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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
,
p
p
.
4
5
6
8
~
4
5
9
1
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
5
.
p
p
4
5
6
8
-
4
5
9
1
4568
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Q
ua
nti
tativ
e ches
t
co
mput
ed t
o
mo
g
ra
phy
-
ba
sed hy
brid dee
p
lea
rning
f
o
r ch
ro
nic lung
disea
se
d
etec
tion
and sta
g
i
ng
G
a
rim
a
J
a
in,
Am
it
K
um
a
r
G
o
el,
Ra
hu
l K
um
a
r
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
i
n
g
a
n
d
I
n
f
o
r
m
a
t
i
c
s,
S
c
h
o
o
l
o
f
En
g
i
n
e
e
r
i
n
g
a
n
d
S
c
i
e
n
c
e
s
,
S
i
r
P
a
d
a
m
p
a
t
S
i
n
g
h
a
n
i
a
U
n
i
v
e
r
s
i
t
y
,
U
d
a
i
p
u
r
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
2
0
,
2
0
2
5
R
ev
is
ed
J
u
l 1
8
,
2
0
2
6
Acc
ep
ted
Au
g
1
2
,
2
0
2
6
Th
is
re
se
a
rc
h
p
ro
p
o
se
s
a
h
y
b
rid
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l
t
o
a
c
c
u
ra
t
e
ly
d
e
tec
t
c
h
ro
n
ic
lu
n
g
d
ise
a
se
s
in
c
h
e
st
c
o
m
p
u
te
d
to
m
o
g
ra
p
h
y
(CT)
sc
a
n
s.
Cu
rre
n
t
d
e
e
p
lea
rn
in
g
m
e
th
o
d
s,
s
u
c
h
a
s
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(
CNN
s)
a
n
d
tran
sfo
rm
e
r
m
o
d
e
ls,
o
ften
su
ffe
r
fro
m
h
ig
h
c
o
m
p
u
tati
o
n
a
l
re
q
u
ire
m
e
n
ts
a
n
d
lo
w
in
ter
p
re
tab
il
it
y
.
T
h
is
re
se
a
rc
h
c
o
m
b
i
n
e
s
a
p
re
-
train
e
d
re
sid
u
a
l
n
e
two
rk
5
0
la
y
e
rs
(
Re
sN
e
t5
0
)
fo
r
d
e
e
p
f
e
a
tu
re
e
x
trac
ti
o
n
wi
th
h
a
n
d
-
c
ra
ft
e
d
tex
tu
r
e
fe
a
tu
re
s.
Th
e
fe
a
tu
re
s
a
re
c
o
m
b
in
e
d
v
ia
h
y
b
ri
d
c
o
n
c
a
ten
a
ti
o
n
,
a
n
d
th
e
m
o
d
e
l
is
c
las
sified
u
sin
g
a
n
o
p
ti
m
ize
d
e
x
trem
e
g
ra
d
ien
t
b
o
o
stin
g
(
XG
Bo
o
st)
e
n
se
m
b
le
with
h
y
p
e
r
p
a
ra
m
e
ter
tu
n
i
n
g
t
o
im
p
ro
v
e
p
e
rfo
rm
a
n
c
e
a
n
d
re
d
u
c
e
o
v
e
rfit
ti
n
g
.
T
h
is
re
se
a
rc
h
c
o
m
p
a
re
d
th
e
p
r
o
p
o
se
d
m
o
d
e
l
with
ra
d
io
m
ics
-
o
n
l
y
,
Re
sN
e
t5
0
CNN
,
v
isi
o
n
tran
sfo
rm
e
r
(ViT
-
B1
6
),
a
n
d
S
wi
n
tr
a
n
sfo
rm
e
r
m
e
th
o
d
s.
Th
is
re
se
a
rc
h
u
se
d
a
p
u
b
li
c
d
a
tas
e
t
c
o
n
tai
n
in
g
a
d
e
n
o
c
a
rc
in
o
m
a
,
sq
u
a
m
o
u
s
c
e
ll
c
a
rc
in
o
m
a
,
larg
e
c
e
ll
c
a
rc
in
o
m
a
,
a
n
d
n
o
rm
a
l
c
las
se
s.
Th
e
c
las
sifica
ti
o
n
a
c
c
u
ra
c
y
o
f
th
e
h
y
b
ri
d
fra
m
e
wo
rk
wa
s
9
2
.
3
8
%
,
with
b
e
tt
e
r
p
re
c
isio
n
,
se
n
sit
iv
it
y
,
sp
e
c
ifi
c
it
y
,
F
1
-
sc
o
re
,
a
n
d
re
c
e
iv
e
r
o
p
e
ra
ti
n
g
c
h
a
ra
c
teristic
-
a
re
a
u
n
d
e
r
t
h
e
c
u
r
v
e
(ROC
-
AU
C)
th
a
n
t
h
e
b
a
se
li
n
e
s.
A
d
u
a
l
-
m
o
d
e
e
x
p
lain
a
b
le
a
rti
ficia
l
i
n
telli
g
e
n
c
e
(XA
I)
fra
m
e
wo
rk
i
n
teg
ra
ti
n
g
g
ra
d
ien
t
-
we
ig
h
ted
c
las
s
a
c
ti
v
a
ti
o
n
m
a
p
p
i
n
g
(
G
ra
d
-
CAM
)
a
n
d
S
h
a
p
ley
a
d
d
it
i
v
e
e
x
p
lan
a
ti
o
n
(S
HA
P
)
w
a
s
imp
lem
e
n
ted
to
e
n
a
b
le
sp
a
ti
a
l
les
io
n
lo
c
a
li
z
a
ti
o
n
a
n
d
fe
a
tu
re
a
tt
rib
u
ti
o
n
fo
r
c
li
n
ica
l
tran
sp
a
re
n
c
y
.
G
ra
d
-
CAM
wa
s
a
b
le
to
lo
c
a
li
z
e
th
e
les
i
o
n
s
su
c
c
e
ss
fu
ll
y
,
a
n
d
S
HA
P
re
v
e
a
led
th
e
i
n
f
lu
e
n
c
e
o
f
e
n
tro
p
y
-
b
a
se
d
tex
tu
re
fe
a
tu
re
s
a
n
d
se
m
a
n
ti
c
e
m
b
e
d
d
i
n
g
s
o
n
th
e
p
r
e
d
ictio
n
s.
Ov
e
ra
ll
,
th
e
fra
m
e
wo
rk
d
e
v
e
lo
p
s
a
d
v
a
n
c
e
d
XA
I
sy
ste
m
s
fo
r
c
h
e
st
CT
a
n
a
ly
sis.
K
ey
w
o
r
d
s
:
C
h
est co
m
p
u
ted
to
m
o
g
r
ap
h
y
E
x
p
lain
ab
le
ar
tific
ial
in
tellig
en
ce
E
x
tr
em
e
g
r
a
d
ien
t b
o
o
s
tin
g
Featu
r
e
-
lev
el
f
u
s
io
n
Gr
ad
ien
t
-
weig
h
ted
class
ac
tiv
atio
n
m
ap
p
i
n
g
R
ad
io
m
ics
Sh
ap
ley
ad
d
itiv
e
e
x
p
lan
atio
n
s
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
:
Gar
im
a
J
ain
Dep
ar
tm
en
t o
f
C
o
m
p
u
tin
g
an
d
I
n
f
o
r
m
atics,
Sch
o
o
l o
f
E
n
g
in
ee
r
in
g
an
d
Scien
ce
s
Sir
Pad
am
p
at
Sin
g
h
an
ia
Un
i
v
er
s
ity
B
h
atew
ar
,
Ud
aip
u
r
-
3
1
3
6
0
1
,
I
n
d
ia
E
m
ail:
g
ar
im
a.
p
h
d
2
0
2
3
@
s
p
s
u
.
ac
.
in
1.
I
NT
RO
D
UCT
I
O
N
L
u
n
g
ca
n
ce
r
is
s
till
o
n
e
o
f
th
e
m
ain
ca
u
s
es
o
f
ca
n
ce
r
-
r
elate
d
d
ea
th
s
ar
o
u
n
d
th
e
g
l
o
b
e
,
an
d
its
p
u
b
lic
h
ea
lth
im
p
ac
t
r
em
ain
s
s
ig
n
if
ican
t,
d
esp
ite
ad
v
an
ce
m
en
ts
m
ad
e
in
m
ed
ical
im
ag
i
n
g
an
d
tr
ea
tm
en
t
m
eth
o
d
s
[
1
]
,
[
2
]
.
Acc
u
r
ate
an
d
ea
r
ly
d
etec
tio
n
o
f
lu
n
g
n
o
d
u
les
f
r
o
m
co
m
p
u
ted
to
m
o
g
r
a
p
h
y
(
CT
)
im
ag
es
is
cr
itical
f
o
r
p
r
o
g
n
o
s
is
,
tim
ely
in
ter
v
en
tio
n
,
m
itig
atin
g
d
is
ea
s
e
p
r
o
g
r
ess
io
n
,
an
d
p
e
r
s
o
n
alizin
g
tr
ea
tm
en
t;
h
o
wev
er
,
d
u
e
to
o
v
er
lap
p
in
g
v
is
u
al
ch
ar
ac
ter
is
tics
,
s
u
b
tle
d
if
f
er
en
ce
s
in
tex
tu
r
e
,
an
d
t
im
e
f
r
o
m
th
e
f
ir
s
t
d
iag
n
o
s
is
,
it
ca
n
b
e
d
if
f
icu
lt f
o
r
clin
ician
s
to
ac
c
u
r
ately
d
is
tin
g
u
is
h
b
etwe
en
ce
r
tain
ty
p
es o
f
lu
n
g
tu
m
o
r
s
s
u
ch
as
ad
en
o
ca
r
cin
o
m
as,
s
q
u
am
o
u
s
ce
ll
ca
r
cin
o
m
as,
an
d
lar
g
e
ce
ll
ca
r
cin
o
m
as
[
3
]
.
W
ith
th
e
g
r
o
win
g
u
s
e
o
f
ar
tific
ial
in
tellig
en
ce
(
AI
)
-
b
a
s
ed
co
m
p
u
ter
-
ass
is
ted
d
iag
n
o
s
is
(
C
AD)
s
y
s
tem
s
in
clin
i
ca
l
p
r
ac
tice,
r
ec
en
t
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
Qu
a
n
tita
tive
ch
est co
mp
u
ted
to
mo
g
r
a
p
h
y
-
b
a
s
ed
h
yb
r
id
d
ee
p
lea
r
n
in
g
fo
r
ch
r
o
n
ic
l
u
n
g
…
(
Ga
r
ima
Ja
in
)
4569
s
tu
d
ies
h
av
e
s
h
o
w
n
th
at
C
AD
s
y
s
tem
s
m
ay
b
e
ef
f
ec
ti
v
e
to
o
ls
t
o
ass
is
t
r
ad
io
lo
g
is
t
s
in
th
e
au
to
m
ated
class
if
icatio
n
o
f
lu
n
g
d
is
ea
s
e
b
y
way
o
f
AI
.
Sev
er
al
d
ee
p
lear
n
in
g
-
b
ased
m
eth
o
d
s
,
in
clu
d
in
g
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs),
h
av
e
b
ee
n
u
s
ed
s
u
cc
ess
f
u
lly
f
o
r
th
e
ex
tr
ac
tio
n
o
f
h
ier
ar
c
h
ical
s
p
atial
r
ep
r
esen
tatio
n
s
f
r
o
m
C
T
im
ag
es
[
4
]
,
[
5
]
.
Va
r
io
u
s
C
NN
ar
ch
itectu
r
es
i.e
.
,
r
esid
u
al
n
etwo
r
k
(
R
esNet
)
,
d
en
s
ely
co
n
n
ec
te
d
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
(
De
n
s
eNe
t
)
,
ef
f
icien
t
n
etwo
r
k
(
E
f
f
ici
en
tNet
)
,
an
d
m
o
b
ile
n
etwo
r
k
(
Mo
b
ileNet
)
,
h
a
v
e
ac
ce
ler
ated
th
e
lear
n
in
g
o
f
a
u
to
m
ated
f
e
atu
r
es
in
th
e
a
n
aly
s
is
o
f
th
o
r
ac
ic
im
a
g
es
[
6
]
–
[
8
]
.
Mo
r
e
r
ec
e
n
t
in
ter
est
h
as
also
b
ee
n
g
en
er
ated
with
r
esp
ec
t
to
tr
an
s
f
o
r
m
er
-
b
ased
v
is
io
n
m
o
d
els
i.e
.
,
v
is
io
n
tr
a
n
s
f
o
r
m
er
s
(
ViT
)
an
d
Swin
tr
an
s
f
o
r
m
er
s
as
th
ese
m
o
d
els
af
f
o
r
d
th
e
ca
p
ab
ilit
y
o
f
lear
n
in
g
lo
n
g
-
r
an
g
e
co
n
tex
t
u
a
l
d
ep
en
d
e
n
cies
an
d
g
lo
b
ally
m
o
d
elin
g
im
a
g
es
[
9
]
,
[
1
0
]
.
E
v
e
n
t
h
o
u
g
h
th
ese
a
r
ch
itectu
r
es
h
a
v
e
d
eliv
er
ed
p
r
o
m
is
in
g
r
esu
lts
in
ter
m
s
o
f
t
h
eir
class
if
icatio
n
a
cc
u
r
ac
ies,
th
er
e
r
em
ain
s
a
n
a
b
u
n
d
a
n
ce
o
f
c
u
r
r
en
t
s
y
s
tem
s
th
at
co
n
tin
u
e
to
b
e
co
m
p
u
tatio
n
ally
d
em
an
d
in
g
,
a
r
e
n
o
t
clin
ically
in
ter
p
r
eta
b
le
,
an
d
ar
e
d
ep
e
n
d
en
t
o
n
lar
g
e
n
u
m
b
er
s
o
f
tr
ain
i
n
g
d
ata
[
1
1
]
.
At
th
e
s
am
e
tim
e,
r
a
d
io
m
ics
-
b
ased
m
eth
o
d
s
h
av
e
a
ls
o
s
tar
ted
to
b
ec
o
m
e
r
ec
o
g
n
iz
ed
as a
n
im
p
o
r
tan
t
an
d
co
m
p
lem
e
n
tar
y
way
o
f
r
e
p
r
esen
tin
g
q
u
a
n
titativ
e
d
ata
with
in
th
e
co
n
tex
t
o
f
m
e
d
ical
im
ag
in
g
.
R
ad
io
m
ics
p
r
o
v
id
es
m
eth
o
d
s
o
f
e
x
tr
ac
ti
n
g
h
a
n
d
-
c
r
af
ted
d
escr
ip
to
r
s
f
r
o
m
C
T
s
ca
n
s
th
at
d
escr
ib
e
p
r
o
p
er
ties
,
s
u
c
h
as
tex
tu
r
e
h
eter
o
g
en
eity
,
in
ten
s
i
ty
d
is
tr
ib
u
tio
n
s
,
s
h
ap
es,
a
n
d
s
tr
u
ctu
r
al
s
h
ap
es
o
f
tis
s
u
es.
T
h
ese
q
u
an
titativ
e
b
io
m
ar
k
e
r
s
ar
e
o
f
ten
u
s
ed
to
d
ef
in
e
lay
er
s
o
f
tis
s
u
e
th
at
h
av
e
clin
ical
r
elev
an
ce
b
u
t
c
an
n
o
t
b
e
s
ee
n
wh
en
d
ev
elo
p
in
g
d
o
m
ain
-
s
p
ec
if
ic
o
r
d
ee
p
laten
t
r
e
p
r
esen
tatio
n
s
.
I
n
ad
d
itio
n
to
th
e
lim
itatio
n
s
o
f
c
o
n
v
e
n
tio
n
a
l
r
ad
io
m
ics p
ip
elin
es,
in
d
ep
en
d
en
t
d
ee
p
lear
n
in
g
m
o
d
els ten
d
to
ac
t a
s
“
b
lack
b
o
x
es
”
lim
itin
g
clin
ician
s
'
ab
ilit
y
to
in
ter
p
r
et
th
e
m
an
d
th
eir
tr
u
s
t in
th
e
o
u
t
p
u
t a
cc
u
r
a
cy
.
Nu
m
er
o
u
s
s
tu
d
ies
th
at
h
av
e
i
n
teg
r
ated
d
ee
p
lear
n
in
g
with
r
ad
io
m
ics
to
ass
ess
p
u
lm
o
n
ar
y
d
is
ea
s
e
h
av
e
u
s
ed
v
ar
io
u
s
tech
n
iq
u
es
to
p
r
o
v
id
e
in
s
ig
h
ts
in
to
p
u
l
m
o
n
ar
y
d
is
ea
s
es;
h
o
wev
er
,
th
er
e
ar
e
s
till
s
o
m
e
cr
itical
lim
itatio
n
s
in
th
ese
attem
p
ts
.
An
o
v
er
wh
elm
i
n
g
n
u
m
b
er
o
f
ex
is
tin
g
f
r
am
ewo
r
k
s
in
cu
r
h
ig
h
co
m
p
u
tatio
n
al
co
s
ts
ass
o
ciate
d
with
th
ei
r
m
u
lti
-
s
tag
e
ar
ch
itectu
r
es,
f
ea
tu
r
e
o
p
tim
izatio
n
p
r
o
ce
d
u
r
es
,
an
d
tr
an
s
f
o
r
m
er
-
b
ased
a
r
ch
itectu
r
e
s
th
at
m
ay
n
o
t
m
a
k
e
th
e
m
p
r
a
ctica
l
f
o
r
r
o
u
tin
e
clin
ical
u
s
e
[
1
2
]
–
[
1
4
]
.
