I
nte
rna
t
io
na
l J
o
urna
l o
f
I
nfo
rm
a
t
ics a
nd
Co
m
m
un
ica
t
io
n T
ec
hn
o
lo
g
y
(
I
J
-
I
CT
)
Vo
l.
15
,
No
.
2
,
J
u
n
e
20
26
,
p
p
.
497
~
5
0
7
I
SS
N:
2252
-
8
7
7
6
,
DOI
:
1
0
.
1
1
5
9
1
/iji
ct
.
v15
i
2
.
p
p
4
9
7
-
5
0
7
497
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ict.
ia
esco
r
e.
co
m
Enha
nced t
ra
nsfe
r learning
f
ra
me
wo
rk f
o
r b
ra
in t
u
mo
r
detec
tion from
M
RI sca
ns usi
ng
att
ention
-
ba
sed
feat
ure f
usio
n
Sm
it
a
B
ha
rne
1
,
E
k
t
a
Sa
rda
2
,
Sh
a
m
a
l Sa
lun
k
he
2
1
D
e
p
a
r
t
me
n
t
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
T
e
c
h
n
o
l
o
g
y
,
B
h
a
r
a
t
i
V
i
d
y
a
p
e
e
t
h
D
e
e
m
e
d
t
o
b
e
U
n
i
v
e
r
s
i
t
y
,
N
a
v
i
M
u
m
b
a
i
,
I
n
d
i
a
2
R
a
mr
a
o
A
d
i
k
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
D
.
Y
.
P
a
t
i
l
d
e
e
me
d
t
o
b
e
U
n
i
v
e
r
s
i
t
y
,
N
a
v
i
M
u
m
b
a
i
,
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
J
u
n
24
,
2
0
2
5
R
ev
is
ed
Dec
18
,
2
0
2
5
Acc
ep
ted
Mar
30
,
2
0
2
6
Du
e
to
th
e
c
o
m
p
lex
i
ty
o
f
th
e
d
iffere
n
t
t
u
m
o
r
ty
p
e
s
in
m
e
d
ica
l
ima
g
in
g
d
e
tec
ti
o
n
o
f
b
ra
i
n
tu
m
o
r
is
stil
l
a
s
p
ro
m
i
n
e
n
t
c
h
a
ll
e
n
g
e
.
T
h
is
p
a
p
e
r
p
re
se
n
t
th
e
in
n
o
v
a
ti
v
e
tec
h
n
iq
u
e
e
n
h
a
n
c
e
d
tran
sfe
r
lea
rn
in
g
fra
m
e
wo
r
k
(ET
LF
)
wh
ich
i
n
teg
ra
ti
n
g
th
e
a
d
v
a
n
c
e
d
p
re
-
p
ro
c
e
ss
in
g
with
h
y
b
ri
d
fin
e
-
tu
n
e
d
m
e
th
o
d
f
o
r
a
c
c
u
ra
te
b
ra
in
tu
m
o
r
d
e
tec
ti
o
n
fr
o
m
m
a
g
n
e
ti
c
re
so
n
a
n
c
e
ima
g
in
g
(
M
RI
)
sc
a
n
s.
Th
e
p
ro
p
o
se
d
m
o
d
e
l
c
o
m
b
in
e
th
e
stre
n
g
th
o
f
p
re
-
train
e
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
(CNN
s)
su
c
h
a
s
Eff
icie
n
tNe
tB
0
t
h
ro
u
g
h
d
o
m
a
in
sp
e
c
ifi
c
t
ra
n
sfe
r
lea
rn
in
g
a
n
d
a
tt
e
n
ti
o
n
b
a
se
d
fi
n
e
t
u
n
i
n
g
.
A
n
o
v
e
l
fe
a
tu
re
fu
sio
n
lay
e
r
a
n
d
a
d
a
p
ti
v
e
lea
rn
in
g
ra
te
sc
h
e
d
u
ler
a
re
k
e
y
in
d
ica
to
r
s
fo
r
m
o
d
e
l
p
e
rfo
rm
a
n
c
e
a
n
d
p
re
v
e
n
t
o
v
e
rfit
ti
n
g
.
T
h
e
m
e
th
o
d
o
l
o
g
y
i
s
a
ss
e
ss
e
d
o
n
t
h
e
b
e
n
c
h
m
a
rk
d
a
tas
e
t
Bra
TS
a
n
d
Ka
g
g
le
b
ra
i
n
tu
m
o
r
d
a
tas
e
ts.
Th
e
m
a
in
c
o
n
tri
b
u
ti
o
n
o
f
wo
r
k
li
e
s
in
d
e
v
e
lo
p
m
e
n
t
o
f
d
o
m
a
in
-
a
d
a
p
ti
v
e
tran
sfe
r
lea
rn
in
g
with
d
iffere
n
t
d
a
tas
e
ts.
Th
e
E
TL
F
sh
o
ws
th
e
h
ig
h
a
c
c
u
ra
c
y
o
f
9
8
.
7
6
%
wh
ich
a
b
le
o
u
tp
e
rf
o
rm
s
e
ffe
c
ti
v
e
ly
in
d
iag
n
o
si
n
g
t
u
m
o
r
su
it
a
b
le
o
f
c
li
n
ica
l
p
u
rp
o
s
e.
K
ey
w
o
r
d
s
:
B
r
ain
tu
m
o
r
d
etec
tio
n
Dee
p
lear
n
in
g
E
f
f
icien
tNet
Fin
e
-
tu
n
in
g
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
:
Sm
ita
B
h
ar
n
e
Dep
ar
tm
en
t o
f
E
n
g
in
ee
r
in
g
a
n
d
T
ec
h
n
o
lo
g
y
,
B
h
ar
ati
Vid
y
ap
ee
th
Dee
m
ed
to
b
e
Un
iv
er
s
ity
Nav
i M
u
m
b
ai,
I
n
d
ia
E
m
ail:
s
m
ita1
4
6
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
B
r
ain
tu
m
o
r
s
r
e
p
r
esen
t
a
c
r
itical
m
ed
ical
ch
allen
g
e
d
u
e
to
th
eir
ag
g
r
ess
iv
e
n
at
u
r
e
a
n
d
h
ig
h
m
o
r
tality
r
ates.
E
ar
ly
d
etec
tio
n
a
n
d
ac
cu
r
ate
class
if
icatio
n
ar
e
ess
en
tial
f
o
r
g
u
id
in
g
t
h
er
ap
e
u
tic
d
ec
is
io
n
s
an
d
im
p
r
o
v
i
n
g
p
atien
t
o
u
tc
o
m
es
[
1
]
,
[
2
]
.
Du
e
to
e
x
ce
p
tio
n
al
co
n
tr
ast
r
eso
lu
tio
n
an
d
th
e
ca
p
ac
ity
to
v
ie
w
th
e
s
o
f
t
tis
s
u
es
w
ith
o
u
t
io
n
izin
g
r
ad
iatio
n
,
m
a
g
n
etic
r
eso
n
an
ce
im
ag
in
g
(
M
R
I
)
is
th
e
r
ec
o
m
m
en
d
ed
im
a
g
in
g
f
o
r
b
r
ain
tu
m
o
r
d
iag
n
o
s
is
[
3
]
.
B
u
t
in
ad
d
itio
n
to
b
ei
n
g
tim
e
co
n
s
u
m
in
g
,
m
an
u
al
MRI
s
ca
n
in
ter
p
r
etatio
n
is
p
r
o
n
e
to
er
r
o
r
s
wh
en
d
if
f
e
r
en
tiatin
g
b
etwe
en
b
r
ain
tu
m
o
r
ty
p
es a
n
d
g
r
ad
es
[
4
]
.
Ad
v
an
ce
d
m
eth
o
d
o
lo
g
y
lik
e
d
ee
p
lear
n
in
g
(
DL
)
,
tr
an
s
f
er
lear
n
in
g
h
as
d
em
o
n
s
tr
ated
o
u
ts
tan
d
in
g
p
er
f
o
r
m
an
ce
in
v
ar
io
u
s
m
e
d
ical
im
ag
e
an
aly
s
is
task
s
,
in
clu
d
in
g
tu
m
o
r
s
eg
m
e
n
tatio
n
an
d
class
if
icatio
n
[
5
]
,
[
6
]
.
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs
)
ar
e
ca
p
ab
le
o
f
au
to
m
atica
lly
lear
n
in
g
h
ier
ar
ch
ical
f
ea
tu
r
es
f
r
o
m
im
ag
in
g
d
ata,
s
u
r
p
ass
in
g
tr
ad
it
io
n
al
m
eth
o
d
s
th
at
r
ely
o
n
h
an
d
cr
af
ted
f
ea
t
u
r
es
[
7
]
.
Nev
er
th
e
less
,
tr
ain
in
g
d
ee
p
n
eu
r
al
n
etwo
r
k
s
f
r
o
m
s
cr
atch
ty
p
ically
d
em
an
d
s
la
r
g
e
a
n
n
o
tated
d
atasets
,
wh
ich
ar
e
o
f
ten
s
ca
r
ce
in
th
e
m
ed
ical
d
o
m
ain
d
u
e
to
t
h
e
c
o
m
p
lex
ity
an
d
co
s
t
o
f
e
x
p
er
t
lab
ellin
g
[
8
]
.
T
o
ad
d
r
ess
th
is
ch
allen
g
e,
tr
an
s
f
e
r
lear
n
in
g
h
as
em
er
g
e
d
as
an
ef
f
ec
tiv
e
s
tr
ateg
y
th
at
lev
er
ag
e
s
k
n
o
wled
g
e
f
r
o
m
m
o
d
els
p
r
e
-
tr
ain
ed
o
n
lar
g
e
-
s
ca
le
d
atasets
lik
e
I
m
ag
eNe
t
an
d
a
d
ap
ts
th
em
to
s
p
ec
if
ic
m
ed
ical
im
ag
in
g
task
s
[
9
]
.
T
r
an
s
f
er
lear
n
in
g
n
o
t
o
n
ly
ac
ce
ler
ates m
o
d
el
co
n
v
e
r
g
en
ce
b
u
t a
ls
o
im
p
r
o
v
es g
en
e
r
aliza
tio
n
o
n
lim
ited
lab
elled
d
atasets
.
Am
o
n
g
th
e
a
v
ailab
le
C
NN
ar
ch
itectu
r
es,
E
f
f
icien
tNet
h
as
g
ai
n
ed
atte
n
tio
n
f
o
r
its
s
u
p
er
io
r
p
e
r
f
o
r
m
an
ce
-
to
-
co
m
p
lex
ity
r
atio
,
m
ak
in
g
it we
ll
-
s
u
ited
f
o
r
m
ed
ical
im
ag
e
an
al
y
s
is
u
n
d
er
co
m
p
u
tatio
n
al
co
n
s
tr
ain
ts
[
1
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
2
,
J
u
n
e
20
26
:
4
9
7
-
5
0
7
498
Desp
ite
th
e
p
r
o
g
r
ess
,
s
ev
er
al
lim
itatio
n
s
r
em
ain
in
ex
is
tin
g
ap
p
r
o
ac
h
es.
