TELKOM
NIKA
, Vol. 11, No. 2, Februa
ry 2013, pp. 1047
~10
5
3
ISSN: 2302-4
046
1047
Re
cei
v
ed Se
ptem
ber 28, 2012; Revi
se
d Jan
uary 6, 2012; Accept
ed Ja
nua
ry 1
7
, 2013
Resear
ch About Three Dimensional Reconstruction of
Medical Image
Lina
n Fa
n*
1
, X
i
n Wan
g
2
, G
u
a
ngy
ua
n Zha
n
g
2
, Qiechun
Chen
3
1
School of Infor
m
ation En
gi
ne
erin
g, Shen
ya
n
g
Univ
ersit
y
, S
hen
ya
n
g
110
0
44, Chi
na.
2
School of Infor
m
ation Sci
enc
e & Electronic
Engi
neer
in
g, Shan
do
ng Jia
o
tong U
n
ivers
i
t
y
,
Jinan 2
5
0
357,
Chin
a.
3
Engin
eeri
ng,
T
ohoku Univer
sit
y
, Sen
dai 9
8
0
-85
76, Jap
an.
*Corres
pon
di
n
g
author, e-ma
i
l
: lina
n
fan@
16
3.
com, tx_
x
i
n
xi
n20
06@
16
3.co
m, kangue
@16
3
.com,
chen
qiec
hu
n@
gmail.c
o
m
Ab
stra
ct
In this p
aper, t
h
rou
gh c
o
mpar
ison
of differe
n
t
re
constructio
n
al
gorit
hms
fo
r volu
me re
nde
ring, w
e
put forw
ard
Ra
y Castin
g
alg
o
r
i
thm as th
e sc
h
e
me
of 3D
rec
onstructio
n
of
me
dic
a
l i
m
ag
e. W
e
i
m
pr
oved
th
e
imag
e synthes
is op
erator, an
d co
mbi
n
e
d
s
e
ction s
a
mpl
i
n
g
mod
e
to rec
onstruct the i
m
a
ge. F
i
n
a
lly,
w
e
rend
ered
i
m
ag
es o
n
GPU.
By us
in
g i
m
prove
d
oper
ator, w
e
n
o
t o
n
ly
made
the
ren
deri
n
g
sp
e
e
d
accel
e
rated, b
u
t also
mad
e
the qu
ality of re
nder
ing i
m
ag
e
s
impr
ove
d
.
Key
w
or
ds
: medical im
ag
e, 3D reconstructi
o
n
, ra
y
casti
n
g
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
In the p
a
st f
e
w
de
cad
e
s,
three
-
dim
e
n
s
ion
a
l me
dical imag
e
re
constructio
n
h
a
s
bee
n
unde
r comp
u
t
er
g
r
aphi
cs on
the ba
sis of
the
devel
o
p
ment of
a n
e
w di
sciplin
e. After the 1
9
80s,
with the d
e
velopme
n
t of
medical ima
g
e
tech
nolo
g
y, althoug
h the
com
puted to
mography (CT)
scan a
nd ma
gnetic
re
son
a
n
ce im
agin
g
(MRI)
can
pro
v
ide high
re
solution a
bout
two-di
men
s
io
nal
image [1], howeve
r
, these medical instru
ment(s) can only p
r
ovide hu
ma
n internal t
w
o-
dimen
s
ion
a
l image. CT is
a medical imaging p
r
o
c
ed
ure that utilizes compute
r
-pro
ce
ssed X-rays
to produ
ce sl
ice
s
of spe
c
ific are
a
s of the
body. Doct
ors
can ima
g
e
a quasi three dimen
s
io
n
a
l
orga
n while readin
g
the
s
e
slices. Th
e cross-se
cti
onal
slices
create
d
by the ea
rly CT ma
chi
ne
is
thick a
nd the
image inte
rva
l
is large, a
c
cordin
g to tradi
tional tech
nol
ogy, CT usu
a
lly have arou
nd
10 pictu
r
e
s
a
nd, less than
40 in most ca
se
s,
due to factors such as sl
ow
scannin
g
sp
ee
d.
