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
0
4
~
5
6
1
5
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
6
.
pp
5
6
0
4
-
5
6
1
5
5604
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Dete
c
ting lung
no
dules in
co
mpute
d t
o
mo
g
ra
phy
ima
g
es ba
sed
o
n
d
eep learni
ng
L
a
m
T
ha
nh
H
ien
1
,
L
e
A
nh
T
u
2
,
P
ha
m
T
rung
H
ieu
3
,
P
h
a
m
M
inh
Duc
4
,
Ng
uy
en
Va
n Na
ng
3
,
Do
Na
ng
T
o
a
n
3
1
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
La
c
H
o
n
g
U
n
i
v
e
r
si
t
y
,
B
i
e
n
H
o
a
,
V
i
e
t
n
a
m
2
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
H
a
L
o
n
g
U
n
i
v
e
r
s
i
t
y
,
U
o
n
g
B
i
,
V
i
e
t
n
a
m
3
I
n
st
i
t
u
t
e
o
f
I
n
f
o
r
ma
t
i
o
n
Te
c
h
n
o
l
o
g
y
,
V
i
e
t
n
a
m
A
c
a
d
e
my
o
f
S
c
i
e
n
c
e
a
n
d
Te
c
h
n
o
l
o
g
y
,
H
a
n
o
i
,
V
i
e
t
n
a
m
4
C
o
m
p
u
t
e
r
V
i
s
i
o
n
a
n
d
R
o
b
o
t
i
c
s
La
b
o
r
a
t
o
r
y
,
I
n
t
e
r
n
a
t
i
o
n
a
l
S
c
h
o
o
l
,
V
i
e
t
n
a
m
N
a
t
i
o
n
a
l
U
n
i
v
e
r
si
t
y
,
H
a
n
o
i
,
V
i
e
t
n
a
m
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
5
,
2
0
2
5
R
ev
is
ed
J
u
l 1
9
,
2
0
2
5
Acc
ep
ted
Sep
1
4
,
2
0
2
5
Lu
n
g
c
a
n
c
e
r
is
c
u
rre
n
t
ly
re
c
o
g
n
i
z
e
d
a
s
o
n
e
o
f
t
h
e
m
o
st
d
a
n
g
e
ro
u
s
c
a
n
c
e
rs,
with
h
i
g
h
m
o
rtalit
y
ra
te.
In
o
r
d
e
r
to
d
e
a
l
with
l
u
n
g
c
a
n
c
e
r,
a
n
imp
o
rtan
t
tas
k
is
to
d
e
tec
t
l
u
n
g
n
o
d
u
les
e
a
rl
y
t
o
imp
r
o
v
e
p
a
ti
e
n
t
su
r
v
iv
a
l
ra
tes
,
a
n
d
c
o
m
p
u
ted
t
o
m
o
g
ra
p
h
y
(
CT
)
sc
a
n
s
a
re
c
ru
c
ial
d
a
ta
fo
r
th
is.
In
th
is
re
se
a
r
c
h
,
we
p
ro
p
o
se
a
d
e
e
p
lea
rn
in
g
-
b
a
se
d
m
e
th
o
d
f
o
r
d
e
tec
ti
n
g
lu
n
g
n
o
d
u
les
in
th
e
CT
ima
g
e
s with
t
h
e
g
o
a
l
o
f
in
c
re
a
sin
g
t
h
e
li
k
e
li
h
o
o
d
o
f
n
o
d
u
le ap
p
e
a
ra
n
c
e
in
th
e
in
p
u
t
d
a
ta
o
f
t
h
e
n
e
two
rk
,
m
a
k
in
g
it
e
a
sie
r
fo
r
th
e
m
o
d
e
l
t
o
fo
c
u
s
o
n
re
lev
a
n
t
a
re
a
s
wh
il
e
re
d
u
c
in
g
n
o
ise
fro
m
a
re
a
s
u
n
re
late
d
t
o
th
e
re
su
lt
.
S
p
e
c
ifi
c
a
ll
y
,
we
p
r
o
p
o
se
a
sim
p
le
lu
n
g
re
g
io
n
se
g
m
e
n
tatio
n
p
r
o
c
e
ss
a
n
d
o
p
ti
m
ize
t
h
e
h
y
p
e
rp
a
ra
m
e
ters
o
f
t
h
e
fa
ste
r
re
g
i
o
n
-
b
a
se
d
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(
fa
ste
r
R
-
CNN
)
m
o
d
e
l
b
a
se
d
o
n
th
e
a
n
a
ly
sis
o
f
n
o
d
u
le
c
h
a
ra
c
teristics
in
CT
ima
g
e
d
a
ta.
In
o
u
r
e
x
p
e
rime
n
ts,
to
e
v
a
lu
a
te
th
e
e
ffe
c
ti
v
e
n
e
ss
o
f
o
u
r
p
ro
p
o
sa
ls,
we
c
o
n
d
u
c
ted
tes
ts
o
n
t
h
e
sta
n
d
a
rd
LUNA
1
6
d
a
tas
e
t
with
d
iffere
n
t
b
a
c
k
b
o
n
e
c
o
n
fig
u
ra
ti
o
n
s
fo
r
th
e
m
o
d
e
l,
n
a
m
e
ly
R
es
N
e
t5
0
,
Re
sN
e
t5
0
v
2
,
a
n
d
M
o
b
il
e
Ne
t
.
Th
e
b
e
st
re
su
lt
s
a
c
h
i
e
v
e
d
we
re
0
.
8
6
m
AP5
0
a
n
d
0
.
9
1
Re
c
a
ll
fo
r
th
e
Re
s
N
e
t5
0
,
a
n
d
0
.
8
4
m
AP5
0
a
n
d
0
.
9
4
Re
c
a
ll
fo
r
th
e
Re
sN
e
t5
0
v
2
.
Th
e
se
imp
re
ss
iv
e
o
u
tco
m
e
s
u
n
d
e
r
sc
o
re
th
e
su
c
c
e
ss
o
f
o
u
r
m
e
th
o
d
a
n
d
e
sta
b
li
sh
a
r
o
b
u
st
b
a
sis
fo
r
f
u
tu
re
stu
d
ies
t
o
fu
rth
e
r
i
n
teg
ra
te AI
i
n
to
h
e
a
lt
h
c
a
r
e
so
lu
ti
o
n
s.
K
ey
w
o
r
d
s
:
C
T
im
ag
e
Dee
p
lear
n
in
g
Fin
e
-
tu
n
ed
f
aster
R
-
C
NN
L
u
n
g
n
o
d
u
le
d
etec
tio
n
L
u
n
g
s
eg
m
e
n
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
:
Do
Nan
g
T
o
a
n
I
n
s
titu
te
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
,
Vietn
a
m
Aca
d
em
y
o
f
S
cien
ce
an
d
T
ec
h
n
o
lo
g
y
1
8
Ho
an
g
Qu
o
c
Viet,
C
au
Gia
y
Dis
tr
ict,
Han
o
i1
0
0
7
2
,
Vietn
am
E
m
ail:
d
n
to
an
@
io
it.a
c.
v
n
1.
I
NT
RO
D
UCT
I
O
N
On
e
o
f
th
e
lead
in
g
ca
u
s
es
o
f
d
ea
th
wo
r
ld
wid
e
f
o
r
b
o
th
w
o
m
en
an
d
m
e
n
is
lu
n
g
ca
n
ce
r
[
1
]
.
T
h
is
is
a
s
er
io
u
s
illn
ess
ch
ar
ac
ter
ized
b
y
a
h
ig
h
r
ate
o
f
m
o
r
tality
,
m
a
in
ly
b
ec
au
s
e
it
is
o
f
ten
d
iag
n
o
s
ed
at
a
late
s
tag
e
wh
en
m
etastas
is
h
as o
cc
u
r
r
ed
o
r
th
er
e
is
s
ev
er
e
d
am
ag
e
to
o
th
er
o
r
g
a
n
s
in
th
e
b
o
d
y
.
Ho
we
v
er
,
ea
r
ly
d
etec
tio
n
o
f
lu
n
g
ca
n
ce
r
ca
n
lead
to
a
s
u
b
s
tan
tial
in
cr
ea
s
e
in
th
e
p
atie
n
t’
s
s
u
r
v
iv
al
r
ate
[
2
]
.
On
e
o
f
t
h
e
ea
r
lies
t
s
ig
n
s
o
f
lu
n
g
ca
n
ce
r
is
th
e
a
p
p
ea
r
a
n
c
e
o
f
r
o
u
n
d
o
r
o
v
al
-
s
h
ap
ed
tu
m
o
r
s
o
r
n
o
d
u
les
in
th
e
lu
n
g
s
[
3
]
.
T
h
ese
tu
m
o
r
s
g
en
er
ally
v
ar
y
in
s
ize,
r
an
g
in
g
f
r
o
m
j
u
s
t
a
f
ew
m
illi
m
eter
s
to
s
ev
er
al
ce
n
tim
eter
s
in
d
ia
m
eter
.
T
h
e
tu
m
o
r
o
r
lu
n
g
n
o
d
u
le
m
ay
h
a
v
e
ir
r
e
g
u
lar
o
r
u
n
clea
r
b
o
r
d
e
r
s
an
d
s
o
m
etim
es
ex
h
ib
it
a
s
u
r
r
o
u
n
d
in
g
s
h
a
d
o
w.
T
h
e
in
cid
en
ce
o
f
in
d
eter
m
in
ate
lu
n
g
n
o
d
u
les
h
as
b
ee
n
s
tead
ily
i
n
cr
ea
s
in
g
o
v
er
th
e
p
ast
f
ew
y
ea
r
s
[
4
]
.
Acc
o
r
d
in
g
to
th
e
Un
iv
er
s
ity
o
f
T
ex
as
So
u
th
wester
n
Me
d
ical
C
en
ter
[
5
]
,
ab
o
u
t
4
0
%
o
f
lu
n
g
n
o
d
u
les
ar
e
ca
n
ce
r
o
u
s
.
As
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
Dete
ctin
g
lu
n
g
n
o
d
u
les in
co
m
p
u
ted
to
mo
g
r
a
p
h
y
ima
g
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
(
La
m
Th
a
n
h
Hien
)
5605
m
en
tio
n
ed
i
n
th
e
d
ata
f
r
o
m
t
h
e
W
o
r
ld
Hea
lth
Or
g
a
n
izatio
n
(
W
HO)
[
6
]
,
a
p
p
r
o
x
im
ately
1
.
8
m
illi
o
n
d
ea
th
s
o
cc
u
r
r
e
d
d
u
e
to
lu
n
g
ca
n
ce
r
in
2
0
2
0
.
A
lu
n
g
n
o
d
u
le
is
d
ef
in
ed
as
a
s
h
ad
o
wed
ar
ea
,
wh
ich
ca
n
b
e
r
o
u
n
d
o
r
ir
r
eg
u
lar
i
n
s
h
ap
e,
with
a
d
iam
eter
r
an
g
in
g
f
r
o
m
3
m
m
to
3
cm
,
an
d
lo
ca
ted
with
in
th
e
air
-
f
illed
lu
n
g
[
7
]
.
Sh
ad
o
ws
with
a
d
iam
eter
s
m
aller
th
an
3
m
m
ar
e
r
ef
er
r
ed
to
as
m
icr
o
n
o
d
u
les,
wh
ile
s
h
ad
o
ws
with
a
d
iam
eter
lar
g
er
th
an
3
cm
ar
e
class
if
ied
as
tu
m
o
r
s
[
8
]
.
I
t is
wo
r
th
n
o
tin
g
th
at
ac
co
r
d
in
g
to
[
5
]
,
ab
o
u
t h
alf
o
f
p
atien
ts
tr
ea
ted
f
o
r
lu
n
g
n
o
d
u
le
s
with
ca
n
ce
r
ca
n
s
u
r
v
iv
e
at
least
f
iv
e
y
ea
r
s
af
ter
d
iag
n
o
s
is
.
F
u
r
th
er
m
o
r
e,
if
th
e
lu
n
g
n
o
d
u
le
is
1
cm
o
r
les
s
in
d
iam
eter
,
th
en
t
h
e
p
atien
t’
s
f
i
v
e
-
y
ea
r
s
u
r
v
iv
al
r
ate
ca
n
r
is
e
to
as
h
ig
h
as
8
0
%.
T
h
at
is
wh
y
th
e
task
o
f
ea
r
ly
d
etec
tio
n
o
f
l
u
n
g
n
o
d
u
les is
s
o
im
p
o
r
tan
t.
An
im
p
o
r
tan
t
to
o
l
f
o
r
th
e
task
o
f
d
etec
tin
g
lu
n
g
n
o
d
u
les
is
c
o
m
p
u
ted
to
m
o
g
r
ap
h
y
(
C
T
)
im
ag
in
g
.
C
T
im
ag
es
h
av
e
th
e
a
b
ilit
y
to
p
r
o
v
id
e
d
etailed
v
is
u
als
o
f
t
h
e
in
t
er
n
al
s
tr
u
ctu
r
es
o
f
t
h
e
b
o
d
y
,
wh
ich
h
elp
s
d
o
ct
o
r
s
id
en
tify
a
b
n
o
r
m
alities
,
in
clu
d
i
n
g
lu
n
g
n
o
d
u
les.
C
T
im
ag
es
o
f
f
er
h
ig
h
r
eso
lu
tio
n
a
n
d
m
u
lti
p
le
s
lices,
m
ak
in
g
it
p
o
s
s
ib
le
to
d
eter
m
in
e
th
e
l
o
ca
tio
n
,
s
ize,
an
d
s
h
ap
e
o
f
t
h
e
lu
n
g
n
o
d
u
les.
T
h
er
ef
o
r
e,
C
T
im
ag
es
ar
e
also
cr
u
cial
d
ata
f
o
r
in
f
o
r
m
atio
n
tech
n
o
lo
g
y
r
esear
ch
aim
ed
at
th
e
au
to
m
atic
d
etec
tio
n
o
f
lu
n
g
n
o
d
u
les,
th
er
eb
y
h
el
p
in
g
to
ac
ce
ler
ate
an
d
im
p
r
o
v
e
t
h
e
ac
cu
r
ac
y
o
f
m
ed
ical
d
iag
n
o
s
es.
