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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
, pp.
516
~
526
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
1
4
.i
1
.pp
516
-
526
516
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
A
r
e
vol
u
t
i
on
ar
y
c
on
vol
u
t
i
on
al
n
e
u
r
al
n
e
t
w
or
k
ar
c
h
i
t
e
c
t
u
r
e
f
or
m
or
e
ac
c
u
r
at
e
l
u
n
g c
an
c
e
r
c
l
ass
i
f
i
c
at
i
on
M
u
li
ad
i
1
, A
gu
s
P
e
r
d
an
a Win
d
ar
t
o
2
,
S
ol
ik
h
u
n
3
,
P
u
t
r
am
a A
lk
h
ai
r
i
2
1
D
e
pa
r
t
m
e
nt
o
f
C
om
pu
t
e
r
S
c
i
e
nc
e
,
F
a
c
u
l
t
y
of
M
a
t
he
m
a
t
i
c
s
a
nd
N
a
t
u
r
a
l
S
c
i
e
nc
e
s
,
L
a
m
bu
ng
M
a
n
gku
r
a
t
U
ni
ve
r
s
i
t
y,
B
a
nj
a
r
m
a
s
i
n,
I
n
do
ne
s
i
a
2
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
on
S
ys
t
e
m
s
, S
T
I
K
O
M
T
una
s
B
a
ngs
a
, P
e
m
a
t
a
ngs
i
a
nt
a
r
,
N
or
t
h S
um
a
t
r
a
,
I
ndone
s
i
a
3
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
c
s
E
ngi
ne
e
r
i
ng
,
S
T
I
K
O
M
T
una
s
B
a
ngs
a
, P
e
m
a
t
a
ngs
i
a
nt
a
r
,
N
or
t
h S
um
a
t
r
a
,
I
ndone
s
i
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
M
a
r
12
2024
R
e
vi
s
e
d
J
ul
19
,
2024
A
c
c
e
pt
e
d
J
ul
26
,
2024
This
research
aimed
to
inv
estigate
a
breakthrough
in
convolut
ional
neural
network
(CNN)
architecture
with
the
potential
to
revolutionize
lung
cancer
classifi
cation.
The
proposed
method
is
a
comparativ
e
optimi
zation
m
odel
of
ResNet
architecture,
with
accuracy
rate
of
99.68%
in
identifyin
g
and
categorizin
g lung
cancer types
. The
results
showed that
the use of
inn
ovative
methods
in
CNN
architec
ture,
such
as
multi
-
dimensional
convolutional
layers
and
the
integration
of
specific
lung
cancer
features,
effectively
pr
ovided
highly
accurate
and
reliable
outcomes.
These
showed
a
positive
impact
on
the
development
of
medical
diagnostic
technology,
offering
promising
potential
to
enhance
prognosis
and
response
to
treatment
for
lung
cancer
patients
.
With
high
accuracy
ra
te,
this
breakthrough
presents
opportunities
for
further
advancement
s
in
lung
cancer
management
through
artificial
intell
i
gence
-
based methods.
K
e
y
w
o
r
d
s
:
C
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
a
r
c
hi
te
c
tu
r
e
H
ig
h a
c
c
ur
a
c
y
L
ung c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on
R
e
vol
ut
io
na
r
y
T
e
c
hnol
ogy innovation
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
:
M
ul
ia
di
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
S
c
ie
nc
e
,
F
a
c
ul
ty
of
M
a
th
e
m
a
ti
c
s
a
nd
N
a
tu
r
a
l
S
c
ie
nc
e
s
L
a
m
bung M
a
ngkur
a
t
U
ni
ve
r
s
it
y
J
l.
A
. Y
a
ni
, K
M
. 36,
L
a
m
bung M
a
ngkur
a
t
U
ni
ve
r
s
it
y, B
ui
ld
in
g
I
I
, F
l.
3, B
a
nj
a
r
ba
r
u
, I
ndone
s
ia
.
E
m
a
il
:
m
ul
ia
di
@
ul
m
.a
c
.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
L
ung
c
a
nc
e
r
is
a
m
ong
th
e
m
os
t
pr
e
va
le
nt
a
nd
de
a
dl
y
ty
pe
s
of
c
a
nc
e
r
gl
oba
ll
y,
oc
c
ur
r
in
g
f
r
om
unc
ont
r
ol
le
d gr
ow
th
i
n l
ung
t
is
s
ue
f
or
m
in
g t
um
o
r
s
[
1]
–
[
3
]
.
T
hi
s
di
s
e
a
s
e
doe
s
not
of
te
n s
how
s
ym
pt
om
s
i
n t
he
e
a
r
ly
s
ta
ge
s
,
le
a
di
ng
to
la
te
di
a
gnos
i
s
w
he
n
tr
e
a
tm
e
nt
be
c
om
e
s
m
or
e
c
ha
ll
e
ngi
ng.
T
he
m
a
in
r
is
k
f
a
c
to
r
s
f
or
lu
ng c
a
nc
e
r
i
nc
lu
de
e
xpos
ur
e
t
o c
a
r
c
in
oge
ni
c
s
ub
s
ta
nc
e
s
s
uc
h a
s
c
ig
a
r
e
tt
e
s
m
oke
, a
ir
pol
lu
ti
on, a
nd
r
a
don
[
4]
,
[
5]
.
T
he
r
e
f
or
e
,
th
e
r
e
is
a
ne
e
d
f
or
c
ont
in
uous
r
e
s
e
a
r
c
h
a
nd
in
nova
ti
on
in
th
e
f
ie
ld
of
l
ung
c
a
nc
e
r
di
a
gnos
ti
c
a
nd
tr
e
a
tm
e
nt
a
im
in
g
a
t
im
pr
ovi
ng
th
e
pr
ognos
is
a
s
w
e
ll
a
s
li
f
e
qua
li
ty
of
pa
ti
e
nt
s
.
S
e
ve
r
a
l
s
tr
a
te
gi
e
s
ha
ve
b
e
e
n
im
pl
e
m
e
nt
e
d
to
de
ve
lo
p
m
or
e
s
e
n
s
it
i
ve
a
nd
a
c
c
ur
a
t
e
di
a
gnos
t
ic
m
e
th
ods
,
a
lo
ng
w
it
h
e
f
f
e
c
ti
ve
a
nd
c
ur
a
ti
ve
th
e
r
a
pi
e
s
t
o a
ddr
e
s
s
t
h
e
gl
oba
l
bur
de
n of
l
ung c
a
nc
e
r
[
4]
–
[
7]
. O
th
e
r
pr
e
ve
nt
iv
e
a
nd ma
na
ge
m
e
nt
s
te
p
s
i
nc
lu
de
c
a
m
pa
ig
ns
to
a
voi
d
m
a
jo
r
r
is
k
f
a
c
to
r
s
s
uc
h
a
s
s
m
oki
ng
a
nd
in
c
r
e
a
s
in
g
a
w
a
r
e
ne
s
s
of
e
a
r
ly
de
te
c
ti
on
[
8]
.
C
ons
e
que
nt
ly
, t
hi
s
r
e
s
e
a
r
c
h f
oc
us
e
s
on t
he
a
ppl
ic
a
ti
on of
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
t
e
c
hnol
ogy in t
he
de
ve
lo
pm
e
nt
of
a
r
e
vol
ut
io
na
r
y
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
)
a
r
c
h
it
e
c
tu
r
e
to
im
p
r
ove
a
c
c
ur
a
c
y
in
lu
ng
c
a
nc
e
r
di
a
gnos
is
of
te
n
hi
nde
r
e
d
by
m
or
phol
ogi
c
a
l
a
nd
ge
ne
ti
c
va
r
ia
ti
ons
th
a
t
a
r
e
di
f
f
ic
ul
t
to
id
e
nt
i
f
y
us
in
g
c
onve
nt
io
na
l
m
e
th
ods
[
9]
–
[
11]
.
T
hi
s
r
e
s
e
a
r
c
h
a
ppl
ie
s
va
r
io
us
m
e
th
ods
in
c
lu
di
ng
th
e
us
e
of
m
ul
ti
-
di
m
e
ns
io
na
l
c
onvolut
io
na
l
la
ye
r
s
a
nd
th
e
in
te
gr
a
ti
on
of
s
pe
c
if
ic
lu
ng
c
a
n
c
e
r
f
e
a
tu
r
e
s
in
to
C
N
N
a
r
c
hi
te
c
tu
r
e
.
I
n
th
e
de
ve
lo
pm
e
nt
of
C
N
N
,
s
e
ve
r
a
l
popula
r
a
r
c
hi
te
c
tu
r
e
t
ha
t
ha
ve
s
uc
c
e
s
s
f
ul
ly
be
e
n e
f
f
e
c
ti
ve
i
n va
r
io
us
i
m
a
ge
pr
oc
e
s
s
in
g a
nd c
la
s
s
if
ic
a
ti
on t
a
s
ks
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
A
r
e
v
ol
ut
io
nar
y
c
onv
ol
ut
io
nal
ne
u
r
al
ne
tw
or
k
ar
c
hi
te
c
tu
r
e
f
or
m
or
e
ac
c
u
r
at
e
…
(
M
ul
ia
di
)
517
in
c
lu
de
L
e
N
e
t
-
5,
A
le
xN
e
t,
vi
s
ua
l
ge
om
e
tr
y
gr
oup
ne
twor
k
(
V
G
G
N
e
t)
,
G
oogL
e
N
e
t
(
I
nc
e
pt
io
n)
,
r
e
s
id
ua
l
ne
twor
k
(
R
e
s
N
e
t)
,
M
obi
le
N
e
t,
E
f
f
ic
ie
nt
N
e
t,
a
nd
D
e
ns
e
N
e
t
[
12]
–
[
17]
.
L
e
N
e
t
-
5
ha
s
m
a
de
s
ig
ni
f
ic
a
nt
c
ont
r
ib
ut
io
ns
to
C
N
N
de
ve
lo
pm
e
nt
,
w
hi
le
A
le
xN
e
t,
de
ve
lo
p
e
d
by
A
le
x
K
r
iz
he
vs
ky,
I
ly
a
S
ut
s
ke
v
e
r
,
a
nd
G
e
of
f
r
e
y
H
in
to
n,
is
known
f
or
it
s
d
e
e
pe
r
a
r
c
hi
te
c
tu
r
e
a
nd
la
r
g
e
r
c
onvolut
io
n
la
ye
r
s
.
F
ur
th
e
r
m
or
e
,
V
G
G
N
e
t,
de
ve
lo
pe
d
a
t
th
e
U
ni
ve
r
s
it
y
of
O
xf
or
d,
is
kno
w
n
f
or
ve
r
y
de
e
p
s
tr
uc
tu
r
e
us
in
g
r
e
la
ti
ve
ly
s
m
a
ll
c
onvolut
io
n
bl
oc
ks
,
c
ons
is
ti
ng
of
16
or
19
la
ye
r
s
,
de
pe
ndi
ng
on
th
e
va
r
i
a
nt
us
e
d.
G
oogL
e
N
e
t,
de
ve
lo
pe
d
a
t
G
oogl
e
,
in
tr
oduc
e
s
th
e
"
I
nc
e
pt
io
n"
m
odul
e
w
hi
c
h
us
e
s
m
ul
ti
pl
e
f
il
te
r
s
iz
e
s
in
one
la
ye
r
to
c
a
pt
ur
e
f
e
a
tu
r
e
s
a
t
va
r
io
us
s
c
a
le
s
a
nd
im
pr
ove
r
e
pr
e
s
e
nt
a
ti
on
s
ki
ll
s
[
18]
,
[
19]
.
R
e
s
N
e
t
w
a
s
de
ve
lo
pe
d
by
K
a
im
in
g
H
e
a
nd
th
e
te
a
m
a
t
M
ic
r
os
of
t
R
e
s
e
a
r
c
h,
us
in
g
r
e
s
id
ua
l
bl
oc
ks
th
a
t
a
ll
ow
de
e
pe
r
m
ode
l
tr
a
in
in
g
w
it
hou
t
e
xpe
r
ie
nc
in
g
va
ni
s
hi
ng
gr
a
di
e
nt
pr
obl
e
m
s
.
M
obi
le
N
e
t
is
de
s
ig
ne
d
f
or
m
obi
le
de
vi
c
e
a
ppl
ic
a
ti
ons
,
opt
im
iz
in
g
s
pe
e
d,
a
nd
c
om
put
a
ti
on
a
l
r
e
qui
r
e
m
e
nt
s
us
in
g
de
pt
hw
is
e
s
e
pa
r
a
bl
e
c
onvolut
io
ns
.
E
f
f
ic
ie
nt
N
e
t
c
om
bi
ne
s
s
c
a
li
ng
m
e
th
od
s
to
opt
im
iz
e
m
ode
l
e
f
f
ic
ie
nc
y
r
e
ga
r
di
ng
s
iz
e
,
s
pe
e
d,
a
nd
a
c
c
ur
a
c
y.
D
e
ns
e
ly
c
onne
c
te
d
c
onvolut
io
na
l
ne
twor
k
(
D
e
ns
e
N
e
t
)
is
de
ve
lo
pe
d
to
a
ddr
e
s
s
th
e
v
a
ni
s
hi
n
g
a
nd
e
xpl
odi
ng
gr
a
di
e
nt
pr
obl
e
m
s
,
us
in
g
c
lo
s
e
r
c
onne
c
ti
ons
be
tw
e
e
n
la
ye
r
s
in
th
e
f
or
m
of
"
de
ns
e
bl
oc
ks
"
.
E
a
c
h
a
r
c
hi
te
c
tu
r
e
ha
s
s
pe
c
if
ic
c
ha
r
a
c
te
r
is
ti
c
s
,
w
hi
c
h
a
r
e
us
e
d
a
c
c
or
di
ng
to
r
e
qui
r
e
d
ta
s
ks
a
nd
ne
e
ds
.
M
e
a
nw
hi
l
e
,
th
e
s
e
le
c
ti
on
of
C
N
N
a
r
c
hi
te
c
tu
r
e
c
a
n
s
ig
ni
f
ic
a
nt
ly
in
f
lu
e
nc
e
th
e
pe
r
f
or
m
a
nc
e
a
nd r
e
s
ul
ts
of
t
he
de
ve
lo
pe
d m
ode
l
[
18]
–
[
22]
.
R
e
ga
r
di
ng
th
e
popula
r
a
r
c
hi
te
c
tu
r
e
in
pr
e
vi
ous
r
e
s
e
a
r
c
h
[
23]
,
s
e
ve
r
a
l
s
ig
ni
f
ic
a
nt
a
dva
nt
a
ge
s
ha
v
e
be
e
n
doc
um
e
nt
e
d.
