I
nte
rna
t
io
na
l J
o
urna
l o
f
Adv
a
nces in Applie
d Science
s
(
I
J
AAS)
Vo
l.
10
,
No
.
1
,
Ma
r
ch
2
0
2
1
,
p
p
.
88
~
98
I
SS
N:
2
2
5
2
-
8
8
1
4
,
DOI
: 1
0
.
1
1
5
9
1
/ijaas.v
10
.
i
1
.
p
p
88
-
98
88
J
o
ur
na
l
ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
a
s
.
ia
esco
r
e.
co
m
Deep learni
ng
mo
del f
o
r
g
lio
ma
,
m
ening
io
ma
,
a
nd
p
i
tuit
a
ry
cla
ss
ificatio
n
T
o
qa
A.
Sa
do
o
n
,
M
o
ha
m
m
e
d H
.
Ali
De
p
a
rtme
n
t
o
f
El
e
c
tro
n
ic an
d
Co
m
m
u
n
ica
ti
o
n
s E
n
g
in
e
e
rin
g
,
C
o
ll
e
g
e
o
f
E
n
g
in
e
e
rin
g
,
Al
-
Na
h
ra
i
n
U
n
iv
e
rsit
y
,
Ba
g
h
d
a
d
,
Ira
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
1
5
,
2
0
2
0
R
ev
is
ed
Dec
2
3
,
2
0
2
0
Acc
ep
ted
J
an
1
3
,
2
0
21
On
e
o
f
th
e
c
o
m
m
o
n
c
a
u
se
s
o
f
d
e
a
th
is
a
b
ra
in
tu
m
o
r
.
Be
c
a
u
se
o
f
th
e
a
b
o
v
e
m
e
n
ti
o
n
e
d
,
e
a
rly
d
e
tec
ti
o
n
o
f
a
b
r
a
in
t
u
m
o
r
is crit
ica
l
fo
r
fa
ste
r
trea
tme
n
t,
a
n
d
th
e
re
fo
re
th
e
re
a
re
m
a
n
y
tec
h
n
iq
u
e
s
u
se
d
t
o
v
isu
a
li
z
e
a
b
ra
in
tu
m
o
r.
On
e
o
f
th
e
se
tec
h
n
i
q
u
e
s
is
m
a
g
n
e
ti
c
re
so
n
a
n
c
e
ima
g
i
n
g
(M
RI)
.
O
n
t
h
e
o
th
e
r
h
a
n
d
,
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
in
g
,
a
n
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
(CNN
)
a
re
th
e
sta
te
o
f
a
rt
tec
h
n
o
l
o
g
ies
in
r
e
c
e
n
t
y
e
a
rs
u
se
d
in
so
lv
i
n
g
m
a
n
y
m
e
d
ica
l
ima
g
e
-
re
late
d
p
ro
b
lem
s
su
c
h
a
s
c
las
sifica
ti
o
n
.
I
n
t
h
is
re
se
a
rc
h
,
t
h
r
e
e
ty
p
e
s
o
f
b
ra
in
tu
m
o
rs
we
re
c
las
sified
u
sin
g
m
a
g
n
e
ti
c
re
so
n
a
n
c
e
ima
g
in
g
n
a
m
e
ly
g
li
o
m
a
,
m
e
n
i
n
g
io
m
a
,
a
n
d
p
it
u
it
a
r
y
g
lan
d
o
n
th
e
b
a
se
d
o
f
CNN
.
T
h
e
d
a
tas
e
t
u
se
d
i
n
t
h
is
wo
r
k
in
c
lu
d
e
s
2
3
3
p
a
ti
e
n
ts
fo
r
a
t
o
tal
o
f
3
,
0
6
4
c
o
n
tras
t
-
e
n
h
a
n
c
e
d
T1
ima
g
e
s.
I
n
t
h
is
p
a
p
e
r,
a
c
o
m
p
a
riso
n
is
p
re
se
n
ted
b
e
twe
e
n
th
e
p
re
se
n
ted
m
o
d
e
l
a
n
d
o
t
h
e
r
m
o
d
e
ls
to
d
e
m
o
n
stra
te
th
e
su
p
e
rio
rit
y
o
f
o
u
r
m
o
d
e
l
o
v
e
r
t
h
e
o
th
e
rs.
M
o
re
o
v
e
r,
th
e
d
iffere
n
c
e
in
o
u
tco
m
e
b
e
twe
e
n
p
re
-
a
n
d
p
o
s
t
-
d
a
ta
p
re
-
p
ro
c
e
ss
in
g
a
n
d
a
u
g
m
e
n
tatio
n
wa
s
d
isc
u
ss
e
d
.
Th
e
h
i
g
h
e
st
a
c
c
u
ra
c
y
m
e
tri
c
s
e
x
trac
ted
fro
m
c
o
n
f
u
sio
n
m
a
tri
c
e
s a
re
a
p
re
c
isi
o
n
o
f
9
9
.
1
%
fo
r
t
h
e
p
it
u
i
tary
,
a
se
n
siti
v
it
y
o
f
9
8
.
7
%
f
o
r
g
li
o
m
a
,
a
sp
e
c
ifi
c
it
y
o
f
9
9
.
1
%
,
a
n
d
a
n
a
c
c
u
ra
c
y
o
f
9
9
.
1
%
fo
r
t
h
e
p
it
u
i
tary
.
T
h
e
o
v
e
ra
ll
a
c
c
u
ra
c
y
o
b
tain
e
d
is 9
6
.
1
%
.
K
ey
w
o
rd
s
:
B
r
ain
tu
m
o
r
C
NN
Dee
p
lear
n
in
g
I
m
ag
e
class
if
icatio
n
I
m
ag
e
p
r
o
ce
s
s
in
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
T
o
q
a
A.
Sad
o
o
n
Dep
ar
tm
en
t o
f
E
lectr
ical
an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
i
n
g
Al
-
Nah
r
ain
Un
iv
er
s
ity
PO Bo
x
6
4
0
4
0
,
B
ag
h
d
a
d
,
I
r
a
q
E
m
ail:
to
q
a.
am
.
wd
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
s
ec
o
n
d
m
o
s
t
im
p
o
r
tan
t
ca
u
s
e
o
f
d
ea
th
ac
co
r
d
in
g
to
t
h
e
wo
r
ld
h
ea
lth
o
r
g
an
izatio
n
is
c
an
ce
r
,
an
d
th
er
e
ar
e
ab
o
u
t
9
.
6
m
illi
o
n
d
e
ath
s
in
2
0
1
8
b
ec
au
s
e
o
f
ca
n
c
er
.
Glo
b
ally
,
ab
o
u
t
1
o
u
t
o
f
e
v
er
y
6
d
ea
th
s
f
r
o
m
ca
n
ce
r
.
On
th
e
o
th
e
r
h
a
n
d
,
th
e
b
r
ain
co
n
s
id
er
s
o
n
e
o
f
th
e
m
o
s
t
co
m
p
lex
o
r
g
a
n
s
in
th
e
h
u
m
an
b
o
d
y
th
at
wo
r
k
s
with
b
illi
o
n
s
o
f
ce
lls
.
T
h
e
ac
c
u
m
u
latio
n
o
f
ab
n
o
r
m
al
ce
lls
i
n
th
e
b
r
ain
lead
s
to
th
e
s
o
-
ca
l
led
b
r
ain
tu
m
o
r
.
A
b
r
ain
tu
m
o
r
is
d
iv
id
e
d
in
to
tw
o
ca
teg
o
r
ies,
p
r
im
ar
y
a
n
d
s
ec
o
n
d
ar
y
.
T
h
e
f
ir
s
t
o
n
e
a
r
is
es
in
th
e
b
r
ain
,
wh
i
le
th
e
s
ec
o
n
d
o
n
e
ar
is
es
f
r
o
m
o
t
h
er
p
ar
ts
o
f
th
e
b
o
d
y
.
T
h
e
tu
m
o
r
s
ca
n
b
e
ca
n
ce
r
o
u
s
(
m
alig
n
an
t
)
o
r
n
o
n
-
ca
n
ce
r
o
u
s
(
b
en
ig
n
)
.
C
an
ce
r
o
u
s
b
r
ain
tu
m
o
r
s
g
r
o
w
r
ap
id
l
y
an
d
s
p
r
ea
d
to
o
th
er
ar
ea
s
o
f
th
e
b
r
ain
co
m
p
ar
ed
to
n
o
n
-
ca
n
ce
r
o
u
s
tu
m
o
r
s
.
Glio
m
a,
m
en
in
g
io
m
a,
a
n
d
p
itu
itar
y
ar
e
o
th
er
d
if
f
er
en
t
ty
p
es
o
f
b
r
ain
tu
m
o
r
s
[
1
]
.
On
a
lar
g
er
s
ca
le,
g
lio
m
a
tu
m
o
r
wh
ich
is
th
e
m
o
s
t
co
m
m
o
n
ty
p
e
o
f
p
r
im
ar
y
b
r
ain
tu
m
o
r
[
2
]
ar
e
class
if
ied
in
to
f
o
u
r
g
r
ad
es,
an
d
th
e
h
ig
h
er
th
e
g
r
ad
e,
th
e
m
o
r
e
m
alig
n
a
n
t
th
e
tu
m
o
r
,
an
d
o
r
ig
i
n
ate
in
th
e
g
lial
ce
lls
o
f
th
e
b
r
ain
[
3
]
.
Me
n
in
g
io
m
as,
wh
ic
h
o
r
ig
in
at
e
f
r
o
m
a
lay
er
o
f
tis
s
u
e
ca
lled
th
e
m
e
n
in
g
es,
a
r
e
s
o
m
etim
es
co
n
s
id
er
e
d
b
e
n
ig
n
tu
m
o
r
s
.
T
h
e
g
r
o
wth
o
f
th
is
s
p
e
cies
is
s
lo
w
an
d
less
wid
esp
r
ea
d
.
W
h
ile
th
e
p
itu
itar
y
tu
m
o
r
g
r
o
ws
o
n
t
h
e
p
itu
itar
y
g
lan
d
.
T
h
ese
tu
m
o
r
s
ar
e
also
b
en
ig
n
an
d
less
wid
esp
r
ea
d
[
4
]
,
[
5
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2
2
5
2
-
8
8
1
4
Dee
p
lea
r
n
in
g
fo
r
g
lio
ma
,
me
n
in
g
io
ma
,
a
n
d
p
itu
ita
r
y
cla
s
s
ifica
tio
n
(
To
g
a
A
.
S
a
d
o
o
n
)
89
I
n
th
e
s
am
e
f
ield
,
th
e
m
ed
ical
im
ag
e
co
n
s
id
er
s
o
n
e
o
f
th
e
b
es
t
tech
n
iq
u
es
u
s
ed
to
im
ag
in
g
th
e
in
ter
n
al
p
ar
t
o
f
th
e
b
o
d
y
u
s
in
g
cr
o
s
s
-
s
ec
tio
n
al
s
lices
to
m
o
n
ito
r
an
d
d
iag
n
o
s
e
th
e
m
ed
ical
ca
s
es.
