TELKOM
NIKA
, Vol.12, No
.3, Septembe
r 2014, pp. 6
31~638
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v12i3.97
631
Re
cei
v
ed Ap
ril 4, 2014; Re
vised July 1
1
, 2014; Accept
ed Jul
y
27, 2
014
Matrix Mask O
verlapping and Convolution Eight
Directions for Blood Vessel Segmentation on Fundus
Retinal Image
Arif Muntasa*,
Indah Agu
s
tien Sirajud
i
n, Mochammad Kau
t
sar
Sophan
Informatics En
gin
eeri
ng D
epa
rtment, Engine
ring F
a
cu
lt
y
,
U
n
iversit
y
of T
r
unoj
o
y
o
R
y
T
e
lan
g
Po. Box
2 Kamal,
Bangk
ala
n
*Corres
p
onding author, e-mail:
arifmuntasa@if.trunojoy
o.ac.id
A
b
st
r
a
ct
Dia
betic R
e
tin
opathy is o
ne
of t
he diseas
e
s
that have the effect
of a high
mortal
ity rate after
heart dis
eas
e and ca
ncer. How
e
ver, the dise
ase can
b
e
early d
e
tected throu
gh b
l
o
od vesse
ls an
d th
e
optic n
e
rve
he
ad i
n
F
u
n
dus
i
m
a
ges. Bl
oo
d
vessels s
e
p
a
ra
tion of th
e o
p
ti
c nerve
he
ad r
equ
ired
hi
gh
effort
w
hen it is c
o
nducte
d
ma
nu
ally, t
her
efore
it is nec
ess
a
ry that
the appr
opri
a
te method
t
o
p
e
rfor
m
seg
m
e
n
tatio
n
of the object. Leve
l
Set met
hod is w
e
ll-kn
ow
n as object
segmentati
o
n
metho
d
bas
e
d
on
obj
ect defor
mabl
e. How
e
ver
,
the
meth
ods
hav
e the
dis
adva
n
tag
e
; it
requ
ires i
n
iti
a
li
z
a
ti
on
befor
e
th
e
seg
m
e
n
tatio
n
process. In
thi
s
rese
arch, s
e
gmentati
o
n
me
thod w
i
th
out i
n
itiali
z
a
t
i
o
n
proc
ess is
pro
pos
e
d
.
T
he seg
m
entat
ion is c
ond
ucte
d by usi
ng th
e max
i
mu
m va
lu
e selecti
on r
e
s
u
lts of conv
olut
ion 8
directi
o
n
s
.
Experi
m
ental r
e
sults sh
ow
that, propos
ed
method
has
obta
i
ne
d 89.4
8
%
a
ccuracy. Seg
m
entatio
n errors
ar
e
cause
d
by sma
ll bra
n
ches, w
here they are
no
t connec
te
d, so that the object
s
are supp
ose
d
as nois
e
s
.
Ke
y
w
ords
: fun
dus i
m
a
ge, seg
m
e
n
tatio
n
, 8 di
rections co
nvol
ution, over
la
ppi
ng
1. Introduc
tion
One
of dia
b
e
t
ic retin
opath
y
indication
s is
eyes reti
nal d
e
fect.
Hard
defe
c
t wi
ll ca
use
blindn
ess [1]-[4], therefore, it is n
e
cessa
r
y to
be
cond
ucted
ea
rly d
e
tection
on
F
undu
s. In thi
s
ca
se, segm
e
n
tation of blo
od vessel i
n
Fundu
s
pla
ys importa
nt rul
e
in dete
c
ting
eye bloo
d ve
ssel
damag
e [5].
The se
gment
ation re
sults
su
ch as le
ngt
h, wide, sha
p
e
of blood vessel bra
n
ch will
assist to det
ermin
e
kind
of the diabet
ic reti
no
path
y
disea
s
e
s
[6]. The probl
em occu
rre
d
on
medical wo
rld
related to blo
od vessel on
Fundu
s is
difficult to seg
m
ent manually
blood ve
ssel, it
is requi
re
hig
h
co
st a
nd a
c
curacy.
Ho
we
ver, bloo
d
ve
ssel segm
ent
ation is not
p
r
ofession
al if it is
con
d
u
c
ted m
anually [7].
Difficulty seg
m
entation on
the Fundu
s i
m
age bl
o
od vessel
s ca
use
d
by optic nerve head
damag
e ha
s
gray level th
a
t
is almo
st
si
milar to
th
e o
b
ject
s a
r
ou
nd
and
overla
pp
ing bet
wee
n
the
blood ve
ssel
s an
d the opt
ic ne
rve hea
d. Therefore
i
t
is necessa
ry to build ap
prop
riate m
e
thod
for blood ve
ssel
s se
gment
ation in the Fundu
s imag
e.
