TELKOM
NIKA
, Vol. 11, No. 9, September 20
13, pp.
4997
~50
0
4
ISSN: 2302-4
046
4997
Re
cei
v
ed Fe
brua
ry 21, 20
13; Re
vised
May 31, 20
13
; Accepte
d
Ju
ne 14, 201
3
Automatic Segmentation Framework of Building
Anatomical Mo
use Mo
del for Bioluminescen
ce
Tomography
Abdullah Ala
l
i
Schoo
l of Com
puter Scie
nce
and En
gi
neer
in
g, Beiha
ng U
n
i
v
ersit
y
, Bei
j
i
ng,
Chin
a
e-mail: Ab
dul
la
h.M.F.Alali@g
m
ail.com
A
b
st
r
a
ct
Biol
umin
escen
c
e to
mo
grap
hy
is kn
ow
n as
a hi
gh
ly il
l-p
o
s
ed i
n
vers
e pr
o
b
le
m. T
o
i
m
pr
ove th
e
reconstructi
on
perfor
m
a
n
ce b
y
introduc
ing
anato
m
ical
str
u
ctures as a
priori
kn
ow
led
ge, an aut
omatic
seg
m
e
n
tatio
n
framew
ork has
bee
n pr
opos
ed
in this
pa
per t
o
extract the
mo
use w
h
o
l
e-
body
org
ans
a
n
d
tissues, w
h
ich
ena
bles
to
bui
l
d
u
p
a
h
e
terog
ene
ous
mous
e
mod
e
l for
rec
onstructio
n
of
biol
u
m
i
nesce
n
c
e
tomo
gra
phy. F
i
nally, a
n
in viv
o
mo
use ex
per
iment has b
e
e
n
cond
ucted to
evalu
a
te this framew
ork by us
i
n
g
an X-r
a
y computed tomogr
ap
hy system
and a
multi-view biolumin
escenc
e imaging syst
em
. T
he findings
sugg
est that the pr
op
osed
meth
od c
an r
eali
z
e
fast
a
u
t
omatic s
e
g
m
entatio
n
of mouse an
ato
m
i
c
a
l
structures, ulti
mate
ly en
ha
nci
ng the reco
nstr
uc
tion p
e
rfor
mance of bi
ol
u
m
i
nesce
nce to
mo
grap
hy.
Ke
y
w
ords
: me
dical i
m
ag
e pro
c
essin
g
, image
segmentati
on,
anato
m
ic
al
mo
use mod
e
l
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Biolumine
s
ce
nce tomo
gra
phy is one
of the optical molecular imaging m
odalitie
s,
enabli
ng real
-time no
n-inv
a
sive in vivo
imaging
of labelled m
o
le
cules in
biolo
g
i
cal o
r
ga
nism
s.
Due
to its hi
gh
sen
s
itivity and
lo
w
co
st, biolumi
n
e
s
cen
c
e tom
o
grap
hy ha
s
attracte
d mu
ch
attention over the past d
e
cade [1-3]. It is capabl
e
of
obtainin
g
thre
e-dim
e
n
s
iona
l distrib
u
tion
of
the internal b
i
olumine
s
cent
signal
sou
r
ce, provid
ing
an effective way of inform
ation acqui
sition
and q
uantitati
v
e analysi
s
f
o
r di
sea
s
e
progre
s
sion, tu
mour dete
c
tio
n
as well
a
s
drug effica
cy
[4
-
6]. One of the chall
eng
es
of biolumine
s
cen
c
e to
mo
graphy is that multiple scattering of ph
oton
s
prop
agatin
g throu
gh bi
olo
g
ical ti
ssue
s makes
re
co
nstru
c
tion
a
highly ill-p
o
sed p
r
oble
m
. As
repo
rted
in related existin
g
literatu
r
e, rese
ar
che
r
s h
a
ve propo
se
d some
meth
ods to
solve
the
inverse pro
b
l
e
m, includi
ng
permi
ssibl
e source
region
approa
ch, mu
lti-spe
c
tral informatio
n based
algorith
m
and
so on [7-10]. The ba
sic id
ea is to re
du
ce the n
u
mb
er of un
kno
w
n variable
s
o
r
to
increa
se the
amount of
kno
w
n bo
unda
ry me
a
s
ureme
n
ts for re
co
nst
r
u
c
tion that i
s
mathemati
c
al
ly a set of undetermi
ned li
near e
quatio
n
s
.
Similarly, to attain more inform
ation as
a p
r
io
ri
kno
w
ledg
e
for biol
umi
nesce
nce
tomography, we utilized an
atomical
stru
cture
s
of the experim
ental
mouse in this paper. Th
e g
o
ld
stand
ard
to
extract the
a
natomic
al
structures is u
s
ing ma
nual
segmentatio
n. It usu
a
lly ta
ke
s
hours to
co
mplete the
e
n
tire inte
ra
cti
v
e pro
c
e
d
u
r
e
even
by a
skill
ed
user.