T
h
e
b
e
s
t
im
ag
in
g
tech
n
iq
u
e
u
s
ed
f
o
r
im
ag
in
g
th
e
tu
m
o
r
is
m
ag
n
etic
r
eso
n
an
ce
i
m
ag
in
g
.
B
ec
au
s
e
it
p
r
o
v
id
es
in
f
o
r
m
atio
n
in
d
etail
ab
o
u
t
d
if
f
e
r
en
t
tis
s
u
es
b
o
d
y
with
h
ig
h
r
eso
lu
tio
n
an
d
c
o
n
tr
ast,
an
d
th
u
s
it
is
wid
ely
u
s
ed
in
an
ato
m
ical
ad
ju
n
ctiv
e
ex
am
i
n
atio
n
o
f
b
r
ai
n
tis
s
u
e
[
6
]
,
[
7
]
.
T
h
e
MRI
tech
n
o
lo
g
y
h
as
s
o
m
e
b
en
e
f
its
s
u
ch
as
th
e
s
u
p
er
-
r
eso
lu
tio
n
o
f
s
o
f
t
-
tis
s
u
e
co
n
tr
ast,
s
eq
u
en
ce
s
o
f
d
if
f
er
e
n
t
p
u
ls
es,
h
ig
h
-
r
eso
lu
tio
n
im
ag
in
g
,
an
d
n
o
n
-
io
n
izin
g
r
ad
iatio
n
.
B
esid
es,
an
MRI
s
c
an
p
r
o
v
i
d
es
a
s
et
o
f
im
ag
es
o
f
tis
s
u
es
with
d
if
f
er
en
t
co
n
tr
a
s
ts
,
an
d
th
ese
h
elp
clin
ician
s
m
ak
e
an
ac
cu
r
ate
d
iag
n
o
s
is
[
8
]
.
So
,
t
o
tr
ea
t
th
e
t
u
m
o
r
f
aster
a
n
d
ac
cu
r
ately
,
it
m
u
s
t
b
e
d
iag
n
o
s
ed
ea
r
ly
an
d
f
o
r
th
is
r
ea
s
o
n
,
co
m
p
u
ter
-
aid
ed
d
iag
n
o
s
is
(
C
AD)
s
y
s
tem
is
u
s
ed
to
s
p
o
t
th
e
d
is
ea
s
e
ea
r
ly
.
T
h
is
s
y
s
te
m
ca
n
lo
ca
lize
th
e
tu
m
o
r
an
d
co
n
clu
d
e;
wh
eth
e
r
it
is
a
t
u
m
o
r
o
r
n
o
t
alo
n
g
with
its
ty
p
e
an
d
d
eg
r
ee
,
if
a
n
y
.
I
n
co
n
tr
ast,
th
e
d
iag
n
o
s
is
o
f
d
o
ct
o
r
s
is
d
is
p
en
s
ed
with
,
wh
ich
ar
e
s
u
b
ject
to
er
r
o
r
s
,
o
m
is
s
io
n
s
,
an
d
tak
e
a
lo
n
g
tim
e
an
d
ef
f
o
r
t
d
u
e
to
a
lar
g
e
n
u
m
b
er
o
f
d
ata.
Hen
ce
,
t
h
e
p
r
o
ce
s
s
es
o
f
class
if
icatio
n
,
s
eg
m
en
tatio
n
,
lo
ca
lizatio
n
,
an
d
d
etec
tio
n
o
f
a
b
r
ain
tu
m
o
r
in
d
iag
n
o
s
in
g
th
e
d
is
ea
s
e
h
av
e
b
ec
o
m
e
th
e
m
o
s
t d
if
f
icu
lt task
[
9
]
.
Ar
tific
ial
in
tellig
en
ce
(
AI
)
,
m
ac
h
in
e
lear
n
in
g
(
ML
)
,
an
d
d
ee
p
lear
n
in
g
(
DL
)
a
r
e
th
r
ee
co
n
ce
p
ts
r
elativ
e
to
ea
ch
o
th
er
.
I
n
g
en
e
r
al,
DL
is
a
ty
p
e
o
f
ML
,
an
d
th
e
last
is
a
ty
p
e
o
f
AI
.
T
o
b
eg
in
with
,
ar
tific
ial
in
tellig
en
ce
r
ef
er
s
to
a
n
y
tech
n
o
l
o
g
y
th
at
h
as
s
o
m
e
in
tellig
en
ce
.
O
n
th
e
o
t
h
er
h
a
n
d
,
ML
is
a
tec
h
n
iq
u
e
u
s
ed
to
f
o
r
m
a
m
o
d
el
ou
t
o
f
d
ata,
an
d
h
as
m
a
n
y
tech
n
o
lo
g
ies;
o
n
e
o
f
t
h
ese
tech
n
o
l
o
g
ies
is
d
ee
p
lear
n
in
g
,
wh
ic
h
i
s
o
u
r
in
ter
est
in
th
is
r
esear
ch
.
Dep
en
d
in
g
o
n
th
e
ap
p
licatio
n
th
at
u
s
ed
m
ac
h
in
e
lear
n
in
g
,
an
d
th
e
tr
ain
in
g
m
eth
o
d
ML
tech
n
iq
u
es
ar
e
class
if
ied
in
to
th
r
ee
k
in
d
s
,
s
u
p
er
v
is
ed
lear
n
i
n
g
,
u
n
s
u
p
e
r
v
is
ed
lear
n
in
g
,
a
n
d
r
ein
f
o
r
ce
m
en
t
l
ea
r
n
in
g
.
I
n
th
e
f
ir
s
t
o
n
e,
all
th
e
tr
ain
in
g
d
ata
s
et
m
u
s
t c
o
n
s
is
t o
f
in
p
u
ts
with
th
e
l
ab
els to
g
eth
er
,
an
d
wh
at
th
e
m
o
d
el
m
u
s
t
b
e
f
o
u
n
d
in
th
e
test
in
g
s
tep
f
o
r
a
g
iv
en
in
p
u
t
is
th
e
p
r
ed
icted
o
u
tp
u
t.
F
o
r
ex
am
p
le
,
r
e
g
r
ess
io
n
,
an
d
class
if
icatio
n
wh
ich
is
o
u
r
in
ter
est.
W
h
ile
th
e
tr
ain
in
g
d
ata
in
u
n
s
u
p
er
v
is
ed
lear
n
in
g
co
n
tain
in
p
u
ts
alo
n
e
with
o
u
t
v
alid
o
u
tp
u
ts
.
I
n
th
is
ty
p
e
o
f
lear
n
in
g
,
it
is
u
s
ed
f
o
r
clu
s
ter
in
g
.
R
ein
f
o
r
ce
m
en
t
lear
n
in
g
u
s
es
in
p
u
t,
o
u
tp
u
t,
a
n
d
g
r
ad
e
in
th
e
tr
ain
in
g
d
ata.
Fo
r
ex
a
m
p
le,
s
u
ch
as
co
n
tr
o
l
an
d
g
am
ep
la
y
s
.
T
h
is
p
ap
er
o
n
ly
co
v
er
s
s
u
p
er
v
is
ed
lear
n
in
g
.
Mo
r
eo
v
er
,
th
r
ee
co
n
ce
p
ts
m
u
s
t
b
e
u
n
d
er
s
to
o
d
in
s
u
p
e
r
v
is
ed
lear
n
in
g
wh
ich
ar
e;
f
ir
s
tly
,
tr
a
in
in
g
d
ata
r
ef
er
s
to
th
e
d
ata
u
s
ed
t
h
r
o
u
g
h
th
e
t
r
ain
i
n
g
s
tep
to
tr
ai
n
th
e
m
o
d
el.
Sec
o
n
d
ly
,
v
alid
atio
n
d
ata,
th
is
c
o
llectio
n
o
f
d
ata
u
s
ed
to
co
m
p
ar
e
th
e
tr
ain
in
g
d
at
a
with
p
r
ed
icted
o
u
tp
u
t.
Fin
ally
,
test
in
g
d
ata
u
s
ed
to
m
ea
s
u
r
e
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
af
ter
th
e
tr
ain
in
g
p
r
o
ce
s
s
[
1
0
]
.
DL
is
o
n
e
o
f
th
e
M
L
tech
n
iq
u
es
[
1
0
]
,
it
co
n
s
is
ts
o
f
m
u
ltip
le
lay
er
s
f
o
r
lear
n
in
g
d
ata.
T
h
es
e
m
eth
o
d
s
h
av
e
g
r
ea
tly
im
p
r
o
v
ed
th
e
lat
est
in
s
p
ee
ch
r
ec
o
g
n
itio
n
tech
n
o
lo
g
y
,
I
m
ag
e
r
ec
o
g
n
itio
n
,
o
b
ject
d
etec
tio
n
,
an
d
m
an
y
o
th
e
r
ar
ea
s
s
u
ch
as
d
is
ea
s
e
d
etec
tio
n
.
Dee
p
lear
n
in
g
co
n
s
is
t
s
o
f
two
ty
p
es,
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN)
h
as
m
ad
e
b
r
ea
k
th
r
o
u
g
h
s
in
s
p
ee
ch
,
im
a
g
e,
an
d
v
i
d
eo
p
r
o
ce
s
s
in
g
,
a
n
d
r
ec
u
r
r
e
n
t
n
eu
r
al
n
etwo
r
k
s
(
R
NN)
h
ig
h
lig
h
ted
s
eq
u
en
tial d
ata
s
u
ch
as tex
t a
n
d
s
p
ee
c
h
[
1
1
]
.
C
o
n
v
Net
is
a
s
p
ec
ialized
k
in
d
o
f
n
eu
r
al
n
etwo
r
k
f
o
r
d
ata
p
r
o
ce
s
s
in
g
.
C
NN
r
ef
er
s
to
th
at
th
e
n
etwo
r
k
u
s
es
a
p
r
o
ce
s
s
f
r
o
m
m
ath
n
am
ely
co
n
v
o
l
u
tio
n
in
s
tead
o
f
th
e
g
en
er
al
m
atr
ix
m
u
ltip
licatio
n
[
1
2
]
.
C
NN
b
ec
am
e
m
o
r
e
f
am
o
u
s
af
ter
Alex
Net
[
1
3
]
h
as
b
ee
n
a
r
ec
o
r
d
i
n
2
0
1
2
w
h
ich
is
d
esig
n
ed
b
y
Kr
izh
e
v
s
k
y
et
a
l
.
,
a
n
d
s
h
o
we
d
o
n
I
L
SVR
C
ex
ce
llen
t
p
er
f
o
r
m
an
ce
[
1
4
]
.
Alex
Net'
s
s
u
cc
es
s
in
s
ett
in
g
th
e
way
f
o
r
th
e
in
v
en
tio
n
o
f
d
i
f
f
er
en
t
C
NN
m
o
d
els
[
1
5
]
in
ad
d
itio
n
to
ap
p
ly
in
g
th
o
s
e
m
o
d
els
in
v
ar
io
u
s
ar
ea
s
o
f
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
an
d
co
m
p
u
ter
v
is
io
n
[
1
6
]
.
