C
omp
u
te
r
S
c
i
e
n
c
e
an
d
I
n
for
mati
on
T
e
c
h
n
ol
ogi
e
s
V
ol
.
6
,
N
o
.
3
,
N
ove
m
b
e
r
20
25
,
p
p.
315
~
325
IS
S
N
:
2722
-
3221
,
D
O
I
:
10.
1
1591
/
c
s
i
t
.
v6
i
3
.
pp
31
5
-
325
315
Jou
r
n
al
h
o
m
e
pa
ge
:
ht
t
p:
/
/
i
ae
s
pr
i
m
e
.
c
om
/
i
nd
e
x
.
php
/
c
s
i
t
Hy
b
r
i
d
f
e
a
t
u
r
e
f
u
si
o
n
f
r
o
m
mul
t
i
p
l
e
CNN
m
o
d
e
l
s with
b
a
y
e
si
a
n
-
o
p
t
i
mi
z
e
d
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
c
l
a
ssi
f
i
e
r
s
D
e
w
i
R
i
s
maw
at
i
,
S
u
gi
ya
r
to
S
u
r
on
o
,
A
r
i
s
Th
ob
i
r
i
n
D
e
p
a
rt
m
e
n
t
o
f
M
a
t
h
e
m
a
t
i
c
s
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
a
n
d
A
p
p
l
i
e
d
T
e
c
h
n
o
l
o
g
y
,
A
h
m
a
d
D
a
h
l
a
n
U
n
i
v
e
rs
i
t
y
,
Y
o
g
y
a
k
a
rt
a
,
In
d
o
n
e
s
i
a
A
r
ti
c
l
e
I
n
fo
A
BS
TR
A
C
T
Ar
t
i
c
l
e
h
i
s
t
or
y
:
Re
c
e
i
v
e
d
A
pr
11
,
202
5
Re
vi
s
e
d
J
un
21
,
2025
A
c
c
e
pt
e
d
J
u
l
4
,
2025
I
nf
or
m
a
t
i
on
t
e
c
hn
ol
o
gy
a
d
va
nc
e
m
e
n
t
s
h
a
ve
c
r
e
a
t
e
d
bi
g
d
a
t
a
,
n
e
c
e
s
s
i
t
a
t
i
ng
e
f
f
i
c
i
e
nt
t
e
c
hni
que
s
t
o
r
e
t
r
i
e
v
e
he
l
p
f
u
l
i
n
f
o
r
m
a
t
i
on
.
W
i
t
h
i
t
s
c
a
pa
c
i
t
y
t
o
r
e
c
o
gn
i
z
e
a
nd
c
a
t
e
go
r
i
z
e
p
a
t
t
e
r
ns
i
n
da
t
a
,
e
s
pe
c
i
a
l
l
y
t
he
g
r
ow
i
ng
a
m
ou
nt
o
f
pi
c
t
u
r
e
d
a
t
a
,
de
e
p
l
e
a
r
n
i
ng
i
s
b
e
c
o
m
i
ng
a
v
i
a
bl
e
op
t
i
on
.
T
h
i
s
r
e
s
e
a
r
c
h
a
i
m
s
t
o
de
v
e
l
o
p
a
m
e
di
c
a
l
i
m
a
ge
c
l
a
s
s
i
f
i
c
a
t
i
on
m
ode
l
us
i
ng
c
h
e
s
t
X
-
R
a
y
w
i
t
h
f
our
c
l
a
s
s
e
s
,
na
m
e
l
y
C
ov
i
d
-
19
,
P
ne
u
m
o
ni
a
,
T
ube
r
c
ul
o
s
i
s
,
a
n
d
N
o
r
m
a
l
.
T
he
pr
op
os
e
d
m
e
t
h
od
c
o
m
b
i
n
e
s
t
he
a
d
va
n
t
a
g
e
s
o
f
de
e
p
l
e
a
r
ni
ng
a
nd
m
a
c
hi
n
e
l
e
a
r
n
i
ng
.
T
h
r
e
e
pr
e
-
t
r
a
i
ne
d
C
N
N
m
o
de
l
s
,
V
G
G
16
,
D
e
n
s
e
N
e
t
201,
a
nd
I
nc
e
pt
i
onV
3,
e
x
t
r
a
c
t
f
e
a
t
ur
e
s
f
r
om
i
m
a
ge
s
.
T
he
f
e
a
t
u
r
e
s
g
e
ne
r
a
t
e
d
f
r
o
m
e
a
c
h
m
od
e
l
a
r
e
f
us
e
d
t
o
e
n
ha
n
c
e
t
h
e
r
e
l
e
va
n
t
i
n
f
o
r
m
a
t
i
on
.
F
u
r
t
h
e
r
m
o
r
e
,
pr
i
n
c
i
p
a
l
c
om
pon
e
nt
a
na
l
y
s
i
s
(
P
C
A
)
w
a
s
a
p
pl
i
e
d
t
o
r
e
duc
e
t
he
d
i
m
e
n
s
i
o
na
l
i
t
y
of
t
h
e
f
e
a
t
u
r
e
s
,
a
nd
B
a
ye
s
i
a
n
o
pt
i
m
i
z
a
t
i
o
n
w
a
s
us
e
d
t
o
op
t
i
m
i
z
e
t
he
hype
r
pa
r
a
m
e
t
e
r
s
o
f
t
he
m
a
c
h
i
ne
l
e
a
r
n
i
n
g
a
l
go
r
i
t
h
m
s
s
upp
or
t
ve
c
t
or
m
a
c
hi
ne
(
S
V
M
)
,
d
e
c
i
s
i
on
t
r
e
e
(
D
T
)
,
a
nd
k
-
ne
a
r
e
s
t
n
e
i
g
hbo
r
s
(
k
-
NN
)
.
T
h
e
r
e
s
u
l
t
i
n
g
c
l
a
s
s
i
f
i
c
a
t
i
on
m
ode
l
w
a
s
e
va
l
ua
t
e
d
ba
s
e
d
on
a
c
c
ur
a
c
y
,
p
r
e
c
i
s
i
on,
r
e
c
a
l
l
,
a
nd
F1
-
s
c
or
e
.
T
he
r
e
s
u
l
t
s
s
how
e
d
t
ha
t
F
F
-
S
V
M
,
w
h
i
c
h
i
s
t
he
p
r
op
os
e
d
m
od
e
l
,
a
c
h
i
e
v
e
d
a
n
a
c
c
u
r
a
c
y
of
98
.
79
%
w
i
t
h
p
r
e
c
i
s
i
on
,
r
e
c
a
l
l
,
a
nd
F
1
-
s
c
or
e
o
f
98.
8
5%
,
98
.
8
2%
,
a
n
d
9
8
.
84
%
,
r
e
s
pe
c
t
i
ve
l
y
.
I
n
c
onc
l
u
s
i
on,
f
us
i
n
g
f
e
a
t
u
r
e
e
xt
r
a
c
t
i
on
f
r
om
m
ul
t
i
pl
e
C
N
N
m
od
e
l
s
i
m
pr
o
ve
d
t
h
e
c
l
a
s
s
i
f
i
c
a
t
i
on
a
c
c
ur
a
c
y
of
e
a
c
h
m
a
c
hi
ne
-
l
e
a
r
n
i
n
g
m
od
e
l
.
I
t
p
r
ov
i
de
d
r
e
l
i
a
bl
e
a
n
d
a
c
c
ur
a
t
e
pr
e
d
i
c
t
i
ons
f
or
l
ung
i
m
a
ge
d
i
a
g
nos
i
s
u
s
i
n
g
c
he
s
t
X
-
R
a
y
.
Ke
y
w
or
d
s
:
Ba
ye
s
i
a
n
op
t
i
m
i
z
a
t
i
on
Bi
g
da
t
a
Convol
u
t
i
ona
l
n
e
ur
a
l
ne
t
w
ork
F
us
i
ng
f
e
a
t
ur
e
M
a
c
h
i
ne
l
e
a
r
ni
ng
T
hi
s
i
s
an
op
e
n
ac
c
e
s
s
ar
t
i
c
l
e
u
nde
r
t
he
C
C
B
Y
-
SA
l
i
c
e
n
s
e
.
Cor
r
e
s
pon
di
n
g
Au
t
h
or
:
S
ugi
ya
r
t
o
S
urono
D
e
pa
r
t
m
e
nt
of
M
a
t
he
m
a
t
i
c
s
,
F
a
c
ul
t
y
of
S
c
i
e
nc
e
a
nd
A
ppl
i
e
d
T
e
c
hn
ol
og
y
,
A
hm
a
d
D
a
h
l
a
n
U
n
i
ve
rs
i
t
y
S
out
h
Ri
ng
Ro
a
d
S
t
re
e
t
,
T
a
m
a
na
n,
Ba
ngun
t
a
p
a
n
,
B
a
nt
ul
,
S
p
e
c
i
a
l
R
e
gi
o
n
o
f
Y
og
ya
k
a
rt
a
55191
,
I
ndon
e
s
i
a
E
m
a
i
l
:
s
ugi
y
a
rt
o
@
m
a
t
h.
ua
d
.
a
c
.
i
d
1.
I
N
TR
O
D
U
C
TI
O
N
P
ul
m
on
a
ry
di
s
e
a
s
e
s
s
u
c
h
a
s
CO
V
ID
-
19
,
p
ne
u
m
on
i
a
,
a
nd
t
ube
rc
ul
os
i
s
a
r
e
g
l
oba
l
h
e
a
l
t
h
prob
l
e
m
s
t
ha
t
s
i
gni
f
i
c
a
nt
l
y
i
m
pa
c
t
m
o
rbi
d
i
t
y
a
nd
m
or
t
a
l
i
t
y
r
a
t
e
s
[1]
.
Re
du
c
i
n
g
t
he
s
pr
e
a
d
a
n
d
i
m
p
rovi
ng
p
a
t
i
e
n
t
prognos
i
s
r
e
qui
r
e
s
a
n
e
a
r
l
y
a
nd
pr
e
c
i
s
e
di
a
gnos
i
s
.
H
ow
e
ve
r,
t
h
e
re
a
re
dra
w
b
a
c
ks
t
o
m
a
nu
a
l
l
y
i
n
t
e
r
pre
t
i
ng
c
he
s
t
m
e
di
c
a
l
i
m
a
gi
ng
,
s
uc
h
a
s
C
T
s
c
a
ns
a
nd
r
a
di
o
gra
p
hs
,
i
n
c
l
ud
i
ng
t
he
ne
e
d
for
ra
d
i
ol
o
gi
s
t
s
'
know
l
e
dg
e
,
t
h
e
pos
s
i
bi
l
i
t
y
of
hu
m
a
n
m
i
s
t
a
ke
s
,
a
nd
a
l
a
c
k
of
r
e
s
ourc
e
s
i
n
pl
a
c
e
s
w
i
t
h
l
ow
h
e
a
l
t
h
c
a
r
e
a
c
c
e
s
s
[2]
.
S
i
nc
e
t
he
m
i
d
-
20t
h
c
e
n
t
ury
,
t
h
e
w
orl
d
h
a
s
unde
rgon
e
profou
nd
c
ha
n
ge
s
t
h
a
nks
t
o
r
a
pi
d
a
dv
a
nc
e
s
i
n
i
nfor
m
a
t
i
on
t
e
c
hn
ol
ogy
,
p
a
r
t
i
c
ul
a
rl
y
c
o
m
pu
t
e
r
t
e
c
hno
l
ogy
.
Bi
g
D
a
t
a
,
t
he
re
s
u
l
t
o
f
t
h
e
s
e
de
v
e
l
o
pm
e
nt
s
i
n
t
he
i
nfor
m
a
t
i
on
a
g
e
,
i
s
now
a
n
i
m
port
a
nt
for
c
e
dri
v
i
ng
t
he
t
ra
ns
f
orm
a
t
i
on
o
f
va
ri
ous
s
e
c
t
ors
,
i
n
c
l
ud
i
ng
he
a
l
t
h
c
a
r
e
.
A
s
t
h
e
nu
m
be
r
a
nd
c
om
p
l
e
xi
t
y
of
m
e
di
c
a
l
i
m
a
ge
s
i
nc
re
a
s
e
i
n
t
he
B
i
g
D
a
t
a
e
ra
,
c
on
ve
n
t
i
on
a
l
a
ppro
a
c
h
e
s
t
o
da
t
a
pro
c
e
s
s
i
ng
b
e
c
o
m
e
i
na
d
e
qu
a
t
e
t
o
m
e
e
t
t
he
ne
e
d
for
a
c
c
u
ra
t
e
a
nd
e
ffi
c
i
e
nt
a
na
l
ys
i
s
[
3]
.
Re
c
e
n
t
d
e
ve
l
op
m
e
n
t
s
,
e
s
p
e
c
i
a
l
l
y
i
n
d
e
e
p
l
e
a
r
ni
ng
(
DL
)
,
ha
ve
e
na
b
l
e
d
t
he
i
d
e
nt
i
fy
i
ng,
m
e
a
s
uri
n
g,
a
n
d
c
l
a
s
s
i
fyi
n
g
of
p
a
t
t
e
rns
i
n
i
m
a
g
e
d
a
t
a
[4]
.
DL
i
s
a
n
a
l
gor
i
t
h
m
i
ns
pi
re
d
by
how
t
h
e
hu
m
a
n
br
a
i
n
w
orks
,
w
h
e
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
,
V
o
l
.
6
,
N
o
.
3
,
N
ov
e
m
be
r
20
25
:
315
-
325
316
a
rt
i
fi
c
i
a
l
n
e
ura
l
n
e
t
w
or
ks
func
t
i
o
n
l
i
ke
t
he
ne
t
w
ork
of
ne
ur
o
ns
i
n
our
b
ra
i
n
t
o
pro
c
e
s
s
i
nf
orm
a
t
i
on
a
nd
l
e
a
rn
from
e
x
pe
r
i
e
n
c
e
[5]
,
[6]
.
O
n
e
of
t
h
e
m
os
t
po
pul
a
r
DL
a
r
c
hi
t
e
c
t
ure
s
i
n
i
m
a
ge
d
a
t
a
p
roc
e
s
s
i
ng
i
s
c
on
vol
u
t
i
o
na
l
ne
ura
l
ne
t
w
orks
(CN
N
).
CN
N
i
s
s
pe
c
i
f
i
c
a
l
l
y
de
s
i
g
ne
d
t
o
e
xt
ra
c
t
fe
a
t
ure
s
re
pre
s
e
nt
i
ng
va
r
i
ous
c
on
t
e
x
t
s
of
i
m
a
ge
da
t
a
w
i
t
hou
t
f
e
a
t
ure
e
ngi
n
e
e
ri
ng
t
h
rough
c
o
nvol
ut
i
on
l
a
ye
rs
[7]
.
