Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol
.
5,
N
o
.
1
,
F
e
br
uary
2
0
1
5
,
pp
. 78~
8
3
I
S
SN
: 208
8-8
7
0
8
78
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Feature Selection
of
the Combinati
o
n of P
o
rous
Trabecular
with Ant
h
ropom
e
tric F
e
at
ures
for Osteoporosis Screening
Enny I Sel
a
*,
Sri Har
t
ati*,
Agus
Har
j
oko, R
e
ta
ntyo
Wardoy
o*
, M Mudjo
s
emedi**
*Department of Computer
Scien
ce and Electron
ics,
Universitas
Gadjah Mad
a
, Y
o
g
y
ak
arta, Indon
esia.
** Departmen
t
o
f
Dentomaxillof
acial R
a
diolog
y
,
Universitas Gad
j
ah Mada, Yog
y
a
k
arta, Indon
esia.
Article Info
A
B
STRAC
T
Article histo
r
y:
ReceivedOct 19,
2014
R
e
vi
sed Dec 4,
2
0
1
4
Accepted Dec 26, 2014
This
s
t
ud
y
a
i
m
s
to s
e
lect
the i
m
portant featu
r
es
from
the com
b
ination of
porous trabecular pattern w
ith
anth
ropometr
i
c featu
r
es for o
s
teoporosis
screening
.
The stud
y
sample has thei
r bone miner
a
l density
(BMD) measured
at th
e proximal
femur/lumbar spine using du
al-
e
nerg
y
X-ray
absorptiometr
y
(DXA).
Morpho
logical porous features su
ch as porosity
,
the size of porous,
and the or
ient
ati
on of porous
are
obtained
from
each den
t
al r
a
dio
g
raph us
ing
digital image processing. The anthropom
etric f
eatur
es consider
ed are age,
height, weigh
t
,
and bod
y
mass index (B
MI). Decision tr
ee (J
.48
method) is
used to evalu
a
te the accuracy
of
morphological
porous and anthropometric
featur
es
for s
e
lection dat
a
. Th
e s
t
ud
y
s
hows that the most importa
nt featur
e is
age and
the
co
nsidered featur
es for
osteoporosis screening
ar
e porosity
,
vertical po
re, and oblique po
re. The
decision
tree h
a
s consid
erably
h
i
gh
accur
a
c
y
,
s
e
ns
i
t
i
v
it
y,
and
s
p
ecif
i
cit
y
.
Keyword:
Ant
h
ropom
etric features
Dental peria
p
ical
X-Ray
O
s
teopo
ro
sis
Porous e
x
tracti
o
n
Copyright ©
201
5Institute of
Ad
v
anced
Engineeri
ng and Scien
c
e.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Enny Itje
Sela
Depa
rt
m
e
nt
of
C
o
m
put
er Sci
e
nce a
n
d
El
ect
r
oni
cs
,
Uni
v
ersi
t
a
s Ga
dja
h
M
a
da,
Y
ogy
a
k
art
a
, I
n
do
nesi
a
Em
a
il: en
n
y
sela@ak
a
ko
m
.
ac.id
1.
INTRODUCTION
Ost
e
o
p
o
r
o
si
s, a
com
m
on
m
e
tabol
i
c
disease
characte
r
izes by reduce
d
bone
m
a
ss and thinni
ng
of the
trabec
ular m
i
croarc
hitecture
,
freque
ntly
resul
t
s in
fract
ures
of ve
rtebrae, hi
p, or forearm
.
There
is
a
cons
ens
u
s
th
at BMD shou
ld b
e
u
s
ed fo
r op
eration
a
l
d
e
fin
itio
n of
t
h
e
d
e
gree
of
o
s
teo
poro
s
is. Measu
r
em
en
t BMD is th
e
pri
n
ci
pal
m
e
t
hod
o
f
di
ag
n
o
si
s o
f
ost
e
o
p
o
r
o
si
s beca
use
pat
i
ent
s
wi
t
h
l
o
w
B
M
D
val
u
es
have
el
evat
e
d
ri
sk
o
f
devel
opi
ng
a b
one
fract
ure
.
DX
A i
s
t
h
e
st
anda
r
d
t
ech
ni
q
u
e f
o
r
det
e
rm
ini
n
g B
M
D.
H
o
we
ve
r, B
M
D
t
e
st
i
n
g
using DXA
for all postm
enopausal wo
m
e
n is not practic
al. Because of
the relative high cost and lim
ited
av
ailab
ility o
f
DXA equ
i
p
m
en
t, DXA can
n
o
t
d
e
tect
m
i
cr
o
s
tru
c
ture of bon
e arch
itectu
r
e, wh
ich
is th
e k
e
y to
b
o
n
e
qu
ality [1].
