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.
1
4
,
No
.
3
,
Dec
em
b
er
20
2
5
,
p
p
.
1
0
0
2
~
1
0
1
4
I
SS
N:
2252
-
8
7
7
6
,
DOI
:
1
0
.
1
1
5
9
1
/iji
ct
.
v
1
4
i
3
.
pp
1
0
0
2
-
1
0
1
4
1002
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ict.
ia
esco
r
e.
co
m
Adv
a
ncements
i
n
bra
in t
umo
r
cla
ss
ificatio
n:
a
su
rv
ey
of
trans
fer
l
ea
rning
techniqu
es
Sn
eha
l J
a
dh
a
v
,
Sm
it
a
B
ha
rne,
Va
ibh
a
v
Na
ra
wa
de
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
R
a
mr
a
o
A
d
i
k
I
n
st
i
t
u
t
e
o
f
Te
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
Sep
1
7
,
2
0
2
4
R
ev
is
ed
Ma
r
1
9
,
2
0
2
5
Acc
ep
ted
J
u
n
9
,
2
0
2
5
Th
is
su
r
v
e
y
a
rti
c
le
p
re
se
n
ts
a
c
ri
ti
c
a
l
re
v
iew
o
f
th
e
sta
te
-
of
-
th
e
-
a
r
t
tran
sfe
r
lea
rn
in
g
(
TL
)
m
e
th
o
d
o
lo
g
ies
a
p
p
li
e
d
in
t
h
e
field
o
f
b
ra
i
n
tu
m
o
r
c
las
sifica
ti
o
n
,
wit
h
a
s
p
e
c
ial
e
m
p
h
a
sis
o
n
th
e
ir
v
a
rio
u
s
c
o
n
tri
b
u
ti
o
n
s
a
n
d
a
ss
o
c
iate
d
p
e
rfo
rm
a
n
c
e
m
e
tri
c
s.
We
will
d
isc
u
ss
v
a
ri
o
u
s
p
re
-
p
ro
c
e
ss
in
g
a
p
p
ro
a
c
h
e
s,
th
e
u
n
d
e
rly
i
n
g
fi
n
e
-
t
u
n
i
n
g
stra
teg
ies
,
wh
e
t
h
e
r
u
se
d
p
u
re
ly
o
r
in
a
n
e
n
d
-
to
-
e
n
d
train
i
n
g
m
a
n
n
e
r,
a
n
d
m
u
lt
i
-
m
o
d
a
l
a
p
p
li
c
a
ti
o
n
s.
T
h
e
c
u
rre
n
t
stu
d
y
sp
e
c
ifi
c
a
ll
y
h
ig
h
li
g
h
ts
th
e
a
p
p
li
c
a
ti
o
n
o
f
VG
G
1
6
a
n
d
re
sid
u
a
l
n
e
tw
o
rk
(
Re
sN
e
t
)
m
e
th
o
d
s
fo
r
fe
a
tu
re
e
x
trac
ti
o
n
,
d
e
m
o
n
stra
ti
n
g
th
a
t
lev
e
ra
g
in
g
h
i
g
h
-
o
rd
e
r
fe
a
tu
re
s
in
m
a
g
n
e
ti
c
re
so
n
a
n
c
e
ima
g
in
g
(
M
RI
)
ima
g
e
s
c
a
n
e
n
h
a
n
c
e
a
c
c
u
ra
c
y
wh
il
e
re
d
u
c
i
n
g
train
i
n
g
.
We
fu
rt
h
e
r
a
n
a
ly
z
e
fi
n
e
-
tu
n
in
g
m
e
th
o
d
s
i
n
re
latio
n
to
t
h
e
ir
r
o
le
i
n
o
p
ti
m
izi
n
g
m
o
d
e
l
lay
e
rs
fo
r
sm
a
ll
,
d
o
m
a
in
-
sp
e
c
ifi
c
d
a
tas
e
ts,
fin
d
in
g
th
e
m
p
a
rti
c
u
larl
y
e
ffe
c
ti
v
e
in
e
n
h
a
n
c
in
g
p
e
rfo
rm
a
n
c
e
o
n
th
e
sm
a
ll
b
ra
in
tu
m
o
r
d
a
tas
e
t.
It
wil
l
lo
o
k
in
to
e
n
d
-
to
-
e
n
d
train
i
n
g
,
wh
i
c
h
m
e
a
n
s
fin
e
-
tu
n
in
g
m
o
d
e
ls
t
h
a
t
h
a
v
e
a
lrea
d
y
b
e
e
n
trai
n
e
d
o
n
larg
e
d
a
tas
e
ts
to
m
a
k
e
th
e
m
b
e
tt
e
r.
I
t
will
a
lso
p
re
se
n
t
m
u
lt
imo
d
a
l
TL
a
s
a
wa
y
t
o
u
se
b
o
t
h
M
RI
a
n
d
c
o
m
p
u
ted
t
o
m
o
g
ra
p
h
y
(
CT
)
sc
a
n
d
a
ta
to
g
e
t
b
e
tt
e
r
c
las
sifica
ti
o
n
re
su
lt
s.
Co
m
p
a
rin
g
d
iffere
n
t
p
re
-
train
e
d
m
o
d
e
ls
c
a
n
p
ro
v
id
e
a
b
e
tt
e
r
u
n
d
e
rsta
n
d
in
g
o
f
th
e
stre
n
g
th
s
a
n
d
we
a
k
n
e
ss
e
s
a
ss
o
c
iate
d
with
th
e
p
a
rt
icu
lar
b
r
a
in
tu
m
o
r
c
las
sifica
ti
o
n
tas
k
.
T
h
is
re
v
iew
a
ims
to
a
n
a
ly
z
e
t
h
e
a
d
v
a
n
c
e
m
e
n
ts
in
TL
fo
r
m
e
d
ica
l
ima
g
e
a
n
a
ly
sis a
n
d
e
x
p
lo
re
p
o
ten
ti
a
l
a
v
e
n
u
e
s fo
r
fu
t
u
re
re
se
a
rc
h
a
n
d
d
e
v
e
lo
p
m
e
n
t
in
th
is cru
c
ial
field
o
f
m
e
d
ica
l
d
iag
n
o
st
ics
.
K
ey
w
o
r
d
s
:
B
r
ain
tu
m
o
r
class
if
icatio
n
Dee
p
lear
n
in
g
Ma
ch
in
e
lear
n
in
g
MRI
im
ag
es
T
r
an
s
f
er
lear
n
i
n
g
T
u
m
o
r
d
etec
tio
n
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
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
R
am
r
a
o
Ad
ik
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
D.
Y.
Patil d
ee
m
ed
to
b
e
Un
iv
er
s
ity
Nav
i M
u
m
b
ai,
I
n
d
ia
E
m
ail: sm
ita1
4
6
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
On
e
o
f
th
e
ess
en
tial
d
iag
n
o
s
tic
task
s
in
m
ed
icin
e
is
th
e
clas
s
if
icatio
n
o
f
b
r
ain
tu
m
o
r
s
,
wh
i
ch
aim
s
to
r
ec
o
g
n
ize
v
ar
io
u
s
ty
p
es
o
f
b
r
ain
tu
m
o
r
s
a
n
d
d
is
tin
g
u
is
h
th
em
f
r
o
m
ea
ch
o
th
er
to
estab
lis
h
an
ap
p
r
o
p
r
iate
tr
ea
tm
en
t
p
lan
.
T
h
e
ty
p
e
an
d
s
tag
e
o
f
b
r
ai
n
tu
m
o
r
d
eter
m
in
e
th
e
tr
ea
tm
en
t
an
d
th
e
p
r
o
g
n
o
s
is
;
th
er
ef
o
r
e,
p
r
o
p
er
d
iag
n
o
s
is
is
cr
u
cial
in
en
h
an
cin
g
p
atien
t
o
u
tco
m
es
[
1
]
,
[
2
]
.
M
a
g
n
etic
r
eso
n
a
n
ce
i
m
ag
in
g
(
MRI
)
a
n
d
co
m
p
u
ted
to
m
o
g
r
ap
h
y
(
CT
)
s
ca
n
s
ar
e
th
e
m
o
s
t
p
r
ef
er
r
ed
m
o
d
alities
u
s
ed
in
th
e
d
iag
n
o
s
is
o
f
co
n
d
i
tio
n
s
af
f
ec
tin
g
th
e
b
r
ai
n
b
ec
a
u
s
e
o
f
th
e
h
ig
h
r
eso
lu
tio
n
a
n
d
n
o
n
-
in
v
asiv
e
tech
n
i
q
u
e
u
s
ed
in
th
eir
ex
ec
u
tio
n
[
3
]
.
MRI
is
esp
ec
ially
h
elp
f
u
l
b
ec
au
s
e
o
f
th
e
h
ig
h
co
n
t
r
ast
o
f
s
o
f
t
tis
s
u
es
[
4
]
a
n
d
is
th
er
ef
o
r
e
v
er
y
v
alu
ab
le
in
th
e
d
etec
tio
n
o
f
b
r
ain
tu
m
o
r
s
,
th
e
i
r
s
ize,
lo
ca
tio
n
,
an
d
m
alig
n
a
n
c
y
.
It
s
ab
ilit
y
to
p
r
o
v
id
e
h
ig
h
-
d
ef
in
itio
n
im
ag
es o
f
an
ato
m
ical
s
tr
u
ctu
r
es,
esp
ec
ially
th
e
b
r
ain
,
m
ak
es
it
th
e
g
o
ld
s
tan
d
a
r
d
f
o
r
b
r
ain
tu
m
o
r
d
iag
n
o
s
is
.
C
T
,
o
r
co
m
p
u
ter
ized
to
m
o
g
r
a
p
h
ic
i
m
ag
in
g
,
p
r
o
v
id
es
cr
o
s
s
-
s
ec
tio
n
p
ictu
r
es
o
f
th
e
b
r
ain
u
s
in
g
X
-
r
ay
s
an
d
is
f
aster
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
A
d
va
n
ce
men
ts
in
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
:
a
s
u
r
ve
y
o
f tra
n
s
fer lea
r
n
in
g
tech
n
iq
u
es
(
S
n
eh
a
l J
a
d
h
a
v
)
1003
an
d
m
o
r
e
ac
ce
s
s
ib
le
th
an
MRI.
I
t
ca
n
also
b
e
h
elp
f
u
l
in
ca
s
es
o
f
h
ea
d
tr
a
u
m
a.
T
h
ey
ar
e
als
o
u
s
ef
u
l
in
v
iewin
g
ca
lcif
icatio
n
s
,
wh
ich
ca
n
s
o
m
etim
es
b
e
d
if
f
icu
lt
to
v
is
u
alize
in
an
MRI
s
ca
n
,
an
d
in
id
en
tify
in
g
b
o
n
e
in
v
o
lv
em
e
n
t
.
Desp
ite
th
e
d
if
f
e
r
en
ce
in
th
e
g
iv
en
a
d
v
an
tag
es
o
f
b
o
t
h
m
o
d
alities
,
in
clin
ical
p
r
ac
tice,
th
ey
wo
r
k
to
g
eth
er
a
n
d
g
iv
e
a
n
o
v
er
all
v
iew
o
f
th
e
tu
m
o
r
an
d
its
b
o
u
n
d
ar
ies.
R
ec
en
t
im
p
r
o
v
em
en
ts
i
n
ML
,
DL
,
an
d
T
L
h
av
e
p
r
o
v
en
t
o
b
e
u
s
ef
u
l in
im
p
r
o
v
in
g
th
e
p
r
o
ce
s
s
o
f
au
t
o
m
a
tic
d
etec
tio
n
an
d
c
h
ar
ac
ter
izati
o
n
o
f
b
r
ain
t
u
m
o
r
s
f
r
o
m
t
h
ese
im
ag
in
g
tech
n
i
q
u
e
s
.
B
y
co
m
b
in
i
n
g
MRI
a
n
d
/o
r
C
T
im
ag
es,
em
er
g
in
g
co
m
p
u
t
atio
n
al
tech
n
iq
u
es
en
ab
le
m
o
r
e
ac
cu
r
ate
tu
m
o
r
class
if
icatio
n
[
5
]
,
[
6
]
th
er
e
b
y
r
e
d
u
cin
g
th
e
r
ad
io
lo
g
is
ts
’
wo
r
k
l
o
ad
.
T
h
ese
co
m
b
in
ed
im
ag
i
n
g
m
o
d
alities
with
n
ew
co
m
p
u
tatio
n
al
co
n
ce
p
ts
h
av
e
ex
ten
d
e
d
ap
p
licatio
n
s
an
d
th
e
p
o
s
s
ib
ilit
y
o
f
i
n
cr
ea
s
ed
d
iag
n
o
s
is
ac
cu
r
ac
y
a
n
d
f
aster
,
less
in
v
asiv
e
d
iag
n
o
s
is
o
f
b
r
ain
tu
m
o
r
s
,
wh
ic
h
h
as
p
o
s
itiv
e
im
p
ac
ts
o
n
p
atien
ts
’
c
ar
e
an
d
tr
ea
tm
e
n
t.
T
h
e
r
elev
an
t
d
etailed
liter
atu
r
e
s
u
r
v
ey
is
d
escr
ib
ed
h
er
e.
T
h
e
f
ir
s
t
f
iv
e
in
v
esti
g
atio
n
s
in
th
e
B
C
b
r
ain
tu
m
o
r
class
i
f
icatio
n
liter
atu
r
e
p
r
esen
t
v
ar
io
u
s
n
o
v
el
s
tr
ateg
ies
f
o
r
e
n
h
an
cin
g
th
e
p
r
ec
is
io
n
an
d
s
p
ee
d
o
f
B
C
d
iag
n
o
s
tic
r
esu
lts
.
S
tu
d
y
[
1
]
s
p
ec
if
ically
d
esig
n
ed
a
n
o
v
el
c
o
n
v
o
lu
ti
o
n
al
n
e
u
r
al
n
etwo
r
k
(
C
NN)
ar
ch
itectu
r
e
to
class
if
y
th
r
ee
t
y
p
es
o
f
b
r
ain
tu
m
o
r
s
f
r
o
m
MRI
im
ag
e
s
:
m
en
in
g
io
m
as,
g
lio
m
as,
a
n
d
p
itu
itar
y
tu
m
o
r
s
.
T
h
e
s
tu
d
y
u
tili
ze
d
co
n
t
r
ast
-
en
h
an
ce
d
T
1
MRI
im
a
g
es
an
d
d
em
o
n
s
tr
ated
th
at
th
ei
r
p
r
o
p
o
s
ed
C
NN
m
o
d
el
class
if
ies
th
e
im
ag
es
with
an
ac
cu
r
ac
y
o
f
9
2
%,
o
u
tp
e
r
f
o
r
m
in
g
th
e
p
r
ev
i
o
u
s
m
eth
o
d
s
.
T
h
e
y
ac
h
iev
ed
a
5
0
%
im
p
r
o
v
em
e
n
t
in
ac
cu
r
ac
y
u
s
in
g
r
elen
t
-
less
ten
-
f
o
l
d
cr
o
s
s
-
v
alid
atio
n
o
n
en
h
a
n
ce
d
p
ictu
r
e
r
ep
o
s
ito
r
ies.
T
h
is
m
o
d
el
-
b
ased
s
tr
ateg
y
’
s
f
ea
tu
r
es
also
s
h
o
wca
s
e
its
p
o
ten
tial
as
an
in
s
tr
u
m
e
n
t
f
o
r
m
e
d
ical
d
iag
n
o
s
is
.
I
n
lin
e
with
s
tu
d
y
[
7
]
,
w
h
ich
id
e
n
tifie
d
th
e
n
ee
d
f
o
r
au
to
m
atic
class
if
icatio
n
o
f
b
r
ain
tu
m
o
r
ty
p
es
.
T
h
ey
p
r
esen
ted
a
n
au
to
m
ated
ap
p
r
o
ac
h
t
h
at
in
v
o
lv
ed
en
h
an
ci
n
g
im
ag
es
f
o
r
b
e
tter
v
is
u
aliza
tio
n
,
f
o
llo
wed
b
y
f
ea
tu
r
e
ex
tr
ac
tio
n
u
s
in
g
two
p
r
e
-
tr
ai
n
ed
d
ee
p
le
ar
n
in
g
(
DL
)
m
o
d
els
p
r
esen
ted
an
au
to
m
ated
ap
p
r
o
ac
h
th
at
i
n
v
o
lv
e
d
en
h
an
cin
g
im
ag
es
f
o
r
b
etter
v
is
u
aliza
tio
n
,
f
o
llo
wed
b
y
f
ea
t
u
r
e
ex
tr
a
ctio
n
u
s
in
g
two
p
r
e
-
tr
ain
ed
d
ee
p
lear
n
i
ng
(
DL
)
m
o
d
els.
T
h
e
PLS
co
m
p
iles
all
th
ese
f
ea
tu
r
es
in
to
a
s
in
g
le
h
y
b
r
id
v
ec
to
r
,
a
n
d
a
g
g
lo
m
er
ativ
e
clu
s
ter
in
g
id
en
tifie
s
th
e
tu
m
o
r
lo
ca
tio
n
.
Ultim
ately
,
em
p
lo
y
E
f
f
icien
tNet
-
B
0
f
o
r
th
e
f
in
al
class
if
ic
atio
n
.
T
h
is
m
eth
o
d
s
u
cc
ess
f
u
lly
class
if
ied
th
e
d
atasets
w
i
th
a
h
ig
h
ac
cu
r
ac
y
o
f
ap
p
r
o
x
im
ately
9
5
%.
T
h
e
ac
c
u
r
ac
y
in
d
iag
n
o
s
in
g
m
en
in
g
io
m
a
,
g
lio
m
a
,
an
d
p
it
u
itar
y
tu
m
o
r
s
was
p
ar
ticu
lar
l
y
h
ig
h
,
with
9
8
%
ac
cu
r
ac
y
,
r
esp
ec
tiv
ely
.
3
1
%,
9
8
.
7
2
%,
an
d
9
9
.