S
e
ve
r
a
l
a
d
va
n
c
e
d
CN
N
a
r
c
hi
t
e
c
t
ur
e
s
w
e
re
de
v
e
l
ope
d
a
nd
a
pp
l
i
e
d
i
n
i
m
a
g
e
da
t
a
c
l
a
s
s
i
f
i
c
a
t
i
on
a
nd
pa
t
t
e
rn
r
e
c
og
ni
t
i
on
[8]
.
P
re
vi
o
us
re
s
e
a
rc
h
by
G
upt
a
a
nd
Cha
w
l
a
[
9]
e
va
l
ua
t
e
d
t
he
e
ffi
c
i
e
nc
y
of
s
e
ve
ra
l
pre
-
t
ra
i
ne
d
CN
N
m
ode
l
s
,
i
nc
l
u
di
ng
V
G
G
16,
V
G
G
19,
X
c
e
pt
i
on,
a
n
d
Re
s
N
e
t
50
,
a
l
ong
w
i
t
h
s
up
port
ve
c
t
or
m
a
c
hi
ne
(S
V
M
)
a
nd
l
ogi
s
t
i
c
re
gre
s
s
i
on
(L
R)
f
or
bre
a
s
t
c
a
nc
e
r
c
l
a
s
s
i
fi
c
a
t
i
on
us
i
ng
hi
s
t
opa
t
hol
og
y
i
m
a
ge
s
.
T
he
re
s
ul
t
s
s
ho
w
e
d
t
ha
t
Re
s
N
e
t
50+
L
R
a
c
hi
e
ve
d
t
he
be
s
t
a
c
c
ura
c
y
(93.
27%
),
out
pe
rfo
rm
i
ng
t
he
ot
he
r
m
o
de
l
s
.
Re
s
e
a
rc
h
by
A
s
l
a
n
e
t
al
.
[10]
de
ve
l
ope
d
a
c
l
a
s
s
i
fi
c
a
t
i
on
m
e
t
hod
for
CO
V
I
D
-
19
di
a
gn
os
i
s
us
i
n
g
c
he
s
t
c
om
put
e
d
t
om
ogra
phy
(
CT
)
i
m
a
ge
s
.
T
he
y
ut
i
l
i
z
e
d
s
e
v
e
ra
l
a
dva
nc
e
d
CN
N
a
rc
hi
t
e
c
t
u
re
s
(
A
l
e
x
N
e
t
,
Re
s
N
e
t
1
8,
Re
s
N
e
t
50,
Inc
e
pt
i
o
nv3,
D
e
ns
e
ne
t
201,
Inc
e
pt
i
o
nre
s
ne
t
v
2,
M
o
bi
l
e
N
e
t
v
2,
G
o
ogl
e
N
e
t
)
t
ha
t
ha
d
be
e
n
pre
-
t
ra
i
ne
d
t
o
e
xt
ra
c
t
fe
a
t
ure
s
a
nd
c
l
a
s
s
i
f
y
t
he
m
u
s
i
ng
s
e
ve
r
a
l
m
a
c
hi
ne
l
e
a
rni
ng
a
l
go
ri
t
hm
s
,
w
he
re
t
h
e
D
e
ns
e
N
e
t
-
S
V
M
a
rc
hi
t
e
c
t
ure
ga
ve
t
he
hi
g
he
s
t
a
c
c
u
ra
c
y
of
96.
29%.
Re
s
e
a
rc
h
b
y
Bi
s
w
a
s
a
nd
Is
l
a
m
[
11
]
c
l
a
s
s
i
fi
e
s
bra
i
n
t
um
or
s
t
hrou
gh
a
hybri
d
m
ode
l
ba
s
e
d
o
n
de
e
p
CN
N
(D
C
N
N
)
a
nd
S
V
M
.
CN
N
+
S
V
M
obt
a
i
ne
d
96.
0%
a
c
c
ura
c
y,
9
8.
0%
s
pe
c
i
fi
c
i
t
y,
a
nd
95.
71
%
s
e
ns
i
t
i
vi
t
y,
hi
g
he
r
t
ha
n
ot
he
r
t
ra
ns
fe
r
l
e
a
rni
ng
m
ode
l
s
(A
l
e
xN
e
t
,
G
o
ogL
e
N
e
t
,
a
nd
V
G
G
16).
P
re
vi
ous
r
e
s
e
a
rc
h
ha
s
fo
c
us
e
d
on
ut
i
l
i
z
i
ng
i
ndi
v
i
du
a
l
CN
N
a
rc
h
i
t
e
c
t
ure
s
fo
r
fe
a
t
u
re
e
xt
r
a
c
t
i
o
n,
w
hi
c
h
a
re
t
h
e
n
c
om
bi
n
e
d
w
i
t
h
c
l
a
s
s
i
c
a
l
m
a
c
hi
n
e
l
e
a
rni
n
g
a
l
gori
t
h
m
s
for
c
l
a
s
s
i
f
i
c
a
t
i
on.
T
h
e
re
fore
,
t
hi
s
s
t
u
dy
propos
e
s
a
f
e
a
t
ure
fus
i
on
a
pp
roa
c
h
by
fus
i
n
g
f
e
a
t
ur
e
s
fr
om
s
e
v
e
ra
l
pre
-
t
r
a
i
n
e
d
CN
N
a
rc
hi
t
e
c
t
ur
e
s
,
n
a
m
e
l
y
V
G
G
16,
D
e
ns
e
N
e
t
201,
a
nd
I
nc
e
pt
i
onV
3
.
T
h
i
s
a
ppr
oa
c
h
a
i
m
s
t
o
ut
i
l
i
z
e
t
he
a
dv
a
nt
a
ge
s
of
e
a
c
h
a
rc
h
i
t
e
c
t
ur
e
i
n
e
xt
r
a
c
t
i
ng
d
i
ffe
re
n
t
v
i
s
ua
l
re
pr
e
s
e
n
t
a
t
i
ons
fro
m
m
e
di
c
a
l
i
m
a
g
e
s
,
re
s
u
l
t
i
ng
i
n
m
ore
i
n
form
a
t
i
ve
a
nd
ri
c
h
fe
a
t
ure
s
[12]
.
T
he
pr
i
nc
i
pa
l
c
om
pone
nt
a
n
a
l
ys
i
s
(P
CA
)
d
i
m
e
ns
i
ona
l
i
t
y
r
e
du
c
t
i
on
t
e
c
hni
que
i
s
t
he
n
us
e
d
t
o
re
duc
e
t
h
e
d
i
m
e
ns
i
on
a
l
i
t
y
of
t
h
e
d
a
t
a
a
nd
t
r
a
ns
for
m
t
he
fe
a
t
ure
s
i
n
t
o
a
m
or
e
c
om
pa
c
t
s
u
bs
pa
c
e
w
hi
l
e
re
t
a
i
n
i
ng
i
m
port
a
n
t
i
nfor
m
a
t
i
on
[13
]
.
F
urt
he
r
m
or
e
,
c
l
a
s
s
i
fi
c
a
t
i
on
i
s
p
e
rfor
m
e
d
us
i
ng
s
e
ve
ra
l
m
a
c
hi
ne
l
e
a
rni
ng
a
l
go
ri
t
hm
s
,
n
a
m
e
l
y
S
V
M
,
de
c
i
s
i
on
t
re
e
(D
T
)
,
a
nd
K
-
n
e
a
re
s
t
n
e
i
g
hbors
(k
-
N
N
)
,
w
h
i
c
h
a
re
know
n
for
t
h
e
i
r
r
e
s
pe
c
t
i
ve
a
dva
nt
a
ge
s
.
S
V
M
offe
rs
s
i
gni
f
i
c
a
nt
a
dv
a
n
t
a
g
e
s
w
i
t
h
i
t
s
a
bi
l
i
t
y
t
o
pro
c
e
s
s
hi
gh
-
di
m
e
ns
i
ona
l
d
a
t
a
a
nd
i
t
s
c
om
p
ut
a
t
i
ona
l
e
ffi
c
i
e
nc
y
[14]
.
D
T
i
s
c
o
m
m
o
nl
y
a
pp
l
i
e
d
b
e
c
a
us
e
i
t
i
s
e
a
s
y
t
o
i
n
t
e
r
pre
t
,
t
r
a
i
ns
r
a
pi
dl
y
,
a
nd
c
a
n
m
a
na
ge
b
ot
h
num
e
ri
c
a
l
a
nd
c
a
t
e
gor
i
c
a
l
v
a
ri
a
b
l
e
s
[15]
.
T
he
b
e
ne
f
i
t
s
of
k
-
N
N
a
ppro
a
c
h
e
s
a
r
e
s
t
r
a
i
gh
t
forw
a
rd
t
o
c
om
p
re
h
e
nd
a
nd
e
xe
c
ut
e
[
16]
.
H
yp
e
rp
a
ra
m
e
t
e
r
a
dj
us
t
m
e
nt
i
s
a
n
i
m
port
a
n
t
c
om
pone
nt
i
n
t
r
a
i
n
i
ng
s
upe
rv
i
s
e
d
an
d
u
ns
upe
r
vi
s
e
d
M
L
m
ode
l
s
.
T
h
e
re
f
ore
,
M
L
m
e
t
hods
m
us
t
b
e
c
onfi
gure
d
b
e
for
e
t
ra
i
ni
ng
t
o
g
e
t
m
a
xi
m
um
re
s
ul
t
s
.
T
h
i
s
i
s
be
c
a
us
e
c
onfi
g
ura
t
i
on
va
r
i
a
b
l
e
s
a
ffe
c
t
m
o
de
l
p
e
rfor
m
a
nc
e
a
nd
a
c
c
ur
a
c
y
[
17]
.
Ba
y
e
s
i
a
n
opt
i
m
i
z
a
t
i
on
(BO
)
i
s
s
e
l
e
c
t
e
d
for
hyp
e
rp
a
ra
m
e
t
e
r
t
uni
ng
ow
i
ng
t
o
i
t
s
c
ons
i
s
t
e
nt
a
dv
a
nt
a
g
e
i
n
r
e
du
c
i
ng
c
om
p
ut
a
t
i
ona
l
t
i
m
e
re
l
a
t
i
v
e
t
o
bot
h
G
r
i
d
a
nd
ra
ndo
m
s
e
a
r
c
h
m
e
t
hods
[18]
.
S
e
ve
r
a
l
s
t
u
di
e
s
e
v
a
l
u
a
t
e
d
t
h
e
f
e
a
t
ure
f
us
i
on
of
m
u
l
t
i
pl
e
pre
-
t
ra
i
ne
d
CN
N
s
b
e
for
e
a
p
pl
y
i
ng
c
l
a
s
s
i
c
a
l
M
L
c
l
a
s
s
i
f
i
e
rs
.
A
l
z
a
h
e
m
[1
9]
us
e
d
D
e
m
ps
t
e
r
-
S
ha
f
e
r
fus
i
o
n
on
m
u
l
t
i
pl
e
CN
N
s
b
ut
r
e
l
i
e
d
o
n
e
ns
e
m
b
l
e
t
he
ory
ra
t
h
e
r
t
h
a
n
c
l
a
s
s
i
c
a
l
M
L
.
Z
h
a
ng
e
t
al
.
[
20]
c
o
m
bi
ne
d
CN
N
fe
a
t
u
re
s
but
fo
c
us
e
d
on
l
y
on
opt
i
m
i
z
e
d
CN
N
s
w
i
t
hou
t
e
xp
l
or
i
ng
t
h
e
fus
i
on
of
c
l
a
s
s
i
c
a
l
c
l
a
s
s
i
f
i
e
rs
w
i
t
h
B
a
ye
s
i
a
n
t
un
i
ng
.
O
ve
r
a
l
l
,
f
e
a
t
ure
fus
i
on
c
om
bi
n
e
d
w
i
t
h
B
a
ye
s
i
a
n
-
o
pt
i
m
i
z
e
d
c
l
a
s
s
i
c
a
l
m
a
c
h
i
ne
l
e
a
rn
i
ng
c
l
a
s
s
i
f
i
e
rs
re
m
a
i
ns
une
x
pl
or
e
d
.
T
h
e
m
a
i
n
c
on
t
ri
b
ut
i
ons
of
t
h
i
s
r
e
s
e
a
rc
h
a
re
s
um
m
a
ri
z
e
d
a
s
fol
l
ow
s
:
−
M
ul
t
i
-
CN
N
f
e
a
t
ure
fus
i
on
:
f
e
a
t
ur
e
s
a
r
e
e
xt
r
a
c
t
e
d
fro
m
t
hr
e
e
p
re
-
t
ra
i
ne
d
CN
N
a
rc
hi
t
e
c
t
ur
e
s
-
V
G
G
16,
D
e
ns
e
N
e
t
201
,
a
nd
In
c
e
pt
i
onV
3
-
a
nd
c
o
m
bi
ne
d
us
i
ng
a
fe
a
t
ur
e
fus
i
on
t
e
c
hni
q
ue
t
o
e
nha
nc
e
f
e
a
t
ur
e
ri
c
hn
e
s
s
a
nd
di
v
e
rs
i
t
y.
−
D
i
m
e
ns
i
on
a
l
i
t
y
r
e
du
c
t
i
on
us
i
ng
P
CA
:
P
CA
i
s
a
p
pl
i
e
d
t
o
t
h
e
fus
e
d
fe
a
t
ur
e
s
t
o
re
du
c
e
di
m
e
ns
i
ona
l
i
t
y
,
t
he
r
e
by
i
m
prov
i
ng
c
o
m
pu
t
a
t
i
on
a
l
e
ffi
c
i
e
nc
y
a
nd
m
i
ni
m
i
z
i
ng
ove
rf
i
t
t
i
ng
.
−
Cl
a
s
s
i
f
i
c
a
t
i
on
w
i
t
h
opt
i
m
i
z
e
d
m
a
c
h
i
n
e
l
e
a
rni
ng
m
o
de
l
s
:
t
h
e
di
m
e
ns
i
on
a
l
i
t
y
-
r
e
du
c
e
d
f
e
a
t
ur
e
s
a
re
c
a
t
e
gor
i
z
e
d
i
nt
o
fou
r
t
ype
s
of
l
ung
di
s
e
a
s
e
s
us
i
ng
opt
i
m
i
z
e
d
ve
rs
i
ons
of
S
V
M
,
DT
,
a
nd
k
-
N
N
a
l
gor
i
t
h
m
s
,
w
h
e
re
t
h
e
opt
i
m
a
l
hyp
e
rp
a
ra
m
e
t
e
rs
a
r
e
de
t
e
r
m
i
n
e
d
t
hroug
h
BO
.
−
Cl
i
n
i
c
a
l
l
y
a
ppl
i
c
a
bl
e
c
l
a
s
s
i
f
i
c
a
t
i
on
p
i
pe
l
i
ne
:
t
h
e
pro
pos
e
d
a
pproa
c
h
a
i
m
s
t
o
d
e
v
e
l
op
a
h
i
gh
l
y
a
c
c
ura
t
e
a
nd
e
ff
i
c
i
e
n
t
c
l
a
s
s
i
f
i
c
a
t
i
on
s
ys
t
e
m
t
h
a
t
c
a
n
b
e
i
n
t
e
gr
a
t
e
d
i
nt
o
c
l
i
ni
c
a
l
d
e
c
i
s
i
on
s
upp
ort
s
ys
t
e
m
s
for
ra
di
o
l
og
i
c
a
l
di
a
gnos
i
s
.