T
he
pr
opos
e
d
e
n
s
e
m
bl
e
le
a
r
ni
ng
of
R
e
s
N
e
t,
E
f
f
ic
ie
nt
N
e
t,
a
nd
I
nc
e
pt
io
n
-
v3
w
it
h
w
e
ig
ht
e
d
vot
in
g
(
E
L
R
E
I
)
m
e
th
od
s
uc
c
e
s
s
f
ul
ly
a
ddr
e
s
s
e
s
li
m
it
a
ti
ons
a
nd
le
ve
r
a
ge
s
th
e
s
tr
e
ngt
hs
of
e
a
c
h
C
N
N
a
r
c
hi
te
c
tu
r
e
,
na
m
e
ly
R
e
s
N
e
t,
E
f
f
ic
ie
nt
N
e
t,
a
nd
I
nc
e
pt
io
n
-
v3.
T
hi
s
m
e
th
od
a
c
hi
e
ve
s
e
x
c
e
ll
e
nt
r
e
s
ul
ts
of
m
or
e
th
a
n
98%
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
,
w
it
h
th
e
pot
e
nt
ia
l
to
ove
r
c
om
e
ove
r
f
it
ti
ng,
w
hi
c
h
is
of
te
n
e
nc
ount
e
r
e
d
in
C
N
N
a
r
c
hi
te
c
tu
r
e
.
T
hi
s
r
e
s
e
a
r
c
h
a
ls
o
pr
ovi
de
s
a
pe
r
f
or
m
a
nc
e
c
om
pa
r
is
on
of
th
e
E
L
R
E
I
m
e
th
od
w
it
h
pr
e
vi
ous
r
e
s
ul
ts
,
s
how
in
g
e
x
c
e
ll
e
nt
pe
r
f
or
m
a
nc
e
in
c
la
s
s
if
yi
ng
lu
ng
di
s
e
a
s
e
b
a
s
e
d
on
c
he
s
t
X
-
r
a
y
(
C
X
R
)
im
a
ge
s
.
H
ow
e
ve
r
,
one
id
e
nt
if
ie
d
li
m
it
a
ti
on
is
th
e
la
c
k
of
di
s
c
us
s
io
n
r
e
g
a
r
di
ng
m
ode
l
te
s
ti
ng
on
di
f
f
e
r
e
nt
da
ta
s
e
ts
or
c
r
os
s
-
da
ta
s
e
t
te
s
ti
ng.
A
lt
hough
th
e
r
e
s
ul
ts
obt
a
in
e
d
a
r
e
m
or
e
th
a
n
98%
,
th
e
r
e
is
no
in
f
or
m
a
ti
on
r
e
ga
r
di
ng mode
l
te
s
ti
ng on e
xt
e
r
na
l
da
ta
s
e
ts
or
c
r
os
s
-
da
ta
s
e
ts
.
M
e
a
nw
hi
le
, i
n ot
he
r
r
e
s
e
a
r
c
h m
a
c
hi
ne
l
e
a
r
ni
ng
m
ode
l,
pa
r
ti
c
ul
a
r
ly
S
V
M
a
nd de
c
is
io
n t
r
e
e
c
la
s
s
if
ie
r
s
ha
ve
be
e
n us
e
d t
o c
la
s
s
if
y a
bnor
m
a
l
br
e
a
th
s
ounds
s
u
c
h
a
s
w
he
e
z
e
s
a
nd
c
r
a
c
kl
e
s
[
24]
. T
hi
s
m
e
th
od
s
how
s
pot
e
nt
ia
l
in
d
ia
gnos
in
g r
e
s
pi
r
a
to
r
y di
s
e
a
s
e
s
m
or
e
a
c
c
ur
a
te
ly
a
nd
obj
e
c
ti
ve
ly
us
in
g
pa
ti
e
nt
d
e
m
ogr
a
phi
c
da
ta
a
nd
c
he
s
t
r
e
gi
on
to
obt
a
in
m
or
e
c
om
pr
e
he
ns
iv
e
in
f
or
m
a
ti
on.
H
ow
e
ve
r
,
th
e
r
e
s
e
a
r
c
h
ha
s
s
e
ve
r
a
l
id
e
nt
if
ia
bl
e
li
m
it
a
ti
ons
in
c
l
udi
ng
th
e
a
bs
e
nc
e
of
in
-
de
pt
h
e
xpl
a
na
ti
on
in
c
om
pa
r
is
on
w
it
h
pr
e
vi
ous
r
e
s
e
a
r
c
h.
I
nf
or
m
a
ti
on
on
li
m
it
a
ti
ons
or
c
ons
tr
a
in
ts
is
a
ls
o
not
a
va
il
a
bl
e
,
w
hi
c
h
c
a
n
he
lp
r
e
a
de
r
s
be
tt
e
r
unde
r
s
ta
nd t
he
c
ont
e
xt
a
nd i
nt
e
r
pr
e
t
th
e
r
e
s
u
lt
s
.
T
hi
s
r
e
s
e
a
r
c
h i
nve
s
ti
ga
te
s
t
he
e
f
f
e
c
t
of
i
m
pr
ovi
ng
th
e
R
e
s
N
e
t5
0 m
ode
l
C
N
N
a
r
c
hi
te
c
tu
r
e
t
o
i
m
pr
ove
im
a
ge
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
in
lu
ng
c
a
n
c
e
r
,
w
hi
le
pr
e
vi
ous
s
tu
di
e
s
ha
ve
e
xpl
or
e
d
th
e
im
pa
c
t
of
us
in
g
th
e
R
e
s
N
e
t
m
ode
l
C
N
N
a
r
c
hi
te
c
tu
r
e
,
but
th
e
y
ha
ve
not
e
x
pl
ic
it
ly
d
is
c
us
s
e
d
th
e
e
f
f
e
c
t
on
im
pr
ovi
ng
c
l
a
s
s
if
ic
a
ti
on
a
nd
th
e
us
e
of
th
e
R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
m
ode
l
is
not
e
xpl
a
in
e
d.
S
o
th
a
t
,
th
is
r
e
s
e
a
r
c
h
pr
e
s
e
nt
s
a
br
e
a
kt
hr
ough,
id
e
nt
if
yi
ng
th
e
s
tr
e
ngt
hs
a
nd
w
e
a
kne
s
s
e
s
of
e
a
c
h
m
e
th
od.
S
tr
e
ngt
hs
in
c
lu
de
a
c
c
ur
a
c
y
a
nd
e
f
f
ic
ie
nc
y
im
pr
ove
m
e
nt
s
,
w
hi
le
w
e
a
kne
s
s
e
s
c
on
s
is
t
of
c
om
put
a
ti
ona
l
c
om
pl
e
xi
ty
a
nd
la
r
ge
da
ta
r
e
qui
r
e
m
e
nt
s
f
or
m
ode
l
tr
a
in
in
g.
T
o
a
ddr
e
s
s
th
e
s
e
w
e
a
kne
s
s
e
s
,
e
f
f
e
c
ti
ve
s
ol
ut
io
ns
a
r
e
r
e
qui
r
e
d,
s
how
in
g
th
e
pot
e
nt
ia
l
a
ppl
ic
a
ti
on
of
th
e
pr
opos
e
d
m
e
th
od
m
or
e
e
f
f
i
c
ie
nt
ly
[
25
]
,
[
26]
.
I
s
s
ue
s
in
th
e
s
e
le
c
te
d
m
e
th
od
in
th
is
r
e
s
e
a
r
c
h
in
c
lu
de
th
e
pot
e
nt
ia
l
li
m
it
a
ti
on
of
tr
a
in
in
g
da
ta
r
e
f
le
c
ti
ng
th
e
f
ul
l
va
r
ia
ti
on
of
lu
ng
c
a
nc
e
r
.
T
h
e
pr
opos
e
d
m
e
th
od
i
s
de
vi
s
e
d
a
s
a
n
im
pr
ove
m
e
nt
pr
opos
a
l
th
a
t
in
c
lu
de
s
s
tr
a
te
gi
e
s
to
e
nh
a
nc
e
tr
a
in
in
g
da
ta
di
v
e
r
s
it
y a
nd
m
in
im
iz
e
pot
e
nt
ia
l
bi
a
s
e
s
a
r
is
in
g
f
r
om
th
e
s
e
li
m
it
a
ti
ons
.
T
he
r
e
s
ul
ts
a
r
e
e
xpe
c
te
d
to
pr
ovi
de
hi
ghe
r
a
c
c
ur
a
c
y
in
c
la
s
s
if
ic
a
ti
on,
ope
ni
ng
oppor
tu
ni
ti
e
s
f
or
s
ig
ni
f
ic
a
nt
im
pr
ove
m
e
nt
s
in
a
c
c
ur
a
te
l
y
id
e
nt
if
yi
ng
lu
ng
c
a
nc
e
r
ty
pe
s
[
21]
,
[
26]
–
[
28]
.
T
he
r
e
f
or
e
,
th
is
r
e
s
e
a
r
c
h
a
im
e
d
to
de
ve
lo
p
C
N
N
a
r
c
hi
te
c
tu
r
e
c
a
pa
bl
e
of
ove
r
c
om
in
g
di
a
gnos
ti
c
c
ons
tr
a
in
ts
of
lu
ng
c
a
nc
e
r
,
im
pr
ovi
ng
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y,
a
nd
pos
it
iv
e
ly
c
ont
r
ib
ut
in
g
to
th
e
a
dva
nc
e
m
e
nt
of
m
e
di
c
a
l
di
a
gnos
ti
c
te
c
hnol
ogy.
T
he
de
ve
lo
pm
e
nt
i
s
c
a
r
r
ie
d
out
to
a
c
h
ie
ve
s
ig
ni
f
ic
a
nt
im
pr
ove
m
e
nt
s
in
lu
ng
c
a
nc
e
r
de
te
c
ti
on,
e
xpe
di
te
th
e
di
a
gnos
is
pr
oc
e
s
s
,
a
nd
e
nha
nc
e
th
e
pr
og
nos
is
a
s
w
e
ll
a
s
tr
e
a
tm
e
nt
of
pa
ti
e
nt
s
a
f
f
e
c
te
d
by l
ung c
a
nc
e
r
.
2.
M
E
T
H
O
D
T
hi
s
r
e
s
e
a
r
c
h
e
va
lu
a
te
s
C
N
N
opt
im
iz
a
ti
on
te
c
hni
que
s
f
or
c
la
s
s
if
yi
ng
bone
f
r
a
c
tu
r
e
s
us
in
g
X
-
r
a
y
C
T
s
c
a
n
d
a
ta
.
T
he
s
tu
dy
pr
oc
e
s
s
e
s
f
r
a
c
tu
r
e
da
ta
f
or
C
N
N
m
od
e
ls
to
d
e
te
c
t
a
nd
c
la
s
s
if
y
f
r
a
c
tu
r
e
s
,
s
how
in
g
s
ig
ni
f
ic
a
nt
pe
r
f
or
m
a
nc
e
im
pr
ove
m
e
nt
s
w
it
h
a
ppr
opr
ia
te
op
ti
m
i
z
a
ti
ons
.
C
om
pa
r
a
ti
ve
a
n
a
ly
s
is
r
e
ve
a
ls
in
s
ig
ht
s
in
to
e
a
c
h
te
c
hni
que
'
s
s
tr
e
ngt
hs
a
nd
w
e
a
kne
s
s
e
s
.
T
h
e
f
in
di
ngs
a
d
va
nc
e
im
a
ge
-
ba
s
e
d
m
e
di
c
a
l
di
a
gnos
i
s
a
nd
of
f
e
r
pr
a
c
ti
c
a
l
gui
da
nc
e
f
or
opt
im
iz
in
g C
N
N
m
ode
ls
i
n he
a
lt
hc
a
r
e
.
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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
:
516
-
526
518
2.1.
D
at
a
d
e
s
c
r
ip
t
io
n
T
he
da
ta
s
e
t
u
s
e
d
in
th
i
s
r
e
s
e
a
r
c
h
c
ons
i
s
te
d
of
C
T
s
c
a
n
s
f
r
om
pa
ti
e
nt
s
w
it
h
lu
ng
di
s
e
a
s
e
s
,
v
a
r
io
us
s
ta
g
e
s
of
c
a
nc
e
r
, a
nd he
a
lt
hy s
ubj
e
c
t
s
. I
Q
-
O
T
H
/NC
C
D
s
li
de
s
, a
nnot
a
te
d by e
xpe
r
ts
, w
e
r
e
s
our
c
e
d f
r
om
K
a
ggl
e
[
29]
.
T
hi
s
da
ta
s
e
t,
c
ol
le
c
te
d
in
a
ut
um
n
2019,
in
c
lu
de
s
150
lu
ng
im
a
ge
s
f
r
om
th
e
O
nc
ol
ogy
T
e
a
c
hi
ng
H
os
pi
ta
l
of
I
r
a
q/
N
a
ti
ona
l
C
e
nt
e
r
f
or
C
a
nc
e
r
D
is
e
a
s
e
s
.
T
he
da
ta
w
a
s
s
pl
it
80:
20
f
or
pr
oc
e
s
s
in
g.
T
he
w
or
kf
lo
w
of
th
is
r
e
s
e
a
r
c
h i
s
s
how
n i
n F
ig
ur
e
1.
