I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l
E
ng
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
4
3
,
No
.
1
,
Ju
ly
2
0
2
6
,
p
p
.
281
~
29
8
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
4
3
.i
1
.
pp
281
-
29
8
281
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
M
o
bileNetV2 wi
t
h t
ra
nsfe
r learnin
g
for bra
in
tumo
r
cla
ss
ificatio
n
Aziz
Sra
i
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
s a
n
d
T
e
c
h
n
i
q
u
e
s Al
H
o
c
e
i
ma
,
A
b
d
e
l
m
a
l
e
k
Essa
a
d
i
U
n
i
v
e
r
si
t
y
,
Té
t
o
u
a
n
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
6
,
2
0
2
6
R
ev
is
ed
J
u
n
2
,
2
0
2
6
Acc
ep
ted
J
u
n
2
7
,
2
0
2
6
Th
is
d
o
c
u
m
e
n
t
p
re
se
n
ts
a
d
e
e
p
lea
rn
in
g
-
b
a
se
d
a
p
p
ro
a
c
h
f
o
r
t
h
e
a
u
to
m
a
ti
c
c
las
sifica
ti
o
n
o
f
b
ra
in
t
u
m
o
rs
f
ro
m
m
a
g
n
e
ti
c
re
so
n
a
n
c
e
ima
g
i
n
g
(
M
RI
)
ima
g
e
s,
u
sin
g
th
e
l
ig
h
twe
ig
h
t
M
o
b
il
e
Ne
tV2
m
o
d
e
l
c
o
m
b
i
n
e
d
wi
t
h
tran
sfe
r
lea
rn
in
g
a
n
d
fi
n
e
-
tu
n
in
g
tec
h
n
iq
u
e
s.
T
h
e
stu
d
y
a
ims
to
a
d
d
re
ss
th
e
c
o
n
stra
in
ts
o
f
re
so
u
rc
e
-
li
m
it
e
d
m
e
d
ica
l
e
n
v
iro
n
m
e
n
ts,
w
h
e
re
m
o
d
e
l
sp
e
e
d
a
n
d
li
g
h
twe
ig
h
t
d
e
sig
n
a
re
a
s
im
p
o
rtan
t
a
s
a
c
c
u
ra
c
y
.
T
h
e
d
a
tas
e
t
u
se
d
is
a
fu
sio
n
o
f
th
re
e
p
u
b
l
ic
d
a
tab
a
se
s
(F
ig
sh
a
re
,
S
ARTAJ
,
a
n
d
BR3
5
H),
c
o
m
p
risin
g
4
,
4
8
0
ima
g
e
s
fo
r
trai
n
in
g
a
n
d
1
,
6
0
0
fo
r
tes
ti
n
g
,
d
iv
i
d
e
d
in
to
f
o
u
r
c
las
se
s:
g
li
o
m
a
,
m
e
n
in
g
io
m
a
,
p
it
u
i
tary
t
u
m
o
r,
a
n
d
h
e
a
lt
h
y
b
ra
in
.
T
h
e
m
e
th
o
d
o
lo
g
y
in
c
lu
d
e
s
se
v
e
ra
l
ste
p
s:
re
siz
in
g
th
e
ima
g
e
s
t
o
2
2
4
×
2
2
4
p
ix
e
ls,
n
o
rm
a
li
z
in
g
p
ix
e
l
v
a
lu
e
s
b
e
twe
e
n
0
a
n
d
1
,
a
u
g
m
e
n
t
in
g
t
h
e
d
a
t
a
th
ro
u
g
h
ra
n
d
o
m
ro
tati
o
n
s,
sh
if
ts,
a
n
d
z
o
o
m
s
to
a
v
o
i
d
o
v
e
rfi
tt
in
g
,
a
n
d
t
h
e
n
e
x
trac
ti
n
g
fe
a
tu
re
s
u
sin
g
M
o
b
il
e
Ne
tV2
p
re
-
train
e
d
o
n
Im
a
g
e
Ne
t.
T
h
e
stra
teg
y
a
d
o
p
ted
c
o
m
p
rise
s
two
p
h
a
se
s:
first,
tran
sfe
r
lea
rn
in
g
wh
e
re
o
n
l
y
th
e
lay
e
r
s
a
d
d
e
d
a
t
th
e
to
p
o
f
t
h
e
m
o
d
e
l
a
re
trai
n
e
d
f
o
r
1
0
e
p
o
c
h
s,
t
h
e
n
f
in
e
-
t
u
n
i
n
g
c
o
n
sistin
g
o
f
u
n
fre
e
z
in
g
t
h
e
las
t
2
0
la
y
e
rs
o
f
t
h
e
b
a
se
m
o
d
e
l
a
n
d
re
train
i
n
g
t
h
e
m
with
a
re
d
u
c
e
d
lea
rn
i
n
g
ra
te
fo
r
5
e
p
o
c
h
s.
Th
e
re
su
lt
s
o
b
tai
n
e
d
s
h
o
w
a
n
o
v
e
ra
ll
a
c
c
u
ra
c
y
o
f
8
6
.
5
6
%
a
fter
fi
n
e
-
t
u
n
i
n
g
,
with
a
m
a
c
ro
-
m
e
a
n
a
re
a
u
n
d
e
r
th
e
ROC
c
u
rv
e
(AU
C)
o
f
0
.
9
6
4
5
,
i
n
d
ica
ti
n
g
e
x
c
e
ll
e
n
t
d
isc
rimin
a
t
o
r
y
p
o
we
r.
Th
e
c
o
n
fu
si
o
n
m
a
tri
x
re
v
e
a
ls
t
h
a
t
t
h
e
"
n
o
tu
m
o
r"
c
las
s
a
c
h
ie
v
e
s
p
e
rfe
c
t
p
e
rfo
rm
a
n
c
e
(4
0
0
o
u
t
o
f
4
0
0
),
wh
il
e
th
e
"
m
e
n
in
g
io
m
a
"
c
las
s
re
m
a
in
s
th
e
m
o
st
d
iffi
c
u
l
t
t
o
c
las
sify
,
o
ften
c
o
n
f
u
se
d
with
g
li
o
m
a
s
a
n
d
p
it
u
it
a
ry
tu
m
o
rs.
Co
m
p
a
re
d
to
m
o
re
re
so
u
rc
e
-
in
te
n
siv
e
m
o
d
e
ls
li
k
e
VG
G
1
6
,
Re
sN
e
t5
0
,
o
r
Eff
icie
n
tNe
t,
M
o
b
il
e
Ne
tV2
o
ffe
rs
a
n
o
p
ti
m
a
l
b
a
lan
c
e
b
e
twe
e
n
p
e
rfo
rm
a
n
c
e
a
n
d
l
ig
h
twe
ig
h
t
d
e
sig
n
,
wit
h
a
sig
n
ifi
c
a
n
t
ly
lo
we
r
n
u
m
b
e
r
o
f
p
a
ra
m
e
ters
,
m
a
k
in
g
it
p
a
rti
c
u
larly
we
ll
-
su
it
e
d
to
re
so
u
rc
e
-
c
o
n
stra
in
e
d
e
n
v
iro
n
m
e
n
ts.
Th
e
a
u
th
o
rs
c
o
n
c
lu
d
e
th
a
t
th
is
a
p
p
r
o
a
c
h
p
ro
v
id
e
s
re
li
a
b
le
d
ia
g
n
o
stic
su
p
p
o
rt
f
o
r
ra
d
io
l
o
g
ists,
a
c
c
e
lera
ti
n
g
t
u
m
o
r
d
e
tec
ti
o
n
wi
th
o
u
t
re
p
lac
in
g
m
e
d
ica
l
e
x
p
e
rti
se
.
F
u
t
u
re
d
irec
ti
o
n
s
in
c
l
u
d
e
c
li
n
ica
l
v
a
li
d
a
ti
o
n
o
n
m
u
lt
i
-
c
e
n
ter
d
a
ta,
in
teg
ra
ti
o
n
o
f
a
n
a
u
t
o
m
a
ted
se
g
m
e
n
tatio
n
ste
p
,
a
n
d
e
x
p
l
o
ra
ti
o
n
o
f
n
e
we
r
a
rc
h
it
e
c
tu
re
s su
c
h
a
s a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
s.
K
ey
w
o
r
d
s
:
Ar
tific
ial
in
tellig
en
ce
Dee
p
lear
n
in
g
Fin
e
tu
n
in
g
Neu
r
o
-
o
n
co
l
o
g
y
T
r
an
s
f
er
lear
n
i
n
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Aziz
Sra
i
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
,
Facu
lty
o
f
Scien
ce
s
an
d
T
ec
h
n
iq
u
es Al
Ho
ce
im
a
Ab
d
elm
alek
E
s
s
aa
d
i U
n
iv
er
s
ity
T
éto
u
an
,
M
o
r
o
cc
o
E
m
ail:
a.
s
r
ai@
u
ae
.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
b
r
ai
n
is
m
ad
e
u
p
o
f
s
ev
er
al
b
illi
o
n
ce
lls
:
n
eu
r
o
n
s
,
w
h
ich
cr
ea
te,
d
ir
ec
t,
a
n
d
c
o
n
tr
o
l
in
f
o
r
m
atio
n
,
an
d
g
lial
ce
lls
,
wh
ich
s
u
p
p
o
r
t
n
e
u
r
o
n
s
an
d
en
a
b
le
th
e
ir
d
ev
elo
p
m
en
t
an
d
n
u
tr
itio
n
,
an
d
c
o
n
tr
ib
u
te
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
281
-
29
8
282
s
ig
n
if
ican
tly
to
b
r
ain
tis
s
u
e
h
o
m
eo
s
tasi
s
.