T
h
e
s
tu
d
y
in
th
is
p
a
p
er
f
o
cu
s
es
o
n
th
e
is
s
u
e
o
f
d
etec
tin
g
l
u
n
g
n
o
d
u
les
in
C
T
im
ag
es
u
s
in
g
a
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
.
I
n
d
ee
p
lear
n
in
g
,
c
o
n
v
o
lu
tio
n
al
lay
e
r
s
with
in
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
ar
e
p
ar
ticu
lar
ly
ef
f
ec
tiv
e
at
ca
p
tu
r
in
g
s
p
atial
f
ea
tu
r
es
f
r
o
m
im
a
g
es,
in
clu
d
in
g
C
T
s
ca
n
s
.
T
h
is
is
a
k
ey
ad
v
an
tag
e
f
o
r
id
en
tify
i
n
g
lu
n
g
n
o
d
u
les
in
C
T
s
ca
n
s
,
wh
er
e
ch
allen
g
es
ar
is
e
in
d
is
tin
g
u
is
h
in
g
r
eg
io
n
s
d
u
e
to
co
m
p
lex
s
tr
u
ctu
r
al
co
n
d
itio
n
s
an
d
co
n
tr
ast
v
ar
iatio
n
s
b
etwe
en
r
eg
io
n
s
.
Ad
d
itio
n
ally
,
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
s
h
av
e
b
ee
n
p
r
o
v
en
to
b
e
ef
f
ec
tiv
e
to
o
ls
in
v
ar
io
u
s
m
ac
h
in
e
v
is
io
n
task
s
,
with
s
p
ec
if
ic
n
etwo
r
k
co
n
f
ig
u
r
atio
n
s
an
d
ap
p
licatio
n
co
n
d
itio
n
s
d
ep
en
d
in
g
o
n
th
e
ch
ar
ac
ter
is
tics
o
f
th
e
d
ata.
I
n
th
is
s
tu
d
y
,
th
e
f
o
cu
s
is
o
n
d
ev
elo
p
in
g
a
p
r
o
ce
s
s
f
o
r
d
etec
t
in
g
lu
n
g
n
o
d
u
les
in
C
T
im
ag
e
s
with
th
e
g
o
al
o
f
e
n
h
an
cin
g
th
e
lik
elih
o
o
d
o
f
n
o
d
u
le
ap
p
ea
r
an
ce
in
th
e
n
etwo
r
k
'
s
in
p
u
t
d
ata,
allo
win
g
th
e
m
o
d
el
t
o
m
o
r
e
ea
s
ily
f
o
c
u
s
o
n
r
elev
an
t
ar
ea
s
wh
ile
r
ed
u
ci
n
g
n
o
is
e
f
r
o
m
r
eg
i
o
n
s
u
n
r
elate
d
to
th
e
r
esu
lt.
Sp
ec
if
ically
,
th
e
s
tu
d
y
in
v
o
lv
es
f
in
e
-
t
u
n
in
g
th
e
fa
ste
r
re
g
io
n
-
b
a
se
d
CNN
(
F
aster
R
-
C
NN
)
m
o
d
el
with
in
th
e
co
n
tex
t
o
f
lu
n
g
n
o
d
u
le
d
ata
co
n
d
itio
n
s
in
C
T
im
ag
es.
W
e
also
ad
d
r
ess
th
e
d
ev
elo
p
m
en
t
o
f
im
ag
e
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
th
e
d
e
tectio
n
r
esu
lts
.
T
h
e
m
ain
co
n
t
r
ib
u
tio
n
s
o
f
th
e
p
a
p
er
in
clu
d
e
:
i
)
E
n
h
an
cin
g
th
e
ef
f
ec
tiv
en
ess
o
f
lu
n
g
n
o
d
u
le
d
etec
tio
n
b
y
p
r
o
p
o
s
in
g
a
lu
n
g
r
eg
io
n
s
eg
m
en
tatio
n
p
r
o
ce
s
s
to
h
elp
d
ee
p
lear
n
in
g
m
o
d
els
f
o
cu
s
o
n
p
r
o
ce
s
s
in
g
t
h
e
lu
n
g
ar
ea
;
ii
)
O
p
tim
izin
g
t
h
e
h
y
p
e
r
p
ar
am
eter
s
o
f
th
e
Fas
ter
R
-
C
N
N
m
o
d
el
b
ased
o
n
th
e
an
aly
s
is
o
f
lu
n
g
n
o
d
u
le
ch
a
r
ac
ter
is
tics
in
C
T
i
m
ag
e
d
ata;
an
d
iii
)
C
lar
if
y
in
g
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
als
b
y
im
p
lem
en
tin
g
an
d
ev
alu
atin
g
th
e
m
o
d
el
with
d
if
f
er
en
t
b
ac
k
b
o
n
es,
s
p
ec
if
ically
R
es
N
et5
0
,
R
esNet5
0
v
2
,
an
d
Mo
b
ileNet
.
T
h
e
s
tr
u
ctu
r
e
o
f
th
e
r
em
ain
i
n
g
p
ap
er
is
as:
s
ec
tio
n
2
co
v
er
s
r
elate
d
wo
r
k
s
,
s
ec
tio
n
3
p
r
esen
ts
th
e
d
ata
u
s
ed
an
d
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
o
lo
g
y
o
f
th
e
p
a
p
er
,
Secti
o
n
4
d
is
cu
s
s
es
th
e
ex
p
er
im
en
ts
an
d
th
e
ev
al
u
atio
n
o
f
th
e
r
esu
lts
o
b
tain
ed
.
Fin
ally
,
th
e
co
n
clu
s
io
n
is
p
r
esen
ted
.
2.
RE
L
AT
E
D
WO
RK
S
On
e
s
tu
d
y
o
n
l
u
n
g
n
o
d
u
le
d
et
ec
tio
n
in
C
T
im
ag
es
is
m
en
tio
n
ed
in
[
9
]
,
wh
er
e
th
e
r
esear
ch
in
tr
o
d
u
ce
s
a
m
eth
o
d
th
at
u
s
es
f
ilter
s
d
e
s
i
g
n
ed
to
en
h
a
n
ce
th
e
q
u
ality
o
f
d
o
t
-
lik
e
an
d
lin
e
-
lik
e
o
b
jects
in
two
-
d
im
en
s
io
n
al
s
p
ac
e.
T
h
e
s
tu
d
y
u
s
es
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
m
etr
ics
to
ev
alu
ate
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
f
ilter
s
.
r
esu
lts
o
n
th
e
lo
w
d
o
s
e
C
T
(
L
DC
T
)
d
at
aset
s
h
o
w
th
at
9
3
.
4
%
o
f
7
6
n
o
d
u
les
wer
e
d
etec
ted
,
with
a
f
alse
p
o
s
itiv
e
r
ate
o
f
4
.
2
.
T
h
e
s
tu
d
y
[
1
0
]
d
is
cu
s
s
es
an
au
t
o
m
ated
m
eth
o
d
f
o
r
d
etec
tin
g
lu
n
g
n
o
d
u
les
th
r
o
u
g
h
m
u
ltip
le
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
an
d
th
e
u
s
e
o
f
wav
elet
t
r
an
s
f
o
r
m
an
d
b
io
r
th
o
g
o
n
al
wav
elet
tech
n
iq
u
e
s
to
en
h
a
n
ce
im
a
g
e
q
u
ality
.
First,
th
e
bi
-
h
is
to
g
r
a
m
eq
u
aliza
tio
n
alg
o
r
ith
m
is
u
s
ed
to
b
ala
n
ce
th
e
c
o
n
tr
ast
o
f
th
e
im
a
g
e.
Nex
t,
m
o
r
p
h
o
lo
g
ical
tr
an
s
f
o
r
m
atio
n
s
ar
e
ap
p
lied
to
clea
n
an
d
s
eg
m
en
t
th
e
im
ag
e,
h
elp
in
g
t
o
s
ep
ar
ate
im
p
o
r
tan
t
s
tr
u
ctu
r
es
in
th
e
lu
n
g
s
.
T
h
e
s
e
g
m
en
ted
r
e
g
io
n
s
ar
e
t
h
en
ex
t
r
ac
ted
an
d
f
e
d
in
to
a
f
u
zz
y
in
f
e
r
en
ce
s
y
s
tem
(
FIS)
.
T
h
is
FIS
is
u
s
ed
to
d
eter
m
in
e
th
e
s
ev
er
ity
o
f
l
u
n
g
n
o
d
u
le
s
,
aid
in
g
in
th
e
class
if
icatio
n
an
d
e
v
alu
atio
n
o
f
tu
m
o
r
s
.
T
h
is
m
et
h
o
d
was
test
ed
o
n
d
ata
f
r
o
m
2
5
p
atien
ts
.
T
h
e
r
esu
lts
s
h
o
w
th
at
th
e
m
eth
o
d
h
as
th
e
ab
ilit
y
to
d
etec
t
lu
n
g
ca
n
ce
r
ea
r
ly
,
c
o
n
f
ir
m
in
g
th
e
p
o
ten
tial
o
f
w
av
elet
tr
an
s
f
o
r
m
atio
n
tech
n
iq
u
es,
B
i
-
His
to
g
r
am
E
q
u
aliza
tio
n
,
an
d
m
o
r
p
h
o
lo
g
i
ca
l f
ilter
s
in
th
e
au
to
m
atic
id
e
n
tific
atio
n
o
f
l
u
n
g
n
o
d
u
les.
I
n
ad
d
itio
n
t
o
th
e
s
tu
d
ies
m
en
tio
n
ed
ab
o
v
e,
th
er
e
a
r
e
s
ev
er
a
l
s
tu
d
ies
th
at
u
s
e
d
ee
p
lear
n
in
g
to
d
etec
t
lu
n
g
n
o
d
u
les
in
C
T
im
ag
es.
I
n
2
0
1
6
,
a
s
tu
d
y
[
1
1
]
was
co
n
d
u
cted
to
d
etec
t
lu
n
g
n
o
d
u
les
in
C
T
im
ag
es
u
s
in
g
a
3
D
C
NN.
T
h
is
m
eth
o
d
ap
p
lies
d
ee
p
lear
n
in
g
tech
n
iq
u
es
to
an
aly
ze
co
m
p
lex
m
e
d
ical
im
ag
e
d
ata,
s
p
ec
if
ically
ch
est
C
T
im
ag
es.
T
h
e
d
ataset
u
s
ed
i
n
th
e
s
tu
d
y
is
th
e
L
I
D
C
-
I
DR
I
d
ataset,
wh
ich
i
n
clu
d
es
1
,
0
1
8
c
h
est
C
T
im
ag
es.
T
h
e
p
er
f
o
r
m
an
ce
m
et
r
ics
u
s
ed
to
ev
alu
ate
th
e
m
eth
o
d
in
clu
d
e
s
en
s
itiv
ity
,
f
alse
p
o
s
itiv
e
r
ate,
an
d
th
e
f
r
ee
-
r
esp
o
n
s
e
r
ec
eiv
er
o
p
e
r
atin
g
ch
ar
ac
ter
is
tic
(
FR
OC
)
cu
r
v
e.
T
h
e
r
esu
lts
in
d
icate
d
th
at
th
e
s
y
s
tem
r
ea
ch
ed
a
s
en
s
itiv
ity
o
f
7
8
.
9
%,
with
an
av
er
ag
e
o
f
2
0
f
alse
p
o
s
itiv
es
p
er
s
ca
n
(
FP
s
/s
ca
n
)
.
T
h
is
d
em
o
n
s
tr
ates
th
at
th
e
s
y
s
tem
h
as
a
g
o
o
d
s
en
s
itiv
ity
f
o
r
d
etec
tin
g
lu
n
g
n
o
d
u
les,
alth
o
u
g
h
th
e
n
u
m
b
er
o
f
f
als
e
p
o
s
itiv
es
m
ay
b
e
h
ig
h
er
c
o
m
p
ar
e
d
to
s
o
m
e
o
th
er
m
eth
o
d
s
.
A
n
o
tewo
r
t
h
y
p
o
in
t
is
th
at
th
is
s
y
s
tem
d
o
es
n
o
t
u
s
e
ad
d
itio
n
a
l
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
0
4
-
5
6
1
5
5606
s
eg
m
en
tatio
n
p
r
o
ce
s
s
es
o
r
f
al
s
e
p
o
s
itiv
e
r
e
d
u
ctio
n
tech
n
i
q
u
es,
wh
ich
m
ay
co
n
tr
ib
u
te
to
th
e
h
ig
h
e
r
n
u
m
b
er
o
f
f
alse
p
o
s
itiv
es.
Ho
wev
er
,
th
e
s
tu
d
y
co
n
f
ir
m
ed
th
e
p
o
te
n
tial
o
f
th
e
3
D
C
NN
in
lu
n
g
n
o
d
u
le
d
etec
tio
n
,
o
p
e
n
in
g
u
p
a
n
ew
d
ir
ec
tio
n
f
o
r
im
p
r
o
v
in
g
au
to
m
ated
d
iag
n
o
s
tic
s
y
s
tem
s
.
[
1
2
]
d
is
cu
s
s
es
a
s
tu
d
y
o
n
th
e
d
ev
el
o
p
m
en
t
an
d
ev
alu
atio
n
o
f
a
c
o
m
p
u
ter
-
aid
ed
d
etec
tio
n
(
C
AD
)
s
y
s
tem
b
ased
o
n
2
D
C
NN
an
d
C
T
i
m
ag
e
s
eg
m
en
tatio
n
tech
n
iq
u
es
to
d
etec
t
lu
n
g
n
o
d
u
les.
T
h
e
d
ataset
u
s
ed
in
th
e
s
tu
d
y
is
th
e
L
I
DC
-
I
DR
I
d
ata
s
et,
wh
ich
in
clu
d
es
1
,
0
1
8
ch
est
C
T
im
ag
es.
T
o
ass
es
s
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
C
AD
s
y
s
tem
,
th
e
s
tu
d
y
u
s
es
m
etr
ics
s
u
ch
as
s
en
s
itiv
ity
,
ac
cu
r
ac
y
,
an
d
FP
s
/s
ca
n
.