Do
ctors find i
t
easie
r to
re
ad a
small
n
u
mbe
r
of
ima
ges,
but on
the othe
r h
a
n
d
, limited by
the
con
d
ition
of
softwa
r
e
an
d
ha
rd
ware, realizi
ng
3D
rec
o
ns
truc
tion is
also difficult. But in
rec
ent
years [2], wit
h
the devel
o
p
ment of co
mputer te
ch
n
o
logy, co
ntin
uou
s imp
r
ove
m
ent of scan
ning
techni
que
an
d the e
m
erg
ence of a
spi
r
al
CT
scann
ing
way, faults
can
be ve
ry thin and
la
yer
spa
c
in
g can
be very
smal
l. View of
so
me organi
c f
ault (se
c
tion) can
re
ach from hu
ndred
s to
thousand
s of
images, the
traditional re
ading tablet
way makes readin
g
these
fault by doctors
become difficult, and three
dimen
s
ional
recon
s
tru
c
tio
n
re
sea
r
ch is develope
d. In addition, th
e
medical clini
c
al appli
c
ation
requi
rem
ents
prom
oted the
prog
re
ss of 3
D
re
con
s
truct
i
on.
Radi
ation the
r
apy of medi
cal ima
g
e
s
is the or
ientati
on of 3D
re
cons
tructio
n
a
pplication
requi
rem
ents.
The develo
p
m
ent of medi
cal ima
g
i
ng t
e
ch
nolo
g
y for 3D recon
s
truction
re
sea
r
ch
provide
d
the
necessa
ry se
curity [3], 3D recon
s
tr
u
c
tio
n
tech
niqu
e for the a
ppli
c
ation of medi
cal
image p
r
ovid
ed impo
rtant
techni
cal
su
pport a
nd
b
r
oad a
ppli
c
ati
on prospe
ct. After years of
developm
ent, the three-di
mensi
onal
m
edical ima
ge
recon
s
tru
c
tio
n
from
auxilia
ry diag
no
sis
has
become a
n
i
m
porta
nt me
ans
of adjuv
ant thera
p
y.
Use of 3
D
reco
nstructio
n
techn
o
logy
on
medical imag
e pro
c
e
ssi
ng,
3D model
structure,
and the three dim
ensi
onal mo
d
e
l from different
dire
ction p
r
oj
ection
displ
a
y, the extracted rel
e
vant o
r
gan
s info
rm
ation, ca
n m
a
ke
quantitati
v
e
descri
p
tion of
the
si
ze, sh
ape
a
nd sp
a
c
e po
sition
o
f
the do
ctor i
n
tere
sted
org
ans. An
d three
dimen
s
ion
a
l
reco
nstructio
n
tech
niqu
e m
a
ke
s it
po
ssib
le for do
cto
r
t
o
di
re
ctly, ob
serve
o
r
ga
ns
o
f
the three dim
ensi
onal st
ru
cture [4] qua
ntitatively.
It
also st
ren
g
th
ens the imag
e variou
s details,
so a
s
to help
the docto
rs m
a
ke the
corre
c
t diagn
osi
s
. Therefore, th
e 3D visu
alization pro
c
e
ssing
of medical i
m
age data,
whi
c
h provid
es a hig
h
ly
efficient intuitive auxilia
ry diagno
stic tool
has
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 2, Februa
ry 2013 : 1047 – 1053
1048
profou
nd si
g
n
ifican
ce, thi
s
ma
ke
s me
dical ima
ge
3D visu
alization studi
es b
e
com
e
one o
f
the
mos
t
ac
tive field of research.
3D Re
co
nst
r
uction of Me
dical Imag
e is one of the key techn
o
lo
gies for me
di
cal imag
e
visuali
z
ation
[5]. Since
the
last
centu
r
y,
with
the
rapid
develo
p
ment
of
comp
uter
hard
w
a
r
e
an
d
softwa
r
e te
ch
nology, a vari
ety of new di
gital image
d
e
vice
s have
e
m
erg
ed in
an
endle
s
s strea
m
.
Ho
wever,
wit
h
the
in
cre
a
sing of
the im
age
data
qua
ntity, the trad
itional di
agn
o
s
tic
metho
d
of
observing
th
e two
-
dim
e
n
s
ion
a
l top
ographi
c im
age
has be
en
insufficient to
m
eet
up
with
the
diagn
osi
s
of
dise
ase n
eed
s, an
d thi
s
di
agno
stic
mod
a
lity with ve
ry stro
ng
su
bj
ectivity factors
largely de
pen
ds on the p
h
y
sicia
n's
clini
c
al expe
rien
ce, which hav
e very large
uncertainty. So,
medical image data for 3D recons
truction, which prov
ide an effici
ent auxiliary vi
sual diagnost
ic
tool, has far-reaching si
gnif
i
can
c
e. Volu
me rend
eri
n
g
is an importa
nt part of the
visuali
z
ation [6],
becau
se it
provide
s
hi
g
h
quality im
age
rend
eri
n
g re
sult, re
ceiving scie
ntific re
se
arche
r
s
attention, and
thus h
a
ve de
veloped
a lot
of different vo
lume rend
eri
ng alg
o
rithm
s
. Among tho
s
e
algorith
m
s, Ray Castin
g is widely used
by resear
ch
e
r
s, be
ca
use of its high im
age qu
ality,
but
becau
se of sl
ow re
nde
ring
spe
ed; som
e
improvem
ent
s of Ray Ca
st
ing are p
u
t forwa
r
d.