Mo
s
t
co
n
v
en
tio
n
al
tr
an
s
f
er
lear
n
in
g
p
i
p
elin
es
ig
n
o
r
e
d
o
m
ain
d
is
cr
ep
an
cies
b
etwe
en
n
atu
r
al
an
d
m
ed
ical
im
ag
es,
lea
d
in
g
to
s
u
b
o
p
tim
al
p
er
f
o
r
m
an
ce
.
Mo
r
eo
v
er
,
cr
iti
c
al
s
p
atial
an
d
co
n
tex
t
u
al
tu
m
o
r
f
ea
t
u
r
es
m
ay
b
e
d
ilu
ted
d
u
r
in
g
d
o
wn
s
am
p
lin
g
in
d
ee
p
ar
ch
itectu
r
es.
T
o
o
v
er
co
m
e
th
ese
ch
allen
g
es,
th
is
w
o
r
k
in
tr
o
d
u
ce
s
a
n
o
v
el
en
h
a
n
c
ed
tr
an
s
f
er
lear
n
in
g
f
r
am
ewo
r
k
(
E
T
L
F)
th
at
in
teg
r
ates
d
o
m
ain
-
s
p
ec
if
ic
f
in
e
-
tu
n
in
g
,
atten
tio
n
-
b
ased
f
ea
tu
r
e
f
u
s
io
n
,
an
d
ad
ap
tiv
e
lear
n
in
g
s
tr
ateg
ies
f
o
r
im
p
r
o
v
ed
b
r
ain
tu
m
o
r
d
etec
tio
n
f
r
o
m
MRI
s
ca
n
s
.
T
h
e
p
r
im
ar
y
co
n
tr
ib
u
tio
n
s
o
f
th
is
r
esear
ch
ar
e:
i
)
a
d
o
m
ai
n
-
ad
ap
tiv
e
f
in
e
-
tu
n
in
g
s
tr
ateg
y
f
o
r
E
f
f
icien
tNet
to
b
r
id
g
e
th
e
g
ap
b
etwe
en
s
o
u
r
ce
a
n
d
tar
g
et
d
o
m
ain
s
;
ii)
a
n
atten
ti
o
n
-
d
r
i
v
en
f
ea
tu
r
e
f
u
s
io
n
m
o
d
u
le
th
at
en
h
a
n
ce
s
d
is
cr
im
in
ativ
e
lear
n
in
g
f
r
o
m
tu
m
o
r
r
eg
i
o
n
s
;
iii)
ex
ten
s
iv
e
v
alid
atio
n
u
s
in
g
B
r
aT
S
2
0
2
0
an
d
a
p
u
b
licly
av
aila
b
le
Kag
g
le
d
ataset
to
d
em
o
n
s
tr
ate
g
e
n
er
aliza
b
ilit
y
a
n
d
r
o
b
u
s
tn
ess
.
2.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
Sev
er
al
ap
p
r
o
ac
h
es
h
a
v
e
b
ee
n
ex
p
l
o
r
ed
f
o
r
b
r
ain
tu
m
o
r
class
if
icatio
n
f
r
o
m
MRI
im
a
g
es
u
s
in
g
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
a
n
d
d
ee
p
lear
n
in
g
m
eth
o
d
s
.
E
a
r
ly
tech
n
iq
u
es
r
elied
o
n
h
an
d
c
r
af
ted
f
ea
tu
r
es
an
d
class
if
ier
s
lik
e
SV
M,
k
-
NN,
an
d
d
ec
is
io
n
tr
ee
s
.
Ho
we
v
er
,
th
ese
m
eth
o
d
s
lack
ed
r
o
b
u
s
tn
ess
an
d
ad
a
p
tab
ilit
y
.
R
asto
g
i
et
a
l.
[
1
1
]
in
v
esti
g
a
ted
th
e
e
f
f
icac
y
o
f
f
in
e
-
t
u
n
e
d
tr
an
s
f
er
lear
n
in
g
m
o
d
els,
in
clu
d
in
g
I
n
ce
p
tio
n
R
esNetV2
,
VGG1
9
,
Xce
p
tio
n
,
an
d
Mo
b
ileNet
V2
,
f
o
r
b
r
ain
tu
m
o
r
d
etec
tio
n
.
T
h
eir
s
tu
d
y
d
em
o
n
s
tr
ated
th
at
th
e
Xce
p
tio
n
m
o
d
el
ac
h
ie
v
ed
h
ig
h
ac
cu
r
a
cy
f
in
e
-
tu
n
ed
DL
lear
n
in
g
m
o
d
els
in
en
h
an
cin
g
d
iag
n
o
s
tic
p
r
ec
is
io
n
.
A
s
tu
d
y
f
r
o
m
Dis
ci
et
a
l.
[
1
2
]
e
x
p
lo
r
e
d
th
e
u
s
e
o
f
p
r
e
-
tr
ain
ed
d
ee
p
l
ea
r
n
in
g
m
o
d
els
f
o
r
class
if
y
in
g
b
r
ain
MRI
im
ag
es
in
to
ca
te
g
o
r
ies
s
u
ch
as
g
lio
m
a,
m
en
i
n
g
io
m
a,
p
it
u
itar
y
tu
m
o
r
s
,
an
d
n
o
t
u
m
o
r
.
T
h
e
Xce
p
tio
n
m
o
d
el
o
u
tp
er
f
o
r
m
ed
o
th
er
s
h
ig
h
lig
h
tin
g
th
e
ef
f
ec
tiv
en
ess
o
f
tr
an
s
f
er
lear
n
in
g
in
m
ed
ical
im
ag
e
class
if
icatio
n
.
Pan
d
e
an
d
C
h
a
k
i
[
1
3
]
p
r
o
p
o
s
ed
a
n
o
v
el
tr
ip
le
-
m
o
d
u
le
a
p
p
r
o
a
ch
f
o
r
a
u
to
m
ated
b
r
ain
tu
m
o
r
class
if
icatio
n
f
r
o
m
MRI
im
a
g
es.
T
h
e
f
ir
s
t
m
o
d
u
le
u
tili
ze
d
p
r
e
-
tr
ai
n
ed
d
ee
p
lear
n
in
g
m
o
d
els
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
f
o
llo
we
d
b
y
f
ea
t
u
r
e
s
elec
tio
n
an
d
class
if
icati
o
n
m
o
d
u
les,
r
esu
ltin
g
in
im
p
r
o
v
e
d
d
iag
n
o
s
tic
ac
cu
r
ac
y
.
I
s
lam
et
a
l
.
[
1
4
]
f
o
cu
s
ed
o
n
th
e
d
ee
p
lear
n
in
g
alg
o
r
ith
m
o
n
MRI
s
ca
n
s
wi
th
t
h
e
in
teg
r
atio
n
o
f
2
D
C
NNs
wh
ich
in
cr
ea
s
e
th
e
m
o
d
el
p
er
f
o
r
m
an
ce
.
Ali
et
a
l.
[
1
5
]
wo
r
k
e
d
o
n
class
if
icatio
n
o
f
b
r
ain
tu
m
o
r
with
U
-
Net
ar
ch
itectu
r
e
ap
p
lied
to
M
R
I
s
ca
n
s
.
T
h
is
s
tu
d
y
also
ex
p
l
o
r
ed
th
e
v
a
r
io
u
s
C
N
Ns lik
e
I
n
ce
p
tio
n
-
V3
,
VGG1
9
th
r
o
u
g
h
tr
an
s
f
er
lear
n
in
g
ac
h
i
ev
ed
th
e
im
p
r
o
v
e
d
p
e
r
f
o
r
m
an
ce
.
A
s
tu
d
y
f
r
o
m
An
a
n
th
ar
ajan
et
a
l.
[
1
6
]
p
r
esen
ted
th
e
w
o
r
k
o
n
p
r
e
-
p
r
o
ce
s
s
in
g
MRI
im
ag
es
with
ad
ap
tiv
e
en
h
a
n
ce
m
en
t
alg
o
r
it
h
m
an
d
m
e
d
ian
f
ilter
in
g
f
o
llo
wed
b
y
th
e
d
ee
p
lear
n
in
g
m
o
d
els.
A
s
tu
d
y
f
r
o
m
R
ez
k
et
a
l.
[
1
7
]
p
r
esen
ted
t
h
e
tech
n
iq
u
e
with
h
y
b
r
id
d
ee
p
le
ar
n
in
g
m
o
d
el
with
in
te
g
r
atio
n
m
ed
ical
in
ter
n
et
o
f
th
in
g
s
(
I
o
T
s
)
.
T
o
en
s
u
r
e
th
e
p
atien
t
d
ata
s
ec
u
r
ity
t
h
ey
e
n
cr
y
p
t
th
e
MRI
im
ag
es
b
ef
o
r
e
class
if
icatio
n
.
Gu
p
ta
e
t
a
l.
[7
]
u
s
ed
th
e
C
NNs
alg
o
r
ith
m
to
d
etec
t
b
r
ain
tu
m
o
r
.
T
h
e
m
eth
o
d
h
el
p
ed
to
ass
is
t
th
e
r
ad
io
lo
g
is
t
in
d
ec
is
io
n
-
m
ak
i
n
g
t
h
r
o
u
g
h
t
h
is
ac
cu
r
ate
d
etec
tio
n
.
A
h
am
ed
et
a
l.
[
1
8
]
p
r
o
v
id
ed
th
e
c
o
m
p
lete
r
e
v
iew
o
f
d
ee
p
lear
n
in
g
a
p
p
licatio
n
s
f
o
c
u
s
in
g
o
n
s
eg
m
e
n
tatio
n
.
I
t
also
h
ig
h
lig
h
t
th
e
ef
f
ec
t
o
f
DL
m
o
d
els
in
au
to
m
ated
tu
m
o
r
s
eg
m
en
tatio
n
f
r
o
m
m
ed
ical
i
m
ag
es.
Ma
th
iv
an
a
n
et
a
l.
[
9
]
an
d
Dis
ci
et
a
l.
[
1
2
]
in
v
esti
g
ated
th
e
ef
f
icien
c
y
o
f
th
e
DL
tr
an
s
f
er
lear
n
in
g
(
T
L
)
m
o
d
els
f
o
r
ac
cu
r
ate
p
er
f
o
r
m
an
ce
in
b
r
ai
n
tu
m
o
r
d
iag
n
o
s
is
.
T
h
e
r
esear
ch
h
ig
h
lig
h
ted
th
e
p
e
r
f
o
r
m
an
ce
o
f
Mo
b
ileNetV3
an
d
Xce
p
tio
n
m
o
d
el
o
u
tp
e
r
f
o
r
m
ed
b
est
r
e
s
p
ec
tiv
ely
th
an
th
e
ex
is
tin
g
m
o
d
el.
Desp
ite
s
ig
n
if
ican
t
p
r
o
g
r
ess
in
ap
p
ly
in
g
DL
an
d
T
L
f
o
r
b
r
ain
tu
m
o
r
d
etec
tio
n
f
r
o
m
MRI
im
ag
es,
s
ev
er
al
g
ap
s
p
er
s
is
t.