B
u
t
ac
t
u
all
y
,
I
t
was
an
an
cien
t
tec
h
n
o
lo
g
y
th
at
was
d
e
v
elo
p
e
d
in
t
h
e
1
9
8
0
s
an
d
1
9
9
0
s
.
I
n
o
th
e
r
wo
r
d
s
,
C
NN
is
a
d
ee
p
n
eu
r
al
n
etwo
r
k
with
m
an
y
h
i
d
d
en
la
y
er
s
.
I
t
is
wo
r
t
h
m
en
tio
n
in
g
,
it
is
n
o
t
g
o
o
d
to
u
s
e
th
e
o
r
ig
in
al
im
a
g
es
d
u
r
i
n
g
tr
ain
in
g
,
b
ec
au
s
e
th
is
lead
s
to
p
o
o
r
r
esu
lts
.
T
h
er
ef
o
r
e,
to
ex
tr
ac
t
th
e
f
ea
tu
r
es
th
e
im
ag
es
m
u
s
t
b
e
p
r
o
ce
s
s
ed
.
T
o
d
o
th
at,
th
er
e
a
r
e
m
an
y
tech
n
iq
u
es
u
s
ed
f
o
r
th
is
p
u
r
p
o
s
e,
wh
ic
h
ar
e
in
d
ep
en
d
en
t
o
f
m
ac
h
in
e
lear
n
in
g
,
an
d
th
is
is
tak
in
g
a
g
r
ea
t
d
ea
l
o
f
tim
e
an
d
c
o
s
t,
wh
ile
o
n
C
NN
th
e
s
to
r
y
is
d
if
f
er
en
t.
T
h
e
r
e
ar
e
s
o
m
e
o
f
th
e
h
id
d
en
lay
er
s
in
C
o
n
v
Net
th
at
ar
e
r
esp
o
n
s
ib
le
to
ex
tr
ac
t
th
e
f
ea
tu
r
e
th
r
o
u
g
h
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
an
d
th
e
o
t
h
er
r
esp
o
n
s
ib
le
f
o
r
ex
am
p
le
class
if
icatio
n
.
Mo
r
eo
v
er
,
wh
en
C
NN
co
n
tain
s
m
o
r
e
d
ee
p
h
id
d
en
lay
e
r
s
,
lead
s
to
m
o
r
e
f
ea
tu
r
e
ex
tr
ac
tio
n
,
as a
r
esu
lt,
b
etter
p
er
f
o
r
m
a
n
ce
.
T
h
ese
d
ay
s
,
th
e
p
o
wer
o
f
d
ee
p
lear
n
in
g
a
n
d
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
it
(
GPU)
ca
n
b
e
an
im
p
o
r
tan
t
to
o
l
to
d
ev
elo
p
m
an
y
n
etwo
r
k
s
th
at
u
s
e
to
s
o
lv
e
d
if
f
e
r
en
t
p
r
o
b
lem
s
.
T
h
ese
n
etwo
r
k
s
ca
n
b
e
r
u
n
th
r
o
u
g
h
a
h
i
g
h
-
lev
e
l
p
r
o
g
r
a
m
m
in
g
in
ter
f
ac
e
b
ased
o
n
NVI
DI
A
GPUs
ac
ce
ler
ated
lib
r
ar
ies.
T
e
n
s
o
r
Flo
w,
Py
T
o
r
ch
,
MA
T
L
AB
,
MX
Net,
NAVI
DA
C
af
f
e,
Pad
d
lePad
d
le,
C
h
ain
er
ar
e
co
m
m
o
n
d
ee
p
lear
n
in
g
f
r
am
ew
o
r
k
s
.
I
n
th
is
s
tu
d
y
,
we
u
s
e
MA
T
L
AB
2
0
1
8
f
r
am
ew
o
r
k
a
n
d
th
e
p
o
wer
o
f
GPU
an
d
C
UDA
f
r
o
m
NA
VI
DA
Gef
o
r
ce
9
2
0
M.
MRI
B
r
ain
tu
m
o
r
class
if
icatio
n
b
ased
o
n
a
m
ac
h
in
e
lear
n
i
n
g
tech
n
i
q
u
e
h
as
b
ee
n
p
er
f
o
r
m
in
g
o
v
er
m
an
y
y
ea
r
s
.
C
h
ar
f
i
et
a
l
.
[
1
7
]
p
r
esen
ted
a
tech
n
iq
u
e
f
o
r
b
r
ain
tu
m
o
r
class
if
icatio
n
in
to
n
o
r
m
al
o
r
m
alig
n
an
t
an
d
ab
n
o
r
m
al
o
r
b
en
i
g
n
.
T
h
ey
u
s
e
d
th
e
h
is
to
g
r
am
d
e
p
en
d
e
n
t
th
r
esh
o
ld
in
g
f
o
r
im
ag
e
s
eg
m
en
ta
tio
n
.
Mo
r
eo
v
er
,
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
th
e
au
th
o
r
s
u
s
ed
d
is
cr
ete
wav
elet
tr
an
s
f
o
r
m
,
f
o
r
r
ed
u
cin
g
t
h
e
d
im
e
n
s
io
n
ality
o
f
th
e
wa
v
elet
co
ef
f
icien
ts
u
s
ed
p
r
i
n
cip
al
c
o
m
p
o
n
e
n
t
an
al
y
s
is
an
d
t
h
e
f
ee
d
-
f
o
r
war
d
b
ack
-
p
r
o
p
ag
atio
n
n
eu
r
al
n
etwo
r
k
f
o
r
class
if
icatio
n
.
T
h
e
class
if
icati
o
n
ac
cu
r
ac
y
o
n
b
o
th
tr
ain
i
n
g
a
n
d
test
im
ag
es
is
9
0
%.
C
h
en
g
et
a
l
.
[
1
]
p
r
esen
ted
a
m
o
d
el
to
en
h
a
n
ce
th
e
p
e
r
f
o
r
m
an
ce
o
f
MRI
class
if
icatio
n
o
f
b
r
ai
n
tu
m
o
r
s
with
th
r
ee
ca
teg
o
r
ies,
m
en
in
g
io
m
a,
g
lio
m
a,
an
d
p
itu
itar
y
.
Firstl
y
t
h
ey
u
s
ed
a
r
eg
io
n
o
f
in
ter
est (
R
OI
)
f
r
o
m
th
e
au
g
m
en
ted
tu
m
o
r
r
eg
io
n
v
ia
im
ag
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
88
–
98
90
d
ilatio
n
.
Seco
n
d
ly
,
th
ey
s
p
lit
t
h
e
au
g
m
e
n
ted
tu
m
o
r
in
to
f
in
e
r
in
g
-
f
o
r
m
s
u
b
r
e
g
io
n
s
.
Fin
ally
,
t
h
ey
h
av
e
b
ee
n
u
s
ed
b
ag
-
of
-
wo
r
d
s
(
B
o
W
)
,
in
ten
s
it
y
h
is
to
g
r
am
,
a
n
d
g
r
ay
lev
el
c
o
-
o
cc
u
r
r
en
ce
m
atr
i
x
(
GL
C
M)
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
b
est ac
cu
r
ac
y
h
as b
ee
n
ev
alu
ated
is
9
1
.
2
8
%.
Pau
l
et
a
l
.
[
1
8
]
p
r
esen
ted
a
m
o
d
el
to
class
if
y
MRI
b
r
ain
tu
m
o
r
s
in
to
th
r
ee
ca
teg
o
r
ies,
p
itu
itar
y
,
m
en
in
g
io
m
a
,
an
d
g
lio
m
a.
T
h
e
y
u
s
ed
ju
s
t
a
x
ial
s
lices
an
d
tw
o
ty
p
es
o
f
n
eu
r
al
n
etwo
r
k
s
(
f
u
lly
co
n
n
ec
ted
n
eu
r
al
n
etwo
r
k
an
d
C
NN)
wh
ich
co
n
tain
two
lay
er
s
o
f
(
co
n
v
o
lu
tio
n
al
lay
er
s
,
m
ax
-
p
o
o
lin
g
la
y
er
s
f
o
llo
wed
b
y
f
u
lly
co
n
n
ec
ted
lay
e
r
s
)
an
d
f
in
ally
ac
h
iev
ed
m
a
x
im
u
m
ac
cu
r
ac
y
o
f
9
1
.
4
3
%.
Af
s
h
ar
et
a
l
.
[
19]
u
s
ed
C
ap
s
Net
ar
ch
itectu
r
e
to
class
if
y
MRI
b
r
ain
tu
m
o
r
in
to
g
lio
m
a,
m
en
in
g
io
m
a,
a
n
d
p
itu
itar
y
.
T
h
ey
ta
k
e
th
e
tu
m
o
r
co
ar
s
e
b
o
u
n
d
ar
ies as e
x
tr
a
in
p
u
ts
f
o
r
th
e
tr
ain
in
g
p
r
o
ce
s
s
.
T
h
e
ac
c
u
r
ac
y
o
f
th
is
m
o
d
el
is
9
0
.
8
9
%.
Kab
ir
et
a
l
.
[
2
0
]
em
p
lo
y
th
e
g
en
etic
al
g
o
r
ith
m
(
GA)
to
ac
h
iev
e
th
e
b
est
p
e
r
f
o
r
m
an
ce
o
f
th
e
C
NN
m
o
d
el
f
o
r
class
if
icatio
n
MRI
in
to
f
o
u
r
g
lio
m
a
g
r
ad
es
an
d
th
r
ee
ty
p
es
o
f
t
h
e
b
r
ai
n
(
g
lio
m
a,
m
en
in
g
io
m
a,
an
d
p
itu
itar
y
)
,
u
n
lik
e
o
th
er
m
eth
o
d
s
r
est
o
n
tr
ial
a
n
d
e
r
r
o
r
.
T
h
e
ac
cu
r
ac
y
o
f
th
is
m
o
d
el
is
,
s
tu
d
y
I
:
9
0
.
0
%,
s
tu
d
y
I
I
:9
4
.
2
%.
Swatia
et
a
l
.
[
2
1
]
p
r
o
p
o
s
ed
a
m
o
d
el
to
class
if
y
MRI
o
f
b
r
ain
tu
m
o
r
s
in
to
t
h
r
ee
ty
p
es
(
g
lio
m
a,
p
itu
itar
y
,
an
d
m
e
n
in
g
io
m
a
)
.
T
h
ey
ar
e
u
s
ed
VGG1
9
to
in
itialized
weig
h
ts
.
Af
ter
th
at,
th
ey
ap
p
lied
f
in
e
-
tu
n
ed
VGG1
9
o
n
th
e
d
ataset.
T
h
e
av
er
ag
e
ac
cu
r
ac
y
was 9
4
.
8
2
%.
T
alo
a
et
a
l.