P
r
e
p
r
o
c
e
s
s
i
n
g
I
m
a
ge
E
nha
nc
e
m
e
nt
G
r
a
ys
c
a
l
e
C
ha
ng
e
I
m
a
g
e
S
i
z
e
D
a
t
a
A
ugm
e
nt
a
t
i
on
F
l
i
ppi
ng
R
o
t
a
t
e
C
h
e
s
t
X
r
a
y
d
a
t
a
s
e
t
C
l
e
a
n
D
a
t
a
(
R
a
t
i
o
D
a
t
a
s
e
t
80
:
20
)
C
l
a
s
s
i
f
i
c
a
t
i
o
n
(
A
)
R
e
s
N
e
t
50
T
r
a
i
ni
ng D
a
t
a
s
e
t
(
B
)
R
e
s
N
e
t
50
+
C
l
a
s
s
W
e
i
ght
e
d A
ppr
oa
c
h
(
C
)
R
e
s
N
e
t
50
+
M
od
e
l
S
M
O
T
E
D
a
t
a
(
D
)
R
e
s
N
e
t
50
+
C
l
a
s
s
W
e
i
ght
e
d
A
pp
r
oa
c
h
+
M
od
e
l
S
M
O
T
E
D
a
t
a
(
P
r
op
os
e
d
M
e
t
ho
d
)
P
r
oc
e
s
s
e
d T
r
a
i
ni
ng
(
A
)
A
c
c
ur
a
c
y
,
P
r
e
c
i
s
i
on
,
R
e
c
a
l
l
a
nd
F
1
-
S
c
or
e
(
B
)
A
c
c
u
r
a
c
y
,
P
r
e
c
i
s
i
on
,
R
e
c
a
l
l
a
nd
F
1
-
S
c
or
e
(
C
)
A
c
c
ur
a
c
y
,
P
r
e
c
i
s
i
on
,
R
e
c
a
l
l
a
nd
F
1
-
S
c
or
e
(
D
)
A
c
c
ur
a
c
y
,
P
r
e
c
i
s
i
o
n
,
R
e
c
a
l
l
a
nd
F
1
-
S
c
or
e
B
e
s
t
C
l
a
s
s
i
f
i
c
a
t
i
o
n
A
l
g
o
r
i
t
h
m
T
e
s
t
i
ng
D
a
t
a
s
e
t
P
r
e
p
a
r
i
n
g
D
a
t
a
(
A
)
A
c
c
u
r
a
c
y
,
P
r
e
c
i
s
i
on
,
R
e
c
a
l
l
a
nd
F
1
-
S
c
or
e
(
B
)
A
c
c
ur
a
c
y
,
P
r
e
c
i
s
i
on
,
R
e
c
a
l
l
a
nd
F
1
-
S
c
or
e
(
C
)
A
c
c
u
r
a
c
y
,
P
r
e
c
i
s
i
on
,
R
e
c
a
l
l
a
nd
F
1
-
S
c
or
e
(
D
)
A
c
c
ur
a
c
y
,
P
r
e
c
i
s
i
on
,
R
e
c
a
l
l
a
nd
F
1
-
S
c
or
e
P
r
oc
e
s
s
e
d T
e
s
t
i
ng
E
v
a
l
u
a
t
i
o
n
B
e
s
t
C
l
a
s
s
i
f
i
c
a
t
i
o
n
A
l
g
o
r
i
t
h
m
F
ig
ur
e
1.
R
e
s
e
a
r
c
h
w
or
kf
lo
w
i
n l
ung dis
e
a
s
e
i
m
a
ge
c
la
s
s
if
ic
a
ti
o
n pr
oc
e
s
s
F
ig
ur
e
1
il
lu
s
tr
a
te
s
th
e
da
t
a
a
n
a
ly
s
is
pr
oc
e
s
s
f
or
c
h
e
s
t
x
-
r
a
y
im
a
ge
s
,
in
c
lu
di
ng
pr
e
pr
oc
e
s
s
in
g,
m
ode
l
te
s
ti
ng,
a
nd
pe
r
f
or
m
a
nc
e
e
va
lu
a
ti
on
of
va
r
io
us
c
la
s
s
if
ic
a
ti
on
m
e
th
ods
.
M
e
tr
ic
s
s
u
c
h
a
s
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F
1
s
c
or
e
a
s
s
e
s
s
m
ode
l
pe
r
f
or
m
a
nc
e
.
M
e
th
od
s
us
e
d
in
c
lu
de
R
e
s
N
e
t5
0,
R
e
s
N
e
t5
0
w
it
h
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h,
R
e
s
N
e
t5
0
w
it
h
s
ynt
he
ti
c
m
in
or
it
y
ove
r
s
a
m
pl
in
g
te
c
hni
que
(
S
M
O
T
E
)
,
a
nd
a
c
om
bi
na
ti
on
o
f
R
e
s
N
e
t5
0
w
it
h
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
a
nd
S
M
O
T
E
(
pr
opos
e
d
m
e
th
od
)
.
T
he
da
ta
s
e
t
w
a
s
c
le
a
ne
d
a
nd
s
pl
i
t
80:
20 f
or
t
r
a
in
in
g. T
he
e
va
lu
a
ti
on i
de
nt
if
ie
s
t
he
be
s
t
a
lg
or
it
hm
f
or
opt
im
a
l
di
a
gnos
ti
c
r
e
s
ul
ts
.
2.2.
E
xp
la
n
at
io
n
of
t
h
e
R
e
s
n
e
t
50, Re
s
N
e
t
50+
m
od
e
l
b
u
il
d
in
g
w
it
h
S
M
O
T
E
d
at
a
an
d
R
e
s
N
e
t
50+
m
od
e
l
b
u
il
d
in
g w
i
t
h
c
la
s
s
w
e
ig
h
t
e
d
ap
p
r
oac
h
R
e
s
N
e
t
is
one
of
th
e
a
r
c
hi
te
c
tu
r
a
l
m
ode
ls
ba
s
e
d
on
th
e
c
on
c
e
pt
of
r
e
s
id
ua
l
bl
oc
ks
,
w
hi
c
h
a
ll
ow
s
th
e
ne
twor
k
to
ove
r
c
om
e
th
e
c
ha
ll
e
nge
s
of
tr
a
in
in
g
ve
r
y
de
e
p
m
o
de
ls
.
R
e
s
id
ua
l
bl
oc
ks
s
e
r
ve
a
s
b
a
s
ic
uni
ts
th
a
t
a
ll
ow
di
r
e
c
t
in
f
or
m
a
ti
on
f
lo
w
f
r
om
in
put
to
out
put
w
it
hout
m
a
ny
c
om
pl
e
x
tr
a
ns
f
or
m
a
ti
ons
,
f
a
c
il
it
a
ti
ng
th
e
pr
e
s
e
r
va
ti
on
of
id
e
nt
it
y
in
f
or
m
a
ti
on
dur
in
g
th
e
tr
a
in
in
g
pr
oc
e
s
s
[
23]
.
T
hi
s
m
ode
l
us
e
s
s
hor
tc
ut
c
onn
e
c
ti
ons
or
"
r
e
s
id
ua
ls
"
th
a
t
a
ll
ow
tr
a
in
in
g
of
ve
r
y
de
e
p
ne
twor
ks
w
it
hout
de
gr
a
da
ti
on
pr
obl
e
m
s
,
w
hi
c
h
us
u
a
ll
y
oc
c
ur
in
de
e
p ne
twor
ks
.
T
h
e
f
ol
lo
w
in
g
w
e
w
il
l
e
xpl
a
in
t
he
3 opti
m
iz
e
d
R
e
s
ne
t
m
ode
l
s
.
F
ig
ur
e
2 s
how
s
t
he
a
r
c
hi
te
c
tu
r
e
o
f
F
ig
u
r
e
2
(
a
)
R
e
s
N
e
t5
0
m
ode
l
,
F
ig
ur
e
2
(
b)
R
e
s
N
e
t5
0 w
it
h S
M
O
T
E
da
ta
,
a
nd
F
ig
ur
e
2
(
c
)
R
e
s
N
e
t5
0
w
it
h
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
,
a
ll
of
w
hi
c
h
us
e
di
f
f
e
r
e
nt
bl
oc
ks
w
it
h
th
e
a
s
s
um
pt
io
n
th
a
t
th
e
y
a
ll
ow
th
e
ne
twor
k
to
le
a
r
n
a
be
tt
e
r
da
t
a
r
e
pr
e
s
e
nt
a
ti
on.
T
h
e
s
e
la
ye
r
s
w
or
k
to
g
e
th
e
r
to
c
onve
r
t
im
a
ge
in
put
in
to
out
put
th
a
t
c
a
n
be
us
e
d
f
or
c
la
s
s
if
ic
a
ti
o
n
or
ot
he
r
ta
s
ks
.
R
e
s
N
e
t5
0
is
a
pow
e
r
f
ul
ne
ur
a
l
ne
twor
k
m
ode
l
a
nd
is
f
r
e
que
nt
ly
us
e
d
in
im
a
ge
r
e
c
ogni
ti
on
a
n
d
c
la
s
s
if
ic
a
ti
on
ta
s
k
s
due
to
it
s
a
bi
li
ty
to
le
a
r
n
be
tt
e
r
r
e
pr
e
s
e
nt
a
ti
ons
of
da
ta
.
A
dva
nt
a
ge
s
of
R
e
s
N
e
t5
0:
ne
two
r
k
de
pt
h:
W
it
h
50
la
ye
r
s
,
th
i
s
m
ode
l
is
a
bl
e
to
c
a
pt
ur
e
ve
r
y
c
om
pl
e
x
a
nd
d
e
ta
il
e
d
f
e
a
tu
r
e
s
.
R
e
s
id
ua
l
c
onne
c
ti
ons
:
R
e
duc
e
s
th
e
va
ni
s
hi
ng
gr
a
di
e
nt
pr
obl
e
m
,
m
a
ki
ng t
r
a
in
in
g m
or
e
e
f
f
ic
ie
nt
. P
r
e
-
tr
a
in
e
d
m
ode
ls
:
P
r
e
-
t
r
a
in
e
d m
ode
ls
a
r
e
a
va
il
a
bl
e
i
n I
m
a
ge
N
e
t,
s
o t
he
y c
a
n
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
A
r
e
v
ol
ut
io
nar
y
c
onv
ol
ut
io
nal
ne
u
r
al
ne
tw
or
k
ar
c
hi
te
c
tu
r
e
f
or
m
or
e
ac
c
u
r
at
e
…
(
M
ul
ia
di
)
519
be
a
da
pt
e
d
to
va
r
io
us
ta
s
ks
qui
c
kl
y.
D
is
a
dva
nt
a
ge
s
of
R
e
s
N
e
t
50:
O
ve
r
f
it
ti
ng:
C
a
n
oc
c
ur
if
th
e
da
ta
s
e
t
is
not
la
r
ge
e
nough or
va
r
ie
d. T
r
a
in
in
g
time
:
R
e
qui
r
e
s
qui
te
a
l
ong tr
a
in
in
g t
im
e
a
nd high c
om
put
in
g r
e
s
our
c
e
s
.
Z
e
r
o
p
a
d
d
i
n
g
C
o
n
v
(
64
,
(
3
,
3
))
B
a
t
c
h
S
i
z
e
,
8
R
e
l
u
M
a
x
P
o
o
l
(
p
o
o
l
_
s
i
z
e
=(
2
,
2
)
)
C
o
n
v
B
l
o
c
k
,
1
2
8
I
d
B
l
o
c
k
x
3
C
o
n
v
B
l
o
c
k
,
2
5
6
I
d
B
l
o
c
k
x
2
A
v
e
r
a
g
e
P
o
o
l
F
l
a
t
t
e
n
D
e
n
se
,
16
F
u
l
l
y
C
o
n
n
e
c
t
e
d
I
npu
t
I
m
a
ge
L
ung
512
x
512
3
va
r
i
a
bl
e
s
O
ut
put
C
l
a
s
s
s
i
f
i
c
a
t
i
on
Ze
r
o
p
a
d
d
i
n
g
C
o
n
v
(
64
,
(
3
,
3
))
B
a
t
c
h
S
i
z
e
R
e
l
u
M
a
x
P
o
o
l
(
p
o
o
l
_
s
i
z
e
=(
2
,
2
)
)
C
o
n
v
B
l
o
c
k
I
d
B
l
o
c
k
x
3
A
v
e
r
a
g
e
P
o
o
l
F
l
a
t
t
e
n
D
e
n
s
e
,
16
F
u
l
l
y
C
o
n
n
e
c
t
e
d
Z
e
r
o P
a
ddi
ng
C
onv
ol
ut
i
ona
l
L
a
ye
r
B
a
t
c
h N
or
m
a
l
i
z
a
t
i
on
A
c
t
i
va
t
i
on
M
a
x P
ool
i
ng
C
onv
ol
ut
i
ona
l
B
l
oc
k
I
de
nt
i
t
y B
l
oc
k
A
ve
r
a
ge
P
ool
F
l
a
t
t
e
n
D
e
ns
e
F
ul
l
y C
onn
e
c
t
e
d
F
e
a
t
u
r
e
s
(
a
)
Z
e
ro p
a
ddi
ng
Conv
(
64
,
(
3
,
3
))
Ba
t
c
h S
i
z
e
,
8
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
Co
nv
Bl
oc
k
125
Id B
l
oc
k x
3
Con
v Bl
oc
k
254
Id Bl
oc
k x
2
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
D
e
ns
e
,
32
F
ul
l
y C
on
ne
c
t
e
d
I
n
p
u
t
I
m
a
g
e
L
u
n
g
5
1
2
x
5
1
2
3
v
a
r
i
a
b
l
e
s
O
u
t
p
u
t
C
l
a
s
s
s
i
f
i
c
a
t
i
o
n
Z
e
ro
pa
ddi
ng
Conv
(
64
,
(
3
,
3
))
Ba
t
c
h S
i
z
e
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
Co
nv
Bl
oc
k
Id B
l
oc
k
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
D
e
ns
e
,
32
F
ul
l
y
Con
ne
c
t
e
d
Z
e
r
o
P
a
d
d
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
La
y
e
r
B
a
t
c
h
N
o
r
m
a
l
i
z
a
t
i
o
n
A
c
t
i
v
a
t
i
o
n
M
a
x
P
o
o
l
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
B
l
o
c
k
I
d
e
n
t
i
t
y
B
l
o
c
k
A
v
e
r
a
g
e
P
o
o
l
F
l
a
t
t
e
n
D
e
n
s
e
F
u
l
l
y
C
o
n
n
e
c
t
e
d
F
e
a
t
ur
e
s
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
3
,
3
))
(
b)
Z
e
ro p
a
ddi
ng
Conv
(
64
,
(
3
,
3
))
256
Ba
t
c
h S
i
z
e
,
8
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
125
Co
nv
Bl
oc
k
127
Id B
l
oc
k x
3
Con
v Bl
oc
k
254
Id Bl
oc
k x
2
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
62
D
e
ns
e
,
32
F
ul
l
y C
on
ne
c
t
e
d
I
n
p
u
t
I
m
a
g
e
L
u
n
g
5
1
2
x
5
1
2
3
v
a
r
i
a
b
l
e
s
O
u
t
p
u
t
C
l
a
s
s
s
i
f
i
c
a
t
i
o
n
Z
e
ro
pa
ddi
ng
Conv
(
64
,
(
3
,
3
))
Ba
t
c
h S
i
z
e
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
Co
nv
Bl
oc
k
Id B
l
oc
k
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
D
e
ns
e
,
32
F
ul
l
y
Con
ne
c
t
e
d
Z
e
r
o
P
a
d
d
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
La
y
e
r
B
a
t
c
h
N
o
r
m
a
l
i
z
a
t
i
o
n
A
c
t
i
v
a
t
i
o
n
M
a
x
P
o
o
l
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
B
l
o
c
k
I
d
e
n
t
i
t
y
B
l
o
c
k
A
v
e
r
a
g
e
P
o
o
l
F
l
a
t
t
e
n
D
e
n
s
e
F
u
l
l
y
C
o
n
n
e
c
t
e
d
F
e
a
t
ur
e
s
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
3
,
3
))
62
(
c
)
F
ig
ur
e
2
.
