I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
,
p
p
.
3226
~
3
2
3
7
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
3
.
pp
3
2
2
6
-
3
2
3
7
3226
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Ro
bust dee
p lear
ning
appro
a
ch f
o
r accura
te
det
e
cti
o
n of brain
tumo
r and a
na
ly
sis
L
a
nk
e
P
a
lla
v
i,
T
ha
t
i
Ra
m
y
a
,
Sin
g
up
ura
pu
Sa
i
Cha
ra
n,
Si
rig
a
dh
a
Am
it
h,
T
ho
du
pu
n
ur
i
Ak
s
ha
y
K
u
ma
r
D
e
p
a
r
t
me
n
t
of
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
B
V
R
a
j
u
I
n
st
i
t
u
t
e
of
Te
c
h
n
o
l
o
g
y
,
N
a
r
sa
p
u
r
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 1
5
,
2
0
2
4
R
ev
is
ed
Dec
1
0
,
2
0
2
4
Acc
ep
ted
Dec
1
9
,
2
0
2
4
Us
u
a
ll
y
,
o
n
e
o
f
t
h
e
f
o
re
m
o
st
p
re
d
o
m
in
a
n
t
a
n
d
in
tri
c
a
te
t
h
e
ra
p
e
u
ti
c
c
o
n
d
i
ti
o
n
s.
As
b
r
o
a
d
ly
p
e
rc
e
iv
e
d
,
b
ra
in
tu
m
o
rs
a
re
a
m
o
n
g
th
e
fo
re
m
o
st
sig
n
ifi
c
a
n
t
ly
h
a
rm
fu
l
c
ircu
m
sta
n
c
e
s
th
a
t
c
a
n
ra
d
ica
ll
y
a
b
b
re
v
iate
a
p
e
rso
n
’s
li
fe
e
x
p
e
c
tan
c
y
.
Va
rio
u
s
m
e
th
o
d
s
a
re
lac
k
in
g
f
o
r
o
b
se
rv
in
g
th
e
a
ss
o
rtme
n
t
o
f
tu
m
o
r
siz
e
s,
s
h
a
p
e
s,
a
n
d
a
re
a
s.
Wh
e
n
m
e
rg
e
d
wit
h
stra
teg
ies
o
f
p
r
o
fo
u
n
d
lea
rn
in
g
,
g
e
n
e
ra
ti
v
e
a
d
v
e
rsa
rial
n
e
two
rk
s
(G
AN
s)
a
re
c
o
m
p
e
ten
t
o
f
c
a
tch
in
g
th
e
m
e
a
su
re
m
e
n
ts,
a
re
a
s,
a
n
d
stru
c
tu
re
s
o
f
tu
m
o
rs.
P
ro
f
o
u
n
d
lea
rn
in
g
fra
m
e
wo
rk
s
will
m
o
v
e
fo
rwa
rd
u
p
o
n
th
e
s
h
o
rta
g
e
o
f
d
a
tas
e
ts.
It
c
a
n
m
o
re
o
v
e
r
p
ro
g
re
ss
p
h
o
to
g
ra
p
h
s
with
d
e
term
in
a
ti
o
n
.
Clas
sify
in
g
a
n
d
p
a
rti
ti
o
n
i
n
g
b
ra
i
n
tu
m
o
rs
p
ro
d
u
c
ti
v
e
ly
is
sig
n
ifi
c
a
n
t.
G
AN
s
a
r
e
u
se
d
i
n
c
o
n
ju
n
c
ti
o
n
with
a
n
o
v
e
ra
rc
h
i
n
g
lea
rn
in
g
h
a
n
d
le.
A
p
r
o
fo
u
n
d
lea
rn
in
g
d
e
sig
n
c
a
ll
e
d
Ne
u
ro
Ne
t1
9
,
c
o
u
ld
b
e
a
n
in
terc
ro
ss
o
f
v
isu
a
l
g
e
o
m
e
try
g
ro
u
p
(VG
G
1
9
)
a
n
d
in
v
e
rted
p
y
ra
m
id
p
o
o
li
n
g
m
o
d
u
le
(IP
P
M
)
wh
ich
is
u
ti
li
z
e
d
t
o
re
c
o
g
n
ize
b
ra
in
tu
m
o
rs.
I
t
i
s
c
lea
r
th
a
t,
Ne
u
ro
Ne
t1
9
e
m
p
lo
y
m
e
n
ts
th
e
fo
re
m
o
st
e
x
a
c
t
tec
h
n
i
q
u
e
in
c
o
m
p
a
riso
n
to
a
ll
m
o
d
e
ls
(De
n
se
Ne
t1
2
1
,
M
o
b
i
leN
e
t,
Re
sN
e
t5
0
,
VG
G
1
6
).
Th
e
e
x
a
c
tn
e
ss
e
x
a
m
in
a
ti
o
n
g
a
v
e
a
Co
h
e
n
Ka
p
p
a
c
o
e
fficie
n
t
o
f
9
9
%
a
n
d
a
F
1
-
sc
o
re
o
f
9
9
.
2
%
K
ey
w
o
r
d
s
:
Den
s
eNe
t1
2
1
G
e
n
e
r
a
t
i
v
e
a
d
v
e
r
s
a
r
i
al
n
e
t
w
o
r
k
s
I
n
v
er
ted
p
y
r
a
m
id
p
o
o
lin
g
m
o
d
u
le
Me
d
ical
im
ag
in
g
b
r
ain
tu
m
o
r
m
en
in
g
io
m
a
p
itu
itar
y
t
u
m
o
r
Ma
g
n
etic
r
eso
n
an
ce
im
a
g
in
g
Neu
r
o
N
et1
9
g
lio
m
a
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
:
T
h
ati
R
am
y
a
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
B
V
R
aju
I
n
s
titu
te
o
f
T
ec
h
n
o
l
o
g
y
Nar
s
ap
u
r
,
T
elan
g
an
a,
I
n
d
ia
E
m
ail: r
am
y
ath
ati1
1
0
2
@
g
m
ai
l.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
B
r
ain
tu
m
o
r
s
m
ay
b
e
d
ef
in
e
d
as
a
g
r
o
win
g
ab
n
o
r
m
ality
o
f
b
r
ain
ce
lls
.
Su
ch
a
b
io
lo
g
ical
co
m
p
lex
ar
ch
itectu
r
e
r
eq
u
ir
es
q
u
ite
a
c
h
allen
g
in
g
m
e
d
icin
e
an
d
also
th
e
p
o
s
s
ib
ilit
y
o
f
im
p
air
in
g
b
r
ain
f
u
n
ctio
n
s
[
1
]
.
T
h
ey
m
a
y
ex
e
r
t
p
r
ess
u
r
e
o
n
s
u
r
r
o
u
n
d
in
g
tis
s
u
es,
h
en
ce
e
v
e
n
tu
ally
lead
in
g
to
i
n
tr
ac
r
an
ial
p
r
ess
u
r
e
an
d
f
lu
i
d
ac
cu
m
u
latio
n
[
2
]
.
Sin
ce
th
e
y
c
o
u
ld
co
m
e
i
n
an
y
p
ar
t
o
f
t
h
e
b
r
ain
an
d
m
a
y
v
ar
y
in
s
ize
an
d
s
ev
er
ity
,
d
iag
n
o
s
is
an
d
tr
ea
tm
en
t tu
r
n
o
u
t q
u
ite
ch
allen
g
in
g
.
B
r
ain
tu
m
o
r
s
ar
e
b
asically
c
lass
if
ied
in
to
two
ca
teg
o
r
ies
ac
co
r
d
in
g
to
t
h
eir
o
r
ig
in
:
b
en
ig
n
an
d
m
alig
n
an
t
[
3
]
.
T
h
ese
b
en
ig
n
ty
p
es
ar
e
g
e
n
er
ally
n
o
n
-
ag
g
r
ess
iv
e,
ch
ar
ac
ter
ized
b
y
s
lo
w
g
r
o
wth
a
n
d
p
o
s
e
r
elativ
ely
less
th
r
ea
t
to
h
ea
lth
[
4
]
.
On
t
h
e
o
th
er
h
an
d
,
m
alig
n
an
t
tu
m
o
r
s
a
r
e
a
g
g
r
ess
iv
e;
th
ey
ca
n
in
v
a
d
e
s
u
r
r
o
u
n
d
in
g
ar
ea
s
as
well
as
s
p
r
ea
d
to
alm
o
s
t
all
p
ar
ts
o
f
th
e
b
o
d
y
an
d
p
o
s
e
a
s
er
io
u
s
th
r
ea
t
to
h
u
m
an
h
ea
lth
,
if
n
o
t
tr
ea
ted
with
ex
ce
llen
t
ef
f
ec
tiv
en
ess
[
5
]
.
T
h
u
s
,
ea
r
ly
b
r
ain
tu
m
o
r
d
etec
tio
n
is
im
p
o
r
tan
t
f
o
r
p
atien
ts
im
p
r
o
v
em
e
n
t th
r
o
u
g
h
th
e
p
r
o
v
is
io
n
o
f
ea
r
ly
m
ed
ical
th
er
a
p
y
.
M
a
n
y
t
e
c
h
n
i
q
u
es
h
a
v
e
b
e
e
n
d
es
i
g
n
e
d
f
o
r
t
h
e
d
e
t
e
ct
i
o
n
of
b
r
a
in
t
u
m
o
r
s
,
a
l
l
of
w
h
i
c
h
h
a
v
e
an
a
p
p
l
i
c
a
t
i
o
n
an
d
a
d
r
aw
b
ac
k
[
6
]
,
[
7
]
.
Ho
wev
er
,
in
m
o
s
t
co
n
v
en
tio
n
al
im
ag
e
d
iag
n
o
s
tics
,
th
e
m
o
d
els
r
ely
on
m
ac
h
i
n
e
lear
n
in
g
-
b
ased
m
o
d
els
th
at
ar
e
ea
s
ily
co
n
f
u
s
ed
with
th
e
in
h
er
e
n
t
c
o
m
p
lex
ity
of
b
r
ain
tu
m
o
r
s
s
in
ce
of
v
ar
iatio
n
s
in
g
r
o
wth
p
atter
n
s
an
d
ch
ar
ac
ter
i
s
tics
[
8
]
,
[
9
]
.
R
ec
en
t r
esear
ch
h
as f
o
cu
s
ed
o
n
d
ee
p
lear
n
in
g
t
ec
h
n
iq
u
es
[
1
0
]
th
at
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
o
b
u
s
t d
ee
p
lea
r
n
in
g
a
p
p
r
o
a
ch
fo
r
a
cc
u
r
a
te
d
etec
tio
n
o
f b
r
a
in
tu
mo
r
a
n
d
a
n
a
lysi
s
(
La
n
ke
P
a
lla
vi
)
3227
m
ay
p
r
o
v
e
q
u
ite
p
r
o
m
is
in
g
in
au
g
m
en
tin
g
d
etec
tio
n
ca
p
ab
ilit
ies.
Kar
am
eh
ić
et
a
l.
[
1
1
]
p
r
esen
ted
th
e
en
s
em
b
le
d
ee
p
lear
n
in
g
m
o
d
e
l
as
an
ap
p
r
o
ac
h
to
o
v
er
c
o
m
e
th
ese
is
s
u
es
b
u
t
m
ay
b
e
a
v
er
y
co
m
p
u
tatio
n
ally
ex
p
en
s
iv
e
m
o
d
el
c
h
allen
g
in
g
to
im
p
lem
e
n
t
in
t
h
e
clin
ica
l
s
ettin
g
.
Pan
d
iy
an
et
a
l.
[
1
2
]
h
a
v
e
u
s
ed
m
o
r
e
ad
v
an
ce
d
ar
ch
itectu
r
es
o
f
co
n
v
o
lu
tio
n
al
s
y
s
tem
s
f
o
r
th
e
ac
k
n
o
wled
g
m
e
n
t
an
d
class
if
icati
o
n
wo
r
k
.
