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
.
6
,
Decem
b
er
20
25
,
p
p
.
5
6
3
3
~
5
6
4
6
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
6
.
pp
5
6
3
3
-
5
6
4
6
5633
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
B
reas
t
c
a
ncer
d
et
ection
using
e
nse
mble met
ho
ds
Ala
a
M
o
ha
m
ed
G
ha
zy
1
,
H
a
l
a
B
a
hy
Na
f
ea
2
,
F
a
y
ez
Wa
nis
Z
a
k
i
2
,
H
a
na
n
M
o
ha
m
ed
Amer
2
1
D
e
p
a
r
t
me
n
t
o
f
C
o
mm
u
n
i
c
a
t
i
o
n
s
a
n
d
El
e
c
t
r
o
n
i
c
s
En
g
i
n
e
e
r
i
n
g
,
M
a
n
s
o
u
r
a
H
i
g
h
e
r
I
n
st
i
t
u
t
e
of
E
n
g
i
n
e
e
r
i
n
g
a
n
d
Te
c
h
n
o
l
o
g
y
,
M
a
n
s
o
u
r
a
,
E
g
y
p
t
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
s
En
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
En
g
i
n
e
e
r
i
n
g
,
M
a
n
so
u
r
a
U
n
i
v
e
r
si
t
y
,
M
a
n
s
o
u
r
a
,
Eg
y
p
t
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
2
4
,
2
0
2
5
R
ev
is
ed
Au
g
2
2
,
2
0
2
5
Acc
ep
ted
Sep
1
4
,
2
0
2
5
Bre
a
st
c
a
n
c
e
r
(BC)
is
o
n
e
o
f
t
h
e
m
o
st
c
o
m
m
o
n
c
a
n
c
e
rs
a
m
o
n
g
w
o
m
e
n
.
Th
is
stu
d
y
'
s
fra
m
e
wo
rk
is
d
iv
i
d
e
d
i
n
to
t
h
re
e
p
h
a
se
s.
F
irst
ly
,
a
m
a
jo
rit
y
h
a
rd
v
o
ti
n
g
a
p
p
r
o
a
c
h
is
u
se
d
t
o
a
p
p
ly
a
n
e
n
se
m
b
le
c
las
sifica
ti
o
n
m
e
c
h
a
n
ism
a
s
a
d
e
c
isio
n
fu
si
o
n
tec
h
n
i
q
u
e
on
t
h
e
lev
e
l
o
f
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
(CNN
s).
F
iv
e
p
re
-
train
e
d
CNN
s
—
v
isu
a
l
g
e
o
m
e
try
g
ro
u
p
1
9
(
VG
G
1
9
),
d
e
n
se
ly
c
o
n
n
e
c
ted
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
two
rk
2
0
1
(De
n
se
Ne
t2
0
1
)
,
re
sid
u
a
l
n
e
two
rk
5
0
(Re
sN
e
t5
0
)
,
m
o
b
il
e
n
e
two
r
k
v
e
rsio
n
2
(M
o
b
il
e
Ne
t
V2
),
a
n
d
in
c
e
p
ti
o
n
v
e
rsio
n
3
(In
c
e
p
ti
o
n
V
3
)
—
a
re
e
v
a
lu
a
ted
,
u
si
n
g
a
d
a
ta
sp
l
it
ti
n
g
tes
t
ra
ti
o
re
p
re
se
n
ts
3
0
%
o
f
t
h
e
to
tal
d
a
tas
e
t.
S
e
c
o
n
d
l
y
,
th
e
c
las
sifica
ti
o
n
re
su
lt
s
o
f
t
h
e
fiv
e
CNN
s
a
re
c
o
m
p
a
re
d
to
g
e
t
t
h
e
b
e
st
-
p
e
rf
o
rm
a
n
c
e
m
o
d
e
l.
Th
e
n
,
se
v
e
n
sta
te
o
f
a
rt
m
a
c
h
in
e
c
las
sifiers
—
d
e
c
isio
n
tree
(DT),
h
ist
o
g
ra
m
-
b
a
se
d
g
ra
d
ien
t
b
o
o
sti
n
g
c
las
sifie
r
(HG
B),
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
(S
V
M
),
ra
n
d
o
m
fo
re
st
(RF
),
l
o
g
isti
c
re
g
re
ss
io
n
(
LR),
g
ra
d
ien
t
b
o
o
st
in
g
(G
B),
a
n
d
e
x
trem
e
g
ra
d
ien
t
b
o
o
st
in
g
(XG
B)
—
a
re
u
se
d
to
imp
r
o
v
e
sy
ste
m
p
e
rf
o
rm
a
n
c
e
o
n
t
h
e
fe
a
tu
re
v
e
c
to
r
th
a
t
wa
s
tak
e
n
fr
o
m
th
is
CNN
m
o
d
e
l.
T
h
ir
d
ly
,
to
imp
ro
v
e
ro
b
u
stn
e
ss
,
a
m
a
jo
rit
y
h
a
rd
v
o
ti
n
g
tec
h
n
i
q
u
e
is
u
se
d
a
t
t
h
e
e
x
tern
a
l
c
las
sifier
lev
e
l
u
si
n
g
th
e
h
i
g
h
e
st
fo
u
r
c
la
ss
ifi
e
rs
se
lec
ted
b
a
se
d
o
n
th
e
ir
a
c
c
u
ra
c
y
.
S
e
v
e
ra
l
e
x
p
e
rime
n
ts
we
re
c
o
n
d
u
c
ted
in
t
h
is
st
u
d
y
,
a
n
d
t
h
e
re
su
lt
s
sh
o
we
d
th
a
t
Re
sN
e
t5
0
p
r
o
d
u
c
e
d
th
e
b
e
st
re
su
lt
s
in
term
s
o
f
p
re
c
isio
n
a
n
d
a
c
c
u
ra
c
y
.
Th
e
m
a
jo
rit
y
v
o
ti
n
g
m
e
c
h
a
n
ism
imp
ro
v
e
s
t
h
e
sy
ste
m
’s
a
c
c
u
ra
c
y
to
9
9
.
8
5
%
th
ro
u
g
h
CNN
s a
n
d
to
1
0
0
%
th
r
o
u
g
h
tra
d
it
i
o
n
a
l
c
las
sifiers
.
K
ey
w
o
r
d
s
:
B
r
ea
s
t c
an
ce
r
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
E
x
ter
n
al
class
if
ier
s
M
a
j
o
r
i
t
y
h
a
r
d
v
o
t
i
n
g
m
e
c
h
a
n
is
m
Ma
m
m
o
g
r
am
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
:
Alaa
Gh
az
y
Dep
ar
tm
en
t o
f
C
o
m
m
u
n
icatio
n
s
an
d
E
lectr
o
n
ics E
n
g
in
ee
r
i
n
g
,
Ma
n
s
o
u
r
a
Hig
h
er
I
n
s
titu
te
o
f
E
n
g
i
n
ee
r
in
g
an
d
T
ec
h
n
o
lo
g
y
Ma
n
s
o
u
r
a,
E
g
y
p
t
E
m
ail
:
alaa
m
o
h
am
ed
g
h
az
y
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
B
C
is
th
e
m
o
s
t
co
m
m
o
n
d
is
ea
s
e
in
th
e
wo
r
ld
,
with
7
.
8
m
illi
o
n
p
eo
p
le
liv
in
g
with
a
d
iag
n
o
s
is
as
o
f
th
e
en
d
o
f
2
0
2
0
.
I
t
o
cc
u
r
s
in
ev
er
y
co
u
n
tr
y
o
f
th
e
wo
r
ld
in
wo
m
en
at
an
y
ag
e
af
ter
p
u
b
er
ty
b
u
t
with
in
cr
ea
s
in
g
r
ates
in
later
life
,
ac
co
r
d
in
g
to
th
e
W
o
r
ld
Hea
lth
Or
g
an
izatio
n
(
W
HO
)
[
1
]
.
Ma
m
m
o
g
r
ap
h
y
is
th
e
m
o
s
t
r
eliab
le
an
d
ac
cu
r
ate
s
cr
ee
n
in
g
tech
n
iq
u
e
an
d
r
em
ain
s
th
e
g
o
ld
s
tan
d
ar
d
f
o
r
co
m
m
u
n
ity
b
r
ea
s
t
ca
n
ce
r
(
B
C
)
s
cr
ee
n
in
g
[
2
]
.
C
u
r
r
en
tly
,
alth
o
u
g
h
it
ca
n
no
t
b
e
r
ep
lace
d
,
m
am
m
o
g
r
ap
h
y
is
s
till
u
tili
ze
d
co
m
b
in
ed
with
MRI
an
d
u
ltra
s
o
u
n
d
,
p
ar
ticu
lar
ly
in
ca
s
es
wh
er
e
th
e
d
en
s
ity
o
f
b
r
ea
s
t
tis
s
u
e
is
s
ig
n
if
ican
t.
I
t
is
th
e
m
ajo
r
clin
ical
test
in
g
B
ec
au
s
e
it
ca
n
b
e
in
ter
p
r
eted
in
s
ev
er
al
m
an
n
er
s
to
p
r
o
v
id
e
ad
d
itio
n
al
in
f
o
r
m
atio
n
p
r
io
r
to
d
iag
n
o
s
is
o
r
d
etec
tio
n
,
h
ig
h
p
r
ec
is
e,
an
d
ca
n
id
en
tify
ar
o
u
n
d
8
0
%
–
9
0
% o
f
th
e
d
an
g
er
o
f
ca
n
ce
r
.
Acc
o
r
d
in
g
to
p
r
ev
io
u
s
s
tu
d
ies,
ea
r
ly
d
etec
tio
n
(
th
e
tu
m
o
r
is
at
least 2
0
m
m
in
s
ize)
o
f
b
r
ea
s
t c
an
ce
r
is
cr
u
cial
b
ec
au
s
e
it
ca
n
co
n
tr
ib
u
te
to
u
p
to
a
4
0
% d
ec
r
ea
s
e
in
m
o
r
tality
r
ate
[
3
]
.
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
.
6
,
Decem
b
e
r
20
25
:
5
6
3
3
-
5
6
4
6
5634
T
h
e
g
o
al
o
f
th
is
r
esear
ch
is
to
in
cr
ea
s
e
th
e
p
r
ec
is
io
n
o
f
B
C
p
r
ed
ictio
n
m
o
d
els.
Fu
ll
ar
ticle
u
s
u
ally
f
o
llo
ws
a
s
tan
d
ar
d
s
tr
u
ctu
r
e:
T
h
e
r
elate
d
wo
r
k
is
p
r
esen
ted
in
s
ec
tio
n
2
.
Sectio
n
3
p
r
o
v
id
es
a
d
etailed
d
escr
ip
tio
n
o
f
th
e
f
r
am
ewo
r
k
'
s
s
u
g
g
ested
m
eth
o
d
o
lo
g
y
,
em
p
lo
y
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
,
class
if
icatio
n
m
eth
o
d
s
,
an
d
a
m
ajo
r
ity
v
o
tin
g
s
y
s
tem
.