An
f
r
ee
-
r
esp
o
n
s
e
o
p
e
r
atin
g
ch
ar
ac
ter
is
tic
(
FR
OC
)
an
aly
s
is
was
al
s
o
co
n
d
u
cte
d
to
m
ea
s
u
r
e
s
en
s
itiv
ity
an
d
th
e
FP
s
/s
ca
n
r
ate,
e
n
ab
lin
g
a
co
m
p
ar
is
o
n
o
f
p
er
f
o
r
m
an
ce
b
etwe
en
d
if
f
er
en
t CAD sy
s
tem
s
.
T
h
e
r
esu
lts
o
f
th
e
s
tu
d
y
s
h
o
w
th
at
t
h
e
p
r
o
p
o
s
ed
C
AD
s
y
s
tem
ac
h
iev
ed
a
s
en
s
itiv
ity
o
f
9
2
.
8
%
with
8
f
alse
p
o
s
itiv
es
p
er
s
ca
n
.
T
h
is
is
a
v
er
y
p
r
o
m
is
in
g
r
esu
lt
wh
en
co
m
p
a
r
ed
to
o
th
e
r
s
tu
d
ies,
d
em
o
n
s
tr
atin
g
th
at
th
e
s
y
s
tem
ca
n
d
etec
t
lu
n
g
n
o
d
u
les
with
h
ig
h
ac
cu
r
ac
y
an
d
f
ewe
r
f
alse
p
o
s
itiv
es.
T
h
e
s
tu
d
y
also
co
m
p
ar
es
th
e
p
er
f
o
r
m
an
ce
o
f
d
if
f
er
e
n
t
C
NN
ar
ch
itectu
r
es,
in
clu
d
in
g
Alex
Ne
t,
Go
o
g
L
eNe
t,
an
d
R
esNet.
I
n
2
0
2
1
,
a
s
tu
d
y
[
1
3
]
co
n
d
u
cted
in
Vietn
am
u
s
ed
v
ar
io
u
s
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
s
u
ch
as
AT
T
,
ASS,
an
d
AST
t
o
d
etec
t
t
u
m
o
r
s
in
lu
n
g
C
T
im
ag
es.
T
h
e
s
tu
d
y
u
s
ed
th
e
L
I
D
C
-
I
DR
I
d
ataset
f
o
r
ev
alu
atio
n
.
Pre
cisi
o
n
,
r
ec
all,
an
d
s
p
ec
if
icity
wer
e
em
p
lo
y
e
d
to
ev
alu
ate
t
h
e
ef
f
ec
tiv
e
n
ess
o
f
th
e
C
NNs.
T
h
e
r
esu
lts
o
f
th
e
s
tu
d
y
s
h
o
wed
im
p
r
ess
iv
e
o
u
tco
m
es,
with
a
p
r
e
cisi
o
n
o
f
9
5
%,
r
ec
all
o
f
8
6
.
4
%
,
an
d
s
p
ec
if
icity
o
f
9
8
.
9
%.
T
h
is
o
u
tco
m
e
r
e
p
r
esen
ts
a
m
ajo
r
s
tep
f
o
r
war
d
in
d
ev
elo
p
in
g
au
to
m
ated
tec
h
n
iq
u
es
f
o
r
tu
m
o
r
d
etec
tio
n
in
lu
n
g
C
T
s
ca
n
s
,
co
n
tr
ib
u
tin
g
t
o
en
h
a
n
ce
d
ac
c
u
r
a
cy
in
d
iag
n
o
s
is
an
d
im
p
r
o
v
ed
tr
ea
tm
en
t
p
lan
n
in
g
.
Als
o
in
Vietn
am
,
a
m
eth
o
d
u
s
i
n
g
th
e
L
d
cNe
t
n
etwo
r
k
[
1
4
]
w
as
im
p
lem
en
ted
to
class
if
y
l
u
n
g
C
T
im
ag
es
f
r
o
m
th
e
L
I
DC
-
I
DR
I
d
ataset.
T
o
ass
ess
th
e
m
o
d
el’
s
ef
f
ec
ti
v
en
ess
,
ev
alu
atio
n
m
etr
ics
i
n
clu
d
in
g
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
an
d
s
p
ec
if
icity
wer
e
u
tili
ze
d
.
T
h
e
r
esu
lts
o
f
th
e
s
t
u
d
y
s
h
o
wed
ac
cu
r
ac
y
,
s
p
ec
if
ic
ity
,
an
d
s
en
s
itiv
ity
v
alu
es
o
f
9
7
.
2
%,
9
7
.
3
%,
an
d
9
6
.
0
%,
r
esp
ec
tiv
el
y
.
T
h
is
d
e
m
o
n
s
tr
ates
th
at
th
e
L
d
cNe
t
n
e
two
r
k
ac
h
iev
e
d
v
e
r
y
h
ig
h
p
e
r
f
o
r
m
an
ce
in
class
if
y
in
g
lu
n
g
im
ag
es
f
r
o
m
th
e
L
I
DC
-
I
DR
I
d
ataset.
T
h
ese
r
esu
lts
co
u
ld
p
r
o
v
id
e
s
ig
n
if
ican
t b
en
ef
its
in
s
u
p
p
o
r
ti
n
g
th
e
d
iag
n
o
s
is
an
d
tr
ea
tm
e
n
t
o
f
lu
n
g
d
is
ea
s
es.
3.
M
AT
E
R
I
AL
S AN
D
M
E
T
H
O
DS
3
.
1
.
Da
t
a
s
et
us
ed
I
n
th
is
s
tu
d
y
,
t
h
e
d
ataset
u
s
e
d
is
L
UNA1
6
[
1
5
]
,
wh
ic
h
co
n
s
is
ts
o
f
8
8
8
C
T
s
lices.
T
h
e
L
UNA1
6
d
ataset
was
d
ev
elo
p
ed
f
r
o
m
th
e
lu
n
g
im
a
g
e
d
atab
ase
c
o
n
s
o
r
tiu
m
an
d
im
a
g
e
d
atab
ase
r
eso
u
r
ce
in
itiativ
e
(
L
I
DC
-
I
DR
I
)
d
ataset,
a
lar
g
e
d
atab
ase
co
n
tain
in
g
lu
n
g
C
T
i
m
ag
es
f
o
r
an
aly
s
is
.
L
I
DC
-
I
DR
I
is
u
s
ed
f
o
r
lu
n
g
ca
n
ce
r
d
ia
g
n
o
s
is
r
esear
ch
,
an
d
L
UNA1
6
is
a
n
ex
ten
s
io
n
o
f
th
is
p
r
o
ject,
f
o
c
u
s
in
g
o
n
lu
n
g
n
o
d
u
le
d
etec
tio
n
.
E
ac
h
im
ag
e
in
L
UNA1
6
h
as
p
ix
el
v
alu
es
r
an
g
in
g
f
r
o
m
−
1
0
0
0
t
o
3
0
0
0
,
m
ea
s
u
r
e
d
in
Ho
u
n
s
f
ield
u
n
its
(
HU)
,
wh
ich
r
ep
r
esen
t
th
e
d
en
s
ity
o
f
tis
s
u
es
in
th
e
b
o
d
y
.
T
h
e
L
UNA1
6
d
ataset
is
wid
ely
u
s
ed
in
s
tu
d
ies
o
n
d
etec
tin
g
m
alig
n
an
t
l
u
n
g
n
o
d
u
les
f
r
o
m
C
T
s
ca
n
d
ata.
T
h
ese
s
lices
h
av
e
a
r
elativ
ely
h
ig
h
s
p
atial
r
eso
lu
tio
n
,
with
a
s
ize
o
f
5
1
2
×5
1
2
p
ix
els.
Ad
d
itio
n
ally
,
th
is
d
ataset
in
cl
u
d
es
lab
els
in
d
icatin
g
th
e
lo
ca
tio
n
an
d
s
ize
o
f
th
e
lu
n
g
n
o
d
u
les,
wh
ich
wer
e
a
n
n
o
tated
b
y
s
ev
er
al
m
e
d
ical
ex
p
er
ts
,
s
p
ec
if
ically
r
ad
io
lo
g
is
ts
.
T
h
is
en
ab
les
d
ee
p
lear
n
in
g
m
o
d
els
to
b
e
tr
ain
e
d
an
d
ev
alu
ated
o
n
ac
cu
r
ate
m
e
d
ical
d
iag
n
o
s
tic
d
ata.
I
n
d
etail,
th
er
e
ar
e
a
to
tal
o
f
1
,
1
8
6
lu
n
g
n
o
d
u
les in
th
e
8
8
8
C
T
im
ag
es.
Fig
u
r
e
1
illu
s
tr
ates a
C
T
s
lice.
Fig
u
r
e
1
.
A
C
T
s
lice
o
f
a
lu
n
g
f
r
o
m
th
e
L
UNA1
6
d
ataset
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
Dete
ctin
g
lu
n
g
n
o
d
u
les in
co
m
p
u
ted
to
mo
g
r
a
p
h
y
ima
g
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
(
La
m
Th
a
n
h
Hien
)
5607
3
.
2
.
Da
t
a
prepro
ce
s
s
ing
3
.
2
.
1
.
L
un
g
s
eg
m
ent
a
t
io
n
Acc
o
r
d
in
g
to
th
e
d
ef
i
n
itio
n
s
o
f
lu
n
g
n
o
d
u
les
p
r
esen
ted
in
s
ec
tio
n
1
,
lu
n
g
n
o
d
u
les
ar
e
alwa
y
s
f
u
lly
lo
ca
ted
with
in
t
h
e
ar
ea
o
f
th
e
two
lu
n
g
s
.
Fo
r
t
h
is
r
ea
s
o
n
,
ac
cu
r
ately
id
e
n
tify
in
g
th
e
p
o
s
itio
n
o
f
th
e
two
lu
n
g
s
in
C
T
im
ag
es
b
ec
o
m
es
a
cr
u
cial
f
ac
to
r
in
d
etec
tin
g
lu
n
g
n
o
d
u
les.
T
o
en
h
an
ce
th
e
ab
il
ity
to
d
etec
t
lu
n
g
n
o
d
u
les,
we
p
r
o
p
o
s
e
a
lu
n
g
r
eg
io
n
s
eg
m
en
tatio
n
tech
n
i
q
u
e
,
with
th
e
m
ain
g
o
al
b
ein
g
to
s
ep
ar
ate
th
e
lu
n
g
r
eg
io
n
f
r
o
m
th
e
s
u
r
r
o
u
n
d
in
g
ar
ea
s
.
L
u
n
g
s
eg
m
en
tatio
n
allo
ws
th
e
m
o
d
el
to
f
o
cu
s
o
n
im
p
o
r
tan
t
f
ea
tu
r
es
an
d
elim
in
ate
u
n
n
ec
ess
ar
y
ar
ea
s
,
th
er
eb
y
im
p
r
o
v
in
g
th
e
a
cc
u
r
ac
y
o
f
lo
ca
tin
g
lu
n
g
n
o
d
u
les.
T
h
e
l
u
n
g
s
eg
m
en
tatio
n
p
r
o
ce
s
s
is
ca
r
r
ied
o
u
t in
th
e
f
o
llo
win
g
s
tep
s
in
Fig
u
r
e
2
.
Fig
u
r
e
2
.
Diag
r
a
m
o
f
th
e
lu
n
g
s
eg
m
en
tatio
n
s
tep
s
Step
1
is
co
n
v
er
tin
g
th
e
C
T
i
m
ag
e
in
to
a
b
in
ar
y
im
ag
e
b
y
u
s
in
g
a
s
p
ec
if
ic
th
r
esh
o
ld
.
I
n
th
is
ca
s
e,
th
e
th
r
esh
o
ld
v
alu
e
is
s
elec
ted
as
-
6
0
4
HU.
Pix
els
with
v
alu
es
lo
wer
th
an
th
is
th
r
esh
o
ld
ar
e
ass
ig
n
ed
a
v
alu
e
o
f
1
,
an
d
th
e
r
em
ain
in
g
p
ix
els
ar
e
a
s
s
ig
n
ed
a
v
alu
e
o
f
0
(
b
ac
k
g
r
o
u
n
d
)
.
T
h
e
p
ix
els
with
a
v
alu
e
o
f
1
ar
e
tem
p
o
r
ar
ily
co
n
s
id
er
ed
as
th
e
lu
n
g
r
eg
io
n
.
T
h
e
alg
o
r
ith
m
tak
es
a
C
T
i
m
ag
e
an
d
th
e
th
r
esh
o
l
d
as
in
p
u
t
an
d
r
et
u
r
n
s
th
e
co
r
r
esp
o
n
d
in
g
b
in
ar
y
im
a
g
e.
First,
cr
ea
te
an
em
p
ty
b
in
ar
y
im
ag
e
o
f
th
e
s
am
e
s
ize
as
th
e
in
p
u
t
C
T
im
ag
e.
T
h
en
,
iter
ate
th
r
o
u
g
h
ea
ch
p
ix
el,
an
d
if
th
e
p
i
x
el
v
alu
e
is
s
m
aller
th
an
th
e
th
r
esh
o
l
d
,
ass
ig
n
a
v
alu
e
o
f
1
to
th
e
co
r
r
esp
o
n
d
in
g
p
ix
el
in
th
e
b
in
ar
y
im
ag
e;
o
t
h
er
wis
e,
ass
ig
n
a
v
alu
e
o
f
0
.
Nex
t
is
s
tep
2
.
Af
ter
b
in
ar
izin
g
th
e
im
ag
e,
th
e
lu
n
g
r
eg
i
o
n
i
s
tem
p
o
r
ar
ily
co
n
s
id
er
e
d
to
b
e
th
e
ar
ea
wh
er
e
th
e
p
ix
el
v
alu
es a
r
e
eq
u
al
to
1
.
Ho
wev
er
,
ar
ea
s
o
u
ts
id
e
th
e
b
o
d
y
will a
ls
o
h
av
e
p
i
x
el
v
alu
es o
f
1
,
as th
is
r
eg
io
n
is
also
air
,
lik
e
th
e
l
u
n
g
s
.
T
h
er
ef
o
r
e,
th
ese
ar
ea
s
n
ee
d
to
b
e
co
n
v
er
ted
to
b
ac
k
g
r
o
u
n
d
,
co
r
r
esp
o
n
d
in
g
to
p
ix
el
v
alu
es
o
f
0
.