2.
M
ET
H
O
D
T
hi
s
r
e
s
e
a
rc
h
us
e
s
a
hybr
i
d
m
e
t
hod
ol
ogy
i
n
t
e
gra
t
i
ng
DL
a
nd
m
a
c
hi
ne
l
e
a
rn
i
ng
t
o
a
n
a
l
y
z
e
l
u
ng
i
m
a
ge
d
a
t
a
.
T
h
e
s
t
a
ge
s
of
t
h
i
s
r
e
s
e
a
rc
h
c
a
n
b
e
s
e
e
n
i
n
F
i
g
ure
1
.
It
s
t
a
r
t
s
w
i
t
h
i
m
a
g
e
i
n
put
,
w
hi
c
h
i
s
t
he
n
re
duc
e
d
a
nd
nor
m
a
l
i
z
e
d.
N
e
xt
,
fe
a
t
ur
e
e
xt
r
a
c
t
i
on
i
s
pe
rf
o
r
m
e
d
us
i
ng
t
h
re
e
CN
N
m
o
de
l
s
,
na
m
e
l
y
V
G
G
16
,
D
e
ns
e
N
e
t
201
,
a
nd
In
c
e
p
t
i
o
nV
3.
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
IS
S
N
:
2722
-
3221
H
y
br
i
d
f
e
at
ur
e
f
us
i
on
f
r
om
m
ul
t
i
p
l
e
CNN
m
od
e
l
s
wi
t
h
bay
e
s
i
an
-
opt
i
m
i
z
e
d
m
a
c
hi
n
e
…
(D
e
wi
R
i
s
m
awat
i
)
317
F
i
gure
1
.
R
e
s
e
a
rc
h
m
e
t
hod
T
he
r
e
s
ul
t
i
ng
f
e
a
t
ur
e
s
a
r
e
fus
e
d
us
i
ng
on
e
of
t
he
fus
i
on
ru
l
e
s
,
na
m
e
l
y
c
onc
a
t
e
na
t
e
.
T
he
r
e
s
ul
t
s
of
t
h
e
fe
a
t
ure
fus
i
o
n
w
i
l
l
b
e
r
e
du
c
e
d
i
n
di
m
e
ns
i
on
us
i
ng
P
CA
.
T
h
e
da
t
a
i
s
t
he
n
di
vi
d
e
d
i
nt
o
t
r
a
i
n
i
ng
da
t
a
a
nd
t
e
s
t
da
t
a
for
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
S
V
M
,
DT
,
a
nd
k
-
N
N
,
w
i
t
h
BO
.
In
t
he
e
v
a
l
u
a
t
i
on
p
ha
s
e
of
t
hi
s
s
t
udy
,
pe
rfor
m
a
nc
e
i
ndi
c
a
t
ors
i
n
c
l
ud
i
ng
a
c
c
u
ra
c
y,
pr
e
c
i
s
i
on
,
re
c
a
l
l
,
a
nd
F
1
-
s
c
ore
a
re
ut
i
l
i
z
e
d
t
o
a
s
s
e
s
s
t
h
e
i
m
p
a
c
t
of
t
he
i
m
pl
e
m
e
n
t
e
d
hybr
i
d
a
pp
roa
c
h
.
2.
1
.
D
atas
e
t
Che
s
t
X
-
ra
y
i
s
a
m
e
di
c
a
l
i
m
a
g
e
t
ha
t
re
s
u
l
t
s
from
t
he
pro
c
e
s
s
of
r
a
d
i
ogra
phy
,
w
h
i
c
h
i
s
a
n
i
m
a
g
i
ng
t
e
c
hni
q
ue
t
h
a
t
us
e
s
X
-
r
a
ys
t
o
v
i
s
ua
l
i
z
e
t
h
e
i
n
t
e
r
na
l
s
t
ruc
t
ure
s
of
t
h
e
c
h
e
s
t
.
A
l
l
i
m
a
g
e
s
h
a
v
e
be
e
n
pr
e
-
proc
e
s
s
e
d
a
n
d
s
c
a
l
e
d
t
o
224
×
224
p
i
x
e
l
s
.
T
h
e
da
t
a
s
e
t
w
a
s
c
re
a
t
e
d
b
y
m
e
r
gi
ng
m
u
l
t
i
pl
e
da
t
a
s
e
t
s
fro
m
t
h
e
K
a
ggl
e
pl
a
t
f
orm
,
a
n
op
e
n
on
l
i
n
e
re
pos
i
t
ory
for
da
t
a
s
e
t
s
.
T
he
da
t
a
s
e
t
s
s
how
e
d
m
i
ni
m
a
l
i
m
ba
l
a
n
c
e
but
c
ont
a
i
n
e
d
dup
l
i
c
a
t
e
i
m
a
ge
s
,
s
o
d
a
t
a
s
e
t
c
l
e
a
ni
ng
w
a
s
p
e
rfor
m
e
d.
T
a
b
l
e
1
pr
e
s
e
n
t
s
t
h
e
da
t
a
s
e
t
bot
h
p
ri
or
t
o
a
nd
fol
l
ow
i
ng
t
h
e
d
a
t
a
c
l
e
a
n
i
ng
proc
e
dur
e
,
w
hi
l
e
F
i
gu
re
2
s
h
ow
s
a
s
a
m
p
l
e
c
h
e
s
t
X
-
ra
y
i
m
a
ge
of
t
h
e
l
ungs
.
S
ubs
e
que
n
t
l
y,
t
he
d
a
t
a
s
e
t
i
s
ra
nd
om
l
y
s
pl
i
t
i
nt
o
85
%
for
t
r
a
i
n
i
ng
a
n
d
15%
fo
r
t
e
s
t
i
ng
,
e
ns
uri
ng
t
h
a
t
c
l
a
s
s
di
s
t
ri
but
i
on
r
e
m
a
i
ns
ba
l
a
n
c
e
d
.
T
r
a
i
n
i
ng
d
a
t
a
s
e
r
ve
s
t
o
bu
i
l
d
t
he
m
od
e
l
,
w
h
i
l
e
t
e
s
t
d
a
t
a
i
s
ut
i
l
i
z
e
d
t
o
m
e
a
s
ur
e
how
w
e
l
l
t
h
e
m
od
e
l
p
e
rfor
m
s
on
uns
e
e
n
i
n
put
s
.
T
a
b
l
e
1
.
N
u
m
be
r
of
i
m
a
ge
s
pe
r
c
l
a
s
s
a
f
t
e
r
d
a
t
a
c
l
e
a
n
i
ng
pro
c
e
s
s
Cl
a
s
s
e
s
N
o
rm
al
CO
V
ID
-
19
P
n
e
u
m
o
n
i
a
T
u
b
e
rc
u
l
o
s
i
s
T
o
t
a
l
Be
fo
re
c
l
e
a
n
i
n
g
1
,
802
1
,
626
1
,
800
1
,
600
6
,
828
A
ft
e
r
c
l
e
a
n
i
n
g
1
,
671
1
,
537
1
,
791
1
,
600
6
,
599
F
i
gure
2
.
C
he
s
t
X
-
r
a
y
o
f
t
he
l
u
ngs
2.
2
.
D
e
e
p
f
e
atu
r
e
e
xt
r
ac
ti
on
A
c
c
ord
i
ng
t
o
H
e
a
t
on
[21]
,
CN
N
i
s
a
s
pe
c
i
f
i
c
a
r
c
h
i
t
e
c
t
ure
i
n
DL
i
nt
e
nd
e
d
t
o
pr
oc
e
s
s
da
t
a
orga
ni
z
e
d
i
n
a
gri
d
for
m
a
t
,
s
u
c
h
a
s
pi
c
t
ure
da
t
a
,
re
pre
s
e
nt
e
d
a
s
a
2D
gri
d
of
pi
x
e
l
s
group
e
d
i
n
row
s
a
nd
c
ol
u
m
ns
.
A
CN
N
ha
s
l
a
ye
rs
of
n
e
uro
ns
t
h
a
t
fa
c
i
l
i
t
a
t
e
pa
t
t
e
rn
re
c
ogn
i
t
i
on
a
nd
f
e
a
t
ur
e
e
xt
r
a
c
t
i
o
n
fr
om
pi
c
t
ur
e
s
[
22]
.
O
ne
of
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
,
V
o
l
.
6
,
N
o
.
3
,
N
ov
e
m
be
r
20
25
:
315
-
325
318
t
he
m
a
i
n
f
e
a
t
ure
s
of
CN
N
i
s
t
h
e
c
onvo
l
ut
i
on
l
a
y
e
r,
t
hi
s
l
a
y
e
r
e
x
t
r
a
c
t
s
c
ha
ra
c
t
e
ri
s
t
i
c
vi
s
ua
l
i
n
form
a
t
i
on
hi
e
r
a
rc
hi
c
a
l
l
y
fro
m
t
h
e
i
np
ut
i
m
a
ge
.
T
h
i
s
p
roc
e
dur
e
e
n
t
a
i
l
s
a
n
i
np
ut
m
a
t
r
i
x
t
h
a
t
s
e
rve
s
a
s
a
num
e
ri
c
a
l
de
pi
c
t
i
on
of
t
he
i
m
a
ge
a
l
ongs
i
d
e
a
fi
l
t
e
r
,
a
di
m
i
nu
t
i
v
e
m
a
t
r
i
x
of
a
p
a
r
t
i
c
ul
a
r
d
i
m
e
ns
i
o
n.
T
h
e
fi
l
t
e
r
fun
c
t
i
ons
a
s
a
fe
a
t
ure
de
t
e
c
t
or
,
t
ra
v
e
rs
i
n
g
t
he
i
npu
t
m
a
t
ri
x
a
nd
e
x
e
c
ut
i
ng
a
n
e
l
e
m
e
nt
-
w
i
s
e
m
ul
t
i
p
l
i
c
a
t
i
on
w
i
t
h
t
he
re
l
e
v
a
nt
s
e
gm
e
nt
of
t
he
i
n
put
,
w
i
t
h
t
he
pr
oduc
t
s
t
he
n
s
u
m
m
e
d
t
o
yi
e
l
d
a
s
i
ng
l
e
ou
t
put
v
a
l
ue
.
T
h
e
o
ut
pu
t
va
l
ue
s
s
ubs
e
que
n
t
l
y
c
r
e
a
t
e
a
f
e
a
t
ur
e
m
a
p
t
h
a
t
i
l
l
us
t
r
a
t
e
s
t
he
c
h
a
ra
c
t
e
ri
s
t
i
c
s
i
de
n
t
i
f
i
e
d
b
y
t
h
e
fi
l
t
e
r.
S
t
r
i
d
e
i
s
t
he
s
t
e
p
de
t
e
r
m
i
ni
ng
t
h
e
m
a
g
ni
t
u
d
e
of
t
h
e
f
i
l
t
e
r
s
h
i
f
t
;
i
f
t
h
e
s
t
ri
de
s
i
z
e
i
s
1
,
t
he
f
i
l
t
e
r
w
i
l
l
a
d
va
n
c
e
by
o
ne
pi
xe
l
w
i
t
h
e
a
c
h
i
t
e
ra
t
i
on
.
T
h
e
c
o
nvo
l
u
t
i
on
pro
c
e
s
s
m
a
y
b
e
m
a
t
h
e
m
a
t
i
c
a
l
l
y
e
xp
re
s
s
e
d
a
s
(1)
,
w
h
e
r
e
(
,
)
i
s
t
he
o
u
t
pu
t
va
l
u
e
a
t
p
os
i
t
i
on
(
,
)
,
i
s
t
h
e
i
n
pu
t
m
a
t
r
i
x
,
i
s
t
h
e
c
o
nv
ol
ut
i
o
n
fi
l
t
e
r
of
s
i
z
e
×
,
a
n
d
i
s
t
he
s
t
r
i
d
e
v
a
l
ue
.
(
,
)
=
∑
∑
(
−
1
)
+
,
(
−
1
)
+
×
(
,
)
=
1
=
1
(1)
Input
i
m
a
ge
s
a
re
proc
e
s
s
e
d
b
y
t
h
e
pr
e
t
r
a
i
n
e
d
m
ode
l
v
i
a
a
s
e
ri
e
s
of
c
onvo
l
ut
i
on
a
l
a
nd
p
ool
i
ng
l
a
y
e
rs
de
s
i
gn
e
d
t
o
e
xt
r
a
c
t
h
i
e
r
a
rc
hi
c
a
l
fe
a
t
ure
s
.
E
a
c
h
c
onvol
ut
i
on
a
l
b
l
o
c
k
a
ugm
e
n
t
s
t
h
e
f
i
l
t
e
r
c
oun
t
,
e
n
a
bl
i
ng
t
h
e
m
ode
l
t
o
di
s
c
e
r
n
p
rogr
e
s
s
i
ve
l
y
i
nt
ri
c
a
t
e
pa
t
t
e
rn
s
.
A
t
t
h
e
s
a
m
e
t
i
m
e
,
t
h
e
p
ool
i
ng
l
a
ye
r
r
e
du
c
e
s
t
h
e
di
m
e
ns
i
on
a
l
i
t
y
of
t
h
e
i
m
a
g
e
.
M
a
x
P
oo
l
i
ng
[23
]
s
e
l
e
c
t
s
t
h
e
m
a
x
i
m
u
m
a
c
t
i
va
t
i
on
v
a
l
u
e
t
o
re
pr
e
s
e
n
t
t
he
w
ho
l
e
re
gi
o
n,
pr
e
s
e
rv
i
ng
e
s
s
e
n
t
i
a
l
i
n
form
a
t
i
on
.
T
h
i
s
t
e
c
hni
q
ue
e
f
fe
c
t
i
v
e
l
y
g
e
ne
r
a
t
e
s
a
s
p
a
rs
e
r
r
e
pr
e
s
e
nt
a
t
i
on
by
s
e
l
e
c
t
i
ng
on
l
y
t
he
h
i
ghe
s
t
a
c
t
i
va
t
i
on
va
l
ue
f
rom
e
a
c
h
po
ol
e
d
re
gi
on
.
T
hi
s
s
t
ud
y
e
m
pl
o
ys
t
hr
e
e
pr
e
-
t
ra
i
ne
d
CN
N
m
ode
l
s
:
V
G
G
16
,
Inc
e
pt
i
onV
3
,
a
n
d
D
e
ns
e
N
e
t
2
01,
w
i
t
h
a
n
i
np
ut
pi
c
t
ure
s
i
z
e
o
f
224
×
224
.
F
i
gu
re
3
pre
s
e
n
t
s
t
he
s
t
ru
c
t
u
ra
l
d
e
s
i
gn
of
e
a
c
h
r
e
s
pe
c
t
i
ve
m
o
de
l
.