Ho
wev
e
r
,
th
e
ab
o
v
e
tech
n
i
q
u
es
ar
e
c
h
allen
g
in
g
b
ec
au
s
e
o
f
th
e
s
p
a
r
s
ity
o
f
th
e
av
ailab
le
lab
ele
d
m
ed
i
ca
l
d
atasets
,
wh
ich
m
ay
ca
u
s
e
o
v
e
r
f
itti
n
g
an
d
p
o
o
r
g
en
er
aliza
tio
n
.
Naje
eb
an
d
Dah
l
[
1
3
]
r
ev
iewe
d
tr
an
s
f
er
lear
n
in
g
tech
n
i
q
u
es
to
u
s
e
p
r
e
-
tr
ain
ed
m
o
d
els
lik
e
VGG
an
d
R
esNet
f
o
r
im
p
r
o
v
em
en
t.
M
ah
m
u
d
et
a
l.
[
1
4
]
h
av
e
em
p
lo
y
ed
p
r
o
f
o
u
n
d
lear
n
in
g
to
m
o
v
e
f
o
r
war
d
th
e
class
if
icatio
n
o
f
b
r
ain
tu
m
o
r
s
.
W
h
ile
th
ese
ad
v
an
ce
m
en
ts
ar
e
f
ea
s
ib
le,
th
e
tech
n
iq
u
es
wo
u
ld
s
till
in
h
er
it
th
e
is
s
u
e
o
f
h
ig
h
in
te
r
-
class
v
ar
iab
ilit
y
o
f
tu
m
o
r
m
o
r
p
h
o
l
o
g
y
a
n
d
v
ar
ied
s
ce
n
ar
io
s
in
im
ag
in
g
.
Gen
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
s
(
GANs)
ar
e
th
e
em
er
g
in
g
tech
n
o
lo
g
ies
in
m
ed
ical
im
ag
in
g
[
1
5
]
,
as
it
ca
n
g
en
er
ate
v
e
r
y
r
ea
lis
tic
s
y
n
th
etic
d
ata
f
o
r
a
u
g
m
e
n
tatio
n
p
u
r
p
o
s
es.
T
h
e
wo
r
k
o
f
Sh
o
aib
et
a
l.
[
1
6
]
a
n
d
Am
b
esh
war
et
a
l.
[
1
7
]
s
h
o
w
th
e
s
tr
en
g
th
o
f
GANs
to
o
v
er
co
m
e
lim
itatio
n
s
f
ac
ed
with
s
m
all
-
s
ized
lab
eled
d
atasets
.
Ho
wev
er
,
m
o
s
t
GAN
-
b
ased
m
o
d
els
f
ail
to
en
s
u
r
e
a
n
ato
m
ical
f
id
elity
in
g
e
n
er
ate
d
im
ag
es
an
d
f
ail
to
co
m
p
o
s
e
with
ea
s
y
class
if
icatio
n
task
s
.
T
h
is
lan
d
s
ca
p
e
u
n
d
er
lin
es
th
e
n
ee
d
f
o
r
in
n
o
v
ativ
e
ap
p
r
o
ac
h
es
d
esig
n
ed
t
o
b
e
ab
le
to
ef
f
ec
t
iv
ely
ad
d
r
ess
th
e
lim
itatio
n
s
o
f
th
e
ex
is
tin
g
m
eth
o
d
o
lo
g
i
es
f
o
r
b
r
ain
tu
m
o
r
d
etec
tio
n
.
T
h
e
in
co
r
p
o
r
atio
n
o
f
GAN
-
b
ased
en
h
an
ce
d
tec
h
n
iq
u
es
in
to
q
u
ite
r
o
b
u
s
t
ar
ch
itectu
r
es
o
f
n
eu
r
al
n
etwo
r
k
s
s
ig
n
if
ican
tly
im
p
r
o
v
es d
iag
n
o
s
tic
ca
p
ab
ilit
ies an
d
en
h
an
ce
s
th
e
o
u
tco
m
e
f
o
r
p
atien
ts
[
1
8
]
.
T
o
o
v
er
c
o
m
e
th
ese
ch
allen
g
es
,
we
p
r
o
p
o
s
e
th
e
in
n
o
v
ativ
e
a
r
ch
itectu
r
e
ca
lled
Neu
r
o
Net1
9
s
h
o
wn
in
Fig
u
r
e
1
,
wh
ich
in
teg
r
ates th
e
p
r
o
s
o
f
VGG1
9
with
in
v
er
ted
p
y
r
am
id
p
o
o
lin
g
m
o
d
u
le
(
I
PP
M)
.
L
ev
er
ag
in
g
th
e
p
r
o
v
e
n
f
ea
tu
r
e
e
x
tr
ac
tio
n
ab
ili
ty
o
f
VGG1
9
,
th
e
in
teg
r
atio
n
o
f
I
PP
M
f
ac
ilit
ates
iter
ativ
ely
r
ef
in
in
g
an
im
ag
e
p
atch
.
T
h
e
h
y
b
r
id
ap
p
r
o
ac
h
th
er
ef
o
r
e
f
u
n
ctio
n
s
to
im
p
r
o
v
e
th
e
ac
cu
r
ac
y
o
f
class
if
icatio
n
f
o
r
tu
m
o
r
s
b
y
s
im
u
ltan
eo
u
s
ly
f
o
cu
s
in
g
o
n
lo
ca
l
an
d
g
lo
b
al
tu
m
o
r
ch
ar
ac
te
r
is
tics
.
W
e
f
u
r
th
er
in
co
r
p
o
r
ate
a
GAN
-
b
ased
d
ata
au
g
m
en
tatio
n
p
ip
elin
e
th
at
g
e
n
er
ates
n
ea
r
r
ea
lis
tic
s
y
n
th
etic
m
ag
n
etic
r
eso
n
a
n
ce
im
ag
in
g
(
MRI)
im
ag
es
an
d
ca
n
m
itig
ate
is
s
u
es
lik
e
lim
ite
d
d
atasets
an
d
en
h
a
n
ce
m
o
d
el
r
o
b
u
s
tn
ess
an
d
Ma
s
k
-
r
e
g
io
n
-
b
ased
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
Ma
s
k
R
-
C
NN)
to
p
o
i
n
t th
e
ex
ac
t tu
m
o
r
l
o
ca
tio
n
.
O
u
r
a
p
p
r
o
a
c
h
f
i
l
l
s
s
t
il
l
m
i
s
s
i
n
g
g
a
p
s
in
e
x
i
s
t
i
n
g
m
e
t
h
o
d
o
l
o
g
i
e
s
by
u
s
i
n
g
a
GAN
-
a
u
g
m
e
n
t
e
d
f
r
a
m
e
w
o
r
k
to
b
o
o
s
t
m
o
d
el
tr
ai
n
in
g
o
n
s
p
ar
s
e
d
atasets
.
Usi
n
g
th
e
n
ew
I
PP
M
m
o
d
u
le,
en
r
ich
e
d
f
e
atu
r
e
ex
tr
ac
tio
n
a
n
d
class
if
icatio
n
.
I
t
is
a
co
m
p
u
tatio
n
ally
ef
f
icien
t
s
o
lu
tio
n
t
h
at
ad
ap
ts
to
clin
ical
wo
r
k
f
l
o
ws.
I
t
s
ig
n
if
ies
a
r
ev
o
lu
tio
n
a
r
y
ap
p
r
o
ac
h
th
at
d
iag
n
o
s
es
b
r
ain
tu
m
o
r
s
u
s
in
g
ad
v
an
ce
d
d
ee
p
lear
n
in
g
an
d
GAN
tech
n
iq
u
es.
Neu
r
o
Net1
9
h
as
a
g
o
o
d
ch
a
n
ce
o
f
tack
lin
g
ea
r
ly
tu
m
o
r
d
etec
tio
n
th
at
wo
u
ld
r
esu
lt
in
th
e
ap
p
r
o
p
r
iate
in
ter
v
en
tio
n
s
at
th
e
r
ig
h
t tim
e,
th
u
s
u
p
g
r
ad
in
g
th
e
s
u
r
v
i
v
al
r
a
tes o
f
p
atien
ts
.
Fig
u
r
e
1.
Pro
p
o
s
ed
a
r
c
h
i
t
e
c
t
u
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
3
2
2
6
-
3
2
3
7
3228
2.
M
E
T
H
O
D
I
n
th
is
s
tu
d
y
,
th
e
d
etec
tio
n
o
f
b
r
ain
tu
m
o
r
s
wo
u
ld
c
o
m
b
in
e
ad
v
an
ce
d
p
r
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
with
th
e
n
o
v
el
ar
ch
itectu
r
e
ca
lled
Neu
r
o
Net1
9
an
d
p
r
o
g
r
ess
iv
e
g
r
o
win
g
GANs.
Ou
r
f
r
am
ewo
r
k
s
ig
n
if
ican
tly
en
h
an
ce
s
th
e
s
p
ee
d
an
d
ac
cu
r
ac
y
o
f
th
e
d
etec
tio
n
o
f
b
r
ai
n
t
u
m
o
r
b
ec
au
s
e
o
f
th
e
r
o
b
u
s
t
f
e
atu
r
es
ex
tr
ac
ted
b
y
VGG1
9
,
en
h
an
ce
d
e
v
en
b
etter
b
y
th
e
I
PP
M.
Mo
r
eo
v
er
,
p
r
ec
is
e
s
eg
m
en
tatio
n
is
ac
co
m
p
lis
h
ed
th
r
o
u
g
h
Ma
s
k
R
-
C
NN
wh
ile
en
s
u
r
in
g
p
r
o
p
er
lo
ca
lizatio
n
o
f
th
e
tu
m
o
r
s
.
T
h
ese
m
o
d
els
wer
e
s
elec
ted
d
u
e
to
th
eir
illu
s
tr
ated
ef
f
ec
tiv
en
ess
in
tak
in
g
ca
r
e
o
f
co
m
p
lex
d
atasets
an
d
g
iv
in
g
p
r
ec
is
e
ex
p
ec
tatio
n
s
.
2.
1
.
Det
a
ils
o
f
d
a
t
a
s
et
T
h
e
d
ataset
co
n
s
is
ts
o
f
7
,
0
2
3
r
esto
r
ativ
e
MRI
im
ag
es
b
elo
n
g
in
g
to
tr
ain
in
g
,
v
alid
atio
n
an
d
test
s
et
s
to
en
s
u
r
e
th
e
th
o
r
o
u
g
h
test
in
g
o
f
a
well
-
tr
ain
ed
m
o
d
el.
Fu
r
th
er
s
ep
ar
ated
in
to
f
o
u
r
d
if
f
er
e
n
t
class
e
s
:
g
lio
m
a,
m
en
in
g
io
m
a
,
p
itu
itar
y
tu
m
o
r
,
an
d
n
o
tu
m
o
r
.
T
h
e
tr
ain
in
g
s
et
is
m
ea
n
t
f
o
r
th
e
tr
ain
in
g
o
f
th
e
Neu
r
o
Net1
9
m
o
d
el,
s
o
th
at
it
g
ain
s
r
o
b
u
s
t
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
.
T
h
e
v
alid
atio
n
s
et
is
m
ain
ly
u
s
ef
u
l
in
th
e
p
r
o
ce
s
s
o
f
tr
ain
in
g
t
o
f
i
n
e
-
tu
n
e
th
e
m
o
d
e
l
f
u
r
th
e
r
in
ar
r
an
g
e
to
av
o
id
o
v
er
f
itti
n
g
.
A
p
r
o
p
e
r
ch
ec
k
o
f
th
e
m
o
d
el
o
n
s
u
c
h
u
n
s
ee
n
d
ata
is
f
in
ally
in
te
n
d
ed
to
b
e
d
o
n
e
b
y
th
e
test
in
g
s
et.
2
.
2
.
