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
.
2
,
A
p
r
il
20
25
,
p
p
.
1
6
7
0
~
1
6
8
5
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
2
.
pp
1
6
7
0
-
1
6
8
5
1670
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
A uniqu
e Y
O
L
O
-
ba
sed
g
a
ted
a
tt
e
n
tion deep
conv
o
lution
network
-
Li
chte
n
berg
optimiza
tion
alg
o
rithm
mo
del
for a
precise b
reas
t
ca
n
cer
seg
menta
tion
a
nd clas
sifica
tion
Vino
t
h Ra
t
hin
a
m
1
,
Sa
s
irek
a
Ra
j
endra
n
2
,
Va
la
rm
a
t
hi K
ri
s
hn
a
s
a
m
y
1
1
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
En
g
i
n
e
e
r
i
n
g
,
P
.
S
.
R
.
E
n
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
,
T
a
mi
l
n
a
d
u
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
B
i
o
t
e
c
h
n
o
l
o
g
y
,
M
e
p
c
o
S
c
h
l
e
n
k
En
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
,
Ta
m
i
l
n
a
d
u
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 1
5
,
2
0
2
4
R
ev
is
ed
No
v
3
0
,
2
0
2
4
Acc
ep
ted
Dec
1
4
,
2
0
2
4
A
n
o
v
e
l
y
o
u
o
n
l
y
l
o
o
k
o
n
c
e
(YO
LO)
-
b
a
se
d
g
a
ted
a
tt
e
n
ti
o
n
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
n
e
two
rk
(G
AD
CN)
c
las
sifica
ti
o
n
a
lg
o
rit
h
m
is
d
e
v
e
lo
p
e
d
a
n
d
u
ti
li
z
e
d
in
th
is
p
re
se
n
t
stu
d
y
fo
r
th
e
d
e
tec
ti
o
n
o
f
b
re
a
st
c
a
n
c
e
r.
In
th
is
fra
m
e
wo
rk
,
c
o
n
tras
t
e
n
h
a
n
c
e
m
e
n
t
-
b
a
se
d
h
isto
g
ra
m
e
q
u
a
li
z
a
ti
o
n
is
a
p
p
l
ied
in
it
ially
to
p
r
o
d
u
c
e
th
e
n
o
rm
a
li
z
e
d
b
re
a
st i
m
a
g
e
with
re
d
u
c
e
d
n
o
ise
a
rti
fa
c
ts.
Th
e
n
,
t
h
e
b
re
a
st r
e
g
io
n
is
a
c
c
u
ra
tely
se
g
m
e
n
ted
fr
o
m
th
e
p
re
p
r
o
c
e
ss
e
d
ima
g
e
s
with
l
o
w
c
o
m
p
lex
it
y
a
n
d
se
g
m
e
n
tatio
n
e
rro
r
u
sin
g
th
e
YO
LO
-
b
a
se
d
a
tt
e
n
ti
o
n
n
e
two
r
k
m
o
d
e
l.
T
o
d
iag
n
o
se
b
re
a
st
c
a
n
c
e
r
with
b
e
tt
e
r
a
c
c
u
ra
c
y
,
th
e
G
AD
CN
m
o
d
e
l
is
u
se
d
t
o
p
re
d
ict
t
h
e
e
x
a
c
t
c
las
s
o
f
ima
g
e
(
i.
e
.
,
b
e
n
ig
n
o
r
m
a
li
g
n
a
n
t
).
Du
rin
g
c
las
sifica
ti
o
n
,
t
h
e
a
c
ti
v
a
ti
o
n
f
u
n
c
ti
o
n
is
o
p
ti
m
a
ll
y
c
o
m
p
u
ted
wi
th
th
e
u
se
o
f
th
e
Li
c
h
ten
b
e
rg
o
p
ti
m
iza
ti
o
n
a
lg
o
rit
h
m
(LOA).
I
t
a
id
s
in
a
c
h
iev
in
g
imp
ro
v
e
d
c
las
sifica
ti
o
n
p
e
rfo
rm
a
n
c
e
with
li
tt
le
c
o
m
p
le
x
it
y
in
train
in
g
a
n
d
a
ss
e
ss
m
e
n
t.
Th
e
sig
n
ifi
c
a
n
c
e
o
f
th
e
p
re
se
n
t
stu
d
y
in
c
lu
d
e
s
th
e
u
se
o
f
a
u
n
i
q
u
e
,
YO
LO
-
b
a
se
d
G
AD
CN
-
LOA
m
o
d
e
l
th
a
t
h
e
lp
s
i
n
t
h
e
p
r
e
d
ictio
n
o
f
b
re
a
st
c
a
n
c
e
r
with
h
i
g
h
e
r
a
c
c
u
ra
c
y
.
It
wa
s
o
b
se
rv
e
d
th
a
t
th
e
m
o
d
e
l
e
x
h
i
b
it
e
d
9
9
%
a
c
c
u
ra
c
y
f
o
r
t
h
e
d
a
tas
e
ts
u
ti
li
z
e
d
.
In
a
d
d
it
i
o
n
,
t
h
e
se
lec
ted
m
o
d
e
l
o
u
t
p
e
rfo
rm
s
we
ll
with
se
n
siti
v
it
y
,
sp
e
c
ifi
c
it
y
,
p
re
c
isi
o
n
,
a
n
d
F
1
-
sc
o
re
.
He
n
c
e
th
e
p
ro
p
o
se
d
m
o
d
e
l
c
o
u
ld
b
e
e
x
p
lo
it
e
d
fo
r
t
h
e
d
iag
n
o
sis
o
f
b
re
a
st
c
a
n
c
e
r
a
t
a
n
e
a
rly
sta
g
e
t
o
e
n
a
b
le
p
re
v
e
n
t
iv
e
c
a
re
.
K
ey
w
o
r
d
s
:
B
r
ea
s
t c
an
ce
r
d
etec
tio
n
C
o
n
tr
ast
lim
ited
ad
ap
tiv
e
h
is
to
g
r
am
eq
u
aliza
tio
n
Dee
p
lear
n
in
g
,
an
d
c
o
m
p
u
ted
aid
ed
d
iag
n
o
s
is
Gate
d
atten
tio
n
d
ee
p
co
n
v
o
l
u
tio
n
n
etwo
r
k
class
if
icatio
n
L
ich
ten
b
er
g
o
p
tim
izatio
n
alg
o
r
ith
m
YOL
O
-
b
ased
atten
tio
n
n
etwo
r
k
s
eg
m
en
tatio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Vin
o
th
R
ath
in
am
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
i
n
g
,
P.S.R.
E
n
g
in
ee
r
in
g
C
o
lleg
e
Siv
ak
asi,
T
am
iln
ad
u
,
I
n
d
ia
E
m
ail:
v
in
o
th
@
p
s
r
.
ed
u
.
i
n
1.
I
NT
RO
D
UCT
I
O
N
Acc
o
r
d
in
g
to
th
e
W
o
r
ld
Hea
lt
h
Or
g
an
izatio
n
(
W
HO)
r
e
p
o
r
t,
b
r
ea
s
t
ca
n
ce
r
[
1
]
,
[
2
]
is
r
eg
ar
d
ed
as
th
e
s
ec
o
n
d
b
i
g
g
est
ca
u
s
e
o
f
m
o
r
b
id
ity
f
o
r
wo
m
e
n
,
a
n
d
n
ea
r
ly
8
.
2
m
illi
o
n
p
eo
p
le
d
ie
ea
ch
y
e
ar
f
r
o
m
ca
n
ce
r
a
n
d
p
r
ed
icts
th
e
s
tatis
tic
will
g
r
o
w
to
2
7
m
illi
o
n
b
y
2
0
3
0
[
3
]
.
T
h
er
ef
o
r
e,
ea
r
ly
d
iag
n
o
s
is
,
tim
ely
an
d
ac
cu
r
ate
d
etec
tio
n
,
an
d
p
r
o
ac
tiv
e
p
r
e
v
e
n
tio
n
ar
e
ess
en
tial
elem
en
ts
i
n
lo
wer
in
g
th
e
m
o
r
tality
r
ate
f
o
r
wo
m
en
[
4
]
,
[
5
]
.
Fu
r
th
er
,
it
is
ess
en
tial
to
p
r
e
d
ict
th
e
d
is
ea
s
e
at
an
ea
r
ly
s
tag
e
s
o
th
at
th
e
tr
ea
tm
en
t
ca
n
b
e
g
iv
e
n
well
i
n
ad
v
an
ce
.
Fo
r
lo
ca
tin
g
an
d
p
r
e
cisely
id
en
tify
in
g
th
e
tu
m
o
r
-
a
f
f
ec
ted
ar
ea
,
m
an
y
im
ag
in
g
m
o
d
alities
ar
e
u
s
ed
.
Me
d
ical
p
r
o
f
ess
io
n
als
o
f
ten
u
s
e
m
am
m
o
g
r
ap
h
y
im
ag
es
in
tan
d
em
with
o
th
er
im
ag
in
g
m
eth
o
d
s
to
d
iag
n
o
s
e
an
d
tr
ea
t
p
atien
ts
with
ac
cu
r
a
cy
[
6
]
–
[
8
]
.
Var
i
o
u
s
m
ed
ical
i
m
ag
e
p
r
o
ce
s
s
in
g
tech
n
i
q
u
es
h
av
e
b
ee
n
em
p
l
o
y
e
d
in
th
e
e
x
is
tin
g
r
esear
ch
with
t
h
e
in
ten
t
o
f
d
etec
tin
g
b
r
ea
s
t
c
an
ce
r
s
f
r
o
m
m
am
m
o
g
r
am
s
[
9
]
,
[
1
0
]
.
C
o
m
p
u
ter
-
aid
ed
d
iag
n
o
s
is
(
C
AD)
[
1
0
]
,
[
1
1
]
s
y
s
tem
s
r
eq
u
ir
e
th
e
d
etec
tio
n
,
s
eg
m
en
tatio
n
,
an
d
class
if
icatio
n
o
f
m
ed
ica
l
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
A
u
n
iq
u
e
YOLO
-
b
a
s
ed
g
a
ted
a
tten
tio
n
d
ee
p
co
n
v
o
lu
tio
n
n
et
w
o
r
k
-
Lich
ten
b
erg
…
(
V
in
o
th
R
a
th
in
a
m
)
1671
im
ag
es
to
ass
is
t
r
ad
io
lo
g
is
ts
in
ex
ac
tly
lo
ca
tin
g
th
e
ca
n
ce
r
-
af
f
ec
ted
b
r
ea
s
t
r
eg
io
n
.
As
a
r
esu
lt,
a
p
r
ec
is
e
m
eth
o
d
o
f
b
r
ea
s
t
ca
n
ce
r
i
d
e
n
tific
atio
n
an
d
ca
teg
o
r
izatio
n
is
s
u
g
g
ested
f
o
r
m
am
m
o
g
r
a
p
h
y
s
cr
ee
n
in
g
.