F
i
gure
3
.
CN
N
m
ode
l
a
rc
hi
t
e
c
t
ur
e
A
s
s
how
n
i
n
F
i
g
ure
3,
e
a
c
h
m
od
e
l
us
e
d
i
nt
e
gra
t
e
s
g
l
ob
a
l
a
v
e
ra
ge
pool
i
ng
a
s
a
n
a
d
di
t
i
on
a
l
l
a
y
e
r
a
ft
e
r
t
he
m
a
i
n
m
od
e
l
.
G
l
ob
a
l
a
ve
r
a
g
e
pool
i
ng
r
e
du
c
e
s
da
t
a
d
i
m
e
ns
i
ons
w
h
i
l
e
re
t
a
i
ni
n
g
i
m
port
a
nt
i
nfor
m
a
t
i
o
n
by
ut
i
l
i
z
i
ng
t
he
i
n
va
r
i
a
n
c
e
of
f
e
a
t
ur
e
m
a
p
a
v
e
ra
g
e
s
,
t
he
r
e
by
r
e
duc
i
n
g
ove
r
fi
t
t
i
n
g
a
nd
i
m
pro
vi
ng
c
om
put
a
t
i
ona
l
e
ffi
c
i
e
nc
y
.
T
he
V
G
G
16
m
o
de
l
h
a
s
a
s
t
a
b
l
e
a
nd
de
e
p
a
r
c
h
i
t
e
c
t
ur
e
w
i
t
h
a
s
t
ra
i
ght
forw
a
r
d
s
e
que
nt
i
a
l
s
t
r
uc
t
ure
,
m
a
k
i
ng
i
t
s
ui
t
a
bl
e
for
e
xt
r
a
c
t
i
n
g
ro
bus
t
s
pa
t
i
a
l
fe
a
t
ur
e
s
.
It
pro
duc
e
s
51
2
-
di
m
e
ns
i
on
a
l
fe
a
t
u
r
e
o
ut
pu
t
s
.
D
e
ns
e
N
e
t
2
01,
de
s
i
g
ne
d
t
o
op
t
i
m
i
z
e
i
nfor
m
a
t
i
on
a
nd
gra
di
e
n
t
f
l
ow
t
hroug
h
d
e
ns
e
c
o
nne
c
t
i
ons
b
e
t
w
e
e
n
l
a
y
e
rs
,
ge
ne
ra
t
e
s
ri
c
he
r
f
e
a
t
ure
re
pre
s
e
nt
a
t
i
ons
w
i
t
h
up
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
IS
S
N
:
2722
-
3221
H
y
br
i
d
f
e
at
ur
e
f
us
i
on
f
r
om
m
ul
t
i
p
l
e
CNN
m
od
e
l
s
wi
t
h
bay
e
s
i
an
-
opt
i
m
i
z
e
d
m
a
c
hi
n
e
…
(D
e
wi
R
i
s
m
awat
i
)
319
1
,
920
out
pu
t
c
ha
n
ne
l
s
.
M
e
a
nw
h
i
l
e
,
In
c
e
pt
i
onV
3
i
nc
or
pora
t
e
s
m
u
l
t
i
pl
e
c
on
vol
u
t
i
o
na
l
fi
l
t
e
r
s
i
z
e
s
w
i
t
hi
n
t
h
e
s
a
m
e
m
odu
l
e
,
a
l
l
ow
i
ng
i
t
t
o
c
a
p
t
ur
e
m
u
l
t
i
-
s
c
a
l
e
f
e
a
t
ur
e
s
e
ff
e
c
t
i
ve
l
y.
T
hi
s
m
ode
l
prod
uc
e
s
f
i
na
l
fe
a
t
ure
s
w
i
t
h
a
d
i
m
e
ns
i
o
n
of
2
,
048
[24]
–
[26]
.
2.
3
.
F
e
atu
r
e
fu
s
i
on
an
d
d
i
m
e
n
s
i
on
al
i
ty
r
e
d
u
c
ti
on
O
ne
f
e
a
t
ure
t
y
pe
i
s
fr
e
qu
e
nt
l
y
i
ns
uff
i
c
i
e
n
t
t
o
a
c
c
om
m
oda
t
e
a
n
e
xt
e
ns
i
v
e
r
a
ng
e
o
f
d
a
t
a
i
n
c
l
a
s
s
i
fi
c
a
t
i
on
i
s
s
ue
s
.
In
o
rde
r
t
o
pro
vi
d
e
a
m
ore
r
e
l
i
a
b
l
e
a
n
d
a
c
c
ura
t
e
r
e
pr
e
s
e
n
t
a
t
i
on
o
f
t
h
e
da
t
a
,
t
e
c
hn
i
qu
e
s
t
ha
t
i
n
t
e
gr
a
t
e
c
ha
r
a
c
t
e
r
i
s
t
i
c
s
fro
m
s
e
ve
r
a
l
e
xt
r
a
c
t
ors
a
re
t
h
us
re
qui
re
d
[27]
.
In
or
de
r
t
o
i
nt
e
gra
t
e
a
l
l
of
t
he
fe
a
t
ure
s
i
nt
o
a
one
-
di
m
e
ns
i
ona
l
(1D
)
ve
c
t
or
for
m
,
f
l
a
t
t
e
ni
ng
i
s
n
e
c
e
s
s
a
ry
b
e
c
a
us
e
e
a
c
h
m
od
e
l
ge
n
e
ra
t
e
s
fe
a
t
ure
r
e
pre
s
e
nt
a
t
i
ons
i
n
a
t
hr
e
e
-
di
m
e
ns
i
ona
l
t
e
ns
or
f
orm
a
t
.
T
hi
s
g
ua
r
a
nt
e
e
s
t
h
a
t
c
h
a
ra
c
t
e
ri
s
t
i
c
s
fro
m
s
e
v
e
r
a
l
m
od
e
l
s
m
a
y
b
e
c
ons
i
s
t
e
nt
l
y
bl
e
nde
d.
T
h
e
f
e
a
t
ure
re
pre
s
e
nt
a
t
i
ons
f
or
e
a
c
h
s
a
m
pl
e
a
re
fl
a
t
t
e
n
e
d
by
t
ra
ns
f
orm
i
ng
t
h
e
m
fro
m
hi
g
h
d
i
m
e
ns
i
ons
ℎ
,
,
i
nt
o
1D
ve
c
t
ors
w
i
t
h
a
s
i
ng
l
e
di
m
e
ns
i
on
of
ℎ
⋅
⋅
.
T
hi
s
proc
e
dur
e
m
a
y
b
e
s
t
a
t
e
d
m
a
t
h
e
m
a
t
i
c
a
l
l
y
a
s
(2)
.
Be
s
i
de
s
i
n
(2)
a
nd
(3),
re
p
re
s
e
n
t
s
t
h
e
m
ode
l
w
he
re
,
ℎ
,
a
nd
a
re
t
h
e
nu
m
b
e
r
of
s
a
m
p
l
e
s
,
fe
a
t
ur
e
h
e
i
gh
t
,
w
i
dt
h,
a
nd
c
ha
n
ne
l
s
.
A
f
t
e
r
a
v
e
ra
g
i
ng
,
t
he
f
e
a
t
ur
e
s
of
t
he
t
hr
e
e
m
od
e
l
s
a
re
f
us
e
d
us
i
ng
t
he
c
on
c
a
t
e
na
t
i
on
m
e
t
hod
,
re
s
ul
t
i
ng
i
n
a
f
e
a
t
ure
fus
i
on
m
a
t
r
i
x
(
)
of
di
m
e
ns
i
on
(
,
3
+
201
+
16
)
.
,
∈
ℝ
×
(2)
W
e
r
e
,
=
ℎ
∙
∙
(3)
Be
s
i
d
e
s
be
i
ng
w
i
de
l
y
us
e
d
i
n
m
od
e
rn
n
e
t
w
o
rk
a
rc
h
i
t
e
c
t
ure
s
for
fe
a
t
u
re
f
us
i
on
,
c
onc
a
t
e
n
a
t
i
on
m
e
t
hods
a
re
us
e
d
fo
r
t
h
e
i
r
s
i
m
pl
i
c
i
t
y
a
nd
a
bi
l
i
t
y
t
o
re
t
a
i
n
i
n
form
a
t
i
on
fro
m
m
ul
t
i
p
l
e
fe
a
t
ur
e
s
our
c
e
s
w
i
t
hou
t
l
os
i
ng
i
m
por
t
a
nt
d
e
t
a
i
l
s
[28
]
.
U
nl
i
k
e
a
v
e
ra
g
i
ng
or
m
a
x
i
m
i
z
i
ng,
w
h
i
c
h
c
a
n
r
e
du
c
e
d
i
m
e
ns
i
o
na
l
i
t
y
o
r
i
gno
re
va
ri
a
t
i
ons
be
t
w
e
e
n
f
e
a
t
ur
e
s
,
c
o
nc
a
t
e
na
t
i
on
pr
e
s
e
r
ve
s
t
h
e
r
i
c
h
ne
s
s
of
t
he
fe
a
t
ure
re
pre
s
e
nt
a
t
i
on
obt
a
i
n
e
d
fro
m
e
a
c
h
m
od
e
l
or
l
a
ye
r
.
T
h
i
s
fus
i
on
c
orr
e
s
ponds
t
o
t
he
fe
a
t
u
re
c
onc
a
t
e
n
a
t
i
o
n
m
e
t
h
od
e
xpr
e
s
s
e
d
i
n
(4)
.
=
[
3
,
16
,
201
]
(4)
H
ow
e
ve
r
,
t
h
e
c
on
c
a
t
e
na
t
i
on
m
e
t
hod
i
nc
r
e
a
s
e
s
t
he
f
i
na
l
fe
a
t
ure
v
e
c
t
or'
s
d
i
m
e
ns
i
o
n,
i
nc
r
e
a
s
i
ng
c
om
p
ut
a
t
i
ona
l
c
om
pl
e
xi
t
y.
T
h
e
r
e
fore
,
P
CA
i
s
a
pp
l
i
e
d
t
o
t
he
fus
e
d
m
a
t
r
i
x
(F
F
)
t
o
re
d
uc
e
t
he
di
m
e
ns
i
on
a
l
i
t
y
of
l
a
rg
e
a
nd
c
o
m
p
l
e
x
f
e
a
t
ur
e
s
w
h
i
l
e
re
t
a
i
ni
ng
t
h
e
m
os
t
i
m
po
rt
a
n
t
i
nf
orm
a
t
i
on
,
s
i
m
p
l
i
f
yi
ng
t
he
re
pr
e
s
e
n
t
a
t
i
on
of
hi
gh
-
di
m
e
ns
i
on
a
l
da
t
a
.
I
n
g
e
ne
ra
l
,
P
CA
w
orks
by
c
a
l
c
ul
a
t
i
ng
t
he
e
i
ge
nva
l
ue
s
of
t
h
e
d
a
t
a
c
ova
ri
a
nc
e
m
a
t
ri
x
t
o
de
t
e
rm
i
ne
t
h
e
m
a
i
n
di
r
e
c
t
i
on
o
f
d
a
t
a
di
s
pe
rs
i
on
t
hr
ough
a
de
c
o
m
pos
i
t
i
on
p
roc
e
s
s
,
a
s
s
h
ow
n
i
n
(
5).
=
(5)
P
a
i
rs
of
e
i
ge
n
va
l
ue
a
nd
e
i
ge
nve
c
t
o
r
a
re
obt
a
i
n
e
d
,
w
hi
c
h
c
om
p
os
e
t
he
pri
n
c
i
p
a
l
c
o
m
pon
e
n
t
s
.
T
he
s
e
c
om
p
one
n
t
s
a
re
t
h
e
n
s
or
t
e
d
by
t
h
e
va
l
ue
of
i
n
de
s
c
e
ndi
ng
ord
e
r.
T
h
i
s
s
t
u
dy
c
hoos
e
s
t
h
e
m
i
ni
m
um
num
b
e
r
o
f
p
ri
n
c
i
p
a
l
c
o
m
po
ne
n
t
s
t
h
a
t
s
a
t
i
s
fy
(6)
.
∑
=
1
∑
=
1
≥
0
.
95
(6)
W
he
r
e
i
s
t
he
-
t
h
e
i
ge
nv
a
l
ue
re
pre
s
e
nt
i
ng
t
h
e
v
a
r
i
a
n
c
e
e
xpl
a
i
ne
d
by
t
h
e
-
t
h
pri
nc
i
pa
l
c
o
m
pon
e
nt
,
a
nd
i
s
t
he
-
t
h
e
i
g
e
nv
a
l
u
e
of
a
l
l
i
ni
t
i
a
l
f
e
a
t
ur
e
s
,
c
u
m
ul
a
t
i
ve
l
y
re
pre
s
e
nt
i
ng
t
he
t
o
t
a
l
v
a
ri
a
n
c
e
of
t
he
ori
gi
n
a
l
da
t
a
.
T
he
95%
c
u
m
ul
a
t
i
ve
e
xp
l
a
i
ne
d
v
a
ri
a
nc
e
t
hre
s
hol
d
i
s
c
hos
e
n
t
o
e
ns
ure
t
h
a
t
t
he
re
d
uc
e
d
f
e
a
t
ure
s
pa
c
e
r
e
t
a
i
ns
m
os
t
of
t
h
e
ori
g
i
n
a
l
i
nfor
m
a
t
i
on
w
h
i
l
e
s
i
gni
f
i
c
a
nt
l
y
r
e
du
c
i
ng
d
i
m
e
ns
i
on
a
l
i
t
y
.
By
pre
s
e
rvi
n
g
a
t
l
e
a
s
t
95%
of
t
he
t
ot
a
l
v
a
ri
a
nc
e
,
t
he
t
ra
ns
fo
rm
e
d
fe
a
t
u
re
s
r
e
m
a
i
n
h
i
ghl
y
re
pr
e
s
e
n
t
a
t
i
v
e
of
t
h
e
ori
gi
n
a
l
d
a
t
a
s
t
ruc
t
ure
,
m
a
ki
ng
t
he
m
s
u
i
t
a
bl
e
f
or
s
ubs
e
qu
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
t
a
s
ks
.
2.
4
.
Bay
e
s
i
an
op
ti
m
i
z
ati
on
for
h
yp
e
r
p
ar
ame
t
e
r
tu
n
i
n
g
i
n
c
l
as
s
i
fi
c
ati
on
m
od
e
l
s
Cl
a
s
s
i
f
i
c
a
t
i
on
w
a
s
p
e
rfo
rm
e
d
us
i
ng
t
hr
e
e
m
a
c
hi
n
e
l
e
a
rn
i
ng
a
l
gori
t
h
m
s
,
S
V
M
,
DT
,
a
nd
k
-
N
N
,
t
o
e
xpl
o
re
di
ff
e
r
e
nt
a
pp
roa
c
he
s
a
n
d
s
e
e
t
he
m
os
t
e
ff
e
c
t
i
v
e
m
o
de
l
.