To ove
r
com
e
these problem
s,
m
a
ny studies
have
de
velope
d m
e
thods to
assess bone quality using
m
i
crost
r
uct
u
re
of
b
o
n
e a
r
chi
t
e
ct
ure i
ndi
cat
or
s. It
has
b
een de
m
onstrated that trabec
ular
bone
pattern cha
ngi
ng
o
f
th
e m
a
n
d
i
b
l
e m
a
y b
e
asso
ciated
with sk
el
etal lo
w BM
D
o
r
o
s
teo
poro
s
is. It im
p
lies th
at th
e
p
o
s
sib
ility th
at
t
r
abec
ul
ar b
o
n
e
pat
t
e
rn o
f
t
h
e
m
a
ndi
bl
e det
ect
ed on
peri
api
cal
dent
al
radi
og
ra
phs m
a
y
b
e
usef
ul
i
ndi
ca
t
o
r f
o
r
id
en
tifying
women
with
l
o
w
sk
eletal BMD
[2
] [3
].
Peri
a
p
ical radi
ographs are
relatively inexpe
nsive
a
n
d are
fre
que
nt
l
y
t
a
ken i
n
dent
al
o
ffi
ces as an ai
d t
o
di
ag
nosi
s
[3]
.
Si
nce t
h
e t
r
a
b
e
c
ul
ar
bo
ne ca
n
be easi
l
y
vi
sua
l
i
zed
i
n
peri
a
p
i
cal
ra
di
o
g
ra
p
h
s, m
a
ny
im
port
a
nt
i
n
f
o
rm
at
i
on
ab
out
t
h
e
bo
ne
’s
con
d
i
t
i
on
o
n
m
i
crost
r
uct
u
re l
e
vel
can
be extracte
d
.
On the trabec
ular bon
e, it can be e
x
tracte
d
the tra
b
ecul
a
r segm
ent
s
(r
ods
) an
d i
t
’
s
po
r
ous
(p
lates). Bo
n
e
streng
th
is also d
e
term
in
ed
fro
m
th
e
po
ro
us
st
ruct
u
r
e p
r
o
p
e
r
t
y
such as p
o
r
o
si
t
y
, hom
oge
nei
t
y
,
and
ani
s
ot
r
opy
[4]
.
Acc
o
r
d
i
n
g [
5
]
,
one
o
f
t
h
e s
u
s
p
i
c
i
on
o
f
o
s
t
e
o
p
o
r
o
s
i
s
can
be e
n
f
o
rce
d
i
f
t
h
e
po
re i
n
t
h
e
lo
wer
j
a
w loo
k
s a little, irregular, an
d low con
n
ectiv
ity.
There i
s
a gr
o
w
i
n
g co
nse
n
su
s t
h
at
screeni
n
g fo
r ost
e
o
p
o
r
osi
s
sh
oul
d i
n
cl
ude ant
h
r
o
po
m
e
t
r
i
c
and
clinical feature
s
such as age
,
heig
ht, weight, calciu
m
intake, exercise
and s
m
oking
habit
s
[6]. Recently, there
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE I
S
SN
:
208
8-8
7
0
8
Fea
t
u
r
e S
e
lecti
o
n o
f
t
h
e C
o
mbin
a
tion
o
f
Po
rou
s
Trab
ecu
l
a
r
with
An
thropom
etric Fea
t
u
r
es … (Enn
y
I
S
e
la
)
79
have
bee
n
rep
o
rt
e
d
l
i
nki
ng
o
s
t
e
op
or
osi
s
t
o
com
b
i
n
i
ng
trab
ecu
lar seg
m
e
n
ts with
an
t
h
ro
po
m
e
tric an
d clin
ical
feat
ure
s
[
2
]
[
3
]
[
6]
.
It
i
s
not
k
n
o
w
n t
h
e
i
m
p
o
rt
a
n
t
feat
u
r
es
o
n
t
h
e c
o
m
b
i
n
at
i
on
o
f
po
r
ous
t
r
a
b
ecul
a
r
pat
t
e
rn
w
ith
an
thr
opometr
i
c f
eatu
r
es f
o
r
osteop
or
osis scr
een
ing
.
A
cco
rd
ing
l
y, th
e p
u
r
p
o
s
e of th
is stu
d
y
selected
feat
ure
s
fr
om
the radi
o
g
ra
phi
c of p
o
r
o
us t
r
abecul
a
r pa
ttern and the ant
h
ropom
etric features.Feat
ure sel
ection
im
proves class
i
fi
cation
by sea
r
chi
n
g for the s
ubs
et of
features,
w
h
i
c
h best
cl
assi
fy
t
h
e
t
r
ai
ni
n
g
dat
a
[
7
]
.
