I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
14
,
No
.
2
,
A
p
r
il
20
25
,
p
p
.
1
4
9
6
~
1
5
0
6
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
14
.i
2
.
p
p
1
4
9
6
-
1
5
0
6
1496
J
o
ur
na
l
ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
H
y
brid mo
del
de
t
ection a
nd c
la
ss
ifi
ca
tion o
f
lung
can
cer
Ra
m
i Y
o
us
ef
1
,
E
m
a
n
Ya
s
er
Da
ra
g
hm
i
2
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
y
st
e
ms
En
g
i
n
e
e
r
i
n
g
,
P
a
l
e
s
t
i
n
e
Te
c
h
n
i
c
a
l
U
n
i
v
e
r
si
t
y
–
K
a
d
o
o
r
i
e
,
T
u
l
k
a
r
m
,
P
a
l
e
s
t
i
n
e
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
P
a
l
e
st
i
n
e
Te
c
h
n
i
c
a
l
U
n
i
v
e
r
si
t
y
–
K
a
d
o
o
r
i
e
,
T
u
l
k
a
r
m,
P
a
l
e
s
t
i
n
e
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
21
,
2
0
2
4
R
ev
is
ed
No
v
4
,
2
0
2
4
Acc
ep
ted
No
v
1
4
,
2
0
2
4
L
u
n
g
c
a
n
c
e
r
ra
n
k
s
a
m
o
n
g
t
h
e
m
o
st
p
re
v
a
len
t
m
a
li
g
n
a
n
c
ies
wo
rld
w
id
e
.
Early
d
e
tec
ti
o
n
is
p
iv
o
tal
to
imp
ro
v
in
g
trea
tme
n
t
o
u
tc
o
m
e
s
fo
r
v
a
rio
u
s
c
a
n
c
e
r
ty
p
e
s
.
Th
e
i
n
teg
ra
ti
o
n
of
a
rti
fici
a
l
in
telli
g
e
n
c
e
(AI)
in
t
o
ima
g
e
p
ro
c
e
ss
in
g
,
c
o
u
p
le
d
with
t
h
e
a
v
a
il
a
b
i
li
ty
of
c
o
m
p
re
h
e
n
si
v
e
h
ist
o
rica
l
lu
n
g
c
a
n
c
e
r
d
a
tas
e
ts,
p
ro
v
id
e
s
t
h
e
c
h
a
n
c
e
to
c
re
a
te
a
c
las
sific
a
ti
o
n
m
o
d
e
l
b
a
se
d
on
d
e
e
p
lea
rn
in
g
,
th
u
s
imp
r
o
v
i
n
g
th
e
p
re
c
isio
n
a
n
d
e
ffe
c
ti
v
e
n
e
ss
of
d
e
tec
ti
n
g
lu
n
g
c
a
n
c
e
r.
Th
is
not
o
n
l
y
a
id
s
lab
o
ra
to
ry
tea
m
s
but
a
lso
c
o
n
tri
b
u
tes
to
re
d
u
c
i
n
g
th
e
ti
m
e
to
d
iag
n
o
sis
a
n
d
a
ss
o
c
iate
d
c
o
sts.
Co
n
se
q
u
e
n
tl
y
,
e
a
rly
d
e
tec
ti
o
n
se
rv
e
s
to
c
o
n
se
rv
e
re
so
u
rc
e
s
a
n
d
,
m
o
re
si
g
n
i
fica
n
tl
y
,
h
u
m
a
n
l
i
v
e
s.
Th
is
stu
d
y
p
r
o
p
o
se
s
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
(CNN
)
m
o
d
e
ls
a
n
d
tran
sfe
r
lea
rn
in
g
-
b
a
se
d
a
rc
h
it
e
c
t
u
re
s,
i
n
c
lu
d
i
n
g
Re
sN
e
t5
0
,
VG
G
1
9
,
De
n
se
Ne
t1
6
9
,
a
n
d
I
n
c
e
p
ti
o
n
V3
,
f
o
r
l
u
n
g
c
a
n
c
e
r
c
las
sifica
ti
o
n
.
An
e
n
se
m
b
le
a
p
p
ro
a
c
h
is
u
se
d
to
e
n
h
a
n
c
e
o
v
e
ra
ll
c
a
n
c
e
r
d
e
tec
ti
o
n
p
e
rfo
rm
a
n
c
e
.
Th
e
p
ro
p
o
se
d
e
n
se
m
b
le m
o
d
e
l,
c
o
m
p
o
se
d
o
f
fi
v
e
e
ffe
c
ti
v
e
m
o
d
e
ls,
a
c
h
iev
e
s a
n
F
1
-
sc
o
re
o
f
9
7
.
7
7
%
a
n
d
a
n
a
c
c
u
ra
c
y
ra
te
o
f
9
7
.
5
%
o
n
th
e
IQ
-
OTH/NCCD
tes
t
d
a
tas
e
t.
Th
e
se
fin
d
in
g
s
h
ig
h
li
g
h
t
t
h
e
e
ff
e
c
ti
v
e
n
e
ss
a
n
d
d
e
p
e
n
d
a
b
i
li
ty
of
our
n
o
v
e
l
m
o
d
e
l
in
a
u
to
m
a
ti
n
g
th
e
c
las
sifi
c
a
ti
o
n
of
lu
n
g
c
a
n
c
e
r,
o
u
tp
e
rf
o
r
m
in
g
p
ri
o
r
re
se
a
rc
h
e
ffo
rts,
stre
a
m
li
n
in
g
d
iag
n
o
sis
p
ro
c
e
ss
e
s,
a
n
d
u
lt
ima
tely
c
o
n
tri
b
u
ti
n
g
to
th
e
p
re
se
rv
a
ti
o
n
of
p
a
ti
e
n
ts'
li
v
e
s.
K
ey
w
o
r
d
s
:
Ar
tific
ial
in
telleg
en
ce
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
E
n
s
em
b
le
lear
n
in
g
L
u
n
g
ca
n
ce
r
T
r
an
s
f
er
lear
n
i
n
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
R
am
i Y
o
u
s
e
f
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Sy
s
t
em
s
E
n
g
in
ee
r
in
g
,
Palest
in
e
T
ec
h
n
ical
Un
iv
er
s
ity
J
af
f
a
Stre
et,
T
u
lk
ar
m
,
Palest
in
e
E
m
ail: r
.
y
o
u
s
u
f
@
p
tu
k
.
ed
u
.
p
s
1.
I
NT
RO
D
UCT
I
O
N
C
an
ce
r
is
a
d
is
ea
s
e
th
at
m
an
i
f
ests
in
m
an
y
way
s
an
d
is
m
o
s
tly
lin
k
ed
to
ab
er
r
an
t
ce
ll
p
o
p
u
latio
n
s
.
T
h
ese
ca
n
ce
r
ce
lls
k
ee
p
d
iv
i
d
in
g
a
n
d
ex
p
a
n
d
in
g
to
b
ec
o
m
e
tu
m
o
r
s
.
L
u
n
g
c
an
ce
r
is
c
an
ce
r
th
at
p
o
s
es
th
e
g
r
ea
test
r
is
k
to
h
u
m
a
n
life
g
l
o
b
ally
.
As
p
er
t
h
e
W
o
r
ld
He
alth
Or
g
an
izatio
n
[
1
]
,
lu
n
g
ca
n
ce
r
is
th
e
lead
in
g
ca
u
s
e
of
m
o
r
tality
wo
r
l
d
wid
e.
In
2
0
0
8
,
lu
n
g
ca
n
ce
r
ac
co
u
n
t
ed
f
o
r
1
.
3
7
m
illi
o
n
d
e
ath
s
g
lo
b
ally
[
2
]
.
B
ased
on
th
e
av
ailab
le
d
ata,
lu
n
g
ca
n
ce
r
co
m
p
r
is
es
th
e
m
ajo
r
ity
of
n
ew
ca
n
ce
r
d
iag
n
o
s
es
wo
r
ld
wid
e,
with
1
,
3
5
0
,
0
0
0
n
ew
ca
s
es,
r
ep
r
esen
tin
g
1
2
.
4
%
of
all
n
ew
ca
n
ce
r
ca
s
es.
Ad
d
itio
n
ally
,
it
c
o
n
s
titu
tes
th
e
m
ajo
r
ity
of
ca
n
ce
r
-
r
elate
d
f
atalities,
with
1
,
1
8
0
,
0
0
0
d
ea
th
s
,
ac
co
u
n
tin
g
f
o
r
1
7
.
6
%
of
all
ca
n
ce
r
d
ea
th
s
[
3
]
.
L
u
n
g
ca
n
ce
r
r
an
k
ed
f
ir
s
t
am
o
n
g
ca
u
s
es
of
d
ea
th
f
o
r
m
en
a
n
d
th
ir
d
am
o
n
g
ca
u
s
es
of
d
ea
th
f
o
r
wo
m
en
in
th
e
Glo
b
al
C
an
ce
r
Ob
s
er
v
ato
r
y
d
atab
ase
cr
ea
ted
by
th
e
I
n
ter
n
atio
n
al
Ag
en
c
y
f
o
r
R
esear
ch
on
C
an
ce
r
(
I
AR
C
)
in
2018.
T
h
e
d
atab
ase
en
co
m
p
ass
ed
r
ates
of
b
o
th
i
n
cid
en
ce
an
d
m
o
r
tality
f
o
r
36
ca
n
ce
r
ty
p
es
ac
r
o
s
s
1
8
5
co
u
n
tr
ies.
Nea
r
l
y
1
.
8
m
illi
o
n
f
atalities
f
r
o
m
ca
n
ce
r
wer
e
r
ec
o
r
d
ed
in
2
0
1
8
,
ac
co
u
n
tin
g
f
o
r
a
b
o
u
t
1
8
.
4
%
of
all
ca
n
ce
r
-
r
elate
d
d
ea
th
s
[
4
]
.
Du
e
to
th
e
alar
m
in
g
in
cr
ea
s
e
in
lu
n
g
ca
n
ce
r
f
atal
ities
an
d
th
e
d
is
ea
s
e's
ex
ce
s
s
i
v
ely
h
ig
h
in
ci
d
en
ce
by
n
at
u
r
e,
m
an
y
s
tu
d
ies
f
o
c
u
s
in
g
on
ca
n
ce
r
co
n
tr
o
l
a
n
d
p
r
o
m
p
t
id
en
tific
atio
n
m
eth
o
d
s
h
av
e
em
e
r
g
ed
to
r
ed
u
ce
m
o
r
tality
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
d
etec
tio
n
a
n
d
c
la
s
s
ifica
tio
n
o
f lu
n
g
ca
n
ce
r
(
R
a
mi
Yo
u
s
ef
)
1497
T
h
e
p
o
ten
tial
of
a
s
u
cc
ess
f
u
l
cu
r
e
f
o
r
lu
n
g
ca
n
ce
r
is
c
o
n
ti
n
g
en
t
u
p
o
n
th
e
tim
ely
d
etec
tio
n
of
th
e
d
is
ea
s
e
an
d
th
e
ac
cu
r
ac
y
of
d
i
ag
n
o
s
tic
p
r
o
ce
d
u
r
es.
E
f
f
ec
tiv
e
d
iag
n
o
s
tic
m
eth
o
d
s
co
n
tr
ib
u
t
e
to
a
r
e
d
u
ctio
n
in
th
e
in
cid
en
ce
of
lu
n
g
ca
n
ce
r
s
,
an
d
th
e
ea
r
ly
id
en
tific
atio
n
of
th
e
ailm
en
t
is
f
u
n
d
am
e
n
t
ally
im
p
er
ativ
e
f
o
r
ac
h
iev
in
g
a
f
av
o
r
ab
le
p
r
o
g
n
o
s
is
in
lu
n
g
ca
n
ce
r
tr
ea
tm
en
t.
Pre
s
en
tly
,
th
er
e
ex
is
t
s
ev
en
m
o
d
alities
f
o
r
th
e
m
an
ag
em
en
t
of
lu
n
g
ca
n
ce
r
,
wh
ich
en
co
m
p
ass
cy
to
lo
g
y
s
p
u
tu
m
a
n
d
b
r
ea
th
an
aly
s
es,
p
o
s
itro
n
em
is
s
io
n
to
m
o
g
r
a
p
h
y
(
PET
)
,
m
ag
n
etic
r
eso
n
an
ce
im
ag
in
g
(
MRI)
,
an
d
ch
est
r
ad
io
g
r
ap
h
s
(
C
XR
s
)
[
2
]
.
It
is
n
o
tewo
r
th
y
th
at
C
X
R
s
,
s
p
u
tu
m
cy
to
lo
g
y
,
an
d
co
m
p
u
te
d
to
m
o
g
r
ap
h
y
(
C
T
)
s
ca
n
s
en
tail
ex
p
o
s
u
r
e
to
io
n
izin
g
r
ad
iatio
n
,
wh
ile
MRI
an
d
PET
s
ca
n
s
im
p
o
s
e
ce
r
tain
lim
itatio
n
s
on
th
e
p
r
ec
is
e
id
en
tific
atio
n
a
n
d
s
ta
g
in
g
of
lu
n
g
ca
n
ce
r
.
It
is
ess
en
tial
to
ac
k
n
o
wled
g
e
th
at
th
ese
d
iag
n
o
s
tic
tech
n
i
q
u
es
ar
e
not
with
o
u
t
th
eir
in
h
er
en
t
lim
itatio
n
s
.
Fu
r
th
er
m
o
r
e
,
it
is
ess
en
tial
to
co
n
s
id
er
th
at
th
e
ad
m
in
is
tr
atio
n
of
a
s
er
u
m
test
is
an
in
v
asiv
e
m
ed
ical
p
r
o
ce
d
u
r
e,
a
n
d
its
lim
ited
ca
p
ac
ity
f
o
r
ea
r
ly
d
etec
tio
n
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
r
e
n
d
er
s
it
u
n
s
u
itab
le
as
a
p
r
im
ar
y
d
iag
n
o
s
tic
to
o
l
[
5
]
.
C
o
n
v
er
s
ely
,
th
e
ass
ess
m
en
t
of
s
p
u
tu
m
n
ec
ess
itates
ad
d
itio
n
al
in
v
esti
g
atio
n
d
u
e
to
th
e
p
r
esen
ce
of
g
en
e
p
r
o
m
o
ter
m
eth
y
latio
n
,
as
in
d
icate
d
by
a
s
tu
d
y
on
www.
ieee
c.
ir
[
5
]
.
D
esp
ite
th
is
n
ee
d
f
o
r
ad
d
itio
n
al
s
cr
u
tin
y
,
s
p
u
tu
m
an
aly
s
is
s
h
o
ws
p
o
ten
tial
to
f
ac
ilit
ate
tim
ely
id
en
tific
atio
n
of
lu
n
g
ca
n
ce
r
.
Ad
d
itio
n
ally
,
v
o
latile
o
r
g
an
ic
co
m
p
o
u
n
d
s
(
VOCs
)
d
etec
ted
in
u
r
in
e
h
av
e
d
e
m
o
n
s
tr
ated
n
o
t
ewo
r
th
y
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
,
alth
o
u
g
h
a
la
r
g
er
s
am
p
le
s
ize
is
n
ec
ess
ar
y
f
o
r
m
o
r
e
r
o
b
u
s
t
r
esu
lts
[
5
]
.
C
o
n
v
er
s
ely
,
ch
est
X
-
r
ay
s
(
C
XR
)
ex
h
ib
it
r
elativ
ely
lo
w
s
en
s
itiv
ity
an
d
a
r
e
p
r
o
n
e
to
p
r
o
d
u
cin
g
f
alse
-
n
eg
ativ
e
o
u
tco
m
es,
as
r
ep
o
r
ted
in
p
r
ev
i
o
u
s
s
tu
d
ies
[
6
]
,
[
7
]
.
