I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
15
, N
o.
1
,
F
e
br
ua
r
y
2026
, pp.
592
~
603
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
15
.i
1
.pp
592
-
603
592
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
Pn
e
u
m
on
i
a c
l
ass
i
f
i
c
at
i
on
f
r
om
c
h
e
st
X
-
r
ay
s u
si
n
g si
gn
i
f
i
c
an
t
f
e
at
u
r
e
s
e
l
e
c
t
i
on
an
d
m
ac
h
i
n
e
l
e
ar
n
i
n
g
Y
u
gan
d
h
ar
C
h
od
agam
, M
an
j
u
n
at
h
a H
ir
e
m
at
h
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
, C
e
nt
r
a
l
C
a
m
pus
,
C
hr
i
s
t
U
ni
ve
r
s
i
t
y, B
e
nga
l
ur
u, I
ndi
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
A
ug 22, 2025
R
e
vi
s
e
d
D
e
c
27, 2025
A
c
c
e
pt
e
d
J
a
n 10, 2026
The
chest
X
-
ray
images
of
normal
lungs
differ
only
subtly
from
th
ose
of
lungs
with
pneumonia,
making
image
-
based
diagnosis
highly
challe
nging.
To
address
this
issue,
we
developed
a
machine
learning
(ML)
-
based,
lightweight,
end
-
to
-
end
Python
package
that
processes
ch
est
X
-
ray
i
mages,
implements
robust
feature
selection
methods,
and
classifies
the
image
s
using
various
algorithms.
While
many
studies
have
focused
on
imp
roving
classifi
cation
accuracy
using
newer
methods
,
few
have
address
ed
the
interpreta
bility
of
the
extracte
d
features
or
the
growing
comput
ational
demands
of
complex
models.
We
used
four
publicly
available
datase
ts
and
extracted
first
-
order,
textural,
and
transform
-
based
radiomic
features
to
test
our
package.
Features
were
selected
using
the
Shapley
additive
expla
nations
(SHAP)
combined
with
recursive
feature
elimin
ation
(RFE)
and
stability
selection
algorithms.
Our
final
solution
contains
a
method
that
ext
racts
a
finite
set
of
features
identified
by
stability
selection
and
feeds
them
as
inputs
into
classical
ML
a
lgorithms.
Our
model
achieve
d
98%
accura
cy
on
the
primary
dataset,
and
97%±
1,
96%±
2,
and
94%±
2%
accuracy
on
th
e
other
three
datasets.
Our
approach
is
f
ast,
self
-
contained
,
and
requires
o
nly
an
ideal
set
of
features,
making
it
suitable
for
resource
-
constrain
ed
c
linical
environm
ents.
K
e
y
w
o
r
d
s
:
F
e
a
tu
r
e
s
e
le
c
ti
on
M
a
c
hi
ne
l
e
a
r
ni
ng
P
ne
um
oni
a
R
a
di
om
ic
s
W
a
ve
le
t
f
e
a
tu
r
e
s
X
-
r
a
y
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
Y
uga
ndha
r
C
hoda
ga
m
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
S
c
ie
nc
e
, C
e
nt
r
a
l
C
a
m
pu
s
, C
hr
is
t
U
ni
v
e
r
s
it
y
B
e
nga
lu
r
u, I
ndi
a
E
m
a
il
:
yuga
ndha
r
.c
h@
r
e
s
.c
hr
is
tu
ni
ve
r
s
it
y.i
n
1.
I
N
T
R
O
D
U
C
T
I
O
N
P
ne
um
oni
a
r
e
m
a
in
s
one
of
th
e
m
os
t
s
e
r
io
us
he
a
lt
h
th
r
e
a
ts
w
or
ld
w
id
e
.
A
c
c
or
di
ng
to
a
r
e
por
t
publ
is
he
d
by
th
e
C
e
nt
e
r
f
or
D
is
e
a
s
e
C
ont
r
ol
a
nd
p
r
e
ve
nt
io
n
i
n
2023,
pne
um
oni
a
c
l
a
im
s
m
or
e
th
a
n
41,000
li
ve
s
e
ve
r
y
ye
a
r
in
th
e
U
ni
te
d
S
ta
te
s
,
a
nd
th
is
bur
de
n
c
oul
d
b
e
he
a
vi
e
r
in
de
ve
lo
pi
ng
c
ount
r
ie
s
w
it
h
poor
e
r
m
e
di
c
a
l
f
a
c
il
it
ie
s
.
D
is
ti
ngui
s
hi
ng
a
he
a
lt
hy
lu
ng
f
r
om
one
w
it
h
pne
um
oni
a
on
a
c
he
s
t
X
-
r
a
y
is
of
te
n
m
or
e
di
f
f
ic
ul
t
th
a
n
it
a
ppe
a
r
s
,
e
ve
n
f
or
a
tr
a
in
e
d
pr
a
c
ti
ti
one
r
.
F
ig
ur
e
1
s
how
s
r
e
pr
e
s
e
nt
a
ti
ve
c
he
s
t
X
-
r
a
y
im
a
ge
s
us
e
d
in
th
is
s
tu
dy:
F
ig
ur
e
1(
a
)
il
lu
s
tr
a
te
s
a
nor
m
a
l
c
a
s
e
,
w
hi
le
F
ig
ur
e
1(
b)
il
lu
s
tr
a
te
s
pne
um
oni
a
.
V
is
ua
l
c
lu
e
s
li
ke
pa
tc
hy
opa
c
it
ie
s
or
f
a
in
t
c
on
s
ol
id
a
ti
ons
c
a
n
be
s
ubt
le
a
nd
e
a
s
y
to
m
is
s
,
e
s
p
e
c
ia
ll
y
f
or
le
s
s
e
xpe
r
ie
n
c
e
d
r
a
di
ol
ogi
s
ts
.
P
a
ti
e
nt
s
w
ho
vi
s
it
hos
pi
ta
ls
w
it
h
li
m
it
e
d
r
e
s
our
c
e
s
f
a
c
e
m
a
ny
hur
dl
e
s
w
he
n
e
nough
e
xpe
r
ie
nc
e
d
r
a
di
ol
ogi
s
ts
a
r
e
una
va
il
a
bl
e
.
A
r
a
di
ol
ogi
s
t
s
houl
d
ha
ve
e
no
ugh
e
xpe
r
ti
s
e
in
r
e
a
di
ng
c
he
s
t
X
-
r
a
y
s
a
nd
id
e
nt
if
yi
ng
a
pa
th
ol
ogy
w
i
th
a
n
a
c
c
e
pt
a
bl
e
le
ve
l
of
c
onf
id
e
nc
e
.
T
hi
s
de
pe
nde
nc
y
on
e
xpe
r
ie
nc
e
d
r
a
di
ol
ogi
s
ts
c
a
n
be
m
it
ig
a
te
d
th
r
ough
a
ut
om
a
te
d
im
a
ge
di
a
gnos
is
or
a
s
s
is
te
d
di
a
gnos
is
.
D
e
e
p
le
a
r
ni
ng
(
D
L
)
s
ys
te
m
s
ha
ve
a
lr
e
a
dy
de
m
ons
tr
a
te
d
im
pr
e
s
s
iv
e
a
c
c
ur
a
c
y
in
m
a
ny
m
e
di
c
a
l
im
a
gi
ng
ta
s
ks
.
H
o
w
e
ve
r
,
th
e
y
of
te
n
r
e
ly
on
m
a
s
s
iv
e
,
w
e
ll
-
a
nnot
a
te
d
da
ta
s
e
ts
a
nd
pow
e
r
f
ul
ha
r
dw
a
r
e
[
1]
,
[
2]
.
A
not
he
r
pr
om
is
in
g
s
ol
ut
io
n
is
to
c
om
bi
ne
m
a
c
hi
ne
le
a
r
ni
ng
(
M
L
)
a
lg
or
it
hm
s
w
it
h
r
a
di
om
ic
s
m
e
th
ods
to
a
c
c
ur
a
te
ly
c
la
s
s
if
y
X
-
r
a
y
im
a
ge
s
by
e
xt
r
a
c
ti
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
P
ne
um
oni
a c
la
s
s
if
ic
at
io
n f
r
om
c
he
s
t
x
-
r
ay
s
us
in
g
s
ig
ni
fi
c
ant
f
e
at
ur
e
s
e
le
c
ti
on …
(
Y
ugandhar
C
hodagam)
593
m
e
a
s
ur
a
bl
e
a
nd
unde
r
s
ta
nda
bl
e
f
e
a
tu
r
e
s
,
s
uc
h
a
s
te
xt
ur
e
,
s
ha
pe
,
a
nd
tr
a
ns
f
or
m
.
S
om
e
pow
e
r
f
ul
M
L
a
lg
or
it
hm
s
c
a
n
le
a
r
n
f
r
om
th
e
s
e
in
f
o
r
m
a
ti
ve
s
e
ts
of
de
s
c
r
ip
to
r
s
a
nd
a
c
hi
e
ve
r
e
li
a
bl
e
pe
r
f
or
m
a
nc
e
w
it
h
le
s
s
c
om
put
a
ti
ona
l
de
m
a
nds
[
2]
.
T
hi
s
r
e
s
e
a
r
c
h
a
im
s
t
o
c
om
bi
n
e
th
e
s
tr
e
n
gt
h
s
of
r
a
di
om
i
c
s
a
n
d M
L
to
pi
np
oi
nt
t
he
m
os
t
di
s
c
r
im
i
na
ti
v
e
f
e
a
tu
r
e
s
u
s
in
g a
dva
nc
e
d f
e
a
t
ur
e
s
e
le
c
ti
on m
e
th
od
s
. W
e
pr
opo
s
e
a
pr
a
c
ti
c
a
l
, hi
gh
-
p
e
r
f
or
m
in
g p
a
c
k
a
ge
d s
ol
u
ti
on
f
or
pne
um
o
ni
a
c
la
s
s
if
i
c
a
ti
on
w
it
h
a
s
y
s
te
m
a
ti
c
a
ppr
o
a
c
h
a
nd
e
xt
e
n
s
iv
e
e
v
a
lu
a
ti
on
a
c
r
o
s
s
m
ul
ti
pl
e
d
a
ta
s
e
t
s
.
V
in
od
e
t
al
.
[
3]
r
e
por
t
t
ha
t
t
he
ir
d
e
e
p
C
ovi
x
-
N
e
t
c
on
vol
ut
io
na
l
n
e
ur
a
l
n
e
twor
ks
(
C
N
N
s
)
obt
a
in
s
p
a
tt
e
r
n
s
di
r
e
c
tl
y f
r
om
t
he
i
m
a
ge
p
ix
e
l
s
, a
nd a
s
e
p
a
r
a
t
e
pi
p
e
li
ne
f
ir
s
t
e
xt
r
a
c
ts
t
e
xt
ur
e
a
nd w
a
v
e
le
t
f
e
a
tu
r
e
s
a
nd t
h
e
n l
e
t
s
a
r
a
ndom f
or
e
s
t
a
l
gor
it
hm
l
e
a
r
n f
r
om
t
h
os
e
de
s
c
r
ip
t
or
s
.
C
N
N
s
l
e
a
r
n pa
tt
e
r
n
s
f
r
om
r
a
w
pi
xe
l
s
, w
he
r
e
a
s
r
a
di
om
i
c
s
m
e
th
od
s
e
xt
r
a
c
t
te
x
tu
r
e
a
n
d
tr
a
ns
f
or
m
f
e
a
tu
r
e
s
,
w
hi
c
h
c
a
n
be
f
e
d
i
nt
o
M
L
a
lg
or
i
th
m
s
.
B
ot
h
th
e
s
e
a
ppr
o
a
c
h
e
s
ha
ve
t
he
ir
s
tr
e
ng
th
s
a
nd
li
m
it
a
ti
o
ns
.
A
nd
e
r
s
o
n
e
t
al
.
[
4]
de
s
ig
ne
d
a
DL
–
ba
s
e
d
s
ys
t
e
m
a
nd
tr
a
in
e
d
it
o
n
490,000
c
h
e
s
t
X
-
r
a
y
s
,
a
c
hi
e
vi
n
g a
n
a
r
e
a
und
e
r
t
h
e
c
ur
ve
(
A
U
C
)
of
0.976 f
or
pa
th
ol
ogy
de
t
e
c
ti
o
n a
n
d i
m
pr
o
ve
d
a
c
c
ur
a
c
y
o
n
u
ns
e
e
n
h
e
ld
-
out
d
a
ta
s
e
t
s
.