BO
w
a
s
a
pp
l
i
e
d
a
s
a
hy
pe
rp
a
ra
m
e
t
e
r
t
uni
n
g
m
e
t
hod
t
o
i
m
pro
ve
c
l
a
s
s
i
fi
c
a
t
i
o
n
pe
r
form
a
n
c
e
.
H
yp
e
rp
a
ra
m
e
t
e
rs
[29]
r
e
f
e
r
t
o
s
e
t
t
i
n
gs
t
h
a
t
d
e
t
e
r
m
i
n
e
how
a
m
a
c
hi
n
e
-
l
e
a
rni
n
g
m
od
e
l
i
s
t
r
a
i
n
e
d
.
H
ype
r
pa
r
a
m
e
t
e
rs
a
r
e
no
t
l
e
a
rne
d
from
t
h
e
da
t
a
b
ut
a
re
s
e
t
be
fore
t
ra
i
n
i
ng
a
nd
c
a
n
s
i
g
ni
f
i
c
a
nt
l
y
a
ff
e
c
t
m
od
e
l
pe
rf
orm
a
nc
e
.
T
he
b
a
s
i
s
of
BO
c
o
m
e
s
fro
m
Ba
y
e
s
'
t
h
e
ore
m
i
n
(7)
,
i
.
e
.
w
he
n
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
,
V
o
l
.
6
,
N
o
.
3
,
N
ov
e
m
be
r
20
25
:
315
-
325
320
e
va
l
ua
t
i
ng
e
v
i
de
nc
e
da
t
a
,
t
he
p
os
t
e
r
i
or
p
roba
bi
l
i
t
y
(
|
)
of
m
od
e
l
i
s
e
qua
l
t
o
t
h
e
prob
a
bi
l
i
t
y
(
|
)
of
e
va
l
ua
t
i
ng
e
vi
d
e
nc
e
da
t
a
w
i
t
h
m
ode
l
m
ul
t
i
pl
i
e
d
b
y
t
he
pr
i
or
p
roba
bi
l
i
t
y
(
)
.
(
|
)
∝
(
|
)
(
)
(7)
T
he
m
a
i
n
ob
j
e
c
t
i
ve
of
BO
i
s
t
o
f
i
nd
t
he
hy
pe
rp
a
r
a
m
e
t
e
r
va
l
u
e
∗
t
h
a
t
m
a
x
i
m
i
z
e
s
t
h
e
obj
e
c
t
i
ve
func
t
i
on
(
)
,
a
s
s
how
n
i
n
(8)
.
∗
=
a
r
g
m
ax
∈
(
)
(8)
In
(8)
,
re
pre
s
e
nt
s
t
he
c
om
b
i
na
t
i
o
n
of
hyp
e
rp
a
ra
m
e
t
e
rs
t
o
be
opt
i
m
i
z
e
d
.
M
e
a
nw
h
i
l
e
,
(
)
re
pre
s
e
nt
s
t
h
e
obj
e
c
t
i
v
e
f
unc
t
i
on
a
s
t
he
a
v
e
ra
ge
a
c
c
u
ra
c
y
s
c
ore
ob
t
a
i
ne
d
fro
m
t
h
e
t
ra
i
ni
ng
a
nd
e
v
a
l
ua
t
i
on
proc
e
s
s
us
i
ng
k
-
fol
d
c
r
os
s
-
va
l
i
da
t
i
o
n
w
i
t
h
a
v
a
l
u
e
of
=
5
.
BO
i
n
t
hi
s
s
t
udy
c
ons
i
s
t
s
of
t
w
o
i
m
por
t
a
n
t
e
l
e
m
e
nt
s
.
F
i
rs
t
,
t
h
e
Ba
y
e
s
i
a
n
G
a
us
s
i
a
n
pro
c
e
s
s
i
ng
(G
P
)
s
t
a
t
i
s
t
i
c
a
l
m
od
e
l
i
s
a
prob
a
b
i
l
i
s
t
i
c
re
pr
e
s
e
n
t
a
t
i
on
of
t
he
ob
j
e
c
t
i
ve
f
unc
t
i
on
.
G
P
i
s
a
c
ol
l
e
c
t
i
on
of
r
a
ndo
m
va
ri
a
bl
e
s
;
e
a
c
h
s
ubs
e
t
of
ra
ndo
m
va
r
i
a
b
l
e
s
w
i
l
l
fo
l
l
ow
a
m
ul
t
i
v
a
ri
a
t
e
G
a
us
s
i
a
n
d
i
s
t
r
i
but
i
on
[
30]
.
In
(9)
de
s
c
ri
b
e
s
a
G
a
us
s
i
a
n
proc
e
s
s
t
hro
ugh
t
w
o
c
o
re
c
om
pone
nt
s
:
t
he
m
e
a
n
fu
nc
t
i
on
(
)
a
nd
t
h
e
c
ov
a
ri
a
n
c
e
fun
c
t
i
on
(
,
′
)
.
(
)
~
(
(
)
,
(
,
′
)
)
(9)
T
he
s
e
c
ond
e
l
e
m
e
n
t
i
s
t
he
e
xpe
c
t
e
d
i
m
prov
e
m
e
nt
(
E
I)
a
c
q
ui
s
i
t
i
on
fun
c
t
i
on,
w
hi
c
h
de
t
e
r
m
i
ne
s
t
h
e
ne
xt
s
e
a
rc
h
poi
n
t
i
n
hyp
e
rp
a
ra
m
e
t
e
r
s
p
a
c
e
.
E
I
i
s
a
m
e
t
hod
us
e
d
t
o
e
v
a
l
ua
t
e
how
m
uc
h
po
t
e
n
t
i
a
l
i
m
pr
ove
m
e
n
t
t
he
r
e
i
s
a
t
a
po
i
nt
i
n
opt
i
m
i
z
a
t
i
on
[31]
.
T
he
m
a
i
n
i
d
e
a
i
s
t
o
e
va
l
u
a
t
e
how
m
u
c
h
po
t
e
n
t
i
a
l
i
m
pr
ove
m
e
nt
i
n
t
he
obj
e
c
t
i
ve
fun
c
t
i
on
va
l
ue
c
a
n
be
c
om
pa
r
e
d
t
o
t
h
e
c
urr
e
nt
b
e
s
t
va
l
u
e
(
∗
)
.
In
i
t
s
i
m
p
l
e
m
e
nt
a
t
i
on,
e
a
c
h
m
a
c
hi
ne
l
e
a
r
ni
ng
a
l
gor
i
t
h
m
'
s
BO
be
gi
ns
w
i
t
h
a
r
a
ndo
m
s
e
l
e
c
t
i
o
n
o
f
hype
rp
a
ra
m
e
t
e
r
c
om
b
i
na
t
i
o
ns
a
s
t
he
s
t
a
rt
i
ng
po
i
nt
of
e
xp
l
ora
t
i
on
.
N
e
xt
,
BO
pe
rfo
rm
s
a
n
o
pt
i
m
i
z
a
t
i
on
pro
c
e
s
s
for
1
00
i
t
e
r
a
t
i
ons
t
o
e
x
pl
or
e
t
h
e
s
e
a
rc
h
s
p
a
c
e
de
s
c
r
i
b
e
d
i
n
T
a
b
l
e
2
a
nd
i
d
e
n
t
i
fy
t
he
hy
pe
rp
a
r
a
m
e
t
e
r
c
om
b
i
n
a
t
i
on
t
h
a
t
produ
c
e
s
t
he
be
s
t
p
e
rfo
rm
a
nc
e
.
T
h
e
i
ni
t
i
a
l
c
o
m
bi
na
t
i
on
a
nd
t
he
fi
n
a
l
r
e
s
ul
t
of
t
he
opt
i
m
i
z
a
t
i
on
pro
c
e
s
s
a
r
e
c
o
m
pre
he
ns
i
ve
l
y
p
re
s
e
nt
e
d
i
n
T
a
bl
e
3
.
T
h
e
s
e
opt
i
m
i
z
e
d
hyp
e
rpa
r
a
m
e
t
e
rs
a
r
e
ut
i
l
i
z
e
d
i
n
bo
t
h
t
he
t
ra
i
ni
ng
a
nd
t
e
s
t
i
ng
pro
c
e
s
s
e
s
t
o
e
nha
nc
e
c
l
a
s
s
i
fi
c
a
t
i
on
m
ode
l
p
e
rfor
m
a
nc
e
.
T
a
b
l
e
2
.
S
e
a
r
c
h
s
p
a
c
e
M
L
a
l
g
.
H
y
p
e
rp
a
ra
m
e
t
e
r
S
e
a
rc
h
s
p
a
c
e
P
ri
o
r
d
i
s
t
ri
b
u
t
i
o
n
S
V
M
Bo
x
c
o
n
s
t
ra
i
n
t
s
(C)
[1
e
-
5
,
1
e
+
1
]
L
o
g
-
u
n
i
fo
rm
K
e
rn
e
l
t
y
p
e
['l
i
n
e
a
r',
'p
o
l
y
',
'rb
f',
's
i
g
m
o
i
d
']
Ca
t
e
g
o
ri
c
a
l
K
e
rn
e
l
s
c
a
l
e
[1
e
-
3
,
1
e
+
1
]
L
o
g
-
u
n
i
fo
rm
P
o
l
y
n
o
m
i
a
l
d
e
g
re
e
[2
,
5
]
U
n
i
fo
rm
(d
i
s
c
re
t
e
)
K
e
rn
e
l
c
o
e
ffi
c
i
e
n
t
[0
,
1
]
U
n
i
fo
rm
DT
M
a
x
i
m
u
m
t
re
e
d
e
p
t
h
[1
,
5
0
]
L
o
g
-
u
n
i
fo
rm
M
i
n
i
m
u
m
s
p
l
i
t
s
i
z
e
{
2
,
7
,
1
2
,
.
.
.
,
1
9
7
}
Ca
t
e
g
o
ri
c
a
l
M
i
n
i
m
u
m
l
e
a
f
s
i
z
e
[1
,
1
0
0
]
U
n
i
fo
rm
M
i
n
i
m
u
m
l
e
a
f
w
e
i
g
h
t
[1
e
-
2
,
5
e
-
1]
L
o
g
-
u
n
i
fo
rm
S
p
l
i
t
t
i
n
g
c
ri
t
e
ri
o
n
{
‘g
i
n
i
’,
‘e
n
t
ro
p
y
’}
Ca
t
e
g
o
ri
c
a
l
K
-
NN
N
u
m
b
e
r
o
f
n
e
i
g
h
b
o
rs
[1
,
2
0
]
U
n
i
fo
rm
W
e
i
g
h
t
s
{
‘u
n
i
fo
rm
’,
‘d
i
s
t
a
n
c
e
’}
Ca
t
e
g
o
ri
c
a
l
S
e
a
rc
h
a
l
g
o
ri
t
h
m
{
‘a
u
t
o
’,
‘b
a
l
l
_
t
re
e
’,
‘k
d
_
t
re
e
’,
‘b
ru
t
e
’}
Ca
t
e
g
o
ri
c
a
l
T
a
b
l
e
3
.
H
yp
e
rp
a
ra
m
e
t
e
r
s
e
t
t
i
ngs
f
or
m
a
c
h
i
ne
l
e
a
r
ni
ng
m
ode
l
s
M
L
a
l
g
.
H
y
p
e
rp
a
ra
m
e
t
e
r
VGG
-
16
D
e
n
s
e
N
e
t
2
0
1
In
c
e
p
t
i
o
n
V
3
F
e
a
t
u
re
f
u
s
i
o
n
S
V
M
Bo
x
c
o
n
s
t
ra
i
n
t
s
(C)
4
.
9
7
8
8
3
10
4
.
4
2
0
1
4
10
K
e
rn
e
l
t
y
p
e
L
i
n
e
a
r
RBF
RBF
P
o
l
y
K
e
rn
e
l
s
c
a
l
e
-
0
.
0
0
1
0
.
0
0
1
0
.
0
0
1
P
o
l
y
n
o
m
i
a
l
d
e
g
re
e
-
-
-
2
K
e
rn
e
l
c
o
e
ffi
c
i
e
n
t
-
-
-
1
DT
M
a
x
i
m
u
m
t
re
e
d
e
p
t
h
50
50
37
23
M
i
n
i
m
u
m
s
p
l
i
t
s
i
z
e
7
7
2
87
M
i
n
i
m
u
m
l
e
a
f
s
i
z
e
1
1
1
26
M
i
n
i
m
u
m
l
e
a
f
w
e
i
g
h
t
0
.
0
1
0
.
0
1
0
.
0
1
1
0
4
0
.
0
1
S
p
l
i
t
t
i
n
g
c
ri
t
e
ri
o
n
E
n
t
ro
p
y
E
n
t
ro
p
y
G
i
n
i
E
n
t
ro
p
y
K
-
NN
N
u
m
b
e
r
o
f
n
e
i
g
h
b
o
rs
7
8
3
3
W
e
i
g
h
t
s
D
i
s
t
a
n
c
e
D
i
s
t
a
n
c
e
D
i
s
t
a
n
c
e
D
i
s
t
a
n
c
e
S
e
a
rc
h
a
l
g
o
ri
t
h
m
Ba
l
l
T
re
e
K
d
T
re
e
K
d
T
re
e
Ba
l
l
T
re
e
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
IS
S
N
:
2722
-
3221
H
y
br
i
d
f
e
at
ur
e
f
us
i
on
f
r
om
m
ul
t
i
p
l
e
CNN
m
od
e
l
s
wi
t
h
bay
e
s
i
an
-
opt
i
m
i
z
e
d
m
a
c
hi
n
e
…
(D
e
wi
R
i
s
m
awat
i
)
321
3.
R
ES
U
LTS
A
N
D
D
I
S
C
U
S
S
I
O
N
3.
1
.
F
e
atu
r
e
e
n
h
an
c
e
me
n
t
r
e
s
u
l
t
T
hre
e
di
f
fe
r
e
nt
CN
N
-
b
a
s
e
d
m
od
e
l
s
w
e
r
e
e
m
pl
oye
d
t
o
e
x
t
ra
c
t
fe
a
t
ur
e
s
from
t
he
pr
oc
e
s
s
e
d
l
u
ng
i
m
a
ge
s
.
T
he
re
s
u
l
t
i
ng
o
ut
pu
t
s
fro
m
e
a
c
h
m
od
e
l
w
e
r
e
f
us
e
d
u
s
i
ng
c
onc
a
t
e
na
t
i
o
n
t
o
i
nc
r
e
a
s
e
t
he
robus
t
ne
s
s
a
nd
di
ve
rs
i
t
y
of
t
he
f
e
a
t
ur
e
re
pr
e
s
e
n
t
a
t
i
on
.
F
e
a
t
ur
e
fus
i
on
y
i
e
l
ds
h
i
gh
-
d
i
m
e
ns
i
ona
l
da
t
a
,
w
hi
c
h
m
a
y
l
e
a
d
t
o
i
s
s
ue
s
s
u
c
h
a
s
c
o
m
put
a
t
i
on
a
l
c
o
m
pl
e
x
i
t
y
.