[
2
2
]
em
p
lo
y
th
e
Vg
g
-
1
6
,
Alex
Net,
R
esNet
-
3
4
,
R
esNet
-
1
8
,
an
d
R
esNet
-
5
0
p
r
e
-
tr
ain
ed
m
o
d
els
to
class
if
y
MRI
b
r
ain
in
to
f
iv
e
class
es
wh
ich
ar
e
n
o
r
m
al,
in
f
lam
m
ato
r
y
,
d
e
g
en
er
ati
v
e,
n
eo
p
last
ic,
an
d
ce
r
eb
r
o
v
asc
u
lar
d
is
ea
s
es c
las
s
es.
T
h
e
b
est ac
cu
r
ac
y
o
f
class
if
icatio
n
o
b
tain
ed
was 9
5
.
2
3
%
±
0
.
6
in
th
e
ca
s
e
o
f
th
e
R
esNet
-
5
0
m
o
d
el
a
m
o
n
g
o
th
er
m
o
d
els.
I
n
th
is
p
ap
er
,
a
m
o
d
el
o
f
C
NN
h
as
b
ee
n
p
r
esen
t
ed
to
class
if
y
MRI
o
f
b
r
ain
tu
m
o
r
s
in
to
th
r
ee
ty
p
es,
g
lio
m
a,
p
itu
itar
y
,
an
d
m
e
n
in
g
io
m
a.
T
h
e
n
etwo
r
k
ar
c
h
itectu
r
e
with
v
ar
io
u
s
n
u
m
b
er
s
o
f
lay
er
s
an
d
p
a
r
am
eter
s
is
d
ev
elo
p
ed
o
n
a
tr
ial
-
an
d
-
er
r
o
r
b
asis
to
ar
r
iv
e
at
th
e
b
est
m
o
d
el.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
co
n
s
is
ts
o
f
t
h
e
f
o
llo
win
g
s
tag
es:
Data
p
r
ep
r
o
ce
s
s
in
g
,
d
ata
au
g
m
en
tatio
n
,
lo
ca
l
izatio
n
o
f
b
r
ain
tu
m
o
r
s
,
C
NN
f
o
r
f
ea
tu
r
e
ex
tr
a
ctio
n
,
an
d
class
if
icatio
n
.
T
h
e
r
em
ain
d
er
o
f
th
e
p
a
p
er
is
o
r
g
an
ized
as
f
o
llo
ws;
th
e
m
ater
ials
an
d
m
eth
o
d
s
s
ec
tio
n
to
d
escr
ib
e
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
d
etail.
Af
ter
th
at,
th
e
r
es
u
lts
an
d
d
is
cu
s
s
io
n
s
ec
tio
n
to
s
u
m
m
a
r
ize
all
th
e
r
esu
lt
an
d
th
e
co
m
p
ar
is
o
n
o
b
t
ain
ed
f
r
o
m
t
h
e
m
o
d
el.
Fin
ally
,
th
e
co
n
clu
s
io
n
to
d
escr
ib
e
all
th
e
wo
r
k
b
r
ief
ly
.
2.
RE
S
E
ARCH
M
E
T
H
O
D
2
.
1
.
Da
t
a
s
et
prepa
ra
t
io
n
Me
d
ical
im
ag
es
th
at
h
a
v
e
b
ee
n
u
s
ed
in
th
is
wo
r
k
co
n
s
is
t
o
f
3
0
6
4
T
1
-
weig
h
ted
co
n
tr
ast
-
en
h
a
n
ce
d
MRI
(
C
E
-
MRI)
f
r
o
m
2
3
3
p
atien
ts
o
f
eith
er
s
ag
ittal,
ax
ial,
o
r
co
r
n
al
v
iews.
T
h
ese
d
ata
s
ets
wer
e
u
s
ed
in
C
h
en
g
et
a
l
.
[
1
]
f
o
r
class
if
icatio
n
an
d
was
c
o
llected
f
r
o
m
T
ian
jin
g
Me
d
ica
l
Un
iv
er
s
ity
,
Nan
f
a
n
g
Ho
s
p
ital
,
Gen
er
al
Ho
s
p
ital,
an
d
Gu
an
g
zh
o
u
in
C
h
in
a
f
r
o
m
2
0
0
5
to
2
0
1
2
.
2
.
2
.
Da
t
a
pre
-
pro
ce
s
s
ing
I
m
ag
e
p
r
o
ce
s
s
in
g
to
o
ls
h
av
e
b
ee
n
u
s
ed
ex
ten
s
iv
ely
in
m
ed
i
ca
l
im
ag
in
g
tec
h
n
iq
u
es
a
n
d
ca
n
im
p
r
o
v
e
th
e
ac
cu
r
ac
y
o
f
d
iag
n
o
s
tic
p
r
o
c
ess
es,
an
d
ty
p
ically
in
clu
d
e
im
ag
e
en
h
a
n
ce
m
en
t
to
r
e
d
u
ce
th
e
ef
f
ec
ts
o
f
co
r
r
u
p
ti
o
n
th
at
ca
n
co
n
tam
i
n
a
te
m
ed
ical
im
ag
es
d
u
r
in
g
th
e
ac
q
u
is
itio
n
o
r
tr
an
s
f
er
p
r
o
ce
s
s
[
2
3
]
.
I
n
th
is
th
esis
,
th
e
p
r
e
-
p
r
o
ce
s
s
in
g
o
n
MRI
b
r
ain
s
ca
n
s
lices
in
v
o
lv
es
im
p
le
m
en
tin
g
m
an
y
a
lg
o
r
it
h
m
s
as
a
p
r
ep
a
r
atio
n
f
o
r
th
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
in
th
e
c
o
n
v
o
lu
tio
n
al
la
y
er
.
T
h
is
p
r
ep
ar
ati
o
n
in
cl
u
d
es
MRI
d
im
en
s
io
n
s
r
esizin
g
an
d
u
s
in
g
a
Gau
s
s
ian
f
ilter
f
o
r
MRI
s
lice
en
h
an
ce
m
e
n
t.
I
n
th
e
ca
s
e
o
f
r
esizin
g
,
th
e
s
ca
n
o
f
MRI
was
r
esized
to
1
2
8
×1
2
8
p
ix
el
s
,
as
a
r
esu
lt,
th
e
al
g
o
r
ith
m
p
r
o
p
o
s
ed
in
th
is
wo
r
k
was
im
p
lem
en
ted
b
ased
o
n
u
s
in
g
s
q
u
ar
ed
s
lices.
On
th
e
o
th
er
h
a
n
d
,
im
a
g
e
en
h
a
n
ce
m
e
n
t is a
co
m
p
lex
task
th
at
is
h
ig
h
ly
d
ep
e
n
d
en
t
o
n
th
e
n
atu
r
e
o
f
th
e
im
ag
e.
Sev
er
al
ty
p
es
o
f
n
o
is
e
ca
n
b
e
f
o
u
n
d
in
im
ag
es
th
a
t
r
eq
u
ir
e
d
if
f
er
e
n
t
im
ag
e
en
h
an
ce
m
en
t
tech
n
iq
u
es.
T
h
e
v
is
u
al
q
u
ality
o
f
th
e
m
ed
ical
im
ag
e
p
lay
s
an
im
p
o
r
tan
t
r
o
le
in
th
e
ac
cu
r
ac
y
o
f
th
e
clin
ical
d
iag
n
o
s
is
b
ec
au
s
e
d
o
cto
r
s
ar
e
u
s
u
ally
tr
ain
ed
an
d
h
a
v
e
ex
p
e
r
ien
ce
in
s
p
ec
if
ic,
h
ig
h
-
q
u
ality
m
ed
ical
im
ag
es.
A
lo
w
-
p
ass
f
ilter
Gau
s
s
ian
f
ilter
was a
p
p
lied
f
o
r
n
o
is
e
r
em
o
v
al
[
2
4
]
.
2
.
3
.
B
ra
in t
um
o
r
lo
ca
t
io
n
Fo
r
f
aster
an
d
m
o
r
e
ac
cu
r
ate
d
iag
n
o
s
is
au
to
m
atica
lly
,
th
e
t
u
m
o
r
h
as
b
ee
n
d
etec
ted
.
T
h
is
o
p
er
atio
n
h
elp
s
th
e
m
o
d
el
to
f
o
c
u
s
o
n
a
s
p
ec
if
ic
r
eg
io
n
(
a
tu
m
o
r
ju
s
t)
in
th
e
im
ag
e
an
d
n
o
t a
ll th
e
im
ag
e
d
im
en
s
io
n
s
.
B
y
f
ee
d
in
g
th
e
n
e
u
r
al
n
etwo
r
k
with
th
e
im
ag
e
o
f
d
etec
ted
t
u
m
o
r
s
,
th
e
s
tr
u
ctu
r
e
ca
n
b
e
b
etter
lea
r
n
ed
,
an
d
s
tep
s
ar
e
tak
en
to
d
is
tin
g
u
is
h
b
r
ain
s
with
an
d
with
o
u
t tu
m
o
r
s
.
2
.
4
.
Da
t
a
a
ug
m
ent
a
t
io
n
W
h
en
u
s
in
g
m
u
lti
-
lay
er
ed
d
ee
p
n
ets
o
r
h
an
d
lin
g
a
lim
ited
n
u
m
b
er
o
f
tr
ain
i
n
g
im
ag
es,
th
e
r
e
is
a
r
is
k
o
f
o
v
er
f
itti
n
g
.
T
h
e
s
tan
d
a
r
d
s
o
lu
tio
n
to
r
ed
u
c
e
o
v
er
f
itti
n
g
is
to
in
cr
ea
s
e
d
ata
th
at
ar
tific
ially
ex
ten
d
th
e
d
ata
s
et
[
1
0
]
.
C
o
m
m
o
n
au
g
m
e
n
tatio
n
t
ec
h
n
iq
u
es
th
at
h
a
v
e
b
ee
n
u
s
ed
in
th
is
wo
r
k
i
n
clu
d
e
s
u
b
tr
an
s
f
o
r
m
atio
n
s
s
u
ch
as
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2
2
5
2
-
8
8
1
4
Dee
p
lea
r
n
in
g
fo
r
g
lio
ma
,
me
n
in
g
io
ma
,
a
n
d
p
itu
ita
r
y
cla
s
s
ifica
tio
n
(
To
g
a
A
.
S
a
d
o
o
n
)
91
r
o
tatio
n
with
4
5
an
g
les,
f
lip
p
i
n
g
,
m
ir
r
o
r
,
an
d
n
o
is
e
ad
d
itio
n
.
Fig
u
r
e
1
s
h
o
w
s
an
e
x
am
p
le
o
f
th
e
d
ata
s
et
af
ter
th
e
ab
o
v
e
p
r
o
ce
s
s
in
g
.
I
n
a
d
d
itio
n
,
T
ab
le
1
s
u
m
m
ar
ize
s
th
e
n
u
m
b
er
o
f
d
atasets
b
ef
o
r
e
a
n
d
af
ter
au
g
m
en
tatio
n
.