T
he
a
r
c
hi
te
c
tu
r
e
of
(
a
)
R
e
s
N
e
t5
0, (
b)
R
e
s
N
e
t5
0 w
it
h
S
M
O
T
E
da
ta
, a
nd (
c
)
R
e
s
N
e
t5
0 w
it
h c
la
s
s
w
e
ig
ht
e
d a
ppr
oa
c
h
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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
:
516
-
526
520
R
e
s
N
e
t5
0
+
S
M
O
T
E
:
T
hi
s
m
ode
l
is
th
e
r
e
s
ul
t
of
a
ppl
yi
ng
th
e
S
M
O
T
E
te
c
hni
que
to
th
e
da
ta
s
e
t
be
f
or
e
tr
a
in
in
g
th
e
R
e
s
N
e
t5
0
m
ode
l.
S
M
O
T
E
is
a
da
ta
ba
la
nc
in
g
t
e
c
hni
que
c
r
e
a
te
d
to
a
ddr
e
s
s
th
e
pr
obl
e
m
o
f
im
ba
la
nc
e
d
da
ta
s
e
ts
,
w
he
r
e
m
in
or
it
y
c
la
s
s
e
s
a
r
e
im
pr
ove
d
by
c
r
e
a
ti
ng
ne
w
s
ynt
he
ti
c
s
a
m
pl
e
s
ba
s
e
d
on
e
xi
s
ti
ng
m
in
or
it
y
s
a
m
pl
e
s
.
A
dva
nt
a
g
e
s
of
R
e
s
N
e
t5
0
+
S
M
O
T
E
:
D
a
ta
b
a
la
nc
in
g:
I
m
pr
ove
s
r
e
pr
e
s
e
nt
a
ti
on
of
m
in
or
it
y
c
la
s
s
e
s
,
h
e
lp
s
m
ode
ls
to
le
a
r
n
be
tt
e
r
a
nd
r
e
duc
e
s
bi
a
s
.
A
c
c
ur
a
c
y
im
pr
ove
m
e
nt
:
T
ypi
c
a
ll
y
im
pr
ove
s
m
ode
l
pe
r
f
or
m
a
nc
e
on
m
in
o
r
it
y
c
la
s
s
e
s
,
r
e
duc
in
g
m
is
c
l
a
s
s
if
yi
ng
pr
obl
e
m
s
.
C
ons
of
R
e
s
N
e
t5
0
+
S
M
O
T
E
:
P
r
oc
e
s
s
in
g
ti
m
e
:
I
nc
r
e
a
s
in
g
s
ynt
he
ti
c
da
ta
r
e
qui
r
e
s
a
ddi
ti
ona
l
pr
oc
e
s
s
in
g
t
im
e
.
O
ve
r
f
it
ti
ng:
E
ve
n
if
th
e
da
ta
is
ba
la
nc
e
d,
ove
r
f
it
ti
ng c
a
n s
ti
ll
oc
c
ur
i
f
t
he
s
ynt
he
ti
c
da
ta
i
s
not
di
ve
r
s
e
e
no
ugh.
R
e
s
N
e
t5
0
+
c
l
a
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
th
is
a
ppr
oa
c
h
in
vol
ve
s
gi
v
in
g
gr
e
a
te
r
w
e
ig
ht
to
m
in
o
r
it
y
c
la
s
s
e
s
dur
in
g
th
e
R
e
s
N
e
t5
0
m
ode
l
tr
a
in
in
g
pr
oc
e
s
s
.
I
ns
te
a
d
of
in
c
r
e
a
s
in
g
th
e
s
a
m
pl
e
of
m
in
or
i
ty
c
la
s
s
e
s
,
th
e
m
ode
l
is
gi
ve
n
a
la
r
ge
r
pe
na
lt
y
f
o
r
e
r
r
or
s
in
th
e
s
e
c
la
s
s
e
s
.
A
dva
n
ta
ge
s
of
R
e
s
N
e
t5
0
+
c
la
s
s
w
e
ig
ht
e
d:
ba
la
nc
in
g
w
it
hout
a
ddi
ng
da
ta
:
N
o
ne
e
d
to
in
c
r
e
a
s
e
th
e
a
m
ount
of
da
ta
,
th
e
r
e
by
r
e
duc
in
g
th
e
r
is
k
of
ove
r
f
it
ti
ng.
E
f
f
ic
ie
nc
y:
U
s
ua
ll
y
f
a
s
te
r
th
a
n
ove
r
s
a
m
pl
in
g
m
e
th
ods
be
c
a
u
s
e
th
e
r
e
is
no
ne
e
d
to
c
r
e
a
te
a
ddi
ti
o
na
l
s
a
m
pl
e
s
.
D
is
a
dva
nt
a
ge
s
of
R
e
s
N
e
t5
0
+
c
la
s
s
w
e
ig
ht
e
d:
C
om
pl
e
x
tu
ni
ng:
C
hoos
in
g
th
e
r
ig
ht
w
e
ig
ht
s
f
or
e
a
c
h
c
la
s
s
c
a
n
b
e
a
c
om
pl
e
x
ta
s
k
a
nd
r
e
qui
r
e
s
a
lo
t
of
e
xpe
r
im
e
nt
a
ti
on.
S
ta
bi
li
ty
:
I
f
th
e
w
e
ig
ht
s
gi
ve
n
a
r
e
to
o
e
xt
r
e
m
e
,
it
c
a
n
c
a
us
e
in
s
ta
bi
li
ty
in
m
ode
l
tr
a
in
in
g
.
2.3.
R
e
s
N
e
t
50+
m
od
e
l
b
u
il
d
in
g w
it
h
c
la
s
s
w
e
ig
h
t
e
d
ap
p
r
oac
h
+
m
od
e
l
b
u
il
d
in
g
w
it
h
S
M
O
T
E
d
at
a
I
nnova
ti
ve
m
e
th
ods
a
r
e
ne
e
d
e
d
to
im
pr
ove
th
e
pe
r
f
or
m
a
nc
e
of
lu
ng
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on
m
ode
l
s
by
c
om
bi
ni
ng
S
M
O
T
E
da
ta
,
da
ta
a
ugm
e
nt
a
ti
on,
a
nd
m
ode
l
bui
ld
in
g
w
it
h
a
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
.
T
hi
s
m
e
th
od
a
im
s
to
ove
r
c
om
e
c
ha
ll
e
ng
e
s
in
de
v
e
lo
pi
ng
lu
ng
c
a
nc
e
r
di
a
gno
s
ti
c
m
ode
ls
.
B
y
c
om
bi
ni
ng
th
e
s
e
th
r
e
e
m
e
th
od
s
,
th
e
m
ode
l
is
e
xpe
c
te
d
to
b
e
a
bl
e
to
r
e
c
ogni
z
e
c
om
pl
e
x
pa
tt
e
r
ns
in
lu
ng
c
a
nc
e
r
im
a
ge
s
,
in
c
lu
di
ng
th
os
e
in
th
e
m
in
or
it
y c
la
s
s
.
F
ig
ur
e
3
s
how
s
th
e
a
r
c
hi
te
c
tu
r
e
of
th
e
pr
opos
e
d
m
e
th
od,
c
om
b
in
in
g
R
e
s
N
e
t5
0,
m
ode
l
bui
ld
in
g
w
it
h
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h,
a
nd
m
ode
l
bui
ld
in
g
w
it
h
S
M
O
T
E
da
ta
,
a
s
w
e
ll
a
s
a
ddi
ng
s
of
tm
a
x
be
f
or
e
th
e
f
ul
ly
c
onne
c
te
d
la
ye
r
.
T
he
r
e
d
li
ne
s
s
how
th
e
p
a
r
ts
of
th
e
pr
opo
s
e
d
m
e
th
od.
T
he
u
s
e
of
S
M
O
T
E
da
ta
,
da
t
a
a
ugm
e
nt
a
ti
on,
a
nd
m
ode
l
bui
ld
in
g
w
it
h
a
c
la
s
s
w
e
ig
ht
e
d
a
ppr
o
a
c
h
in
th
is
r
e
s
e
a
r
c
h
w
a
s
pr
ove
n
to
in
c
r
e
a
s
e
th
e
a
c
c
ur
a
c
y
a
nd
s
e
ns
it
iv
it
y
of
th
e
lu
ng
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on
m
ode
l.
T
he
pr
opos
e
d
m
e
th
od
is
e
f
f
e
c
ti
ve
f
o
r
r
e
c
ogni
z
in
g
c
om
pl
e
x
p
a
tt
e
r
ns
in
lu
ng
c
a
n
c
e
r
im
a
ge
s
,
th
e
r
e
b
y
im
pr
ovi
ng
th
e
pe
r
f
or
m
a
nc
e
of
lu
ng
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on
m
ode
ls
.
Z
e
ro p
a
ddi
ng
Conv
(
64
,
(
3
,
3
))
256
Ba
t
c
h S
i
z
e
,
8
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
125
Conv B
l
oc
k
128
Id B
l
oc
k x
3
Conv
Bl
oc
k
256
Id Bl
oc
k x
2
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
62
D
e
ns
e
,
32
F
ul
l
y C
on
ne
c
t
e
d
I
n
p
u
t
I
m
a
g
e
Lu
n
g
5
1
2
x
5
1
2
3
v
a
r
i
a
b
l
e
s
O
u
t
p
u
t
C
l
a
s
s
s
i
f
i
c
a
t
i
o
n
Z
e
ro p
a
ddi
ng
Conv
(
64
,
(
3
,
3
))
Ba
t
c
h S
i
z
e
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
Co
nv
Bl
oc
k
Id B
l
oc
k
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
D
e
ns
e
,
32
F
ul
l
y Con
ne
c
t
e
d
Ze
r
o
P
a
d
d
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
La
y
e
r
B
a
t
c
h
N
o
r
m
a
l
i
z
a
t
i
o
n
A
c
t
i
v
a
t
i
o
n
M
a
x
P
o
o
l
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
B
l
o
c
k
I
d
e
n
t
i
t
y
B
l
o
c
k
A
v
e
r
a
g
e
P
o
o
l
F
l
a
t
t
e
n
D
e
n
s
e
F
u
l
l
y
C
o
n
n
e
c
t
e
d
P
r
opo
s
e
d M
e
t
hod
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
3
,
3
))
62
S
of
t
m
a
x
S
oft
m
a
x
S
o
f
t
m
a
x
F
e
a
t
ur
e
s
F
ig
ur
e
3
.
I
ll
us
tr
a
ti
on of
th
e
pr
opos
e
d m
e
th
od i
n l
ung
C
T
s
c
a
n i
m
a
ge
c
la
s
s
if
ic
a
ti
on pr
oc
e
s
s
2.4.
E
val
u
at
io
n
I
n
th
e
e
va
lu
a
ti
on
s
ta
ge
,
th
e
p
e
r
f
or
m
a
nc
e
r
e
s
ul
ts
of
s
in
gl
e
-
c
la
s
s
if
ic
a
ti
on
m
ode
ls
a
r
e
c
om
pa
r
e
d
us
in
g
R
e
s
N
e
t5
0,
R
e
s
N
e
t5
0
bui
lt
w
it
h
S
M
O
T
E
da
ta
,
a
nd
R
e
s
N
e
t5
0
bui
lt
w
it
h
a
c
la
s
s
-
w
e
ig
ht
e
d
a
ppr
oa
c
h.
S
ubs
e
que
nt
ly
, t
he
pr
opos
e
d m
e
th
od,
w
hi
c
h i
s
a
r
e
vol
ut
io
na
r
y C
N
N
a
r
c
hi
te
c
tu
r
e
f
or
m
or
e
a
c
c
ur
a
te
l
ung c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on,
is
c
r
e
a
te
d
a
nd
e
va
lu
a
te
d
u
s
in
g
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
s
s
uc
h
a
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
F
1
-
s
c
or
e
.
T
he
r
e
s
ul
ts
of
th
is
m
e
th
od
a
r
e
f
ur
th
e
r
c
om
pa
r
e
d
w
it
h
ot
he
r
r
e
s
e
a
r
c
h
in
th
e
f
ie
ld
,
de
m
ons
tr
a
ti
ng
it
s
e
f
f
e
c
ti
ve
ne
s
s
i
n
e
nha
nc
in
g t
he
pr
e
c
i
s
io
n of
l
ung c
a
nc
e
r
de
te
c
ti
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
A
r
e
v
ol
ut
io
nar
y
c
onv
ol
ut
io
nal
ne
u
r
al
ne
tw
or
k
ar
c
hi
te
c
tu
r
e
f
or
m
or
e
ac
c
u
r
at
e
…
(
M
ul
ia
di
)
521
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
he
m
ode
l
tr
a
in
in
g
r
e
s
ul
ts
pr
e
s
e
nt
e
d
in
th
is
a
r
ti
c
le
in
c
lu
de
c
r
uc
i
a
l
in
s
ig
ht
s
in
to
th
e
pe
r
f
or
m
a
nc
e
of
th
e
opt
im
iz
e
d
C
N
N
m
ode
l
f
or
lu
ng
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on,
s
pe
c
if
ic
a
ll
y
us
in
g
th
e
C
N
N
a
r
c
hi
te
c
tu
r
e
f
or
hi
gh
-
a
c
c
ur
a
c
y
im
a
ge
c
la
s
s
if
ic
a
ti
on
w
it
h
th
e
R
e
s
N
e
t5
0
P
lu
s
m
ode
l
a
s
th
e
pr
opos
e
d
m
ode
l,
c
om
pa
r
e
d
to
th
e
R
e
s
N
e
t5
0
s
ta
nd
a
r
d.
I
n
th
is
r
e
vol
ut
io
na
r
y
C
N
N
a
r
c
hi
te
c
tu
r
e
f
or
m
or
e
a
c
c
ur
a
t
e
lu
ng
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on
.
F
ig
ur
e
3
pr
ovi
de
s
a
de
ta
il
e
d
e
xpl
a
na
ti
on
of
th
e
r
e
s
ul
ts
obt
a
in
e
d
f
or
e
a
c
h
e
poc
h,
hi
ghl
ig
h
ti
ng t
he
a
dva
n
c
e
m
e
nt
s
a
nd
im
pr
ove
m
e
nt
s
in
m
ode
l
pe
r
f
or
m
a
nc
e
t
hr
oughout t
he
t
r
a
in
in
g pr
oc
e
s
s
.
3.1.
I
m
age
q
u
al
it
y i
m
p
r
ove
m
e
n
t
I
n
th
is
r
e
s
e
a
r
c
h,
th
e
in
it
ia
l
s
ta
ge
i
s
im
pr
ovi
ng
im
a
ge
qua
li
ty
to
e
nha
nc
e
th
e
qua
li
ty
of
lu
ng
C
T
s
c
a
n
im
a
ge
s
.