First,
d
o
m
ain
s
h
if
t
ch
allen
g
es
r
em
ain
lar
g
ely
u
n
r
eso
lv
ed
,
with
m
a
n
y
m
o
d
els
s
tr
u
g
g
lin
g
to
g
en
er
alize
ac
r
o
s
s
d
if
f
er
e
n
t
MRI
ac
q
u
is
itio
n
p
r
o
to
co
ls
an
d
in
s
titu
tio
n
s
.
Seco
n
d
,
ex
p
lai
n
ab
ilit
y
an
d
clin
ical
in
ter
p
r
etab
ilit
y
o
f
d
ee
p
m
o
d
e
ls
ar
e
u
n
d
er
ex
p
l
o
r
ed
,
lim
itin
g
th
eir
ad
o
p
tio
n
in
r
ea
l
-
wo
r
ld
d
iag
n
o
s
tics
.
T
h
ir
d
,
d
ata
s
ca
r
city
an
d
a
n
n
o
tatio
n
c
o
s
t
h
in
d
er
th
e
d
ev
elo
p
m
en
t
o
f
r
o
b
u
s
t
m
o
d
els,
esp
ec
ially
f
o
r
r
ar
e
tu
m
o
r
ty
p
es.
Fo
u
r
th
,
f
ew
s
tu
d
ies
em
p
h
asize
cr
o
s
s
-
d
ataset
v
alid
atio
n
,
wh
ich
is
cr
itical
f
o
r
e
n
s
u
r
in
g
g
e
n
er
aliza
tio
n
.
L
astl
y
,
r
ea
l
-
tim
e
an
d
lig
h
tweig
h
t
d
e
p
lo
y
m
en
t
o
n
ed
g
e
d
ev
ices
r
em
ain
s
u
n
d
er
-
ad
d
r
ess
ed
,
im
p
ac
tin
g
th
eir
u
s
e
i
n
telem
ed
icin
e
o
r
lo
w
-
r
eso
u
r
ce
s
ettin
g
s
.
T
o
ad
d
r
ess
th
ese
lim
itatio
n
s
,
th
is
wo
r
k
i
n
tr
o
d
u
c
es
a
n
o
v
el
tr
an
s
f
er
lear
n
in
g
p
ip
elin
e
with
d
o
m
ai
n
-
s
p
ec
if
ic
tu
n
in
g
,
lig
h
tweig
h
t
ar
ch
itectu
r
e,
an
d
a
d
ap
tiv
e
l
ea
r
n
in
g
s
ch
ed
u
les
-
s
ig
n
if
ican
tly
im
p
r
o
v
in
g
class
if
icatio
n
m
etr
ics an
d
r
ea
l
-
w
o
r
ld
f
ea
s
ib
il
ity
.
3.
RE
S
E
ARCH
M
E
T
H
O
D
3.
1
.
Da
t
a
s
et
des
cr
iptio
n
W
e
u
tili
ze
two
b
en
ch
m
ar
k
d
a
tasets
:
B
r
aT
S
2
0
2
0
:
C
o
n
tain
s
m
u
lti
-
m
o
d
al
MRI
s
ca
n
s
(
T
1
,
T
1
c,
T
2
,
FLAI
R
)
with
g
r
o
u
n
d
tr
u
th
s
eg
m
en
tatio
n
s
f
o
r
g
lio
m
a
tu
m
o
r
s
[
1
9
]
–
[
2
2
]
.
Kag
g
le
B
r
ain
T
u
m
o
r
Data
s
et,
in
clu
d
es
3
9
0
3
T
1
-
weig
h
ted
c
o
n
tr
ast
-
e
n
h
an
ce
d
im
a
g
es
ca
teg
o
r
ized
in
to
g
lio
m
a,
m
en
i
n
g
io
m
a,
p
itu
itar
y
tu
m
o
r
,
a
n
d
n
o
r
m
al
[
2
3
]
.
All
im
ag
es
u
n
d
er
wen
t
r
esizin
g
to
2
2
4
×
2
2
4
p
ix
els,
n
o
r
m
aliza
tio
n
,
a
n
d
au
g
m
en
tatio
n
th
r
o
u
g
h
r
an
d
o
m
r
o
tatio
n
s
,
z
o
o
m
s
,
an
d
f
lip
s
to
im
p
r
o
v
e
g
en
e
r
aliza
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
n
h
a
n
ce
d
tr
a
n
s
fer lea
r
n
in
g
fr
a
mewo
r
k
fo
r
b
r
a
in
tu
mo
r
d
etec
tio
n
fr
o
m
MRI
s
ca
n
s
u
s
in
g
…
(
S
mita
B
h
a
r
n
e
)
499
3.
2
.
P
r
o
po
s
ed
enha
nced
t
ra
ns
f
er
lea
rning
f
ra
m
ewo
rk
(
E
T
L
F
)
Fig
u
r
e
1
s
h
o
ws
th
e
ar
ch
itectu
r
e
o
f
p
r
o
p
o
s
ed
s
y
s
tem
.
I
t
in
cl
u
d
es
th
e
f
o
llo
win
g
c
o
m
p
o
n
en
ts
:
i)
b
ase
mod
el:
E
f
f
icien
tNetB
0
p
r
e
-
tr
a
in
ed
o
n
I
m
a
g
eNe
t
is
u
s
ed
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
.
I
n
itial
lay
er
s
ar
e
f
r
o
ze
n
d
u
r
in
g
ea
r
ly
tr
ain
in
g
s
tag
es
;
ii)
d
o
m
ain
-
ad
ap
tiv
e
f
in
e
-
tu
n
in
g
:
t
o
p
lay
er
s
ar
e
f
in
e
-
tu
n
ed
u
s
in
g
a
s
m
all
lear
n
in
g
r
ate,
wh
ile
in
ter
m
ed
iate
lay
er
s
ar
e
s
elec
tiv
ely
u
n
f
r
o
ze
n
u
s
in
g
a
co
s
in
e
a
n
n
ea
lin
g
s
ch
ed
u
le
r
;
i
ii)
f
ea
tu
r
e
f
u
s
io
n
m
o
d
u
le:
a
tten
tio
n
-
b
ased
f
u
s
i
o
n
lay
er
co
m
b
in
es
s
p
atial
an
d
ch
an
n
el
-
wis
e
atten
tio
n
to
em
p
h
asize
tu
m
o
r
-
s
p
ec
if
ic
r
eg
io
n
s
iv
)
c
lass
if
icatio
n
h
ea
d
:
a
d
e
n
s
e
b
lo
ck
with
d
r
o
p
o
u
t
(
0
.
3
)
,
b
atch
n
o
r
m
aliz
atio
n
,
an
d
s
o
f
tm
a
x
ac
tiv
atio
n
p
er
f
o
r
m
s
m
u
lti
-
cl
as
s
class
if
icat
io
n
.
Fig
u
r
e
1
.
Ar
c
h
itectu
r
e
o
f
p
r
o
p
o
s
ed
s
y
s
tem
3.
3
.
Da
t
a
pre
-
pro
ce
s
s
ing
T
h
e
in
itial
p
h
ase
in
v
o
lv
es
s
y
s
tem
atic
p
r
e
-
p
r
o
ce
s
s
in
g
to
n
o
r
m
alize
an
d
s
tan
d
ar
d
ize
b
r
ain
M
R
I
im
ag
es
ac
r
o
s
s
m
o
d
alities
an
d
ac
q
u
is
itio
n
s
o
u
r
ce
s
.
T
h
e
s
tep
s
ar
e
as f
o
llo
ws:
a.
R
esizin
g
:
all
im
ag
es a
r
e
r
esize
d
to
2
2
4
×2
2
4
p
ix
els to
m
atch
t
h
e
in
p
u
t sh
a
p
e
b
y
E
f
f
icien
tNet
B
0
.
b.
No
r
m
aliza
t
io
n
:
p
ix
el
in
te
n
s
ity
v
alu
es a
r
e
s
ca
led
to
th
e
[
0
,
1
]
r
an
g
e
f
o
r
u
n
i
f
o
r
m
ity
ac
r
o
s
s
b
at
ch
es.
c.
C
o
n
tr
ast
E
n
h
an
ce
m
e
n
t:
h
is
to
g
r
am
e
q
u
aliza
tio
n
a
n
d
co
n
tr
ast
lim
ited
ad
ap
tiv
e
h
is
to
g
r
a
m
eq
u
aliza
tio
n
(
C
L
AHE
)
ar
e
ap
p
lied
to
ac
ce
n
tu
ate
tu
m
o
r
r
eg
io
n
s
.
d.
Data
au
g
m
en
tatio
n
:
to
r
ed
u
ce
o
v
er
f
itti
n
g
an
d
s
im
u
late
im
a
g
in
g
v
ar
iab
ilit
y
,
we
ap
p
ly
r
a
n
d
o
m
r
o
tatio
n
s
(
±
1
5
°),
h
o
r
iz
o
n
tal/v
er
tical
f
lip
s
,
r
an
d
o
m
z
o
o
m
(
0
.
8
x
–
1
.
2
x
)
,
Gau
s
s
ian
n
o
is
e
ad
d
itio
n
.
3.
4
.
B
a
s
e
net
wo
r
k
s
elec
t
io
n a
nd
ini
t
ia
liza
t
io
n
W
e
ad
o
p
t
E
f
f
icien
tNetB
0
,
a
h
ig
h
ly
ef
f
icien
t
C
NN
ar
ch
itectu
r
e
k
n
o
wn
f
o
r
its
co
m
p
o
u
n
d
s
ca
lin
g
o
f
wid
th
,
d
ep
th
,
an
d
r
eso
l
u
tio
n
.
I
t
is
p
r
e
-
tr
ain
ed
o
n
I
m
ag
e
Net
an
d
s
elec
ted
d
u
e
to
,
lo
w
p
ar
am
eter
co
u
n
t
(
~5
.
3
m
illi
o
n
)
,
b
alan
ce
d
p
e
r
f
o
r
m
an
ce
-
to
-
c
o
m
p
u
tatio
n
r
atio
,
an
d
p
r
o
v
en
s
u
cc
ess
in
m
ed
ical
im
ag
in
g
co
n
tex
ts
.
T
h
e
b
ase
lay
er
s
u
p
to
th
e
p
en
u
ltima
te
co
n
v
o
lu
tio
n
al
b
lo
ck
ar
e
in
itially
f
r
o
ze
n
t
o
r
etain
g
en
er
al
f
ea
tu
r
es
.
3.
5
.
Do
m
a
in
-
a
da
ptiv
e
f
ine
-
t
un
ing
s
t
ra
t
eg
y
T
o
tack
le
th
e
d
if
f
e
r
en
ce
b
etwe
en
n
atu
r
al
an
d
m
ed
ical
im
ag
es
,
we
em
p
lo
y
p
r
o
g
r
ess
iv
e
u
n
f
r
e
ez
in
g
an
d
co
s
in
e
an
n
ea
lin
g
lear
n
in
g
r
at
e
s
ch
ed
u
lin
g
,
wh
ich
co
n
s
is
ts
o
f
lay
e
r
-
wis
e
u
n
f
r
ee
zin
g
.