Da
t
a
p
re
pro
ce
s
s
ing
Pre
p
r
o
ce
s
s
in
g
m
ay
b
e
a
p
iv
o
tal
m
u
lti
-
s
tep
s
tr
ateg
y
th
at
i
m
p
r
o
v
es
th
e
q
u
ality
o
f
MRI
p
ictu
r
es
b
y
d
ec
r
ea
s
in
g
clam
o
r
,
ad
ju
s
tin
g
ar
tifa
cts,
an
d
s
tan
d
ar
d
izin
g
th
e
in
f
o
r
m
atio
n
f
o
r
d
ep
e
n
d
ab
le
in
v
esti
g
atio
n
.
T
h
is
p
r
ep
ar
e
in
clu
d
es
a
n
ar
r
an
g
e
m
en
t
o
f
ch
a
n
g
es,
s
u
ch
as
co
n
ce
n
tr
ated
n
o
r
m
aliza
tio
n
,
cr
a
n
iu
m
s
tr
ip
p
in
g
,
a
n
d
p
ictu
r
e
en
r
o
llm
en
t,
to
g
u
ar
an
t
ee
co
n
s
is
ten
cy
o
v
er
f
ilter
s
.
T
h
ese
r
ef
in
ed
p
ictu
r
es
s
er
v
e
as
t
h
e
estab
lis
h
m
en
t
f
o
r
p
r
ec
is
e
s
y
m
p
to
m
atic
ap
p
r
aisals
an
d
AI
-
b
ased
f
o
r
ec
asts
in
th
er
ap
eu
tic
im
ag
in
g
[
1
9
]
.
2
.
2
.
1
.
No
rma
liza
t
io
n
Nu
m
er
o
u
s
s
tr
ateg
ies
ar
e
ac
ce
s
s
ib
le
f
o
r
n
o
r
m
alizin
g
in
f
o
r
m
atio
n
,
co
u
n
tin
g
z
-
s
co
r
e,
m
in
-
m
ax
,
an
d
s
ca
lin
g
o
f
d
ec
im
al
[
2
0
]
.
B
y
u
t
ilizin
g
th
is
m
in
-
m
ax
s
tr
ateg
y
,
th
e
co
n
ce
n
t
r
ated
v
alu
es
in
th
e
in
p
u
t
MRI
im
ag
es
ar
e
s
ca
led
to
th
e
ex
ten
d
(
1
,
1
)
o
r
(
0
,
1
)
.
“
z
-
s
co
r
e
s
tan
d
ar
d
iz
atio
n
”
is
a
m
eth
o
d
th
at
n
o
r
m
a
lizes
ea
ch
escalate
d
esteem
s
ee
n
in
an
attr
ac
tiv
e
r
e
v
er
b
er
atio
n
p
ictu
r
in
g
s
u
ch
t
h
at
th
e
s
tan
d
ar
d
d
ev
iatio
n
a
n
d
m
e
an
ar
e
‘
0
’
[
2
1
]
.
2
.
2
.
2
.
Sk
ull
s
t
rippi
ng
MRI
o
f
b
r
ain
c
h
ec
k
s
h
ab
itu
all
y
u
n
co
v
er
z
o
n
es
o
f
n
o
n
-
b
r
ain
tis
s
u
es
s
u
ch
as
th
e
cr
an
iu
m
,
d
u
r
a
m
ater
,
s
ca
lp
an
d
m
en
in
g
es.
W
h
en
th
e
s
e
th
in
g
s
ar
e
in
co
r
p
o
r
ated
in
t
h
e
s
h
o
w,
th
e
o
p
e
r
atio
n
o
f
clas
s
if
icatio
n
o
r
d
in
ar
ily
p
r
o
ce
ed
s
.
So
,
t
h
e
b
r
ain
tu
m
o
r
ca
teg
o
r
izatio
n
ass
ig
n
m
en
ts
r
e
g
u
lar
ly
u
tili
ze
th
is
m
eth
o
d
as
a
to
b
eg
in
with
s
tep
to
im
p
r
o
v
e
ex
ec
u
tio
n
wh
er
e
th
e
n
o
n
-
b
r
ain
z
o
n
es a
r
e
e
x
p
elled
.
2
.
2
.
3
.
Resizing
W
e
g
o
t
to
s
ca
le
all
th
e
p
h
o
t
o
g
r
ap
h
s
s
o
m
e
tim
e
r
ec
e
n
tly
n
o
u
r
is
h
in
g
in
to
th
e
m
o
d
els
o
f
n
e
u
r
al
n
etwo
r
k
class
if
icatio
n
,
s
in
ce
d
ee
p
n
eu
r
al
s
y
s
tem
s
n
ee
d
r
eliab
le
esti
m
ate
in
p
u
t
[
2
2
]
.
I
n
ca
s
e
a
p
i
ctu
r
e
is
b
ig
g
er
th
a
n
f
u
n
d
am
e
n
tal,
it
m
ay
b
e
s
ca
led
d
o
wn
b
y
alter
in
g
th
e
p
ix
els
in
ter
io
r
th
e
o
u
tlin
e
o
r
b
y
ad
j
u
s
tin
g
th
e
p
ictu
r
e
.
T
h
u
s
,
all
MRI
p
ictu
r
es
ar
e
r
es
ized
to
a
s
tead
y
m
ea
s
u
r
em
en
t,
o
f
2
2
4
×
2
2
4
p
i
x
els
to
m
ee
t
in
p
u
t
p
r
er
e
q
u
is
ites
o
f
th
e
VGG1
9
en
g
in
ee
r
in
g
.
2
.
2
.
4
.
T
ra
ns
f
o
rm
ed
pict
ure
T
ak
in
g
af
ter
th
e
p
r
o
ce
s
s
o
f
tr
i
m
m
in
g
,
p
i
x
el
v
alu
es
o
f
th
is
tr
im
m
ed
p
ictu
r
e
a
r
e
n
o
r
m
alize
d
s
u
ch
th
at
th
e
v
alu
e
d
r
o
p
s
b
etwe
en
‘
0
’
an
d
‘
1
’
.
I
t
g
u
ar
an
tees
d
e
p
en
d
ab
le
a
n
d
c
o
n
tin
u
o
u
s
e
x
c
h
an
g
e
o
f
co
m
p
le
x
p
r
o
g
r
ess
iv
e
d
ata.
T
h
is
im
ag
e
at
th
at
p
o
in
t
ex
p
e
r
ien
ce
s
a
b
et
a
(
o
r
p
o
wer
-
law)
r
ed
r
ess
to
f
o
cu
s
o
n
it
i
s
s
p
ec
ial
f
ea
tu
r
es.
I
t is sh
o
wn
in
(
1
)
.
=
×
(
1
)
I
n
th
e
s
p
ec
if
ic
s
ce
n
ar
io
,
an
es
teem
o
f
1
.
5
is
allo
tted
to
alter
v
alu
e.
T
h
e
p
r
o
p
o
r
tio
n
ality
s
tead
y
is
r
ep
r
esen
ted
by
,
in
itial e
s
teem
o
f
p
ix
el
b
y
,
an
d
ch
a
n
g
ed
o
v
er
esteem
o
f
p
ix
el
b
y
.
2
.
2
.
5
.
Deno
is
ing
No
is
e
o
f
h
ig
h
f
r
eq
u
en
cy
,
f
r
eq
u
en
tly
ca
lled
“
s
p
ec
k
clam
o
r
,
”
p
r
e
v
en
ts
ac
cu
r
ac
y
ab
o
u
t
af
ter
war
d
ex
am
in
atio
n
a
n
d
el
u
cid
atio
n
s
am
id
p
r
e
p
ar
in
g
o
f
r
esto
r
ativ
e
im
ag
es.
T
h
er
ef
o
r
e,
G
au
s
s
ian
o
b
s
cu
r
e
is
u
tili
ze
d
u
p
o
n
p
h
o
to
s
f
o
r
d
im
in
is
h
in
g
th
e
d
o
t
clam
o
r
an
d
p
r
o
g
r
ess
it
i
s
clar
ity
a
n
d
s
m
o
o
th
n
ess
.
T
h
is
G
au
s
s
ian
b
lu
r
s
er
v
es a
s
estab
lis
h
m
en
t f
o
r
th
e
ex
ten
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
o
b
u
s
t d
ee
p
lea
r
n
in
g
a
p
p
r
o
a
ch
fo
r
a
cc
u
r
a
te
d
etec
tio
n
o
f b
r
a
in
tu
mo
r
a
n
d
a
n
a
lysi
s
(
La
n
ke
P
a
lla
vi
)
3229
(
,
)
=
1
2
2
−
2
+
2
2
2
(
2
)
We
s
p
ea
k
to
th
e
Gau
s
s
ian
p
ar
t
esteem
at
p
o
in
t
(
,
)
u
tili
zin
g
im
ag
e
GB
(
,
)
as
ap
p
ea
r
ed
in
(
2
)
.
11
×
1
1
p
ar
t
m
ea
s
u
r
e
is
u
tili
ze
d
ac
co
r
d
in
g
to
o
u
r
e
x
am
in
atio
n
.
Stan
d
ar
d
d
ev
iatio
n
is
ca
lcu
lated
by
p
r
ep
ar
in
g
p
r
o
g
r
am
in
s
id
e,
with
th
e
eq
u
atio
n
.
=
−
1
6
(
3
)
‘
n’
s
p
ea
k
s
to
b
it
m
ea
s
u
r
em
e
n
t
in
(
3
)
.
2
.
2
.
6
.
Da
t
a
e
nla
rg
e
m
ent
W
e
u
tili
ze
th
e
“in
f
o
r
m
atio
n
in
cr
ea
s
e”
m
eth
o
d
[
2
3
]
to
e
x
ten
d
th
e
d
ataset
b
y
ap
p
ly
in
g
d
if
f
er
e
n
t
ch
an
g
es
to
e
x
is
tin
g
in
f
o
r
m
at
io
n
test
s
,
co
u
n
tin
g
tu
r
n
,
f
lip
p
in
g
,
z
o
o
m
in
g
,
tr
im
m
i
n
g
,
a
n
d
th
r
esh
o
ld
i
n
g
,
as
o
u
tlin
ed
i
n
Fig
u
r
e
2
.
T
h
ese
a
d
ju
s
tm
en
ts
p
r
esen
t
d
if
f
er
e
n
ce
s
in
s
id
e
th
e
p
r
ep
a
r
in
g
s
et,
im
p
r
o
v
in
g
t
h
e
m
o
d
el'
s
ca
p
ac
ity
to
g
e
n
er
alize
o
v
e
r
d
iv
er
s
e
v
ar
ieties
o
f
i
n
p
u
t
in
f
o
r
m
atio
n
.
B
y
co
n
s
o
lid
atin
g
a
wid
e
r
u
n
o
f
m
o
d
if
icatio
n
s
,
th
is
ap
p
r
o
ac
h
p
r
o
g
r
ess
es
v
is
u
al
q
u
ality
,
f
o
r
t
if
ies
s
h
o
w
ex
ec
u
tio
n
,
an
d
co
n
tr
ib
u
tes
to
a
m
o
r
e
p
r
o
f
o
u
n
d
h
y
p
o
th
etica
l u
n
d
er
s
t
an
d
in
g
o
f
th
e
i
n
f
o
r
m
atio
n
.
Fig
u
r
e
2.
Data
au
g
m
en
tatio
n
f
o
r
b
r
ain
tu
m
o
r
s
e
g
m
e
n
t
a
t
i
o
n
2
.
3
.
Neuro
Net
1
9
a
rc
hite
ct
ure
T
h
e
Neu
r
o
Net1
9
m
o
d
el
is
o
n
e
o
f
th
e
d
ee
p
l
y
s
p
ec
ialized
p
r
o
f
o
u
n
d
lear
n
in
g
s
tr
u
ctu
r
es
th
at
h
av
e
b
ee
n
o
u
tlin
ed
to
class
if
y
p
ictu
r
es
o
f
b
r
ain
tu
m
o
r
s
f
r
o
m
a
d
ataset
o
f
MRI
im
ag
es.
