TELKOM
NIKA
, Vol. 11, No. 6, June 20
13, pp. 3159
~
3
164
e-ISSN: 2087
-278X
3159
Re
cei
v
ed
De
cem
ber 2
9
, 2012; Re
vi
sed
April 2, 2013;
Accept
ed Ap
ril 17, 2013
Diagnosis of Hepatocellular Carcinoma Spectroscopy
Based on the Feature Selection Approach of the
Genetic Algorithm
Shao-qing Wang*
1
, Qiang Liu
2
, Dong-
y
u
e Yu
4
, Guang-ju Liang
4
1,2
Shando
ng M
edic
a
l Imagi
ng
Rese
arch Instit
ute, Shan
don
g
Jina
n 25
002
1, Chin
a
3
Shand
on
g Z
i
b
o
Prison H
o
spit
al, Shan
do
ng Z
i
bo 2
5
5
129, Ch
ina
4
Shand
on
g Pro
v
ince
Xi
nT
ai Cit
y
Peo
p
le'
s
Ho
spital, Sha
n
d
o
ng
Xinta
i
27
12
00, Chi
n
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: 2006
w
s
q@
1
63.com*, 20
02
md@16
3
.com, fishlet@
163.co
m,
l
i
a
ng
gu
an
g
j
u
1
@
1
63
.co
m
A
b
st
r
a
ct
This pa
per a
i
ms to study the
app
licati
on
of me
dic
a
l i
m
agi
n
g
techn
o
l
ogy w
i
th artifici
al i
n
te
llig
enc
e
techno
lo
gy on
how
to improv
e the dia
g
n
o
sti
c
accuracy
rate for hepatoc
el
lular carc
in
oma
.
T
he
recognit
i
o
n
meth
od
b
a
sed
on
ge
netic
al
gorith
m
(GA) a
nd
Neur
al
Net
w
ork are
pres
ented. GA w
a
s
use
d
to s
e
l
e
ct
2
0
opti
m
a
l
featur
es from the 4
01
in
itial fe
atu
r
es. BP (Back-prop
agat
i
on
Neur
al Netw
or
k, BP) and PNN
(Proba
bil
i
stic Neur
al Netw
or
k, PNN) w
e
re used to
clas
sify tested samp
les b
a
sed
on these
opti
m
i
z
e
d
features, and
mak
e
c
o
mpar
i
s
on betw
een
r
e
sults
bas
ed
o
n
2
0
opti
m
a
l
f
eatures
a
nd t
h
e a
ll
40
1 fe
atu
r
es.
T
he results of the exp
e
ri
me
nt
show
that the meth
od ca
n i
m
prove the rec
o
gniti
on rate.
Ke
y
w
ords
:
h
epatoc
ell
u
l
a
r carcin
oma, bac
k-prop
agati
on
neur
al n
e
tw
ork, probab
ilistic
neura
l
netw
o
rk,
gen
etic al
gorith
m
, 31P-
hosp
h
o
r
us ma
gn
etic reson
anc
e spec
troscopy
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 clini
c
al wo
rk, 31
P Magnetic
Re
so
n
a
n
c
e
Spectrosco
py (31-P
h
o
s
ph
oru
s
an
d
Magneti
c
Re
son
a
n
c
e S
p
e
c
tro
s
copy, 3
1
P-MRS
)
te
chnolo
g
y [1,
2] ca
n
use
slight chan
ge
s of
chemi
c
al
shif
t in info
rmati
on
colle
ction
todet
e
r
mine
the huma
n
energy metabolism a
nd b
ody
chemi
c
al
s. T
hat i
s
cu
rre
n
t
ly the only n
oninva
s
ive
a
ppro
a
ch
in
studying physi
ologi
cal path
o
logy
cha
nge
s of e
m
ergi
ng tech
nologi
es in vi
vo. So
the evaluation a
nd
31P-M
R sp
ectrum of dise
a
s
e
diagn
osi
s
an
d treatment a
r
e impo
rt
ant clinical
significance [3, 4].
Artificial neu
ral netwo
rk is
an imitation o
f
bi
ological brain in the inf
o
rmatio
n pro
c
e
ssi
ng
method. Thi
s
techniq
ue
ca
n be a very g
ood de
al
with
multivariable
nonline
a
r rel
a
tion. It can be
use
d
for i
denti
f
ication
and
cla
ssifi
cation
throug
h the
trainin
g
of
co
mplex mo
de.