4
6
%,
r
esp
ec
tiv
ely
.
T
h
is
is
an
ef
f
ec
tiv
e
s
o
lu
tio
n
to
th
e
p
r
o
b
lem
s
ass
o
c
iated
with
m
an
u
al
class
if
icatio
n
s
in
ce
th
i
s
m
et
h
o
d
o
f
f
er
s
an
alm
o
s
t
1
0
0
%
au
to
m
ated
p
r
o
ce
s
s
.
T
h
e
s
tu
d
y
[
8
]
d
esig
n
ed
an
ap
p
r
o
ac
h
th
at
in
co
r
p
o
r
ated
DL
an
d
im
p
o
r
ta
n
t
im
ag
e
p
r
o
ce
s
s
in
g
m
eth
o
d
s
b
ased
o
n
th
e
E
f
f
icien
tNet
m
o
d
el
to
im
p
r
o
v
e
th
e
p
e
r
f
o
r
m
an
ce
o
f
b
r
ain
t
u
m
o
r
class
if
icatio
n
.
Pr
ep
r
o
ce
s
s
in
g
o
f
MRI
im
ag
es
was
p
er
f
o
r
m
e
d
u
s
in
g
cr
o
p
p
in
g
,
r
esizin
g
,
d
en
o
is
in
g
,
an
d
n
o
r
m
aliza
tio
n
tech
n
i
q
u
es;
f
ea
tu
r
e
ex
tr
ac
tio
n
was
d
o
n
e
u
s
in
g
Den
s
eNe
t1
2
1
,
an
d
th
e
m
o
d
el
u
s
es
s
ig
m
o
id
ac
tiv
atio
n
f
o
r
th
e
class
if
icatio
n
.
T
h
e
o
b
tain
e
d
r
esu
lts
s
h
o
wed
f
air
ly
h
ig
h
r
ec
alls
,
r
an
g
in
g
f
r
o
m
8
7
%
to
a
p
ea
k
o
f
9
2
%,
p
r
ec
is
io
n
o
f
9
3
.
8
2
%,
F1
-
s
co
r
e
o
f
9
3
.
1
5
%,
an
d
o
v
er
all
ac
cu
r
ac
y
o
f
9
4
%.8
3
%.
T
h
is
wo
r
k
s
h
o
wca
s
es
th
at,
th
r
o
u
g
h
in
teg
r
atin
g
a
d
v
an
ce
d
im
ag
e
an
al
y
s
is
an
d
d
ee
p
r
ein
f
o
r
ce
m
e
n
t
lear
n
in
g
,
o
n
e
ca
n
o
b
ta
i
n
a
r
e
lativ
ely
h
ig
h
lev
el
o
f
g
r
u
eso
m
en
ess
in
tu
m
o
r
id
en
tific
atio
n
an
d
d
elin
ea
tio
n
.
Stu
d
y
[
9
]
in
teg
r
ated
d
ee
p
an
d
s
h
allo
w
f
ea
tu
r
e
e
x
tr
ac
tio
n
to
d
is
tin
g
u
is
h
b
r
ain
tu
m
o
r
s
an
d
f
o
r
ec
ast
th
e
1
p
/1
9
q
co
-
d
eletio
n
s
tatu
s
o
f
L
GG
tu
m
o
r
s
.
Featu
r
e
ex
tr
ac
tio
n
was
p
er
f
o
r
m
ed
u
s
in
g
p
r
e
-
tr
ain
ed
n
etwo
r
k
s
in
clu
d
in
g
Alex
Net,
R
esNet
-
1
8
,
Go
o
g
L
eNe
t,
an
d
Sh
u
f
f
leNe
t
to
ex
t
r
ac
t
th
e
d
ee
p
f
ea
tu
r
es,
wh
ile
a
s
im
p
le
s
h
allo
w
n
etwo
r
k
ca
p
tu
r
e
d
th
e
lo
w
-
lev
el
d
etail.
T
h
ese
f
ea
tu
r
es
wer
e
co
u
p
led
,
an
d
th
e
class
if
icatio
n
was
d
o
n
e
with
t
h
e
h
elp
o
f
SVM
as
well
as
th
e
k
-
NN
class
if
ier
s
.
T
h
is
co
m
m
o
n
in
te
g
r
atio
n
d
em
o
n
s
tr
ated
th
at
th
e
f
u
s
io
n
ap
p
r
o
ac
h
,
wh
en
co
m
b
i
n
ed
wi
th
th
e
en
lar
g
e
m
en
t
o
f
th
e
tu
m
o
r
r
eg
i
o
n
o
f
in
ter
est
(
R
OI
)
,
e
n
h
an
ce
d
s
en
s
itiv
ity
b
y
1
1
%.
T
h
ese
r
esu
lts
s
h
o
w
t
h
at
b
o
th
i
n
f
o
r
m
ativ
e
a
n
d
n
o
n
-
in
f
o
r
m
ativ
e
f
ea
tu
r
es
ar
e
im
p
o
r
tan
t
f
o
r
im
p
r
o
v
in
g
class
if
icatio
n
ac
cu
r
ac
y
an
d
,
as
a
r
esu
lt,
m
ak
in
g
a
b
etter
d
iag
n
o
s
tic
s
y
s
tem
.
A
s
tu
d
y
[
1
0
]
lo
o
k
ed
in
to
an
alg
o
r
ith
m
th
at
u
s
ed
d
ee
p
C
NNs
an
d
a
n
atu
r
e
-
in
s
p
ir
ed
R
esNet1
5
2
tr
a
n
s
f
er
lear
n
in
g
(TL)
m
o
d
el
to
h
elp
f
in
d
b
r
ain
tu
m
o
r
s
a
n
d
tell
th
e
m
ap
ar
t.
Pre
p
r
o
ce
s
s
in
g
was
d
o
n
e
to
th
e
im
ag
es
t
o
elim
in
ate
n
o
is
e
an
d
to
in
c
r
ea
s
e
th
e
q
u
ality
o
f
t
h
e
ac
q
u
i
r
ed
v
ec
t
o
r
s
u
s
in
g
Ots
u
b
in
ar
izatio
n
,
wh
i
le
f
ea
tu
r
e
ex
tr
ac
tio
n
u
s
ed
GL
C
M
m
eth
o
d
s
.
W
ith
an
ac
cu
r
a
cy
o
f
9
9
%,
t
h
e
C
o
v
id
-
1
9
alg
o
r
ith
m
o
f
weig
h
t
t
u
n
in
g
is
th
en
ap
p
lied
with
R
esNet1
5
2
,
wh
ich
is
r
ec
o
g
n
ized
as
a
h
y
b
r
id
m
o
d
e
l.
C
o
m
p
ar
ed
to
ex
is
tin
g
tech
n
iq
u
es,
it
ac
h
iev
es
a
lo
w
er
r
o
r
r
ate
o
f
5
7
%.
I
t
illu
s
tr
ates
th
e
er
r
o
r
r
ates
an
d
tim
e
co
m
p
lex
ity
ass
o
ciate
d
with
b
r
ain
t
u
m
o
r
d
etec
tio
n
,
an
d
p
r
o
p
o
s
es
m
o
r
e
ac
cu
r
ate
an
d
e
f
f
icien
t
s
o
lu
tio
n
s
.
T
h
ese
s
tu
d
ies
ar
e
g
r
o
u
n
d
b
r
ea
k
in
g
in
th
e
class
if
icatio
n
o
f
b
r
ai
n
tu
m
o
r
s
,
d
em
o
n
s
tr
atin
g
a
s
h
if
t t
o
war
d
s
au
to
m
atin
g
c
u
r
r
en
t c
lass
if
icatio
n
s
y
s
tem
s
an
d
u
tili
zin
g
a
v
ar
i
ety
o
f
DL
m
eth
o
d
s
to
en
h
an
ce
th
e
p
r
ec
is
io
n
an
d
e
f
f
ec
tiv
en
ess
o
f
th
is
d
iag
n
o
s
tic
f
ield
.
T
h
e
s
tu
d
y
[
1
1
]
p
r
esen
ts
a
g
en
er
al
f
r
a
m
ewo
r
k
f
o
r
b
r
ain
tu
m
o
r
class
if
icatio
n
,
lo
ca
liz
atio
n
,
an
d
s
eg
m
en
tatio
n
u
s
in
g
T
1
-
weig
h
ted
co
n
tr
ast
-
en
h
an
ce
d
(
T
1
W
-
C
E
)
MRI
im
ag
es.
Data
s
p
lits
in
to
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
p
ar
ts
an
d
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
es
s
u
ch
as
wav
elet
d
ec
o
m
p
o
s
itio
n
an
d
g
eo
m
etr
ical
tr
an
s
f
o
r
m
atio
n
s
wer
e
u
s
ed
in
th
is
wo
r
k
,
as
wer
e
t
wo
Dar
k
Net
m
o
d
els
(
Dar
k
Net
-
1
9
an
d
Dar
k
Net
-
5
3
)
th
at
wer
e
p
r
e
-
tr
ain
ed
o
n
o
th
er
d
atasets
.
T
h
e
Dar
k
Net
-
5
3
m
o
d
el
ac
h
iev
ed
u
n
p
r
ec
ed
en
te
d
s
u
cc
ess
in
test
in
g
,
ac
h
iev
in
g
9
8
.
5
4
%
ac
cu
r
ac
y
,
an
ar
ea
u
n
d
er
cu
r
v
e
(
AUC)
o
f
0
.
9
9
,
an
d
a
Dice
in
d
ex
o
f
0
.
9
4
f
o
r
tu
m
o
r
s
eg
m
en
tatio
n
.
T
h
is
ap
p
r
o
ac
h
s
h
o
ws
a
r
em
ar
k
ab
le
im
p
r
o
v
e
m
en
t
in
th
e
o
p
p
o
r
tu
n
ity
to
p
e
r
f
o
r
m
tu
m
o
r
a
n
aly
s
is
th
r
o
u
g
h
co
m
p
u
ter
-
aid
ed
s
y
s
te
m
s
as
we
ll
as
it
s
ap
p
licab
ilit
y
to
clin
ical
p
r
ac
tice.
DL
u
s
i
n
g
th
e
E
f
f
icien
tNet
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.
1
4
,
No
.
3
,
Dec
em
b
er
20
2
5
:
1
0
0
2
-
1
0
1
4
1004
f
am
ily
was
p
r
o
p
o
s
ed
in
s
tu
d
y
[
1
2
]
as
a
wa
y
o
f
im
p
r
o
v
in
g
th
e
class
if
icatio
n
an
d
d
etec
tio
n
o
f
b
r
ai
n
tu
m
o
r
s
.
T
h
e
ap
p
r
o
ac
h
es
u
tili
ze
d
a
d
at
aset
o
f
3
0
6
4
T
1
-
weig
h
ted
C
E
MRI
im
ag
es,
al
o
n
g
with
p
r
e
-
p
r
o
ce
s
s
in
g
an
d
d
ata
au
g
m
en
tatio
n
tec
h
n
iq
u
es,
to
i
m
p
r
o
v
e
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
n
u
m
er
ical
m
o
d
els.
Am
o
n
g
a
ll
th
e
m
o
d
els
u
s
ed
,
th
e
m
o
s
t
ac
cu
r
ate
m
o
d
el
was
th
e
E
f
f
icien
tNetB
3
m
o
d
el,
r
ea
ch
in
g
an
ac
cu
r
ac
y
o
f
9
9
.
6
9
%,
an
d
it
p
er
f
o
r
m
s
co
n
s
id
er
ab
ly
b
ett
er
th
an
m
an
y
o
th
e
r
e
x
is
tin
g
s
tate
-
of
-
t
h
e
-
a
r
t
m
eth
o
d
s
.
T
o
r
ea
c
h
t
h
is
co
n
clu
s
io
n
,
t
h
is
s
tu
d
y
s
h
o
ws
th
at
en
g
r
av
in
g
E
f
f
icien
tNet
en
ab
led
f
aster
,
m
o
r
e
r
eli
ab
le
d
iag
n
o
s
is
,
en
h
a
n
cin
g
th
e
ea
r
ly
d
etec
tio
n
o
f
b
r
ain
tu
m
o
r
s
.
I
n
s
tu
d
y
[
1
3
]
,
t
h
e
r
esear
ch
er
s
aim
ed
to
r
e
d
u
ce
th
e
er
r
o
r
r
ate
an
d
co
m
p
u
t
atio
n
al
tim
e
in
th
e
d
ev
elo
p
m
e
n
t
o
f
a
b
r
ain
t
u
m
o
r
d
etec
tio
n
a
p
p
r
o
ac
h
,
u
tili
zin
g
t
h
e
d
ee
p
C
NN
an
d
th
e
n
at
u
r
e
-
i
n
s
p
ir
ed
R
esNet1
5
2
tr
an
s
f
er
lear
n
in
g
(
Hy
b
-
DC
NN
-
R
esNet1
5
2
T
L
)
.
T
h
e
s
tu
d
y
in
clu
d
ed
im
ag
e
d
en
o
is
in
g
an
d
i
m
ag
e
en
h
a
n
ce
m
en
t
u
s
in
g
Ots
u
b
in
ar
izatio
n
o
f
im
ag
es
as
well
a
s
th
e
ex
tr
ac
tio
n
o
f
f
ea
tu
r
es
u
s
in
g
GL
C
M
m
eth
o
d
s
.
T
h
e
h
y
b
r
id
m
o
d
el,
o
p
tim
ized
u
s
in
g
a
C
o
v
id
-
1
9
o
p
tim
izatio
n
alg
o
r
ith
m
,
s
h
o
wed
s
tr
ik
in
g
ac
cu
r
ac
y
r
ate
s
th
at
ca
m
e
with
in
th
e
r
an
g
e
o
f
9
4
.
3
1
%
to
9
9
.
7
%
s
u
cc
ess
r
ate
an
d
lo
wer
er
r
o
r
co
m
p
ar
ed
to
th
e
e
x
is
tin
g
tech
n
iq
u
es.
T
h
is
ap
p
r
o
ac
h
ef
f
ec
tiv
ely
r
e
d
u
ce
s
er
r
o
r
s
an
d
en
h
an
ce
s
co
m
p
u
ta
tio
n
al
ca
p
ab
ilit
ies
in
th
e
clas
s
if
icatio
n
o
f
b
r
ain
tu
m
o
r
s
.
I
n
th
e
s
tu
d
y
[
1
4
]
,
t
h
e
au
th
o
r
s
d
ev
elo
p
ed
a
n
ew
tech
n
iq
u
e
to
class
if
y
th
r
ee
d
if
f
er
en
t
ty
p
e
s
o
f
b
r
ain
tu
m
o
r
s
.
T
h
ey
em
p
lo
y
ed
n
o
r
m
aliza
tio
n
,
d
en
s
e,
s
p
ee
d
ed
-
u
p
r
o
b
u
s
t
f
ea
tu
r
es,
an
d
g
r
ad
ien
t
h
i
s
to
g
r
am
m
eth
o
d
s
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
th
e
M
R
I
im
ag
e
an
d
en
h
an
ce
th
e
f
ea
tu
r
es
f
o
r
m
o
r
e
d
etailed
class
if
icatio
n
.
T
h
e
m
eth
o
d
ad
o
p
ted
in
th
e
class
if
icatio
n
u
s
ed
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
with
an
ac
cu
r
ac
y
o
f
9
0
%.
T
h
is
wo
r
k
also
d
em
o
n
s
tr
ates
h
o
w
f
ea
tu
r
e
en
h
an
ce
m
en
t
tec
h
n
iq
u
es,
wh
en
co
m
b
in
ed
with
r
eliab
le
class
if
icatio
n
alg
o
r
ith
m
s
,
en
h
an
ce
th
e
d
ia
g
n
o
s
tic
ab
ilit
y
b
ey
o
n
d
wh
at
was
p
r
ev
io
u
s
ly
p
o
s
s
ib
le.
Fin
ally
,
s
tu
d
y
[
1
5
]
d
is
cu
s
s
ed
b
r
ain
tu
m
o
r
class
if
icatio
n
,
wh
ich
is
a
d
if
f
icu
lt
task
u
s
in
g
a
n
ew
tech
n
iq
u
e,
C
NN
with
TL
ap
p
r
o
ac
h
es.
T
h
e
s
tu
d
y
em
p
lo
y
ed
m
ix
e
d
C
NN,
wh
ic
h
was
au
g
m
e
n
ted
with
a
R
e
s
Net1
5
2
lay
er
an
d
o
p
tim
ized
b
y
th
e
C
o
v
id
-
1
9
o
p
tim
izatio
n
alg
o
r
ith
m
.
T
h
e
a
p
p
r
o
a
c
h
ac
h
iev
ed
h
ig
h
ac
cu
r
a
cy
r
ates,
r
ea
ch
in
g
u
p
to
9
9
p
e
r
ce
n
t.
I
n
cid
e
n
tally
,
th
ese
h
av
e
b
ee
n
r
e
p
o
r
ted
t
o
b
e
b
etwe
en
5
7
% a
n
d
s
ig
n
if
ica
n
tly
lo
wer
er
r
o
r
r
ates th
an
th
o
s
e
o
f
th
e
co
n
v
en
tio
n
al
ap
p
r
o
ac
h
es.
T
h
is
s
tu
d
y
f
o
cu
s
es
o
n
th
e
en
h
an
ce
d
ca
p
ab
ilit
i
es
o
f
ad
v
an
ce
d
n
e
u
r
al
n
etwo
r
k
ar
ch
itectu
r
es
an
d
o
p
tim
izatio
n
alg
o
r
ith
m
s
,
with
th
e
aim
o
f
r
ed
u
cin
g
er
r
o
r
s
a
n
d
en
h
an
cin
g
class
if
icatio
n
ab
ilit
y
.