Fig
u
r
e
1
p
r
esen
ts
f
o
u
r
m
ag
n
etic
r
eso
n
a
n
ce
im
ag
in
g
(
MRI
)
b
r
ain
im
ag
e
class
es:
g
lio
m
a,
m
en
in
g
io
m
a,
p
itu
itar
y
t
u
m
o
r
,
an
d
n
o
tu
m
o
r
.
T
h
er
e
a
r
e
s
ev
er
al
ty
p
es
o
f
g
lial
ce
lls
(
astro
cy
tes,
o
lig
o
d
e
n
d
r
o
cy
tes,
m
icr
o
g
lia,
e
p
en
d
y
m
al
ce
lls
)
,
ea
ch
with
s
p
ec
if
ic
r
o
les.
T
h
r
o
u
g
h
o
u
t
life
,
b
r
ain
ce
lls
ca
n
ac
q
u
ir
e
m
u
tatio
n
s
in
th
eir
d
eo
x
y
r
ib
o
n
u
cleic
ac
id
(
DNA
)
d
u
e
to
ag
in
g
,
en
v
ir
o
n
m
en
tal
f
ac
to
r
s
,
an
d
g
en
etics,
th
u
s
tr
an
s
f
o
r
m
in
g
in
to
ca
n
ce
r
o
u
s
ce
lls
th
at
escap
e
ce
llu
lar
co
n
t
r
o
l
m
ec
h
an
is
m
s
.
B
r
ain
tu
m
o
r
s
a
r
e
s
u
b
d
iv
id
ed
i
n
to
two
d
is
tin
ct
ty
p
es:
−
Prim
ar
y
tu
m
o
r
s
:
th
ese
o
r
ig
i
n
a
te
in
th
e
b
r
ain
m
ain
l
y
f
r
o
m
g
lial
ce
lls
th
at
m
ak
e
u
p
th
e
n
e
r
v
o
u
s
s
y
s
tem
[
1
]
.
T
h
ey
ca
n
b
e
b
e
n
ig
n
(
two
-
th
ir
d
s
o
f
p
r
im
ar
y
tu
m
o
r
ca
s
es)
o
r
m
alig
n
an
t
[
2
]
.
−
Seco
n
d
ar
y
tu
m
o
r
s
,
also
ca
lled
b
r
ain
m
etastas
es,
o
r
ig
in
ate
f
r
o
m
a
p
r
e
-
ex
is
tin
g
ca
n
ce
r
l
o
ca
ted
in
a
n
o
th
er
o
r
g
an
an
d
m
ig
r
ate
v
ia
th
e
b
l
o
o
d
s
tr
ea
m
to
b
r
ain
tis
s
u
e
[
3
]
.
T
h
e
ca
n
ce
r
s
m
o
s
t
f
r
eq
u
e
n
tly
a
s
s
o
ciate
d
with
th
e
d
ev
elo
p
m
e
n
t
o
f
b
r
ain
m
etasta
s
es
ar
e
lu
n
g
ca
n
ce
r
,
b
r
ea
s
t
ca
n
ce
r
,
an
d
m
elan
o
m
a
[
4
]
.
Seco
n
d
ar
y
tu
m
o
r
s
,
o
r
b
r
ain
m
etastas
es,
ar
e
th
e
m
o
s
t
co
m
m
o
n
ty
p
e
o
f
b
r
ain
tu
m
o
r
,
o
cc
u
r
r
in
g
3
to
1
0
tim
es
m
o
r
e
f
r
eq
u
en
tly
th
a
n
p
r
im
ar
y
b
r
ain
tu
m
o
r
s
[
5
]
.
−
Glio
m
as
:
Glio
m
a
is
th
e
g
en
er
al
n
am
e
g
iv
e
n
to
th
e
tu
m
o
r
g
r
o
u
p
e
n
co
m
p
ass
in
g
p
r
im
ar
y
b
r
ain
t
u
m
o
r
s
o
r
ig
in
atin
g
f
r
o
m
g
lial
ce
lls
[
6
]
.
T
h
e
r
e
ar
e
d
if
f
er
e
n
t
ty
p
es
o
f
g
lio
m
as
ass
o
ciate
d
with
th
e
th
r
ee
ty
p
es
o
f
g
lial
ce
lls
:
astro
cy
to
m
as
(
f
o
r
astro
cy
tes),
o
lig
o
d
e
n
d
r
o
g
lio
m
as
(
f
o
r
o
lig
o
d
en
d
r
o
cy
tes),
an
d
e
p
en
d
y
m
o
m
as
(
f
o
r
ep
en
d
y
m
al
ce
lls
)
[
7
]
.
−
M
en
in
g
io
m
a:
A
m
e
n
in
g
io
m
a
is
a
tu
m
o
r
th
at
d
e
v
elo
p
s
f
r
o
m
th
e
m
en
in
g
es,
th
e
m
em
b
r
a
n
es
th
at
co
v
er
t
h
e
b
r
ain
an
d
s
p
in
al
co
r
d
[
8
]
.
I
t
a
cc
o
u
n
ts
f
o
r
ap
p
r
o
x
im
ately
3
5
%
o
f
p
r
im
ar
y
b
r
ain
tu
m
o
r
s
,
m
ak
in
g
it
th
e
m
o
s
t
co
m
m
o
n
[
9
]
.
Me
n
in
g
io
m
as
a
r
e
b
e
n
ig
n
in
8
0
%
o
f
ca
s
es
(
g
r
ad
e
I
)
,
b
u
t
aty
p
ical
(
g
r
ad
e
I
I
)
a
n
d
m
alig
n
a
n
t
an
a
p
last
ic
(
g
r
ad
e
I
I
I
)
f
o
r
m
s
also
ex
is
t
[
1
0
]
.
T
h
ey
o
cc
u
r
m
o
r
e
f
r
eq
u
en
tly
in
wo
m
e
n
,
d
u
e
to
h
o
r
m
o
n
e
r
ec
ep
to
r
s
[
1
1
]
.
M
o
s
t
m
en
in
g
i
o
m
as
r
em
ain
asy
m
p
t
o
m
atic
an
d
ar
e
d
is
co
v
e
r
ed
in
ci
d
en
tally
d
u
r
in
g
a
n
im
ag
i
n
g
ex
am
in
atio
n
[
1
2
]
.
W
h
en
th
e
y
b
ec
o
m
e
s
y
m
p
to
m
atic,
t
h
ey
c
au
s
e
s
eizu
r
es,
h
ea
d
ac
h
es,
o
r
f
o
ca
l
n
eu
r
o
lo
g
ical
d
ef
icits
[
1
3
]
.
T
h
r
o
u
g
h
m
ass
ef
f
ec
t,
th
ey
ca
n
c
o
m
p
r
ess
a
b
r
ai
n
r
eg
i
o
n
,
a
cr
an
ial
n
er
v
e
,
o
r
a
b
lo
o
d
v
ess
el
[
1
4
]
.
On
MRI,
a
ty
p
ical
m
en
in
g
io
m
a
is
a
well
-
d
ef
in
ed
,
ex
t
r
a
-
ax
ial
m
ass
,
o
f
ten
o
n
i
o
n
-
s
h
ap
e
d
[
1
5
]
.
−
Pit
u
itar
y
:
A
p
itu
itar
y
tu
m
o
r
,
o
r
p
itu
itar
y
ad
en
o
m
a,
is
a
b
en
ig
n
tu
m
o
r
th
at
d
ev
el
o
p
s
in
th
e
p
itu
itar
y
g
lan
d
,
th
e
"c
o
n
d
u
cto
r
"
o
f
h
o
r
m
o
n
es
[
1
6
]
.
T
h
e
p
itu
itar
y
g
lan
d
is
lo
ca
ted
in
th
e
s
ella
tu
r
cica
,
a
s
m
all
b
o
n
y
ca
v
ity
at
th
e
b
ase
o
f
th
e
s
k
u
ll
[
1
7
]
.
Pit
u
itar
y
ad
e
n
o
m
as
a
r
e
v
er
y
co
m
m
o
n
,
f
o
u
n
d
in
ap
p
r
o
x
im
ately
1
0
%
o
f
th
e
p
o
p
u
latio
n
at
au
to
p
s
y
[
1
8
]
.
Mo
s
t
ar
e
m
icr
o
s
co
p
ic
an
d
co
m
p
letely
a
s
y
m
p
to
m
atic.
T
h
e
y
ca
n
b
ec
o
m
e
s
y
m
p
to
m
atic
th
r
o
u
g
h
two
m
ec
h
an
is
m
s
:
h
o
r
m
o
n
al
ex
ce
s
s
o
r
lo
ca
l
co
m
p
r
ess
io
n
.
So
-
ca
lled
f
u
n
ctio
n
al
ad
en
o
m
as
s
ec
r
et
e
h
o
r
m
o
n
es
in
ex
ce
s
s
iv
e
am
o
u
n
ts
.
Pro
lactin
is
th
e
h
o
r
m
o
n
e
m
o
s
t
o
f
ten
f
o
u
n
d
in
ex
ce
s
s
,
ca
u
s
in
g
a
s
y
n
d
r
o
m
e
in
wo
m
en
(
g
alac
to
r
r
h
ea
,
m
en
s
tr
u
al
ir
r
eg
u
lar
ities
)
an
d
in
m
e
n
(
d
ec
r
ea
s
ed
lib
i
d
o
,
im
p
o
ten
c
e)
.
E
x
ce
s
s
g
r
o
wth
h
o
r
m
o
n
e
in
a
d
u
lts
ca
u
s
es
ac
r
o
m
eg
aly
,
with
e
n
lar
g
em
e
n
t
o
f
th
e
h
an
d
s
,
f
ee
t,
an
d
f
ac
ial
f
ea
tu
r
es.