2.
R
E
SEARC
H M
ETHOD
All periapical X-Ray im
ages and ant
h
ropometric f
eatures
were collected fr
om
t
h
e Depart
m
e
nt
of
Den
t
o
m
ax
illo
facial Rad
i
o
l
o
gy o
f
Pro
f
. So
ed
o
m
o
Den
t
al
Ho
sp
ital, Faculty o
f
Den
tistry, Un
iv
ersitas Gadj
ah
M
a
da. The a
n
t
h
r
o
p
o
m
e
t
r
i
c
f
eat
ures f
o
r
os
t
e
op
or
osi
s
, s
u
ch as:
age, he
i
ght
, wei
ght
,
ho
rm
onal
con
d
i
t
i
on,
cal
ci
um
i
n
t
a
ke, sm
oki
ng ha
b
i
t
s
, were o
b
t
a
i
n
ed
fr
om
a quest
i
o
n
n
ai
re. T
h
en, B
M
I i
s
re
prese
n
t
e
d
by
wei
g
ht
di
vi
de
d
by
sq
u
a
re o
f
hei
ght
.
The a
n
t
h
r
o
pom
et
ri
c feat
ure
s
con
s
i
d
ere
d
i
n
t
h
i
s
st
u
d
y
we
re
age (
U
)
,
hei
ght
(TB
)
,
wei
g
ht
(B
B
)
, a
nd B
M
I.
Asses
s
m
e
nt
of B
M
D
and
l
u
m
b
ar sp
i
n
e col
l
ect
ed
fr
om
t
h
e Depa
rt
m
e
nt
of R
a
di
ol
ogy
Dr.
Sar
d
ji
t
o
H
o
spi
t
a
l
usi
ng
DX
A.
Su
b
j
ect
s were cl
assi
fi
ed i
n
t
o
o
n
e o
f
t
h
ree
gr
ou
ps c
ont
ai
ne
d
wom
e
n
w
h
o
were classifie
d
according to t
h
e
WHO
classi
fication: osteoporotic, oste
open
ia, a
nd
normal [3][6]. The
ove
rall
m
o
d
e
l selects t
h
e im
p
o
r
tan
t
featu
r
es
f
r
o
m
t
h
e co
m
b
in
atio
n
o
f
po
ro
us tr
ab
ecu
lar
p
a
ttern w
ith
an
thr
opometr
i
c
feat
ure
s
f
o
r
o
s
t
e
op
or
osi
s
sc
reeni
n
i
n
cl
u
d
e
s
sel
ect
i
on
of
R
O
Is
, segm
en
tation, porous feature ext
r
action,
sel
ect
i
on
of t
h
e feat
u
r
es,
dec
i
si
on t
r
ee
vi
s
u
al
i
zat
i
on,
an
d m
easuri
n
g per
f
o
r
m
a
nce
t
o
classify
(Fi
g
u
r
e
1
)
.
We
will ex
p
l
ai
n
each
o
f
th
ese i
n
brief as b
e
l
o
w.
Fi
gu
re
1.
Feat
u
r
e sel
ect
i
o
n m
odel
An al
go
ri
t
h
m
was
devel
ope
d
t
o
pe
rf
orm
a seque
nt
i
a
l
pr
oc
edu
r
e t
o
sel
ect
regi
on
o
f
i
n
t
e
r
e
st
(R
OI
)
f
o
r
each patient
and segm
entation on
these
R
O
Is [8].
The
s
e
gm
ented im
a
g
es are
proce
s
sed to
obtain
porous
feature
s
from
each ROI
using
m
o
rphologic a
n
alysis. Por
ous
features are t
h
en c
o
m
b
ined with
a
n
thropometric
feat
ure
s
t
o
acq
ui
re a
k
n
o
wl
e
d
ge
base i
n
a
de
ci
si
on t
r
e
e
. T
h
e decision t
r
ee
shows
the
i
m
port
a
nt
feat
ures
of
t
h
e
co
m
b
in
atio
n
of po
rou
s
and
anth
ro
po
m
e
tric featu
r
es. Th
e m
o
rphol
ogic
feat
ures
that c
h
a
r
a
c
terize the
porous
on
a R
O
I
are
p
o
r
o
si
t
y
, hom
oge
ne
i
t
y
, and t
h
e
ori
e
nt
at
i
on
o
f
po
r
e
.
Po
ro
sity (Po
r) is ratio
b
e
tween
po
rou
s
area an
d
to
tal area o
n
a ROI. The p
o
rou
s
area is th
e to
tal
num
ber o
f
bac
k
pi
xel
i
n
bi
na
r
y
im
age and t
o
t
a
l
area on a R
O
I i
s
t
h
e
num
ber o
f
pi
xel
s
on
a R
O
I.