C
o
llectiv
ely
,
th
ese
wo
r
k
s
p
r
esen
t
v
ar
io
u
s
s
tate
-
of
-
th
e
-
ar
t
a
p
p
r
o
ac
h
es
to
class
if
y
in
g
b
r
ain
tu
m
o
r
s
u
s
in
g
DL
f
r
am
ewo
r
k
s
,
s
elec
tin
g
an
d
ap
p
ly
i
n
g
d
ata
a
u
g
m
en
tatio
n
s
tr
ateg
ies,
an
d
m
et
h
o
d
s
o
f
f
ea
tu
r
e
s
elec
tio
n
an
d
c
lass
if
icatio
n
.
E
ac
h
s
tu
d
y
h
o
ld
s
s
ig
n
if
ica
n
ce
an
d
r
elev
an
ce
as
it
co
n
tr
i
b
u
tes
to
th
e
o
n
g
o
in
g
r
ef
in
em
e
n
t
o
f
d
iag
n
o
s
tic
to
o
ls
an
d
m
eth
o
d
o
l
o
g
ies,
wh
ich
i
n
tu
r
n
lead
s
to
en
h
a
n
ce
d
e
f
f
icien
cy
in
th
e
id
e
n
tific
atio
n
a
n
d
tr
ea
tm
en
t
o
f
b
r
ain
tu
m
o
r
s
.
T
ab
le
1
s
h
o
ws th
e
co
m
p
ar
ativ
e
an
aly
s
is
o
f
th
e
b
r
ain
tu
m
o
r
c
lass
if
icatio
n
s
tu
d
ies in
th
e
liter
atu
r
e.
T
ab
le
1
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
o
f
s
tr
en
g
th
s
an
d
wea
k
n
ess
es o
f
b
r
ai
n
tu
m
o
r
class
if
icatio
n
s
tu
d
ies
S
t
r
e
n
g
t
h
s
W
e
a
k
n
e
sses
C
i
t
a
t
i
o
n
s
H
i
g
h
c
l
a
ss
i
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
(
9
8
.
0
4
%)
a
c
r
o
ss
t
u
m
o
r
t
y
p
e
s
;
Ef
f
e
c
t
i
v
e
t
u
m
o
r
l
o
c
a
l
i
z
a
t
i
o
n
a
n
d
r
e
f
i
n
e
me
n
t
u
s
i
n
g
a
g
g
l
o
mer
a
t
i
v
e
c
l
u
st
e
r
i
n
g
;
R
e
f
i
n
e
d
p
r
o
p
o
s
a
l
s
i
n
c
r
e
a
se
a
c
c
u
r
a
c
y
.
R
e
l
i
a
n
c
e
o
n
a
g
g
l
o
mera
t
i
v
e
c
l
u
s
t
e
r
i
n
g
may
i
n
t
r
o
d
u
c
e
v
a
r
i
a
b
i
l
i
t
y
i
n
t
u
mo
r
p
r
o
p
o
sa
l
a
c
c
u
r
a
c
y
;
M
a
y
r
e
q
u
i
r
e
e
x
t
e
n
s
i
v
e
c
o
m
p
u
t
a
t
i
o
n
a
l
r
e
so
u
r
c
e
s
f
o
r
p
r
o
c
e
ss
i
n
g
.
[
6
]
I
mp
r
e
ssi
v
e
p
e
r
f
o
r
ma
n
c
e
m
e
t
r
i
c
s (r
e
c
a
l
l
:
9
2
.
8
7
%
,
p
r
e
c
i
si
o
n
:
9
3
.
8
2
%,
a
c
c
u
r
a
c
y
:
9
4
.
8
3
%)
;
Ef
f
e
c
t
i
v
e
u
se
o
f
D
e
n
seN
e
t
1
2
1
f
o
r
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
;
D
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
i
mp
r
o
v
e
s m
o
d
e
l
r
o
b
u
st
n
e
ss
.
M
a
y
n
o
t
a
d
d
r
e
ss
v
a
r
i
a
b
i
l
i
t
y
i
n
t
u
mo
r
c
h
a
r
a
c
t
e
r
i
st
i
c
s
o
r
d
a
t
a
se
t
b
i
a
s;
F
o
c
u
s
o
n
a
si
n
g
l
e
m
o
d
e
l
a
r
c
h
i
t
e
c
t
u
r
e
mi
g
h
t
l
i
mi
t
f
l
e
x
i
b
i
l
i
t
y
.
[
7
]
U
t
i
l
i
z
e
s fe
a
t
u
r
e
f
u
si
o
n
o
f
d
e
e
p
a
n
d
s
h
a
l
l
o
w
f
e
a
t
u
r
e
s
;
R
O
I
e
x
p
a
n
s
i
o
n
i
m
p
r
o
v
e
s
se
n
si
t
i
v
i
t
y
(
1
1
.
7
2
%
i
n
c
r
e
a
se)
;
C
o
m
p
e
t
i
t
i
v
e
r
e
s
u
l
t
s wi
t
h
R
e
sN
e
t
-
1
8
.
R
O
I
e
x
p
a
n
s
i
o
n
ma
y
i
n
c
r
e
a
s
e
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
m
p
l
e
x
i
t
y
;
S
h
a
l
l
o
w
n
e
t
w
o
r
k
d
e
si
g
n
mi
g
h
t
n
o
t
c
a
p
t
u
r
e
a
l
l
r
e
l
e
v
a
n
t
f
e
a
t
u
r
e
s
.
[
8
]
H
i
g
h
a
c
c
u
r
a
c
y
(
9
9
.
6
0
%) w
i
t
h
D
a
r
k
N
e
t
m
o
d
e
l
s
;
Ef
f
e
c
t
i
v
e
u
se
o
f
d
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
;
E
x
c
e
l
l
e
n
t
p
e
r
f
o
r
m
a
n
c
e
i
n
se
g
m
e
n
t
a
t
i
o
n
w
i
t
h
a
D
i
c
e
i
n
d
e
x
o
f
0
.
9
4
.
P
e
r
f
o
r
ma
n
c
e
ma
y
b
e
d
a
t
a
se
t
-
sp
e
c
i
f
i
c
;
D
a
r
k
N
e
t
mo
d
e
l
s
’
p
r
e
-
t
r
a
i
n
i
n
g
m
i
g
h
t
l
i
mi
t
g
e
n
e
r
a
l
i
z
a
b
i
l
i
t
y
t
o
o
t
h
e
r
d
a
t
a
s
e
t
s
.
[
9
]
O
u
t
s
t
a
n
d
i
n
g
p
e
r
f
o
r
ma
n
c
e
w
i
t
h
Ef
f
i
c
i
e
n
t
N
e
t
B
3
(
9
9
.
6
9
%
a
c
c
u
r
a
c
y
)
;
C
o
m
p
r
e
h
e
n
si
v
e
p
r
e
p
r
o
c
e
ss
i
n
g
a
n
d
a
u
g
me
n
t
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
;
H
i
g
h
r
e
l
i
a
b
i
l
i
t
y
f
o
r
c
l
i
n
i
c
a
l
set
t
i
n
g
s
.
M
a
y
r
e
q
u
i
r
e
e
x
t
e
n
s
i
v
e
c
o
m
p
u
t
a
t
i
o
n
a
l
r
e
so
u
r
c
e
s
;
D
e
p
e
n
d
e
n
c
e
o
n
Ef
f
i
c
i
e
n
t
N
e
t
ma
y
l
i
m
i
t
e
x
p
l
o
r
a
t
i
o
n
o
f
o
t
h
e
r
a
r
c
h
i
t
e
c
t
u
r
e
s.
[
1
0
]
H
i
g
h
a
c
c
u
r
a
c
y
(
9
8
.
5
4
%) a
n
d
A
U
C
(
0
.
9
9
)
w
i
t
h
D
a
r
k
N
e
t
mo
d
e
l
s
;
Ef
f
e
c
t
i
v
e
d
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
a
n
d
s
e
g
m
e
n
t
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s;
D
e
m
o
n
s
t
r
a
t
e
s
c
l
i
n
i
c
a
l
a
p
p
l
i
c
a
b
i
l
i
t
y
.
D
a
r
k
N
e
t
m
o
d
e
l
s
ma
y
n
o
t
g
e
n
e
r
a
l
i
z
e
w
e
l
l
t
o
o
t
h
e
r
d
a
t
a
se
t
s;
D
a
t
a
se
t
-
sp
e
c
i
f
i
c
p
e
r
f
o
r
ma
n
c
e
ma
y
n
o
t
r
e
f
l
e
c
t
b
r
o
a
d
e
r
a
p
p
l
i
c
a
b
i
l
i
t
y
.
[
1
1
]
Ex
c
e
p
t
i
o
n
a
l
a
c
c
u
r
a
c
y
w
i
t
h
Ef
f
i
c
i
e
n
t
N
e
t
B
3
(
9
9
.
6
9
%);
A
d
v
a
n
c
e
d
p
r
e
p
r
o
c
e
ssi
n
g
a
n
d
a
u
g
me
n
t
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s;
S
i
g
n
i
f
i
c
a
n
t
i
m
p
r
o
v
e
me
n
t
o
v
e
r
st
a
t
e
-
of
-
t
h
e
-
a
r
t
m
e
t
h
o
d
s
.
H
i
g
h
c
o
m
p
u
t
a
t
i
o
n
a
l
r
e
q
u
i
r
e
m
e
n
t
s;
Ef
f
i
c
i
e
n
t
N
e
t
B
3
’
s
p
e
r
f
o
r
m
a
n
c
e
mi
g
h
t
n
o
t
b
e
a
s
e
f
f
e
c
t
i
v
e
f
o
r
a
l
l
t
y
p
e
s
o
f
M
R
I
d
a
t
a
s
e
t
s.
[
1
2
]
H
i
g
h
a
c
c
u
r
a
c
y
(
u
p
t
o
9
9
.
5
7
%) a
n
d
l
o
w
e
r
r
o
r
r
a
t
e
s
;
Ef
f
e
c
t
i
v
e
u
se
o
f
h
y
b
r
i
d
C
N
N
w
i
t
h
R
e
s
N
e
t
1
5
2
;
O
p
t
i
mi
z
a
t
i
o
n
t
h
r
o
u
g
h
C
o
v
i
d
-
1
9
a
l
g
o
r
i
t
h
m
.
P
o
t
e
n
t
i
a
l
l
y
h
i
g
h
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
s
t
;
H
y
b
r
i
d
mo
d
e
l
may
b
e
c
o
m
p
l
e
x
a
n
d
h
a
r
d
e
r
t
o
f
i
n
e
-
t
u
n
e
.
[
1
3
]
S
t
r
o
n
g
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
u
s
i
n
g
n
o
r
ma
l
i
z
a
t
i
o
n
a
n
d
d
e
n
s
e
f
e
a
t
u
r
e
s;
G
o
o
d
p
e
r
f
o
r
m
a
n
c
e
w
i
t
h
S
V
M
c
l
a
ss
i
f
i
c
a
t
i
o
n
(
9
0
.
2
7
%
a
c
c
u
r
a
c
y
)
;
S
u
r
p
a
sses
p
r
e
v
i
o
u
s me
t
h
o
d
s
.
M
a
y
l
a
c
k
D
L
a
d
v
a
n
t
a
g
e
s;
S
V
M
m
a
y
n
o
t
b
e
a
s
r
o
b
u
st
a
s
mo
d
e
r
n
n
e
u
r
a
l
n
e
t
w
o
r
k
a
p
p
r
o
a
c
h
e
s fo
r
a
l
l
d
a
t
a
se
t
s.
[
1
4
]
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
A
d
va
n
ce
men
ts
in
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
:
a
s
u
r
ve
y
o
f tra
n
s
fer lea
r
n
in
g
tech
n
iq
u
es
(
S
n
eh
a
l J
a
d
h
a
v
)
1005
2.
AL
G
O
RI
T
H
M
US
E
D
T
h
is
s
ec
tio
n
h
ig
h
lig
h
ts
th
e
v
a
r
io
u
s
ty
p
es
o
f
b
r
ai
n
tu
m
o
r
cla
s
s
if
icatio
n
alg
o
r
ith
m
s
b
ased
o
n
m
ac
h
in
e
lear
n
in
g
(
ML
)
,
DL
,
a
n
d
T
L
.
2
.
1
.
M
a
chine
lea
rning
ba
s
ed
cla
s
s
if
ica
t
io
n
T
h
er
e
h
a
d
b
ee
n
p
r
o
g
r
ess
in
u
s
in
g
ML
,
esp
ec
ially
in
th
e
class
if
icatio
n
o
f
b
r
ai
n
tu
m
o
r
s
t
h
r
o
u
g
h
t
h
e
an
aly
s
is
o
f
co
m
p
lex
m
ed
ical
i
m
ag
in
g
d
ata.
Pre
v
io
u
s
ML
ap
p
r
o
ac
h
es,
in
cl
u
d
in
g
SVMs,
r
an
d
o
m
f
o
r
ests
,
an
d
k
-
NN,
h
a
v
e
b
ee
n
em
p
lo
y
ed
t
o
class
if
y
b
r
ain
tu
m
o
r
s
b
y
le
ar
n
in
g
f
r
o
m
MR
im
ag
es
e
x
tr
ac
ted
f
ea
tu
r
es
[
1
6
]
.
T
h
ese
m
o
d
els
r
eq
u
ir
e
e
x
ten
s
iv
e
p
r
ep
r
o
c
ess
in
g
an
d
n
o
n
-
au
t
o
m
atic
f
ea
tu
r
e
s
elec
tio
n
,
wh
ich
in
v
o
lv
es
s
ea
r
ch
in
g
f
o
r
s
p
ec
if
ic
f
ea
tu
r
es
with
in
th
e
d
ata
s
ets
to
id
en
tify
a
s
p
ec
if
i
c
tu
m
o
r
[
1
7
]
.
SVM
as
an
alg
o
r
ith
m
th
at
ex
ce
ls
in
wo
r
k
in
g
with
h
ig
h
-
d
im
e
n
s
io
n
al
s
p
ac
e
h
as
b
ee
n
u
s
ef
u
l
in
class
if
y
in
g
class
es
with
well
-
s
ep
ar
ated
m
ar
g
in
s
,
th
o
u
g
h
th
e
p
er
f
o
r
m
a
n
ce
d
ep
e
n
d
s
h
ea
v
ily
o
n
th
e
f
ea
tu
r
e
s
p
ac
e
u
s
ed
.
R
an
d
o
m
f
o
r
ests
,
b
ased
o
n
d
ec
is
io
n
tr
ee
s
,
ca
n
h
an
d
le
lar
g
e
d
atasets
an
d
p
r
o
v
id
e
a
q
u
a
n
titativ
e
m
ea
s
u
r
e
o
f
f
ea
tu
r
e
im
p
o
r
tan
ce
,
m
ak
in
g
th
em
u
s
ef
u
l
i
n
d
eter
m
in
in
g
w
h
ich
asp
ec
ts
o
f
th
e
im
ag
in
g
d
ata
co
n
tr
ib
u
te
t
o
th
e
class
if
icatio
n
[
1
8
]
.
Ho
w
ev
er
,
th
ese
m
o
d
els
ca
n
in
f
ac
t
b
ec
o
m
e
v
e
r
y
in
tr
ic
ate
an
d
less
u
n
d
e
r
s
tan
d
ab
le
wh
en
th
e
n
u
m
b
e
r
o
f
t
r
ee
s
ca
r
r
ie
d
is
h
i
g
h
.
k
-
NN
is
a
m
u
ch
s
im
p
ler
tech
n
iq
u
e
th
at
ca
n
b
e
p
er
u
s
ed
f
o
r
class
if
y
in
g
tu
m
o
r
s
b
ased
o
n
th
ei
r
m
ajo
r
ity
o
f
n
ea
r
est
n
eig
h
b
o
r
s
in
th
e
f
ea
tu
r
e
s
p
a
ce
,
wh
ich
ca
n
b
e
ea
s
ily
u
n
d
er
s
to
o
d
an
d
im
p
lem
e
n
ted
,
p
a
r
ticu
lar
ly
f
o
r
s
m
all
d
atasets
.
Ho
wev
er
,
th
ese
tr
a
d
itio
n
al
m
o
d
els
h
av
e
s
o
m
e
d
r
awb
ac
k
s
d
u
e
to
t
h
eir
r
elian
ce
o
n
m
an
u
ally
d
esig
n
e
d
f
ea
tu
r
es,
wh
ich
m
a
y
n
o
t
ac
c
u
r
ately
r
ef
lect
th
e
ac
t
u
al
p
att
er
n
s
o
f
b
r
ain
tu
m
o
r
s
.
T
o
a
d
d
r
ess
th
is
ch
allen
g
e,
r
esear
ch
er
s
h
av
e
d
ev
el
o
p
ed
TL
to
f
in
e
-
tu
n
e
m
o
d
els
tr
ain
ed
o
n
s
im
ilar
task
s
o
r
lar
g
er
d
atasets
f
o
r
b
r
ain
tu
m
o
r
class
if
icatio
n
.
T
h
is
ap
p
r
o
ac
h
h
elp
s
in
a
d
d
r
ess
in
g
th
e
p
r
o
b
l
em
o
f
r
estricte
d
am
o
u
n
ts
o
f
l
ab
eled
d
ata
s
in
ce
it
allo
ws k
n
o
wled
g
e
tr
a
n
s
f
er
ac
r
o
s
s
r
elate
d
d
o
m
ain
s
,
an
d
th
u
s
t
h
e
r
esu
ltin
g
m
o
d
el
p
er
f
o
r
m
s
w
ell.
2
.
2
.
Deep
lea
rning
ba
s
ed
cla
s
s
s
if
ica
t
io
n
DL
h
as
s
ig
n
if
ican
tly
im
p
ac
ted
th
e
m
ed
ical
f
ield
,
p
ar
ticu
lar
ly
in
th
e
class
if
icatio
n
o
f
f
ea
tu
r
es
in
b
r
ain
tu
m
o
r
im
ag
es
[
1
9
]
b
y
a
u
to
m
at
ically
ex
tr
ac
tin
g
f
ea
t
u
r
es
f
r
o
m
th
e
im
ag
e
d
ata.