2. The T
y
pic
a
l Method
s of Volume Re
ndering
2.1. The Process o
f
Volume Rende
rin
g
Firstly, we ge
t the initial
co
ntinuou
s 3
D
data
field
ba
sed on
recon
s
tructing
the
d
i
screte
3D d
a
ta field
.
Then, we g
e
t the functio
nal value
of the sam
p
ling
point by re
sampling i
n
th
e
contin
uou
s 3
D
data field.
For the n
e
w
sampl
e
poi
nts, cla
s
sificati
on and
col
o
r assignm
ent are
according to
the set o
p
a
c
ity and col
o
r
model. Fin
a
lly, acco
rdin
g to the volume
rend
erin
g of
the
optical mo
del
, we get the entire data fiel
d
proje
c
tion i
m
age throug
h the image synthesi
s
.
The process
of volume ren
derin
g is sho
w
n in Figu
re
1.
Figure 1. The
Proce
s
s of Volume Rend
e
r
ing
Volume re
nd
ering meth
od
can map o
u
t the data field in subtle
and difficult to use
geomet
ric re
pre
s
entatio
n
detail
s
, fully reflectin
g
the data
fiel
d of the
wh
ole info
rmati
on.
Therefore, th
e metho
d
b
r
i
ngs to attenti
on, a va
riety
of sp
ecifi
c
m
e
thod
s. Base
d on
re
sam
p
l
i
ng
and
synthe
si
s meth
od
s, the meth
od of
volume
ren
d
e
ring
can b
e
divided into t
he ima
ge
sp
ace
scan body drawin
g metho
d
[7], object spa
c
e sca
n
volume ren
d
e
ring meth
od
and freque
n
cy
domain vol
u
me re
nde
ring
. Frequ
en
cy domain vol
u
me re
nde
ring
method
due
to its own p
r
i
n
cipl
e
limits whi
c
h inclu
de sin
g
le
rende
ring m
ode and difficulty on achie
v
ing compl
e
x display effect is
still in the sta
ge of laborato
r
y research.
2.2. The Co
mparison of
Se
v
e
ral Volume Rende
rin
g
Reg
u
lar
data
field volume
rend
erin
g re
sea
r
ch mainl
y
has 4
kind
s of algo
rith
ms: Ray
Ca
sting, Spla
tting, Shear-warp
and 3
D
texture-ma
ppi
ng hardware.
Ray Ca
sting [8]: It only relates to the amount
of proje
c
tion light an
d it can also
prom
ote
synthe
sis.
Th
erefo
r
e, it i
s
particula
rly suitable fo
r th
e volum
e
ren
derin
g of
3
D
data field
whi
c
h
feature la
rge
volume data
field and tight
distributio
n.
Ray Ca
sting'
s prin
cipl
e is
simple a
nd it is
easy to reali
z
e. It can ea
si
ly achieve a
perspe
c
ti
ve p
r
oje
c
tion
with
high ima
ge
quality, so it i
s
approp
riate for the medi
ca
l 3D re
con
s
truction al
gorit
hm.
Splatting [9]:
Thre
e-di
men
s
ion
a
l data p
o
ints a
r
e proj
ected o
n
to the scre
en to achi
eve
data re
sampli
ng and imag
e synthesi
s
. Therefore, th
e dra
w
ing tim
e
, which i
s
highly sen
s
itive to
the size of the data field, depe
nd
s pri
m
arily
on the
quantity of opaqu
e vowel
s
and the
size of
gene
ric footp
r
int table, a
n
d
, as
a re
sult
, not suit
a
b
le
for la
rge
-
sca
l
e data field
reco
nstructio
n
of
medical imag
e.
Shear-warp [
10]: A software
re
nde
rin
g
meth
od,
which
is abl
e
to o
b
tain i
n
teractive
rend
eri
ng sp
eed in real time by software, po
ss
e
s
ses the fa
stest rende
ring
speed by far.