E
x
ce
s
s
co
r
tis
o
l
(
C
u
s
h
in
g
'
s
d
is
ea
s
e)
lead
s
to
weig
h
t
g
ai
n
,
s
tr
etch
m
ar
k
s
,
h
y
p
er
te
n
s
io
n
,
a
n
d
s
k
in
f
r
a
g
ilit
y
.
No
n
-
f
u
n
ctio
n
in
g
ad
en
o
m
as
d
o
n
o
t
s
ec
r
ete
h
o
r
m
o
n
es
an
d
a
r
e
r
ev
ea
led
b
y
th
eir
m
ass
ef
f
ec
t.
C
o
m
p
r
ess
io
n
o
f
th
e
o
p
tic
ch
iasm,
wh
ich
cr
o
s
s
es
ju
s
t
ab
o
v
e
th
e
p
itu
itar
y
g
l
an
d
,
ca
u
s
es
b
item
p
o
r
al
h
e
m
ian
o
p
s
ia
(
lo
s
s
o
f
p
er
ip
h
er
al
v
is
io
n
in
b
o
th
ey
es)
[
1
9
]
.
I
n
s
u
m
m
a
r
y
,
t
h
ese
th
r
ee
tu
m
o
r
s
d
if
f
er
in
th
ei
r
o
r
ig
in
,
b
eh
a
v
io
r
,
an
d
p
r
o
g
n
o
s
is
,
b
u
t
s
h
ar
e
t
h
e
n
ee
d
f
o
r
p
r
ec
is
e
im
ag
in
g
an
d
m
u
ltid
is
cip
lin
ar
y
m
an
ag
em
e
n
t.
On
th
e
o
th
er
h
an
d
,
th
e
in
tr
o
d
u
ctio
n
o
f
MRI
in
th
e
f
ield
o
f
o
n
co
l
o
g
y
was
in
itiated
b
y
p
h
y
s
ician
R
ay
m
o
n
d
Dam
alia
n
in
1
9
7
1
wh
en
h
e
d
is
co
v
e
r
ed
t
h
at
tu
m
o
r
tis
s
u
e
h
as
a
d
if
f
er
e
n
t
r
ela
x
atio
n
tim
e
th
an
h
ea
lth
y
tis
s
u
e.
Sin
ce
th
en
,
s
ev
er
al
im
ag
in
g
tech
n
iq
u
es
h
a
v
e
em
er
g
ed
,
n
o
ta
b
ly
d
if
f
u
s
io
n
-
weig
h
te
d
im
ag
in
g
(
DW
I
)
,
cr
ea
ted
in
1
9
8
6
b
y
p
h
y
s
ician
an
d
p
h
y
s
icis
t
Den
is
L
e
B
ih
an
,
wh
ich
allo
ws
f
o
r
th
e
ex
p
lo
r
atio
n
o
f
th
e
b
r
ain
'
s
an
ato
m
ical
co
n
n
ec
tiv
ity
b
y
m
ea
s
u
r
in
g
a
s
ig
n
al
s
en
s
itiv
e
to
th
e
m
o
v
e
m
en
t
o
f
wate
r
m
o
lec
u
les
[
2
0
]
.
I
n
1
9
9
0
,
th
e
wo
r
k
o
f
n
eu
r
o
s
cien
tis
t
a
n
d
b
io
p
h
y
s
icis
t
Seiji
Og
awa
b
r
o
u
g
h
t
ab
o
u
t
co
n
s
id
er
ab
le
p
r
o
g
r
ess
with
th
e
d
ev
elo
p
m
e
n
t
o
f
f
u
n
ctio
n
al
MRI,
s
h
o
win
g
th
e
b
r
ain
"in
ac
tio
n
.
"
T
h
is
m
ag
n
etic
s
u
s
ce
p
tib
ilit
y
im
ag
in
g
tech
n
iq
u
e
ex
p
lo
its
th
e
o
x
y
g
e
n
atio
n
p
r
o
p
er
ties
o
f
b
lo
o
d
,
also
k
n
o
wn
as
th
e
B
OL
D
ef
f
e
ct
–
B
lo
o
d
-
Ox
y
g
en
L
ev
el
-
Dep
en
d
e
n
t.
Fin
ally
,
th
e
2
0
0
0
s
m
ar
k
e
d
a
n
ew
ad
v
a
n
c
e
in
MRI
with
th
e
u
s
e
o
f
co
n
tr
ast
ag
en
ts
h
av
in
g
m
ag
n
etic
s
u
s
ce
p
tib
ilit
y
f
o
r
th
e
s
tu
d
y
o
f
v
ascu
lar
izatio
n
,
th
e
b
est
k
n
o
wn
an
d
u
s
ed
b
ei
n
g
g
ad
o
lin
iu
m
c
h
elate
,
allo
win
g
in
p
ar
ticu
lar
th
e
a
s
s
es
s
m
en
t
o
f
v
ascu
lar
p
er
m
ea
b
ilit
y
[
2
1
]
.
I
n
s
u
m
m
ar
y
,
MRI
an
d
ar
tific
ial
in
tellig
en
ce
p
lay
an
im
p
o
r
tan
t
r
o
le
to
d
ay
in
th
e
f
ield
o
f
o
n
co
l
o
g
y
.
I
n
th
is
ar
ticle,
we
u
s
ed
d
ee
p
lear
n
in
g
as
th
e
p
r
im
ar
y
alg
o
r
it
h
m
f
o
r
b
r
ain
tu
m
o
r
d
etec
tio
n
[
2
2
]
.
B
r
ain
tu
m
o
r
d
etec
tio
n
ty
p
ically
o
cc
u
r
s
in
two
s
tag
es:
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
an
d
class
if
icatio
n
[
2
3
]
.
Fo
r
class
if
icatio
n
,
we
u
s
ed
a
p
r
e
-
tr
ain
ed
Mo
b
ileNetV2
m
o
d
el.
W
e
th
en
em
p
lo
y
ed
f
i
n
e
-
tu
n
i
n
g
to
m
o
n
ito
r
th
e
p
er
f
o
r
m
an
ce
o
f
o
u
r
m
o
d
el.
T
wo
im
p
o
r
tan
t
co
n
ce
p
ts
in
d
e
ep
lear
n
in
g
ar
e
u
s
ed
in
th
is
d
o
cu
m
e
n
t:
tr
an
s
f
er
lear
n
in
g
a
n
d
f
in
e
-
t
u
n
in
g
.
Fig
u
r
e
2
,
f
o
r
ex
am
p
le,
p
r
es
en
ts
an
im
ag
e
p
r
e
d
ictio
n
u
s
in
g
t
r
a
n
s
f
er
lear
n
i
n
g
,
an
d
Fig
u
r
e
3
s
h
o
ws
an
ex
a
m
p
le
o
f
a
p
r
ed
ictio
n
u
s
in
g
f
i
n
e
-
tu
n
i
n
g
.
T
h
e
r
em
ain
d
e
r
o
f
t
h
e
ar
ti
cle
is
o
r
g
a
n
ized
as
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Mo
b
ileN
etV
2
w
ith
tr
a
n
s
fe
r
lea
r
n
in
g
fo
r
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
(
A
z
iz
S
r
a
i
)
283
f
o
llo
ws:
Sectio
n
2
d
escr
ib
e
s
th
e
m
eth
o
d
o
lo
g
y
f
o
llo
we
d
in
th
is
d
o
cu
m
e
n
t
,
Sectio
n
3
d
escr
ib
es
th
e
ex
p
er
im
en
tal
r
esu
lts
o
b
tain
e
d
,
s
ec
tio
n
4
co
n
clu
d
es th
e
ar
ticle
an
d
p
r
o
p
o
s
es r
esear
ch
p
er
s
p
e
ctiv
es.
Fig
u
r
e
1
.
Fo
u
r
(
4
)
ty
p
es
o
f
M
R
I
im
ag
es o
f
b
r
ain
tu
m
o
r
Fig
u
r
e
2
.
E
x
am
p
le
o
f
a
p
r
ed
ict
io
n
with
o
u
t f
i
n
e
-
tu
n
i
n
g
,
u
s
in
g
o
n
ly
tr
an
s
f
e
r
lear
n
in
g
Fig
u
r
e
3
. E
x
am
p
le
o
f
a
p
r
ed
ict
io
n
u
s
in
g
Fin
e
-
tu
n
in
g
I
n
o
th
er
wo
r
d
s
,
an
d
b
ef
o
r
e
p
r
e
s
en
tin
g
th
e
r
esu
lts
o
f
th
is
s
cie
n
tific
ar
ticle,
we
m
u
s
t e
x
p
lain
o
u
r
ch
o
ic
e
o
f
th
e
Mo
b
ileNetV2
m
o
d
el.
I
n
d
ee
d
,
th
e
ch
o
ice
o
f
Mo
b
ileNetV2
as
a
h
ig
h
-
p
er
f
o
r
m
an
ce
a
p
p
r
o
ac
h
is
b
ased
o
n
its
ex
ce
llen
t
b
alan
ce
b
etwe
en
ac
cu
r
ac
y
,
s
p
ee
d
,
a
n
d
lig
h
twe
ig
h
t
n
atu
r
e,
m
a
k
in
g
it
p
ar
ticu
lar
ly
well
-
s
u
ited
to
co
n
s
tr
ain
ed
e
n
v
ir
o
n
m
en
ts
s
u
c
h
as
m
o
b
ile
o
r
em
b
ed
d
ed
s
y
s
t
em
s
.