At pre
s
e
n
t, the
neural n
e
two
r
k i
n
3
1
P m
agneti
c
reso
nan
ce
sp
ectroscopy (31P
-MRS)
study has
be
en wi
dely
use
d
. Amon
g them, the reverse tra
n
smi
ssi
on
Neural Net-wo
rk(Back
p
r
op
-agatio
n Ne
u
r
al
Network, BP) model is a m
o
re impo
rtant
artifici
al Ne
ural Net-wo
rk
model. BP is the advantag
e of
netwo
rk opti
m
ization
with
accu
ra
cy. Proba-bilisti
c n
eural
network (proba
bilisti
c neu
ral net
work,
PNN) is
a va
riation fo
rm o
f
the radi
al b
a
si
s fun
c
tion,
also
with th
e ch
aracte
rist
ic of the
sim
p
le
stru
cture, trai
ning qui
ckly and so o
n
.
2. Magnetic
Reso
nanc
e Phosphoru
s
Spectr
u
m
All
ca
se
s we
re sel
e
cte
d
randomly
fro
m
S
han
don
g
medi
cal
ima
g
ing
re
sea
r
ch in
stitute
from Jan. 20
08 to Jan. 20
09. The
r
e a
r
e
130
sam
p
le
data, incl
udin
g
45
ca
se
s fo
r he
patocellul
a
r
carcin
oma, a
nd
2
8
ca
se
s for
liver
ci
rrh
o
si
s, 57
case
s for the
no
rmal. In the
n
o
rmal
group,
with
the co
nventio
nal che
ck,
n
o
histo
r
y of li
ver di
sea
s
e i
s
recogni
zed,
all liver
cirrh
o
-si
s
and
HCC
patients i
n
31
P-MRS after
resea
r
ch all case
s
a
r
e
conf
irmed
by biop
sy pathol
ogy. 31P-M
RS ca
n
measure the seven form
ants (F
i
gure 1): singl
e ph
osp
h
-ate e
s
t
e
r(pho
sp
hom
onoe
ster, PME),
inor-ga
n
ic p
h
o
sp
horus
(in
o
rga
n
ic p
h
o
s
phorus, Pi
),
pho
sph
o
ri
c a
c
id two fat (pho
sph
o
-di
e
ster
,
PDE), p
hosp
hori
c
a
c
id
creatine
(ph
o
spho
cre
a
tine,
PCr),
ade
no
sine trip
ho
-sp
hate
(
α
-ATP,
β
-
ATP,
γ
-ATP)
[5]. 31P MRS curve
descri
bs the mai
n
index: chemi
c
al
shift, wav
e
integral area,
Evaluation Warning : The document was created with Spire.PDF for Python.
TEL
K
3160
sev
e
n
deno
milli
o
tratio
the c
u
3. R
e
divid
e
1
2
,
x
x
to th
hidd
e
the r
e
K
OM
NIKA
V
n
pa
ram
e
te
r
t
e
s t
he co
m
o
n (PPM). W
a
n, a
nd
reso
n
u
rve y-coor
d
Figure 1.
e
v
e
rse Tran
s
BP netw
o
e
d into i
npu
t
2
,
...,
i
x
in-di
c
ree exp
e
ri
m
e
n node
s ch
o
Hypot
h
e
lationship b
()
[
k
N
k
j
i
yf
1
1e
x
p
f
V
ol. 11, No.
6
r
s of th
e
rati
o
m
p
oun
ds t
o
t
a
ve integral
n
an
ce
is
pro
p
d
i
nate, re
pre
s
31P Spectr
u
s
mission
N
e
o
rk
[6, 7] is
t
layer, hi
d
d
c
ate the
inp
u
m
e
n
ts of out
o
os
e
nu
mb
e
r
inp
u
F
h
esis the fir
s
etwe
en in
pu
1
(1
)
(
1
)
1
k
kk
ij
i
wy
1
p
()
x
，
6
, June 20
1
3
o
of th
e va
l
u
he loc
a
tion,
area (li
ne th
p
ortio
nal t
o
t
s
entative
s
o
f
u
m Di
agra
m
e
ural Net
w
o
a layered
s
t
d
en laye
r a
n
u
t of the
net
w
put: hepat
o
-
r
25, tak
e
Si
g
u
t layer
F
igure 2. BP
s
t
k
-
1 layer i
n
t and outp
u
t
()
]
k
j
，
3
: 3159 – 3
1
u
e. Chemi
c
a
with the m
a
e following
a
he
num
be
r
o
f
the compo
u
of the Hep
a
t
o
rk (BP
)
t
ru
cture of f
e
d out-put la
y
w
or
k
,
12
,,
.