This
method,
however, exhibit
s
many i
nhe
re
nt defe
c
ts
: its two li
nea
r
re
sampli
ng
ma
y re
sult in
lo
ss
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch Ab
out Thre
e Di
m
ensional
Re
con
s
tru
c
tion
of Medical Im
age (Li
nan F
an)
1049
and in
di
stin
ction of
detai
ls; si
nce the
r
e is
no i
n
terl
amina
r
resa
mpling, the
q
uality of ima
ges
decrea
s
e
s
si
gnifica
ntly du
ring m
agnifi
cation; the
di
stribution
of the inputted
da
ta in the th
re
e
axial dire
ctio
ns mu
st be
equally spa
c
ed, othe
rwi
s
e, interpolat
ion sh
ould
be executed
in
advan
ce; be
cause the offset enco
d
ing
of three
face
ts' dire
ction
need
s to be
pre
-
sto
r
ed, a
nd
con
s
e
que
ntly, result in larg
e amount of
memory o
c
cu
pation.
3D texture
-
m
appin
g
hard
w
are [11]: is bas
ed o
n
h
a
rd
wa
re to improve the rende
ring
spe
ed, it nee
ds
spe
c
ial
graphi
cs
hardware.
Re
sampl
i
ng inte
rpolati
on op
eratio
n
in texture
spa
c
e
and
ope
ratio
n
of the
ima
g
e
synth
e
si
s
with opa
city va
lues, i
s
co
mp
leted by
ha
rd
ware, thu
s
ta
ke
up a lot of memory
spa
c
e. And re
n
derin
g qualit
y
is also po
or. Evaluatio
n of four kin
d
s of
algorith
m
is shown in Tabl
e 1.
Table 1. Algo
rithms Pe
rformance Evalu
a
tion
Volum
e
rende
ring algorithm
Algorithm
fe
ature
s
Im
ag
e
qualit
y
Rending
speed
Ra
y
C
a
sting
Splatting
Shear-
w
arp
3D text
ure-
mappi
ng Hard
ware
Each voxel has different color
and opacit
y
Footprint spline sampling
Data access is continuous
Calculation required Ha
rd
w
a
re
Highest
High
Medium
Lo
w
e
r
Slow
Fas
t
Fas
t
es
t
Fas
t
In
su
mma
ry,
this pap
er use
s
Ray Ca
stin
g
alg
o
rithm for three-dimens
ional reconstruction
of the medica
l image.
3. Ra
y
Casting
3.1. The Principle of Ray
Cas
t
ing
Ray Castin
g
[11] is a
ki
nd
of imag
es
order
vol
u
me
rende
ring
met
hod. It set
s
o
u
t from
each pixel poi
nt F (x, y) of the im
age
spa
c
e, acco
rdin
g
to the direct
i
on of line sig
h
t cast ray I, the
ray g
o
th
rou
gh the
3
D
d
a
ta field
at a
ce
rtain
ste
p
m, alo
ng th
e ray sele
ction K
equi
distant
sampli
ng poi
nts, by the distan
ce
of the color value
s
and the opa
city values of the eight data
points in a rece
nt sampli
ng point for the three
lin
e
a
r interpolati
on [10], then find out opa
city
value and
color value of
this sampl
e
point. Synthesi
z
ed
colo
r and op
acity values of ea
ch
sampli
ng
poi
nt from
front t
o
ba
ck
can
g
e
t the
colo
r v
a
lue
of pixel
s
point f
r
om
th
e ray, then
g
e
ts
the final imag
e on the scre
en. The pri
n
ci
pl
e of Ray Ca
sting is
sho
w
n in Figure 2.
Figure 2. The
Principl
e of Ray Ca
sting
3.2. The pro
cess o
f
Ray
Cas
t
ing
Ray castin
g
algorith
m
which i
s
ba
se
d
on imag
e sp
ace i
s
often
use
d
in di
re
ct volume
rend
eri
ng alg
o
rithm, the b
a
si
c ide
a
is t
hat ea
ch pixe
l point on th
e
scree
n
send
s a sa
mpling
rays
whi
c
h goe
s throu
gh a dat
a field, resa
mpled a
c
cording to the sampling
step
[12]. Color and
opa
city of ea
ch
re
sampli
n
g
point
comp
ose
d
is
ca
l
c
ul
ated by vario
u
s inte
rpol
ation alg
o
rithm,
and
then image
synthesi
s
from
front to back or from
re
ar
to front, until the light passes thro
ugh th
e
volume d
a
ta
field or is ab
sorbed
compl
e
tely
[13]. Th
e process
of
algor
ith
m
in
cl
ude
s: gradie
n
t
cal
c
ulatio
n, material
cla
s
sificatio
n
, intensity
cal
c
ul
ation, re
sam
p
ling an
d im
age synth
e
si
s. The
detailed p
r
o
c
ess of algorit
hm
is shown in Figure 3.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 2, Februa
ry 2013 : 1047 – 1053
1050
Figure 3. The
Proce
s
s of Ray Castin
g
The alg
o
rithm
assume
s tha
t
the 3D sp
atial data
are di
stribute
d
in a
point of uniform grid
or
reg
u
lar g
r
id, data
pre
p
ro
ce
ssi
ng
flow
diag
ram
inclu
d
e
s
fo
rmat conversi
on of
ra
w
d
a
ta
,
can
c
eli
ng of redun
dant dat
a,
and image
pre
-
processin
g
.