Un
lik
e
m
o
r
e
r
eso
u
r
ce
-
in
t
en
s
iv
e
m
o
d
els
s
u
ch
as
VGG1
6
o
r
R
esNet5
0
,
o
r
E
f
f
icien
tNet,
E
f
f
icien
tNetV2
,
C
o
n
v
NeX
t,
v
is
io
n
tr
an
s
f
o
r
m
er
(
ViT
)
,
Swi
n
T
r
an
s
f
o
r
m
e
r
,
R
eg
Net,
Mo
b
ileViT
,
an
d
Mo
b
ileNetV3
,
Mo
b
ileNetV2
u
s
es
s
ep
ar
ab
le
d
ep
th
co
n
v
o
l
u
tio
n
s
an
d
in
v
er
s
e
lin
ea
r
b
o
ttlen
ec
k
b
lo
c
k
s
,
d
r
asti
ca
lly
r
ed
u
ci
n
g
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
an
d
th
e
co
m
p
u
tatio
n
al
l
o
ad
with
o
u
t
s
ig
n
if
ican
tly
s
ac
r
if
icin
g
ac
cu
r
ac
y
.
I
t
i
s
h
ig
h
ly
r
esis
tan
t
to
o
v
er
f
itti
n
g
o
n
s
m
all
d
atasets
th
an
k
s
to
its
ef
f
icien
t
s
tr
u
ctu
r
e
an
d
allo
ws
f
o
r
r
ap
id
f
in
e
-
tu
n
i
n
g
.
C
o
m
p
ar
ed
to
ar
ch
itectu
r
es
lik
e
I
n
ce
p
tio
n
V3
o
r
Den
s
eNe
t1
2
1
,
it o
f
f
er
s
s
ig
n
if
i
ca
n
tly
f
aster
i
n
f
er
en
ce
s
an
d
a
s
m
aller
m
em
o
r
y
f
o
o
tp
r
i
n
t,
wh
i
le
m
ain
tain
in
g
h
ig
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
281
-
29
8
284
p
er
f
o
r
m
an
ce
f
o
r
task
s
s
u
ch
as
class
if
icatio
n
an
d
d
etec
tio
n
.
T
h
is
en
er
g
y
an
d
co
m
p
u
tatio
n
a
l
ef
f
icien
cy
m
ak
e
it
s
u
p
er
io
r
in
r
ea
l
-
tim
e
en
v
ir
o
n
m
en
ts
o
r
o
n
lim
ited
h
ar
d
war
e.
T
h
e
co
n
tr
ib
u
tio
n
o
f
th
is
ar
ticle
ca
n
b
e
s
u
m
m
ar
ize
d
in
th
e
f
o
llo
win
g
p
o
i
n
ts
:
−
W
e
ap
p
ly
Mo
b
ileNetV2
with
t
r
an
s
f
er
lear
n
in
g
an
d
f
in
e
-
t
u
n
in
g
to
a
m
u
lti
-
class
b
r
ain
tu
m
o
r
MRI
d
ataset.
−
W
e
p
r
o
v
id
e
a
d
etailed
c
o
m
p
ar
ativ
e
an
aly
s
is
with
ex
is
tin
g
m
o
d
els
(
E
DC
NN,
I
n
ce
p
t
io
n
V3
,
VGG1
6
en
s
em
b
le,
B
r
ain
Net,
R
o
b
u
s
t CNN+U
-
Net,
g
en
er
ic
C
NN)
.
−
W
e
an
aly
ze
m
is
cla
s
s
if
icatio
n
ca
u
s
es
(
esp
ec
ially
m
en
in
g
io
m
a
v
s
.
p
itu
itar
y
an
d
g
lio
m
a)
an
d
d
is
cu
s
s
th
e
p
r
ac
tical
ad
v
an
ta
g
es o
f
Mo
b
ileNetV2
f
o
r
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
.
−
W
e
r
ep
o
r
t
n
o
t
o
n
ly
ac
cu
r
ac
y
b
u
t
also
s
en
s
it
iv
ity
,
s
p
ec
if
ic
ity
,
p
r
ec
is
io
n
,
F1
-
s
co
r
e,
co
n
f
u
s
io
n
m
atr
ices,
R
OC
cu
r
v
es,
an
d
ef
f
icien
cy
c
o
n
s
id
er
atio
n
s
.
2.
M
E
T
H
O
D
T
h
e
m
eth
o
d
o
lo
g
y
s
ec
tio
n
ex
p
lain
s
th
e
d
ata
(
d
ataset,
p
r
ep
r
o
ce
s
s
in
g
)
,
m
o
d
el
f
in
d
in
g
s
,
an
d
ev
alu
atio
n
m
etr
ics u
s
ed
in
th
is
s
tu
d
y
.
E
v
e
r
y
s
tep
is
cr
u
cial
to
ac
h
ie
v
in
g
h
ig
h
ac
cu
r
ac
y
in
b
r
ain
t
u
m
o
r
c
lass
if
icatio
n
.
2
.1
.
Da
t
a
s
et
d
escript
io
n
Fo
r
th
is
r
esear
ch
,
we
u
s
ed
a
f
r
ee
ly
ac
ce
s
s
ib
le
M
R
I
im
ag
e
d
ataset
ca
lled
Mso
u
d
[
1
]
,
d
o
w
n
lo
ad
ab
le
f
r
o
m
th
e
Kag
g
le
p
latf
o
r
m
.
I
t
co
m
p
r
is
es
im
ag
es
d
iv
id
e
d
in
t
o
f
o
u
r
ca
teg
o
r
ies:
g
lio
m
a,
m
e
n
in
g
io
m
a,
p
itu
itar
y
tu
m
o
r
,
an
d
h
ea
lth
y
b
r
ain
illu
s
tr
ated
in
Fig
u
r
e
1
.
T
h
e
d
ata
ar
e
o
r
g
an
ize
d
in
to
th
r
ee
m
ai
n
s
u
b
s
ets:
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
in
g
.
T
h
e
in
itial
s
p
litt
in
g
o
f
th
e
d
ataset
was
p
er
f
o
r
m
e
d
b
y
th
e
au
t
h
o
r
s
,
with
4
,
4
8
0
im
ag
es
f
o
r
tr
ain
in
g
,
1
,
1
2
0
im
ag
es
f
o
r
v
alid
atio
n
,
an
d
1
,
6
0
0
im
a
g
es
f
o
r
test
in
g
,
ea
ch
class
b
ein
g
p
r
o
p
o
r
tio
n
ally
r
ep
r
esen
ted
.
T
h
is
p
r
ed
ef
in
e
d
p
ar
titi
o
n
in
g
was
im
p
lem
en
ted
b
y
th
e
cr
ea
to
r
s
o
f
th
e
o
r
i
g
in
al
d
ataset
to
en
s
u
r
e
f
air
ev
alu
atio
n
an
d
av
o
id
an
y
r
is
k
o
f
cr
o
s
s
-
co
n
tam
in
atio
n
b
etwe
en
th
e
s
ets.
T
h
e
ex
ac
t
d
is
tr
ib
u
tio
n
b
etwe
en
th
e
tr
ain
in
g
,
v
alid
atio
n
,
an
d
t
est
d
ata
was
r
etr
o
s
p
ec
tiv
ely
a
n
aly
ze
d
to
en
s
u
r
e
ad
eq
u
ate
b
alan
ce
Fig
u
r
es
4
-
6
.
No
d
ata
leak
ag
e
was o
b
s
er
v
ed
,
as th
e
p
ar
titi
o
n
in
g
was f
ix
ed
b
y
th
e
o
r
ig
in
al
au
t
h
o
r
s
.
Fig
u
r
e
4
.
E
x
am
p
le
o
f
a
p
r
ed
ict
io
n
p
er
ce
n
tile d
is
tr
ib
u
tio
n
o
f
tr
ain
d
ata
Fig
u
r
e
5
.
E
x
am
p
le
o
f
a
p
r
ed
ict
io
n
p
er
ce
n
tile d
is
tr
ib
u
tio
n
o
f
v
alid
atio
n
d
ata
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Mo
b
ileN
etV
2
w
ith
tr
a
n
s
fe
r
lea
r
n
in
g
fo
r
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
(
A
z
iz
S
r
a
i
)
285
Fig
u
r
e
6
. E
x
am
p
le
o
f
a
p
r
ed
ict
io
n
p
er
ce
n
tile d
is
tr
ib
u
tio
n
o
f
t
est
d
ata
2
.
2
.
P
re
pro
ce
s
s
ing
W
e
h
av
e
p
r
ep
ar
ed
a
m
ed
ical
im
ag
e
p
r
o
ce
s
s
in
g
p
ip
elin
e
(
b
r
ain
MRI)
f
o
r
a
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
Mo
b
ileNetV2
)
,
u
s
in
g
au
g
m
en
tatio
n
tech
n
iq
u
es
to
p
r
ev
en
t
o
v
e
r
f
itti
n
g
an
d
im
p
r
o
v
e
g
en
er
aliza
tio
n
.
W
e
f
ir
s
t
d
ef
in
ed
s
ev
er
al
ess
en
tial
p
ar
am
eter
s
f
o
r
p
r
e
p
ar
in
g
th
e
tr
ain
in
g
d
ata
f
o
r
th
is
m
o
d
el
d
ed
icate
d
t
o
an
aly
zin
g
m
ed
ical
b
r
ain
im
ag
es.
T
h
e
v
ar
iab
le
I
MG
_
SIZ
E
i
s
s
et
to
2
2
4
b
y
2
2
4
p
ix
els,
a
s
tan
d
ar
d
r
eso
lu
tio
n
s
u
itab
le
f
o
r
m
o
s
t
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
e
s
s
u
ch
as
R
e
s
Net
o
r
VGG,
an
d
p
ar
ticu
lar
ly
well
-
s
u
ited
to
o
u
r
s
u
b
ject,
Mo
b
ileNetV2
.