Pre
s
en
tly
,
th
e
m
o
s
t
d
e
p
en
d
ab
le
ap
p
r
o
ac
h
to
d
etec
tin
g
lu
n
g
ca
n
ce
r
is
th
e
u
tili
za
tio
n
of
CT
im
a
g
in
g
.
T
h
is
im
ag
in
g
m
o
d
ality
o
f
f
er
s
p
r
ec
is
e
in
f
o
r
m
atio
n
r
eg
ar
d
in
g
t
h
e
lo
ca
tio
n
an
d
s
ize
of
p
u
lm
o
n
ar
y
n
o
d
u
les,
en
ab
lin
g
th
e
ea
r
ly
d
etec
tio
n
of
ca
n
ce
r
o
u
s
g
r
o
wth
s
.
L
o
w
-
d
o
s
e
CT
s
c
r
ee
n
in
g
h
as
p
r
o
v
en
ef
f
ec
tiv
e
in
id
en
tif
y
in
g
ea
r
ly
-
s
tag
e
ca
n
ce
r
t
u
m
o
r
s
,
r
esu
ltin
g
in
a
n
o
ta
b
le
2
0
.
0
%
r
e
d
u
ctio
n
in
m
o
r
tality
wh
e
n
co
m
p
ar
ed
to
co
n
v
en
tio
n
al
r
ad
i
o
g
r
ap
h
ic
tech
n
iq
u
es,
an
d
an
in
cr
ea
s
ed
r
ate
of
p
o
s
itiv
e
s
cr
ee
n
in
g
r
esu
lts
[
8
]
.
Deep
lear
n
in
g
is
a
s
p
ec
ialized
b
r
an
ch
of
m
ac
h
i
n
e
lear
n
in
g
(
ML
)
,
wh
ich
its
elf
f
alls
with
in
th
e
lar
g
er
f
ield
of
ar
tific
ial
in
tellig
en
ce
(
AI
)
.
T
h
e
o
v
er
ar
ch
i
n
g
o
b
jecti
v
e
of
AI
is
to
f
u
r
n
is
h
a
co
lle
ctio
n
of
alg
o
r
ith
m
s
an
d
m
eth
o
d
o
l
o
g
ies
d
esig
n
ed
to
ad
d
r
ess
p
r
o
b
lem
s
t
h
at
h
u
m
a
n
s
ef
f
o
r
tles
s
ly
an
d
in
tu
itiv
ely
u
n
d
er
tak
e
but
p
o
s
e
s
ig
n
if
ican
t
co
m
p
u
tatio
n
al
ch
al
len
g
es.
ML
,
as
a
d
is
c
i
p
li
n
e
,
is
h
a
r
n
e
s
s
e
d
f
o
r
t
h
e
p
u
r
p
o
s
e
of
p
a
t
t
e
r
n
r
e
c
o
g
n
i
t
i
o
n
,
w
i
t
h
d
e
e
p
le
a
r
n
i
n
g
c
o
m
p
r
i
s
i
n
g
a
c
a
t
e
g
o
r
y
of
ML
al
g
o
r
i
t
h
m
s
c
o
n
c
e
i
v
e
d
by
d
r
a
wi
n
g
i
n
s
p
i
r
a
t
i
o
n
f
r
o
m
t
h
e
s
t
r
u
c
t
u
r
a
l
a
n
d
o
p
e
r
a
t
i
o
n
a
l
p
r
in
c
i
p
l
e
s
of
t
h
e
h
u
m
a
n
b
r
a
i
n
.
Deep
l
e
a
r
n
i
n
g
e
n
d
e
a
v
o
r
s
to
e
m
u
l
a
t
e
t
h
e
h
u
m
a
n
p
e
r
c
e
p
t
u
a
l
p
r
o
c
e
s
s
by
e
s
t
a
b
l
is
h
i
n
g
a
r
t
i
f
i
c
ia
l
n
e
u
r
o
n
s
or
n
o
d
e
s
w
i
t
h
i
n
l
a
y
e
r
e
d
a
r
c
h
i
t
e
ct
u
r
e
s
,
wh
i
c
h
a
r
e
c
a
p
a
b
l
e
of
f
e
a
t
u
r
e
e
x
t
r
a
c
ti
o
n
f
r
o
m
o
b
j
e
c
ts
.
T
h
i
s
i
m
p
l
i
es
t
h
at
w
h
e
n
a
p
p
l
ied
to
i
m
a
g
e
c
la
s
s
i
f
ic
a
t
i
o
n
ta
s
k
s
,
d
e
e
p
l
e
a
r
n
i
n
g
a
i
m
s
to
d
i
s
c
e
r
n
p
a
tt
e
r
n
s
f
r
o
m
a
s
e
t
of
i
m
a
g
e
s
f
o
r
t
h
e
p
u
r
p
o
s
e
of
d
i
s
t
i
n
g
u
is
h
i
n
g
b
e
t
w
ee
n
d
i
v
e
r
s
e
c
l
a
s
s
es
or
o
b
j
e
c
ts
.
S
i
g
n
i
f
i
c
a
n
t
l
y
,
t
h
e
n
e
u
r
a
l
n
et
w
o
r
k
'
s
t
r
a
i
n
i
n
g
p
r
o
c
es
s
i
n
v
o
l
v
es
th
e
a
u
t
o
m
a
t
i
c
e
x
t
r
ac
t
i
o
n
of
i
m
ag
e
f
e
a
t
u
r
e
s
[
9
]
.
In
th
e
m
e
d
ical
d
o
m
ai
n
,
s
p
ec
if
ically
with
in
th
e
co
n
te
x
t
of
lu
n
g
ca
n
ce
r
d
iag
n
o
s
is
,
th
e
p
r
in
cip
al
d
iag
n
o
s
tic
tech
n
iq
u
e
r
elies
on
th
e
e
x
am
in
atio
n
of
tis
s
u
e
s
am
p
les.
Ho
wev
er
,
it
is
wo
r
t
h
ac
k
n
o
wled
g
i
n
g
t
h
at
th
is
d
iag
n
o
s
tic
p
r
o
ce
d
u
r
e
en
ta
ils
a
tim
e
-
co
n
s
u
m
in
g
p
r
o
ce
s
s
.
Utilizin
g
an
ar
r
ay
of
d
eep
le
ar
n
in
g
m
o
d
els
an
d
tr
an
s
f
er
lear
n
in
g
-
b
ased
m
o
d
els,
in
clu
d
in
g
R
esNet5
0
,
E
f
f
icien
tNetB
7
,
Den
s
eNe
t1
6
9
,
VGG1
6
,
VGG1
9
,
Xce
p
tio
n
,
a
n
d
I
n
ce
p
tio
n
V3
,
a
p
p
lied
to
th
e
lu
n
g
ca
n
ce
r
d
ata
s
et
IQ
-
OT
H/NC
C
D,
we
in
tr
o
d
u
ce
a
d
eep
lear
n
i
n
g
m
eth
o
d
o
l
o
g
y
in
th
is
r
esear
ch
to
tr
ain
a
n
d
test
th
e
m
o
d
els
a
t:
h
ttp
s
://www.
k
ag
g
le.
co
m
/co
d
e/k
er
n
eler
/s
tar
ter
-
th
e
-
iq
-
o
th
-
n
cc
d
-
lu
n
g
-
ca
n
ce
r
-
0
9
c3
a8
c9
-
4
/d
ata
.
2.
R
E
L
AT
E
D
WO
RK
T
h
e
id
e
n
tific
atio
n
of
p
u
lm
o
n
a
r
y
ir
r
e
g
u
lar
ities
co
n
s
titu
tes
a
s
u
b
s
tan
tial
h
az
ar
d
to
h
u
m
an
well
-
b
ein
g
,
an
d
th
e
tim
ely
r
ec
o
g
n
itio
n
th
er
eo
f
ass
u
m
es
a
p
iv
o
tal
r
o
le
in
r
is
k
m
itig
atio
n
.
T
im
ely
d
iag
n
o
s
is
f
ac
ilit
ates
ex
p
ed
itio
u
s
an
d
ef
f
icac
io
u
s
in
ter
v
en
tio
n
,
th
er
e
b
y
r
e
d
u
cin
g
p
o
ten
tial
co
m
p
licatio
n
s
an
d
en
h
an
cin
g
p
atien
t
o
u
tco
m
es.
Am
o
n
g
th
e
d
iag
n
o
s
tic
m
o
d
es,
CT
em
er
g
es
as
a
n
o
tewo
r
th
y
to
o
l
f
o
r
d
et
ec
tin
g
p
u
lm
o
n
ar
y
ab
n
o
r
m
alities
.
Nev
er
th
eless
,
th
e
in
ter
p
r
etatio
n
of
lu
n
g
CT
s
ca
n
s
p
r
esen
ts
ch
allen
g
es,
ev
en
f
o
r
s
ea
s
o
n
ed
r
ad
io
lo
g
is
ts
[
1
0
]
.
Ov
er
th
e
p
ast
f
ew
y
ea
r
s
,
in
v
esti
g
ato
r
s
h
av
e
d
elv
ed
i
n
to
th
e
ap
p
licatio
n
of
d
ee
p
lear
n
in
g
m
eth
o
d
o
l
o
g
ies
to
au
to
m
ate
th
e
d
iag
n
o
s
tic
p
r
o
ce
s
s
f
o
r
p
u
lm
o
n
ar
y
ir
r
e
g
u
lar
ities
,
aim
in
g
to
en
h
an
ce
d
iag
n
o
s
tic
p
r
ec
is
io
n
an
d
p
o
ten
tially
s
av
e
liv
es.
As
an
illu
s
tr
atio
n
,
Asu
n
th
a
an
d
Srin
iv
asan
[
1
1
]
in
tr
o
d
u
ce
d
a
p
io
n
ee
r
i
n
g
ap
p
r
o
ac
h
ter
m
ed
f
ast
an
d
p
o
wer
-
ef
f
icie
n
t
s
y
s
tem
-
on
-
c
h
ip
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
wo
r
k
(
FP
SOC
NN)
,
s
h
o
wca
s
in
g
th
e
o
n
g
o
in
g
ex
p
l
o
r
atio
n
of
in
n
o
v
ativ
e
m
eth
o
d
o
lo
g
ies
in
th
is
d
o
m
ain
.
T
h
e
FP
SOC
NN
d
ev
elo
p
ed
by
[
1
1
]
s
ee
k
s
to
allev
iate
th
e
co
m
p
u
tatio
n
al
in
tr
icac
ies
in
h
e
r
en
t
in
co
n
v
en
tio
n
al
C
NNs.
In
th
eir
s
tu
d
y
,
th
ey
ex
am
in
ed
v
ar
io
u
s
f
ea
tu
r
e
ex
t
r
ac
tio
n
tech
n
i
q
u
es,
in
cl
u
d
in
g
Z
er
n
ik
e
m
o
m
en
t,
h
is
to
g
r
am
of
o
r
ie
n
ted
g
r
ad
ien
ts
(
Ho
G)
,
wav
elet
tr
an
s
f
o
r
m
-
b
as
ed
f
ea
tu
r
es,
lo
ca
l
b
in
ar
y
p
atte
r
n
(
L
B
P),
wav
elet
tr
an
s
f
o
r
m
-
b
ased
f
ea
tu
r
es,
an
d
s
ca
le
in
v
ar
ian
t
f
ea
tu
r
e
tr
an
s
f
o
r
m
(
SIFT
)
.
T
h
e
p
r
o
p
o
s
ed
FP
SO
C
NN
m
eth
o
d
o
lo
g
y
not
o
n
ly
d
em
o
n
s
tr
ated
o
u
ts
tan
d
in
g
ac
h
ie
v
em
en
t
but
also
ef
f
ec
tiv
ely
ad
d
r
ess
ed
th
e
co
m
p
u
tatio
n
al
co
m
p
lex
iti
es
ass
o
ciate
d
wi
th
tr
ad
itio
n
al
C
NNs.
In
a
s
ep
ar
ate
s
tu
d
y
by
[
1
2
]
a
m
u
lti
-
p
a
th
C
NN
was
in
tr
o
d
u
ce
d
,
le
v
er
ag
in
g
c
o
n
tex
tu
al
f
ea
tu
r
es,
b
o
th
lo
ca
l
an
d
b
r
o
a
d
,
to
id
en
tify
l
u
n
g
ca
n
ce
r
au
t
o
n
o
m
o
u
s
ly
.
By
in
teg
r
atin
g
th
is
m
eth
o
d
,
th
ei
r
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
2
,
Ap
r
il
20
25
:
1
4
9
6
-
1
5
0
6
1498
d
em
o
n
s
tr
ates
in
cr
ea
s
ed
ad
ap
ta
b
ilit
y
in
h
an
d
lin
g
v
ar
iatio
n
s
in
n
o
d
u
le
s
ize
an
d
s
h
ap
e.
T
h
is
ch
ar
ac
ter
is
tic
lead
s
to
en
h
an
c
ed
d
etec
tio
n
r
esu
lt
s
co
m
p
ar
ed
to
m
o
d
er
n
s
tate
-
of
-
th
e
-
ar
t
tech
n
i
q
u
es
[
1
2
]
.
A
co
m
p
r
eh
en
s
iv
e
d
em
o
n
s
tr
atio
n
of
th
e
u
s
e
of
c
o
m
p
u
ter
-
aid
ed
d
ia
g
n
o
s
is
(
C
AD)
m
eth
o
d
s
f
o
r
t
h
e
id
en
tific
a
tio
n
of
ea
r
ly
-
s
tag
e
lu
n
g
ca
n
ce
r
is
p
r
esen
ted
in
th
e
s
tu
d
y
by
[
1
3
]
.
C
o
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs),
a
m
o
n
g
v
ar
i
o
u
s
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es,
h
av
e
b
e
en
ex
ten
s
iv
ely
ap
p
lied
to
th
e
task
s
in
v
o
lv
i
n
g
co
m
p
u
ter
v
i
s
io
n
.
T
h
e
au
th
o
r
s
em
p
h
asize
th
e
s
u
p
er
io
r
ity
of
3
-
d
im
en
s
io
n
al
C
NNs
o
v
er
2
-
d
im
en
s
io
n
al
C
NNs
f
o
r
im
p
r
o
v
ed
ef
f
ec
tiv
en
ess
in
d
etec
tin
g
lu
n
g
ca
n
ce
r
.
In
c
o
n
tr
ast,
Sh
y
n
i
a
n
d
C
h
itra
[
1
4
]
h
ig
h
lig
h
t
th
e
wid
esp
r
ea
d
u
s
e
of
C
NN
as
th
e
p
r
im
a
r
y
d
eep
lear
n
i
n
g
alg
o
r
ith
m
f
o
r
d
etec
tin
g
C
OVI
D
-
19
f
r
o
m
m
e
d
ical
im
ag
es.
T
h
ese
ar
ticles
n
o
t
o
n
ly
p
r
o
m
o
te
th
e
ex
ten
s
iv
e
ad
o
p
tio
n
of
C
NN
but
also
p
r
o
v
id
e
v
alu
ab
le
in
s
ig
h
ts
,
in
s
p
ir
in
g
em
er
g
in
g
r
e
s
ea
r
ch
er
s
to
cr
ea
te
h
ig
h
ly
ef
f
ec
tiv
e
C
NN
m
o
d
els
th
at
u
s
e
m
ed
ical
im
ag
es
to
d
etec
t
d
is
ea
s
es
ea
r
ly
.
In
a
d
is
tin
ct
in
v
esti
g
atio
n
,
R
ah
m
an
et
a
l.
[
1
5
]
co
n
d
u
cted
a
s
tu
d
y
u
tili
zin
g
C
NN
f
o
r
task
s
r
elate
d
to
class
if
icatio
n
th
at
in
v
o
lv
e
two
,
th
r
ee
,
an
d
m
u
ltip
le
class
es.