yy
-
cellular car
c
g
moid func
ti
o
hidden laye
r
Neu
r
al N
e
t
w
n
the firs
t i
n
is:
1
64
l shif
t
sp
ect
r
a
gneti
c re
s
o
a
rea
s
) rep
r
e
s
o
f the nuc
l
e
u
u
nd
s sig
nal
s
t
ocellular C
a
e
edforward
n
y
er, as
sho
w
..,
j
y
indi
cat
e
c
inom
a, live
o
n as n
euro
n
r
out
pu
t
w
o
r
k S
t
ru
ct
u
r
n
eu
ron
s
(
k
i
y
e
-
r
os
co
p
y
c
u
r
v
o
n
a
nc
e
fr
eq
u
s
ents
the c
o
m
u
s;
P
e
a
k
ref
e
s
t
r
ength u
n
it.
a
rci
noma
Or
g
n
eu
r
a
l netw
o
w
n in Figu
r
e
e
o
u
tput vec
t
r
cirr
ho
sis
a
n
s t
r
an
sf
er f
u
t
layer
r
e
1)
as the i
n
-
ISSN: 2087
v
e i
n
the
ab
s
u
ency in pa
r
m
poun
d's c
o
e
rs to the he
i
g
anization
o
rk
, the net
w
e
2
.
Among
t
or, co
rre
s
p
o
a
n
d
norm
al.
u
n
c
tion.
put,
k
j
y
as
o
(
-278X
s
c
i
ssa
ts per
o
nce
n
-
i
gh
t o
f
w
ork is
them,
o
ndi
ng
Her
e
o
utput,
(1)
(
2)
Evaluation Warning : The document was created with Spire.PDF for Python.
TEL
K
the fi
trans
4. Pr
o
com
p
necti
n
for in
P is
f
layer
funct
layer
vec
t
o
of th
e
cat
e
g
The
d
prob
a
trans
K
OM
NIKA
Diag
no
si
s
The
(1
k
ij
w
rst
k lay
e
r i
n
fer func
tion
o
o
babilistic
N
Proba
bili
s
p
etition l
a
ye
r
n
g input lay
e
put , that is
t
f
or waiting
c
, whi
c
h
belo
n
ion in the ra
; C is a t
r
an
s
Whe
n
th
e
o
r an
d the in
p
(|
|
DP
I
Gene
rat
e
e
radial b
a
si
s
11
(|
|
nb
P
Then, th
e
1
a
R
adb
a
1,
0
i
a
g
ory su
m
1
a
2
2,
(
nn
d
imen
s
i
on
o
f
Finally, t
h
a
bility of ea
c
fer func
tion
e
s
of
Hepatoc
)
in ty
pe (1
)
i
n
firs
t j neu
r
o
f neuron
s.
N
eu
ral Ne
t
w
s
ti
c neural
r
s. T
he n
e
tw
o
e
r an
d rad
i
a
l
t
he n
u
mb
er
o
c
haracte
ri
sti
c
n
gs
to the t
h
dial b
a
si
s;
L
s
fer func
tion
e
network p
r
p
ut vector of
12
||
,
|
|
|
|,
,
I
PI
e
vector
D
m
s
tran
sfer f
u
n
1
12
||
,
|
|
|
Ib
P
I
e
ra
d
i
al basi
s
1
1,
1
1
()
(
,
a
sn
a
a
(1
,
2
,
ii
, get probab
12
,
2
2
,,
,
nn
f
2
n
is
K
,
e
Input
R
Figu
r
h
e vector
p
a
c
h pattern a
p
e
l
e
ment
s co
r
e-I
ellular Carci
n
i
s the lin
k
w
e
r
on
s
;
k
N
is t
h
w
ork
(
P
NN
)
n
e
twork[8]
o
r
k
st
ru
ct
ur
e
l
ba
sis f
u
nc
t
o
f n
euron
s i
n
c
v
e
ct
or
s (
R
h
re
shol
d vec
t
2,
1
L
W
is the
for the com
p
r
ovide
s
an i
n
the sampl
e
s
,
||
||)
T
Q
PI
m
ultiply b
y
b
n
ct
io
n:
1
|,
,
|
|
|
|
)
Q
bP
I
s
func
tion fo
r
1
,2
1
,
,,
)
T
Q
a
)
, In the w
il
ity vector:
,
)
T
K
Ta
e
ach
compo
n
R
adial gra
s
s
r
r
e 3
.