Ma
n
y
o
f
th
e
ex
is
tin
g
s
tu
d
ies
o
n
co
m
p
u
ter
-
b
ased
an
al
y
s
is
h
av
e
f
o
cu
s
ed
p
r
im
ar
ily
o
n
b
i
n
ar
y
c
lass
if
icatio
n
task
s
.
T
h
er
e
is
also
lim
ited
r
esear
ch
av
ailab
le
th
at
an
aly
ze
s
th
e
cli
n
ical
asp
ec
ts
o
f
m
u
lti
-
class
d
i
s
ea
s
e
d
if
f
er
en
tiatio
n
with
r
esp
ec
t
to
f
alse
n
eg
ativ
e
an
aly
s
is
,
ea
r
ly
-
s
tag
e
lesi
o
n
s
en
s
itiv
ity
,
co
m
p
u
tatio
n
al
f
ea
s
ib
ilit
y
,
ex
p
lain
ab
ilit
y
v
alid
atio
n
,
an
d
d
em
o
g
r
a
p
h
ic
r
o
b
u
s
tn
ess
as
well
as
h
o
w
t
h
ey
ca
n
b
e
d
ep
lo
y
e
d
with
in
a
p
r
ac
tical
r
ad
io
lo
g
y
s
ettin
g
.
T
o
ad
d
r
ess
th
ese
lim
itatio
n
s
,
t
h
is
s
tu
d
y
p
r
esen
ts
a
h
y
b
r
id
d
e
ep
lear
n
in
g
f
r
am
ewo
r
k
in
a
q
u
an
titativ
e
asp
ec
t
f
o
r
m
u
ltip
le
class
if
icatio
n
an
d
d
etec
tin
g
c
h
r
o
n
ic
lu
n
g
d
is
ea
s
e
th
r
o
u
g
h
c
h
est
C
T
u
tili
zin
g
b
o
th
d
ee
p
s
p
atial
em
b
ed
d
in
g
an
d
r
ad
io
m
ics
tex
tu
r
e
d
escr
ip
to
r
s
with
th
e
u
s
e
o
f
h
an
d
c
r
af
ted
s
tatis
tical
p
r
o
p
er
ties
.
T
o
d
ev
elo
p
th
is
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
,
a
R
esNet5
0
p
r
e
-
tr
ain
ed
b
ac
k
b
o
n
e
ar
c
h
itectu
r
e
will
b
e
u
s
ed
to
ex
tr
ac
t
h
ig
h
-
d
im
e
n
s
io
n
al
s
p
atial
r
ep
r
esen
tatio
n
s
f
r
o
m
ch
est
C
T
im
ag
es.
Par
allel
to
f
ea
tu
r
e
e
x
tr
a
ctio
n
p
er
f
o
r
m
ed
b
y
th
e
R
esNet5
0
n
etwo
r
k
,
th
e
cl
ass
if
icatio
n
will
em
p
lo
y
a
s
ec
o
n
d
b
r
an
ch
o
f
r
ad
i
o
m
ics
wh
er
eb
y
th
e
co
m
p
u
ted
tis
s
u
e
h
eter
o
g
en
eity
will
b
e
d
escr
ib
ed
u
s
in
g
g
r
ay
-
lev
el
co
-
o
cc
u
r
r
en
ce
m
atr
ix
(
GL
C
M)
s
tatis
tics
an
d
en
tr
o
p
y
-
b
ased
an
aly
s
es.
T
h
ese
m
u
tu
ally
co
m
p
lem
e
n
tar
y
f
e
atu
r
e
r
ep
r
esen
tatio
n
s
will
th
en
b
e
f
u
s
ed
to
g
eth
e
r
o
n
a
f
ea
tu
r
e
lev
el
b
ef
o
r
e
b
ein
g
class
if
ied
u
s
in
g
an
o
p
t
im
ized
ex
tr
em
e
g
r
a
d
ien
t
b
o
o
s
tin
g
(
XGBo
o
s
t
)
en
s
em
b
le
class
if
ier
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
l
o
g
y
is
d
is
tin
ct
f
r
o
m
th
at
o
f
m
an
y
t
r
a
d
itio
n
al
d
ee
p
lear
n
in
g
a
p
p
r
o
ac
h
es
in
th
at
it
f
ea
tu
r
es
n
o
t
o
n
ly
n
o
n
-
d
e
ter
m
in
is
tic
laten
t
em
b
ed
d
in
g
s
as
in
p
u
t
to
th
e
class
if
ier
b
u
t
will
in
clu
d
e
d
eter
m
in
is
tic
r
ad
io
m
ic
f
ea
tu
r
e
s
d
ir
ec
tly
v
ia
th
eir
in
clu
s
io
n
i
n
th
e
o
v
e
r
all
f
ea
t
u
r
e
s
et
,
th
er
e
b
y
e
n
h
an
cin
g
b
o
th
d
is
cr
im
in
atio
n
ca
p
ab
ilit
y
a
n
d
in
ter
p
r
etab
ilit
y
.
T
h
e
g
o
al
o
f
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
th
r
o
u
g
h
s
y
s
tem
atic
g
r
id
s
ea
r
ch
a
n
aly
s
is
was
ac
h
iev
ed
b
y
b
ein
g
ab
le
to
o
b
tai
n
m
o
d
el
s
tab
ilit
y
an
d
r
o
b
u
s
tn
ess
ac
r
o
s
s
v
ar
io
u
s
ex
p
er
im
en
tal
co
n
f
ig
u
r
atio
n
s
.
I
n
ad
d
itio
n
to
cr
ea
tin
g
b
aseli
n
es
with
o
n
ly
C
NN
s
,
we
b
e
n
ch
m
ar
k
e
d
ag
ain
s
t
tr
an
s
f
o
r
m
er
-
lik
e
ar
ch
itectu
r
es
s
u
ch
as
th
e
ViT
an
d
Swin
tr
an
s
f
o
r
m
er
to
m
ea
s
u
r
e
h
o
w
well
th
ey
class
if
y
im
ag
es a
n
d
g
e
n
er
alize
b
ey
o
n
d
th
eir
tr
ain
in
g
s
ets.
T
h
is
wo
r
k
'
s
ad
d
itio
n
al
co
n
tr
ib
u
tio
n
is
th
e
u
s
e
o
f
a
d
u
al
-
m
o
d
e
ex
p
lain
ab
le
ar
tific
ial
in
tellig
en
ce
(
XAI
)
f
r
am
ewo
r
k
th
at
en
h
an
ce
s
clin
ical
tr
an
s
p
ar
en
c
y
an
d
p
h
y
s
ician
tr
u
s
t.
T
h
is
wo
r
k
em
p
lo
y
g
r
ad
ien
t
-
weig
h
ted
class
ac
tiv
atio
n
m
ap
p
in
g
(
Gr
ad
-
C
AM
)
-
b
ased
v
is
u
aliza
tio
n
to
l
o
ca
te
th
e
d
iag
n
o
s
tically
im
p
o
r
tan
t
p
u
lm
o
n
a
r
y
r
eg
i
o
n
s
an
d
u
tili
ze
Sh
ap
ley
a
d
d
itiv
e
ex
p
lan
atio
n
s
(
SHAP)
to
q
u
an
tify
b
o
th
d
ee
p
em
b
ed
d
in
g
s
an
d
h
an
d
-
cr
af
te
d
r
ad
io
m
ic
d
escr
ip
to
r
s
'
co
n
tr
ib
u
tio
n
s
at
th
e
f
ea
tu
r
e
lev
el
[
1
5
]
.
T
h
e
jo
in
t
ex
p
lain
ab
ilit
y
m
ec
h
a
n
is
m
allo
ws
f
o
r
s
p
atial
an
d
q
u
an
titativ
e
in
ter
p
r
etatio
n
s
o
f
th
e
d
iag
n
o
s
t
ic
p
r
o
ce
s
s
,
th
er
eb
y
p
r
o
m
o
tin
g
tr
an
s
p
ar
e
n
cy
in
th
e
AI
-
ass
is
ted
clin
ical
d
ec
is
io
n
-
m
ak
in
g
p
r
o
ce
s
s
[
1
6
]
.
A
m
u
lti
-
class
ch
est
C
T
d
ataset
co
n
s
is
tin
g
o
f
n
o
r
m
al,
ad
en
o
ca
r
cin
o
m
a,
s
q
u
am
o
u
s
ce
ll
ca
r
cin
o
m
a,
a
n
d
lar
g
e
ce
ll
ca
r
cin
o
m
a
ca
teg
o
r
ies
is
ass
e
s
s
ed
u
s
in
g
s
tr
atif
ied
tr
ain
in
g
-
v
alid
atio
n
-
test
in
g
s
p
lits
,
th
en
ev
alu
ated
u
s
in
g
class
if
icatio
n
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e
,
s
en
s
itiv
ity
,
ar
ea
u
n
d
e
r
t
h
e
c
u
r
v
e
(
AUC
)
b
y
r
ec
eiv
er
o
p
er
atin
g
ch
a
r
ac
ter
is
tic
(
R
OC
)
cu
r
v
es,
an
d
co
n
f
u
s
io
n
m
atr
ix
an
aly
s
es.
Ad
d
itio
n
ally
,
to
ev
alu
ate
th
e
p
r
ac
ticality
o
f
im
p
lem
en
tin
g
th
is
p
r
o
p
o
s
ed
m
o
d
el
in
f
u
tu
r
e
r
a
d
io
lo
g
ical
s
cr
ee
n
i
n
g
s
y
s
tem
s
,
telem
ed
icin
e
s
y
s
tem
s
,
an
d
in
tellig
en
t
clin
ical
d
ec
is
io
n
s
u
p
p
o
r
t sy
s
tem
s
,
b
o
th
co
m
p
u
tatio
n
al
ef
f
icien
cy
an
d
ex
p
lain
ab
ilit
y
an
aly
s
es we
r
e
p
er
f
o
r
m
e
d
.
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
5
6
8
-
4
5
9
1
4570
T
h
is
s
tu
d
y
estab
lis
h
es
th
e
f
o
llo
win
g
k
ey
co
n
t
r
ib
u
tio
n
s
:
i)
a
h
y
b
r
id
f
ea
tu
r
e
-
lev
el
f
u
s
io
n
f
r
am
ewo
r
k
th
at
in
co
r
p
o
r
ates
d
ee
p
C
NN
s
p
atial
em
b
ed
d
in
g
s
with
tr
ad
itio
n
al
h
an
d
cr
a
f
ted
r
ad
i
o
m
ic
tex
tu
r
e
f
ea
tu
r
es
to
f
ac
ilit
ate
th
e
class
if
icat
io
n
o
f
C
T
im
ag
es
o
f
th
e
ch
es
t
in
to
m
u
ltip
le
ca
teg
o
r
ies
;
ii)
a
co
m
p
ar
ativ
e
b
en
ch
m
ar
k
in
g
with
o
th
er
d
e
ep
lear
n
in
g
m
o
d
els,
s
u
ch
as
th
e
ViT
,
Swin
tr
an
s
f
o
r
m
er
,
an
d
C
NN
s
u
ch
as
R
esNet,
o
n
ly
o
n
m
u
ltiv
ar
iat
e
p
er
f
o
r
m
a
n
ce
m
et
r
ics
;
iii)
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
an
d
co
m
p
u
tatio
n
al
ef
f
icien
cy
a
n
aly
s
is
f
o
r
im
p
r
o
v
ed
r
ep
r
o
d
u
cib
ilit
y
,
r
o
b
u
s
tn
ess
,
an
d
ea
s
e
o
f
d
ep
l
o
y
m
en
t
;
i
v
)
t
h
e
in
teg
r
atio
n
o
f
a
n
XAI
f
r
am
ewo
r
k
th
r
o
u
g
h
b
o
t
h
Gr
ad
-
C
AM
an
d
SHAP
f
o
r
s
p
atial
-
lev
el
an
d
f
ea
tu
r
e
-
lev
el
in
ter
p
r
etatio
n
o
f
r
esu
lts
;
an
d
v
)
a
th
o
r
o
u
g
h
e
v
alu
atio
n
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
with
r
esp
ec
t
to
m
u
ltip
le
class
if
icatio
n
m
etr
ics,
ab
latio
n
an
aly
s
is
,
R
OC
-
AUC an
aly
s
i
s
,
an
d
clin
ically
in
ter
p
r
eta
b
le
ex
p
lan
atio
n
s
o
f
r
esu
lts
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
I
n
t
h
is
w
o
r
k
by
T
h
an
o
o
n
et
a
l.
[
1
7
]
d
ee
p
lea
r
n
i
n
g
–
b
as
e
d
m
eth
o
d
s
f
o
r
C
T
im
a
g
i
n
g
o
f
t
h
e
ch
est
f
o
r
lu
n
g
ca
n
c
e
r
s
c
r
ee
n
i
n
g
a
n
d
d
ia
g
n
o
s
is
ar
e
p
r
ese
n
te
d
.
T
h
e
a
u
t
h
o
r
s
o
r
g
an
iz
e
a
n
d
co
m
p
a
r
e
v
a
r
i
o
u
s
d
e
e
p
le
ar
n
i
n
g
tec
h
n
i
q
u
es
i
n
c
lu
d
i
n
g
(
b
u
t n
o
t l
im
it
e
d
t
o
)
C
NN,
r
ec
u
r
r
e
n
t
n
e
u
r
al
n
etw
o
r
k
s
(
R
NNs
)
as
we
ll
a
s
n
e
w
h
y
b
r
i
d
-
b
as
e
d
ar
ch
ite
ct
u
r
es
t
h
a
t
h
a
v
e
b
ee
n
p
r
o
p
o
s
ed
d
u
r
i
n
g
t
h
e
last
h
al
f
-
d
e
ca
d
e
(
2
0
1
8
–
2
0
2
2
)
b
y
g
r
o
u
p
i
n
g
d
i
f
f
er
e
n
t
m
et
h
o
d
s
o
n
cl
ass
i
f
ic
ati
o
n
/s
e
g
m
e
n
t
ati
o
n
.
T
h
is
r
e
v
i
ew
h
i
g
h
li
g
h
ts
t
h
e
p
o
te
n
ti
al
o
f
d
ee
p
l
ea
r
n
i
n
g
f
o
r
ad
d
r
ess
i
n
g
d
i
v
e
r
s
e
ty
p
es
o
f
i
n
p
u
ts
(
2
D,
3
D
,
a
n
d
l
o
w
-
d
o
s
e
C
T
)
b
u
t
als
o
c
r
it
ic
all
y
e
v
al
u
at
es
p
r
o
b
l
em
s
f
r
o
m
d
a
ta
d
i
v
e
r
s
it
y
,
i
m
a
g
e
ar
t
if
ac
ts
t
o
la
c
k
o
f
l
a
b
el
e
d
d
atas
ets.
M
o
r
e
o
v
e
r
,
t
h
e
p
a
p
e
r
als
o
i
d
e
n
ti
f
i
es
p
o
t
en
tia
l
r
es
e
ar
ch
a
r
e
as
s
u
ch
as
s
tan
d
ar
d
i
ze
d
p
r
e
-
p
r
o
ce
s
s
i
n
g
an
d
in
te
g
r
ati
o
n
o
f
cli
n
i
ca
l
d
a
ta
t
o
e
n
h
a
n
c
e
au
to
m
a
te
d
tu
m
o
r
s
e
g
m
e
n
ta
ti
o
n
.
Ozd
em
ir
et
a
l.
[
1
8
]
p
r
o
p
o
s
ed
a
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
el
f
o
r
lu
n
g
ca
n
ce
r
d
etec
tio
n
u
s
in
g
C
NNs,
b
ased
o
n
I
n
ce
p
tio
n
NeX
t
b
lo
c
k
s
an
d
ViT
,
u
s
in
g
g
r
i
d
an
d
b
l
o
ck
atten
tio
n
m
eth
o
d
s
.
T
h
is
m
o
d
el
is
d
esig
n
ed
to
ca
p
tu
r
e
lo
ca
l
d
etailed
in
f
o
r
m
a
tio
n
an
d
o
v
er
all
c
o
n
tex
tu
al
d
at
a
f
r
o
m
C
T
s
ca
n
s
o
f
th
e
lu
n
g
s
s
o
it
ca
n
b
e
u
s
ed
in
a
clin
ical
s
ettin
g
.
T
h
e
m
o
d
el
was
tr
ain
ed
u
s
in
g
two
d
if
f
er
en
t
p
u
b
lic
d
atasets
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
(
I
Q
-
OT
H/NC
C
D)
d
ataset
an
d
th
e
c
h
est
C
T
d
ataset.
T
h
e
m
o
d
el
was
test
ed
ag
ain
s
t
b
o
th
d
atasets
an
d
ac
h
iev
ed
an
ac
c
u
r
ac
y
o
f
9
9
.
5
4
%
ac
cu
r
ately
an
d
9
8
.
4
1
%
ac
cu
r
ately
,
r
esp
ec
tiv
ely
.
I
t
is
im
p
o
r
tan
t
to
n
o
te
th
at
th
e
au
th
o
r
s
wer
e
ab
le
to
ac
h
iev
e
th
is
s
tate
-
of
-
th
e
-
ar
t
r
esu
lt
with
an
ef
f
icien
t
an
d
lig
h
tweig
h
t
m
o
d
el
d
esig
n
o
f
o
n
ly
1
8
.
1
m
illi
o
n
p
ar
am
eter
s
.