T
h
ey
em
p
lo
y
ed
elec
tr
o
ca
r
d
io
g
r
am
(
E
C
G)
s
ig
n
als
as
in
p
u
t,
ac
h
iev
in
g
p
r
o
m
is
in
g
o
u
tco
m
es
in
th
eir
ex
p
er
im
en
ts
.
No
tab
ly
,
th
e
y
u
s
ed
th
e
Gr
ad
-
C
AM
m
eth
o
d
to
p
i
n
p
o
in
t
c
r
i
tical
an
d
p
ar
ticu
lar
ar
ea
s
in
th
e
in
p
u
t
s
ig
n
als,
th
er
eb
y
f
ac
ilit
atin
g
in
f
o
r
m
e
d
d
ec
is
io
n
-
m
ak
in
g
d
u
r
in
g
th
e
cl
ass
if
icatio
n
p
r
o
ce
s
s
[
1
5
]
.
Gif
an
i
et
a
l.
[
1
6
]
c
o
n
d
u
cted
r
esear
ch
wh
er
ein
th
ey
d
e
v
is
ed
an
ass
em
b
lag
e
d
eep
l
ea
r
n
in
g
m
o
d
el
f
o
r
th
e
au
to
m
ated
id
en
tific
atio
n
of
C
OVI
D
-
19
f
r
o
m
CT
s
ca
n
s
.
E
m
p
lo
y
in
g
C
NN
alo
n
g
s
id
e
tr
an
s
f
er
lear
n
in
g
tech
n
iq
u
es,
th
e
r
esear
ch
e
r
s
in
teg
r
ated
15
p
r
e
-
tr
ain
e
d
C
NN
m
o
d
els,
lev
er
a
g
in
g
th
e
c
o
llectiv
e
ex
p
er
tis
e
a
n
d
ca
p
ab
ilit
ies.
T
h
e
e
n
s
em
b
le
a
cc
ess
r
esu
lted
in
e
n
h
an
ce
d
r
o
b
u
s
tn
ess
an
d
ac
cu
r
ac
y
in
t
h
e
id
en
tific
atio
n
of
C
OVI
D
-
19
f
r
o
m
CT
s
ca
n
s
.
In
th
eir
s
tu
d
y
,
Do
r
j
et
a
l.
[
1
7
]
co
n
ce
n
tr
ated
on
l
u
n
g
c
an
ce
r
class
if
icatio
n
u
s
in
g
er
r
o
r
-
c
o
r
r
ec
tin
g
o
u
tp
u
t
co
d
es
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
E
C
OC
SV
M)
an
d
d
ee
p
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
DC
NN)
.
An
E
C
OC
SVM
cl
ass
if
ier
was
em
p
lo
y
ed
f
o
r
ca
teg
o
r
izin
g
s
ev
er
al
ty
p
es
of
lu
n
g
ca
n
ce
r
,
with
t
h
e
alg
o
r
ith
m
ap
p
lied
to
a
d
at
aset
co
m
p
r
is
in
g
3
7
5
3
im
a
g
es
r
ep
r
esen
tin
g
f
o
u
r
lu
n
g
ca
n
ce
r
ty
p
es.
T
h
e
im
p
lem
en
tatio
n
y
iel
d
ed
n
o
tab
le
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
a
n
d
s
p
ec
if
icity
,
with
p
ea
k
v
alu
es
r
ep
o
r
ted
as
9
4
.
1
7
%
f
o
r
s
q
u
am
o
u
s
ce
ll
ca
r
cin
o
m
a,
9
8
.
9
%
f
o
r
ac
ti
n
ic
k
e
r
ato
s
is
,
an
d
,
9
5
.
1
%
f
o
r
s
q
u
am
o
u
s
ce
ll
ca
r
cin
o
m
a
,
r
esp
ec
tiv
ely
.
A
DC
NN
m
o
d
el
was
p
r
o
p
o
s
ed
by
[
1
8
]
,
e
m
p
lo
y
in
g
a
d
eep
lear
n
in
g
a
p
p
r
o
ac
h
f
o
r
p
r
ec
is
e
class
if
icatio
n
b
etwe
en
b
en
ig
n
an
d
m
alig
n
an
t
lu
n
g
lesi
o
n
s
.
W
ith
a
test
in
g
ac
cu
r
ac
y
of
9
1
.
9
3
%
an
d
a
tr
ai
n
in
g
ac
cu
r
ac
y
of
9
3
.
1
6
%,
t
h
e
ev
al
u
atio
n
of
th
e
HAM
1
0
0
0
0
d
at
aset
r
ev
ea
led
r
em
ar
k
ab
le
r
es
u
lts
.
T
h
ese
r
esu
lts
h
ig
h
lig
h
t
how
well
th
e
DC
NN
m
o
d
el
d
is
tin
g
u
is
h
es
b
e
n
ig
n
f
r
o
m
m
alig
n
a
n
t
lu
n
g
lesi
o
n
s
.
P
a
l
et
a
l
.
[
1
9
]
c
r
e
a
t
e
d
an
i
n
t
e
r
p
r
e
t
a
b
l
e
ML
m
o
d
e
l
f
o
r
l
u
n
g
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
c
al
l
e
d
AI
C
AD
.
I
n
t
e
g
r
a
t
i
n
g
e
x
p
l
a
i
n
a
b
l
e
a
r
ti
f
i
c
ia
l
i
n
te
l
l
i
g
en
c
e
(
X
A
I
)
m
e
c
h
a
n
is
m
s
,
t
h
e
m
o
d
e
l
f
u
r
n
i
s
h
e
s
c
o
m
p
r
e
h
e
n
s
iv
e
e
x
p
l
a
n
a
t
i
o
n
s
f
o
r
c
r
u
c
i
a
l
f
e
a
t
u
r
es
i
d
e
n
t
i
f
i
e
d
by
th
e
A
I
/
M
L
a
l
g
o
r
i
t
h
m
s
.
E
n
c
o
u
r
ag
i
n
g
c
o
n
f
i
d
e
n
c
e
in
its
a
p
p
l
i
c
a
tio
n
,
t
h
e
m
o
d
e
l
of
AI
C
A
D
d
e
m
o
n
s
t
r
a
te
d
e
x
c
e
l
le
n
t
in
t
e
r
p
r
e
t
a
t
i
o
n
in
d
e
te
c
t
i
n
g
l
u
n
g
c
a
n
c
e
r
.
T
h
e
m
o
d
e
l
s
w
e
r
e
c
o
n
s
tr
u
c
t
e
d
u
s
i
n
g
d
i
v
e
r
s
e
a
l
g
o
r
i
t
h
m
s
,
i
n
cl
u
d
i
n
g
k
-
n
e
a
r
e
s
t
n
e
i
g
h
b
o
r
s
(
K
N
N
)
,
s
u
p
p
o
r
t
v
e
c
t
o
r
m
a
c
h
i
n
e
(
S
V
M
)
,
g
r
a
d
i
e
n
t
b
o
o
s
t
i
n
g
m
a
c
h
i
n
e
(
G
B
M
)
,
X
GB
o
o
s
t
,
r
a
n
d
o
m
f
o
r
e
s
t
c
l
ass
i
f
i
e
r
(
R
FC
)
,
a
n
d
f
e
e
d
-
f
o
r
w
a
r
d
a
r
c
h
i
t
e
c
t
u
r
es
.
T
h
e
i
n
t
e
r
p
r
e
t
a
b
il
i
t
y
of
m
o
d
e
l
s
w
i
t
h
s
u
p
e
r
i
o
r
p
e
r
f
o
r
m
a
n
c
e
in
n
e
u
r
a
l
n
e
t
w
o
r
k
s
was
h
i
g
h
l
i
g
h
t
e
d
t
h
r
o
u
g
h
XAI
o
u
t
p
u
t
s
,
r
e
v
e
a
l
i
n
g
t
h
a
t
t
h
es
e
m
o
d
e
l
s
p
r
i
o
r
i
ti
z
e
d
a
c
o
n
s
i
s
te
n
t
s
e
t
of
i
n
p
u
t
f
e
at
u
r
e
s
wi
t
h
e
l
e
v
ate
d
i
m
p
o
r
t
a
n
c
e
.
An
in
ter
p
r
etab
le
s
y
s
tem
f
o
r
d
iag
n
o
s
in
g
l
u
n
g
ca
n
ce
r
was
cr
ea
ted
in
th
e
wo
r
k
of
[
2
0
]
u
tili
zin
g
n
u
m
er
o
u
s
m
o
d
els
of
ML
,
in
clu
d
in
g
n
aiv
e
B
ay
es
class
if
ier
,
lo
g
is
tic
r
eg
r
ess
io
n
,
d
ec
is
io
n
tr
ee
,
a
n
d
r
an
d
o
m
f
o
r
est
.
B
ased
on
a
d
ataset
of
lu
n
g
ca
n
ce
r
ca
s
es,
th
e
a
n
aly
s
is
y
ield
ed
a
97%
ac
cu
r
ac
y
r
ate.
Ho
wev
er
,
it
is
n
o
tewo
r
th
y
th
at
th
e
v
alid
atio
n
of
th
e
m
o
d
el
was
r
estricte
d
to
a
C
SV
f
ile,
lack
in
g
im
ag
e
v
al
id
atio
n
.
In
o
r
d
e
r
to
in
cr
ea
s
e
in
ter
p
r
eta
b
ilit
y
,
XAI
tech
n
iq
u
es
lik
e
lim
e
a
n
d
SHAP
wer
e
u
s
ed
.
B
h
an
d
ar
i
et
a
l
.
[
2
1
]
in
t
r
o
d
u
ce
d
a
d
ee
p
lear
n
in
g
m
o
d
el
th
at
was
v
alid
ated
u
s
in
g
a
d
ataset
to
p
r
ed
ict
f
o
u
r
ca
te
g
o
r
ies
co
m
p
r
is
in
g
v
is
u
als
of
7
1
3
2
C
XR
.
T
h
e
m
o
d
el,
lev
e
r
ag
in
g
Gr
ad
-
C
AM
,
SHAP,
an
d
L
I
ME
m
eth
o
d
s
f
o
r
in
ter
p
r
etati
o
n
th
r
o
u
g
h
10
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
,
ac
h
iev
e
d
an
av
er
ag
e
of
9
4
.
5
4
%
(
±
1
.
3
3
%)
v
alid
atio
n
ac
cu
r
ac
y
an
d
9
4
.
3
1
%
(
±
1
.
0
1
%)
of
test
ac
cu
r
ac
y
.
To
d
eter
m
in
e
b
io
m
a
r
k
er
s
in
non
-
s
m
all
-
ce
ll
lu
n
g
c
ar
cin
o
m
a
(
NSC
L
C
)
s
u
b
ty
p
es,
Dwiv
ed
i
et
a
l.
[
2
2
]
s
u
g
g
ested
a
f
r
a
m
ewo
r
k
of
XAI
-
b
ased
d
eep
lear
n
in
g
.
T
h
e
f
r
am
ewo
r
k
i
n
co
r
p
o
r
ated
an
au
to
en
co
d
er
,
a
b
io
m
ar
k
e
r
d
is
co
v
er
y
m
o
d
u
le,
an
d
a
class
if
icatio
n
n
eu
r
al
n
etwo
r
k
.
52
r
elate
d
b
io
m
ar
k
er
s
wer
e
d
is
co
v
er
ed
u
s
in
g
XAI
tech
n
iq
u
es;
of
th
e
s
e,
14
wer
e
d
r
u
g
g
ab
le
an
d
28
wer
e
s
u
r
v
iv
al
p
r
ed
ictiv
e.
9
5
.
7
4
%
ac
cu
r
a
cy
in
NSC
L
C
s
u
b
ty
p
e
class
if
icatio
n
was
attain
ed
with
th
e
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
.
We
r
ec
o
m
m
en
d
u
tili
zin
g
C
NN
m
o
d
els
in
our
ap
p
r
o
ac
h
an
d
s
ev
er
al
tr
an
s
f
er
lear
n
in
g
-
b
as
ed
ar
ch
itectu
r
es,
in
clu
d
in
g
R
esNet5
0
,
VGG1
9
,
Den
s
eNe
t1
6
9
,
an
d
I
n
ce
p
tio
n
V3
,
f
o
r
lu
n
g
ca
n
ce
r
class
if
ica
tio
n
task
s
.
Ad
d
itio
n
ally
,
we
e
m
p
lo
y
an
en
s
em
b
l
e
ap
p
r
o
ac
h
by
co
m
b
i
n
in
g
v
ar
io
u
s
m
o
d
el
co
m
b
i
n
atio
n
s
to
en
h
a
n
ce
th
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
of
ca
n
ce
r
d
etec
tio
n
.
3.
DATAS
E
T
T
o
ass
ess
th
e
ef
f
ec
tiv
en
ess
of
our
p
r
o
p
o
s
ed
m
o
d
el,
we
u
t
ilized
th
e
IQ
-
OT
H/NC
C
D
lu
n
g
ca
n
ce
r
d
ataset
[
2
3
]
wh
ich
c
o
m
p
r
is
es
2073
CT
im
ag
es
f
r
o
m
1
1
0
p
atien
ts
,
in
clu
d
in
g
b
o
th
th
o
s
e
in
g
o
o
d
h
ea
lth
an
d
th
o
s
e
wh
o
h
av
e
b
ee
n
d
ia
g
n
o
s
ed
with
lu
n
g
ca
n
ce
r
.
T
h
is
d
at
aset
was
g
ath
er
ed
at
th
e
I
r
a
q
-
On
co
lo
g
y
T
ea
ch
in
g
Ho
s
p
ital/Natio
n
al
C
en
ter
f
o
r
C
an
ce
r
Dis
ea
s
es
d
u
r
in
g
th
r
ee
m
o
n
th
s
in
2
0
1
9
.
A
Siem
en
s
s
ca
n
n
er
in
DI
C
OM
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
d
etec
tio
n
a
n
d
c
la
s
s
ifica
tio
n
o
f lu
n
g
ca
n
ce
r
(
R
a
mi
Yo
u
s
ef
)
1499
f
o
r
m
at
was
u
s
ed
to
o
b
tain
t
h
e
CT
s
ca
n
s
an
d
each
s
ca
n
co
n
t
ain
ed
80
–
2
0
0
s
lices
with
a
1
m
m
s
lice
th
ick
n
ess
.
In
o
r
d
er
to
g
u
ar
an
tee
d
iv
er
s
it
y
,
th
e
d
ataset
was
co
llected
f
r
o
m
d
i
v
er
s
e
r
eg
i
o
n
s
in
I
r
aq
,
wh
ich
r
ep
r
esen
t
a
r
an
g
e
of
d
em
o
g
r
a
p
h
ics.
T
h
e
s
tu
d
y
was
ap
p
r
o
v
ed
et
h
ically
by
th
e
in
s
titu
tio
n
al
r
ev
iew
b
o
ar
d
,
e
n
s
u
r
in
g
th
e
r
ig
h
ts
an
d
p
r
iv
ac
y
of
p
ar
ticip
a
n
ts
ar
e
u
p
h
el
d
d
u
r
in
g
th
e
p
r
o
c
ess
of
co
llectin
g
d
ata.
4.
M
E
T
H
O
D
T
h
e
t
h
o
r
o
u
g
h
m
e
t
h
o
d
o
l
o
g
y
e
m
p
l
o
y
e
d
in
t
h
i
s
s
t
u
d
y
is
d
e
s
c
r
i
b
e
d
in
t
h
i
s
s
e
c
t
i
o
n
.