V
e
r
m
a
e
t
al
.
[
1]
u
s
e
d
t
r
a
ns
f
or
m
f
e
a
tu
r
e
s
s
u
c
h
a
s
w
a
v
e
le
ts
to
e
x
tr
a
c
t
f
e
a
tu
r
e
s
s
u
c
h
a
s
e
nt
r
o
py
a
nd
e
ne
r
g
y,
a
c
hi
e
vi
ng
a
n
a
c
c
ur
a
c
y
of
96.5%
u
s
in
g
a
s
upp
or
t
ve
c
to
r
m
a
c
hi
ne
a
lg
or
it
hm
,
K
h
a
tt
a
b
e
t
al
.
[
5]
pr
opo
s
e
d
f
oc
a
l
lo
s
s
–
tu
n
e
d
I
n
c
e
pt
io
n
m
o
de
l
s
,
w
hi
c
h
de
m
ons
tr
a
te
d
a
n
ov
e
r
a
ll
a
c
c
ur
a
c
y
of
97.6
7%
w
it
h
r
e
c
a
ll
r
a
te
s
a
bove
96%
.
R
a
bb
a
h
e
t
al
.
[
6]
a
dde
d
th
r
e
e
de
n
s
e
l
a
y
e
r
s
ov
e
r
th
e
I
nc
e
pt
i
on
-
v3 e
xt
r
a
c
to
r
w
it
h
a
ppr
oxi
m
a
te
l
y 22.9 mi
ll
io
n w
e
ig
ht
s
a
nd r
e
por
t
e
d a
n a
c
c
ur
a
c
y of
97
.23%
f
or
bi
na
r
y
pne
um
o
ni
a
d
e
te
c
ti
on.
E
m
pl
oyi
ng
h
e
te
r
o
ge
n
e
ou
s
e
n
s
e
m
bl
e
s
,
e
f
f
ic
ie
nt
-
V
G
G
16
ne
ur
a
l
ne
t
w
or
k
a
c
hi
e
ve
d
99.46%
a
c
c
ur
a
c
y
on
th
e
c
o
vi
d
-
xr
a
y
-
5k
be
nc
hm
a
r
k
[
7]
,
w
hi
le
J
a
g
hda
m
e
t
al
.
[
8]
a
c
hi
e
v
e
d
a
n
e
v
e
n
hi
ghe
r
a
c
c
ur
a
c
y
of
99
.59%
b
y
hyp
e
r
-
opt
im
i
z
in
g
a
n
e
f
f
ic
i
e
nt
D
e
n
s
e
N
e
t
ba
c
kbon
e
.
S
ur
e
ndr
a
e
t
al
.
[
9]
C
X
N
e
t
m
ode
l
de
m
on
s
tr
a
t
e
d
a
n
a
c
c
ur
a
c
y
of
98%
,
w
it
h
a
96%
r
e
c
a
ll
r
a
te
f
or
de
te
c
ti
ng
c
or
ona
vi
r
us
di
s
e
a
s
e
201
9
c
a
s
e
s
in
a
m
ul
ti
c
l
a
s
s
c
la
s
s
if
i
c
a
ti
on.
H
ow
e
ve
r
,
a
non
–
DL
a
ppr
oa
c
h
c
a
n
b
e
e
qu
a
ll
y
e
f
f
e
c
ti
ve
in
m
e
d
ic
a
l
im
a
ge
c
la
s
s
if
ic
a
ti
on.
Ö
z
c
a
n
[
10]
l
ig
ht
G
B
M
a
lg
or
it
hm
,
tr
a
in
e
d
on
ju
s
t
97
m
ul
ti
s
c
a
le
r
a
di
om
ic
f
e
a
tu
r
e
s
,
s
ti
ll
de
li
ve
r
e
d
a
n
a
c
c
ur
a
c
y,
s
e
ns
it
iv
it
y,
a
nd
s
pe
c
if
ic
it
y
of
97.5,
97.5,
a
nd
98.75%
,
r
e
s
pe
c
ti
ve
ly
.
G
ua
n
e
t
al
.
[
11]
pr
opos
e
d
th
e
us
e
of
a
c
om
pa
c
t
ha
s
h
la
ye
r
f
or
e
xt
r
a
c
ti
ng
e
m
be
ddi
ngs
f
r
om
th
e
D
e
n
s
e
N
e
t
-
121
ne
two
r
k,
a
c
hi
e
vi
ng
a
n
a
v
e
r
a
ge
pr
e
c
is
io
n
of
0.84
f
or
r
e
tr
ie
vi
ng
pne
um
oni
a
c
a
s
e
s
.
P
a
l
e
t
al
.
[
12]
de
s
ig
ne
d
a
hyb
r
id
C
N
N
-
tr
a
ns
f
or
m
e
r
m
ode
l
th
a
t
yi
e
ld
e
d
a
n
a
c
c
ur
a
c
y
of
95.14%
w
he
n
te
s
te
d
on
th
e
s
a
m
e
th
r
e
e
-
c
la
s
s
be
n
c
hm
a
r
k.
T
he
s
e
r
e
s
ul
ts
s
how
th
a
t
w
hi
le
m
ode
r
n
C
N
N
s
or
C
N
N
-
vi
s
io
n
tr
a
n
s
f
or
m
e
r
e
ns
e
m
bl
e
s
of
te
n
a
c
hi
e
ve
a
c
c
ur
a
c
ie
s
of
98
-
99%
,
f
e
a
tu
r
e
s
id
e
nt
if
ie
d
th
r
ough
s
ta
bi
li
ty
s
e
le
c
ti
on
a
nd
ot
he
r
c
a
r
e
f
ul
ly
c
hos
e
n
r
a
di
om
ic
f
e
a
tu
r
e
s
c
a
n
m
a
tc
h
th
a
t
pe
r
f
or
m
a
nc
e
w
it
h
f
a
r
le
s
s
c
om
put
a
ti
ona
l
ove
r
he
a
d.
(
a
)
(
b)
F
ig
ur
e
1. R
e
pr
e
s
e
nt
a
ti
ve
c
he
s
t
X
-
r
a
y
s
a
m
pl
e
s
of
(
a
)
nor
m
a
l
a
nd (
b)
pne
um
oni
a
2.
M
E
T
H
O
D
R
e
s
e
a
r
c
h
e
r
s
w
or
ld
w
id
e
ha
ve
e
xpe
nde
d
c
on
s
id
e
r
a
bl
e
e
f
f
or
ts
f
or
tr
a
in
in
g
c
om
put
e
r
s
to
de
te
c
t
pne
um
oni
a
a
nd
ot
he
r
th
or
a
c
ic
di
s
e
a
s
e
s
di
r
e
c
tl
y
f
r
om
X
-
r
a
y
s
us
in
g
DL
a
nd
M
L
m
e
th
ods
.
W
hi
le
th
e
pr
og
r
e
s
s
ha
s
be
e
n
e
nc
our
a
gi
ng,
th
e
f
ol
lo
w
in
g
is
s
ue
s
ha
ve
pe
r
s
is
te
d:
i
)
m
ode
ls
tr
a
in
e
d
on
one
da
ta
s
e
t
do
not
a
lwa
ys
pe
r
f
or
m
w
e
ll
on
a
not
he
r
;
ii
)
th
ous
a
nds
of
ove
r
la
ppi
ng
f
e
a
t
ur
e
s
c
a
n
m
a
s
k
th
e
tr
ul
y
us
e
f
ul
s
ig
na
ls
;
a
nd
iii
)
he
a
vyw
e
ig
ht
a
r
c
hi
te
c
tu
r
e
s
c
a
n
be
to
o
s
lo
w
or
c
os
tl
y
f
or
e
v
e
r
yda
y
c
li
ni
c
a
l
us
e
. T
o
a
ddr
e
s
s
th
e
s
e
is
s
u
e
s
, w
e
de
ve
lo
pe
d
a
s
e
lf
-
c
ont
a
in
e
d
pa
c
ka
g
e
th
a
t
c
a
n
f
a
c
il
it
a
te
th
e
e
x
tr
a
c
ti
on
of
s
ig
ni
f
ic
a
nt
f
e
a
tu
r
e
s
,
e
na
bl
in
g
th
e
c
la
s
s
if
ic
a
ti
on
of
c
he
s
t
X
-
r
a
y
s
a
s
nor
m
a
l
or
in
di
c
a
ti
ve
of
pne
um
oni
a
w
it
h
id
e
a
l
a
c
c
ur
a
c
y
le
ve
ls
.
T
he
de
s
ig
n
e
d
pa
c
ka
ge
s
houl
d
id
e
a
ll
y
de
m
ons
tr
a
te
s
ta
bl
e
a
nd
opt
im
a
l
pe
r
f
or
m
a
nc
e
a
c
r
os
s
m
ul
ti
pl
e
da
ta
s
e
ts
, e
ve
n
in
s
e
tt
in
gs
w
it
h
li
m
it
e
d
c
om
put
a
ti
ona
l
r
e
s
our
c
e
s
.
T
he
m
a
in
a
dva
nt
a
g
e
s
of
th
is
a
ppr
oa
c
h
a
r
e
th
a
t
it
:
i
)
e
xpe
di
te
s
im
a
ge
-
ba
s
e
d
di
a
gnos
is
,
ii
)
r
e
li
e
s
on
ge
ne
r
a
l
c
om
put
e
r
s
th
a
t
a
r
e
e
a
s
il
y
a
va
il
a
bl
e
,
a
nd
iii
)
a
s
s
is
ts
be
gi
nne
r
r
a
di
ol
ogi
s
ts
/a
nnot
a
to
r
s
w
hi
le
r
e
duc
in
g t
he
bur
de
n on e
xpe
r
t
r
a
d
io
lo
gi
s
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
15
, N
o.
1
,
F
e
br
ua
r
y
20
26
:
592
-
603
594
F
ir
s
t,
r
a
di
om
ic
de
s
c
r
ip
to
r
s
w
e
r
e
in
it
ia
ll
y
e
xt
r
a
c
te
d
us
in
g
th
e
P
yR
a
di
om
ic
s
pa
c
ka
g
e
[
13]
,
w
hi
c
h
f
ol
lo
w
s
th
e
im
a
ge
bi
om
a
r
ke
r
s
ta
nda
r
di
z
a
ti
on
in
it
ia
ti
ve
.
S
ubs
e
que
nt
ly
,
th
e
m
os
t
in
f
or
m
a
ti
ve
f
e
a
tu
r
e
s
w
e
r
e
id
e
nt
if
ie
d
a
nd
r
e
ta
in
e
d
by
a
ppl
yi
ng
two
r
ig
or
ous
f
e
a
tu
r
e
s
e
le
c
ti
on
te
c
hni
que
s
na
m
e
ly
,
S
ha
pl
e
y
a
ddi
ti
ve
e
xpl
a
na
ti
on
(
S
H
A
P
)
c
om
bi
ne
d
w
it
h
r
e
c
ur
s
iv
e
f
e
a
tu
r
e
e
li
m
in
a
ti
on
(
R
F
E
)
a
nd
s
ta
bi
li
ty
s
e
le
c
ti
on.
A
s
e
t
of
im
a
ge
s
w
a
s
r
e
a
d i
nt
o m
e
m
or
y i
n t
he
G
oogl
e
C
ol
a
b e
nvi
r
onm
e
nt
a
lo
ng w
it
h t
he
P
yR
a
di
om
ic
s
pa
c
ka
g
e
. B
e
f
or
e
th
e
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
a
ll
th
e
im
a
ge
s
w
e
r
e
c
onve
r
te
d
in
to
ne
a
r
ly
r
a
w
r
a
s
te
r
da
ta
(
nr
r
d
)
f
il
e
s
.