With the
rap
i
d
developm
ent
of automatic se
gm
entati
on algo
rithm
s
, the pro
c
e
ssi
ng efficie
n
cy ha
s be
en
signifi
cantly i
m
prove
d
[11-15]. Ho
wever, most
metho
d
s a
r
e
develo
ped for some
spe
c
ific
org
a
n
s
or tissue
s, re
quirin
g
intera
ctive operatio
ns, wh
i
c
h a
r
e
not appro
p
ri
ate for wh
ole
-
body auto
m
atic
appli
c
ation
s
. Therefore, a
gene
ral aut
o
m
atic segm
e
n
tation frame
w
ork h
a
s b
e
e
n
pro
p
o
s
ed in
this
pape
r to
extract the
whole
-
body
anato
m
ical
structu
r
es of
the exp
e
rime
ntal
mo
use, whi
c
h
could
be ap
plied
to build
a
hetero
gen
eo
us m
o
u
s
e
model to
e
nhan
ce th
e
perfo
rma
n
ce of
biolumin
esce
nce tomo
grap
hy.
2. Rese
arch
Metho
d
To acq
u
ire the experi
m
e
n
tal datasets, the
Caliper Life Science’s Spe
c
trum
CT wa
s
use
d
, whi
c
h i
s
an
integ
r
ati
v
e platform th
at com
b
ine
s
biolumin
esce
nce
and
fluorescen
c
e i
m
a
g
ing
with X-Ray CT scannin
g
. Here, two m
odalitie
s inclu
d
ing biol
umin
escen
c
e ima
g
ing an
d X-Ray
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 9, September 201
3: 499
7 – 5004
4998
CT
were a
p
p
lied to th
e
followin
g
e
x
perime
n
t. For X-Ray im
aging, the
cone-bea
m X-ray
gene
rato
r wa
s ope
rated in
a continu
ous mode wi
th the tube volta
ge being 4
5
kVp, where 36
0°
proje
c
tion
s were
sca
nned.
For bi
olumin
escen
c
e
ima
g
ing, a
coole
d
CCD
with
13.5
μ
m×13.5
μ
m
pixel size wa
s involved in
taking the mu
lti-views
of th
e optical ima
ges. The exp
e
rime
ntal mo
use
wa
s inje
cted
throug
h the
caudal tail vei
n
with Fe
ne
stra L
C
which is an io
dinate
d
lipid em
ulsi
on
blood
pool
co
ntrast a
gent
helpin
g
overcome the p
r
o
b
l
em of inhe
re
ntly soft tissu
e co
ntra
st in
CT
imaging, follo
wed by 0.3
m
l
of anesth
etic at a 0.15g/ml
con
c
e
n
tratio
n via intrap
eri
t
oneal inje
ctio
n.
Then, a
hom
e-ma
de lumi
nesce
nt bea
d
wa
s impla
n
t
ed into the
mouse. Since the lumin
e
s
cent
bead was
wrappe
d in a plastic mate
rial
, it coul
d be easily detect
e
d
by CT, enabl
ing examinati
on
of the reconst
r
uctio
n
accu
racy
of biolumi
nesce
nce imaging.
2.1. Automa
tic Segmenta
tion Frame
w
ork
2
.
1
.
1
.
B
o
n
e
st
r
u
ct
u
r
es
To a
u
tomatically extract
b
one
stru
ctu
r
e
s
, a th
re
sh
old
i
ng meth
od
b
a
se
d o
n
the
prin
ciple
of maximum
entropy
wa
s utili
zed
he
re, be
cau
s
e
the skeleton
gene
rally sh
ows
the
hig
h
est
contrast on
CT imag
es.
This app
ro
ach d
e
termi
nes the opti
m
al seg
m
en
tation thresh
old
automatically, where the ba
sic id
ea is to
maximize
the
total information entro
py of the object an
d
backg
rou
nd
after segme
n
t
ation. The p
r
ocedu
re
ca
n
be math
em
atically de
scribed a
s
follo
ws.
Suppo
sing a discrete
ra
nd
om
varia
b
le
v
stand
s fo
r a
thre
shold
value by which the pixel
s
on
a
CT im
age
are divided
int
o
two
group
s, obj
ect
and
ba
ckgro
und,
wh
ere
the
obje
c
t is
always
brighte
r
th
an
the b
a
ckg
r
o
und i
n
this case. A
s
su
min
g
the
pro
babi
lity distributio
ns
of the
obj
ect
and b
a
ckg
r
ou
nd,
D
O
a
nd
D
B
, are defin
ed
as Eq
uation
(1), the
corre
s
po
ndin
g
entropie
s
,
H
O
and
H
B
, are expre
s
sed a
s
Equa
tion (2)
respe
c
tively.