(
a)
(
b
)
(
c)
Fig
u
r
e
1
.
Sam
p
le
o
f
d
i
f
f
er
en
t t
y
p
es o
f
b
r
ain
tu
m
o
r
,
(
a)
g
lio
m
a,
(
b
)
m
en
in
g
i
o
m
a
,
an
d
(
c)
p
it
u
itar
y
af
ter
p
r
e
-
p
r
o
ce
s
s
in
g
an
d
d
ata
au
g
m
en
ta
tio
n
wh
ich
ar
e
t
h
e
o
r
i
g
in
al
im
ag
e,
r
o
tatin
g
b
y
4
5
d
eg
r
ee
s
,
ad
d
s
alt
&
p
ap
er
n
o
is
e,
m
ir
r
o
r
an
d
u
p
-
d
o
wn
f
lip
p
in
g
f
r
o
m
le
f
t to
r
ig
h
t
T
ab
le
1
.
T
h
e
n
u
m
b
er
o
f
d
ata
s
et
b
ef
o
r
e
a
n
d
af
te
r
au
g
m
en
tatio
n
C
a
t
e
g
o
r
y
N
u
mb
e
r
o
f
sl
i
c
e
b
e
f
o
r
e
a
u
g
me
n
t
a
t
i
o
n
N
u
mb
e
r
o
f
sl
i
c
e
s
a
f
t
e
r
a
u
g
m
e
n
t
a
t
i
o
n
G
l
i
o
ma
1
4
2
6
7
1
3
0
M
i
n
i
n
g
i
o
ma
7
0
8
3
5
4
0
P
i
t
u
i
t
a
r
y
9
3
0
4
6
5
0
To
t
a
l
3
0
6
4
1
5
3
2
0
2
.
5
.
CNN
cla
s
s
if
ica
t
io
n m
o
del
As we
m
en
tio
n
ed
in
th
e
p
r
ev
i
o
u
s
s
ec
tio
n
s
,
C
NN
is
a
n
eu
r
al
n
etwo
r
k
u
s
in
g
f
o
r
class
if
y
in
g
i
m
ag
es a
n
d
d
is
p
lay
ed
g
o
o
d
p
er
f
o
r
m
a
n
ce
in
ca
teg
o
r
izin
g
d
if
f
er
en
t
s
u
p
er
v
i
s
ed
lear
n
in
g
task
s
[
2
5
]
.
W
e
h
a
v
e
b
ee
n
test
ed
m
an
y
lay
er
s
an
d
p
ar
am
eter
s
in
th
is
wo
r
k
,
an
d
th
e
b
est
o
n
e
f
o
r
th
is
m
o
d
el
was
th
e
f
o
llo
win
g
.
T
h
e
m
o
d
el
in
clu
d
es
2
8
lay
e
r
s
th
at
b
eg
a
n
with
th
e
in
p
u
t
lay
er
th
at
tak
es
th
e
im
ag
e
with
1
2
8
×
1
2
8
×
3
s
ize
af
ter
p
r
e
-
p
r
o
ce
s
s
in
g
an
d
au
g
m
en
tatio
n
.
T
h
ese
im
ag
es
ar
e
p
ass
in
g
th
r
o
u
g
h
s
ix
lay
er
s
o
f
(
co
n
v
o
lu
tio
n
al
lay
er
,
r
ec
tifie
d
l
in
ea
r
u
n
it
(
R
eL
U
)
,
an
d
m
ax
-
p
o
o
lin
g
lay
er
)
r
esp
e
ctiv
ely
.
Ab
s
o
lu
tely
,
we
u
s
e
f
i
v
e
d
r
o
p
o
u
t
lay
er
s
to
p
r
ev
en
t
f
r
o
m
o
v
e
r
f
itti
n
g
.
T
h
e
last
th
r
ee
lay
er
s
ar
e
in
s
eq
u
e
n
ce
f
u
lly
co
n
n
ec
ted
la
y
er
,
th
e
s
o
f
tm
ax
la
y
er
,
a
n
d
f
in
ally
class
if
icatio
n
lay
er
.
Mo
r
eo
v
er
,
we
u
s
e
af
ter
th
e
f
ir
s
t
c
o
n
v
o
lu
tio
n
al
lay
er
b
atc
h
n
o
r
m
aliza
tio
n
lay
er
.
T
h
e
f
o
llo
win
g
s
en
ten
ce
s
d
escr
ib
e
th
e
b
eh
av
io
r
o
f
ea
ch
lay
er
in
d
etails
.
T
h
e
in
p
u
t
lay
er
is
u
s
ed
to
en
ter
th
e
tr
ain
in
g
d
ata
to
th
e
m
o
d
el
with
in
p
u
t size
1
2
8
×
128
×
3.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
88
–
98
92
T
h
e
co
n
v
o
lu
tio
n
al
la
y
er
is
u
s
ed
to
ex
tr
ac
t
th
e
f
ea
tu
r
e
f
r
o
m
th
e
in
p
u
t
im
a
g
e.
I
n
th
is
lay
er
,
th
er
e
is
a
f
ilter
ca
lled
k
er
n
el
th
at
h
as
b
e
en
co
n
v
o
lv
e
d
with
th
e
i
n
p
u
t
i
m
ag
e.
An
d
we
m
u
s
t
m
e
n
tio
n
to
th
at,
th
e
k
er
n
el
in
th
e
ea
r
ly
lay
er
s
u
s
ed
t
o
ex
tr
ac
t
lo
w
-
lev
el
f
ea
tu
r
es
f
r
o
m
th
e
i
m
ag
e
s
u
ch
as
lin
es
an
d
e
d
g
es.
W
h
ile
th
e
k
er
n
el
in
th
e
ad
v
an
ce
d
lay
er
s
is
u
s
ed
to
ex
tr
ac
t c
o
m
p
lex
f
ea
tu
r
es [
2
6
]
.
T
h
e
o
u
tp
u
t
o
f
th
is
lay
er
is
a
n
ew
s
et
o
f
im
ag
es c
alled
f
ea
tu
r
e
m
ap
s
,
wh
ich
is
eq
u
al
to
th
e
n
u
m
b
er
o
f
a
k
er
n
el
th
at
h
as
b
ee
n
u
s
ed
i
n
th
is
lay
er
[
1
0
]
.
I
n
th
is
wo
r
k
,
t
h
e
n
u
m
b
er
s
o
f
f
ilter
s
ar
e
6
4
,
6
4
,
9
6
,
9
6
,
1
2
8
,
an
d
1
2
8
with
k
er
n
el
s
ize
7
×
7
,
9
×
9
,
9
×
9
,
9
×
9
,
1
1
×
1
1
,
an
d
1
1
×
1
1
r
esp
ec
tiv
ely
.
Strid
e
is
m
o
v
in
g
alo
n
g
th
e
v
er
tical
o
r
h
o
r
izo
n
tal
p
o
s
itio
n
o
f
th
e
im
ag
e
b
y
o
n
e
o
r
m
o
r
e
s
tep
s
ize
th
r
o
u
g
h
a
co
n
v
o
l
u
tio
n
al
o
p
er
ati
o
n
.
T
h
e
s
tr
id
e
s
ize
is
o
n
e
f
o
r
all
th
e
co
n
v
o
lu
tio
n
al
la
y
er
s
.
B
u
t
wh
en
we
g
iv
e
a
b
o
r
d
er
s
ize
o
f
th
e
im
ag
e
m
o
r
e
im
p
o
r
tan
ce
th
is
is
ca
lled
b
y
p
ad
d
in
g
an
d
th
is
is
d
o
n
e
b
y
ad
d
in
g
e
x
tr
a
r
o
w
an
d
co
lu
m
n
ar
o
u
n
d
th
e
im
ag
e
m
atr
i
x
.
T
h
e
p
ad
d
in
g
s
ize
u
s
ed
in
th
is
wo
r
k
i
s
0
,
1
,
1
,
1
,
1
,
an
d
1
.
Fig
u
r
e
2
s
h
o
w
s
an
ex
am
p
l
e
o
f
a
co
n
v
o
l
u
tio
n
al
lay
er
o
p
er
atio
n
with
a
k
er
n
el
s
ize
o
f
3
×
3.
Fig
u
r
e
2
.
C
o
n
v
o
lu
tio
n
al
lay
er
T
h
e
b
atch
n
o
r
m
aliza
tio
n
lay
e
r
is
u
s
ed
to
n
o
r
m
alize
th
e
tr
ain
i
n
g
d
ata
d
u
r
in
g
t
r
ain
in
g
p
r
o
ce
s
s
in
g
r
ath
er
th
an
n
o
r
m
alizin
g
all
th
e
d
ata
s
et
in
th
e
p
r
e
-
p
r
o
ce
s
s
in
g
s
tep
a
n
d
th
is
p
r
o
ce
s
s
will
d
ec
r
ea
s
e
t
h
e
tr
ain
in
g
tim
e
[
2
7
]
.
E
ac
h
co
n
v
o
lu
tio
n
al
la
y
er
is
f
o
l
lo
wed
b
y
a
n
ac
tiv
atio
n
f
u
n
ctio
n
u
s
ed
to
d
eter
m
in
e
th
e
b
eh
a
v
io
r
o
f
t
h
e
co
n
n
ec
tio
n
n
o
d
e.
I
n
o
u
r
m
o
d
el
we
u
s
e
r
e
ctif
ier
lin
ea
r
u
n
it
(
R
eL
U)
,
th
e
o
u
tp
u
t
o
f
th
is
ac
tiv
atio
n
is
a
p
o
s
itiv
e
n
u
m
b
er
a
n
d
ze
r
o
.
T
h
e
f
o
llo
win
g
e
q
u
atio
n
is
th
e
m
ath
em
atica
l
r
ep
r
esen
tatio
n
f
o
r
th
is
f
u
n
ctio
n
in
(
1
)
.
Fig
u
r
e
3
s
h
o
w
s
th
e
b
eh
av
io
r
o
f
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
.
(
)
=
(
0
,
)
(
1
)
Fig
u
r
e
3
.
R
eL
U
[
1
3
]
As
f
o
r
th
e
m
ax
-
p
o
o
lin
g
lay
er
,
it
is
u
s
ed
to
r
ed
u
ce
th
e
f
ea
tu
r
e
m
ap
s
d
im
en
s
io
n
af
ter
a
c
o
n
v
o
lu
tio
n
al
o
p
er
atio
n
.
Similar
to
th
e
co
n
v
o
lu
tio
n
al
lay
er
,
p
o
o
lin
g
lay
e
r
also
h
as
a
f
ilter
m
o
v
in
g
o
n
th
e
f
ea
tu
r
e
m
ap
an
d
as
a
r
esu
lt r
e
d
u
ce
s
th
e
co
m
p
u
tati
o
n
o
f
t
h
e
n
etwo
r
k
[
16
]
,
[
28
]
,
[
2
9
]
.