T
h
e
b
atch
s
ize,
B
AT
C
H_
SIZ
E
,
is
s
et
to
3
2
,
m
ea
n
in
g
th
at
3
2
im
a
g
es
will
b
e
p
r
o
ce
s
s
ed
s
im
u
ltan
eo
u
s
ly
in
ea
ch
tr
ain
in
g
iter
atio
n
,
a
g
o
o
d
co
m
p
r
o
m
is
e
b
etwe
en
g
r
a
d
ien
t
s
tab
ilit
y
an
d
m
em
o
r
y
u
s
ag
e.
Fin
ally
,
th
e
p
ar
am
eter
NUM
_
C
L
ASSES
i
s
s
et
to
4
,
co
r
r
esp
o
n
d
in
g
t
o
th
e
f
o
u
r
d
iag
n
o
s
tic
ca
teg
o
r
ies
th
at
th
e
m
o
d
el
m
u
s
t
d
is
tin
g
u
is
h
:
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
.
Nex
t,
th
e
co
d
e
cr
ea
tes
an
au
g
m
en
ted
im
ag
e
g
en
e
r
ato
r
,
a
cr
u
cial
tech
n
i
q
u
e
f
o
r
a
r
tific
ially
en
r
ich
in
g
t
h
e
tr
ain
in
g
s
et
an
d
im
p
r
o
v
in
g
m
o
d
el
r
o
b
u
s
tn
ess
.
E
ac
h
lo
ad
ed
im
a
g
e
is
f
ir
s
t
n
o
r
m
alize
d
with
a
r
escale
f
ac
to
r
o
f
1
:2
5
5
,
co
n
v
er
tin
g
p
ix
el
v
alu
es
f
r
o
m
th
e
r
an
g
e
o
f
0
to
2
5
5
t
o
th
e
r
a
n
g
e
o
f
0
to
1
,
th
u
s
ac
ce
ler
atin
g
lear
n
in
g
co
n
v
er
g
en
ce
.
T
h
e
g
en
er
ato
r
th
en
a
p
p
lies
v
ar
io
u
s
r
an
d
o
m
t
r
an
s
f
o
r
m
atio
n
s
:
r
o
tatio
n
lim
ited
to
2
0
d
e
g
r
ee
s
in
eith
er
d
ir
ec
tio
n
,
h
o
r
izo
n
tal
an
d
v
er
tical
s
h
if
ts
o
f
u
p
to
1
0
p
er
ce
n
t
o
f
th
e
im
ag
e'
s
wid
th
o
r
h
eig
h
t,
zo
o
m
s
o
f
u
p
to
1
0
p
er
ce
n
t,
an
d
r
an
d
o
m
h
o
r
iz
o
n
tal
f
lip
s
.
All
th
ese
v
ar
iatio
n
s
s
im
u
late
t
h
e
d
if
f
er
en
ce
s
in
o
r
ien
tatio
n
an
d
p
o
s
itio
n
th
at
ca
n
b
e
en
co
u
n
ter
ed
in
clin
ical
p
r
a
ctice
.
Fin
ally
,
v
alid
atio
n
is
s
e
t
at
2
0
p
er
ce
n
t,
m
ea
n
in
g
th
at
th
e
g
en
er
at
o
r
will
au
to
m
atica
ll
y
r
eser
v
e
th
is
p
r
o
p
o
r
tio
n
o
f
im
ag
es
to
ev
alu
a
te
m
o
d
el
p
er
f
o
r
m
an
ce
d
u
r
in
g
tr
ain
in
g
,
with
th
e
r
em
ain
in
g
8
0
p
er
ce
n
t
b
ei
n
g
u
s
ed
f
o
r
ac
tu
al
lear
n
in
g
.
W
e
s
u
m
m
ar
ize
th
e
ap
p
r
o
ac
h
to
d
ata
p
r
ep
r
o
ce
s
s
in
g
an
d
au
g
m
en
tatio
n
in
th
e
f
o
llo
win
g
p
o
in
ts
:
−
R
esizin
g
: to
2
2
4
×2
2
4
p
ix
els (
Mo
b
ileNetV2
in
p
u
t size)
.
−
No
r
m
aliza
tio
n
: p
ix
el
v
alu
es scaled
to
[
0
,
1
]
.
−
Data
au
g
m
en
tatio
n
(
tr
ain
in
g
o
n
ly
)
:
r
an
d
o
m
r
o
tatio
n
(
±
2
0
°),
h
o
r
izo
n
tal/v
er
tical
s
h
if
ts
(
±
1
0
%),
zo
o
m
(
±
1
0
%),
h
o
r
izo
n
tal
f
lip
s
.
−
Valid
atio
n
s
p
lit:
2
0
% o
f
tr
ain
i
n
g
d
ata
u
s
ed
f
o
r
v
alid
atio
n
.
2
.
3
.
M
o
bil
eNe
t
V2
Arc
hite
ct
ure
T
h
e
f
ir
s
t
la
y
er
r
ep
r
esen
ts
a
2
2
4
x
2
2
4
p
ix
el
in
p
u
t
im
a
g
e
with
3
co
lo
r
ch
a
n
n
els
(
R
GB
)
,
f
o
ll
o
wed
b
y
a
s
ec
o
n
d
lay
er
s
p
ec
if
y
in
g
th
e
u
s
e
o
f
th
e
Mo
b
ileNetV2
ar
c
h
i
tectu
r
e
as
th
e
f
ea
tu
r
e
ex
tr
ac
t
o
r
.
T
h
e
th
ir
d
lay
e
r
p
er
f
o
r
m
s
a
Glo
b
alAv
er
ag
ePo
o
lin
g
2
D
th
at
tr
an
s
f
o
r
m
s
th
e
f
e
atu
r
e
m
ap
s
in
to
a
f
ea
tu
r
e
v
ec
t
o
r
.
T
h
e
f
o
u
r
th
la
y
er
is
a
d
en
s
e
lay
er
o
f
1
2
8
n
eu
r
o
n
s
with
R
eL
U
ac
tiv
atio
n
,
f
o
llo
wed
b
y
a
3
0
%
d
r
o
p
o
u
t
lay
e
r
to
r
eg
u
lar
ize
th
e
m
o
d
el
an
d
p
r
ev
en
t
o
v
er
f
itti
n
g
.
T
h
e
s
ix
th
lay
e
r
is
a
d
en
s
e
la
y
er
o
f
4
n
e
u
r
o
n
s
with
So
f
tm
ax
ac
tiv
atio
n
,
wh
ich
u
ltima
tely
p
r
o
d
u
ce
s
p
r
o
b
ab
ilit
ies f
o
r
th
e
f
o
u
r
class
es: Glio
m
a,
Me
n
in
g
io
m
a
,
Pit
u
itar
y
T
u
m
o
r
,
an
d
Hea
lth
y
.
Mo
b
ileNetV2
u
s
es
d
ep
th
wis
e
s
ep
ar
ab
le
co
n
v
o
l
u
tio
n
s
an
d
in
v
er
ted
lin
ea
r
b
o
ttlen
ec
k
s
,
r
ed
u
cin
g
p
ar
am
eter
s
an
d
c
o
m
p
u
tatio
n
al
co
s
t.
Ou
r
cu
s
to
m
h
ea
d
:
Glo
b
alAv
er
ag
ePo
o
lin
g
2
D
→
Den
s
e
(
1
2
8
,
R
eL
U)
→
Dr
o
p
o
u
t
(
3
0
%)
→
Den
s
e
(
4
,
So
f
tm
ax
)
.
Af
ter
p
r
esen
tin
g
th
e
ar
ch
itectu
r
e
o
f
th
e
M
o
b
ile
NetV2
m
o
d
el,
we
p
r
esen
t b
elo
w
a
f
l
o
w
d
iag
r
a
m
o
f
th
e
s
tu
d
y
p
r
o
p
o
s
ed
in
th
is
d
o
cu
m
en
t.
Fig
u
r
e
7
s
h
o
ws
t
h
e
p
r
o
p
o
s
ed
Mo
b
i
leN
etV
2
a
r
c
h
i
te
ct
u
r
e
,
c
o
n
s
is
ti
n
g
o
f
f
ea
t
u
r
e
e
x
t
r
a
cti
o
n
,
Glo
b
a
lA
v
e
r
a
g
eP
o
o
li
n
g
2
D,
De
n
s
e
,
D
r
o
p
o
u
t
,
a
n
d
S
o
f
t
m
a
x
la
y
er
s
f
o
r
f
o
u
r
-
c
lass
c
lass
i
f
i
ca
t
io
n
.
Fi
g
u
r
e
8
p
r
es
e
n
ts
th
e
o
v
e
r
al
l
ex
p
e
r
i
m
e
n
t
al
w
o
r
k
f
lo
w
,
f
r
o
m
d
at
a
p
r
e
p
a
r
at
io
n
a
n
d
m
o
d
el
t
r
a
in
in
g
t
o
ev
al
u
a
ti
o
n
a
n
d
f
i
n
al
r
es
u
lts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
281
-
29
8
286
Fig
u
r
e
7
.
Fu
n
ctio
n
al
d
etails o
f
th
e
b
asic M
o
b
ileNetv
2
ar
ch
itectu
r
e
Fig
u
r
e
8
.
C
o
m
p
lete
f
lo
w
d
iag
r
am
2
.
4
.