Hom
o
g
e
nei
t
y
can b
e
re
pres
e
n
t
e
d
by
JK a
n
d
JB
. JK
is
ratio b
e
tween
th
e
nu
m
b
er of sm
all
p
o
re an
d th
e t
o
tal nu
m
b
er o
f
p
o
re
on a R
O
I w
h
i
l
e
JB
i
s
rat
i
o
bet
w
een t
h
e n
u
m
ber of l
a
rge
po
re an
d t
h
e t
o
t
a
l
num
ber of
por
e o
n
a R
O
I. Th
e
sm
a
ll p
o
r
e is t
h
e to
tal nu
m
b
er of
p
o
re
wh
ich
h
a
s area le
ss
th
an
72
p
i
x
e
ls; th
e larg
e pore
is th
e to
tal num
b
e
r o
f
po
re
w
h
i
c
h
has
area m
o
re t
h
a
n
a
n
d
eq
ual
t
o
72
pi
xel
s
[
9
]
.
On
JK
an
d
JB
feat
ure
s
, l
a
bel
i
n
g
o
p
e
r
at
i
o
n
w
a
s use
d
to obtain the
num
b
er of
pore
in each R
O
I.
The
num
ber
of pore is
t
h
e
num
b
er of pore
whic
h
have a
r
ea 39
p
i
x
e
ls
o
r
larg
er. Th
ere are three step
s to ob
tain
th
e
orie
nt
at
i
o
n:
fi
n
d
t
h
e ce
nt
roi
d
[
1
0]
, fi
nd
t
h
e o
r
i
e
nt
at
i
o
n
[1
1]
and vi
sual
i
ze t
h
e
ori
e
nt
at
i
o
n
.
The ori
e
nt
at
i
on of p
o
re
fe
at
ure
i
s
re
pres
ent
e
d by
V, H,
a
nd O. V
i
s
rat
i
o
bet
w
ee
n t
h
e
n
u
m
ber of
vert
i
cal
po
re an
d t
h
e
t
o
t
a
l
num
be
r
of
po
re
on a R
O
I
.
H i
s
rat
i
o bet
w
een t
h
e
num
ber
o
f
horizontal pore
and t
h
e t
o
tal num
b
er of
pore
on a R
O
I.
O
i
s
rat
i
o
bet
w
ee
n t
h
e n
u
m
b
er of o
b
l
i
que
po
re an
d t
h
e
to
tal n
u
m
b
e
r of po
re
on
a ROI.Th
e
fin
a
l resu
lt is th
e po
r
o
us feat
ures c
o
nsi
s
t
i
ng
o
f
P, J
B
, JK,
V,
H, a
nd
M
.
These
features
are store
d
in a
data for st
atistical analysis usi
n
g the
decision tree.
Deci
si
o
n
t
r
ee
(C
4.
5
or
J.
48
m
e
t
hod
) i
s
u
s
ed t
o
cl
assi
fy
po
r
ous
an
d a
n
t
h
ro
p
o
m
e
t
r
i
c
feat
ure
s
a
s
bel
o
ngi
ng t
o
n
o
rm
al
, ost
e
ope
ni
a, o
r
ost
e
o
p
o
r
ot
i
c
. C
4
.5 al
g
o
ri
t
h
m
i
s
one o
f
dat
a
cl
assi
fi
c
a
t
i
on al
g
o
ri
t
h
m
s
wi
t
h
segment
ation
por
o
uus
extract
i
on
featu
res
s
e
lect
ion
evaluati
on
an
trop
ometricf
eatu
res
: ag
e,
h
e
igh
t
,
weigh
t
, BMI
class : normal,
osteopenia,
osteoporosis
Morphologic a
n
alysis
Statistical analysis
Por
o
us
Feature
s
d
ecis
i
on
tree
ROI
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:20
88-
870
8
I
JECE Vo
l. 5
,
N
o
. 1
,
Febru
a
ry
2
015
:78–
83
80
deci
si
o
n
t
ech
ni
que
w
h
i
c
h i
s
p
o
p
u
l
a
r a
n
d fa
v
o
re
d
due t
o
i
t
s
adva
nt
age
s
[
1
2
]
. To
per
f
o
r
m
deci
si
o
n
t
r
ee,
Weka
soft
ware
is us
ed.
The
decisi
on tree m
e
thod em
ploys a
recursi
v
e al
gorithm
.
At th
e st
art, all of
dat
a
are
co
nsid
ered
togeth
er at th
e root o
f
a
p
r
ed
iction
tree. Th
e d
a
t
a
are sp
lit o
n
th
e v
a
riab
le th
at resu
lts in
th
e larg
est
di
ffe
re
nce am
on
g t
h
e succ
e
ssi
ve n
o
d
es.
