
a
n
d
W
eb
of
Sci
en
ce.
S
h
e
can
be
c
o
n
t
act
e
d
at
emai
l:
l
een
a.
r
o
s
e
5
2
7
@
g
mai
l
.
co
m.
Evaluation Warning : The document was created with Spire.PDF for Python.