Then, d
a
ta
classificatio
n
,
who
s
e
obje
c
t
i
ve is
to
cla
s
sify co
rrectly
all data
poi
nts a
n
d
different valu
es
of colo
r a
n
d
op
acity fo
r
each d
a
ta a
c
cording
to
differen
c
e
of
rel
a
ted info
rmati
on
of the data v
a
lue, in o
r
d
e
r to sho
w
co
rrectly the
diffe
rent di
strib
u
tion of a va
riet
y of sub
s
tan
c
es
or multiple attribute
s
of a single materi
al
[14
]. Then resam
p
ling, n
a
mely a ray is laun
che
d
from
each pixel of
the scre
en a
c
cordi
ng to t
he set di
re
ction of ob
serv
ation,
that ra
y passes th
ro
ugh
the 3D d
a
ta field, cho
o
si
n
g
K equidi
sta
n
t resa
mplin
g
point alon
g the ray, and t
he interpolati
o
n
cal
c
ulatio
n accordi
ng to the col
o
r valu
e and opa
city
of eight data point whi
c
h is the nea
re
st to
sampli
ng poi
nt, calcul
atin
g colo
r value
and op
acity o
f
the samplin
g point. Obviously, before
the
resampli
ng, colo
r and
op
acity
value
s
of
3D
d
a
ta f
i
eld
sho
u
ld
be tran
sform
ed from m
o
del
coo
r
din
a
te sp
ace into
spa
c
e of the corre
s
po
ndin
g
ima
ge co
ordi
nate
s
.
Lastly, Algori
t
hm whi
c
h i
s
image
synth
e
si
s,
nam
ely colo
r a
nd
o
pacity value
s
of ea
ch
sampli
ng poi
nt on every ray is com
pou
nded from fro
n
t
to back or
from re
ar to front, colo
r value
s
and op
acity o
f
screen pixel
for whi
c
h ray can b
e
obtai
ned [15]. Re
sampling
and i
m
age
synthe
sis
are
co
ndu
cte
d
a
c
cordi
ng
to every
pixe
l of ea
ch
scan lin
e o
n
th
e screen,
so
this
algo
rith
m
belon
gs to vo
lume re
nde
rin
g
algorith
m
which i
s
scan
n
ed by image
spa
c
e.
4. The Basic
Idea of Improv
ing Ra
y
Casting
Algorithm
Although
ray
-
ca
sting i
s
an
effective m
e
thod
fo
r th
re
e-dim
e
n
s
iona
l re
co
nstructi
on, we
discovered
some of its d
r
awba
ck
s aft
e
r the te
st. Its rend
erin
g
spe
ed is
sl
ow. The
r
efo
r
e
,
con
s
id
erin
g t
h
is
dra
w
b
a
ck, the re
nde
ri
ng
spe
ed
sh
ould
be im
proved to
aco
mmplish volu
me
rend
eri
ng ne
eds li
ke b
a
si
c resampli
ng, image
cla
ssifi
cation
and
re
nderi
ng, an
d i
m
age
synthe
sis,
we co
ncentra
ted our re
se
a
r
ch o
n
re
sam
p
ling and
ima
ge synthe
sis,
and improve
d
the tradition
al
ray-ca
sting al
gorithm a
s
be
low.
4.1. The Improv
ement of the Sampling
The p
o
int
sa
mpling i
s
ch
ange
d to th
e
se
gmente
d
sampli
ng
bet
wee
n
two a
d
jacent
se
ction
s
(t
wo
se
ction
s
co
mpose a thi
n
plate)
wh
en
light pa
ssing
throu
gh. Th
us in
the mix
i
ng
pro
c
e
ss,
th
e mixing
of sa
mpling point RGBA
valu
e is
turned
to b
e
the mixin
g
of co
rrespon
ding
segm
ents re
spectively. Th
e cal
c
ul
ation
of the pre-
int
egratio
n tabl
e is
often time-con
sumi
ng,
in
orde
r to
me
et the n
eed
s
of re
al-time
re
n
derin
g,
the
pre-integ
r
atio
n
algorith
m
i
s
a
c
celerated.