O
pera
t
io
na
l
m
o
del
T
h
e
o
v
er
all
ar
ch
itectu
r
e
o
f
t
h
e
o
p
er
atio
n
al
m
o
d
el
f
o
r
th
e
Mo
b
ileNetV2
th
at
is
u
tili
ze
d
f
o
r
th
e
class
if
icatio
n
o
f
b
r
ain
tu
m
o
r
s
b
ased
o
n
MRI
im
ag
es
is
p
r
es
en
ted
in
Fig
u
r
e
9
.
T
h
e
p
r
o
ce
d
u
r
e
ca
n
b
e
d
iv
id
e
d
in
to
th
r
ee
p
r
im
a
r
y
s
tag
es:
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
,
Mo
b
i
leN
etV2
ar
ch
itectu
r
e,
an
d
f
in
al
class
if
icatio
n
id
en
tific
atio
n
.
−
Data
au
g
m
e
n
tatio
n
:
Data
au
g
m
en
tatio
n
was
a
p
p
lied
to
th
e
t
r
ain
in
g
s
et
to
im
p
r
o
v
e
m
o
d
el
g
en
er
aliza
tio
n
b
y
ar
tific
ially
ex
p
an
d
i
n
g
it.
T
ec
h
n
iq
u
es
u
s
ed
in
cl
u
d
ed
h
o
r
iz
o
n
tal
an
d
v
er
tical
f
lip
s
,
as
well
a
s
r
an
d
o
m
r
o
tatio
n
s
with
in
a
±
2
0
° r
an
g
e,
to
s
im
u
late
v
ar
iatio
n
s
in
im
ag
e
ac
q
u
is
itio
n
p
o
s
itio
n
.
−
I
m
ag
e
R
esizin
g
:
All
o
f
th
e
im
ag
es
ar
e
r
esized
s
o
th
a
t
th
ey
co
n
f
o
r
m
to
t
h
e
s
p
e
cif
icatio
n
s
th
at
Mo
b
ileNetV2
r
eq
u
ir
es f
o
r
in
p
u
t (
f
o
r
ex
am
p
le,
2
2
4
b
y
2
2
4
p
i
x
els).
−
No
r
m
alis
ati
o
n
:
I
m
a
g
es
ar
e
n
o
r
m
al
is
e
d
t
o
en
s
u
r
e
th
at
th
e
i
n
t
e
n
s
it
ies
o
f
e
ac
h
p
i
x
e
l a
r
e
c
o
n
s
is
ten
t
f
o
r
ef
f
e
cti
v
e
lea
r
n
i
n
g
.
As
in
p
u
t
,
t
h
e
i
m
a
g
es
th
a
t
h
a
v
e
b
e
e
n
p
r
e
p
r
o
ce
s
s
e
d
a
r
e
i
n
t
r
o
d
u
ce
d
in
to
t
h
e
M
o
b
il
eN
etV
2
m
o
d
el
.
−
I
n
th
e
s
ec
o
n
d
s
tag
e
o
f
t
h
e
p
r
o
ce
s
s
,
th
e
Mo
b
ileNetV2
ar
ch
itectu
r
e
is
m
ad
e
u
p
o
f
s
ev
er
a
l
co
n
v
o
lu
tio
n
al
lay
er
s
th
at
ar
e
in
ter
s
p
er
s
ed
wi
th
b
atch
n
o
r
m
alis
atio
n
,
R
eL
U
ac
tiv
atio
n
s
,
an
d
m
ax
p
o
o
lin
g
lay
er
s
.
On
e
o
f
th
e
m
o
s
t
well
-
k
n
o
wn
ch
a
r
ac
ter
is
tics
o
f
th
e
Mo
b
ileNetV2
m
o
d
el
is
its
ef
f
ec
tiv
en
ess
in
f
ea
tu
r
e
ex
tr
ac
tio
n
th
r
o
u
g
h
th
e
u
tili
s
atio
n
o
f
d
ep
t
h
wis
e
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Mo
b
ileN
etV
2
w
ith
tr
a
n
s
fe
r
lea
r
n
in
g
fo
r
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
(
A
z
iz
S
r
a
i
)
287
a)
Featu
r
e
E
x
tr
ac
tio
n
: E
x
tr
ac
tin
g
f
ea
tu
r
es a
t m
u
ltip
le
lev
els is
wh
at
co
n
v
o
l
u
tio
n
al
lay
er
s
ar
e
all
ab
o
u
t.
b)
B
atch
No
r
m
aliza
tio
n
:
B
atch
No
r
m
aliza
tio
n
is
a
p
r
o
ce
s
s
th
a
t
n
o
r
m
alize
s
th
e
o
u
tp
u
ts
o
f
la
y
er
s
to
g
u
ar
an
tee
s
tab
le
an
d
ef
f
ec
tiv
e
tr
ai
n
in
g
.
c)
R
eL
U6
ac
tiv
atio
n
:
T
h
e
R
eL
U
ac
tiv
atio
n
in
cr
ea
s
es
th
e
a
m
o
u
n
t
o
f
n
o
n
-
lin
ea
r
ity
wh
ile
p
r
eser
v
in
g
th
e
n
u
m
er
ical
p
r
ed
ictab
ilit
y
.
d)
Ma
x
Po
o
lin
g
:
Ma
x
Po
o
lin
g
L
ay
er
s
ca
n
r
ed
u
ce
th
e
s
p
atial
d
im
en
s
io
n
s
wh
ile
s
ti
l
l
p
r
eser
v
in
g
th
e
ess
en
tia
l
ch
ar
ac
ter
is
tics
.
e)
Den
s
e
lay
er
s
:
T
o
p
ass
o
n
th
e
f
ea
tu
r
es
th
at
wer
e
ex
tr
ac
te
d
f
r
o
m
th
ese
lay
er
s
to
th
e
d
en
s
e
l
ay
er
s
,
th
ey
a
r
e
f
ir
s
t f
latten
ed
.
Fig
u
r
e
9
.
Deta
iled
o
v
er
v
iew
o
f
th
e
Mo
b
ilen
etV2
ar
ch
itectu
r
e
2.
5
.
T
ra
ns
f
er
lea
rning
a
nd
f
ine
-
t
un
ing
s
t
ra
t
eg
y
−
T
r
an
s
f
er
lear
n
in
g
:
W
e
lo
ad
ed
Mo
b
ileNetV2
p
r
e
-
tr
ain
ed
o
n
I
m
ag
eNe
t,
f
r
o
ze
all
b
ase
lay
er
s
,
an
d
tr
ain
ed
o
n
ly
th
e
n
ewly
ad
d
ed
to
p
lay
e
r
s
f
o
r
1
0
ep
o
c
h
s
(
o
p
tim
izer
: A
d
am
,
lear
n
in
g
r
ate:
0
.
0
0
1
,
b
atc
h
s
ize:
3
2
)
.
−
Fin
e
-
tu
n
i
n
g
:
W
e
u
n
f
r
o
ze
th
e
last
2
0
lay
er
s
o
f
th
e
b
ase
m
o
d
el
an
d
r
etr
ain
e
d
with
a
lo
wer
lear
n
in
g
r
ate
(
0
.
0
0
0
1
)
f
o
r
5
ep
o
ch
s
to
ad
a
p
t
f
ea
tu
r
es to
th
e
m
e
d
ical
d
o
m
ain
.
T
h
is
s
tr
ateg
y
av
o
id
s
o
v
er
f
itti
n
g
an
d
le
v
er
ag
es g
e
n
er
al
f
ea
tu
r
es f
r
o
m
n
atu
r
al
im
ag
es.
2
.
6
.
E
v
a
lua
t
i
o
n
m
et
rics
I
n
th
is
s
tu
d
y
,
th
e
p
er
f
o
r
m
a
n
c
e
o
f
th
e
p
r
o
p
o
s
ed
b
r
ain
tu
m
o
r
class
if
icatio
n
m
o
d
el
is
an
aly
ze
d
u
s
in
g
m
u
ltip
le
ev
alu
atio
n
m
etr
ics.
T
h
ese
m
etr
ics
ar
e
d
ef
in
e
d
an
d
ex
p
lain
e
d
f
u
r
th
er
in
th
is
d
o
cu
m
en
t.
W
e
u
s
ed
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
(
s
e
n
s
iti
v
ity
)
,
F1
-
s
co
r
e,
s
p
ec
if
icit
y
,
co
n
f
u
s
io
n
m
atr
i
x
,
an
d
R
OC
-
AUC.
T
h
ese
ar
e
cr
itical
in
m
ed
ical
im
ag
in
g
to
b
alan
ce
f
alse p
o
s
itiv
es a
n
d
f
al
s
e
n
eg
ativ
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
281
-
29
8
288
2
.
6
.
1
.
Acc
ura
cy
Acc
u
r
ac
y
r
e
f
er
s
to
th
e
p
r
o
p
o
r
tio
n
o
f
s
am
p
les
th
at
a
m
o
d
el
class
if
ies
co
r
r
ec
tly
.
T
o
co
m
p
u
te
it,
we
d
ef
in
e
tr
u
e
p
o
s
itiv
es
(
T
P)
an
d
tr
u
e
n
eg
ativ
es
(
T
N)
as
th
e
s
am
p
les
co
r
r
ec
tly
class
if
ied
,
wh
ile
f
alse
p
o
s
itiv
es
(
FP
)
an
d
f
alse
n
eg
ativ
es
(
FN)
co
r
r
esp
o
n
d
to
m
is
class
if
ied
s
a
m
p
les.