A
ppr
oxi
m
a
te
l
y
955
f
e
a
tu
r
e
s
w
e
r
e
e
xt
r
a
c
te
d,
na
m
e
ly
,
or
ig
in
a
l
s
ta
ti
s
ti
c
a
l
f
e
a
tu
r
e
s
f
r
om
r
a
w
im
a
ge
s
(
95)
;
tr
a
ns
f
or
m
f
e
a
tu
r
e
s
s
uc
h
a
s
w
a
v
e
le
t
-
lo
w
-
pa
s
s
,
hi
gh
-
pa
s
s
(
L
H
)
(
86)
,
w
a
ve
le
t
-
hi
gh
-
pa
s
s
,
lo
w
-
pa
s
s
(
H
L
)
(
86)
,
w
a
ve
le
t
-
hi
gh
-
pa
s
s
,
hi
gh
-
pa
s
s
(
H
H
)
(
86)
,
w
a
ve
le
t
-
lo
w
-
pa
s
s
(
L
L
)
(
86)
,
e
xpone
nt
i
a
l
(
86)
,
gr
a
di
e
nt
(
86)
,
lo
ga
r
i
th
m
(
86)
,
s
qua
r
e
(
86)
,
s
qua
r
e
r
oot
(
86)
,
a
nd
lb
p
-
2d
(
86)
.
T
he
s
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
w
e
r
e
f
e
d
in
to
a
lg
or
it
hm
s
s
uc
h
a
s
X
G
B
oos
t,
gr
a
di
e
nt
boos
ti
ng,
a
nd
r
a
ndom
f
or
e
s
t
to
s
e
le
c
t
th
e
f
e
a
tu
r
e
s
of
hi
gh
im
por
ta
nc
e
a
c
r
os
s
two
da
ta
s
e
ts
.
A
f
lo
w
c
ha
r
t
de
pi
c
ti
ng t
he
c
la
s
s
if
ic
a
ti
on of
i
m
a
ge
s
u
s
in
g S
H
A
P
a
n
d R
F
E
i
s
i
ll
us
tr
a
te
d i
n F
ig
ur
e
2.
F
ig
ur
e
2.
F
lo
w
c
ha
r
t
of
i
m
a
ge
c
la
s
s
if
ic
a
ti
on
us
in
g t
he
S
H
A
P
a
n
d R
F
E
m
e
th
ods
T
he
f
our
da
ta
s
e
ts
u
s
e
d
in
th
is
s
tu
dy
w
e
r
e
obt
a
in
e
d
f
r
om
th
e
f
ol
lo
w
in
g
s
our
c
e
s
:
G
ua
ngz
hou
W
om
e
n
a
nd
C
hi
ld
r
e
n’
s
M
e
di
c
a
l
C
e
nt
e
r
(
G
W
C
M
C
)
,
R
a
di
ol
ogy
A
I
G
r
oup
(
R
A
I
G
)
,
N
a
ti
ona
l
I
ns
ti
tu
te
of
G
e
ne
r
a
l
M
e
di
c
a
l
S
c
ie
n
c
e
s
(
N
I
G
M
S
)
,
a
nd
J
a
ga
nna
th
U
ni
ve
r
s
it
y.
S
H
A
P
c
om
bi
ne
d
w
it
h
R
F
E
e
xhi
bi
te
d
e
xc
e
ll
e
nt
pe
r
f
or
m
a
nc
e
in
id
e
nt
if
yi
ng
th
e
f
e
a
tu
r
e
s
f
or
e
a
c
h
da
ta
s
e
t
s
e
pa
r
a
te
ly
.
A
lt
hough
th
e
m
ode
ls
a
c
hi
e
ve
d
s
a
ti
s
f
a
c
to
r
y
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y,
th
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
e
ts
a
c
r
os
s
a
ll
th
e
da
ta
s
e
ts
s
ha
r
e
d
onl
y
30%
c
om
m
on f
e
a
tu
r
e
s
. S
H
A
P
e
xpl
a
in
s
a
s
in
gl
e
p
r
e
di
c
ti
on by s
pl
it
ti
n
g t
he
di
f
f
e
r
e
nc
e
be
twe
e
n t
he
m
ode
l
out
put
f
o
r
th
a
t
in
s
ta
nc
e
,
f
(
x)
,
a
nd
th
e
m
ode
l’
s
e
xpe
c
te
d
out
put
,
E
(
f
(
z
)
)
,
ove
r
a
r
e
f
e
r
e
nc
e
da
ta
s
e
t
[
14]
.
A
lt
hough
th
e
S
H
A
P
a
nd
R
F
E
te
c
hni
que
s
yi
e
ld
e
d
good
a
c
c
ur
a
c
y
f
or
e
a
c
h
da
ta
s
e
t
(
G
W
C
M
C
98%
;
R
A
I
G
98%
;
N
I
G
M
S
97%
;
J
a
ga
nna
th
94%
)
,
th
e
S
H
A
P
f
e
a
tu
r
e
s
va
r
ie
d
a
c
r
os
s
da
ta
s
e
ts
.
T
he
num
be
r
of
c
om
m
on
f
e
a
tu
r
e
s
w
a
s
lo
w
,
a
nd t
he
S
H
A
P
va
lu
e
s
of
t
he
f
e
a
tu
r
e
s
w
e
r
e
t
ie
d t
o t
he
da
ta
s
e
t.
C
onve
r
s
e
ly
,
th
e
f
lo
w
c
ha
r
t
in
F
ig
ur
e
3
a
nd
th
e
a
lg
or
it
hm
in
F
ig
ur
e
4
a
c
hi
e
ve
d
opt
im
a
l
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
a
nd
id
e
nt
if
ie
d
f
e
a
tu
r
e
s
e
ts
w
it
h
ove
r
90%
c
om
m
ona
l
it
y
a
c
r
os
s
a
ll
da
ta
s
e
ts
.
O
pt
im
a
l
a
c
c
ur
a
c
y
w
a
s
a
c
hi
e
ve
d
b
e
c
a
u
s
e
s
ta
bi
li
ty
s
e
l
e
c
ti
on
dr
a
w
s
m
a
ny
boot
s
tr
a
p
s
u
bs
e
ts
,
f
it
s
a
s
pa
r
s
e
m
ode
l
on
e
a
c
h,
a
nd
ke
e
p
s
onl
y
th
os
e
va
r
ia
bl
e
s
w
hos
e
s
e
le
c
ti
on
f
r
e
que
nc
y
e
xc
e
e
ds
a
u
s
e
r
-
de
f
in
e
d
th
r
e
s
hol
d,
w
hi
c
h
w
e
s
e
t
a
s
20%
[
15]
.
T
a
bl
e
1 pr
e
s
e
nt
s
da
ta
on t
he
pe
r
c
e
nt
a
ge
of
c
om
m
on f
e
a
tu
r
e
s
i
d
e
nt
if
ie
d a
c
r
os
s
t
he
f
our
da
ta
s
e
ts
us
in
g di
f
f
e
r
e
n
t
e
xt
r
a
c
ti
on me
th
ods
.
T
he
m
e
th
ods
u
s
e
d
f
or
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
r
e
de
s
c
r
ib
e
d
in
th
e
f
ol
lo
w
in
g
s
ubs
e
c
ti
ons
.
T
o
ke
e
p
th
e
doc
um
e
nt
c
onc
is
e
,
a
ll
th
e
f
e
a
tu
r
e
s
e
xt
r
a
c
te
d
u
s
in
g
S
H
A
P
c
o
m
bi
ne
d
w
it
h
R
F
E
a
r
e
not
li
s
te
d
he
r
e
.
T
hi
s
i
s
be
c
a
us
e
t
h
e
y a
r
e
ge
ne
r
a
ll
y a
s
ubs
e
t
of
t
he
f
e
a
tu
r
e
s
e
xt
r
a
c
te
d us
i
ng t
he
s
ta
bi
li
ty
s
e
le
c
ti
on a
lg
or
it
hm
.
F
ig
ur
e
3
.
I
m
a
ge
c
la
s
s
if
ic
a
ti
on us
in
g s
ta
bi
li
ty
s
e
le
c
ti
on
S
t
a
r
t
pyr
a
di
om
i
c
s
C
onve
r
t
i
m
a
ge
s
c
om
bi
ne
d f
r
om
2
da
t
a
s
e
t
s
t
o .nr
r
d f
i
l
e
s
E
xt
r
a
c
t
~
1000
f
e
a
t
ur
e
s
F
i
r
s
t
or
de
r
a
nd
W
a
ve
le
ts
S
t
a
bi
l
i
t
y s
e
l
e
c
t
i
on
I
m
a
ge
c
l
a
s
s
i
f
i
c
a
t
i
on
S
t
a
r
t
pyr
a
di
om
i
c
s
C
onve
r
t
i
m
a
ge
s
t
o .nr
r
d f
i
l
e
s
E
xt
r
a
c
t
~
1000 f
e
a
t
ur
e
s
F
i
r
s
t
or
de
r
a
nd W
a
ve
l
e
t
S
H
A
P
+R
F
E
I
m
a
ge
c
l
a
s
s
i
f
i
c
a
t
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
P
ne
um
oni
a c
la
s
s
if
ic
at
io
n f
r
om
c
he
s
t
x
-
r
ay
s
us
in
g
s
ig
ni
fi
c
ant
f
e
at
ur
e
s
e
le
c
ti
on …
(
Y
ugandhar
C
hodagam)
595
F
ig
ur
e
4. S
ta
bi
li
ty
s
e
le
c
ti
on a
lg
or
it
hm
T
a
bl
e
1.
C
om
pa
r
is
on of
c
om
m
on f
e
a
tu
r
e
pe
r
c
e
nt
a
ge
s
a
c
r
os
s
f
ou
r
da
ta
s
e
ts
us
in
g di
f
f
e
r
e
nt
m
e
th
ods
M
e
t
hod us
e
d
F
e
a
t
ur
e
m
a
t
c
h pe
r
c
e
nt
a
ge
a
c
r
os
s
da
t
a
s
e
t
s
(
%
)
C
om
m
e
nt
s
S
H
A
P
+R
F
E
~
30
F
e
a
t
ur
e
s
c
om
m
on
a
c
r
o
s
s
t
hr
e
e
da
t
a
s
e
t
s
w
e
nt
up t
o 30%
S
t
a
bi
l
i
t
y s
e
l
e
c
t
i
on us
i
ng
l
ogi
s
t
i
c
r
e
gr
e
s
s
i
on
~
70
L
ogi
s
t
i
c
r
e
gr
e
s
s
i
on
i
n
s
t
a
bi
l
i
t
y
s
e
l
e
c
t
i
on
ga
ve
good r
e
s
ul
t
s
S
t
a
bi
l
i
t
y s
e
l
e
c
t
i
on w
i
t
h
X
G
B
oos
t
>90
X
G
B
oos
t
pr
oduc
e
d
t
he
b
e
s
t
r
e
s
ul
t
s
w
he
n
t
he
s
t
a
bi
l
i
t
y s
e
l
e
c
t
i
on a
l
gor
i
t
hm
w
a
s
u
s
e
d
2.1.
S
h
ap
le
y ad
d
i
t
iv
e
e
xp
la
n
at
io
n
s
an
d
r
e
c
u
r
s
iv
e
f
e
at
u
r
e
e
li
m
in
at
io
n
al
gor
it
h
m
s
2.1.1.
S
h
ap
le
y ad
d
i
t
iv
e
e
xp
la
n
at
io
n
s
T
he
S
ha
pl
e
y
f
unc
ti
on
c
a
lc
ul
a
t
e
s
a
va
lu
e
f
or
e
a
c
h
f
e
a
tu
r
e
e
xt
r
a
c
te
d
f
r
om
e
a
c
h
im
a
g
e
.
T
hi
s
va
lu
e
r
e
f
le
c
ts
th
e
im
por
ta
nc
e
of
th
a
t
pa
r
ti
c
ul
a
r
f
e
a
tu
r
e
in
th
e
ove
r
a
ll
i
m
a
ge
c
la
s
s
if
ic
a
ti
on.
I
t
is
th
e
a
ve
r
a
ge
m
a
r
gi
na
l
c
ont
r
ib
ut
io
n of
a
f
e
a
tu
r
e
a
c
r
os
s
a
ll
pos
s
ib
le
c
om
bi
na
ti
ons
.
ϕ
(
,
)
=
∑
|
|
!
(
|
|
−
|
|
−
1
)
!
|
|
!