I
n
eac
h da
u
ght
er
node,
varia
b
les are a
g
ai
n exam
ined to
find the
p
r
ed
icto
r th
at
resu
lts in th
e
b
e
st sp
lit a
m
o
n
g
no
rm
al, o
s
teo
p
e
n
i
a,
o
r
o
s
t
e
o
poro
s
is. Sp
littin
g
con
tinu
e
s un
til
stopping c
r
iteria are
reache
d
or
un
til
further
splitting node doe
s not
i
m
pr
ove classi
fication [13]. Term
inal
no
des
(“l
eave
s
”) are
cl
assi
fi
e
d
as
n
o
rm
al
, os
t
e
ope
ni
a,
or
ost
e
op
o
r
osi
s
.
3.
R
E
SU
LTS AN
D ANA
LY
SIS
M
o
r
p
h
o
l
o
gi
cal
po
ro
us anal
y
s
i
s
perf
o
r
m
e
d on t
h
e i
m
ages usi
n
g t
h
e i
m
age pr
ocessi
ng s
o
ft
ware
.
Fi
gu
re 2
dem
onst
r
at
es m
o
rph
o
l
o
gi
cal
p
o
r
ous a
n
al
y
s
i
s
resul
t
based
o
n
o
u
r f
r
am
ewor
k ab
o
v
e. It
sho
w
s
m
o
rph
o
l
o
gi
cal
di
ffe
re
nt
i
a
t
i
on
of
o
u
r
cl
assi
fi
c
a
t
i
on cl
ass
(
n
o
r
m
a
l
,
ost
e
o
p
eni
a
, an
d
ost
e
op
or
osi
s
).
Step
No
rm
al
Osteo
p
e
n
ia
Osteo
p
o
r
o
sis
Selectio
n
ROI
Seg
m
en
tatio
n
Clo
s
in
g
ope
rat
i
o
n t
o
per
f
o
r
m
p
o
ro
sity
Filled
po
rou
s
to
ob
tain
t
h
e
area of pore
The
ori
e
nt
at
i
on o
f
po
r
ous
Fi
gu
re 2.
M
o
rp
hol
ogi
c A
n
al
y
s
i
s
A
decision tree
analysis consi
d
eri
ng m
o
rphological
por
ou
s
an
d an
t
h
ro
po
m
e
tr
ic f
eatur
es f
i
n
d
th
at th
e
im
port
a
nt
feat
ures
fo
r cl
assi
fy
i
ng
dat
a
we
re U, P
o
r, V
,
and M
.
as s
h
o
w
n i
n
Fi
g
u
r
e 3. T
h
e i
m
po
rt
ant
anthropom
e
tric
features were
U
a
n
d Por; t
h
e
im
portant m
o
rphologic
features we
re
V a
nd M.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE I
S
SN
:
208
8-8
7
0
8
Fea
t
u
r
e S
e
lecti
o
n o
f
t
h
e C
o
mbin
a
tion
o
f
Po
rou
s
Trab
ecu
l
a
r
with
An
thropom
etric Fea
t
u
r
es … (Enn
y
I
S
e
la
)
81
Fi
gu
re
3. Deci
s
i
on
t
r
ee 9-
fol
d
cross
-
val
i
d
at
i
o
n
Tab
l
e 1 shows
p
e
rcen
tag
e
accu
r
acy, sen
s
itiv
i
t
y, an
d
sp
ecificity o
f
d
ecision
tree testin
g
for
9
,
6
,
and
3
cro
ss
v
a
lid
ation
.
It sho
w
s th
at th
is test
in
g
resu
lts th
e best
eval
uat
i
on f
o
r
9-cr
oss val
i
d
a
t
i
on t
e
st
. Thi
s
m
ode
test h
a
s 8
7
.
04
fo
r accu
r
acy test; 8
7
.
80
% for sen
s
itiv
ity te
st,
an
d
8
8
.21
%
for sp
ecificity tes
t
. Th
e weigh
t
kap
p
a
i
nde
x, a
m
easure
of t
h
e a
g
ree
m
ent
bet
w
ee
n t
h
e
pre
d
i
c
t
e
d a
n
d act
ual
b
o
n
e
a
g
reem
ent
,
i
s
0.
79
8
2
.
Tab
l
e 1
.
Percen
tag
e
of
accu
r
acy,
sen
s
itifity, an
d
sp
escificity
test
K-Fold Accurac
y
(%
)
Kappa
statistic
Class
Sensitivity (%
)
Specificity (%
)
9 87.
04
0.
7982
Osteopor
osis
90.
90
86.
05
Osteopenia 77.
27
96.
77
Norm
al 95.
24
81.
81
Aver
age 87.