An
au
to
m
ated
C
AD
-
b
ased
d
iag
n
o
s
is
to
o
l
[
2
]
,
[
1
2
]
,
[
1
3
]
h
as
r
ec
en
tly
b
ee
n
u
s
ed
in
ce
r
tain
ex
is
tin
g
s
tu
d
ies
with
d
ee
p
lear
n
in
g
tech
n
i
q
u
es.
Dee
p
lear
n
in
g
is
th
e
m
o
s
t
p
r
ac
tical
tech
n
iq
u
e
wid
ely
u
s
ed
f
o
r
id
e
n
tify
in
g
ab
n
o
r
m
al
ities
f
r
o
m
m
ed
ical
im
ag
es
in
th
e
r
ea
l
wo
r
ld
.
Mo
r
e
o
v
er
,
th
e
s
tan
d
a
r
d
im
ag
e
p
r
o
ce
s
s
in
g
s
tag
es
ar
e
in
v
o
lv
e
d
in
th
e
au
to
m
ated
s
y
s
tem
th
at
e
n
ab
les
th
e
d
ee
p
lear
n
in
g
m
o
d
el
to
p
e
r
f
o
r
m
well.
Su
b
s
tan
tial h
ier
ar
c
h
ical
f
ea
tu
r
e
m
ap
s
ca
n
b
e
p
r
o
d
u
ce
d
b
y
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
f
r
o
m
th
e
id
e
n
tical
m
am
m
o
g
r
a
m
b
r
ea
s
t
im
ag
e.
An
am
p
le
tr
u
e
p
o
s
itiv
e
r
ate
an
d
d
is
ea
s
e
d
iag
n
o
s
is
r
esu
lts
ar
e
d
ep
en
d
en
t
u
p
o
n
th
e
ac
cu
r
ate
d
iag
n
o
s
is
o
f
s
u
s
p
icio
u
s
b
r
ea
s
t
lesi
o
n
s
[
1
4
]
–
[
1
6
]
,
wh
ic
h
co
m
e
in
a
wid
e
v
ar
iety
o
f
s
h
a
p
es,
tex
tu
r
es,
an
d
p
o
s
itio
n
s
th
at
m
ak
e
th
e
d
etec
tio
n
ch
allen
g
in
g
.
Mo
r
e
s
p
ec
if
ically
,
a
n
u
m
b
er
o
f
s
cien
tis
ts
h
av
e
u
s
ed
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
m
o
d
els,
in
clu
d
in
g
I
n
ce
p
tio
n
R
esNet
-
V2
,
I
n
ce
p
tio
n
-
V
3
,
VGG1
6
an
d
1
9
,
Go
o
g
leNe
t,
R
esNet
-
1
8
,
5
0
,
an
d
1
0
1
,
to
d
iag
n
o
s
e
b
r
ea
s
t
ca
n
ce
r
.
Si
m
ilar
to
th
is
,
th
e
ap
p
licatio
n
o
f
ar
tific
ial
in
tellig
en
ce
h
as
b
ee
n
cr
u
cial
in
th
e
d
iag
n
o
s
is
o
f
b
r
ea
s
t tu
m
o
r
s
th
r
o
u
g
h
th
e
u
s
e
o
f
s
ev
er
al
m
ac
h
i
n
e
lear
n
in
g
alg
o
r
ith
m
s
.
Alth
o
u
g
h
v
a
r
io
u
s
m
eth
o
d
s
tr
y
to
f
o
r
ec
ast
th
e
t
u
m
o
r
an
d
ca
teg
o
r
ize
it
as
b
e
n
ig
n
o
r
m
al
ig
n
an
t,
th
e
cu
r
r
en
t
ap
p
r
o
ac
h
es
h
av
e
a
n
u
m
b
er
o
f
s
h
o
r
t
c
o
m
i
n
g
s
.
Usi
n
g
wi
d
el
y
a
cc
ess
ib
le
b
r
e
ast im
ag
e
d
a
tase
ts
,
t
h
e
c
u
r
r
e
n
t
s
t
u
d
y
aim
s
t
o
cr
ea
te
a
n
o
v
el
y
o
u
o
n
ly
lo
o
k
o
n
ce
(
Y
OL
O
)
-
b
ase
d
g
a
ted
at
te
n
ti
o
n
d
e
e
p
c
o
n
v
o
lu
ti
o
n
n
etw
o
r
k
(
GADCN
)
-
L
i
c
h
t
en
b
e
r
g
o
p
ti
m
iz
ati
o
n
alg
o
r
it
h
m
(
L
OA
)
m
o
d
el
f
o
r
t
h
e
p
r
e
d
ic
ti
o
n
o
f
b
r
ea
s
t
c
a
n
ce
r
i
n
o
r
d
e
r
to
g
et
b
e
y
o
n
d
t
h
e
r
est
r
ic
tio
n
s
.
T
h
e
g
o
al
o
f
t
h
e
p
r
o
p
o
s
ed
r
esear
ch
p
r
o
ject
is
to
c
r
ea
te
a
n
e
w
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
m
o
d
el
b
ased
o
n
d
ee
p
lear
n
in
g
f
o
r
p
r
ec
is
e
b
r
ea
s
t
ca
n
ce
r
d
etec
tio
n
.
B
elo
w
is
th
e
s
u
g
g
ested
f
r
am
e
wo
r
k
f
o
r
th
e
cu
r
r
en
t
in
v
esti
g
atio
n
:
a.
T
o
cr
ea
te
th
e
n
o
r
m
alize
d
im
ag
e,
a
co
n
tr
ast
lim
ited
ad
ap
t
iv
e
h
is
to
g
r
am
eq
u
aliza
tio
n
(
C
L
AHE
)
-
b
ased
im
ag
e
p
r
e
p
r
o
ce
s
s
in
g
m
o
d
el
h
as
b
ee
n
u
s
ed
,
wh
ich
ca
r
r
i
es
o
u
t
h
is
to
g
r
am
e
q
u
aliza
tio
n
an
d
co
n
tr
ast
en
h
an
ce
m
e
n
t p
r
o
ce
d
u
r
es.
b.
T
h
e
YOL
O
-
b
ased
atten
tio
n
n
etwo
r
k
m
o
d
el
is
im
p
lem
en
ted
to
ac
cu
r
ately
s
eg
m
en
t
th
e
b
r
ea
s
t
r
eg
io
n
f
r
o
m
th
e
p
r
ep
r
o
ce
s
s
ed
im
ag
es with
lo
w
co
m
p
lex
ity
a
n
d
s
eg
m
e
n
tatio
n
er
r
o
r
.
c.
T
h
e
GADCN
m
o
d
el
is
ap
p
li
ed
to
p
r
ed
ict
th
e
ac
cu
r
ate
cla
s
s
o
f
im
ag
e
(
i.e
.
,
b
e
n
ig
n
o
r
m
alig
n
an
t)
f
o
r
a
p
r
o
p
er
ab
n
o
r
m
ality
id
e
n
tific
atio
n
f
r
o
m
th
e
b
r
ea
s
t im
ag
e.
d.
T
o
d
eter
m
i
n
e
th
e
id
ea
l
v
al
u
e
f
o
r
ca
lcu
latin
g
th
e
ac
tiv
atio
n
f
u
n
ctio
n
u
tili
ze
d
in
th
e
GAD
C
N
m
o
d
el,
th
e
L
OA
is
u
tili
ze
d
.
L
o
w
tr
ain
in
g
an
d
test
in
g
co
m
p
lex
ity
co
n
tr
ib
u
tes to
b
etter
ca
teg
o
r
izatio
n
p
e
r
f
o
r
m
a
n
ce
.
e.
T
h
is
s
tu
d
y
u
s
es
a
n
u
m
b
er
o
f
co
m
m
o
n
an
d
well
-
k
n
o
wn
b
r
e
ast
im
ag
e
d
ataset
s
f
o
r
s
y
s
tem
im
p
lem
en
tatio
n
an
d
p
er
f
o
r
m
an
ce
v
alid
atio
n
,
in
clu
d
i
n
g
th
e
d
ig
ital
d
ata
b
ase
f
o
r
s
cr
ee
n
in
g
m
am
m
o
g
r
ap
h
y
(
DDSM)
,
m
am
m
o
g
r
a
p
h
ic
im
ag
e
an
aly
s
i
s
s
o
ciety
(
MI
AS)
,
I
NB
r
ea
s
t,
W
is
co
n
s
in
d
iag
n
o
s
tic
b
r
ea
s
t
ca
n
ce
r
(
W
DB
C
)
,
an
d
W
is
co
n
s
in
b
r
ea
s
t c
an
ce
r
d
ataset
(
W
B
C
D)
.
Her
e,
an
ex
h
a
u
s
tiv
e
liter
atu
r
e
r
ev
iew
is
ca
r
r
ied
o
u
t
to
ex
a
m
in
e
v
ar
io
u
s
m
eth
o
d
o
lo
g
ies
u
s
ed
in
th
e
b
io
-
m
ed
ical
f
iel
d
f
o
r
th
e
i
d
en
tific
atio
n
o
f
a
b
n
o
r
m
alities
f
r
o
m
th
e
b
r
ea
s
t
im
a
g
e.
T
h
r
o
u
g
h
th
eir
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
p
r
o
ce
d
u
r
es,
it
in
v
esti
g
ates
th
e
ad
v
an
tag
es
an
d
d
if
f
icu
lties
ass
o
ciate
d
with
th
e
cu
r
r
en
t
m
o
d
els.
Ham
ed
et
a
l.
[
1
7
]
i
n
v
esti
g
ated
th
e
r
ec
en
t
lear
n
in
g
-
b
ased
class
if
icatio
n
ap
p
r
o
a
ch
es
f
o
r
p
r
e
d
ictin
g
ab
n
o
r
m
alities
f
r
o
m
th
e
b
r
ea
s
t
im
ag
e.
T
h
e
au
th
o
r
s
in
ten
d
to
id
en
tify
th
e
p
r
e
m
atu
r
e
s
ig
n
s
o
f
b
r
ea
s
t
ca
n
ce
r
f
r
o
m
th
e
m
am
m
o
g
r
am
im
ag
es
b
y
d
ev
elo
p
in
g
a
C
AD
m
o
d
el.
Kh
an
et
a
l.
[
1
8
]
u
tili
ze
d
a
tr
an
s
f
er
lear
n
in
g
m
ec
h
an
is
m
f
o
r
a
n
ac
cu
r
ate
d
eter
m
in
atio
n
an
d
ca
teg
o
r
izatio
n
o
f
b
r
ea
s
t
ab
n
o
r
m
alities
f
r
o
m
m
am
m
o
g
r
a
m
im
ag
es.
I
n
th
is
f
r
am
ewo
r
k
,
a
co
m
b
in
atio
n
o
f
m
u
ltip
le
C
NN
ar
ch
itectu
r
es
is
em
p
lo
y
ed
to
o
b
tain
f
ast
an
d
ac
cu
r
ate
d
etec
tio
n
r
esu
lts
.
So
u
lam
i
et
a
l.
[
1
9
]
im
p
lem
en
te
d
a
UNe
t
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
f
o
r
d
ev
el
o
p
in
g
an
au
to
m
ated
b
r
ea
s
t
ca
n
ce
r
d
etec
tio
n
s
y
s
tem
.