In rece
nt years, many re
se
arche
r
s h
a
ve co
n
d
u
c
ted re
sea
r
ch on the
segme
n
tatio
n
of the
blood
vessel
s of the
Fu
ndu
s [7]-[1
0], Opt
i
c
Ne
rve
Hea
d
[11]-[1
4
]. T
he
re
sea
r
ch
condu
cted
is a
n
attempt to im
prove th
e
se
gmentation
result F
und
us image
bloo
d
vessel
s. Ho
wever, th
e re
sults
of segmentation error
rate
are still far from the expect
ed.
The meth
od i
s
widely u
s
e
d
to dete
c
t blo
od vessel
s i
s
Level Set [14]
, but the met
hod h
a
s
a limitation
when th
e seg
m
entation
pr
oce
s
s
which
requi
re
s a
p
p
r
opriate
initiali
zation
re
gion,
if it
doe
s not m
a
tch a
c
tual
outcom
e
s
re
sult in hig
h
error
rates.
In this study
, the prop
osed
segm
entation
method
with
out usi
ng a
Fundu
s bl
oo
d vessel regi
on initialization, the prop
ose
d
method
ba
se
d on
the
ma
ximum value
of the
conv
olution
matrix
8
dire
ction
s
and
overl
app
ing
mask matrix. The propo
se
d method works q
u
ickly.
Segmentatio
n of blood v
e
ssel
s and t
he dete
r
mina
tion of the optic disc i
s
a
seri
es of
Fundu
s imag
e biom
edi
cal
resea
r
ch p
r
o
c
ess u
s
e
d
to
d
e
tect level
s
of diab
etic
dise
ase
retinopat
hy
[15]. Object separation Fu
ndu
s image b
l
ood vessel
s
are very co
m
p
licate
d
pro
c
ess. The pro
b
l
em
that ari
s
e
s
when th
e
seg
m
entation
proce
s
s i
s
ve
ry little differe
n
c
e
between
t
he bl
ood
vessel
s
and othe
r o
b
ject
s that a
r
e aroun
d, it will cau
s
e difficulty
in
performi
ng se
gmentation. The
numbe
r of
pi
xels on
the b
l
ood ve
ssels
and the
opti
c
disc ove
r
lap
p
ing al
so
ca
use
difficulty
in
perfo
rming
segmentatio
n. Ho
wever, ma
ny method
s have bee
n d
e
velope
d by the re
sea
r
che
r
to
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 63
1 – 638
632
segm
ent the
blood
vessel
s and
the
opti
c
di
sc o
n
Fu
ndu
s ima
ge.
The p
r
e
s
e
n
ce of
cotton
wools
spot
s, microa
nuery
s
m
s
, edema an
d exudate
s
are al
so an imp
edi
ment to sepa
rate the bloo
d
vessels a
nd the optic n
e
rv
e disc.
2. Rese
arch
Metho
d
Fundu
s im
ag
e ha
s bla
c
k
backg
rou
nd,
whe
r
ea
s bl
oo
d vessel a
n
d
optic di
sk h
a
ve the
little differen
c
e to th
e b
a
ckgrou
nd. T
h
is
probl
em
ca
uses
difficulty to
se
pa
rate
bet
wee
n
the
blo
od
vessel and t
he ba
ckgro
u
nd. In this rese
arch
, the
propo
s
e
d
method Fu
n
d
us bl
ood v
e
ssel
segm
entation
by determini
ng the maximum value
o
f
the convolu
t
ion matrix followe
d by an 8-
dire
ction
s
ov
erlap
p
ing
ma
sk
matrix. Suppo
se
an
RGB fun
d
u
s
image a
s
i
n
the form of
the
followin
g
equ
ation
w
h
f
h
f
h
f
w
f
f
f
w
f
f
f
F
rgb
rgb
rgb
rgb
rg
b
rg
b
rg
b
rgb
rg
b
RGB
,
2
,
1
,
,
2
2
,
2
1
,
2
,
1
2
,
1
1
,
1
(1)
For th
e n
e
xt pro
c
e
ss,
gre
en valu
e of t
he Equ
a
tion
(1) is processed
a
s
see
n
on the
follo
wing
equatio
n
w
h
f
h
f
h
f
w
f
f
f
w
f
f
f
F
g
g
g
g
g
g
g
g
g
G
,
2
,
1
,
,
2
2
,
2
1
,
2
,
1
2
,
1
1
,
1
(2)
In order to
create m
a
sk i
m
age, it i
s
n
e
ce
ss
ary
to
determi
ne th
e threshold
value. It i
s
con
d
u
c
ted to
se
parate b
e
twee
n Fu
ndu
s obje
c
t an
d b
a
ckgroun
d. T
he follo
wing
are
equ
ation
to
cre
a
te the ma
sk ima
ge u
s
in
g the thre
shol
d value
else
thresho
l
d
F
if
Mas
k
G
255
0
(3)
Mask imag
e
can b
e
create
d
from Equati
on (2
) by
usi
ng Equatio
n (3) a
s
se
en in
Figure 1. In this
ca
se, the thre
shol
d value u
s
ed i
s
50.