T
h
e
f
ilter
'
s
s
ize
is
2
×
2
wit
h
o
n
e
s
tr
id
e
s
ize
f
o
r
all
m
ax
-
p
o
o
lin
g
lay
e
r
s
.
Fig
u
r
e
4
s
h
o
w
s
th
e
m
ax
-
p
o
o
lin
g
lay
e
r
with
a
2
×
2
f
ilter
s
ize.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2
2
5
2
-
8
8
1
4
Dee
p
lea
r
n
in
g
fo
r
g
lio
ma
,
me
n
in
g
io
ma
,
a
n
d
p
itu
ita
r
y
cla
s
s
ifica
tio
n
(
To
g
a
A
.
S
a
d
o
o
n
)
93
Fig
u
r
e
4
.
E
x
am
p
le
o
f
m
ax
p
o
o
lin
g
lay
er
On
e
o
f
th
e
c
o
m
m
o
n
p
r
o
b
lem
s
th
r
o
u
g
h
tr
ain
in
g
is
o
v
e
r
f
itti
n
g
.
Ov
er
f
itti
n
g
is
m
ea
n
th
at,
b
e
s
t
lear
n
in
g
p
er
f
o
r
m
an
ce
ag
ain
s
t
b
a
d
test
in
g
p
er
f
o
r
m
a
n
ce
.
I
n
o
r
d
er
t
o
p
r
e
v
en
t
th
e
m
o
d
el
f
r
o
m
o
v
e
r
f
itti
n
g
,
we
u
s
e
a
d
r
o
p
o
u
t
lay
er
.
I
n
th
is
la
y
er
,
s
o
m
e
n
o
d
es
h
av
e
b
ee
n
s
elec
ted
r
an
d
o
m
ly
an
d
s
et
to
ze
r
o
,
th
e
n
u
m
b
e
r
o
f
s
elec
ted
n
o
d
es
d
ep
en
d
e
d
o
n
a
p
er
ce
n
ta
g
e
v
al
u
e.
I
n
th
e
p
r
o
p
o
s
ed
m
o
d
el,
we
f
o
u
n
d
t
h
at
th
e
b
est d
r
o
p
o
u
t
p
r
o
b
ab
ilit
ies ar
e
1
0
%,
1
0
%,
2
0
%,
2
0
%,
an
d
2
0
%
r
esp
ec
tiv
ly
f
o
r
t
h
e
f
iv
e
d
r
o
p
o
u
t
l
ay
er
s
.
An
ex
am
p
le
o
f
th
e
d
r
o
p
o
u
t
lay
er
s
h
o
wn
in
Fig
u
r
e
5.
Fig
u
r
e
5
.
E
x
am
p
le
o
f
d
r
o
p
o
u
t
lay
er
[
3
0
]
Fin
ally
,
th
e
last
t
h
r
ee
la
y
er
s
wh
ich
ar
e
f
u
lly
co
n
n
ec
ted
la
y
er
,
t
h
e
s
o
f
tm
a
x
la
y
er
f
o
llo
w
ed
b
y
th
e
class
if
icatio
n
lay
er
.
T
h
e
f
ir
s
t
l
ay
er
(
f
u
lly
c
o
n
n
ec
ted
lay
er
)
is
u
s
ed
to
co
n
v
er
t
th
e
two
-
d
im
e
n
s
io
n
al
im
ag
e
i
n
to
1
D.
I
n
th
is
lay
er
,
ea
ch
n
eu
r
o
n
i
s
co
n
n
ec
ted
to
th
e
p
r
e
v
io
u
s
n
e
u
r
o
n
an
d
th
e
n
ex
t
n
e
u
r
o
n
.
T
h
e
o
u
tp
u
t
o
f
th
is
lay
er
is
th
e
s
am
e
n
u
m
b
er
o
f
ca
teg
o
r
ies
wh
ich
ar
e
th
r
ee
class
es
in
o
u
r
ca
s
e.
T
h
e
last
lay
er
o
f
f
u
ll
y
co
n
n
ec
ted
lay
er
s
u
s
es a
s
p
ec
ial
f
u
n
ctio
n
to
p
r
ed
ict
th
e
p
r
o
b
ab
le
o
u
tc
o
m
e
f
o
r
ea
ch
ca
teg
o
r
y
,
a
n
d
th
e
b
ig
g
est v
alu
e
o
f
p
r
o
b
ab
ilit
y
r
ep
r
esen
ted
t
h
e
co
r
r
ec
t
class
.
I
n
th
is
m
o
d
el,
a
s
o
f
tm
ax
f
u
n
cti
o
n
h
as
b
ee
n
u
s
ed
.
T
o
ca
lcu
lat
e
th
e
o
u
tp
u
t
o
f
t
h
is
lay
er
we
ca
n
u
s
e
(
2
)
.
Fig
u
r
e
6
s
h
o
w
s
an
ex
am
p
le
o
f
th
e
last
t
h
r
ee
lay
er
s
.
f(x
i
)
=
∑
=
1
(
2
)
Fig
u
r
e
6
.
So
f
tm
a
x
lay
er
with
n
eu
r
al
n
etwo
r
k
[
3
1
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
88
–
98
94
L
astl
y
,
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
c
tio
n
h
as
b
ee
n
u
s
ed
in
th
e
class
if
icatio
n
lay
er
to
d
eter
m
i
n
e
th
e
er
r
o
r
(
th
e
d
if
f
er
en
ce
b
etwe
en
th
e
ac
tu
al
o
u
tp
u
t
an
d
th
e
p
r
ed
icted
o
u
t
p
u
t)
o
f
th
e
class
if
icatio
n
an
d
p
r
o
d
u
ce
th
e
f
in
al
p
r
ed
icted
class
f
o
r
ea
c
h
in
p
u
t im
ag
e.
T
h
e
(
3
)
u
s
ed
to
ca
lcu
lat
e
th
e
er
r
o
r
:
J=
∑
{
−
ln
(
)
−
(
1
−
)
ln
(
1
−
)
}
+
1
2
‖
‖
2
=
1
(3
)
w
h
er
e
is
th
e
o
u
t
p
u
t
o
f
th
e
n
etwo
r
k
,
is
t
h
e
c
o
r
r
ec
t
o
u
tp
u
t,
λ
i
s
a
co
ef
f
icien
t r
elate
d
to
th
e
c
o
n
n
ec
tio
n
weig
h
t a
n
d
co
s
t
f
u
n
ctio
n
an
d
r
e
p
r
esen
t
th
e
o
u
tp
u
t
n
o
d
e
u
n
til
M
n
o
d
es.
Mo
r
eo
v
er
,
to
r
ed
u
ce
th
e
er
r
o
r
we
u
s
e
s
to
ch
asti
c
g
r
ad
ien
t d
escen
t w
ith
m
o
m
en
t
u
m
(
SGDM
)
as a
n
o
p
tim
izer
m
eth
o
d
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
s
h
o
ws
in
Fig
ur
e
7
.
Fig
u
r
e
7
.
Pro
p
o
s
ed
wo
r
k
b
l
o
c
k
d
iag
r
am
3.
RE
SU
L
T
S
A
ND
AN
AL
Y
SI
S
W
e
d
iv
id
ed
th
e
d
ata
s
et
in
to
7
7
%
f
o
r
tr
ai
n
in
g
,
2
0
%
f
o
r
v
alid
atio
n
,
an
d
th
e
r
em
ain
d
e
r
u
s
ed
f
o
r
m
o
d
el
test
in
g
.
I
n
th
e
tr
ain
in
g
p
r
o
ce
s
s
f
o
r
d
ee
p
lear
n
in
g
,
m
o
m
en
tu
m
is
s
et
to
0
.
9
,
th
e
m
ax
im
u
m
iter
atio
n
3
6
8
0
0
,
th
e
ep
o
ch
is
1
0
0
,
th
e
i
n
itial
lear
n
in
g
r
ate
is
0
.
0
0
0
1
an
d
th
e
m
in
i
-
b
atch
s
ize
is
3
2
.
Fig
u
r
e
8
s
h
o
w
th
e
ac
cu
r
ac
y
a
n
d
er
r
o
r
f
o
r
b
o
th
tr
ain
in
g
v
alid
ati
o
n
p
r
o
g
r
ess
.
Fig
u
r
e
8
s
h
o
w
s
th
at
af
ter
1
0
0
0
0
iter
atio
n
s
th
e
ac
cu
r
ac
y
b
ec
am
e
n
ea
r
to
1
0
0
%
an
d
in
th
e
f
in
al,
th
e
b
est
v
alid
atio
n
ac
cu
r
ac
y
o
b
tai
n
ed
is
9
6
.
1
%.
W
h
ile
th
e
lo
s
s
f
u
n
ctio
n
is
less
th
an
0
.
2
.
W
e
m
u
s
t
m
en
tio
n
th
at
b
ec
au
s
e
o
f
u
s
in
g
3
2
im
ag
es
as
a
m
in
i
-
b
atch
s
ize
th
e
cu
r
v
e
f
ir
s
tly
d
r
o
p
s
s
h
ar
p
ly
with
s
o
m
e
f
lu
ctu
atio
n
s
[
3
2
]
b
u
t
th
e
s
e
ten
d
to
d
is
ap
p
ea
r
af
ter
1
0
0
0
0
iter
atio
n
s
f
o
r
b
o
th
cu
r
v
es.
On
th
e
o
th
e
r
h
a
n
d
,
f
o
r
m
o
d
el
test
in
g
,
we
u
s
e
4
5
9
s
lices a
n
d
th
e
m
o
d
el
s
h
o
ws th
e
te
s
t a
cc
u
r
ac
y
o
f
9
3
.
2
%.
Fig
u
r
e
8
.
T
r
ain
in
g
p
r
o
ce
s
s
3
.
1
.
N
um
ber
o
f
la
y
er
a
nd
hy
per
pa
ra
m
et
er
s
I
n
th
is
s
u
b
s
ec
tio
n
,
th
e
d
if
f
er
e
n
t
p
ar
am
eter
s
an
d
n
u
m
b
er
o
f
l
ay
er
s
o
f
th
e
m
o
d
el
th
at
h
av
e
b
ee
n
test
ed
u
n
til r
ea
ch
in
g
th
e
b
est m
o
d
el
ar
e
p
r
esen
ted
in
T
ab
le
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2
2
5
2
-
8
8
1
4
Dee
p
lea
r
n
in
g
fo
r
g
lio
ma
,
me
n
in
g
io
ma
,
a
n
d
p
itu
ita
r
y
cla
s
s
ifica
tio
n
(
To
g
a
A
.
S
a
d
o
o
n
)
95
T
ab
le
2
.