I
n
a
b
in
ar
y
m
ed
ical
cla
s
s
if
icatio
n
task
f
o
r
ex
am
p
le,
d
is
tin
g
u
is
h
in
g
h
ea
lt
h
y
i
n
d
iv
i
d
u
als
f
r
o
m
s
ick
p
atien
ts
a
T
P
o
cc
u
r
s
wh
en
a
s
ick
p
atien
t
is
co
r
r
ec
tly
id
en
tifie
d
as
s
ick
,
an
d
a
T
N
wh
en
a
h
ea
lth
y
p
er
s
o
n
is
c
o
r
r
ec
tly
id
en
tifie
d
as
h
ea
lth
y
.
C
o
n
v
er
s
ely
,
a
FP
co
r
r
esp
o
n
d
s
to
a
h
ea
lt
h
y
p
e
r
s
o
n
m
is
tak
en
ly
class
if
ied
as
s
ick
,
an
d
a
FN
to
a
s
ick
p
atien
t
in
co
r
r
ec
tly
p
r
ed
icted
as h
ea
lth
y
.
T
h
e
f
o
r
m
u
la
b
elo
w
s
h
o
ws h
o
w
to
ca
lcu
late
th
e
ac
cu
r
ac
y
s
co
r
e
o
n
a
g
iv
en
d
ataset:
=
(
TP
+
TN
)
(
TP
+
TN
+
FP
+
FN
)
T
h
is
m
etr
ic
ca
n
b
e
co
m
p
u
te
d
d
u
r
in
g
th
e
tr
ai
n
in
g
p
h
ase,
an
d
Fig
u
r
e
10
illu
s
tr
ates
a
ty
p
ical
ex
am
p
le.
Un
lik
e
th
e
lo
s
s
f
u
n
ctio
n
,
wh
ic
h
u
s
u
ally
d
ec
r
ea
s
es,
th
e
ac
cu
r
ac
y
cu
r
v
e
f
o
llo
ws a
lo
g
a
r
ith
m
i
c
in
cr
ea
s
e.
As
with
th
e
lo
s
s
cu
r
v
es,
c
o
m
p
ar
in
g
t
h
e
ac
c
u
r
ac
y
ac
h
iev
e
d
o
n
th
e
tr
ain
in
g
d
ata
with
t
h
at
o
b
tain
ed
o
n
th
e
test
d
ata
m
ak
es it p
o
s
s
ib
le
to
d
etec
t o
v
er
f
itti
n
g
o
r
u
n
d
er
f
i
ttin
g
.
Ho
w
ev
er
,
t
h
is
ac
cu
r
ac
y
m
e
tr
i
c
h
as
ce
r
ta
in
l
im
i
tat
io
n
s
,
as
i
t
p
r
o
v
i
d
es
o
n
l
y
a
g
l
o
b
a
l
s
co
r
e
.
R
e
tu
r
n
in
g
t
o
th
e
ex
am
p
l
e
o
f
b
i
n
a
r
y
class
if
ic
ati
o
n
in
t
h
e
m
e
d
i
ca
l
f
i
el
d
,
ac
c
u
r
ac
y
a
lo
n
e
d
o
es
n
o
t
i
n
d
ica
te
h
o
w
wel
l
t
h
e
m
o
d
e
l
id
e
n
t
if
ies
h
e
alt
h
y
s
u
b
je
cts
o
r
,
co
n
v
e
r
s
el
y
,
s
i
ck
p
ati
en
ts
.
T
o
ad
d
r
ess
t
h
is
,
w
e
c
a
n
e
x
a
m
i
n
e
th
e
s
e
n
s
it
iv
it
y
a
n
d
s
p
e
ci
f
ici
ty
s
c
o
r
es
.
S
en
s
iti
v
i
ty
m
ea
s
u
r
es
t
h
e
m
o
d
el'
s
ab
ili
ty
t
o
co
r
r
e
ctl
y
d
ete
ct
s
ic
k
p
a
tie
n
t
s
,
w
h
i
le
s
p
ec
i
f
i
cit
y
r
e
f
le
cts
its
a
b
il
it
y
to
c
o
r
r
ec
tl
y
i
d
e
n
ti
f
y
h
ea
l
th
y
i
n
d
iv
id
u
als.
T
h
ese
tw
o
m
e
tr
i
cs
all
o
w
t
h
e
m
o
d
el
to
b
e
f
in
e
-
tu
n
ed
ac
c
o
r
d
in
g
t
o
th
e
s
p
ec
i
f
i
c
n
e
e
d
s
o
f
t
h
e
u
s
er
.
T
h
e
e
q
u
at
io
n
s
b
el
o
w
d
eta
il
h
o
w
t
h
es
e
tw
o
s
co
r
e
s
a
r
e
c
alc
u
l
ate
d
.
Se
n
s
itivit
y
=
TP
TP
+
FN
Sp
e
c
ifi
c
ity
=
TN
TN
+
FP
Fig
u
r
e
10
.
Acc
u
r
ac
y
lea
r
n
in
g
cu
r
v
e
o
n
th
e
tr
ai
n
in
g
a
n
d
test
d
ataset
2
.
6
.
2
.
P
re
cisi
o
n
T
o
m
in
im
ize
f
alse
p
o
s
itiv
es,
p
r
ec
is
io
n
ev
alu
ates
a
m
o
d
el'
s
p
o
s
it
iv
e
p
r
ed
ictiv
e
ac
cu
r
ac
y
b
y
ca
lcu
latin
g
th
e
p
r
o
p
o
r
tio
n
o
f
c
o
r
r
ec
tly
p
r
e
d
icted
p
o
s
itiv
e
s
am
p
les
am
o
n
g
all
s
am
p
les
p
r
ed
icted
as
p
o
s
itiv
e.
I
ts
m
ath
em
atica
l e
x
p
r
ess
io
n
is
:
Pr
e
c
ision
=
TP
TP
+
FP
I
n
th
e
co
n
tex
t
o
f
tu
m
o
r
class
if
icatio
n
,
p
r
ec
is
io
n
p
lay
s
a
cr
u
cial
r
o
le
in
r
ed
u
cin
g
f
alse
-
p
o
s
itiv
e
d
iag
n
o
s
es.
Fo
r
in
s
tan
ce
,
m
is
tak
en
ly
i
d
en
tify
in
g
a
tu
m
o
r
o
n
a
s
tan
d
ar
d
b
r
ain
s
ca
n
ca
n
l
ea
d
to
u
n
n
ec
ess
ar
y
m
ed
ical
in
ter
v
e
n
tio
n
s
a
n
d
s
ig
n
if
ican
t
d
is
tr
ess
f
o
r
th
e
p
atien
t.
A
m
o
d
el
with
h
ig
h
p
r
ec
is
io
n
is
th
er
e
f
o
r
e
m
o
r
e
r
eliab
le
wh
en
p
r
ed
ictin
g
t
h
e
p
r
esen
ce
o
f
a
tu
m
o
r
.
2
.
6
.
3
.
Rec
a
ll/
s
ens
it
iv
it
y
R
ec
all,
also
co
m
m
o
n
ly
r
ef
e
r
r
ed
to
as
s
en
s
itiv
ity
,
m
ea
s
u
r
es
th
e
m
o
d
el'
s
ab
ilit
y
to
co
r
r
ec
t
ly
id
en
tify
p
o
s
itiv
e
s
am
p
les.
I
t is d
ef
in
ed
b
y
th
e
f
o
llo
win
g
f
o
r
m
u
la:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Mo
b
ileN
etV
2
w
ith
tr
a
n
s
fe
r
lea
r
n
in
g
fo
r
b
r
a
in
tu
mo
r
cla
s
s
ifica
tio
n
(
A
z
iz
S
r
a
i
)
289
R
e
c
a
l
l
=
Se
n
s
itivity
=
TP
TP
+
FN
I
n
m
ed
ical
im
ag
in
g
,
r
ec
all
is
ar
g
u
ab
ly
th
e
m
o
s
t
cr
itical
m
et
r
ic
to
co
n
s
id
er
,
s
in
ce
f
ailin
g
to
d
etec
t
a
tu
m
o
r
(
r
esu
ltin
g
in
a
f
alse
n
eg
ativ
e)
ca
n
h
av
e
f
atal
co
n
s
eq
u
e
n
ce
s
f
o
r
a
p
atien
t.
A
h
ig
h
r
ec
all
en
s
u
r
es
th
at
th
e
m
o
d
el
ca
p
t
u
r
es
n
ea
r
l
y
all
g
e
n
u
in
e
tu
m
o
r
ca
s
es,
m
ak
i
n
g
it
j
u
s
t
as
im
p
o
r
tan
t
as
p
r
ec
is
io
n
.
T
h
is
an
aly
s
is
p
lace
s
s
tr
o
n
g
em
p
h
asis
o
n
r
ec
all
b
ec
au
s
e
th
e
in
h
er
en
t
co
s
t
o
f
m
is
s
in
g
a
tu
m
o
r
is
f
ar
g
r
ea
ter
an
d
m
u
ch
m
o
r
e
s
ev
er
e
th
an
th
at
o
f
a
f
alse p
o
s
itiv
e.
2
.
6
.
4
.
F1
-
s
co
re
T
h
e
F1
-
s
co
r
e
c
o
m
p
u
tes
th
e
weig
h
ted
av
er
a
g
e
o
f
p
r
ec
is
io
n
an
d
r
ec
all
u
s
in
g
t
h
eir
h
a
r
m
o
n
ic
m
ea
n
,
co
m
b
in
in
g
b
o
t
h
m
etr
ics in
to
a
s
in
g
le
v
alu
e.