O
n
the pre
m
ise
of the linea
r i
n
terpol
ation, firstl
y, we
cal
c
ulated the
RGBA values
on the o
ne ra
nge
interval
0-1,
1
-
2…
9-10. T
h
en afte
r a
c
cu
mulati
ng
the
RGBA valu
es we
could
obt
ain the
value
on
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch Ab
out Thre
e Di
m
ensional
Re
con
s
tru
c
tion
of Medical Im
age (Li
nan F
an)
1051
interval 0
-
1,
1-2…
9
-
10.
Next, wh
en
calcul
ating
the
value
on
1-1
1
ra
nge,
we
j
u
st n
eed
to a
d
d
the value of
1-10
and 1
0
-11 togethe
r; the cal
c
ul
atio
n method o
n
other inte
rva
l
s is a
nalo
g
o
u
s.
Pre point
s accele
rated al
g
o
rithm pri
n
ci
p
l
e is sh
own in
Figure 4.
Figure 4. The
Principl
e of Pre Points Accele
rated Alg
o
rithm
4.2. The Improv
ement of the Image Composition
Opera
t
or
For
com
p
leti
ng the volu
me re
nde
rin
g
nee
ded
st
eps li
ke
ba
sic resamplin
g, image
cla
ssifi
cation
and
ren
deri
n
g, image
synt
hesi
s
,
we
did
sp
eci
a
l rese
arch o
n
the
i
m
age
synthe
sis.
We fo
und
th
at the im
age
co
mpo
s
ition
ope
rato
r
co
uld b
e
simplif
ied
whe
n
u
s
i
ng a
fro
n
t en
d to
rea
r
en
d ima
ge
synthe
sis
operator. It n
eed
s five ti
m
e
s
of multipli
cation, additio
n
and
subtraction.
Every sam
p
li
ng p
o
int ne
e
d
s to
be
syn
t
hesi
z
ed.
T
h
e traditio
nal
image
synth
e
si
s o
perator is
sho
w
n in for
m
ula (1
), (2).
)
1
(
i
in
i
now
i
now
i
in
i
in
i
out
i
out
C
C
C
(1)
)
1
(
i
in
i
now
i
in
i
out
(2)
The re
nde
rin
g
spe
ed will
be furthe
r im
proving
if the
comp
ositio
n operator i
s
si
mplified.
Assu
ming th
e
colo
r valu
e o
f
the i individ
ual elem
ent i
s
i
now
C
, the opacity
value is
i
now
.The c
o
lor
value whi
c
h
e
n
ter the i indi
vidual elem
e
n
t is
i
in
C
, the opacity value is
i
in
. The colo
r value which
pass through
i individual element is
i
out
C
, the opa
city value is
i
out
. Rea
c
hin
g
to the point n,
1
n
out
a
.So the impro
v
ed operator i
s
sh
own in (3
).
n
i
i
in
i
now
i
now
n
out
a
C
C
C
1
)
1
(
(3)
This im
prov
ed imag
e synthesi
s
op
e
r
ator
only n
eed
s thre
e
multiplicatio
n
s
an
d a
subtractio
n o
peratio
n, it can effe
ctively improv
e
s
th
e imag
e syn
t
hesi
s
efficie
n
cy. Using
thi
s
improve
d
Ra
y Castin
g al
gorithm.
Usi
ng imp
r
ov
ed
ope
rator,
we not o
n
ly accele
rated
the
rend
eri
ng spe
ed, but also i
m
prove
d
the quality of ren
derin
g image
s.
4.3. The Improv
ement on the GPU
For tra
d
itiona
l dra
w
ing on
the CPU, the ren
deri
ng
spe
ed is n
o
t fast enoug
h
and the
image qu
ality is also not very high. As kno
w
n,
GPU is a spe
c
iali
zed p
r
o
c
e
ssi
ng hardware
of
image p
r
o
c
e
s
sor a
nd it can bri
ng extremely high
i
m
age q
uality. By using th
e improve
d
li
ght
proje
c
tion al
g
o
rithm on GP
U, the softwa
r
e and
hardware a
c
cele
rati
on ca
n be a
c
hieved at the
same time.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 2, Februa
ry 2013 : 1047 – 1053
1052
5. Anal
y
s
is o
f
Test Result
In this p
ape
r,
we
used th
e
improve
d
Ra
y Ca
sting al
g
o
rithm fo
r 3
D
re
con
s
tructio
n
of the
medical ima
g
e
. Thre
e g
r
o
ups
of expe
ri
mental d
a
ta
were
cho
s
e
n
to prove
tha
t
this imp
r
ov
e
d
algorith
m
is succe
ssful.