⊆
∖
{
}
(
(
∪
{
}
)
−
(
)
)
(
1)
W
he
r
e
f
de
not
e
s
t
he
M
L
m
ode
l;
x
de
not
e
s
t
he
i
ns
ta
nc
e
f
or
w
hi
c
h w
e
w
a
nt
t
o e
xpl
a
in
t
he
pr
e
di
c
ti
on;
S
de
not
e
s
a
f
e
a
tu
r
e
s
ubs
e
t;
f
x
(
S
)
de
not
e
s
t
he
pr
e
di
c
ti
on f
or
i
ns
ta
nc
e
x
us
in
g only t
he
S
f
e
a
tu
r
e
s
;
a
nd
ϕ
i
de
not
e
s
t
he
S
H
A
P
va
lu
e
f
or
f
e
a
tu
r
e
I
;
(
(
∪
{
}
)
−
(
)
)
r
e
pr
e
s
e
nt
s
th
e
m
a
r
gi
na
l
c
ont
r
ib
ut
io
n
of
f
e
a
t
ur
e
i
w
he
n
it
is
a
dde
d
to
th
e
s
ubs
e
t
S
;
a
nd
|
|
!
(
|
|
−
|
|
−
1
)
!
/
|
|
!
de
not
e
s
th
e
w
e
ig
ht
in
g
f
a
c
to
r
f
or
th
e
num
be
r
of
w
a
ys
s
ubs
e
t
S
c
a
n
be
f
or
m
e
d, e
ns
ur
in
g a
ve
r
a
ge
c
ont
r
ib
ut
io
n ove
r
a
ll
pos
s
ib
le
s
ub
s
e
ts
.
2.1.2.
R
e
c
u
r
s
iv
e
f
e
at
u
r
e
e
li
m
in
at
io
n
R
F
E
i
s
a
f
e
a
tu
r
e
s
e
le
c
ti
on a
lg
or
it
hm
t
ha
t
a
im
s
t
o i
de
nt
if
y
th
e
m
os
t
im
por
ta
nt
f
e
a
tu
r
e
s
f
or
a
pr
e
di
c
ti
ve
m
ode
l
by
r
e
m
ovi
ng
th
e
le
a
s
t
im
por
ta
nt
f
e
a
tu
r
e
s
r
e
c
ur
s
iv
e
ly
.
T
r
a
in
th
e
m
ode
l
a
s
f
ol
lo
w
s
:
tr
a
in
a
m
ode
l
us
in
g
a
ll
t
he
da
ta
s
e
t
f
e
a
tu
r
e
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
15
, N
o.
1
,
F
e
br
ua
r
y
20
26
:
592
-
603
596
(
)
=
(
(
)
,
)
(
2)
W
he
r
e
X
(
t
)
de
not
e
s
t
he
f
e
a
tu
r
e
m
a
tr
ix
a
t
it
e
r
a
ti
on
t
, a
nd
M
(
t
)
de
not
e
s
t
he
t
r
a
in
e
d m
ode
l
a
t
it
e
r
a
ti
on
t
.
R
a
nk
f
e
a
tu
r
e
s
:
c
om
put
e
th
e
im
por
ta
nc
e
of
e
a
c
h
f
e
a
tu
r
e
in
th
e
tr
a
in
e
d
m
ode
l
M
(
t
)
.
L
e
t
ω
(
t
)
ϵ
ℝ
p
be
th
e
ve
c
to
r
of
f
e
a
tu
r
e
i
m
por
ta
nc
e
or
c
oe
f
f
ic
ie
nt
s
.
ω
(
)
=
(
(
)
)
(
3)
E
li
m
in
a
ti
ng
th
e
le
a
s
t
im
por
ta
nt
f
e
a
tu
r
e
:
f
e
a
tu
r
e
w
it
h
th
e
s
m
a
ll
e
s
t
im
por
ta
nc
e
s
c
or
e
is
id
e
nt
if
ie
d
a
nd
r
e
m
ove
d
f
r
om
t
he
f
e
a
tu
r
e
s
e
t.
j
(
t
)
=
arg
m
in
(
ω
j
(
t
)
)
(
4)
2.2.
S
t
ab
il
it
y s
e
le
c
t
io
n
T
he
pr
oc
e
s
s
of
pne
um
oni
a
c
la
s
s
if
ic
a
ti
on
us
in
g
th
e
s
ta
bi
li
ty
s
e
le
c
ti
on
a
lg
or
it
hm
is
de
pi
c
te
d
in
F
ig
ur
e
3.
T
he
w
or
kf
lo
w
is
a
lm
os
t
id
e
nt
ic
a
l
to
th
a
t
a
s
s
oc
ia
te
d
w
it
h
th
e
S
H
A
P
+
R
F
E
m
e
th
od
e
xc
e
pt
it
in
vol
ve
s
two
da
ta
s
e
ts
—
G
W
C
M
C
a
nd
a
K
a
ggl
e
da
ta
s
e
t
—
a
nd
th
e
a
l
gor
it
hm
to
s
e
le
c
t
th
e
s
ig
ni
f
ic
a
nt
f
e
a
tu
r
e
s
is
c
ha
nge
d,
de
not
e
d
in
bl
ue
c
ol
or
.
A
de
ta
il
e
d
f
lo
w
c
ha
r
t
de
s
c
r
ib
in
g
th
e
s
ta
bi
li
ty
s
e
le
c
ti
on
a
lg
or
it
hm
is
il
lu
s
tr
a
te
d
in
F
ig
ur
e
4.
T
he
a
lg
or
it
hm
w
or
ks
by
r
e
pe
a
te
dl
y
tr
a
in
in
g
a
n
M
L
c
la
s
s
if
ie
r
on
r
a
ndom
da
ta
s
ubs
e
ts
.
W
e
us
e
d
th
e
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
a
nd
X
G
B
oos
t
a
lg
or
it
hm
s
a
nd
t
r
a
c
ke
d
how
of
te
n
e
a
c
h
f
e
a
tu
r
e
w
a
s
id
e
nt
if
ie
d
a
s
im
por
ta
nt
.
T
hi
s
he
lp
s
id
e
nt
if
y
f
e
a
tu
r
e
s
th
a
t
c
on
s
is
te
nt
ly
c
ont
r
i
but
e
to
c
la
s
s
if
ic
a
ti
on
a
c
r
os
s
di
f
f
e
r
e
nt
s
a
m
pl
e
s
.
T
he
f
e
a
tu
r
e
s
e
xt
r
a
c
t
e
d
us
in
g
th
e
X
G
B
oos
t
a
lg
or
it
hm
w
e
r
e
f
o
und
to
f
a
c
il
it
a
te
th
e
c
la
s
s
if
ic
a
ti
on
of
im
a
ge
s
a
c
r
os
s
a
ll
f
our
da
ta
s
e
ts
a
s
s
how
n
in
T
a
bl
e
2.
T
hi
s
m
e
th
od
he
lp
s
s
e
le
c
t
s
ta
bl
e
f
e
a
tu
r
e
s
a
c
r
os
s
m
a
ny
it
e
r
a
ti
ons
,
in
c
r
e
a
s
in
g
th
e
r
obus
tn
e
s
s
of
th
e
m
ode
l
w
hi
le
r
e
duc
in
g
ove
r
f
i
tt
in
g.
T
he
a
lg
or
it
hm
pr
e
s
e
nt
e
d
in
F
ig
u
r
e
4
is
e
xpl
a
in
e
d
a
s
A
lg
or
it
hm
1.
T
a
bl
e
2. L
is
t
of
f
e
a
tu
r
e
s
i
de
nt
if
ie
d t
hr
ough s
ta
bi
li
ty
s
e
le
c
ti
on
N
o.
F
e
a
t
ur
e
na
m
e
N
o.
F
e
a
t
ur
e
na
m
e
1
w
a
ve
l
e
t
-
L
H
_gl
c
m
_C
l
us
t
e
r
T
e
nde
nc
y
13
w
a
ve
l
e
t
-
H
H
_gl
c
m
_I
dn
2
w
a
ve
l
e
t
-
H
H
_gl
r
l
m
_R
unE
nt
r
opy
14
w
a
ve
l
e
t
-
H
H
_gl
c
m
_I
dm
n
3
w
a
ve
l
e
t
-
L
H
_gl
r
l
m
_S
hor
t
R
unE
m
pha
s
i
s
15
w
a
ve
l
e
t
-
H
H
_gl
c
m
_I
dm
4
w
a
ve
l
e
t
-
L
H
_gl
r
l
m
_R
unE
nt
r
opy
16
w
a
ve
l
e
t
-
H
H
_gl
c
m
_I
m
c
1
5
w
a
ve
l
e
t
-
L
H
_gl
c
m
_S
um
E
nt
r
opy
17
w
a
ve
l
e
t
-
H
H
_gl
c
m
_I
m
c
2
6
w
a
ve
l
e
t
-
H
H
_gl
r
l
m
_R
unL
e
ngt
hN
onU
ni
f
or
m
i
t
yN
or
m
a
l
i
z
e
d
18
w
a
ve
l
e
t
-
L
H
_gl
c
m
_I
d
7
w
a
ve
l
e
t
-
L
H
_gl
r
l
m
_R
unL
e
ngt
hN
onU
ni
f
or
m
i
t
yN
or
m
a
l
i
z
e
d
19
w
a
ve
l
e
t
-
L
H
_gl
c
m
_I
dn
8
w
a
ve
l
e
t
-
L
H
_s
t
d
20
w
a
ve
l
e
t
-
L
H
_gl
c
m
_I
dm
n
9
w
a
ve
l
e
t
-
H
H
_e
ne
r
gy
21
w
a
ve
l
e
t
-
L
H
_gl
c
m
_I
dm
10
w
a
ve
l
e
t
-
H
H
_s
t
d
22
w
a
ve
l
e
t
-
L
H
_gl
c
m
_I
m
c
1
11
w
a
ve
l
e
t
-
H
H
_m
e
a
n
23
w
a
ve
l
e
t
-
L
H
_gl
c
m
_I
m
c
2
12
w
a
ve
l
e
t
-
H
H
_gl
c
m
_I
d
A
lg
or
it
hm
1.
S
ta
bi
li
ty
s
e
le
c
ti
on
i)
I
nput
:
s
ta
r
t
w
it
h t
he
m
e
r
ge
d da
ta
s
e
ts
X
(
f
e
a
tu
r
e
s
)
a
nd y (
la
be
l
s
)
.
ii)
S
ubs
a
m
pl
in
g:
a
r
a
ndom ha
lf
of
t
he
da
ta
s
e
t
(
s
e
t
to
50%
)
i
s
s
e
le
c
t
e
d f
or
e
a
c
h i
te
r
a
ti
on.
iii)
M
ode
l
tr
a
in
in
g:
a
n
M
L
m
ode
l
(
X
G
B
oos
t)
is
tr
a
in
e
d
on
th
e
s
ub
s
a
m
pl
e
w
it
h
f
ix
e
d
pa
r
a
m
e
te
r
s
(
e
.g., 50 r
ounds
, de
pt
h 3)
.
iv
)
F
e
a
tu
r
e
i
m
por
ta
nc
e
:
a
f
te
r
t
r
a
in
in
g, t
he
m
ode
l
f
e
a
tu
r
e
s
a
r
e
c
ol
le
c
te
d.
v)
T
hr
e
s
hol
di
ng:
a
lo
c
a
l
th
r
e
s
hol
d i
s
c
om
put
e
d (
s
e
t
a
t
20%
of
t
he
m
a
xi
m
um
i
m
por
ta
nc
e
i
n t
hi
s
s
tu
dy)
.
vi
)
B
in
a
r
y
m
a
s
k:
a
bi
na
r
y
ve
c
to
r
is
c
r
e
a
te
d
th
a
t
e
va
lu
a
te
s
a
s
e
le
c
t
e
d
f
e
a
tu
r
e
a
s
1
if
it
e
xc
e
e
d
s
th
e
th
r
e
s
hol
d
a
nd 0 othe
r
w
is
e
.
vi
i)
R
e
pe
a
t:
t
hi
s
pr
oc
e
s
s
is
r
e
p
e
a
te
d
f
or
m
ul
ti
pl
e
it
e
r
a
ti
ons
(
e
.g.,
100
ti
m
e
s
)
,
a
nd
a
ll
bi
na
r
y
ve
c
to
r
s
a
r
e
a
c
c
um
ul
a
te
d.
vi
ii
)
F
in
a
l
s
e
le
c
ti
on:
t
he
s
e
le
c
ti
on
f
r
e
que
nc
y
f
o
r
e
a
c
h
f
e
a
tu
r
e
is
c
om
put
e
d,
a
nd
th
os
e
th
a
t
c
ons
is
te
nt
ly
a
ppe
a
r
a
r
e
r
a
nke
d a
nd s
e
le
c
te
d.