Var
io
u
s
p
r
ev
i
o
u
s
ly
te
s
ted
p
ar
am
eter
s
an
d
la
y
er
s
Tr
i
a
l
a
n
d
e
r
r
o
r
p
a
r
a
me
t
e
r
s
P
o
o
l
i
n
g
l
a
y
e
r
s
t
r
i
d
e
N
u
mb
e
r
o
f
(
C
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
e
r
+
R
e
LU
+
P
o
o
l
i
n
g
l
a
y
e
r
)
1
,
2
,
3
,
4
,
5
,
6
P
o
o
l
i
n
g
l
a
y
e
r
M
a
x
,
a
v
e
r
a
g
e
p
o
o
l
i
n
g
D
r
o
p
o
u
t
l
a
y
e
r
1
,
2
,
3
,
4
D
r
o
p
o
u
t
r
a
t
i
o
0
.
1
,
0
.
1
5
,
0
.
2
,
0
.
2
5
,
0
.
3
,
0
.
4
,
0
.
5
Ep
o
c
h
3
0
,
5
0
,
7
0
,
1
0
0
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
1
,
0
,
0
0
1
,
0
.
0
0
0
1
,
1
e
-
4
,
1e
-
5
O
p
t
i
mi
z
a
t
i
o
n
me
t
h
o
d
s
S
G
D
,
A
d
a
m
M
i
n
i
-
b
a
t
c
h
s
i
z
e
1
6
,
3
2
N
u
mb
e
r
o
f
k
e
r
n
e
l
4
8
,
6
4
,
9
6
,
1
2
8
A
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
R
e
LU
,
l
e
a
k
y
R
e
LU
K
e
r
n
e
l
si
z
e
3
,
5
,
7
,
9
,
1
1
C
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
e
r
p
a
d
d
i
n
g
0
,
1
,
2
,
3
,
4
C
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
e
r
st
r
i
d
e
1
,
2
P
o
o
l
i
n
g
l
a
y
e
r
p
a
d
d
i
n
g
0
,
1
P
o
o
l
i
n
g
l
a
y
e
r
st
r
i
d
e
1
,
2
3
.
2
.
Co
nf
us
io
n m
a
t
rix
T
h
e
co
n
f
u
s
io
n
m
atr
ices h
a
v
e
b
ee
n
u
s
ed
to
m
ea
s
u
r
e
th
e
m
o
d
e
l's
p
er
f
o
r
m
an
ce
f
o
r
o
u
r
s
tu
d
y
.
Pre
cisi
o
n
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
an
d
ac
cu
r
ac
y
h
a
v
e
b
ee
n
d
eter
m
i
n
ed
u
s
in
g
(
4
)
-
(
7
)
.
=
/
(
+
)
(
4
)
=
/
(
+
)
(
5
)
=
/
(
+
)
(
6
)
=
(
+
)
/
(
)
(
7
)
W
h
er
e,
T
P,
FP
,
T
N,
FN
ar
e
tr
u
e
p
o
s
itiv
e,
f
alse
p
o
s
itiv
e,
tr
u
e
n
eg
ativ
e,
an
d
f
alse
n
eg
ativ
e,
r
esp
ec
tiv
ely
.
T
o
d
escr
ib
e
th
e
co
n
f
u
s
io
n
m
at
r
ix
,
we
will m
en
tio
n
to
p
e
o
p
le
with
tu
m
o
r
b
y
p
o
s
itiv
e
an
d
p
eo
p
l
e
with
o
u
t
tu
m
o
r
b
y
n
eg
ativ
e.
Mo
r
eo
v
er
,
tr
u
e
a
n
d
f
alse f
o
r
co
r
r
ec
tly
an
d
i
n
co
r
r
ec
t
ly
d
iag
n
o
s
e
r
esp
ec
tiv
ely
.
So
,
-
T
r
u
e
p
o
s
itiv
e
(
T
P):
p
eo
p
le
wit
h
tu
m
o
r
co
r
r
ec
tly
id
en
tifie
d
(
c
o
r
r
ec
tly
d
ia
g
n
o
s
e)
.
-
Fals
e
p
o
s
itiv
e
(
FP
)
: p
eo
p
le
with
o
u
t tu
m
o
r
in
c
o
r
r
ec
tly
id
e
n
tifi
ed
p
eo
p
le
with
tu
m
o
r
(
in
co
r
r
e
ctly
d
iag
n
o
s
e)
.
-
T
r
u
e
n
e
g
ativ
e
(
T
N)
: p
e
o
p
le
wi
th
o
u
t tu
m
o
r
co
r
r
ec
tly
id
en
tifie
d
as h
ea
lth
y
(
c
o
r
r
ec
tly
d
iag
n
o
s
e)
.
-
Fals
e
n
eg
ativ
e
(
FN)
:
p
eo
p
le
with
tu
m
o
r
in
co
r
r
ec
tly
id
e
n
tifie
d
as
p
eo
p
le
with
o
u
t
tu
m
o
r
(
in
co
r
r
ec
tly
d
iag
n
o
s
e)
.
Fig
u
r
e
9
s
h
o
w
s
th
e
ac
cu
r
ac
ies
th
at
ar
e
f
o
u
n
d
f
r
o
m
t
h
e
co
n
f
u
s
io
n
m
atr
ix
an
d
s
u
m
m
ar
ize
d
in
T
ab
le
3
.
Pre
cisi
o
n
o
f
9
9
.
1
% f
o
r
p
itu
itar
y
,
s
en
s
itiv
ity
o
f
9
8
.
7
% f
o
r
g
lio
m
a,
s
p
ec
if
icity
o
f
9
9
.
1
%,
an
d
ac
cu
r
ac
y
o
f
9
9
.
1
%
f
o
r
p
itu
itar
y
ar
e
th
e
h
ig
h
est p
e
r
f
o
r
m
a
n
ce
.
Fig
u
r
e
9
.
C
o
n
f
u
s
io
n
m
atr
i
x
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
88
–
98
96
T
ab
le
3
.
C
o
n
f
u
s
io
n
m
atr
ix
ac
c
u
r
ac
y
Tu
m
o
r
t
y
p
e
TP
TN
FP
FN
P
r
e
c
i
s
i
o
n
%
S
e
n
s
i
t
i
v
i
t
y
%
S
p
e
c
i
f
i
c
i
t
y
%
A
c
c
u
r
a
c
y
%
G
l
i
o
ma
1
3
3
7
1
6
2
0
89
18
9
3
.
8
9
8
.
7
9
4
.
8
9
6
.
5
M
e
n
i
n
g
i
o
ma
6
8
5
2
2
7
5
23
81
9
6
.
8
8
9
.
4
9
6
.
8
9
6
.
6
P
i
t
u
i
t
a
r
y
9
2
2
2
1
1
3
8
21
9
9
.
1
9
7
.
8
9
9
.
1
9
9
.
1
3
.
3
.
T
ra
ini
ng
t
o
o
ls
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
b
r
ain
tu
m
o
r
class
if
icatio
n
is
tr
ain
ed
o
n
I
n
tel
(
R
)
C
o
r
e
(
T
M)
i3
-
4
0
0
5
U
C
PU
@
1
.
7
GHz
,
R
AM
(
4
GB
)
,
NVI
DI
A
GeFo
r
ce
9
2
0
M
GPU,
N
VI
DI
A
C
UDA
1
0
.
1
.
2
3
6
,
an
d
Ma
tlab
2
0
1
8
b
.
3
.
4
.
Co
m
pa
riso
n
wit
h
o
t
her
cla
s
s
if
ica
t
io
n m
o
dels
a
nd
dis
cus
s
i
o
n
Fo
r
th
e
p
u
r
p
o
s
e
o
f
c
o
m
p
ar
is
o
n
an
d
to
p
r
o
v
e
th
e
s
u
p
er
io
r
ity
o
f
o
u
r
m
o
d
el
o
v
er
th
e
r
est
o
f
t
h
e
m
o
d
els
u
s
ed
s
im
ilar
im
ag
es o
f
b
r
ain
tu
m
o
r
ty
p
es,
we
u
s
ed
th
r
ee
ca
s
es f
o
r
co
m
p
ar
is
o
n
:
Firstl
y
,
we
co
m
p
ar
e
o
u
r
m
o
d
el
with
th
e
p
r
ev
io
u
s
C
NN
m
o
d
el
s
,
an
d
we
s
ee
th
r
o
u
g
h
c
o
m
p
r
e
s
s
io
n
th
at
o
u
r
m
o
d
el
s
h
o
ws
th
e
b
est
p
er
f
o
r
m
a
n
ce
.
T
h
en
we
u
s
e
two
ty
p
es
o
f
p
r
e
-
tr
ain
ed
m
o
d
els,
R
esNet
-
5
0
an
d
Alex
Net,
an
d
we
s
ee
th
at
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
o
v
er
co
m
es
th
e
p
r
ev
io
u
s
ly
tr
ai
n
ed
m
o
d
els
in
th
e
ca
s
e
o
f
th
e
d
ata
s
et
th
at
was
u
s
ed
in
th
is
wo
r
k
.
T
ab
el
4
s
h
o
ws
th
e
co
m
p
r
ess
io
n
am
o
n
g
th
e
p
r
o
p
o
s
ed
m
o
d
el
an
d
th
e
o
th
er
m
o
d
els.
Fin
ally
Fig
u
r
es
1
0
an
d
1
1
s
h
o
w
th
e
t
r
ain
in
g
p
r
o
g
r
ess
a
n
d
c
o
n
f
u
s
io
n
m
at
r
ix
o
f
th
e
o
r
i
g
in
al
d
ataset
r
esp
ec
tiv
ely
.
I
t
is
clea
r
f
o
r
u
s
t
o
s
ee
th
e
lar
g
e
g
ap
b
etwe
e
n
r
esu
lts
b
ef
o
r
e
an
d
af
ter
d
ate
p
r
e
-
p
r
o
ce
s
s
in
g
a
n
d
a
u
g
m
e
n
tatio
n
wh
ich
is
s
h
o
wn
i
n
Fig
u
r
e
8
,
Fig
u
r
e
9
,
Fig
u
r
e
1
0
,
an
d
Fig
u
r
e
1
1
.
I
t
is
wo
r
th
m
e
n
tio
n
in
g
th
at;
in
th
e
tu
m
o
r
lo
ca
lizatio
n
s
tep
we
h
av
e
b
ee
n
u
s
ed
th
e
s
eg
m
e
n
ted
im
ag
e
th
at
is
av
ailab
le
with
th
e
d
ata
s
et,
b
u
t
in
f
u
tu
r
e,
we
ca
n
m
ak
e
a
C
AD
to
s
eg
m
en
t
th
e
im
ag
e,
d
etec
ted
th
e
tu
m
o
r
,
a
n
d
f
i
n
ally
class
if
icatio
n
.
T
ab
le
4
.
A
co
m
p
ar
is
o
n
b
etwe
en
th
e
p
r
e
v
io
u
s
r
elate
d
w
o
r
k
s
a
n
d
th
e
p
r
o
p
o
s
ed
o
u
r
m
o
d
el
M
o
d
e
l
A
c
c
u
r
a
c
y
%
C
h
e
n
g
e
t
a
l
.
[
1
]
9
1
.
2
8
P
a
u
l
e
t
a
l
.
[
1
8
]
9
1
.