Un
lik
e
s
tan
d
a
r
d
ML
m
o
d
els
th
at
in
v
o
lv
e
h
an
d
cr
af
te
d
f
ea
tu
r
es,
m
o
s
t
DL
m
o
d
els,
p
ar
ticu
lar
ly
C
NN
s
,
lear
n
h
ier
ar
ch
ical
r
ep
r
esen
tatio
n
s
o
f
th
e
f
ea
tu
r
es
f
r
o
m
t
h
e
r
aw
d
ata,
m
ak
in
g
th
em
i
d
ea
l
f
o
r
im
a
g
e
cl
ass
if
icatio
n
in
g
en
er
al.
T
h
r
o
u
g
h
lay
er
s
o
f
co
n
v
o
l
u
tio
n
al
f
ilter
s
[
20
],
[
2
1
]
.
C
NNs
aim
to
lear
n
a
s
et
o
f
s
p
atial
p
y
r
am
id
s
to
m
o
d
el
th
e
h
ier
ar
ch
ies
p
r
esen
t
in
b
r
ain
MRI
s
ca
n
s
,
d
etec
tin
g
ed
g
es,
tex
tu
r
es,
an
d
h
i
g
h
er
-
le
v
el
s
tr
u
ctu
r
es
in
th
e
s
ca
n
n
ed
im
ag
e.
T
h
ese
lay
er
s
en
ca
p
s
u
late
th
e
im
a
g
e
f
ea
tu
r
e
s
in
a
s
tep
-
by
-
s
tep
m
a
n
n
er
s
o
as
to
e
n
ab
le
th
e
n
etwo
r
k
to
lear
n
c
o
m
p
licated
f
ea
tu
r
es
th
at
ar
e
r
elev
an
t
to
tu
m
o
r
d
if
f
er
e
n
tiatio
n
.
B
y
a
u
to
-
lear
n
in
g
f
r
o
m
v
ast
a
m
o
u
n
ts
o
f
d
ata,
it
h
as
d
em
o
n
s
tr
ated
s
tate
-
of
-
th
e
-
ar
t
p
er
f
o
r
m
an
ce
in
b
r
ain
tu
m
o
r
class
if
icatio
n
,
o
u
tp
er
f
o
r
m
in
g
th
e
tr
ad
itio
n
al
ap
p
r
o
ac
h
in
ter
m
s
o
f
p
r
ec
is
i
o
n
an
d
r
eliab
ilit
y
.
T
h
e
o
th
er
n
o
tab
le
d
ev
elo
p
m
en
t
in
DL
f
o
r
b
r
ain
t
u
m
o
r
class
if
icatio
n
is
TL
,
wh
ich
m
ak
es
u
s
e
o
f
p
r
e
-
tr
ain
e
d
m
o
d
e
ls
o
n
lar
g
e
d
atasets
lik
e
I
m
ag
eNe
t,
f
o
r
in
s
tan
ce
,
VGGN
et,
R
e
s
Net,
E
f
f
icien
tNet,
an
d
o
th
e
r
s
in
th
e
class
if
icatio
n
task
[
2
2
]
.
T
h
is
m
eth
o
d
ca
n
b
e
u
s
ef
u
l
f
o
r
f
u
r
th
er
s
tu
d
ies
s
in
ce
,
b
y
tu
n
i
n
g
th
ese
m
o
d
els
o
n
th
e
b
r
ain
tu
m
o
r
d
atasets
,
th
ey
h
av
e
s
h
o
wn
h
ig
h
ac
cu
r
ac
y
with
a
lim
ited
am
o
u
n
t
o
f
lab
eled
m
e
d
ical
d
ata.
T
h
is
m
a
k
es
th
e
a
p
p
r
o
ac
h
m
o
r
e
s
u
ita
b
le
th
an
m
o
s
t
o
th
er
tr
ain
in
g
ap
p
r
o
ac
h
es
s
in
ce
it
n
o
t
o
n
ly
s
h
o
r
te
n
s
tr
ain
in
g
tim
e
b
u
t
also
im
p
r
o
v
es
th
e
m
o
d
el
’
s
ab
ilit
y
to
g
en
er
alize
o
v
er
d
if
f
er
e
n
t
ty
p
es
o
f
tu
m
o
r
s
an
d
im
ag
in
g
s
ce
n
ar
io
s
.
TL
h
as b
ee
n
esp
ec
ial
ly
h
e
lp
f
u
l
in
ad
d
r
ess
in
g
th
e
p
r
o
b
lem
o
f
s
ca
r
ce
m
ed
ica
l
d
ata,
wh
ich
r
em
ain
s
a
co
m
m
o
n
is
s
u
e
i
n
r
esear
ch
i
n
th
e
s
ec
to
r
.
TL
,
wh
eth
er
th
r
o
u
g
h
th
e
u
s
e
o
f
p
r
e
-
t
r
ain
ed
n
etwo
r
k
s
as
f
ea
tu
r
e
s
u
b
s
tr
ate
s
o
r
f
ea
tu
r
e
tu
n
i
n
g
,
h
as
en
ab
l
ed
DL
s
tr
u
ctu
r
es
to
im
p
r
o
v
e
d
iag
n
o
s
ti
c
p
er
f
o
r
m
a
n
ce
f
r
o
m
b
r
ain
t
u
m
o
r
im
ag
es
at
a
f
aster
p
ac
e
a
n
d
with
g
r
ea
ter
r
eliab
ilit
y
.
I
n
ad
d
itio
n
,
o
th
er
tech
n
o
l
o
g
ies,
in
clu
d
in
g
en
s
em
b
le
lear
n
i
n
g
,
d
ata
au
g
m
en
tatio
n
,
an
d
ex
p
licit
atten
tio
n
m
ec
h
an
is
m
s
,
h
av
e
co
n
tr
i
b
u
te
d
p
o
s
itiv
ely
to
b
etter
class
if
icatio
n
o
f
th
e
b
r
ai
n
tu
m
o
r
u
s
in
g
DL
m
o
d
els
[
2
3
]
.
E
n
s
em
b
le
lear
n
in
g
,
wh
ich
i
n
v
o
lv
es
u
s
in
g
m
o
r
e
th
a
n
o
n
e
m
o
d
el
to
m
a
k
e
a
p
r
ed
ictio
n
,
in
cr
ea
s
es
r
o
b
u
s
tn
ess
an
d
ac
c
u
r
ac
y
,
wh
er
ea
s
d
ata
a
u
g
m
en
tatio
n
ar
tific
ially
i
n
cr
e
ases
th
e
tr
ain
in
g
d
ata
s
et
b
y
ap
p
ly
in
g
tr
an
s
f
o
r
m
s
lik
e
r
o
tatio
n
an
d
s
ca
lin
g
,
th
er
eb
y
r
e
d
u
cin
g
o
v
e
r
f
itti
n
g
.
I
n
c
o
n
tr
ast,
atten
tio
n
m
ec
h
an
is
m
s
b
r
in
g
th
e
f
o
cu
s
o
f
th
e
m
o
d
els
o
n
th
e
p
a
r
ts
o
f
t
h
e
im
ag
e
th
at
ar
e
r
elev
an
t,
s
u
ch
as
th
e
tu
m
o
r
m
ass
,
an
d
h
en
ce
y
ield
b
etter
class
if
icatio
n
r
esu
lts
[
24
].
2.
3
.
T
ra
ns
f
er
l
ea
rning
-
ba
s
ed
cla
s
s
if
ica
t
io
n
Fo
llo
win
g
s
ec
tio
n
d
ec
s
r
ib
e
th
e
in
-
d
ep
th
TL
b
ased
class
if
icatio
n
ap
p
r
o
ac
h
es.
2
.
3
.
1
.
F
ea
t
ure
e
x
t
a
rc
t
i
o
n
Featu
r
e
ex
tr
ac
tio
n
in
b
r
ain
t
u
m
o
r
class
if
icatio
n
f
o
cu
s
es
o
n
ex
p
l
o
r
in
g
n
ec
ess
ar
y
f
ea
t
u
r
e
s
f
r
o
m
a
n
MRI
im
ag
e
wit
h
tr
ain
in
g
o
f
t
h
e
p
r
e
-
tr
ain
e
d
m
o
d
els
f
o
r
d
et
ec
tin
g
s
ig
n
if
ican
t
f
ea
tu
r
es.
T
h
is
m
eth
o
d
lev
er
ag
es
th
e
f
ac
t
th
at
th
e
m
o
d
els
th
at
ar
e
u
s
u
ally
tr
ain
e
d
o
n
lar
g
e
a
n
d
d
iv
e
r
s
e
d
ata
ca
n
b
e
u
s
ed
to
co
n
v
er
t
r
aw
MR
I
d
ata
in
to
h
i
g
h
-
lev
el
f
ea
tu
r
es.
Firstl
y
,
we
ch
o
o
s
e
a
p
r
o
p
er
p
r
e
-
tr
ain
ed
m
o
d
el
lik
e
VGG1
6
o
r
R
esNet
s
in
ce
th
eir
f
ea
tu
r
e
ex
tr
ac
tio
n
a
b
ilit
y
h
as
b
ee
n
wid
ely
ac
k
n
o
wled
g
ed
.
T
h
is
p
r
o
ce
s
s
s
tar
ts
w
ith
f
ee
d
in
g
b
r
ain
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.
1
4
,
No
.
3
,
Dec
em
b
er
20
2
5
:
1
0
0
2
-
1
0
1
4
1006
to
th
is
m
o
d
el
to
g
et
f
ea
tu
r
e
m
ap
s
o
r
em
b
ed
d
in
g
s
f
r
o
m
th
e
h
i
d
d
en
lay
er
s
.
T
h
ese
f
ea
tu
r
e
r
ep
r
ese
n
tatio
n
s
,
wh
ich
co
n
tain
th
e
in
t
r
icate
d
etails
o
f
ev
en
th
e
in
te
r
n
al
s
tr
u
ctu
r
es
o
f
th
e
im
ag
es,
ar
e
th
e
n
u
s
ed
in
th
e
id
en
tific
atio
n
o
f
th
e
ty
p
e
o
f
tu
m
o
r
th
r
o
u
g
h
a
s
ec
o
n
d
class
if
ier
th
at
m
ay
b
e
a
SVM
o
r
a
f
u
lly
co
n
n
ec
ted
n
eu
r
al
n
etwo
r
k
.
Fo
r
in
s
tan
ce
,
in
th
e
VGG1
6
b
r
ain
s
,
r
elev
an
t
f
ea
tu
r
es
ar
e
ex
tr
ac
ted
f
r
o
m
b
r
ain
MRI
s
ca
n
s
an
d
lo
g
is
tic
r
eg
r
ess
io
n
f
o
r
a
class
if
ier
,
m
o
r
e
p
r
ec
is
e
R
esNet
h
ig
h
h
ier
ar
ch
ical
f
ea
t
u
r
es
in
ad
d
i
tio
n
to
u
s
in
g
o
th
er
s
u
cc
ee
d
in
g
ML
f
o
r
class
if
icatio
n
.
T
h
e
m
ajo
r
b
e
n
ef
its
o
f
t
h
is
ap
p
r
o
ac
h
i
n
clu
d
e
th
e
f
ac
t
th
at
p
ass
in
g
t
h
e
f
ea
tu
r
es th
r
o
u
g
h
lay
e
r
s
r
ed
u
ce
s
th
e
am
o
u
n
t o
f
tim
e
th
at
is
tak
en
in
tr
ain
i
n
g
s
in
ce
th
e
im
p
o
r
tan
t f
ea
tu
r
es c
an
b
e
o
b
tain
ed
f
r
o
m
th
e
p
r
e
-
tr
ain
ed
f
ea
tu
r
e
m
a
p
s
.
Ad
d
itio
n
ally
,
t
h
e
lear
n
ed
r
ep
r
esen
tatio
n
s
f
r
o
m
s
u
ch
lar
g
e
d
atasets
p
r
o
v
id
e
b
etter
class
if
icatio
n
ac
cu
r
ac
y
f
o
r
th
e
p
ar
ticu
lar
tas
k
o
f
tu
m
o
r
r
ec
o
g
n
itio
n
.
Nev
e
r
th
eless
,
wh
at
h
as
b
ee
n
e
x
p
lain
ed
ab
o
v
e
ab
o
u
t
t
h
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
p
r
o
ce
s
s
is
n
o
t
f
r
ee
f
r
o
m
ce
r
tain
lim
itatio
n
s
.
T
h
e
s
o
u
r
ce
o
f
d
eg
en
er
atio
n
is
r
o
o
ted
in
th
e
f
ea
tu
r
e
r
ep
r
esen
tatio
n
lim
itatio
n
o
f
th
e
p
r
e
-
tr
ain
e
d
m
o
d
e
l,
wh
ich
m
ay
n
o
t
co
n
tain
s
u
f
f
icien
t
f
ea
tu
r
es to
r
ep
r
esen
t d
if
f
er
e
n
t ty
p
es o
f
b
r
ai
n
tu
m
o
r
s
[
2
5
]
,
[
2
6
]
.
2
.
3
.
2
.
F
ine
t
uning
Fin
e
-
tu
n
in
g
is
an
ess
en
tial
asp
ec
t
o
f
TL
wh
e
r
e
th
e
in
itial
m
o
d
el
is
tr
ain
ed
o
n
a
ce
r
tain
tar
g
et
d
ataset,
lik
e
th
e
im
ag
es
o
f
b
r
ain
tu
m
o
r
s
f
r
o
m
MRI
s
ca
n
s
,
to
im
p
r
o
v
e
its
p
er
f
o
r
m
an
ce
[
2
0
]
in
th
e
n
ew
task
at
h
an
d
[
2
7
]
.
T
h
e
p
r
o
ce
s
s
s
tar
ts
w
ith
t
h
e
ab
ilit
y
to
lo
ad
a
m
o
d
el
th
at
h
as
b
ee
n
tr
ain
ed
with
a
h
u
g
e
d
ataset
o
f
co
m
m
o
n
ob
jects,
s
u
ch
as
I
m
ag
eNe
t,
t
h
at
p
r
o
v
id
es
th
e
m
o
d
el
with
g
o
o
d
f
ea
tu
r
es
to
s
tar
t
with
.
Fo
r
th
is
m
o
d
el
to
b
e
ap
p
lied
in
th
e
class
if
icatio
n
o
f
b
r
ain
tu
m
o
r
s
,
th
e
ea
r
ly
lay
er
s
th
at
g
en
er
alize
f
ea
tu
r
es su
ch
a
s
ed
g
es a
n
d
tex
tu
r
e
ar
e
lef
t
u
n
f
r
o
ze
n
in
o
r
d
er
t
o
r
etain
p
r
e
v
io
u
s
ly
lea
r
n
ed
k
n
o
w
led
g
e.
T
o
f
in
e
-
tu
n
e
th
e
later
l
ay
er
s
o
f
th
e
m
o
d
el,
as
th
o
s
e
lay
er
s
a
r
e
m
o
r
e
p
r
ec
is
e
in
t
h
eir
n
atu
r
e
an
d
d
ir
ec
tly
h
elp
u
s
in
ex
tr
ac
tin
g
r
elev
an
t
f
e
atu
r
es
r
eq
u
ir
ed
f
o
r
th
e
class
if
icatio
n
[
2
8
]
.
T
h
is
is
co
r
r
ec
ted
b
y
r
etr
ai
n
in
g
th
ese
lay
e
r
s
u
s
in
g
th
e
b
r
ain
tu
m
o
r
d
ataset
in
o
r
d
er
to
en
ab
le
th
e
m
o
d
el
to
id
e
n
tify
s
p
ec
if
ic
f
ea
tu
r
es
th
at
ar
e
p
ar
tic
u
lar
ly
r
elate
d
to
d
if
f
er
i
n
g
ty
p
e
s
o
f
tu
m
o
r
s
.
I
n
th
is
p
h
ase,
a
s
m
aller
lear
n
in
g
r
at
e
is
u
s
ed
to
a
v
o
id
lar
g
e
c
h
a
n
g
es
to
th
e
p
r
e
-
tr
ain
in
g
weig
h
ts
as
th
e
m
o
d
el
is
f
in
etu
n
ed
f
o
r
th
e
n
ew
task
wh
i
le
r
etain
in
g
th
e
lear
n
ed
g
en
er
a
lized
f
ea
tu
r
es.
Fin
e
-
tu
n
in
g
h
as
b
ec
o
m
e
p
r
o
v
en
in
a
n
u
m
b
er
o
f
ap
p
licatio
n
s
o
r
u
s
es.
Fo
r
in
s
tan
ce
,
th
e
R
esN
et5
0
m
o
d
el,
wh
ile
ap
p
lied
to
class
if
y
b
r
ain
MRI
im
ag
es
to
co
r
r
esp
o
n
d
in
g
tu
m
o
r
ty
p
es,
h
as
b
ee
n
s
h
o
wn
to
en
jo
y
h
ig
h
e
r
class
if
icatio
n
ac
cu
r
ac
y
co
m
p
a
r
ed
to
o
th
er
m
o
d
els
b
y
u
s
in
g
a
d
ee
p
ar
ch
itectu
r
e
a
n
d
lear
n
ed
f
ea
tu
r
es
to
m
ak
e
th
e
d
if
f
er
en
tial
d
iag
n
o
s
is
[
2
9
]
.
I
n
a
s
im
ilar
m
an
n
er
,
I
n
ce
p
tio
n
V3
h
as
b
ee
n
a
p
p
lied
t
o
b
r
ain
tu
m
o
r
d
atasets
b
ec
au
s
e,
d
u
e
to
its
h
ig
h
ly
s
tack
e
d
lay
er
s
,
it
ca
n
ca
p
tu
r
e
m
u
lti
-
s
ca
le
f
ea
tu
r
es.