T
o
m
i
t
i
ga
t
e
t
h
e
s
e
c
ha
l
l
e
nge
s
,
P
CA
i
s
e
m
pl
oye
d
t
o
proj
e
c
t
t
he
fe
a
t
ure
s
i
n
t
o
a
n
e
w
s
ubs
p
a
c
e
w
i
t
h
r
e
du
c
e
d
d
i
m
e
ns
i
o
ns
.
T
h
e
o
ut
c
om
e
s
o
f
t
hi
s
s
e
r
i
e
s
of
m
e
t
ho
dol
o
gi
e
s
a
re
di
s
pl
a
ye
d
i
n
T
a
b
l
e
4
.
T
a
b
l
e
4
.
F
e
a
t
ure
e
x
t
ra
c
t
i
on
a
n
d
d
i
m
e
ns
i
o
na
l
i
t
y
r
e
du
c
t
i
on
re
s
ul
t
s
F
e
a
t
u
re
V
G
G
1
6
D
e
n
s
e
N
e
t
2
0
1
In
c
e
p
t
i
o
n
V
3
FF
FF
-
P
CA
F
e
a
t
u
re
d
i
m
e
n
s
i
o
n
512
1
,
920
2
,
048
4
,
480
1
,
217
M
e
t
h
o
d
G
l
o
b
a
l
a
v
e
ra
g
e
p
o
o
l
i
n
g
G
l
o
b
a
l
a
v
e
ra
g
e
p
o
o
l
i
n
g
G
l
o
b
a
l
a
v
e
ra
g
e
p
o
o
l
i
n
g
Co
n
c
a
t
e
n
a
t
i
o
n
P
ri
n
c
i
p
a
l
c
o
m
p
o
n
e
n
t
a
n
a
l
y
s
i
s
T
he
t
a
b
l
e
i
ndi
c
a
t
e
s
t
ha
t
e
a
c
h
CN
N
m
od
e
l
g
e
ne
r
a
t
e
s
a
d
i
s
t
i
n
c
t
qu
a
nt
i
t
y
of
fe
a
t
ur
e
s
.
V
G
G
16
g
e
n
e
ra
t
e
s
512
f
e
a
t
ur
e
s
,
D
e
ns
e
N
e
t
2
01
ge
n
e
ra
t
e
s
1
,
920
fe
a
t
u
re
s
,
a
nd
In
c
e
p
t
i
onV
3
g
e
ne
ra
t
e
s
2
,
0
48
fe
a
t
ur
e
s
.
In
c
e
p
t
i
onV
3
produc
e
s
t
he
m
os
t
fe
a
t
u
re
s
,
s
i
gn
i
fyi
ng
t
ha
t
t
he
m
od
e
l
e
x
t
ra
c
t
s
m
or
e
c
o
m
pr
e
h
e
ns
i
v
e
i
nfor
m
a
t
i
on
t
h
a
n
V
G
G
16
a
nd
D
e
ns
e
N
e
t
201
.
F
o
l
l
ow
i
ng
t
he
c
onc
a
t
e
n
a
t
i
on
of
fe
a
t
ure
s
,
t
he
d
i
m
e
ns
i
ona
l
i
t
y
i
n
c
re
a
s
e
d
t
o
4
,
480.
P
CA
s
ubs
t
a
nt
i
a
l
l
y
l
ow
e
r
e
d
t
h
e
f
e
a
t
ure
d
i
m
e
ns
i
on
fro
m
4
,
480
t
o
1
,
217
f
e
a
t
ure
s
.
3.
2
.
Eva
l
u
ati
on
of
c
l
as
s
i
fi
c
at
i
on
Cl
a
s
s
i
f
i
c
a
t
i
on
t
a
s
ks
w
e
r
e
c
a
rri
e
d
ou
t
us
i
ng
t
hr
e
e
m
a
c
hi
ne
l
e
a
rni
ng
m
e
t
hods
,
na
m
e
l
y
S
V
M
,
D
e
c
i
s
i
on
D
T
,
a
nd
k
-
N
N
,
w
i
t
h
t
he
s
pe
c
i
f
i
c
hyp
e
rp
a
ra
m
e
t
e
rs
out
l
i
n
e
d
i
n
T
a
bl
e
3
.
T
h
e
p
e
r
for
m
a
nc
e
of
e
a
c
h
m
ode
l
w
a
s
a
s
s
e
s
s
e
d
us
i
ng
e
va
l
ua
t
i
on
m
e
t
r
i
c
s
s
uc
h
a
s
a
c
c
ur
a
c
y
,
pr
e
c
i
s
i
on
,
re
c
a
l
l
,
a
nd
F
1
s
c
ore
,
a
s
re
pr
e
s
e
n
t
e
d
i
n
(10)
-
(13)
.
T
h
e
r
e
s
ul
t
s
d
e
ri
ve
d
fro
m
t
he
e
va
l
ua
t
i
o
n
a
re
s
um
m
a
ri
z
e
d
i
n
T
a
bl
e
5
.
=
+
+
+
+
(10)
=
+
(11)
=
+
(12)
1
−
=
2
×
×
+
(13)
T
he
m
ode
l
e
va
l
u
a
t
i
on
us
e
s
s
t
a
nd
a
rd
m
e
t
r
i
c
s
s
uc
h
a
s
t
ru
e
p
os
i
t
i
v
e
(
TP
)
,
t
ru
e
n
e
ga
t
i
v
e
(
TN
)
,
fa
l
s
e
pos
i
t
i
ve
(
FP
),
a
n
d
f
a
l
s
e
n
e
ga
t
i
v
e
(
FN
)
for
p
e
rfor
m
a
n
c
e
a
s
s
e
s
s
m
e
nt
.
Ba
s
e
d
on
T
a
b
l
e
5
,
t
he
hybr
i
d
a
ppr
oa
c
h
ba
s
e
d
on
fe
a
t
u
re
fus
i
o
n
(F
F
)
w
i
t
h
t
he
c
om
b
i
n
a
t
i
on
of
F
F
-
S
V
M
s
how
s
t
h
e
be
s
t
c
l
a
s
s
i
fi
c
a
t
i
on
p
e
rfor
m
a
n
c
e
c
om
p
a
r
e
d
t
o
s
i
n
gl
e
m
ode
l
s
s
uc
h
a
s
V
G
G
16,
D
e
ns
e
N
e
t
201
,
a
nd
I
nc
e
pt
i
onV
3
.
F
F
-
S
V
M
re
c
ord
e
d
t
h
e
h
i
ghe
s
t
a
c
c
ura
c
y
o
f
98
.
79%
,
a
l
on
g
w
i
t
h
pr
e
c
i
s
i
on
,
r
e
c
a
l
l
,
a
nd
F
1
-
s
c
ore
,
a
l
l
a
bo
ve
9
8%.
T
h
e
s
e
re
s
ul
t
s
r
e
fl
e
c
t
t
h
e
s
t
a
bi
l
i
t
y
of
t
he
m
od
e
l
i
n
c
l
a
s
s
i
fy
i
ng
m
ul
t
i
c
l
a
s
s
m
e
d
i
c
a
l
i
m
a
g
e
s
.
Co
m
p
a
re
d
t
o
b
a
s
e
l
i
ne
s
s
u
c
h
a
s
D
e
ns
e
N
e
t
201
-
S
V
M
(97.
47%)
a
n
d
V
G
G
16
-
S
V
M
(97
.
07%)
,
fe
a
t
ur
e
fus
i
on
f
rom
m
u
l
t
i
pl
e
CN
N
s
prov
i
de
s
ri
c
he
r
i
nfor
m
a
t
i
o
n.
T
a
b
l
e
5
.
E
va
l
ua
t
i
o
n
re
s
ul
t
M
od
e
l
ML
a
l
gor
i
t
m
s
A
c
c
ur
a
c
y
(
%
)
P
r
e
c
i
s
i
on
(
%
)
R
e
c
a
l
l
(%)
F1
-
S
c
or
e
(
%
)
T
i
m
e
t
a
ke
n
(
s
)
V
G
G
16
S
V
M
97.
0
7
97.
1
6
97.
1
7
97.
1
6
244.
6789
DT
83.
3
3
83.
8
1
83.
4
4
83.
5
3
310.
1874
K
-
NN
95.
8
6
95.
9
1
96.
0
0
95.
9
4
202.
0312
D
e
ns
e
N
e
t
2
01
S
V
M
97.
4
7
97.
5
5
97.
4
6
97.
4
9
648.
1935
DT
88.
5
9
88.
7
3
88.
7
3
88.
6
9
520.
6981
K
-
NN
96.
4
6
96.
5
9
96.
4
8
96.
5
1
727.
9218
I
nc
e
pt
i
onV
3
S
V
M
95.
4
5
95.
5
4
95.
5
3
95.
5
3
1008
.
927
6
DT
79.
8
0
79.
8
8
79.
8
3
79.
8
1
678.
3621
K
-
NN
91.
5
2
91.
6
2
91.
6
6
91.
6
2
715.
0327
FF
S
V
M
98.
7
9
98.
8
5
98.
8
2
98.
8
4
638.
0538
DT
88.
7
9
88.
8
5
88.
8
8
88.
8
4
508.
6961
K
-
NN
97.
2
7
97.
4
0
97.
2
9
97.
3
4
401.
2894
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
,
V
o
l
.
6
,
N
o
.
3
,
N
ov
e
m
be
r
20
25
:
315
-
325
322
F
e
a
t
u
re
fus
i
o
n
a
l
s
o
i
m
pr
ove
s
t
h
e
pe
rf
orm
a
n
c
e
of
ot
he
r
a
l
gor
i
t
h
m
s
.
T
h
e
a
c
c
u
ra
c
y
of
k
-
N
N
i
nc
re
a
s
e
d
from
95
.
86%
t
o
97.
27%
,
a
nd
D
T
s
l
i
gh
t
l
y
i
nc
r
e
a
s
e
d
from
8
8
.
59%
t
o
8
8.
7
9%.
H
ow
e
ve
r,
D
T
s
t
i
l
l
s
how
e
d
t
h
e
l
ow
e
s
t
p
e
rfor
m
a
n
c
e
a
m
ong
t
h
e
t
hr
e
e
a
l
gor
i
t
h
m
s
,
i
ndi
c
a
t
i
n
g
i
t
s
l
i
m
i
t
a
t
i
ons
i
n
r
e
c
ogni
z
i
n
g
c
o
m
pl
e
x
vi
s
u
a
l
pa
t
t
e
rns
.
F
e
a
t
ur
e
fus
i
on
re
s
ul
t
i
n
g
i
n
h
i
gh
-
d
i
m
e
ns
i
on
a
l
f
e
a
t
ure
ve
c
t
ors
s
how
s
t
ha
t
t
h
e
a
pp
l
i
c
a
t
i
on
of
P
CA
s
i
gni
f
i
c
a
nt
l
y
i
m
pr
ove
s
c
o
m
put
a
t
i
on
a
l
e
ff
i
c
i
e
n
c
y
w
i
t
hou
t
d
e
gra
d
i
ng
m
o
de
l
a
c
c
u
ra
c
y.
T
h
i
s
e
ffe
c
t
i
ve
n
e
s
s
i
s
re
fl
e
c
t
e
d
i
n
t
h
e
r
e
l
a
t
i
ve
l
y
e
ffi
c
i
e
nt
t
r
a
i
n
i
ng
t
i
m
e
of
t
h
e
F
F
-
S
V
M
m
od
e
l
,
w
hi
c
h
i
s
a
bout
638
s
e
c
onds
.
T
h
i
s
dura
t
i
on
i
s
c
o
m
pa
ra
b
l
e
t
o
a
s
i
n
gl
e
m
ode
l
s
u
c
h
a
s
D
e
ns
e
N
e
t
201
-
SVM
,
w
hi
c
h
t
a
k
e
s
6
48
s
e
c
on
ds
of
t
r
a
i
n
i
ng
t
i
m
e
a
nd
i
s
s
i
gn
i
fi
c
a
n
t
l
y
f
a
s
t
e
r
t
ha
n
In
c
e
pt
i
onV
3
-
S
V
M
,
w
hi
c
h
t
a
ke
s
up
t
o
1
,
008
s
e
c
onds
.
P
i
pe
l
i
ne
e
ff
i
c
i
e
n
c
y
i
s
a
l
s
o
i
m
prov
e
d
b
y
a
p
pl
y
i
ng
BO
i
n
t
he
h
ype
rp
a
ra
m
e
t
e
r
a
d
j
us
t
m
e
n
t
pro
c
e
s
s
.
T
hi
s
a
ppro
a
c
h
offe
rs
a
dv
a
nt
a
ge
s
o
ve
r
c
onv
e
nt
i
on
a
l
t
ri
a
l
-
a
nd
-
e
rr
or
m
e
t
hods
,
a
s
i
t
i
s
a
bl
e
t
o
i
d
e
nt
i
fy
t
h
e
op
t
i
m
a
l
c
om
b
i
n
a
t
i
on
o
f
p
a
ra
m
e
t
e
rs
w
i
t
h
f
e
w
e
r
i
t
e
r
a
t
i
ons
a
u
t
om
a
t
i
c
a
l
l
y
.
A
s
a
r
e
s
ul
t
,
t
h
e
m
od
e
l
t
r
a
i
n
i
ng
pro
c
e
s
s
be
c
o
m
e
s
fa
s
t
e
r
,
a
nd
t
h
e
c
o
m
put
a
t
i
on
a
l
burd
e
n
c
a
n
b
e
r
e
d
uc
e
d
s
i
gn
i
fi
c
a
nt
l
y.
T
h
i
s
i
s
c
ruc
i
a
l
,
e
s
p
e
c
i
a
l
l
y
i
n
s
ys
t
e
m
i
m
pl
e
m
e
n
t
a
t
i
on
i
n
c
l
i
ni
c
a
l
e
nvi
ronm
e
n
t
s
t
ha
t
d
e
m
a
nd
hi
g
h
e
ffi
c
i
e
nc
y
a
nd
re
l
i
a
bi
l
i
t
y
.
A
c
c
ord
i
ng
t
o
t
h
e
o
ut
c
om
e
s
of
t
h
e
e
rro
r
a
n
a
l
ys
i
s
,
t
h
e
c
on
fu
s
i
on
m
a
t
ri
x
re
v
e
a
l
s
t
ha
t
m
os
t
i
ns
t
a
nc
e
s
w
e
re
c
l
a
s
s
i
fi
e
d
c
orr
e
c
t
l
y
,
re
fl
e
c
t
i
n
g
s
t
rong
ove
r
a
l
l
m
ode
l
pe
rfor
m
a
nc
e
.
H
ow
e
ve
r,
s
o
m
e
m
i
s
c
l
a
s
s
i
fi
c
a
t
i
o
n
c
a
s
e
s
r
e
m
a
i
n
,
e
s
p
e
c
i
a
l
l
y
i
n
c
l
a
s
s
e
s
w
i
t
h
h
i
gh
v
i
s
ua
l
s
i
m
i
l
a
ri
t
y.