I
n
ad
d
itio
n
to
cr
ea
tin
g
an
ef
f
ec
tiv
e
class
if
ier
,
th
e
au
th
o
r
s
u
s
ed
G
r
ad
-
C
AM
to
cr
ea
te
h
ea
tm
ap
s
o
f
d
if
f
e
r
en
t
ar
ea
s
o
f
th
e
lu
n
g
s
th
at
wer
e
b
ein
g
em
p
h
asized
o
r
d
e
-
em
p
h
asized
wh
en
th
e
m
o
d
el
was m
ak
in
g
i
ts
p
r
ed
ictio
n
.
Ku
m
ar
i
an
d
Go
el
[
1
6
]
,
g
iv
e
an
ex
ten
s
iv
e
r
ev
iew
o
f
co
m
p
u
tatio
n
al
in
tellig
en
ce
tech
n
iq
u
es
f
o
r
d
iag
n
o
s
in
g
as
well
as
class
if
y
in
g
lu
n
g
ca
n
ce
r
with
in
ch
est
C
T
im
ag
es.
T
h
is
s
tu
d
y
c
ateg
o
r
izes
an
d
co
m
p
a
r
e
d
if
f
er
en
t m
eth
o
d
s
s
u
ch
as c
las
s
ical
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es (
e.
g
.
s
u
p
p
o
r
t v
ec
t
o
r
m
ac
h
i
n
e
(
SVM
)
,
r
an
d
o
m
f
o
r
ests
(
R
F)
)
,
d
ee
p
lear
n
in
g
f
am
ilies
lik
e
C
NN
s
,
R
N
Ns
,
o
r
v
ar
iatio
n
al
a
u
to
en
co
d
er
s
(
VAE
s
)
g
en
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
s
(
GANs
)
as
well
as
h
y
b
r
id
alg
o
r
ith
m
s
th
at
co
m
b
in
e
n
atu
r
e
in
s
p
ir
e
d
o
p
tim
izatio
n
.
T
h
e
p
ap
er
d
o
es
well
to
f
o
cu
s
o
n
g
r
o
win
g
tr
en
d
s
lik
e
tr
an
s
f
er
lear
n
in
g
,
f
e
d
er
ated
lear
n
in
g
f
o
r
p
r
iv
ac
y
,
an
d
XAI
to
ad
d
r
ess
th
e
“b
lack
-
b
o
x
”
lim
it
atio
n
s
o
f
m
o
d
els.
I
t
also
s
y
s
te
m
atica
lly
s
tu
d
ies
th
e
p
er
f
o
r
m
an
ce
b
ased
o
n
lar
g
e
p
u
b
lic
d
atab
ases
o
f
th
e
l
u
n
g
i
m
ag
e
d
atab
ase
co
n
s
o
r
tiu
m
im
ag
e
d
atab
ase
r
eso
u
r
ce
i
n
itiativ
e
(
L
I
DC
-
I
DR
I
)
an
d
th
e
lu
n
g
n
o
d
u
le
an
aly
s
is
2
0
1
6
(
L
UNA1
6
)
,
a
n
d
d
em
o
n
s
tr
ates
th
e
u
r
g
en
t
ch
allen
g
es
i
n
g
r
o
u
p
b
ias
an
d
class
im
b
alan
ce
th
at
h
in
d
er
g
en
er
ali
zin
g
AI
s
o
lu
tio
n
s
to
b
r
o
a
d
clin
ical
co
n
tex
ts
.
T
h
is
s
tu
d
y
b
y
Su
n
et
a
l.
[
1
9
]
in
tr
o
d
u
ce
s
a
wea
k
ly
s
u
p
e
r
v
is
ed
d
ee
p
lear
n
i
n
g
alg
o
r
ith
m
f
o
r
th
e
au
to
m
ated
d
etec
tio
n
a
n
d
s
tag
in
g
p
r
o
g
r
ess
io
n
o
f
c
h
r
o
n
ic
o
b
s
tr
u
ctiv
e
p
u
lm
o
n
ar
y
d
is
ea
s
e
(
C
OPD)
f
r
o
m
v
o
lu
m
etr
ic
C
T
s
ca
n
s
.
T
o
ad
d
r
ess
lim
ita
tio
n
s
in
tr
ad
itio
n
al
d
en
s
ito
m
etr
ic
an
aly
s
is
,
wh
er
e
n
o
is
e
-
s
en
s
itiv
e
p
ix
el
th
r
esh
o
ld
s
an
d
tim
e
-
co
n
s
u
m
i
n
g
m
a
n
u
al
s
eg
m
e
n
tatio
n
s
ar
e
u
s
ed
,
th
e
y
p
r
esen
t
a
m
u
ltip
l
e
in
s
tan
ce
lear
n
in
g
(
MI
L
)
a
p
p
r
o
ac
h
.
T
h
is
ar
ch
it
ec
tu
r
e
u
s
es
a
3
D
R
esNet
m
o
d
el
f
o
r
C
T
s
lice
en
co
d
in
g
an
d
an
atten
tio
n
m
ec
h
an
is
m
f
o
r
p
o
o
lin
g
,
b
ei
n
g
tr
ain
ed
at
p
atien
t
-
lev
el
(
G
OL
D
s
tag
e
f
r
o
m
s
p
ir
o
m
etr
y
)
in
s
tead
o
f
d
e
n
s
e
p
ix
el
-
lev
el
an
n
o
tatio
n
s
.
T
h
e
m
o
d
el
d
em
o
n
s
tr
ates
ex
ce
llen
t
d
is
cr
im
in
atio
n
f
o
r
C
OPD
s
tag
es
(
GOL
D
1
,
2
,
3
,
an
d
4
)
an
d
p
r
o
v
id
es
clin
ical
in
ter
p
r
etab
ilit
y
th
r
o
u
g
h
th
e
g
en
er
atio
n
o
f
atten
tio
n
h
ea
tm
a
p
s
wh
ich
h
ig
h
lig
h
t
d
is
ea
s
e
-
r
elate
d
p
ath
o
lo
g
ies
s
u
ch
as
em
p
h
y
s
em
a
an
d
b
r
o
n
ch
i
ec
tasi
s
.
Seg
m
en
tatio
n
,
co
m
p
u
t
er
-
aid
ed
d
etec
tio
n
,
an
d
co
m
p
u
ter
-
aid
ed
d
iag
n
o
s
i
s
(
Seg
-
C
ADe
-
C
ADx
)
was
d
ev
elo
p
ed
b
y
Su
b
r
am
a
n
y
am
et
a
l.
[
2
0
]
as
a
d
u
al
-
s
tag
e
d
ee
p
lear
n
in
g
s
y
s
tem
th
at
ca
n
d
etec
t
p
u
lm
o
n
ar
y
n
o
d
u
le
m
alig
n
an
cy
r
is
k
as
an
in
teg
r
ated
p
r
o
ce
s
s
f
r
o
m
lo
w
-
d
o
s
e
C
T
s
ca
n
s
.
T
h
e
f
ir
s
t
s
tag
e
u
s
es
a
3
D
U
-
Net
to
a
s
s
is
t
with
s
eg
m
en
tatio
n
-
b
ased
n
o
d
u
le
d
etec
tio
n
,
wh
ich
is
f
o
llo
we
d
b
y
a
m
u
lt
i
-
s
tr
ea
m
h
y
b
r
id
class
if
ier
c
o
n
s
is
tin
g
o
f
lo
ca
l
v
o
lu
m
etr
ic
f
e
atu
r
es
f
r
o
m
a
3
D
Den
s
eNe
t,
g
lo
b
al
co
n
tex
t
f
r
o
m
an
Swin
tr
an
s
f
o
r
m
er
,
an
d
late
-
s
tag
e
co
n
ca
ten
atio
n
o
f
h
an
d
cr
af
ted
r
a
d
io
m
ic
f
ea
tu
r
es
f
r
o
m
th
e
v
o
lu
m
et
r
ic
i
m
ag
es.
R
esu
lts
f
r
o
m
b
o
t
h
th
e
L
UNA1
6
d
atab
ase
an
d
th
e
L
I
DC
-
I
DR
I
d
atab
ase
p
r
o
v
id
e
d
a
co
m
p
etitio
n
p
er
f
o
r
m
an
ce
m
etr
ic
(
C
PM
)
f
o
r
n
o
d
u
le
d
etec
tio
n
o
f
0
.
9
4
4
with
an
ex
ce
llen
t
R
OC
-
AU
C
o
f
0
.
9
8
8
f
o
r
b
in
ar
y
class
if
icatio
n
o
f
m
alig
n
an
cy
.
Ad
d
itio
n
ally
,
th
e
p
a
p
er
d
is
cu
s
s
es
im
p
r
o
v
in
g
t
h
e
clin
ical
ac
cu
r
ac
y
o
f
th
e
m
o
d
el
b
y
in
tr
o
d
u
cin
g
tem
p
er
at
u
r
e
s
ca
lin
g
f
o
r
th
e
ca
lib
r
ati
o
n
o
f
m
alig
n
a
n
cy
p
r
o
b
a
b
ilit
y
o
u
tp
u
ts
,
th
er
eb
y
p
r
o
v
id
in
g
m
o
r
e
cl
o
s
ely
r
elate
d
e
m
p
ir
ical
r
is
k
esti
m
ates th
an
p
r
ev
io
u
s
ly
av
ailab
le.
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
Qu
a
n
tita
tive
ch
est co
mp
u
ted
to
mo
g
r
a
p
h
y
-
b
a
s
ed
h
yb
r
id
d
ee
p
lea
r
n
in
g
fo
r
ch
r
o
n
ic
l
u
n
g
…
(
Ga
r
ima
Ja
in
)
4571
Su
i
et
a
l.
[
2
1
]
p
r
o
p
o
s
ed
a
m
ac
h
in
e
lear
n
in
g
m
eth
o
d
to
d
iag
n
o
s
e
an
d
g
r
ad
e
th
e
s
ev
er
ity
o
f
C
OP
D
in
p
atien
ts
f
r
o
m
C
T
s
ca
n
s
.
Un
li
k
e
a
lar
g
e
p
r
o
p
o
r
tio
n
o
f
p
r
ev
io
u
s
wo
r
k
s
1
–
2
2
th
at
s
im
p
ly
f
o
cu
s
o
n
is
o
lated
f
ea
tu
r
es,
it
co
n
s
id
er
s
m
u
ltid
im
en
s
io
n
al
ch
ar
ac
ter
is
tics
in
clu
d
in
g
th
e
lu
n
g
p
ar
e
n
ch
y
m
a,
wh
o
le
air
way
tr
ee
,
an
d
p
u
lm
o
n
a
r
y
v
ascu
latu
r
e
-
r
elate
d
o
n
es.
T
h
e
p
a
p
er
u
tili
ze
d
th
e
SVM
class
if
ier
f
o
r
3
4
9
s
u
b
jects
an
d
ac
h
iev
e
d
b
etter
d
iag
n
o
s
tic
p
er
f
o
r
m
an
ce
(
AUC
=
0
.
9
6
5
o
n
ex
ter
n
al
t
esti
n
g
)
an
d
g
r
ad
in
g
,
s
u
cc
ess
f
u
lly
d
is
tin
g
u
is
h
in
g
C
VD
f
r
o
m
HC
(
AUC
=
0
.
8
1
3
)
.
T
h
e
n
o
v
el
an
d
in
ter
esti
n
g
f
in
d
in
g
in
th
e
s
tu
d
y
was
th
e
im
p
o
r
tan
ce
o
f
v
ascu
lar
d
er
iv
ativ
es
o
n
C
OPD
+
as
o
p
p
o
s
ed
to
th
e
cu
r
r
en
tly
h
ig
h
lig
h
ted
co
n
tr
ib
u
tio
n
o
f
r
em
o
d
elin
g
in
th
e
tr
u
n
k
o
f
ar
ter
ies
an
d
b
r
o
n
c
h
u
s
.
Ho
wev
er
,
o
u
r
s
tu
d
y
is
lim
ited
b
y
th
e
s
m
all
s
am
p
le
s
ize
an
d
an
o
v
er
ly
s
im
p
lifie
d
d
ich
o
to
m
izatio
n
(
m
ild
v
er
s
u
s
s
ev
er
e)
th
at
m
ay
n
o
t f
u
lly
ac
c
o
u
n
t f
o
r
th
e
clin
ical
h
eter
o
g
e
n
eity
o
f
th
e
d
is
ea
s
e.
Als
u
r
ay
h
i
[
2
2
]
h
as
p
r
o
p
o
s
ed
an
d
v
alid
ate
d
m
o
d
els
b
ased
o
n
d
ee
p
lear
n
i
n
g
f
o
r
au
to
m
atic
s
tag
in
g
o
f
C
OPD
with
ch
est
C
T
.
T
h
e
s
tu
d
y
o
v
er
co
m
es
t
h
e
d
r
aw
b
ac
k
s
o
f
co
n
v
en
tio
n
al
p
u
lm
o
n
ar
y
f
u
n
ctio
n
test
s
(
PF
T
s
)
b
y
in
tr
o
d
u
cin
g
im
ag
e
-
b
ased
class
if
icatio
n
s
y
s
tem
s
th
at
ar
e
co
n
s
is
ten
t
with
n
o
t
o
n
ly
t
h
e
h
ig
h
-
r
eso
lu
tio
n
eig
h
t
-
s
tag
e
s
y
s
tem
b
u
t
also
th
e
n
ew
GOL
D
2
0
2
3
cr
iter
ia.
T
h
e
m
eth
o
d
in
co
r
p
o
r
ates
n
eu
r
a
l
n
etwo
r
k
s
an
d
3
D
C
NNs
with
p
r
e
-
d
ef
in
e
d
f
ea
t
u
r
es
ex
tr
ac
ted
f
r
o
m
p
air
ed
in
s
p
ir
atio
n
/ex
p
ir
atio
n
C
T
s
c
an
s
(
e.
g
.
,
o
n
e
-
s
h
o
t
tr
an
s
f
er
-
b
ased
f
ea
tu
r
es
an
d
GL
C
M
-
b
ased
tex
tu
r
e
p
atter
n
s
)
as
well
a
s
en
d
-
to
-
en
d
d
ee
p
C
NN
lear
n
in
g
.
A
f
u
r
th
er
k
e
y
d
ev
el
o
p
m
en
t
is
in
th
e
g
en
er
atio
n
o
f
'
h
y
b
r
id
'
m
o
d
els
wh
ich
co
m
b
in
e
im
ag
in
g
f
e
atu
r
es
with
clin
ical
in
f
o
r
m
atio
n
,
s
u
ch
as
s
y
m
p
to
m
s
ev
er
ity
an
d
ex
ac
er
b
ati
o
n
r
is
k
,
r
esu
ltin
g
in
a
s
ig
n
if
ican
t
g
ain
in
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
Her
e
,
th
is
s
tu
d
y
s
h
o
w
th
at
th
ese
d
ee
p
lea
r
n
in
g
m
o
d
els
ca
n
ac
cu
r
ately
p
r
ed
ict
C
OPD
s
ev
er
ity
f
ac
to
r
s
d
ir
ec
tly
f
r
o
m
C
T
im
ag
es,
wh
ich
co
u
ld
h
elp
d
im
in
is
h
th
e
r
eq
u
ir
em
e
n
t
o
f
p
atien
t
-
d
e
p
en
d
en
t
s
p
ir
o
m
etr
y
m
ea
s
u
r
em
en
ts
[
2
3
]
,
[
2
4
]
.
T
ab
le
1
s
u
m
m
ar
izes
th
e
m
aj
o
r
s
tu
d
ies
ap
p
ly
in
g
a
d
v
an
ce
d
d
ee
p
lear
n
in
g
an
d
h
y
b
r
id
AI
m
o
d
els
f
o
r
d
etec
tio
n
,
class
if
icatio
n
,
an
d
s
ev
er
ity
ass
ess
m
en
t
o
f
lu
n
g
d
is
ea
s
es
f
r
o
m
ch
est
C
T
im
ag
e
s
.
T
ab
le
1
p
r
o
v
id
es
au
th
o
r
s
,
m
ai
n
m
eth
o
d
o
l
o
g
ica
l
ap
p
r
o
ac
h
,
d
ataset
n
am
es,
s
am
p
le
s
izes,
co
r
e
co
n
tr
ib
u
t
io
n
s
,
p
er
f
o
r
m
a
n
ce
h
ig
h
lig
h
ts
,
an
d
r
esear
ch
g
ap
s
f
o
r
ea
ch
r
ef
e
r
en
ce
.
T
h
is
co
m
p
ar
ativ
e
s
u
m
m
ar
y
h
ig
h
lig
h
ts
th
e
cu
r
r
en
t
s
tatu
s
an
d
r
em
ain
in
g
c
h
allen
g
es
in
m
e
d
ical
im
ag
in
g
AI
,
in
clu
d
in
g
in
ter
p
r
etab
ilit
y
,
m
u
lti
-
class
d
if
f
er
en
tiatio
n
,
an
d
cr
o
s
s
-
d
ataset
v
alid
atio
n
.
T
ab
le
1
.
L
iter
atu
r
e
r
ev
iew
A
u
t
h
o
r
(
s)
A
r
c
h
i
t
e
c
t
u
r
e
D
a
t
a
s
e
t
n
a
m
e
S
a
mp
l
e
si
z
e
M
a
i
n
c
o
n
t
r
i
b
u
t
i
o
n
O
u
t
c
o
m
e
R
e
se
a
r
c
h
g
a
p
s
O
z
d
e
m
i
r
e
t
a
l
.