80
88.
21
6 87.
04
0.
7978
Osteopor
osis
90.
90
86.
05
Osteopenia 90.
48
87.
5
Norm
al 85.
71
87.
88
Aver
age 89.
03
87.
14
3 83.
33
0.
7422
Osteopor
osis
90.
90
81.
40
Osteopenia 68.
18
93.
75
Norm
al 95.
24
75.
75
Aver
age 84.
77
83.
33
The prese
n
t study fi
nds that the co
m
b
in
ation
of ag
e,
h
e
ight, weig
h
t
,
BMI and feat
ures
of the porous
of t
r
a
b
ecul
a
r m
o
rph
o
l
o
gy
of
i
n
t
e
rde
n
t
a
l
bo
ne i
s
usef
ul
i
n
i
d
ent
i
f
y
i
n
g
p
o
s
t
-m
enopausal
wom
a
n wi
t
h
l
o
w-
bo
ne
mass. In t
h
is study, t
h
e age of subject
s
(U) is conside
r
ed t
o
be the
one of
t
h
e m
o
st
im
port
a
nt
feat
ure
fo
r l
o
ss
o
f
bon
e m
a
ss. Th
is fi
n
d
i
n
g
is co
n
s
isten
t
with
[3
][6
]. In ad
d
itio
n, an
thropo
m
e
tric
featu
r
es BMI was n
o
t
i
m
p
o
r
tan
t
for id
en
tifying
women
with
low
BMD. It is no
t
con
s
i
s
t
e
nt
wi
t
h
[3]
.
T
h
e i
m
por
t
a
nt
po
r
ous
fea
t
ure i
s
po
r
o
si
t
y
(P
or
),
t
h
e
o
b
l
i
que
p
o
r
o
us
(M
),
an
d t
h
e
vert
i
cal
po
r
ous
(
V
)
.
T
h
e M
a
n
d
V
f
eat
ures
we
re c
onsi
s
t
e
nt
wi
t
h
[
14]
[
1
5]
. The Po
r feat
ure i
s
co
nsi
s
t
e
nt
wi
t
h
[
7
]
.
Se
veral
feat
ures
(
J
K, JB
, TB
, a
n
d B
B
)
ha
ve u
s
ed i
n
this study and they are
not
conside
r
ed
fa
ctors
for
os
teo
poro
s
is. In
t
h
is stud
y, th
e an
thropo
m
e
tr
ic an
d
radi
ographic fe
atures a
r
e a
n
alysis separately. The
accur
acy
of
ant
h
ropometric features a
nd
the
radi
ographic
f
eatu
r
es
h
a
v
e
a go
od
testing
with
80
.3
3% and 87
.04
%
accu
r
acy. Th
is
find
i
n
g is no
t co
nsisten
t
with
[3
].
Tab
l
e
2 s
h
ows
this
resea
r
ch can prove t
h
at features
se
l
ect
i
o
n
usi
n
g C
4
.
5
on
t
h
e c
o
m
b
i
n
at
i
on o
f
p
o
r
o
us
anda
nt
r
o
pom
eti
c
feat
ur
es r
e
su
l
t
s
a si
gni
fi
cant
val
u
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:20
88-
870
8
I
J
ECE Vo
l. 5
,
N
o
. 1
,
Febru
a
ry
2
015
:78–
83
82
Tabl
e
2. T
h
e
c
o
m
p
ari
s
on
o
f
resu
lts with o
t
her research
s
Pa
ra
m
e
ter
Lick
s
d
kk
(
2010)
Lee dan Whi
t
e
(
2005)
This research
M
e
thod of decision
tree
Classification and
R
e
gression Tree Analysis
(
C
ART
)
Classification and
R
e
gression Tree
A
nalysis (
C
A
R
T)
C4
.5
(
3
,
6
,
9
-
f
old validati
on)
Nu
m
b
er
of tr
aining
data
60 (
22 norm
a
l and
osteopenia; 38
Osteopor
osis)
65
54 (
11 norm
a
l,
22
osteopenia,
21
osteopor
osis)
Nu
m
b
e
r
of testing
data
60 (
22 norm
a
l and
Osteopenia; 38
Osteopor
osis)
28 (
7
Norm
al,
17
Osteopenia,
4
Osteopor
osis)
54 (
11 norm
a
l,
22
osteopenia,
21
osteopor
osis)
T
e
stiing featur
es
Antr
opo
m
e
tr
ic: height,
weight, BMI
,
age,
M1
, M2
,
M3
,
M4
M5
, M6
,
M7
,
M8
M9
, M1
0
,
M1
1
,
M
12,
M
13,
M
14.