A
to
ta
l
of
th
r
e
e
lu
ng
C
T
s
c
a
n
im
a
ge
s
is
s
how
n
s
id
e
by
s
id
e
w
it
h
la
be
ls
"
N
or
m
a
l,
"
"
M
a
li
gna
nt
,"
a
nd
"
B
e
ni
gn"
a
s
s
how
n
in
F
ig
ur
e
4
.
T
he
im
a
ge
la
be
le
d
"
N
or
m
a
l
"
s
how
s
a
n
a
xi
a
l
(
c
r
os
s
-
s
e
c
ti
ona
l)
s
li
c
e
of
th
e
c
he
s
t
C
T
s
c
a
n
w
it
hout
s
ig
ni
f
ic
a
nt
a
bnor
m
a
li
ti
e
s
.
M
or
e
ove
r
,
lu
ng
s
tr
uc
tu
r
e
s
a
ppe
a
r
c
l
e
a
n
w
it
h
a
nor
m
a
l
a
ir
pa
tt
e
r
n,
no nodule
s
or
m
a
s
s
e
s
, a
nd
a
n a
b
s
e
nc
e
of
f
lu
id
bui
ld
up.
T
he
i
m
a
ge
l
a
be
le
d "
M
a
li
gna
nt
"
s
how
s
a
n
a
xi
a
l
s
li
c
e
of
th
e
c
he
s
t
C
T
s
c
a
n
w
it
h
ir
r
e
gul
a
r
m
a
s
s
e
s
or
nodule
s
in
lu
ng.
T
he
m
a
s
s
e
s
a
ppe
a
r
d
e
ns
e
r
th
a
n
th
e
s
ur
r
ounding
nor
m
a
l
lu
ng
ti
s
s
ue
,
s
how
in
g
a
pos
s
ib
le
m
a
li
gna
nt
tu
m
or
.
T
he
i
m
a
ge
la
be
le
d
"
B
e
ni
gn"
s
how
s
a
n
a
xi
a
l
s
li
c
e
of
th
e
c
he
s
t
C
T
s
c
a
n
w
it
h
le
s
io
ns
or
nodule
s
th
a
t
a
r
e
m
or
e
de
f
in
e
d
a
nd
a
ppe
a
r
r
e
gul
a
r
c
om
pa
r
e
d
to
th
e
"
M
a
li
gna
nt
"
im
a
ge
. T
he
s
e
l
e
s
io
ns
ha
ve
c
le
a
r
e
r
bounda
r
ie
s
,
s
how
in
g t
he
t
e
nd
e
nc
y f
or
be
ni
gn gr
ow
th
s
.
T
he
s
e
i
m
a
ge
s
c
a
n be
us
e
d
f
or
e
duc
a
ti
ona
l
or
de
m
on
s
tr
a
ti
on
pur
pos
e
s
on
how
m
e
d
i
c
a
l
im
a
ge
s
a
r
e
u
s
e
d
to
di
f
f
e
r
e
nt
ia
te
be
twe
e
n
nor
m
a
l
lu
ng c
ondi
ti
ons
, m
a
li
gna
nc
y (
c
a
nc
e
r
)
, a
nd be
ni
gn gr
ow
th
s
.
F
ig
ur
e
5
s
how
s
th
r
e
e
lu
ng
s
e
gm
e
nt
a
ti
on
im
a
ge
s
s
e
e
m
in
gl
y
ge
ne
r
a
te
d
f
r
om
a
C
T
s
c
a
n.
T
he
F
ig
ur
e
5
(
a
)
is
la
be
le
d
"
B
e
nga
li
c
a
s
e
”
,
w
hi
c
h
s
how
s
lu
ng
s
li
c
e
w
it
h
a
c
ol
or
s
c
a
le
r
a
ngi
ng
f
r
om
bl
ue
to
ye
ll
ow
.
B
r
ig
ht
e
r
a
r
e
a
s
(
ye
ll
ow
)
s
how
hi
ghe
r
t
is
s
ue
d
e
ns
it
y or
a
r
e
a
s
of
f
oc
us
f
or
a
na
ly
s
i
s
, w
it
h t
he
s
e
gm
e
nt
a
ti
on e
m
pha
s
iz
in
g s
pe
c
if
ic
te
xt
ur
e
s
or
pa
tt
e
r
ns
w
it
hi
n
lu
ng.
T
he
F
ig
ur
e
5
(
b)
is
la
be
le
d
"
P
ol
yc
ys
ti
c
c
a
s
e
"
,
w
he
r
e
lu
ng
s
e
gm
e
nt
a
ti
on
is
pe
r
f
or
m
e
d
to
a
ll
ow
f
o
r
a
c
le
a
r
e
r
vi
e
w
of
th
e
in
te
r
na
l
lu
ng
s
tr
uc
tu
r
e
s
.
F
ur
th
e
r
m
or
e
,
th
e
r
e
a
r
e
c
ont
our
li
ne
s
f
ol
lo
w
in
g
th
e
s
ha
pe
of
lu
ng
a
nd
bl
ood
v
e
s
s
e
ls
(
br
onc
hi
)
,
w
it
h
a
di
ve
r
s
e
c
ol
or
r
a
ng
e
s
ho
w
in
g
va
r
io
us
ti
s
s
ue
de
ns
it
ie
s
or
ty
pe
s
,
w
he
r
e
br
ig
ht
gr
e
e
n
a
r
e
a
s
s
how
c
ys
ts
or
ot
he
r
a
bnor
m
a
li
ti
e
s
.
T
he
F
ig
ur
e
5
(
c
)
is
la
be
le
d
"
N
or
m
a
l
c
a
s
e
"
,
s
ho
w
in
g
lu
ng
s
li
c
e
th
a
t
a
ppe
a
r
s
nor
m
a
l
w
it
h
s
e
gm
e
nt
a
ti
on
in
di
c
a
ti
ng
uni
f
or
m
ti
s
s
ue
di
s
tr
ib
ut
io
n.
T
he
c
ol
or
s
c
a
le
i
s
s
im
il
a
r
to
th
e
f
ir
s
t
im
a
g
e
but
w
it
h
f
e
w
e
r
va
r
ia
ti
ons
,
s
ho
w
in
g
a
po
s
s
ib
le
la
c
k
of
s
ig
ni
f
ic
a
nt
a
bnor
m
a
li
ti
e
s
. A
l
l
th
r
e
e
i
m
a
ge
s
ha
ve
c
oor
di
na
te
d
s
c
a
le
s
, w
it
h va
lu
e
s
s
how
in
g pi
xe
l
pos
it
io
n
s
. T
hi
s
is
c
om
m
on f
or
m
e
di
c
a
l
im
a
ge
a
na
ly
s
is
,
w
he
r
e
t
he
l
oc
a
ti
on
a
nd
s
iz
e
of
s
p
e
c
if
ic
a
r
e
a
s
a
r
e
e
s
s
e
nt
ia
l
f
or
di
a
gnos
i
s
a
nd
e
va
lu
a
ti
on.
T
he
s
e
im
a
g
e
s
c
a
n
be
us
e
d
by
h
e
a
lt
hc
a
r
e
pr
of
e
s
s
i
ona
ls
to
a
s
s
e
s
s
lu
ng
c
ondi
ti
ons
,
de
te
c
t
di
s
e
a
s
e
s
or
a
bnor
m
a
li
ti
e
s
,
a
nd
m
oni
to
r
th
e
pr
og
r
e
s
s
io
n
of
m
e
di
c
a
l
c
ondi
ti
ons
.
T
he
s
e
gm
e
nt
a
ti
on
s
how
n
a
id
s
in
id
e
nt
if
yi
ng a
nd vis
ua
li
z
in
g l
ung s
tr
uc
tu
r
e
s
i
n m
or
e
de
ta
il
.
N
or
m
a
l
M
a
li
gna
nt
B
e
ni
gn
F
ig
ur
e
4
.
D
a
ta
a
ugm
e
nt
a
ti
on r
e
s
ul
ts
us
in
g r
ot
a
ti
on t
r
a
ns
f
or
m
a
ti
on
(
a
)
(
b)
(
c
)
F
ig
ur
e
5
.
E
xa
m
pl
e
of
c
le
a
n da
ta
i
m
a
ge
s
(
a
)
be
nga
li
c
a
s
e
, (
b)
pol
yc
ys
ti
c
c
a
s
e
, (
c
)
nor
m
a
l
c
a
s
e
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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
:
516
-
526
522
3.2.
C
T
s
c
an
i
m
ag
e
c
la
s
s
if
ic
at
io
n
I
n
im
a
ge
c
la
s
s
if
ic
a
ti
on
s
ta
g
e
,
th
e
d
a
ta
us
e
d
in
c
lu
de
th
e
s
e
gm
e
nt
a
ti
on
r
e
s
ul
t
of
C
T
s
c
a
n
im
a
ge
,
w
hi
c
h
a
r
e
di
vi
de
d
in
to
tr
a
in
in
g
a
nd
te
s
ti
ng
da
ta
.
T
h
e
tr
a
in
in
g
da
ta
pr
oc
e
s
s
i
s
f
ur
th
e
r
di
vi
de
d
in
to
tr
a
in
in
g
a
nd
va
li
d
a
ti
on
da
ta
. A
t
th
is
s
ta
ge
, 1097 d
a
ta
a
r
e
us
e
d c
ons
i
s
ti
ng of
70%
(
768 tr
a
in
in
g da
ta
)
a
nd 30%
(
329 va
li
da
ti
on da
ta
)
.
3.2.1.
C
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
c
la
s
s
if
ic
at
io
n
ar
c
h
it
e
c
t
u
r
e
A
t
th
is
s
t
a
ge
,
im
a
ge
c
la
s
s
if
ic
a
ti
on
is
pe
r
f
or
m
e
d
s
e
p
a
r
a
te
ly
us
in
g
a
r
c
hi
te
c
tu
r
e
R
e
s
N
e
t5
0,
R
e
s
N
e
t5
0
+
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
,
R
e
s
N
e
t5
0+
m
ode
l
S
M
O
T
E
da
ta
,
a
nd
R
e
s
N
e
t5
0+
c
l
a
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
+
S
M
O
T
E
da
ta
.
F
ig
ur
e
6
s
how
s
th
e
c
om
pa
r
is
on
of
a
c
c
ur
a
c
y
r
e
s
ul
t
s
obt
a
in
e
d
by
e
a
c
h
a
r
c
hi
te
c
tu
r
e
on
tr
a
in
in
g
a
nd
va
li
da
ti
on
da
ta
.
T
h
e
bl
ue
li
ne
r
e
pr
e
s
e
nt
s
a
c
c
ur
a
c
y
v
a
lu
e
f
or
tr
a
in
in
g
da
ta
,
w
hi
le
th
e
or
a
ng
e
li
ne
s
ho
w
s
v
a
li
da
ti
on
da
ta
a
c
c
ur
a
c
y
a
s
s
h
ow
n
i
n
F
ig
ur
e
s
6
(
a
)
to
6
(
d)
.
(
a
)
(
b)
(
c
)
(
d)
F
ig
ur
e
6
.
G
r
a
ph of
a
c
c
ur
a
c
y a
nd l
os
s
f
or
t
he
C
X
R
i
m
a
g
e
c
la
s
s
if
ic
a
ti
on t
r
a
in
in
g pr
oc
e
s
s
us
in
g a
r
c
hi
te
c
tu
r
e
:
(
a
)
R
e
s
N
e
t5
0, (
b)
R
e
s
N
e
t5
0+
S
M
O
T
E
da
ta
, (
c
)
R
e
s
N
e
t5
0+
c
la
s
s
w
e
ig
ht
e
d a
ppr
oa
c
h
, (
d)
R
e
s
N
e
t5
0+
S
M
O
T
E
da
ta
+
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
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
A
r
e
v
ol
ut
io
nar
y
c
onv
ol
ut
io
nal
ne
u
r
al
ne
tw
or
k
ar
c
hi
te
c
tu
r
e
f
or
m
or
e
ac
c
u
r
at
e
…
(
M
ul
ia
di
)
523
F
ig
ur
e
6
c
ont
a
in
s
f
our
li
ne
gr
a
phs
,
e
a
c
h
r
e
pr
e
s
e
nt
in
g
m
o
de
l
a
c
c
ur
a
c
y
ove
r
s
e
ve
r
a
l
e
poc
h
s
.
S
pe
c
if
ic
a
ll
y, e
a
c
h gr
a
ph i
s
l
a
be
le
d di
f
f
e
r
e
nt
ly
, s
how
in
g t
he
t
ype
of
m
ode
l
or
da
ta
pr
oc
e
s
s
in
g us
e
d.
F
ig
ur
e
6
(
a
)
is
ti
tl
e
d
"
R
e
s
N
e
t5
0”
,
s
how
in
g
two
li
ne
s
w
it
h
one
r
e
pr
e
s
e
nt
in
g
tr
a
in
in
g
a
c
c
ur
a
c
y
(
o
r
a
nge
)
a
nd
th
e
ot
he
r
in
di
c
a
ti
ng
va
li
da
ti
on
a
c
c
ur
a
c
y
(
bl
ue
)
.
B
ot
h
li
ne
s
s
how
a
n
upw
a
r
d
tr
e
nd,
s
how
in
g
a
n
in
c
r
e
a
s
e
in
a
c
c
ur
a
c
y
a
s
th
e
num
be
r
of
e
poc
hs
r
is
e
s
.
T
r
a
in
in
g a
c
c
ur
a
c
y s
ta
r
ts
s
li
ght
ly
a
bove
0.6326 a
nd e
nds
ne
a
r
0.8783, while va
li
d
a
ti
on
a
c
c
ur
a
c
y
c
om
m
e
nc
e
s
a
ppr
oxi
m
a
te
ly
b
e
lo
w
0.6145
a
nd
c
onc
l
ude
s
a
bov
e
th
e
tr
a
in
in
g
a
c
c
ur
a
c
y
a
t
0.8436.
F
ig
ur
e
6
(
b)
is
ti
tl
e
d
"
S
M
O
T
E
d
a
ta
"
,
s
how
in
g
li
ne
s
f
or
tr
a
in
in
g
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y.
T
r
a
in
in
g
a
c
c
ur
a
c
y
c
om
m
e
nc
e
s
s
li
ght
ly
a
bove
0.8246
a
nd
e
xpe
r
ie
nc
e
s
a
li
tt
le
de
c
r
e
a
s
e
b
e
lo
w
0.9968.
M
e
a
nw
hi
le
,
va
li
da
ti
on
a
c
c
ur
a
c
y c
om
m
e
n
c
e
s
a
t
a
ppr
oxi
m
a
te
ly
0.9673, r
is
in
g t
o m
e
e
t
th
e
t
r
a
in
in
g a
c
c
ur
a
c
y i
n t
he
m
id
dl
e
of
t
he
gr
a
ph,
but
di
ve
r
ge
s
a
nd
e
nds
s
li
ght
ly
be
lo
w
0.9273.