Pro
babi
a
s
s
ed th
r
o
u
g
p
pe
ars, thro
u
r
re
sp
ondin
g
SSN: 2087
-
2
n
om
a Spect
r
e
ight bet
wee
h
e numb
e
r
o
is ma
de u
e
as
s
h
ow
n i
n
ion (RB
F
) l
a
n
hidd
en lay
e
x 1)
;
1
b
is
t
or (Q
×1
));
a
weight mat
r
p
etition.
n
put m
ode v
s
, between t
w
1
b
and m
u
ltip
)
T
r
the out
put:
eig
h
t of the
1
n
ent corresp
o
r
oots
li
stic Neural
g
h a
compe
t
u
gh the com
p
out-put 1,
o
t
2
78
X
r
osco
py bas
e
n the first
k-
1
o
f neuron
s i
n
p of th
e i
n
n
Figu
re 3.
L
a
yer
,
1,
1
IQ
e
r ; R i
ndi
ca
t
the thres
h
ol
1
a
is the out
p
r
ix connectin
e
cto
r
, radial
w
o v
e
ct
o
r
s t
h
li
cation, get
radi
al gra
s
s
o
n
d
ing to
a
m
Competitio
Net
w
o
r
k S
t
r
u
t
itive transf
e
p
etition for t
h
herwi
se
the
e
d on the…
(
1
lay
e
r fo
r th
n
the firs
t
k
n
pu
t la
yer
,
1,
1
L
W
is the
w
R
Q
,
Q
is th
e
t
es the
input
d o
f
the rad
p
ut vector of
g radial
ba
s
grassro
o
ts
c
h
e distan
ce i
s
1
n
, t
h
at is
t
s
ro
ots
ac
cor
d
m
odel ca
teg
o
n layer
u
c
t
ur
e
e
r func
tion
C
h
e probabilit
y
output i
s
0.
H
(
Shao
-qin
g
W
e fi
rst i
neu
r
o
lay
e
r
;
F (x
)
ra
dial ba
s
e
w
eight matri
x
e
numbe
r
of
v
e
ctor dime
d
ial basi
s
fu
n
the radial
tr
a
e a
nd
co
mp
c
om
pute th
e
s
:
(
he
inpu
t ve
c
(
(
d
ing to
the
(
o
ry
.
C
, cal
c
ul
a
t
e
t
y
of the larg
e
H
ere, we ad
o
W
an
g
)
3161
o
ns
to
is
the
e
an
d
x
co
n-
ta
rget
nsi
on;
n
-c
tion
a
ns
fer
etition
e
input
(
3)
c
tor
(
4)
(
5)
m
odel
(
6)
t
he
e
st
o
pt
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3159 – 3
164
3162
Euclide
an n
o
r
m to mea
s
u
r
e weight
s o
f
the distan
ce between t
he inp
u
t vector and th
e
weig
hts vecto
r
. At the same time, using
reflect p
r
ob
ab
ility density of gaussia
n
fun
c
tion a
s
the
transfe
r fun
c
t
i
on of hid
den
layer (Be
2
n
e
, inclu
d
ing n i
s
the input val
ue of the rad
i
al ba
sis
function n
eurons ).
5. Results
This
expe
rim
ent is ba
sed
on three
data
set: I.
the
m
e
dical
feature set colle
cting from
the
liver 3
1
P-M
R
S of the
20
feature
s
out,
20
cha
r
a
c
te
ri
stic se
pa
ratel
y
refer
r
ed to
PME, Pi, PDE,
PCr, ATP(
α
,
β
,
γ
) chemi
c
a
l
shift a
nd th
e
are
a
u
nde
r t
he p
e
a
k
a
nd
PME/PDE, Pi/PDE, PCr/P
D
E,
α
-ATP/PDE,
β
-ATP/PDE,
γ
-ATP/ PDE; II. 31P-MRS data of 401 al
l the spect
r
um characteri
stics;
III.