T
a
bl
e
2
pr
e
s
e
nt
s
a
li
s
t
of
23
di
s
c
r
im
in
a
ti
ve
f
e
a
tu
r
e
s
(
out
of
th
e
to
p
30)
id
e
nt
if
ie
d
us
in
g
th
e
s
ta
bi
li
ty
s
e
le
c
ti
on
a
lg
or
it
hm
.
T
h
e
f
ir
s
t
s
e
v
e
n
f
e
a
tu
r
e
s
a
r
e
in
di
c
a
to
r
s
of
th
e
di
s
or
de
r
or
s
pr
e
a
d
in
th
e
w
a
ve
le
t
s
ub
-
ba
nd
s
.
T
he
ne
xt
f
our
f
e
a
tu
r
e
s
a
r
e
r
e
la
te
d
to
th
e
id
e
nt
if
ic
a
ti
on
o
f
s
tr
o
ng
e
dge
s
in
th
e
in
f
e
c
te
d
f
ie
ld
.
R
e
c
ogni
z
in
g
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
P
ne
um
oni
a c
la
s
s
if
ic
at
io
n f
r
om
c
he
s
t
x
-
r
ay
s
us
in
g
s
ig
ni
fi
c
ant
f
e
at
ur
e
s
e
le
c
ti
on …
(
Y
ugandhar
C
hodagam)
597
pa
tt
e
r
n,
w
e
lo
oke
d
f
or
ot
he
r
de
s
c
r
ip
to
r
s
th
a
t
f
ol
lo
w
th
e
s
a
m
e
m
a
th
e
m
a
ti
c
a
l
pr
in
c
ip
le
s
but
w
e
r
e
not
in
c
lu
de
d
in
th
e
or
ig
in
a
l
li
s
t
o
f
955
f
e
a
tu
r
e
s
.
T
hi
s
w
a
s
ne
c
e
s
s
a
r
y
be
c
a
u
s
e
th
e
a
c
c
ur
a
c
y
of
im
a
ge
c
la
s
s
if
ic
a
ti
on
in
th
e
J
a
ga
nna
th
da
ta
s
e
t
w
a
s
j
us
t
a
r
ound 91%
.
T
he
r
e
f
or
e
,
w
e
s
y
s
te
m
a
ti
c
a
ll
y
s
e
a
r
c
he
d
f
or
a
ddi
ti
ona
l
f
e
a
tu
r
e
s
t
o
im
pr
ove
th
e
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y.
W
e
f
ound
th
a
t
G
r
a
di
e
nt
_e
nt
r
opy_me
a
n
is
s
im
il
a
r
to
ot
he
r
e
nt
r
opy
m
e
a
s
ur
e
s
,
e
xc
e
pt
th
a
t
it
is
c
a
lc
ul
a
te
d
ove
r
th
e
S
obe
l
gr
a
di
e
nt
m
a
gni
tu
de
hi
s
to
gr
a
m
.
I
n
a
ddi
ti
on,
S
obe
l_
e
dge
_m
e
a
n
c
a
pt
ur
e
s
th
e
a
ve
r
a
ge
gr
a
di
e
nt
m
a
gni
tu
de
,
w
he
r
e
a
s
F
r
a
ngi
_m
e
a
n
e
m
pha
s
i
z
e
s
e
lo
nga
te
d
r
id
ge
-
li
ke
s
tr
uc
tu
r
e
s
.
I
nc
or
por
a
ti
ng
G
r
a
di
e
nt
_e
nt
r
opy_me
a
n
a
nd
F
r
a
ngi
_m
e
a
n
in
to
ou
r
m
ode
l
in
c
r
e
a
s
e
d
th
e
a
c
c
ur
a
c
y
of
im
a
ge
c
la
s
s
if
ic
a
ti
on
in
th
e
J
a
ga
nna
th
da
ta
s
e
t
to
94%
.
E
xc
e
pt
f
or
G
r
a
di
e
nt
_e
nt
r
opy_m
e
a
n,
S
obe
l_
e
dge
_m
e
a
n,
a
nd
F
r
a
ngi
_m
e
a
n,
a
ll
th
e
s
e
le
c
te
d
f
e
a
tu
r
e
s
a
r
e
c
a
lc
ul
a
te
d
a
f
te
r
pe
r
f
or
m
in
g
w
a
ve
le
t
tr
a
ns
f
or
m
a
ti
on.
T
he
w
a
ve
le
t
f
e
a
tu
r
e
s
a
r
e
s
pe
c
if
ic
a
ll
y
f
r
om
th
e
L
H
a
nd
H
H
s
ub
-
ba
nd
s
a
nd
pr
im
a
r
il
y
f
r
om
two
ty
pe
s
of
te
xt
ur
e
s
:
gr
a
y
le
ve
l
c
o
-
oc
c
ur
r
e
nc
e
m
a
tr
ix
a
nd gr
a
y l
e
ve
l
r
un l
e
ngt
h m
a
tr
ix
.
U
s
in
g
t
-
di
s
tr
ib
ut
e
d
s
to
c
ha
s
ti
c
ne
ig
hbor
e
m
be
ddi
ng,
w
e
p
r
oj
e
c
te
d
th
e
to
p
w
a
ve
le
t
f
e
a
tu
r
e
s
f
r
om
th
e
G
W
C
M
C
da
ta
s
e
t
in
to
a
two
-
di
m
e
ns
io
na
l
s
pa
c
e
a
s
pr
e
s
e
nt
e
d
i
n
F
ig
ur
e
5.
T
h
e
bl
ue
a
nd
r
e
d
poi
nt
s
r
e
pr
e
s
e
nt
c
he
s
t
X
-
r
a
ys
be
lo
ngi
ng
to
th
e
nor
m
a
l
a
nd
pne
um
oni
a
c
a
te
gor
ie
s
,
r
e
s
pe
c
ti
ve
ly
.
W
e
c
a
n
c
le
a
r
ly
obs
e
r
ve
th
a
t
th
e
s
e
le
c
te
d f
e
a
tu
r
e
s
r
obus
tl
y c
a
pt
ur
e
t
he
di
s
c
r
im
in
a
ti
ve
s
tr
uc
tu
r
e
.
F
ig
ur
e
5.
T
-
di
s
tr
ib
ut
e
d s
to
c
ha
s
ti
c
ne
ig
hbor
e
m
be
ddi
ng pr
oj
e
c
ti
o
n of
t
he
t
op w
a
ve
le
t
f
e
a
tu
r
e
s
2.3.
F
e
at
u
r
e
u
n
d
e
r
s
t
an
d
in
g
A
nom
a
li
e
s
or
le
s
io
ns
a
r
e
ty
pi
c
a
ll
y
lo
c
a
li
z
e
d
in
m
e
di
c
a
l
im
a
g
e
s
,
a
nd
te
c
hni
c
ia
ns
,
r
a
di
ol
ogi
s
ts
,
a
nd
phys
ic
ia
ns
f
oc
us
on
th
os
e
r
e
gi
ons
to
id
e
nt
if
y
pa
th
ol
og
ic
a
l
bi
o
m
a
r
ke
r
s
.
S
e
ve
r
a
l
te
c
hni
que
s
a
r
e
e
m
pl
oye
d
in
m
a
c
hi
ne
vi
s
io
n
–
ba
s
e
d
di
a
gnos
ti
c
s
,
s
u
c
h
a
s
f
a
s
t
f
our
ie
r
tr
a
ns
f
or
m
s
,
G
a
bor
f
il
te
r
s
,
a
nd
w
a
ve
le
t
tr
a
ns
f
or
m
s
.
W
a
ve
le
t
tr
a
ns
f
or
m
s
a
r
e
th
e
m
os
t
w
id
e
ly
a
dopt
e
d
be
c
a
us
e
th
e
y
c
a
pt
ur
e
bot
h
f
r
e
que
nc
y
a
nd
s
pa
ti
a
l
in
f
or
m
a
ti
on,
w
he
r
e
a
s
f
a
s
t
f
ou
r
ie
r
t
r
a
ns
f
or
m
s
c
a
pt
ur
e
onl
y
th
e
gl
oba
l
f
r
e
que
nc
y
c
ont
e
nt
by
de
c
om
pos
in
g
s
ig
na
ls
i
nt
o s
in
e
a
nd
c
os
in
e
c
om
pone
nt
s
[
16]
, [
17]
.
F
ir
s
t
-
or
de
r
s
ta
ti
s
ti
c
s
e
xt
r
a
c
te
d
f
r
om
th
e
r
a
w
pi
xe
ls
of
a
n
im
a
ge
do
not
pr
ovi
de
th
e
d
e
ta
il
e
d
in
f
or
m
a
ti
on
r
e
qui
r
e
d
f
or
im
a
ge
c
la
s
s
if
ic
a
ti
on
be
c
a
us
e
th
e
y
la
c
k
s
pa
ti
a
l
a
nd
f
r
e
que
nc
y
c
ont
e
xt
s
.
R
a
w
pi
xe
ls
a
r
e
f
la
t
di
s
tr
ib
ut
io
ns
of
in
te
n
s
it
ie
s
a
nd
f
ir
s
t
-
or
de
r
s
ta
ti
s
ti
c
s
on
s
uc
h
a
di
s
tr
ib
ut
io
n
ig
nor
e
e
dge
s
,
te
xt
ur
e
s
,
a
nd
f
in
e
-
s
c
a
le
pa
tt
e
r
ns
,
f
e
a
tu
r
e
s
th
a
t
a
r
e
c
r
it
ic
a
l
f
or
m
e
di
c
a
l
im
a
ge
-
ba
s
e
d
di
a
gno
s
is
.
I
s
ol
a
ti
ng
th
e
f
in
e
e
dg
e
s
,
te
xt
ur
e
s
,
or
s
ubt
le
va
r
ia
ti
ons
in
pi
xe
l
in
te
ns
it
y
is
ke
y
to
de
te
c
ti
ng
a
ny
a
bnor
m
a
li
ti
e
s
or
pa
th
ol
ogi
e
s
in
a
n
im
a
ge
.
W
a
ve
le
ts
de
c
om
pos
e
th
e
im
a
ge
in
to
a
s
e
t
o
f
s
ub
-
ba
nds
us
in
g
w
a
ve
le
ts
or
w
a
ve
-
li
ke
os
c
il
la
ti
ons
.
V
a
r
io
us
f
r
e
que
nc
ie
s
a
nd
or
ie
nt
a
ti
ons
a
r
e
c
a
pt
ur
e
d
by
e
a
c
h
s
ub
-
ba
nd,
a
nd
s
om
e
of
th
e
s
e
d
e
ta
il
s
m
a
y
c
a
pt
ur
e
th
e
obs
e
r
ve
d
s
ig
na
l
[
18]
.
T
he
im
a
ge
i
s
f
ir
s
t
de
c
om
pos
e
d
a
lo
ng
r
ow
s
a
nd
th
e
n
a
lo
ng
c
ol
um
ns
,
pr
oduc
in
g
f
our
s
ub
-
ba
nds
:
L
L
,
w
hi
c
h
c
a
pt
ur
e
s
th
e
c
oa
r
s
e
s
tr
uc
tu
r
e
of
th
e
im
a
g
e
;
L
H
,
w
hi
c
h
c
a
pt
ur
e
s
th
e
ve
r
ti
c
a
l
e
dge
s
;
H
L
,
w
hi
c
h
c
a
pt
ur
e
s
th
e
hor
iz
ont
a
l
e
dge
s
;
a
nd
H
H
,
w
hi
c
h
c
a
pt
ur
e
s
th
e
di
a
gona
l
e
dge
s
a
nd
noi
s
e
.
T
he
lo
w
-
pa
s
s
f
il
te
r
s
r
e
m
ove
hi
gh
-
f
r
e
que
nc
y
de
ta
il
s
s
uc
h a
s
e
dge
s
a
nd
noi
s
e
, w
he
r
e
a
s
th
e
hi
gh
-
pa
s
s
f
il
te
r
s
e
m
pha
s
iz
e
e
dge
s
,
bounda
r
ie
s
, a
nd f
in
e
de
ta
il
s
by r
e
m
ovi
ng t
he
l
ow
-
f
r
e
que
nc
y, s
m
oot
h ba
c
kgr
ound.
2.4.