v
1
-
L
v
3
v
v
2
v
v
1
v
v
v
v
2
v
1
v
0
P
-
1
P
...,
,
P
-
1
P
,
P
-
1
P
,
P
-
1
P
:
P
P
...,
,
P
P
,
P
P
,
P
P
:
B
O
D
D
(1)
ln
)
(
1
)
1
ln(
)
(
v
v
v
B
v
v
v
O
P
H
P
v
H
P
H
H
P
v
H
(2)
Whe
r
e
v
0
i
P
i
v
P
,
v
0
i
v
ln
P
-
i
i
P
H
,
1
-
L
0
i
ln
P
-
i
i
P
H
with
P
b
e
ing
the p
r
ob
abil
i
ty distributio
n
,
variable
s
v
a
nd
i
being n
a
tural nu
mbe
r
s,
and co
nsta
nts
V
and
L
bei
ng po
sitive integer. Here
by
,
the entropy o
f
the whole
CT imag
e
can
be
wri
tten
as Equation
(3), and th
e o
p
timal threshol
d
value sh
own as Equatio
n (4) will be o
b
tained
when E
quation (3) at
tains its maxi
mum value.
v
v
v
v
v
v
B
O
P
H
H
P
H
P
P
v
H
v
H
v
H
1
)
1
(
ln
)
(
)
(
)
(
(3)
))
(
(
max
arg
v
H
T
v
optimal
(4)
2.1.2. Bod
y
Outline and
Lungs
Although th
e
body o
u
tline
and l
ung
s d
o
not exhi
bit hi
gher contrast
on
CT i
m
ag
es, they
occupy a
rel
a
tively large
r
are
a
. Thu
s
,
a re
gion
gro
w
ing
method
com
b
ine
d
wi
th thre
shol
din
g
is
applie
d he
re
to auto
m
ati
c
ally
segm
e
n
t these
two st
ru
cture
s
.
The
r
e
usu
a
lly exist
so
me
disa
dvantag
e
s
when
imag
es
are
proce
s
sed o
n
ly
by
usin
g a
n
in
dividual m
e
th
od. On th
e o
n
e
hand, ba
ckg
r
ound n
o
ise whose greysca
l
e value is
cl
ose to the o
b
j
ect will be in
evitably left o
n
the segm
enta
t
ion results u
s
ing the
singl
e thre
shol
di
n
g
approa
ch. On the othe
r hand, it will result
in over-segm
entation o
r
u
nder-segm
en
tation us
i
ng t
he si
ngle
regi
on g
r
owi
ng
method
whe
n
the
obje
c
t po
sse
s
ses an
une
ven gray lev
e
l dist
ribut
io
n or a fu
zzy
boun
da
ry. Whe
n
integ
r
ating
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Autom
a
tic Segm
entation Fram
ewo
r
k of
Buildi
ng Anat
om
ical Mou
s
e Model…
(Abdulla
h Alali)
4999
thresholdi
ng and
region g
r
owi
ng,
we could
avoid
th
e above p
r
ob
lems by taki
n
g
the followin
g
step
s. Suppo
sing
f
(
x
,
y
) is the greysca
l
e value function for an i
m
age, an o
p
timal greysca
l
e
threshold
T
1
can b
e
achie
v
ed usin
g the maximum
cla
ss
di
stan
ce method, on
e of the cla
s
sical
thresholdi
ng algorith
m
s.
Afterwa
r
d
s
, we
therefo
r
e get
Equation (5).
1
1
T
y)
f(x,
,
0
T
y)
f(x,
,
1
)
,
(
y
x
g
(5)
Furthe
rmo
r
e,
assumin
g
R
stand
s for t
he seed
regi
on, the mea
n
grey
scale
value
m
and the
stand
ard d
e
viation
of the region
R
are
resp
ectively d
e
fined a
s
:
R
l
k
R
l
k
m
l
k
f
n
l
k
f
n
m
)
,
(
2
)
,
(
)
,
(
1
)
,
(
1
(6)
With
f
(
k
,
l
) be
ing the g
r
ayscale val
ue fu
nction
and
n
being th
e n
u
mbe
r
of pix
e
ls in th
e se
ed
regio
n
. The criterio
n for growth with
a
constraint co
n
d
ition is given
in:
)
,
(
)
,
(
1
2
T
l
k
f
T
m
l
k
f
(7)
Whe
r
e
C
T
m
T
1
2
with
T
C
being a
co
ntrol varia
b
le.
The initial value of
T
C
can
be estimate
d
according to t
he initial
seed region,
and it w
ill
be adaptively adjusted du
ring the regi
on
growi
n
g
pro
c
ed
ure un
til the target regi
on i
s
com
p
letely extracted.
2.1.3. Other
Organ
s
Other maj
o
r organ
s in
cl
uding he
art
and liver will
be extracte
d by an atlas ba
sed
automatic
se
gmentation a
ppro
a
ch. The
atlas used in
this study was devel
ope
d
from ten set
s
of
mice trainin
g
data a
c
qui
re
d by CT,
wh
ose
anatom
i
c
al structu
r
e
s
have already
been
manu
a
l
ly
segm
ented. T
he follo
wing f
our
step
s a
r
e
mainly in
volv
ed in m
a
them
atically de
riving this
metho
d
.