Step
2
is
u
s
ed
to
elim
in
ate
t
h
e
b
o
u
n
d
ar
y
ar
ea
s
an
d
r
etain
th
e
lu
n
g
r
eg
io
n
.
T
h
e
alg
o
r
ith
m
tak
es
a
b
in
ar
y
im
a
g
e
an
d
a
p
ad
d
in
g
s
ize
as
in
p
u
t
a
n
d
r
etu
r
n
s
th
e
co
r
r
esp
o
n
d
in
g
im
ag
e
with
th
e
b
o
u
n
d
ar
y
r
eg
io
n
s
r
e
m
o
v
e
d
.
First,
we
n
e
ed
to
ch
ec
k
if
th
e
p
ad
d
in
g
s
iz
e
is
n
eg
ativ
e;
if
it
is
,
th
r
o
w
a
n
ex
ce
p
tio
n
.
T
h
en
,
ad
d
p
a
d
d
in
g
to
th
e
im
ag
e
wit
h
th
e
co
r
r
esp
o
n
d
in
g
s
ize,
a
n
d
s
et
th
e
p
ix
el
v
alu
es
o
f
th
e
a
d
d
ed
p
ad
d
in
g
to
1
.
Nex
t,
we
u
s
e
th
e
alg
o
r
ith
m
b
y
W
u
et
a
l.
[
1
6
]
to
id
en
tif
y
t
h
e
co
n
n
ec
ted
c
o
m
p
o
n
en
ts
in
th
e
im
ag
e
o
b
tain
ed
f
r
o
m
th
e
p
r
ev
io
u
s
s
tep
.
T
h
e
alg
o
r
ith
m
b
y
Kesh
en
g
W
u
is
f
ast
an
d
p
r
o
v
i
d
es
h
ig
h
ac
cu
r
ac
y
f
o
r
c
o
n
n
ec
te
d
co
m
p
o
n
en
ts
.
T
h
en
,
th
e
p
ix
els
at
th
e
b
o
u
n
d
ar
y
ar
e
ass
ig
n
ed
a
v
alu
e
o
f
0
to
b
ec
o
m
e
th
e
b
ac
k
g
r
o
u
n
d
.
T
h
e
b
o
u
n
d
ar
y
ar
ea
is
d
ef
in
ed
as th
e
ar
ea
wh
er
e
at
least o
n
e
p
ix
el
to
u
ch
es th
e
ed
g
e
o
f
th
e
im
ag
e
(
wh
ich
co
u
ld
b
e
at
th
e
to
p
,
b
o
tto
m
,
lef
tm
o
s
t,
o
r
r
i
g
h
tm
o
s
t
ed
g
e
)
.
Fin
ally
,
i
f
p
ad
d
in
g
was
ad
d
e
d
in
th
e
p
r
ev
io
u
s
s
tep
,
th
e
p
ad
d
in
g
will
b
e
r
em
o
v
ed
to
r
etu
r
n
th
e
im
ag
e
to
its
o
r
ig
in
al
s
ize.
T
h
e
alg
o
r
ith
m
will
r
etu
r
n
th
e
co
r
r
esp
o
n
d
i
n
g
im
ag
e
af
ter
th
e
b
o
u
n
d
ar
y
r
eg
io
n
s
h
a
v
e
b
ee
n
r
em
o
v
ed
.
Nex
t
is
s
tep
3
.
L
ab
el
th
e
co
n
n
ec
ted
co
m
p
o
n
en
ts
to
id
e
n
tif
y
th
e
r
em
ain
i
n
g
f
o
r
eg
r
o
u
n
d
a
r
ea
s
in
th
e
im
ag
e
af
ter
r
em
o
v
in
g
th
e
b
o
u
n
d
ar
y
r
eg
i
o
n
s
.
Step
3
p
er
f
o
r
m
s
th
is
task
.
First,
Kesh
en
g
W
u
’
s
alg
o
r
ith
m
is
u
s
ed
ag
ain
to
id
en
tif
y
th
e
co
n
n
ec
te
d
co
m
p
o
n
e
n
ts
in
t
h
e
b
i
n
ar
y
i
m
ag
e.
T
h
e
r
esu
lt
in
clu
d
es
a
li
s
t
o
f
r
e
g
io
n
s
,
la
b
els
f
o
r
ea
c
h
c
o
r
r
esp
o
n
d
in
g
r
e
g
io
n
,
in
f
o
r
m
atio
n
ab
o
u
t
th
e
s
ize,
a
n
d
th
e
co
o
r
d
in
ates o
f
t
h
e
b
o
u
n
d
in
g
b
o
x
es o
f
th
ese
r
eg
io
n
s
.
B
ased
o
n
th
e
in
f
o
r
m
a
tio
n
f
r
o
m
th
e
b
o
u
n
d
in
g
b
o
x
es,
th
e
co
n
n
ec
te
d
co
m
p
o
n
e
n
ts
ar
e
d
r
awn
f
o
r
a
v
is
u
al
r
ep
r
esen
tatio
n
.
T
h
e
lis
t
o
f
c
o
n
n
ec
ted
c
o
m
p
o
n
en
ts
is
iter
ate
d
th
r
o
u
g
h
,
a
n
d
a
co
r
r
esp
o
n
d
i
n
g
b
o
u
n
d
in
g
b
o
x
is
d
r
awn
ar
o
u
n
d
ea
ch
r
eg
i
o
n
o
n
th
e
b
i
n
ar
y
im
ag
e.
T
h
is
is
d
o
n
e
b
y
u
s
in
g
a
r
ec
ta
n
g
le
to
s
u
r
r
o
u
n
d
ea
ch
r
eg
i
o
n
,
with
th
e
r
ec
tan
g
le
co
l
o
r
s
et
to
g
r
ee
n
an
d
th
e
lin
e
wid
t
h
s
et
to
2
p
ix
els.
Nex
t
is
s
tep
4
.
I
t
ca
n
b
e
s
ee
n
th
at
r
e
m
o
v
in
g
th
e
b
o
r
d
er
m
ig
h
t
n
o
t
n
ec
ess
ar
ily
k
ee
p
o
n
ly
th
e
lu
n
g
in
f
o
r
m
atio
n
,
as
it
co
u
l
d
also
c
o
n
tain
o
th
e
r
r
e
g
io
n
s
.
Fo
r
ex
a
m
p
le,
f
at
ar
ea
s
u
s
u
ally
h
a
v
e
l
o
w
p
ix
el
v
al
u
es,
s
o
th
r
o
u
g
h
th
e
b
i
n
ar
y
im
a
g
e
p
r
o
c
ess
,
th
ese
r
eg
io
n
s
will
h
av
e
p
ix
el
v
alu
es
o
f
1
,
m
ix
in
g
with
th
e
ar
ea
s
co
n
tain
in
g
air
.
T
h
e
lu
n
g
r
eg
i
o
n
is
id
en
tif
ied
b
y
ca
lcu
latin
g
th
e
ar
ea
o
f
ea
ch
co
n
n
ec
te
d
r
eg
i
o
n
an
d
o
n
ly
k
ee
p
in
g
th
e
two
lar
g
est ar
ea
s
,
co
r
r
esp
o
n
d
in
g
to
th
e
two
lu
n
g
s
.
Nex
t
is
s
tep
5
.
Fil
lin
g
th
e
h
o
les
in
th
e
lu
n
g
r
eg
io
n
h
elp
s
en
s
u
r
e
th
at
th
e
lu
n
g
ar
ea
is
co
n
t
in
u
o
u
s
an
d
n
o
t f
r
ag
m
e
n
ted
,
allo
win
g
th
e
r
esu
ltin
g
lu
n
g
m
ask
to
co
v
e
r
all
th
e
in
f
o
r
m
atio
n
with
in
th
e
lu
n
g
s
.
Step
5
f
ills
th
e
h
o
les
with
in
th
e
lu
n
g
ar
ea
,
wh
ich
co
r
r
esp
o
n
d
to
p
o
s
itio
n
s
with
p
ix
el
v
alu
es
o
f
0
.
First,
t
h
e
C
an
n
y
alg
o
r
ith
m
[
1
7
]
is
ap
p
lied
to
th
e
m
ask
to
cr
ea
te
a
n
ew
im
ag
e
c
o
n
tain
in
g
o
n
ly
th
e
o
u
tlin
es.
T
h
e
n
,
m
o
r
p
h
o
lo
g
ical
clo
s
in
g
is
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
0
4
-
5
6
1
5
5608
ap
p
lied
to
clo
s
e
th
e
s
m
all
h
o
les
in
th
e
m
ask
.
T
h
e
alg
o
r
ith
m
co
n
tin
u
es
b
y
f
in
d
in
g
c
o
n
n
ec
ted
r
eg
io
n
s
in
th
e
m
ask
an
d
f
illi
n
g
th
e
s
m
all
h
o
l
es
b
y
co
m
p
ar
in
g
th
eir
ar
ea
wi
th
th
e
ma
x
_
h
o
le_
s
iz
e
(
a
v
alu
e
th
at
is
p
r
ed
ef
i
n
ed
)
.
T
h
e
r
esu
lt is
a
lu
n
g
m
ask
with
th
e
s
m
all
h
o
les f
illed
.
Nex
t
is
s
tep
6
.
T
h
e
cr
ea
tio
n
o
f
a
co
n
v
ex
h
u
ll
f
o
r
th
e
lu
n
g
s
h
elp
s
elim
in
ate
s
m
all
in
d
en
tati
o
n
s
ar
o
u
n
d
th
e
s
u
r
f
ac
e
o
f
th
e
l
u
n
g
s
.
First,
we
f
in
d
t
h
e
co
n
t
o
u
r
s
in
t
h
e
i
m
ag
e
u
s
in
g
t
h
e
alg
o
r
ith
m
b
y
Sato
s
h
i
Su
zu
k
i
et
al.
[
1
8
]
.
Sin
ce
n
o
n
-
lu
n
g
ar
ea
s
o
r
n
o
is
e
with
in
th
e
lu
n
g
s
h
av
e
b
ee
n
r
em
o
v
ed
in
p
r
ev
io
u
s
s
tep
s
,
o
n
ly
th
e
two
co
n
to
u
r
s
o
f
t
h
e
two
lu
n
g
s
r
e
m
ain
.
T
h
e
n
,
t
h
e
alg
o
r
ith
m
ite
r
ates
th
r
o
u
g
h
ea
c
h
co
n
to
u
r
f
o
u
n
d
i
n
th
e
p
r
e
v
io
u
s
s
tep
an
d
co
m
p
u
tes
th
e
co
n
v
ex
h
u
ll
f
o
r
ea
ch
c
o
r
r
esp
o
n
d
in
g
c
o
n
to
u
r
.
T
h
is
co
n
v
ex
h
u
ll
ca
lcu
latio
n
is
p
er
f
o
r
m
ed
u
s
in
g
Sk
lan
s
k
y
'
s
alg
o
r
ith
m
[
1
9
]
.
Af
ter
war
d
,
th
e
p
i
x
els
with
in
th
e
co
n
v
ex
h
u
ll
ar
e
ass
ig
n
ed
a
v
alu
e
o
f
1
to
cr
ea
te
a
lu
n
g
m
ask
.
Fin
ally
,
th
e
b
in
ar
y
m
ask
is
o
v
er
laid
o
n
th
e
o
r
i
g
in
al
C
T
im
ag
e
to
elim
in
ate
ar
ea
s
o
u
ts
id
e
th
e
lu
n
g
s
,
r
etain
in
g
o
n
ly
th
e
r
ele
v
an
t
lu
n
g
r
eg
i
o
n
s
.
T
h
r
o
u
g
h
t
h
e
lu
n
g
s
eg
m
en
tatio
n
s
tep
s
a
b
o
v
e,
th
e
r
esu
ltin
g
im
ag
e
will
co
n
tain
o
n
ly
t
h
e
two
l
u
n
g
s
,
wh
ich
allo
ws
m
o
d
els
to
f
o
c
u
s
ex
clu
s
iv
ely
o
n
th
e
lu
n
g
ar
e
a
an
d
en
h
a
n
ce
s
th
e
ab
ilit
y
to
d
etec
t th
e
lo
ca
tio
n
o
f
lu
n
g
n
o
d
u
les.
Fig
u
r
e
3
illu
s
tr
a
tes ea
ch
s
tep
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
.
Fig
u
r
e
3
.
I
ll
u
s
tr
ate
th
e
s
tep
s
p
er
f
o
r
m
e
d
o
n
a
s
lice
3
.
2
.
2
.
Da
t
a
no
rm
a
liza
t
io
n
Ad
d
itio
n
ally
,
b
ef
o
r
e
th
e
d
ata
ca
n
b
e
in
p
u
t
in
t
o
th
e
m
o
d
el,
th
e
C
T
im
ag
e
n
ee
d
s
to
b
e
n
o
r
m
alize
d
.
T
h
e
p
ix
el
v
alu
es
m
ay
co
n
tain
d
is
c
r
ep
an
cies
d
u
e
to
er
r
o
r
s
f
r
o
m
th
e
im
ag
in
g
m
ac
h
in
e,
an
d
n
o
r
m
alizin
g
th
e
im
ag
e
b
ef
o
r
e
f
ee
d
in
g
it
in
t
o
th
e
m
o
d
el
ca
n
h
elp
r
ed
u
ce
th
ese
d
is
cr
ep
an
cies
an
d
g
iv
e
th
e
m
o
d
e
l
an
o
v
er
all
v
iew
o
f
th
e
d
ataset.
T
h
e
p
ix
el
v
alu
es
will
b
e
s
ca
led
to
th
e
r
an
g
e
[
0
-
1
]
.
First,
th
e
m
in
im
u
m
v
a
lu
e
(
min
_
v
a
l
)
a
n
d
m
ax
im
u
m
v
al
u
e
(
ma
x
_
va
l
)
o
f
th
e
im
ag
e
ar
e
d
eter
m
in
ed
.
T
h
en
,
th
e
p
ix
el
v
alu
es
will
b
e
n
o
r
m
alize
d
to
th
e
r
an
g
e
[
0
-
1
]
u
s
in
g
th
e
f
o
llo
win
g
f
o
r
m
u
la:
_
=
−
_
_
−
_
(
1
)
3
.
3
.