Us
ing GA
algorithm to selec
t
the bes
t
out of the 20 features
.
5.1. Based o
n
BP Neur
al Net
w
o
r
k Ex
p
e
riment
Table 1. 3-fol
d
Experiment
al Re
sults
Feature
Set
Carcino
ma(%)
Liver
c
i
rrhos
i
s
(%)
Normal(
%
)
Running
time(S)
I
II
III
84
75.1
82.2
73.6
64.3
87.1
91.2
92.9
95.1
56.8
37.5
28.9
Table 2.
5-fol
d
Experiment
al Re
sults
Feature
Set
Carcinoma(
%)
Liver
c
i
rrhos
i
s
(%)
Normal(
%
)
Running
time(S)
I
II
III
80.4
77.8
82.2
73.6
70
80.7
89.8
89.8
90.9
125.48
78.2
62.8
Table 3. 10-f
o
ld Experime
n
tal Re
sults
Feature
Set
Carcinoma(
%)
Liver
c
i
rrhos
i
s
(%)
Normal(
%
)
Running
time(S)
I
II
III
81.5
74.2
84.4
79.3
72.1
87.1
90.2
93.7
95.8
172
133
89
5.2. PNN Ne
ural Ne
t
w
o
r
k
based on E
x
periments
Table 4.
3-fol
d
Experiment
al Re
sults
Feature
Set
Carcinoma(
%
)
Liver
c
i
rrhos
i
s
(%)
Normal(
%
)
Running
time(S)
I
II
III
71.1
77.8
78.2
70.7
71.4
67.8
92.6
91.2
92.9
3.14
3.69
2.83
Table 5.
5-fol
d
Experiment
al Re
sults
Feature
Set
Carcinoma(
%)
Liver
c
i
rrhos
i
s
(%)
Normal(
%
)
Running
time(S)
I
II
III
68.6
78.2
78.7
70
70.7
72.1
92.9
90.5
90.9
5.48
5.73
4.14
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Diag
no
sis of
Hep
a
tocellula
r Ca
rci
nom
a Spec
trosco
py base
d
on the
…
(Shao
-qin
g Wan
g
)
3163
To test an
d verify the validity of the method
of GA-B
P, this pape
r
respe
c
tively use
s
the
3-fold, fold
5-10-fold
cro
s
s validation m
e
thod
s, to
ra
ndomly divid
e
the featu
r
e
setsi
n
to trai
ni
ng
set a
nd te
st
set, an
d to
count the
test
set a
n
ave
r
a
ge of
10 time
s id
entificatio
n a
c
cura
cy.
The
results
sho
w
n in table 1,
2, 3,
and the
experim
ent in
three diffe
re
nt feature
set
the re
sults
were
comp
ared.
Table 6. 10-f
o
ld Experime
n
tal Re
sults
Feature
Set
Carcinoma(
%)
Liver
c
i
rrhos
i
s
(%)
Normal(
%
)
Running
time(S)
I
II
III
80
80
82.2
75
92.8
82.1
89.5
80.7
87.7
25.48
11.61
7.23
6. Discussio
n
Geneti
c
alg
o
rithm (Ge
netic Algori
-
thms,
the
GA) [9, 1
0
, 11] was
propo
sed fi
rstly
inthe
university of Michig
an by
Joh
n
Holl
and
in 1975, it
is a kind of bi
ol
ogical by natural
sele
ction
and
natural
Gen
e
tic me
cha
n
ism of ran
dom
sea
r
ch alg
o
rit
h
m. It is a ki
n
d
of group, o
peratio
n obj
e
c
ts
are
all individ
uals i
n
the
group. Th
rou
g
h
the choi
ce, a
new ge
neration of g
r
ou
ps
is p
r
od
uced b
y
cro
s
s an
d variation
ope
ration. As a
kind
of effici
ent parallel,
its main
characteri
stic i
s
the
sea
r
ching
strategy in gro
up and i
ndivi
dual in
fo
rmat
ion exch
ang
e, automatic acqui
sition
and
accumul
a
tion
of the kno
w
ledge
of the
sea
r
ch
for th
e sp
ace
can
be a
c
hieve
d
in the
sea
r
ch
pro
c
e
ss, an
d the optimal solution can b
e
got
in the adaptive co
ntrol sea
r
ch pro
c
e
ss.