F
ig
ur
e
6
(
c
)
is
ti
tl
e
d
"
c
la
s
s
w
e
ig
ht
e
d a
ppr
oa
c
h
"
,
w
hi
c
h
is
s
im
il
a
r
to
th
e
f
ir
s
t
g
r
a
ph
c
ha
r
a
c
te
r
iz
e
d
by
two
li
ne
s
f
or
t
r
a
in
in
g
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y.
T
r
a
in
in
g
a
c
c
ur
a
c
y
c
om
m
e
n
c
e
s
at
0.5572
a
nd
r
is
e
s
to
s
li
ght
ly
be
lo
w
0.9903,
w
hi
le
va
li
da
ti
on
a
c
c
ur
a
c
y
s
t
a
r
ts
a
r
ound
0.7636
a
nd
in
c
r
e
a
s
e
s
to
0.9818.
F
ig
ur
e
6
(
d)
s
how
s
two
gr
a
phs
de
pi
c
ti
ng
a
c
c
ur
a
c
y
a
nd
l
os
s
of
th
e
pr
opos
e
d
m
ode
l.
A
c
c
ur
a
c
y
on
bot
h
tr
a
in
in
g
a
nd
va
li
da
ti
on
da
ta
te
nds
to
in
c
r
e
a
s
e
w
it
h
a
n
in
c
r
e
a
s
in
g
num
be
r
of
e
poc
hs
.
T
r
a
in
in
g
a
c
c
ur
a
c
y
is
0.9968
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y
i
s
in
th
e
hi
gh
r
a
nge
of
0.9891.
F
ur
th
e
r
m
or
e
,
th
e
lo
s
s
va
lu
e
s
f
or
bot
h
tr
a
in
in
g
a
nd
va
li
da
ti
on
da
ta
te
nd
to
de
c
r
e
a
s
e
w
it
h
a
n
in
c
r
e
a
s
in
g
num
be
r
of
e
poc
hs
,
w
it
h
a
lo
s
s
of
0.0075
a
nd
a
va
l_
lo
s
s
of
0.0577. T
hi
s
s
how
s
t
ha
t
th
e
m
ode
l
is
c
a
pa
bl
e
of
r
e
duc
in
g l
os
s
v
a
lu
e
a
s
t
he
t
r
a
in
in
g pr
ogr
e
s
s
e
s
.
T
he
gr
a
p
hs
s
h
ow
th
a
t
th
e
pr
o
pos
e
d
m
ode
l
is
c
a
pa
bl
e
of
i
m
pr
ov
in
g
a
c
c
ur
a
c
y
a
nd
r
e
du
c
in
g
lo
s
s
a
s
th
e
tr
a
in
in
g
pr
ogr
e
s
s
e
s
.
T
h
e
va
li
da
ti
on
a
c
c
ur
a
c
y
a
ppr
oa
c
hi
ng
or
e
xc
e
e
di
ng
th
e
tr
a
in
i
ng
a
c
c
ur
a
c
y
s
how
s
th
a
t
th
e
m
ode
l
c
a
n
ge
ne
r
a
l
iz
e
w
e
ll
t
o
ne
w
d
a
ta
,
s
h
ow
in
g
goo
d
p
e
r
f
or
m
a
n
c
e
.
A
ll
gr
a
ph
s
h
a
ve
t
he
x
-
a
x
is
l
a
b
e
le
d
"
E
poc
h"
a
nd
th
e
y
-
a
xi
s
la
be
l
e
d
"
A
c
c
ur
a
c
y."
T
he
x
-
a
xi
s
s
h
ow
s
th
e
pr
ogr
e
s
s
of
e
po
c
hs
,
a
lt
ho
ugh
th
e
e
x
a
c
t
n
um
be
r
i
s
not
s
pe
c
if
ic
.
T
he
y
-
a
xi
s
s
how
s
a
c
c
ur
a
c
y,
r
a
ngi
ng f
r
om
a
r
o
und
0.65
t
o s
l
ig
ht
ly
a
b
ove
0.9
i
n
a
ll
gr
a
phs
w
it
h
a
c
c
ur
a
c
y
va
lu
e
s
f
lu
c
tu
a
ti
ng
a
nd
l
e
ge
nd
s
ho
w
in
g
a
li
n
e
th
a
t
r
e
pr
e
s
e
n
ts
tr
a
in
in
g
a
nd
va
li
d
a
ti
on
.
T
h
e
s
e
gr
a
ph
s
a
r
e
c
om
m
o
nl
y
us
e
d
to
e
v
a
lu
a
t
e
t
he
pe
r
f
or
m
a
nc
e
of
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
l
,
f
o
c
us
in
g
on
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
g
e
ne
r
a
ti
ng
un
s
e
e
n
ne
w
da
t
a
(
v
a
li
da
ti
o
n
a
c
c
ur
a
c
y)
c
om
p
a
r
e
d
to
th
e
pe
r
f
or
m
a
nc
e
of
tr
a
in
in
g
da
t
a
. T
he
pr
op
os
e
d
m
od
e
l
ove
r
c
om
e
s
li
m
it
a
ti
on
s
a
nd
le
v
e
r
a
g
e
s
th
e
a
d
va
nt
a
ge
s
of
di
f
f
e
r
e
nt
C
N
N
a
r
c
h
it
e
c
tu
r
e
(
R
e
s
N
e
t5
0,
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h,
a
nd
m
ode
l
S
M
O
T
E
d
a
ta
)
by
c
om
bi
ni
ng
va
r
io
us
c
l
a
s
s
if
ic
a
t
io
n
r
e
s
ul
t
s
.
F
ur
th
e
r
m
or
e
,
th
e
pr
op
os
e
d
m
ode
l
w
or
k
s
dur
in
g
th
e
t
r
a
in
i
ng pha
s
e
, w
hi
c
h a
ll
o
w
s
c
he
c
ki
ng t
h
e
pe
r
f
or
m
a
nc
e
r
e
s
ul
ts
of
t
he
w
e
ig
ht
s
a
t
e
a
c
h e
po
c
h on tr
a
i
ni
ng a
n
d
va
li
da
ti
on
d
a
ta
t
o
ha
ndl
e
ove
r
f
it
ti
n
g.
O
ur
r
e
s
e
a
r
c
h
s
how
s
th
a
t
th
e
hi
ghe
s
t
p
e
r
f
or
m
a
n
c
e
a
c
hi
e
ve
d
by
th
e
pr
op
os
e
d
m
e
th
od
i
s
n
ot
onl
y
li
m
it
e
d t
o ne
a
r
-
p
e
r
f
e
c
t
a
c
c
ur
a
c
y l
e
v
e
ls
,
b
ut
a
ls
o
to
ve
r
y
hi
g
h
pr
e
c
i
s
io
n
,
r
e
c
a
ll
,
a
nd
F
1
s
c
or
e
.
T
hi
s
c
a
n
b
e
in
te
r
pr
e
t
e
d
t
o
m
e
a
n
t
ha
t
hi
gh
a
c
c
ur
a
c
y
i
s
no
t
a
l
w
a
ys
a
s
s
oc
ia
t
e
d
w
it
h
poor
pe
r
f
or
m
a
nc
e
in
ot
he
r
a
s
p
e
c
t
s
s
uc
h
a
s
pr
e
c
i
s
io
n
,
r
e
c
a
ll
,
a
nd
F
1
s
c
or
e
.
T
he
pr
o
pos
e
d
m
e
th
od
c
a
n
b
e
ne
f
it
f
r
om
th
e
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
,
S
M
O
T
E
da
t
a
,
a
n
d
R
e
s
N
e
t5
0
a
r
c
h
it
e
c
tu
r
e
w
it
h
out
a
dv
e
r
s
e
ly
im
p
a
c
ti
n
g
th
e
m
ode
l
p
e
r
f
or
m
a
n
c
e
in
te
r
m
s
of
lu
ng
c
a
nc
e
r
d
e
te
c
ti
on
.
T
h
e
e
v
a
lu
a
ti
on
r
e
s
ul
t
s
s
how
th
a
t
th
i
s
c
o
m
bi
na
ti
o
n
of
m
e
th
o
ds
i
s
a
bl
e
to
im
pr
o
ve
ov
e
r
a
ll
pe
r
f
or
m
a
n
c
e
w
it
hout
s
a
c
r
if
i
c
in
g o
th
e
r
i
m
por
ta
n
t
a
s
pe
c
t
s
in
l
ung
c
a
n
c
e
r
c
la
s
s
if
i
c
a
ti
on.
3.2.2.
C
on
f
u
s
io
n
m
at
r
ix
I
n
tr
a
in
in
g
C
N
N
m
ode
l
w
it
h
R
e
s
N
e
t5
0,
R
e
s
N
e
t5
0+
S
M
O
T
E
da
ta
,
R
e
s
N
e
t5
0
+
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
,
a
nd
R
e
s
N
e
t5
0
+
S
M
O
T
E
da
ta
+
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
(
pr
opos
e
d
m
e
th
od
)
,
a
n
im
po
r
ta
nt
e
va
lu
a
ti
on
to
ol
c
a
ll
e
d
th
e
c
onf
us
io
n
m
a
tr
ix
i
s
u
s
e
d
f
or
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
m
e
nt
.
T
h
e
c
onf
us
io
n
m
a
tr
ix
pr
ovi
de
s
a
de
t
a
il
e
d
ove
r
vi
e
w
of
how
w
e
ll
th
e
m
ode
l
c
a
n
c
la
s
s
if
y
e
a
c
h
c
a
t
e
gor
y.
T
hi
s
m
a
tr
ix
c
ons
is
ts
of
f
our
m
a
in
pa
r
ts
,
na
m
e
ly
tr
ue
pos
it
iv
e
(
T
P
)
,
tr
ue
ne
ga
ti
ve
(
T
N
)
,
f
a
ls
e
pos
it
iv
e
(
F
P
)
,
a
nd
f
a
ls
e
n
e
ga
ti
v
e
(
F
N
)
.
TP
r
e
pr
e
s
e
nt
s
th
e
num
be
r
of
s
a
m
pl
e
s
c
or
r
e
c
tl
y
c
la
s
s
if
ie
d
in
to
th
e
c
or
r
e
c
t
c
la
s
s
,
w
hi
le
TN
in
c
lu
de
s
s
a
m
pl
e
s
c
or
r
e
c
tl
y
c
la
s
s
if
ie
d
in
to
th
e
ne
ga
ti
ve
c
la
s
s
.
FP
is
th
e
num
be
r
of
s
a
m
pl
e
s
m
is
ta
ke
nl
y c
la
s
s
if
ie
d
a
s
pos
it
iv
e
, w
hi
le
FN
is
th
e
num
be
r
of
s
a
m
pl
e
s
m
i
s
ta
ke
nl
y
c
la
s
s
if
ie
d a
s
ne
ga
ti
ve
.
I
n
F
ig
ur
e
7
(
a
)
to
(
d
)
you c
a
n s
e
e
t
he
f
ol
lo
w
in
g
.
A
s
s
how
n
in
F
ig
ur
e
7,
th
e
us
e
of
c
onf
us
io
n
m
a
tr
ix
f
a
c
il
it
a
te
s
th
e
c
a
lc
ul
a
ti
on
of
s
e
v
e
r
a
l
im
por
ta
nt
e
va
lu
a
ti
on
m
e
tr
ic
s
s
uc
h
a
s
a
c
c
ur
a
c
y,
pr
e
c
i
s
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
.
A
c
c
ur
a
c
y
m
e
a
s
ur
e
s
th
e
a
bi
li
ty
of
m
ode
l
to
c
or
r
e
c
tl
y
pr
e
di
c
t
c
la
s
s
e
s
ov
e
r
a
ll
.
P
r
e
c
is
io
n
m
e
a
s
ur
e
s
a
c
c
ur
a
c
y
of
m
ode
l'
s
pos
it
iv
e
pr
e
di
c
ti
ons
,
w
hi
le
r
e
c
a
ll
m
e
a
s
ur
e
s
th
e
a
bi
li
ty
to
f
in
d
a
ll
a
c
tu
a
l
pos
it
iv
e
c
a
s
e
s
.
F
ur
th
e
r
m
or
e
,
th
e
F
1
-
s
c
or
e
is
a
c
om
bi
na
ti
on
of
p
r
e
c
is
io
n
a
nd
r
e
c
a
ll
.
T
hr
ough
th
e
a
na
ly
s
is
of
c
onf
us
io
n
m
a
tr
ix
a
nd
e
va
lu
a
t
io
n
m
e
tr
ic
s
,
de
e
p
in
s
ig
ht
s
in
to
th
e
pe
r
f
o
r
m
a
nc
e
of
C
N
N
m
ode
l
w
it
h
R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
c
a
n
be
obt
a
in
e
d.
M
e
a
nw
hi
le
,
a
dj
us
tm
e
nt
or
f
in
e
-
tu
ni
ng
is
m
a
de
to
im
pr
ove
c
la
s
s
if
ic
a
ti
on r
e
s
ul
ts
w
h
e
r
e
ne
c
e
s
s
a
r
y, a
s
s
how
n i
n
T
a
b
le
1.
T
a
bl
e
1. C
om
pa
r
is
on of
l
ung C
T
s
c
a
n i
m
a
ge
c
la
s
s
if
ic
a
ti
on pe
r
f
or
m
a
nc
e
us
in
g R
e
s
N
e
t
M
e
t
hod
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
R
e
s
N
e
t
50
0.8783
0.7400
0.9700
0.8400
R
e
s
N
e
t
c
l
a
s
s
w
e
i
ght
e
d a
ppr
oa
c
h
0.9903
0.9800
0.9800
0.9800
R
e
s
N
e
t
S
M
O
T
E
da
t
a
0.9952
0.9800
0.9800
0.9800
P
ro
pos
e
d
m
e
t
hod
0.9968
0.9900
0.9900
0.9900
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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
:
516
-
526
524
(
a
)
(
b)
(
c
)
(
d)
F
ig
ur
e
7
.
C
onf
us
io
n
m
a
tr
i
x
(
a
)
R
e
s
N
e
t
50,
(
b)
R
e
s
N
e
t
50
+
S
M
O
T
E
da
ta
, (
c
)
R
e
s
N
e
t5
0
+
c
la
s
s
w
e
ig
ht
e
d
a
p
pr
o
a
c
h
,
(
d)
R
e
s
N
e
t5
0+
S
M
O
T
E
da
ta
+
c
la
s
s
w
e
ig
ht
e
d a
ppr
oa
c
h
3.2.3. E
val
u
at
io
n
I
n
th
is
r
e
s
e
a
r
c
h,
lu
ng C
T
s
c
a
n
im
a
ge
c
la
s
s
if
ic
a
ti
on
is
pe
r
f
or
m
e
d
us
in
g R
e
s
N
e
t
m
ode
l,
a
nd
th
e
r
e
s
ul
ts
a
r
e
c
om
pa
r
e
d
w
it
h
th
e
p
r
e
vi
ous
in
ve
s
ti
ga
ti
on.