F
e
at
u
r
e
e
xp
la
n
at
io
n
B
u
i
l
d
i
ng
o
n
t
he
ob
s
e
r
v
a
t
io
n
s
m
e
n
t
i
o
ne
d
a
b
o
ve
,
w
e
c
r
e
a
te
d
a
P
y
t
h
o
n
pa
c
ka
g
e
t
o
e
x
t
r
a
c
t
o
n
l
y
t
he
2
6
a
f
o
r
e
m
e
n
t
i
o
ne
d
f
e
a
tu
r
e
s
f
r
o
m
t
he
i
m
a
ge
s
a
n
d
c
la
s
s
i
f
y
t
h
e
i
m
a
g
e
s
us
i
n
g
t
h
e
X
G
B
oo
s
t
a
l
g
o
r
i
t
h
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
15
, N
o.
1
,
F
e
br
ua
r
y
20
26
:
592
-
603
598
(
F
i
g
u
r
e
4
)
.
O
t
he
r
a
l
g
o
r
i
th
m
s
s
u
c
h
a
s
r
a
n
d
o
m
f
o
r
e
s
t
w
e
r
e
a
l
s
o
t
e
s
t
e
d
,
b
u
t
X
G
B
oo
s
t
pe
r
f
o
r
m
e
d
s
l
i
g
ht
l
y
b
e
t
t
e
r
.
T
h
i
s
m
i
r
r
o
r
s
t
h
e
f
i
n
d
i
n
gs
i
n
l
un
g
d
is
e
a
s
e
i
m
a
g
i
ng
s
t
u
di
e
s
w
he
r
e
X
G
B
o
os
t
a
c
h
ie
v
e
d
h
i
gh
e
r
a
c
c
u
r
a
c
y
a
n
d
b
e
tt
e
r
o
v
e
r
a
l
l
p
e
r
f
o
r
m
a
n
c
e
m
e
t
r
i
c
s
t
h
a
n
r
a
n
d
om
f
o
r
e
s
t
[
1
9
]
.
P
n
e
um
o
w
a
v
e
p
a
c
ka
g
e
e
x
t
r
a
c
ts
t
h
e
2
6
f
e
a
t
u
r
e
s
a
f
te
r
m
i
n
i
m
a
l
p
r
e
p
r
oc
e
s
s
in
g
o
f
im
a
ge
s
a
n
d
a
l
s
o
do
e
s
t
he
c
la
s
s
i
f
ic
a
t
i
on
o
f
i
m
a
ge
s
a
n
d
t
h
e
f
l
o
w
c
ha
r
t
i
s
d
e
p
i
c
te
d
i
n
F
i
g
u
r
e
6
.
F
ig
ur
e
6.
P
ne
um
ow
a
ve
pa
c
ka
ge
us
e
d t
o e
xt
r
a
c
t
26 f
e
a
tu
r
e
s
a
nd
pe
r
f
or
m
i
m
a
ge
c
la
s
s
if
ic
a
ti
on
S
om
e
of
t
he
s
ig
ni
f
ic
a
nt
f
e
a
tu
r
e
s
a
r
e
de
s
c
r
ib
e
d
a
s
f
ol
lo
w
s
:
i)
w
a
ve
le
t
-
L
H
_gl
c
m
_C
lu
s
te
r
T
e
nd
e
nc
y:
c
lu
s
t
e
r
te
nde
nc
y
qu
a
nt
if
ie
s
th
e
e
xt
e
nt
to
w
hi
c
h
a
pa
ir
of
pi
xe
l
s
w
it
h s
im
il
a
r
i
nt
e
ns
it
ie
s
gr
oup toge
th
e
r
i
n a
s
m
a
ll
r
e
gi
on a
t
a
s
p
e
c
if
ic
di
s
ta
nc
e
a
nd or
ie
nt
a
ti
on.
=
∑
∑
(
+
−
−
)
2
=
1
=
1
(
,
)
(
5)
W
he
r
e
de
not
e
s
th
e
m
e
a
n
gr
a
y
-
le
ve
l
of
th
e
s
ub
-
ba
nd
,
(
,
)
de
not
e
s
nor
m
a
li
z
e
d
gr
a
y
-
le
ve
l
co
-
oc
c
ur
r
e
nc
e
m
a
tr
ix
va
lu
e
,
a
nd
μ
,
μ
de
not
e
s
m
e
a
n
gr
a
y
-
le
ve
l
of
th
e
r
e
f
e
r
e
nc
e
pi
xe
ls
,
ne
ig
hbor
pi
xe
ls
r
e
s
pe
c
ti
ve
ly
.
ii)
w
a
ve
le
t
-
H
H
_gl
r
lm
_R
unE
nt
r
opy:
e
nt
r
opy
is
a
m
e
a
s
ur
e
of
th
e
unpr
e
di
c
ta
bi
li
ty
or
r
a
ndomne
s
s
of
a
s
ys
te
m
.
R
un
e
nt
r
opy
d
e
s
c
r
ib
e
s
th
e
di
s
tr
ib
ut
io
n
of
r
uns
f
or
s
e
que
nc
e
s
of
c
ons
e
c
ut
iv
e
pi
x
e
ls
w
it
h
th
e
s
a
m
e
in
te
ns
it
y.
T
hi
s
is
u
s
e
f
ul
f
or
de
te
c
ti
ng
r
e
gi
ons
w
it
h
di
s
to
r
ti
on,
w
hi
c
h
m
ig
ht
in
di
c
a
te
a
pa
th
ol
ogy.
T
he
s
e
n
s
it
iv
it
y of
w
a
ve
le
t
-
L
H
_gl
r
lm
_R
unE
nt
r
opy s
hi
f
ts
t
ow
a
r
d
hor
iz
ont
a
l
s
tr
uc
tu
r
e
s
.
=
−
∑
∑
(
,
)
=
1
(
,
)
=
1
(
6)
W
he
r
e
de
not
e
s
num
be
r
of
gr
a
y
le
ve
l
s
a
f
te
r
you
qua
nt
iz
e
th
e
i
m
a
ge
,
de
not
e
s
m
a
xi
m
um
r
un
le
ngt
h
a
nd
(
,
)
de
not
e
s
t
he
pr
oba
bi
li
ty
of
e
nc
ount
e
r
in
g a
r
un of
l
e
ngt
h j
a
t
gr
a
y
-
le
ve
l
.
iii)
w
a
ve
le
t
-
L
H
_gl
r
lm
_S
hor
tR
unE
m
pha
s
is
(
S
R
E
)
:
S
R
E
id
e
nt
if
ie
s
f
in
e
-
gr
a
in
e
d
te
xt
ur
e
pa
tt
e
r
ns
,
s
uc
h
a
s
m
ic
r
ot
e
xt
ur
a
l
c
ha
nge
s
a
nd s
ubt
le
e
dge
s
, by
a
s
s
ig
ni
ng highe
r
w
e
ig
ht
s
t
o s
hor
te
r
r
uns
.
ℎ
ℎ
=
1
∑
∑
(
,
)
2
=
1
=
1
(
7)
W
he
r
e
de
not
e
s
numbe
r
of
gr
a
y l
e
ve
ls
a
f
te
r
you qua
nt
iz
e
t
he
i
m
a
ge
,
de
not
e
s
m
a
xi
m
um
r
un l
e
ngt
h
,
a
nd
2
de
not
e
s
s
qu
a
r
e
d r
un l
e
ngt
h i
n t
he
de
nom
in
a
to
r
, gi
vi
ng s
hor
t
r
uns
a
l
a
r
ge
r
w
e
ig
ht
t
ha
n l
ong r
uns
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
P
ne
um
oni
a c
la
s
s
if
ic
at
io
n f
r
om
c
he
s
t
x
-
r
ay
s
us
in
g
s
ig
ni
fi
c
ant
f
e
at
ur
e
s
e
le
c
ti
on …
(
Y
ugandhar
C
hodagam)
599
iv
)
w
a
ve
le
t
-
L
H
_gl
c
m
_S
um
E
nt
r
opy:
s
um
e
nt
r
opy
is
s
im
il
a
r
to
r
un
e
nt
r
opy
e
xc
e
pt
th
a
t
it
is
th
e
s
um
of
gr
a
y
-
le
ve
l
va
lu
e
s
of
ne
ig
hbor
in
g pi
xe
ls
, w
hi
c
h c
a
pt
ur
e
s
t
h
e
t
e
xt
ur
e
c
om
pl
e
xi
ty
i
n t
he
i
m
a
ge
.
=
−
∑
+
(
)
2
=
2
+
(
)
(
8)
W
he
r
e
de
not
e
s
num
be
r
of
gr
a
y
le
ve
ls
a
f
te
r
you
qua
nt
iz
e
th
e
im
a
ge
,
a
nd
+
(
)
de
not
e
s
s
um
di
s
tr
ib
ut
io
n of
t
he
G
L
C
M
.
v)
w
a
ve
le
t
-
H
H
_gl
r
lm
_R
unL
e
ngt
hN
onU
ni
f
or
m
it
yN
or
m
a
li
z
e
d
(
R
L
N
U
N
)
:
r
un
le
ngt
h
non
-
uni
f
or
m
it
y
nor
m
a
li
z
e
d
m
e
a
s
ur
e
s
th
e
va
r
ia
ti
on
in
th
e
le
ngt
h
s
of
pi
xe
l
r
uns
in
a
n
im
a
ge
.
I
t
te
ll
s
us
w
he
th
e
r
th
e
r
uns
(
i.
e
.,
s
e
que
nc
e
s
of
pi
xe
ls
w
it
h
th
e
s
a
m
e
in
te
n
s
it
y)
a
r
e
m
os
tl
y
of
th
e
s
a
m
e
le
ngt
h
or
s
pr
e
a
d
a
c
r
os
s
di
f
f
e
r
e
nt
le
ngt
hs
.
T
he
w
a
ve
le
t
-
L
H
_gl
r
lm
_R
unL
e
ngt
hN
onU
ni
f
or
m
it
yN
or
m
a
li
z
e
d
f
e
a
tu
r
e
c
a
pt
ur
e
s
th
e
s
a
m
e
m
e
a
s
ur
e
m
e
nt
f
or
t
he
L
H
s
ub
-
ba
nd.
=
1
∑
(
∑
(
,
)
=
1
)
2
=
1
(
9)
W
he
r
e
de
not
e
s
numbe
r
of
gr
a
y l
e
ve
ls
a
f
te
r
you qua
nt
iz
e
t
he
i
m
a
ge
,
de
not
e
s
m
a
xi
m
um
r
un l
e
ngt
h
,
a
nd
(
,
)
de
not
e
s
t
he
pr
oba
bi
li
ty
of
e
nc
ount
e
r
in
g a
r
un of
l
e
ngt
h j
a
t
gr
a
y
-
le
ve
l
.
vi
)
w
a
ve
le
t
-
L
H
_s
td
(
S
T
D
)
:
S
T
D
qua
nt
if
ie
s
how
w
id
e
ly
pi
xe
l
va
lu
e
s
de
vi
a
te
f
r
om
th
e
m
e
a
n,
m
a
ki
ng
i
t
s
e
ns
it
iv
e
t
o l
oc
a
l
va
r
ia
ti
ons
i
n br
ig
ht
ne
s
s
a
nd t
e
xt
ur
e
.
σ
=
√
1
∑
(
−
μ
)
2
=
1
(
10)
W
he
r
e
σ
de
not
e
s
s
ta
nda
r
d de
vi
a
ti
on
,
de
not
e
s
t
he
i
th
obs
e
r
va
ti
on i
n t
he
s
a
m
pl
e
, a
nd
μ
de
not
e
s
a
ve
r
a
g
e
.
vi
i)
w
a
ve
le
t
-
H
H
_e
ne
r
gy:
e
ne
r
gy s
e
r
ve
s
a
s
a
n
in
di
c
a
to
r
of
th
e
pr
e
s
e
nc
e
a
nd
m
a
gni
tu
de
of
de
t
a
il
e
d
s
tr
uc
tu
r
e
s
,
s
uc
h a
s
f
in
e
e
dg
e
s
,
s
pe
c
kl
e
noi
s
e
, or
t
e
xt
ur
a
l
pa
tt
e
r
ns
.
=
∑
2
=
1
(
11)
W
he
r
e
N
de
not
e
s
t
he
t
ot
a
l
num
b
e
r
of
s
a
m
pl
e
s
, a
nd
de
not
e
s
t
he
i
th
obs
e
r
va
ti
on i
n t
he
s
a
m
pl
e
.
vi
ii
)
w
a
ve
le
t
-
H
H
_m
e
a
n:
m
e
a
n
pr
ovi
de
s
in
f
or
m
a
ti
on
a
bout
th
e
o
ve
r
a
ll
br
ig
ht
ne
s
s
or
s
ig
na
l
le
ve
l
in
th
e
di
a
gona
l
te
xt
ur
e
r
e
gi
ons
.