Her
e,
th
e
p
i
x
el
-
to
-
p
ix
el
class
if
icatio
n
is
p
er
f
o
r
m
e
d
to
o
b
tain
ac
cu
r
ate
d
etec
tio
n
r
esu
lts
.
J
ab
ee
n
et
a
l.
[
2
0
]
d
ep
lo
y
e
d
a
p
r
o
b
ab
ilit
y
-
b
ased
d
ee
p
lear
n
in
g
a
r
ch
itectu
r
e
m
o
d
el
f
o
r
th
e
id
en
tific
atio
n
o
f
ab
n
o
r
m
alities
f
r
o
m
th
e
u
ltra
s
o
u
n
d
b
r
ea
s
t
im
ag
es.
T
h
is
f
r
am
ewo
r
k
in
clu
d
e
s
th
e
o
p
er
atio
n
s
o
f
d
ata
au
g
m
e
n
tatio
n
,
p
r
e
-
tr
ain
i
n
g
,
f
ea
t
u
r
e
ex
t
r
ac
tio
n
,
o
p
tim
izatio
n
,
an
d
p
r
e
d
ictio
n
.
Her
e
,
th
e
p
r
o
b
a
b
ilis
tic
m
eth
o
d
is
u
tili
ze
d
to
m
er
g
e
th
e
b
est
-
ch
o
s
en
ch
ar
ac
ter
is
tics
.
Sh
ah
id
i
et
a
l.
[
2
1
]
p
r
esen
ted
a
co
m
p
ar
is
o
n
s
tu
d
y
to
ex
am
in
e
v
a
r
io
u
s
class
if
icatio
n
m
eth
o
d
o
lo
g
ies
u
s
ed
in
t
h
e
f
ield
o
f
m
e
d
ical
im
ag
in
g
.
T
h
is
s
tu
d
y
s
et
o
u
t
to
s
h
o
w
h
o
w
d
ee
p
-
lear
n
in
g
tec
h
n
iq
u
es
co
u
l
d
b
e
u
s
ed
to
ca
te
g
o
r
ize
h
is
to
lo
g
ical
im
ag
es
o
f
b
r
ea
s
t
ca
n
ce
r
.
T
h
e
d
if
f
icu
lties
in
class
if
y
in
g
b
r
ea
s
t
ca
n
ce
r
p
ath
o
lo
g
y
im
ag
es
w
er
e
n
o
ted
,
an
d
p
o
s
s
ib
le
s
o
lu
tio
n
s
wer
e
c
o
n
s
id
er
ed
in
th
is
s
tu
d
y
.
Su
n
n
y
et
a
l.
[
2
2
]
ca
r
r
ied
o
u
t
a
co
m
p
a
r
is
o
n
s
tu
d
y
t
o
v
er
if
y
th
e
ef
f
ec
tiv
e
n
ess
o
f
t
h
e
c
o
m
m
o
n
m
ac
h
i
n
e
lear
n
in
g
class
if
ier
s
.
T
o
u
s
e
th
e
class
if
icatio
n
alg
o
r
ith
m
s
,
th
e
d
ataset
was
s
p
lit
in
to
tr
ain
in
g
an
d
test
in
g
p
h
ases
.
T
h
e
m
eth
o
d
th
at
y
ield
s
th
e
b
e
s
t
r
esu
lts
will
b
e
u
s
ed
o
n
th
e
web
s
ite
'
s
b
ac
k
en
d
,
an
d
th
e
p
r
e
d
icted
o
u
tco
m
e
will
lab
el
th
e
tu
m
o
r
as
eith
er
b
e
n
ig
n
o
r
m
alig
n
an
t.
Ar
tific
ial
in
tellig
en
ce
(
AI
)
ap
p
licatio
n
s
s
u
ch
as
m
ac
h
in
e
lear
n
in
g
let
co
m
p
u
te
r
s
lear
n
f
r
o
m
th
eir
p
ast
ac
tio
n
s
an
d
b
ec
o
m
e
b
etter
as
tim
e
p
ass
es
with
o
u
t
ex
p
licit
p
r
o
g
r
a
m
m
in
g
.
Ma
ch
in
e
lear
n
i
n
g
is
p
r
im
ar
ily
c
o
n
ce
r
n
ed
with
s
o
f
twar
e
a
p
p
licatio
n
s
th
at
r
e
tr
iev
e
av
ailab
le
d
ata
an
d
u
s
e
it
to
g
ain
k
n
o
wled
g
e
b
y
th
em
s
elv
es.
Kr
ith
ik
a
an
d
Gee
th
a
[
2
3
]
d
id
a
s
y
s
tem
atic
r
ev
iew
to
lo
o
k
at
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
.
2
,
Ap
r
il
20
25
:
1
6
7
0
-
1
6
8
5
1672
v
ar
io
u
s
m
eth
o
d
s
to
lo
ca
te
ca
n
ce
r
in
b
r
ea
s
t
tis
s
u
e.
Ad
d
itio
n
al
ly
,
it
lo
o
k
s
in
to
a
f
ew
o
f
th
e
w
ell
-
k
n
o
wn
m
ac
h
in
e
lear
n
in
g
m
o
d
els em
p
lo
y
ed
in
t
h
e
co
n
v
en
tio
n
al
s
y
s
tem
s
ar
e
li
s
ted
in
T
ab
le
1
.
T
ab
le
1
.
L
is
t o
f
c
o
n
v
e
n
tio
n
al
c
lass
if
icatio
n
m
o
d
els
Ref
C
l
a
s
si
f
i
c
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
M
e
r
i
t
s
C
h
a
l
l
e
n
g
e
s
[
2
4
]
S
u
p
p
o
r
t
v
e
c
t
o
r
mac
h
i
n
e
(
S
V
M
)
I
t
w
o
r
k
s we
l
l
i
n
h
i
g
h
d
i
me
n
s
i
o
n
a
l
s
p
a
c
e
a
n
d
e
f
f
i
c
i
e
n
t
m
e
m
o
r
y
u
t
i
l
i
z
a
t
i
o
n
.
I
t
i
s n
o
t
c
a
p
a
b
l
e
o
f
h
a
n
d
l
i
n
g
h
u
g
e
d
a
t
a
set
s
a
n
d
f
a
l
l
s
i
n
l
o
c
a
l
o
p
t
i
ma.
[
2
5
]
D
e
c
i
s
i
o
n
t
r
e
e
(
D
T)
Ea
sy
t
o
u
n
d
e
r
st
a
n
d
,
a
n
d
i
t
c
a
n
h
a
n
d
l
e
mu
l
t
i
-
o
u
t
p
u
t
p
r
o
b
l
e
ms.
A
smal
l
c
h
a
n
g
e
i
n
t
h
e
t
r
e
e
s
t
r
u
c
t
u
r
e
c
o
u
l
d
a
f
f
e
c
t
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
t
h
e
e
n
t
i
r
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
a
n
d
t
h
e
p
r
e
d
i
c
t
i
o
n
h
i
g
h
l
y
d
e
p
e
n
d
s
o
n
t
h
e
s
e
l
e
c
t
i
o
n
a
t
t
r
i
b
u
t
e
s.
[
2
6
]
F
u
z
z
y
l
o
g
i
c
(
F
L)
B
e
t
t
e
r
r
o
b
u
s
t
n
e
ss,
a
n
d
c
a
p
a
b
i
l
i
t
y
o
f
h
a
n
d
l
i
n
g
i
mp
r
e
c
i
s
e
d
a
t
a
.
I
n
a
c
c
u
r
a
t
e
p
r
e
d
i
c
t
i
o
n
s,
a
n
d
n
o
t
s
u
i
t
e
d
f
o
r
a
l
l
a
p
p
l
i
c
a
t
i
o
n
s.
[
2
7
]
D
e
e
p
l
e
a
r
n
i
n
g
(
D
L)
W
o
r
k
s
w
e
l
l
f
o
r
l
a
r
g
e
d
i
me
n
s
i
o
n
a
l
d
a
t
a
,
a
n
d
h
i
g
h
a
c
c
u
r
a
c
y
.
R
e
q
u
i
r
e
s
c
o
mp
l
e
x
m
a
t
h
e
m
a
t
i
c
a
l
o
p
e
r
a
t
i
o
n
s
t
o
p
r
e
d
i
c
t
t
h
e
d
e
s
i
r
e
d
r
e
s
u
l
t
,
a
n
d
l
a
c
k
o
f
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
.
[
2
8
]
N
a
ï
v
e
B
a
y
e
s
(
N
B
)
H
i
g
h
p
r
o
c
e
ss
i
n
g
s
p
e
e
d
,
a
n
d
p
e
r
f
o
r
ms
w
e
l
l
f
o
r
l
a
r
g
e
-
s
i
z
e
d
a
t
a
.
O
v
e
r
f
i
t
t
i
n
g
,
a
n
d
c
o
mp
l
e
x
i
t
y
.
2.
M
E
T
H
O
D
T
h
is
s
ec
tio
n
p
r
o
v
id
es
a
c
o
m
p
lete
ex
p
lan
atio
n
o
f
th
e
p
r
o
p
o
s
ed
YOL
O
-
b
ased
GADCN
-
L
OA
b
r
ea
s
t
ca
n
ce
r
d
etec
tio
n
s
y
s
tem
.
T
h
is
p
ap
er
'
s
k
ey
co
n
tr
ib
u
tio
n
is
th
e
d
ev
elo
p
m
e
n
t
o
f
a
n
in
n
o
v
ativ
e
C
AD
f
r
am
ewo
r
k
f
o
r
t
h
e
p
r
ec
is
e
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
o
f
tu
m
o
r
s
f
r
o
m
th
e
in
p
u
t
m
am
m
o
g
r
a
p
h
y
b
r
ea
s
t
im
ag
es.
T
h
e
p
r
o
p
o
s
ed
YOL
O
-
b
ased
GADCN
-
L
OA
f
r
am
ewo
r
k
’
s
f
lo
w
is
p
o
r
tr
ay
e
d
in
Fig
u
r
e
1
,
wh
ich
en
co
m
p
ass
es
th
e
f
o
llo
win
g
o
p
er
atio
n
s
:
i)
C
o
n
tr
ast
en
h
an
ce
m
e
n
t
an
d
e
q
u
aliza
tio
n
,
ii)
YOL
O
-
b
ased
atten
tio
n
n
etwo
r
k
m
o
d
el
f
o
r
s
eg
m
en
tatio
n
,
iii)
GADCN
m
o
d
el
f
o
r
d
is
ea
s
e
class
if
icatio
n
,
an
d
iv
)
L
OA
f
o
r
ac
tiv
atio
n
f
u
n
ctio
n
esti
m
atio
n
.