F
i
g
u
r
e
1
s
h
o
w
s
t
h
e
s
c
h
e
m
a
t
i
c
w
o
r
k
f
l
o
w
of
o
u
r
p
r
o
c
e
d
u
r
e
,
g
i
v
i
n
g
d
e
t
a
i
l
e
d
s
t
e
p
s
of
t
h
e
e
x
p
e
r
i
m
e
n
t
a
l
p
r
o
c
e
d
u
r
e
.
S
e
v
e
r
a
l
p
r
e
p
r
o
c
e
s
s
i
n
g
t
e
c
h
n
i
q
u
e
s
a
r
e
c
o
m
p
a
r
e
d
to
s
t
r
u
c
t
u
r
a
l
d
a
t
a
a
n
d
t
a
k
e
n
i
n
t
o
c
o
n
s
i
d
e
r
a
t
i
o
n
w
h
e
n
d
i
s
c
u
s
s
i
n
g
i
m
a
g
e
d
a
t
a
.
By
s
c
a
l
i
n
g
d
a
t
a
a
n
d
r
e
d
u
c
i
n
g
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
m
p
l
e
x
i
t
y
,
t
h
e
s
e
t
e
c
h
n
i
q
u
e
s
h
e
l
p
to
d
e
c
r
e
a
s
e
m
o
d
e
l
i
n
g
c
o
s
t
s
w
h
i
l
e
i
n
c
r
e
a
s
i
n
g
t
h
e
q
u
a
n
t
i
t
y
or
q
u
a
l
i
t
y
of
t
h
e
d
a
t
a
s
e
t
.
S
i
n
c
e
t
h
e
u
s
e
of
d
e
e
p
l
e
a
r
n
i
n
g
h
a
s
i
n
c
r
e
a
s
e
d
,
s
c
i
e
n
t
i
s
t
s
h
a
v
e
w
o
r
k
e
d
on
t
h
i
s
p
r
o
b
l
e
m
,
a
n
d
n
o
w
a
d
a
y
s
t
h
e
r
e
a
r
e
a
v
a
r
i
e
t
y
of
a
p
p
r
o
a
c
h
e
s
to
d
e
a
l
i
n
g
w
i
t
h
t
h
i
s
a
s
p
e
c
t
.
P
r
e
p
r
o
c
e
s
s
i
n
g
p
r
o
c
e
d
u
r
e
s
a
r
e
u
s
e
d
in
t
h
i
s
w
o
r
k
to
g
e
t
t
h
e
d
a
t
a
in
F
i
g
u
r
e
s
2
(
a
)
a
n
d
2
(
b
)
r
e
a
d
y
f
o
r
u
s
e
in
o
u
r
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s
.
T
h
e
f
o
l
l
o
w
i
n
g
a
c
t
i
o
n
s
w
e
r
e
r
e
q
u
i
r
e
d
to
g
u
a
r
a
n
t
e
e
t
h
e
d
a
t
a
'
s
c
o
m
p
a
t
i
b
i
l
i
t
y
a
n
d
e
n
h
a
n
c
e
its
q
u
a
l
i
t
y
:
−
Data
au
g
m
en
tatio
n
:
to
ad
d
r
ess
th
e
im
b
alan
ce
d
n
at
u
r
e
of
th
e
d
ataset,
d
ata
au
g
m
e
n
tatio
n
is
em
p
lo
y
ed
.
T
h
is
in
v
o
lv
es
th
e
cr
ea
tio
n
of
n
ew
i
m
ag
es
u
s
in
g
tech
n
iq
u
es
s
u
ch
as
cr
o
p
p
in
g
,
r
o
tatin
g
,
f
lip
p
in
g
,
an
d
zo
o
m
i
n
g
.
In
th
is
s
tu
d
y
,
a
two
-
p
h
ase
a
u
g
m
en
tatio
n
a
p
p
r
o
ac
h
is
im
p
lem
en
ted
.
I
n
itially
,
v
ar
io
u
s
tech
n
i
q
u
es,
in
clu
d
in
g
r
o
tatin
g
,
zo
o
m
in
g
,
r
a
n
d
o
m
d
is
to
r
tio
n
,
co
n
tr
ast
an
d
b
r
ig
h
tn
es
s
ad
ju
s
tm
en
ts
,
r
an
d
o
m
cr
o
p
p
in
g
,
an
d
f
lip
p
in
g
,
ar
e
ap
p
lied
to
im
ag
es
in
th
e
m
alig
n
an
t
class
ac
r
o
s
s
th
e
tr
a
in
in
g
,
v
alid
atio
n
,
an
d
test
s
et
s
.
Su
b
s
eq
u
en
tly
,
r
an
d
o
m
f
lip
p
in
g
in
b
o
th
h
o
r
iz
o
n
tal
an
d
v
e
r
tical
d
ir
ec
tio
n
s
is
ap
p
lied
to
im
ag
es
in
b
o
th
cla
s
s
es,
en
s
u
r
in
g
a
d
iv
er
s
e
s
et
of
im
ag
es
to
aid
m
o
d
el
g
en
e
r
aliza
tio
n
.
−
I
m
ag
e
s
ize
s
tan
d
ar
d
izatio
n
:
th
e
ch
o
s
en
C
NN
ar
ch
itectu
r
e
can
ac
co
m
m
o
d
ate
im
ag
es
of
v
ar
y
in
g
s
izes;
h
o
wev
er
,
f
o
r
tr
an
s
f
er
lear
n
in
g
,
it
is
ad
v
is
ab
le
to
alig
n
wi
th
p
r
e
-
tr
ain
ed
m
o
d
els
th
at
ty
p
ically
p
r
o
ce
s
s
im
ag
es
in
224
×
224
d
im
en
s
io
n
s
.
T
h
er
ef
o
r
e,
th
e
im
ag
e
s
ize
is
r
ed
u
ce
d
to
2
2
4
×
2
2
4
,
e
n
ab
lin
g
a
d
ir
ec
t
co
m
p
ar
is
o
n
with
p
r
e
-
tr
ain
ed
m
o
d
els
an
d
r
ed
u
cin
g
co
m
p
u
t
atio
n
al
o
v
er
h
ea
d
.
Su
b
s
eq
u
e
n
tly
,
n
o
r
m
aliza
tio
n
is
ap
p
lied
to
f
u
r
th
er
s
tr
ea
m
lin
e
co
m
p
u
tatio
n
al
co
m
p
le
x
ity
.
−
Data
n
o
r
m
aliza
tio
n
:
in
o
r
d
e
r
to
p
r
ep
a
r
e
d
atasets
f
o
r
f
u
r
th
er
an
aly
s
is
an
d
m
o
d
elin
g
,
d
ata
n
o
r
m
aliza
tio
n
is
an
ess
en
tial
s
tep
.
A
s
p
ec
tr
u
m
of
tech
n
iq
u
es
is
av
ailab
le
f
o
r
n
o
r
m
aliza
tio
n
,
en
co
m
p
a
s
s
in
g
m
in
–
m
ax
n
o
r
m
aliza
tio
n
,
z
-
s
co
r
e
n
o
r
m
a
lizatio
n
,
an
d
d
ec
im
al
s
ca
lin
g
n
o
r
m
aliza
tio
n
[
2
4
]
.
Data
n
o
r
m
aliza
tio
n
is
p
r
im
ar
ily
u
s
ed
to
im
p
r
o
v
e
d
ata
q
u
ality
,
m
ak
e
d
ata
co
m
p
ar
ab
le
ac
r
o
s
s
v
ar
io
u
s
r
ec
o
r
d
s
an
d
f
ield
s
,
an
d
im
p
r
o
v
e
e
n
tr
y
t
y
p
e
u
n
if
o
r
m
ity
an
d
co
n
s
is
ten
cy
.
−
C
o
n
v
er
s
io
n
of
d
ata:
th
e
L
I
DC
-
I
DR
I
d
ataset'
s
CT
s
ca
n
s
wer
e
co
n
v
e
r
ted
f
r
o
m
th
eir
o
r
ig
in
al
DI
C
OM
f
o
r
m
at
to
th
e
more
co
m
m
o
n
ly
u
s
ed
NI
f
T
I
f
o
r
m
at.
T
h
is
co
n
v
er
s
i
o
n
was
im
p
er
ativ
e
to
en
s
u
r
e
h
ar
m
o
n
izatio
n
b
etwe
en
our
d
eep
lea
r
n
in
g
f
r
a
m
ewo
r
k
an
d
th
e
d
ata,
f
ac
ilit
atin
g
s
m
o
o
th
in
teg
r
atio
n
an
d
p
r
o
ce
s
s
in
g
[
2
5
]
.
In
th
e
r
ea
lm
of
m
o
d
elin
g
,
two
ap
p
r
o
ac
h
es
a
r
e
a
v
ailab
le:
th
e
im
p
lem
en
tatio
n
of
a
C
NN
ar
ch
itectu
r
e
f
r
o
m
s
cr
atch
or
th
e
u
tili
za
tio
n
of
tr
an
s
f
er
lear
n
i
n
g
.
T
h
e
f
o
r
m
er
is
ty
p
ically
f
a
v
o
r
e
d
w
h
en
an
am
p
le
am
o
u
n
t
of
d
ata
is
av
ailab
le,
en
s
u
r
in
g
th
at
th
e
r
is
k
of
o
v
er
f
itti
n
g
is
m
in
im
ized
.
C
o
n
v
er
s
ely
,
th
e
l
atter
is
em
p
lo
y
ed
in
s
ce
n
ar
io
s
ch
ar
ac
ter
ized
by
d
ata
s
ca
r
city
or
t
h
e
im
p
e
r
ativ
e
to
r
ed
u
ce
m
o
d
elin
g
co
s
ts
.
T
r
an
s
f
er
lea
r
n
in
g
lev
er
ag
es
p
r
e
-
tr
ain
e
d
m
o
d
els,
b
en
ef
itin
g
f
r
o
m
weig
h
ts
d
er
iv
ed
f
r
o
m
ex
te
n
s
iv
e
tr
ain
in
g
on
d
iv
er
s
e
d
atasets
en
co
m
p
ass
in
g
th
o
u
s
an
d
s
of
class
es,
th
er
eb
y
r
ed
u
cin
g
th
e
r
e
q
u
is
ite
n
u
m
b
er
of
tr
ai
n
in
g
e
p
o
ch
s
.
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
2
,
Ap
r
il
20
25
:
1
4
9
6
-
1
5
0
6
1500
(
a)
(
b
)
Fig
u
r
e
2.
Diag
n
o
s
tic
im
ag
er
y
:
(
a)
in
p
u
t
CT
im
ag
es
an
d
(
b
)
p
r
ep
r
o
ce
s
s
ed
im
ag
es
Sin
ce
th
e
ad
v
e
n
t
of
Alex
Net
in
2
0
1
2
,
w
h
ich
d
e
m
o
n
s
tr
ated
s
u
p
er
io
r
ac
cu
r
ac
y
o
v
er
tr
a
d
itio
n
al
ML
alg
o
r
ith
m
s
,
C
NNs
h
av
e
b
ec
o
m
e
p
er
v
asiv
e.
T
h
is
d
ee
p
f
ee
d
-
f
o
r
war
d
ar
c
h
itectu
r
e
f
in
d
s
w
id
e
ap
p
licatio
n
s
in
im
ag
e
class
if
icatio
n
,
im
ag
e
s
e
g
m
en
tatio
n
,
an
d
o
b
ject
d
etec
tio
n
.
In
m
ed
ical
im
ag
i
n
g
d
atasets
,
wh
er
e
p
r
ec
is
io
n
is
cr
u
cial
f
o
r
h
u
m
an
life
,
co
n
v
en
tio
n
al
ML
alg
o
r
ith
m
s
o
f
ten
f
all
s
h
o
r
t.
Su
b
s
eq
u
en
t
to
th
e
in
tr
o
d
u
ctio
n
of
Alex
Net,
s
u
b
s
tan
tial
r
esear
ch
ef
f
o
r
ts
h
av
e
b
ee
n
d
ed
icate
d
to
in
n
o
v
ativ
ely
in
v
esti
g
atin
g
an
d
im
p
lem
en
tin
g
C
NNs,
r
esu
ltin
g
in
m
o
d
els
ex
h
ib
itin
g
s
u
p
er
i
o
r
p
e
r
f
o
r
m
an
ce
ev
en
in
c
o
m
p
ar
is
o
n
to
Alex
Ne
t.
A
s
tan
d
ar
d
C
NN
m
o
d
el
co
m
p
r
is
es
one
or
f
ew
co
n
v
o
l
u
tio
n
al
an
d
p
o
o
lin
g
lay
er
s
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
as
s
h
o
wn
in
F
ig
u
r
e
3,
cu
lm
in
atin
g
in
one
or
m
o
r
e
f
u
lly
co
n
n
ec
ted
lay
er
s
f
o
r
g
en
er
atin
g
class
if
ied
o
u
tp
u
ts
.
To
ex
t
r
ac
t
f
ea
tu
r
es
f
r
o
m
ev
e
r
y
v
is
u
al
,
th
e
co
n
v
o
lu
t
io
n
al
lay
er
co
n
v
o
lv
es
it
u
s
in
g
a
lear
n
a
b
le
k
e
r
n
el.
T
h
e
k
er
n
el,
r
ep
r
esen
ted
by
a
m
atr
ix
of
d
is
cr
ete
weig
h
ts
,
is
in
itialized
r
an
d
o
m
l
y
an
d
u
p
d
ated
iter
ativ
ely
to
m
in
im
ize
er
r
o
r
s
.
T
h
e
s
tr
id
e
p
a
r
am
eter
g
o
v
er
n
s
th
e
k
er
n
el'
s
m
o
v
em
en
t
th
r
o
u
g
h
th
e
im
ag
e,
with
v
alu
es
u
p
d
ated
th
r
o
u
g
h
a
co
m
p
u
tatio
n
al
p
r
o
ce
s
s
.
T
h
e
o
u
tp
u
t
of
a
co
n
v
o
lu
tio
n
al
lay
er
,
k
n
o
wn
as
a
f
ea
tu
r
e
m
ap
,
is
s
u
b
s
eq
u
en
tly
f
o
r
war
d
e
d
to
th
e
n
ex
t
lay
er
as
in
p
u
t.
T
h
e
o
u
tp
u
t
s
ize
(
ℎ
×
×
d)
of
a
c
o
n
v
o
lu
ti
o
n
al
lay
er
is
co
m
p
u
te
d
u
s
in
g
th
e
f
o
r
m
u
la:
ℎ
=
ℎ
−
+
2
+
1
(
1
)
=
−
+
2
+
1
(
2
)
W
h
er
e
h
an
d
w
ar
e
th
e
h
eig
h
t
an
d
wid
th
of
th
e
in
p
u
t
im
ag
e,
r
esp
ec
tiv
ely
.
f
is
th
e
s
ize
of
th
e
f
ilter
(
o
r
k
er
n
el)
.
p
is
th
e
am
o
u
n
t
of
ze
r
o
-
p
ad
d
in
g
ap
p
lied
to
th
e
in
p
u
t
im
ag
e
(
i
f
an
y
)
.
s
is
th
e
s
tr
id
e
of
th
e
c
o
n
v
o
lu
tio
n
.
Fig
u
r
e
3
.
Sam
p
le
c
o
n
v
o
lu
tio
n
al
n
eu
r
o
n
n
etwo
r
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
d
etec
tio
n
a
n
d
c
la
s
s
ifica
tio
n
o
f lu
n
g
ca
n
ce
r
(
R
a
mi
Yo
u
s
ef
)
1501
As
th
e
d
ep
th
of
a
C
NN
in
cr
e
ases
,
more
d
etails
can
be
e
x
t
r
ac
ted
,
r
esu
ltin
g
in
in
cr
ea
s
ed
ac
cu
r
ac
y
.