Gr
a
d
u
al
u
n
f
r
ee
zin
g
o
f
lay
er
s
s
tar
tin
g
f
r
o
m
d
ee
p
er
la
y
er
s
to
war
d
ea
r
lier
o
n
es.
Dy
n
am
ic
lear
n
in
g
r
ate
th
at
b
eg
in
s
at
1
e
-
3
an
d
d
ec
ay
s
f
o
llo
win
g
a
co
s
in
e
cu
r
v
e
to
p
r
ev
en
t
ea
r
ly
c
o
n
v
e
r
g
en
ce
.
W
ith
d
is
cr
im
in
ativ
e
f
in
e
-
tu
n
in
g
,
d
if
f
e
r
en
t
lear
n
in
g
r
ates
ar
e
ass
ig
n
ed
to
d
if
f
er
en
t
lay
er
s
(
h
ig
h
er
f
o
r
later
lay
er
s
,
lo
wer
f
o
r
ea
r
lier
o
n
es)
to
o
p
t
im
ize
task
-
s
p
ec
if
ic
f
ea
tu
r
e
lear
n
in
g
.
3.
6
.
F
e
a
t
ure
f
us
io
n m
o
du
le
wit
h a
t
t
ent
io
n m
ec
ha
nis
m
A
n
o
v
el
f
ea
tu
r
e
f
u
s
io
n
m
o
d
u
le
(
FF
M)
is
p
r
o
p
o
s
ed
to
en
h
an
ce
tu
m
o
r
r
eg
io
n
d
et
ec
tio
n
b
y
em
p
h
asizin
g
d
is
cr
im
in
ativ
e
f
ea
tu
r
es.
Fig
u
r
e
2
s
h
o
ws
th
e
f
ea
tu
r
e
f
u
s
io
n
m
o
d
u
le
with
att
en
tio
n
m
ec
h
an
is
m
.
T
h
e
f
ea
tu
r
e
f
u
s
io
n
m
o
d
u
le
(
FF
M)
en
h
an
ce
s
th
e
d
is
cr
im
i
n
ativ
e
ca
p
ab
ilit
y
o
f
ex
t
r
ac
ted
f
ea
tu
r
e
m
ap
s
b
y
in
co
r
p
o
r
atin
g
c
h
an
n
el
a
n
d
s
p
atial
atten
tio
n
m
ec
h
an
is
m
s
,
e
n
ab
lin
g
th
e
n
etwo
r
k
to
f
o
cu
s
m
o
r
e
ef
f
ec
tiv
ely
o
n
tu
m
o
r
-
s
p
ec
if
ic
r
e
g
io
n
s
in
b
r
ai
n
MRI
im
ag
es.
Fo
llo
win
g
ar
e
th
e
co
m
p
o
n
en
ts
o
f
th
e
FF
M.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
2
,
J
u
n
e
20
26
:
4
9
7
-
5
0
7
500
-
I
n
p
u
t
f
ea
t
u
r
es:
∈
ℝ
×
×
th
e
in
p
u
t
to
th
e
m
o
d
u
le
is
a
3
D
f
ea
tu
r
e
m
a
p
f
r
o
m
th
e
b
ac
k
b
o
n
e
C
NN
(
e.
g
.
,
E
f
f
icien
tNetB
0
)
,
wh
er
e
is
Hei
g
h
t,
is
W
id
th
an
d
Nu
m
b
er
o
f
ch
an
n
els.
-
Ou
tp
u
t
f
ea
t
u
r
es:
∈
ℝ
×
×
th
e
o
u
tp
u
t
r
et
ain
s
th
e
s
am
e
d
im
en
s
io
n
ality
as
th
e
in
p
u
t
b
u
t
is
r
ich
er
in
tu
m
o
r
-
r
ele
v
an
t
co
n
tex
tu
al
an
d
s
p
atial
in
f
o
r
m
atio
n
,
im
p
r
o
v
in
g
d
o
wn
s
tr
ea
m
class
if
icatio
n
a
cc
u
r
ac
y
.
-
C
h
an
n
el
atten
tio
n
m
o
d
u
le
(
C
AM
)
:
ca
p
tu
r
es
in
ter
-
ch
an
n
el
d
ep
en
d
e
n
cies
b
y
co
m
p
u
tin
g
weig
h
ted
f
ea
tu
r
e
im
p
o
r
tan
ce
ac
r
o
s
s
ch
an
n
els.
-
Sp
atial
atten
tio
n
m
o
d
u
le
(
SAM)
:
lear
n
s
s
p
atial
lo
ca
tio
n
im
p
o
r
tan
ce
to
lo
ca
lize
t
u
m
o
r
r
eg
io
n
s
p
r
ec
is
ely
.
-
Fu
s
io
n
lay
er
:
C
AM
an
d
SAM
o
u
tp
u
ts
ar
e
f
u
s
ed
v
ia
elem
en
t
-
wis
e
m
u
lt
ip
licatio
n
an
d
ad
d
e
d
to
th
e
f
ea
tu
r
e
m
ap
,
im
p
r
o
v
in
g
f
o
c
u
s
o
n
tu
m
o
r
r
eg
i
o
n
s
.
Fig
u
r
e
2
.
Featu
r
e
f
u
s
io
n
m
o
d
u
le
with
atten
tio
n
m
ec
h
an
is
m
3.
7
.
Cla
s
s
if
ica
t
io
n hea
d
Af
ter
f
ea
tu
r
e
e
x
tr
ac
tio
n
a
n
d
en
h
an
ce
m
en
t,
t
h
e
class
if
icatio
n
h
ea
d
p
er
f
o
r
m
s
tu
m
o
r
ca
teg
o
r
i
za
tio
n
:
-
Glo
b
al
av
er
ag
e
p
o
o
lin
g
: r
ed
u
c
es
s
p
atial
d
im
en
s
io
n
s
an
d
o
v
e
r
f
itti
n
g
r
is
k
s
.
-
Fu
lly
co
n
n
ec
ted
lay
er
: 2
5
6
-
n
e
u
r
o
n
d
en
s
e
lay
er
with
R
eL
U
a
ctiv
atio
n
.
-
Dr
o
p
o
u
t la
y
er
: Set
at
0
.
3
to
r
e
d
u
ce
co
-
a
d
ap
tatio
n
o
f
n
eu
r
o
n
s
.
-
Ou
tp
u
t
lay
er
:
a
s
o
f
tm
ax
clas
s
if
ier
with
f
o
u
r
n
eu
r
o
n
s
(
g
li
o
m
a,
m
en
i
n
g
io
m
a,
p
itu
itar
y
,
n
o
tu
m
o
r
)
f
o
r
m
u
lti
-
class
o
u
tp
u
t.
T
h
e
m
o
d
el
is
tr
ain
ed
o
n
th
e
B
r
aT
S
2
0
2
0
d
ataset
an
d
ev
al
u
ated
o
n
Kag
g
le
B
r
ain
MRI
d
ataset
(
an
d
v
ice
v
er
s
a)
t
o
m
ea
s
u
r
e
p
er
f
o
r
m
an
ce
.
Dr
o
p
in
p
e
r
f
o
r
m
an
c
e
m
etr
ics
ac
r
o
s
s
d
atasets
is
k
ep
t
with
in
1
.
5
%,
s
h
o
win
g
h
ig
h
d
o
m
ain
ad
a
p
tab
ilit
y
.
Alg
o
r
ith
m
Let
=
{
(
,
)
}
=
1
be the dataset of MRI images
∈
ℝ
×
×
with class labels
∈
{
0
,
1
,
2
,
3
}
be the pre
-
trained base model (EfficientNetB0) with parameters
be the classification head parameters.
,
r
epresent parameters of attention modules (channel and spatial attention).
ℒ
be the total loss function.
a.
Featu
r
e
ex
tr
ac
tio
n
T
h
e
im
ag
e
p
ass
ed
th
r
o
u
g
h
th
e
p
r
e
-
tr
ain
ed
C
NN
to
o
b
tain
h
i
g
h
-
lev
el
f
ea
tu
r
es
=
(
)
,
∈
ℝ
ℎ
×
×
wh
er
e
th
e
f
ea
tu
r
e
m
ap
e
x
tr
ac
t
ed
b
y
E
f
f
icien
tNetB
0
.
b.
Atten
tio
n
-
b
ased
f
ea
tu
r
e
f
u
s
io
n
-
C
h
an
n
el
atten
tio
n
m
o
d
u
le
(
C
AM
)
:
=
(
2
⋅
R
eL
U
(
1
⋅
GAP
(
)
)
)
=
⊙
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
n
h
a
n
ce
d
tr
a
n
s
fer lea
r
n
in
g
fr
a
mewo
r
k
fo
r
b
r
a
in
tu
mo
r
d
etec
tio
n
fr
o
m
MRI
s
ca
n
s
u
s
in
g
…
(
S
mita
B
h
a
r
n
e
)
501
wh
e
r
e:
GAP
Glo
b
al
Av
er
ag
e
Po
o
lin
g
,
1
,
2
: le
ar
n
ed
weig
h
ts
,
: sig
m
o
id
fun
c
tion
,
⊙
: e
lem
en
t
-
wis
e
m
u
ltip
licatio
n
-
Sp
atial
atten
tio
n
m
o
d
u
le
(
SAM)
=
(
C
on
v
7
×
7
(
[
A
vgPool
(
)
;
M
a
xPo
ol
(
)
]
)
)
=
⊙
-
Fin
al
atten
tio
n
-
f
u
s
ed
f
ea
t
u
r
e
m
ap
=
+
T
h
is
f
u
s
io
n
r
etain
s
th
e
o
r
ig
in
al
f
ea
tu
r
es wh
ile
en
h
a
n
cin
g
r
elev
an
t o
n
es v
ia
atten
tio
n
.
c.
C
las
s
if
icatio
n
h
ea
d
=
GAP
(
)
ℎ
=
R
eL
U
(
1
+
1
)
ℎ
′
=
Dr
o
p
o
u
t
(
ℎ
,
=
0
.
3
)
̂
=
So
f
tm
ax
(
2
ℎ
′
+
2
)
wh
er
e:
1
,
2
an
d
1
,
2
ar
e
tr
ain
ab
le
weig
h
ts
an
d
b
iases
,
̂
∈
ℝ
4
is
th
e
p
r
ed
icted
class
p
r
o
b
ab
ilit
y
v
ec
t
o
r
.
d.
L
o
s
s
f
u
n
ctio
n
W
e
u
s
e
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
f
o
r
m
u
lti
-
class
class
if
icati
o
n
:
ℒ
=
−
∑
∗
=
1
∑
_
(
̂
_
)
4
=
1
wh
er
e:
is
th
e
tr
u
e
lab
el
(
o
n
e
-
h
o
t)
f
o
r
class
̂
is
th
e
p
r
ed
icted
p
r
o
b
a
b
ilit
y
f
o
r
class
e.