Sectio
n
4
d
elv
es in
to
d
ataset
d
escr
ip
tio
n
,
Sy
s
tem
s
p
ec
if
icatio
n
s
,
Fin
e
-
tu
n
ed
p
r
e
-
tr
ain
ed
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
,
d
ata
p
r
ep
r
o
ce
s
s
in
g
,
an
d
th
e
ex
p
er
im
en
tal
r
esu
lts
.
I
t
also
in
clu
d
es
th
e
s
tan
d
ar
d
s
b
y
wh
ich
th
e
m
o
d
els'
p
er
f
o
r
m
an
ce
is
ev
alu
ated
.
T
h
e
co
n
clu
s
io
n
is
s
h
o
wn
in
s
ec
tio
n
5
.
T
h
e
p
r
in
cip
al
co
n
tr
ib
u
tio
n
s
o
f
th
e
s
u
g
g
ested
wo
r
k
ar
e:
a.
C
o
m
p
ar
e
th
e
ac
cu
r
ac
y
an
d
ev
alu
atio
n
m
ea
s
u
r
e
p
er
f
o
r
m
an
ce
o
f
in
d
iv
id
u
al
p
r
etr
ain
ed
C
NNs
th
en
ap
p
ly
th
e
m
ajo
r
ity
h
ar
d
v
o
tin
g
to
in
cr
ea
s
e
th
e
b
en
ef
it
f
r
o
m
f
ea
tu
r
es
o
f
ea
ch
C
NN
an
d
th
u
s
in
cr
ea
s
e
ac
cu
r
ac
y
o
f
th
e
s
y
s
tem
.
b.
Stu
d
y
th
e
ef
f
ec
t
o
f
u
s
in
g
e
x
ter
n
al
m
ac
h
in
e
lear
n
in
g
class
if
ier
s
o
n
th
e
p
e
r
f
o
r
m
an
ce
o
f
p
r
etr
ain
ed
C
NN
m
o
d
el
b
y
ap
p
ly
i
n
g
s
ev
en
m
ac
h
in
e
lear
n
in
g
class
if
ier
s
to
th
e
f
ea
tu
r
e
v
ec
to
r
ex
tr
ac
ted
f
r
o
m
b
est
ar
ch
itectu
r
e
o
u
t o
f
th
e
f
iv
e
p
r
e
-
tr
ain
ed
m
o
d
el
f
r
o
m
th
e
f
ir
s
t stag
e
an
d
c
o
m
p
ar
e
its
p
er
f
o
r
m
a
n
ce
.
c.
S
tu
d
y
in
g
th
e
e
f
f
ec
t
o
f
a
p
p
ly
in
g
two
-
lev
el
en
s
em
b
le
f
r
am
ewo
r
k
th
at
co
m
b
in
es
d
ee
p
l
ea
r
n
in
g
m
o
d
els
(
C
NNs)
an
d
ex
ter
n
al
class
if
ier
s
to
en
s
u
r
e
r
o
b
u
s
tn
ess
an
d
in
cr
ea
s
e
s
y
s
tem
p
er
f
o
r
m
an
ce
.
2.
RE
L
AT
E
D
WO
RK
Ma
m
m
o
g
r
am
im
a
g
es
wer
e
wid
ely
s
tu
d
ied
f
o
r
B
C
d
etec
tio
n
an
d
class
if
icatio
n
u
s
in
g
v
ar
io
u
s
C
NN
s
.
Nu
m
er
o
u
s
r
esear
ch
s
tu
d
ies
h
av
e
r
ep
o
r
ted
s
ig
n
if
ican
t
r
esu
lts
u
s
in
g
d
if
f
er
en
t
class
if
ier
s
an
d
d
ee
p
lear
n
in
g
tech
n
iq
u
es.
A
s
u
m
m
ar
y
o
f
th
e
m
o
s
t
s
ig
n
if
ican
t
r
elate
d
wo
r
k
is
in
clu
d
ed
in
T
ab
le
1
,
s
h
o
wca
s
in
g
d
iv
er
s
e
ap
p
r
o
ac
h
es
ap
p
lied
to
d
if
f
e
r
e
n
t
d
atasets
.
L
ag
h
m
ati
et
a
l.
[
4
]
ap
p
lied
f
o
u
r
tr
ad
itio
n
al
cl
ass
if
ier
s
—
ar
tific
ia
l
n
eu
r
al
n
etwo
r
k
(
ANN)
,
k
-
n
e
ar
est
n
eig
h
b
o
r
s
(
KNN)
,
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es
(
SVM)
,
an
d
d
ec
is
io
n
tr
ee
(
DT
)
—
o
n
th
e
Ma
m
m
o
g
r
ap
h
i
c
Ma
s
s
d
ataset,
ac
h
iev
in
g
a
m
ax
im
u
m
ac
c
u
r
ac
y
o
f
8
4
%
with
ANN.
Ho
wev
er
,
th
e
s
tu
d
y
was
lim
ited
b
y
its
r
elian
ce
o
n
class
ical
m
o
d
els
an
d
r
elativ
ely
lo
w
ac
cu
r
ac
y
.
Ng
u
y
en
et
a
l.
[
5
]
u
tili
ze
d
b
o
th
s
u
p
er
v
is
ed
a
n
d
u
n
s
u
p
er
v
is
ed
lear
n
in
g
ap
p
r
o
a
ch
es
with
p
r
in
cip
al
co
m
p
o
n
e
n
t
an
aly
s
is
(
PC
A)
—
b
ased
f
ea
tu
r
e
s
elec
tio
n
.
Alth
o
u
g
h
en
s
em
b
le
v
o
tin
g
ac
h
i
ev
ed
ar
o
u
n
d
9
0
%
ac
cu
r
ac
y
,
th
e
s
tu
d
y
lack
ed
in
teg
r
atio
n
with
d
ee
p
f
ea
tu
r
e
ex
tr
ac
to
r
s
an
d
d
id
n
o
t
ex
p
lo
r
e
C
NN
ar
ch
itectu
r
es,
wh
ich
ar
e
k
n
o
wn
to
p
er
f
o
r
m
b
etter
in
im
ag
e
-
b
ased
task
s
.
T
ab
le
1
.
Sh
o
w
s
u
m
m
ar
y
o
f
r
e
lated
wo
r
k
N
o
.
A
u
t
h
o
r
Y
e
a
r
M
e
t
h
o
d
o
l
o
g
y
R
e
s
u
l
t
s
1
La
g
h
m
a
t
i
e
t
a
l
.
[
4
]
2
0
1
9
A
N
N
,
S
V
M
,
D
T
a
n
d
K
N
N
W
i
t
h
8
4
%
a
c
c
u
r
a
c
y
,
t
h
e
A
N
N
m
o
d
e
l
p
r
o
d
u
c
e
d
t
h
e
b
e
st
r
e
s
u
l
t
s
2
N
g
u
y
e
n
e
t
a
l
.
[
5
]
2
0
1
9
S
V
M
,
A
d
a
B
o
o
st
,
L
R
a
n
d
e
n
s
e
m
b
l
e
v
o
t
i
n
g
c
l
a
ss
i
f
i
e
r
En
se
mb
l
e
v
o
t
i
n
g
c
l
a
ssi
f
i
e
r
s
a
c
h
i
e
v
e
d
a
r
o
u
n
d
9
0
%
a
c
c
u
r
a
c
y
3
G
o
p
a
l
e
t
a
l
.
[
6
]
2
0
2
1
R
F
,
M
LP
a
n
d
LR
Th
e
a
p
p
l
i
e
d
R
F
mo
d
e
l
a
c
h
i
e
v
e
d
9
5
%
a
c
c
u
r
a
c
y
4
B
a
t
a
i
n
e
h
[
7
]
2
0
1
9
K
-
N
N
,
N
B
,
M
LP,
a
n
d
S
V
M
t
h
r
o
u
g
h
W
B
C
D
d
a
t
a
se
t
Th
e
ma
x
i
mu
m
a
c
c
u
r
a
c
y
o
f
9
9
.
1
2
%
w
a
s
a
t
t
a
i
n
e
d
b
y
M
LP.
5
M
a
n
g
u
k
i
y
a
[
8
]
2
0
2
2
R
F
,
N
B
,
S
V
M
,
D
T
,
K
N
N
,
X
G
B
o
o
st
,
a
n
d
A
d
a
B
o
o
st
t
h
r
o
u
g
h
W
B
C
D
X
G
B
o
o
st
a
c
h
i
e
v
e
d
9
8
.
2
4
%
a
c
c
u
r
a
c
y
6
O
sman
a
n
d
A
l
j
a
h
d
a
l
i
[
9
]
2
0
2
0
En
se
mb
l
e
b
o
o
s
t
i
n
g
w
i
t
h
R
B
F
n
e
u
r
a
l
n
e
t
w
o
r
k
En
se
mb
l
e
b
o
o
s
t
i
n
g
a
c
h
i
e
v
e
d
9
8
.
4
%
a
c
c
u
r
a
c
y
7
K
u
mar
e
t
a
l
.
[
1
0
]
2
0
1
9
1
2
c
l
a
ss
i
f
i
e
r
s
i
n
c
l
u
d
i
n
g
t
r
e
e
a
n
d
l
a
z
y
a
l
g
o
r
i
t
h
ms
Tr
e
e
a
n
d
l
a
z
y
c
l
a
ssi
f
i
e
r
s r
e
a
c
h
e
d
9
9
%
a
c
c
u
r
a
c
y
8
M
o
h
a
me
d
e
t
a
l
.
[
1
1
]
2
0
2
2
C
N
N
(
U
-
N
e
t
)
o
n
D
M
R
-
I
R
d
a
t
a
s
e
t
S
p
e
c
i
f
i
c
i
t
y
=
9
8
.
6
7
%,
S
e
n
s
i
t
i
v
i
t
y
=
1
0
0
%,
a
n
d
A
c
c
u
r
a
c
y
=
9
9
.
3
3
%
9
S
i
n
g
h
e
t
a
l
.
[
1
2
]
2
0
2
3
H
y
b
r
i
d
a
l
g
o
r
i
t
h
ms
o
n
W
B
C
D
d
a
t
a
set
A
c
h
i
e
v
e
d
9
8
.
9
6
%
a
c
c
u
r
a
c
y
.
10
C
h
e
n
e
t
a
l
.
[
1
3
]
2
0
2
3
X
G
B
,
R
F
,
L
R
,
a
n
d
K
N
N
X
G
B
a
c
h
i
e
v
e
d
b
e
s
t
a
c
c
u
r
a
c
y
o
f
9
7
.
4
%.
11
H
a
mza
a
n
d
M
e
z
l
[
1
4
]
2
0
2
4
C
N
N
(
U
-
N
e
t
+
+
)
w
i
t
h
M
o
b
i
l
e
N
e
t
V
2
a
n
d
I
n
c
e
p
t
i
o
n
V
3
o
n
u
l
t
r
a
so
u
n
d
M
o
b
i
l
e
N
e
t
V
2
a
c
h
i
e
v
e
d
9
6
.
5
8
%
a
c
c
u
r
a
c
y
;
I
n
c
e
p
t
i
o
n
V
3
a
c
h
i
e
v
e
d
7
2
.
8
0
%.
Go
p
al
et
a
l.
[
6
]
em
p
lo
y
ed
R
F,
L
R
,
an
d
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
(
ML
P)
o
n
th
e
W
B
C
D
d
ataset,
r
ep
o
r
tin
g
a
m
ax
im
u
m
ac
c
u
r
a
cy
o
f
9
5
%.