In
t
he
pro
pos
e
d
F
F
-
S
V
M
m
o
de
l
,
fo
r
e
x
a
m
p
l
e
,
t
he
r
e
a
re
5
c
a
s
e
s
of
t
ub
e
r
c
ul
os
i
s
,
4
c
a
s
e
s
of
p
ne
u
m
on
i
a
c
l
a
s
s
i
fi
e
d
a
s
CO
V
ID
-
19
,
a
nd
on
e
no
rm
a
l
c
a
s
e
c
l
a
s
s
i
fi
e
d
a
s
p
ne
u
m
on
i
a
.
T
hi
s
fi
ndi
n
g
re
fl
e
c
t
s
t
he
c
h
a
l
l
e
ng
e
s
i
n
di
s
t
i
ng
ui
s
hi
ng
d
i
s
e
a
s
e
s
t
h
a
t
h
a
v
e
ov
e
rl
a
ppi
ng
ra
di
o
l
og
i
c
a
l
c
h
a
ra
c
t
e
ri
s
t
i
c
s
.
F
i
g
ure
4
pr
e
s
e
n
t
s
a
vi
s
u
a
l
re
p
re
s
e
nt
a
t
i
on
of
t
h
e
m
i
s
c
l
a
s
s
i
fi
c
a
t
i
on
pa
t
t
e
rn
t
hroug
h
t
he
c
onfus
i
o
n
m
a
t
ri
x
f
or
e
a
c
h
m
ode
l
,
w
h
i
c
h
pro
vi
d
e
s
a
c
om
p
re
h
e
ns
i
v
e
ove
rvi
e
w
o
f
t
he
d
i
s
t
ri
but
i
on
of
pre
di
c
t
i
ons
be
t
w
e
e
n
c
l
a
s
s
e
s
.
F
i
gure
4
.
C
onfus
i
o
n
m
a
t
ri
c
e
s
for
di
ff
e
r
e
nt
c
l
a
s
s
i
f
i
e
rs
a
n
d
a
rc
hi
t
e
c
t
ure
s
F
rom
t
he
re
c
e
i
ve
r
op
e
ra
t
i
n
g
c
ha
r
a
c
t
e
r
i
s
t
i
c
(
RO
C
)
c
urv
e
s
i
n
F
i
gure
5,
F
F
-
S
V
M
re
c
ord
e
d
A
U
C
=
1
.
00
for
a
l
l
c
l
a
s
s
e
s
,
i
nd
i
c
a
t
i
ng
a
v
e
ry
hi
gh
c
l
a
s
s
i
f
i
c
a
t
i
on
a
bi
l
i
t
y.
D
T
a
n
d
k
-
N
N
w
e
re
s
l
i
ght
l
y
l
ow
e
r
,
w
i
t
h
a
v
e
ra
g
e
A
U
Cs
of
0
.
98
a
nd
0
.
99
,
r
e
s
pe
c
t
i
ve
l
y.
T
h
e
s
e
re
s
u
l
t
s
c
o
nfi
rm
t
h
e
s
u
pe
r
i
ori
t
y
of
S
V
M
,
e
s
p
e
c
i
a
l
l
y
i
n
di
s
t
i
n
gui
s
h
i
ng
c
l
a
s
s
e
s
s
u
c
h
a
s
T
B
,
a
s
i
ndi
c
a
t
e
d
b
y
i
t
s
pe
rf
e
c
t
A
U
C
of
1.
0
0
a
c
r
os
s
a
l
l
c
l
a
s
s
e
s
.
T
he
propos
e
d
m
e
t
hod'
s
s
up
e
ri
or
i
t
y
i
s
a
l
s
o
a
pp
a
r
e
nt
c
o
m
pa
re
d
t
o
pr
e
vi
o
us
s
t
ud
i
e
s
i
n
T
a
bl
e
6
.
P
re
vi
ous
a
ppro
a
c
he
s
,
s
u
c
h
a
s
V
G
G
N
e
t
a
nd
A
BO
-
CN
N
,
onl
y
a
c
h
i
e
v
e
d
95
.
11
%
a
n
d
96.
95%
a
c
c
ura
c
y
.
E
v
e
n
m
ode
rn
e
ns
e
m
bl
e
a
pp
roa
c
he
s
s
u
c
h
a
s
V
G
G
-
19
+
V
i
s
i
on
T
ra
ns
form
e
r
onl
y
yi
e
l
d
e
d
9
4.
52
%
a
c
c
ura
c
y.
W
i
t
h
a
n
a
c
c
ura
c
y
of
98
.
79%
,
t
h
e
F
F
-
S
V
M
m
e
t
h
od
s
urp
a
s
s
e
s
t
h
e
s
e
re
s
u
l
t
s
a
nd
off
e
rs
b
e
t
t
e
r
t
i
m
e
e
ff
i
c
i
e
n
c
y
a
nd
pe
rfor
m
a
nc
e
s
t
a
bi
l
i
t
y
.
T
he
s
e
r
e
s
ul
t
s
hi
gh
l
i
gh
t
t
he
propos
e
d
a
pproa
c
h'
s
s
ubs
t
a
n
t
i
a
l
c
on
t
ri
bu
t
i
o
ns
a
nd
re
l
e
v
a
n
c
e
for
c
he
s
t
X
-
r
a
y
i
m
a
g
e
-
ba
s
e
d
l
ung
d
i
s
e
a
s
e
d
e
t
e
c
t
i
o
n
s
ys
t
e
m
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
IS
S
N
:
2722
-
3221
H
y
br
i
d
f
e
at
ur
e
f
us
i
on
f
r
om
m
ul
t
i
p
l
e
CNN
m
od
e
l
s
wi
t
h
bay
e
s
i
an
-
opt
i
m
i
z
e
d
m
a
c
hi
n
e
…
(D
e
wi
R
i
s
m
awat
i
)
323
(a
)
(b)
(c
)
F
i
gure
5
.
RO
C
c
urv
e
of
F
F
m
od
e
l
s
(a
)
S
uppo
rt
ve
c
t
or
m
a
c
hi
ne
,
(b)
D
e
c
i
s
i
on
t
re
e
,
a
nd
(c
)
K
-
n
e
a
r
e
s
t
ne
i
ghbors
T
a
b
l
e
6
.
Co
m
p
a
ri
s
on
w
i
t
h
pre
vi
ous
s
t
ud
i
P
re
v
i
o
u
s
s
t
u
d
y
M
e
t
h
o
d
s
D
a
t
a
s
e
t
A
c
c
u
ra
c
y
(%
)
[3
2
]
V
G
G
n
e
t
Ch
e
s
t
X
-
Ra
y
9
5
.
1
1
[3
3
]
M
o
d
e
l
E
n
s
e
m
b
l
e
(V
G
G
-
1
9
,
Re
s
N
e
t
5
0
,
v
i
s
i
o
n
t
ra
n
s
fo
rm
e
r)
Ch
e
s
t
X
-
Ra
y
9
2
.
4
6
9
4
.
5
2
[3
4
]
A
BO
-
CN
N
Ch
e
s
t
X
-
Ra
y
9
6
.
9
5
4.
C
O
N
C
LU
S
I
O
N
T
hi
s
s
t
udy
s
how
s
t
h
a
t
f
us
i
ng
fe
a
t
ur
e
s
fro
m
s
e
ve
r
a
l
CN
N
-
b
a
s
e
d
a
rc
h
i
t
e
c
t
ure
s
c
on
t
ri
but
e
s
t
o
i
m
prov
e
d
c
l
a
s
s
i
fi
c
a
t
i
on
a
c
c
ur
a
c
y
i
n
CX
R
i
m
a
g
e
a
na
l
ys
i
s
us
i
ng
m
a
c
hi
ne
l
e
a
rn
i
ng
m
od
e
l
s
.
T
he
de
ve
l
ope
d
F
F
-
SVM
m
ode
l
a
c
hi
e
v
e
d
t
h
e
hi
g
he
s
t
a
c
c
ur
a
c
y
of
9
8.
7
9%
i
n
c
l
a
s
s
i
fyi
n
g
fo
ur
l
ung
di
s
e
a
s
e
c
a
t
e
gor
i
e
s
:
N
orm
a
l
,
CO
V
ID
-
19,
P
n
e
u
m
oni
a
,
a
nd
T
ub
e
rc
ul
os
i
s
.
T
he
s
e
r
e
s
ul
t
s
i
n
di
c
a
t
e
t
h
e
gre
a
t
pot
e
n
t
i
a
l
of
t
hi
s
a
ppr
oa
c
h
t
o
b
e
a
pp
l
i
e
d
i
n
m
e
d
i
c
a
l
d
i
a
gn
os
i
s
s
u
pport
s
ys
t
e
m
s
,
e
s
p
e
c
i
a
l
l
y
i
n
a
pp
l
i
c
a
t
i
on
s
t
h
a
t
a
re
e
a
s
i
l
y
a
c
c
e
s
s
i
bl
e
i
n
a
re
a
s
w
i
t
h
l
i
m
i
t
e
d
he
a
l
t
h
f
a
c
i
l
i
t
i
e
s
.
H
ow
e
ve
r
,
t
h
i
s
s
t
udy
h
a
s
s
om
e
l
i
m
i
t
a
t
i
ons
.
T
he
m
od
e
l
w
a
s
t
r
a
i
ne
d
us
i
ng
on
l
y
CX
R
i
m
a
g
e
s
w
i
t
hou
t
c
ons
i
d
e
ri
n
g
ot
he
r
m
e
di
c
a
l
d
a
t
a
s
uc
h
a
s
C
T
s
c
a
ns
,
M
RIs
,
or
a
d
di
t
i
on
a
l
c
l
i
ni
c
a
l
d
a
t
a
.
In
a
ddi
t
i
on
,
pot
e
n
t
i
a
l
b
i
a
s
e
s
i
n
t
h
e
da
t
a
s
e
t
a
n
d
r
e
l
i
a
nc
e
on
i
m
a
g
e
qua
l
i
t
y
m
a
y
l
i
m
i
t
t
he
m
od
e
l
'
s
g
e
n
e
ra
l
i
z
a
t
i
on
t
o
a
bro
a
d
e
r
popul
a
t
i
on.
T
e
s
t
i
ng
h
a
s
no
t
b
e
e
n
c
ondu
c
t
e
d
on
e
xt
e
rna
l
d
a
t
a
s
e
t
s
nor
i
n
t
he
c
o
nt
e
xt
of
r
e
a
l
us
e
i
n
a
c
l
i
n
i
c
a
l
e
nvi
r
onm
e
nt
.
A
s
a
d
e
ve
l
op
m
e
n
t
d
i
re
c
t
i
on,
e
x
pl
or
i
ng
m
ul
t
i
m
oda
l
da
t
a
f
u
s
i
on
s
t
ra
t
e
g
i
e
s
a
nd
va
l
i
d
a
t
i
on
i
n
re
a
l
-
w
orl
d
s
c
e
na
r
i
os
c
o
ul
d
b
e
a
n
i
m
port
a
n
t
s
t
e
p
t
o
i
m
pro
ve
r
e
l
i
a
b
i
l
i
t
y
a
nd
bro
a
de
r
a
pp
l
i
c
a
b
i
l
i
t
y
.
T
he
m
e
t
hodo
l
ogy
a
l
s
o
s
uppor
t
s
t
he
us
e
of
a
rt
i
fi
c
i
a
l
i
n
t
e
l
l
i
ge
n
c
e
(A
I)
s
ys
t
e
m
s
i
n
h
e
a
l
t
h
c
a
r
e
,
w
h
i
c
h
c
a
n
b
e
e
xt
e
nd
e
d
t
o
o
t
he
r
m
e
d
i
c
a
l
i
m
a
g
i
ng
m
oda
l
i
t
i
e
s
a
n
d
d
i
ff
e
re
n
t
di
s
e
a
s
e
c
a
t
e
gor
i
e
s
.
A
C
K
N
O
WL
ED
G
M
EN
TS
D
uri
n
g
t
hi
s
re
s
e
a
rc
h,
t
he
a
ut
h
ors
t
ha
nk
ou
r
fri
e
nd
s
f
o
r
t
he
i
r
e
nc
oura
ge
m
e
nt
,
s
upp
ort
,
a
nd
m
ot
i
va
t
i
on.
T
ha
nks
t
o
t
he
i
r
t
re
m
e
nd
ous
a
s
s
i
s
t
a
nc
e
,
t
he
a
ut
hor
s
ha
ve
be
e
n
a
bl
e
t
o
s
t
a
y
m
ot
i
va
t
e
d
a
nd
foc
us
e
d.
T
he
a
ut
hor
s
a
l
s
o
a
c
kn
ow
l
e
d
ge
t
he
i
nva
l
ua
bl
e
a
dvi
c
e
,
know
l
e
dge
,
a
nd
s
ug
ge
s
t
i
o
ns
from
o
ur
s
upe
r
vi
s
o
rs
,
w
hi
c
h
s
i
g
ni
fi
c
a
nt
l
y
ra
i
s
e
d
t
he
c
a
l
i
be
r
o
f
ou
r
s
t
udy.
A
d
di
t
i
ona
l
l
y,
t
he
a
ut
ho
rs
a
re
gra
t
e
f
ul
t
o
t
h
e
j
ourna
l
e
di
t
o
rs
a
n
d
re
vi
e
w
e
rs
for
t
he
i
r
he
l
pf
ul
c
ri
t
i
c
i
s
m
a
nd
re
c
om
m
e
nda
t
i
on
s
,
w
hi
c
h
e
nha
nc
e
d
t
h
e
m
a
nus
c
ri
pt
'
s
c
on
s
i
s
t
e
nc
y
a
nd
c
l
a
ri
t
y.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t
S
c
i
Inf
T
e
c
h
nol
,
V
o
l
.
6
,
N
o
.
3
,
N
ov
e
m
be
r
20
25
:
315
-
325
324
F
U
N
D
I
N
G
I
N
F
O
R
M
A
TI
O
N
A
ut
hors
s
t
a
t
e
no
fund
i
ng
i
nvo
l
v
e
d.
A
U
TH
O
R
C
O
N
TR
I
BU
TI
O
N
S
S
TA
T
EM
EN
T
T
hi
s
j
ou
rna
l
us
e
s
t
h
e
Cont
r
i
bu
t
or
Rol
e
s
T
a
xono
m
y
(C
Re
di
T
)
t
o
r
e
c
o
gni
z
e
i
ndi
v
i
du
a
l
a
u
t
hor
c
ont
r
i
bu
t
i
ons
,
r
e
du
c
e
a
ut
hors
h
i
p
di
s
pu
t
e
s
,
a
nd
fa
c
i
l
i
t
a
t
e
c
o
l
l
a
bora
t
i
on
.