Op
tim
izatio
n
T
h
e
to
tal
o
b
jectiv
e
is
to
m
i
n
im
ize:
ℒ
(
,
,
,
)
=
ℒ
+
⋅
ℛ
(
,
)
wh
er
e:
ℛ
is
an
ℓ
2
r
eg
u
lar
izatio
n
ter
m
.
co
n
tr
o
ls
th
e
r
e
g
u
lar
izatio
n
s
tr
en
g
th
.
Op
tim
izatio
n
is
co
n
d
u
cted
u
s
in
g
th
e
ad
am
o
p
tim
izer
with
lear
n
in
g
r
ate
ad
ju
s
ted
th
r
o
u
g
h
co
s
in
e
an
n
ea
lin
g
:
=
+
1
2
(
−
)
(
1
+
c
os
(
)
)
f.
Do
m
ain
-
Ad
ap
tiv
e
Fin
e
-
T
u
n
in
g
Stra
teg
y
L
et
=
{
,
}
wh
er
e:
s
h
allo
w
lay
er
s
(
f
r
o
ze
n
o
r
m
in
im
ally
u
p
d
ated
)
,
d
ee
p
er
lay
e
r
s
(
f
in
e
-
tu
n
ed
m
o
r
e
ag
g
r
ess
iv
ely
)
.
T
h
en
:
≈
0
;
∝
T
h
is
h
ier
ar
ch
ical
f
in
e
-
tu
n
i
n
g
s
tab
ilizes
tr
ain
in
g
wh
ile
ad
ap
tin
g
th
e
d
ee
p
er
lay
er
s
to
th
e
d
o
m
ain
-
s
p
ec
if
ic
f
ea
tu
r
es in
MRI
im
ag
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
2
,
J
u
n
e
20
26
:
4
9
7
-
5
0
7
502
4.
RE
SU
L
T
AND
DI
SCUS
SI
O
N
Pro
p
o
s
ed
E
T
L
F
o
u
tp
er
f
o
r
m
s
a
ll
b
aselin
e
m
o
d
els
ac
r
o
s
s
ev
er
y
ev
alu
atio
n
m
etr
ic
with
an
ac
cu
r
ac
y
o
f
9
8
.
7
6
%,
it
s
h
o
ws
a
+6
.
5
%
i
m
p
r
o
v
e
m
en
t
o
v
er
VGG1
6
a
n
d
~2
%
g
ain
o
v
er
E
f
f
icien
tN
etB
0
alo
n
e,
d
u
e
to
en
h
an
ce
d
f
in
e
-
t
u
n
in
g
,
atten
ti
o
n
f
u
s
io
n
,
a
n
d
lea
r
n
in
g
r
ate
ad
ap
ta
tio
n
.
Fig
u
r
e
3
s
h
o
ws
th
e
p
e
r
f
o
r
m
an
ce
co
m
p
ar
is
o
n
o
f
b
r
ai
n
tu
m
o
r
d
etec
tio
n
m
o
d
el.
T
h
ese
r
esu
lts
co
llectiv
ely
s
h
o
ws
th
e
s
u
p
er
i
o
r
ity
a
n
d
g
en
er
aliza
b
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el.
Per
f
o
r
m
an
ce
o
f
cr
o
s
s
v
alid
atio
n
o
n
tr
ain
in
g
o
n
B
r
aT
S
an
d
test
in
g
o
n
Kag
g
le
(
a
n
d
v
ic
e
v
er
s
a)
r
esu
l
ted
in
o
n
ly
~1
.
5
%
d
r
o
p
,
s
h
o
w
ca
s
in
g
th
e
m
o
d
el’
s
d
o
m
ai
n
g
e
n
er
aliza
b
ilit
y
.
T
h
e
r
esu
lts
ar
e
co
m
p
ar
ed
to
e
x
is
tin
g
m
o
d
els
lik
e
VGG1
6
(
B
aselin
e)
,
R
esNet5
0
,
I
n
ce
p
tio
n
V3
,
E
f
f
icien
tNetB
0
an
d
th
e
p
r
o
p
o
s
ed
E
T
L
F
ac
h
ie
v
ed
s
u
p
er
io
r
r
esu
lts
with
f
ewe
r
p
a
r
am
e
ter
s
an
d
f
aster
in
f
er
en
ce
.
T
ab
le
1
s
h
o
ws
th
e
co
m
p
ar
ativ
e
an
aly
s
is
o
f
d
if
f
er
en
t
m
o
d
els
with
b
aselin
e
m
o
d
els.
I
t
in
d
icate
s
th
e
p
o
s
s
ib
le
f
o
r
in
teg
r
atio
n
in
to
r
ea
l
-
tim
e
clin
ical
s
y
s
tem
s
.
Fig
u
r
e
3
.
Per
f
o
r
m
an
c
e
co
m
p
ar
is
o
n
o
f
b
r
ain
tu
m
o
r
d
etec
tio
n
m
o
d
els
T
ab
le
1
.
C
o
m
p
a
r
is
o
n
o
f
r
esu
lts
b
etwe
en
p
r
o
p
o
s
ed
E
T
L
F a
n
d
b
aselin
e
m
o
d
els
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
S
c
o
r
e
(
%)
AUC
V
G
G
1
6
(
ba
s
e
l
i
n
e
)
9
1
.
2
3
9
0
.
8
1
8
9
.
6
7
9
0
.
2
0
0
.
9
4
R
e
sN
e
t
5
0
9
4
.
3
7
9
3
.
7
0
9
2
.
8
9
9
3
.
2
9
0
.
9
6
I
n
c
e
p
t
i
o
n
V
3
9
5
.
8
0
9
5
.
2
1
9
4
.
7
0
9
4
.
9
5
0
.
9
7
Ef
f
i
c
i
e
n
t
N
e
t
B
0
(
B
e
st
B
a
se
l
i
n
e
)
9
6
.
9
2
9
6
.
3
0
9
5
.
7
0
9
6
.
0
0
0
.
9
8
P
r
o
p
o
se
d
E
TLF
9
8
.
7
6
9
8
.
5
4
9
8
.
3
4
9
8
.
4
3
0
.
9
9
Fig
u
r
e
4
s
h
o
ws
th
e
p
r
o
p
o
s
ed
m
o
d
el’
s
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
E
T
L
F
m
o
d
el
ac
r
o
s
s
th
e
ep
o
ch
s
.
Fro
m
F
ig
u
r
e
5
we
ca
n
an
aly
ze
th
e
ac
cu
r
ac
y
cu
r
v
e
s
h
o
ws
a
r
is
in
g
tr
en
d
an
d
th
e
lo
s
s
cu
r
v
e
is
d
ec
r
ea
s
in
g
co
n
s
is
ten
tly
,
s
h
o
win
g
th
e
e
f
f
ec
tiv
e
lear
n
in
g
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el.
Fig
u
r
e
5
(
a
)
s
h
o
ws
th
e
tr
ain
in
g
an
d
v
alid
atio
n
test
ac
cu
r
ac
y
.
T
h
e
ac
cu
r
ac
y
cu
r
v
es
s
h
o
w
a
co
n
s
is
ten
t
in
cr
ea
s
e
in
b
o
th
tr
ai
n
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
,
p
latea
u
i
n
g
af
ter
a
p
p
r
o
x
im
ately
1
5
ep
o
ch
s
.
T
h
e
f
in
al
v
alid
atio
n
ac
cu
r
ac
y
clo
s
ely
f
o
llo
ws
th
e
tr
a
in
in
g
ac
cu
r
ac
y
,
r
ea
ch
in
g
a
b
o
v
e
9
8
%,
wh
ich
s
ig
n
if
ies
h
ig
h
class
if
icatio
n
p
er
f
o
r
m
an
c
e
ac
r
o
s
s
all
tu
m
o
r
class
es.
T
h
e
p
ar
allel
b
eh
av
io
u
r
o
f
th
e
two
c
u
r
v
es
co
n
f
ir
m
s
th
at
th
e
m
o
d
el
m
ain
tain
s
a
g
o
o
d
b
ias
-
v
ar
ian
c
e
tr
ad
e
-
o
f
f
.
I
n
Fig
u
r
e
5
(
b
)
,
th
e
tr
ain
in
g
an
d
v
alid
a
tio
n
lo
s
s
cu
r
v
es
d
em
o
n
s
tr
ate
a
s
m
o
o
t
h
an
d
co
n
s
is
ten
t
d
o
wn
war
d
tr
en
d
,
s
h
o
win
g
t
h
e
ef
f
ec
tiv
e
lear
n
in
g
an
d
c
o
n
v
er
g
e
n
ce
o
f
th
e
m
o
d
el.
B
o
th
lo
s
s
es
d
ec
r
ea
s
e
p
r
o
g
r
ess
iv
ely
ac
r
o
s
s
ep
o
ch
s
,
with
m
in
im
al
g
ap
b
etwe
en
th
em
,
s
u
g
g
esti
n
g
th
at
th
e
m
o
d
el
g
en
er
alize
s
well
to
u
n
s
ee
n
d
ata
a
n
d
d
o
es
n
o
t
o
v
er
f
it.
Fro
m
Fig
u
r
e
6
,
t
h
e
a
r
ea
u
n
d
er
c
u
r
v
e
(
AUC
)
s
co
r
e
o
f
0
.
9
9
co
n
f
ir
m
s
ex
ce
llen
t
class
s
ep
ar
ab
ilit
y
,
w
h
ile
th
e
h
i
g
h
p
r
ec
is
io
n
a
n
d
r
e
ca
ll
s
co
r
es
v
alid
ate
its
r
o
b
u
s
tn
ess
in
b
o
th
tu
m
o
r
p
r
esen
ce
d
etec
tio
n
a
n
d
co
r
r
ec
t
class
if
icatio
n
.
An
a
b
latio
n
s
tu
d
y
was
co
n
d
u
cted
to
ass
ess
th
e
co
n
tr
ib
u
tio
n
o
f
ea
ch
co
m
p
o
n
en
t,
with
o
u
t
f
in
e
-
tu
n
in
g
;
ac
cu
r
ac
y
d
r
o
p
p
ed
to
9
4
.
0
2
%,
with
o
u
t
atten
tio
n
f
u
s
io
n
;
ac
cu
r
ac
y
d
r
o
p
p
ed
to
9
5
.
1
8
%
an
d
with
o
u
t
lear
n
in
g
r
ate
(
L
R
)
s
ch
ed
u
le
r
;
ac
cu
r
ac
y
d
r
o
p
p
e
d
to
9
6
.
7
2
%.
Fig
u
r
e
7
clea
r
ly
illu
s
tr
ates
h
o
w
ea
ch
co
m
p
o
n
en
t
—
f
in
e
-
tu
n
i
n
g
,
atten
tio
n
f
u
s
io
n
,
an
d
lear
n
i
n
g
r
ate
s
c
h
ed
u
ler
a
f
f
ec
ts
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el.
T
h
e
f
u
ll E
T
L
F m
o
d
el
lead
s
with
th
e
h
ig
h
est ac
cu
r
ac
y
at
9
8
.