I
t
is
an
ad
v
an
ce
m
e
n
t
o
v
e
r
th
e
VGG1
9
ar
ch
itectu
r
e
an
d
co
m
p
r
is
es
a
n
o
v
el
f
ea
tu
r
e
ex
tr
ac
tio
n
m
o
d
u
le
k
n
o
wn
as
th
e
in
v
er
ted
p
y
r
a
m
i
d
p
o
o
lin
g
m
o
d
u
l
e
or
I
P
P
M
.
I
ts
h
y
b
r
i
d
i
n
n
o
v
a
t
i
v
e
d
e
s
i
g
n
a
i
m
s
at
a
c
h
i
e
v
i
n
g
e
x
c
el
l
e
n
t
ca
p
a
b
i
l
iti
e
s
f
o
r
b
o
t
h
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
an
d
class
if
icatio
n
ac
cu
r
ac
y
,
m
ai
n
ly
in
th
e
ap
p
licati
o
n
s
p
ac
e
o
f
b
r
ain
tu
m
o
r
lo
ca
ti
o
n
.
2
.
3
.
1
.
VG
G
1
9
B
a
c
k
bo
ne
T
h
e
Neu
r
o
Net1
9
m
o
d
el
f
u
n
d
am
en
tally
b
u
ild
s
on
th
e
ar
c
h
itectu
r
e
of
VGG1
9
.
VGG1
9
is
th
e
o
n
e
am
o
n
g
p
r
o
m
i
n
en
t
n
e
u
r
al
n
et
wo
r
k
s
u
s
ed
f
o
r
im
ag
e
class
if
icatio
n
.
T
h
e
VGG1
9
m
o
d
el
c
o
m
p
r
is
es
m
ain
ly
1
9
weig
h
ted
lay
e
r
s
co
u
n
tin
g
co
n
v
o
lu
tio
n
al
lay
e
r
s
(
1
6
lay
er
s
),
f
u
lly
co
n
n
ec
ted
lay
e
r
s
(3
la
y
e
r
s
),
an
d
th
e
p
o
o
lin
g
lay
er
s
.
Stru
ctu
r
e
is
s
h
o
wn
in
Fig
u
r
e
3
.
C
o
n
v
o
lu
tio
n
al
lay
er
s
(
1
6
lay
e
r
s
)
:
VGG1
9
b
ac
k
b
o
n
es
s
tar
t
with
1
6
co
n
v
o
lu
tio
n
al
lay
er
s
.
I
t
is
th
ese
lay
er
s
th
at
ca
n
b
e
u
s
ed
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
[
2
4
]
.
C
o
n
v
o
lu
t
io
n
al
f
ilter
s
ar
e
ap
p
lie
d
to
th
e
in
p
u
t
MRI
im
ag
es
,
wh
ich
p
er
m
its
to
d
etec
t
f
u
n
d
am
en
tal
v
is
u
al
p
atter
n
s
-
ed
g
es,
s
u
r
f
ac
es,
an
d
s
tr
aig
h
tf
o
r
war
d
s
h
ap
es
with
in
th
e
in
itial
lay
er
s
.
Ho
wev
er
,
as
th
is
d
ata
g
o
es
d
ee
p
er
th
r
o
u
g
h
th
e
n
etwo
r
k
,
th
ese
lay
e
r
s
s
tar
t
to
ca
p
tu
r
e
m
o
r
e
co
m
p
le
x
an
d
ab
s
tr
ac
ted
p
atter
n
s
[
2
5
]
.
T
h
is
p
r
o
ce
s
s
lets
th
e
m
o
d
el
s
tead
ily
co
n
s
tr
u
ct
a
h
i
g
h
-
lev
el
u
n
d
er
s
tan
d
in
g
o
f
th
e
im
ag
e
th
at
is
in
ev
itab
le
to
r
ec
o
g
n
ize
tu
m
o
r
-
r
elate
d
f
ea
tu
r
es
in
b
r
ain
MRIs.
Max
-
p
o
o
lin
g
lay
er
s
(
5
lay
er
s
)
:
Af
ter
ea
ch
co
n
v
o
lu
tio
n
al
la
y
er
,
m
ax
-
p
o
o
lin
g
lay
er
s
ar
e
u
s
ed
.
T
h
e
m
ax
-
p
o
o
lin
g
lay
er
s
d
ec
r
ea
s
e
s
p
atial
m
ea
s
u
r
em
en
ts
ab
o
u
t
i
n
clu
d
e
m
ap
s
o
r
ig
in
ated
f
r
o
m
th
e
c
o
n
v
o
lu
tio
n
al
lay
er
s
.
Max
-
p
o
o
lin
g
k
ee
p
s
th
e
m
o
s
t
p
r
o
m
in
en
t
f
ea
tu
r
es
an
d
d
o
wn
s
am
p
les
th
e
d
ata.
T
h
is
r
ed
u
ce
s
th
e
am
o
u
n
t
o
f
c
o
m
p
u
tatio
n
r
eq
u
ir
e
d
to
p
r
ed
ict
th
e
v
alu
es
o
f
th
e
d
ata
p
o
in
ts
wh
ile
also
en
s
u
r
in
g
th
e
r
e
is
less
o
v
er
f
itti
n
g
.
Max
-
p
o
o
lin
g
also
im
p
r
o
v
es in
v
ar
ian
ce
o
f
th
e
m
o
d
el
to
s
m
all
s
h
if
ts
o
r
d
is
to
r
tio
n
s
o
f
t
h
e
in
p
u
t im
ag
es
[
2
6
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
3
2
2
6
-
3
2
3
7
3230
Fu
lly
co
n
n
ec
ted
lay
er
s
(3
lay
e
r
s
)
:
No
w,
af
ter
f
ea
tu
r
e
ex
tr
ac
ti
o
n
,
t
h
e
y
ield
f
r
o
m
th
e
co
n
v
o
lu
tio
n
al
an
d
p
o
o
lin
g
lay
er
s
is
s
tr
aig
h
ten
ed
an
d
p
ass
ed
th
r
o
u
g
h
th
r
ee
c
o
m
p
letely
ass
o
ciate
d
lay
er
s
.
T
h
ese
lay
er
s
p
er
f
o
r
m
h
ig
h
-
lev
el
th
i
n
k
in
g
an
d
d
ec
is
io
n
-
m
ak
i
n
g
b
ased
o
n
t
h
e
h
ig
h
li
g
h
ts
ex
tr
icate
d
f
r
o
m
th
e
MRI
p
ictu
r
es
[
2
7
]
.
Af
te
r
th
ese
p
r
o
ce
s
s
es,
p
r
o
p
er
f
u
ll e
x
tr
ac
tio
n
tak
es p
lace
to
m
a
p
th
e
f
in
al
f
ea
tu
r
es to
war
d
tu
m
o
r
cl
ass
if
icatio
n
.
Fig
u
r
e
3.
Ar
c
h
itectu
r
e
of
V
G
G
1
9
C
o
n
v
o
l
u
t
i
o
n
h
a
n
d
l
e
i
s
c
o
n
n
e
c
t
i
n
g
to
t
h
e
i
n
p
u
t
p
i
c
t
u
r
e
‘
M’
u
s
i
n
g
c
h
a
n
n
e
l
‘
F
’
.
R
e
g
a
r
d
i
n
g
e
v
e
r
y
c
o
o
r
d
i
n
a
t
e
(
,
):
,
=
(
⊙
)
(4
)
,
allu
d
es
to
esteem
d
etec
ted
at
ar
r
a
n
g
es
(
,
)
in
(
4
)
,
with
in
co
n
s
eq
u
en
t
n
etwo
r
k
‘
C
’
,
w
h
er
ea
s
s
p
ea
k
s
to
th
e
s
eg
m
e
n
t
o
f
in
p
u
t
g
r
i
d
‘
M
’
wh
ic
h
is
p
r
esen
tly
d
e
f
en
d
e
d
th
r
o
u
g
h
th
e
r
o
u
te.
E
v
e
r
y
co
n
v
o
lu
tio
n
lay
er
in
ter
f
ac
es d
if
f
er
en
t r
o
u
te
s
f
o
r
ex
tr
icatin
g
u
n
m
is
tak
ab
le
f
ea
tu
r
es a
m
o
n
g
in
p
u
t in
f
o
r
m
at
io
n
.
So
f
tMa
x
a
ctiv
atio
n
:
Af
ter
a
p
p
ly
in
g
th
e
last
f
u
lly
c
o
n
n
ec
t
ed
lay
er
,
th
er
e
is
a
So
f
tMa
x
a
ctiv
atio
n
o
p
er
atio
n
.
T
h
is
So
f
tMa
x
o
p
er
atio
n
is
a
m
ath
em
atica
l
o
p
er
at
io
n
wh
ich
tu
r
n
s
th
e
o
u
tp
u
t
o
f
t
h
is
n
eu
r
al
n
etwo
r
k
in
to
a
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
.
Her
e,
to
o
,
it
cr
ea
tes
a
p
r
o
b
ab
ilit
y
f
o
r
ea
ch
o
f
th
e
th
r
ee
class
es
an
d
th
e
n
eg
ativ
e
class
,
wh
ich
is
n
o
tu
m
o
r
.
T
h
e
m
o
d
el
co
n
clu
d
es its
p
r
ed
ictio
n
b
y
tak
in
g
th
e
class
with
m
ax
i
m
u
m
p
r
o
b
ab
ilit
y
.
(
)
=
∑
=
(5
)
W
e
u
tili
ze
th
e
ca
lcu
latio
n
to
ch
an
g
e
o
v
er
s
o
r
ted
o
u
t
y
ield
s
in
to
K
-
class
lik
elih
o
o
d
co
n
v
e
y
an
ce
s
.
I
n
(
5
)
th
e
n
u
m
er
ical
s
tead
y
E
u
ler
s
n
u
m
b
er
(
e
)
s
p
ea
k
s
to
in
p
u
t
t
o
ev
er
y
p
ath
,
in
d
icate
d
as
.
E
x
p
o
n
e
n
tial
esteem
ab
o
u
t
a
b
d
icate
s
ap
p
r
aise
f
o
r
ea
ch
less
o
n
is
d
ec
id
ed
u
tili
zin
g
th
e
e
q
u
atio
n
.
Ad
d
in
g
in
s
titu
tio
n
alize
s
th
e
So
f
tMa
x
y
ield
to
e
v
er
y
class
‘
’
(
to
o
ca
lled
(
)
)
in
co
n
n
ec
tio
n
with
y
ield
s
to
ev
er
y
o
th
er
cl
a
ss
.
2
.
3
.
2
.
I
P
P
M
T
h
e
in
v
e
r
ted
p
y
r
am
i
d
p
o
o
lin
g
m
o
d
u
le
is
p
r
o
p
o
s
ed
f
o
r
f
u
r
th
er
im
p
r
o
v
em
en
t
u
p
o
n
th
e
ab
il
ity
o
f
t
h
e
f
ea
tu
r
e
ex
tr
ac
to
r
VGG1
9
.
Fu
r
th
er
ex
ten
s
io
n
o
f
th
is
m
o
d
el
’
s
ex
p
er
tis
e
to
ca
p
tu
r
e
f
ea
tu
r
es
t
h
r
o
u
g
h
o
u
t
s
ca
les
is
also
im
p
o
r
tan
t
f
o
r
tu
m
o
r
i
d
en
tific
atio
n
m
ain
ly
b
ased
o
n
s
ize
an
d
s
h
ap
e
d
if
f
er
en
ce
s
.
T
h
e
I
PP
M
em
p
lo
y
s
th
e
p
o
o
lin
g
with
th
e
win
d
o
w
s
izes
of
2
×
2,
3
×
3,
4
×
4,
an
d
6
×
6,
wh
ich
en
ab
les
m
u
lti
-
s
ca
le
p
o
o
lin
g
.
T
h
is
allo
ws
th
e
m
o
d
el
to
ca
p
tu
r
e
s
ev
er
al
lay
er
s
o
f
s
p
atial
in
f
o
r
m
atio
n
.