So
,
t
h
e
ad
v
an
tag
es
o
f
f
in
e
-
t
u
n
in
g
ar
e
a
p
p
ar
e
n
t
in
g
en
er
al
with
th
e
im
p
r
o
v
e
d
ap
titu
d
es
o
f
th
e
m
o
d
el
to
d
eliv
er
th
e
p
latf
o
r
m
th
e
b
en
ef
it
o
f
s
p
ec
ialized
lear
n
i
n
g
,
wh
er
e
th
e
b
asic
m
o
tiv
a
tio
n
is
to
en
ab
le
th
e
m
o
d
el
to
s
p
ec
if
y
in
th
e
s
alien
t
ch
ar
ac
ter
is
tics
o
f
th
e
b
r
ain
tu
m
o
r
s
,
wh
ich
in
ef
f
ec
t
en
h
an
ce
s
th
e
p
r
o
d
u
ctiv
e
cla
s
s
if
icatio
n
[
3
0
]
.
Als
o
,
it
is
an
ef
f
ec
tiv
e
u
tili
za
tio
n
o
f
la
b
eled
d
ata
as
th
e
f
r
am
ewo
r
k
a
v
o
id
s
th
e
r
eq
u
ir
e
m
en
t
o
f
lar
g
e
d
a
tasets
,
wh
ich
is
m
ad
e
p
o
s
s
ib
le
b
y
p
r
e
-
tr
ain
e
d
m
o
d
els.
Sti
ll,
lik
e
an
y
o
th
er
p
r
o
ce
s
s
,
f
in
e
-
t
u
n
in
g
co
m
es
with
its
o
wn
s
et
o
f
p
r
o
b
lem
s
.
Ov
er
f
itti
n
g
is
a
m
ajo
r
p
r
o
b
lem
f
ac
in
g
th
e
ap
p
licatio
n
o
f
ML
,
esp
ec
ially
wh
en
wo
r
k
in
g
with
s
m
all
d
at
a
s
ets,
s
in
ce
th
e
m
o
d
el
m
ig
h
t
en
d
u
p
lear
n
in
g
th
e
n
o
is
e
r
ath
er
th
a
n
ca
p
tu
r
in
g
t
h
e
u
n
d
er
ly
in
g
p
atter
n
s
.
Fu
r
t
h
er
m
o
r
e
,
f
in
e
-
t
u
n
in
g
m
ay
b
e
a
co
m
p
u
tatio
n
ally
ex
p
en
s
iv
e
p
r
o
ce
s
s
an
d
n
ee
d
s
a
lar
g
e
a
m
o
u
n
t
o
f
r
eso
u
r
ce
s
an
d
tim
e,
esp
ec
ially
wh
en
wo
r
k
in
g
with
lar
g
e
m
o
d
els.
Nev
er
th
eless
,
f
in
e
-
t
u
n
in
g
s
tay
s
th
e
m
ai
n
in
s
tr
u
m
en
t
in
TL
an
d
h
el
p
s
to
p
r
o
g
r
ess
in
th
e
class
if
icatio
n
o
f
b
r
ai
n
tu
m
o
r
s
[
3
1
]
.
2
.
3
.
3
.
E
nd
t
o
end t
ra
ini
ng
E
n
d
-
to
-
e
n
d
tr
ain
in
g
m
ea
n
s
r
eu
s
in
g
p
r
e
-
tr
ain
ed
m
o
d
els
b
y
s
t
ar
tin
g
with
a
b
an
k
o
f
weig
h
ts
lear
n
ed
b
y
tr
ain
in
g
o
n
a
lar
g
e
s
et
o
f
d
ata
co
llected
f
r
o
m
an
im
ag
e
d
ata
b
ase
to
class
if
y
th
e
n
o
v
el
s
et
o
f
b
r
ain
tu
m
o
r
s
an
d
f
in
e
-
tu
n
in
g
th
is
m
o
d
el
o
n
a
n
ew
s
et
o
f
d
ata
[
3
2
]
.
T
h
is
m
eth
o
d
o
lo
g
y
co
m
b
in
es
th
e
id
ea
o
f
u
s
in
g
p
r
e
-
tr
ai
n
e
d
weig
h
ts
ac
q
u
ir
ed
b
y
lear
n
i
n
g
f
r
o
m
a
lar
g
e
d
atab
ase,
wh
ich
i
n
co
r
p
o
r
ates
th
e
g
en
er
al
co
n
ce
p
t
o
f
th
e
im
a
g
es
t
o
lear
n
f
r
o
m
th
e
n
ew
d
ataset,
esp
ec
ially
r
elatin
g
t
o
b
r
ain
tu
m
o
r
s
.
T
h
e
p
r
o
ce
s
s
s
tar
ts
with
th
e
lo
ad
in
g
o
f
th
e
b
ase
m
o
d
el
an
d
th
e
in
itial
weig
h
ts
,
wh
ich
in
clu
d
e
E
f
f
icien
tNet
an
d
Den
s
eNe
t,
am
o
n
g
o
th
er
s
,
wh
ich
ar
e
in
itially
tr
ain
ed
o
n
d
atasets
I
m
ag
eNe
t
[
3
3
]
.
Af
ter
th
at,
tr
ain
in
g
is
p
er
f
o
r
m
ed
o
n
th
e
b
r
ain
t
u
m
o
r
d
ataset,
a
n
d
all
weig
h
ts
o
f
th
e
m
o
d
el
’
s
lay
er
s
ar
e
tu
n
ed
to
ac
h
ie
v
e
th
e
b
est
ac
cu
r
ac
y
in
r
e
p
r
es
en
tin
g
th
e
n
e
w
d
ataset.
T
h
is
all
-
en
co
m
p
ass
in
g
tr
ain
i
n
g
p
r
o
ce
d
u
r
e
h
as
th
e
ad
v
a
n
tag
e
o
f
d
is
till
atio
n
f
r
o
m
ex
te
n
s
iv
e
p
r
e
-
tr
ain
ed
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
an
d
ac
cu
r
ate,
d
o
wn
s
tr
ea
m
twea
k
ed
m
o
d
if
i
ca
tio
n
s
.
So
p
h
is
ticated
m
eth
o
d
s
ar
e
th
e
n
u
s
ed
to
im
p
r
o
v
e
th
e
tr
ain
in
g
r
ates
an
d
q
u
ality
an
d
to
g
u
ar
a
n
tee
co
n
v
er
g
en
ce
to
th
e
f
in
al
s
o
lu
tio
n
.
A
ca
s
e
o
f
u
tili
zin
g
E
f
f
icien
tNet
in
en
d
-
to
-
e
n
d
lea
r
n
in
g
s
h
o
ws
it
h
as
ce
r
tain
b
en
ef
its
b
y
o
p
tim
izin
g
th
e
in
teg
r
atio
n
o
f
p
r
et
r
ain
ed
d
ata
an
d
s
p
ec
if
ic
task
d
ata
ac
q
u
ir
ed
in
th
e
class
if
ica
tio
n
o
f
b
r
ain
tu
m
o
r
s
.
L
ik
ewise,
th
e
Den
s
eNe
t
ar
ch
itectu
r
e
im
p
r
o
v
es
o
n
th
e
a
b
ilit
y
o
f
th
e
n
etwo
r
k
to
class
if
y
th
e
in
p
u
t
im
ag
e
s
in
ce
,
wh
en
tr
ain
ed
en
d
-
to
-
en
d
,
it
d
r
aws
o
n
p
r
e
-
tr
ain
ed
weig
h
ts
to
u
p
d
ate
u
p
o
n
ex
p
o
s
u
r
e
to
n
ew
d
ata
[
3
4
]
.
T
r
ain
i
n
g
f
r
o
m
e
n
d
to
en
d
is
co
m
p
u
tatio
n
ally
ex
p
en
s
iv
e
an
d
tim
e
-
co
n
s
u
m
in
g
,
esp
ec
ially
with
elab
o
r
ate
m
o
d
els an
d
d
ata
b
ases
.
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
A
d
va
n
ce
men
ts
in
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
:
a
s
u
r
ve
y
o
f tra
n
s
fer lea
r
n
in
g
tech
n
iq
u
es
(
S
n
eh
a
l J
a
d
h
a
v
)
1007
2
.
3
.
4
.
M
ulti
-
m
o
da
l t
ra
ns
f
er
l
ea
rning
Usi
n
g
in
f
o
r
m
atio
n
f
r
o
m
d
if
f
er
en
t
m
o
d
es
o
f
im
ag
in
g
lik
e
MRI
an
d
C
T
s
ca
n
s
,
m
u
lti
-
m
o
d
al
TL
im
p
r
o
v
es
th
e
id
en
tific
atio
n
o
f
b
r
ain
tu
m
o
r
s
f
r
o
m
TL
,
wh
ic
h
i
s
ca
p
ab
le
o
f
f
u
n
ctio
n
in
g
ac
r
o
s
s
th
e
d
if
f
er
e
n
t
d
ata
s
ets
[
3
5
]
.
Fig
u
r
e
1
s
h
o
ws
th
e
TL
m
o
d
el
ap
p
r
o
ac
h
.
TL
m
o
d
el
s
lev
er
ag
e
th
e
k
n
o
wled
g
e
g
ain
ed
b
y
a
m
o
d
el
f
r
o
m
p
r
e
-
tr
ain
in
g
o
n
a
lar
g
e
d
ataset
an
d
tr
an
s
f
er
th
is
to
a
s
im
ilar
task
with
lim
ited
d
ata,
th
u
s
allo
win
g
f
aster
co
n
v
er
g
en
ce
a
n
d
im
p
r
o
v
ed
p
er
f
o
r
m
an
ce
.
B
y
r
eu
s
in
g
th
e
lear
n
ed
f
ea
tu
r
es,
TL
m
in
i
m
izes
th
e
n
ee
d
f
o
r
ex
t
en
s
iv
e
tr
ain
in
g
o
n
s
m
aller
d
atasets
an
d
h
en
ce
is
o
f
v
er
y
h
ig
h
v
alu
e
i
n
s
p
ec
ialized
a
p
p
licatio
n
s
s
u
ch
as
m
ed
ical
im
ag
e
class
if
icatio
n
[
3
6
]
.
Fig
u
r
e
1
.
T
r
an
s
f
er
lear
n
in
g
m
o
d
els [
37
]
T
h
is
ap
p
r
o
ac
h
is
m
ea
n
t
to
en
h
an
ce
th
e
m
o
d
el
’
s
s
tab
ilit
y
an
d
p
er
f
o
r
m
an
ce
th
r
o
u
g
h
th
e
ag
g
r
eg
atio
n
o
f
s
y
n
er
g
is
tic
in
f
o
r
m
atio
n
f
r
o
m
d
if
f
er
en
t
s
o
u
r
ce
s
.
Data
in
teg
r
atio
n
co
m
es
f
ir
s
t,
wh
er
e
two
s
ets
o
f
im
ag
es,
MRI
an
d
C
T
s
ca
n
s
,
ar
e
f
u
s
ed
to
o
b
tain
a
m
o
r
e
im
p
r
o
v
e
d
s
et
o
f
d
ata.
T
h
is
in
teg
r
atio
n
o
f
th
e
d
ata
f
r
o
m
th
e
two
m
o
d
alities
m
ay
allo
w
f
o
r
a
b
e
tter
d
escr
ip
tio
n
o
f
t
h
e
tu
m
o
r
in
its
en
tire
ty
an
d
m
ay
e
x
h
ib
it
f
ea
tu
r
es
th
at
wo
u
ld
n
o
t b
e
o
b
s
er
v
e
d
wh
en
u
s
in
g
o
n
ly
o
n
e
m
eth
o
d
.
Nex
t,
th
er
e
is
m
o
d
el
s
elec
tio
n
,
in
wh
ich
m
o
d
els ap
p
r
o
p
r
iate
f
o
r
p
r
o
ce
s
s
in
g
m
u
ltimo
d
al
d
at
a
ar
e
s
elec
ted
.
T
h
ese
m
o
d
els,
wh
i
ch
ar
e
tr
ain
ed
o
n
a
lar
g
e
a
n
d
d
iv
er
s
e
d
ataset,
ar
e
th
en
f
in
e
-
tu
n
ed
f
o
r
h
a
n
d
lin
g
an
d
lear
n
in
g
f
r
o
m
th
e
i
n
teg
r
a
ted
d
ata.
T
h
e
tr
ain
i
n
g
p
h
ase
i
n
v
o
lv
es
eith
er
f
in
e
-
tu
n
in
g
o
r
tr
ai
n
in
g
th
e
m
o
d
el
o
n
th
is
en
r
ich
ed
d
ataset,
en
ab
lin
g
it
to
f
u
n
ctio
n
as
a
class
if
icatio
n
m
o
d
el
th
at
class
if
ies
th
e
v
ar
io
u
s
ty
p
es
o
f
in
f
o
r
m
atio
n
it
r
ec
eiv
es.
So
m
e
o
f
th
e
ex
am
p
les
o
f
th
is
ap
p
r
o
ac
h
ar
e
MRI
an
d
C
T
s
ca
n
s
,
wh
er
e
th
e
i
n
f
o
r
m
atio
n
o
f
b
o
th
m
o
d
alities
in
tr
ai
n
in
g
m
o
d
els
wo
r
k
s
b
etter
i
n
t
u
m
o
r
class
if
icat
io
n
th
an
th
e
u
tili
za
tio
n
o
f
s
in
g
le
m
o
d
ality
in
f
o
r
m
atio
n
.
T
h
e
ad
v
an
tag
es
o
f
m
u
lti
-
m
o
d
al
TL
ar
e
s
ig
n
if
ican
t:
in
ad
d
itio
n
t
o
im
p
r
o
v
i
n
g
t
h
e
m
o
d
el
’
s
ac
cu
r
ac
y
,
th
e
u
tili
za
tio
n
o
f
m
u
ltip
le
d
ata
s
ets
en
ab
les
t
h
e
m
o
d
el
to
an
aly
ze
m
an
y
asp
ec
ts
o
f
tu
m
o
r
s
,
wh
ich
wo
u
ld
o
th
er
wis
e
n
o
t
b
e
p
o
s
s
ib
le
if
o
n
ly
d
ea
lin
g
with
o
n
e
d
ata
s
et
[
3
8
]
.
T
h
is
im
p
r
o
v
es
th
e
s
tab
ilit
y
o
f
th
e
m
o
d
el
b
y
en
a
b
lin
g
th
e
p
r
o
p
o
s
ed
m
o
d
el
to
d
eliv
er
ac
cu
r
ate
d
iag
n
o
s
tics
,
ir
r
esp
ec
tiv
e
o
f
t
h
e
d
ata
ty
p
e
u
n
d
er
an
aly
s
is
.
Ho
wev
er
,
th
er
e
ar
e
ce
r
tain
d
r
awb
ac
k
s
t
h
at
ca
n
n
o
t
b
e
lef
t
u
n
n
o
ticed
.
I
n
ter
-
m
o
d
ality
d
at
a
co
m
p
atib
ilit
y
:
r
elatin
g
to
d
ata
f
r
o
m
d
if
f
er
e
n
t
m
o
d
alities
at
d
if
f
er
en
t
f
o
r
m
ats
an
d
/o
r
r
eso
lu
tio
n
s
,
th
e
p
r
o
b
l
em
ca
n
b
e
q
u
ite
co
m
p
lex
,
e
s
p
ec
ially
s
in
ce
an
aly
tics
h
as
to
m
er
g
e
th
e
d
ata
to
g
eth
er
in
to
o
n
e
m
ea
n
in
g
f
u
l f
o
r
m
at
to
b
e
e
x
p
lo
ited
.
Fu
r
th
er
,
th
e
m
u
lti
-
m
o
d
al
d
ata
ar
e
m
u
c
h
m
o
r
e
c
o
m
p
lex
t
o
m
an
ag
e
a
n
d
p
r
o
ce
s
s
as
co
m
p
ar
ed
ag
ain
s
t
th
e
s
in
g
le
ty
p
e
o
f
d
ata,
as
it
d
e
m
an
d
s
m
u
ch
m
o
r
e
r
eso
u
r
ce
s
an
d
b
etter
ap
p
r
o
ac
h
es
to
h
an
d
le
a
ll
th
e
g
ath
er
ed
v
ar
ieties
o
f
in
f
o
r
m
atio
n
.
Ho
wev
er
,
r
esear
ch
s
h
o
ws
th
at
m
u
lti
-
m
o
d
al
TL
s
ig
n
i
f
ican
tly
ad
v
a
n
ce
s
th
e
f
ield
o
f
b
r
ain
tu
m
o
r
class
if
icatio
n
b
y
h
a
r
n
ess
in
g
th
e
ad
v
a
n
tag
es
o
f
im
ag
in
g
m
o
d
alities
to
cr
ea
te
m
o
r
e
ac
cu
r
ate
an
d
r
eliab
le
d
ia
g
n
o
s
tic
m
o
d
els.
2
.
4
.
Co
m
pa
ra
r
t
iv
e
a
na
ly
s
is
o
f
t
he
bra
in
t
um
o
r
cl
a
s
s
if
ica
t
io
n a
lg
o
rit
hm
s
T
h
is
T
ab
le
2
s
h
o
ws
th
e
co
m
p
ar
ativ
e
a
n
aly
is
o
f
t
h
e
ea
ch
m
eth
o
d
o
lo
g
y
,
em
p
h
a
s
izin
g
th
e
ap
p
r
o
p
r
iaten
ess
o
f
TL
in
a
r
ea
s
s
u
ch
as
b
r
ain
tu
m
o
r
class
if
icatio
n
,
wh
e
r
e
a
n
n
o
tated
m
ed
ic
al
d
ata
is
f
r
eq
u
en
tly
s
ca
r
ce
.
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.
1
4
,
No
.
3
,
Dec
em
b
er
20
2
5
:
1
0
0
2
-
1
0
1
4
1008
T
ab
le
2
.
C
o
m
ap
a
r
ativ
e
an
aly
s
i
s
f
o
r
th
e
ML
,
DL
,
an
d
T
L
f
o
r
b
r
ain
tu
m
o
r
class
if
icatio
n
3.