7
6
%.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
n
h
a
n
ce
d
tr
a
n
s
fer lea
r
n
in
g
fr
a
mewo
r
k
fo
r
b
r
a
in
tu
mo
r
d
etec
tio
n
fr
o
m
MRI
s
ca
n
s
u
s
in
g
…
(
S
mita
B
h
a
r
n
e
)
503
Fig
u
r
e
4
.
Ov
e
r
all
tr
ain
in
g
p
er
f
o
r
m
an
ce
E
T
L
F m
o
d
el
(
a)
(
b
)
Fig
u
r
e
5
.
T
r
ain
in
g
a
n
d
v
alid
atio
n
(
a)
ac
cu
r
ac
y
an
d
(
b
)
l
o
s
s
Fig
u
r
e
6
.
AUC
v
alid
atio
n
v
s
test
d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
2
,
J
u
n
e
20
26
:
4
9
7
-
5
0
7
504
Fig
u
r
e
7
.
I
m
p
ac
t o
f
k
ey
co
m
p
o
n
en
ts
o
n
E
T
L
F a
cc
u
r
ac
y
5.
ST
A
T
I
S
T
I
CAL
SI
G
N
I
F
I
CA
NCE O
F
T
H
E
ST
U
DY
T
ab
le
2
s
h
o
ws
th
e
s
tatis
t
ic
al
c
o
m
p
ar
is
o
n
b
etwe
en
p
r
o
p
o
s
ed
E
T
L
F
an
d
E
f
f
icien
tN
etB
0
(
B
est
B
aselin
e
Mo
d
el)
.
A
Statis
tical
s
ig
n
if
ican
ce
an
aly
s
is
was
co
n
d
u
cted
to
v
alid
ate
th
e
r
o
b
u
s
t
n
ess
an
d
r
eliab
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
E
T
L
F.
T
h
e
s
tatis
tical
ev
alu
atio
n
f
o
llo
ws
with
th
e
m
o
d
er
n
ML
b
ased
co
m
p
ar
is
o
n
in
m
e
d
ical
im
ag
in
g
[
2
4
]
.
T
h
e
p
r
im
ar
y
g
o
al
is
to
d
eter
m
in
e
if
th
e
o
b
s
er
v
ed
i
m
p
r
o
v
em
en
ts
d
u
e
t
o
E
T
L
F
o
v
er
b
aselin
e
m
o
d
els we
r
e
ca
u
s
ed
b
y
th
e
m
e
th
o
d
o
lo
g
y
b
ein
g
s
u
p
e
r
io
r
,
n
o
t
r
an
d
o
m
v
ar
iatio
n
s
in
t
h
e
tr
ain
i
n
g
d
y
n
am
ics.
T
o
en
s
u
r
e
th
e
s
tab
ilit
y
o
f
th
e
m
o
d
el
tr
ain
in
g
/tes
t
d
atasets
,
a
5
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
ex
p
er
i
m
en
t
was
p
er
f
o
r
m
ed
.
T
h
e
E
T
L
F
s
h
o
wed
th
e
m
ea
n
ac
cu
r
ac
y
o
f
9
8
.
7
6
%
,
s
tan
d
ar
d
d
ev
iatio
n
(
σ
)
o
f
0
.
2
7
%
an
d
co
ef
f
icien
t
o
f
v
ar
iatio
n
is
0
.
0
0
2
7
.
T
h
e
s
m
all
d
ev
iatio
n
ac
r
o
s
s
th
e
f
o
ld
s
in
d
icate
s
th
at
th
e
E
T
L
F
p
r
o
d
u
ce
s
r
eliab
le
p
r
ed
ictio
n
s
an
d
h
as
th
e
p
o
t
en
tial
f
o
r
g
o
o
d
g
en
er
alis
atio
n
ca
p
ab
ilit
ies.
T
o
m
ea
s
u
r
e
im
p
r
o
v
e
m
en
ts
in
p
er
f
o
r
m
an
ce
with
th
e
b
est
b
aselin
e
m
o
d
el
i.e
.
E
f
f
icien
tNetB
0
,
a
p
air
ed
two
-
tailed
t
-
test
was
co
n
d
u
cted
with
th
e
1
0
in
d
ep
en
d
e
n
t
r
u
n
s
f
o
r
ea
ch
m
o
d
el.
T
h
e
E
f
f
icien
tNe
tB
0
m
ea
n
ac
cu
r
ac
y
is
9
6
.
9
2
%.
W
e
g
o
t
p
-
v
alu
e
(
ac
cu
r
ac
y
)
as
0
.
0
0
4
1
an
d
p
-
v
a
lu
e
(
AUC)
as
0
.
0
0
9
4
.
Sin
ce
b
o
th
th
e
p
-
v
alu
es
a
r
e
less
th
an
0
.
0
5
,
th
e
d
if
f
er
en
ce
h
ig
h
ly
s
ig
n
if
ica
n
t r
esu
lts
o
f
p
r
o
p
o
s
ed
E
T
L
F’s m
o
d
el
im
p
r
o
v
em
en
ts
is
n
o
t r
an
d
o
m
.
Fig
u
r
e
8
s
h
o
ws
th
e
v
is
u
al
r
ep
r
esen
tatio
n
o
f
th
e
9
5
%
co
n
f
id
en
ce
in
ter
v
al
f
o
r
k
ey
m
etr
ics
o
f
E
T
L
F.
Usi
n
g
cr
o
s
s
v
alid
atio
n
s
co
r
es,
th
e
m
o
d
el
h
ad
a
9
5
%
co
n
f
i
d
e
n
ce
in
ter
v
al
(
C
I
)
f
o
r
its
class
if
icatio
n
ac
cu
r
ac
y
o
f
E
T
L
F
ac
cu
r
ac
y
C
I
(
9
5
%)
s
h
o
ws 9
8
.
7
6
% with
an
im
p
r
o
v
em
en
ts
o
f
±
0
.
2
9
%.
T
h
e
n
ar
r
o
w
C
I
in
d
icate
s
s
tab
ilit
y
o
f
th
e
m
o
d
el
an
d
lo
w
v
ar
ian
c
e,
f
u
r
th
er
s
u
p
p
o
r
tin
g
th
e
r
eliab
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
.
T
o
m
ea
s
u
r
e
th
e
ef
f
ec
t
s
ize
m
ea
s
u
r
em
en
t
with
C
o
h
en
’
s
d
ef
f
ec
t
s
ize
was
c
alcu
late
b
etwe
en
th
e
E
T
L
F
an
d
E
f
f
icien
tNetB
0
s
h
o
ws
th
e
C
o
h
en
’
s
d
(
Acc
u
r
ac
y
)
is
1
.
8
5
C
o
h
e
n
’
s
d
AUC
is
1
.
6
7
r
esp
ec
t
iv
ely
.
Acc
o
r
d
in
g
to
co
n
v
en
tio
n
al
th
r
esh
o
ld
s
,
a
C
o
h
en
’
s
d
>1
.
6
r
ep
r
esen
ts
a
lar
g
e
ef
f
ec
t
s
ize
s
h
o
ws
E
T
L
F
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
n
o
t
ju
s
t
s
tatis
t
ically
d
etec
tab
le
with
th
e
b
aselin
e
m
o
d
el.
T
o
ch
ec
k
th
e
er
r
o
r
p
atter
n
a
s
tu
d
y
o
f
th
e
d
is
tr
ib
u
tio
n
o
f
m
is
clas
s
i
f
icatio
n
er
r
o
r
s
h
as
d
em
o
n
s
tr
ated
th
at
th
e
m
ajo
r
ity
o
f
m
is
class
i
f
icatio
n
s
in
b
aselin
e
m
o
d
els
o
cc
u
r
w
h
en
class
if
y
in
g
b
etwe
en
th
e
g
lio
m
a
an
d
m
en
in
g
io
m
a
class
es.
T
h
is
is
p
r
im
ar
ily
d
u
e
to
th
e
f
ac
t
th
a
t
th
ese
two
class
es
h
av
e
s
ev
er
al
m
o
r
p
h
o
lo
g
ical
ch
ar
ac
ter
is
tic
s
th
at
o
v
er
lap
wi
th
o
n
e
an
o
th
er
.
T
h
er
e
f
o
r
e,
b
y
ap
p
ly
in
g
an
atten
tio
n
f
u
s
io
n
m
o
d
u
le
to
im
p
r
o
v
e
th
e
lear
n
in
g
o
f
d
is
cr
im
in
ativ
e
s
p
atial
f
ea
tu
r
es,
we
r
ed
u
ce
d
th
e
n
u
m
b
e
r
o
f
f
alse
p
o
s
itiv
e
r
esu
lts
an
d
in
cr
ea
s
ed
th
e
p
er
-
class
r
eliab
ilit
y
o
f
r
esu
lts
.
T
ab
le
2
.
St
atis
tical
co
m
p
ar
is
o
n
b
etwe
en
th
e
p
r
o
p
o
s
ed
E
T
L
F
an
d
E
f
f
icien
tNetB
0
(
b
est b
aselin
e
m
o
d
el)
M
e
t
r
i
c
Ef
f
i
c
i
e
n
t
N
e
t
B
0
P
r
o
p
o
se
d
ETLF
A
b
so
l
u
t
e
i
mp
r
o
v
e
me
n
t
9
5
%
C
I
(
ETLF)
p
-
v
a
l
u
e
(
P
a
i
r
e
d
t
-
t
e
st
)
Ef
f
e
c
t
si
z
e
(
C
o
h
e
n
’
s
d
)
A
c
c
u
r
a
c
y
(
%)
9
6
.
9
2
9
8
.
7
6
+
1
.
8
4
%
±
0
.
2
9
%
0
.
0
0
4
1
1
.
8
5
P
r
e
c
i
s
i
o
n
(
%)
9
6
.
3
0
9
8
.
5
4
+
2
.
2
4
%
±
0
.
3
1
%
0
.
0
0
6
3
1
.
7
2
R
e
c
a
l
l
(
%)
9
5
.
7
0
9
8
.
3
4
+
2
.
6
4
%
±
0
.
2
7
%
0
.
0
0
5
7
1
.
9
1
F1
-
S
c
o
r
e
(
%)
9
6
.
0
0
9
8
.
4
3
+
2
.
4
3
%
±
0
.
2
4
%
0
.
0
0
5
1
1
.
8
8
AUC
0
.
9
8
0
.
9
9
+
0
.
0
1
±
0
.
0
0
4
0
.
0
0
9
4
1
.
6
7
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
n
h
a
n
ce
d
tr
a
n
s
fer lea
r
n
in
g
fr
a
mewo
r
k
fo
r
b
r
a
in
tu
mo
r
d
etec
tio
n
fr
o
m
MRI
s
ca
n
s
u
s
in
g
…
(
S
mita
B
h
a
r
n
e
)
505
Fig
u
r
e
8
.
Statis
tical
s
ig
n
if
ican
ce
o
f
co
n
f
id
en
ce
in
ter
v
al
o
f
E
T
L
F
p
er
f
o
r
m
an
ce
6.