T
h
e
in
p
u
t
b
r
ea
s
t
im
ag
e
o
b
tain
ed
f
r
o
m
th
e
g
iv
e
n
d
ataset
is
i
n
itially
p
r
ep
r
o
ce
s
s
ed
with
th
e
u
s
e
o
f
t
h
e
C
L
AHE
m
o
d
el,
wh
er
e
th
e
i
m
ag
e
en
h
a
n
ce
m
en
t,
h
is
to
g
r
a
m
eq
u
aliza
tio
n
,
n
o
is
e
r
em
o
v
a
l,
an
d
n
o
r
m
aliza
tio
n
o
p
er
atio
n
s
a
r
e
ca
r
r
ied
o
u
t
[
2
9
]
.
T
h
is
k
in
d
o
f
p
r
e
p
r
o
ce
s
s
in
g
alg
o
r
ith
m
h
elp
s
to
r
ed
u
ce
t
h
e
class
if
icatio
n
er
r
o
r
wh
ile
m
in
im
izin
g
th
e
tr
ain
in
g
an
d
test
in
g
tim
e.
T
h
e
n
,
a
n
o
v
el
YOL
O
-
b
ased
atten
tio
n
n
etwo
r
k
m
o
d
el
is
u
tili
ze
d
to
s
eg
m
en
t
th
e
b
r
ea
s
t
r
eg
io
n
f
r
o
m
th
e
p
r
ep
r
o
ce
s
s
ed
im
ag
e
with
a
r
ed
u
ce
d
o
v
er
-
s
e
g
m
en
tatio
n
r
ate.
T
o
ac
cu
r
ately
p
r
ed
ict
th
e
h
ea
lth
y
an
d
ca
n
ce
r
-
af
f
ec
ted
im
a
g
es
with
h
ig
h
p
r
ec
is
io
n
an
d
d
ete
ctio
n
ac
cu
r
ac
y
,
th
e
GADCN
-
b
ased
class
if
icat
io
n
alg
o
r
ith
m
is
u
s
ed
in
th
is
s
y
s
tem
[
3
0
]
.
D
u
r
in
g
class
if
icatio
n
,
th
e
ac
tiv
atio
n
f
u
n
ctio
n
is
esti
m
ated
b
ased
o
n
th
e
o
p
tim
al
s
o
lu
tio
n
o
b
tain
ed
f
r
o
m
th
e
L
OA.
T
h
e
m
ajo
r
m
er
its
o
f
th
e
p
r
o
p
o
s
ed
YOL
O
-
b
ased
GADCN
-
L
OA
f
r
am
ewo
r
k
a
r
e
r
ed
u
ce
d
class
if
icatio
n
er
r
o
r
,
im
p
r
o
v
ed
ac
c
u
r
ac
y
,
lo
w
s
y
s
tem
co
m
p
lex
ity
,
m
in
im
ized
tim
e
c
o
n
s
u
m
p
tio
n
,
an
d
r
ed
u
ce
d
o
v
er
-
s
eg
m
en
tatio
n
.
2
.
1
.
P
re
-
pro
ce
s
s
ing
I
t
is
th
e
in
itia
l
s
tag
e
o
f
th
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
,
wh
e
r
e
th
e
im
ag
e
f
ilter
in
g
an
d
n
o
r
m
aliza
tio
n
p
r
o
ce
s
s
es
ar
e
ca
r
r
ied
o
u
t
to
g
en
er
ate
th
e
co
n
t
r
ast
-
en
h
an
ce
d
q
u
ality
o
u
tp
u
t
im
ag
e
.
A
te
ch
n
iq
u
e
k
n
o
wn
as
im
ag
e
en
h
an
ce
m
e
n
t
is
a
way
to
tr
ea
t
an
im
ag
e
s
o
t
h
at
th
e
o
u
tco
m
e
is
s
ig
n
if
ica
n
tly
m
o
r
e
ap
p
r
o
p
r
iate
f
o
r
a
g
iv
en
ap
p
licatio
n
th
a
n
th
e
o
r
i
g
in
al
im
ag
e.
T
h
e
n
u
m
e
r
o
u
s
f
o
r
m
s
o
f
n
o
is
e
p
r
esen
t
in
th
e
u
n
p
r
o
ce
s
s
ed
im
ag
es
g
ath
er
ed
f
r
o
m
th
e
s
ca
n
n
er
p
o
r
t
an
d
web
s
ites
m
ak
e
th
em
u
n
s
u
itab
le
f
o
r
im
m
ed
iate
p
r
o
ce
s
s
in
g
.
C
o
n
s
eq
u
en
tly
,
it
n
ee
d
s
to
b
e
tr
an
s
f
o
r
m
ed
b
ef
o
r
e
b
ein
g
ex
am
i
n
ed
.
An
im
p
o
r
tan
t
s
tep
in
im
ag
e
p
r
o
ce
s
s
in
g
is
s
ca
l
in
g
an
im
ag
e
to
ch
an
g
e
its
p
ix
el
s
ize.
He
r
e,
th
e
C
L
AHE
p
r
ep
r
o
ce
s
s
in
g
m
o
d
el
is
u
s
ed
to
g
en
e
r
ate
t
h
e
q
u
ality
-
en
h
a
n
ce
d
b
r
ea
s
t
im
ag
e
with
r
e
d
u
ce
d
n
o
is
e.
W
h
en
co
m
p
a
r
ed
to
th
e
o
th
er
p
r
ep
r
o
ce
s
s
in
g
m
o
d
els,
t
h
e
k
ey
b
en
e
f
its
o
f
u
s
in
g
C
L
AHE
ar
e
s
im
p
le
to
i
m
p
lem
en
t,
e
n
h
an
ce
d
im
a
g
e
co
n
tr
ast,
an
d
b
etter
v
is
ib
ilit
y
.
Du
r
in
g
th
is
o
p
e
r
atio
n
,
th
e
im
ag
e
n
o
r
m
aliza
tio
n
,
an
d
h
is
to
g
r
am
eq
u
aliza
tio
n
p
r
o
c
ess
es
ar
e
ca
r
r
ied
o
u
t.
Af
ter
o
b
tain
in
g
th
e
in
p
u
t
im
ag
e,
th
e
n
o
r
m
aliza
tio
n
is
p
er
f
o
r
m
e
d
b
ased
o
n
th
e
m
in
i
m
u
m
an
d
m
ax
im
u
m
v
alu
es
a
s
r
ep
r
esen
ted
in
th
e
f
o
llo
win
g
m
o
d
el:
=
(
−
)
∗
∑
∑
(
,
)
−
=
1
=
1
+
(
1
)
wh
er
e
is
th
e
to
tal
n
u
m
b
er
o
f
im
ag
e
s
izes,
(
,
)
is
th
e
im
ag
e
p
ix
el
at
ea
ch
co
o
r
d
in
ate,
&
ar
e
th
e
n
u
m
b
er
o
f
r
o
ws
an
d
co
lu
m
n
s
o
f
th
e
im
ag
e,
d
en
o
tes
th
e
in
p
u
t
im
ag
e,
d
en
o
tes
th
e
m
in
im
u
m
v
alu
e
o
f
th
e
in
p
u
t
im
ag
e,
a
n
d
d
e
n
o
te
s
th
e
m
ax
im
u
m
v
alu
e
o
f
th
e
in
p
u
t
im
ag
e
.
T
h
u
s
,
as
illu
s
tr
ated
b
elo
w,
t
h
e
h
is
to
g
r
am
eq
u
aliza
tio
n
is
u
s
ed
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
u
n
iq
u
e
YOLO
-
b
a
s
ed
g
a
ted
a
tten
tio
n
d
ee
p
co
n
v
o
lu
tio
n
n
et
w
o
r
k
-
Lich
ten
b
erg
…
(
V
in
o
th
R
a
th
in
a
m
)
1673
ℎ
=
ℎ
(
)
(
2
)
wh
er
e
ℎ
r
ep
r
esen
ts
th
e
h
is
to
g
r
am
-
tr
an
s
f
o
r
m
e
d
im
ag
e
.
Fin
all
y
,
th
e
q
u
ality
-
en
h
an
ce
d
im
ag
e
is
o
b
tain
ed
as
s
h
o
wn
in
th
e
f
o
llo
win
g
f
o
r
m
:
=
{
ℎ
(
,
)
|
∀
(
,
)
∈
}
(
3
)
wh
er
e
is
th
e
en
h
an
ce
d
b
r
ea
s
t
im
ag
e,
d
en
o
tes
th
e
i
n
p
u
t
im
a
g
e,
an
d
,
ar
e
th
e
p
i
x
el
co
o
r
d
in
ates.
T
h
e
o
u
tp
u
t
q
u
ality
en
h
an
ce
d
im
a
g
e
is
u
s
ed
f
o
r
s
eg
m
e
n
tatio
n
an
d
class
if
icatio
n
o
p
er
atio
n
s
.
Fig
u
r
e
1
.
Flo
wch
ar
t
o
f
th
e
p
r
o
p
o
s
ed
YOL
O
-
b
ased
GADCN
-
L
OA
b
r
ea
s
t c
an
ce
r
d
etec
tio
n
s
y
s
tem
2
.
2
.
YO
L
O
ba
s
ed
a
t
t
ent
io
n
net
wo
rk
s
eg
m
ent
a
t
io
n
Af
ter
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
,
t
h
e
YOL
O
-
b
ased
atten
tio
n
n
e
two
r
k
s
eg
m
en
tatio
n
m
o
d
el
i
s
u
s
ed
to
s
eg
m
en
t
th
e
b
r
ea
s
t
im
a
g
e
f
r
o
m
th
e
q
u
ality
-
e
n
h
an
ce
d
im
a
g
e
.
I
n
g
e
n
er
al,
t
h
e
d
if
f
er
e
n
t
ty
p
e
s
o
f
t
h
r
esh
o
ld
-
b
ased
,
en
co
d
in
g
-
b
ased
,
an
d
d
ee
p
ar
c
h
itectu
r
e
-
b
ased
s
eg
m
en
tatio
n
m
o
d
els
ar
e
im
p
lem
en
te
d
in
t
h
e
ex
is
tin
g
s
tu
d
ies
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
.
2
,
Ap
r
il
20
25
:
1
6
7
0
-
1
6
8
5
1674
f
o
r
b
r
ea
s
t
ca
n
ce
r
d
iag
n
o
s
is
.
H
o
wev
er
,
m
o
s
t
o
f
th
e
tech
n
iq
u
e
s
h
av
e
th
e
m
ai
n
d
r
aw
b
ac
k
s
o
f
o
v
er
-
s
eg
m
e
n
tatio
n
,
h
ig
h
d
esig
n
in
g
co
m
p
lex
ity
,
a
n
d
in
cr
ea
s
ed
tim
e
co
n
s
u
m
p
tio
n
.
T
h
er
ef
o
r
e,
th
e
p
r
o
p
o
s
ed
w
o
r
k
in
ten
d
s
to
u
s
e
a
s
o
p
h
is
ticated
d
ee
p
s
eg
m
en
tati
o
n
m
o
d
el
f
o
r
p
r
e
d
ictin
g
ca
n
ce
r
in
b
r
ea
s
ts
.
T
h
e
atten
tio
n
m
e
ch
an
is
m
m
ak
es
u
s
e
o
f
th
e
h
u
m
an
ca
p
ac
ity
f
o
r
s
ele
ctiv
e
atten
tio
n
.