Ho
wev
er
,
a
tr
ad
e
-
o
f
f
e
x
is
ts
,
as
d
ee
p
er
ar
c
h
itectu
r
es
d
em
a
n
d
more
c
o
m
p
u
tatio
n
al
p
r
o
ce
s
s
es,
th
er
eb
y
in
cu
r
r
in
g
h
ig
h
er
co
s
ts
.
T
h
e
o
p
tim
al
n
u
m
b
er
of
lay
er
s
n
ec
ess
itates
c
ar
ef
u
l
co
n
s
id
er
atio
n
,
as
ex
ce
s
s
iv
e
d
ep
th
m
ay
n
o
t
alwa
y
s
tr
an
s
late
to
im
p
r
o
v
e
d
p
er
f
o
r
m
a
n
ce
co
m
p
ar
ed
to
s
h
allo
wer
ar
ch
itectu
r
es.
In
t
h
e
s
u
b
s
e
q
u
e
n
t
s
ta
g
es
of
n
e
u
r
a
l
n
et
wo
r
k
o
p
e
r
at
io
n
s
,
er
r
o
r
c
alc
u
l
ati
o
n
b
e
co
m
es
im
p
e
r
at
iv
e.
Po
s
t
-
o
u
t
p
u
t
g
en
er
ati
o
n
,
t
h
e
lo
s
s
f
u
n
cti
o
n
is
em
p
l
o
y
e
d
to
c
o
m
p
a
r
e
esti
m
a
te
d
l
a
b
els
wit
h
t
r
u
e
la
b
els
,
f
a
cili
tat
in
g
er
r
o
r
ass
ess
m
e
n
t
.
C
o
m
m
o
n
l
o
s
s
f
u
n
cti
o
n
s
i
n
cl
u
d
e
c
r
o
s
s
e
n
t
r
o
p
y
,
e
u
cl
id
ea
n
,
a
n
d
h
in
g
e.
W
eig
h
t
u
p
d
at
es
f
o
r
s
u
b
s
e
q
u
e
n
t
e
p
o
c
h
s
ar
e
o
r
ch
est
r
at
e
d
by
o
p
ti
m
i
ze
r
f
u
n
c
ti
o
n
s
,
wit
h
A
d
a
m
b
ei
n
g
a
wi
d
el
y
ad
o
p
te
d
ch
o
i
ce
.
T
h
is
ite
r
at
iv
e
p
r
o
c
ess
is
r
e
p
e
at
ed
a
cr
o
s
s
e
p
o
c
h
s
,
al
lo
wi
n
g
f
o
r
e
r
r
o
r
c
o
m
p
a
r
is
o
n
wi
th
th
e
b
est
p
r
e
v
i
o
u
s
e
p
o
ch
a
n
d
s
av
i
n
g
m
o
d
el
i
m
p
r
o
v
e
m
e
n
ts
w
h
e
n
o
b
s
er
v
e
d
.
An
i
ll
u
s
tr
ati
v
e
f
ig
u
r
e
of
a
C
NN
wit
h
tw
o
h
i
d
d
en
l
ay
er
s
,
f
ea
t
u
r
i
n
g
k
e
r
n
el
s
i
ze
s
of
(9
×
9
×
1)
a
n
d
(5
×
5
×
4)
in
t
h
e
f
i
r
s
t
a
n
d
s
ec
o
n
d
l
ay
er
s
,
r
esp
ec
t
iv
el
y
,
is
p
r
o
v
i
d
e
d
f
o
r
r
e
f
e
r
en
ce
.
5.
T
RANSF
E
R
L
E
A
RNING
Deep
lear
n
in
g
alg
o
r
ith
m
s
,
n
o
tab
ly
C
NNs,
d
em
o
n
s
tr
ate
ex
c
ep
tio
n
al
p
e
r
f
o
r
m
an
ce
wh
e
n
c
o
n
f
r
o
n
ted
with
a
s
u
b
s
tan
tial
v
o
lu
m
e
of
im
ag
es
p
er
class
.
Ho
wev
er
,
th
e
co
m
p
u
tatio
n
al
d
em
an
d
s
i
m
p
o
s
ed
by
ex
te
n
s
iv
e
lay
er
u
s
ag
e
ca
n
r
esu
lt
in
p
r
o
tr
ac
ted
tr
ain
in
g
tim
es,
s
p
an
n
in
g
d
ay
s
or
wee
k
s
on
co
n
tem
p
o
r
ar
y
h
a
r
d
war
e.
T
h
is
b
ec
o
m
es
p
ar
ticu
lar
ly
im
p
r
ac
t
ical
f
o
r
n
u
m
e
r
o
u
s
p
r
o
b
lem
d
o
m
ain
s
,
esp
ec
ially
th
o
s
e
p
er
t
ain
in
g
to
m
ed
ical
ap
p
licatio
n
s
.
C
o
n
s
eq
u
en
tly
,
a
d
o
p
tin
g
a
tr
an
s
f
er
lear
n
in
g
p
ar
ad
ig
m
,
wh
e
r
ein
p
r
ec
o
m
p
u
t
ed
weig
h
ts
f
r
o
m
a
p
r
e
-
tr
ain
ed
m
o
d
el
on
a
n
alo
g
o
u
s
d
ata
ar
e
r
ep
u
r
p
o
s
ed
,
p
r
o
v
e
s
ad
v
an
tag
eo
u
s
in
ter
m
s
of
c
o
s
t
ef
f
icien
cy
.
T
h
is
ap
p
r
o
ac
h
lev
er
ag
es
k
n
o
wled
g
e
g
lean
ed
d
u
r
in
g
t
h
e
p
r
io
r
tr
ain
in
g
of
a
m
o
d
el
on
co
m
p
a
r
ab
le
d
atasets
.
Fo
r
ex
am
p
le,
a
m
o
d
el
in
itially
tr
a
in
ed
f
o
r
h
is
to
p
ath
o
lo
g
y
im
ag
e
-
b
ased
ca
n
ce
r
d
iag
n
o
s
is
can
be
r
ep
u
r
p
o
s
ed
f
o
r
lu
n
g
ca
n
ce
r
d
iag
n
o
s
is
by
tr
an
s
f
er
r
in
g
t
h
e
ac
q
u
i
r
ed
weig
h
ts
.
T
r
an
s
f
er
lear
n
i
n
g
f
in
d
s
u
tili
ty
ac
r
o
s
s
m
u
ltip
le
d
o
m
ain
s
,
in
clu
d
in
g
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P),
s
o
u
n
d
,
im
a
g
e,
an
d
v
id
eo
p
r
o
ce
s
s
in
g
.
In
th
e
c
o
n
t
ex
t
of
C
NNs,
th
e
tr
ain
in
g
p
r
o
ce
s
s
b
eg
in
s
by
d
is
ce
r
n
in
g
im
a
g
e
e
d
g
es
an
d
b
o
r
d
er
s
,
p
r
o
g
r
ess
in
g
to
s
h
ap
e
id
en
tific
atio
n
.
Dee
p
er
la
y
er
s
c
ap
tu
r
e
in
c
r
ea
s
in
g
ly
in
tr
icate
d
etails,
cu
lm
in
atin
g
in
attem
p
ts
to
class
if
y
im
ag
es
b
ased
on
ass
ig
n
ed
lab
els.
Su
b
s
eq
u
en
tly
,
em
p
lo
y
in
g
tr
a
n
s
f
er
lear
n
in
g
in
v
o
lv
es
u
tili
zin
g
th
e
in
itial
lay
er
s
r
esp
o
n
s
ib
le
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
wh
ile
r
etr
ain
in
g
th
e
m
o
d
el'
s
latter
l
ay
er
s
with
n
ew
d
ata.
In
t
h
is
s
tu
d
y
,
C
NN
m
o
d
els
an
d
s
ev
er
al
tr
an
s
f
er
lear
n
in
g
-
b
ased
ar
ch
itectu
r
es,
in
clu
d
in
g
R
esNet5
0
,
VGG1
9
,
Den
s
eNe
t
1
6
9
,
a
n
d
I
n
ce
p
tio
n
V
3
.
5
.
1
.
ResNet
5
0
T
h
e
ar
ch
itectu
r
e
is
s
ep
ar
ated
in
to
two
b
lo
ck
s
,
ea
c
h
with
a
s
k
ip
co
n
n
ec
tio
n
.
T
h
e
o
u
t
p
u
t
of
th
e
p
r
ev
io
u
s
b
lo
c
k
is
s
u
m
m
ed
with
th
e
cu
r
r
en
t
b
lo
ck
’
s
o
u
t
p
u
t,
an
d
th
e
ac
tiv
atio
n
f
u
n
c
tio
n
is
ap
p
lied
to
f
(
x
)
+
x
r
ath
e
r
th
a
n
ju
s
t
f
(
x
)
.
T
h
is
g
en
e
r
ates
th
e
b
lo
ck
'
s
o
u
tp
u
t,
wh
ic
h
is
th
e
n
f
o
r
war
d
e
d
to
t
h
e
s
u
b
s
eq
u
en
t
b
lo
ck
.
In
R
esNet
50,
th
r
ee
c
o
n
v
o
lu
ti
o
n
al
lay
er
s
ar
e
u
tili
z
ed
with
a
s
p
ec
if
ic
co
n
f
ig
u
r
atio
n
,
as
illu
s
tr
ated
in
Fig
u
r
e
4,
in
s
tead
of
th
e
t
y
p
ical
two
co
n
v
o
lu
tio
n
al
lay
e
r
s
.
Fig
u
r
e
4
.
R
esidua
l
b
lo
ck
5
.
2
.
VG
G
1
9
T
h
e
VGG1
9
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
e
co
m
p
r
is
es
19
lay
er
s
an
d
f
ea
tu
r
es
a
s
u
b
s
tan
tial
n
u
m
b
er
of
p
ar
am
eter
s
.
T
h
e
m
o
d
el'
s
s
ize,
s
p
ec
if
ically
in
ter
m
s
of
f
u
lly
co
n
n
ec
ted
n
o
d
es,
is
5
7
4
MB.
An
in
cr
ea
s
e
in
th
e
n
u
m
b
er
of
lay
er
s
ty
p
ically
co
r
r
elate
s
with
en
h
an
ce
d
ac
cu
r
ac
y
in
d
ee
p
n
e
u
r
al
n
etwo
r
k
s
(
D
NNs).
T
h
e
VGG1
9
ar
ch
itectu
r
e
i
n
clu
d
es
19
tr
ain
a
b
le
co
n
v
o
lu
tio
n
al
lay
e
r
s
,
wh
ic
h
ar
e
in
ter
s
p
er
s
ed
with
m
a
x
p
o
o
lin
g
an
d
d
r
o
p
o
u
t
lay
er
s
,
as
d
ep
icted
in
Fig
u
r
e
5.
VGG1
9
is
h
ig
h
ly
a
d
v
an
tag
e
o
u
s
due
to
its
u
s
e
of
a
s
er
ies
of
3
×3
C
o
n
v
Net
to
in
c
r
ea
s
e
n
etwo
r
k
d
ep
t
h
.
To
r
ed
u
ce
f
ea
tu
r
e
m
ap
d
im
e
n
s
io
n
s
,
m
ax
-
p
o
o
lin
g
lay
er
s
ar
e
i
n
co
r
p
o
r
ated
.
T
h
e
f
u
lly
co
n
n
ec
ted
n
etwo
r
k
(
FC
N)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
2
,
Ap
r
il
20
25
:
1
4
9
6
-
1
5
0
6
1502
co
m
p
r
is
es
two
lay
er
s
,
ea
c
h
co
n
tain
in
g
4096
n
eu
r
o
n
s
.
VGG1
9
was
tr
ain
ed
on
in
d
iv
id
u
al
lesi
o
n
s
am
p
les,
wh
ile
f
o
r
test
in
g
,
all
lesi
o
n
ty
p
es
wer
e
in
clu
d
ed
to
m
in
im
ize
t
h
e
o
c
cu
r
r
en
ce
of
f
alse
p
o
s
itiv
es.
Fig
u
r
e
5
.
VGG1
9
a
r
ch
itectu
r
e
5
.
3
.
DenseNet
1
6
9
T
h
e
Den
s
eNe
t
ar
ch
itectu
r
e
is
co
m
p
o
s
ed
of
m
u
ltip
le
d
en
s
e
b
lo
ck
s
,
wh
ich
a
r
e
in
ter
c
o
n
n
ec
ted
by
tr
an
s
itio
n
b
lo
c
k
s
th
at
c
o
n
s
is
t
of
co
n
v
o
lu
tio
n
al
an
d
p
o
o
lin
g
la
y
er
s
.
E
ac
h
d
en
s
e
b
lo
ck
is
s
tr
u
ctu
r
ed
f
r
o
m
s
ev
er
al
u
n
its
co
n
tain
in
g
c
o
n
v
o
lu
tio
n
a
l
lay
er
s
,
wh
e
r
e
ea
c
h
u
n
it
r
ec
eiv
es
th
e
o
u
tp
u
ts
of
all
p
r
ec
e
d
in
g
u
n
its
with
t
h
e
s
am
e
f
ea
tu
r
e
m
ap
s
ize.
Un
lik
e
R
esNet,
wh
ich
p
er
f
o
r
m
s
s
u
m
m
atio
n
on
th
e
o
u
tp
u
ts
,
Den
s
eNe
t
co
n
ca
ten
ates
th
em
.
Fig
u
r
e
6
illu
s
tr
ates
th
e
s
tr
u
ctu
r
e
of
a
d
en
s
e
b
lo
c
k
with
a
g
r
o
wth
r
ate
of
k
=4
.
I
n
th
is
co
n
tex
t,
'
k
'
r
ep
r
esen
ts
th
e
n
u
m
b
e
r
o
f
f
ea
t
u
r
e
m
ap
s
g
en
er
ate
d
b
y
th
e
f
u
n
ctio
n
H.
I
f
ea
c
h
H
f
u
n
ctio
n
p
r
o
d
u
ce
s
k
f
ea
tu
r
e
m
ap
s
‘
k
’
f
ea
tu
r
e
m
ap
s
,
th
en
t
h
e
n
-
th
lay
er
will
h
av
e
k
0
+k
×(
n
−1
)
in
p
u
t
f
ea
tu
r
e
m
ap
s
,
w
h
er
e
k
0
d
en
o
tes
th
e
n
u
m
b
er
o
f
c
h
an
n
els in
th
e
in
p
u
t la
y
er
.
Fig
u
r
e
6
.
Den
s
eNe
t b
lo
c
k
with
a
g
r
o
wth
r
ate
o
f
k
=4
5
.
4
.
I
ncept
io
nV3
As
s
h
o
wn
in
T
ab
le
1
,
th
is
ar
c
h
itectu
r
e
in
clu
d
es
a
s
tem
b
lo
c
k
co
n
s
is
tin
g
o
f
s
ix
co
n
v
o
l
u
tio
n
al
lay
er
s
,
ea
ch
with
a
3
×3
k
er
n
el
s
ize,
with
a
p
o
o
lin
g
lay
er
f
o
llo
win
g
th
e
t
h
ir
d
co
n
v
o
lu
tio
n
al
lay
e
r
.
Pad
d
in
g
is
s
et
to
ze
r
o
f
o
r
t
h
e
s
p
ec
if
ied
co
n
v
o
lu
tio
n
al
lay
er
s
,
wh
ile
f
o
r
th
e
o
th
er
s
,
n
o
p
a
d
d
in
g
is
ap
p
lied
.
T
o
r
ed
u
ce
g
r
id
s
ize,
a
r
ed
u
ctio
n
m
eth
o
d
is
em
p
lo
y
ed
b
etwe
en
I
n
ce
p
tio
n
b
lo
ck
s
.