T
h
e
s
m
all
s
ize
s
o
f
th
e
p
o
o
lin
g
ar
e
lik
ely
to
ca
p
tu
r
e
th
e
f
in
er
d
etails,
in
th
is
ca
s
e,
th
e
e
d
g
es
of
th
e
tu
m
o
r
,
wh
er
ea
s
th
e
lar
g
er
p
o
o
l
s
izes
ca
p
tu
r
e
in
f
o
r
m
atio
n
at
a
g
lo
b
a
l
lev
el,
th
u
s
o
u
tlin
in
g
th
e
g
en
er
al
s
h
ap
e
o
f
th
e
tu
m
o
r
[
2
8
]
.
T
h
is
is
p
ar
ticu
lar
ly
im
p
o
r
t
an
t
in
b
r
ain
tu
m
o
r
d
etec
tio
n
s
in
ce
s
izes c
o
u
ld
b
e
g
r
ea
tly
v
ar
ie
d
an
d
s
o
ca
n
s
tr
u
c
tu
r
es.
(
,
)
=
′
′
,
′
′
(6
)
‘
’
s
tan
d
s
f
o
r
t
h
e
ten
s
o
r
in
p
u
t
i
n
(
6
)
,
in
th
is
ca
s
e
it
is
eq
u
ally
m
atch
in
g
with
th
e
v
alu
e
o
f
p
i
x
el
at
th
e
p
o
s
itio
n
(
,
)
of
u
p
s
am
p
led
r
esu
lt.
‘
’
is
u
p
s
am
p
lin
g
o
p
er
atio
n
.
Fo
r
th
e
p
u
r
p
o
s
es
of
th
e
o
p
er
atio
n
,
th
e
s
izes
o
f
p
o
o
l
f
u
r
n
is
h
ed
b
y
Neu
r
o
Net1
9
a
r
e
u
s
ed
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
o
b
u
s
t d
ee
p
lea
r
n
in
g
a
p
p
r
o
a
ch
fo
r
a
cc
u
r
a
te
d
etec
tio
n
o
f b
r
a
in
tu
mo
r
a
n
d
a
n
a
lysi
s
(
La
n
ke
P
a
lla
vi
)
3231
Featu
r
e
r
ef
in
em
e
n
t:
f
ea
tu
r
es
ex
tr
ac
ted
b
y
th
e
VGG1
9
b
ac
k
b
o
n
e
co
n
tain
f
u
n
d
am
e
n
tal
v
is
u
al
in
f
o
r
m
atio
n
.
H
o
we
v
er
,
th
ey
m
ay
n
o
t
b
e
en
o
u
g
h
to
in
c
o
r
p
o
r
ate
all
th
e
p
o
s
s
ib
le
p
atter
n
s
f
o
r
th
e
p
r
o
p
er
class
if
icatio
n
o
f
tu
m
o
r
s
.
T
h
a
t
is
wh
er
e
t
h
e
I
PP
M
o
p
er
ate
d
m
u
lti
-
s
ca
le
p
o
o
lin
g
to
war
d
th
e
e
x
p
r
ess
io
n
a
n
d
co
m
b
in
atio
n
o
f
f
ea
tu
r
es
lear
n
ed
ac
r
o
s
s
d
if
f
e
r
en
t
s
ca
les
with
a
v
iew
t
o
im
p
r
o
v
is
in
g
th
e
c
o
n
ce
p
t
ca
p
ab
ilit
y
o
f
th
is
m
o
d
el
o
v
e
r
v
ar
io
u
s
class
es o
f
tu
m
o
r
s
an
d
MRI
s
ca
n
v
ar
i
atio
n
s
.
I
n
(
7
)
,
t
h
e
co
m
m
u
n
icatio
n
o
f
m
ix
o
f
in
co
r
p
o
r
ate
m
ap
s
a
m
o
n
g
u
n
m
is
tak
ab
le
lay
e
r
s
f
r
o
m
I
PP
M
is
s
h
o
wed
u
p
:
=
(
1
,
2
,
3
,
.
.
.
,
)
(
7
)
Her
e,
th
e
in
teg
r
ated
ab
d
icate
is
d
em
o
n
s
tr
ated
as
‘
C
’
,
an
d
m
ix
in
g
o
f
s
p
o
tlig
h
t
m
ap
s
co
n
s
o
lid
ates
1
,
2
,
3
,
.
.
.
,
.
T
h
ese
m
ap
s
ar
e
f
r
o
m
s
o
m
e
lev
els
o
f
wo
r
ld
wid
e
o
p
e
n
ar
r
an
g
em
e
n
t
o
r
g
a
n
izatio
n
.
T
h
e
u
tili
ze
o
f
a
co
u
p
le
o
f
u
p
s
am
p
lin
g
s
tag
es
an
d
r
ea
s
o
n
ab
le
s
tr
ateg
ies
o
f
p
o
o
lin
g
in
t
h
e
I
PP
M
o
f
Neu
r
o
N
et1
9
g
r
an
ts
th
em
to
r
ea
s
o
n
ab
ly
ca
tch
a
co
n
tr
asti
n
g
am
p
lify
a
b
o
u
t
h
ig
h
lig
h
ts
.
2
.
3
.
3
.
P
G
G
AN
f
o
r
da
t
a
a
u
g
m
e
n
t
a
t
i
o
n
T
h
e
m
ajo
r
c
h
allen
g
e
in
tr
ai
n
in
g
th
e
d
ee
p
lear
n
in
g
m
o
d
els
in
m
ed
ical
im
ag
e
an
al
y
s
is
is
th
at
no
an
n
o
tated
d
ata
is
av
ailab
le
[
2
9
]
.
Fo
r
th
is
p
r
o
b
lem
,
PGGAN
is
u
tili
ze
d
to
cr
ea
te
s
y
n
th
etic
MRI
im
ag
es
to
ac
t
as
an
au
g
m
en
tatio
n
f
o
r
th
e
ex
i
s
tin
g
tr
ain
in
g
d
ataset.
T
h
e
r
e
ar
e
two
m
ajo
r
co
m
p
o
n
en
ts
to
P
GGAN
ar
ch
itectu
r
e
as
s
h
o
wn
in
Fig
u
r
e
4:
th
e
g
en
er
ato
r
an
d
th
e
d
is
cr
im
in
ato
r
.
T
h
ese
co
m
p
lem
en
t
ea
ch
o
t
h
er
p
er
f
ec
tly
wh
e
n
cr
ea
tin
g
an
d
ev
alu
atin
g
ar
tific
i
ally
p
r
o
d
u
ce
d
MRI
-
s
ca
n
n
e
d
i
m
ag
es.
Fig
u
r
e
4.
Ar
c
h
itectu
r
e
of
P
G
G
A
N
T
h
e
g
e
n
e
r
a
t
o
r
h
a
s
a
n
i
n
f
l
u
e
n
t
i
a
l
r
o
l
e
i
n
g
e
n
e
r
a
ti
n
g
n
o
v
e
l
,
s
y
n
t
h
e
t
i
c
MR
I
i
m
a
g
es
w
h
i
ch
c
l
o
s
e
l
y
r
e
s
e
m
b
l
e
l
e
g
it
i
m
at
e
M
R
I
d
a
t
a;
i
t
b
e
g
i
n
s
w
i
t
h
a
l
o
w
-
r
es
o
l
u
t
i
o
n
i
m
a
g
e
a
n
d
d
e
v
e
l
o
p
s
f
r
o
m
t
h
i
s
o
v
e
r
t
i
m
e
.
D
u
r
i
n
g
t
h
e
c
o
u
r
s
e
o
f
t
h
e
t
r
ai
n
i
n
g
p
r
o
c
es
s
,
t
h
e
g
e
n
e
r
a
t
o
r
l
e
a
r
n
s
h
o
w
t
o
d
e
v
e
l
o
p
v
e
r
y
r
e
a
l
is
t
ic
i
m
a
g
es
,
w
h
e
r
e
i
t g
a
i
n
s
f
i
n
e
-
g
r
a
i
n
e
d
p
o
i
n
t
s
o
f
i
n
t
e
r
e
s
t
a
n
d
c
o
m
p
l
e
x
p
a
t
t
e
r
n
s
i
n
M
R
I
p
i
ct
u
r
e
s
.
T
h
e
d
i
s
c
r
i
m
i
n
at
o
r
a
s
s
e
s
s
es
t
h
e
i
m
a
g
e
s
c
r
e
a
t
e
d
b
y
t
h
e
g
e
n
e
r
a
t
o
r
.
I
t
d
i
f
f
e
r
e
n
tia
t
e
s
b
et
w
e
e
n
a
u
t
h
e
n
t
i
c
a
n
d
s
y
n
t
h
e
t
i
c
i
m
a
g
es
,
o
f
f
e
r
i
n
g
c
r
u
ci
a
l
f
e
e
d
b
a
c
k
t
o
t
h
e
g
e
n
e
r
a
t
o
r
.
T
h
e
p
r
i
m
a
r
y
a
i
m
o
f
t
h
e
d
i
s
c
r
i
m
i
n
a
t
o
r
is
t
o
c
o
r
r
e
c
tl
y
i
d
e
n
t
i
f
y
i
m
a
g
es
a
s
ei
t
h
e
r
r
ea
l
o
r
f
a
k
e
,
t
h
e
r
e
b
y
a
s
s
i
s
t
i
n
g
t
h
e
g
e
n
e
r
a
t
o
r
i
n
h
o
n
i
n
g
i
t
s
o
u
t
p
u
t
a
n
d
e
n
h
a
n
c
i
n
g
t
h
e
r
e
a
l
i
s
m
o
f
t
h
e
i
m
a
g
es
i
t
p
r
o
d
u
ce
s
[
3
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
3
2
2
6
-
3
2
3
7
3232
PGGAN
u
s
es
p
r
o
g
r
ess
iv
e
tr
ain
in
g
.
I
t
b
e
g
in
s
s
o
lv
in
g
g
e
n
er
ated
im
ag
es
at
lo
w
r
eso
lu
tio
n
an
d
g
r
ad
u
ally
in
cr
ea
s
es
th
e
r
eso
lu
tio
n
.
T
h
is
way
,
it
a
v
o
id
s
tr
ain
i
n
g
in
s
tab
ilit
y
but
lets
th
e
m
o
d
el
ca
p
tu
r
e
co
m
p
le
x
f
ea
tu
r
es
o
n
h
ig
h
er
r
eso
lu
tio
n
s
.
Yet
an
o
th
er
g
u
ar
an
tee
wh
ic
h
is
p
r
o
g
r
ess
iv
e
f
o
r
PGGAN
is
th
at
th
e
g
en
er
ato
r
im
p
r
o
v
es
t
h
e
a
b
i
l
i
t
y
to
g
e
n
e
r
a
te
q
u
a
l
i
t
y
i
m
a
g
e
s
o
v
e
r
t
i
m
e
.
T
h
e
i
m
a
g
es
t
h
a
t
r
e
s
u
l
t
f
r
o
m
P
G
GA
N
a
p
p
e
a
r
to
b
l
e
n
d
v
e
r
y
w
e
l
l
w
i
t
h
th
e
tr
ain
in
g
s
et,
th
er
eb
y
s
u
p
p
lem
e
n
tin
g
it
a
n
d
m
ak
i
n
g
it
r
ic
h
er
wh
ile
als
o
m
ak
in
g
a
g
r
ea
ter
v
ar
iety
of
ex
a
m
p
les
av
ailab
le
f
o
r
Ne
u
r
o
Net1
9
to
a
b
s
o
r
b
.
T
h
is
au
g
m
e
n
tatio
n
s
ee
m
s
to
aid
in
b
etter
g
en
er
aliza
tio
n
esp
ec
ially
wh
e
n
f
ac
ed
b
y
d
e
f
icien
cy
o
f
ac
tu
a
l d
ata
av
ailab
le.
2
.
4
.
Seg
m
ent
a
t
io
n
wit
h
M
a
s
k
R
-
CNN
Ma
s
k
R
-
C
NN
is
a
s
tr
o
n
g
m
o
d
el
f
o
r
ca
n
ce
r
o
u
s
tu
m
o
r
d
etec
tio
n
an
d
s
p
atial
s
eg
m
e
n
tatio
n
b
y
u
s
in
g
MRI
im
ag
es.