N
ame
of
A
u
th
o
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
D
e
w
i
Ri
s
m
a
w
a
t
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
S
ugi
ya
r
t
o
S
urono
✓
✓
✓
✓
✓
✓
✓
A
ri
s
T
hobi
r
i
n
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
ft
w
a
re
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
rm
a
l
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
s
o
u
rc
e
s
D
:
D
a
t
a
Cu
ra
t
i
o
n
O
:
W
ri
t
i
n
g
-
O
ri
g
i
n
a
l
D
ra
ft
E
:
W
ri
t
i
n
g
-
Re
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
s
u
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
rv
i
s
i
o
n
P
:
P
ro
j
e
c
t
a
d
m
i
n
i
s
t
ra
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
s
i
t
i
o
n
C
O
N
F
LI
C
T
O
F
I
N
T
ER
ES
T
S
TA
T
EM
EN
T
A
ut
hors
s
t
a
t
e
no
c
onf
l
i
c
t
of
i
n
t
e
r
e
s
t
.
D
A
TA
A
V
A
I
LA
BI
LI
TY
T
he
da
t
a
t
ha
t
s
upp
ort
t
h
e
fi
nd
i
ngs
of
t
hi
s
s
t
udy
a
re
op
e
nl
y
a
v
a
i
l
a
bl
e
i
n
ka
ggl
e
a
t
t
h
e
fol
l
ow
i
ng
l
i
nks
:
‒
D
a
t
a
s
e
t
1
-
CO
V
ID
-
19
,
P
n
e
u
m
oni
a
,
a
nd
N
orm
a
l
CX
R
Im
a
g
e
s
‒
D
a
t
a
s
e
t
2
-
T
ub
e
r
c
ul
os
i
s
CX
R
(S
he
nz
h
e
n)
‒
D
a
t
a
s
e
t
3
-
T
ub
e
r
c
ul
os
i
s
CX
R
(M
on
t
go
m
e
ry
)
‒
D
a
t
a
s
e
t
4
-
T
BX
1
1
CX
R
(us
m
a
ns
h
a
m
s
)
R
EF
ER
EN
C
ES
[1
]
W
o
rl
d
H
e
a
l
t
h
O
rg
a
n
i
z
a
t
i
o
n
,
W
o
r
l
d
h
e
a
l
t
h
s
t
a
t
i
s
t
i
c
s
2
0
2
4
:
m
o
n
i
t
o
r
i
n
g
h
e
a
l
t
h
f
o
r
t
h
e
S
D
G
s
s
u
s
t
a
i
n
a
b
l
e
d
e
v
e
l
o
p
m
e
n
t
g
o
a
l
s
.
G
e
n
e
v
a
,
S
w
i
t
z
e
rl
a
n
d
:
W
o
rl
d
H
e
a
l
t
h
O
rg
a
n
i
z
a
t
i
o
n
,
2
0
2
4
.
[2
]
S
.
K
.
Z
h
o
u
e
t
a
l
.
,
“
A
re
v
i
e
w
o
f
d
e
e
p
l
e
a
rn
i
n
g
i
n
m
e
d
i
c
a
l
i
m
a
g
i
n
g
:
i
m
a
g
i
n
g
t
ra
i
t
s
,
t
e
c
h
n
o
l
o
g
y
t
re
n
d
s
,
c
a
s
e
s
t
u
d
i
e
s
w
i
t
h
p
ro
g
r
e
s
s
h
i
g
h
l
i
g
h
t
s
,
a
n
d
fu
t
u
re
p
ro
m
i
s
e
s
,
”
i
n
P
r
o
c
e
e
d
i
n
g
s
o
f
t
h
e
IE
E
E
,
2
0
2
1
,
v
o
l
.
1
0
9
,
n
o
.
5
,
p
p
.
8
2
0
–
838.
[3
]
W
.
M
a
a
n
d
B.
G
u
o
,
“
A
d
v
a
n
c
e
d
n
e
u
ro
l
o
g
i
c
a
l
a
c
t
i
v
i
t
y
s
t
a
t
u
s
o
f
a
t
h
l
e
t
e
s
b
a
s
e
d
o
n
b
i
g
d
a
t
a
t
e
c
h
n
o
l
o
g
y
,
”
H
e
l
i
y
o
n
,
v
o
l
.
1
0
,
n
o
.
1
7
,
2
0
2
4
.
[4
]
M
.
F
.
I
j
a
z
a
n
d
M
.
W
o
ź
n
i
a
k
,
“
E
d
i
t
o
ri
a
l
:
Re
c
e
n
t
a
d
v
a
n
c
e
s
i
n
d
e
e
p
l
e
a
rn
i
n
g
a
n
d
m
e
d
i
c
a
l
i
m
a
g
i
n
g
fo
r
c
a
n
c
e
r
t
re
a
t
m
e
n
t
,
”
Ca
n
c
e
r
s
,
v
o
l
.
1
6
,
n
o
.
4
,
2
0
2
4
.
[5
]
S
.
T
.
H
.
K
i
e
u
,
A
.
Ba
d
e
,
M
.
H
.
A
.
H
i
j
a
z
i
,
a
n
d
H
.
K
o
l
i
v
a
n
d
,
“
A
s
u
rv
e
y
o
f
d
e
e
p
l
e
a
rn
i
n
g
fo
r
l
u
n
g
d
i
s
e
a
s
e
d
e
t
e
c
t
i
o
n
o
n
m
e
d
i
c
a
l
i
m
a
g
e
s
:
s
t
a
t
e
-
of
-
t
h
e
-
a
rt
,
t
a
x
o
n
o
m
y
,
i
s
s
u
e
s
a
n
d
fu
t
u
re
d
i
re
c
t
i
o
n
s
,
”
J
o
u
r
n
a
l
o
f
Im
a
g
i
n
g
,
v
o
l
.
6
,
n
o
.
1
2
,
2
0
2
0
.
[6
]
G
.
G
a
u
t
a
m
a
n
d
A
.
K
h
a
n
n
a
,
“
Co
n
t
e
n
t
b
a
s
e
d
i
m
a
g
e
re
t
ri
e
v
a
l
s
y
s
t
e
m
u
s
i
n
g
CN
N
b
a
s
e
d
d
e
e
p
l
e
a
rn
i
n
g
m
o
d
e
l
s
,
”
i
n
P
r
o
c
e
d
i
a
Co
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
2
3
5
,
p
p
.
3
1
3
1
–
3
1
4
1
,
2
0
2
4
.
[7
]
G
.
H
a
m
e
d
,
M
.
A
.
E
.
R.
M
a
re
y
,
S
.
E
.
S
.
A
m
i
n
,
a
n
d
M
.
F
.
T
o
l
b
a
,
“
D
e
e
p
l
e
a
rn
i
n
g
i
n
b
re
a
s
t
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
s
s
i
fi
c
a
t
i
o
n
,
”
A
d
v
a
n
c
e
s
i
n
In
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s
a
n
d
Co
m
p
u
t
i
n
g
,
p
p
.
3
2
2
–
3
3
3
,
2
0
2
0
.
[8
]
A
.
B.
Ra
m
a
k
ri
s
h
n
a
n
,
M
.
S
r
i
d
e
v
i
,
S
.
K
.
V
a
s
u
d
e
v
a
n
,
R.
M
a
n
i
k
a
n
d
a
n
,
a
n
d
A
.
H
.
G
a
n
d
o
m
i
,
“
O
p
t
i
m
i
z
i
n
g
b
ra
i
n
t
u
m
o
r
c
l
a
s
s
i
fi
c
a
t
i
o
n
w
i
t
h
h
y
b
ri
d
CN
N
a
rc
h
i
t
e
c
t
u
re
:
b
a
l
a
n
c
i
n
g
a
c
c
u
ra
c
y
a
n
d
e
ffi
c
i
e
n
c
y
t
h
ro
u
g
h
o
n
e
A
P
I
o
p
t
i
m
i
z
a
t
i
o
n
,
”
In
f
o
r
m
a
t
i
c
s
i
n
M
e
d
i
c
i
n
e
U
n
l
o
c
k
e
d
,
v
o
l
.
4
4
,
2
0
2
4
.
[9
]
K
.
G
u
p
t
a
a
n
d
N
.
Ch
a
w
l
a
,
“
A
n
a
l
y
s
i
s
o
f
h
i
s
t
o
p
a
t
h
o
l
o
g
i
c
a
l
i
m
a
g
e
s
fo
r
p
r
e
d
i
c
t
i
o
n
o
f
b
re
a
s
t
c
a
n
c
e
r
u
s
i
n
g
t
ra
d
i
t
i
o
n
a
l
c
l
a
s
s
i
fi
e
rs
w
i
t
h
p
re
-
t
ra
i
n
e
d
CN
N
,
”
i
n
P
r
o
c
e
d
i
a
Co
m
p
u
t
e
r
S
c
i
e
n
c
e
,
2
0
2
0
,
v
o
l
.
1
6
7
,
p
p
.
8
7
8
–
889.
[1
0
]
M
.
F
.
A
s
l
a
n
,
K
.
S
a
b
a
n
c
i
,
A
.
D
u
rd
u
,
a
n
d
M
.
F
.
U
n
l
e
rs
e
n
,
“
CO
V
I
D
-
1
9
d
i
a
g
n
o
s
i
s
u
s
i
n
g
s
t
a
t
e
-
of
-
t
h
e
-
a
rt
CN
N
a
rc
h
i
t
e
c
t
u
re
fe
a
t
u
re
s
a
n
d
Ba
y
e
s
i
a
n
o
p
t
i
m
i
z
a
t
i
o
n
,
”
Co
m
p
u
t
e
r
s
i
n
B
i
o
l
o
g
y
a
n
d
M
e
d
i
c
i
n
e
,
v
o
l
.
1
4
2
,
2
0
2
2
.
[1
1
]
A
.
Bi
s
w
a
s
a
n
d
M
.
S
.
Is
l
a
m
,
“
A
h
y
b
ri
d
d
e
e
p
CN
N
-
S
V
M
a
p
p
ro
a
c
h
fo
r
b
ra
i
n
t
u
m
o
r
c
l
a
s
s
i
fi
c
a
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
In
f
o
r
m
a
t
i
o
n
S
y
s
t
e
m
s
E
n
g
i
n
e
e
r
i
n
g
a
n
d
B
u
s
i
n
e
s
s
In
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
9
,
n
o
.
1
,
2
0
2
3
.
[1
2
]
X
.
Z
h
a
o
,
C.
T
a
n
g
,
H
.
H
u
,
W
.
W
a
n
g
,
S
.
Q
i
a
o
,
a
n
d
A
.
T
o
n
g
,
“
A
t
t
e
n
t
i
o
n
m
e
c
h
a
n
i
s
m
b
a
s
e
d
m
u
l
t
i
m
o
d
a
l
fe
a
t
u
re
fu
s
i
o
n
n
e
t
w
o
rk
fo
r
h
u
m
a
n
a
c
t
i
o
n
re
c
o
g
n
i
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
V
i
s
u
a
l
Co
m
m
u
n
i
c
a
t
i
o
n
a
n
d
Im
a
g
e
R
e
p
r
e
s
e
n
t
a
t
i
o
n
,
v
o
l
.
1
1
0
,
2
0
2
5
.
[1
3
]
I.
T
.
J
o
l
l
i
ffe
,
“
P
r
i
n
c
i
p
a
l
c
o
m
p
o
n
e
n
t
a
n
a
l
y
s
i
s
,
”
i
n
S
p
r
i
n
g
e
r
s
e
r
i
e
s
i
n
s
t
a
t
i
s
t
i
c
s
,
2
n
d
e
d
.
,
N
e
w
Y
o
rk
,
U
S
A
:
S
p
ri
n
g
e
r,
2
0
0
2
.
[1
4
]
S
.
A
.
S
a
l
l
e
h
e
t
a
l
.
,
“
S
u
p
p
o
rt
v
e
c
t
o
r
m
a
c
h
i
n
e
(S
V
M
)
a
n
d
o
b
j
e
c
t
-
b
a
s
e
d
c
l
a
s
s
i
fi
c
a
t
i
o
n
i
n
e
a
rt
h
l
i
n
e
a
r
fe
a
t
u
re
s
e
x
t
ra
c
t
i
o
n
:
a
c
o
m
p
a
ri
s
o
n
,
”
R
e
v
u
e
In
t
e
r
n
a
t
i
o
n
a
l
e
d
e
G
é
o
m
a
t
i
q
u
e
,
v
o
l
.
3
3
,
n
o
.
1
,
p
p
.
1
8
3
–
1
9
9
,
2
0
2
4
.
[1
5
]
J
.
W
.
H
a
n
,
M
.
K
a
m
b
e
r,
a
n
d
J
.
P
e
i
,
“
Cl
a
s
s
i
fi
c
a
t
i
o
n
:
b
a
s
i
c
c
o
n
c
e
p
t
s
,
”
i
n
D
a
t
a
m
i
n
i
n
g
:
c
o
n
c
e
p
t
s
a
n
d
t
e
c
h
n
i
q
u
e
s
,
Bu
rl
i
n
g
t
o
n
,
M
a
s
s
a
c
h
u
s
e
t
t
s
:
M
o
rg
a
n
K
a
u
fm
a
n
n
,
2
0
1
2
.
[1
6
]
Z
.
E
.
Ra
s
j
i
d
a
n
d
R.
S
e
t
i
a
w
a
n
,
“
P
e
rfo
rm
a
n
c
e
c
o
m
p
a
ri
s
o
n
a
n
d
o
p
t
i
m
i
z
a
t
i
o
n
o
f
t
e
x
t
d
o
c
u
m
e
n
t
c
l
a
s
s
i
fi
c
a
t
i
o
n
u
s
i
n
g
k
-
N
N
a
n
d
n
a
ï
v
e
Ba
y
e
s
c
l
a
s
s
i
fi
c
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
,
”
P
r
o
c
e
d
i
a
Co
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
1
6
,
p
p
.
1
0
7
–
1
1
2
,
2
0
1
7
.
[1
7
]
Y
.
Y
a
o
,
J
.
Ca
o
,
a
n
d
Z
.
M
a
,
“
A
c
o
s
t
-
e
ffe
c
t
i
v
e
d
e
a
d
l
i
n
e
-
c
o
n
s
t
ra
i
n
e
d
s
c
h
e
d
u
l
i
n
g
s
t
ra
t
e
g
y
fo
r
a
h
y
p
e
r
p
a
ra
m
e
t
e
r
o
p
t
i
m
i
z
a
t
i
o
n
w
o
rk
fl
o
w
fo
r
m
a
c
h
i
n
e
l
e
a
rn
i
n
g
a
l
g
o
ri
t
h
m
s
,
”
i
n
S
e
r
v
i
c
e
-
O
r
i
e
n
t
e
d
Co
m
p
u
t
i
n
g
,
v
o
l
.
1
1
2
3
6
,
N
e
w
Y
o
rk
,
U
S
A
:
S
p
ri
n
g
e
r,
Ch
a
m
,
2018.
Evaluation Warning : The document was created with Spire.PDF for Python.