μ
=
1
∑
=
1
(
12)
W
he
r
e
μ
de
ont
e
s
t
he
a
ve
r
a
ge
a
nd
de
not
e
s
t
he
i
th
obs
e
r
va
ti
on i
n t
he
s
a
m
pl
e
.
ix
)
gr
a
di
e
nt
_e
nt
r
opy_me
a
n
(
G
E
M
)
:
G
E
M
c
a
pt
ur
e
s
th
e
unpr
e
di
c
t
a
bi
li
ty
of
gr
a
di
e
nt
c
ha
nge
s
in
th
e
im
a
ge
.
C
he
s
t
X
-
r
a
y
s
of
lu
ngs
w
it
h
pne
um
oni
a
pr
e
s
e
nt
di
s
or
de
r
e
d
te
x
tu
r
e
s
a
nd
une
ve
n
tr
a
ns
it
io
ns
,
w
hi
c
h
a
r
e
de
te
c
te
d w
e
ll
by t
hi
s
f
e
a
tu
r
e
.
=
1
⁄
·
∑
(
−
∑
ᵢ
(
)
·
2
(
ᵢ
(
)
)
)
=
1
(
13)
W
he
r
e
N
d
e
not
e
s
th
e
to
ta
l
num
b
e
r
of
ne
ig
hbor
hoods
ov
e
r
w
hi
c
h
e
nt
r
opy
is
a
ve
r
a
ge
d
a
nd
ᵢ
(
)
de
not
e
s
nor
m
a
li
z
e
d hi
s
to
gr
a
m
bi
n
.
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
he
G
W
C
M
C
da
ta
s
e
t
w
a
s
our
pr
im
a
r
y
da
ta
s
e
t
f
or
im
a
ge
c
la
s
s
if
ic
a
ti
on
a
s
s
how
n
in
T
a
bl
e
3.
W
e
na
m
e
d
th
e
pa
c
ka
g
e
w
e
d
e
s
ig
ne
d
a
s
P
ne
um
ow
a
ve
.
P
ne
um
ow
a
ve
in
c
lu
de
d
m
in
im
a
l
pr
e
pr
oc
e
s
s
in
g
of
da
ta
(
im
a
ge
s
w
e
r
e
r
e
s
iz
e
d
to
224×
224
a
nd
pi
xe
l
in
te
ns
it
ie
s
w
e
r
e
no
r
m
a
li
z
e
d)
,
e
xt
r
a
c
ti
on
of
26
f
e
a
tu
r
e
s
(
ba
s
e
d
on
th
e
f
e
a
tu
r
e
s
id
e
nt
if
ie
d
by
th
e
s
ta
bi
li
ty
s
e
l
e
c
ti
on
a
lg
or
it
hm
a
nd
m
or
e
)
,
m
ode
l
bui
ld
in
g
(
X
G
B
oos
t
w
it
h
m
a
x_de
pt
h:
6, l
e
a
r
ni
ng_r
a
te
:
0.05, n_e
s
ti
m
a
to
r
s
:
300)
, a
nd t
e
s
ti
ng. T
he
pa
c
ka
g
e
a
c
hi
e
ve
d a
n
a
c
c
ur
a
c
y of
98%
a
nd
a
s
e
ns
it
iv
it
y
of
99%
on
th
e
G
W
C
M
C
d
a
ta
s
e
t.
T
h
e
to
p
11
o
f
th
e
26
f
e
a
tu
r
e
s
li
s
te
d
in
T
a
bl
e
2
w
e
r
e
us
e
d
in
th
e
m
ode
l.
R
e
s
e
a
r
c
h
e
r
s
ha
ve
w
id
e
ly
u
s
e
d
th
e
G
W
C
M
C
da
t
a
s
e
t
f
or
pne
um
oni
a
de
te
c
ti
on.
T
he
de
e
p
C
N
N
m
ode
l
bui
lt
by
T
a
ng
e
t
al
.
[
20]
de
m
ons
tr
a
te
d
a
n
a
c
c
ur
a
c
y
of
94.6
4%
,
s
e
ns
it
iv
it
y
of
96.5
%
,
a
nd
s
pe
c
if
ic
it
y
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
15
, N
o.
1
,
F
e
br
ua
r
y
20
26
:
592
-
603
600
92.86%
.
W
a
ng
e
t
al
.
[
21]
e
s
ta
bl
is
h
e
d
a
D
L
pi
pe
li
ne
th
a
t
a
c
h
ie
ve
d
a
n
A
U
C
in
th
e
r
a
nge
of
0.94
–
0.98
f
or
pne
um
oni
a
c
la
s
s
if
ic
a
ti
on.
P
r
a
ka
s
h
e
t
al
.
[
22]
pr
opos
e
d
tr
a
n
s
f
e
r
le
a
r
ni
ng
w
it
h
s
ta
c
ke
d
C
N
N
s
,
a
c
hi
e
vi
ng
96.15%
a
c
c
ur
a
c
y,
97.90%
pr
e
c
is
io
n,
a
nd
95.90%
r
e
c
a
ll
.
G
upt
a
e
t
al
.
[
23]
us
e
d
ne
ur
a
l
a
r
c
hi
te
c
tu
r
e
s
e
a
r
c
h
a
nd
r
e
por
te
d
a
good
A
U
C
of
0.976,
but
th
e
ir
a
ppr
oa
c
h
r
e
qui
r
e
d
hi
gh
c
om
put
a
ti
ona
l
e
f
f
or
t.
S
ha
ti
e
t
al
.
[
24
]
c
om
bi
ne
d
w
a
ve
le
t
s
,
gr
a
y
le
ve
l
c
o
-
oc
c
ur
r
e
nc
e
m
a
tr
ix
,
a
nd
k
-
ne
a
r
e
s
t
ne
ig
hbor
s
to
a
c
hi
e
ve
a
hi
gh
A
U
C
a
nd
a
n
a
c
c
ur
a
c
y
of
ne
a
r
ly
99%
;
how
e
ve
r
,
th
e
y
us
e
d
a
ve
r
y
c
om
pl
e
x
m
ode
l.
S
in
gh
e
t
a
l.
[
25
]
a
ppl
ie
d
vi
s
io
n
tr
a
ns
f
or
m
e
r
s
,
a
c
hi
e
vi
ng
a
hi
gh
a
c
c
ur
a
c
y
of
97.61%
a
nd
a
n
A
U
C
of
0.96.
R
a
gha
w
e
t
al
.
[
26]
us
e
d
di
la
te
d
c
onvolut
io
n
a
nd
tr
a
ns
f
or
m
e
r
s
,
obt
a
in
in
g
a
n
a
c
c
ur
a
c
y
of
9
2.87%
,
pr
e
c
is
io
n
of
91.72%
,
a
nd
r
e
c
a
ll
of
a
ppr
oxi
m
a
te
ly
91.59%
.
H
ow
e
ve
r
,
bot
h
th
e
s
e
a
ppr
oa
c
h
e
s
w
e
r
e
c
om
pl
e
x
a
nd
c
om
put
a
ti
ona
ll
y
in
te
ns
iv
e
.
O
ur
ow
n
D
L
m
ode
l
ba
s
e
d
on
D
e
n
s
e
N
e
t
-
121
yi
e
ld
e
d
a
n
a
c
c
ur
a
c
y
o
f
98.2%
w
it
h
m
in
im
a
l
pr
e
pr
oc
e
s
s
in
g,
r
unni
ng
f
or
a
ppr
oxi
m
a
te
ly
8 hour
s
on a
C
P
U
-
ba
s
e
d s
ys
te
m
.
T
a
bl
e
3. D
a
ta
s
e
ts
u
s
e
d i
n t
hi
s
s
tu
dy
D
a
t
a
s
e
t
S
our
c
e
I
m
a
ge
c
ount
C
om
pi
l
e
d by
G
W
C
M
C
G
ua
ngz
hou W
om
e
n a
nd
C
hi
l
dr
e
n’
s
M
e
di
c
a
l
C
e
nt
e
r
, G
ua
ng
z
hou, C
hi
na
5,856
D
a
ni
e
l
e
t
al
.
;
ht
t
ps
:
/
/
da
t
a
.m
e
nde
l
e
y.c
om
/
da
t
a
s
e
t
s
/
r
s
c
bj
br
9s
j
/
2
ht
t
ps
:
/
/
w
w
w
.c
e
l
l
.c
om
/
c
e
l
l
/
f
ul
l
t
e
xt
/
S
0092
-
8674(
18)
30154
-
5
R
A
I
G
I
t
a
l
i
a
n
S
oc
i
e
t
y
of
M
e
di
c
a
l
a
nd
I
nt
e
r
ve
nt
i
ona
l
R
a
di
ol
ogy (
S
I
R
M
)
2,998
R
a
di
ol
ogy A
I
G
r
oup, Q
a
t
a
r
U
ni
ve
r
s
i
t
y a
nd t
he
U
ni
ve
r
s
i
t
y of
D
ha
ka
a
l
ong w
i
t
h c
ol
l
a
bor
a
t
or
s
f
r
om
P
a
ki
s
t
a
n a
nd M
a
l
a
y
s
i
a
a
nd s
om
e
m
e
di
c
a
l
doc
t
or
s
;
ht
t
ps
:
/
/
da
t
a
.ni
a
i
d.ni
h.gov/
r
e
s
our
c
e
s
?
i
d=f
i
gs
ha
r
e
_14151854
ht
t
ps
:
/
/
w
w
w
.ka
ggl
e
.c
om
/
da
t
a
s
e
t
s
/
t
a
w
s
i
f
ur
r
a
hm
a
n/
c
ovi
d19
-
r
a
di
ogr
a
phy
-
da
t
a
ba
s
e
(
I
t
a
l
i
a
n
-
S
I
R
M
)
N
I
G
M
S
N
I
G
M
S
/
N
I
H
(
N
a
t
i
ona
l
I
ns
t
i
t
ut
e
of
G
e
ne
r
a
l
M
e
di
c
a
l
S
c
i
e
nc
e
s
)
(
ht
t
ps
:
/
/
w
w
w
.m
e
dr
xi
v.or
g/
c
ont
e
nt
/
10.1101/
2020.05.09.20096560v2)
240
N
a
t
ur
e
a
nd S
c
i
e
nt
i
f
i
c
R
e
por
t
s
;
ht
t
ps
:
/
/
gi
t
hub.c
om
/
a
bz
a
r
ga
r
/
C
O
V
I
D
-
C
l
a
s
s
i
f
i
e
r
J
a
ga
nna
t
h
U
ni
ve
r
s
i
t
y
V
a
r
i
ous
hos
pi
t
a
l
s
i
n B
a
ngl
a
de
s
h
2,050
ht
t
ps
:
/
/
da
t
a
.m
e
nde
l
e
y.c
om
/
da
t
a
s
e
t
s
/
m
4s
2j
n3c
s
b/
1
W
he
n t
he
t
op 11
f
e
a
tu
r
e
s
l
is
te
d i
n T
a
bl
e
2 w
e
r
e
us
e
d, P
ne
um
o
w
a
ve
a
c
hi
e
ve
d a
n a
c
c
ur
a
c
y of
98%
on
th
e
G
W
C
M
C
a
nd R
A
I
G
da
ta
s
e
ts
a
nd 96%
on t
he
N
I
G
M
S
da
ta
s
e
t.
T
he
pe
r
f
or
m
a
nc
e
i
s
i
ll
us
tr
a
te
d i
n c
onf
us
io
n
m
a
tr
ic
e
s
a
s
s
how
n
in
F
ig
ur
e
7.
T
he
a
c
c
ur
a
c
y
of
im
a
ge
c
la
s
s
if
ic
a
ti
on
in
th
e
N
I
G
M
S
da
ta
s
e
t
im
pr
ove
d
to
98%
w
he
n
w
a
ve
le
t
-
H
H
_gl
c
m
_I
dn
a
nd
w
a
ve
le
t
-
L
H
_gl
r
lm
_gr
a
yl
e
v
e
lNonUnif
or
m
it
yN
or
m
a
li
z
e
d
w
e
r
e
a
dde
d
a
s
f
e
a
tu
r
e
s
.