[
1
8
]
A
t
t
e
n
t
i
o
n
-
e
n
h
a
n
c
e
d
I
n
c
e
p
t
i
o
n
N
e
X
t
-
b
a
s
e
d
h
y
b
r
i
d
d
e
e
p
l
e
a
r
n
i
n
g
P
r
i
v
a
t
e
l
u
n
g
C
T
d
a
t
a
se
t
2
,
5
0
0
i
ma
g
e
s
H
y
b
r
i
d
mo
d
e
l
u
si
n
g
a
t
t
e
n
t
i
o
n
a
n
d
I
n
c
e
p
t
i
o
n
N
e
X
t
f
o
r
l
u
n
g
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
.
H
i
g
h
a
c
c
u
r
a
c
y
a
n
d
r
o
b
u
st
n
e
ss
i
n
d
e
t
e
c
t
i
o
n
Li
mi
t
e
d
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
;
n
o
t
e
v
a
l
u
a
t
e
d
o
n
p
u
b
l
i
c
o
r
mu
l
t
i
-
c
l
a
ss
d
a
t
a
se
t
s
K
a
m
a
l
a
a
n
d
M
o
h
a
n
[
2
5
]
H
y
b
r
i
d
AI
f
r
a
mew
o
r
k
P
u
b
l
i
c
a
n
d
p
r
i
v
a
t
e
C
T
d
a
t
a
se
t
s
1
,
8
0
0
i
ma
g
e
s
Ef
f
i
c
i
e
n
t
h
y
b
r
i
d
A
I
w
i
t
h
d
e
e
p
a
n
d
h
a
n
d
c
r
a
f
t
e
d
f
e
a
t
u
r
e
s
f
o
r
l
u
n
g
c
a
n
c
e
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
En
h
a
n
c
e
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
a
n
d
e
f
f
i
c
i
e
n
c
y
Li
mi
t
e
d
e
x
p
l
a
i
n
a
b
i
l
i
t
y
;
l
a
c
k
s
e
x
t
e
r
n
a
l
v
a
l
i
d
a
t
i
o
n
H
o
sn
y
e
t
a
l
.
[
2
6
]
M
u
l
t
i
-
mo
d
a
l
a
t
t
e
n
t
i
o
n
–
b
a
se
d
I
n
c
e
p
t
i
o
n
-
R
e
sN
e
t
LI
D
C
-
I
D
R
I
1
,
0
0
0
c
a
ses
M
u
l
t
i
-
mo
d
a
l
a
t
t
e
n
t
i
o
n
-
b
a
se
d
f
u
si
o
n
f
o
r
l
u
n
g
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
O
u
t
p
e
r
f
o
r
m
e
d
u
n
i
m
o
d
a
l
mo
d
e
l
s
La
c
k
s
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
a
n
d
c
o
m
p
u
t
a
t
i
o
n
a
l
e
f
f
i
c
i
e
n
c
y
a
n
a
l
y
si
s
P
a
r
i
d
a
e
t
a
l
.
[
2
7
]
3
D
F
u
s
i
o
n
N
e
t
S
y
n
t
h
e
t
i
c
C
T
D
a
t
a
s
e
t
6
0
0
s
c
a
n
s
3
D
f
u
s
i
o
n
mo
d
e
l
f
o
r
l
u
n
g
c
a
n
c
e
r
seg
m
e
n
t
a
t
i
o
n
H
i
g
h
s
e
g
m
e
n
t
a
t
i
o
n
p
r
e
c
i
s
i
o
n
,
b
e
t
t
e
r
b
o
u
n
d
a
r
i
e
s
N
o
t
v
a
l
i
d
a
t
e
d
o
n
r
e
a
l
C
T
d
a
t
a
;
l
a
c
k
s
c
l
i
n
i
c
a
l
o
u
t
c
o
m
e
c
o
r
r
e
l
a
t
i
o
n
D
u
h
a
y
y
i
m
et
a
l
.
[
2
8
]
I
n
t
e
r
p
r
e
t
a
b
l
e
h
y
b
r
i
d
e
n
s
e
m
b
l
e
w
i
t
h
a
t
t
e
n
t
i
o
n
f
u
si
o
n
+
EA
O
O
-
g
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m
(
GA
)
o
p
t
i
m
i
z
a
t
i
o
n
P
r
i
v
a
t
e
C
T
d
a
t
a
se
t
1
,
2
0
0
A
t
t
e
n
t
i
o
n
-
b
a
se
d
f
u
si
o
n
a
n
d
g
e
n
e
t
i
c
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
a
n
d
a
c
c
u
r
a
c
y
S
t
a
t
e
-
of
-
t
h
e
-
a
r
t
a
c
c
u
r
a
c
y
a
n
d
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
N
o
m
u
l
t
i
-
c
l
a
ss
o
r
e
x
t
e
r
n
a
l
d
a
t
a
se
t
v
a
l
i
d
a
t
i
o
n
F
a
i
z
i
e
t
a
l
.
[
2
9
]
C
N
N
-
b
a
s
e
d
d
e
e
p
l
e
a
r
n
i
n
g
LU
N
A
1
6
8
8
8
C
T
sca
n
s
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
CT
-
b
a
se
d
l
u
n
g
c
a
n
c
e
r
R
e
l
i
a
b
l
e
a
n
d
r
o
b
u
st
c
l
a
ss
i
f
i
c
a
t
i
o
n
B
i
n
a
r
y
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
n
l
y
;
l
a
c
k
s
e
x
p
l
a
i
n
a
b
i
l
i
t
y
a
n
d
c
r
o
ss
-
d
a
t
a
se
t
e
v
a
l
u
a
t
i
o
n
A
l
mei
d
a
e
t
a
l
.
[
3
0
]
A
n
o
m
a
l
y
-
b
a
se
d
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
d
i
s
e
a
se
se
v
e
r
i
t
y
C
O
P
D
G
e
n
e
,
N
LST
4
,
5
0
0
sca
n
s
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
q
u
a
n
t
i
t
a
t
i
v
e
C
O
P
D
sev
e
r
i
t
y
p
r
e
d
i
c
t
i
o
n
A
c
c
u
r
a
t
e
d
i
s
e
a
se
se
v
e
r
i
t
y
a
ssessm
e
n
t
N
o
t
m
u
l
t
i
-
d
i
sea
s
e
f
o
c
u
se
d
;
l
i
mi
t
e
d
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
a
ssessm
e
n
t
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
5
6
8
-
4
5
9
1
4572
2
.
1
.
Resea
rc
h g
a
p su
m
m
a
ry
a
nd
m
o
t
iv
a
t
io
n
Desp
ite
th
e
m
ajo
r
ad
v
a
n
ce
m
e
n
ts
s
ee
n
in
r
ec
en
t
y
ea
r
s
,
in
clu
d
in
g
im
p
r
o
v
em
en
ts
in
d
ee
p
le
ar
n
in
g
an
d
tr
an
s
f
o
r
m
er
-
b
ased
ar
ch
itectu
r
es,
r
ad
io
m
ic
an
aly
s
is
tech
n
iq
u
es,
an
d
XAI
f
o
r
p
u
lm
o
n
a
r
y
d
i
s
ea
s
e
clas
s
if
icatio
n
f
r
o
m
ch
est
C
T
s
ca
n
s
,
th
er
e
ar
e
s
till
n
u
m
er
o
u
s
s
ig
n
if
ican
t
d
r
awb
ac
k
s
ac
r
o
s
s
m
an
y
ex
is
tin
g
p
u
b
licatio
n
s
.
I
n
g
en
er
al,
m
o
s
t
o
f
t
h
e
liter
atu
r
e
ten
d
s
to
war
d
s
b
in
ar
y
class
if
icatio
n
s
wh
ile
n
eg
lectin
g
f
in
e
-
g
r
ain
ed
m
u
lti
-
class
s
tag
in
g
.
Ad
d
itio
n
ally
,
tr
ad
itio
n
al
C
NNs
ty
p
ically
p
r
o
v
id
e
l
o
w
lev
els
o
f
f
ea
t
u
r
e
-
lev
el
i
n
ter
p
r
etab
ilit
y
wh
ile
tr
an
s
f
o
r
m
er
-
in
s
p
ir
ed
ar
c
h
itectu
r
es o
f
ten
n
ec
ess
itate
lar
g
e
am
o
u
n
ts
o
f
d
ata
to
ac
h
ie
v
e
s
tab
le
o
p
tim
izatio
n
.
Fu
r
th
er
m
o
r
e
,
s
ev
er
al
r
ad
io
m
ics
-
b
ased
ap
p
r
o
ac
h
es
g
e
n
er
a
te
clin
ically
in
ter
p
r
etab
le
q
u
an
titativ
e
o
u
tp
u
ts
,
b
u
t
m
o
s
t
d
o
litt
le
to
ca
p
tu
r
e
t
h
e
m
u
lti
-
lev
el
s
p
atial
h
ier
ar
c
h
ies
p
r
esen
t
in
th
o
r
ac
ic
C
T
im
ag
in
g
.
I
n
ad
d
itio
n
,
th
e
m
ajo
r
ity
o
f
h
y
b
r
id
f
r
am
ewo
r
k
s
ar
e
h
ig
h
l
y
r
elian
t
o
n
co
m
p
u
tatio
n
ally
in
ten
s
iv
e
f
u
s
io
n
m
eth
o
d
s
o
r
o
f
f
er
o
n
ly
lim
ited
lev
els
o
f
ex
p
lain
ab
ilit
y
,
lim
itin
g
th
eir
ap
p
licab
ilit
y
in
lar
g
e
-
s
ca
le,
r
ea
l
-
wo
r
ld
clin
ical
wo
r
k
f
lo
ws.
Fin
ally
,
th
er
e
is
s
till
lim
i
ted
ex
ter
n
al
v
alid
ati
o
n
,
lo
w
tr
an
s
p
ar
e
n
cy
ac
r
o
s
s
all
p
r
ed
ictio
n
s
,
an
d
lim
ited
co
n
s
id
er
atio
n
o
f
clin
ic
al
d
ep
lo
y
m
e
n
t,
wh
ich
co
n
tin
u
e
to
in
h
ib
it
th
e
o
v
e
r
all
r
eliab
ilit
y
an
d
ad
o
p
tio
n
o
f
cu
r
r
en
tly
a
v
ailab
le
au
to
m
ate
d
s
y
s
tem
s
f
o
r
d
iag
n
o
s
in
g
lu
n
g
d
i
s
ea
s
e.
T
o
o
v
er
c
o
m
e
lim
itatio
n
s
in
e
x
is
tin
g
wo
r
k
s
,
p
r
esen
t
a
h
y
b
r
id
d
ee
p
lear
n
in
g
f
r
am
ewo
r
k
b
ased
o
n
q
u
an
titativ
e
ch
est
C
T
,
w
h
ich
em
p
lo
y
s
a
p
r
e
-
tr
ai
n
ed
R
esNet5
0
b
ac
k
b
o
n
e
at
its
co
r
e
t
o
p
r
o
d
u
ce
d
ee
p
s
p
atial
em
b
ed
d
in
g
s
,
co
m
b
in
e
d
with
tex
tu
r
e
f
ea
tu
r
es
d
er
i
v
ed
f
r
o
m
h
an
d
cr
a
f
ted
GL
C
M
tex
tu
r
e
an
aly
s
es
an
d
en
tr
o
p
ic
m
ea
s
u
r
es
th
at
ch
ar
ac
ter
ize
h
et
er
o
g
en
eity
.
A
d
d
itio
n
ally
in
clu
d
ed
in
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
is
an
o
p
tim
ized
en
s
em
b
le
class
if
ier
im
p
lem
en
ted
v
ia
th
e
XGBo
o
s
t
alg
o
r
ith
m
,
as
well
as
two
f
o
r
m
s
o
f
XAI
th
r
o
u
g
h
Gr
ad
-
C
AM
an
d
SHAP
to
p
r
o
m
o
te
e
n
h
an
ce
d
s
p
atial
a
n
d
f
ea
tu
r
e
-
lev
el
ex
p
lain
ab
ili
ty
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
l
o
g
y
was
b
en
c
h
m
ar
k
ed
ag
ain
s
t
tr
a
n
s
f
o
r
m
e
r
-
b
ased
ar
ch
itectu
r
es.
ViT
-
B
1
6
an
d
Swin
tr
asn
f
o
r
m
er
ar
ch
itectu
r
es
f
o
r
class
if
icatio
n
p
er
f
o
r
m
an
ce
,
co
m
p
u
tatio
n
al
ef
f
icien
cy
,
an
d
clin
ical
in
ter
p
r
etab
ilit
y
.
T
h
e
m
o
tiv
atio
n
o
f
th
is
r
esear
ch
is
th
er
ef
o
r
e
to
d
ev
elo
p
a
co
m
p
u
tatio
n
ally
ef
f
icien
t
m
u
lti
-
class
ch
est
C
T
class
if
icatio
n
f
r
am
ewo
r
k
with
clin
ical
in
ter
p
r
etab
ilit
y
an
d
r
esil
ien
cy
ap
p
licab
le
f
o
r
f
u
tu
r
e
u
s
e
in
AI
-
ass
is
ted
r
ad
io
lo
g
y
an
d
in
tellig
en
t lu
n
g
d
is
ea
s
e
s
cr
ee
n
in
g
ap
p
licatio
n
s
.
3.
P
RO
P
O
SE
D
F
RA
M
E
WO
R
K
I
n
th
is
s
tu
d
y
,
a
n
ew
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
el
is
p
r
o
p
o
s
ed
,
wh
ic
h
in
teg
r
ates
th
e
ab
s
tr
ac
t
r
ep
r
esen
tatio
n
a
r
ch
itectu
r
e
o
f
a
d
ee
p
C
NN
an
d
h
an
d
cr
a
f
ted
r
ad
io
m
ics
to
o
v
er
c
o
m
e
th
e
li
m
itatio
n
s
o
f
u
tili
zin
g
o
n
ly
a
b
s
tr
ac
t
s
tr
u
ctu
r
e
f
r
o
m
a
C
NN
o
r
h
an
d
c
r
af
ted
r
ad
i
o
m
ics
f
o
r
ca
teg
o
r
izin
g
c
h
est
C
T
.
T
h
e
p
r
im
ar
y
m
eth
o
d
o
l
o
g
ical
co
n
tr
ib
u
tio
n
i
s
a
f
ea
tu
r
e
f
u
s
io
n
tech
n
iq
u
e
at
th
e
lev
el
o
f
f
ea
tu
r
es
th
at
c
o
m
b
in
es
h
i
g
h
-
lev
el
s
p
atial
em
b
ed
d
in
g
s
f
r
o
m
a
p
r
e
-
tr
ain
ed
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
an
d
r
ad
io
m
ic
h
an
d
-
cr
a
f
ted
q
u
an
titativ
e
tex
tu
r
e
d
escr
ip
to
r
s
th
at
ar
e
also
clin
i
ca
lly
in
ter
p
r
etab
le.
T
h
e
m
o
d
e
l
co
m
b
in
es
th
ese
s
ep
ar
ate
f
ea
tu
r
e
m
o
d
alities
to
en
h
an
ce
class
if
icatio
n
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
u
lti
-
class
s
ta
g
es
o
f
lu
n
g
d
is
ea
s
es,
as
well
as
o
f
f
er
in
g
r
o
b
u
s
t
m
o
d
el
in
ter
p
r
etab
ilit
y
b
y
m
ea
n
s
o
f
d
u
al
XAI
ap
p
r
o
ac
h
es.
Fi
g
u
r
e
1
p
r
esen
ts
th
e
p
r
o
p
o
s
ed
d
u
al
-
s
tr
ea
m
s
ch
em
e
.
I
n
p
u
t
C
T
d
ata
is
co
n
co
m
itan
tl
y
an
aly
ze
d
b
y
d
ee
p
lear
n
in
g
an
d
r
ad
io
m
ic
p
ip
elin
es.
Sev
er
al
o
f
th
ese
f
ea
tu
r
es
ar
e
th
en
co
m
b
in
ed
u
s
in
g
a
f
u
s
io
n
s
tr
ateg
y
b
ef
o
r
e
class
if
icatio
n
,
s
o
t
h
at
th
e
clin
ical
p
r
ed
ictio
n
s
h
av
e
b
o
t
h
s
p
atial
(
Gr
ad
-
C
AM
)
an
d
f
ea
tu
r
e
-
lev
el
(
SHAP)
in
ter
p
r
etab
ilit
y
.
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
q
u
an
titativ
e
-
s
p
atial
f
r
am
ewo
r
k
3
.
1
.
Da
t
a
s
et
T
h
is
p
u
b
licly
s
h
ar
e
d
d
ataset
c
o
n
tain
s
ap
p
r
o
x
im
ately
o
n
e
t
h
o
u
s
an
d
C
T
s
ca
n
s
,
wh
ich
will
b
e
u
s
ed
to
class
if
y
p
atien
ts
wi
th
lu
n
g
ca
n
ce
r
in
to
th
eir
ca
teg
o
r
ies.
T
h
e
d
ataset
co
n
s
is
t
s
o
f
1
,
0
0
0
c
h
est
C
T
s
ca
n
im
ag
es
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
Qu
a
n
tita
tive
ch
est co
mp
u
ted
to
mo
g
r
a
p
h
y
-
b
a
s
ed
h
yb
r
id
d
ee
p
lea
r
n
in
g
fo
r
ch
r
o
n
ic
l
u
n
g
…
(
Ga
r
ima
Ja
in
)
4573
with
a
to
tal
f
ile
s
ize
o
f
ap
p
r
o
x
im
ately
1
2
4
MB.
T
h
e
im
ag
es
ar
e
p
r
o
v
id
ed
in
h
ig
h
-
r
eso
l
u
tio
n
.
jp
g
an
d
.
p
n
g
f
o
r
m
ats,
en
s
u
r
in
g
v
is
u
al
clar
i
ty
s
u
itab
le
f
o
r
d
etailed
an
aly
s
is
.