T
he
c
om
pa
r
is
on
of
lu
ng
C
T
s
c
a
n
im
a
ge
c
la
s
s
if
ic
a
ti
on
r
e
s
ul
ts
w
it
h
ot
he
r
r
e
s
e
a
r
c
h
s
a
r
e
pr
e
s
e
nt
e
d
in
T
a
bl
e
1.
T
he
e
v
a
lu
a
ti
on
r
e
s
ul
ts
in
T
a
bl
e
1
s
how
th
e
pe
r
f
or
m
a
nc
e
c
om
pa
r
is
on
a
m
ong
th
e
f
our
te
s
te
d
m
e
th
ods
.
B
a
s
e
d
on
th
e
r
e
s
ul
t
s
,
R
e
s
N
e
t5
0
m
ode
l
w
it
hout
s
pe
c
if
ic
a
dj
us
tm
e
nt
s
pr
ovi
de
s
a
c
c
ur
a
c
y of
87.83%
,
pr
e
c
is
io
n of
74.00%
, r
e
c
a
ll
of
97.
00%
, a
nd F
1
-
s
c
or
e
of
84.00%
. W
he
n
a
ppl
yi
ng
th
e
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
to
R
e
s
N
e
t,
a
s
ig
ni
f
ic
a
nt
im
pr
ove
m
e
nt
is
ob
s
e
r
ve
d
w
it
h
99.03%
a
c
c
ur
a
c
y,
a
nd
98.00%
f
or
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F1
-
s
c
or
e
,
r
e
s
pe
c
ti
ve
ly
.
T
he
us
e
of
S
M
O
T
E
da
ta
on
R
e
s
N
e
t
pr
ovi
de
s
th
e
hi
ghe
s
t
a
c
c
ur
a
c
y of
99.52%
w
it
h c
ons
is
te
nt
pr
e
c
is
io
n, r
e
c
a
ll
, a
nd F
1
-
s
c
or
e
a
t
98.00%
.
H
ow
e
ve
r
, t
he
pr
opos
e
d
m
e
th
od,
c
ons
is
ti
ng
of
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
,
S
M
O
T
E
da
ta
,
a
n
d
R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
,
s
how
s
th
e
m
o
s
t
s
up
e
r
io
r
pe
r
f
or
m
a
nc
e
w
it
h
a
c
c
ur
a
c
y
of
99.68%
,
a
nd
99.00%
f
or
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
,
r
e
s
pe
c
ti
ve
ly
.
T
he
s
e
r
e
s
ul
ts
s
how
th
a
t
th
e
pr
opos
e
d
m
e
th
od
c
a
n
pos
it
iv
e
ly
c
ont
r
ib
ut
e
to
im
pr
ovi
ng
lu
ng
c
a
nc
e
r
de
te
c
ti
on
th
r
oug
h
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
l.
T
he
pr
opos
e
d
m
e
th
od
s
how
s
th
e
hi
ghe
s
t
pe
r
f
or
m
a
nc
e
w
it
h
a
ppr
oxi
m
a
te
ly
pe
r
f
e
c
t
a
c
c
ur
a
c
y
a
s
w
e
ll
a
s
v
e
r
y
hi
gh
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
.
T
hi
s
s
how
s
th
e
s
u
c
c
e
s
s
f
ul
c
om
bi
na
ti
on
of
th
e
c
la
s
s
w
e
ig
ht
e
d a
ppr
oa
c
h
, S
M
O
T
E
da
ta
,
a
nd R
e
s
N
e
t5
0
a
r
c
hi
te
c
t
ur
e
i
n i
m
pr
ovi
ng l
ung c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on.
T
hi
s
r
e
s
e
a
r
c
h
e
xpl
or
e
s
th
e
us
e
of
c
om
bi
na
ti
on
m
e
th
ods
s
uc
h
a
s
th
e
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
,
S
M
O
T
E
da
ta
,
a
nd
th
e
c
om
pr
e
he
ns
iv
e
R
e
s
N
e
t5
0
a
r
c
hi
te
c
tu
r
e
in
im
pr
ovi
ng
th
e
pe
r
f
or
m
a
nc
e
of
lu
ng
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on
m
ode
ls
.
H
ow
e
ve
r
,
f
ur
th
e
r
a
nd
in
-
de
pt
h
s
tu
di
e
s
m
a
y
be
ne
e
d
e
d
to
c
onf
ir
m
th
e
s
upe
r
io
r
it
y
of
th
is
m
e
th
od
in
pr
a
c
ti
c
a
l
s
c
e
na
r
io
s
in
vol
vi
ng
m
or
e
c
om
pl
e
x
a
nd
di
ve
r
s
e
da
ta
,
e
s
pe
c
ia
ll
y
r
e
g
a
r
di
ng
th
e
ge
n
e
r
a
li
z
a
ti
on
of
th
e
r
e
s
ul
ts
to
ne
ve
r
-
be
f
or
e
-
s
e
e
n
te
s
t
da
ta
.
O
ur
s
tu
dy
s
how
s
th
a
t
our
pr
opos
e
d
a
ppr
oa
c
h,
na
m
e
ly
a
c
om
bi
na
ti
on
of
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
,
S
M
O
T
E
da
ta
,
a
nd
R
e
s
N
e
t5
0
a
r
c
hi
te
c
tu
r
e
,
is
m
or
e
r
obus
t
to
c
la
s
s
im
ba
la
n
c
e
a
nd
va
r
ia
ti
on
in
lu
ng
c
a
nc
e
r
da
ta
s
e
ts
.
F
ut
ur
e
s
tu
di
e
s
c
oul
d
e
xpl
or
e
de
e
pe
r
in
te
gr
a
ti
on
of
da
ta
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
s
uc
h
a
s
d
a
ta
a
ugm
e
nt
a
ti
on
to
e
xpa
nd
th
e
g
e
ne
r
a
l
iz
a
ti
on
of
th
e
m
ode
l
a
c
r
os
s
va
r
io
us
m
e
di
c
a
l
da
ta
s
e
ts
,
th
e
r
e
by
pr
oduc
in
g
m
or
e
r
e
li
a
bl
e
a
nd
de
f
e
ns
ib
le
r
e
s
ul
ts
.
R
e
c
e
nt
obs
e
r
va
ti
ons
s
ho
w
th
a
t
th
e
in
te
gr
a
ti
on
of
c
la
s
s
im
ba
la
nc
e
ha
ndl
in
g
s
tr
a
te
gi
e
s
s
u
c
h
a
s
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
a
nd
da
ta
S
M
O
T
E
in
to
th
e
R
e
s
N
e
t5
0
a
r
c
hi
te
c
tu
r
e
not
onl
y
i
m
pr
ove
s
m
ode
l
pe
r
f
o
r
m
a
nc
e
in
te
r
m
s
o
f
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
a
nd
F
1
s
c
or
e
,
but
a
ls
o
pr
ovi
de
s
c
onc
lu
s
iv
e
e
vi
d
e
nc
e
th
a
t
th
e
s
e
phe
nom
e
n
a
a
r
e
r
e
la
te
d
to
c
ha
nge
s
in
th
e
s
tr
uc
tu
r
e
of
th
e
da
ta
s
e
t
a
nd
c
la
s
s
di
s
tr
ib
ut
io
n, not due
t
o a
n a
r
ti
f
ic
ia
l
or
s
ynt
he
ti
c
in
c
r
e
a
s
e
i
n t
he
numbe
r
of
m
in
or
it
y c
la
s
s
s
a
m
pl
e
s
.
4.
C
O
N
C
L
U
S
I
O
N
I
n
c
onc
lu
s
io
n,
th
is
r
e
s
e
a
r
c
h
s
how
e
d
th
a
t
th
e
a
ddi
ti
on
of
s
pe
c
if
ic
m
e
th
ods
s
uc
h
a
s
th
e
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
a
nd
S
M
O
T
E
da
ta
to
R
e
s
N
e
t5
0
a
r
c
hi
te
c
tu
r
e
s
ig
ni
f
ic
a
n
tl
y
im
pr
ov
e
d
th
e
pe
r
f
or
m
a
nc
e
of
th
e
pr
opos
e
d
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
A
r
e
v
ol
ut
io
nar
y
c
onv
ol
ut
io
nal
ne
u
r
al
ne
tw
or
k
ar
c
hi
te
c
tu
r
e
f
or
m
or
e
ac
c
u
r
at
e
…
(
M
ul
ia
di
)
525
m
ode
l
.
B
a
s
e
d
on
th
e
r
e
s
ul
ts
,
R
e
s
N
e
t5
0
m
ode
l
w
it
hout
s
pe
c
if
ic
a
dj
us
tm
e
nt
s
s
how
ed
good
p
e
r
f
or
m
a
nc
e
,
s
pe
c
if
ic
a
ll
y
in
r
e
c
a
ll
but
e
nc
ount
e
r
ed
c
ha
ll
e
nge
s
in
pr
e
c
i
s
io
n
.
T
he
im
pl
e
m
e
nt
a
ti
on
of
th
e
c
la
s
s
w
e
ig
ht
e
d
a
ppr
oa
c
h
pr
ov
id
e
d
a
s
ig
ni
f
ic
a
nt
im
pr
ove
m
e
nt
in
a
c
c
ur
a
c
y
a
nd
pr
e
c
is
io
n
w
it
hout
s
a
c
r
if
ic
in
g
r
e
c
a
ll
,
w
hi
le
th
e
us
e
of
S
M
O
T
E
da
ta
r
e
s
ul
t
e
d
in
th
e
hi
ghe
s
t
a
c
c
ur
a
c
y.
T
he
pr
opos
e
d
m
e
th
od
s
how
e
d
th
e
m
os
t
s
upe
r
io
r
pe
r
f
or
m
a
nc
e
w
it
h
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
a
ppr
oa
c
hi
ng
pe
r
f
e
c
t
va
lu
e
s
w
it
h
a
c
c
ur
a
c
y
of
9
9.68%
a
nd
pr
e
c
is
io
n
of
99.00%
.
T
h
e
r
e
f
or
e
,
th
e
in
te
gr
a
ti
on
of
c
la
s
s
im
ba
la
nc
e
ha
ndl
in
g
s
tr
a
te
gi
e
s
a
nd
R
e
s
N
e
t5
0
a
r
c
hi
te
c
tu
r
e
de
ve
lo
pm
e
nt
in
th
e
pr
opos
e
d
m
ode
l
s
how
ed
s
ig
ni
f
ic
a
nt
pot
e
nt
ia
l
in
im
pr
ovi
ng
a
c
c
ur
a
c
y
of
lu
ng
c
a
nc
e
r
di
a
gnos
i
s
t
hr
ough ma
c
hi
ne
l
e
a
r
ni
ng mode
l.
R
E
F
E
R
E
N
C
E
S
[
1]
A
.
R
i
z
a
l
,
R
.
H
i
da
ya
t
,
H
.
A
.
N
ugr
oho,
a
nd
W
.
A
.
C
a
hy
a
di
,
“
L
ung
s
ound
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
m
ul
t
i
r
e
s
ol
ut
i
on
H
i
guc
hi
f
r
a
c
t
a
l
di
m
e
ns
i
on
m
e
a
s
ur
e
m
e
nt
,”
I
nt
e
r
nat
i
onal
J
ou
r
nal
of
E
l
e
c
t
r
i
c
al
and C
om
put
e
r
E
ngi
ne
e
r
i
ng
,
vol
.
13,
no.
5,
pp.
5091
–
5100,
2023,
doi
:
10.11591/
i
j
e
c
e
.v13i
5.pp5091
-
5100.
[
2]
J
.
A
l
s
hudukhi
,
S
.
A
l
j
a
l
oud,
T
.
S
.
A
l
ha
r
bi
,
a
nd
S
.
A
be
ba
w
,
“
C
onvol
ut
i
ona
l
n
e
ur
a
l
ne
t
w
or
k
a
r
c
hi
t
e
c
t
ur
e
s
t
o
s
ol
ve
a
pr
obl
e
m
of
t
ube
r
c
ul
os
i
s
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
x
-
r
a
y
i
m
a
ge
s
of
t
he
l
u
ngs
,
”
J
ou
r
nal
of
N
anom
at
e
r
i
al
s
,
vol
.
2022,
2022,
doi
:
10.1155/
2022/
2509830.
[
3]
F
. L
i
u, C
. Z
a
ng, J
. S
hi
, W
. H
e
, Y
. L
i
a
ng, a
nd L
. L
i
, “
A
n
i
m
pr
ove
d
C
O
V
I
D
-
19
l
ung
X
-
R
a
y i
m
a
ge
c
l
a
s
s
i
f
i
c
a
t
i
on a
l
gor
i
t
hm
ba
s
e
d
on
C
onvN
e
X
t
ne
t
w
or
k
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
I
m
age
and G
r
aphi
c
s
, 2023, doi
:
10.1142/
S
0219467824500360.
[
4]
M
.
V
e
da
r
a
j
,
C
.
S
.
A
ni
t
a
,
A
.
M
ur
a
l
i
dha
r
,
V
.
L
a
v
a
nya
,
K
.
B
a
l
a
s
a
r
a
nya
,
a
nd
P
.
J
a
ga
de
e
s
a
n,
“
E
a
r
l
y
pr
e
di
c
t
i
on
of
l
ung
c
a
nc
e
r
us
i
ng
ga
us
s
i
a
n
n
a
i
ve
ba
ye
s
c
l
a
s
s
i
f
i
c
a
t
i
on
a
l
gor
i
t
hm
,”
I
nt
e
r
nat
i
onal
J
ou
r
nal
of
I
nt
e
l
l
i
ge
nt
Sy
s
t
e
m
s
and
A
ppl
i
c
at
i
ons
i
n
E
ngi
ne
e
r
i
ng
,
vol
.
11, no. 6s
, pp. 838
–
848, 2023.
[
5]
R
.
U
.
O
s
a
r
ogi
a
gbon
e
t
al
.
,
“
T
he
i
nt
e
r
na
t
i
ona
l
a
s
s
oc
i
a
t
i
on
f
or
t
he
s
t
udy
of
l
un
g
c
a
nc
e
r
l
ung
c
a
nc
e
r
s
t
a
gi
ng
pr
oj
e
c
t
:
ove
r
vi
e
w
o
f
c
ha
l
l
e
nge
s
a
nd oppor
t
uni
t
i
e
s
i
n r
e
vi
s
i
ng t
h
e
noda
l
c
l
a
s
s
i
f
i
c
a
t
i
on of
l
ung c
a
nc
e
r
,
”
J
our
nal
of
T
hor
ac
i
c
O
nc
ol
ogy
,
vol
. 18,
no. 4,
pp.