T
h
e
ar
ch
itectu
r
e
h
as
a
d
ep
th
o
f
4
2
lay
er
s
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
d
etec
tio
n
a
n
d
c
la
s
s
ifica
tio
n
o
f lu
n
g
ca
n
ce
r
(
R
a
mi
Yo
u
s
ef
)
1503
in
cu
r
s
a
co
m
p
u
tatio
n
al
co
s
t
2
.
5
tim
es
h
ig
h
er
t
h
an
Go
o
g
leN
et,
th
o
u
g
h
it
r
em
ain
s
s
ig
n
if
ica
n
tly
m
o
r
e
e
f
f
icien
t
th
an
th
e
VGG
ar
ch
itectu
r
e.
T
h
e
p
r
im
a
r
y
d
is
tin
ctio
n
b
etwe
e
n
I
n
ce
p
tio
n
v
2
an
d
v
3
lies
in
th
e
f
ac
to
r
izatio
n
o
f
co
n
v
o
l
u
tio
n
s
,
wh
ich
r
ed
u
ce
s
th
e
n
u
m
b
e
r
o
f
p
ar
am
eter
s
with
o
u
t
af
f
ec
tin
g
n
etwo
r
k
p
er
f
o
r
m
an
ce
.
Fo
r
ex
am
p
le,
r
ep
lacin
g
a
5
×5
co
n
v
o
lu
ti
o
n
a
l
lay
er
with
two
3
×3
lay
er
s
allo
ws
th
e
n
etwo
r
k
to
ex
tr
ac
t
f
ea
tu
r
es
s
im
ilar
ly
wh
ile
r
ed
u
cin
g
co
m
p
u
tatio
n
al
co
m
p
lex
ity
a
n
d
m
o
d
el
co
s
t.
T
ab
le
1.
Ar
c
h
itectu
r
e
o
f
in
ce
p
tio
n
V3
Ty
p
e
P
a
t
c
h
s
i
z
e
/
s
t
r
i
d
e
o
r
r
e
m
a
r
k
s
I
n
p
u
t
si
z
e
c
o
n
v
3
×
3
/
2
2
9
9
×
2
9
9
×
3
c
o
n
v
3
×
3
/
1
1
4
9
×
1
4
9
×
32
c
o
n
v
p
a
d
d
e
d
3
×
3
/
1
1
4
7
×
1
4
7
×
32
p
o
o
l
3
×
3
/
2
1
4
7
×
1
4
7
×
64
c
o
n
v
3
×
3
/
1
73
×
73
×
64
c
o
n
v
3
×
3
/
2
71
×
71
×
80
c
o
n
v
3
×
3
/
1
35
×
35
×
1
9
2
3
×
I
n
c
e
p
t
i
o
n
I
n
c
e
p
t
i
o
n
b
l
o
k
A
35
×
35
×
2
8
8
5
×
I
n
c
e
p
t
i
o
n
I
n
c
e
p
t
i
o
n
b
l
o
k
B
17
×
17
×
7
6
8
2
×
I
n
c
e
p
t
i
o
n
I
n
c
e
p
t
i
o
n
b
l
o
k
C
8
×
8
×
1
2
8
0
p
o
o
l
8
×
8
8
×
8
×
2
0
4
8
l
i
n
e
a
r
l
o
g
i
t
s
1
×
1
×
2
0
4
8
so
f
t
m
a
x
c
l
a
ss
i
f
i
e
r
1
×
1
×
1
0
0
0
6.
E
NS
E
M
B
L
E
AP
P
RO
ACH
E
n
s
em
b
le
lear
n
i
n
g
i
n
v
o
lv
es
am
alg
am
atin
g
t
h
e
o
u
tco
m
es
of
m
u
ltip
le
m
o
d
els,
wh
ich
m
ay
em
p
l
o
y
id
en
tical
or
d
is
tin
ct
alg
o
r
ith
m
s
.
T
h
is
f
ield
en
co
m
p
ass
es
v
ar
io
u
s
tech
n
iq
u
es
ca
te
g
o
r
ize
d
in
to
th
r
ee
p
r
im
a
r
y
g
r
o
u
p
s
:
b
ag
g
in
g
,
b
o
o
s
tin
g
,
an
d
s
tack
in
g
.
B
ag
g
in
g
en
tails
ag
g
r
eg
atin
g
th
e
r
esu
lts
of
a
s
i
n
g
le
m
o
d
el
tr
ain
ed
iter
ativ
ely
on
d
iv
er
s
e
d
atasets
.
B
o
o
s
tin
g
co
m
b
in
es
th
e
o
u
tp
u
ts
of
m
u
ltip
le
m
o
d
els,
wh
ile
s
tack
in
g
in
teg
r
ates
elem
en
ts
of
b
o
t
h
ap
p
r
o
ac
h
es.
In
th
is
s
tu
d
y
,
th
e
b
o
o
s
tin
g
tech
n
iq
u
e
will
be
im
p
lem
e
n
ted
.
Fo
llo
win
g
th
e
tr
ain
in
g
of
C
NN,
R
esNet5
0
,
VGG1
9
,
Den
s
eNe
t
1
6
9
,
an
d
I
n
ce
p
tio
n
v3
m
o
d
els,
p
r
ed
ictio
n
s
will
be
m
ad
e
on
im
ag
es
with
in
th
e
v
alid
atio
n
a
n
d
test
d
atasets
.
Su
b
s
eq
u
en
tly
,
a
m
ax
v
o
tin
g
tech
n
iq
u
e
will
be
ap
p
lied
,
le
v
er
ag
in
g
th
e
s
u
m
f
u
n
ctio
n
,
as
th
e
o
u
tco
m
es
ar
e
co
n
f
in
ed
to
th
e
[
0
,
1]
class
r
an
g
e.
T
h
e
class
with
th
e
h
ig
h
es
t
v
o
te
co
u
n
t
will
be
s
elec
ted
.
T
h
is
ap
p
r
o
ac
h
s
er
v
es
to
m
itig
ate
v
ar
ian
ce
an
d
e
n
h
a
n
ce
er
r
o
r
g
en
e
r
aliza
tio
n
.
7.
E
VA
L
UA
T
I
O
N
W
h
en
d
ea
lin
g
with
s
u
p
er
v
is
e
d
p
r
o
b
lem
s
,
we
ca
n
ev
al
u
ate
th
e
m
o
d
el
in
two
d
if
f
e
r
en
t
w
ay
s
.
First,
d
ata
m
u
s
t
b
e
d
iv
id
e
d
in
to
tr
ai
n
in
g
an
d
test
in
g
s
ets.
Nex
t,
th
e
m
o
d
el
m
u
s
t
b
e
tr
ain
ed
u
s
in
g
th
e
tr
ain
in
g
s
ets,
an
d
test
d
ataset
lab
els
m
u
s
t
b
e
p
r
ed
icted
u
s
in
g
th
e
tr
ain
ed
m
o
d
el.
T
h
is
allo
ws
u
s
to
co
m
p
u
te
th
e
e
r
r
o
r
s
b
y
co
m
p
ar
in
g
th
e
p
r
ed
icted
r
esu
l
ts
with
th
e
tr
u
e
v
alu
es.
K
-
Fo
l
d
cr
o
s
s
-
v
alid
atio
n
is
th
e
s
ec
o
n
d
p
r
o
ce
s
s
.
T
o
tr
ai
n
th
e
m
o
d
el
i
n
th
is
s
ce
n
ar
io
,
th
e
d
ata
will
b
e
s
p
lit
in
to
K
s
u
b
s
ets,
with
th
e
ex
ce
p
tio
n
o
f
o
n
e
t
h
at
will
b
e
k
ep
t
o
u
t
f
o
r
ass
ess
m
en
t.
Af
ter
ea
ch
tr
ain
in
g
r
o
u
n
d
,
we
co
m
p
u
te
m
atr
ices,
an
d
th
e
o
p
tim
al
m
o
d
e
l
will
u
ltima
tely
b
e
ch
o
s
en
.
T
h
o
u
g
h
it r
eq
u
ir
es a
lo
t m
o
r
e
co
m
p
u
tatio
n
an
d
m
o
n
ey
th
an
d
ee
p
lear
n
in
g
,
th
is
ap
p
r
o
ac
h
is
s
u
p
er
io
r
to
th
e
p
r
ev
io
u
s
o
n
e.
Du
e
to
th
e
h
u
g
e
am
o
u
n
t o
f
co
m
p
u
tatio
n
in
n
etwo
r
k
s
,
it is
n
o
t a
d
v
is
ed
u
n
l
ess
it is
af
f
o
r
d
ab
le
.
8.
RE
SU
L
T
S
AND
D
I
SCU
SS
I
O
N
Fo
r
r
esu
lts
an
aly
s
is
,
to
m
ak
e
t
h
e
r
esu
lts
more
a
p
p
r
o
p
r
iate
f
o
r
our
p
u
r
p
o
s
e,
w
h
ich
is
m
in
i
m
izin
g
th
e
in
co
r
r
ec
t
class
if
icatio
n
s
,
we
co
m
p
u
ted
th
e
co
n
f
u
s
io
n
m
at
r
ix
,
wh
ich
d
is
p
lay
s
th
e
n
u
m
b
er
of
co
r
r
ec
t
an
d
in
co
r
r
ec
t
class
if
icatio
n
s
.
Ad
d
itio
n
ally
,
th
e
co
n
f
u
s
io
n
m
atr
i
x
-
ass
is
ted
in
d
eter
m
i
n
in
g
all
of
th
e
s
ig
n
if
ican
t
m
etr
ics;
p
r
ec
is
io
n
,
ac
cu
r
ac
y
,
F1
s
co
r
es,
an
d
r
ec
all
wer
e
co
m
p
u
ted
f
o
r
each
ar
ch
itectu
r
e.
8
.
1
.
E
ns
em
ble
5
m
o
dels
CN
N,
ResNet
5
0
,
VG
G
1
9
,
DenseNet
1
6
9
,
a
nd
I
ncept
io
nV3
An
en
s
em
b
le
m
o
d
el
is
a
co
m
b
in
atio
n
o
f
m
u
ltip
le
in
d
iv
i
d
u
al
m
o
d
els
to
ac
h
iev
e
b
ette
r
p
r
e
d
ictiv
e
p
er
f
o
r
m
an
ce
.
T
h
e
i
d
ea
is
th
at
b
y
c
o
m
b
i
n
in
g
t
h
e
s
tr
en
g
t
h
s
o
f
m
u
ltip
le
m
o
d
els,
t
h
e
en
s
e
m
b
le
ca
n
ca
p
tu
r
e
a
b
r
o
ad
e
r
r
an
g
e
o
f
f
ea
tu
r
es
an
d
r
ed
u
ce
e
r
r
o
r
s
.
In
th
is
m
eth
o
d
,
C
NN,
R
esNet5
0
,
VGG1
9
,
Den
s
eNe
t1
6
9
,
an
d
I
n
ce
p
tio
n
V3
m
o
d
els
wer
e
ch
o
s
en
.
Af
ter
av
er
ag
in
g
th
eir
p
r
ed
ictio
n
s
,
th
e
p
er
f
o
r
m
an
ce
m
etr
ics
f
o
r
th
e
test
d
ataset
wer
e
as
f
o
llo
ws:
9
7
.
5
%
ac
cu
r
ac
y
,
9
6
.
8
%
p
r
ec
is
io
n
,
9
6
.
9
%
r
ec
all,
a
n
d
9
7
.
7
7
%
F1
-
s
co
r
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
2
,
Ap
r
il
20
25
:
1
4
9
6
-
1
5
0
6
1504
8
.
2
.
Co
m
pa
riso
ns
wit
h o
t
her
wo
rk
s
T
h
is
s
ec
tio
n
co
m
p
ar
es
our
p
r
o
p
o
s
ed
m
o
d
el'
s
p
er
f
o
r
m
an
ce
w
ith
a
n
u
m
b
er
of
s
tate
-
of
-
th
e
-
ar
t
m
eth
o
d
s
th
at
h
av
e
b
ee
n
put
f
o
r
th
f
o
r
u
s
in
g
m
icr
o
s
co
p
ic
im
ag
es
to
id
en
tify
lu
n
g
ca
n
ce
r
.
T
ab
le
2
co
m
p
ar
es
th
e
r
esu
lts
of
th
e
co
r
r
esp
o
n
d
in
g
in
tr
o
d
u
ce
d
m
eth
o
d
with
s
ev
er
al
s
tate
-
of
-
th
e
-
ar
t
tech
n
iq
u
es
d
escr
ib
ed
in
th
e
liter
atu
r
e
r
ev
iew.
T
h
e
s
tu
d
y
c
o
m
p
ar
es
t
h
e
p
r
o
p
o
s
ed
m
o
d
el'
s
p
er
f
o
r
m
a
n
ce
to
th
e
s
tate
-
of
-
th
e
-
ar
t
m
o
d
els
f
o
r
th
e
d
etec
tio
n
an
d
ca
teg
o
r
izatio
n
of
lu
n
g
ca
n
ce
r
.
T
ab
le
2.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
b
etwe
en
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
an
d
e
x
is
tin
g
m
o
d
els
ac
r
o
s
s
d
iv
er
s
e
d
ata
s
ets
R
e
f
e
r
e
n
c
e
s
D
a
t
a
s
e
t
M
o
d
e
l
R
e
c
a
l
l
A
c
c
u
r
a
c
y
M
a
f
t
o
u
n
i
et
al
.
[
2
6
]
C
O
V
I
D
1
9
-
CT
En
se
mb
l
e
+
S
V
M
9
0
.
8
0
9
5
.
3
1
G
i
f
a
n
i
et
al
.
[
1
6
]
C
O
V
I
D
1
9
-
CT
C
N
N
+
LSTM
8
5
.
5
0
8
5
.
5
0
B
h
a
n
d
a
r
et
al
.
[
2
1
]
CR
C
u
s
t
o
m
C
N
N
9
6
.
5
6
9
4
.
3
1
C
h
e
n
et
al
.
[
2
7
]
v
IQ
-
O
TH
/
N
C
C
D
C
N
N
+
N
LP
8
7
.
5
8
8
.
0
A
l
i
et
al
.
[
1
8
]
H
A
M
1
0
0
0
0
D
C
N
N
9
3
.
6
6
9
1
.
9
3
Al
-
Y
a
sr
i
y
et
al
.
[
2
8
]
IQ
-
O
TH
/
N
C
C
D
C
N
N
:
A
l
e
x
N
e
t
a
r
c
h
i
t
e
c
t
u
r
e
9
3
.
2
3
9
3
.
5
4
P
r
o
p
o
se
d
m
o
d
e
l
IQ
-
O
TH
/
N
C
C
D
En
se
mb
l
e
5
M
o
d
e
l
s
C
N
N
,
R
e
sN
e
t
5
0
,
VGG
19,
D
e
n
seNe
t
1
6
9
,
a
n
d
I
n
c
e
p
t
i
o
n
V
3
9
7
.
7
7
9
7
.
5
T
h
is
wo
r
k
aim
ed
to
im
p
r
o
v
e
p
er
f
o
r
m
an
ce
b
y
u
s
in
g
f
o
u
r
d
if
f
e
r
en
t
e
n
s
em
b
le
a
p
p
r
o
ac
h
es
an
d
a
cu
s
to
m
ized
C
NN
tr
ain
ed
o
n
m
icr
o
s
co
p
ic
im
ag
es
o
f
l
u
n
g
ca
n
ce
r
.
T
h
e
ac
c
u
r
ac
y
r
esu
lts
o
f
all
m
o
d
els
ar
e
co
m
p
ar
ed
in
T
ab
le
2
.
On
test
d
ata,
th
e
en
s
em
b
le
o
f
th
e
t
o
p
f
o
u
r
m
o
d
els
h
ad
t
h
e
h
ig
h
est
ac
cu
r
ac
y
.