Al
l
s
ca
n
s
ar
e
d
er
iv
ed
f
r
o
m
C
T
,
m
ak
in
g
th
e
d
ataset
well
-
s
u
ited
f
o
r
r
esear
c
h
an
d
d
e
v
elo
p
m
e
n
t
in
au
to
m
a
ted
lu
n
g
d
is
ea
s
e
d
etec
tio
n
an
d
class
if
icatio
n
.
T
h
e
d
ataset
co
n
s
is
ts
o
f
f
o
u
r
p
at
h
o
l
o
g
ical
class
if
icatio
n
s
o
f
p
atien
ts
with
lu
n
g
ca
n
ce
r
,
d
escr
ib
ed
as f
o
llo
ws:
‒
No
r
m
al:
it
is
lu
n
g
tis
s
u
e
th
at
d
em
o
n
s
tr
ates
a
h
ea
lth
y
lu
n
g
with
n
o
clea
r
ev
id
en
ce
o
f
n
o
d
u
les
o
r
o
th
er
lesi
o
n
s
p
r
esen
t
[
3
1
]
.
‒
Ad
en
o
ca
r
cin
o
m
a:
it
is
th
e
m
o
s
t
co
m
m
o
n
s
u
b
ty
p
e
o
f
n
o
n
-
s
m
all
ce
ll
lu
n
g
ca
n
ce
r
(
NSC
L
C
)
an
d
ar
is
es
f
r
o
m
p
er
ip
h
e
r
al
g
lan
d
u
lar
ce
ll
s
[
3
2
]
.
‒
Sq
u
am
o
u
s
ce
ll
ca
r
cin
o
m
a:
it
is
th
e
s
ec
o
n
d
m
o
s
t
c
o
m
m
o
n
ty
p
e
o
f
NSC
L
C
an
d
o
r
ig
i
n
ates
f
r
o
m
th
e
s
q
u
am
o
u
s
ce
lls
lin
in
g
th
e
m
aj
o
r
air
way
s
[
3
3
]
.
‒
L
ar
g
e
ce
ll
ca
r
cin
o
m
a:
a
ty
p
e
o
f
NSC
L
C
,
ar
e
th
e
m
o
s
t
co
m
m
o
n
g
r
o
u
p
o
f
lu
n
g
ca
n
c
er
s
,
an
d
L
C
C
ac
co
u
n
ts
f
o
r
a
s
m
aller
p
er
ce
n
tag
e
o
f
th
ese
ca
s
es
[
3
4
]
–
[
3
6
]
.
T
h
e
d
ataset
is
s
y
s
tem
atica
lly
p
ar
titi
o
n
e
d
in
to
tr
ain
i
n
g
,
v
a
lid
atio
n
,
an
d
test
s
ets
to
ev
alu
ate
th
e
m
o
d
el'
s
g
en
er
aliza
tio
n
p
er
f
o
r
m
an
ce
.
T
r
ai
n
in
g
s
et:
a
to
tal
o
f
6
1
3
im
ag
es
d
i
v
id
ed
i
n
to
ad
e
n
o
ca
r
cin
o
m
a
(
1
9
5
)
,
lar
g
e
ce
ll
ca
r
cin
o
m
a
(
1
1
5
)
,
s
q
u
am
o
u
s
ce
ll
ca
r
cin
o
m
a
(
1
5
5
)
,
an
d
n
o
r
m
al
(
1
4
8
)
.
Valid
a
tio
n
s
et:
a
to
tal
o
f
7
2
im
ag
es
ar
e
g
r
o
u
p
ed
as a
d
e
n
o
ca
r
cin
o
m
a
(
2
3
)
,
lar
g
e
ce
ll c
ar
cin
o
m
a
(
2
1
)
,
s
q
u
am
o
u
s
ce
ll c
ar
cin
o
m
a
(
1
5
)
,
an
d
n
o
r
m
al
(
1
3
)
.
T
esti
n
g
s
et:
a
to
tal
o
f
3
1
5
im
ag
es
ar
e
g
r
o
u
p
e
d
as
ad
en
o
ca
r
cin
o
m
a
(
1
2
0
)
,
l
ar
g
e
ce
ll
ca
r
cin
o
m
a
(
5
1
)
,
s
q
u
am
o
u
s
ce
ll c
ar
cin
o
m
a
(
9
0
)
,
an
d
n
o
r
m
al
(
5
4
)
.
T
ab
le
2
s
u
m
m
ar
izes th
e
m
ain
f
ea
tu
r
e
ca
teg
o
r
ies,
n
am
es
,
an
d
d
escr
ip
tio
n
s
u
s
ed
in
th
e
ch
est C
T
im
ag
e
class
if
icatio
n
p
ip
elin
e.
Key
f
ea
tu
r
es
ar
e:
b
asic
im
ag
e
m
e
tad
ata,
p
ix
el
-
lev
el
s
tatis
tics
,
tex
tu
r
e
d
escr
ip
to
r
s
(
e.
g
.
,
GL
C
M
-
b
ased
m
etr
ics
an
d
en
tr
o
p
y
)
,
s
tr
u
ctu
r
al
an
d
s
h
ap
e
attr
ib
u
tes,
h
is
to
g
r
am
an
d
f
r
eq
u
en
c
y
d
o
m
ai
n
in
f
o
r
m
atio
n
,
an
d
d
ee
p
lear
n
i
n
g
-
b
ased
em
b
e
d
d
in
g
s
.
Au
g
m
en
tatio
n
attr
ib
u
tes
an
d
d
ataset
s
p
lit
ass
ig
n
m
en
ts
ar
e
also
p
r
o
v
id
e
d
f
o
r
r
e
p
r
o
d
u
cib
ili
ty
an
d
m
o
d
el
v
alid
atio
n
p
u
r
p
o
s
es.
T
h
ese
f
ea
tu
r
es
to
g
eth
er
e
n
ab
le
a
h
o
lis
tic
an
d
m
u
lti
-
f
ac
eted
an
aly
s
is
o
f
c
h
est C
T
im
ag
es f
o
r
r
o
b
u
s
t c
lass
if
icatio
n
an
d
i
n
ter
p
r
etab
ilit
y
.
T
ab
le
2
.
Ov
e
r
v
iew
o
f
e
x
tr
ac
te
d
f
ea
tu
r
e
ca
te
g
o
r
ies
F
e
a
t
u
r
e
c
a
t
e
g
o
r
y
F
e
a
t
u
r
e
n
a
me
D
e
scri
p
t
i
o
n
Ty
p
e
/
u
s
e
B
a
si
c
i
ma
g
e
met
a
d
a
t
a
I
mag
e
f
i
l
e
n
a
m
e
F
i
l
e
n
a
me
(
i
d
e
n
t
i
f
i
e
r
)
o
f
t
h
e
C
T
i
ma
g
e
I
d
e
n
t
i
f
i
e
r
F
o
l
d
e
r
/
l
a
b
e
l
G
r
o
u
n
d
-
t
r
u
t
h
c
l
a
ss
l
a
b
e
l
:
a
d
e
n
o
c
a
r
c
i
n
o
m
a
,
l
a
r
g
e
c
e
l
l
c
a
r
c
i
n
o
ma
,
s
q
u
a
m
o
u
s
c
e
l
l
c
a
r
c
i
n
o
ma
,
n
o
r
m
a
l
Ta
r
g
e
t
/
l
a
b
e
l
R
e
s
o
l
u
t
i
o
n
O
r
i
g
i
n
a
l
i
ma
g
e
w
i
d
t
h
×
h
e
i
g
h
t
(
p
i
x
e
l
s
)
N
u
meri
c
(
i
m
a
g
e
p
r
o
p
e
r
t
y
)
F
o
r
mat
F
i
l
e
t
y
p
e
(
e
.
g
.
,
JP
G
,
P
N
G
)
C
a
t
e
g
o
r
i
c
a
l
P
i
x
e
l
-
l
e
v
e
l
f
e
a
t
u
r
e
s
M
e
a
n
p
i
x
e
l
i
n
t
e
n
s
i
t
y
A
v
e
r
a
g
e
g
r
a
y
v
a
l
u
e
a
c
r
o
ss a
l
l
p
i
x
e
l
s
N
u
meri
c
(
i
m
a
g
e
s
t
a
t
i
s
t
i
c
)
S
t
a
n
d
a
r
d
d
e
v
i
a
t
i
o
n
P
i
x
e
l
i
n
t
e
n
si
t
y
s
p
r
e
a
d
N
u
meri
c
(
i
m
a
g
e
s
t
a
t
i
s
t
i
c
)
S
k
e
w
n
e
ss
a
n
d
k
u
r
t
o
si
s
D
i
st
r
i
b
u
t
i
o
n
s
h
a
p
e
o
f
p
i
x
e
l
i
n
t
e
n
s
i
t
i
e
s
N
u
meri
c
(
i
m
a
g
e
s
t
a
t
i
s
t
i
c
)
Te
x
t
u
r
e
f
e
a
t
u
r
e
s
G
LC
M
c
o
n
t
r
a
s
t
Lo
c
a
l
c
o
n
t
r
a
s
t
c
o
m
p
u
t
e
d
f
r
o
m
G
L
C
M
Te
x
t
u
r
e
d
e
s
c
r
i
p
t
o
r
G
LC
M
e
n
e
r
g
y
Te
x
t
u
r
e
u
n
i
f
o
r
m
i
t
y
f
r
o
m G
L
C
M
Te
x
t
u
r
e
d
e
s
c
r
i
p
t
o
r
G
LC
M
h
o
m
o
g
e
n
e
i
t
y
H
o
w
u
n
i
f
o
r
m
t
h
e
t
e
x
t
u
r
e
i
s
a
c
r
o
ss a
n
i
mag
e
Te
x
t
u
r
e
d
e
s
c
r
i
p
t
o
r
En
t
r
o
p
y
I
n
f
o
r
mat
i
o
n
c
o
n
t
e
n
t
/
c
o
m
p
l
e
x
i
t
y
i
n
t
h
e
i
ma
g
e
Te
x
t
u
r
e
d
e
s
c
r
i
p
t
o
r
S
h
a
p
e
/
st
r
u
c
t
u
r
e
f
e
a
t
u
r
e
s
Lu
n
g
a
r
e
a
P
r
o
p
o
r
t
i
o
n
o
f
l
u
n
g
r
e
g
i
o
n
v
s
b
a
c
k
g
r
o
u
n
d
(
f
r
o
m
seg
m
e
n
t
a
t
i
o
n
)
N
u
meri
c
N
o
d
u
l
e
c
o
u
n
t
/
s
i
z
e
N
u
mb
e
r
a
n
d
si
z
e
o
f
d
e
t
e
c
t
e
d
n
o
d
u
l
e
s
N
u
meri
c
H
i
st
o
g
r
a
m
f
e
a
t
u
r
e
s
P
i
x
e
l
i
n
t
e
n
si
t
y
h
i
s
t
o
g
r
a
m
b
i
n
s
F
r
e
q
u
e
n
c
y
o
f
p
i
x
e
l
i
n
t
e
n
s
i
t
i
e
s
i
n
f
i
x
e
d
b
i
n
s
N
u
meri
c
l
i
s
t
/
v
e
c
t
o
r
F
r
e
q
u
e
n
c
y
d
o
m
a
i
n
FFT
sp
e
c
t
r
a
l
f
e
a
t
u
r
e
s
F
r
e
q
u
e
n
c
i
e
s fr
o
m
f
a
s
t
F
o
u
r
i
e
r
t
r
a
n
sf
o
r
m
N
u
meri
c
l
i
s
t
/
v
e
c
t
o
r
D
e
e
p
l
e
a
r
n
i
n
g
f
e
a
t
u
r
e
s
C
N
N
e
m
b
e
d
d
i
n
g
s
(
p
r
e
-
t
r
a
i
n
e
d
)
F
e
a
t
u
r
e
v
e
c
t
o
r
s
f
r
o
m
p
r
e
-
t
r
a
i
n
e
d
C
N
N
l
a
y
e
r
s
(
e
.
g
.
,
R
e
sN
e
t
,
v
i
su
a
l
g
e
o
met
r
y
g
r
o
u
p
(
VGG
)
)
N
u
meri
c
v
e
c
t
o
r
F
e
a
t
u
r
e
m
a
p
s
A
c
t
i
v
a
t
i
o
n
m
a
p
s
f
r
o
m
i
n
t
e
r
me
d
i
a
t
e
C
N
N
l
a
y
e
r
s
N
u
meri
c
a
r
r
a
y
s
A
u
g
m
e
n
t
a
t
i
o
n
a
t
t
r
i
b
u
t
e
s
R
o
t
a
t
i
o
n
R
a
n
d
o
m r
o
t
a
t
i
o
n
a
p
p
l
i
e
d
d
u
r
i
n
g
a
u
g
m
e
n
t
a
t
i
o
n
A
u
g
m
e
n
t
e
d
m
e
t
a
d
a
t
a
F
l
i
p
/
sc
a
l
e
W
h
e
t
h
e
r
i
ma
g
e
w
a
s
f
l
i
p
p
e
d
o
r
s
c
a
l
e
d
i
n
p
r
e
p
r
o
c
e
ssi
n
g
A
u
g
m
e
n
t
e
d
m
e
t
a
d
a
t
a
C
l
a
s
s l
a
b
e
l
C
a
n
c
e
r
t
y
p
e
A
d
e
n
o
c
a
r
c
i
n
o
m
a
-
l
a
r
g
e
c
e
l
l
c
a
r
c
i
n
o
m
a
-
sq
u
a
m
o
u
s
c
e
l
l
c
a
r
c
i
n
o
ma
-
n
o
r
mal
Ta
r
g
e
t
v
a
r
i
a
b
l
e
Tr
a
i
n
/
v
a
l
i
d
a
t
i
o
n
/
t
e
st
sp
l
i
t
S
e
t
a
ssi
g
n
me
n
t
D
e
n
o
t
e
s wh
i
c
h
s
p
l
i
t
t
h
e
i
m
a
g
e
b
e
l
o
n
g
s t
o
C
a
t
e
g
o
r
i
c
a
l
u
se
d
f
o
r
t
r
a
i
n
i
n
g
/
v
a
l
i
d
a
t
i
o
n
/
t
e
st
i
n
g
3
.
2
.
P
r
o
po
s
ed
m
et
ho
do
lo
g
y
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
e
m
p
lo
y
s
a
f
ea
tu
r
e
-
le
v
el
f
u
s
io
n
f
r
am
ewo
r
k
t
h
at
co
m
b
in
es
d
ee
p
lear
n
in
g
an
d
r
ad
io
m
ics to
class
if
y
an
d
d
iag
n
o
s
e
lu
n
g
d
is
ea
s
e
f
r
o
m
ch
est C
T
s
ca
n
s
ac
r
o
s
s
m
u
lti
-
clas
s
d
is
ea
s
e
s
tag
es.
T
o
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
5
6
8
-
4
5
9
1
4574
d
o
th
is
,
d
ata
ar
e
p
r
o
ce
s
s
ed
u
s
i
n
g
a
d
u
al
-
s
tr
ea
m
ar
c
h
itectu
r
e,
with
d
ee
p
a
b
s
tr
ac
t
s
p
atial
f
ea
tu
r
es
d
er
iv
ed
f
r
o
m
th
e
R
esNet5
0
C
NN
p
r
e
-
tr
ain
ed
o
n
im
ag
e
d
ata
alo
n
e,
as
well
as
q
u
an
titativ
e
tis
s
u
e
te
x
tu
r
e
ch
ar
ac
te
r
is
tics
u
s
in
g
h
an
d
c
r
af
ted
r
ad
i
o
m
ic
d
escr
ip
to
r
s
,
s
p
ec
if
ically
,
th
e
GL
C
M
an
d
s
tati
s
tical
en
tr
o
p
y
.
Af
ter
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
b
o
th
s
tr
ea
m
s
o
f
i
n
f
o
r
m
atio
n
ar
e
c
o
n
ca
ten
ated
to
cr
ea
te
o
n
e
h
y
b
r
id
h
ig
h
-
d
im
en
s
io
n
al
f
ea
tu
r
e
v
ec
to
r
,
wh
ich
will
th
en
b
e
n
o
r
m
alize
d
to
a
co
n
s
is
ten
t
s
ca
le
(
Min
Ma
x
)
b
ef
o
r
e
b
ein
g
m
o
d
ell
ed
b
y
an
XGBo
o
s
t
en
s
em
b
le
class
if
ier
f
o
r
g
en
er
a
tin
g
th
e
f
in
al
class
if
icatio
n
p
r
ed
ictio
n
.
T
h
e
m
eth
o
d
o
lo
g
y
in
clu
d
es
a
d
u
al
-
m
o
d
e
XAI
ap
p
r
o
ac
h
,
th
e
u
s
e
o
f
Gr
ad
-
C
AM
f
o
r
s
p
atial
l
o
ca
lizatio
n
an
d
SHAP
f
o
r
f
ea
tu
r
e
im
p
o
r
tan
ce
s
co
r
es
o
f
f
in
d
in
g
s
in
o
r
d
er
to
im
p
r
o
v
e
cl
in
ical
in
ter
p
r
etab
ilit
y
.
T
h
e
tech
n
ical
r
ea
lizatio
n
o
f
t
h
e
p
r
o
p
o
s
ed
p
ip
elin
e
is
s
tr
u
ctu
r
ed
in
to
s
ev
er
al
k
ey
s
tag
es.