To build
an a
v
erage
-sha
pe
atlas
: Firstly, one volum
e
data is
sele
ct
ed from th
e ten sets of mi
ce
training
data
as the
be
n
c
hma
r
k data.
Then, t
he
affine re
gistration is
pe
rforme
d u
s
ing
an
algorith
m
pre
s
ente
d
by Sla
g
molen
d
et al
. [16].
Afterword
s
, a no
n-ri
gid re
gist
ratio
n
ba
sed
on a
B-
spline transf
ormation model is
utilized to
process the datasets, where t
he regist
rati
on
measurement
is determin
ed by
the weighted
sum
of the mutual informati
on and surf
ace
distan
ce. The
non-rigid re
g
i
stration i
s
co
ndu
ct
ed between every two dataset
s, so each d
a
taset
D
i
will be
pro
c
e
s
sed nin
e
times by Equ
a
t
ion (8
), finall
y
generating
90 tran
sform
a
tion fields
F
ij
(i
≠
j)
.
The me
an va
lue of the
tra
n
sformation fi
eld for ea
ch
dataset ca
n b
e
cal
c
ul
ated by
Equation (9),
and sub
s
eq
u
ently we attain the averag
e
-
sh
ape atla
s
sho
w
n a
s
Eq
uation (1
0).
i
ij
j
i
D
F
D
(8)
j
i
ij
i
F
F
9
1
(9)
10
1
10
1
i
i
i
atlas
D
F
D
(10)
To roug
hly l
o
cali
ze th
e
orga
ns
:
Thi
s
pro
c
e
dure
can
be
gen
e
r
ally rega
rde
d
a
s
the
para
m
eter (
x
,
y
,
z
) adju
s
tme
n
t for the 3D transl
a
tion of the regi
stratio
n
, where
x
,
y
,
z
r
e
spec
tively
stand
for th
e
offsets on
the
coronal,
sagittal,
a
n
d
tran
sversal
plane
s. Pri
o
r to sele
cting
an
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 9, September 201
3: 499
7 – 5004
5000
approp
riate v
a
lue fo
r
z
,
a b
i
nary filter [1
7
]
is empl
oyed
to
rou
ghly
se
arch the l
o
cation of a
n
orga
n
from an
exp
e
rime
ntal dat
aset. Th
en,
assign th
e
slice n
u
mbe
r
of the co
ron
a
l view
with
a
maximum gray value for
z
.
After finding t
he bri
ghte
s
t slice fro
m
the
averag
e-sh
ap
e atlas, move
it
to where
z
i
s
, and (
x
,
y
) can be th
eref
ore a
d
ju
sted
by mappin
g
the atla
s to the expe
rime
ntal
dataset.
To m
a
ke
preci
s
e
registration
: A
multi-resol
u
tion regi
stration
[18] is utilized to ensure
the
comp
utationa
l efficien
cy an
d rob
u
stn
e
ss,
who
s
e
ba
sic idea i
s
to ad
d a pyramid fi
lter [19] befo
r
e
makin
g
the registration b
e
twee
n the fixed im
age
a
nd moving i
m
age. Fu
rthe
rmore, a B-spline
transfo
rmatio
n based o
n
mutual information a
s
a
simila
rity measu
r
e i
s
appl
ied at last, which
enabl
es the d
e
formatio
n of an atlas o
r
ga
n to c
onverge
to the one of the experim
e
n
tal dataset.
2.1.4. Whole-bod
y
Integra
t
ion
After obtaini
ng the se
pa
rated volume
s
of org
a
n
s
and tissue
s
based on th
e above
segm
entation
process, we
need to fuse them
into one volume
data. Duri
ng
the automatic
integratio
n, p
r
ioritie
s
have
been
set for different
org
ans a
nd tissu
e
s to elimin
ate the inevita
b
le
overlap
s
and
hole
s
. Fu
rth
e
rmo
r
e, to
d
e
scrib
e
th
e
b
ehaviou
r
of i
n
ternal
biol
u
m
inesce
nt si
gnal
s
traveling in
sid
e
living subje
c
ts, the
co
rre
s
po
ndin
g
opti
c
al p
r
o
pertie
s
for different
parts have
be
en
assign
ed, which a
r
e m
e
a
s
u
r
ed by diffu
si
on opt
i
c
al to
mography. Fi
nally, a heterogen
eou
s mo
use
model ha
s be
en com
p
leted
.