M
E
T
H
O
DS
T
h
is
s
ec
tio
n
d
escr
ib
es th
e
r
esear
ch
m
eth
o
d
s
u
s
in
g
p
r
e
-
tr
ain
e
d
m
o
d
els f
o
r
b
r
ai
n
tu
m
o
r
class
if
icatio
n
.
3
.
1
.
VG
G
Ne
t
VGGN
et
is
y
et
an
o
th
er
DL
ar
ch
itectu
r
e
d
ev
el
o
p
ed
b
y
th
e
Vis
u
al
Geo
m
etr
y
Gr
o
u
p
f
r
o
m
Ox
f
o
r
d
Un
iv
er
s
ity
[
3
8
]
,
it
h
as
g
ain
e
d
a
lo
t
o
f
p
o
p
u
lar
ity
in
m
a
n
y
im
ag
e
class
if
icatio
n
task
s
,
s
u
ch
as
th
e
m
ed
ical
im
ag
in
g
m
o
d
ality
lik
e
b
r
ain
t
u
m
o
r
class
if
icatio
n
.
T
h
er
e
ar
e
th
r
ee
f
ea
tu
r
es
in
VGGN
et:
T
h
e
n
eu
r
al
n
etwo
r
k
’
s
r
ath
er
s
im
p
le
s
tr
u
ctu
r
e
s
u
g
g
es
ts
th
at
d
ee
p
co
n
v
o
lu
tio
n
al
lay
er
s
,
ar
r
an
g
e
d
o
n
e
af
ter
th
e
o
th
er
,
b
u
ild
th
e
n
e
u
r
al
n
etwo
r
k
;
ea
ch
lay
er
u
s
es
th
r
e
e
f
ilter
s
an
d
R
eL
U
ac
tiv
atio
n
.
T
h
is
ap
p
r
o
ac
h
allo
ws
f
o
r
th
e
en
h
an
ce
m
en
t
o
f
n
etwo
r
k
f
ea
tu
r
es,
f
r
o
m
s
im
p
le
f
ea
tu
r
es
lik
e
ed
g
es
i
n
th
e
im
a
g
es
to
h
i
g
h
-
lev
el
ab
s
tr
ac
t
asp
e
cts.
T
h
e
n
etwo
r
k
’
s
v
ar
ian
ts
ar
e
VGG1
6
an
d
VGG1
9
,
a
n
d
th
eir
n
am
es
ar
e
ass
o
ciate
d
with
th
e
n
u
m
b
er
o
f
lay
er
s
:
1
6
an
d
1
9
co
r
r
esp
o
n
d
in
g
ly
.
Nev
er
th
eless
,
th
e
d
ee
p
n
atu
r
e
o
f
VGGN
et
’
s
ar
ch
itectu
r
e
aid
s
in
th
e
ex
t
r
ac
tio
n
o
f
in
tr
icate
s
p
atial
h
ier
ar
ch
ies,
m
ak
in
g
it o
n
e
o
f
th
e
m
o
s
t
ef
f
ec
tiv
e
ex
is
t
in
g
ap
p
r
o
ac
h
es
f
o
r
im
a
g
e
clas
s
if
icatio
n
,
s
im
ilar
to
I
m
ag
eNe
t.
T
h
e
au
th
o
r
s
n
o
te
th
at
o
n
e
o
f
its
m
ajo
r
d
is
ad
v
an
tag
es
is
th
e
h
ig
h
co
m
p
u
tat
io
n
al
an
d
m
em
o
r
y
co
m
p
lex
ity
g
iv
e
n
b
y
a
d
ee
p
s
tr
u
ctu
r
e
an
d
m
an
y
p
ar
am
ete
r
s
[
3
9
]
.
T
h
is
h
as
led
to
th
e
wid
esp
r
ea
d
u
s
e
o
f
VGGN
et
in
m
ed
ical
im
ag
e
co
n
tex
ts
,
p
ar
ticu
lar
l
y
f
o
r
b
r
ain
tu
m
o
r
class
if
icatio
n
.
TL
allo
ws
f
o
r
th
e
m
o
d
if
icatio
n
o
f
p
r
e
-
t
r
ain
ed
m
o
d
els
lik
e
VGGN
et
an
d
VGG1
6
to
class
if
y
tu
m
o
r
s
u
s
in
g
MRI
d
ata
f
r
o
m
d
atasets
lik
e
I
m
ag
eNe
t.
I
n
th
is
s
ettin
g
,
th
e
in
itial
lay
er
s
o
f
co
n
v
o
lu
tio
n
f
ir
s
t
ex
tr
ac
t
lo
w
-
lev
el
f
ea
tu
r
es
co
m
m
o
n
to
all
th
e
im
ag
es,
an
d
th
en
o
p
tim
ize
th
e
o
th
e
r
lay
er
s
to
d
etec
t
tu
m
o
r
-
s
p
ec
if
ic
f
ea
tu
r
es.
I
n
th
is
s
p
ec
if
ic
f
ield
,
th
e
ap
p
licatio
n
o
f
VGGN
et
p
r
o
v
es
ad
v
an
tag
eo
u
s
,
as
its
d
esig
n
allo
ws
f
o
r
th
e
ex
tr
ac
tio
n
an
d
d
ep
ictio
n
o
f
m
i
n
u
te
d
e
tails
in
im
ag
es,
a
f
ea
tu
r
e
th
at
s
ig
n
if
ic
an
tly
aid
s
in
d
is
tin
g
u
is
h
in
g
b
e
twee
n
g
lio
m
as
an
d
m
en
in
g
io
m
as.
P
a
r
a
me
t
e
r
ML
DL
TL
C
i
t
a
t
i
o
n
s
D
a
t
a
d
e
p
e
n
d
e
n
c
y
R
e
q
u
i
r
e
s f
e
a
t
u
r
e
e
n
g
i
n
e
e
r
i
n
g
;
smal
l
t
o
me
d
i
u
m
-
si
z
e
d
d
a
t
a
se
t
s
a
r
e
o
f
t
e
n
su
f
f
i
c
i
e
n
t
.
H
i
g
h
d
e
p
e
n
d
e
n
c
y
o
n
l
a
r
g
e
d
a
t
a
s
e
t
s
t
o
a
u
t
o
m
a
t
i
c
a
l
l
y
l
e
a
r
n
f
e
a
t
u
r
e
s.
R
e
d
u
c
e
s
d
e
p
e
n
d
e
n
c
y
o
n
l
a
r
g
e
d
a
t
a
se
t
s
b
y
l
e
v
e
r
a
g
i
n
g
p
r
e
-
t
r
a
i
n
e
d
mo
d
e
l
s
.
[
2
9
]
,
[
3
0
]
F
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
M
a
n
u
a
l
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
;
r
e
q
u
i
r
e
s
d
o
ma
i
n
e
x
p
e
r
t
i
se
.
A
u
t
o
ma
t
e
d
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
t
h
r
o
u
g
h
h
i
e
r
a
r
c
h
i
c
a
l
l
a
y
e
r
s
o
f
t
h
e
n
e
t
w
o
r
k
.
U
t
i
l
i
z
e
s fe
a
t
u
r
e
s
l
e
a
r
n
e
d
b
y
p
r
e
-
t
r
a
i
n
e
d
m
o
d
e
l
s fr
o
m
l
a
r
g
e
d
a
t
a
se
t
s,
mi
n
i
mi
z
i
n
g
t
h
e
n
e
e
d
f
o
r
man
u
a
l
e
x
t
r
a
c
t
i
o
n
.
[
3
2
]
,
[
3
3
]
M
o
d
e
l
c
o
m
p
l
e
x
i
t
y
Ty
p
i
c
a
l
l
y
,
l
o
w
e
r
c
o
m
p
l
e
x
i
t
y
(
S
V
M
,
d
e
c
i
si
o
n
t
r
e
e
s
).
H
i
g
h
c
o
m
p
l
e
x
i
t
y
(
C
N
N
s,
R
N
N
s)
,
o
f
t
e
n
i
n
v
o
l
v
i
n
g
mi
l
l
i
o
n
s
o
f
p
a
r
a
me
t
e
r
s.
M
o
d
e
r
a
t
e
c
o
m
p
l
e
x
i
t
y
;
b
u
i
l
d
s
o
n
p
r
e
-
t
r
a
i
n
e
d
n
e
t
w
o
r
k
s
l
i
k
e
V
G
G
,
R
e
sN
e
t
,
I
n
c
e
p
t
i
o
n
,
e
t
c
.
[
2
5
]
,
[
4
2
]
Tr
a
i
n
i
n
g
t
i
me
S
h
o
r
t
e
r
t
r
a
i
n
i
n
g
t
i
m
e
,
e
sp
e
c
i
a
l
l
y
w
i
t
h
sma
l
l
e
r
d
a
t
a
se
t
s.
Lo
n
g
t
r
a
i
n
i
n
g
t
i
me
d
u
e
t
o
t
h
e
l
a
r
g
e
n
u
m
b
e
r
o
f
p
a
r
a
m
e
t
e
r
s a
n
d
d
e
e
p
l
a
y
e
r
s.
S
i
g
n
i
f
i
c
a
n
t
l
y
r
e
d
u
c
e
d
t
r
a
i
n
i
n
g
t
i
m
e
a
s
o
n
l
y
t
h
e
f
i
n
a
l
l
a
y
e
r
s
o
r
sp
e
c
i
f
i
c
p
a
r
t
s
o
f
t
h
e
mo
d
e
l
a
r
e
f
i
n
e
-
t
u
n
e
d
.
[
1
7
]
P
e
r
f
o
r
ma
n
c
e
o
n
smal
l
d
a
t
a
se
t
s
Te
n
d
s
t
o
p
e
r
f
o
r
m w
e
l
l
w
i
t
h
t
a
i
l
o
r
e
d
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
.
P
e
r
f
o
r
ms
p
o
o
r
l
y
u
n
l
e
ss
a
u
g
me
n
t
e
d
w
i
t
h
sy
n
t
h
e
t
i
c
d
a
t
a
.
S
u
p
e
r
i
o
r
p
e
r
f
o
r
ma
n
c
e
b
y
t
r
a
n
sf
e
r
r
i
n
g
l
e
a
r
n
e
d
f
e
a
t
u
r
e
s,
f
i
n
e
-
t
u
n
i
n
g
m
o
d
e
l
s
f
o
r
sm
a
l
l
,
d
o
ma
i
n
-
sp
e
c
i
f
i
c
d
a
t
a
se
t
s.
[
2
6
]
,
[
4
3
]
A
c
c
u
r
a
c
y
M
o
d
e
r
a
t
e
a
c
c
u
r
a
c
y
,
d
e
p
e
n
d
s
h
e
a
v
i
l
y
o
n
f
e
a
t
u
r
e
e
n
g
i
n
e
e
r
i
n
g
q
u
a
l
i
t
y
.
H
i
g
h
a
c
c
u
r
a
c
y
w
h
e
n
t
r
a
i
n
e
d
o
n
l
a
r
g
e
,
d
i
v
e
r
se
d
a
t
a
se
t
s.
V
e
r
y
h
i
g
h
a
c
c
u
r
a
c
y
,
e
s
p
e
c
i
a
l
l
y
i
n
med
i
c
a
l
i
ma
g
i
n
g
t
a
s
k
s,
a
s
i
t
b
e
n
e
f
i
t
s fr
o
m
b
o
t
h
d
e
e
p
f
e
a
t
u
r
e
s
a
n
d
d
o
mai
n
-
s
p
e
c
i
f
i
c
t
u
n
i
n
g
.
[
2
9
]
C
o
m
p
u
t
a
t
i
o
n
a
l
r
e
so
u
r
c
e
s
R
e
q
u
i
r
e
s m
o
d
e
r
a
t
e
c
o
m
p
u
t
a
t
i
o
n
a
l
p
o
w
e
r
.
H
i
g
h
c
o
m
p
u
t
a
t
i
o
n
a
l
d
e
m
a
n
d
f
o
r
t
r
a
i
n
i
n
g
l
a
r
g
e
m
o
d
e
l
s
f
r
o
m
scrat
c
h
.
M
o
d
e
r
a
t
e
,
s
i
n
c
e
p
r
e
-
t
r
a
i
n
e
d
mo
d
e
l
s
r
e
d
u
c
e
t
h
e
c
o
mp
u
t
a
t
i
o
n
a
l
l
o
a
d
.
[
3
1
]
,
[
3
2
]
I
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
R
e
l
a
t
i
v
e
l
y
e
a
s
i
e
r
t
o
i
n
t
e
r
p
r
e
t
mo
d
e
l
s
(
e
.
g
.
,
d
e
c
i
si
o
n
t
r
e
e
s
,
S
V
M
)
.
H
a
r
d
e
r
t
o
i
n
t
e
r
p
r
e
t
d
u
e
t
o
d
e
e
p
l
a
y
e
r
s
a
n
d
c
o
m
p
l
e
x
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
.
M
o
d
e
r
a
t
e
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
;
f
e
a
t
u
r
e
s
a
r
e
p
r
e
-
l
e
a
r
n
e
d
,
b
u
t
so
m
e
l
a
y
e
r
s
c
a
n
b
e
f
i
n
e
-
t
u
n
e
d
f
o
r
t
a
s
k
-
sp
e
c
i
f
i
c
u
n
d
e
r
s
t
a
n
d
i
n
g
.
[
3
1
]
,
[
3
2
]
G
e
n
e
r
a
l
i
z
a
t
i
o
n
Te
n
d
s
t
o
o
v
e
r
f
i
t
o
n
sm
a
l
l
d
a
t
a
se
t
s w
i
t
h
o
u
t
r
o
b
u
st
r
e
g
u
l
a
r
i
z
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s.
S
t
r
o
n
g
g
e
n
e
r
a
l
i
z
a
t
i
o
n
w
h
e
n
t
r
a
i
n
e
d
o
n
l
a
r
g
e
d
a
t
a
se
t
s,
b
u
t
r
i
s
k
s
o
v
e
r
f
i
t
t
i
n
g
o
n
sma
l
l
d
a
t
a
se
t
s.
S
t
r
o
n
g
g
e
n
e
r
a
l
i
z
a
t
i
o
n
;
b
a
l
a
n
c
e
s
b
e
t
w
e
e
n
l
a
r
g
e
-
sc
a
l
e
p
r
e
-
t
r
a
i
n
e
d
mo
d
e
l
s
a
n
d
t
a
sk
-
s
p
e
c
i
f
i
c
f
i
n
e
-
t
u
n
i
n
g
.
[
3
6
]
Ex
a
m
p
l
e
s
o
f
t
e
c
h
n
i
q
u
e
s/
m
o
d
e
l
s
S
V
M
,
r
a
n
d
o
m
f
o
r
e
st
,
k
-
N
N
,
N
a
i
v
e
B
a
y
e
s
.
C
N
N
s,
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s (RNN
s)
.
V
G
G
1
6
,
R
e
sN
e
t
,
I
n
c
e
p
t
i
o
n
N
e
t
,
Ef
f
i
c
i
e
n
t
N
e
t
w
i
t
h
f
i
n
e
-
t
u
n
i
n
g
o
r
e
n
d
-
to
-
e
n
d
t
r
a
i
n
i
n
g
.
[
3
1
]
,
[
2
9
]
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
A
d
va
n
ce
men
ts
in
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
:
a
s
u
r
ve
y
o
f tra
n
s
fer lea
r
n
in
g
tech
n
iq
u
es
(
S
n
eh
a
l J
a
d
h
a
v
)
1009
3
.
2
.
ResNet
Mic
r
o
s
o
f
t
R
esear
ch
p
r
o
p
o
s
ed
th
e
r
esid
u
al
n
etwo
r
k
(
R
esNet
)
an
d
p
u
b
lis
h
ed
‘
Dee
p
R
esid
u
a
l
Nets
f
o
r
I
m
ag
e
R
ec
o
g
n
itio
n
’
,
wh
ich
s
ig
n
if
ican
tly
ex
p
an
d
e
d
th
e
p
o
te
n
tial
o
f
DL
b
y
r
eso
lv
in
g
th
e
in
h
er
en
t
v
an
is
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
in
tr
ain
in
g
e
x
tr
em
ely
d
ee
p
n
e
u
r
al
n
etwo
r
k
s
[
4
0
]
.
T
h
e
k
e
y
f
ea
tu
r
e
o
f
R
e
s
Net
is
it
s
r
esid
u
al
lear
n
in
g
,
wh
e
r
e
th
e
c
o
n
ce
p
t o
f
s
h
o
r
tcu
t
co
n
n
ec
tio
n
s
,
also
k
n
o
wn
as
“
s
k
ip
co
n
n
ec
tio
n
s
,
”
all
o
ws
th
e
n
etwo
r
k
t
o
tr
av
er
s
e
o
n
e
o
r
m
o
r
e
lay
er
s
at
a
tim
e.
T
h
is
d
esig
n
allo
ws
f
o
r
th
e
tr
ain
i
n
g
o
f
n
etwo
r
k
s
th
at
co
n
tain
h
u
n
d
r
ed
s
an
d
p
o
s
s
ib
ly
th
o
u
s
an
d
s
o
f
lay
er
s
with
o
u
t
e
x
p
er
ien
ci
n
g
t
h
e
e
ar
lier
ca
s
es
o
f
d
eg
r
ad
atio
n
in
th
e
p
er
f
o
r
m
a
n
ce
o
f
d
ee
p
ar
c
h
itectu
r
es.
R
esNet
in
tr
o
d
u
ce
d
s
ev
er
al
t
y
p
es
o
f
ar
c
h
itectu
r
es:
R
esNet1
8
,
R
esNet
3
4
,
R
esNet5
0
,
a
n
d
R
esNet1
0
1
,
wh
er
e
th
e
n
u
m
b
er
s
y
m
b
o
lizes
th
e
d
ep
th
o
f
t
h
e
n
etwo
r
k
.
T
h
e
d
ee
p
er
n
etwo
r
k
s
,
s
u
ch
as
R
esNet5
0
an
d
R
esNet1
0
1
,
ar
e
m
o
r
e
a
p
p
r
o
p
r
iate
f
o
r
s
u
c
h
task
s
b
e
ca
u
s
e
ea
ch
lay
er
is
ab
le
to
lear
n
an
ab
s
tr
ac
t
r
ep
r
esen
tatio
n
o
f
th
e
in
p
u
t
im
ag
e
[
4
1
]
.