From
Tabl
e 1
,
whi
c
h
can
b
e
seen
after
GA f
eature selectio
n of th
e cla
s
sified
a
c
cura
cy,
all sp
ect
r
um i
s
significantly highe
r tha
n
t
he o
r
igin
al on
e, liver
cirrho
sis re
co
gnitio
n
rate
in
cre
a
sed
from the original 64.3% to
87.1.
Table 3 bas
e
d on data s
e
t III in
the rec
o
gnition ra
te of normal
as hi
gh a
s
9
5
.8%. Comp
ared to
medi
cal 2
0
char
a
c
teri
stics, u
s
i
ng GA 2
0
fe
ature
sele
cti
on,
runni
ng time also g
r
eatly redu
ced. Tabl
e 3, the
use of medical 2
0
feature, ru
n 10 times n
eed
172 second
s,
and use the feat
ure extract
i
on GA 20, ru
n
10 times on
ly 89 second
s.
From Ta
ble
4, 5, 6, which can
be se
en PNN neu
ral net
work b
a
se
d on the
hepato
-
cellul
a
r
ca
rcin
oma di
agn
osi
s
al
so
a
c
hiev
e hig
h
a
c
cura
cy. And m
edi
cal fe
ature
20
, com
pared t
he
choi
ce
of GA
20
cha
r
a
c
teri-stics,
we
ca
n draw
the hi
gher reco
gniti
on rate. In ta
ble 6, b
a
sed
on
GA choi
ce of
20 ch
ara
c
te
ri
stics, the re
cogniti
on rate of can
c
er by
more tha
n
2.2% out.
A com
b
inatio
n of the
expe
rimental
re
sul
t
s, t
he exp
e
ri
ment ba
se
d o
n
BP is sli
ghtl
y
belo
w
the accu
ra
cy of PNN, but from run
n
in
g time to
see
,
the cost of the experim
e
n
t PNN time
be
much l
e
ss. T
o
cont
ra
st Ta
ble 1 and ta
bl
e 4, the
co
st of the experi
m
ent time ba
sed o
n
BP is
15
-
20 times of PNN.
T
h
is
ne
ur
a
l
ne
tw
o
r
k
pa
r
t
icu
l
a
r
ly is
mu
ch
mo
re
suit
a
b
le f
o
r
solv
in
g pat
t
e
rn
cla
ssif
i
cat
i
on
probl
em
s, which
can rea
lize fault det
ection a
nd
d
i
agno
si
s. In the model cl
assificatio
n
, its
advantag
e i
s
that it can
u
s
e li
nea
r le
arning
al
go
rith
m to
compl
e
te befo
r
e
non
linear alg
o
rit
h
m
work, mea
n
-while, it can
keep the chara
c
teri
stics of hi
gh pre
c
i
s
ion
nonlin
ear al
g
o
rithm.
7. Conclusio
n
Thro
ugh the
above anal
ysis, to use
the genetic
algorith
m
in feature sele
ction and
feature
s
del
e
t
ion have
sm
all co
ntributio
n to the
correc
t
c
l
ass
i
fication. Thus
, it influenc
e
the
c
h
ar
ac
te
r
i
s
t
ic c
l
ass
i
fic
a
tion
, an
d
c
an fin
d
o
u
t
the
problem
sp
ace
to
rep
r
e
s
ent th
e
opti
m
al
feature.
The exp
e
rim
ent proves th
at the gen
etic alg
o
rithm
can ove
r
co
me
som
e
of the
pitfalls of
neural n
e
two
r
k in
different
extent, whi
c
h
mean
s
it can
ma
ke
use of
medi
cal i
m
a
g
ing te
ch
nolo
g
y
and a
r
tificial
intelligen
ce
technol
ogy
to improv
e
the com
b
in
ation of sa
mple cl
assifi
cation
accuracy that
is the diagn
o
s
is a
c
curacy rate.