410
–
418, 2023, doi
:
10.1016/
j
.j
t
ho.2022.12.009.
[
6]
A
.
N
a
i
k
a
nd
D
. R
.
E
dl
a
,
“
L
ung
nodul
e
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
c
om
bi
na
t
i
on
of
C
N
N
,
s
e
c
ond
a
nd
hi
ghe
r
or
de
r
t
e
xt
ur
e
f
e
a
t
ur
e
s
,”
J
our
na
l
of
I
nt
e
l
l
i
ge
nt
and F
uz
z
y
Sy
s
t
e
m
s
, vol
. 41, no. 5, pp. 5243
–
5251, 2021, doi
:
10.3
233/
J
I
F
S
-
189847.
[
7]
L
.
P
ha
m
,
H
.
P
ha
n,
R
.
P
a
l
a
ni
a
ppa
n,
A
.
M
e
r
t
i
ns
,
a
nd
I
.
M
c
L
oughl
i
n,
“
C
N
N
-
M
o
E
ba
s
e
d
f
r
a
m
e
w
or
k
f
or
c
l
a
s
s
i
f
i
c
a
t
i
on
of
r
e
s
pi
r
a
t
or
y
a
nom
a
l
i
e
s
a
nd
l
ung
di
s
e
a
s
e
de
t
e
c
t
i
on,”
I
E
E
E
J
our
nal
of
B
i
om
e
di
c
al
and
H
e
al
t
h
I
nf
or
m
at
i
c
s
,
vol
.
25,
no.
8,
pp.
2938
–
2947,
2021,
doi
:
10.1109/
J
B
H
I
.2021.3064237.
[
8]
L
.
N
.
M
a
hdy,
A
.
I
.
B
.
E
l
S
e
dda
w
y,
a
nd
K
.
A
.
E
z
z
a
t
,
“
A
ut
om
a
t
i
c
C
O
V
I
D
-
19
l
ung
i
m
a
ge
s
c
l
a
s
s
i
f
i
c
a
t
i
on
s
ys
t
e
m
ba
s
e
d
on
c
onvol
ut
i
on
ne
ur
a
l
ne
t
w
or
k,”
I
nt
e
r
nat
i
onal
J
our
nal
of
E
l
e
c
t
r
i
c
al
and
C
om
put
e
r
E
ngi
ne
e
r
i
ng
,
vol
.
12,
no.
5,
pp.
5573
–
5579,
2022,
doi
:
10.11591/
i
j
e
c
e
.v12i
5.pp5573
-
5579.
[
9]
B
.
T
.
W
.
P
ut
r
a
,
R
.
A
m
i
r
udi
n,
a
nd
B
.
M
a
r
ha
e
na
nt
o,
“
T
he
e
va
l
ua
t
i
on
of
de
e
p
l
e
a
r
ni
ng
us
i
ng
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
o
r
k
(
C
N
N
)
a
ppr
oa
c
h f
or
i
de
nt
i
f
y
i
n
g A
r
a
bi
c
a
a
nd
r
obus
t
a
c
of
f
e
e
pl
a
nt
s
,”
J
our
nal
of
B
i
os
y
s
t
e
m
s
E
ngi
ne
e
r
i
ng
, vol
. 47, no. 2, pp. 118
–
129, 2022,
doi
:
10.1007/
s
42853
-
022
-
00136
-
y.
[
10]
D
.
R
ua
n,
F
.
Z
ha
ng,
L
.
Z
ha
ng,
a
nd
J
.
Y
a
n,
“
O
pt
i
m
a
l
m
odi
f
i
c
a
t
i
ons
i
n
C
N
N
f
o
r
be
a
r
i
ng
f
a
ul
t
c
l
a
s
s
i
f
i
c
a
t
i
on
a
nd
a
d
a
pt
a
t
i
on
a
c
r
os
s
di
f
f
e
r
e
nt
w
or
k
i
ng
c
ondi
t
i
ons
,”
J
our
nal
of
V
i
br
at
i
on
E
ng
i
ne
e
r
i
ng
and
T
e
c
hnol
ogi
e
s
,
vol
.
12,
no.
3,
pp.
4075
–
4095,
2024,
doi
:
10.1007/
s
42417
-
023
-
01106
-
0.
[
11]
N
. A
l
i
z
a
de
h, S
. A
f
r
a
kht
e
h, a
nd M
. R
. M
os
a
vi
, “
D
e
e
p C
N
N
-
ba
s
e
d c
l
a
s
s
i
f
i
c
a
t
i
on
of
m
ot
or
i
m
a
ge
r
y
t
a
s
ks
f
r
om
E
E
G
s
i
gna
l
s
us
i
ng 2D
w
a
ve
l
e
t
t
r
a
ns
f
or
m
e
d
i
m
a
ge
s
of
a
da
pt
i
ve
l
y
r
e
c
ons
t
r
uc
t
e
d
s
i
gna
l
s
f
r
om
M
V
M
D
de
c
om
pos
e
d
m
ode
s
,”
I
nt
e
r
nat
i
onal
J
our
nal
o
f
I
m
agi
ng Sy
s
t
e
m
s
and T
e
c
hnol
ogy
, vol
. 33, no. 6, pp. 1988
–
2011, 2023, doi
:
10.
1002/
i
m
a
.22913.
[
12]
T
.
Z
hou
e
t
al
.
,
“
C
l
a
s
s
i
f
i
c
a
t
i
on
of
l
ung
a
de
noc
a
r
c
i
nom
a
ba
s
e
d
on
i
m
m
une
c
he
c
kpoi
nt
a
nd
s
c
r
e
e
ni
ng
of
r
e
l
a
t
e
d
ge
ne
s
,”
J
our
nal
o
f
O
nc
ol
ogy
, vol
. 2021, 2021, doi
:
10.1155/
2021/
5512325.
[
13]
P
.
M
.
B
r
unt
ha
,
S
.
I
.
A
.
P
a
ndi
a
n,
J
.
A
ni
t
ha
,
S
.
S
.
A
br
a
ha
m
,
a
nd
S
.
N
.
K
um
a
r
,
“
A
nove
l
hybr
i
di
z
e
d
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on
a
ppr
oa
c
h
f
or
l
ung
nodul
e
c
l
a
s
s
i
f
i
c
a
t
i
on
ba
s
e
d
on
t
r
a
ns
f
e
r
l
e
a
r
ni
ng
t
e
c
hni
que
,”
J
our
nal
of
M
e
di
c
al
P
hy
s
i
c
s
,
vol
.
47,
no.
1,
pp.
1
–
9,
2022,
do
i
:
10.4103/
j
m
p.j
m
p_61_21.
[
14]
Y
.
Y
a
ng,
X
.
L
i
,
Z
.
H
a
n,
J
.
F
u,
a
nd
B
.
G
a
o,
“
R
e
s
e
a
r
c
h
on
c
l
a
s
s
i
f
i
c
a
t
i
on
of
b
e
ni
gn
a
nd
m
a
l
i
gna
nt
l
ung
nodul
e
s
ba
s
e
d
on
t
hr
e
e
-
di
m
e
ns
i
ona
l
m
ul
t
i
-
vi
e
w
s
que
e
z
e
-
a
nd
-
e
xc
i
t
a
t
i
on
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k,”
She
ng
w
u
y
i
x
ue
gong
c
he
ng
x
ue
z
a
z
hi
=
J
our
na
l
of
bi
om
e
di
c
al
e
ngi
ne
e
r
i
ng
=
She
ngw
u
y
i
x
ue
gongc
he
ngx
ue
z
az
hi
,
vol
.
39,
no.
3,
pp.
452
–
461,
2022,
doi
:
10.7507
/
1001
-
5515.202110059.
[
15]
P
ut
r
a
m
a
A
l
kha
i
r
i
a
nd
A
.
P
.
W
i
nda
r
t
o,
“
C
l
a
s
s
i
f
i
c
a
t
i
on
a
na
l
ys
i
s
of
ba
c
k
pr
opa
ga
t
i
on
-
opt
i
m
i
z
e
d
C
N
N
pe
r
f
or
m
a
nc
e
i
n
i
m
a
ge
pr
oc
e
s
s
i
ng
,”
J
our
nal
of
Sy
s
t
e
m
s
E
ngi
ne
e
r
i
ng
and
I
nf
or
m
at
i
on
T
e
c
hnol
ogy
(
J
O
SE
I
T
)
,
vol
.
2,
no.
1,
pp.
8
–
15,
2023,
doi
:
10.29207/
j
os
e
i
t
.v2i
1.5015.
[
16]
T
. I
m
a
nda
s
a
r
i
, M
. G
.
S
a
de
w
o,
A
. P
. W
i
nd
a
r
t
o, A
. W
a
nt
o,
H
. O
.
L
.
W
i
j
a
ya
,
a
nd
R
. K
ur
ni
a
w
a
n,
“
A
na
l
y
s
i
s
of
t
he
s
e
l
e
c
t
i
on f
a
c
t
or
of
onl
i
ne
t
r
a
ns
por
t
a
t
i
on
i
n
t
he
V
I
K
O
R
m
e
t
hod
i
n
P
e
m
a
t
a
ngs
i
a
nt
a
r
C
i
t
y,”
J
our
na
l
of
P
hy
s
i
c
s
:
C
onf
e
r
e
nc
e
Se
r
i
e
s
,
vol
.
1255,
no.
1
,
2019, doi
:
10.1088/
1742
-
6596/
1255/
1/
012008.
[
17]
P
.
A
l
kha
i
r
i
,
E
.
R
.
B
a
t
uba
r
a
,
R
.
R
os
ne
l
l
y,
W
.
W
a
n
a
ya
um
i
ni
,
a
nd
H
.
S
.
T
a
m
b
una
n,
“
E
f
f
e
c
t
of
gr
a
di
e
nt
de
s
c
e
nt
w
i
t
h
m
om
e
nt
um
ba
c
kpr
opa
ga
t
i
on
t
r
a
i
ni
ng
f
unc
t
i
on
i
n
de
t
e
c
t
i
ng
a
l
pha
be
t
l
e
t
t
e
r
s
,”
Si
nk
r
on
,
vol
.
8,
no.
1,
pp.
574
–
583,
2023,
doi
:
10.33395/
s
i
nkr
on.v8i
1.12183.
[
18]
N
.
K
e
s
a
v
a
nd
M
.
G
.
J
i
bukum
a
r
,
“
M
ul
t
i
-
C
ha
nne
l
C
N
N
ba
s
e
d
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
S
K
I
P
c
onne
c
t
i
on
a
nd
M
S
V
M
,
”
I
nt
e
r
nat
i
onal
J
our
nal
of
C
om
put
e
r
s
and A
ppl
i
c
at
i
ons
, vol
. 44, no. 10, pp. 981
–
990, 2022, doi
:
10.1080/
1206212X
.2022.2047443.
[
19]
H
.
F
ı
r
a
t
a
nd
D
.
H
a
nba
y,
“
C
om
pa
r
i
s
on
of
3D
C
N
N
ba
s
e
d
de
e
p
l
e
a
r
ni
ng
a
r
c
hi
t
e
c
t
ur
e
s
us
i
ng
hype
r
s
pe
c
t
r
a
l
i
m
a
ge
s
,”
J
our
nal
of
t
he
F
ac
ul
t
y
of
E
ngi
ne
e
r
i
ng and A
r
c
hi
t
e
c
t
ur
e
of
G
az
i
U
ni
v
e
r
s
i
t
y
, vol
. 38, no. 1, pp.
521
–
534, 2023, doi
:
10.17341/
ga
z
i
m
m
f
d.977688.
[
20]
S
.
S
.
G
a
i
kw
a
d,
S
.
S
.
R
um
m
a
,
a
nd
M
.
H
a
nga
r
ge
,
“
F
ungi
a
f
f
e
c
t
e
d
f
r
ui
t
l
e
a
f
d
i
s
e
a
s
e
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
de
e
p
C
N
N
a
r
c
hi
t
e
c
t
ur
e
,
”
I
nt
e
r
nat
i
onal
J
our
nal
of
I
nf
or
m
at
i
on T
e
c
hnol
ogy
, vol
. 14, no. 7, pp. 3815
–
3824, 2022, doi
:
10.1007/
s
41870
-
022
-
00860
-
w.
[
21]
V
.
S
una
nt
hi
ni
e
t
al
.
,
“
C
om
pa
r
i
s
on
of
C
N
N
a
l
gor
i
t
hm
s
f
or
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on
on
f
undus
i
m
a
ge
s
t
o
de
t
e
c
t
gl
a
uc
om
a
,”
J
our
nal
o
f
H
e
al
t
hc
ar
e
E
ngi
ne
e
r
i
ng
, vol
. 2022, 2022, doi
:
10.1155/
2022/
7873300.
[
22]
Z
. W
a
ng, W
. W
a
ng, Y
. Y
a
ng, Z
. H
a
n,
D
. X
u, a
nd C
.
S
u, “
C
N
N
-
a
nd G
A
N
-
ba
s
e
d c
l
a
s
s
i
f
i
c
a
t
i
on of
m
a
l
i
c
i
ous
c
ode
f
a
m
i
l
i
e
s
:
A
c
od
e
vi
s
ua
l
i
z
a
t
i
on
a
ppr
oa
c
h,”
I
nt
e
r
nat
i
onal
J
our
nal
of
I
nt
e
l
l
i
ge
nt
Sy
s
t
e
m
s
,
vol
.
37,
no.
12,
pp.
12472
–
12489,
2022,
doi
:
10.1002/
i
nt
.23094.
[
23]
L
.
I
.
K
e
s
um
a
,
E
r
m
a
t
i
t
a
,
a
nd
E
r
w
i
n,
“
E
L
R
E
I
:
E
ns
e
m
bl
e
l
e
a
r
ni
ng
of
R
e
s
N
e
t
,
E
f
f
i
c
i
e
nt
N
e
t
,
a
nd
I
nc
e
pt
i
on
-
v3
f
or
l
ung
di
s
e
a
s
e
c
l
a
s
s
i
f
i
c
a
t
i
on
ba
s
e
d
on
c
he
s
t
x
-
r
a
y
i
m
a
ge
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
I
nt
e
l
l
i
ge
nt
E
ngi
ne
e
r
i
ng
and
Sy
s
t
e
m
s
,
vol
.
16,
no.
5,
pp.
149
–
Evaluation Warning : The document was created with Spire.PDF for Python.