Ho
wev
er
,
th
eir
m
eth
o
d
d
i
d
n
o
t
c
o
n
s
id
er
c
o
m
b
i
n
in
g
d
ee
p
lear
n
i
n
g
with
class
ical
m
o
d
els,
lim
it
in
g
its
s
ca
lab
ilit
y
an
d
r
o
b
u
s
tn
ess
o
n
lar
g
er
d
atasets
.
B
atain
eh
[
7
]
co
m
p
ar
ed
ML
P,
KNN,
SVM,
an
d
Naïv
e
B
ay
e
s
(
NB
)
,
wh
er
e
ML
P
ac
h
iev
ed
9
6
.
7
0
%
ac
cu
r
ac
y
.
Desp
ite
th
i
s
im
p
r
o
v
e
m
en
t,
t
h
e
s
tu
d
y
was
co
n
s
tr
ain
ed
b
y
th
e
lack
o
f
en
s
em
b
le
s
tr
ateg
ies
an
d
ab
s
en
ce
o
f
d
ec
is
io
n
f
u
s
io
n
m
ec
h
an
is
m
s
th
at
co
u
ld
i
n
cr
ea
s
e
m
o
d
el
r
eliab
ilit
y
.
Ma
n
g
u
k
i
y
a
[
8
]
test
ed
s
ev
en
class
if
ier
s
,
r
ep
o
r
tin
g
a
p
ea
k
a
cc
u
r
ac
y
o
f
9
8
.
1
4
%
u
s
in
g
XGBo
o
s
t.
Ho
wev
er
,
th
e
lack
o
f
C
NN
-
b
ased
f
ea
tu
r
e
e
x
tr
ac
tio
n
lim
its
th
e
m
o
d
el'
s
ap
p
licab
ilit
y
f
o
r
h
i
g
h
-
d
im
en
s
io
n
al
im
ag
e
d
ata.
Osm
an
an
d
Aljah
d
ali
[
9
]
u
s
ed
a
n
e
n
s
em
b
le
b
o
o
s
tin
g
m
eth
o
d
co
m
b
in
ed
with
a
r
ad
ial
b
asis
f
u
n
ctio
n
(
R
B
F)
n
eu
r
al
n
etwo
r
k
,
r
ea
ch
in
g
9
8
.
4
%
ac
c
u
r
ac
y
.
Yet,
th
ei
r
ap
p
r
o
ac
h
d
i
d
n
o
t
in
c
o
r
p
o
r
ate
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
B
r
ea
s
t c
a
n
ce
r
d
etec
tio
n
u
s
in
g
en
s
emb
le
meth
o
d
s
(
A
la
a
Mo
h
a
med
Gh
a
z
y
)
5635
m
u
ltip
le
C
NNs
o
r
e
v
alu
ate
r
o
b
u
s
tn
ess
th
r
o
u
g
h
ex
te
r
n
al
class
if
ier
s
.
Ku
m
ar
et
a
l.
[
1
0
]
ev
alu
ate
d
twelv
e
class
if
ier
s
an
d
ac
h
iev
ed
u
p
t
o
9
9
%
ac
cu
r
ac
y
u
s
in
g
tr
ee
-
b
ased
m
o
d
els.
W
h
ile
co
m
p
r
eh
en
s
iv
e,
th
eir
s
tu
d
y
lack
ed
a
h
y
b
r
id
e
n
s
em
b
le
s
tr
u
ctu
r
e
th
at
c
o
m
b
in
es
C
NN
f
ea
tu
r
es
with
tr
ad
itio
n
al
class
if
ier
s
.
Mo
h
am
ed
et
a
l.
[
1
1
]
ap
p
lied
a
U
-
Net
C
NN
to
th
er
m
al
im
a
g
es
an
d
ac
h
ie
v
ed
ex
ce
llen
t
p
er
f
o
r
m
a
n
ce
(
ac
cu
r
ac
y
=
9
9
.
3
3
%,
s
en
s
itiv
ity
=
1
0
0
%).
No
n
eth
eless
,
th
e
s
co
p
e
was
lim
ited
to
th
er
m
al
im
ag
in
g
an
d
d
id
n
o
t
ex
ten
d
to
m
am
m
o
g
r
a
p
h
y
o
r
e
n
s
em
b
le
f
u
s
io
n
s
tr
ateg
ies.
Sin
g
h
et
a
l.
[
1
2
]
u
s
ed
a
h
y
b
r
id
f
ea
tu
r
e
s
elec
tio
n
m
eth
o
d
th
at
im
p
r
o
v
e
d
p
er
f
o
r
m
an
ce
to
9
8
.
1
6
%,
b
u
t
th
e
s
tu
d
y
d
i
d
n
o
t
ex
p
lo
r
e
en
s
em
b
le
m
eth
o
d
s
co
m
b
in
in
g
m
u
ltip
le
C
NNs
wi
th
m
ac
h
in
e
lear
n
i
n
g
class
if
ier
s
.
C
h
en
et
a
l.
[
1
3
]
co
m
p
ar
e
d
XGBo
o
s
t,
R
F,
L
R
,
an
d
KNN,
with
XGBo
o
s
t
p
er
f
o
r
m
in
g
b
est
(
9
7
.
4
%
ac
cu
r
ac
y
)
.
Ho
wev
er
,
th
e
a
b
s
en
ce
o
f
C
NN
-
b
ased
f
ea
tu
r
e
ex
tr
ac
tio
n
r
estricts
its
ef
f
ec
tiv
en
ess
f
o
r
co
m
p
lex
im
ag
e
d
ata.
Ham
za
an
d
Me
zl
[
1
4
]
em
p
lo
y
e
d
a
C
NN
(
U
-
Net+
+)
f
o
r
b
r
ea
s
t
r
eg
io
n
s
eg
m
en
tatio
n
f
r
o
m
u
lt
r
aso
u
n
d
im
ag
es
an
d
u
s
ed
M
o
b
ileNetV2
an
d
I
n
ce
p
tio
n
V3
f
o
r
class
if
icatio
n
.
Alth
o
u
g
h
Mo
b
ileNetV2
ac
h
iev
ed
9
6
.
5
8
%
ac
cu
r
ac
y
,
th
e
s
tu
d
y
was
lim
ited
b
y
u
s
in
g
s
in
g
le
C
NN
m
o
d
els
an
d
lack
ed
class
if
ier
-
lev
el
en
s
em
b
le
m
ec
h
an
is
m
s
.
2
.
1
.
I
dentif
ied g
a
ps
a
nd
m
o
t
iv
a
t
io
n
W
h
ile
ex
is
tin
g
s
tu
d
ies
h
av
e
m
ad
e
m
ea
n
in
g
f
u
l
p
r
o
g
r
ess
,
m
o
s
t
s
u
f
f
er
f
r
o
m
at
least
o
n
e
o
f
th
e
f
o
llo
win
g
lim
itatio
n
s
:
i)
Used
o
n
ly
s
in
g
le
m
o
d
el,
m
ak
in
g
th
em
s
u
s
ce
p
tib
le
to
o
v
er
f
i
ttin
g
an
d
lack
o
f
g
en
er
aliza
tio
n
;
ii)
L
im
ited
in
teg
r
atio
n
o
f
d
ee
p
lear
n
in
g
wit
h
tr
ad
itio
n
al
class
if
ier
s
,
r
ed
u
cin
g
th
e
p
o
ten
tial
f
o
r
h
y
b
r
id
lear
n
i
n
g
;
iii)
L
ac
k
o
f
m
u
lti
-
lev
el
en
s
em
b
le
tech
n
i
q
u
es,
wh
ich
c
o
u
ld
en
h
a
n
ce
r
o
b
u
s
tn
ess
an
d
ac
cu
r
ac
y
;
an
d
iv
)
U
s
e
o
f
n
o
n
-
m
am
m
o
g
r
ap
h
ic
d
atasets
(
th
er
m
al
o
r
u
ltra
s
o
u
n
d
im
ag
es),
lim
itin
g
ap
p
licab
ilit
y
to
m
am
m
o
g
r
a
p
h
y
-
b
ased
d
iag
n
o
s
is
.
T
h
ese
lim
itatio
n
s
h
ig
h
lig
h
t
th
e
n
ee
d
f
o
r
a
m
o
r
e
c
o
m
p
r
e
h
en
s
iv
e,
en
s
em
b
le
-
b
ased
f
r
am
ewo
r
k
t
h
at
in
teg
r
at
es
m
u
ltip
le
C
NN
s
an
d
tr
ad
itio
n
al
class
if
ier
s
th
r
o
u
g
h
d
ec
is
io
n
-
lev
el
f
u
s
io
n
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
ad
d
r
ess
es th
ese
g
ap
s
b
y
:
a.
C
o
m
p
ar
in
g
a
n
d
co
m
b
in
in
g
f
iv
e
d
if
f
er
en
t CNN m
o
d
els u
s
in
g
a
h
ar
d
v
o
tin
g
m
ec
h
an
is
m
.
b.
F
ee
d
in
g
th
e
b
est C
NN’
s
ex
tr
ac
ted
f
ea
tu
r
es in
to
m
u
ltip
le
m
a
ch
in
e
lear
n
in
g
class
if
ier
s
.
c.
A
p
p
ly
in
g
a
s
ec
o
n
d
e
n
s
em
b
le
a
t th
e
class
if
ier
lev
el
to
en
h
an
c
e
s
y
s
tem
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
.
3.
M
E
T
H
O
D
T
h
is
r
esear
ch
p
r
o
p
o
s
es
a
r
o
b
u
s
t
h
y
b
r
id
e
n
s
em
b
le
f
o
u
n
d
atio
n
f
o
r
B
C
b
in
ar
y
class
if
icatio
n
with
m
am
m
o
g
r
a
p
h
ic
im
a
g
es.
T
h
e
m
eth
o
d
o
l
o
g
y
is
s
tr
u
ctu
r
e
d
in
t
o
th
r
ee
s
tag
es,
in
teg
r
atin
g
th
e
s
tr
en
g
th
s
o
f
C
NNs
an
d
class
if
ier
s
to
g
eth
e
r
.
T
h
e
f
r
am
ewo
r
k
is
ev
alu
ated
u
s
i
n
g
a
p
u
b
licly
a
v
ailab
le
m
a
m
m
o
g
r
a
p
h
y
d
ataset.
Fig
u
r
es 1
th
r
o
u
g
h
4
illu
s
tr
ate
th
e
wo
r
k
f
l
o
w
an
d
c
o
m
p
o
n
en
ts
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
.
a.
F
ir
s
t s
ta
g
e:
u
tili
z
in
g
p
r
e
-
tr
a
in
ed
C
N
N
s
fo
r
fea
tu
r
e
ex
tr
a
ctio
n
A
t
th
e
b
eg
in
n
in
g
,
f
iv
e
p
r
e
-
tr
ain
ed
m
o
d
els
—
VGG1
9
,
Den
s
eNe
t2
0
1
,
R
esNet5
0
,
Mo
b
ileNetV2
,
an
d
I
n
ce
p
tio
n
V3
—
ar
e
u
s
ed
in
m
am
m
o
g
r
ap
h
y
im
ag
es to
ex
tr
ac
t d
ee
p
f
ea
tu
r
es a
s
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
ese
m
o
d
els
ar
e
f
in
e
-
tu
n
ed
,
u
s
in
g
weig
h
ts
in
itialized
f
r
o
m
th
e
I
m
ag
eNe
t
d
ataset.