First,
d
ata
p
r
ep
r
o
ce
s
s
in
g
is
co
n
d
u
cted
t
o
en
s
u
r
e
im
ag
e
q
u
ality
an
d
c
o
n
s
is
ten
cy
ac
r
o
s
s
th
e
d
ataset.
Nex
t,
th
e
p
r
o
ce
s
s
s
p
ec
if
ically
ad
d
r
ess
es
th
e
id
en
tifie
d
r
esear
ch
g
ap
s
an
d
in
co
r
p
o
r
ates
th
e
s
tu
d
y
’
s
n
o
v
el
co
n
tr
ib
u
tio
n
s
.
T
o
tack
le
th
e
r
is
k
o
f
m
o
d
el
o
v
e
r
f
itti
n
g
,
an
ex
p
licit
s
tag
e
is
d
ed
icate
d
to
an
aly
zin
g
th
e
o
v
er
f
it
tin
g
co
n
d
itio
n
a
n
d
im
p
lem
en
tin
g
p
r
o
p
o
s
ed
m
itig
atio
n
s
.
T
h
e
p
ip
elin
e
th
en
p
r
o
ce
ed
s
with
a
d
u
al
-
b
r
a
n
ch
f
ea
tu
r
e
ex
tr
ac
tio
n
ap
p
r
o
ac
h
,
s
im
u
ltan
eo
u
s
ly
lev
e
r
ag
in
g
b
o
th
d
ee
p
lear
n
in
g
an
d
h
an
d
cr
a
f
ted
f
ea
tu
r
e
tech
n
iq
u
e
s
.
T
h
ese
ex
tr
ac
ted
f
ea
tu
r
es
ar
e
s
u
b
s
eq
u
en
tly
in
teg
r
ated
th
r
o
u
g
h
f
ea
tu
r
e
-
le
v
e
l
f
u
s
io
n
to
m
a
x
im
ize
r
e
p
r
es
en
tatio
n
al
p
o
wer
.
Fin
ally
,
th
e
wo
r
k
f
lo
w
co
n
clu
d
es
with
class
if
icatio
n
p
o
s
t
-
p
r
o
ce
s
s
in
g
an
d
a
th
o
r
o
u
g
h
in
ter
p
r
etab
ilit
y
an
aly
s
is
,
en
s
u
r
in
g
tr
a
n
s
p
ar
en
t a
n
d
clin
i
ca
lly
r
elev
an
t d
ec
is
io
n
-
m
ak
in
g
.
T
h
e
d
etailed
n
etwo
r
k
s
tr
u
ctu
r
e
an
d
th
e
d
im
en
s
io
n
s
o
f
th
e
d
ata
in
ea
ch
lay
er
ar
e
p
r
e
s
en
ted
in
Fig
u
r
e
2
.
T
h
e
i
n
p
u
t
ten
s
o
r
(
2
2
4
×
2
2
4
×
3
)
af
ter
p
r
ep
r
o
ce
s
s
in
g
is
p
ass
ed
to
a
f
r
o
ze
n
R
esN
et5
0
b
ac
k
b
o
n
e
.
T
h
e
f
ea
tu
r
e
m
ap
s
ar
e
f
in
ally
p
o
o
le
d
d
o
wn
v
ia
g
lo
b
al
av
er
ag
e
p
o
o
lin
g
(
GAP)
to
a
1
×
2
0
4
8
v
ec
to
r
.
Me
an
wh
ile,
th
e
1
s
t
–
6
th
v
o
x
els
o
f
th
e
r
ad
io
m
i
c
v
ec
to
r
ar
e
ex
tr
ac
ted
f
r
o
m
t
h
e
g
r
ay
s
ca
le
in
p
u
t.
T
h
ese
ar
e
th
en
co
n
ca
ten
ated
d
escr
ip
tiv
ely
in
to
a
1
×
2
0
5
4
h
y
b
r
id
v
ec
to
r
,
wh
ich
is
t
h
er
e
af
ter
s
ca
led
u
s
in
g
Min
Ma
x
Sc
aler
an
d
f
ed
to
t
h
e
XGBo
o
s
t c
las
s
if
ier
,
wh
ich
o
u
t
p
u
ts
a
1
×
3
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
.
Fig
u
r
e
2
.
Netwo
r
k
ar
ch
itectu
r
e
an
d
d
ata
d
im
en
s
io
n
s
at
ea
ch
s
tag
e
3
.
3
.
Da
t
a
prepro
ce
s
s
ing
All
th
e
r
aw
C
T
im
ag
es
wer
e
p
r
ep
r
o
ce
s
s
ed
id
en
tically
to
a
llo
w
co
n
s
is
ten
t
in
p
u
ts
f
o
r
th
e
d
if
f
er
en
t
f
ea
tu
r
e
ex
tr
ac
tio
n
s
tr
ea
m
s
.
T
h
e
s
ize
o
f
th
e
im
ag
es
is
r
escaled
to
2
2
4
×
2
2
4
.
T
h
is
s
tu
d
y
s
ca
le
s
p
ix
el
v
alu
es
in
r
an
g
e
[
0
,
1
]
with
Min
-
Ma
x
s
ca
lin
g
f
o
r
f
ast
co
n
v
e
r
g
en
ce
.
I
n
th
e
d
ee
p
lea
r
n
in
g
b
r
an
c
h
p
r
e
s
er
v
e
th
e
3
-
c
h
an
n
el
R
GB
f
o
r
m
at
(
r
ep
licatin
g
th
e
g
r
ay
s
ca
le
ch
a
n
n
el
th
r
ee
tim
es
if
r
eq
u
ir
e
d
)
to
f
it
th
e
in
p
u
t
s
p
ec
if
icatio
n
s
o
f
p
r
e
-
tr
ain
ed
b
ac
k
b
o
n
e
.
C
o
n
cu
r
r
en
tly
,
a
s
in
g
le
-
c
h
an
n
el
g
r
ay
s
ca
le
is
p
r
o
d
u
ce
d
f
o
r
th
e
r
ad
io
m
ic
tex
tu
r
e
an
al
y
s
is
b
r
an
ch
.
Du
r
i
n
g
th
e
p
r
ep
r
o
ce
s
s
in
g
s
tag
e
o
f
o
u
r
h
y
b
r
id
f
r
am
ewo
r
k
,
s
ev
er
al
n
o
is
e
r
e
d
u
ctio
n
ap
p
r
o
ac
h
es
wer
e
in
teg
r
ated
with
a
v
iew
to
en
h
an
cin
g
class
if
icatio
n
r
o
b
u
s
tn
ess
an
d
in
h
ib
itin
g
ac
q
u
is
itio
n
-
r
elate
d
ar
tifa
cts.
T
h
e
ch
est
C
T
im
ag
es
f
r
eq
u
en
tly
s
h
o
w
q
u
an
tu
m
n
o
is
e,
r
ec
o
n
s
t
r
u
ctio
n
ar
tifa
cts
,
an
d
s
lig
h
t
i
n
ten
s
ity
v
ar
iatio
n
s
,
wh
ich
ar
e
d
etr
im
e
n
tal
to
d
ee
p
f
ea
tu
r
e
ex
tr
ac
tio
n
as
well
as
h
an
d
cr
af
ted
te
x
tu
r
e
a
n
aly
s
is
.
Acc
o
r
d
in
g
l
y
,
im
ag
es
wer
e
r
esized
to
a
co
m
m
o
n
s
ca
le
2
2
4
×
2
2
4
p
ix
els
an
d
th
e
im
ag
es
wer
e
ex
p
o
s
u
r
e
n
o
r
m
alize
d
f
o
r
u
n
if
o
r
m
co
n
tr
ast s
ca
lin
g
p
r
io
r
to
f
ea
tu
r
e
ex
tr
ac
tio
n
.
3
.
4
.
Address
ing
t
he
re
s
ea
rc
h g
a
p a
nd
no
v
el
co
ntr
ibu
t
io
ns
T
h
er
e
h
as
b
ee
n
a
f
ast
-
m
o
v
in
g
p
r
o
ce
s
s
o
f
d
ev
elo
p
in
g
C
A
D
f
o
r
lu
n
g
ca
n
ce
r
.
Ho
wev
er
,
a
cr
itical
r
ev
iew
o
f
th
e
r
esear
ch
liter
atu
r
e
s
h
o
ws
th
at
th
er
e
s
till
ar
e
a
n
u
m
b
er
o
f
im
p
o
r
tan
t h
o
les
in
th
e
cu
r
r
e
n
t
liter
atu
r
e
av
ailab
le
o
n
th
e
d
e
v
elo
p
m
e
n
t
o
f
th
is
s
u
b
ject
ar
ea
.
Ma
n
y
o
f
th
e
cu
r
r
en
t
m
o
d
els ar
e
b
ased
o
n
to
o
m
an
y
s
ep
ar
ate
an
d
u
n
r
elate
d
s
tag
es
th
at
ca
u
s
e
ad
d
itio
n
al
p
r
o
ce
s
s
in
g
co
s
ts
,
r
eq
u
ir
e
th
e
im
p
r
ac
tical
u
s
e
o
f
m
u
ltip
le
im
ag
in
g
m
o
d
alities
C
T
,
p
o
s
itro
n
em
i
s
s
io
n
to
m
o
g
r
ap
h
y
(
PET
)
/C
T
,
ch
est
x
-
r
ay
(
C
XR
)
)
to
g
et
h
er
,
an
d
/o
r
ar
e
to
o
tim
e
-
co
n
s
u
m
in
g
b
ec
au
s
e
o
f
c
o
m
p
lex
co
m
p
u
tatio
n
s
r
eq
u
ir
e
d
f
o
r
f
ea
tu
r
e
s
elec
tio
n
.
I
n
ad
d
itio
n
,
m
o
s
t
o
f
th
e
b
etter
-
p
er
f
o
r
m
in
g
m
o
d
els
ar
e
also
co
n
s
id
er
ed
"b
lack
-
b
o
x
"
m
o
d
els,
m
ea
n
in
g
th
e
y
d
o
n
o
t
p
r
o
v
id
e
en
o
u
g
h
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
Qu
a
n
tita
tive
ch
est co
mp
u
ted
to
mo
g
r
a
p
h
y
-
b
a
s
ed
h
yb
r
id
d
ee
p
lea
r
n
in
g
fo
r
ch
r
o
n
ic
l
u
n
g
…
(
Ga
r
ima
Ja
in
)
4575
d
etails
ab
o
u
t
h
o
w
th
e
u
n
d
e
r
ly
in
g
alg
o
r
ith
m
s
wo
r
k
to
g
a
in
clin
ician
'
s
tr
u
s
t
in
th
eir
u
s
e.
T
h
e
r
esear
ch
co
n
tr
ib
u
tio
n
s
o
f
th
is
d
is
s
er
tatio
n
ar
e
f
o
cu
s
ed
o
n
r
eso
lv
in
g
th
e
r
esear
ch
p
r
o
b
lem
s
o
u
tli
n
ed
a
b
o
v
e
an
d
ar
e
o
r
g
an
ized
in
to
two
ca
teg
o
r
ies:
th
eo
r
etica
l c
o
n
tr
i
b
u
tio
n
s
an
d
p
r
ac
tical
co
n
tr
ib
u
tio
n
s
.
T
h
eo
r
etica
l
c
o
n
tr
ib
u
tio
n
s
:
i
)
u
n
if
ied
s
tr
u
ctu
r
al
f
ea
tu
r
e
-
lev
el
f
u
s
io
n
:
a
c
o
m
p
letely
n
ew
s
tr
u
ctu
r
e
th
a
t
co
n
n
ec
ts
th
e
s
em
an
tic
g
ap
o
f
ab
s
tr
ac
t
d
ee
p
lear
n
in
g
to
d
eter
m
in
is
tic
r
ad
io
m
ics.
T
h
is
m
eth
o
d
in
teg
r
ates
m
ea
s
u
r
ab
le,
tr
a
d
itio
n
al
r
ad
io
m
ic
d
escr
ip
to
r
s
GL
C
M
an
d
en
tr
o
p
y
)
as
i
n
p
u
ts
in
t
o
th
e
d
ee
p
lear
n
i
n
g
f
ea
tu
r
e
s
p
ac
e,
allo
win
g
f
o
r
a
r
ap
i
d
,
th
o
r
o
u
g
h
s
tu
d
y
o
f
th
e
tu
m
o
r
p
r
o
f
ile
with
o
u
t h
av
in
g
to
u
tili
ze
two
f
u
n
ctio
n
s
f
o
r
th
e
s
am
e
im
ag
e
(
f
ea
tu
r
e
ex
tr
ac
to
r
an
d
class
if
ier
)
in
a
d
ee
p
lear
n
in
g
m
o
d
el
;
an
d
ii)
m
ax
im
izin
g
s
in
g
le
m
o
d
ality
:
d
em
o
n
s
tr
atin
g
th
e
d
iag
n
o
s
tic
an
d
p
r
ed
ictiv
e
ca
p
ab
ilit
ies
o
f
s
in
g
le
-
m
o
d
ality
C
T
s
ca
n
s
,
t
h
er
eb
y
elim
in
atin
g
r
elian
ce
o
n
c
o
m
p
lex
an
d
s
o
m
e
tim
es im
p
r
ac
tical
m
u
lti
-
m
o
d
al
im
ag
in
g
d
ata.
Pra
ctica
l
c
o
n
tr
ib
u
tio
n
s
:
i)
r
eso
u
r
ce
-
ef
f
icien
t
p
ip
elin
e
d
e
s
ig
n
:
th
is
s
y
s
tem
ac
h
iev
es
m
ax
im
u
m
ef
f
icien
cy
in
p
ip
elin
e
d
esig
n
b
y
u
s
in
g
a
r
o
b
u
s
t
h
y
b
r
id
ar
ch
itectu
r
e
co
m
p
r
is
in
g
a
R
esNet5
0
b
ac
k
b
o
n
e
,
an
XGBo
o
s
t
cla
s
s
if
ier
,
an
d
a
n
o
n
-
co
m
p
u
tatio
n
ally
in
ten
s
iv
e
ev
o
lu
tio
n
ar
y
alg
o
r
ith
m
.
Fu
r
th
er
m
o
r
e
,
b
y
elim
in
atin
g
th
e
tr
ad
itio
n
al
s
o
f
t
m
ax
lay
er
f
r
o
m
th
e
a
r
ch
itectu
r
e,
th
er
e
is
v
er
y
litt
le
lik
elih
o
o
d
o
f
o
v
er
f
itti
n
g
o
n
lar
g
e,
co
m
p
lex
d
atasets
;
an
d
ii)
d
u
al
-
m
o
d
e
XAI
f
r
am
ew
o
r
k
:
a
“d
u
al
-
s
id
ed
”
XAI
in
ter
p
r
et
ab
ilit
y
s
y
s
tem
will
b
e
p
io
n
ee
r
e
d
with
v
is
u
al
s
p
atial
lo
ca
lizatio
n
p
r
o
v
id
ed
b
y
Gr
ad
-
C
AM
co
m
b
in
ed
with
f
ea
tu
r
e
attr
ib
u
tio
n
s
tatis
t
ics
g
en
er
ated
b
y
SHAP,
to
r
em
o
v
e
th
e
“
b
lack
-
b
o
x
”
o
f
AI
s
y
s
tem
s
an
d
p
r
o
v
id
e
g
r
an
u
lar
an
d
tr
an
s
p
ar
e
n
t
in
f
o
r
m
atio
n
n
ec
ess
ar
y
to
d
ev
e
lo
p
p
h
y
s
ician
tr
u
s
t a
n
d
s
u
p
p
o
r
t c
lin
ical
im
p
lem
en
tatio
n
.
3
.
5
.
T
he
o
v
er
f
i
t
t
ing
co
nd
it
io
n a
nd
pro
po
s
ed
m
it
ig
a
t
io
ns
On
e
o
f
th
e
m
ain
ch
allen
g
es
wh
en
u
s
in
g
d
ee
p
c
o
n
v
o
lu
tio
n
al
n
etwo
r
k
m
o
d
els
f
o
r
m
e
d
ical
im
ag
e
class
if
icatio
n
,
p
ar
ticu
lar
ly
wh
en
tr
ain
ed
o
n
d
o
m
ain
-
s
p
ec
if
ic
d
atasets
with
r
elativ
ely
f
ew
ex
am
p
les,
is
o
v
er
f
itti
n
g
.
W
h
en
a
m
o
d
el
o
v
er
f
its
,
it
m
em
o
r
izes
s
tatis
t
ical
ar
tifa
cts
f
r
o
m
th
e
tr
ain
in
g
d
ata
r
ath
er
th
an
lear
n
in
g
th
e
u
n
d
er
ly
in
g
an
ato
m
ical
p
ath
o
lo
g
y
,
lead
in
g
to
p
o
o
r
g
en
e
r
aliza
tio
n
to
n
ew
m
e
d
ical
im
ag
es.
Ma
n
y
cu
r
r
en
t
s
tu
d
ies
h
av
e
t
r
ied
to
a
d
d
r
ess
th
is
is
s
u
e
b
y
o
v
er
-
r
el
y
in
g
o
n
ar
tific
ial
d
ata
au
g
m
e
n
ta
tio
n
s
,
wh
ich
d
o
n
o
t
f
u
lly
s
o
lv
e
th
e
p
r
o
b
lem
.