CO
N
CL
U
SI
O
N
I
n
th
is
s
tu
d
y
we
p
r
esen
t
an
in
n
o
v
ativ
e
f
r
am
ewo
r
k
f
o
r
b
r
ain
tu
m
o
r
d
etec
tio
n
with
an
en
h
an
ce
d
tr
an
s
f
er
lear
n
in
g
m
o
d
el.
B
y
i
n
teg
r
atin
g
d
o
m
ain
-
ad
ap
tiv
e
f
i
n
e
-
tu
n
in
g
,
atten
tio
n
-
b
ase
d
f
ea
tu
r
e
f
u
s
io
n
,
an
d
an
ad
ap
tiv
e
tr
ain
in
g
s
tr
ateg
y
,
th
e
p
r
o
p
o
s
ed
m
o
d
el
attain
s
ex
ce
llen
t
p
er
f
o
r
m
a
n
ce
o
n
m
u
ltip
le
d
atasets
.
T
h
is
wo
r
k
b
r
id
g
es
th
e
g
a
p
b
etwe
en
g
e
n
er
al
-
p
u
r
p
o
s
e
C
NNs
an
d
d
o
m
ain
-
s
p
ec
if
ic
m
ed
ical
im
ag
in
g
task
s
b
y
o
f
f
er
in
g
a
p
r
ac
tical,
g
en
er
aliza
b
le,
a
n
d
ef
f
icien
t
d
iag
n
o
s
tic
s
o
lu
tio
n
.
T
h
e
s
tatis
tical
s
ig
n
if
ican
ce
o
f
th
e
s
tu
d
y
with
all
cr
o
s
s
v
alid
atio
n
,
t
-
test
s
,
co
n
f
id
en
ce
in
ter
v
als
an
d
ef
f
ec
t
s
ize
an
aly
s
is
v
alid
ate
th
e
p
er
f
o
r
m
an
ce
g
ain
s
s
h
o
wed
b
y
th
e
p
r
o
p
o
s
ed
E
T
L
F
m
o
d
el
is
s
tatis
tically
s
u
b
s
tan
tia
l.
T
h
e
r
esu
lts
s
h
o
w
im
p
r
o
v
em
en
t
s
o
v
er
th
e
ex
is
tin
g
d
ee
p
lear
n
in
g
tech
n
iq
u
es
f
o
r
b
r
ain
tu
m
o
r
d
etec
tio
n
.
S
tu
d
y
f
u
r
th
er
co
n
f
ir
m
ed
th
e
c
r
itical
co
n
tr
ib
u
tio
n
s
o
f
ea
c
h
m
o
d
u
le,
in
cl
u
d
in
g
atten
tio
n
f
u
s
io
n
an
d
f
in
e
-
tu
n
i
n
g
,
in
b
o
o
s
tin
g
th
e
o
v
e
r
all
ef
f
icien
cy
o
f
th
e
m
o
d
el.
T
h
e
n
o
v
elty
o
f
th
is
s
tu
d
y
lies
in
h
y
b
r
id
f
in
e
-
tu
n
in
g
s
tr
ateg
y
tail
o
r
ed
f
o
r
MRI
-
b
ased
task
s
an
d
a
p
o
wer
f
u
l
f
ea
t
u
r
e
f
u
s
io
n
m
ec
h
an
is
m
.
Fu
tu
r
e
wo
r
k
in
clu
d
es
ex
ten
d
in
g
th
e
f
r
a
m
ewo
r
k
to
m
u
lti
-
m
o
d
al
f
u
s
io
n
(
co
m
b
in
in
g
MRI
m
o
d
alities
)
,
3
D
v
o
l
u
m
etr
ic
an
aly
s
i
s
,
an
d
ex
p
lain
ab
le
AI
co
m
p
o
n
e
n
ts
f
o
r
b
etter
clin
ical
in
ter
p
r
etab
ilit
y
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT
)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Sm
ita
B
h
ar
n
e
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
E
k
ta
Sar
d
a
✓
✓
✓
✓
✓
✓
Sh
am
al
Salu
n
k
h
e
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
2
,
J
u
n
e
20
26
:
4
9
7
-
5
0
7
506
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
I
NF
O
RM
E
D
CO
NS
E
N
T
W
e
h
av
e
o
b
tain
ed
in
f
o
r
m
ed
c
o
n
s
en
t f
r
o
m
all
in
d
iv
id
u
als in
c
lu
d
ed
in
t
h
is
s
tu
d
y
.
E
T
H
I
CAL
AP
P
RO
V
AL
T
h
is
s
tu
d
y
d
o
es n
o
t in
v
o
l
v
e
h
u
m
an
p
a
r
ticip
an
ts
o
r
an
im
als
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
is
av
ailab
le
at
h
t
tp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/p
k
d
ar
ab
i/m
ed
ical
-
im
a
g
e
-
d
at
aset
-
b
r
ain
-
tu
m
o
r
d
etec
tio
n
.
RE
F
E
R
E
NC
E
S
[
1
]
E.
-
S
.
A
.
E
l
-
D
a
h
sh
a
n
,
T.
H
o
s
n
y
,
a
n
d
A
.
-
B
.
M
.
S
a
l
e
m,
“
H
y
b
r
i
d
i
n
t
e
l
l
i
g
e
n
t
t
e
c
h
n
i
q
u
e
s
f
o
r
M
R
I
b
r
a
i
n
i
ma
g
e
s
c
l
a
s
si
f
i
c
a
t
i
o
n
,
”
D
i
g
i
t
a
l
S
i
g
n
a
l
Pro
c
e
ss
i
n
g
,
v
o
l
.
2
0
,
n
o
.
2
,
p
p
.
4
3
3
–
4
4
1
,
M
a
r
.
2
0
1
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
d
s
p
.
2
0
0
9
.
0
7
.
0
0
2
.
[
2
]
B
.
B
a
b
u
V
i
m
a
l
a
,
S
.
S
r
i
n
i
v
a
s
a
n
,
S
.
K
.
M
a
t
h
i
v
a
n
a
n
,
M
a
h
a
l
a
k
s
h
m
i
,
P
.
J
a
y
a
g
o
p
a
l
,
a
n
d
G
.
T
.
D
a
l
u
,
“
D
e
t
e
c
t
i
o
n
a
n
d
c
l
a
s
s
i
f
i
c
a
t
i
o
n
o
f
b
r
a
i
n
t
u
m
o
r
u
s
i
n
g
h
y
b
r
i
d
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s
,
”
S
c
i
e
n
t
i
f
i
c
R
e
p
o
r
t
s
,
v
o
l
.
1
3
,
n
o
.
1
,
p
.
2
3
0
2
9
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
0
2
3
-
5
0
5
0
5
-
6.
[
3
]
K
.
K
a
v
i
n
K
u
mar
e
t
a
l
.
,
“
B
r
a
i
n
t
u
mo
r
i
d
e
n
t
i
f
i
c
a
t
i
o
n
u
si
n
g
d
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
a
n
d
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
C
o
m
p
u
t
e
r
S
y
st
e
m
s
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
4
6
,
n
o
.
2
,
p
p
.
1
8
4
5
–
1
8
6
1
,
2
0
2
3
,
d
o
i
:
1
0
.
3
2
6
0
4
/
c
sse
.
2
0
2
3
.
0
3
3
9
2
7
.
[
4
]
K
.
K
a
m
n
i
t
sas
e
t
a
l
.
,
“
Ef
f
i
c
i
e
n
t
m
u
l
t
i
-
s
c
a
l
e
3
D
C
N
N
w
i
t
h
f
u
l
l
y
c
o
n
n
e
c
t
e
d
C
R
F
f
o
r
a
c
c
u
r
a
t
e
b
r
a
i
n
l
e
s
i
o
n
se
g
me
n
t
a
t
i
o
n
,
”
Me
d
i
c
a
l
I
m
a
g
e
A
n
a
l
y
s
i
s
,
v
o
l
.
3
6
,
p
p
.
6
1
–
7
8
,
F
e
b
.
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
m
e
d
i
a
.
2
0
1
6
.
1
0
.
0
0
4
.
[
5
]
M
.
D
.
Ze
i
l
e
r
a
n
d
R
.
F
e
r
g
u
s
,
“
V
i
su
a
l
i
z
i
n
g
a
n
d
u
n
d
e
r
st
a
n
d
i
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s,”
i
n
L
e
c
t
u
r
e
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
(
i
n
c
l
u
d
i
n
g
su
b
ser
i
e
s
L
e
c
t
u
re
N
o
t
e
s
i
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
L
e
c
t
u
re
N
o
t
e
s
i
n
B
i
o
i
n
f
o
rm
a
t
i
c
s)
,
v
o
l
.
8
6
8
9
LN
C
S
,
n
o
.
P
A
R
T
1
,
D
.
F
l
e
e
t
,
T.
P
a
j
d
l
a
,
B
.
S
c
h
i
e
l
e
,
a
n
d
T.
Tu
y
t
e
l
a
a
r
s,
Ed
s
.
,
2
0
1
4
,
p
p
.
8
1
8
–
8
3
3
.
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
319
-
1
0
5
9
0
-
1
_
5
3
.
[
6
]
M
.
Ta
n
a
n
d
Q
.
V
.
L
e
,
“
Ef
f
i
c
i
e
n
t
N
e
t
:
r
e
t
h
i
n
k
i
n
g
m
o
d
e
l
s
c
a
l
i
n
g
f
o
r
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
s
,
”
a
rX
i
v
p
r
e
p
r
i
n
t
a
rX
i
v
1
9
0
5
.
1
1
9
4
6
,
2
0
1
9
.
[
7
]
B
.
B
.
G
u
p
t
a
,
A
.
G
a
u
r
a
v
,
a
n
d
V
.
A
r
y
a
,
“
D
e
e
p
C
N
N
b
a
s
e
d
b
r
a
i
n
t
u
mo
r
d
e
t
e
c
t
i
o
n
i
n
i
n
t
e
l
l
i
g
e
n
t
s
y
s
t
e
ms
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
N
e
t
w
o
rks
,
v
o
l
.
5
,
p
p
.
3
0
–
3
7
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
j
i
n
.
2
0
2
3
.
1
2
.
0
0
1
.
[
8
]
M
.
Z.
K
h
a
l
i
k
i
a
n
d
M
.
S
.
B
a
şars
l
a
n
,
“
B
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
f
r
o
m
i
ma
g
e
s
a
n
d
c
o
m
p
a
r
i
so
n
w
i
t
h
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
me
t
h
o
d
s
a
n
d
3
-
l
a
y
e
r
C
N
N
,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
rt
s
,
v
o
l
.
1
4
,
n
o
.
1
,
p
.
2
6
6
4
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
5
9
8
-
024
-
5
2
8
2
3
-
9.
[
9
]
S
.
K
.
M
a
t
h
i
v
a
n
a
n
,
S
.
S
o
n
a
i
m
u
t
h
u
,
S
.
M
u
r
u
g
e
sa
n
,
H
.
R
a
j
a
d
u
r
a
i
,
B
.
D
.
S
h
i
v
a
h
a
r
e
,
a
n
d
M
.
A
.
S
h
a
h
,
“
Emp
l
o
y
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
f
o
r
a
c
c
u
r
a
t
e
b
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
rt
s
,
v
o
l
.