Antropo
m
e
t
r
ic :
height,
weight,
BM
I
,
age,
M
o
r
phologic:
number of termini,
nodes per unit are
a
,
number of length o
f
struts segment
between termini and
nodes
.
Antr
opo
m
e
tr
ic : height,
weight, BMI
,
age,
M
o
r
phologic : por
,
JB,
JK, V,
H,
M
Gold-
S
tand
ard
(classtarget)
- lu
m
b
ar/fe
m
o
r
al
BMD
DXA
-
norm
a
l and
osteopenia/osteo
p
o
r
osis
-
l
um
ba
r
/
f
e
m
o
r
a
l
BM
D
DX
A
-
norm
a
l,
osteopenia,
and
osteopor
osis
-
l
um
ba
r
/
f
e
m
o
r
a
l
BM
D
DXA
-
norm
a
l,
osteopenia,
and
osteopor
osis
T
h
e m
o
st
im
por
tant
f
eature
age age
Age
The considered
f
eatures
M
8
,
M
3
, M
12,
B
M
I
Node:term
inus r
a
tio
Por
,
V, M
Accur
a
cy 88.
33%
82%
86.
67%
Accurac
y
of
antropo
m
e
tic
f
eatures testing
-
80.
33%
Accur
a
cy
of por
ous
f
eatures testing
-
-
87.
04%
M1: trabec
ular area / t
o
tal are
a
M2
: p
e
riph
ery
/ to
tal area
M3
: p
e
riph
ery
/ to
tal area
M4
: leng
th
/ t
r
ab
ecu
lar area
M5
: leng
th
/ t
o
tal area
M6
: term
in
al p
o
i
n
t
s / cm
2
M7
: term
in
al p
o
i
n
t
s / len
g
t
h
M
8
:
t
e
rm
i
n
al
poi
nt
s /
peri
ph
er
y
M9
: term
in
al p
o
i
n
t
s / trab
ecu
l
ar area
M
10:
bra
n
c
h
p
o
i
n
t
s
/
cm
2
M1
1
:
b
r
an
ch
po
in
ts
/
len
g
t
h
M
12:
bra
n
c
h
p
o
i
n
t
s
/
peri
ph
er
y
M1
3
:
b
r
an
ch
po
in
ts / trab
ecu
l
ar area
M1
4
:
b
r
an
ch
po
in
ts / term
in
al po
in
ts
Th
ere are so
me li
m
ita
tio
n
s
t
o
d
e
sign
of th
i
s
stud
y.
First,
th
e sam
p
le size is m
o
d
e
st,
particu
l
arly in
term
s o
f
th
e
nu
m
b
er of
d
a
ta with
no
rm
al class. Th
is
stud
y is
n
o
t
i
n
clud
ing
m
a
le p
a
tien
t
s an
d it on
ly to
ok
account
of the subj
ects’ age, weight,
hei
ght
, and BM
I.
Future st
udies shoul
d cons
ider other
anthropometric
feat
ure
s
f
o
r
ost
e
op
o
r
osi
s
, s
u
c
h
as
exe
r
ci
se, s
m
oki
ng,
o
r
use
o
f
m
e
di
cat
i
ons.
4.
CO
NCL
USI
O
N
Th
e stud
y addressed
th
e
u
tilit
y o
f
trab
ecu
lar
p
o
ro
u
s
feat
ures to
con
t
ribu
te
to
th
e feat
u
r
es
selectio
n
of
th
e th
e co
m
b
in
atio
n
o
f
porou
s trab
ecu
l
ar p
a
ttern
o
f
th
e m
a
n
d
i
b
l
e
with
an
th
rop
o
metric featu
r
es for
osteoporosis sc
reeni
ng
with 87
.04% accurac
y
. The selected features a
r
e
U, Por, V, and M. These feat
ures are
capable for classifiying phase
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE I
S
SN
:
208
8-8
7
0
8
Fea
t
u
r
e S
e
lecti
o
n o
f
t
h
e C
o
mbin
a
tion
o
f
Po
rou
s
Trab
ecu
l
a
r
with
An
thropom
etric Fea
t
u
r
es … (Enn
y
I
S
e
la
)
83
REFERE
NC
ES
[1]
Brandi, Mar
i
a L, 2009, Micro
a
rchiture, the k
e
y
to bone qua
lity
, Rheumatolog
y
, 48:iv3-iv8
,
The Oxford
Univers
ity
Press. DOI: 10.1
093/rheumatolo
g
y
/k
ep273
[2]
A
s
ano A
., Tam
b
e T.
, Tagu
chi A
., A
s
ano CM
.,
N
a
kam
o
to T
.,
Tanimoto K., Hin
a
moto T. danM
uney
a
su M., 20
06,
Extraction of tr
abecular structu
r
es of mandible ex
cluding too
t
h roots on dental panoramic radiographs usin
g
mathematica
l
m
o
rphology
, Proceeding
The 18th
International Co
nference
on
Pattern recognition (
I
CPR’06), Tok
yo.