I
n
p
a
r
ticu
lar
,
a
p
er
s
o
n
m
ig
h
t
f
o
cu
s
o
n
t
h
e
a
r
e
as
o
f
i
n
ter
est
wh
ile
q
u
ick
ly
s
ca
n
n
i
n
g
th
e
e
n
tire
i
m
ag
e.
T
h
e
m
o
d
el
m
i
g
h
t
th
en
au
to
m
atica
lly
f
o
c
u
s
o
n
cr
u
cia
l
s
eq
u
en
ce
f
ea
tu
r
es,
im
p
r
o
v
in
g
th
e
ab
ilit
y
to
h
an
d
le
s
eq
u
en
ce
d
ata
with
o
u
t
in
c
r
ea
s
in
g
th
e
c
o
s
t
o
f
co
m
p
u
tatio
n
.
Fo
llo
win
g
th
at,
d
etailed
in
f
o
r
m
atio
n
a
b
o
u
t
s
p
ec
if
ic
r
eg
io
n
s
is
ac
q
u
ir
ed
a
n
d
u
n
n
ec
ess
ar
y
in
f
o
r
m
atio
n
is
s
u
p
p
r
ess
ed
.
I
n
th
e
p
r
o
p
o
s
ed
s
y
s
tem
,
th
e
lig
h
twe
ig
h
t
YOL
O
-
b
ased
atte
n
tio
n
n
etwo
r
k
is
s
p
ec
if
ically
im
p
le
m
en
ted
to
p
e
r
f
o
r
m
s
eg
m
en
tatio
n
with
m
in
im
ized
co
m
p
u
tatio
n
al
c
o
m
p
lex
ity
a
n
d
tim
e
co
n
s
u
m
p
tio
n
.
T
h
e
n
o
v
el
co
n
ce
p
t
o
f
th
is
m
o
d
el
is
,
th
at
a
b
o
u
n
d
ar
y
lo
s
s
f
u
n
ctio
n
is
co
m
p
u
ted
t
o
im
p
r
o
v
e
th
e
d
is
ea
s
e
p
r
ed
ictio
n
r
ate.
Mo
r
eo
v
er
,
t
h
e
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
ar
ch
it
ec
tu
r
e
m
o
d
el
co
m
p
r
is
es
th
e
f
o
llo
win
g
lay
er
s
:
i)
in
p
u
t
l
a
y
er
,
ii)
b
ac
k
b
o
n
e
lay
er
,
iii)
f
ea
tu
r
e
p
y
r
am
id
lay
e
r
(
FP
L
)
,
iv
)
p
ath
ag
g
r
eg
atio
n
l
ay
er
(
PAL)
,
an
d
v
)
p
r
ed
ictio
n
lay
e
r
T
h
e
in
p
u
t
lay
er
is
m
ai
n
ly
u
s
e
d
to
g
at
h
er
th
e
in
p
u
t
im
ag
e
d
a
ta
f
o
r
e
n
h
an
ce
m
e
n
t.
T
h
en
,
t
h
e
b
ac
k
b
o
n
e
lay
er
co
m
p
r
is
es
th
e
s
lice
m
o
d
u
le
th
at
h
elp
s
to
im
p
r
o
v
e
t
h
e
p
r
o
ce
s
s
o
f
co
m
p
u
tatio
n
w
ith
r
ed
u
ce
d
s
p
ee
d
.
Mo
r
eo
v
er
,
th
e
FP
L
an
d
PAL
ar
e
u
s
ed
to
p
er
f
o
r
m
th
e
f
u
s
io
n
an
d
,
c
o
m
p
lem
e
n
tatio
n
o
f
h
ig
h
-
lev
el
f
ea
tu
r
es
an
d
lo
w
-
lev
el
f
ea
tu
r
es
r
esp
ec
tiv
el
y
.
Fin
ally
,
th
e
p
r
ed
ictio
n
lay
e
r
is
u
s
ed
to
g
e
n
er
ate
th
e
o
u
tp
u
t
class
ac
co
r
d
in
g
to
th
e
b
o
u
n
d
a
r
y
lo
s
s
f
u
n
ctio
n
.
Af
ter
g
ettin
g
t
h
e
en
h
an
ce
d
im
ag
e
f
r
o
m
th
e
p
r
ev
io
u
s
s
tag
e,
th
e
lay
e
r
in
itializatio
n
is
p
er
f
o
r
m
e
d
at
f
ir
s
t.
T
h
en
,
th
e
d
y
n
am
ic
an
c
h
o
r
lo
ca
lizatio
n
is
p
er
f
o
r
m
ed
with
th
e
u
s
e
o
f
a
b
ac
k
b
o
n
e
lay
er
,
wh
er
e
th
e
m
atch
in
g
d
eg
r
ee
is
esti
m
ated
b
ased
o
n
th
e
s
p
atial
m
atch
in
g
in
f
o
r
m
atio
n
,
f
ea
tu
r
e
p
lace
m
en
t
ca
p
ab
ilit
y
,
an
d
r
e
g
r
ess
io
n
am
b
ig
u
ity
.
Her
e,
th
e
lo
ca
lizatio
n
ca
p
ac
ity
is
m
ea
s
u
r
ed
ac
co
r
d
in
g
to
th
e
r
eg
r
ess
io
n
am
b
ig
u
ity
,
an
d
is
r
ep
r
esen
ted
in
th
e
f
o
llo
win
g
m
o
d
el:
=
∗
+
(
1
−
)
∗
−
(
4
)
wh
er
e
&
ar
e
th
e
h
y
p
er
p
ar
am
eter
s
u
s
ed
to
weig
h
th
e
in
f
lu
en
ce
o
f
d
if
f
er
e
n
t
d
ata,
is
th
e
s
p
atia
l
m
atch
in
g
in
f
o
r
m
atio
n
,
in
d
icate
s
th
e
f
ea
tu
r
e
p
lace
m
en
t
ca
p
ab
ilit
y
f
o
r
i
n
p
u
t
d
ata,
an
d
is
a
p
en
alty
ter
m
.
Af
ter
th
at,
an
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
(
I
o
U)
is
esti
m
ated
b
ef
o
r
e
an
d
af
ter
r
eg
r
ess
io
n
am
b
i
g
u
ity
as
r
ep
r
esen
ted
in
th
e
f
o
llo
win
g
m
o
d
el:
=
|
−
|
(
5
)
Mo
r
eo
v
er
,
th
e
h
y
p
er
p
ar
am
ete
r
is
co
m
p
u
ted
ac
co
r
d
in
g
to
th
e
p
r
ec
is
e
a
d
ju
s
tm
en
t
s
ch
e
d
u
le
as
s
h
o
wn
in
th
e
f
o
llo
win
g
m
o
d
el:
(
)
=
{
1
,
<
0
.
1
5
∗
(
0
−
1
)
∗
+
(
1
.
5
−
0
.
5
∗
0
)
0
.
1
≤
<
0
.
3
0
≥
0
.
3
(
6
)
=
(
7
)
wh
er
e
in
d
icate
s
th
e
cu
r
r
en
t
iter
atio
n
,
is
th
e
to
tal
n
u
m
b
er
o
f
iter
atio
n
s
,
an
d
0
r
ep
r
esen
ts
th
e
p
r
ev
io
u
s
weig
h
tin
g
f
ac
t
o
r
o
f
ea
c
h
iter
at
io
n
.
Fu
r
t
h
er
m
o
r
e,
th
e
atten
tio
n
m
ec
h
a
n
is
m
is
im
p
lem
en
ted
with
th
e
p
er
ce
p
tr
o
n
m
o
d
el,
wh
er
e
th
e
s
p
atial
atten
tio
n
is
co
m
p
u
ted
to
im
p
r
o
v
e
t
h
e
n
etwo
r
k
’
s
s
en
s
itiv
ity
to
id
en
tify
th
e
d
ef
ec
ted
ar
ea
s
.
Du
r
in
g
th
is
p
r
o
ce
s
s
,
th
e
ch
an
n
el
atten
tio
n
weig
h
t is co
m
p
u
ted
as sh
o
wn
i
n
(
8
)
:
(
)
=
∗
(
(
)
+
(
)
)
(
8
)
T
h
en
,
th
e
s
p
atial
atten
tio
n
weig
h
t is estima
ted
b
y
u
s
in
g
t
h
e
f
o
llo
win
g
m
o
d
el:
(
)
=
∗
(
(
∗
)
)
(
9
)
wh
er
e
,
(
.
)
r
ep
r
esen
ts
th
e
p
er
ce
p
tr
o
n
lay
e
r
,
(
.
)
r
ep
r
esen
ts
th
e
co
n
v
o
lu
tio
n
al
o
p
er
atio
n
with
th
e
k
er
n
el
s
ize
7
∗
7
,
&
in
d
icate
s
th
e
ch
an
n
e
l
d
im
en
s
io
n
o
f
a
v
er
ag
e
a
n
d
m
ax
im
u
m
f
ea
t
u
r
es,
an
d
d
en
o
te
s
th
e
ch
an
n
el
atten
tio
n
p
ar
am
eter
.
Fu
r
th
er
m
o
r
e
,
th
e
f
in
al
atten
ti
o
n
m
ec
h
a
n
is
m
is
ap
p
lied
to
m
ak
e
an
ap
p
r
o
p
r
iate
s
eg
m
en
tatio
n
d
ec
is
io
n
as illu
s
tr
ated
in
(
1
0
)
:
′
=
(
)
⨂
(
1
0
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
u
n
iq
u
e
YOLO
-
b
a
s
ed
g
a
ted
a
tten
tio
n
d
ee
p
co
n
v
o
lu
tio
n
n
et
w
o
r
k
-
Lich
ten
b
erg
…
(
V
in
o
th
R
a
th
in
a
m
)
1675
′′
=
(
)
⨂
′
(
1
1
)
C
o
n
s
eq
u
en
tly
,
th
e
f
u
s
io
n
m
o
d
el
is
ap
p
lied
to
f
u
s
e
th
e
f
e
atu
r
es
f
o
r
o
b
tain
in
g
an
ac
cu
r
ate
d
ef
ec
t
r
ec
o
g
n
itio
n
p
r
o
b
a
b
ilit
y
.
T
h
e
n
o
v
el
co
n
ce
p
t
o
f
th
is
m
o
d
el
is
to
co
m
p
u
te
th
e
b
o
u
n
d
a
r
y
lo
s
s
f
u
n
ctio
n
ac
co
r
d
in
g
to
th
e
g
r
o
u
n
d
tr
u
th
b
o
u
n
d
in
g
b
o
x
as r
ep
r
esen
te
d
in
th
e
f
o
llo
win
g
m
o
d
el:
(
,
)
=
(
,
)
−
|
|
−
|
∪
|
|
|
(
1
2
)
=
1
−
(
,
)
=
1
−
(
,
)
+
|
|
−
|
∪
|
|
|
(
1
3
)
wh
er
e
is
th
e
s
m
allest
b
o
x
co
n
tain
in
g
an
d
,
d
en
o
tes th
e
d
is
j
o
in
t sit
u
atio
n
o
f
an
d
in
th
at
h
as
s
im
ilar
s
ca
le
-
in
v
ar
ian
ce
ch
ar
a
cter
is
tics
.