O
pti
m
izing
t
he
hy
perpa
ra
m
et
er
s
o
f
t
he
f
a
s
t
er
R
-
CN
N
Fas
ter
R
-
C
NN
[
2
0
]
h
as
b
ee
n
ex
ten
s
iv
ely
ap
p
lied
in
m
e
d
ical
im
ag
e
o
b
ject
r
ec
o
g
n
itio
n
a
p
p
licatio
n
s
f
o
r
s
ev
e
r
al
im
p
o
r
tan
t
r
ea
s
o
n
s
r
elate
d
t
o
ac
c
u
r
ac
y
,
th
e
ab
il
ity
to
d
etec
t
co
m
p
lex
o
b
jects
,
an
d
th
e
m
o
d
el's
f
lex
ib
ilit
y
.
I
n
m
ed
ical
im
ag
es,
th
e
o
b
jects
th
at
n
ee
d
to
b
e
d
etec
ted
ar
e
o
f
ten
v
er
y
s
m
all
o
r
h
av
e
co
m
p
lex
s
h
ap
es,
s
u
ch
as
lu
n
g
n
o
d
u
les.
Fas
ter
R
-
C
NN,
w
ith
its
ab
ili
ty
to
lear
n
an
d
o
p
tim
ize
f
ea
tu
r
es
f
r
o
m
m
u
ltip
le
lev
els
with
in
th
e
C
NN
n
et
wo
r
k
,
ca
n
ac
c
u
r
ately
d
etec
t
an
d
class
if
y
th
ese
o
b
jects.
T
h
e
Fas
ter
R
-
C
N
N
ar
ch
itectu
r
e
is
co
m
p
o
s
ed
o
f
two
p
r
im
ar
y
co
m
p
o
n
en
ts
:
th
e
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
(
R
PN)
an
d
th
e
Fas
t R
-
C
NN
d
etec
tio
n
m
o
d
u
le
.
T
h
e
R
PN
p
lay
s
a
cr
u
cial
r
o
le
in
th
e
Fas
ter
R
-
C
NN
m
o
d
el,
as
it
is
r
esp
o
n
s
ib
le
f
o
r
g
en
er
atin
g
ca
n
d
id
ate
r
eg
i
o
n
s
o
f
in
ter
est
f
r
o
m
th
e
in
p
u
t
im
a
g
e,
h
elp
i
n
g
to
id
en
tify
a
r
ea
s
th
at
m
ay
c
o
n
tain
o
b
jects.
R
PN
o
p
er
ates
b
ased
o
n
a
C
NN
to
ex
tr
ac
t
f
ea
tu
r
es
f
r
o
m
th
e
im
a
g
e,
th
en
u
s
es
a
s
lid
in
g
win
d
o
w
to
s
ca
n
th
r
o
u
g
h
p
o
in
ts
o
n
th
e
f
ea
tu
r
e
m
a
p
.
At
ea
ch
p
o
in
t,
R
PN
g
en
er
ates
"a
n
ch
o
r
s
"
(
ass
u
m
ed
r
ec
ta
n
g
les
with
d
if
f
er
en
t
s
ca
les
an
d
asp
ec
t
r
atio
s
)
an
d
class
if
ies
th
em
in
to
two
g
r
o
u
p
s
:
co
n
tain
in
g
an
o
b
ject
o
r
n
o
t
co
n
tain
in
g
an
o
b
ject.
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
Dete
ctin
g
lu
n
g
n
o
d
u
les in
co
m
p
u
ted
to
mo
g
r
a
p
h
y
ima
g
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
(
La
m
Th
a
n
h
Hien
)
5609
Simu
ltan
eo
u
s
ly
,
th
e
R
PN
p
r
e
d
icts
p
o
s
itio
n
al
r
ef
in
em
en
ts
f
o
r
th
e
an
c
h
o
r
s
u
s
in
g
a
b
o
u
n
d
in
g
b
o
x
r
eg
r
ess
io
n
n
etwo
r
k
,
allo
win
g
th
em
to
m
o
r
e
ac
cu
r
ately
alig
n
with
th
e
ac
tu
al
o
b
jects
in
th
e
im
ag
e.
Af
ter
g
en
er
atin
g
th
e
p
r
o
p
o
s
als,
R
P
N
u
s
es
th
e
No
n
-
Ma
x
im
u
m
Su
p
p
r
ess
io
n
(
NM
S)
tech
n
iq
u
e
to
elim
in
ate
o
v
er
l
ap
p
in
g
r
e
g
io
n
s
an
d
r
etain
th
e
b
est
o
n
es.
I
n
teg
r
atin
g
R
PN
in
to
d
ee
p
lear
n
in
g
m
o
d
els
h
elp
s
ac
ce
ler
ate
th
e
o
b
jec
t
d
etec
tio
n
p
r
o
ce
s
s
,
wh
ile
s
im
u
ltan
eo
u
s
ly
o
p
tim
i
zin
g
th
e
g
e
n
er
atio
n
o
f
r
e
g
io
n
p
r
o
p
o
s
als
an
d
o
b
ject
cla
s
s
if
icatio
n
,
th
er
eb
y
im
p
r
o
v
in
g
th
e
e
f
f
icien
cy
a
n
d
a
cc
u
r
ac
y
o
f
th
e
m
o
d
el.
I
n
th
is
s
tu
d
y
,
we
will
o
p
tim
ize
th
e
p
a
r
am
eter
s
o
f
th
e
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
m
o
d
u
le
in
th
e
Fas
ter
R
-
C
NN
m
o
d
el
b
ased
o
n
th
e
an
aly
s
is
o
f
lu
n
g
n
o
d
u
le
c
h
ar
ac
ter
is
tics
in
C
T
im
ag
e
d
ata.
T
h
e
o
r
ig
i
n
al
ar
ch
itectu
r
e
o
f
Fas
ter
R
-
C
NN
p
r
ed
ef
in
es
3
d
if
f
e
r
en
t
s
izes
[
1
2
8
,
2
5
6
,
5
1
2
]
an
d
3
asp
ec
t
r
atio
s
[
0
.
5
,
1
,
2
]
f
o
r
th
e
an
ch
o
r
b
o
x
es.
As
a
r
esu
lt,
Fas
ter
R
-
C
NN
u
tili
ze
s
n
in
e
d
if
f
er
en
t
an
ch
o
r
b
o
x
s
izes
in
to
t
al:
[
6
4
,
1
2
8
]
,
[
1
2
8
,
1
2
8
]
,
[
2
5
6
,
1
2
8
]
,
[
1
2
8
,
2
5
6
]
,
[
2
5
6
,
2
5
6
]
,
[
5
1
2
,
2
5
6
]
,
[
2
5
6
,
5
1
2
]
,
[
5
1
2
,
5
1
2
]
,
an
d
[
1
0
2
4
,
5
1
2
]
.
T
h
ese
p
r
ed
ef
in
e
d
s
izes
ar
e
co
n
s
id
er
a
b
ly
lar
g
e
r
th
an
th
e
t
y
p
ical
d
ia
m
eter
o
f
lu
n
g
n
o
d
u
les,
r
esu
lti
n
g
in
a
lar
g
e
n
u
m
b
er
o
f
u
n
n
ec
ess
ar
y
o
r
ir
r
elev
an
t a
n
ch
o
r
b
o
x
es.
T
ak
in
g
in
to
ac
co
u
n
t th
e
ac
tu
al
s
izes o
f
lu
n
g
n
o
d
u
les an
d
th
e
asp
ec
t
r
atio
d
is
tr
ib
u
tio
n
in
th
e
L
UN
A1
6
d
ataset,
we
ad
o
p
t
th
r
ee
s
m
aller
an
ch
o
r
b
o
x
s
izes
[
3
2
,
6
4
,
1
2
8
]
co
m
b
in
ed
with
two
asp
ec
t
r
atio
s
[
1
,
2
]
,
r
esu
ltin
g
in
s
ix
an
ch
o
r
b
o
x
co
n
f
ig
u
r
atio
n
s
: [
3
2
,
3
2
]
,
[
3
2
,
6
4
]
,
[
6
4
,
6
4
]
,
[
6
4
,
1
2
8
]
,
[
1
2
8
,
1
2
8
]
,
a
n
d
[
1
2
8
,
2
5
6
]
.
T
h
ese
an
ch
o
r
b
o
x
s
izes
ar
e
s
m
a
ller
an
d
b
etter
alig
n
ed
with
th
e
ac
tu
al
d
im
en
s
io
n
s
o
f
lu
n
g
n
o
d
u
les
in
th
e
d
atase
t.
Fig
u
r
e
4
p
r
esen
ts
a
co
m
p
ar
is
o
n
b
etwe
en
th
e
o
r
i
g
in
al
an
c
h
o
r
b
o
x
es
an
d
th
e
m
o
d
if
ied
a
n
ch
o
r
b
o
x
es e
m
p
lo
y
ed
in
o
u
r
a
p
p
r
o
ac
h
.
Fig
u
r
e
4
.
C
o
m
p
a
r
is
o
n
o
f
Fas
ter
R
-
C
NN
u
s
in
g
th
e
o
r
ig
in
al
an
ch
o
r
b
o
x
es (
lef
t)
a
n
d
th
e
cu
s
to
m
ized
an
ch
o
r
b
o
x
es tailo
r
ed
t
o
lu
n
g
n
o
d
u
les (
r
ig
h
t)
T
h
e
Fas
t
R
-
C
NN
d
etec
to
r
in
Fas
ter
R
-
C
N
N
is
th
e
n
ex
t
s
tep
af
ter
th
e
r
eg
io
n
p
r
o
p
o
s
als
ar
e
g
en
er
ated
b
y
th
e
R
PN.
On
ce
th
e
r
eg
io
n
s
th
at
m
ay
co
n
tain
o
b
jects
ar
e
id
en
tifie
d
,
Fas
t
R
-
C
NN
p
r
o
ce
s
s
es
th
em
to
clas
s
if
y
th
e
o
b
jects
a
n
d
ac
cu
r
ately
d
e
ter
m
in
e
th
eir
p
o
s
itio
n
s
in
t
h
e
im
ag
e.
Fas
t
R
-
C
NN
o
p
er
ate
s
b
y
u
s
in
g
r
eg
io
n
p
r
o
p
o
s
als
g
en
er
ated
f
r
o
m
th
e
R
PN
an
d
p
er
f
o
r
m
s
th
e
f
o
llo
w
in
g
s
tep
s
:
f
ir
s
t,
th
ese
p
r
o
p
o
s
als
ar
e
cr
o
p
p
ed
f
r
o
m
th
e
f
ea
tu
r
e
m
a
p
(
th
e
f
ea
tu
r
e
m
ap
ex
tr
ac
ted
b
y
th
e
C
NN
f
r
o
m
th
e
in
p
u
t
im
ag
e
)
.
T
h
e
n
,
a
tech
n
iq
u
e
ca
lled
r
eg
io
n
o
f
in
ter
est
(
R
o
I
)
p
o
o
lin
g
is
u
s
ed
to
r
esize
th
e
p
r
o
p
o
s
als
o
f
d
if
f
er
e
n
t
s
izes
in
to
a
f
ix
ed
s
ize,
m
a
k
in
g
th
em
s
u
itab
le
f
o
r
f
u
r
t
h
er
p
r
o
ce
s
s
in
g
in
th
e
n
etwo
r
k
.
T
h
ese
f
ea
tu
r
es
a
r
e
th
e
n
p
ass
ed
in
t
o
a
f
u
lly
co
n
n
ec
ted
(
FC
)
n
etwo
r
k
f
o
r
o
b
ject
class
if
icatio
n
an
d
b
o
u
n
d
in
g
b
o
x
c
o
o
r
d
in
ate
p
r
ed
ictio
n
(
p
o
s
itio
n
an
d
s
ize)
f
o
r
ea
c
h
p
r
o
p
o
s
al.
T
h
e
Fas
t
R
-
C
NN
d
etec
to
r
u
s
es
a
m
u
lti
-
task
lo
s
s
f
u
n
ctio
n
th
at
co
m
b
i
n
es
class
if
icatio
n
lo
s
s
an
d
b
o
u
n
d
in
g
b
o
x
r
eg
r
ess
io
n
lo
s
s
.
T
h
e
f
o
r
m
u
la
f
o
r
th
e
lo
s
s
f
u
n
ctio
n
is
as:
(
,
,
)
=
1
∑
(
,
∗
)
+
1
∑
∗
(
,
)
,
(
2
)
I
n
th
e
f
o
m
u
la,
d
en
o
tes
th
e
n
u
m
b
er
o
f
R
o
I
s
u
s
ed
f
o
r
class
if
icatio
n
,
wh
ile
r
ef
er
s
to
th
e
n
u
m
b
er
o
f
R
o
I
s
u
s
ed
f
o
r
b
o
u
n
d
in
g
b
o
x
r
eg
r
ess
io
n
.
is
th
e
p
r
ed
icted
p
r
o
b
ab
ilit
y
th
at
th
e
th
R
o
I
c
o
n
tain
s
an
o
b
ject
an
d
∗
is
th
e
g
r
o
u
n
d
tr
u
th
la
b
el,
wh
i
ch
tak
es
v
alu
es
o
f
1
o
r
0
t
o
in
d
icate
wh
eth
er
th
e
ℎ
R
o
I
is
an
o
b
ject
o
r
n
o
t.
r
ep
r
esen
ts
th
e
g
r
o
u
n
d
tr
u
th
b
o
u
n
d
in
g
b
o
x
p
ar
a
m
eter
s
f
o
r
t
h
e
ℎ
R
o
I
an
d
d
en
o
tes
th
e
p
r
e
d
icted
b
o
u
n
d
in
g
b
o
x
p
ar
am
eter
s
f
o
r
th
e
i
th
R
o
I
.
i
s
th
e
class
if
icatio
n
lo
s
s
f
u
n
cti
o
n
(
t
y
p
ically
cr
o
s
s
en
tr
o
p
y
)
,
a
n
d
is
th
e
lo
s
s
f
u
n
ctio
n
f
o
r
b
o
u
n
d
in
g
b
o
x
co
o
r
d
in
ate
r
eg
r
ess
io
n
(
u
s
u
a
lly
s
m
o
o
th
L
1
lo
s
s
)
.