T
o
en
h
an
ce
th
e
ad
ap
tab
ilit
y
o
f
ea
ch
m
o
d
el
to
th
e
b
r
ea
s
t
ca
n
ce
r
class
if
icatio
n
task
,
cu
s
to
m
class
if
icatio
n
h
ea
d
s
ar
e
ap
p
en
d
ed
,
co
n
s
is
tin
g
o
f
Flatten
lay
er
,
f
u
lly
co
n
n
ec
ted
Den
s
e
lay
er
u
s
in
g
R
eL
U
ac
tiv
atio
n
,
Dr
o
p
o
u
t
lay
er
to
av
o
id
o
v
er
f
itti
n
g
,
an
d
f
in
al
So
f
t
m
ax
lay
er
to
p
er
f
o
r
m
b
in
ar
y
class
if
icatio
n
.
Fig
u
r
e
1
.
Prim
e
m
ec
h
an
is
m
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
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
.
6
,
Decem
b
e
r
20
25
:
5
6
3
3
-
5
6
4
6
5636
T
h
e
d
ataset
is
s
p
lit
r
an
d
o
m
ly
in
to
tr
ain
in
g
s
et
r
ep
r
esen
ts
7
0
%
an
d
th
e
test
in
g
s
et
r
ep
r
esen
ts
3
0
%
o
f
th
e
to
tal
d
ataset.
T
h
e
Ad
am
o
p
tim
izer
is
u
s
ed
to
f
in
e
-
tu
n
e
ea
ch
m
o
d
el
ac
r
o
s
s
2
0
ep
o
ch
s
an
d
ea
r
ly
s
to
p
p
in
g
p
atien
ce
=
1
0
to
r
ed
u
ce
u
n
n
ec
ess
ar
y
co
m
p
u
tatio
n
a
n
d
Min
i
m
ize
o
v
er
f
itti
n
g
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
ea
ch
C
NN
is
in
d
ep
en
d
en
tly
ev
alu
ated
a
n
d
co
m
p
ar
ed
to
g
et
th
e
b
est
-
p
er
f
o
r
m
an
ce
m
o
d
el.
T
o
en
h
an
ce
d
ec
is
io
n
r
eliab
ilit
y
an
d
r
e
d
u
ce
m
o
d
el
-
s
p
ec
if
ic
b
i
ases
,
th
e
o
u
tp
u
ts
o
f
all
f
iv
e
C
NNs
ar
e
f
u
s
ed
u
s
in
g
a
m
a
jo
r
ity
h
ar
d
v
o
tin
g
m
ec
h
an
is
m
at
th
e
d
ec
is
io
n
lev
el
as
s
h
o
wn
in
Fig
u
r
e
2
.
T
h
is
en
s
em
b
le
ap
p
r
o
ac
h
ag
g
r
eg
ates
th
e
p
r
ed
ictio
n
s
f
r
o
m
in
d
iv
id
u
al
m
o
d
els an
d
s
elec
ts
th
e
m
o
s
t f
r
eq
u
en
t c
lass
a
s
th
e
f
in
al
d
ec
is
io
n
.
Fig
u
r
e
2
.
Ma
jo
r
ity
-
h
ar
d
v
o
tin
g
tech
n
iq
u
e
o
n
th
e
lev
el
o
f
co
n
v
o
lu
tio
n
n
eu
r
al
n
etwo
r
k
s
b.
S
ec
o
n
d
st
a
ge:
appl
yi
n
g se
ve
n cl
a
ss
i
f
i
er
s
t
h
rough
t
he
be
st
C
N
N
At
s
ec
o
n
d
s
tag
e,
th
e
b
est
-
p
er
f
o
r
m
in
g
C
NN
m
o
d
el
f
r
o
m
th
e
p
r
ev
io
u
s
p
h
ase
is
u
s
ed
as
a
f
ea
tu
r
e
ex
tr
ac
to
r
.
T
h
e
f
ea
tu
r
e
v
ec
to
r
s
o
b
tain
ed
f
r
o
m
th
e
p
en
u
ltima
te
lay
er
o
f
th
is
m
o
d
el
s
er
v
e
as
in
p
u
t
to
s
ev
en
class
if
ier
s
:
DT
,
L
R
,
HGB,
SVM,
R
F,
GB
,
an
d
XGB
as
s
h
o
wn
Fig
u
r
e
3
.
E
ac
h
class
if
ier
is
tr
ain
ed
o
n
th
e
ex
tr
ac
ted
f
ea
tu
r
e
v
ec
to
r
s
u
s
in
g
co
n
s
is
ten
t
h
y
p
er
p
ar
am
eter
tu
n
in
g
.
T
h
is
s
tag
e
is
d
esig
n
ed
to
lev
er
ag
e
th
e
d
is
cr
im
in
ativ
e
p
o
wer
o
f
th
e
C
NN
f
ea
tu
r
es
an
d
th
e
class
if
ier
s
ab
ilit
ies
to
d
is
tin
g
u
is
h
b
etwe
en
b
en
ig
n
tu
m
o
r
an
d
m
alig
n
an
t tu
m
o
r
.
Fig
u
r
e
3
.
Sev
e
n
s
tate
-
of
-
th
e
-
ar
t m
ac
h
in
e
class
if
ier
s
ar
e
u
s
ed
in
th
e
p
r
o
p
o
s
ed
m
eth
o
d
c.
Th
ir
d
st
age
:
U
s
i
ng m
aj
ori
t
y
hard v
o
t
i
ng on
t
he
l
e
ve
l
of
ex
t
e
rna
l
c
l
ass
i
f
i
e
rs
I
n
th
e
f
in
al
s
tag
e,
a
m
ajo
r
ity
h
ar
d
v
o
tin
g
m
ec
h
an
is
m
is
ap
p
lied
o
n
th
e
lev
el
o
f
th
e
ex
ter
n
al
class
if
ier
s
.
T
h
e
b
est
f
o
u
r
class
if
ier
s
th
at
ac
h
iev
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
p
er
f
o
r
m
an
ce
ar
e
s
elec
ted
an
d
in
teg
r
ated
u
s
in
g
a
m
ajo
r
ity
h
ar
d
v
o
tin
g
s
tr
ateg
y
as
s
h
o
wn
in
Fig
u
r
e
4
.
T
h
is
class
if
ier
-
lev
el
en
s
em
b
le
ag
g
r
eg
ates
th
e
p
r
ed
ictio
n
s
o
f
th
e
m
o
s
t
r
eliab
le
m
o
d
els,
aim
in
g
to
r
ed
u
ce
in
d
iv
id
u
al
class
if
ier
v
ar
ian
ce
an
d
im
p
r
o
v
e
o
v
er
all
s
y
s
tem
r
o
b
u
s
tn
ess
an
d
s
tab
ilit
y
.
T
h
e
p
r
o
p
o
s
ed
th
r
ee
-
s
tag
e
f
r
am
ewo
r
k
co
m
p
r
is
in
g
d
ee
p
f
ea
tu
r
e
ex
tr
ac
tio
n
,
th
e
in
co
r
p
o
r
atio
n
o
f
class
if
ier
s
,
an
d
en
s
em
b
le
-
b
ased
d
ec
is
io
n
f
u
s
io
n
s
tr
ateg
ically
co
m
b
in
es
th
e
b
en
ef
its
o
f
d
ee
p
lear
n
in
g
an
d
co
n
v
en
tio
n
al
m
eth
o
d
s
.
T
h
is
h
y
b
r
id
s
tr
ateg
y
im
p
r
o
v
es
th
e
s
y
s
tem
’
s
ab
ilit
y
to
d
eliv
er
h
ig
h
class
if
icatio
n
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
B
r
ea
s
t c
a
n
ce
r
d
etec
tio
n
u
s
in
g
en
s
emb
le
meth
o
d
s
(
A
la
a
Mo
h
a
med
Gh
a
z
y
)
5637
ac
cu
r
ac
y
,
im
p
r
o
v
ed
g
en
er
aliza
tio
n
,
an
d
in
cr
ea
s
ed
r
o
b
u
s
tn
ess
ac
r
o
s
s
d
iv
er
s
e
d
iag
n
o
s
tic
co
n
d
itio
n
s
.
E
x
p
er
im
en
tal
f
in
d
in
g
s
v
alid
ate
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
f
r
am
ewo
r
k
,
h
ig
h
lig
h
tin
g
its
p
r
o
m
is
e
as
an
ef
f
ec
tiv
e
an
d
d
ep
en
d
ab
le
m
eth
o
d
f
o
r
B
C
d
etec
tin
g
in
m
ed
ical
im
ag
in
g
.
Fig
u
r
e
4
.
m
ajo
r
ity
-
h
ar
d
v
o
tin
g
m
ec
h
an
is
m
o
n
th
e
lev
el
o
f
e
x
ter
n
al
class
if
ier
s
3
.
1
.
T
he
a
pp
lied CN
Ns
o
f
pr
o
po
s
ed
m
o
del
3
.
1
.
1
.
VG
G
1
9
a
rc
hite
ct
ure
I
t
is
a
d
ee
p
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
e
p
r
o
p
o
s
ed
b
y
th
e
Vis
u
al
Geo
m
etr
y
Gr
o
u
p
at
th
e
Un
iv
er
s
ity
o
f
Ox
f
o
r
d
,
in
tr
o
d
u
ce
d
b
y
Simo
n
y
an
an
d
Z
is
s
er
m
an
in
2
0
1
4
[
1
5
]
.
I
t
co
n
s
is
ts
o
f
1
9
weig
h
t
lay
er
s
th
at
u
s
e
s
m
all
3
×3
co
n
v
o
lu
tio
n
f
ilter
s
co
n
s
is
ten
tly
ac
r
o
s
s
all
co
n
v
o
lu
tio
n
al
lay
er
s
.
T
h
e
ar
ch
itectu
r
e
f
o
llo
ws
a
f
ix
ed
p
atter
n
o
f
co
n
v
o
lu
tio
n
al
lay
er
s
th
en
R
ec
tifie
d
L
in
ea
r
Un
it
an
d
m
ax
-
p
o
o
lin
g
lay
er
s
,
wh
ich
im
p
r
o
v
es
f
ea
tu
r
e
ex
tr
ac
tio
n
wh
ile
m
ain
tain
in
g
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
T
h
is
u
n
if
o
r
m
ar
ch
itectu
r
e
f
ac
ilit
ates
d
ee
p
er
n
etwo
r
k
s
with
o
u
t
s
ig
n
if
ican
tly
in
cr
ea
s
in
g
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
,
m
ak
in
g
it
well
-
s
u
ited
f
o
r
lar
g
e
-
s
ca
le
im
ag
e
r
ec
o
g
n
itio
n
task
s
[
1
6
]
.
3
.
1
.
2
.