I
t
p
r
o
d
u
ce
s
h
i
g
h
-
ac
cu
r
ac
y
s
eg
m
en
tin
g
m
ask
s
d
elin
ea
tin
g
tu
m
o
r
b
o
u
n
d
ar
ie
s
,
th
er
ef
o
r
e
b
ein
g
v
er
y
v
al
u
ab
le
f
o
r
v
er
y
s
en
s
itiv
e
m
ed
ical
ap
p
licatio
n
s
r
eq
u
ir
i
n
g
ex
ce
llen
t
d
etail.
T
h
is
m
ea
n
s
th
e
m
o
d
el
s
tar
ts
with
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
(
R
PN)
,
wh
ich
id
en
tifie
s
wh
er
e
p
o
s
s
ib
le
tu
m
o
r
s
m
i
g
h
t
be
lo
ca
ted
,
a
n
d
it
p
r
o
d
u
ce
s
b
o
u
n
d
i
n
g
b
o
x
es
b
as
ed
on
f
ea
tu
r
es
f
r
o
m
th
e
im
ag
e,
lik
e
ir
r
e
g
u
lar
ities
o
r
a
b
n
o
r
m
al
p
atter
n
s
.
T
h
e
p
r
o
p
o
s
als ar
e
r
ef
in
ed
in
two
s
tag
es:
B
o
u
n
d
in
g
b
o
x
es a
r
e
ad
j
u
s
ted
p
r
ec
is
ely
f
o
r
ac
cu
r
ac
y
,
t
h
en
a
b
in
ar
y
m
ask
to
estab
lis
h
ex
ac
tly
wh
er
e
th
e
tu
m
o
r
m
ig
h
t
b
e.
T
h
e
p
ix
el
-
le
v
el
s
eg
m
en
tatio
n
m
ask
s
p
r
o
v
i
d
ed
in
th
e
o
u
tp
u
t
en
ab
le
p
r
ec
is
e
id
en
tific
atio
n
o
f
tu
m
o
r
s
,
d
if
f
er
e
n
tiatio
n
of
m
u
ltip
le
tu
m
o
r
s
,
an
d
a
d
v
an
ce
d
an
aly
s
es
lik
e
v
o
lu
m
e
esti
m
atio
n
or
tr
ea
t
m
en
t
p
lan
n
in
g
.
Ma
s
k
R
-
C
NN’
s
ab
ilit
y
to
ac
h
iev
e
h
ig
h
-
ac
cu
r
ac
y
in
s
tan
ce
s
eg
m
en
tatio
n
m
ak
es
it
an
ex
ce
llen
t
ch
o
ice
f
o
r
v
ar
io
u
s
m
e
d
ica
l
im
ag
in
g
ap
p
licatio
n
s
.
F
i
n
a
l
l
y
,
t
h
e
t
u
m
o
r
d
e
t
e
c
t
i
o
n
s
y
s
t
e
m
f
o
l
l
o
w
s
a
s
t
r
u
c
t
u
r
e
d
w
o
r
k
f
l
o
w
,
b
e
g
i
n
n
i
n
g
w
i
t
h
p
r
e
p
r
o
c
e
s
s
i
n
g
s
t
e
p
s
s
u
c
h
a
s
n
o
r
m
a
l
i
z
a
t
i
o
n
,
s
k
u
l
l
s
t
r
i
p
p
i
n
g
,
a
n
d
r
e
s
i
z
i
n
g
o
f
M
R
I
i
m
a
g
e
s
.
T
h
e
N
e
u
r
o
N
e
t
1
9
m
o
d
e
l
i
s
t
h
e
n
t
r
a
i
n
e
d
o
n
t
h
e
s
e
p
r
e
p
r
o
c
e
s
s
e
d
i
m
a
g
e
s
,
i
n
c
o
r
p
o
r
a
t
i
n
g
s
y
n
t
h
e
t
i
c
d
a
t
a
g
e
n
e
r
a
t
e
d
b
y
P
G
G
A
N
t
o
e
n
h
a
n
c
e
d
a
t
a
s
e
t
d
i
v
e
r
s
i
t
y
.
M
a
s
k
R
-
C
N
N
i
s
u
t
i
l
i
z
e
d
f
o
r
p
r
e
c
i
s
e
s
e
g
m
e
n
t
a
t
i
o
n
o
f
t
u
m
o
r
r
e
g
i
o
n
s
w
i
t
h
i
n
t
h
e
i
m
a
g
e
s
,
a
f
t
e
r
w
h
i
c
h
t
h
e
s
y
s
t
e
m
c
l
a
s
s
i
f
i
e
s
t
h
e
d
e
t
e
c
t
e
d
t
u
m
o
r
s
i
n
t
o
a
n
y
o
n
e
o
f
f
o
u
r
c
l
a
s
s
e
s
:
g
l
i
o
m
a
,
m
e
n
i
n
g
i
o
m
a
,
p
i
t
u
i
t
a
r
y
,
a
n
d
n
o
t
u
m
o
r
.
2
.
5
.
P
er
f
o
rma
nce
a
na
ly
s
is
We
ev
alu
ate
p
r
ed
ictio
n
of
ea
c
h
ca
teg
o
r
y
b
ased
on
th
e
ass
ess
m
en
t
cr
iter
ia
co
u
n
tin
g
F1
-
s
co
r
e,
C
o
h
en
’
s
k
ap
p
a,
r
e
-
ca
ll.
W
e
also
ca
lcu
late
ev
er
y
ex
ec
u
tio
n
p
o
in
ter
u
tili
zin
g
th
e
p
e
r
p
lex
ity
o
f
f
o
u
r
-
ce
ll
cr
o
s
s
s
ec
tio
n
:
a.
Gen
u
in
e
p
o
s
itiv
e
(
T
P):
Occ
asio
n
s
in
wh
ich
,
th
e
ap
p
ea
r
p
r
ec
i
s
ely
d
ec
id
es
th
e
tu
m
o
r
’
s
co
r
r
e
ct
lo
ca
tio
n
.
b.
Gen
u
in
e
n
eg
ativ
e
(
T
N)
:
C
ir
cu
m
s
tan
ce
s
wh
er
e,
th
e
co
m
p
u
ter
p
r
o
g
r
am
p
r
ec
is
ely
f
o
r
ec
asts
th
at
th
e
tu
mor
is
not
p
r
e
s
e
n
t
.
c.
Un
tr
u
e
p
o
s
itiv
e
(
FP
)
:
C
ir
cu
m
s
tan
ce
s
wh
er
e,
th
e
ap
p
ea
r
er
r
o
n
e
o
u
s
ly
ca
teg
o
r
ize
t
h
e
c
h
ec
k
of
s
o
u
n
d
as
t
u
m
o
r
.
d.
Un
tr
u
e
n
e
g
ativ
e
(
FN)
:
C
ir
cu
m
s
tan
ce
s
wh
er
e,
th
e
a
p
p
ea
r
er
r
o
n
e
o
u
s
ly
ig
n
o
r
es
th
e
t
u
m
o
r
,
d
r
iv
in
g
it
i
s
ca
teg
o
r
izatio
n
as
s
o
u
n
d
c
h
a
n
n
e
l
.
2
.
5
.
1
.
Acc
ura
cy
Acc
u
r
ac
y
m
ea
s
u
r
es
th
e
p
er
f
o
r
m
an
ce
o
f
a
m
o
d
el
b
y
s
u
g
g
esti
n
g
n
u
m
b
e
r
o
f
p
r
ec
is
ely
f
o
r
ec
asted
in
s
tan
ce
s
an
d
th
e
wh
o
le
in
s
tan
ce
s
.
I
t
is
at
v
ar
io
u
s
p
o
in
ts
in
th
e
d
ataset
an
im
p
o
r
tan
t
m
ea
s
u
r
e
of
m
o
d
el
ef
f
icien
cy
,
an
d
a
h
ig
h
er
v
alu
e
o
f
ac
cu
r
ac
y
m
ea
n
s
b
etter
p
er
f
o
r
m
a
n
ce
an
d
d
ec
r
ea
s
ed
p
r
ed
ictio
n
er
r
o
r
s
.
W
e
ex
ec
u
te
th
e
ca
lcu
latio
n
u
s
in
g
t
h
e
f
o
r
m
u
la
(
8
)
.
=
+
+
+
+
(
8
)
2
.
5
.
2
.
P
re
cisi
o
n
Pre
cisi
o
n
m
ea
s
u
r
es
how
ac
cu
r
ately
th
e
m
o
d
el
id
en
tifie
s
th
e
p
o
s
itiv
e
ca
s
es
in
all
of
th
e
id
en
tifie
d
p
o
s
itiv
es,
wh
ich
is
an
im
p
o
r
tan
t
asp
ec
t
in
r
ed
u
cin
g
f
als
e
p
o
s
itiv
es
(
FP
)
.
Hig
h
p
r
ec
is
io
n
im
p
lies
th
at,
d
em
o
n
s
tr
ate
h
as
p
r
ec
is
ely
r
e
co
g
n
ized
th
e
p
o
s
itiv
e
ca
s
es,
an
d
th
e
r
ef
o
r
e
p
r
e
d
ictio
n
will
b
e
r
elia
b
le.
I
t
is
ch
ar
ac
ter
ized
as r
ep
r
esen
ted
i
n
(
9
)
.
=
+
(
9
)
2
.
5
.
3
.
Rec
a
ll
R
ec
all
o
r
s
en
s
itiv
ity
m
ea
s
u
r
es
h
o
w
well
a
m
o
d
el
class
if
ies
wh
at
ar
e
k
n
o
wn
to
b
e
tr
u
e
p
o
s
itiv
es
-
it
m
in
im
izes
th
e
f
alse
n
eg
ativ
e
(
FN)
o
f
all
ac
tu
al
p
o
s
itiv
es
-
o
u
t
o
f
all.
R
ec
all
is
cr
itical
f
o
r
ap
p
licatio
n
s
wh
er
e
f
ailu
r
e
to
class
if
y
p
o
s
itiv
e
in
s
tan
ce
s
m
ea
n
s
th
e
en
d
as sh
o
wn
in
(
1
0
)
.
=
+
(
1
0
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
o
b
u
s
t d
ee
p
lea
r
n
in
g
a
p
p
r
o
a
ch
fo
r
a
cc
u
r
a
te
d
etec
tio
n
o
f b
r
a
in
tu
mo
r
a
n
d
a
n
a
lysi
s
(
La
n
ke
P
a
lla
vi
)
3233
2
.
5
.
4
.
F1
-
s
co
re
E
q
u
atio
n
(
1
1
)
ca
lc
u
lates
th
e
F1
-
s
co
r
e,
in
wh
ic
h
r
ec
all
an
d
p
r
ec
is
io
n
b
alan
ce
s
o
u
t
ea
ch
o
th
er
,
wh
ich
m
ak
es
it
a
v
er
y
u
s
ef
u
l
m
ea
s
u
r
e
to
ch
ec
k
m
o
d
els
o
n
im
b
alan
ce
d
d
atasets
.
T
h
e
lar
g
er
t
h
e
F1
-
s
co
r
e
,
t
h
e
b
etter
th
e
tr
ad
e
-
o
f
f
b
etwe
en
r
ec
all
an
d
p
r
ec
is
io
n
.
1
−
=
2
×
(
×
)
×
(
1
1
)
2
.
5
.
5
.
Co
hen’s
K
a
pp
a
co
ef
f
ic
ient
It
ev
alu
ates
th
e
d
eg
r
ee
of
c
o
n
co
r
d
an
ce
am
o
n
g
m
o
d
el’
s
f
o
r
ec
asts
an
d
g
en
u
in
e
n
am
es.
Scien
tific
o
p
er
atio
n
is
r
e
p
r
esen
ted
as (
1
2
)
.