2.2. Bioluminesce
nce To
mograph
y
Recons
truc
tio
n
For biol
umin
escen
c
e tom
ogra
phy, the
diffusion e
q
uation an
d the Ro
bin bo
unda
ry
con
d
ition are
employed to model the
light propag
ation in biolo
g
ical tissue
s [20], which are
defined a
s
:
R
B
D
a
x
x
x
x
x
x
,
(11)
R
v
D
r
f
x
x
x
x
x
,
0
2
(12)
Whe
r
e
D
is the diffus
i
on coeffic
i
ent;
a
is the ab
sorption
coeffici
ent;
x
is
th
e
po
s
i
tion
ve
c
t
or
;
B
is
the biolumin
e
s
cent so
urce distrib
u
tion;
is the photo
n
flux density;
R
is the regi
on
of biologi
cal
tis
s
u
es
;
f
is
the boun
da
ry mismat
ch factor
betwe
e
n
the biolo
g
i
c
al tissue
s a
nd air;
r
i
s
t
he
refra
c
tive ind
e
x of the biological tissu
e
s
;
v
is the u
n
it outward n
o
rmal on
R
; and
R
i
s
the
boun
dary
of
the biol
ogical
tissue
s. In
orde
r to
sim
p
lify the follo
wing
comput
ation in
stea
d
of
solving
the a
bove diffu
sio
n
eq
uation
d
i
rectly,
the li
near relation
ship
bet
ween
the me
asured
outgoin
g
phot
on den
sity on
the boun
dary and the un
kn
o
w
n
sou
r
ce distri
bution
is built up in t
he
matrix-vec
tor form:
P
Κ
B
(13)
Whe
r
e
K
i
s
t
he
system
m
a
trix, standi
n
g
for th
e o
p
tical
pro
p
e
r
ties of biol
ogical
tissu
es;
P
i
s
t
he
measured out
going p
hoton
den
sity on the boun
dary.
The recon
s
truction
proce
dure
in bi
olu
m
i
nesce
nce tomography aims
to re
co
ver
the
sign
al so
urce
distributio
n
B
in Equation (13). Ho
weve
r, it is an unde
rdete
r
mine
d system of linear
equatio
ns wit
h
fewer eq
u
a
tions than
u
n
kn
owns
,
wh
ich i
s
known
as a
n
ill-po
sed
p
r
oble
m
. A
popul
ar m
e
th
od [21] to
re
d
u
ce
the ill
-po
s
ed
ne
ss is to
introd
uce a
n
a
tomical
st
ru
cture
info
rma
t
ion
as
a prio
ri
knowl
edge. T
he simpl
e
st solution to Equat
ion (1
3) is expresse
d as Equation (14),
whe
r
e th
e le
a
s
t squa
re
ap
proa
ch
ha
s
b
een
applie
d [
22], but it u
s
u
a
lly magnifie
s
noi
se e
r
ror.
To
further enh
a
n
ce t
he p
r
a
c
ticability, the Ti
kh
onov re
gulari
z
atio
n method ha
s been co
mmo
nly
utilized for an alternative [23]. The o
b
ject function with the
l
2
norm co
nstraint is given
as
Equation (15
)
,
wh
ere
is the re
gula
r
i
z
ation
param
eter a
nd
2
2
st
and
s f
o
r t
h
e
l
2
norm.
Therefore, th
e intern
al sou
r
ce
dist
ributio
n ca
n be fin
a
l
l
y solved by
applying
optimal minimi
zat
i
on
to the object functio
n
.
2
2
2
1
min
arg
P
KB
B
(14)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Autom
a
tic Segm
entation Fram
ewo
r
k of
Buildi
ng Anat
om
ical Mou
s
e Model…
(Abdulla
h Alali)
5001
2
2
2
2
2
1
min
arg
B
P
KB
B
(15)
3. Results a
nd Discu
ssi
on
The
acquired
CT
proje
c
tion d
a
ta
wa
s ori
g
inally in
DICOM fo
rmat, whi
c
h
wa
s the
n
conve
r
ted int
o
3D volume
data by the
FDK algo
rith
m [24]. Afterwards, the vo
lume data wit
h
a
size of
512
×5
12×512
wa
s
automat
ically se
gmente
d
,
whe
r
e th
e tot
a
l a
c
cumulat
ed
segm
entat
ion
time is less t
han 10 mi
nut
es. The
re
sul
t
s are di
spl
a
yed in the foll
owin
g pictu
r
e
s
, amon
g whi
c
h
the sep
a
rate
d volume
s of orga
ns a
nd tissue
s are
de
scribe
d in Fig
u
re 1 to 4
wh
ile the ultimate
whol
e-b
ody integratio
n is
visuali
z
ed in
Figure 5.
The
re
sult in
Figure 1
a
su
gge
sts th
at th
e th
resholdin
g
meth
od
ba
sed
on
the
principl
e of
maximum
ent
ropy can be utilized
to provide
a
prec
ise segmentati
on of bo
ne
st
ructures. It took
less than 5 seco
nd
s to co
mplete the au
tomatic pro
c
e
dure. As sho
w
n in Figu
re 1b, the surfa
c
e of
the experim
e
n
tal mouse b
ody is ren
d
e
r
ed ba
sed o
n
its seg
m
ented
body outline.