No
wad
ay
s
,
R
esNet
h
as
ac
h
iev
ed
s
ig
n
if
ican
t
s
u
cc
ess
in
b
en
ch
m
ar
k
s
lik
e
I
m
ag
eNe
t,
s
u
r
p
ass
in
g
e
v
e
n
its
s
u
p
er
io
r
p
r
e
v
io
u
s
m
o
d
el,
VGGN
et.
I
n
th
e
m
ed
ical
im
a
g
in
g
s
ec
to
r
,
R
esNe
t
h
as
tr
an
s
f
o
r
m
ed
in
t
o
a
r
o
b
u
s
t
to
o
l
wh
en
it
co
m
es
to
task
s
s
u
ch
as
th
e
class
if
icat
io
n
o
f
b
r
ain
tu
m
o
r
s
[
4
2
]
.
TL
s
av
es
r
esear
ch
er
s
tim
e
an
d
r
eso
u
r
ce
s
b
y
en
ab
lin
g
th
em
to
u
tili
ze
lar
g
e
p
r
e
-
tr
ain
ed
m
o
d
els
lik
e
R
esNe
t
f
r
o
m
I
m
ag
eNe
t
a
n
d
r
ef
in
e
th
e
m
o
n
a
s
m
aller
d
ataset
o
f
M
R
I
b
r
ain
s
ca
n
s
,
f
o
r
in
s
tan
ce
.
B
ec
au
s
e
R
esNe
t
h
as
r
esid
u
al
co
n
n
ec
tio
n
s
,
it
is
p
o
s
s
ib
le
to
g
et
h
ig
h
-
lev
el
an
d
m
u
lti
-
lev
el
f
ea
tu
r
e
r
e
p
r
esen
ta
tio
n
s
o
f
th
e
s
h
a
p
e,
tex
tu
r
e,
an
d
ed
g
es
o
f
tu
m
o
r
s
.
T
h
is
m
ak
es
R
e
s
Net
a
g
o
o
d
to
o
l
f
o
r
f
in
d
in
g
an
d
class
if
y
in
g
b
r
ain
tu
m
o
r
s
.
Ad
d
itio
n
ally
,
R
esNet
p
o
s
s
e
s
s
es
a
r
o
b
u
s
t
f
ea
tu
r
e
ex
tr
ac
tio
n
ca
p
ab
ilit
y
,
r
en
d
er
in
g
it
id
ea
l
f
o
r
in
teg
r
atio
n
in
to
m
u
lti
-
lev
el
s
y
s
tem
s
,
allo
win
g
f
o
r
th
e
s
u
b
s
eq
u
e
n
t
ap
p
licatio
n
o
f
o
th
er
tech
n
iq
u
es
s
u
ch
as
s
eg
m
en
tatio
n
o
r
alter
n
ativ
e
class
if
ier
ar
ch
itect
u
r
es.
Fu
r
th
er
m
o
r
e,
s
tu
d
ies
h
av
e
s
h
o
wn
th
e
m
o
d
el
’
s
u
s
ef
u
ln
ess
n
o
t
o
n
ly
in
class
if
icatio
n
b
u
t
also
i
n
tu
m
o
r
s
eg
m
en
tatio
n
,
wh
e
r
e
p
r
ec
is
e
tu
m
o
r
m
ar
g
in
d
elin
ea
ti
o
n
is
cr
u
cial
f
o
r
tr
ea
tm
e
n
t
p
lan
n
in
g
.
3
.
3
.
I
ncept
io
n
n
et
wo
rk
T
h
e
in
ce
p
tio
n
n
etwo
r
k
ca
lled
Go
o
g
L
eNe
t
was
u
n
v
eile
d
b
y
Go
o
g
le
in
th
e
I
L
SV
R
C
2
0
1
4
co
m
p
etitio
n
,
wh
ich
ch
a
n
g
ed
t
h
e
way
h
o
w
to
d
esig
n
an
ef
f
icien
t
DL
n
etwo
r
k
.
T
h
e
I
n
ce
p
tio
n
m
o
d
u
le
is
th
e
m
o
s
t
in
n
o
v
ativ
e
p
ar
t
o
f
th
is
n
etwo
r
k
.
I
t
co
m
b
in
es
m
u
ltip
le
s
ca
le
f
ea
tu
r
e
m
ap
s
th
r
o
u
g
h
th
e
s
am
e
lay
er
b
y
u
s
in
g
co
n
v
o
lu
tio
n
an
d
m
a
x
p
o
o
lin
g
with
f
ilter
s
o
f
d
if
f
er
e
n
t
s
izes
(
1
×1
,
3
×3
,
5
×5
)
.
T
h
is
m
eth
o
d
o
f
e
x
tr
ac
tin
g
m
u
lti
-
s
ca
led
f
ea
tu
r
es
in
a
m
u
lti
-
b
r
an
ch
m
a
n
n
er
is
ef
f
icien
t
f
o
r
co
m
p
u
tatio
n
an
d
ca
p
t
u
r
es
all
ty
p
es
o
f
f
ea
tu
r
es,
f
r
o
m
th
e
f
ir
s
t
lev
el
to
th
e
s
ec
o
n
d
lev
el
p
atter
n
s
o
f
th
e
i
n
p
u
t
im
ag
e.
An
o
th
er
s
ig
n
if
ican
t
co
n
tr
ib
u
tio
n
to
th
e
I
n
ce
p
tio
n
ar
ch
itectu
r
e
is
th
e
a
u
th
o
r
s
’
u
s
e
o
f
1
×
1
c
o
n
v
o
lu
tio
n
s
to
r
e
d
u
ce
co
m
p
u
tatio
n
al
c
o
m
p
lex
ity
,
r
esu
ltin
g
in
a
d
ec
r
ea
s
e
in
th
e
n
u
m
b
er
o
f
p
ar
a
m
eter
s
with
o
u
t
co
m
p
r
o
m
is
in
g
ac
cu
r
ac
y
.
I
n
th
is
co
n
tex
t,
th
e
n
etwo
r
k
h
as
u
n
d
er
g
o
n
e
n
u
m
er
o
u
s
iter
atio
n
s
s
u
ch
as
I
n
ce
p
tio
n
-
v
3
an
d
I
n
ce
p
tio
n
-
v
4
,
in
co
r
p
o
r
atin
g
m
in
o
r
m
o
d
i
f
icatio
n
s
to
th
e
o
r
ig
in
al
d
esig
n
,
s
u
c
h
as
th
e
ad
d
itio
n
o
f
n
ew,
m
o
r
e
ef
f
ici
en
t
lay
er
s
an
d
im
p
r
o
v
ed
o
p
ti
m
izatio
n
tech
n
iq
u
es
[
4
3
]
.
T
h
e
n
etwo
r
k
’
s
ex
ten
t
an
d
co
n
n
ec
tiv
ity
allo
w
it
to
ca
p
tu
r
e
h
ig
h
-
lev
el
s
im
p
lis
tic
f
ea
tu
r
es
o
f
im
ag
es,
an
d
its
h
ier
ar
ch
ical
s
tr
u
ctu
r
e
m
a
k
es
th
e
m
o
d
el
h
ig
h
ly
u
s
ef
u
l
f
o
r
im
ag
e
class
if
icatio
n
.
I
n
m
ed
ical
im
ag
in
g
,
ap
p
licati
o
n
o
f
th
e
I
n
ce
p
tio
n
n
etwo
r
k
h
as
b
ee
n
p
ar
ticu
l
ar
ly
im
p
o
r
ta
n
t
in
task
s
s
u
ch
as
class
if
y
in
g
o
r
s
eg
m
en
tin
g
b
r
ai
n
tu
m
o
r
s
.
T
h
a
t
is
wh
y
TL
h
as
m
ad
e
it
p
o
s
s
ib
le
to
f
in
e
-
tu
n
e
I
n
ce
p
tio
n
m
o
d
els,
f
o
r
e
x
am
p
le,
I
n
ce
p
tio
n
-
v
3
,
f
o
r
m
ed
ical
task
s
with
th
e
h
elp
o
f
f
ewe
r
s
am
p
les,
MRIs,
o
r
C
T
s
ca
n
s
o
f
b
r
ain
tu
m
o
r
s
[
4
4
]
.
T
h
e
n
etwo
r
k
’
s
ab
ilit
y
to
ex
t
r
ac
t
f
ea
tu
r
es
at
m
u
ltip
le
s
ca
les
is
esp
ec
ially
b
en
ef
icial
in
m
ed
ical
ap
p
licatio
n
s
s
u
ch
as
im
ag
e
d
iag
n
o
s
tics
,
in
wh
ich
tu
m
o
r
s
m
ay
b
e
o
f
r
ad
i
ca
lly
d
if
f
er
e
n
t
d
im
e
n
s
io
n
s
.
T
h
e
I
n
ce
p
tio
n
b
r
o
o
k
s
o
f
th
e
n
etwo
r
k
ef
f
icien
tly
ca
p
tu
r
e
th
ese
v
ar
iatio
n
s
,
en
ab
li
n
g
ac
cu
r
ate
id
en
tific
atio
n
an
d
class
if
icatio
n
o
f
tu
m
o
r
s
.
Fo
r
in
s
tan
ce
,
in
ce
p
ti
o
n
n
etwo
r
k
s
ca
n
d
is
tin
g
u
is
h
b
etwe
en
d
if
f
er
en
t
ty
p
es
o
f
b
r
ain
tu
m
o
r
s
s
u
ch
as
g
lio
m
as
an
d
m
en
n
in
g
io
m
as,
wh
ich
s
h
ar
e
m
a
n
y
s
im
ilar
ities
b
u
t
o
n
ly
s
lig
h
tly
d
if
f
e
r
in
t
h
e
ir
ar
ch
itectu
r
al
a
n
d
tex
tu
r
al
f
ea
tu
r
es a
cr
o
s
s
ce
r
tain
s
ca
les an
d
m
u
ltis
ca
le
n
etwo
r
k
s
.
3
.
4
.
E
f
f
iecie
ntN
et
T
h
e
n
ew
m
o
d
el
th
at
Go
o
g
le
c
r
ea
ted
is
ca
lled
E
f
f
icien
tNet.
I
t
is
th
e
n
ex
t
s
tep
in
th
e
d
e
v
el
o
p
m
en
t
o
f
C
NN
ar
ch
itectu
r
es
an
d
m
ak
es
th
e
m
o
d
el
m
u
ch
b
etter
at
f
in
d
in
g
t
h
e
b
est
b
alan
ce
b
etwe
e
n
ac
cu
r
ac
y
a
n
d
n
et
co
m
p
u
tatio
n
al
co
s
t.
T
h
e
p
r
im
ar
y
im
p
r
o
v
em
en
t
o
f
th
e
E
f
f
ic
ien
tNet
m
o
d
el
is
th
at
it
u
s
e
s
co
m
p
o
u
n
d
s
ca
lin
g
,
wh
ich
m
ea
n
s
th
at
o
n
e
s
ca
les
d
ep
th
,
wid
th
,
a
n
d
r
e
s
o
lu
tio
n
at
o
n
ce
.
T
h
e
m
ain
f
ea
tu
r
e
th
at
d
if
f
er
en
tiates
E
f
f
icien
tNet
f
r
o
m
th
e
t
r
ad
itio
n
al
m
o
d
els
th
at
ar
e
s
ca
led
al
o
n
g
o
n
e
o
f
t
h
e
d
im
e
n
s
io
n
s
i
s
th
at
E
f
f
icien
tNet
s
ca
les
al
l
th
r
ee
d
im
en
s
io
n
s
o
f
th
e
m
o
d
el
p
r
o
p
o
r
tio
n
all
y
,
wh
ich
allo
ws
to
in
cr
ea
s
e
th
e
m
o
d
el
’
s
ef
f
icien
cy
wh
ile
in
cr
ea
s
in
g
its
p
er
f
o
r
m
an
ce
.
B
y
u
s
in
g
th
is
ap
p
r
o
ac
h
,
we
ca
n
cr
ea
te
m
o
d
els
th
at
ar
e
s
i
g
n
if
ican
tly
s
m
aller
an
d
f
aster
th
an
s
o
m
e
o
f
th
e
ex
i
s
tin
g
ar
ch
itectu
r
es,
s
u
ch
as R
esNet
an
d
i
n
ce
p
tio
n
.
T
h
e
ef
f
ici
en
t f
am
ily
co
n
tain
s
th
e
n
etwo
r
k
s
o
f
B
0
,
B
1
,
B
2
,
B
3
,
B
4
,
B
5
,
B
6
,
an
d
B
7
,
wh
er
e
th
e
n
u
m
b
er
af
ter
t
h
e
letter
B
r
ep
r
esen
ts
th
e
p
o
wer
an
d
s
ize
o
f
th
e
n
etwo
r
k
[
3
9
]
.
T
h
is
ef
f
icien
cy
co
m
es
f
r
o
m
m
o
b
ile
in
v
er
ted
b
o
ttlen
ec
k
co
n
v
o
lu
tio
n
(
MBC
o
n
v
)
an
d
s
q
u
ee
ze
-
an
d
-
e
x
citatio
n
n
etwo
r
k
s
(
SE
b
lo
ck
s
)
th
at
ad
d
f
u
r
th
e
r
en
h
an
ce
m
en
t
in
co
m
p
u
tatio
n
al
u
tili
za
tio
n
.
T
h
is
ar
ch
itectu
r
e
n
o
t
o
n
ly
a
p
p
lies
to
g
e
n
er
al
im
a
g
e
class
if
icatio
n
b
u
t
is
also
esp
ec
ially
u
s
ef
u
l
f
o
r
th
o
s
e
ap
p
licatio
n
s
th
at
r
e
q
u
ir
e
h
ig
h
ac
c
u
r
ac
y
b
u
t
ca
n
b
e
c
o
m
p
u
ted
in
a
lim
ited
way
.
Am
o
n
g
th
e
ap
p
licatio
n
s
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.
1
4
,
No
.
3
,
Dec
em
b
er
20
2
5
:
1
0
0
2
-
1
0
1
4
1010
o
f
m
e
d
ical
im
ag
i
n
g
,
it
h
as
b
ee
n
f
o
u
n
d
v
alu
ab
le
in
a
r
ea
s
s
u
c
h
as
b
r
ain
tu
m
o
r
class
if
icatio
n
an
d
s
eg
m
en
tatio
n
.
Giv
en
th
e
s
m
all
d
atasets
an
d
lim
ited
co
m
p
u
tatio
n
al
p
o
we
r
,
th
is
s
u
g
g
ests
th
at
Dee
p
C
T
ca
n
s
ca
le
p
ar
ticu
lar
ly
well
f
o
r
m
ed
ical
ap
p
licatio
n
s
[
4
5
]
.
O
n
th
e
o
t
h
er
h
a
n
d
,
TL
s
ch
em
es
ca
n
f
in
e
-
t
u
n
e
E
f
f
icien
tNet
m
o
d
el
tr
ain
in
g
u
s
in
g
s
m
aller
m
ed
ical
im
ag
i
n
g
s
ets
lik
e
MRIs
o
f
b
r
ain
tu
m
o
r
s
.
MRI
im
ag
es
av
ailab
l
e
in
th
e
d
ataset
ar
e
p
r
ep
r
o
ce
s
s
ed
as
s
h
o
wn
in
Fig
u
r
e
2
.
Du
e
to
t
h
e
co
m
p
o
u
n
d
s
ca
lin
g
o
f
E
f
f
icien
tNet,
it
is
ca
p
a
b
le
o
f
id
e
n
tify
in
g
s
m
all
d
etails
o
f
tu
m
o
r
s
,
s
u
ch
as
th
eir
s
h
a
p
es,
tex
tu
r
e
,
an
d
co
n
tr
ast,
wh
ic
h
en
h
an
ce
s
its
d
iag
n
o
s
tic
ab
ilit
y
.
R
esear
ch
er
s
d
is
co
v
er
ed
t
h
at
E
f
f
icien
tNet
o
u
tp
e
r
f
o
r
m
s
o
th
er
ar
ch
itectu
r
es,
s
p
ec
if
ically
VG
GNe
t
an
d
R
esNet,
in
ac
cu
r
ately
class
if
y
in
g
b
r
ai
n
tu
m
o
r
s
s
u
ch
as
g
lio
m
as,
m
en
in
g
io
m
as,
an
d
p
itu
itar
y
t
u
m
o
r
s
,
b
y
s
ev
e
r
al
p
er
ce
n
t.
Fig
u
r
e
2
.
Step
s
f
o
r
MRI
r
esu
lts
d
ataset
b
y
[
46
]
3
.
5
.
Co
m
pa
ra
t
iv
e
a
na
ly
s
is
o
f
t
he
pre
-
t
ra
ined m
o
dels
f
o
r
bra
in t
um
o
r
cla
s
s
if
ica
t
io
n
W
ith
th
e
r
ap
id
d
ev
elo
p
m
e
n
t o
f
im
ag
e
p
r
o
ce
s
s
in
g
an
d
ar
tific
ial
in
tellig
en
ce
,
th
e
d
o
o
r
h
as b
ee
n
th
r
o
wn
wid
e
o
p
en
to
v
a
r
io
u
s
au
to
m
ated
s
y
s
tem
s
f
o
r
th
e
class
if
icati
o
n
o
f
b
r
ai
n
tu
m
o
r
s
.
T
h
u
s
,
t
h
e
r
eq
u
ir
em
e
n
t
to
ass
is
t
th
e
r
ad
io
lo
g
is
t
in
d
iag
n
o
s
is
is
u
s
ef
u
l
in
m
u
ch
f
aster
an
d
ac
cu
r
ate
m
an
n
e
r
.