A
b
alan
c
in
g
p
ar
am
eter
is
in
tr
o
d
u
ce
d
to
weig
h
t t
h
e
co
n
t
r
ib
u
tio
n
s
o
f
t
h
e
class
if
icatio
n
an
d
r
eg
r
ess
io
n
lo
s
s
es in
th
e
o
v
e
r
all
lo
s
s
f
u
n
ctio
n
.
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
0
4
-
5
6
1
5
5610
Fas
t
R
-
C
NN
in
tr
o
d
u
ce
s
s
ev
er
al
im
p
r
o
v
em
en
ts
o
v
er
th
e
p
r
ev
i
o
u
s
m
eth
o
d
,
R
-
C
NN,
in
clu
d
in
g
r
ed
u
cin
g
co
m
p
u
tatio
n
al
tim
e
b
y
p
r
o
ce
s
s
in
g
th
e
im
ag
e
o
n
ly
o
n
ce
th
r
o
u
g
h
th
e
C
NN
to
ex
tr
ac
t
f
ea
tu
r
es,
r
ath
er
th
an
p
r
o
ce
s
s
in
g
it
m
u
ltip
le
tim
es
f
o
r
ea
c
h
p
r
o
p
o
s
al.
T
h
is
m
ak
es
Fas
t
R
-
C
NN
f
aster
an
d
m
o
r
e
ef
f
icien
t
in
o
b
ject
d
etec
tio
n
.
A
d
d
itio
n
ally
,
Fas
t
R
-
C
NN
ca
n
o
p
tim
ize
th
e
en
tire
lear
n
in
g
p
r
o
ce
s
s
,
f
r
o
m
f
ea
tu
r
e
e
x
tr
ac
tio
n
an
d
class
if
icatio
n
to
b
o
u
n
d
in
g
b
o
x
ad
ju
s
tm
en
t,
r
esu
ltin
g
i
n
a
m
o
r
e
p
o
wer
f
u
l
an
d
ac
cu
r
ate
o
b
ject
d
etec
tio
n
m
o
d
el.
I
n
th
is
s
tu
d
y
,
we
u
s
e
th
r
ee
C
NN
b
ac
k
b
o
n
e
ar
ch
itect
u
r
es
in
Fas
ter
R
-
C
NN:
R
e
s
Ne
t5
0
[
2
1
]
,
R
esNet5
0
v2
[
2
2
]
,
a
n
d
Mo
b
ileNet
[
2
3
]
.
Fig
u
r
e
5
d
ep
icts
th
e
o
v
er
all
s
tr
u
ctu
r
e
o
f
th
e
Fas
ter
R
-
C
NN
f
r
am
ewo
r
k
em
p
lo
y
ed
in
th
is
s
tu
d
y
.
Fig
u
r
e
5
.
T
h
e
ar
ch
itectu
r
e
o
f
F
aster
R
-
C
NN
w
ith
R
esNe
t5
0
,
R
esNet5
0
v
2
,
an
d
M
o
b
ileNet
b
ac
k
b
o
n
es
4.
E
XP
E
R
I
M
E
N
T
AND
R
E
SU
L
T
S
4
.
1
.
E
x
perim
ent
T
o
clar
if
y
th
e
ef
f
ec
tiv
e
n
ess
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
,
we
ca
r
r
y
o
u
t
e
x
p
er
im
en
ts
with
two
s
ce
n
ar
io
s
:
u
s
in
g
im
ag
e
d
ata
th
at
in
clu
d
e
s
b
o
th
lu
n
g
s
an
d
u
s
in
g
im
ag
e
d
ata
th
at
h
as
b
ee
n
cr
o
p
p
ed
t
o
in
clu
d
e
o
n
ly
o
n
e
lu
n
g
.
T
h
e
u
s
e
o
f
o
n
e
-
lu
n
g
im
ag
e
d
ata
is
b
ased
o
n
o
u
r
h
y
p
o
th
esis
to
en
h
an
ce
th
e
p
er
f
o
r
m
an
ce
o
f
d
etec
tin
g
lu
n
g
n
o
d
u
les
b
y
d
ir
ec
ti
n
g
th
e
Fas
ter
R
-
C
NN
m
o
d
els
to
f
o
cu
s
o
n
p
r
o
ce
s
s
in
g
th
e
l
u
n
g
a
r
ea
.
T
h
er
e
f
o
r
e,
th
e
r
esu
lts
with
o
n
e
-
lu
n
g
d
ata
a
r
e
ex
p
ec
te
d
to
p
er
f
o
r
m
b
ette
r
.
T
o
en
s
u
r
e
a
f
air
ev
alu
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
’
s
ef
f
ec
tiv
e
n
ess
o
n
th
e
Fas
ter
R
-
C
NN
m
o
d
el,
th
e
m
etr
ics
will
b
e
ca
lcu
lat
ed
b
ased
o
n
th
e
Fas
ter
R
-
C
NN
m
o
d
el
with
d
if
f
er
en
t b
ac
k
b
o
n
e
v
ar
iatio
n
s
,
in
cl
u
d
in
g
R
esNet5
0
,
R
esNet5
0
v
2
,
an
d
Mo
b
ileNet
.
I
n
ter
m
s
o
f
ex
p
er
im
e
n
tal
d
eta
ils
,
th
e
m
o
d
el
was
tr
ain
ed
u
s
i
n
g
2
T
4
GPUs
o
n
th
e
Kag
g
le
p
latf
o
r
m
[
2
4
]
.
T
h
e
d
ata
was
au
g
m
en
ted
u
s
in
g
tech
n
iq
u
es
s
u
ch
as
im
a
g
e
cr
o
p
p
in
g
,
b
r
ig
h
tn
ess
ad
ju
s
t
m
en
t,
a
n
d
b
lu
r
r
in
g
to
in
cr
ea
s
e
th
e
g
en
er
aliza
tio
n
o
f
th
e
d
ataset.
T
h
e
m
o
d
el
was
tr
ain
ed
with
a
lear
n
in
g
r
ate
o
f
0
.
0
0
0
1
,
a
b
atch
s
ize
o
f
1
0
,
an
d
was
r
u
n
f
o
r
1
0
0
ep
o
ch
s
.
Ad
d
itio
n
ally
,
f
r
o
m
th
e
p
r
e
p
r
o
ce
s
s
ed
d
ata,
w
e
p
er
f
o
r
m
ed
im
ag
e
cr
o
p
p
in
g
to
s
p
lit
ea
ch
im
ag
e
in
to
two
n
ew
im
ag
es
co
r
r
esp
o
n
d
in
g
to
th
e
two
s
ep
ar
ate
lu
n
g
s
.
T
h
u
s
,
we
will
ex
p
er
im
en
t
a
n
d
e
v
alu
ate
o
n
t
wo
d
atasets
.
T
h
e
f
ir
s
t
d
ataset
is
th
e
u
n
cr
o
p
p
ed
d
ataset,
wh
ic
h
in
clu
d
es
1
,
1
8
6
C
T
im
ag
e
s
lices
with
a
s
ize
o
f
5
1
2
×5
1
2
.
T
h
e
s
ec
o
n
d
d
ataset
is
th
e
o
n
e
w
h
er
e
ea
ch
im
ag
e
is
s
p
lit
in
to
two
h
al
v
es
ac
co
r
d
in
g
to
th
e
two
lu
n
g
s
.
T
o
allo
w
th
e
m
o
d
el
to
f
o
cu
s
m
o
r
e
o
n
im
ag
es
co
n
tain
in
g
l
u
n
g
n
o
d
u
les,
th
is
d
ataset
will
co
n
s
is
t
o
f
1
0
0
%
im
ag
es
co
n
tain
in
g
n
o
d
u
les
an
d
o
n
ly
2
0
%
o
f
im
a
g
es
with
o
u
t
n
o
d
u
les.
T
h
er
ef
o
r
e,
th
e
d
ataset
will
co
n
s
is
t
o
f
1
,
3
0
4
i
m
ag
es,
ea
ch
with
a
s
ize
o
f
2
5
6
×3
8
4
.
T
h
e
test
s
et
o
f
th
is
d
a
taset
will
b
e
tak
en
f
r
o
m
th
e
test
s
et
o
f
th
e
two
-
lu
n
g
d
ataset
af
ter
b
ein
g
s
p
lit
in
h
alf
.
T
h
e
tr
ain
,
v
alid
atio
n
,
an
d
test
s
ets
ar
e
r
an
d
o
m
l
y
s
p
lit f
r
o
m
th
e
o
r
i
g
in
al
d
ataset
with
a
r
atio
o
f
7
:2
:1
.
Fig
u
r
e
6
illu
s
tr
ates a
n
im
ag
e
with
two
lu
n
g
s
th
at
h
as b
ee
n
cr
o
p
p
e
d
in
to
two
s
ep
ar
ate
lu
n
g
im
a
g
es.
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
Dete
ctin
g
lu
n
g
n
o
d
u
les in
co
m
p
u
ted
to
mo
g
r
a
p
h
y
ima
g
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
(
La
m
Th
a
n
h
Hien
)
5611
Fig
u
r
e
6
.
I
ll
u
s
tr
atio
n
o
f
a
lu
n
g
im
ag
e
s
p
lit in
to
two
s
ep
ar
ate
l
u
n
g
im
a
g
es
Fo
r
th
is
p
r
o
b
lem
,
th
e
m
etr
ics
u
s
ed
to
e
v
alu
ate
th
e
m
o
d
el
ar
e
m
AP5
0
(
m
ea
n
a
v
er
ag
e
p
r
ec
is
io
n
at
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
(
I
o
U
)
t
h
r
esh
o
ld
o
f
5
0
%)
an
d
r
ec
all.
Fo
r
a
p
r
ed
ictio
n
r
esu
lt,
ev
alu
ati
o
n
is
b
ased
o
n
tr
u
e
p
o
s
itiv
e
(
T
P)
an
d
f
alse
p
o
s
itiv
e
(
FP
)
th
r
o
u
g
h
I
o
U
.
I
f
th
e
I
o
U
>
5
0
%,
th
e
r
esu
lt
is
co
n
s
id
er
e
d
a
T
P,
o
th
er
wis
e,
it is
n
o
t.
I
f
a
co
r
r
ec
t r
esu
lt is
n
o
t d
etec
ted
b
y
th
e
m
o
d
el,
it is
co
n
s
id
er
ed
a
f
alse n
eg
ativ
e
(
F
N)
.
I
n
th
e
d
ep
lo
y
m
en
t
s
tag
e,
we
u
s
ed
th
e
o
p
tim
al
m
o
d
el
weig
h
ts
o
b
tain
ed
f
r
o
m
th
e
ex
p
er
i
m
en
tal
p
h
ase
an
d
b
u
ilt
a
g
r
ap
h
ical
u
s
er
in
te
r
f
ac
e
to
f
ac
ilit
ate
s
m
o
o
th
in
te
r
ac
tio
n
with
th
e
m
o
d
el.
T
h
is
d
esk
to
p
a
p
p
licatio
n
was d
ev
elo
p
ed
u
s
in
g
Py
th
o
n
’
s
T
k
in
ter
lib
r
ar
y
,
wh
ich
is
well
-
s
u
ited
f
o
r
cr
ea
tin
g
GUI
ap
p
li
ca
tio
n
s
.
T
h
e
s
y
s
tem
ac
ce
p
ts
a
2
D
C
T
im
a
g
e
s
lice
as
in
p
u
t
a
n
d
p
r
o
ce
s
s
es
it
to
d
e
tect
an
d
h
ig
h
lig
h
t
an
y
lu
n
g
n
o
d
u
les
p
r
esen
t.
T
h
e
o
u
tp
u
t,
co
n
s
is
tin
g
o
f
th
e
o
r
ig
i
n
al
im
ag
e
with
th
e
id
en
tifie
d
n
o
d
u
le
m
ar
k
ed
,
is
d
is
p
lay
ed
with
in
th
e
in
ter
f
ac
e
u
s
in
g
th
e
Ma
tp
lo
tlib
lib
r
ar
y
.
Fo
r
b
etter
v
is
u
aliza
tio
n
,
th
e
C
T
im
ag
es
ar
e
s
h
o
wn
u
s
in
g
a
g
r
ay
s
ca
le
co
lo
r
m
ap
,
wh
ich
is
co
m
m
o
n
l
y
ap
p
lied
in
m
ed
ical
im
ag
in
g
to
em
p
h
asize
s
tr
u
ctu
r
al
f
ea
tu
r
es.
4
.
2
.
Resul
t
a
nd
d
is
cus
s
io
n
First,
ex
p
er
im
en
ts
wer
e
co
n
d
u
cted
u
s
in
g
th
e
d
ataset
co
n
tain
in
g
b
o
th
lu
n
g
s
,
t
h
e
r
esu
lts
ar
e
s
u
m
m
ar
ized
in
T
ab
le
1
.
T
h
e
F
aster
R
-
C
NN
m
o
d
el
with
R
esNet5
0
,
af
ter
b
ei
n
g
t
r
ain
ed
f
o
r
ap
p
r
o
x
im
ately
1
0
0
ep
o
ch
s
,
ac
h
iev
e
d
th
e
b
est
r
esu
lt
with
a
r
ec
all
o
f
0
.
8
0
6
4
an
d
m
AP5
0
o
f
0
.
7
.
T
h
e
lo
s
s
f
u
n
ct
io
n
o
n
t
h
e
tr
ain
in
g
s
et
d
r
o
p
p
ed
b
elo
w
0
.
1
,
in
d
ica
tin
g
th
at
th
e
m
o
d
el
ac
h
iev
ed
a
r
elativ
ely
h
ig
h
s
en
s
itiv
ity
in
d
etec
tin
g
p
o
s
itiv
e
ca
s
es.
Ho
wev
er
,
th
e
av
e
r
ag
e
p
r
ec
is
io
n
(
m
AP5
0
)
o
f
th
e
p
r
e
d
ictio
n
s
s
till
n
ee
d
s
im
p
r
o
v
e
m
en
t.
Similar
ly
,
th
e
Fas
ter
R
-
C
NN
m
o
d
el
with
R
esNet5
0
v
2
,
af
ter
b
ein
g
t
r
ain
ed
f
o
r
ar
o
u
n
d
1
0
0
ep
o
ch
s
,
ac
h
ie
v
ed
th
e
b
est
r
esu
lt
with
a
R
ec
all
o
f
0
.