Mo
r
eo
v
er
,
th
e
d
ir
ec
tio
n
is
g
r
ad
u
ally
co
m
p
u
ted
b
a
s
ed
o
n
th
e
d
is
tan
ce
b
etwe
en
two
m
in
o
r
co
m
p
o
n
e
n
ts
,
an
d
ca
n
ef
f
ec
tiv
ely
r
ed
u
ce
th
e
d
is
tan
ce
b
etwe
en
tw
o
tar
g
et
b
o
x
es,
h
en
ce
it
co
n
v
er
g
es
th
e
s
p
ee
d
m
u
ch
f
aster
.
T
h
en
,
th
e
p
ar
a
m
eter
is
co
m
p
u
ted
b
y
u
s
in
g
th
e
f
o
llo
win
g
m
o
d
el:
=
−
2
(
,
)
(
1
4
)
=
1
−
=
1
−
+
2
(
,
)
(
1
5
)
wh
er
e
,
an
d
in
d
icate
th
e
p
r
ed
i
ctio
n
'
s
p
r
im
ar
y
p
o
in
ts
b
o
x
a
n
d
g
r
o
u
n
d
-
tr
u
th
b
o
x
,
r
esp
ec
tiv
ely
,
in
d
icate
s
th
e
s
q
u
ar
e
o
f
t
h
e
m
i
n
im
al
b
o
u
n
d
in
g
b
o
x
'
s
d
iag
o
n
al
len
g
th
,
a
n
d
2
d
en
o
tes
th
e
E
u
clid
ea
n
d
is
tan
ce
.
I
n
th
e
f
o
llo
win
g
m
o
d
el,
th
e
r
e
ar
e
two
b
o
x
es in
b
o
th
th
e
h
o
r
i
zo
n
tal
an
d
v
er
tical
o
r
ien
tatio
n
s
:
=
-
[
2
(
,
)
+
]
(
1
6
)
=
1
−
+
[
2
(
,
)
+
]
(
1
7
)
wh
er
e
is
th
e
weig
h
t
p
ar
am
et
er
an
d
is
u
s
ed
to
m
ea
s
u
r
e
t
h
e
s
im
ilar
ity
o
f
th
e
asp
ec
t
r
a
tio
.
Fin
ally
,
t
h
e
p
r
o
ce
s
s
h
as
b
ee
n
iter
ated
u
n
til
r
ea
ch
in
g
th
e
lo
west
in
t
er
s
ec
tio
n
o
v
e
r
u
n
io
n
(
I
o
U)
,
wh
ich
r
etu
r
n
s
t
h
e
s
eg
m
en
ted
im
ag
e
as th
e
r
esu
lt.
2
.
3
.
G
a
t
ed
a
da
ptiv
e
deep
co
nv
o
lutio
na
l net
wo
rk
Af
ter
s
eg
m
en
tatio
n
,
th
e
n
o
v
el
GADC
N
-
b
ased
class
if
icatio
n
alg
o
r
ith
m
is
im
p
lem
en
ted
to
ac
cu
r
ately
p
r
ed
ict
th
e
d
is
ea
s
e
f
r
o
m
th
e
s
eg
m
en
ted
im
a
g
e.
T
r
ad
i
tio
n
ally
,
v
ar
io
u
s
d
ee
p
-
lear
n
i
n
g
alg
o
r
ith
m
s
ar
e
im
p
lem
en
ted
f
o
r
b
r
ea
s
t
ca
n
ce
r
d
etec
tio
n
an
d
class
id
en
tific
atio
n
.
Ho
wev
er
,
m
o
s
t
o
f
th
e
m
ec
h
an
is
m
s
h
av
e
m
ajo
r
p
r
o
b
lem
s
in
ter
m
s
o
f
in
cr
ea
s
ed
co
m
p
u
tatio
n
al
co
m
p
lex
ity
,
m
is
-
p
r
ed
ictio
n
r
ate,
an
d
h
ig
h
s
y
s
tem
co
m
p
lex
ity
.
T
h
er
e
f
o
r
e,
th
e
p
r
o
p
o
s
ed
w
o
r
k
aim
s
to
u
s
e
a
n
o
v
el
GADCN
class
if
icatio
n
alg
o
r
ith
m
f
o
r
b
r
ea
s
t
ca
n
ce
r
d
iag
n
o
s
is
,
L
ich
ten
b
er
g
o
p
tim
izatio
n
alg
o
r
ith
m
is
e
m
p
lo
y
ed
t
o
co
m
p
u
te
th
e
ac
ti
v
atio
n
f
u
n
ctio
n
f
o
r
en
h
an
cin
g
p
r
ed
ictio
n
ac
c
u
r
ac
y
.
As
a
r
esu
lt
o
f
s
tu
d
y
in
g
th
e
ch
an
n
el
atten
tio
n
m
ec
h
a
n
is
m
,
th
is
s
y
s
tem
u
s
e
s
a
s
p
ec
ial
atten
tio
n
m
o
d
u
le
to
p
r
o
m
o
te
an
a
d
ap
tiv
e
f
ea
tu
r
e
f
u
s
io
n
th
at
in
co
r
p
o
r
ates
th
e
ch
an
n
el
r
elev
an
c
e
d
escr
ip
tio
n
in
to
th
e
s
tan
d
ar
d
g
ated
atten
tio
n
(
GA)
m
o
d
u
le.
T
h
e
g
ated
m
ec
h
an
is
m
is
th
en
u
s
ed
to
p
er
f
o
r
m
ad
ap
tiv
e
f
ea
tu
r
e
f
u
s
io
n
b
y
t
h
e
g
ated
ch
an
n
el
atten
tio
n
c
o
ef
f
i
cien
ts
,
allo
win
g
th
e
g
ated
m
ec
h
an
is
m
to
tak
e
in
t
o
ac
co
u
n
t
all
o
f
th
e
weig
h
ts
s
u
p
p
lied
to
class
if
icatio
n
f
ea
tu
r
e
m
ap
s
an
d
u
s
e
th
e
f
u
lly
co
n
n
ec
ted
s
u
b
-
n
etwo
r
k
s
weig
h
ts
to
ex
p
lain
th
e
s
ig
n
if
i
ca
n
ce
ac
r
o
s
s
ch
an
n
els.
I
t
d
o
es
th
is
b
y
o
b
tain
in
g
th
e
s
tatis
tic
al
p
ar
am
eter
s
o
f
an
ad
ap
tiv
e
g
ated
ch
an
n
el
u
s
in
g
g
lo
b
al
p
o
o
lin
g
an
d
a
f
u
ll
y
co
n
n
ec
ted
s
u
b
-
n
etwo
r
k
.
Af
ter
o
b
tain
in
g
th
e
s
eg
m
en
ted
im
ag
e
,
th
e
f
ea
tu
r
e
m
ap
is
co
n
s
tr
u
cted
at
f
ir
s
t
b
a
s
ed
o
n
t
h
e
co
n
v
o
lu
tio
n
o
p
er
ati
o
n
as
s
h
o
wn
in
(
1
8
)
:
=
(
)
(
1
8
)
wh
er
e
(
.
)
in
d
icate
s
th
e
co
n
v
o
lu
ti
o
n
al
o
p
er
atio
n
.
T
h
en
,
th
e
ad
ap
tiv
e
weig
h
t is est
im
ated
u
s
in
g
th
e
f
ea
tu
r
e
m
ap
o
f
t
h
e
s
eg
m
en
ted
p
ix
el
as
r
ep
r
esen
ted
in
th
e
f
o
llo
win
g
m
o
d
el:
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
.
2
,
Ap
r
il
20
25
:
1
6
7
0
-
1
6
8
5
1676
̅
=
∗
+
(
1
−
)
∑
∈
̅
(
1
9
)
wh
er
e
th
e
s
eg
m
en
ted
p
ix
el'
s
f
ea
tu
r
e
m
ap
,
,
wh
ich
r
an
g
es
f
r
o
m
0
to
1
,
is
th
e
co
ef
f
icien
t
o
f
.
T
h
e
weig
h
ted
av
er
ag
e
o
f
th
e
ch
ar
ac
ter
is
tics
o
f
p
ix
els
wi
th
in
th
e
s
am
e
s
u
p
-
p
ix
el
is
r
ep
r
esen
ted
b
y
th
e
f
u
n
ctio
n
∑
∈
̅
,
with
(
1
−
)
as
th
e
c
o
r
r
esp
o
n
d
in
g
co
ef
f
icien
t.
Her
e,
th
e
f
ea
tu
r
e
d
is
tan
ce
b
etwe
en
t
h
e
p
ai
r
o
f
p
ix
els
is
u
s
ed
t
o
esti
m
ate
th
e
ad
ap
tiv
e
weig
h
t
th
at
co
r
r
esp
o
n
d
s
to
th
e
n
ea
r
b
y
p
ix
el
o
f
th
e
s
eg
m
en
ted
im
ag
e
,
as illu
s
tr
ated
in
th
e
m
o
d
el
th
at
f
o
llo
ws.
=
e
xp
(
−
|
̅
̅
̅
−
̅
̅
̅
|
2
)
(
2
0
)
wh
er
e
(
.
)
in
d
icate
s
th
e
ex
p
o
n
en
t
f
u
n
ctio
n
,
|
̅
−
̅
|
is
th
e
ab
s
o
lu
te
v
alu
e
o
f
th
e
d
if
f
e
r
en
ce
b
etwe
en
f
ea
tu
r
es
o
f
ℎ
an
d
ℎ
p
ix
el.
2
r
ef
er
s
to
th
e
v
ar
ian
ce
o
f
th
e
in
te
n
s
ity
f
ea
tu
r
es
o
f
p
ix
els
with
in
th
e
ad
jace
n
t
p
ix
el.
Mo
r
e
o
v
er
,
t
h
e
g
lo
b
al
p
o
o
lin
g
o
p
er
atio
n
is
p
er
f
o
r
m
e
d
i
n
th
e
p
o
o
lin
g
la
y
er
as r
ep
r
esen
ted
in
(
2
1
)
:
[
1
,
2
,
…
,
]
=
(
1
,
2
,
…
,
)
(
2
1
)
wh
er
e
r
ep
r
esen
ts
th
e
ch
a
n
n
el
m
atr
ix
o
f
n
u
m
b
e
r
o
f
f
ea
tu
r
es
an
d
is
th
e
co
r
r
esp
o
n
d
in
g
f
ea
t
u
r
e
o
b
tain
e
d
b
y
g
lo
b
al
p
o
o
lin
g
.