Dense
Net
a
rc
hite
ct
ure
Den
s
eNe
t,
o
r
d
en
s
ely
co
n
n
ec
ted
co
n
v
o
lu
tio
n
al
n
etwo
r
k
,
was
in
tr
o
d
u
ce
d
b
y
Hu
an
g
et
a
l.
in
2
0
1
7
[
1
7
]
.
B
y
f
ee
d
-
f
o
r
war
d
ly
co
n
n
ec
tin
g
ea
ch
lay
er
to
ev
er
y
o
th
er
lay
er
,
th
is
ar
ch
itectu
r
e
im
p
r
o
v
es
g
r
ad
ien
t
f
lo
w
an
d
in
f
o
r
m
atio
n
.
Sp
ec
if
ically
,
ev
er
y
lay
er
r
ec
eiv
es
f
ea
tu
r
e
m
ap
s
f
r
o
m
all
p
r
ec
ed
in
g
lay
er
s
,
p
r
o
m
o
tin
g
f
ea
tu
r
e
r
eu
s
e
an
d
m
itig
atin
g
th
e
p
r
o
b
lem
o
f
v
an
is
h
in
g
g
r
ad
ien
t.
I
t is ef
f
icien
t in
ter
m
s
o
f
p
ar
am
eter
u
s
ag
e,
as it a
v
o
id
s
r
ed
u
n
d
an
t
f
ea
tu
r
e
lear
n
in
g
an
d
d
ec
r
ea
s
es
th
e
o
v
er
all
p
ar
am
eter
s
in
co
n
tr
ast
to
tr
ad
itio
n
al
C
NNs.
I
ts
u
s
e
o
f
d
en
s
e
b
lo
ck
s
an
d
tr
an
s
itio
n
lay
er
s
allo
ws
f
o
r
co
m
p
ac
t,
y
et
p
o
wer
f
u
l,
n
etwo
r
k
s
th
at
p
er
f
o
r
m
ex
ce
p
tio
n
ally
well
o
n
class
if
icatio
n
task
s
[
1
8
]
.
3
.
1
.
3
.
ResNet
5
0
a
rc
hite
ct
ure
I
t
is
a
d
ee
p
r
esid
u
al
n
etwo
r
k
in
tr
o
d
u
ce
d
b
y
He
et
a
l.
[
1
8
]
,
[
1
9
]
,
wh
ich
was
d
ev
elo
p
ed
at
Mic
r
o
s
o
f
t
R
esear
ch
Asi
a.
T
h
e
ar
ch
itectu
r
e
is
d
is
tin
g
u
is
h
ed
b
y
its
u
s
e
o
f
r
esid
u
al
co
n
n
ec
tio
n
s
o
r
"sk
ip
co
n
n
ec
tio
n
s
,
"
wh
ich
m
ak
e
it
ea
s
ier
f
o
r
g
r
ad
ien
ts
to
m
o
v
e
th
r
o
u
g
h
th
e
lay
er
s
d
u
r
in
g
th
e
tr
ain
in
g
o
f
d
ee
p
n
etwo
r
k
s
.
T
h
ese
co
n
n
ec
tio
n
s
s
o
lv
in
g
th
e
p
r
o
b
lem
o
f
v
an
is
h
in
g
g
r
ad
ien
t,
wh
ich
co
m
m
o
n
ly
af
f
ec
ts
d
ee
p
C
NNs.
R
esNet5
0
co
n
s
is
ts
o
f
5
0
lay
er
s
an
d
was p
r
etr
ain
ed
o
n
th
e
I
m
ag
eNe
t d
ataset
to
en
h
an
ce
its
g
en
er
aliza
tio
n
ca
p
ab
ilit
ies
.
3
.
1
.
4
.
M
o
bil
e
Net
a
rc
hite
ct
ure
I
t
is
a
lig
h
tweig
h
t
d
ee
p
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
e
d
esig
n
ed
b
y
Go
o
g
le’
s
Mo
b
ile
Vis
io
n
team
,
s
p
ec
if
ically
tailo
r
ed
f
o
r
ef
f
icien
t
in
f
er
en
ce
in
m
o
b
ile
an
d
em
b
ed
d
ed
v
is
io
n
ap
p
licatio
n
s
[
2
0
]
.
T
h
e
m
o
d
el
in
tr
o
d
u
ce
s
Dep
th
wis
e
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
s
an
d
lin
ea
r
b
o
ttlen
ec
k
s
to
m
in
im
ize
m
o
d
el
s
ize
an
d
co
m
p
u
tin
g
co
m
p
lex
ity
wh
ile
p
r
eser
v
in
g
co
m
p
etitiv
e
ac
cu
r
ac
y
.
T
h
e
lin
ea
r
b
o
ttlen
ec
k
lay
er
s
r
ed
u
ce
th
e
d
im
en
s
io
n
ality
o
f
f
ea
tu
r
e
m
ap
s
an
d
en
h
an
ce
n
o
n
-
lin
ea
r
ity
,
en
ab
lin
g
th
e
m
o
d
el
to
o
p
er
ate
ef
f
icien
tly
o
n
d
ev
ices w
ith
lim
ited
p
r
o
ce
s
s
in
g
r
eso
u
r
ce
s
.
3
.
1
.
5
.
I
ncept
io
n
Net
a
rc
hite
ct
ure
I
t
is
also
k
n
o
wn
as
Go
o
g
L
eNe
t,
was
in
tr
o
d
u
ce
d
b
y
Szeg
ed
y
et
a
l.
[
2
0
]
,
[
2
1
]
.
I
t
em
p
lo
y
s
f
ac
to
r
ized
co
n
v
o
lu
tio
n
s
an
d
in
ce
p
tio
n
m
o
d
u
les
to
d
r
am
atica
lly
m
in
im
ize
o
v
er
all
p
ar
am
eter
s
wh
ile
r
etain
in
g
h
ig
h
ac
cu
r
ac
y
in
task
s
o
f
im
ag
e
class
if
icatio
n
.
T
h
e
ar
ch
itectu
r
e
co
m
b
in
es
m
u
ltip
le
co
n
v
o
lu
tio
n
al
f
ilter
s
izes
with
in
a
s
in
g
le
m
o
d
u
le,
wh
ich
allo
ws
it
to
ca
p
tu
r
e
in
f
o
r
m
atio
n
at
d
if
f
er
en
t
s
ca
les.
Du
e
to
its
ef
f
icien
t
d
esig
n
an
d
p
o
wer
f
u
l p
er
f
o
r
m
an
ce
.
I
t is p
r
etr
ain
ed
o
n
th
e
I
m
ag
eNe
t d
ataset
an
d
wid
ely
ad
o
p
ted
in
co
m
p
u
ter
v
is
io
n
task
s
.
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
.
6
,
Decem
b
e
r
20
25
:
5
6
3
3
-
5
6
4
6
5638
3
.
2
.
T
he
us
e
o
f
diff
er
ent
cla
s
s
if
iers t
hro
ug
h pro
po
s
e
d m
o
del.
3
.
2
.
1
.
Ra
nd
o
m
decisi
o
n t
re
e
I
t
is
s
u
p
er
v
is
ed
alg
o
r
ith
m
s
m
ain
ly
u
s
ed
f
o
r
th
e
g
r
ap
h
ical
r
ep
r
esen
tatio
n
o
f
all
th
e
p
o
s
s
ib
le
s
o
lu
tio
n
s
[
2
2
]
.
I
t
is
ch
ar
ac
ter
ized
b
y
th
e
ab
ilit
y
to
id
en
tify
an
d
ch
o
o
s
e
th
e
m
o
s
t
im
p
o
r
tan
t
attr
ib
u
tes
wh
ich
ar
e
u
s
ef
u
l
in
th
e
class
if
icatio
n
s
tag
e.
I
t
ca
n
also
s
elec
t
th
e
attr
ib
u
tes
wh
ich
d
eliv
er
th
e
m
ax
im
u
m
in
f
o
r
m
atio
n
g
ain
(
I
G)
wh
ich
is
d
ef
in
ed
as:
=
(
)
−
(
ℎ
)
(1
)
wh
er
e
E
n
tr
o
p
y
(
E
)
is
d
ef
in
e
d
as:
=
∑
−
(
l
og
2
)
(2
)
an
d
is
th
e
p
r
o
b
a
b
ilit
y
o
f
class
i.
3
.
2
.
2
.
G
ra
dient
bo
o
s
t
ing
cla
s
s
if
ier
I
t
is
p
r
ef
er
r
ed
f
o
r
s
m
all
s
am
p
les
an
d
is
co
n
s
id
er
ed
an
ex
ce
llen
t
m
o
d
el
f
o
r
r
eg
r
ess
io
n
an
d
class
if
icatio
n
,
p
ar
ticu
lar
ly
f
o
r
tab
u
lar
d
ata
[
2
3
]
.
I
t
is
ch
ar
ac
ter
ized
b
y
ea
s
y
im
p
lem
en
tatio
n
,
lo
w
co
m
p
u
tatio
n
al
co
s
t,
an
d
ef
f
icien
cy
.
3
.
2
.
3
.
H
is
t
g
ra
dient
bo
o
s
t
ing
cla
s
s
if
ier
I
t
is
an
en
h
an
ce
d
v
er
s
io
n
o
f
GB
DT
th
at
cr
ea
tes
a
h
is
to
g
r
am
o
f
f
ea
tu
r
e
v
alu
es
d
u
r
in
g
tr
ain
in
g
wh
ile
r
ed
u
cin
g
tr
ain
in
g
tim
e
an
d
m
em
o
r
y
co
n
s
u
m
p
tio
n
an
d
s
p
litt
in
g
th
e
co
n
tin
u
o
u
s
v
ar
iab
le
in
to
b
in
s
.
I
t
is
ch
ar
ac
ter
ized
b
y
s
p
ee
d
with
lar
g
e
n
u
m
b
er
o
f
s
am
p
les.
T
h
e
u
tili
za
tio
n
o
f
h
is
to
g
r
am
s
an
d
b
etter
d
ata
s
tr
u
ctu
r
es
is
p
r
im
ar
ily
r
esp
o
n
s
ib
le
f
o
r
th
is
s
p
ee
d
in
cr
ea
s
e.
T
h
e
alg
o
r
ith
m
lear
n
s
h
o
w
to
h
an
d
le
m
is
s
in
g
d
ata
d
u
r
in
g
tr
ain
in
g
,
m
ak
in
g
th
e
p
r
o
ce
s
s
m
o
r
e
s
tr
aig
h
tf
o
r
war
d
an
d
ef
f
icien
t
[
2
4
]
.
3
.
2
.
4
.
Su
pp
o
rt
v
ec
t
o
r
m
a
c
hi
ne
I
t
co
n
s
id
er
s
th
e
m
o
s
t
d
ep
en
d
ab
le
alg
o
r
ith
m
s
b
ased
o
n
s
tatis
tical
lear
n
in
g
f
r
am
ewo
r
k
s
.
I
t
is
r
eg
ar
d
ed
as
a
Dec
is
io
n
p
lan
e
-
b
ased
m
o
d
el
[
2
5
]
th
at
o
f
f
er
s
a
s
o
lu
tio
n
f
o
r
b
o
th
r
eg
r
ess
io
n
an
d
class
if
icatio
n
p
r
o
b
lem
s
as
well
as
f
o
r
b
o
th
lin
ea
r
an
d
n
o
n
-
lin
ea
r
d
atasets
.