I
n
co
n
tr
ast,
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
f
r
am
ew
o
r
k
s
o
lv
es
an
d
p
r
ev
en
ts
o
v
er
f
itti
n
g
u
s
in
g
th
r
e
e
d
if
f
er
en
t
s
tr
u
ctu
r
al
d
esig
n
d
ec
is
io
n
s
:
i)
T
r
an
s
f
er
lear
n
in
g
with
f
r
o
ze
n
weig
h
ts
:
th
is
s
tu
d
y
u
s
ed
a
p
r
e
-
tr
ain
ed
R
esNet5
0
m
o
d
el
th
a
t
was
tr
ain
ed
u
s
in
g
I
m
ag
eNe
t
as
o
u
r
b
ac
k
b
o
n
e
m
o
d
el
with
o
u
t
m
o
d
if
y
i
n
g
th
e
co
n
v
o
lu
tio
n
al
weig
h
ts
;
th
e
weig
h
ts
o
f
ea
ch
lay
er
r
em
ain
u
n
c
h
an
g
e
d
th
r
o
u
g
h
o
u
t
th
is
p
r
o
ce
s
s
.
T
h
er
e
f
o
r
e,
th
e
m
o
d
el
was
ab
le
to
lev
er
ag
e
r
o
b
u
s
t,
g
en
er
ic
ca
p
ab
ilit
ies
f
o
r
d
etec
ti
n
g
e
d
g
es
an
d
p
atter
n
s
ac
r
o
s
s
im
ag
es
with
o
u
t
m
ass
iv
ely
h
y
p
er
-
s
p
ec
ializin
g
its
p
ar
am
eter
s
o
n
th
e
lim
ited
a
m
o
u
n
t
o
f
d
ata
we
h
ad
av
ailab
l
e.
ii)
Dim
en
s
io
n
ality
co
n
s
tr
ain
t:
th
is
s
tu
d
y
u
s
ed
GAP
to
cr
ea
te
a
s
in
g
le
ca
ll
o
f
th
e
C
NN
m
o
d
el
an
d
o
u
tp
u
t
an
ar
r
ay
o
f
6
-
d
im
en
s
io
n
al
v
ec
to
r
s
o
f
tex
tu
r
e
f
ea
tu
r
es
to
b
e
in
p
u
t
to
th
e
r
ad
io
m
ic
m
o
d
el.
T
h
e
r
ad
io
m
i
c
m
o
d
el
h
as
a
h
ig
h
d
eg
r
ee
o
f
d
im
en
s
io
n
ality
am
o
n
g
its
v
ar
iab
les;
th
er
ef
o
r
e,
im
p
lem
en
tin
g
th
e
GAP
p
r
o
ce
d
u
r
e
p
r
o
v
id
es
a
m
ea
s
u
r
e
o
f
d
im
e
n
s
io
n
ality
r
estrictio
n
,
wh
ich
h
elp
s
u
s
p
r
e
v
en
t
o
v
er
f
itti
n
g
as
ex
p
er
ien
ce
d
th
r
o
u
g
h
th
e
m
o
d
e
l w
ith
an
ex
ce
s
s
o
f
p
a
r
am
eter
s
.
iii)
Alg
o
r
it
h
m
ic
r
e
g
u
la
r
iz
ati
o
n
v
i
a
XGB
o
o
s
t:
th
is
s
tu
d
y
r
e
p
l
ac
ed
th
e
o
u
t
p
u
t d
e
n
s
e
s
o
f
tm
ax
l
ay
e
r
o
f
o
u
r
m
o
d
el
wit
h
a
n
XGB
o
o
s
t
c
lass
i
f
i
ca
t
i
o
n
h
ea
d
d
u
e
t
o
th
e
f
a
ct
th
at
d
e
n
s
e
s
o
f
t
m
a
x
p
r
e
d
i
cti
o
n
l
a
y
er
s
a
r
e
h
i
g
h
l
y
o
v
e
r
f
itti
n
g
w
h
e
n
b
u
il
t
o
u
t
o
f
s
m
al
l
a
m
o
u
n
ts
o
f
d
at
a.
B
y
u
s
i
n
g
XGB
o
o
s
t,
th
is
s
t
u
d
y
was
a
b
le
t
o
a
d
d
a
n
esta
b
lis
h
ed
a
lg
o
r
it
h
m
th
at
al
r
e
ad
y
f
ea
t
u
r
es
a
n
u
m
b
e
r
o
f
m
ec
h
an
is
m
s
d
esi
g
n
e
d
to
p
r
e
v
e
n
t
o
v
er
f
it
ti
n
g
w
h
e
n
p
r
o
d
u
ci
n
g
a
p
r
o
b
a
b
ilis
tic
p
r
ed
i
cti
o
n
b
as
ed
u
p
o
n
h
ig
h
-
d
im
en
s
i
o
n
al
i
n
p
u
t
f
e
at
u
r
e
v
e
ct
o
r
s
.
3
.
6
.
E
x
perim
ent
a
l
s
et
up
a
nd
v
a
lid
a
t
io
n str
a
t
eg
y
T
h
e
s
tu
d
y
was
d
esig
n
ed
to
b
e
r
ep
r
o
d
u
cib
le,
r
o
b
u
s
t
,
an
d
f
air
ly
co
m
p
a
r
ab
le.
Fo
r
th
is
r
ea
s
o
n
,
all
ex
p
er
im
en
ts
wer
e
co
n
d
u
cted
u
s
in
g
a
s
tan
d
ar
d
ized
co
m
p
u
tatio
n
al
a
n
d
v
alid
ati
o
n
f
r
am
ewo
r
k
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
ar
ch
itectu
r
e
co
n
s
is
ted
o
f
th
r
ee
m
ain
co
m
p
o
n
e
n
ts
:
a
C
NN
-
b
ased
b
aselin
e
ar
ch
itectu
r
e,
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
els
f
o
r
b
en
ch
m
a
r
k
in
g
,
an
d
im
p
le
m
en
tatio
n
s
o
f
b
o
th
th
e
b
ase
an
d
b
en
ch
m
ar
k
ed
ar
ch
itectu
r
es u
s
in
g
d
ee
p
lear
n
i
n
g
in
Py
th
o
n
with
g
r
ap
h
ics p
r
o
ce
s
s
in
g
u
n
it (
GPU
)
ac
ce
ler
atio
n
.
Du
r
in
g
th
e
e
x
p
er
im
e
n
tal
p
h
ase,
th
e
d
atasets
wer
e
p
ar
titi
o
n
ed
u
s
in
g
s
tr
atif
ied
r
an
d
o
m
s
am
p
lin
g
to
en
s
u
r
e
eq
u
al
s
am
p
le
s
izes
ac
r
o
s
s
all
an
aly
ze
d
p
ath
o
lo
g
ica
l
d
is
ea
s
e
c
lass
e
s
i.e
.
,
n
o
r
m
al,
ad
en
o
ca
r
cin
o
m
a,
s
q
u
am
o
u
s
ce
ll
ca
r
cin
o
m
a,
a
n
d
lar
g
e
ce
ll
ca
r
ci
n
o
m
a
.
T
o
m
in
im
ize
o
v
er
f
itti
n
g
,
g
iv
en
th
e
li
m
ited
v
ar
ia
b
ilit
y
in
th
e
m
ed
ical
im
ag
es,
t
h
is
s
tu
d
y
u
s
ed
m
u
ltip
le
au
g
m
en
tati
o
n
s
tr
ateg
ies
d
u
r
i
n
g
m
o
d
el
t
r
ain
in
g
.
All
o
f
th
e
au
g
m
en
tatio
n
s
tr
ateg
ies
wer
e
d
esig
n
ed
to
m
im
ic
ac
q
u
is
itio
n
v
ar
iab
ilit
y
,
v
ar
ia
b
ilit
y
in
an
at
o
m
ical
o
r
ien
tatio
n
,
an
d
in
ten
s
ity
v
ar
iatio
n
s
th
at
ar
e
f
r
eq
u
e
n
tly
ex
p
er
ie
n
ce
d
i
n
clin
ical
im
ag
in
g
en
v
ir
o
n
m
en
ts
.
Au
g
m
en
tatio
n
s
tr
ateg
ies
wer
e
d
y
n
am
ically
av
ailab
le
f
o
r
im
p
r
o
v
in
g
th
e
r
o
b
u
s
tn
ess
o
f
m
o
d
els
tr
ain
e
d
o
n
h
eter
o
g
en
eo
u
s
im
ag
in
g
e
n
v
ir
o
n
m
en
ts
.
T
o
en
s
u
r
e
th
at
th
e
m
o
d
els
wer
e
r
ep
r
o
d
u
cib
le
an
d
y
ield
ed
co
n
s
is
ten
t
r
esu
lts
ac
r
o
s
s
all
ex
p
er
im
en
ts
u
s
ed
to
co
m
p
a
r
e
th
em
,
th
is
s
tu
d
y
u
s
ed
a
s
tr
atif
ied
v
alid
atio
n
p
r
o
t
o
co
l
an
d
f
ix
ed
r
a
n
d
o
m
in
itializatio
n
s
ee
d
s
.
T
h
e
ex
p
e
r
im
en
t
co
n
s
is
ted
o
f
p
er
f
o
r
m
an
ce
ev
alu
atio
n
m
etr
ics
o
f
class
if
icatio
n
ac
cu
r
ac
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
5
6
8
-
4
5
9
1
4576
m
ac
r
o
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
c
o
n
f
u
s
io
n
m
atr
ix
r
esu
lts
an
d
R
OC
-
AUC
an
aly
s
is
o
f
one
-
vs
-
r
est cla
s
s
if
icatio
n
r
esu
lts
.
T
ab
le
3
p
r
esen
ts
th
e
co
m
p
u
t
atio
n
al
en
v
ir
o
n
m
en
t
a
n
d
s
o
f
t
war
e
s
etu
p
u
s
ed
f
o
r
im
p
le
m
en
tin
g
an
d
test
in
g
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
f
r
am
ewo
r
k
.
T
o
s
p
ee
d
u
p
th
e
tr
a
in
in
g
o
f
d
ee
p
f
ea
tu
r
es
ex
tr
ac
te
d
f
r
o
m
im
ag
es
an
d
f
o
r
tr
an
s
f
o
r
m
er
b
e
n
ch
m
ar
k
in
g
ex
p
er
im
e
n
ts
,
GPU
co
m
p
u
tatio
n
was
u
s
ed
.
T
h
e
s
o
f
twar
e
lib
r
ar
ies
th
at
wer
e
u
s
ed
an
d
th
e
h
ar
d
war
e
s
etu
p
s
th
at
wer
e
u
s
ed
f
o
r
ea
c
h
o
f
th
e
ex
p
er
im
en
ts
wer
e
k
e
p
t
co
n
s
ta
n
t
ac
r
o
s
s
all
o
f
th
e
ex
p
er
im
en
ts
s
o
th
at
r
esu
lts
c
o
u
ld
b
e
r
ep
licated
an
d
an
u
n
b
iased
co
m
p
ar
is
o
n
co
u
ld
b
e
m
ad
e
b
etwe
en
th
e
d
if
f
er
en
t
v
er
s
io
n
s
o
f
th
e
s
o
f
t
war
e.
T
ab
le
4
s
h
o
ws
th
e
d
at
aset
p
ar
titi
o
n
in
g
an
d
v
alid
atio
n
s
tr
ateg
y
th
at
was
em
p
lo
y
ed
th
r
o
u
g
h
o
u
t
th
e
test
i
n
g
an
d
ev
alu
atio
n
p
r
o
ce
s
s
.
T
o
m
ain
tain
b
alan
ce
b
etwe
en
class
es
in
th
e
tr
ain
in
g
,
v
alid
atio
n
a
n
d
test
d
atasets
,
as
well
as
r
ed
u
ce
th
e
e
f
f
ec
ts
o
f
b
ias
in
th
e
d
if
f
er
e
n
t
test
in
g
m
eth
o
d
s
,
s
tr
atif
ie
d
s
am
p
lin
g
an
d
k
-
f
o
ld
cr
o
s
s
v
alid
atio
n
wer
e
u
s
ed
wh
en
cr
ea
tin
g
th
e
d
if
f
er
en
t
test
s
ets.
T
ab
le
5
o
u
tlin
es
th
e
way
s
in
wh
ich
th
e
tr
ain
in
g
d
ataset
was
au
g
m
en
ted
in
o
r
d
er
to
p
r
o
v
id
e
a
b
etter
a
b
ilit
y
f
o
r
th
e
d
e
v
elo
p
ed
m
o
d
e
l
to
g
en
er
alize
,
as
well
as
r
ed
u
c
e
th
e
p
o
te
n
tial
f
o
r
o
v
er
f
itti
n
g
,
an
d
p
r
o
v
i
d
e
a
m
ea
s
u
r
e
o
f
v
ar
iab
ilit
y
th
at
ca
n
b
e
ex
p
ec
ted
f
r
o
m
h
eter
o
g
e
n
eo
u
s
d
ata
s
ets d
er
iv
ed
f
r
o
m
th
e
im
a
g
in
g
o
f
th
e
ch
est u
s
in
g
C
T
.
T
ab
le
3
.
C
o
m
p
u
tatio
n
al
en
v
ir
o
n
m
en
t a
n
d
s
o
f
twar
e
co
n
f
ig
u
r
atio
n
C
o
m
p
o
n
e
n
t
S
p
e
c
i
f
i
c
a
t
i
o
n
G
P
U
N
V
I
D
I
A
R
TX
3
0
6
0
C
P
U
I
n
t
e
l
C
o
r
e
i
7
R
A
M
1
6
G
B
O
p
e
r
a
t
i
n
g
sy
st
e
m
W
i
n
d
o
w
s 1
1
/
U
b
u
n
t
u
2
2
.
0
4
P
r
o
g
r
a
m
mi
n
g
l
a
n
g
u
a
g
e
P
y
t
h
o
n
3
.
1
0
D
e
e
p
l
e
a
r
n
i
n
g
f
r
a
mew
o
r
k
Te
n
s
o
r
F
l
o
w
/
K
e
r
a
s
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
l
i
b
r
a
r
i
e
s
S
c
i
k
i
t
-
l
e
a
r
n
a
n
d
X
G
B
o
o
st
C
U
D
A
v
e
r
s
i
o
n
C
U
D
A
1
2
.
0
I
mag
e
p
r
o
c
e
ssi
n
g
l
i
b
r
a
r
y
O
p
e
n
C
V
Ex
p
l
a
i
n
a
b
i
l
i
t
y
l
i
b
r
a
r
i
e
s
S
H
A
P
a
n
d
G
r
a
d
-
C
A
M
T
ab
le
4
.
Data
s
et
p
ar
titi
o
n
in
g
a
n
d
v
alid
atio
n
s
tr
ateg
y
V
a
l
i
d
a
t
i
o
n
c
o
m
p
o
n
e
n
t
C
o
n
f
i
g
u
r
a
t
i
o
n
Tr
a
i
n
i
n
g
sp
l
i
t
7
0
%
V
a
l
i
d
a
t
i
o
n
s
p
l
i
t
1
5
%
Te
st
i
n
g
s
p
l
i
t
1
5
%
V
a
l
i
d
a
t
i
o
n
st
r
a
t
e
g
y
S
t
r
a
t
i
f
i
e
d
s
a
mp
l
i
n
g
C
r
o
ss
-
v
a
l
i
d
a
t
i
o
n
me
t
h
o
d
S
t
r
a
t
i
f
i
e
d
k
-
f
o
l
d
N
u
mb
e
r
o
f
f
o
l
d
s
5
R
a
n
d
o
m s
e
e
d
42
C
l
a
s
s
d
i
s
t
r
i
b
u
t
i
o
n
p
r
e
ser
v
a
t
i
o
n
En
a
b
l
e
d
Ev
a
l
u
a
t
i
o
n
me
t
r
i
c
s
A
c
c
u
r
a
c
y
,
p
r
e
c
i
s
i
o
n
,
r
e
c
a
l
l
,
F
1
-
sc
o
r
e
,
sen
s
i
t
i
v
i
t
y
,
sp
e
c
i
f
i
c
i
t
y
,
R
O
C
-
AUC
T
ab
le
5
.
Data
au
g
m
en
tatio
n
tech
n
iq
u
es a
p
p
lied
d
u
r
in
g
tr
ai
n
in
g
A
u
g
m
e
n
t
a
t
i
o
n
t
e
c
h
n
i
q
u
e
C
o
n
f
i
g
u
r
a
t
i
o
n
R
o
t
a
t
i
o
n
r
a
n
g
e
±
1
5
°
H
o
r
i
z
o
n
t
a
l
f
l
i
p
En
a
b
l
e
d
Zo
o
m ra
n
g
e
0
.
1
W
i
d
t
h
sh
i
f
t
0
.
1
H
e
i
g
h
t
s
h
i
f
t
0
.
1
B
r
i
g
h
t
n
e
ss
a
d
j
u
s
t
me
n
t
0
.
1
R
e
sc
a
l
i
n
g
1
/
2
5
5
I
n
t
e
r
p
o
l
a
t
i
o
n
st
r
a
t
e
g
y
B
i
l
i
n
e
a
r
A
u
g
m
e
n
t
a
t
i
o
n
a
p
p
l
i
c
a
t
i
o
n
R
e
a
l
-
t
i
me
d
y
n
a
mi
c
a
u
g
me
n
t
a
t
i
o
n
3
.
7
.
Dua
l
-
bra
nch f
ea
t
ure
ex
t
ra
ct
io
n
T
h
e
d
u
al
-
f
ea
t
u
r
e
ex
tr
ac
tio
n
ar
ch
itectu
r
e
in
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
f
r
am
ew
o
r
k
ex
tr
ac
ts
co
m
p
lem
en
tar
y
in
f
o
r
m
atio
n
ab
o
u
t
th
e
tex
tu
r
e
o
f
ch
est
C
T
im
ag
es,
b
o
t
h
s
p
a
tially
an
d
q
u
a
n
titativ
ely
.