A
C
OVI
D
-
19
-
C
T
d
ataset
co
m
p
r
is
in
g
7
,
5
9
3
im
ag
es
s
o
u
r
ce
d
f
r
o
m
s
ev
e
n
p
u
b
licly
av
ailab
le
d
atasets
,
en
co
m
p
ass
in
g
d
at
a
f
r
o
m
4
6
6
p
atien
ts
is
p
r
o
v
id
ed
b
y
Ma
f
to
u
n
i
et
a
l
.
[
2
6
]
th
r
o
u
g
h
th
e
u
s
e
o
f
an
en
s
em
b
le
d
ee
p
lear
n
i
n
g
m
o
d
el
u
tili
zin
g
p
r
e
-
tr
ai
n
ed
r
esid
u
al
atten
tio
n
an
d
De
n
s
eNe
t
ar
ch
itectu
r
es,
Gif
an
i
et
a
l
.
[
1
6
]
em
p
lo
y
ed
a
n
en
s
em
b
le
d
ee
p
tr
a
n
s
f
er
lear
n
in
g
s
y
s
tem
,
lev
er
a
g
in
g
d
iv
er
s
e
p
r
e
-
tr
ain
ed
C
NN
ar
ch
itectu
r
es,
to
ac
h
iev
e
ef
f
ec
tiv
e
d
iag
n
o
s
is
o
f
C
OVI
D
-
1
9
f
r
o
m
C
T
s
ca
n
s
.
On
th
e
o
th
er
h
a
n
d
,
a
n
o
v
el
lig
h
tweig
h
t
s
in
g
le
C
NN
m
o
d
el
f
o
r
C
OVI
D
-
1
9
im
ag
e
class
if
icatio
n
u
s
in
g
C
XR
im
ag
es wa
s
p
r
o
p
o
s
ed
in
[
2
1
]
.
Fu
r
th
e
r
m
o
r
e
,
o
n
th
e
test
d
ataset,
all
o
f
th
e
en
s
em
b
le
m
o
d
els o
u
tp
er
f
o
r
m
ed
s
in
g
le
m
o
d
els
in
ter
m
s
o
f
ac
c
u
r
ac
y
.
T
h
e
e
n
s
em
b
le
m
o
d
el
o
f
th
e
to
p
f
o
u
r
m
o
d
els,
wh
ich
h
ad
an
ac
cu
r
ac
y
o
f
9
7
.
5
.
3
%,
was
th
e
b
est
m
o
d
el
i
n
th
e
d
ataset
v
alid
atio
n
s
ce
n
ar
io
.
I
n
ad
d
itio
n
t
o
tu
b
er
cu
lo
s
is
an
d
p
n
eu
m
o
n
ia,
an
ex
p
lan
atio
n
g
en
er
atio
n
(
XAI
)
f
r
am
ewo
r
k
is
u
s
ed
.
T
h
e
d
etec
tio
n
o
f
C
OVI
D
-
1
9
,
p
n
eu
m
o
n
i
a,
an
d
t
u
b
er
c
u
lo
s
is
d
is
ea
s
es
u
s
in
g
s
u
ch
an
XAI
-
b
ased
s
in
g
le
C
NN
m
o
d
el
p
r
o
d
u
ce
d
tr
ai
n
in
g
ac
cu
r
ac
y
o
f
9
5
.
7
6
±
1
.
1
5
%,
test
ac
cu
r
ac
y
o
f
9
4
.
3
1
±
1
.
0
1
%,
an
d
v
alid
atio
n
ac
cu
r
ac
y
o
f
9
4
.
5
4
±
1
.
3
3
%.
B
as
i
c
a
n
d
b
e
n
c
h
m
a
r
k
C
N
N
a
r
ch
i
t
e
c
t
u
r
es
w
e
r
e
a
p
p
l
i
e
d
b
y
C
h
en
e
t
a
l
.
[
2
7
]
,
w
h
o
w
e
r
e
k
n
o
w
n
t
o
e
m
p
l
o
y
a
n
y
p
a
r
a
m
e
t
e
r
o
p
t
i
m
i
z
a
ti
o
n
t
e
ch
n
i
q
u
e
.
E
v
e
n
t
h
o
u
g
h
t
h
e
m
a
j
o
r
i
t
y
o
f
h
is
s
t
u
d
y
'
s
r
es
u
l
ts
a
r
e
i
n
t
r
i
g
u
i
n
g
a
n
d
a
c
h
i
e
v
e
a
n
a
c
c
u
r
a
c
y
o
f
8
8
.
0
%
,
w
e
o
b
s
e
r
v
e
t
h
a
t
t
h
e
s
e
m
o
d
e
ls
wi
l
l
p
er
f
o
r
m
p
o
o
r
l
y
i
f
p
e
r
f
o
r
m
a
n
c
e
t
i
l
t
i
n
g
c
o
n
d
i
t
i
o
n
s
a
r
e
a
d
d
e
d
.
A
d
d
i
t
i
o
n
a
l
l
y
,
a
D
C
N
N
m
o
d
e
l
b
a
s
e
d
o
n
a
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
w
a
s
p
r
o
p
o
s
e
d
b
y
A
l
i
e
t
a
l
.
[
1
8
]
f
o
r
p
r
ec
is
e
ca
teg
o
r
izatio
n
o
f
b
en
i
g
n
an
d
m
alig
n
an
t
s
k
in
lesi
o
n
s
.
T
h
e
m
o
d
el
was
ass
ess
ed
u
s
i
n
g
th
e
HAM
1
0
0
0
0
d
ataset
an
d
attain
ed
im
p
r
ess
iv
e
o
u
tco
m
es
with
a
test
in
g
ac
cu
r
ac
y
o
f
9
1
.
9
3
%
an
d
a
tr
ain
in
g
ac
cu
r
ac
y
o
f
9
3
.
1
6
%.
Al
-
Yasr
iy
et
a
l
.
[
2
8
]
u
s
ed
th
e
s
am
e
d
ataset
th
at
we
u
s
ed
in
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
u
s
in
g
C
NN
-
Alex
Net
ar
ch
itectu
r
e
an
d
ac
h
iev
e
d
an
a
cc
u
r
ac
y
o
f
9
3
.
5
4
8
%,
a
s
en
s
itiv
ity
o
f
9
5
.
7
1
4
%,
an
d
a
s
p
ec
if
icity
o
f
9
5
%.
Ou
r
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
els;
C
NN,
R
esNet5
0
,
VGG1
9
,
Den
s
eNe
t1
6
9
,
an
d
I
n
ce
p
tio
n
V3
,
g
av
e
p
er
f
o
r
m
an
ce
m
etr
ics,
af
ter
a
v
er
ag
in
g
p
r
ed
ictio
n
o
f
s
elec
ted
m
o
d
els,
o
f
9
7
.
5
%
ac
c
u
r
ac
y
.
C
o
n
s
eq
u
en
tly
,
th
is
s
tu
d
y
h
as
s
h
o
wn
h
o
w
cr
u
cial
it
is
to
s
o
lv
e
th
e
ch
allen
g
in
g
ta
s
k
o
f
ch
o
o
s
in
g
th
e
id
ea
l
s
et
o
f
weig
h
ts
an
d
b
iases
n
ee
d
ed
f
o
r
tr
ain
i
n
g
a
C
NN
m
o
d
el
b
y
u
tili
zin
g
th
e
h
y
b
r
i
d
m
etah
eu
r
is
tic
en
s
em
b
le
alg
o
r
ith
m
an
d
C
NN
m
o
d
els
[
2
9
]
.
Fu
r
th
er
m
o
r
e,
th
e
m
eth
o
d
d
em
o
n
s
tr
ates
th
at
am
alg
am
atin
g
th
ese
ap
p
r
o
ac
h
es
en
h
an
ce
s
th
e
ac
cu
r
ac
y
o
f
class
if
icatio
n
an
d
o
v
er
all
p
e
r
f
o
r
m
an
ce
wh
e
n
class
if
y
in
g
lu
n
g
ca
n
ce
r
in
C
T
im
ag
es.
9.
CO
NCLU
SI
O
N
Usi
n
g
a
C
NN
m
o
d
el
an
d
f
o
u
r
d
if
f
e
r
en
t
en
s
em
b
le
ap
p
r
o
ac
h
es,
th
is
s
tu
d
y
p
r
esen
ts
a
n
o
v
el
h
y
b
r
id
alg
o
r
ith
m
to
in
cr
ea
s
e
th
e
ac
cu
r
ac
y
o
f
lu
n
g
ca
n
ce
r
class
if
icat
io
n
.
T
o
en
s
u
r
e
a
r
o
b
u
s
t
an
d
ac
cu
r
ate
co
m
p
ar
is
o
n
o
f
r
esu
lts
,
th
e
e
x
p
er
im
e
n
tal
p
r
o
ce
d
u
r
e
in
v
o
l
v
ed
p
ar
titi
o
n
in
g
th
e
d
ataset
in
to
test
s
ets,
v
alid
atio
n
an
d
tr
ain
in
g
.
Du
r
in
g
t
h
e
tr
ain
in
g
p
h
ase,
m
o
d
el
v
alid
atio
n
was
p
e
r
f
o
r
m
e
d
u
s
in
g
a
d
e
d
icate
d
v
alid
atio
n
d
ataset.
Fo
llo
win
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
p
r
ed
ictio
n
s
wer
e
g
e
n
er
ated
f
o
r
b
o
t
h
t
h
e
v
alid
atio
n
an
d
test
.
T
h
e
m
o
d
els
wer
e
tr
ain
e
d
u
s
in
g
a
cu
s
to
m
ized
C
NN
m
o
d
el,
alo
n
g
with
s
ev
er
al
p
r
e
-
tr
ain
ed
tr
an
s
f
er
lear
n
in
g
ar
c
h
itectu
r
es,
n
am
ely
R
esNet5
0
,
VGG1
9
,
Den
s
eNe
t1
6
9
,
a
n
d
I
n
ce
p
tio
n
V
3
.
Su
b
s
eq
u
en
tly
,
th
is
en
s
em
b
le
a
p
p
r
o
ac
h
was
em
p
lo
y
ed
to
ad
d
r
ess
m
o
d
el
wea
k
n
ess
es
an
d
en
h
an
ce
o
v
er
all
p
e
r
f
o
r
m
an
c
e.
Mu
ltip
le
m
o
d
els
wer
e
co
m
b
in
ed
an
d
m
er
g
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
d
etec
tio
n
a
n
d
c
la
s
s
ifica
tio
n
o
f lu
n
g
ca
n
ce
r
(
R
a
mi
Yo
u
s
ef
)
1505
u
s
in
g
th
e
av
er
a
g
e
en
s
em
b
le
te
ch
n
iq
u
e.
T
h
e
o
p
tim
al
co
m
b
in
a
tio
n
,
id
e
n
tifie
d
as
a
b
len
d
of
t
h
e
to
p
f
iv
e
m
o
d
els,
was
ch
o
s
en
as
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
el.
T
h
is
wo
r
k
can
be
ex
ten
d
e
d
to
im
p
lem
en
t
with
d
if
f
er
e
n
t
d
atasets
an
d
u
s
in
g
3D
C
NNs
or
v
o
lu
m
etr
ic
s
eg
m
en
tatio
n
f
o
r
3D
m
e
d
ical
im
ag
es.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
au
th
o
r
/s
wo
u
ld
lik
e
to
th
an
k
th
e
Palest
in
e
T
ec
h
n
ical
Un
iv
er
s
ity
-
Kad
o
o
r
ie
f
o
r
th
ei
r
f
in
an
cial
s
u
p
p
o
r
t to
c
o
n
d
u
ct
th
is
r
esear
c
h
.
RE
F
E
R
E
NC
E
S
[
1]
R
.
Y
o
u
sef
,
“
I
d
e
n
t
i
f
y
i
n
g
i
n
f
o
r
m
a
t
i
v
e
c
o
r
o
n
a
v
i
r
u
s
t
w
e
e
t
s
u
si
n
g
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
d
o
c
u
m
e
n
t
e
mb
e
d
d
i
n
g
,
”
P
a
l
e
st
i
n
e
T
e
c
h
n
i
c
a
l
U
n
i
v
e
rs
i
t
y
R
e
se
a
rc
h
J
o
u
r
n
a
l
,
v
o
l
.
1
0
,
n
o
.
1
,
p
p
.
9
3
–
1
0
2
,
2
0
2
2
,
d
o
i
:
1
0
.
5
3
6
7
1
/
p
t
u
r
j
.
v
1
0
i
1
.
2
2
0
.
[
2
]
W
.
L
i
e
t
a
l
.
,
“
A
d
v
a
n
c
e
s
i
n
t
h
e
e
a
r
l
y
d
e
t
e
c
t
i
o
n
o
f
l
u
n
g
c
a
n
c
e
r
u
s
i
n
g
a
n
a
l
y
si
s
o
f
v
o
l
a
t
i
l
e
o
r
g
a
n
i
c
c
o
m
p
o
u
n
d
s
:
F
r
o
m
i
ma
g
i
n
g
t
o
sen
s
o
r
s,”
Asi
a
n
P
a
c
i
f
i
c
J
o
u
r
n
a
l
o
f
C
a
n
c
e
r
Pr
e
v
e
n
t
i
o
n
,
v
o
l
.
1
5
,
n
o
.
1
1
,
p
p
.
4
3
7
7
–
4
3
8
4
,
2
0
1
4
,
d
o
i
:
1
0
.
7
3
1
4
/
A
P
JC
P
.
2
0
1
4
.
1
5
.
1
1
.
4
3
7
7
.
[
3
]
C
.
S
.
D
e
l
a
C
r
u
z
,
L.
T
.
Ta
n
o
u
e
,
a
n
d
R
.
A
.
M
a
t
t
h
a
y
,
“
Lu
n
g
c
a
n
c
e
r
:
e
p
i
d
e
mi
o
l
o
g
y
,
e
t
i
o
l
o
g
y
,
a
n
d
p
r
e
v
e
n
t
i
o
n
,
”
C
l
i
n
i
c
s
i
n
c
h
e
st
m
e
d
i
c
i
n
e
,
v
o
l
.
3
2
,
n
o
.
4
,
p
p
.
6
0
5
–
6
4
4
,
2
0
1
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
c
m.
2
0
1
1
.
0
9
.
0
0
1
.
[
4
]
F
.
B
r
a
y
e
t
a
l
.
,
“
G
l
o
b
a
l
c
a
n
c
e
r
st
a
t
i
st
i
c
s
2
0
2
2
:
G
LO
B
O
C
A
N
e
st
i
ma
t
e
s
o
f
i
n
c
i
d
e
n
c
e
a
n
d
m
o
r
t
a
l
i
t
y
w
o
r
l
d
w
i
d
e
f
o
r
3
6
c
a
n
c
e
r
s
i
n
1
8
5
c
o
u
n
t
r
i
e
s,”
C
A:
A
C
a
n
c
e
r
J
o
u
r
n
a
l
f
o
r
C
l
i
n
i
c
i
a
n
s
,
v
o
l
.
7
4
,
n
o
.
3
,
p
p
.
2
2
9
–
2
6
3
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
2
2
/
c
a
a
c
.
2
1
8
3
4
.
[
5
]
S
.
A
.
B
e
l
i
n
sk
y
e
t
a
l
.
,
“
G
e
n
e
p
r
o
m
o
t
e
r
m
e
t
h
y
l
a
t
i
o
n
i
n
p
l
a
sma
a
n
d
s
p
u
t
u
m
i
n
c
r
e
a
ses
w
i
t
h
l
u
n
g
c
a
n
c
e
r
r
i
sk
,
”
C
l
i
n
i
c
a
l
C
a
n
c
e
r
Re
se
a
rc
h
,
v
o
l
.
1
1
,
n
o
.
1
8
,
p
p
.