T
h
e
b
asic
id
ea
o
f
SVM
was
to
s
ep
ar
ate
d
if
f
er
en
t
g
r
o
u
p
s
u
s
in
g
h
y
p
er
p
lan
es.
T
h
e
two
m
ain
is
s
u
es
with
SVM
ar
e
th
e
co
r
r
ec
t
s
elec
tio
n
o
f
k
er
n
el
f
u
n
ctio
n
,
an
d
its
p
ar
am
eter
s
[
2
6
]
.
T
h
e
k
er
n
el
f
u
n
ctio
n
allo
ws
SVMs
to
class
if
y
o
n
e
-
d
im
en
s
io
n
al
d
ata
in
a
two
-
d
im
en
s
io
n
al
ap
p
r
o
ac
h
.
T
y
p
ically
,
a
lin
ea
r
k
er
n
el
f
u
n
ctio
n
is
d
ef
in
ed
as f
o
llo
ws:
(
,
)
=
⋅
(
3
)
An
d
Po
ly
n
o
m
ial
k
er
n
el
f
u
n
cti
o
n
s
ar
e
d
ef
i
n
ed
as:
(
,
1
)
=
(
1
+
⋅
)
(
4
)
‘
’
is
d
eg
r
ee
o
f
k
er
n
el
f
u
n
ctio
n
.
3
.
2
.
5
.
Ra
nd
o
m
decisi
o
n f
o
re
s
t
I
t
b
u
ild
s
m
an
y
d
ec
is
io
n
tr
ee
s
d
u
r
in
g
th
e
tr
ain
in
g
p
h
ase
an
d
th
en
g
en
er
ates
class
es
f
o
r
ea
ch
.
I
t
is
m
ain
ly
u
s
ed
in
class
if
icatio
n
an
d
r
eg
r
ess
io
n
.
B
y
cr
ea
tin
g
n
u
m
er
o
u
s
d
ec
is
io
n
tr
ee
s
f
r
o
m
tr
ain
in
g
d
ata
u
s
in
g
b
o
o
ts
tr
ap
p
ed
s
am
p
les
with
a
s
m
all
m
o
d
if
icatio
n
,
th
e
d
e
-
co
r
r
elate
d
tr
ee
v
ia
b
ag
g
in
g
.
T
h
e
im
p
ac
t
o
f
ea
ch
p
r
ed
ictio
n
in
f
o
r
m
s
th
e
f
in
al
p
r
ed
ictio
n
.
I
t c
an
in
ter
p
r
et
ir
r
elev
an
t a
ttrib
u
tes an
d
h
an
d
le
m
is
s
in
g
d
ata
[
2
7
]
.
3
.
2
.
6
.
Sim
ple
lo
g
is
t
ic
re
g
re
s
s
io
n m
o
del
T
h
e
L
R
m
o
d
el
is
a
p
o
p
u
lar
ch
o
ice
f
o
r
b
in
ar
y
class
if
icatio
n
s
[
2
8
]
.
I
t
is
b
eliev
ed
th
at
a
lin
ea
r
co
m
b
in
atio
n
o
f
th
e
in
p
u
t
f
ea
tu
r
es
eq
u
als
th
e
co
n
d
itio
n
al
p
r
o
b
ab
ilit
y
o
f
o
n
e
o
f
th
e
two
o
u
tp
u
t
class
es.
T
h
e
class
if
icatio
n
m
o
d
el'
s
lo
g
is
tic
eq
u
atio
n
is
as f
o
llo
ws:
=
ln
(
1
−
)
(
5
)
wh
er
e
r
ep
r
esen
ts
p
r
o
b
a
b
ilit
y
th
at
ev
en
t
will o
cc
u
r
.
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
B
r
ea
s
t c
a
n
ce
r
d
etec
tio
n
u
s
in
g
en
s
emb
le
meth
o
d
s
(
A
la
a
Mo
h
a
med
Gh
a
z
y
)
5639
3
.
2
.
7
.
E
x
t
re
m
e
g
ra
dient
bo
o
s
t
ing
E
x
tr
em
e
g
r
ad
ien
t
b
o
o
s
tin
g
is
a
s
ca
lab
le
m
ac
h
in
e
lear
n
in
g
s
y
s
tem
f
o
r
tr
ee
b
o
o
s
tin
g
[
2
9
]
wh
ich
is
im
p
lem
en
ted
f
o
r
s
u
p
er
v
is
ed
lear
n
in
g
p
r
o
b
lem
s
an
d
d
ev
elo
p
ed
s
p
ec
if
ically
to
b
o
o
s
t
m
o
d
el
p
er
f
o
r
m
an
ce
an
d
co
m
p
u
tatio
n
al
ef
f
ec
tiv
en
ess
.
I
t
s
o
lv
es
p
r
o
b
lem
s
u
s
in
g
m
in
im
al
r
eso
u
r
ce
s
an
d
in
co
r
p
o
r
ates
a
r
eg
u
lar
ized
m
o
d
el
to
p
r
ev
en
t o
v
er
f
itti
n
g
an
d
is
s
u
ited
f
o
r
class
if
icatio
n
p
r
o
b
lem
s
.
3
.
3
.
M
a
j
o
rit
y
ha
rd
v
o
t
ing
mecha
nis
m
o
f
pro
po
s
ed
m
o
del
I
t
is
wid
ely
u
s
ed
th
r
o
u
g
h
en
s
em
b
le
class
if
icatio
n
.
I
t
is
also
ca
lled
p
lu
r
ality
v
o
tin
g
an
d
u
s
ed
to
im
p
r
o
v
e
th
e
class
if
icatio
n
r
esu
lts
.
I
n
th
is
tech
n
iq
u
e,
th
e
p
r
ed
ictio
n
o
f
class
lab
el
y
p
er
f
o
r
m
s
v
ia
m
ajo
r
ity
v
o
tin
g
o
f
ea
ch
class
if
ier
C
:
=
{
1
(
)
,
2
(
)
,
…
,
(
)
}
(
6
)
4.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S
4
.
1
.
Da
t
a
des
cr
iptio
n
Fo
r
ex
p
er
im
en
tatio
n
,
th
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
is
th
e
b
r
ea
s
t
m
am
m
o
g
r
ap
h
y
d
ataset
with
m
ass
es
in
tr
o
d
u
ce
d
b
y
Hu
an
g
an
d
L
in
[
3
0
]
.
All im
ag
es in
th
is
d
ataset
o
n
th
e
Me
n
d
eley
web
s
ite
wer
e
av
ailab
le
in
PNG
f
o
r
m
at
an
d
s
ized
to
2
2
7
×
2
2
7
p
ix
els.
T
h
e
im
ag
e
d
atasets
o
n
th
is
web
s
ite
wer
e
ar
r
an
g
ed
in
to
th
r
ee
m
ain
d
atasets
:
I
N
b
r
ea
s
t,
MI
AS,
an
d
DDSM
d
atasets
.
I
n
th
is
s
tu
d
y
,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
was
im
p
lem
en
ted
o
n
o
n
ly
th
e
d
ig
ital
d
atab
ase
f
o
r
s
cr
ee
n
in
g
m
am
m
o
g
r
ap
h
y
(
DDSM)
d
atasets
,
wh
ich
co
n
s
is
ted
o
f
2
,
1
8
8
m
ass
im
ag
es
ex
tr
ac
ted
f
r
o
m
1
,
3
1
9
ca
s
es
b
ef
o
r
e
au
g
m
en
tatio
n
th
en
th
e
n
u
m
b
er
o
f
im
ag
es
r
ea
ch
ed
1
3
,
1
2
8
m
ass
im
ag
es
af
ter
au
g
m
en
tatio
n
,
ar
r
an
g
ed
in
two
m
ain
f
o
ld
er
s
.
T
h
e
b
en
ig
n
f
o
ld
er
co
m
p
r
is
ed
5
,
9
7
0
im
ag
es
an
d
th
e
m
alig
n
an
t
f
o
ld
er
co
m
p
r
is
ed
7
,
1
5
8
im
ag
es.
T
h
is
d
atab
ase
is
n
’
t
b
alan
ce
d
.
So
,
in
th
is
s
tu
d
y
,
a
r
an
d
o
m
ly
b
alan
ce
d
s
u
b
s
et
o
f
th
e
DDSM
d
ataset
was
s
elec
ted
,
co
n
tain
in
g
1
,
6
0
0
m
ass
im
ag
es d
iv
id
ed
in
to
8
0
0
b
en
ig
n
m
ass
im
ag
es a
n
d
8
0
0
m
alig
n
an
t m
ass
im
ag
es
in
a
b
alan
ce
d
m
an
n
er
as
illu
s
tr
ated
in
T
ab
le
2
.
A
s
am
p
le
m
am
m
o
g
r
ap
h
y
m
ass
im
ag
e
with
b
en
ig
n
an
d
m
alig
n
an
t tu
m
o
r
s
is
s
h
o
wn
in
Fig
u
r
e
5
.
T
ab
le
2
.
Nu
m
b
er
o
f
m
am
m
o
g
r
ap
h
y
m
ass
im
ag
es u
s
ed
B
e
n
i
g
n
i
m
a
g
e
s
M
a
l
i
g
n
a
n
t
I
mag
e
s
To
t
a
l
8
0
0
8
0
0
1
,
6
0
0
Fig
u
r
e
5
.
B
en
ig
n
an
d
m
alig
n
a
n
t m
ass
tu
m
o
r
f
r
o
m
DDSM
d
a
taset
4
.
1
.
1
.
Da
t
a
s
pli
t
t
ing
T
h
e
d
ataset
is
s
ep
ar
ated
in
to
r
an
d
o
m
(
tr
ain
in
g
an
d
test
in
g
)
s
ets.
T
h
e
tr
ain
in
g
d
ataset
r
ep
r
e
s
en
ts
7
0
%
an
d
th
e
test
in
g
d
ataset
r
ep
r
esen
ts
3
0
%
o
f
t
h
e
to
tal
d
ataset.
T
ab
le
3
illu
s
tr
ates
th
e
n
u
m
b
e
r
o
f
m
am
m
o
g
r
a
p
h
y
m
ass
im
ag
es u
s
ed
in
th
is
s
tu
d
y
.
T
ab
le
3
.
Nu
m
b
er
o
f
m
am
m
o
g
r
ap
h
y
m
ass
im
ag
es u
s
ed
th
r
o
u
g
h
s
p
litt
in
g
r
atio
(
7
0
:3
0
)
S
p
l
i
t
t
i
n
g
r
a
t
i
o
Tr
a
i
n
e
d
i
ma
g
e
s
Te
st
e
d
i
m
a
g
e
s
To
t
a
l
D
a
t
a
sp
l
i
t
t
i
n
g
(
7
0
:
3
0
)
1
1
2
0
4
8
0
1
,
6
0
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
.
6
,
Decem
b
e
r
20
25
:
5
6
3
3
-
5
6
4
6
5640
4
.
2
.