T
h
e
r
atio
n
ale
f
o
r
u
s
in
g
th
e
d
u
al
-
f
ea
tu
r
e
ex
tr
ac
tio
n
ar
c
h
itectu
r
e
is
to
m
er
g
e
th
e
ca
p
a
b
ilit
ies
o
f
a
d
ee
p
C
NN
to
lear
n
at
m
u
ltip
le
lev
els
o
f
s
em
an
tic
h
ier
ar
ch
ical
d
ata,
with
th
e
u
s
e
o
f
h
a
n
d
cr
a
f
ted
r
ad
io
m
ic
f
ea
tu
r
es
lear
n
ed
f
r
o
m
th
e
C
T
im
a
g
es.
Dee
p
lear
n
in
g
em
b
ed
d
in
g
s
ex
ce
l
at
r
ep
r
esen
tin
g
c
o
m
p
lex
p
atter
n
s
in
p
u
lm
o
n
ar
y
s
p
atial
d
is
tr
ib
u
tio
n
s
,
wh
ile
r
ad
io
m
ics
-
b
ased
an
aly
s
es
p
r
o
v
id
e
q
u
an
titativ
e
m
ea
s
u
r
es
o
f
tis
s
u
e
h
eter
o
g
en
eity
,
s
tr
u
ctu
r
al
ir
r
eg
u
lar
ity
,
an
d
in
ten
s
ity
v
ar
iatio
n
o
f
ten
f
o
u
n
d
in
m
alig
n
an
t tis
s
u
e.
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
Qu
a
n
tita
tive
ch
est co
mp
u
ted
to
mo
g
r
a
p
h
y
-
b
a
s
ed
h
yb
r
id
d
ee
p
lea
r
n
in
g
fo
r
ch
r
o
n
ic
l
u
n
g
…
(
Ga
r
ima
Ja
in
)
4577
3
.
7
.
1
.
Dee
p
s
pa
t
ia
l f
ea
t
ure
ex
t
ra
ct
io
n bra
nch
(
CNN
-
ba
s
ed
)
Fo
r
lear
n
in
g
r
ep
r
esen
tatio
n
s
o
f
p
o
ly
g
o
n
s
,
th
e
p
r
im
ar
y
C
NN
b
ac
k
b
o
n
e
was
a
p
r
e
-
tr
ain
e
d
R
esNet5
0
ar
ch
itectu
r
e
with
I
m
ag
eNe
t
weig
h
ts
.
R
es
Net5
0
was
ch
o
s
en
f
o
r
its
ab
ilit
y
to
ex
tr
ac
t
f
ea
tu
r
es
q
u
ick
ly
an
d
ea
s
ily
,
its
s
u
itab
ilit
y
f
o
r
r
esid
u
al
lear
n
in
g
,
a
n
d
its
r
elativ
e
s
u
cc
ess
in
im
ag
e
a
n
aly
s
is
with
in
th
e
h
ea
lth
ca
r
e
in
d
u
s
tr
y
.
T
o
a
v
o
id
o
v
er
f
itti
n
g
th
at
is
ca
u
s
ed
b
y
th
e
s
m
all
am
o
u
n
t
o
f
d
ata,
tr
an
s
f
er
lear
n
in
g
was
u
s
ed
with
R
esNet5
0
to
tak
e
ad
v
a
n
tag
e
o
f
th
e
lo
w
-
lev
el
ed
g
es,
co
n
to
u
r
s
,
an
d
tex
tu
r
es
lear
n
e
d
f
r
o
m
im
ag
es
in
a
lar
g
e
n
u
m
b
er
o
f
ca
teg
o
r
ies,
wh
ile
u
s
in
g
th
e
R
esNet5
0
lo
w
-
lev
el
co
n
v
o
l
u
tio
n
al
lay
er
s
to
h
elp
t
r
an
s
f
er
k
ee
p
with
in
th
e
g
en
e
r
al
ar
ea
o
f
th
e
im
ag
es
,
co
m
b
in
in
g
n
atu
r
al
an
d
m
ed
i
ca
l
im
ag
es
to
g
eth
er
.
T
h
e
p
r
e
p
r
o
ce
s
s
ed
d
ata
we
r
e
r
esized
to
2
2
4
×
2
2
4
×
3
a
n
d
p
ass
ed
th
r
o
u
g
h
th
e
co
n
v
o
lu
tio
n
al
lay
er
s
o
f
R
esNet5
0
.
T
h
e
f
ir
s
t
lay
er
o
f
co
n
v
o
l
u
tio
n
al
lay
er
s
was
f
r
o
ze
n
at
th
e
b
eg
in
n
in
g
o
f
tr
ain
in
g
,
an
d
h
ig
h
e
r
-
lev
el
lay
er
s
wer
e
ad
ju
s
ted
af
ter
en
d
-
s
tag
e
tr
ain
in
g
to
b
etter
r
e
p
r
esen
t
th
e
p
o
ly
g
o
n
s
u
s
in
g
th
e
n
ew
in
f
o
r
m
atio
n
lear
n
e
d
.
Af
ter
ap
p
ly
in
g
a
GAP
o
p
er
atio
n
to
th
e
o
u
t
p
u
t
o
f
th
e
f
in
al
co
n
v
o
lu
tio
n
,
it
is
p
o
s
s
i
b
le
to
co
n
d
en
s
e
th
e
f
ea
tu
r
e
m
ap
s
s
p
atially
wh
il
e
p
r
eser
v
in
g
th
eir
d
is
cr
im
in
ativ
e
s
em
an
tic
co
n
ten
t.
T
h
e
GAP
o
p
er
atio
n
p
r
o
d
u
ce
d
a
co
m
p
ac
t
2
0
4
8
-
d
im
en
s
io
n
al
em
b
ed
d
in
g
v
ec
to
r
f
o
r
c
o
n
v
e
r
tin
g
h
ig
h
-
d
im
en
s
io
n
al
c
o
n
v
o
lu
tio
n
al
f
ea
tu
r
e
m
ap
s
in
to
a
r
ep
r
esen
tatio
n
o
f
m
u
ltid
im
en
s
io
n
al
d
ee
p
em
b
ed
d
in
g
h
ier
ar
ch
ical
an
ato
m
y
o
f
lu
n
g
s
an
d
h
o
w
lesi
o
n
s
ar
e
c
o
n
t
ex
tu
ally
d
e
f
in
ed
b
y
th
e
d
is
ea
s
es
p
r
esen
t.
T
h
ese
2
0
4
8
-
d
im
en
s
io
n
al
em
b
ed
d
i
n
g
s
ca
p
tu
r
e
c
o
m
p
lex
d
if
f
e
r
en
ce
s
in
th
e
s
tr
u
ct
u
r
e
o
f
lu
n
g
s
b
y
ca
teg
o
r
izin
g
th
em
ac
co
r
d
i
n
g
t
o
s
p
ec
if
ic
d
is
ea
s
es,
s
u
ch
as
ad
en
o
ca
r
cin
o
m
a,
s
q
u
am
o
u
s
ce
ll
ca
r
cin
o
m
a
,
a
n
d
lar
g
e
ce
ll c
ar
ci
n
o
m
a.
3
.
7
.
2
.
H
a
nd
cr
a
f
t
ed
qu
a
ntit
a
t
iv
e
t
ex
t
ure
f
e
a
t
ure
bra
nch
(
ra
dio
m
ics
-
ba
s
ed
)
I
n
p
a
r
allel,
th
e
r
ad
io
m
ics
b
r
a
n
ch
was
in
teg
r
ated
in
t
o
th
e
m
o
d
el
to
ex
tr
ac
t
h
an
d
cr
af
ted
q
u
an
titativ
e
m
ea
s
u
r
es
f
r
o
m
in
ten
s
ity
-
n
o
r
m
alize
d
ch
est
C
T
s
lice
im
ag
es.
R
ad
io
m
ic
an
aly
s
is
was
p
er
f
o
r
m
e
d
u
s
in
g
th
e
Scik
it
-
im
ag
e
(
s
k
im
ag
e
)
lib
r
ar
y
to
ass
ess
th
e
s
tatis
t
ical
an
d
tex
tu
r
al
ch
ar
ac
ter
is
tics
ass
o
ciat
ed
with
d
estru
ctiv
e
d
is
ea
s
e
p
r
o
g
r
ess
io
n
an
d
m
alig
n
an
cy
s
tag
in
g
in
th
e
lu
n
g
s
.
Fr
o
m
ea
ch
C
T
im
ag
e,
a
co
m
p
ac
t
s
et
o
f
s
ix
r
ad
io
m
ic
d
escr
ip
to
r
s
was c
alcu
lated
u
s
in
g
GL
C
M
-
b
ased
tex
tu
r
e
a
n
aly
s
is
an
d
g
lo
b
al
s
tatis
tical
in
ten
s
ity
m
etr
ics:
i)
GL
C
M
-
b
ased
tex
tu
r
e
f
ea
tu
r
es
:
th
e
p
u
r
p
o
s
e
o
f
p
e
r
f
o
r
m
in
g
GL
C
M
an
aly
s
is
is
to
q
u
a
n
tif
y
r
elatio
n
s
h
ip
s
am
o
n
g
in
ten
s
ity
v
alu
es
ac
r
o
s
s
p
ix
els
with
in
an
d
am
o
n
g
lu
n
g
tis
s
u
es.
T
h
r
ee
d
escr
ip
to
r
s
wer
e
ca
lcu
lated
as r
ep
r
esen
tatio
n
s
o
f
tex
t
u
r
e
c
h
ar
ac
ter
is
tics
:
‒
C
o
n
tr
ast:
m
ea
s
u
r
es th
e
d
eg
r
ee
o
f
lo
ca
l v
a
r
iatio
n
in
tex
t
u
r
e
in
t
en
s
ity
an
d
ir
r
e
g
u
lar
ity
.
‒
Un
if
o
r
m
ity
:
m
ea
s
u
r
es h
o
w
s
im
ilar
an
d
co
n
s
is
ten
t n
eig
h
b
o
r
i
n
g
p
ix
el
v
alu
e
d
is
tr
ib
u
tio
n
s
ar
e.
‒
E
n
er
g
y
:
m
ea
s
u
r
es
h
o
w
co
n
s
is
ten
tly
tex
tu
r
ed
o
b
jects
ar
e
in
ter
m
s
o
f
co
m
m
o
n
r
ep
ea
tin
g
s
tr
u
ctu
r
es
o
r
p
atter
n
s
.
ii)
Glo
b
al
s
tatis
tical
f
ea
tu
r
es:
o
th
er
ty
p
es
o
f
f
ir
s
t
-
o
r
d
e
r
s
tatis
tical
d
escr
ip
to
r
s
wer
e
u
s
ed
t
o
esti
m
ate
th
e
d
is
tr
ib
u
tio
n
o
f
g
lo
b
al
g
r
ay
s
ca
l
e
in
ten
s
ity
an
d
to
c
h
ar
ac
ter
ize
th
e
im
ag
e
co
m
p
lex
ity
:
‒
Me
an
g
r
ay
le
v
el:
p
r
o
v
id
es
an
av
er
ag
e
o
f
all
g
r
a
y
s
ca
le
lev
els f
o
u
n
d
in
th
e
im
ag
e.
‒
Stan
d
ar
d
d
ev
iatio
n
:
p
r
o
v
id
es
a
m
ea
s
u
r
e
o
f
t
h
e
r
a
n
g
e
o
f
in
t
en
s
ity
o
f
all
v
alu
es
f
o
u
n
d
at
e
ac
h
p
ix
el
an
d
h
o
w
m
u
ch
v
ar
iatio
n
e
x
is
ts
am
o
n
g
th
o
s
e
p
ix
els.
‒
Sh
an
n
o
n
en
tr
o
p
y
:
p
r
o
v
id
es
a
m
ea
s
u
r
e
o
f
h
o
w
r
a
n
d
o
m
l
y
d
is
tr
ib
u
ted
g
r
ay
s
ca
le
lev
els
ar
e
with
in
th
e
im
ag
e,
an
d
h
o
w
h
eter
o
g
e
n
eo
u
s
th
ey
m
ay
b
e
.
W
h
en
co
n
s
id
er
ed
to
g
eth
e
r
,
th
e
co
m
b
in
atio
n
o
f
th
e
r
ad
io
m
ics
d
escr
ip
to
r
s
cr
ea
ted
a
6
-
d
i
m
en
s
io
n
al
q
u
an
titativ
e
tex
tu
r
e
v
ec
to
r
t
h
at
r
ep
r
esen
ts
q
u
an
tifia
b
le,
i
n
ter
p
r
etab
le
s
tatis
tical
p
r
o
p
er
ties
r
eg
ar
d
in
g
th
e
q
u
an
tity
a
n
d
q
u
ality
o
f
s
tr
u
ctu
r
es
with
in
th
e
p
u
lm
o
n
ar
y
tis
s
u
e
lesi
o
n
al
b
o
r
d
er
s
.
T
h
is
v
ec
to
r
p
r
o
v
id
es
a
s
ec
o
n
d
b
r
an
ch
o
f
p
ar
allel
an
aly
s
is
to
d
eliv
er
b
ac
k
g
r
o
u
n
d
in
f
o
r
m
ati
o
n
s
u
p
p
o
r
tin
g
th
e
d
ee
p
C
NN
em
b
ed
d
in
g
an
al
y
s
is
o
f
th
e
s
am
e
d
ata.
Ultim
ately
,
it
in
c
o
r
p
o
r
ates
d
eter
m
i
n
is
tic
tex
tu
r
ed
in
f
o
r
m
atio
n
th
at
is
n
o
t
n
ec
ess
ar
ily
av
ailab
le
f
r
o
m
cu
r
r
e
n
tly
r
etain
ed
laten
t d
ee
p
f
ea
tu
r
e
d
ata.
3
.
8
.
F
e
a
t
ure
-
lev
el
hy
brid f
us
io
n
Featu
r
e
-
lev
el
h
y
b
r
i
d
f
u
s
io
n
i
s
a
tech
n
iq
u
e
wh
er
eb
y
r
a
d
io
m
ic
tex
tu
r
e
d
escr
ip
to
r
s
an
d
d
ee
p
C
NN
r
ep
r
esen
tatio
n
s
ar
e
in
teg
r
ated
to
f
u
lly
ex
p
lo
it
b
o
t
h
th
e
h
ig
h
-
l
ev
el
s
em
an
tic
lear
n
in
g
c
h
ar
ac
t
er
is
tics
in
h
er
en
t
in
d
ee
p
m
o
d
els
an
d
th
e
q
u
an
tit
ativ
e
tis
s
u
e
h
eter
o
g
en
eity
p
r
o
v
id
ed
th
r
o
u
g
h
r
ad
io
m
ic
d
ata.
W
h
ile
d
ee
p
C
NN
ar
ch
itectu
r
es
ef
f
ec
tiv
ely
ca
p
tu
r
e
co
m
p
lex
s
p
atial
an
d
co
n
tex
tu
al
p
u
lm
o
n
ar
y
p
atter
n
s
au
to
m
atica
lly
,
h
an
d
-
c
r
af
ted
r
ad
io
m
ic
tex
tu
r
e
d
escr
ip
to
r
s
y
ield
in
ter
p
r
eta
b
le
s
tatis
tical
ex
p
lan
atio
n
s
o
f
tis
s
u
e
v
ar
iatio
n
s
.
T
h
ese
v
ar
iatio
n
s
i
n
clu
d
e,
b
u
t
ar
e
n
o
t
lim
ited
to
,
p
ix
el
r
eg
u
lar
ity
an
d
ir
r
eg
u
lar
ity
,
in
ten
s
ity
v
ar
iab
ilit
y
,
a
n
d
s
tr
u
ctu
r
al
m
o
r
p
h
o
lo
g
y
.
A
2
0
4
8
D
d
ee
p
-
s
p
atial
em
b
ed
d
in
g
v
ec
to
r
f
r
o
m
an
o
f
f
-
th
e
-
s
h
elf
p
r
e
-
tr
ain
ed
R
esNet5
0
b
ac
k
b
o
n
e
was
h
y
b
r
id
-
f
u
s
ed
with
a
6
D
h
a
n
d
c
r
af
ted
r
a
d
io
m
ic
f
ea
tu
r
e
v
ec
to
r
f
r
o
m
GL
C
M
tex
tu
r
e
s
tatis
tics
an
d
en
tr
o
p
y
-
b
ased
f
ea
tu
r
es
to
cr
ea
te
a
c
o
m
b
i
n
ed
2
0
5
4
D
h
y
b
r
i
d
f
ea
t
u
r
e
v
ec
to
r
f
o
r
d
o
wn
s
tr
ea
m
clas
s
if
icatio
n
.
Min
Ma
x
n
o
r
m
aliza
tio
n
was
a
p
p
lied
to
ad
d
r
ess
s
ca
le
in
co
n
s
is
ten
cies
b
etwe
en
th
e
h
ig
h
-
d
im
en
s
io
n
al
d
ee
p
e
m
b
ed
d
in
g
s
an
d
lo
w
-
d
im
en
s
io
n
al
r
a
d
io
m
ics
p
r
io
r
to
class
if
icatio
n
.
T
h
is
n
o
r
m
aliza
tio
n
m
et
h
o
d
co
n
v
er
ted
all
f
ea
tu
r
e
v
alu
es
to
a
co
m
m
o
n
r
an
g
e
o
f
0
to
1
,
wh
ic
h
im
p
r
o
v
ed
n
u
m
er
ical
s
tab
ilit
y
/co
n
v
e
r
g
en
ce
b
eh
a
v
io
r
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.