Ackn
o
w
l
e
dg
ments
This
work
wa
s suppo
rted i
n
part by Sh
ando
ng p
r
ovi
n
ce n
a
tural scien
c
e fun
d
proje
c
ts
unde
r G
r
ant
No
s. Y200
6C9
6
，
ZR20
10CM
051, Y
2008
G30, a
n
d
Shan
-do
n
g
medi
cal h
e
a
lth
sci
en
ce an
d tech
nolo
g
y de
velopment pl
an proj
ect No
. 2009 HZ0
8
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3159 – 3
164
3164
Referen
ces
[1]
Baop
en
g Li, Qi
ang
Liu. T
he pr
esent res
earch
si
tuatio
n an
d d
e
vel
opme
n
t trend of Ma
gnetic
reson
a
n
c
e
sp
e
c
tro
scopy
te
ch
no
l
ogy
in the liver tumor.
J
ourn
a
l of Medi
cal Ima
g
i
n
g
. 20
08; 18: 14
61-1
463.
[2]
Yuba
o li
u, Cha
ngh
on
g Li
ang,
Z
hong
lin Z
h
an
g. Pre
limi
nar
y
clinic
al a
p
p
licat
ion res
earc
h
of
the
31
P MR
spectral im
agi
n
g
in the liv
er.
Journ
a
l of Practi
cal Ra
dio
l
ogy
.
200
6; 22: 417-
420.
[3]
Jian
g L
i
n, Ka
n
g
ron
g
Z
h
o
u
, P
hosp
ho-rus-
31
magn
etic res
o
nanc
e sp
ectru
m
ana
l
y
sis
in t
he l
i
ver,
J. of
F
udan (
m
e
d
ica
l
scienc
es
). 20
06; 33: 28
3-28
5.
[4]
W
u
Bi, Bin S
o
n
g
, Yang
Yan
g
.
T
he clinica
l
a
p
p
licati
on
of liv
e
r
31
P-MR spectral ima
g
i
ng.
C
h
ina ge
nera
l
surgery basic and clinical maga
z
i
ne.
200
5; 12: 624-6
27.
[5]
Shah
N, Satta
r A, Bena
nti M
,
Magnetic
res
ona
nc
e s
pectr
oscop
y
as a
n
imagi
ng t
ool f
o
r cancer. A
Revie
w
of the Literature.
Jour
nal A
m
Osteop
ath Assoc
. 200
6; 106: 23-
27.
[6]
Juan
jua
n
Lu,
Hon
g
Che
n
, T
he rese
arch pr
ogress of BP neur
al net
w
o
rk
.
Control en
gi
n
eeri
n
g
. 20
06;
13: 449-
45
1.
[7] Negnevitsky
M.
Artificial Intellig
ence: A Guide to In
telligent System
s,
2
nd Ed., Add
i
s
on W
e
sle
y
,
Engl
and. 2
003.
[8]
Bang
hu
a Yan
g
,
Guozhen
g Yan. T
o
improve brai
n co
mp
u
t
er interface th
e midbr
ain
ele
c
trical sig
n
a
l
recog
n
itio
n rate base
d
on ge
netic al
gorithm
and pro
bab
ilis
tic neura
l
net
w
o
rk.
Journa
l of Shang
h
a
i
Jiaoto
ng u
n
iver
sity
. 2005; 1: 1
689-
169
2.
[9]
Xi
ao
pin
g
W
a
n
g
, Lim
i
ng
C
ao.
Genetic
a
l
gor
i
t
hm t
heor
y a
n
d
soft
w
a
r
e
r
eal
izatio
n, Xi
an: Xi
an
Jia
o
ton
g
univ
e
rsit
y
Pr
es
s, 2002
[10]
Minqi
an
g Li, Ji
song K
ou, Da
n
Lin, Gen
e
tic al
gorit
hm th
e ba
sic theor
y a
nd
app
licati
on, Be
i
jing: sci
enc
e
press. 200
2: 9-
11.
[11]
Z
i
xi
ng Cai,
Gu
ang
yo
u Xu.
Artificial i
n
tel
lig
e
n
ce a
nd
its ap
plicati
on.
B
e
ij
i
ng: Ch
in
a un
iv
ersit
y
pres
s
a
w
a
y
. 20
00. l1
7.
Evaluation Warning : The document was created with Spire.PDF for Python.