8
3
an
d
m
A
P5
0
o
f
0
.
7
4
.
T
h
is
m
o
d
el
p
r
o
d
u
ce
d
th
e
b
est r
esu
lts
o
n
th
e
d
ataset
with
b
o
th
lu
n
g
s
b
u
t,
in
r
etu
r
n
,
t
h
e
p
r
ed
ictio
n
tim
e
f
o
r
an
im
ag
e
was
lo
n
g
er
d
u
e
to
th
e
a
r
ch
itectu
r
e
o
f
th
e
m
o
d
el
.
T
h
e
Fas
ter
R
-
C
NN
m
o
d
el
with
Mo
b
ileNet
ac
h
iev
ed
th
e
b
est
r
esu
lt
with
a
R
ec
all
o
f
0
.
7
5
an
d
m
AP5
0
o
f
0
.
6
3
.
T
h
is
m
o
d
el
c
o
n
v
e
r
g
ed
q
u
ite
q
u
ick
l
y
d
u
r
in
g
tr
ain
in
g
b
u
t
h
ad
a
lo
wer
R
ec
all
an
d
m
AP5
0
co
m
p
ar
ed
to
th
e
o
th
e
r
two
m
o
d
els.
Ho
wev
er
,
it
p
er
f
o
r
m
e
d
p
r
ed
ictio
n
s
v
er
y
q
u
ick
ly
d
u
e
to
its
s
m
aller
m
o
d
el
s
tr
u
ctu
r
e.
T
ab
le
1
.
E
x
p
er
im
en
tal
r
esu
lts
o
f
b
ac
k
b
o
n
es o
f
Fas
ter
R
-
C
N
N
with
th
e
two
-
lu
n
g
im
ag
e
d
at
aset
M
e
t
r
i
c
s
R
e
s
n
e
t
5
0
R
e
sN
e
t
5
0
v
2
M
o
b
i
l
e
N
e
t
R
e
c
a
l
l
0
.
8
0
6
4
0
.
8
3
0
.
7
5
mA
P
5
0
0
.
7
0
.
7
4
0
.
6
3
Fo
r
ex
p
e
r
im
en
ts
with
th
e
o
n
e
-
lu
n
g
im
ag
e
d
ataset,
th
e
r
esu
lts
s
h
o
w
a
clea
r
d
if
f
er
en
ce
as
s
h
o
wn
in
T
ab
le
2
.
Af
ter
tr
ain
i
n
g
t
h
e
m
o
d
el
f
o
r
a
p
p
r
o
x
im
ately
1
0
0
e
p
o
ch
s
,
th
e
Fas
ter
R
-
C
NN
m
o
d
el
with
R
esNet5
0
ac
h
iev
ed
a
r
ec
all
o
f
0
.
9
1
a
n
d
m
AP5
0
o
f
0
.
8
6
.
Sp
litt
in
g
t
h
e
t
wo
-
lu
n
g
im
ag
e
in
t
o
in
d
i
v
id
u
al
lu
n
g
s
m
a
d
e
n
o
d
u
le
d
etec
tio
n
ea
s
ier
.
T
h
e
r
ec
all
in
d
icate
s
th
at
th
e
m
o
d
el
ca
n
ac
cu
r
ately
id
en
tify
9
1
%
o
f
th
e
n
o
d
u
les
in
th
e
test
d
ataset,
an
d
th
e
m
AP5
0
o
f
0
.
8
6
s
h
o
ws
th
at
th
e
m
o
d
el
q
u
al
ity
is
n
o
ticea
b
ly
b
etter
co
m
p
a
r
ed
to
t
h
e
two
-
lu
n
g
im
ag
e
d
ataset.
T
h
e
Fas
ter
R
-
C
NN
m
o
d
el
with
R
esNet5
0
v
2
ac
h
iev
ed
th
e
b
est
r
esu
lts
with
a
r
ec
all
o
f
0
.
9
4
an
d
m
AP5
0
o
f
0
.
8
4
.
W
h
ile
it
ca
n
id
en
tify
m
o
r
e
ac
cu
r
ately
th
a
n
th
e
Fas
ter
R
-
C
NN
with
th
e
R
esNet5
0
b
ac
k
b
o
n
e,
th
e
lo
wer
m
AP5
0
in
d
icate
s
th
at
th
e
m
o
d
el
d
etec
ts
m
o
r
e
f
alse
p
o
s
itiv
es
in
th
e
test
s
et.
T
h
e
Fas
ter
R
-
C
NN
m
o
d
el
with
Mo
b
ileNet
,
af
ter
t
r
ain
in
g
f
o
r
ar
o
u
n
d
1
0
0
e
p
o
ch
s
,
ac
h
iev
e
d
a
R
ec
all
o
f
0
.
8
3
an
d
m
AP5
0
o
f
0
.
8
0
.
T
h
is
m
o
d
el
is
th
e
f
astes
t
in
t
er
m
s
o
f
p
r
ed
ictio
n
am
o
n
g
all
s
ix
m
o
d
els
a
n
d
p
er
f
o
r
m
s
b
et
ter
th
an
t
h
e
m
o
d
els
u
s
in
g
th
e
two
-
lu
n
g
im
a
g
e
d
ata
s
et.
T
o
ev
alu
ate
th
e
r
esu
lts
m
o
r
e
o
b
jectiv
ely
,
we
co
m
p
ar
ed
o
u
r
r
esu
lts
with
th
e
s
tu
d
y
in
[
2
5
]
.
I
n
t
h
at
s
tu
d
y
,
th
e
a
u
th
o
r
s
p
r
esen
ted
a
lu
n
g
n
o
d
u
le
d
etec
tio
n
p
ip
elin
e
u
s
in
g
th
e
YOL
Ov
3
m
o
d
el
with
th
e
Dar
k
n
et5
3
b
ac
k
b
o
n
e.
T
h
e
m
o
d
el
was
p
r
e
-
tr
ain
ed
o
n
t
h
e
MS
C
OC
O
d
ataset.
I
n
ex
p
er
im
en
ts
with
th
e
L
UNA1
6
d
ataset,
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
0
4
-
5
6
1
5
5612
th
e
au
th
o
r
s
ac
h
iev
ed
th
eir
b
es
t
r
esu
lts
with
a
m
AP5
0
o
f
ap
p
r
o
x
im
ately
0
.
6
4
a
n
d
a
r
ec
all
o
f
ar
o
u
n
d
0
.
7
5
.
T
h
e
au
th
o
r
s
also
co
m
p
ar
e
d
th
e
p
e
r
f
o
r
m
an
ce
with
s
ev
er
al
o
t
h
er
m
o
d
els,
in
clu
d
in
g
C
ascad
e
R
-
C
NN
with
a
m
A
P5
0
o
f
0
.
4
6
an
d
a
r
ec
all
o
f
0
.
6
0
,
F
C
OS
with
a
m
AP5
0
o
f
0
.
5
0
a
n
d
a
r
ec
all
o
f
0
.
7
2
,
a
n
d
f
i
n
ally
,
YOL
Ov
3
with
th
e
Mo
b
ileNet
b
ac
k
b
o
n
e,
wh
ich
ac
h
iev
ed
a
m
AP5
0
o
f
0
.
5
1
a
n
d
a
r
ec
all
o
f
0
.
8
3
.
T
h
u
s
,
co
m
p
ar
ed
to
th
e
b
est
r
esu
lts
we
ac
h
iev
ed
,
o
u
r
m
o
d
e
ls
with
R
es
Net5
0
an
d
R
esNet5
0
v
2
b
ac
k
b
o
n
es
d
em
o
n
s
tr
ated
s
ig
n
if
ican
tly
b
etter
p
er
f
o
r
m
an
ce
,
with
m
AP5
0
v
a
lu
es
o
f
0
.
8
6
an
d
0
.
8
4
,
an
d
R
e
ca
ll
v
alu
es
o
f
0
.
9
1
an
d
0
.
9
4
,
r
esp
ec
tiv
ely
.
T
h
ese
r
esu
lts
clea
r
ly
h
ig
h
lig
h
t
th
e
ef
f
ec
tiv
en
ess
o
f
ap
p
l
y
in
g
th
e
p
r
o
p
o
s
ed
p
r
o
ce
s
s
f
o
r
s
eg
m
e
n
tin
g
lu
n
g
ar
ea
s
,
as
well
as
o
p
tim
izin
g
th
e
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
m
o
d
u
le
in
th
e
F
aster
R
-
C
N
N
m
o
d
el
b
ased
o
n
th
e
an
aly
s
is
o
f
lu
n
g
n
o
d
u
le
ch
ar
ac
ter
is
tics
in
C
T
im
ag
e
d
ata.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
e
n
h
an
ce
s
lu
n
g
n
o
d
u
le
d
et
ec
tio
n
b
y
d
ir
ec
tin
g
th
e
m
o
d
el’
s
atten
tio
n
to
war
d
r
elev
an
t
r
eg
io
n
s
wh
ile
ef
f
ec
tiv
ely
s
u
p
p
r
ess
in
g
n
o
is
e
f
r
o
m
ir
r
elev
an
t
ar
ea
s
,
th
er
eb
y
im
p
r
o
v
in
g
b
o
th
ac
c
u
r
ac
y
an
d
f
o
cu
s
.
T
ab
le
2
.
E
x
p
er
im
en
tal
r
esu
lts
o
f
b
ac
k
b
o
n
es o
f
Fas
ter
R
-
C
N
N
with
th
e
o
n
e
-
lu
n
g
im
a
g
e
d
at
aset
M
e
t
r
i
c
s
R
e
s
n
e
t
5
0
R
e
sN
e
t
5
0
v
2
M
o
b
i
l
e
N
e
t
R
e
c
a
l
l
0
.
9
1
0
.
9
4
0
.
8
3
mA
P
5
0
0
.
8
6
0
.
8
4
0
.
8
0
Ou
r
f
in
d
in
g
s
d
e
m
o
n
s
tr
ate
t
h
at
in
co
r
p
o
r
atin
g
a
l
u
n
g
r
eg
i
o
n
s
eg
m
en
tatio
n
s
tep
p
r
i
o
r
to
d
ete
ctio
n
ca
n
ef
f
ec
tiv
ely
r
ed
u
ce
b
ac
k
g
r
o
u
n
d
n
o
is
e
an
d
en
h
a
n
ce
t
h
e
m
o
d
el’
s
f
o
cu
s
o
n
r
elev
a
n
t
an
at
o
m
ical
ar
ea
s
.
T
h
is
p
r
ep
r
o
ce
s
s
in
g
ap
p
r
o
ac
h
,
alth
o
u
g
h
lig
h
tweig
h
t,
co
n
tr
ib
u
te
s
to
im
p
r
o
v
ed
p
er
f
o
r
m
an
ce
in
d
etec
tin
g
lu
n
g
n
o
d
u
les
f
r
o
m
C
T
s
ca
n
s
.
T
h
e
s
e
r
esu
lts
ad
d
to
th
e
cu
r
r
e
n
t
b
o
d
y
o
f
r
esear
c
h
b
y
h
ig
h
lig
h
tin
g
th
e
co
m
b
in
ed
im
p
o
r
tan
ce
o
f
d
ata
p
r
e
p
ar
atio
n
an
d
m
o
d
el
co
n
f
i
g
u
r
atio
n
.
W
h
ile
p
r
ev
io
u
s
s
tu
d
ies
h
av
e
p
r
im
ar
ily
f
o
cu
s
ed
o
n
en
d
-
to
-
e
n
d
d
etec
tio
n
f
r
a
m
ewo
r
k
s
,
o
u
r
wo
r
k
s
h
o
ws
th
at
ev
en
m
o
d
est,
d
o
m
ain
-
s
p
ec
if
ic
ad
ju
s
tm
en
ts
in
p
r
ep
r
o
ce
s
s
in
g
an
d
m
o
d
el
tu
n
in
g
ca
n
lea
d
to
s
u
b
s
tan
tial im
p
r
o
v
em
en
ts
.
Ad
d
itio
n
ally
,
we
h
av
e
in
te
g
r
ated
all
th
e
co
m
p
o
n
e
n
ts
o
f
th
is
s
tu
d
y
in
to
a
test
in
g
s
o
f
twar
e,
as
illu
s
tr
ated
in
Fig
u
r
e
7
.
T
h
e
s
o
f
twar
e
allo
ws
th
e
u
s
er
to
s
elec
t
o
n
e
o
f
th
e
m
o
d
els
s
tu
d
ied
in
th
is
p
ap
er
an
d
u
s
e
ch
est
C
T
im
ag
es
as
in
p
u
t
d
ata.
I
t
th
en
o
u
t
p
u
ts
th
e
d
etec
ted
lu
n
g
n
o
d
u
le
r
e
g
io
n
s
alo
n
g
with
th
eir
co
r
r
esp
o
n
d
in
g
p
r
o
b
ab
ilit
ies.
T
h
is
in
f
o
r
m
atio
n
is
d
is
p
lay
ed
v
i
s
u
ally
o
n
th
e
p
r
o
g
r
am
'
s
in
ter
f
ac
e.
Fig
u
r
e
7
.
I
ll
u
s
tr
atio
n
o
f
t
h
e
in
t
er
f
ac
e
o
f
t
h
e
lu
n
g
n
o
d
u
le
d
ete
ctio
n
s
o
f
twar
e
5.
CO
NCLU
SI
O
N
I
n
th
is
p
ap
er
,
we
p
r
esen
ted
a
s
tu
d
y
o
n
lu
n
g
n
o
d
u
le
d
etec
ti
o
n
in
C
T
im
ag
es
u
s
in
g
a
d
ee
p
lear
n
i
n
g
ap
p
r
o
ac
h
,
aim
in
g
to
en
h
an
ce
th
e
v
is
ib
ilit
y
o
f
lu
n
g
n
o
d
u
le
s
an
d
h
elp
th
e
m
o
d
el
f
o
cu
s
m
o
r
e
ef
f
ec
tiv
el
y
o
n
r
elev
an
t
r
eg
i
o
n
s
wh
ile
r
ed
u
cin
g
n
o
is
e
f
r
o
m
ar
ea
s
u
n
r
elate
d
t
o
th
e
r
esu
lts
.