S
im
il
a
r
ly
,
th
e
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
in
th
e
J
a
g
a
nna
th
da
ta
s
e
t
in
c
r
e
a
s
e
d
up
to
94%
w
he
n
G
r
a
di
e
nt
_e
nt
r
opy_me
a
n
a
nd
F
r
a
ngi
_m
e
a
n
w
e
r
e
a
dde
d
to
th
e
to
p
11
f
e
a
tu
r
e
s
.
W
e
tr
ie
d
to
in
c
or
por
a
te
f
e
w
im
a
ge
pr
e
pr
oc
e
s
s
in
g
te
c
hni
que
s
,
in
c
lu
di
ng
de
noi
s
in
g,
c
ont
r
a
s
t
-
li
m
it
e
d
a
da
pt
iv
e
hi
s
to
gr
a
m
e
qua
li
z
a
ti
on
(
a
t
e
c
hni
que
us
e
d t
o e
nha
nc
e
l
oc
a
l
im
a
ge
c
ont
r
a
s
t)
, a
nd O
ts
u
th
r
e
s
hol
di
ng. How
e
ve
r
, t
he
y r
e
s
ul
te
d i
n
r
e
duc
e
d
pe
r
f
or
m
a
nc
e
be
c
a
us
e
t
he
s
e
m
e
th
ods
m
ig
ht
i
na
dve
r
te
nt
ly
r
e
m
ove
t
he
r
e
le
va
nt
de
ta
il
s
or
s
e
gm
e
nt
s
. I
n a
ny c
a
s
e
,
th
e
r
e
duc
e
d
a
c
c
ur
a
c
y
obs
e
r
ve
d
in
th
e
J
a
ga
nna
th
da
ta
s
e
t
c
a
n
be
pa
r
tl
y
a
tt
r
ib
ut
e
d
to
th
e
in
f
e
r
io
r
im
a
ge
qua
li
ty
,
a
n a
s
pe
c
t
e
xa
m
in
e
d i
n de
ta
il
i
n t
he
f
ol
lo
w
in
g s
e
c
ti
on.
3.1.
Q
u
al
it
y of
i
m
age
s
T
he
a
c
c
ur
a
c
y
of
im
a
g
e
c
la
s
s
if
ic
a
ti
on
de
p
e
nds
h
e
a
vi
ly
on
th
e
qua
li
ty
of
im
a
ge
s
.
I
m
a
ge
s
f
r
om
th
e
G
W
C
M
C
,
R
A
I
G
,
a
nd
N
I
G
M
S
da
ta
s
e
ts
ha
d
be
tt
e
r
qua
li
ty
th
a
n
th
os
e
f
r
om
th
e
J
a
ga
nna
th
da
ta
s
e
t.
T
he
m
e
a
n
he
ig
ht
of
im
a
ge
s
f
r
om
th
e
J
a
ga
nna
th
da
ta
s
e
t
w
a
s
299
pi
xe
ls
,
w
he
r
e
a
s
th
a
t
of
im
a
ge
s
f
r
om
th
e
ot
he
r
th
r
e
e
da
ta
s
e
ts
r
a
nge
d
f
r
om
820
to
1,472
pi
xe
ls
.
S
im
il
a
r
ly
,
th
e
w
id
th
of
im
a
ge
s
f
r
om
th
e
J
a
ga
nna
th
da
t
a
s
e
t
w
a
s
299 pixe
ls
,
w
he
r
e
a
s
t
h
a
t
of
i
m
a
ge
s
f
r
om
t
he
ot
he
r
t
hr
e
e
da
ta
s
e
ts
r
a
nge
d f
r
om
1,039 to 1,686 pi
xe
ls
.
T
he
m
e
a
n
s
iz
e
of
im
a
g
e
s
f
r
om
th
e
J
a
ga
nna
th
da
t
a
s
e
t
w
a
s
10
kB
,
w
he
r
e
a
s
th
a
t
of
im
a
ge
s
f
r
om
th
e
ot
he
r
th
r
e
e
da
ta
s
e
ts
r
a
nge
d
f
r
om
81
to
538
kB
.
T
h
e
T
e
ne
ng
r
a
d
s
ha
r
pne
s
s
of
im
a
ge
s
f
r
om
th
e
G
W
C
M
C
,
R
A
I
G
,
a
nd
N
I
G
M
S
da
ta
s
e
ts
w
a
s
0.77,
0.96,
a
nd
0.84,
r
e
s
p
e
c
ti
ve
ly
,
in
di
c
a
ti
ng
a
good
b
a
la
nc
e
be
twe
e
n
im
a
ge
s
in
th
e
nor
m
a
l
a
nd
pne
um
oni
a
c
a
t
e
gor
ie
s
.
T
he
T
e
ne
ng
r
a
d
s
ha
r
pne
s
s
of
im
a
ge
s
f
r
om
th
e
J
a
ga
nna
th
da
ta
s
e
t
w
a
s
a
s
lo
w
a
s
0.29,
in
di
c
a
ti
ng
th
a
t
nor
m
a
l
im
a
g
e
s
f
r
om
t
hi
s
da
ta
s
e
t
w
e
r
e
m
uc
h
le
s
s
s
h
a
r
p
th
a
n
im
a
ge
s
of
lu
ngs
w
it
h
pne
um
oni
a
.
T
hi
s
im
ba
la
n
c
e
c
oul
d
be
r
e
s
pons
ib
le
f
or
th
e
poor
c
la
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
,
a
s
a
ls
o note
d by P
ol
la
to
u a
nd F
e
r
r
a
nt
e
[
27]
.
3.2.
P
n
e
u
m
on
ia
c
la
s
s
if
ic
at
io
n
p
e
r
f
or
m
an
c
e
of
p
n
e
u
m
ow
ave
p
ac
k
age
T
he
f
e
a
tu
r
e
s
id
e
nt
if
ie
d
us
in
g
th
e
s
ta
bi
li
ty
s
e
le
c
ti
on
a
lg
or
it
hm
li
s
te
d
in
T
a
bl
e
2
w
e
r
e
e
xt
r
a
c
te
d
f
or
im
a
ge
s
in
th
r
e
e
ot
he
r
da
ta
s
e
ts
us
in
g
th
e
d
e
ve
lo
pe
d
P
ne
um
ow
a
ve
pa
c
ka
ge
.
T
a
bl
e
4
pr
e
s
e
nt
s
th
e
da
t
a
on
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
P
ne
um
oni
a c
la
s
s
if
ic
at
io
n f
r
om
c
he
s
t
x
-
r
ay
s
us
in
g
s
ig
ni
fi
c
ant
f
e
at
ur
e
s
e
le
c
ti
on …
(
Y
ugandhar
C
hodagam)
601
c
ount
of
im
a
ge
s
,
a
c
c
ur
a
c
y,
s
e
n
s
it
iv
it
y,
s
pe
c
if
ic
it
y,
a
nd
A
U
C
in
th
e
f
in
a
l
da
ta
s
e
t
s
.
I
n
a
ddi
ti
on
to
th
is
,
pe
r
f
or
m
a
nc
e
of
our
m
ode
l
is
e
it
he
r
c
om
pa
r
a
bl
e
to
or
,
in
s
om
e
c
a
s
e
s
,
e
ve
n
b
e
tt
e
r
th
a
n
th
e
m
os
t
a
dv
a
nc
e
d
DL
a
r
c
hi
te
c
tu
r
e
s
m
e
nt
io
ne
d i
n t
he
r
e
s
ul
ts
s
e
c
ti
on f
or
pne
um
oni
a
c
la
s
s
if
ic
a
ti
on on the
G
W
C
M
C
da
t
a
s
e
t.
(
a
)
(
b)
(
c
)
(
d)
F
ig
ur
e
7. C
onf
us
io
n m
a
tr
ic
e
s
of
(
a
)
G
W
C
M
C
, (
b)
R
A
I
G
,
(
c
)
N
I
G
M
S
, a
nd (
d)
J
a
ga
nna
th
T
a
bl
e
4.
V
a
r
io
us
d
a
ta
s
e
t
s
a
nd c
la
s
s
if
ic
a
ti
on r
e
s
ul
ts
D
a
t
a
s
e
t
T
r
a
i
n (
n)
T
e
s
t
(
n)
T
ot
a
l
A
c
c
ur
a
c
y
S
e
ns
i
t
i
vi
t
y
S
pe
c
i
f
i
c
i
t
y
AUC
G
W
C
M
C
4684
1172
5856
0.98
0.99
0.94
0.99
R
A
I
G
2398
600
2998
0.98
0.98
0.97
0.99
N
I
G
M
S
192
48
240
0.96
0.93
1
0.98
J
a
ga
nna
t
h
1640
410
2050
0.94
0.96
0.92
0.98
4.
C
O
N
C
L
U
S
I
O
N
T
hi
s
s
tu
dy
de
m
ons
tr
a
te
s
th
a
t
th
e
e
xt
r
a
c
ti
on
a
nd
s
e
le
c
ti
on
of
a
s
im
pl
e
s
e
t
of
c
us
to
m
f
e
a
tu
r
e
s
c
om
bi
ne
d
w
it
h
c
la
s
s
ic
a
l
M
L
m
e
th
ods
c
a
n
de
li
ve
r
a
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
c
om
pa
r
a
bl
e
to
th
a
t
obt
a
in
e
d
us
in
g
de
e
p
ne
ur
a
l
ne
twor
ks
.
T
hi
s
s
tu
dy
e
xt
r
a
c
te
d
f
ir
s
t
-
or
de
r
s
ta
ti
s
ti
c
s
,
te
xt
ur
e
,
a
nd
w
a
ve
l
e
t
m
e
a
s
ur
e
s
f
r
om
e
a
c
h
X
-
r
a
y
im
a
ge
.
T
w
o
s
e
l
e
c
to
r
s
w
e
r
e
te
s
t
e
d,
a
nd
s
ta
bi
li
ty
s
e
le
c
ti
o
n
pa
ir
e
d
w
it
h
X
G
B
oos
t
pr
ove
d
to
be
th
e
be
s
t
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
od.
S
ta
bi
li
ty
s
e
le
c
ti
on
id
e
nt
if
ie
d
11
f
e
a
tu
r
e
s
th
a
t
w
e
r
e
s
uf
f
ic
ie
nt
a
c
r
os
s
a
ll
f
our
da
ta
s
e
ts
a
nd
de
li
ve
r
e
d
a
c
c
ur
a
c
ie
s
of
98%
,
98%
,
96%
,
a
nd
94%
on
th
e
G
W
C
M
C
,
R
A
I
G
,
N
I
G
M
S
,
a
nd
J
a
ga
nna
th
da
ta
s
e
ts
,
r
e
s
p
e
c
ti
ve
ly
.
M
ode
l
tr
a
in
in
g
w
a
s
c
om
pl
e
te
d
in
unde
r
a
m
in
ut
e
on
s
ta
nda
r
d
ha
r
dw
a
r
e
,
a
nd
in
f
e
r
e
nc
e
r
e
qui
r
e
d
no
m
or
e
th
a
n
13
va
r
ia
bl
e
s
.
T
hi
s
a
ppr
oa
c
h
is
s
ui
ta
bl
e
f
or
s
c
e
na
r
io
s
w
he
r
e
r
e
s
our
c
e
s
a
r
e
s
e
ve
r
e
ly
li
m
it
e
d
a
nd
he
a
vy
m
ode
l
s
a
r
e
e
xpe
n
s
iv
e
to
hos
t.
T
he
r
e
s
ul
ts
w
e
r
e
s
im
il
a
r
to
th
os
e
of
r
e
c
e
nt
s
tu
di
e
s
th
a
t
c
om
bi
ne
d
c
la
s
s
ic
a
l
te
xt
ur
e
c
ue
s
w
it
h
M
L
in
s
te
a
d
of
a
f
ul
l
de
e
p
ne
twor
k.
T
hi
s
a
ppr
oa
c
h
c
a
n
be
e
xt
e
nd
e
d
to
m
ul
ti
c
la
s
s
c
la
s
s
if
ic
a
ti
on t
ha
t
in
vol
ve
s
ba
c
t
e
r
ia
l,
vi
r
a
l,
a
nd c
or
ona
vi
r
us
di
s
e
a
s
e
2019 im
a
ge
c
la
s
s
e
s
or
bi
na
r
y
Evaluation Warning : The document was created with Spire.PDF for Python.