Figure 1. The
Results of Bone an
d Bod
y
Segment
ation: (a) the
se
gmented
b
o
n
e
stru
ctures, (b)
the segm
ente
d
body outlin
e.
Figure 2. The
Results of Lu
ng Segme
n
ta
tion: (a)
a CT slice in tran
sversal view wit
h
a marked
regio
n
in the lung
s, (b) the
segm
ented lu
ngs rend
ered
in 3D.
Figure 3. The
Results of Heart Segme
n
tation: (a)
a
CT slice in tran
sversal view
with a marke
d
regio
n
in the heart, (b
) the
segm
ented h
eart re
nde
red
in 3D.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 9, September 201
3: 499
7 – 5004
5002
The se
gme
n
tation re
sults
of the mouse
orga
ns
a
r
e gi
ven in Figure
2 to 4. The lungs a
r
e
pro
c
e
s
sed by
a
re
gion gro
w
ing method
combi
ned wi
t
h
thre
sh
oldin
g
, whi
c
h i
s
re
latively easy
to
reali
z
e. The
results sho
w
n in Figure
2 sugg
est th
at this appro
a
ch gu
arante
e
s goo
d regi
on
con
s
i
s
ten
c
e.
Since th
e oth
e
r
soft tissue
s
su
ch
as the
hea
rt an
d liv
er a
r
e
in the
l
o
we
r g
r
ay-val
ue
contrast,
a
n
atlas ba
sed automatic
se
gmentatio
n
a
ppro
a
ch i
s
a
pplied to
obt
ain the
re
sul
t
s
exhibited in F
i
gure 3 a
nd Fi
gure 4.
Figure 4. The
Results of Li
ver Segment
ation: (a)
a
CT slice in
co
ronal view
with a marked region
in theliver, (b) the segme
n
ted liver re
nde
red in 3
D
.
Figure 5. The
Hetero
gen
eo
us Mou
s
e M
o
del after wh
ol
e-bo
dy Integration.
Figure 6 sho
w
s th
e two
reco
nstructio
n
results for
bi
olumine
s
cen
c
e tomog
r
aph
y. In th
e
first
ca
se, a
homo
gen
eo
us m
o
u
s
e
m
odel
without
any a
natom
ical i
n
form
ation i
s
utili
zed
to
recover the i
n
ternal
biolu
m
i
nesce
nt source
sig
nal
, a
n
d
the
co
rresp
ondin
g
results
can
be
see
n
in
Figur
e 6
a
.
I
n
t
he
se
co
nd
ca
se,
t
h
e
a
n
a
t
o
mical
st
ru
ct
ure
i
s
t
a
ke
n a
s
a p
r
io
ri
kno
w
le
dge
fo
r the
r
e
co
ns
tr
uc
tion
, w
h
ich
o
b
v
io
u
s
ly le
ad
s
to
mo
r
e
acc
u
ra
te
r
e
c
o
ns
tr
uc
tio
n
re
su
lts
s
h
ow
n
in
F
i
gu
r
e
6b.
Figure 6. The
Recon
s
tru
c
tion Re
sult
s of Biol
umine
s
ce
nce To
mog
r
a
phy in Cro
s
s-se
ctional
Views: (a) the
reco
nst
r
u
c
tio
n
results ba
sed on a hom
o
gene
ou
s mou
s
e mod
e
l, (b)
the
recon
s
tru
c
tio
n
results ba
sed on a hete
r
ogen
eou
s
mo
use mo
del, where the
circl
e
rep
r
e
s
ent
s the
real lo
cation
of the biolumi
nesce
nt sou
r
ce.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Autom
a
tic Segm
entation Fram
ewo
r
k of
Buildi
ng Anat
om
ical Mou
s
e Model…
(Abdulla
h Alali)
5003
4. Conclusio
n
To validate
th
e fea
s
ibility of
this
method,
t
he ra
w
data
s
ets of th
e two
imagin
g
m
o
d
a
lities
have be
en a
c
hieved via an
in vi
vo
mo
use experi
m
ent.
Subse
que
ntly
processe
d
by the pro
p
o
s
ed
segm
entation
method, the
anatomi
c
al m
ouse mod
e
l h
a
s b
een a
u
to
matically built
up, whi
c
h
ca
n
guarantee co
mputational efficien
cy.
Further
valid
at
e
d
by the reconstructio
n
compa
r
ison, this
hetero
gen
eo
us mod
e
l is capabl
e of ensuring t
he a
ccura
cy of biolu
m
inesce
nce tomog
r
ap
hy.
Referen
ces
[1]
Prescher
JA, C
ontag
CH. Gu
i
ded
b
y
th
e
lig
ht: visual
izi
ng
bio
m
olec
ular
proc
esses i
n
livi
ng
anim
a
ls
w
i
t
h
biol
umi
nesce
n
c
e.