The th
ree
sets of
data
are
re
spe
c
tively
on the
re
sol
u
tion of 5
1
2
×
5
12, 48
skull i
m
age
s,
361 ki
dney i
m
age
s an
d 4
63 upp
er im
a
ges. F
r
om th
e re
sult, usin
g improve
d
o
perato
r
, we
n
o
t
only accele
ra
ted the re
nde
ring
spe
ed, b
u
t also
im
pro
v
ed the quali
t
y of renderi
n
g image
s. Th
e
recon
s
tru
c
tio
n
of the image well preserved body in d
e
tail.
From the
he
ad, the ki
dne
y and the u
p
per p
a
rt
of t
he imag
e we
can
prove t
hat the
improve
d
alg
o
rithm
i
s
su
cce
ssful. We can see
th
at rende
ring
spe
ed is a
c
celerated ap
pa
ren
t
ly,
and the imp
r
oved algo
rith
m based on
GPU rende
ri
ng re
sult is
hi
gher th
an in t
he CP
U grap
hics,
and the spee
d is faste
r
.
The time usi
n
g a different a
l
gorithm i
s
sh
own in Ta
ble
2.
Table 2.
The
Comp
ari
s
o
n
of Different Rende
ring Tim
e
Bod
y
pa
rts
Data field
scale
The tra
d
itional Ra
y
Casting algorithm for
rending time(ms)
The improved R
a
y
Casting algorithm based
on CPU re
nding t
i
me(ms)
The improved
ra
y
casting algorithm based
on GPU
rending
time(ms)
Head
Kidne
y
Upper
bod
y
512
512
48
512
512
36
1
512
512
46
3
420
930
1160
270
500
830
170
320
560
From tabl
e 2, we can see t
hat the ren
d
e
r
i
ng time u
s
in
g improve
d
Ray Castin
g al
gorithm
based on GP
U is faste
r
th
an tradition
al Ray Ca
sting
algorith
m
.
6. Conclusio
n
In this
pap
er,
we
intro
d
u
c
e
d
volume
ren
derin
g al
gorit
hm in
detail.
For
medi
cal i
m
age
s,
high qu
ality rende
ring effe
ct is impo
rta
n
t for health
care worke
r
s.
Therefo
r
e, i
n
cont
ra
st to the
cla
ssi
c vol
u
m
e
rend
erin
g a
l
gorithm
s,
we
ch
ose th
e
ra
y-ca
sting
alg
o
rithm
as the
scheme
of th
ree
dimen
s
ion
a
l
recon
s
tru
c
tio
n
of medi
ca
l image, a
n
d
improved
the traditio
n
a
l Ray
Ca
st
ing
algorith
m
, fin
a
lly achievin
g 3
D
me
dica
l image
reco
nstru
c
tion
of
three
g
r
ou
p
s
of
data’
s.
Th
e
rend
eri
ng
spe
ed which u
s
e
d
improved
Ray Ca
sting al
gorithm i
s
fa
ster than th
at usin
g tra
d
itional
Ray Castin
g
algo
rithm, combine
d
with
the GP
U, a
c
hievin
g
the image ren
dering
with
high
er
qualit
y
,
f
a
st
er
re
sult
s.
Wit
h
the develo
p
m
ent of com
puter
hardw
a
r
e system
s,
volume ren
dering
method in 3
D
medical im
a
ge re
con
s
truction will be m
o
re matu
re, a
nd more wid
e
l
y use.
Referen
ces
[1]
Xi
aok
ai Jin, B
a
sed o
n
vol
u
m
e
rend
erin
g thr
ee-
d
i
mens
ion
a
l
image rec
ons
truction al
gorit
hm to stud
y
Engl
ish. Xi'
an: Univers
i
t
y
of Electronic Sci
enc
e and T
e
chno
l
o
g
y
of Chi
na m
a
ster thesis. 20
07.
[2]
Ruij
ua
n W
ang,
Z
hang Ji, Pe
n
g
Ke. Algor
ith
m
for computer
-assisted me
di
cal ima
ge 3
D
reconstructi
on.
Journ
a
l of Cli
ni
cal Re
hab
ilitati
v
e Tissue Eng
i
neer
ing R
e
se
a
r
ch.
2011; 15(
4).
[3]
Z
heshe
ng T
ang.