1
4
,
n
o
.
1
,
p
.
7
2
3
2
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
024
-
5
7
9
7
0
-
7.
[
1
0
]
R
.
P
r
e
e
t
h
a
,
M
.
J.
P
.
P
r
i
y
a
d
a
r
s
i
n
i
,
a
n
d
J.
S
.
N
i
sh
a
,
“
A
u
t
o
ma
t
e
d
b
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
f
r
o
m
ma
g
n
e
t
i
c
r
e
s
o
n
a
n
c
e
i
mag
e
s
u
s
i
n
g
f
i
n
e
-
t
u
n
e
d
Ef
f
i
c
i
e
n
t
N
e
t
-
B
4
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
,
”
I
EE
E
Ac
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
1
1
2
1
8
1
–
1
1
2
1
9
5
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
4
.
3
4
4
2
9
7
9
.
[
1
1
]
D
.
R
a
st
o
g
i
e
t
a
l
.
,
“
B
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
a
n
d
p
r
e
d
i
c
t
i
o
n
i
n
M
R
I
i
ma
g
e
s
u
t
i
l
i
z
i
n
g
a
f
i
n
e
-
t
u
n
e
d
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
mo
d
e
l
i
n
t
e
g
r
a
t
e
d
w
i
t
h
i
n
d
e
e
p
l
e
a
r
n
i
n
g
f
r
a
m
e
w
o
r
k
s,
”
L
i
f
e
,
v
o
l
.
1
5
,
n
o
.
3
,
p
.
3
2
7
,
F
e
b
.
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
l
i
f
e
1
5
0
3
0
3
2
7
.
[
1
2
]
R
.
D
i
sc
i
,
F
.
G
u
r
c
a
n
,
a
n
d
A
.
S
o
y
l
u
,
“
A
d
v
a
n
c
e
d
b
r
a
i
n
t
u
mo
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
i
n
M
R
i
ma
g
e
s
u
si
n
g
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
a
n
d
p
r
e
-
t
r
a
i
n
e
d
d
e
e
p
C
N
N
m
o
d
e
l
s
,
”
C
a
n
c
e
rs
,
v
o
l
.
1
7
,
n
o
.
1
,
p
.
1
2
1
,
Ja
n
.
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
c
a
n
c
e
r
s1
7
0
1
0
1
2
1
.
[
1
3
]
Y
.
P
a
n
d
e
a
n
d
J
.
C
h
a
k
i
,
“
B
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
a
c
r
o
ss
d
i
v
e
r
s
e
M
R
i
m
a
g
e
s:
A
n
a
u
t
o
m
a
t
e
d
t
r
i
p
l
e
-
m
o
d
u
l
e
a
p
p
r
o
a
c
h
i
n
t
e
g
r
a
t
i
n
g
r
e
d
u
c
e
d
f
u
se
d
d
e
e
p
f
e
a
t
u
r
e
s
a
n
d
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
,
”
R
e
s
u
l
t
s
i
n
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
5
,
p
.
1
0
3
8
3
2
,
M
a
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
r
i
n
e
n
g
.
2
0
2
4
.
1
0
3
8
3
2
.
[
1
4
]
M
.
N
.
I
sl
a
m
,
M
.
S
.
A
z
a
m,
M
.
S
.
I
sl
a
m,
M
.
H
.
K
a
n
c
h
a
n
,
A
.
H
.
M
.
S
.
P
a
r
v
e
z
,
a
n
d
M
.
M
.
I
sl
a
m,
“
A
n
i
m
p
r
o
v
e
d
d
e
e
p
l
e
a
r
n
i
n
g
-
b
a
se
d
h
y
b
r
i
d
m
o
d
e
l
w
i
t
h
e
n
sem
b
l
e
t
e
c
h
n
i
q
u
e
s
f
o
r
b
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
f
r
o
m
M
R
I
i
mag
e
,
”
I
n
f
o
rm
a
t
i
c
s
i
n
M
e
d
i
c
i
n
e
U
n
l
o
c
k
e
d
,
v
o
l
.
4
7
,
p
.
1
0
1
4
8
3
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
m
u
.
2
0
2
4
.
1
0
1
4
8
3
.
[
1
5
]
S
.
A
l
i
e
t
a
l
.
,
“
A
n
i
m
p
r
o
v
e
d
3
D
U
-
N
e
t
-
b
a
se
d
d
e
e
p
l
e
a
r
n
i
n
g
s
y
st
e
m
f
o
r
b
r
a
i
n
t
u
m
o
r
s
e
g
me
n
t
a
t
i
o
n
u
s
i
n
g
m
u
l
t
i
-
m
o
d
a
l
M
R
I
,
”
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
.
8
3
,
n
o
.
3
7
,
p
p
.
8
5
0
2
7
–
8
5
0
4
6
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
0
2
4
-
1
9
4
0
6
-
2.
[
1
6
]
S
.
A
n
a
n
t
h
a
r
a
j
a
n
,
S
.
G
u
n
a
s
e
k
a
r
a
n
,
T.
S
u
b
r
a
m
a
n
i
a
n
,
a
n
d
V
.
R
,
“
M
R
I
b
r
a
i
n
t
u
mo
r
d
e
t
e
c
t
i
o
n
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
mac
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s,
”
Me
a
su
r
e
m
e
n
t
:
S
e
n
so
rs
,
v
o
l
.
3
1
,
p
.
1
0
1
0
2
6
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
mea
se
n
.
2
0
2
4
.
1
0
1
0
2
6
.
[
1
7
]
N
.
G
.
R
e
z
k
,
S
.
A
l
sh
a
t
h
r
i
,
A
.
S
a
y
e
d
,
E
.
E.
-
D
.
H
e
m
d
a
n
,
a
n
d
H
.
El
-
B
e
h
e
r
y
,
“
S
e
c
u
r
e
h
y
b
r
i
d
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
M
R
I
-
b
a
se
d
b
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
i
n
smar
t
me
d
i
c
a
l
I
o
T
s
y
st
e
m
s,”
D
i
a
g
n
o
s
t
i
c
s
,
v
o
l
.
1
5
,
n
o
.
5
,
p
.
6
3
9
,
M
a
r
.
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
s
t
i
c
s1
5
0
5
0
6
3
9
.
[
1
8
]
M
.
F
.
A
h
a
m
e
d
e
t
a
l
.
,
“
A
r
e
v
i
e
w
o
n
b
r
a
i
n
t
u
m
o
r
s
e
g
m
e
n
t
a
t
i
o
n
b
a
s
e
d
o
n
d
e
e
p
l
e
a
r
n
i
n
g
m
e
t
h
o
d
s
w
i
t
h
f
e
d
e
r
a
t
e
d
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
i
z
e
d
Me
d
i
c
a
l
I
m
a
g
i
n
g
a
n
d
G
r
a
p
h
i
c
s
,
v
o
l
.
1
1
0
,
p
.
1
0
2
3
1
3
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
m
p
m
e
d
i
ma
g
.
2
0
2
3
.
1
0
2
3
1
3
.
[
1
9
]
S
.
B
a
k
a
s,
“
M
u
l
t
i
m
o
d
a
l
b
r
a
i
n
t
u
m
o
r
s
e
g
me
n
t
a
t
i
o
n
c
h
a
l
l
e
n
g
e
2
0
2
0
:
d
a
t
a
,
”
2
0
2
0
,
MIC
C
AI
2
0
2
0
Br
a
T
S
D
a
t
a
,
C
e
n
t
e
r
f
o
r
Bi
o
m
e
d
i
c
a
l
I
m
a
g
e
C
o
m
p
u
t
i
n
g
& A
n
a
l
y
t
i
c
s
.
[
2
0
]
B
.
H
.
M
e
n
z
e
e
t
a
l
.
,
“
T
h
e
m
u
l
t
i
m
o
d
a
l
b
r
a
i
n
t
u
mo
r
i
m
a
g
e
s
e
g
m
e
n
t
a
t
i
o
n
b
e
n
c
h
m
a
r
k
(
B
R
A
TS)
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
Me
d
i
c
a
l
I
m
a
g
i
n
g
,
v
o
l
.
3
4
,
n
o
.
1
0
,
p
p
.
1
9
9
3
–
2
0
2
4
,
O
c
t
.
2
0
1
5
,
d
o
i
:
1
0
.
1
1
0
9
/
T
M
I
.
2
0
1
4
.
2
3
7
7
6
9
4
.
[
2
1
]
S
.
B
a
k
a
s
e
t
a
l
.
,
“
A
d
v
a
n
c
i
n
g
t
h
e
c
a
n
c
e
r
g
e
n
o
me
a
t
l
a
s
g
l
i
o
ma
M
R
I
c
o
l
l
e
c
t
i
o
n
s
w
i
t
h
e
x
p
e
r
t
s
e
g
me
n
t
a
t
i
o
n
l
a
b
e
l
s
a
n
d
r
a
d
i
o
mi
c
f
e
a
t
u
r
e
s,
”
S
c
i
e
n
t
i
f
i
c
D
a
t
a
,
v
o
l
.
4
,
n
o
.
1
,
p
.
1
7
0
1
1
7
,
S
e
p
.
2
0
1
7
,
d
o
i
:
1
0
.
1
0
3
8
/
sd
a
t
a
.
2
0
1
7
.
1
1
7
.
[
2
2
]
S
.
B
a
k
a
s
e
t
a
l
.
,
“
I
d
e
n
t
i
f
y
i
n
g
t
h
e
b
e
s
t
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
m
s
f
o
r
b
r
a
i
n
t
u
m
o
r
s
e
g
me
n
t
a
t
i
o
n
,
p
r
o
g
r
e
ssi
o
n
a
sses
sme
n
t
,
a
n
d
o
v
e
r
a
l
l
s
u
r
v
i
v
a
l
p
r
e
d
i
c
t
i
o
n
i
n
t
h
e
B
R
A
TS c
h
a
l
l
e
n
g
e
,
”
a
rXi
v
p
re
p
r
i
n
t
a
rXi
v
:
1
8
1
1
.
0
2
6
2
9
,
2
0
1
9
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
8
1
1
.
0
2
6
2
9
.
[
2
3
]
P
.
K
.
D
a
r
a
b
i
,
“
M
e
d
i
c
a
l
i
m
a
g
e
d
a
t
a
s
e
t
:
b
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
,
”
2
0
2
4
,
K
a
g
g
l
e
.
[
2
4
]
B
.
D
.
W
i
l
l
i
a
mso
n
,
P
.
B
.
G
i
l
b
e
r
t
,
M
.
C
a
r
o
n
e
,
a
n
d
N
.
S
i
m
o
n
,
“
N
o
n
p
a
r
a
met
r
i
c
v
a
r
i
a
b
l
e
i
mp
o
r
t
a
n
c
e
a
ssess
me
n
t
u
si
n
g
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s,
”
B
i
o
m
e
t
r
i
c
s
,
v
o
l
.
7
7
,
n
o
.
1
,
p
p
.
9
–
2
2
,
M
a
r
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
1
1
/
b
i
o
m.
1
3
3
9
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.