T
o
d
em
o
n
s
tr
ate
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
,
we
co
n
d
u
cte
d
ex
p
er
im
en
ts
o
n
th
e
s
tan
d
ar
d
L
UNA1
6
d
ataset
an
d
ac
h
iev
ed
th
e
b
est
r
esu
lts
o
f
0
.
8
6
m
AP5
0
an
d
0
.
9
1
R
ec
all
f
o
r
th
e
R
esNet5
0
b
ac
k
b
o
n
e
m
o
d
el,
an
d
0
.
8
4
m
AP5
0
an
d
0
.
9
4
R
ec
all
f
o
r
th
e
R
esNet5
0
v
2
b
ac
k
b
o
n
e
m
o
d
el.
T
h
ese
r
esu
lts
r
ef
lec
t
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
an
d
h
ig
h
lig
h
t
th
e
p
o
ten
tial
ap
p
lica
tio
n
o
f
th
e
s
tu
d
y
in
s
u
p
p
o
r
tin
g
lu
n
g
ca
n
ce
r
d
iag
n
o
s
is
.
Des
p
ite
ac
h
iev
in
g
h
ig
h
ev
alu
atio
n
m
etr
ics,
th
e
r
esear
ch
s
till
r
eq
u
ir
es
f
u
r
th
er
im
p
r
o
v
em
en
t
an
d
test
in
g
in
t
h
e
co
n
tex
t
o
f
r
ea
l
-
wo
r
ld
d
ata,
with
m
o
r
e
d
iv
er
s
e
an
d
c
o
m
p
lex
ca
s
es
o
f
lu
n
g
C
T
im
ag
es.
Ou
r
f
u
tu
r
e
r
esear
ch
will
aim
to
im
p
r
o
v
e
th
e
g
en
er
aliza
tio
n
ca
p
ab
ilit
y
o
f
th
e
em
p
lo
y
ed
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es,
wh
ile
also
in
v
esti
g
atin
g
th
e
3
D
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Dete
ctin
g
lu
n
g
n
o
d
u
les in
co
m
p
u
ted
to
mo
g
r
a
p
h
y
ima
g
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
(
La
m
Th
a
n
h
Hien
)
5613
s
tr
u
ctu
r
al
ch
ar
ac
te
r
is
tics
o
f
C
T
im
ag
e
d
ata.
Ad
d
itio
n
ally
,
with
th
e
g
o
al
o
f
tr
a
n
s
latin
g
th
e
r
esear
ch
in
to
p
r
ac
tical
h
ea
lth
ca
r
e
ap
p
licatio
n
s
,
we
will
wo
r
k
o
n
s
p
ec
if
ic
t
ask
s
to
in
teg
r
ate
th
is
m
eth
o
d
i
n
to
th
e
wo
r
k
f
lo
w
o
f
r
ad
io
lo
g
is
ts
in
h
o
s
p
itals
,
allo
win
g
f
o
r
o
b
jectiv
e
a
n
d
p
r
ac
tical
ass
es
s
m
en
ts
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
r
esear
ch
co
n
ten
t
in
th
is
p
ap
er
is
s
p
o
n
s
o
r
ed
b
y
th
e
I
n
s
titu
te
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
(
Viet
Nam
Aca
d
em
y
o
f
Scien
ce
an
d
T
ec
h
n
o
lo
g
y
)
u
n
d
er
th
e
p
r
o
jec
t c
o
d
e
“T
XKHT
E
.
0
2
/
2
5
-
25
-
0
2
”.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT
)
to
r
ec
o
g
n
ize
in
d
iv
i
d
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
L
am
T
h
an
h
Hien
✓
✓
✓
✓
✓
✓
✓
L
e
An
h
T
u
✓
✓
✓
✓
✓
Ph
am
T
r
u
n
g
Hieu
✓
✓
✓
✓
Ph
am
Min
h
Du
c
✓
✓
Ng
u
y
en
Van
Nan
g
✓
✓
✓
Do
Nan
g
T
o
a
n
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
a
r
e
o
p
en
l
y
av
ailab
le
in
L
UNA1
6
at
h
ttp
s
://l
u
n
a1
6
.
g
r
a
n
d
-
c
h
allen
g
e
.
o
r
g
/Ho
m
e/
RE
F
E
R
E
NC
E
S
[
1
]
K
.
V
.
H
i
n
i
sh
a
a
n
d
A
.
Li
j
i
y
a
,
“
L
u
n
g
n
o
d
u
l
e
i
d
e
n
t
i
f
i
c
a
t
i
o
n
,
”
i
n
2
0
1
9
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
I
n
t
e
l
l
i
g
e
n
t
C
o
m
p
u
t
i
n
g
,
I
n
st
r
u
m
e
n
t
a
t
i
o
n
a
n
d
C
o
n
t
r
o
l
T
e
c
h
n
o
l
o
g
i
e
s
(
I
C
I
C
I
C
T
)
,
Ju
l
.
2
0
1
9
,
p
p
.
4
5
1
–
4
5
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
C
I
C
T4
6
0
0
8
.
2
0
1
9
.
8
9
9
3
2
1
8
.
[
2
]
D
.
E.
M
i
d
t
h
u
n
,
“
E
a
r
l
y
d
e
t
e
c
t
i
o
n
o
f
l
u
n
g
c
a
n
c
e
r
,
”
F
1
0
0
0
Re
s
e
a
rc
h
,
v
o
l
.
5
,
p
.
7
3
9
,
A
p
r
.
2
0
1
6
,
d
o
i
:
1
0
.
1
2
6
8
8
/
f
1
0
0
0
r
e
sea
r
c
h
.
7
3
1
3
.
1
.
[
3
]
A
.
P
a
n
u
n
z
i
o
a
n
d
P
.
S
a
r
t
o
r
i
,
“
Lu
n
g
c
a
n
c
e
r
a
n
d
r
a
d
i
o
l
o
g
i
c
a
l
i
m
a
g
i
n
g
,
”
C
u
rr
e
n
t
R
a
d
i
o
p
h
a
rm
a
c
e
u
t
i
c
a
l
s
,
v
o
l
.
1
3
,
n
o
.
3
,
p
p
.
2
3
8
–
2
4
2
,
N
o
v
.
2
0
2
0
,
d
o
i
:
1
0
.
2
1
7
4
/
1
8
7
4
4
7
1
0
1
3
6
6
6
2
0
0
5
2
3
1
6
1
8
4
9
.
[
4
]
A
.
R
.
La
r
i
c
i
e
t
a
l
.
,
“
L
u
n
g
n
o
d
u
l
e
s:
si
z
e
s
t
i
l
l
m
a
t
t
e
r
s,
”
E
u
r
o
p
e
a
n
Re
sp
i
ra
t
o
ry
R
e
v
i
e
w
,
v
o
l
.
2
6
,
n
o
.
1
4
6
,
p
.
1
7
0
0
2
5
,
D
e
c
.
2
0
1
7
,
d
o
i
:
1
0
.
1
1
8
3
/
1
6
0
0
0
6
1
7
.
0
0
2
5
-
2
0
1
7
.
[
5
]
“
Lo
n
g
N
o
d
u
l
e
,
”
U
T
S
o
u
t
h
w
e
st
e
rn
Me
d
i
c
a
l
C
e
n
t
e
r
.
h
t
t
p
s
:
/
/
u
t
sw
me
d
.
o
r
g
/
c
o
n
d
i
t
i
o
n
s
-
t
r
e
a
t
me
n
t
s
/
l
u
n
g
-
n
o
d
u
l
e
s
(
a
c
c
e
sse
d
D
e
c
.
1
2
,
2
0
2
4
)
.
[
6
]
“
Lu
n
g
c
a
n
c
e
r
,
”
W
o
r
l
d
H
e
a
l
t
h
O
r
g
a
n
i
za
t
i
o
n
.
h
t
t
p
s
:
/
/
w
w
w
.
w
h
o
.
i
n
t
/
n
e
w
s
-
r
o
o
m/
f
a
c
t
-
s
h
e
e
t
s
/
d
e
t
a
i
l
/
l
u
n
g
-
c
a
n
c
e
r
(
a
c
c
e
sse
d
D
e
c
.
1
2
,
2
0
2
4
)
.
[
7
]
D
.
M
.
H
a
n
se
l
l
,
A
.
A
.
B
a
n
k
i
e
r
,
H
.
M
a
c
M
a
h
o
n
,
T
.
C
.
M
c
L
o
u
d
,
N
.
L.
M
ü
l
l
e
r
,
a
n
d
J
.
R
e
m
y
,
“
F
l
e
i
sc
h
n
e
r
s
o
c
i
e
t
y
:
g
l
o
ssar
y
o
f
t
e
r
ms
f
o
r
t
h
o
r
a
c
i
c
i
ma
g
i
n
g
,
”
R
a
d
i
o
l
o
g
y
,
v
o
l
.
2
4
6
,
n
o
.
3
,
p
p
.
6
9
7
–
7
2
2
,
M
a
r
.
2
0
0
8
,
d
o
i
:
1
0
.
1
1
4
8
/
r
a
d
i
o
l
.
2
4
6
2
0
7
0
7
1
2
.
[
8
]
M
.
Le
d
e
r
l
i
n
,
M
.
-
P
.
R
e
v
e
l
,
A
.
K
h
a
l
i
l
,
G
.
F
e
r
r
e
t
t
i
,
B
.
M
i
l
l
e
r
o
n
,
a
n
d
F
.
La
u
r
e
n
t
,
“
M
a
n
a
g
e
me
n
t
s
t
r
a
t
e
g
y
o
f
p
u
l
m
o
n
a
r
y
n
o
d
u
l
e
i
n
2
0
1
3
,
”
D
i
a
g
n
o
st
i
c
a
n
d
I
n
t
e
r
v
e
n
t
i
o
n
a
l
I
m
a
g
i
n
g
,
v
o
l
.
9
4
,
n
o
.
1
1
,
p
p
.
1
0
8
1
–
1
0
9
4
,
N
o
v
.
2
0
1
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
d
i
i
i
.
2
0
1
3
.
0
5
.
0
0
7
.
[
9
]
Q
.
L
i
,
S
.
S
o
n
e
,
a
n
d
K
.
D
o
i
,
“
S
e
l
e
c
t
i
v
e
e
n
h
a
n
c
e
me
n
t
f
i
l
t
e
r
s
f
o
r
n
o
d
u
l
e
s,
v
e
ssel
s,
a
n
d
a
i
r
w
a
y
w
a
l
l
s
i
n
t
w
o
‐
a
n
d
t
h
r
e
e
‐
d
i
m
e
n
s
i
o
n
a
l
C
T
s
c
a
n
s,”
M
e
d
i
c
a
l
P
h
y
s
i
c
s
,
v
o
l
.
3
0
,
n
o
.
8
,
p
p
.
2
0
4
0
–
2
0
5
1
,
A
u
g
.
2
0
0
3
,
d
o
i
:
1
0
.
1
1
1
8
/
1
.
1
5
8
1
4
1
1
.
[
1
0
]
C
.
C
.
S
a
m
u
e
l
,
V
.
S
a
r
a
v
a
n
a
n
,
a
n
d
M
.
R
.
V
.
D
e
v
i
,
“
Lu
n
g
n
o
d
u
l
e
d
i
a
g
n
o
si
s
f
r
o
m
C
T
i
ma
g
e
s
u
s
i
n
g
f
u
z
z
y
l
o
g
i
c
,
”
i
n
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
Mu
l
t
i
m
e
d
i
a
A
p
p
l
i
c
a
t
i
o
n
s
(
I
C
C
I
MA
2
0
0
7
)
,
D
e
c
.
2
0
0
7
,
p
p
.
1
5
9
–
1
6
3
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
I
M
A
.
2
0
0
7
.
2
3
6
.
[
1
1
]
R
.
G
o
l
a
n
,
C
.
J
a
c
o
b
,
a
n
d
J.
D
e
n
z
i
n
g
e
r
,
“
Lu
n
g
n
o
d
u
l
e
d
e
t
e
c
t
i
o
n
i
n
C
T
i
ma
g
e
s
u
si
n
g
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
i
n
2
0
1
6
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
r
e
n
c
e
o
n
N
e
u
ra
l
N
e
t
w
o
rks
(
I
J
C
N
N
)
,
Ju
l
.
2
0
1
6
,
p
p
.
2
4
3
–
2
5
0
,
d
o
i
:
1
0
.
1
1
0
9
/
I
JC
N
N
.
2
0
1
6
.
7
7
2
7
2
0
5
.
[
1
2
]
Q
.
W
a
n
g
,
F
.
S
h
e
n
,
L.
S
h
e
n
,
J.
H
u
a
n
g
,
a
n
d
W
.
S
h
e
n
g
,
“
Lu
n
g
n
o
d
u
l
e
d
e
t
e
c
t
i
o
n
i
n
C
T
i
m
a
g
e
s u
s
i
n
g
a
r
a
w
p
a
t
c
h
-
b
a
sed
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
J
o
u
r
n
a
l
o
f
D
i
g
i
t
a
l
I
m
a
g
i
n
g
,
v
o
l
.
3
2
,
n
o
.
6
,
p
p
.
9
7
1
–
9
7
9
,
D
e
c
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
2
7
8
-
0
1
9
-
0
0
2
2
1
-
3.
[
1
3
]
K
.
D
.
L
a
i
,
T
.
T.
N
g
u
y
e
n
,
a
n
d
T
.
H
.
L
e
,
“
D
e
t
e
c
t
i
o
n
o
f
l
u
n
g
n
o
d
u
l
e
s
o
n
C
T
i
mag
e
s
b
a
se
d
o
n
t
h
e
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
w
i
t
h
a
t
t
e
n
t
i
o
n
mec
h
a
n
i
sm,
”
An
n
a
l
s
o
f
Em
e
r
g
i
n
g
T
e
c
h
n
o
l
o
g
i
e
s
i
n
C
o
m
p
u
t
i
n
g
,
v
o
l
.
5
,
n
o
.
2
,
p
p
.
7
8
–
8
9
,
A
p
r
.
2
0
2
1
,
Evaluation Warning : The document was created with Spire.PDF for Python.