Sy
s
t
e
m
s
pecif
ica
t
io
ns
T
h
e
ex
p
er
im
en
tal
wo
r
k
is
ca
r
r
ied
o
u
t
u
s
in
g
Py
th
o
n
3
in
th
e
Go
o
g
le
C
o
llab
o
r
ato
r
y
with
o
n
lin
eT
4
GPU
u
s
in
g
6
4
GB
o
f
R
AM
an
d
an
I
n
tel
C
o
r
e
i7
p
r
o
ce
s
s
o
r
ar
e
f
ea
tu
r
es
o
f
th
is
lap
to
p
.
I
n
th
is
ca
s
e,
th
e
m
o
d
el
is
o
p
tim
ized
u
s
in
g
Ad
am
o
p
tim
izer
,
lear
n
in
g
r
ate
is
0
.
0
0
0
1
,
m
in
im
u
m
b
atch
s
ize
is
2
3
,
an
d
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
ep
o
ch
s
is
2
0
.
4
.
3
.
F
ine
-
t
un
ed
t
he
pre
-
t
ra
i
ned (
CNNs)
I
n
o
r
d
er
to
m
ax
im
ize
t
h
e
ad
v
an
tag
es
o
f
tr
a
n
s
f
er
lear
n
i
n
g
,
e
v
er
y
m
o
d
el
u
tili
ze
d
i
n
th
is
in
v
esti
g
atio
n
was
f
ir
s
t
tr
ain
ed
o
n
th
e
I
m
ag
e
Net
d
ataset.
T
o
en
h
a
n
ce
th
eir
ab
ilit
y
to
ad
a
p
t
to
th
e
task
o
f
B
C
clas
s
if
icatio
n
,
ar
ch
itectu
r
al
m
o
d
if
icatio
n
s
wer
e
m
ad
e
b
y
ad
d
i
n
g
s
o
m
e
lay
er
s
s
u
ch
as
F
latten
lay
er
,
Den
s
e
lay
er
,
a
Dr
o
p
o
u
t
lay
er
=
0
.
2
,
an
d
So
f
tm
ax
o
u
tp
u
t
lay
er
.
T
h
ese
m
o
d
if
ied
m
o
d
els
wer
e
th
en
f
in
e
-
tu
n
e
d
o
n
a
d
ataset
co
n
s
is
tin
g
o
f
1
,
6
0
0
im
ag
es
o
v
e
r
2
0
tr
ain
in
g
ep
o
c
h
s
,
with
e
ar
ly
s
to
p
p
in
g
i
m
p
lem
en
ted
u
s
in
g
a
p
atien
ce
v
alu
e
o
f
1
0
to
av
o
id
o
v
er
f
itti
n
g
.
4
.
4
.
Da
t
a
p
re
pro
ce
s
s
ing
No
r
m
aliza
tio
n
,
s
ca
lin
g
,
an
d
au
g
m
en
tatio
n
o
f
th
e
tr
ain
in
g
im
ag
es
ar
e
all
p
ar
ts
o
f
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es.
T
h
e
tech
n
i
q
u
e
o
f
n
o
r
m
aliza
tio
n
in
v
o
lv
es
tr
an
s
f
o
r
m
in
g
p
ix
el
v
alu
es
to
f
all
with
in
a
p
r
e
d
ef
in
ed
r
an
g
e,
b
o
u
n
d
ed
b
etwe
en
0
a
n
d
1
,
to
en
h
a
n
ce
m
o
d
el
g
en
er
aliza
b
ilit
y
an
d
s
tr
ea
m
lin
e
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
.
T
h
en
,
th
e
im
ag
es
ar
e
ad
ju
s
ted
to
a
ty
p
ical
s
ize
o
f
2
2
4
×2
2
4
×
3
p
ix
els
to
co
n
f
ir
m
th
e
r
eq
u
ir
ed
in
p
u
t
s
ize
o
f
all
ap
p
lied
p
r
e
-
tr
ain
ed
m
o
d
els.
T
o
im
p
r
o
v
e
th
e
d
i
v
er
s
ity
o
f
tr
ai
n
in
g
d
ata
a
n
d
b
o
o
s
t
th
e
m
o
d
el
'
s
p
er
f
o
r
m
an
ce
an
d
g
en
er
aliza
tio
n
ab
ilit
y
,
d
ata
au
g
m
en
tatio
n
is
m
eth
o
d
icall
y
ap
p
lied
to
t
h
e
tr
ain
in
g
d
ataset.
T
h
e
ap
p
lied
au
g
m
en
tatio
n
f
u
n
ctio
n
is
an
im
ag
e
d
ata
g
en
e
r
ato
r
f
u
n
ctio
n
in
clu
d
in
g
th
e
f
o
llo
win
g
:
s
h
ea
r
r
an
g
e
=
0
.
1
,
wid
t
h
s
h
if
t r
an
g
e
=
0
.
1
,
h
eig
h
t sh
if
t
r
an
g
e
=
0
.
1
,
r
o
tatio
n
r
a
n
g
e
=
2
0
an
d
z
o
o
m
r
an
g
e
=
0
.
1
.
4
.
5
.
P
er
f
o
r
m
a
nce
ev
a
lua
t
io
n m
ea
s
ures
Per
f
o
r
m
an
ce
e
v
alu
atio
n
m
etr
i
cs
wer
e
co
m
p
u
ted
b
ased
o
n
th
e
n
u
m
b
e
r
s
o
f
tr
u
e
p
o
s
itiv
es
(
T
P),
f
alse
p
o
s
itiv
es (
FP
)
,
tr
u
e
n
eg
ativ
es (
T
N)
,
an
d
f
alse n
eg
ativ
es (
FN)
:
=
+
+
+
+
(7
)
Pr
=
+
(
8
)
=
+
(
9
)
1
−
=
2
⋅
⋅
+
(
1
0
)
An
d
th
e
ar
ea
u
n
d
er
th
e
r
ec
ei
v
er
o
p
er
atin
g
ch
a
r
ac
ter
is
tic
(
R
OC
)
cu
r
v
e
an
d
AUC
wer
e
ca
l
cu
lated
to
ev
alu
ate
th
e
p
er
f
o
r
m
an
ce
o
f
B
C
class
if
i
ca
tio
n
.
4
.
6
.
E
x
perim
ent
a
l
re
s
ults a
nd
dis
cu
s
s
io
n
I
n
th
is
p
ar
t,
th
e
s
u
g
g
ested
B
C
class
if
icatio
n
s
y
s
tem
's
p
er
f
o
r
m
an
ce
is
ass
es
s
ed
u
s
in
g
a
d
ata
s
p
litt
in
g
r
atio
o
f
3
0
%
f
o
r
test
in
g
an
d
7
0
%
f
o
r
tr
ain
in
g
.
T
h
e
s
y
s
tem
'
s
d
esig
n
in
clu
d
es
a
th
r
ee
-
s
tag
e
f
r
am
ewo
r
k
to
en
h
an
ce
s
y
s
tem
ac
cu
r
ac
y
.
E
v
er
y
s
tep
is
e
s
s
en
tial
to
en
h
an
cin
g
th
e
m
o
d
el'
s
ca
p
ac
ity
f
o
r
p
r
ed
ictio
n
u
s
in
g
a
co
m
b
in
atio
n
o
f
co
n
v
en
tio
n
al
m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
le
ar
n
in
g
m
eth
o
d
s
.
I
n
th
e
f
ir
s
t
s
tag
e,
f
iv
e
p
r
e
-
tr
ain
ed
m
o
d
els
—
VGG1
9
,
Mo
b
ileNet
V2
,
Den
s
eNe
t2
0
1
,
R
esNet5
0
,
an
d
I
n
ce
p
tio
n
V3
—
wer
e
tr
ai
n
ed
an
d
e
v
alu
ated
u
s
in
g
d
ata
s
p
litt
in
g
r
atio
7
0
:
3
0
.
T
h
eir
in
d
iv
id
u
al
class
if
icatio
n
p
e
r
f
o
r
m
an
ce
s
ar
e
ev
al
u
ated
,
a
f
ter
th
at
a
m
ajo
r
ity
h
ar
d
v
o
tin
g
m
ec
h
a
n
is
m
is
ap
p
lied
to
co
m
b
in
e
t
h
eir
p
r
ed
ictio
n
s
.
T
h
is
en
s
em
b
le
tech
n
iq
u
e
m
itig
ate
s
th
e
d
is
ad
v
an
tag
es o
f
an
y
o
n
e
C
NN
wh
ile
u
tili
zin
g
th
e
b
en
ef
its
o
f
ea
ch
m
o
d
el
to
in
cr
ea
s
e
o
v
er
all
ac
cu
r
ac
y
.
T
h
e
s
ec
o
n
d
s
tep
is
to
im
p
r
o
v
e
s
y
s
tem
p
er
f
o
r
m
an
ce
b
y
d
eter
m
in
in
g
wh
ich
C
NN
m
o
d
el
p
er
f
o
r
m
s
b
est
in
ter
m
s
o
f
class
if
icatio
n
ac
cu
r
ac
y
.
Sev
en
class
if
ier
s
—
DT
,
R
F,
L
R
,
HGB,
SVM,
GB
,
an
d
XGB
—
ar
e
f
ed
f
ea
tu
r
e
v
ec
to
r
s
th
at
ar
e
tak
en
f
r
o
m
th
e
p
en
u
ltima
te
lay
er
o
f
th
is
b
est
-
p
er
f
o
r
m
in
g
C
NN
m
o
d
el.
E
ac
h
class
if
ier
is
tr
ain
ed
an
d
o
p
tim
ized
to
ass
ess
its
ab
ilit
y
to
g
en
er
alize
an
d
en
h
an
ce
d
etec
tio
n
ac
cu
r
ac
y
.
I
n
th
e
th
ir
d
s
tag
e,
th
e
class
if
icatio
n
r
esu
lts
o
f
th
e
s
ev
en
class
if
ier
s
ar
e
co
m
p
ar
ed
,
an
d
th
e
f
o
u
r
class
if
ier
s
with
th
e
h
ig
h
est
p
er
f
o
r
m
an
ce
ar
e
s
elec
ted
.
A
m
ajo
r
ity
h
ar
d
v
o
tin
g
m
ec
h
an
is
m
is
th
en
ap
p
lied
to
th
e
o
u
tp
u
ts
o
f
th
ese
f
o
u
r
class
if
ier
s
,
tr
ea
tin
g
th
em
as
a
d
ec
is
io
n
-
m
ak
in
g
co
m
m
ittee.
T
h
is
en
s
em
b
le
s
tr
ateg
y
is
d
esig
n
ed
to
co
r
r
ec
t
in
d
iv
id
u
al
class
if
ier
er
r
o
r
s
an
d
im
p
r
o
v
e
th
e
f
in
al
class
if
icatio
n
o
u
tco
m
e.
T
h
e
in
teg
r
atio
n
o
f
th
is
v
o
tin
g
m
ec
h
an
is
m
d
em
o
n
s
tr
ates
a
s
ig
n
if
ican
t
en
h
an
ce
m
en
t
in
th
e
r
o
b
u
s
tn
ess
an
d
ac
cu
r
ac
y
o
f
th
e
o
v
er
all
s
y
s
tem
.
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
B
r
ea
s
t c
a
n
ce
r
d
etec
tio
n
u
s
in
g
en
s
emb
le
meth
o
d
s
(
A
la
a
Mo
h
a
med
Gh
a
z
y
)
5641
4
.
6
.
1
.