=
(
−
)
(
1
−
)
(
1
2
)
T
h
e
Kap
p
a
co
e
f
f
icien
t
ca
lcu
la
tes
th
e
d
eg
r
ee
in
wh
ich
g
e
n
u
i
n
e
co
m
p
r
e
h
en
s
io
n
o
u
tf
lan
k
s
i
r
r
eg
u
lar
p
o
s
s
ib
ilit
y
.
Gen
u
in
e
v
alu
es
d
em
o
n
s
tr
ate
tall
s
tag
e
ab
o
u
t
in
f
o
r
m
atio
n
,
s
o
lid
ce
r
tific
atio
n
,
an
d
to
tal
n
ee
d
ab
o
u
t
m
is
co
n
ce
p
tio
n
;
less
v
al
u
es
s
h
o
w
p
r
o
f
itab
le
a
b
n
o
r
m
ality
.
E
x
p
ec
ted
co
n
s
en
t
is
s
p
o
k
e
n
to
b
y
th
e
v
a
r
iab
le
p
e
,
th
o
u
g
h
watc
h
ed
u
n
d
er
s
tan
d
in
g
is
s
p
o
k
en
to
by
th
e
v
ar
iab
le
p
o
.
T
h
e
p
er
f
o
r
m
an
ce
an
al
y
s
is
is
s
h
o
wn
in
Fig
u
r
e
5
.
Fig
u
r
e
5.
C
o
n
f
u
s
io
n
m
atr
ix
f
o
r
N
e
u
r
o
N
e
t
1
9
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
Neu
r
o
Net1
9
can
id
en
tif
y
t
u
m
o
r
s
in
t
h
e
b
r
ain
with
e
x
ce
llen
t
a
d
v
an
tag
es,
esp
ec
ially
wh
e
n
c
o
m
p
ar
ed
to
o
th
er
lea
d
in
g
m
o
d
els.
T
h
is
f
r
am
ewo
r
k
g
ets
test
a
cc
u
r
ac
y
o
f
9
9
.
5
%
a
n
d
v
alid
atio
n
ac
cu
r
ac
y
o
f
9
7
.
9
%.
I
t
is
in
d
ee
d
th
e
f
ir
s
t
ar
ch
itectu
r
e
ab
le
to
o
u
tp
er
f
o
r
m
lead
er
s
lik
e
Den
s
eNe
t1
2
1
,
Mo
b
ileNetV2
,
R
esNet5
0
,
VGG1
6
,
an
d
VGG1
9
.
Fu
r
th
er
m
o
r
e,
ea
ch
o
f
its
ev
alu
atio
n
m
etr
ics:
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e,
k
ee
p
s
r
is
in
g
ab
o
v
e
9
8
% o
n
all
tu
m
o
r
ty
p
es,
n
am
el
y
m
en
in
g
io
m
a,
g
lio
m
a,
p
itu
ita
r
y
,
an
d
n
o
tu
m
o
r
Oth
er
m
o
d
els,
s
u
ch
as
Den
s
e
Net1
2
1
an
d
Mo
b
ileNetV2
,
p
e
r
f
o
r
m
d
if
f
er
e
n
tly
o
n
test
s
,
y
iel
d
in
g
9
0
.
2
%
an
d
9
2
.
3
%,
wh
ile
R
esNe
t5
0
an
d
VGG1
6
ar
e
m
o
d
er
atel
y
p
e
r
f
o
r
m
in
g
,
with
test
ac
cu
r
ac
ies
9
3
.
5
%
an
d
9
5
.
6
%.
Desp
ite
th
e
p
r
ev
alen
ce
o
f
s
u
c
h
m
o
d
els
in
task
s
ap
p
lied
in
m
ed
ical
im
ag
in
g
,
Ne
u
r
o
Net1
9
s
ee
m
s
to
o
u
tp
er
f
o
r
m
th
em
by
p
er
f
o
r
m
in
g
well
on
a
cc
o
u
n
t
of
its
o
p
tim
ized
ar
ch
ite
ctu
r
e,
wh
ich
in
teg
r
ally
in
clu
d
e
s
th
e
ap
p
licatio
n
o
f
VGG1
9
f
ea
tu
r
es
an
d
in
v
e
r
ted
p
y
r
am
id
p
o
o
lin
g
m
o
d
u
le
.
T
h
is
u
n
iq
u
e
d
esig
n
en
h
an
ce
s
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
s
p
atial
r
eso
lu
tio
n
,
wh
ich
ar
e
c
r
itical
f
o
r
ac
cu
r
ately
id
en
tify
i
n
g
tu
m
o
r
s
,
e
v
en
in
co
m
p
le
x
s
ce
n
ar
io
s
.
T
h
e
s
o
p
h
is
ticated
ar
ch
itectu
r
e,
esp
ec
ially
th
e
I
PP
M
of
N
eu
r
o
Net1
9
,
m
u
ch
en
h
an
ce
s
m
u
lti
-
s
ca
le
f
ea
tu
r
e
ag
g
r
e
g
atio
n
f
o
r
th
e
id
en
tific
atio
n
o
f
all
s
izes
o
f
tu
m
o
r
s
.
Ad
d
in
g
PGGAN
-
g
en
er
ate
d
s
y
n
th
etic
im
ag
es
m
ak
es
th
e
m
o
d
el
m
o
r
e
r
o
b
u
s
t
by
en
lar
g
in
g
d
ataset’
s
s
ize
wh
er
e,
at
s
tar
t,
th
e
lack
of
a
n
n
o
tated
m
ed
ical
d
ata
is
o
v
er
co
m
e
.
W
ith
th
is
e
n
h
an
ce
m
en
t,
Neu
r
o
Net1
9
o
b
tain
s
b
etter
ac
cu
r
ac
y
an
d
ev
alu
atio
n
m
etr
ics
th
a
n
o
th
e
r
m
o
d
e
l
s
.
Ho
wev
er
,
th
e
q
u
ality
an
d
d
iv
e
r
s
ity
ab
o
u
t
i
n
p
u
t d
ata
will
s
way
th
is
f
r
am
ewo
r
k
’
s
ap
tn
ess
an
d
ab
ilit
y
to
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
3
2
2
6
-
3
2
3
7
3234
ea
s
ily
id
en
tify
s
m
all
o
r
aty
p
ic
al
tu
m
o
r
s
.
A
co
m
p
ar
ativ
e
an
a
ly
s
is
is
g
iv
en
in
th
e
Fig
u
r
e
6
t
h
at
s
h
o
ws
m
o
d
els,
s
u
c
h
a
s
Den
s
eNe
t1
2
1
,
Mo
b
ileNetV2
,
R
esNet5
0
,
VGG1
6
h
av
e
s
h
o
wn
g
o
o
d
p
er
f
o
r
m
an
ce
b
u
t
lack
at
in
tr
icate
tu
m
o
r
d
etails
an
d
co
n
f
r
o
n
ts
Neu
r
o
Net1
9
’
s
b
r
illi
an
t
f
ea
t
u
r
es
th
at
it
h
as
s
h
o
wn
in
d
eliv
er
in
g
co
n
s
is
ten
t
an
d
r
eliab
le
o
u
tco
m
es f
o
r
all
tu
m
o
r
ca
teg
o
r
ies.
Fig
u
r
e
6.
C
o
m
p
a
r
is
o
n
of
Neu
r
o
Net1
9
with
o
th
er
m
o
d
e
l
s
4.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
estab
lis
h
es
Neu
r
o
Net1
9
as
a
s
tate
-
of
-
th
e
-
a
r
t
d
e
ep
m
o
d
el
f
o
r
a
p
ar
ticu
lar
task
o
f
b
r
ain
tu
m
o
r
d
etec
tio
n
th
at
o
u
tp
er
f
o
r
m
s
o
th
er
m
o
d
els,
s
u
ch
as
M
o
b
ileNetV2
,
R
esNet5
0
,
an
d
Den
s
eNe
t1
2
1
,
with
in
cr
ea
s
ed
ef
f
ec
tiv
e
n
ess
.
I
n
o
r
d
er
to
g
et
9
9
.
5
%
test
ac
cu
r
ac
y
,
I
PP
M
f
o
r
m
u
lti
-
s
ca
le
f
ea
tu
r
e
ag
g
r
e
g
atio
n
with
d
ata
au
g
m
en
tatio
n
th
r
o
u
g
h
P
GGAN
is
u
s
ed
;
th
is
also
g
iv
es
an
ev
alu
atio
n
th
at
is
more
t
h
an
98%
in
m
o
s
t
ca
teg
o
r
ies
o
f
tu
m
o
r
s
o
b
s
er
v
e
d
.
T
h
ese
r
esu
lts
ca
n
cr
o
s
s
th
e
is
s
u
e
o
f
lim
ited
an
n
o
tated
d
ata
an
d
d
escr
ib
e
s
u
b
tle
tu
m
o
r
c
h
ar
ac
ter
is
tics
; t
h
er
ef
o
r
e
,
it is
a
u
s
ef
u
l so
lu
tio
n
f
o
r
m
e
d
ical
im
ag
in
g
an
aly
s
is
.
T
h
e
f
u
r
th
er
im
p
licatio
n
s
f
r
o
m
th
is
wo
r
k
a
r
e
its
ab
ilit
y
to
im
p
r
o
v
e
d
iag
n
o
s
tic
ac
cu
r
ac
y
an
d
aid
d
o
cto
r
s
in
th
e
s
p
ec
ial
lo
ca
liza
tio
n
an
d
class
if
icatio
n
of
tu
m
o
r
s
.
I
t
also
h
as
it
s
l
im
itatio
n
in
th
at
it
d
em
an
d
s
p
r
o
p
er
h
ig
h
-
q
u
ality
tr
ai
n
in
g
d
ata,
an
d
it
is
n
o
t
p
o
s
s
ib
le
to
d
etec
t
s
m
all
or
aty
p
ical
tu
m
o
r
s
.
T
h
ese
lim
itatio
n
s
can
be
r
e
m
o
v
e
d
i
n
f
u
t
u
r
e
s
t
u
d
i
e
s
b
y
a
d
d
i
n
g
d
i
v
e
r
s
i
f
i
e
d
a
n
d
h
i
g
h
-
r
e
s
o
l
u
t
i
o
n
d
a
t
a
s
et
s
,
i
m
p
r
o
v
i
n
g
t
h
e
m
o
r
e
a
d
v
a
n
c
e
d
a
u
g
m
e
n
t
a
t
i
o
n
m
eth
o
d
s
,
o
r
r
esear
ch
in
g
th
e
d
o
m
ain
ad
ap
tatio
n
th
at
ten
d
s
to
f
in
d
th
e
o
u
tlier
s
.
Ad
d
itio
n
al
wo
r
k
on
o
p
tim
izi
n
g
Neu
r
o
Net1
9
f
o
r
clin
ical
wo
r
k
f
lo
ws
or
u
s
e
th
r
o
u
g
h
a
clo
u
d
-
co
m
p
u
tin
g
p
latf
o
r
m
m
a
y
also
p
ay
o
f
f
in
th
e
f
u
tu
r
e
.
Fu
tu
r
e
a
d
v
an
ce
s
in
th
e
d
ir
ec
tio
n
o
f
i
n
ter
p
r
etab
ilit
y
an
d
ex
p
lain
ab
ilit
y
m
ay
f
u
r
t
h
er
s
p
e
ed
u
p
t
h
is
tech
n
o
lo
g
y
to
p
o
ten
tially
m
ak
e
it e
n
d
u
p
in
clin
ica
l p
r
ac
tice
.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
th
o
r
s
ex
ten
d
th
eir
s
in
ce
r
e
g
r
atitu
d
e
to
L
an
k
e
Pallav
i
a
n
d
C
h
.
Ma
d
h
u
B
ab
u
f
o
r
th
eir
in
v
alu
ab
le
s
u
p
p
o
r
t
an
d
g
u
i
d
an
ce
th
r
o
u
g
h
o
u
t
th
is
r
esear
ch
.