T
he visualiz
ation of three-
dime
ns
io
nal d
a
t
a field. Beiji
ng
:
T
s
inghua U
n
i
v
ersit
y
Press
.
199
9; 9-10
9.
[4]
Haib
o L
i
u, Jin
g
She
n
, Song
Guo. Digita
l
Im
age Proc
essi
ng Usi
ng Vis
u
al C+
+
.
Beijin
g: Machin
e
r
y
Industr
y
Pr
ess. 2010; 6.
[5]
Xu
eso
ng Yi
n, Qian Z
h
a
ng, Guoh
ua W
u
, Z
h
i
gen
g Pan.
R
e
v
i
e
w
o
n
F
our V
o
lume R
e
n
deri
n
g Alg
o
rithms.
Co
mp
uter Engi
neer
ing a
nd Ap
plicati
ons.
2
0
0
4
; 40(16).
[6]
Hu Yi
ng. R
e
s
earch
on
ke
y techn
o
lo
g
y
o
f
medi
ca
l im
a
ge vis
u
a
lizati
o
n. Dissertati
o
n
.
Shen
ya
ng
:
Doctora
l
Disse
rtation of North
eastern U
n
iver
sit
y
. 20
04; 7.
[7]
Avidan S. E
n
s
e
mble T
r
acking
.
IEEE
Conf. on Computer Vision
an
d Pat
t
ern Recognition. Sandiego:
200
5; 2: 494-5
01.
[8]
Yang
X
,
Pei J
H
. Rapid and Robust Medic
a
l Image Elast
i
c Registration using Mean
Shift.
Chines
e
Optics Letters
. 200
8; 6(12): 95
0.
[9] Goshtasby
A.
2-D
an
d 3-
D Imag
e R
e
g
i
stration
for
Medic
a
l, R
e
mote Se
nsin
g,
and
Ind
u
stria
l
Appl
icatio
ns
. W
ile
y
Pr
ess. 2005.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch Ab
out Thre
e Di
m
ensional
Re
con
s
tru
c
tion
of Medical Im
age (Li
nan F
an)
1053
[10]
HU Z
h
a
n
-li. E
x
tractio
n
of A
n
y
Ang
l
e V
i
rtu
a
l Sl
ice o
n
3
D
CT
Image.
Internatio
na
l S
y
mp
osi
u
m o
n
Intelli
gent Infor
m
ati
on T
e
ch
no
logy Ap
plic
atio
n
. 2008; 1: 35
6
-
360.
[11]
J Sun, Y W
a
n
g
, X W
u
et. a
l
. A Ne
w
Im
ag
e Se
gm
entati
o
n Alg
o
rithm
an
d It`s Appl
icati
on i
n
l
e
ttuc
e
obj
ect segme
n
t
ation.
T
E
LKOMNIKA Indon
e
s
ian Jo
urn
a
l of
Electrical E
n
g
i
ne
erin
g
. 20
12;
10(3): 49
7-
504.
[12]
J Z
han
g, CH
Yan, CK
Ch
ui,
SH On
g. F
a
st s
egm
entatio
n
of b
one
in
C
T
images us
in
g 3D
a
daptiv
e
thresho
l
di
ng.
C
o
mputers i
n
bi
olo
g
y an
d medi
cine
. 20
10; 40(
2): 231-2
36.
[13]
Carlo
Col
o
mb
o, Alberto
De
l Bimbo, F
e
d
e
rico
P
e
rnic
i. Metric 3D
Reco
nstruction
and T
e
xtur
e
Acquis
i
tion
of
Surfaces of R
e
voluti
on from
a
Sing
le U
n
ca
lib
rated Vi
e
w
.
IE
EE T
r
ansactio
n
s on P
a
ttern
Analys
is and M
a
chi
ne Intel
lig
e
n
ce
. 200
5; 27(
1): 99-11
4.
[14]
Richar
d J Gar
d
ner, Markus
Ki
derl
en. A
Ne
w
Algorit
hm for
3
D
Rec
onstructi
on from S
u
p
p
o
r
t F
unctions.
IEEE Transactions on Patter
n
Analys
is and M
a
chi
ne Intel
lig
e
n
ce
. 200
9; 31(
3): 556-5
62.
[15]
Shan-s
h
a
n
F
a
n, Xua
n
Ya
ng.
3D C
o
rresp
on
din
g
Co
ntrol P
o
ints Estimati
o
n
usi
ng M
ean
Shift Iteration.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(5): 1
040-
105
0.
Evaluation Warning : The document was created with Spire.PDF for Python.