C
o
n
s
eq
u
e
n
tly
,
th
e
f
u
n
ctio
n
n
a
m
ed
r
ec
ti
f
ied
lin
ea
r
u
n
its
(
R
eL
U)
is
a
p
p
lied
as
th
e
ac
tiv
e
f
u
n
ctio
n
o
f
th
e
f
ir
s
t
f
u
lly
co
n
n
ec
ted
lay
er
to
m
ak
e
th
e
p
a
r
a
m
eter
s
o
f
GADCN
a
s
il
lu
s
tr
at
ed
in
th
e
f
o
llo
win
g
m
o
d
el:
=
(
∗
+
)
(
2
2
)
wh
er
e
in
d
icate
s
th
e
b
ias
o
f
t
h
e
n
eu
r
o
n
.
T
h
en
,
th
e
o
u
tp
u
t
v
alu
es
ar
e
r
ef
in
e
d
b
y
u
s
in
g
th
e
s
i
g
m
o
id
f
u
n
ctio
n
,
wh
ich
in
d
icate
s
th
e
g
ated
atten
tio
n
co
ef
f
icien
t v
alu
e
r
an
g
i
n
g
as (
0
,
1
)
:
ƛ
=
(
(
)
∗
(
)
+
(
)
)
(
2
3
)
wh
er
e
is
th
e
o
u
tp
u
t
o
f
lig
h
tw
eig
h
t
f
ea
tu
r
e
o
p
tim
izin
g
(
L
F
O
)
.
Mo
r
eo
v
er
,
th
e
f
ea
t
u
r
e
f
u
s
io
n
is
p
er
f
o
r
m
e
d
b
y
th
e
g
ated
atten
tio
n
m
ec
h
an
i
s
m
as r
ep
r
esen
ted
in
th
e
f
o
llo
win
g
m
o
d
el:
=
ƛ
⨂
1
+
⋯
+
(
1
−
ƛ
)
⨂
(
2
4
)
Fin
ally
,
th
e
class
if
ied
lab
el
is
p
r
ed
icted
as th
e
o
u
tp
u
t w
ith
t
h
e
co
s
t f
u
n
ctio
n
as sh
o
wn
in
(
2
5
)
:
=
−
1
∑
∑
(
=
)
{
∑
(
)
=
1
}
=
1
=
1
(
2
5
)
wh
er
e
is
th
e
n
u
m
b
er
o
f
v
al
u
es in
f
ea
tu
r
e
f
u
s
io
n
a
n
d
n
u
m
b
e
r
o
f
class
es.
2
.
4
.
L
icht
enberg
o
ptim
iza
t
i
o
n a
lg
o
rit
hm
Du
r
in
g
class
if
icatio
n
,
th
e
ac
ti
v
atio
n
f
u
n
ctio
n
is
o
p
tim
ally
c
o
m
p
u
ted
b
ased
o
n
th
e
b
est
o
p
tim
al
v
alu
e
o
b
tain
ed
f
r
o
m
t
h
e
L
OA.
W
h
en
co
m
p
a
r
ed
to
th
e
tr
ad
itio
n
a
l
o
p
tim
izatio
n
tech
n
iq
u
es,
th
e
L
OA
p
r
o
v
i
d
es
an
im
p
r
o
v
e
d
p
e
r
f
o
r
m
an
ce
o
u
tco
m
e,
wh
ich
h
elp
s
to
o
b
tain
th
e
m
ax
im
u
m
d
is
ea
s
e
p
r
e
d
ictio
n
ac
cu
r
ac
y
.
I
n
t
h
is
m
o
d
el,
th
e
o
b
jectiv
e
f
u
n
ctio
n
i
n
th
e
s
ea
r
ch
in
g
s
p
ac
e
at
f
ir
s
t w
ith
th
e
u
p
p
er
an
d
lo
wer
b
o
u
n
d
v
alu
es.
Similar
ly
,
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
iter
a
tio
n
s
an
d
n
u
m
b
er
o
f
p
o
p
u
latio
n
s
ar
e
also
in
itialized
.
T
h
en
,
t
h
e
r
a
n
d
o
m
s
ca
lin
g
an
d
r
o
tatio
n
o
p
e
r
atio
n
s
ar
e
ca
r
r
ied
o
u
t to
esti
m
ate
th
e
f
itn
ess
v
alu
e
f
o
r
th
e
g
i
v
en
p
r
o
b
lem
.
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
an
d
r
esu
lts
o
f
th
e
YOL
O
-
b
ased
GADCN
-
L
OA
m
ec
h
an
is
m
f
o
r
th
e
b
r
e
ast
ca
n
ce
r
d
iag
n
o
s
is
o
b
tain
ed
u
s
in
g
p
o
p
u
lar
b
en
c
h
m
ar
k
d
atasets
is
d
escr
ib
ed
in
th
is
s
ec
tio
n
.
T
h
e
e
v
alu
atio
n
m
ea
s
u
r
es
s
u
ch
as
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
an
d
F1
-
s
co
r
e
ar
e
an
aly
ze
d
to
v
alid
ate
t
h
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
cu
r
r
en
t
s
tu
d
y
u
tili
ze
s
th
e
p
u
b
lic
an
d
m
o
s
t
p
o
p
u
lar
d
at
asets
f
o
r
s
y
s
tem
v
ali
d
atio
n
a
n
d
an
aly
s
is
[
3
1
]
an
d
ar
e
r
ep
r
esen
ted
in
T
ab
le
2
.
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
A
u
n
iq
u
e
YOLO
-
b
a
s
ed
g
a
ted
a
tten
tio
n
d
ee
p
co
n
v
o
lu
tio
n
n
et
w
o
r
k
-
Lich
ten
b
erg
…
(
V
in
o
th
R
a
th
in
a
m
)
1677
W
ith
lin
k
ed
g
r
o
u
n
d
tr
u
th
o
f
m
ass
r
eg
io
n
s
,
th
e
DDSM
d
atab
ase
h
as
r
o
u
g
h
l
y
2
,
6
2
0
ca
s
e
s
o
f
ca
n
ce
r
,
b
en
ig
n
,
an
d
n
o
r
m
al
b
r
ea
s
t
im
a
g
es.
Dec
o
m
p
r
ess
ed
to
a
s
ize
o
f
5
0
0
0
×
3
0
0
0
p
ix
els,
th
e
s
elec
ted
p
ictu
r
es
f
o
r
th
e
C
B
I
S
-
DD
SM
d
ataset
h
av
e
b
ee
n
tr
an
s
lated
in
to
DI
C
OM
f
o
r
m
at.
T
h
e
M
I
AS
d
atab
ase
c
o
n
tain
s
3
2
2
d
ig
ital
m
am
m
o
g
r
a
p
h
y
im
ag
es,
ea
ch
m
ea
s
u
r
in
g
1
,
0
2
4
b
y
1
,
0
2
4
p
ix
els
an
d
in
PGM
f
o
r
m
at.
T
h
e
en
tire
I
N
B
r
ea
s
t
d
atab
ase
co
n
tain
s
4
1
0
p
h
o
to
g
r
ap
h
s
,
r
ep
r
esen
tin
g
b
en
ig
n
,
m
al
ig
n
an
t,
an
d
n
o
r
m
al
s
tates.
I
t c
o
n
tain
s
1
1
2
m
ass
es
in
to
tal,
all
o
f
wh
ich
a
r
e
lin
k
e
d
to
ac
cu
r
ate
g
r
o
u
n
d
tr
u
t
h
o
u
tlin
es
cr
ea
ted
b
y
p
r
o
f
ess
io
n
als.
T
h
e
im
ag
es
in
th
is
d
atab
ase
ar
e
r
ec
o
r
d
ed
in
th
e
DI
C
OM
f
ile
f
o
r
m
at
an
d
r
a
n
g
e
in
s
ize
f
r
o
m
2
5
6
0
×
3
3
2
8
p
i
x
els
to
3
3
2
8
×4
0
8
4
p
ix
els,
d
ep
en
d
in
g
o
n
th
e
s
ize
o
f
th
e
p
atien
t'
s
b
r
ea
s
t.
Fig
u
r
e
2
(
a)
to
2
(
f
)
d
is
p
lay
th
e
s
am
p
le
in
p
u
t
an
d
o
u
t
p
u
t
b
r
ea
s
t
im
ag
es,
wh
ich
in
clu
d
e
th
e
in
p
u
t
p
h
o
to
s
,
g
r
o
u
n
d
tr
u
th
im
ag
es,
p
r
ep
r
o
ce
s
s
ed
o
u
t
p
u
ts
,
clu
s
ter
ed
o
u
t
p
u
ts
,
b
in
ar
y
o
u
tp
u
ts
,
a
n
d
o
u
tp
u
t im
a
g
es with
tu
m
o
r
s
d
etec
ted
.
T
ab
le
2
.
Data
s
ets u
s
ed
in
th
is
s
tu
d
y
D
a
t
a
s
e
t
s
D
e
scri
p
t
i
o
n
S
e
t
1
D
D
S
M
S
e
t
2
I
N
B
r
e
a
st
S
e
t
3
M
I
A
S
S
e
t
4
W
D
B
C
S
e
t
5
W
B
C
D
(
i)
(
ii)
(
iii)
(
iv
)
(
v
)
(
v
i)
(
v
ii)
(
v
iii)
(
ix
)
(
x
)
(
i)
(
ii)
(
iii)
(
iv
)
(
v
)
(
v
i)
(
v
ii)
(
v
iii)
(
ix
)
(
x
)
(
a)
(
b
)
(
i)
(
ii)
(
iii)
(
iv
)
(
v
)
(
v
i)
(
v
ii)
(
v
iii)
(
ix
)
(
x
)
(
i)
(
ii)
(
iii)
(
iv
)
(
v
)
(
v
i)
(
v
ii)
(
v
iii)
(
ix
)
(
x
)
(
c)
(
d
)
(
i)
(
ii)
(
iii)
(
iv
)
(
v
)
(
v
i)
(
v
ii)
(
v
iii)
(
ix
)
(
x
)
(
i)
(
ii)
(
iii)
(
iv
)
(
v
)
(
v
i)
(
v
ii)
(
v
iii)
(
ix
)
(
x
)
(
e)
(f)
Fig
u
r
e
2
.
B
r
ea
s
t
tu
m
o
r
d
etec
tio
n
s
tag
es (
a)
in
p
u
t b
r
ea
s
t im
ag
es,
(
b
)
g
r
o
u
n
d
tr
u
t
h
b
r
ea
s
t im
a
g
es,
(
c)
p
r
e
p
r
o
ce
s
s
ed
b
r
ea
s
t im
ag
e
s
,
(
d
)
clu
s
ter
ed
o
u
tp
u
t im
a
g
es,
(
e)
b
in
a
r
y
o
u
tp
u
t im
a
g
es,
an
d
(
f
)
tu
m
o
r
d
etec
ted
im
ag
es
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
.
2
,
Ap
r
il
20
25
:
1
6
7
0
-
1
6
8
5
1678
Mo
r
eo
v
er
,
th
e
p
er
f
o
r
m
an
ce
m
ea
s
u
r
es
u
s
ed
to
ass
ess
th
e
r
esu
lts
ar
e
co
m
p
u
ted
b
y
u
s
in
g
th
e
f
o
llo
win
g
m
o
d
els:
=
+
+
+
+
(
2
6
)
=
+
(
2
7
)
=
+
(
2
8
)
=
+
(
2
9
)
1
−
=
2
×
×
+
(
3
0
)
wh
er
e
,
is
tr
u
e
p
o
s
itiv
es
,
T
n
is
tr
u
e
n
eg
ativ
e
,
is
f
alse
p
o
s
it
iv
e,
an
d
is
f
alse
n
eg
ativ
e.