4
3
A
f
sh
a
r
e
t
a
l
.
[
1
9
]
9
0
.
8
9
K
a
b
i
r
e
t
a
l
.
[
2
0
]
9
4
.
2
S
w
a
t
i
e
t
a
l
.
[
2
1
]
9
4
.
8
2
M
u
h
a
mm
e
d
e
t
a
l
.
[
2
2
]
9
5
.
2
3
±
0
.
6
A
l
e
x
N
e
t
8
2
.
2
R
e
s
n
e
t
-
50
7
5
.
6
P
r
o
p
o
se
d
m
o
d
e
l
9
6
.
1
Fig
u
r
e
1
0
.
T
h
e
tr
ain
i
n
g
p
r
o
ce
s
s
o
f
th
e
o
r
i
g
in
al
d
ata
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2
2
5
2
-
8
8
1
4
Dee
p
lea
r
n
in
g
fo
r
g
lio
ma
,
me
n
in
g
io
ma
,
a
n
d
p
itu
ita
r
y
cla
s
s
ifica
tio
n
(
To
g
a
A
.
S
a
d
o
o
n
)
97
Fig
u
r
e
1
1
.
C
o
n
f
u
s
io
n
m
atr
ix
o
f
th
e
o
r
i
g
in
al
d
ata
4.
CO
NCLU
SI
O
N
I
n
th
is
r
esear
ch
,
a
b
r
ain
tu
m
o
r
class
if
icatio
n
m
o
d
el
was
p
r
o
p
o
s
ed
with
m
ag
n
etic
r
eso
n
an
c
e
I
m
ag
in
g
to
class
if
y
th
e
b
r
ain
tu
m
o
r
in
to
th
r
ee
ty
p
es
wh
ich
ar
e
m
en
in
g
io
m
a,
g
lio
m
a,
an
d
p
itu
itar
y
g
la
n
d
b
ased
o
n
C
NN.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
co
n
s
is
ts
o
f
2
8
lay
er
s
s
tar
tin
g
with
an
in
p
u
t
lay
er
th
at
tak
es
t
h
e
in
p
u
t
im
a
g
es,
6
co
n
v
o
l
u
tio
n
al
lay
er
s
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
a
n
o
r
m
aliza
tio
n
lay
er
f
o
r
n
o
r
m
alizin
g
im
a
g
es,
6
R
eL
U
lay
er
s
f
u
n
ctio
n
,
6
lay
er
s
f
o
r
m
ax
-
p
o
o
lin
g
to
r
ed
u
ce
t
h
e
d
im
en
s
io
n
s
o
f
f
ea
tu
r
e
m
a
p
s
,
5
d
r
o
p
o
u
t la
y
er
s
to
p
r
ev
en
t f
r
o
m
o
v
er
f
itti
n
g
,
a
f
u
lly
co
n
n
ec
te
d
lay
er
as
a
f
lattin
g
lay
er
,
s
o
f
tm
ax
lay
er
to
f
in
d
f
o
r
ea
ch
class
i
t’
s
p
r
o
b
ab
ilit
y
an
d
f
in
ally
th
e
class
if
icatio
n
lay
er
to
p
r
e
d
ict
th
e
o
u
tp
u
t.
B
esid
es,
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
a
n
d
au
g
m
en
tatio
n
h
el
p
ed
o
u
r
m
o
d
el
to
s
h
o
w
b
etter
ac
c
u
r
a
cy
,
an
d
th
is
h
as
b
ee
n
illu
s
tr
ated
in
t
h
e
p
a
p
er
a
b
o
v
e
.
Mo
r
eo
v
er
,
to
p
r
o
v
e
th
e
s
u
p
er
io
r
ity
o
f
o
u
r
m
o
d
el
o
v
e
r
t
h
e
r
est
o
f
th
e
m
o
d
els,
we
p
r
es
en
ted
a
co
m
p
ar
is
o
n
am
o
n
g
t
h
e
m
.
T
h
e
ac
cu
r
ac
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
u
p
to
9
6
.
1
%.
RE
F
E
R
E
NC
E
S
[1
]
J.
Ch
e
n
g
,
e
t
a
l
.,
“
En
h
a
n
c
e
d
p
e
rfo
r
m
a
n
c
e
o
f
b
ra
in
tu
m
o
r
c
las
sifica
ti
o
n
v
ia t
u
m
o
r
re
g
i
o
n
a
u
g
m
e
n
tatio
n
a
n
d
p
a
rti
ti
o
n
,”
P
lo
S
o
n
e
,
v
o
l.
10
,
n
o
.
1
2
,
Oc
t.
2
0
1
5
.
[2
]
Wen
P
.
Y
.
,
Ke
sa
ri
S
.
,
“
M
a
li
g
n
a
n
t
g
li
o
m
a
s in
a
d
u
lt
s
,”
N E
n
g
l
J
M
e
d
,
v
o
l
.
3
5
9
,
n
o
.
5
,
p
p
.
4
9
2
-
5
0
7
,
2
0
0
8
.
[3
]
M
.
L.
G
o
o
d
e
n
b
e
rg
e
r
a
n
d
R.
B
.
J
e
n
k
in
s,
“
G
e
n
e
ti
c
s
o
f
a
d
u
lt
g
li
o
m
a
,
”
Ca
n
c
e
r
Ge
n
e
t
.
,
v
o
l.
2
0
5
,
n
o
.
1
2
,
p
p
.
6
1
3
-
6
2
1
,
De
c
.
2
0
1
2
.
[4
]
Law
s E
.
R
.
,
e
t
a
l
.,
“
P
it
u
it
a
ry
d
iso
r
d
e
rs:
d
iag
n
o
sis a
n
d
m
a
n
a
g
e
m
e
n
t
,”
J
o
h
n
W
il
e
y
&
S
o
n
s
,
2
0
1
3
.
[5
]
Blac
k
P
.
M.
,
“
Bra
in
tu
m
o
rs
,”
N
E
n
g
l
J
M
e
d
,
v
o
l.
3
2
4
,
n
o
.
2
2
,
p
p
.
1
5
5
5
-
1
5
6
4
,
1
9
9
1
.
[6
]
As
wa
th
y
,
S
.
U.,
et
a
l.
,
“
A
su
rv
e
y
o
n
d
e
tec
ti
o
n
o
f
b
ra
in
t
u
m
o
r
fro
m
M
RI
b
ra
in
ima
g
e
s
,”
2
0
1
4
in
ter
n
a
ti
o
n
a
l
c
o
n
fer
e
n
c
e
o
n
c
o
n
tro
l,
I
n
stru
me
n
t
a
ti
o
n
,
C
o
m
mu
n
ica
t
io
n
a
n
d
C
o
mp
u
ta
ti
o
n
a
l
T
e
c
h
n
o
l
o
g
ies
(ICCICCT
)
,
Ju
l.
2
0
1
4
.
[7
]
A.
R.
Ab
d
u
lraq
e
b
,
e
t
a
l
.
,
“
An
a
u
to
m
a
ted
m
e
th
o
d
fo
r
se
g
m
e
n
t
i
n
g
b
ra
in
t
u
m
o
rs
o
n
M
RI
ima
g
e
s
,”
B
io
me
d
ica
l
E
n
g
i
n
e
e
rin
g
,
v
o
l.
51
,
n
o
.
2
,
p
p
.
97
-
1
0
1
,
Ju
l.
2
0
1
7
.
[8
]
R.
G
u
ru
sa
m
y
a
n
d
V.
S
u
b
ra
m
a
n
iam
,
“
A
m
a
c
h
in
e
lea
rn
in
g
a
p
p
r
o
a
c
h
fo
r
M
RI
b
ra
i
n
tu
m
o
r
c
las
sifica
ti
o
n
,
”
Co
m
p
u
t
.
,
M
a
ter
.
C
o
n
ti
n
u
a
,
v
o
l.
5
3
,
n
o
.
2
,
p
p
.
9
1
-
1
0
8
,
2
0
1
7
.
[9
]
Ya
z
d
a
n
i,
e
t
a
l.
,
“
Im
a
g
e
se
g
m
e
n
tat
io
n
m
e
th
o
d
s
a
n
d
a
p
p
li
c
a
ti
o
n
s
i
n
M
RI
b
ra
in
ima
g
e
s
,
”
IET
E
T
e
c
h
n
i
c
a
l
Rev
iew
,
v
o
l
.
32
,
n
o
.
6
,
4
1
3
-
4
2
7
.
2
0
1
5
.
[1
0
]
P
h
il
Kim
,
“
M
a
tl
a
b
d
e
e
p
lea
rn
in
g
:
with
m
a
c
h
in
e
lea
rn
in
g
,”
Ne
u
ra
l
N
e
two
rk
s
a
n
d
Arti
fi
c
ia
l
In
telli
g
e
n
c
e
,
v
o
l.
28
,
2
0
1
7
.
[1
1
]
Y.
LeCu
n
,
“
De
e
p
lea
rn
in
g
,”
N
a
t
u
re
,
v
o
l.
5
2
1
,
n
o
.
7
5
5
3
,
p
p
.
4
3
6
-
444
,
2
0
1
5
.
[1
2
]
I.
G
o
o
d
fe
ll
o
w,
e
t
a
l
., “
A:
De
e
p
L
e
a
rn
in
g
,”
M
IT
Pr
e
ss
,
2
0
1
6
.
[1
3
]
A.
Kriz
h
e
v
sk
y
,
I.
S
u
tsk
e
v
e
r,
a
n
d
G
.
E.
Hin
to
n
,
“
Im
a
g
e
Ne
t
c
las
sic
a
ti
o
n
wit
h
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s,
”
Pro
c
.
Ad
v
.
Ne
u
ra
l
In
f.
Pro
c
e
ss
.
S
y
st.
(NIPS
)
,
Ja
n
.
2
0
1
2
.
[1
4
]
O.
Ru
ss
a
k
o
v
s
k
y
,
e
t
a
l
.
,
“
Im
a
g
e
Ne
t
larg
e
sc
a
le
v
isu
a
l
re
c
o
g
n
it
i
o
n
c
h
a
ll
e
n
g
e
,”
In
t.
J
.
C
o
mp
u
t.
Vi
s.
(IJ
C
V)
,
v
o
l.
1
1
5
,
n
o
.
3
,
p
p
.
2
1
1
–
2
5
2
,
2
0
1
5
.
[1
5
]
F.
S
u
lt
a
n
a
,
A.
S
u
fia
n
,
P.
D
u
tt
a
,
“
Ad
v
a
n
c
e
m
e
n
ts
in
ima
g
e
c
las
sifica
ti
o
n
u
sin
g
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
,”
2
0
1
8
Fo
u
rt
h
I
n
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Res
e
a
rc
h
in
C
o
mp
u
ta
ti
o
n
a
l
I
n
telli
g
e
n
c
e
a
n
d
C
o
mm
u
n
ica
ti
o
n
Ne
two
rk
s
(ICRCICN
),
2
0
1
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.