E
x
perim
ent
a
l
re
s
ults o
f
pro
po
s
ed
m
o
del
wh
ile
im
p
lem
en
tin
g
th
e
s
u
g
g
ested
class
if
icatio
n
s
y
s
tem
3
0
%
o
f
th
e
d
ata
was
u
s
ed
f
o
r
t
esti
n
g
an
d
7
0
%
f
o
r
tr
ain
in
g
.
T
h
e
e
x
p
er
i
m
en
tal
r
esu
lts
ar
e
d
is
cu
s
s
ed
in
d
etail
f
o
r
ea
ch
o
f
th
e
th
r
e
e
m
ain
s
tag
es
th
at
co
m
p
r
is
e
th
e
f
r
am
ewo
r
k
.
E
a
ch
s
tag
e
co
n
tr
ib
u
tes
to
r
ef
in
in
g
th
e
class
if
icatio
n
p
r
o
ce
s
s
an
d
co
llectiv
ely
d
em
o
n
s
tr
ates th
e
ef
f
ec
tiv
e
n
ess
o
f
th
e
s
u
g
g
ested
h
y
b
r
id
e
n
s
em
b
le
ap
p
r
o
ac
h
in
B
C
d
etec
tio
n
.
a.
T
h
e
ex
p
er
im
en
tal
r
esu
lt o
f
th
e
f
ir
s
t stag
e
R
esu
lts
f
r
o
m
th
e
f
ir
s
t
s
tep
o
f
th
e
ex
p
er
im
en
t
ar
e
p
r
esen
ted
in
T
ab
le
4
,
wh
er
e
th
e
f
iv
e
p
r
e
-
tr
ain
ed
C
NN
m
o
d
els'
class
if
icatio
n
p
er
f
o
r
m
an
ce
is
h
ig
h
lig
h
ted
.
T
h
ese
m
o
d
els
—
VGG1
9
,
Den
s
eNe
t2
0
1
,
R
esNet5
0
,
Mo
b
ileNetV2
,
an
d
I
n
ce
p
tio
n
V3
—
wer
e
ev
alu
ated
in
d
iv
id
u
ally
.
I
n
ad
d
itio
n
to
th
e
in
d
iv
id
u
al
p
er
f
o
r
m
an
ce
s
,
T
ab
le
4
also
r
ep
o
r
ts
th
e
o
v
er
all
class
if
icatio
n
ac
cu
r
ac
y
ac
h
iev
ed
wh
en
th
e
m
ajo
r
ity
h
ar
d
v
o
tin
g
m
ec
h
an
is
m
is
ap
p
lied
to
th
e
co
m
b
in
ed
o
u
tp
u
ts
o
f
th
ese
m
o
d
els.
T
h
e
s
tr
en
g
th
s
o
f
ea
ch
C
NN
ar
e
u
s
ed
in
th
is
en
s
em
b
le
tech
n
iq
u
e
to
g
et
a
p
r
ed
ictio
n
r
esu
lt th
at
is
m
o
r
e
r
eliab
le
an
d
ac
cu
r
ate.
T
h
e
R
esNet5
0
m
o
d
el
p
r
o
d
u
ce
d
th
e
b
est
class
if
icatio
n
r
esu
lt
s
am
o
n
g
th
e
f
iv
e
p
r
e
-
t
r
ain
ed
C
NNs,
a
s
in
d
icate
d
in
T
ab
le
4
.
I
t
r
ec
o
r
d
ed
an
ac
cu
r
ac
y
an
d
p
r
ec
is
io
n
o
f
9
9
.
5
8
%,
alo
n
g
with
an
F1
-
s
co
r
e
an
d
r
ec
all
o
f
9
9
.
5
8
%,
an
d
R
OC
-
AUC
v
al
u
e
o
f
9
9
.
9
8
%.
Fu
r
th
er
m
o
r
e,
wh
en
th
e
m
ajo
r
ity
h
ar
d
v
o
ti
n
g
m
ec
h
an
is
m
was
ap
p
lied
to
th
e
o
u
tp
u
ts
o
f
all
f
iv
e
C
NN
m
o
d
els,
th
e
s
y
s
tem
'
s
o
v
er
all
class
if
ica
tio
n
ac
cu
r
ac
y
im
p
r
o
v
e
d
to
9
9
.
8
5
%,
s
h
o
win
g
h
o
w
e
n
s
em
b
le
lear
n
in
g
ca
n
im
p
r
o
v
e
d
iag
n
o
s
tic
r
eliab
ilit
y
.
Fig
u
r
es
6
to
1
0
illu
s
tr
ate
t
h
e
f
i
v
e
d
ee
p
lear
n
in
g
m
o
d
els'
ac
cu
r
ac
y
an
d
l
o
s
s
d
u
r
in
g
t
h
e
tr
ain
in
g
an
d
test
in
g
s
tep
s
o
v
er
twen
t
y
ep
o
ch
s
.
T
ab
le
4
.
T
h
e
f
iv
e
p
r
e
-
tr
ain
e
d
m
o
d
els'
p
er
f
o
r
m
an
ce
m
etr
ics an
d
th
e
m
aj
o
r
ity
h
a
r
d
v
o
te'
s
cla
s
s
if
icatio
n
ac
cu
r
ac
y
o
n
th
e
m
o
d
els'
o
u
tp
u
t
Tr
a
i
n
e
d
m
o
d
e
l
s
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
S
c
o
r
e
(
%)
R
o
c
-
A
U
C
(
%)
Ti
me
(
S
e
c
)
V
G
G
1
9
9
9
.
5
7
9
9
.
1
7
1
0
0
9
9
.
5
8
9
9
.
9
9
1
3
5
0
.
2
9
D
e
n
s
e
N
e
t
9
7
.
5
0
9
8
.
3
0
9
6
.
6
6
9
7
.
4
7
9
9
.
7
1
4
8
4
.
7
2
R
e
sN
e
t
5
0
9
9
.
5
8
9
9
.
5
8
9
9
.
5
8
9
9
.
5
8
9
9
.
9
8
4
3
8
.
1
8
M
o
b
i
l
e
N
e
t
9
5
.
6
2
9
4
.
3
3
9
7
.
0
8
9
5
.
6
8
9
9
.
1
2
3
3
3
.
6
5
I
n
c
e
p
t
i
o
n
N
e
t
9
3
.
3
3
9
3
.
3
3
9
3
.
3
3
9
3
.
3
3
9
7
.
7
5
4
4
4
.
3
4
M
a
j
o
r
i
t
y
h
a
r
d
v
o
t
i
n
g
b
a
se
d
o
n
t
h
e
p
r
e
v
i
o
u
s
l
y
t
r
a
i
n
e
d
m
o
d
e
l
s
I
n
c
r
e
a
se
t
h
e
s
y
st
e
m's
c
l
a
ssi
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
t
o
9
9
.
8
5
%.
Fig
u
r
e
6
.
Acc
u
r
a
cy
an
d
lo
s
s
d
u
r
in
g
tr
ai
n
in
g
a
n
d
test
in
g
u
s
in
g
VGG
1
9
p
r
etr
ain
ed
C
NN
Fig
u
r
e
7
.
Acc
u
r
ac
y
a
n
d
lo
s
s
d
u
r
in
g
tr
ai
n
in
g
a
n
d
test
in
g
u
s
in
g
Den
s
e
Net
p
r
etr
ain
ed
C
NN
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
.
6
,
Decem
b
e
r
20
25
:
5
6
3
3
-
5
6
4
6
5642
Fig
u
r
e
8
.
Acc
u
r
ac
y
a
n
d
lo
s
s
d
u
r
in
g
tr
ai
n
in
g
a
n
d
test
in
g
u
s
in
g
R
esNet5
0
p
r
etr
ain
ed
C
NN
Fig
u
r
e
9
.
Acc
u
r
ac
y
a
n
d
lo
s
s
d
u
r
in
g
tr
ai
n
in
g
a
n
d
test
in
g
u
s
in
g
Mo
b
ileNet
p
r
etr
ain
e
d
C
NN
Fig
u
r
e
1
0
.
Acc
u
r
ac
y
an
d
l
o
s
s
d
u
r
in
g
tr
ain
in
g
an
d
test
in
g
u
s
in
g
I
n
ce
p
tio
n
Net
p
r
etr
ain
ed
C
NN
Fig
u
r
es
6
to
10
s
h
o
w
co
m
p
ar
is
o
n
b
etwe
en
(
T
r
ain
in
g
-
T
esti
n
g
)
ac
cu
r
ac
y
an
d
l
o
s
s
f
o
r
th
e
f
iv
e
d
ee
p
lear
n
in
g
m
o
d
els
o
v
er
twen
ty
ep
o
ch
s
.
Fig
u
r
e
s
11
to
1
5
o
u
tlin
e
th
e
co
n
f
u
s
io
n
m
atr
ic
es
ac
q
u
ir
ed
d
u
r
in
g
ex
p
er
im
en
ts
,
wh
er
e
C
lass
es
"
0
"
an
d
"1
"
d
en
o
te
"b
e
n
ig
n
"
an
d
"m
alig
n
a
n
t,"
r
esp
ec
tiv
ely
.
Fig
u
r
es
11
to
15
s
h
o
w
co
m
p
ar
is
o
n
b
etwe
en
th
e
C
o
n
f
u
s
io
n
m
atr
ices
th
r
o
u
g
h
d
if
f
er
en
t
f
iv
e
d
ee
p
lear
n
in
g
m
o
d
els
in
class
if
y
in
g
0
an
d
1
wh
ich
r
ef
er
to
B
en
ig
n
an
d
Ma
lig
n
an
t t
u
m
o
r
in
a
test
in
g
d
ataset,
r
esp
ec
tiv
ely
.
b.
T
h
e
ex
p
er
im
en
tal
r
esu
lt o
f
th
e
s
ec
o
n
d
s
tag
e
T
h
e
r
esu
lts
o
f
th
e
s
ec
o
n
d
s
tag
e'
s
ex
p
er
im
en
t
ar
e
s
h
o
wn
in
T
ab
le
5
.
wh
er
e
a
p
r
elim
in
ar
y
an
aly
s
is
was
co
n
d
u
cted
to
co
m
p
ar
e
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
th
e
f
iv
e
p
r
o
p
o
s
ed
p
r
e
-
tr
ain
ed
m
o
d
els wh
ich
led
to
th
e
R
esNet
5
0
ac
h
iev
in
g
th
e
h
ig
h
est
class
if
icatio
n
r
esu
lt.
T
h
en
,
in
th
e
s
ec
o
n
d
s
tag
e
o
f
th
e
s
u
g
g
ested
s
ch
em
e,
s
ev
en
class
if
ier
s
(
DT
,
HGB,
SVM,
R
F,
L
R
,
GB
,
an
d
XGB)
ar
e
ap
p
lied
to
th
e
f
ea
tu
r
e
v
ec
to
r
ex
tr
ac
ted
f
r
o
m
R
es
N
et
5
0
p
r
e
-
tr
ain
ed
m
o
d
el,
an
d
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
th
ese
m
ac
h
in
e
lear
n
in
g
class
if
ier
s
is
illu
s
tr
ated
in
T
ab
le
5.
Evaluation Warning : The document was created with Spire.PDF for Python.