6
5
0
5
–
6
5
1
1
,
2
0
0
5
,
d
o
i
:
1
0
.
1
1
5
8
/
1
0
7
8
-
0
4
3
2
.
C
C
R
-
05
-
0
6
2
5
.
[
6
]
V
.
P
.
D
o
r
i
a
-
R
o
se,
P
.
M
.
M
a
r
c
u
s,
E
.
S
z
a
b
o
,
M
.
S
.
T
o
c
k
ma
n
,
M
.
R
.
M
e
l
a
m
e
d
,
a
n
d
P
.
C
.
P
r
o
r
o
k
,
“
R
a
n
d
o
m
i
z
e
d
c
o
n
t
r
o
l
l
e
d
t
r
i
a
l
s
o
f
t
h
e
e
f
f
i
c
a
c
y
o
f
l
u
n
g
c
a
n
c
e
r
scree
n
i
n
g
b
y
s
p
u
t
u
m
c
y
t
o
l
o
g
y
r
e
v
i
si
t
e
d
:
A
c
o
m
b
i
n
e
d
m
o
r
t
a
l
i
t
y
a
n
a
l
y
si
s
f
r
o
m
t
h
e
J
o
h
n
s
H
o
p
k
i
n
s
l
u
n
g
p
r
o
j
e
c
t
a
n
d
t
h
e
m
e
m
o
r
i
a
l
s
l
o
a
n
-
k
e
t
t
e
r
i
n
g
l
u
n
g
st
u
d
y
,
”
C
a
n
c
e
r
,
v
o
l
.
1
1
5
,
n
o
.
2
1
,
p
p
.
5
0
0
7
–
5
0
1
7
,
2
0
0
9
,
d
o
i
:
1
0
.
1
0
0
2
/
c
n
c
r
.
2
4
5
4
5
.
[
7
]
M
.
M
.
O
k
e
n
e
t
a
l
.
,
“
S
c
r
e
e
n
i
n
g
b
y
c
h
e
st
r
a
d
i
o
g
r
a
p
h
a
n
d
l
u
n
g
c
a
n
c
e
r
m
o
r
t
a
l
i
t
y
:
T
h
e
p
r
o
st
a
t
e
,
l
u
n
g
,
c
o
l
o
r
e
c
t
a
l
,
a
n
d
o
v
a
r
i
a
n
(
P
L
C
O
)
r
a
n
d
o
m
i
z
e
d
t
r
i
a
l
,
”
J
a
m
a
,
v
o
l
.
3
0
6
,
n
o
.
1
7
,
p
p
.
1
8
6
5
–
1
8
7
3
,
2
0
1
1
,
d
o
i
:
1
0
.
1
0
0
1
/
j
a
m
a
.
2
0
1
1
.
1
5
9
1
.
[
8
]
Th
e
N
a
t
i
o
n
a
l
L
u
n
g
S
c
r
e
e
n
i
n
g
Tr
i
a
l
R
e
se
a
r
c
h
Te
a
m
,
“
R
e
d
u
c
e
d
l
u
n
g
-
c
a
n
c
e
r
m
o
r
t
a
l
i
t
y
w
i
t
h
l
o
w
-
d
o
s
e
c
o
m
p
u
t
e
d
t
o
m
o
g
r
a
p
h
i
c
scree
n
i
n
g
,
”
N
e
w
En
g
l
a
n
d
J
o
u
rn
a
l
o
f
Me
d
i
c
i
n
e
,
v
o
l
.
3
6
5
,
n
o
.
5
,
p
p
.
3
9
5
–
4
0
9
,
2
0
1
1
,
d
o
i
:
1
0
.
1
0
5
6
/
N
EJ
M
o
a
1
1
0
2
8
7
3
.
[
9
]
K
.
D
a
s
e
t
a
l
.
,
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
i
t
s
a
p
p
l
i
c
a
t
i
o
n
i
n
s
k
i
n
c
a
n
c
e
r
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
En
v
i
r
o
n
m
e
n
t
a
l
Re
s
e
a
r
c
h
a
n
d
P
u
b
l
i
c
H
e
a
l
t
h
,
v
o
l
.
1
8
,
n
o
.
2
4
,
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
i
j
e
r
p
h
1
8
2
4
1
3
4
0
9
.
[
1
0
]
M
.
M
.
A
h
sa
n
e
t
a
l
.
,
“
D
e
e
p
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s
f
o
r
M
o
n
k
e
y
p
o
x
d
i
s
e
a
se
d
i
a
g
n
o
si
s
,
”
Ex
p
e
rt
S
y
st
e
m
s
w
i
t
h
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
2
1
6
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
2
.
1
1
9
4
8
3
.
[
1
1
]
A
.
A
su
n
t
h
a
a
n
d
A
.
S
r
i
n
i
v
a
sa
n
,
“
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
l
u
n
g
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
Mu
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
7
9
,
n
o
.
1
1
,
p
p
.
7
7
3
1
–
7
7
6
2
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
019
-
0
8
3
9
4
-
3
.
[
1
2
]
W
.
J.
S
o
r
i
,
J.
F
e
n
g
,
a
n
d
S
.
Li
u
,
“
M
u
l
t
i
-
p
a
t
h
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
l
u
n
g
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
,
”
Mu
l
t
i
d
i
m
e
n
si
o
n
a
l
S
y
st
e
m
s
a
n
d
S
i
g
n
a
l
Pro
c
e
ss
i
n
g
,
v
o
l
.
3
0
,
p
p
.
1
7
4
9
–
1
7
6
8
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
5
-
0
1
8
-
0
6
2
6
-
9
.
[
1
3
]
L
.
N
.
G
u
m
m
a
,
R
.
T
h
i
r
u
v
e
n
g
a
t
a
n
a
d
h
a
n
,
L
.
K
u
r
a
k
u
l
a
,
a
n
d
T
.
S
i
v
a
p
r
a
k
a
s
a
m
,
“
A
s
u
r
v
e
y
o
n
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
(
d
e
e
p
-
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
)
-
b
a
s
e
d
l
u
n
g
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
,
”
S
N
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
3
,
n
o
.
1
,
p
p
.
1
–
7
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
2
9
7
9
-
021
-
0
0
8
8
7
-
z.
[
1
4
]
H
.
M
.
S
h
y
n
i
a
n
d
E
.
C
h
i
t
r
a
,
“
A
c
o
m
p
a
r
a
t
i
v
e
s
t
u
d
y
o
f
X
-
R
a
y
a
n
d
C
T
i
ma
g
e
s
i
n
C
O
V
I
D
-
1
9
d
e
t
e
c
t
i
o
n
u
si
n
g
i
ma
g
e
p
r
o
c
e
ssi
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s,”
C
o
m
p
u
t
e
r
M
e
t
h
o
d
s
a
n
d
Pro
g
r
a
m
s
i
n
B
i
o
m
e
d
i
c
i
n
e
U
p
d
a
t
e
,
v
o
l
.
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
m
p
b
u
p
.
2
0
2
2
.
1
0
0
0
5
4
.
[
1
5
]
T.
R
a
h
ma
n
e
t
a
l
.
,
“
C
O
V
-
E
C
G
N
ET
:
C
O
V
I
D
-
1
9
d
e
t
e
c
t
i
o
n
u
s
i
n
g
EC
G
t
r
a
c
e
i
m
a
g
e
s
w
i
t
h
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
a
n
d
S
y
st
e
m
s
,
v
o
l
.
1
0
,
n
o
.
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
3
7
5
5
-
0
2
1
-
0
0
1
6
9
-
1.
[
1
6
]
P
.
G
i
f
a
n
i
,
A
.
S
h
a
l
b
a
f
,
a
n
d
M
.
V
a
f
a
e
e
z
a
d
e
h
,
“
A
u
t
o
ma
t
e
d
d
e
t
e
c
t
i
o
n
o
f
C
O
V
I
D
-
1
9
u
si
n
g
e
n
s
e
m
b
l
e
o
f
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
w
i
t
h
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
b
a
se
d
o
n
C
T
sc
a
n
s
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
C
o
m
p
u
t
e
r
Ass
i
st
e
d
Ra
d
i
o
l
o
g
y
a
n
d
S
u
r
g
e
ry
,
v
o
l
.
1
6
,
n
o
.
1
,
p
p
.
1
1
5
–
1
2
3
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
5
4
8
-
0
2
0
-
0
2
2
8
6
-
w.
[
1
7
]
U
.
O
.
D
o
r
j
,
K
.
K
.
Le
e
,
J.
Y
.
C
h
o
i
,
a
n
d
M
.
Le
e
,
“
T
h
e
sk
i
n
c
a
n
c
e
r
c
l
a
s
si
f
i
c
a
t
i
o
n
u
s
i
n
g
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
Mu
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
7
7
,
n
o
.
8
,
p
p
.
9
9
0
9
–
9
9
2
4
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
0
1
8
-
5
7
1
4
-
1.
[
1
8
]
M
.
S
.
A
l
i
,
M
.
S
.
M
i
a
h
,
J
.
H
a
q
u
e
,
M
.
M
.
R
a
h
ma
n
,
a
n
d
M
.
K
.
I
sl
a
m,
“
A
n
e
n
h
a
n
c
e
d
t
e
c
h
n
i
q
u
e
o
f
s
k
i
n
c
a
n
c
e
r
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
w
i
t
h
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
m
o
d
e
l
s,”
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
5
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
ml
w
a
.
2
0
2
1
.
1
0
0
0
3
6
.
[
1
9
]
M
.
P
a
l
,
S
.
M
i
st
r
y
,
a
n
d
D
.
D
e
,
“
I
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
a
p
p
r
o
a
c
h
e
s
o
f
e
x
p
l
a
i
n
a
b
l
e
A
I
i
n
a
n
a
l
y
z
i
n
g
f
e
a
t
u
r
e
s
f
o
r
l
u
n
g
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
,
”
i
n
Fro
n
t
i
e
rs
o
f
I
C
T
i
n
H
e
a
l
t
h
c
a
re
,
S
i
n
g
a
p
o
r
e
:
S
p
r
i
n
g
e
r
,
2
0
2
3
,
p
p
.
2
7
7
–
2
8
7
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
19
-
5
1
9
1
-
6
_
2
3
.
[
2
0
]
M
.
S
.
A
h
me
d
,
K
.
N
.
I
q
b
a
l
,
a
n
d
M
.
G
.
R
.
A
l
a
m,
“
I
n
t
e
r
p
r
e
t
a
b
l
e
l
u
n
g
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
u
s
i
n
g
e
x
p
l
a
i
n
a
b
l
e
A
I
met
h
o
d
s,
”
i
n
2
0
2
3
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
f
o
r
A
d
v
a
n
c
e
m
e
n
t
i
n
T
e
c
h
n
o
l
o
g
y
,
I
C
O
N
AT
2
0
2
3
,
2
0
2
3
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
O
N
A
T5
7
1
3
7
.
2
0
2
3
.
1
0
0
8
0
4
8
0
.
[
2
1
]
M
.
B
h
a
n
d
a
r
i
,
T
.
B
.
S
h
a
h
i
,
B
.
S
i
k
u
,
a
n
d
A
.
N
e
u
p
a
n
e
,
“
E
x
p
l
a
n
a
t
o
r
y
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
C
X
R
i
ma
g
e
s
i
n
t
o
C
O
V
I
D
-
1
9
,
P
n
e
u
m
o
n
i
a
a
n
d
Tu
b
e
r
c
u
l
o
s
i
s
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
X
A
I
,
”
C
o
m
p
u
t
e
rs
i
n
Bi
o
l
o
g
y
a
n
d
Me
d
i
c
i
n
e
,
v
o
l
.
1
5
0
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
mp
b
i
o
m
e
d
.
2
0
2
2
.
1
0
6
1
5
6
.
[
2
2
]
K
.
D
w
i
v
e
d
i
,
A
.
R
a
j
p
a
l
,
S
.
R
a
j
p
a
l
,
M
.
A
g
a
r
w
a
l
,
V
.
K
u
m
a
r
,
a
n
d
N
.
K
u
mar,
“
A
n
e
x
p
l
a
i
n
a
b
l
e
A
I
-
d
r
i
v
e
n
b
i
o
mark
e
r
d
i
sc
o
v
e
r
y
f
r
a
mew
o
r
k
f
o
r
n
o
n
-
sm
a
l
l
c
e
l
l
l
u
n
g
c
a
n
c
e
r
c
l
a
s
si
f
i
c
a
t
i
o
n
,
”
C
o
m
p
u
t
e
r
s
i
n
Bi
o
l
o
g
y
a
n
d
Me
d
i
c
i
n
e
,
v
o
l
.
1
5
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
mp
b
i
o
m
e
d
.
2
0
2
3
.
1
0
6
5
4
4
.
[
2
3
]
W
.
N
.
H
e
n
n
e
s,
“
L
u
n
g
c
a
n
c
e
r
d
a
t
a
se
t
(
I
Q
-
O
TH
/
N
C
C
D
)
,
”
K
a
g
g
l
e
,
2
0
2
2
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m
/
d
a
t
a
se
t
s/
w
a
s
e
e
m
n
a
g
a
h
h
e
n
e
s/
l
u
n
g
-
c
a
n
c
e
r
-
d
a
t
a
set
-
iq
-
o
t
h
n
c
c
d
[
2
4
]
M
.
A
z
i
z
j
o
n
,
A
.
J
u
ma
b
e
k
,
a
n
d
W
.
K
i
m,
“
1
D
C
N
N
b
a
s
e
d
n
e
t
w
o
r
k
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
w
i
t
h
n
o
r
m
a
l
i
z
a
t
i
o
n
o
n
i
m
b
a
l
a
n
c
e
d
d
a
t
a
,
”
i
n
2
0
2
0
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
i
n
I
n
f
o
rm
a
t
i
o
n
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
(
I
C
AI
I
C
)
,
2
0
2
0
,
p
p
.
2
1
8
–
2
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
I
I
C
4
8
5
1
3
.
2
0
2
0
.
9
0
6
4
9
7
6
.
[
2
5
]
L.
H
u
a
n
g
,
X
.
H
o
n
g
,
Z.
Y
a
n
g
,
Y
.
Li
u
,
a
n
d
B
.
Z
h
a
n
g
,
“
C
N
N
-
LST
M
n
e
t
w
o
r
k
-
b
a
se
d
d
a
ma
g
e
d
e
t
e
c
t
i
o
n
a
p
p
r
o
a
c
h
f
o
r
c
o
p
p
e
r
p
i
p
e
l
i
n
e
u
si
n
g
l
a
ser
u
l
t
r
a
s
o
n
i
c
sc
a
n
n
i
n
g
,
”
U
l
t
r
a
so
n
i
c
s
,
v
o
l
.
1
2
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
u
l
t
r
a
s.
2
0
2
2
.
1
0
6
6
8
5
.
[
2
6
]
M
.
M
a
f
t
o
u
n
i
,
A
.
C
.
C
.
L
a
w
,
B
.
S
h
e
n
,
Z.
K
.
G
r
a
d
o
,
Y
.
Z
h
o
u
,
a
n
d
N
.
A
.
Y
a
z
d
i
,
“
A
r
o
b
u
s
t
e
n
s
e
m
b
l
e
-
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
f
o
r
C
O
V
I
D
-
1
9
d
i
a
g
n
o
s
i
s
b
a
s
e
d
o
n
a
n
i
n
t
e
g
r
a
t
e
d
C
T
s
c
a
n
i
m
a
g
e
s
d
a
t
a
b
a
s
e
,
”
i
n
I
I
S
E
A
n
n
u
a
l
C
o
n
f
e
r
e
n
c
e
a
n
d
E
x
p
o
2
0
2
1
,
2
0
2
1
,
p
p
.
6
3
2
–
637.
Evaluation Warning : The document was created with Spire.PDF for Python.