T
h
eir
in
s
ig
h
tf
u
l
d
is
cu
s
s
io
n
s
an
d
co
n
s
tr
u
ct
iv
e
f
ee
d
b
ac
k
h
av
e
s
ig
n
if
ican
tly
co
n
tr
ib
u
ted
to
th
e
s
u
cc
ess
f
u
l c
o
m
p
letio
n
o
f
th
i
s
s
tu
d
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
E
ac
h
au
th
o
r
h
as
co
n
tr
i
b
u
ted
s
i
g
n
if
ican
tly
t
o
th
is
s
tu
d
y
,
en
s
u
r
in
g
its
co
m
p
letio
n
an
d
ac
cu
r
a
cy
.
L
an
k
e
Pallav
i
was
r
esp
o
n
s
ib
le
f
o
r
c
o
n
ce
p
tu
aliza
tio
n
,
s
o
f
twar
e
,
r
eso
u
r
ce
s
,
v
alid
atio
n
,
f
o
r
m
al
an
aly
s
is
,
wr
itin
g
r
ev
iew
&
ed
itin
g
,
s
u
p
e
r
v
is
io
n
,
a
n
d
p
r
o
ject
ad
m
in
is
tr
atio
n
.
T
h
ati
R
a
m
y
a
co
n
tr
i
b
u
ted
to
m
eth
o
d
o
l
o
g
y
,
s
o
f
twar
e
,
d
ata
cu
r
atio
n
,
wr
itin
g
o
r
ig
i
n
al
d
r
af
t
,
wr
itin
g
r
ev
iew
&
e
d
itin
g
,
a
n
d
v
is
u
aliza
tio
n
.
Sin
g
u
p
u
r
a
p
u
Sai
C
h
ar
an
p
lay
ed
a
k
ey
r
o
le
in
m
eth
o
d
o
lo
g
y
,
s
o
f
twar
e,
v
alid
atio
n
,
f
o
r
m
al
an
aly
s
is
,
wr
it
in
g
o
r
ig
in
al
d
r
af
t,
an
d
wr
itin
g
r
ev
iew
&
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
o
b
u
s
t d
ee
p
lea
r
n
in
g
a
p
p
r
o
a
ch
fo
r
a
cc
u
r
a
te
d
etec
tio
n
o
f b
r
a
in
tu
mo
r
a
n
d
a
n
a
lysi
s
(
La
n
ke
P
a
lla
vi
)
3235
ed
i
ti
n
g
.
Si
r
i
g
a
d
h
a
Am
i
th
p
r
o
v
i
d
e
d
r
es
o
u
r
ce
s
,
c
o
n
d
u
c
te
d
in
v
e
s
tig
ati
o
n
,
m
a
n
a
g
e
d
d
ata
c
u
r
ati
o
n
,
a
n
d
c
o
n
t
r
i
b
u
te
d
to
w
r
i
ti
n
g
r
e
v
iew
&
e
d
i
ti
n
g
a
n
d
v
is
u
al
iza
ti
o
n
.
T
h
o
d
u
p
u
n
u
r
i
A
k
s
h
a
y
K
u
m
a
r
f
o
cu
s
ed
o
n
v
al
id
ati
o
n
,
in
v
est
ig
ati
o
n
,
w
r
iti
n
g
o
r
i
g
i
n
al
d
r
a
f
t,
a
n
d
w
r
it
in
g
r
ev
iew
&
e
d
iti
n
g
.
A
ll
au
th
o
r
s
h
a
v
e
r
ea
d
a
n
d
ap
p
r
o
v
ed
t
h
e
f
i
n
al
m
a
n
u
s
c
r
i
p
t.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
L
an
k
e
Pallav
i
✓
✓
✓
✓
✓
✓
✓
✓
T
h
ati
R
am
y
a
✓
✓
✓
✓
✓
✓
Sin
g
u
p
u
r
a
p
u
Sai Ch
ar
an
✓
✓
✓
✓
✓
✓
Sirig
ad
h
a
Am
ith
✓
✓
✓
✓
✓
T
h
o
d
u
p
u
n
u
r
i A
k
s
h
ay
Ku
m
ar
✓
✓
✓
✓
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.
I
NF
O
RM
E
D
CO
NS
E
N
T
No
t a
p
p
licab
le,
as n
o
h
u
m
a
n
p
ar
ticip
an
ts
wer
e
in
v
o
lv
e
d
in
th
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
o
r
an
im
al
s
u
b
jects; th
er
e
f
o
r
e,
eth
ical
a
p
p
r
o
v
al
was n
o
t
r
eq
u
ir
ed
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
t
h
at
s
u
p
p
o
r
t
t
h
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
a
v
ailab
le
f
r
o
m
th
e
co
r
r
esp
o
n
d
i
n
g
a
u
th
o
r
,
TR
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
A
.
B
.
A
b
d
u
sa
l
o
m
o
v
,
M
.
M
u
k
h
i
d
d
i
n
o
v
,
a
n
d
T
.
K
.
W
h
a
n
g
b
o
,
“
B
r
a
i
n
t
u
mo
r
d
e
t
e
c
t
i
o
n
b
a
s
e
d
o
n
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s
a
n
d
mag
n
e
t
i
c
r
e
s
o
n
a
n
c
e
i
m
a
g
i
n
g
,
”
C
a
n
c
e
r
s
,
v
o
l
.
1
5
,
n
o
.
1
6
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
c
a
n
c
e
r
s
1
5
1
6
4
1
7
2
.
[
2
]
A
.
A
.
N
a
y
a
n
e
t
a
l
.
,
“
A
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
f
o
r
b
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
u
s
i
n
g
m
a
g
n
e
t
i
c
r
e
s
o
n
a
n
c
e
i
m
a
g
i
n
g
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
E
l
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r E
n
g
i
n
e
e
r
i
n
g
(
I
J
EC
E)
,
v
o
l
.
1
3
,
n
o
.
1
,
p
p
.
1
0
3
9
–
1
0
4
7
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
3
i
1
.
p
p
1
0
3
9
-
1
0
4
7
.
[
3
]
S
.
A
l
su
b
a
i
,
H
.
U
.
K
h
a
n
,
A
.
A
l
q
a
h
t
a
n
i
,
M
.
S
h
a
,
S
.
A
b
b
a
s
,
a
n
d
U
.
G
.
M
o
h
a
mm
a
d
,
“
E
n
sem
b
l
e
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
b
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
,
”
Fr
o
n
t
i
e
rs
i
n
C
o
m
p
u
t
a
t
i
o
n
a
l
N
e
u
r
o
sc
i
e
n
c
e
,
v
o
l
.
1
6
,
2
0
2
2
,
d
o
i
:
1
0
.
3
3
8
9
/
f
n
c
o
m.
2
0
2
2
.
1
0
0
5
6
1
7
.
[
4
]
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
.
Ja
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
ssi
f
i
c
a
t
i
o
n
o
f
b
r
a
i
n
t
u
mo
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
Re
p
o
rt
s
,
v
o
l
.
1
3
,
n
o
.
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
023
-
5
0
5
0
5
-
6.
[
5
]
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
.
[
6
]
H
.
C
h
e
n
,
Z.
Q
i
n
,
Y
.
D
i
n
g
,
a
n
d
T
.
L
a
n
,
“
B
r
a
i
n
t
u
mo
r
se
g
me
n
t
a
t
i
o
n
w
i
t
h
g
e
n
e
r
a
t
i
v
e
a
d
v
e
r
s
a
r
i
a
l
n
e
t
s,
”
i
n
2
0
1
9
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
Bi
g
D
a
t
a
,
I
C
AI
BD
2
0
1
9
,
2
0
1
9
,
p
p
.
3
0
1
–
3
0
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
I
B
D
.
2
0
1
9
.
8
8
3
6
9
6
8
.
[
7
]
G
.
B
.
M
o
h
a
n
e
t
a
l
.
,
“
En
h
a
n
c
i
n
g
b
r
a
i
n
t
u
m
o
r
d
i
a
g
n
o
s
i
s
w
i
t
h
g
e
n
e
r
a
t
i
v
e
a
d
v
e
r
s
a
r
i
a
l
n
e
t
w
o
r
k
s,”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
1
4
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
l
o
u
d
C
o
m
p
u
t
i
n
g
,
D
a
t
a
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
ri
n
g
,
C
o
n
f
l
u
e
n
c
e
2
0
2
4
,
2
0
2
4
,
p
p
.
8
4
6
–
8
5
1
,
d
o
i
:
1
0
.
1
1
0
9
/
C
o
n
f
l
u
e
n
c
e
6
0
2
2
3
.
2
0
2
4
.
1
0
4
6
3
3
8
4
.
[
8
]
C
.
G
e
,
I
.
Y
.
H
.
G
u
,
A
.
S
.
Ja
k
o
l
a
,
a
n
d
J.
Y
a
n
g
,
“
D
e
e
p
se
mi
-
s
u
p
e
r
v
i
se
d
l
e
a
r
n
i
n
g
f
o
r
b
r
a
i
n
t
u
mo
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
BM
C
Me
d
i
c
a
l
I
m
a
g
i
n
g
,
v
o
l
.
2
0
,
n
o
.
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
8
6
/
s
1
2
8
8
0
-
0
2
0
-
0
0
4
8
5
-
0.
[
9
]
S
.
G
u
p
t
a
,
M
.
A
g
g
r
a
w
a
l
,
A
.
S
h
u
k
l
a
,
a
n
d
A
.
Ty
a
g
i
,
“
B
r
a
i
n
t
u
m
o
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
t
e
c
h
n
i
q
u
e
s
b
a
s
e
d
o
n
M
R
I
i
mag
e
s,”
L
e
c
t
u
re
N
o
t
e
s
i
n
N
e
t
w
o
r
k
s
a
n
d
S
y
st
e
m
s
,
v
o
l
.
7
2
6
LN
N
S
,
p
p
.
5
6
9
–
5
7
7
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
981
-
99
-
3
7
1
6
-
5
_
4
6
.
[
1
0
]
C
.
H
a
n
e
t
a
l
.
,
“
C
o
m
b
i
n
i
n
g
n
o
i
s
e
-
to
-
i
mag
e
a
n
d
i
m
a
g
e
-
to
-
i
m
a
g
e
G
A
N
s
:
B
r
a
i
n
M
R
i
m
a
g
e
a
u
g
me
n
t
a
t
i
o
n
f
o
r
t
u
mo
r
d
e
t
e
c
t
i
o
n
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
7
,
p
p
.
1
5
6
9
6
6
–
1
5
6
9
7
7
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
1
9
.
2
9
4
7
6
0
6
.
[
1
1
]
S
.
K
a
r
a
me
h
i
ć
a
n
d
S
.
J
u
k
i
ć
,
“
B
r
a
i
n
t
u
mo
r
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
V
G
G
1
6
d
e
e
p
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
m
a
n
d
p
y
t
h
o
n
i
ma
g
i
n
g
l
i
b
r
a
r
y
,
”
Bi
o
e
n
g
i
n
e
e
ri
n
g
S
t
u
d
i
e
s
,
v
o
l
.
4
,
n
o
.
2
,
p
p
.
1
–
1
3
,
N
o
v
.
2
0
2
3
,
d
o
i
:
1
0
.
3
7
8
6
8
/
b
e
s.
v
4
i
2
.
i
d
2
5
2
.
[
1
2
]
P
a
n
d
i
y
a
n
M
,
Je
n
i
s
h
a
E
,
M
a
d
h
u
r
a
n
j
a
n
i
G
,
B
.
D
h
a
r
sh
i
n
i
S
,
“
B
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
u
s
i
n
g
d
e
e
p
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
I
n
t
e
ra
n
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
S
c
i
e
n
t
i
f
i
c
Re
s
e
a
rc
h
i
n
En
g
i
n
e
e
ri
n
g
a
n
d
Ma
n
a
g
e
m
e
n
t
,
v
o
l
.
0
8
,
n
o
.
0
3
,
p
p
.
1
–
5
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
5
5
0
4
1
/
I
JS
R
E
M
2
9
7
0
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.