B
y
u
s
in
g
th
e
DDSM,
I
N
B
r
ea
s
t,
an
d
MI
AS
d
atasets
,
Fig
u
r
e
3
v
er
if
ies
th
e
r
ec
eiv
er
o
p
er
atin
g
ch
a
r
ac
ter
is
tics
(
R
O
C
)
o
f
th
e
p
r
o
p
o
s
ed
YOL
O
-
b
ased
GADCN
-
L
OA
m
o
d
el.
Fig
u
r
e
3
.
R
OC
an
aly
s
is
T
h
e
ar
ea
u
n
d
er
cu
r
v
e
(
AUC)
co
n
ce
r
n
i
n
g
th
e
T
P
r
ate
an
d
FP
r
ate
is
g
en
er
ated
in
th
is
a
n
aly
s
is
to
d
em
o
n
s
tr
ate
th
e
b
r
ea
s
t
ca
n
ce
r
d
iag
n
o
s
is
m
o
d
el'
s
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
im
p
lem
en
ts
a
n
o
v
el
YOL
O
-
b
ased
atten
tio
n
n
etwo
r
k
s
eg
m
e
n
tatio
n
tec
h
n
iq
u
e
th
at
ac
c
u
r
ately
s
eg
m
en
ts
th
e
b
r
ea
s
t
r
e
g
io
n
f
o
r
a
s
u
cc
ess
f
u
l
illn
ess
p
r
ed
ictio
n
.
T
h
e
YOL
O
atten
tio
n
s
eg
m
en
t
atio
n
m
o
d
el
g
r
ea
tly
e
n
h
an
ce
s
t
h
e
ac
cu
r
ac
y
o
f
th
e
s
u
g
g
ested
m
o
d
el'
s
class
if
ier
d
etec
tio
n
.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
p
r
o
p
o
s
ed
GADCN
-
L
OA
clas
s
if
ier
an
d
co
n
v
en
tio
n
al
d
ee
p
lear
n
i
n
g
is
th
en
v
alid
ated
a
n
d
c
o
m
p
a
r
ed
u
s
in
g
th
e
M
I
AS
d
ataset,
as
s
h
o
w
n
in
Fig
u
r
es
4
t
o
6
an
d
t
h
e
er
r
o
r
r
ate
is
s
h
o
wn
in
Fig
u
r
e
7
.
Sev
er
al
m
etr
ics
h
a
v
e
b
ee
n
u
s
ed
in
th
is
in
q
u
ir
y
to
ass
ess
p
er
f
o
r
m
an
ce
.
T
h
e
en
h
an
ce
d
F1
-
s
co
r
e
v
alu
es,
s
en
s
itiv
ity
,
ac
cu
r
ac
y
,
a
n
d
p
r
ec
is
io
n
a
r
e
f
r
e
q
u
en
tly
u
s
ed
to
g
au
g
e
th
e
class
if
ier
'
s
ef
f
ec
tiv
en
ess
.
Ov
er
all,
th
e
f
in
d
in
g
s
d
em
o
n
s
tr
ate
th
at
th
e
p
r
o
p
o
s
ed
GADCN
-
L
OA
in
co
n
ju
n
ctio
n
with
a
YOL
O
-
atten
tio
n
n
etwo
r
k
s
eg
m
en
tatio
n
m
o
d
el
p
r
o
d
u
ce
s
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
r
esu
l
ts
wh
en
co
m
p
ar
ed
to
th
e
o
th
er
d
ee
p
lear
n
in
g
m
o
d
els.
T
h
e
o
v
e
r
all
p
er
f
o
r
m
a
n
ce
ev
al
u
atio
n
r
esu
lts
o
f
th
e
s
u
g
g
ested
YOL
O
-
b
ased
GADCN
-
L
OA
tech
n
iq
u
e
em
p
lo
y
in
g
th
e
DDSM,
I
NB
r
ea
s
t,
an
d
MI
AS
d
atasets
ar
e
s
h
o
wn
in
Fig
u
r
es
8
an
d
9
.
T
o
as
ce
r
tain
th
e
ac
c
u
r
ate
d
etec
tio
n
o
u
tco
m
es
o
f
th
e
s
u
g
g
ested
class
if
ier
,
p
er
f
o
r
m
an
ce
p
ar
am
eter
s
ar
e
cr
ea
ted
a
n
d
e
x
am
in
ed
f
o
r
e
v
er
y
b
en
ch
m
ar
k
in
g
d
ataset
th
at
is
ac
ce
s
s
ib
le
to
th
e
g
e
n
er
al
p
u
b
lic.
Fo
r
ev
er
y
b
r
ea
s
t
im
ag
e
d
ataset
tak
en
in
t
o
co
n
s
id
er
atio
n
in
th
is
wo
r
k
,
th
e
esti
m
ated
f
i
n
d
in
g
s
d
e
m
o
n
s
tr
ate
th
at
t
h
e
s
u
g
g
ested
GADCN
-
L
OA
in
co
n
ju
n
ctio
n
with
a
YOL
O
s
eg
m
en
tatio
n
m
o
d
el
f
u
n
ctio
n
s
b
r
illi
an
tly
.
Usi
n
g
th
e
p
r
o
p
er
i
m
ag
e
n
o
r
m
aliza
tio
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
A
u
n
iq
u
e
YOLO
-
b
a
s
ed
g
a
ted
a
tten
tio
n
d
ee
p
co
n
v
o
lu
tio
n
n
et
w
o
r
k
-
Lich
ten
b
erg
…
(
V
in
o
th
R
a
th
in
a
m
)
1679
an
d
s
eg
m
e
n
tatio
n
tec
h
n
iq
u
es
g
r
ea
tly
e
n
h
an
ce
s
th
e
d
etec
tio
n
o
u
tco
m
es
in
t
h
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
.
Fig
u
r
e
1
0
v
alid
ates
an
d
co
m
p
a
r
es
th
e
p
r
ec
is
io
n
o
f
s
tan
d
ar
d
m
ac
h
in
e
lear
n
in
g
[
3
2
]
an
d
th
e
p
r
o
p
o
s
ed
GADCN
m
o
d
el
u
s
in
g
th
e
W
DB
C
d
ataset.
Si
m
ilar
to
th
is
,
t
h
e
W
DB
C
an
d
W
B
C
D
d
atasets
ar
e
u
s
ed
to
c
o
m
p
ar
e
th
e
e
x
is
tin
g
an
d
p
r
o
p
o
s
ed
class
if
icatio
n
m
o
d
els,
as
s
h
o
wn
in
Fig
u
r
e
1
1
.
I
n
th
is
p
ap
er
,
class
if
ier
ac
cu
r
a
cy
is
co
n
f
ir
m
e
d
an
d
co
m
p
ar
ed
f
o
r
th
is
an
aly
s
is
.
Fig
u
r
es
1
2
to
1
6
v
alid
ate
a
n
d
c
o
m
p
ar
e
th
e
ac
cu
r
ac
y
o
f
th
e
e
x
is
tin
g
[
3
2
]
,
[
3
3
]
a
n
d
p
r
o
p
o
s
ed
alg
o
r
ith
m
s
b
y
u
s
in
g
th
e
DDSM,
MI
AS,
an
d
I
NB
r
ea
s
t
d
atas
ets
r
esp
ec
tiv
ely
.
T
h
is
co
m
p
ar
is
o
n
r
esear
ch
r
ev
ea
ls
th
at
th
e
YOL
O
atten
tio
n
n
etw
o
r
k
s
eg
m
e
n
tatio
n
in
teg
r
ated
GADCN
-
L
OA
m
o
d
el
y
ield
s
b
etter
o
u
tco
m
es
th
an
th
e
o
th
er
m
o
d
els.
Ad
d
itio
n
ally
,
th
e
p
r
o
p
o
s
ed
GADCN
-
L
AO
m
o
d
el
is
co
n
tr
asted
with
s
o
m
e
o
th
er
co
n
tem
p
o
r
ar
y
b
r
ea
s
t
ca
n
ce
r
d
etec
tio
n
m
eth
o
d
o
lo
g
ies
u
s
in
g
th
e
DDSM
a
n
d
MI
AS
d
atas
ets,
as
d
ep
icted
in
Fig
u
r
es
1
5
an
d
1
6
,
r
esp
ec
ti
v
ely
.
Sin
ce
th
e
ac
tiv
atio
n
f
u
n
ctio
n
in
th
e
s
u
g
g
ested
class
if
icatio
n
m
o
d
el
is
ca
lcu
lated
u
s
in
g
th
e
b
est
p
o
s
s
ib
le
L
OA
o
p
tim
al
v
alu
e.
As
a
r
esu
lt,
th
e
s
u
g
g
ested
f
r
am
ewo
r
k
s
ig
n
if
ican
tly
im
p
r
o
v
es
t
h
e
class
if
ier
'
s
tr
ain
in
g
a
n
d
test
in
g
o
u
tco
m
es,
an
d
th
e
YOL
O
-
b
ased
GADCN
-
L
AO
ap
p
r
o
ac
h
o
u
tp
er
f
o
r
m
s
th
e
p
r
ev
i
o
u
s
m
o
d
els.
Als
o
,
it
s
u
p
p
o
r
ts
o
b
tain
i
n
g
an
av
er
a
g
e
ac
cu
r
ac
y
o
f
u
p
to
9
9
%
f
o
r
all
th
e
d
atasets
.
Fro
m
th
e
r
esu
lt
s
,
it
is
clea
r
th
at
th
e
p
r
o
p
o
s
ed
m
e
th
o
d
o
lo
g
y
ef
f
icien
tly
p
r
ed
icts
b
r
ea
s
t
ca
n
ce
r
b
y
em
p
lo
y
in
g
n
o
v
el
tech
n
iq
u
es
a
t
d
if
f
er
e
n
t
s
tag
es.
T
h
o
u
g
h
it
p
r
o
v
id
es
b
etter
r
esu
lts
,
th
e
n
ee
d
o
f
lar
g
e
d
ataset
is
r
eq
u
ir
ed
to
p
er
f
o
r
m
th
e
a
n
aly
s
is
.
Fu
r
th
er
,
a
d
v
an
ce
d
YOL
O
m
eth
o
d
s
ca
n
b
e
o
p
te
d
f
o
r
t
r
ain
in
g
a
n
d
p
r
o
ce
s
s
in
g
with
m
in
im
al
tim
e.
Fig
u
r
e
4
.
Per
f
o
r
m
an
c
e
ev
alu
at
io
n
u
s
in
g
th
e
MI
AS d
ataset
Fig
u
r
e
5
.
Pre
cisi
o
n
,
F1
-
s
co
r
e,
an
d
AUC an
aly
s
is
u
s
in
g
th
e
MI
AS d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.