Current Opi
n
io
n in Ch
e
m
ic
al Bio
l
ogy.
2
0
1
0
; 14(1): 80-8
9
.
[2]
Lep
pa
nen
O, Ekstrand M, B
r
aese
n
JH. B
i
olumi
nesc
enc
e
imag
in
g of
e
nerg
y
de
pl
etio
n i
n
vasc
ula
r
patho
lo
g
y
: p
a
tent ductus arte
riosus a
nd ath
e
roscl
erosis.
J
ourn
a
l of Bio
p
h
o
tonics
. 20
12;
5(4): 336-
34
4.
[3]
Hashiz
ume R,
Oza
w
a
T
,
D
i
nca EB. A h
u
man
brai
nste
m gliom
a
xen
ograft mod
e
l
ena
ble
d
fo
r
biol
umi
nesce
n
c
e imag
ing.
Jo
urna
l of Neuro-
Oncolo
gy
. 201
0; 96(2): 15
1-1
59.
[4]
Hu H, L
i
u J, Y
ao L. R
e
a
l
-time bi
olum
inesc
e
nce a
nd tom
o
g
r
aph
ic ima
g
in
g
of gastric ca
n
c
er in
a n
o
ve
l
orthotop
ic mou
s
e mode
l.
Oncolo
gy Rep
o
rts
. 201
2; 27(6): 19
37-1
943.
[5]
Shavku
nov A,
Panov
a N, Pra
s
ai A. Bio
l
umi
n
escenc
e meth
odo
log
y
f
o
r the
detectio
n
of pr
otein-
protei
n
interacti
ons
w
i
thin the v
o
ltag
e-gate
d
so
d
i
u
m
chan
nel m
a
cromol
ecu
l
ar
comple
x.
Ass
a
y an
d Dru
g
Devel
o
p
m
ent T
e
chn
o
lo
gi
es
. 2012; 10(
2): 148
-160.
[6]
T
u
rner D, S
un J,
Nels
on
S.
Quantitat
ive ana
l
y
sis of
bacte
r
i
o
p
h
age
pl
aq
ue
expa
nsio
n
b
y
biol
umi
nesce
n
c
e imag
ing.
L
u
m
i
nesc
ence
. 2
012; 27(
2): 167
-169.
[7]
Con
g
W
,
W
ang G.
Iterative meth
od for b
i
ol
umin
escenc
e tomogr
aphy
ba
sed on th
e rad
i
ative transp
o
rt
equ
atio
n
. Proc
eed
ings
of the
Societ
y
of Ph
oto-optic
al
Inst
rumentati
on E
ngi
neers, S
an
Dieg
o
. 2
006
;
631
8: 318
26-3
182
6.
[
8
]
Cong W
,
W
a
n
g
G. Boundary integral metho
d
for bioluminescence t
o
m
o
g
r
a
p
h
y
.
Jour
nal of
Biomed
ica
l
Optics.
2006; 1
:
02050
3-1-
020
503-
3.
[9]
Klose A
D
. Ntzi
achristos V, H
i
e
lsch
er AH,
T
he inv
e
rse so
ur
ce pro
b
lem
ba
sed o
n
the ra
di
ative transfer
equ
atio
n in opti
c
al mol
e
cul
a
r i
m
agi
ng.
Journ
a
l of Co
mp
utati
ona
l Physics
. 2
005; 20
2: 323-
345.
[10]
Slavi
ne NV, L
e
w
i
s
MA. Rich
e
r
E,
Antich PP. Iterative
reco
n
s
truction meth
od for li
ght emi
tting sourc
e
s
base
d
on the d
i
ffusion e
q
u
a
tio
n
.
Medica
l Phy
s
ics
. 2006; 3
3
: 61-6
8
.
[11]
Smeets D, Lo
eckx
D, Stijn
e
n
B.
Semi-a
utomatic l
e
vel s
e
t segme
n
tatio
n
of liv
er tumo
rs combi
n
in
g
a
spiral-sc
a
n
n
in
g
techniq
ue
w
i
t
h
superv
i
sed f
u
zz
y
p
i
xel cl
as
sificatio
n
.
Medi
cal Ima
ge An
a
l
ysis
. 200
9;
14(1): 13-
20.
[12]
Piccini
D, Litt
mann A,
Nie
ll
es-Vall
e
sp
in S
.
Resp
ir
ator
y self-nav
igati
o
n
for
w
h
o
l
e-
hea
rt bright-b
loo
d
coron
a
r
y
MRI:
Methods
for r
o
bust is
olati
o
n
and
a
u
tomatic
segme
n
tatio
n
of the
bl
oo
d p
ool.
M
agn
eti
c
Reso
nanc
e in
Medici
ne
. 20
12
; 68(2): 571-5
7
9
.