I
t c
a
n
b
e
e
x
p
r
ess
ed
as f
o
llo
ws:
F1
=
2
∗
Pr
ecis
i
o
n
∗
Recal
l
Pr
ecis
i
o
n
+
Recal
l
I
n
class
if
icatio
n
task
s
s
u
ch
as
b
r
ain
tu
m
o
r
d
etec
tio
n
,
w
h
er
e
d
atasets
o
f
te
n
s
u
f
f
er
f
r
o
m
class
im
b
alan
ce
,
th
e
F1
-
s
co
r
e
p
r
o
v
id
es
a
m
o
r
e
m
ea
n
in
g
f
u
l
ass
ess
m
en
t
o
f
m
o
d
el
p
er
f
o
r
m
a
n
ce
.
T
h
is
m
etr
ic
is
p
ar
ticu
lar
ly
u
s
ef
u
l
wh
e
n
th
e
r
e
is
a
tr
ad
e
-
o
f
f
b
etwe
en
p
r
ec
is
io
n
an
d
r
ec
all,
as
it
g
iv
es
eq
u
al
weig
h
t
to
b
o
t
h
.
A
h
ig
h
F1
-
s
co
r
e
in
d
icate
s
th
at
th
e
m
o
d
el
n
o
t
o
n
ly
co
r
r
ec
tly
id
en
tifie
s
tr
u
e
p
o
s
itiv
es
(
e.
g
.
,
ac
tu
al
tu
m
o
r
s
)
b
u
t
also
av
o
id
s
g
en
er
atin
g
a
lar
g
e
n
u
m
b
er
o
f
f
alse p
o
s
itiv
es.
2
.
6
.
5
.
Sp
ec
if
icit
y
Sp
ec
if
icity
is
a
k
ey
ev
alu
atio
n
m
etr
ic
in
m
ed
ical
d
iag
n
o
s
ti
c
s
y
s
tem
s
.
I
t
q
u
an
tifie
s
th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
id
e
n
tifie
d
n
e
g
ativ
e
c
ases
TN
r
elativ
e
to
all
ac
tu
al
n
eg
ativ
e
ca
s
es.
Sp
ec
if
icity
is
f
o
r
m
ally
d
ef
in
e
d
as:
Sp
e
c
ifi
c
ity
=
TN
TN
+
FP
Hig
h
s
p
ec
if
icity
is
e
s
p
ec
iall
y
cr
u
cial
in
m
ed
ical
im
ag
e
cl
ass
if
icatio
n
to
p
r
ev
en
t
f
alse
p
o
s
itiv
es,
wh
ich
ca
n
lead
t
o
u
n
n
ec
ess
ar
y
clin
ical
p
r
o
ce
d
u
r
es,
p
atien
t a
n
x
iety
,
an
d
h
ig
h
e
r
h
ea
lth
ca
r
e
e
x
p
en
s
es.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
wa
s
ev
alu
ated
o
n
th
e
Go
o
g
le
C
o
lab
p
latf
o
r
m
b
ased
o
n
th
e
f
ac
to
r
o
f
th
e
ev
alu
ati
o
n
m
atr
ix
an
d
th
e
m
o
d
el
p
er
f
o
r
m
an
ce
.
3
.
1
.
E
v
a
lua
t
io
n
m
et
rics
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
ev
alu
ated
u
s
in
g
v
ar
io
u
s
m
etr
ics,
in
clu
d
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
th
e
F1
s
c
o
r
e.
Ou
r
m
o
d
el
is
co
m
p
ar
ed
t
o
a
s
et
o
f
m
o
d
els
T
ab
le
1
[
2
4
]
.
T
h
e
co
m
p
a
r
is
o
n
o
f
o
u
r
m
o
d
el
with
ex
is
tin
g
m
o
d
el
s
,
ac
co
r
d
in
g
to
th
ese
d
if
f
e
r
en
t
m
etr
ics,
is
p
r
esen
ted
in
Fig
u
r
e
s
1
1
an
d
1
2
[
2
5
]
.
Fig
u
r
e
1
1
.
C
o
m
p
ar
is
o
n
o
f
m
o
d
el
p
er
f
o
r
m
an
ce
,
lin
e
g
r
a
p
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
281
-
29
8
290
Fig
u
r
e
12
.
C
o
m
p
ar
is
o
n
o
f
m
o
d
el
p
er
f
o
r
m
an
ce
,
b
ar
g
r
ap
h
T
ab
le
1
.
C
o
m
p
a
r
is
o
n
o
f
th
e
ac
cu
r
ac
y
o
f
o
u
r
m
o
d
el
with
o
th
e
r
ex
is
tin
g
m
o
d
els
M
o
d
e
l
F1
-
S
c
o
r
e
(
%)
A
c
c
u
r
a
c
y
(
%)
P
r
o
p
o
se
d
m
o
d
e
l
(
M
o
b
i
l
e
N
e
t
V
2
)
8
6
.
0
8
6
.
5
6
En
se
mb
l
e
d
e
e
p
l
e
a
r
n
i
n
g
(
ED
C
N
N
)
9
7
.
0
9
7
.
7
7
I
n
c
e
p
t
i
o
n
V
3
9
6
.
0
9
7
.
1
2
V
G
G
1
6
b
a
s
e
d
e
n
sem
b
l
e
+
C
N
N
9
2
.
0
9
8
.
4
1
B
r
a
i
n
N
e
t
9
8
.
0
9
8
.
1
8
R
o
b
u
st
C
N
N
+
U
-
N
e
t
(
n
o
s
e
g
m
e
n
t
a
t
i
o
n
)
9
8
.
0
9
8
.
7
0
R
o
b
u
st
C
N
N
+
U
-
N
e
t
(
w
i
t
h
s
e
g
me
n
t
a
t
i
o
n
)
9
8
.
0
9
8
.
8
0
G
e
n
e
r
i
c
C
N
N
(
b
i
n
a
r
y
c
l
a
ss
i
f
i
c
a
t
i
o
n
)
8
1
.
0
8
1
.
0
5
3
.
2
.
M
o
del
p
er
f
o
rma
nce
T
h
e
m
o
d
el
w
as
tr
ai
n
e
d
o
n
4
,
4
8
0
MR
I
im
ag
es
d
i
v
i
d
e
d
i
n
t
o
f
o
u
r
c
lass
es:
g
li
o
m
a
,
m
e
n
i
n
g
i
o
m
a,
p
i
tu
ita
r
y
tu
m
o
r
,
a
n
d
n
o
t
u
m
o
r
.
A
f
te
r
i
n
itial
t
r
ai
n
i
n
g
,
v
ali
d
ati
o
n
a
cc
u
r
ac
y
r
ea
c
h
e
d
8
8
%
;
a
f
t
er
f
in
e
-
t
u
n
i
n
g
,
t
est
ac
cu
r
a
c
y
was
8
6
.
5
6
%
T
a
b
l
e
1
.
T
h
e
m
ac
r
o
-
av
er
a
g
e
AUC
was
0
.
9
6
4
5
F
ig
u
r
e
25
,
i
n
d
i
ca
ti
n
g
e
x
c
ell
en
t
d
is
cr
im
i
n
ab
ilit
y
.
I
n
th
is
s
tu
d
y
,
an
d
to
s
u
p
p
o
r
t
o
u
r
ch
o
ice
o
f
th
e
Mo
b
ileN
etV2
m
o
d
el
as
th
e
b
est
m
o
d
el
f
o
r
th
e
class
if
icatio
n
an
d
d
etec
tio
n
o
f
b
r
ai
n
tu
m
o
r
s
,
we
d
ev
el
o
p
e
d
s
ev
er
a
l
m
o
d
els,
in
clu
d
in
g
a
C
NN
m
o
d
el
a
n
d
o
th
er
s
s
u
ch
as
VGG1
6
,
VGG
1
9
,
R
esNet5
0
,
Xce
p
tio
n
,
an
d
E
f
f
icien
tNet.
W
e
m
o
n
ito
r
e
d
t
h
e
p
er
f
o
r
m
an
ce
o
f
ea
ch
m
o
d
el,
s
o
m
etim
es
ch
an
g
in
g
th
e
n
u
m
b
er
o
f
ep
o
c
h
s
f
o
r
s
o
m
e.
W
e
co
n
f
ir
m
ed
t
h
at
th
e
Mo
b
ileNetV2
m
o
d
el
p
r
o
v
id
es
th
e
b
est
p
er
f
o
r
m
an
ce
with
a
s
im
p
le
ar
ch
itectu
r
e
[
2
6
]
.
T
h
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
m
o
d
els
u
s
ed
in
th
is
s
tu
d
y
is
p
r
esen
ted
in
th
e
g
r
ap
h
s
,
u
s
in
g
th
e
C
NN
m
o
d
el
ar
ch
itectu
r
e.
Fig
u
r
e
1
3
co
m
p
ar
es
th
e
p
r
ec
is
io
n
ac
h
iev
ed
b
y
ea
c
h
m
o
d
el,
wh
ile
Fig
u
r
e
1
4
illu
s
tr
ates
th
e
o
v
e
r
all
p
er
f
o
r
m
a
n
ce
tr
e
n
d
ac
r
o
s
s
d
if
f
er
e
n
t
ar
ch
itectu
r
es.
Fig
u
r
e
1
5
p
r
esen
ts
th
e
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
C
NN
m
o
d
el
u
s
ed
f
o
r
co
m
p
ar
is
o
n
,
a
n
d
Fig
u
r
e
1
6
c
o
m
p
a
r
es th
e
ac
cu
r
a
cy
an
d
F1
-
s
co
r
e
o
f
all
m
o
d
els
b
ef
o
r
e
a
n
d
af
te
r
f
in
e
-
t
u
n
in
g
.
Fig
u
r
e
1
3
.
p
r
ec
is
io
n
b
y
m
o
d
el
Fig
u
r
e
1
4
.
p
er
f
o
r
m
an
ce
tr
en
d
Evaluation Warning : The document was created with Spire.PDF for Python.