[3]
Licks R.,
Licks
V., Ourique F.,
Bitten
c
ourt H.R
.
, Fontanel
la V., 2010, Developm
ent of predi
c
tion
tool for low bone
mass based on clinical
data an
d periap
ical rad
i
ograph
y
,
Den
t
o
m
axillofa
cia
l
Ra
diology
39
, 224
-230, The Br
itis
h
Institute of R
a
di
olog
y
.
DOI:
10.1259/dm
fr/23760876
[4]
Buie H.R., Campbell G.M., Klin
ck R.J
., MacNeil Joshua A, Bo
y
d
S.K., 2007,
A
u
tomatic segmentation of
cortical
and trabecular compartments based on a dual threshold t
echn
i
que for in vivo micro-CT bone analysis, Bone 4, 505-
515, Elsevier In
c. DOI:
10
.1016/j.bone 2007
.07.0
07
[5]
Da
vison K.
S.
Josse
H.
,
Simi
noski K.
, Adachi J.
D.,
Hanley
D.
A.,
Goltzma
n D., H
odsman A.B., 20
06, Bone Strength
:
The Whole is g
r
eater th
en
the sum of its p
a
rts,
Elsevier In
c. DOI;
10.
1016/j.semarthrit 2006
.04.002
[6]
Lee B.D., and
White S.C, 200
5, Age
and tr
ab
ecular feat
ur
es
of alv
e
olar
bone asso
ciated with
osteoporosis,
O
r
al
Surg Oral Med
Oral Pathol Oral Rad
i
olEndod
2
005; 100:92-8
.
DOI: 10.1016/j.tripleo
2004.11
.0
20
[7]
Hlaing, Thuzar, 2012,
Featur
e
Selec
tion
and Fuzzy
D
ecision
Tree for
Network I
n
trusion Detection,
Internat
ion
a
l
Journal of Infor
m
atics and Com
m
unication
Tech
nology
(IJ-ICT)
1(2):109~118.DOI: 10.11591/ij-ict.v1
i2.591
[8]
Sela E.I
., Har
t
ati S., Harjoko A.,
Ward
o
y
o R., Mudjosemedi M., 2
013, Segmentation on the dental periap
ical X-R
a
y
images for osteo
porosis screenin
g,
Intl
Journal of Advan
ced
Com
puter Science an
d Application
,4(
7
): 147-151.
[9]
Padilla F., 2005, Influen
ce of
the precision of spectral back
scatter
m
easurem
ents o
n
the
es
tim
at
ion
of stutter
e
rs si
ze
in cancellous bo
ne, Bon
e
Quan
tita
tiv
e Ultr
asound, pp
123-14, Springer.
[10]
Stojm
e
novic, M
ilos and
Nay
a
k, Am
i
y
a, 2007
, Direct
el
lipse f
itting
and m
e
asuring based on
shape boundar
i
es,
Advances in
Image and
Video
Technolog
y
Lectu
r
e Notes
in Comp
uter
Science 487
2: 221-235, Springer
[11]
Asatr
y
an D., Eg
iazarian K
., Kur
k
chiy
an
, 2010,
Orientati
on Estimation with Applicati
ons to Image Analy
s
is an
d
Registration,
International Journ
a
l Informa
tion
Theories and
App
lications
, 17(4)
:
303-311.
[12]
S
udarm
a M
and Hos
tiadi, The
Es
tablis
hm
ent o
f
Decis
i
on
Tree Model in Network Traffic In
ciden
t
Using C4.5
Method,
International Jo
urnal of Informatics and Communication Techn
o
logy
(IJ-IC
T)
3(1):23~29.DOI: 10.
11591/ij-ict.v3i1.5179
[13]
Han J.
and Kam
b
er., M., 2001
,
Data mining
con
cepts and
technique
, A
cadem
ic
P
r
es
s
,
US
A
[14]
Fields J.A., Lee
G.L., Liu X.
S.,
Jekir M.G., Guo
X.E., Keav
er
y
T.M., 2011, Influence of
v
e
rtical trab
ecular on
th
e
compressive strength of the human vertebr
a
,
Journal of Bon
e
and
Mineral Research
, 26(2): 263-26
9. DOI: 10.1002
/j
bmr.2007
[15]
Mosekilde L,
19
88,
Age-r
e
lated
changes in
v
e
rtebral tr
ab
ecular b
one ar
chitecture-
a
ssessed b
y
a new method, J
Bon
e
1998; 9; 247-50.
Evaluation Warning : The document was created with Spire.PDF for Python.