T
ec
h
n
iq
u
es
s
u
ch
as
VGG1
6
,
R
esNet,
I
n
ce
p
tio
n
Net,
an
d
E
f
f
icien
tNet
h
av
e
e
m
er
g
e
d
as
s
o
m
e
o
f
th
e
s
tr
o
n
g
d
ee
p
m
o
d
els
f
o
r
im
ag
e
r
ec
o
g
n
itio
n
task
s
.
T
ab
le
3
s
h
o
ws
th
e
C
o
m
p
ar
is
o
n
o
f
th
e
m
o
s
t
p
o
p
u
lar
p
r
e
-
tr
ain
e
d
b
r
ain
tu
m
o
r
class
if
icatio
n
m
o
d
els o
n
a
r
ch
itectu
r
e,
c
o
m
p
l
ex
ity
,
ac
cu
r
ac
y
,
tr
an
s
f
er
a
b
ilit
y
,
an
d
s
m
all
d
ataset
p
er
f
o
r
m
an
c
e
.
T
ab
le
3
.
C
o
m
ap
a
r
ativ
e
an
aly
s
i
s
o
f
th
e
p
r
et
r
ain
ed
m
o
d
els
M
o
d
e
l
n
a
m
e
A
r
c
h
i
t
e
c
t
u
re
M
o
d
e
l
c
o
m
p
l
e
x
i
t
y
P
e
r
f
o
r
ma
n
c
e
o
n
s
mal
l
d
a
t
a
se
t
s
A
c
c
u
r
a
c
y
i
n
b
r
a
i
n
t
u
m
o
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
C
i
t
a
t
i
o
n
s
V
G
G
1
6
16
-
l
a
y
e
r
d
e
e
p
C
N
N
w
i
t
h
3
×
3
c
o
n
v
o
l
u
t
i
o
n
f
i
l
t
e
r
s.
M
o
d
e
r
a
t
e
;
1
3
8
mi
l
l
i
o
n
p
a
r
a
m
e
t
e
r
s.
P
e
r
f
o
r
ms w
e
l
l
w
i
t
h
f
i
n
e
-
t
u
n
i
n
g
,
b
u
t
r
e
q
u
i
r
e
s
d
a
t
a
a
u
g
me
n
t
a
t
i
o
n
.
H
i
g
h
a
c
c
u
r
a
c
y
w
h
e
n
f
i
n
e
-
t
u
n
e
d
f
o
r
me
d
i
c
a
l
i
ma
g
i
n
g
t
a
sk
s
.
[
3
9
]
,
[
4
7
]
R
e
sN
e
t
R
e
si
d
u
a
l
N
e
t
w
o
r
k
w
i
t
h
s
k
i
p
c
o
n
n
e
c
t
i
o
n
s
t
o
a
v
o
i
d
v
a
n
i
s
h
i
n
g
g
r
a
d
i
e
n
t
s (R
e
sN
e
t
-
5
0
,
R
e
sN
e
t
-
1
0
1
)
.
H
i
g
h
;
2
5
-
4
4
m
i
l
l
i
o
n
p
a
r
a
m
e
t
e
r
s (R
e
sN
e
t
-
5
0
/
1
0
1
)
.
S
t
r
o
n
g
p
e
r
f
o
r
ma
n
c
e
o
n
sma
l
l
d
a
t
a
set
s
w
i
t
h
f
i
n
e
-
t
u
n
i
n
g
.
H
i
g
h
;
c
o
n
s
i
st
e
n
t
l
y
p
e
r
f
o
r
ms
w
e
l
l
o
n
me
d
i
c
a
l
i
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
.
[
4
1
]
,
[
4
8
]
,
[
49
]
I
n
c
e
p
t
i
o
n
N
e
t
w
o
r
k
I
n
c
e
p
t
i
o
n
mo
d
u
l
e
s
w
i
t
h
m
u
l
t
i
p
l
e
c
o
n
v
o
l
u
t
i
o
n
f
i
l
t
e
r
s
a
t
d
i
f
f
e
r
e
n
t
si
z
e
s
.
M
o
d
e
r
a
t
e
t
o
h
i
g
h
;
~
2
3
mi
l
l
i
o
n
p
a
r
a
m
e
t
e
r
s
(
I
n
c
e
p
t
i
o
n
V
3
)
.
G
o
o
d
p
e
r
f
o
r
ma
n
c
e
o
n
smal
l
d
a
t
a
se
t
s wi
t
h
TL
a
n
d
a
u
g
me
n
t
a
t
i
o
n
.
H
i
g
h
;
p
e
r
f
o
r
ms w
e
l
l
w
h
e
n
p
r
e
-
t
r
a
i
n
e
d
o
n
I
mag
e
N
e
t
a
n
d
f
i
n
e
-
t
u
n
e
d
f
o
r
b
r
a
i
n
t
u
m
o
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
.
[
4
4
]
,
[
4
9
]
,
[
5
0
]
Ef
f
i
c
i
e
n
t
N
e
t
S
c
a
l
a
b
l
e
C
N
N
t
h
a
t
b
a
l
a
n
c
e
s
d
e
p
t
h
,
w
i
d
t
h
,
a
n
d
r
e
s
o
l
u
t
i
o
n
.
H
i
g
h
e
f
f
i
c
i
e
n
c
y
;
f
e
w
e
r
p
a
r
a
m
e
t
e
r
s t
h
a
n
R
e
sN
e
t
,
b
u
t
s
t
r
o
n
g
p
e
r
f
o
r
ma
n
c
e
(
~
5
.
3
-
1
9
mi
l
l
i
o
n
p
a
r
a
m
e
t
e
r
s)
.
Ex
c
e
l
l
e
n
t
;
d
e
s
i
g
n
e
d
f
o
r
e
f
f
i
c
i
e
n
c
y
a
n
d
sca
l
e
s we
l
l
o
n
s
mal
l
e
r
d
a
t
a
se
t
s.
H
i
g
h
;
o
f
t
e
n
o
u
t
p
e
r
f
o
r
ms
o
t
h
e
r
m
o
d
e
l
s
i
n
b
r
a
i
n
t
u
m
o
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
t
a
s
k
s.
[
4
5
]
,
[
4
6
]
Fro
m
T
ab
le
2
,
it is
an
aly
ze
d
th
at
VGG1
6
’
s
s
im
p
licity
an
d
f
ea
tu
r
e
ex
tr
ac
tio
n
ar
e
o
f
f
s
et
b
y
i
ts
m
em
o
r
y
u
tili
za
tio
n
an
d
len
g
th
y
in
f
er
e
n
ce
tim
es.
R
esNe
t
u
s
es
r
e
s
id
u
al
co
n
n
ec
tio
n
s
to
tr
ain
d
ee
p
e
r
n
etwo
r
k
s
,
y
ield
in
g
ex
ce
llen
t
ac
cu
r
ac
y
b
u
t
in
cr
ea
s
in
g
co
m
p
u
tin
g
lo
a
d
.
T
h
e
in
ce
p
tio
n
n
etwo
r
k
p
er
f
o
r
m
s
well
o
n
s
m
all
d
atasets
f
o
r
m
u
lti
-
s
ca
le
f
ea
tu
r
e
ex
tr
ac
tio
n
,
b
u
t
its
co
m
p
lex
ity
m
a
k
es
f
in
e
-
tu
n
in
g
d
if
f
icu
lt.
Fin
ally
,
E
f
f
icien
tNet
b
alan
ce
s
ac
cu
r
ac
y
a
n
d
c
o
m
p
u
tin
g
e
f
f
icien
cy
,
o
f
ten
o
u
tp
er
f
o
r
m
in
g
b
r
ain
tu
m
o
r
class
if
icatio
n
test
s
,
b
u
t
it
m
ay
r
eq
u
ir
e
ca
r
e
f
u
l
tu
n
in
g
.
T
h
ese
m
o
d
els
d
em
o
n
s
tr
ate
th
e
ev
o
lu
tio
n
o
f
DL
ar
c
h
itectu
r
es
in
m
ed
ical
im
ag
in
g
,
p
r
o
g
r
ess
iv
ely
im
p
r
o
v
i
n
g
d
ia
g
n
o
s
tic
ca
p
ab
ilit
ies.
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
A
d
va
n
ce
men
ts
in
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
:
a
s
u
r
ve
y
o
f tra
n
s
fer lea
r
n
in
g
tech
n
iq
u
es
(
S
n
eh
a
l J
a
d
h
a
v
)
1011
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
NS
R
ec
en
t
s
tu
d
ies
o
f
th
e
id
en
tifi
ca
tio
n
o
f
b
r
ain
tu
m
o
r
s
th
r
o
u
g
h
TL
s
u
g
g
est
th
at
th
e
lev
el
o
f
ac
cu
r
ac
y
an
d
p
r
o
p
en
s
ity
o
f
d
iag
n
o
s
in
g
b
r
ain
tu
m
o
r
s
h
as
b
ee
n
en
h
an
ce
d
b
y
im
ag
in
g
tech
n
i
q
u
es
o
f
v
ar
io
u
s
ty
p
es.
T
h
e
u
s
e
o
f
C
NNs lik
e
R
es
Net,
VGGN
et,
an
d
Den
s
eNe
t,
am
o
n
g
o
th
er
DL
m
o
d
els,
h
as p
r
o
v
en
v
er
y
e
f
f
icien
t in
th
e
class
if
icatio
n
o
f
b
r
ain
tu
m
o
r
s
f
r
o
m
MRI
a
n
d
C
T
im
a
g
es.
T
h
r
o
u
g
h
TL
,
th
ese
tech
n
iq
u
es
ca
n
f
in
e
-
tu
n
e
o
n
s
u
ch
s
m
all
d
atasets
th
at
b
elo
n
g
to
a
p
ar
tic
u
lar
d
o
m
ain
,
an
d
th
ey
ca
n
ac
h
iev
e
g
o
o
d
class
if
icatio
n
ac
cu
r
ac
ies,
wh
ich
ar
e
n
o
r
m
ally
h
ig
h
e
r
th
an
9
0
%.
T
h
is
a
p
p
r
o
ac
h
n
o
t
o
n
ly
im
p
r
o
v
es
th
e
p
r
ed
ictio
n
o
f
m
o
d
els
wh
en
b
i
g
lab
eled
d
ata
s
ets
ar
e
r
ar
e
b
u
t
also
cu
ts
th
e
am
o
u
n
t
o
f
co
m
p
u
tatio
n
an
d
tim
e
tak
en
in
to
h
alf
.
T
h
ese
m
o
d
els
h
av
e
b
ee
n
f
o
u
n
d
to
p
er
f
o
r
m
well
f
o
r
ch
ar
ac
ter
izatio
n
o
f
v
ar
io
u
s
tu
m
o
r
ty
p
es
an
d
s
u
b
ty
p
es
with
h
ig
h
er
ac
cu
r
a
cy
th
a
n
tr
ad
itio
n
al
ML
m
o
d
els,
wh
ich
r
eq
u
i
r
e
f
ea
tu
r
e
en
g
in
ee
r
in
g
.
Ad
d
itio
n
al
tr
ea
tm
en
ts
lik
e
d
ata
au
g
m
en
tatio
n
,
th
e
u
s
e
o
f
atte
n
tio
n
,
a
n
d
e
n
s
em
b
le
lear
n
in
g
p
r
o
v
id
e
an
ad
d
itio
n
al
g
u
ar
an
t
ee
f
o
r
th
e
r
eliab
ilit
y
an
d
v
er
s
atility
o
f
t
h
ese
m
o
d
els
.
Data
au
g
m
en
tati
o
n
r
eso
lv
es
th
e
p
r
o
b
lem
o
f
s
m
all
tr
ain
in
g
d
a
ta
b
y
cr
ea
tin
g
m
o
r
e
d
ata
wh
ile
atten
tio
n
in
cr
ea
s
es
th
e
m
o
d
els
’
f
o
cu
s
o
n
s
p
ec
if
ic
tu
m
o
r
p
ar
ts
,
h
en
ce
in
cr
ea
s
in
g
t
h
e
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
.
First,
en
s
em
b
le
lear
n
in
g
a
p
p
l
ies
s
ev
er
al
m
o
d
els
an
d
r
ef
in
es
th
eir
o
u
tp
u
ts
to
en
h
a
n
ce
t
h
e
ac
cu
r
ac
y
o
f
th
e
m
o
d
el
p
r
ed
ictio
n
s
.
Ho
wev
er
,
th
er
e
ar
e
s
till
s
o
m
e
is
s
u
e
s
;
th
e
DL
m
o
d
els
ar
e
co
m
p
u
tatio
n
-
in
ten
s
iv
e,
an
d
i
n
clin
ical
p
r
ac
tice,
we
o
f
ten
r
eq
u
ir
e
an
ex
p
la
n
atio
n
o
f
th
e
o
u
tp
u
t
m
o
d
els.
T
h
e
ap
p
licatio
n
o
f
TL
with
MRI
an
d
C
T
im
a
g
es
h
as
s
o
lv
ed
th
e
c
h
allen
g
es,
im
p
r
o
v
ed
th
e
b
r
ain
tu
m
o
r
class
if
icatio
n
,
an
d
c
o
n
tr
ib
u
ted
to
m
o
r
e
ac
cu
r
ate,
ef
f
icien
t,
an
d
s
ca
lab
l
e
s
o
lu
tio
n
s
f
o
r
th
e
p
atien
ts
’
m
an
ag
em
en
t a
n
d
tr
ea
tm
e
n
t stra
teg
ies.
5.
CO
N
CL
U
SI
O
N
T
h
e
TL
ap
p
r
o
ac
h
to
th
e
b
r
ain
tu
m
o
r
class
if
icatio
n
h
as
s
ig
n
if
ican
tly
im
p
r
o
v
ed
th
e
d
iag
n
o
s
tic
ac
cu
r
ac
y
ac
r
o
s
s
MRI
an
d
C
T
m
ed
ical
im
ag
es.
Au
th
o
r
itativ
e
ly
,
it
h
as
b
ee
n
n
o
ted
th
at
f
i
n
e
-
tu
n
ed
th
r
o
u
g
h
TL
,
DL
m
o
d
els
lik
e
C
NN,
R
esNet,
VGGN
et,
an
d
Den
s
eNe
t
h
a
d
p
ass
ed
th
e
9
0
%
ac
cu
r
ac
y
r
ate
in
class
if
y
in
g
th
e
b
r
ain
tu
m
o
r
s
.
T
h
e
ad
v
an
ce
m
en
t
o
f
th
is
tech
n
o
l
o
g
y
m
ee
ts
s
ig
n
if
ican
t
p
r
o
b
lem
s
o
f
th
e
f
ield
,
s
u
ch
as
th
e
r
eq
u
ir
em
e
n
t
o
f
lar
g
e
lab
eled
d
atasets
o
f
ex
am
p
les
an
d
th
e
tim
e
-
co
n
s
u
m
in
g
co
m
p
u
tatio
n
th
at
ar
is
es
wh
en
r
etr
ain
in
g
m
o
d
el
s
f
r
o
m
th
e
p
r
im
ar
y
ar
e
r
eq
u
ir
ed
.
W
ith
th
e
h
elp
o
f
s
u
ch
tactics
as
d
at
a
au
g
m
en
tatio
n
an
d
f
ea
tu
r
es
lik
e
atten
tio
n
,
th
ese
m
o
d
els
en
h
a
n
ce
th
e
class
if
icatio
n
ac
cu
r
ac
y
an
d
m
ak
e
th
e
m
o
d
els
m
o
r
e
r
o
b
u
s
t
an
d
u
s
ef
u
l
in
clin
ical
p
r
ac
tice
an
d
f
u
r
th
e
r
r
esear
c
h
.
Ho
wev
er
,
th
er
e
ar
e
s
ev
e
r
al
is
s
u
es
th
at
r
em
ain
with
DL
m
o
d
els,
in
clu
d
in
g
th
e
tim
e
c
o
n
s
u
m
p
tio
n
in
co
m
p
u
tatio
n
o
f
h
ig
h
-
co
m
p
lex
ity
m
o
d
els,
m
o
d
el
ex
p
lan
atio
n
o
r
u
n
d
er
s
tan
d
i
n
g
,
a
n
d
p
atien
t
p
r
i
v
ac
y
.
Po
ten
tial
im
p
r
o
v
em
en
t
o
f
th
e
alg
o
r
ith
m
s
,
in
cr
ea
s
in
g
th
e
in
ter
p
r
etab
ilit
y
o
f
th
ese
m
o
d
els,
an
d
th
e
in
teg
r
atio
n
o
f
th
ese
s
o
p
h
is
ticated
s
y
s
tem
s
in
cl
in
ical
p
r
ac
tice
f
o
r
d
aily
u
s
e
in
th
e
f
u
tu
r
e.
TL
,
DL
,
an
d
o
th
er
ad
v
an
ce
m
e
n
ts
in
th
e
f
ield
co
n
t
in
u
e
to
s
ee
en
h
a
n
ce
m
en
ts
;
h
e
n
ce
,
th
e
f
u
tu
r
e
h
o
ld
s
b
etter
s
o
lu
tio
n
s
to
ac
cu
r
ately
d
iag
n
o
s
e
b
r
ain
tu
m
o
r
s
.
T
h
ese
tech
n
o
lo
g
ies,
cu
r
r
e
n
tly
in
d
ev
elo
p
m
e
n
tal
p
h
ases
,
p
o
s
s
ess
th
e
ca
p
ac
ity
to
r
ev
o
lu
tio
n
ize
m
ed
ical
i
m
ag
in
g
,
th
er
eb
y
e
n
h
an
cin
g
p
atien
t
o
u
tco
m
es
th
r
o
u
g
h
im
p
r
o
v
ed
d
ia
g
n
o
s
tic
ca
p
ab
ilit
ies
an
d
m
o
r
e
p
r
ec
is
e
tr
ea
tm
en
t
d
e
cisi
o
n
s
.
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
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Sn
eh
al
J
ad
h
av
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Sm
ita
B
h
ar
n
e
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Vaib
h
av
Nar
awa
d
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
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.
Evaluation Warning : The document was created with Spire.PDF for Python.