[13]
Bai J, T
r
inh T
H
, Chu
ang
KH.
Atlas-bas
ed
au
toma
tic mo
use
brai
n im
age
se
gmentati
o
n
rev
i
sited: m
o
d
e
l
compl
e
xit
y
vs. i
m
age re
gistrati
on.
Magn
etic R
e
son
anc
e Ima
g
in
g
. 201
2; 30(
6): 789-7
98.
[14]
Lai J,
Xie M. S
egme
n
tatio
n
lu
ng fiel
ds
in th
or
acic CT
scans usin
g man
i
fold
method.
Te
l
k
om
ni
ka
. 20
12;
10(5): 10
05-
10
14.
[15]
Sun J. A
n
e
w
imag
e s
egm
entat
io
n
alg
o
ri
thm an
d its
a
pplic
atio
n i
n
l
e
ttuce o
b
j
e
ct
segme
n
tatio
n
.
Te
lkom
n
i
ka
. 20
12; 10(3): 5
57-
563.
[16]
Slagm
ole
n
P,
Elen
A, Seg
her
s D,
Lo
eck
x
D,
Maes F
,
H
aus
termans K.
Atla
s ba
se
d l
i
v
e
r
se
gm
en
ta
ti
o
n
usin
g no
nrig
id
registratio
n
w
i
th a B-splin
e transfor
m
ati
on
mod
e
l
, Pr
ocee
din
g
s of Internatio
na
l
Confer
ence
o
n
Medic
a
l Im
ag
e Com
puti
ng
a
nd C
o
mp
ut
er
Assisted Interv
entio
n. Brisb
a
n
e
. 20
07; 1
97-
206.
[17]
Martinez-C
orra
l M, Andres P,
Z
apata-Ro
dri
guez CJ. T
h
ree-dime
nsi
ona
l superr
e
sol
u
tio
n
b
y
ann
ul
a
r
bin
a
r
y
fi
lters.
Optics Co
mmu
n
i
c
ations
. 19
99;
165(
4-6): 26
7-278.
[18] Alhic
h
ri HS, K
a
mel M. Mu
lti-r
e
sol
u
tion
ima
g
e
reg
i
stratio
n
u
s
ing m
u
lti-cl
as
s Haus
dorff fra
c
tion.
Pattern
recog
n
itio
n lett
ers
. 2002; 2
3
(1
-3): 279-2
86.
[19]
Aiazzi B, A
l
p
a
r
one
L, Bar
o
n
t
i S. P
y
r
a
mid-
base
d
mu
ltires
oluti
on
ad
aptiv
e
filters for
a
dditiv
e
a
n
d
multipl
i
cativ
e
i
m
age n
o
is
e.
IEEE Transactions on
Circuits and Syst
em
s
II: Analog and Digital Signal
Processi
ng
. 19
98; 45(8): 1
092
-109
7.
[20]
Arridgey
SR,
Hebdenz
JC.
Optical im
agin
g in medicine
II: modelling
and rec
onstruction.
Phys
ics i
n
Medici
ne a
nd
Biol
ogy.
19
97; 42: 841-
85
3.
[21]
Z
a
charo
p
o
u
los
AD, Svenmar
ker P, Axelsso
n J,
S
c
h
w
e
i
g
e
r
M
,
Arridge
S
R
. And
e
rsson
-
Enge
ls S,
A
matr
ix-fr
ee algor
i
thm
for
m
u
lt
iple
wa
v
e
l
e
n
g
t
h
f
l
u
o
r
e
s
c
e
n
c
e
t
o
m
o
g
r
a
p
h
y
.
Optics Expres
s.
200
9; 17
:
302
5-30
35.
[22]
Paig
e CC, S
a
u
nders MA.
LS
QR – a
n
al
gor
i
t
hm fo
r spars
e
line
a
r-e
quati
o
n
s
and
spars
e
l
east-squ
a
res.
ACM T
r
ansacti
ons on Math
e
m
atic
al Softw
are
. 1982; 8(
1): 43-7
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 9, September 201
3: 499
7 – 5004
5004
[23]
Calam
ante
F
,
Gadia
n
DG, C
onn
ell
y
A. Qua
n
tifica
tio
n
of
bo
lus-trackin
g
M
R
I: Improved c
haracter
i
zatio
n
of the tissue re
sidu
e functio
n
usin
g T
i
khonov
regul
arizati
on.
Magnetic
Res
ona
nce i
n
Med
i
cine
. 20
03
;
50(6): 12
37-
12
47.
[24]
Pan
X, Sidk
y
EY, Vannier M
.
W
h
y
d
o
com
m
ercial CT
scann
ers still
em
plo
y
trad
ition
a
l,
filtered back-
proj
ection for i
m
age rec
onstr
uction
?
In
ve
rse Pro
b
l
em
s
. 20
0
9
; 25(12): 1
230
09.
Evaluation Warning : The document was created with Spire.PDF for Python.