Figure 1. Fun
dus Imag
e DRIVE Databa
se an
d Mask
Image
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Matrix
M
a
sk Overlapping and
Co
nvoluti
on Eight Directions
for Blood .... (Arif M
untasa)
633
The next ste
p
is to calcu
l
ate the resu
lt of
the con
v
olution matrix 8 direction
s
. This
pro
c
e
ss i
s
sta
r
ted by defini
ng the integ
e
r numbe
r
x
and
x
{-t, -5, -4, . .
. .4, 5 }. T
he value of
x
is
comp
uted by usin
g equ
atio
n
2
2
*
.
x
x
Y
(4)
The results o
f
Equation (4
) are shifted the va
lue b
a
sed on the
ma
ximum value
as
see
n
on t
h
e
followin
g
equ
ation
Y
Y
V
)
max(
(5)
The results o
f
Equation (5
)
a
r
e
d
upli
c
at
ed
n
li
ne
s an
d cal
c
ul
ated t
he ave
r
ag
e value
s
by u
s
in
g
equatio
n
n
j
m
k
k
j
V
MN
U
11
,
1
(6)
Zero m
ean
s value can be
calcul
ated ba
sed on
Equati
on (5
) and (6) as the followi
ng equ
ation
U
V
Z
(7)
The re
sult
s of Equation (7
) are no
rmali
z
e
d
by using e
q
uation
h
n
w
m
m
n
w
h
Z
Z
R
11
,
,
(8)
In this
c
a
s
e
, the 1
st
row of
R ha
s the sa
me value wit
h
the 2
nd
row until the end row. The valu
e of
R ca
n be see
n
in equatio
n
t
t
t
t
t
t
R
R
R
R
R
R
R
R
R
R
,
4
2
,
4
1
,
4
,
2
2
,
2
1
,
2
,
1
2
,
1
1
,
1
(9)
The
re
sults o
f
Equation
(9
) i
s
u
s
e
d
to
created
ma
sk matrix
by
u
s
i
ng
the
follo
wi
ng rule as se
en
on the followi
ng equ
ation
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 63
1 – 638
634
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
,
4
2
,
4
1
,
4
,
2
2
,
2
1
,
2
,
1
2
,
1
1
,
1
t
t
t
t
t
t
R
R
R
R
R
R
R
R
R
R
(10
)
Mask matrix i
n
Equation
(10) i
s
rotate
d
by using
an
gles
20
0
until
160
0
(M
20
0
, M
40
0
, M
60
0
, M
80
0
,
M
100
0
, M
120
0
, M
140
0
, and M
16
0
0
)
j
G
j
M
F
Q
, j
20
0
, 40
0
,
.
. .
. .
. 160
0
(11
)
The
result of
the convoluti
on of
Equation (11) i
s
checked
every
pi
xel, the largest value
will
be
taken a
nd pla
c
ed o
n
the po
sition of the correspon
ding
pixel
j
Q
R
max
(12
)
The results o
f
sele
cting th
e maximum v
a
lue of t
he
convolution i
s
taken th
e mo
st minimal va
lue
and no
rmali
z
ed by usin
g e
quation
R
S
min
(13
)
))
max(max(
255
*
S
R
S
R
H
(14
)
Value of the matrix H is the scre
enin
g
result
s of
co
nvolution mat
r
ix 8 directio
ns. Furth
e
rm
ore,
the re
sults of
Equation (14
)
is used to de
termi
ne the t
h
re
shol
d valu
e of the imag
e by comp
ari
ng
sum of the largest value by
using e
quati
on
i
i
T
T
T
2
1
max
(15
)
256
1
,
256
1
,
1
,
1
,
*
2
log
*
*
2
1
2
and
*
2
log
*
*
2
1
1
i
n
m
n
i
n
m
n
i
i
n
m
n
i
n
m
n
i
H
H
T
H
H
T
(16
)
The results o
f
Equation (1
6) a
r
e u
s
ed t
o
ch
ange
Fu
ndu
s obje
c
t, but ba
ckgro
u
nd imag
e is
not
cha
nge
d as
seen on the fol
l
owin
g equati
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Matrix
M
a
sk Overlapping and
Co
nvoluti
on Eight Directions
for Blood .... (Arif M
untasa)
635
else
Mask
and
T
H
if
B
1
255
0
(17)
Binary image
results in E
quation (17
)
are al
so
con
t
ain som
e
n
o
ise
s
, therefore it is
necessa
ry to remove the b
r
eadth of its
noise is
le
ss
than the thre
shol
d va
lue. Obje
cts that are
smalle
r than t
he thre
shol
d value is chan
ged to bla
ck
as the follo
wi
ng equ
ation
else
ld
ObjThresho
O
if
E
i
1
0
(18
)
The re
sult
s o
f
equation (1
9) is then ov
erlap
ped
wi
th
mask image
of Equation (3) by u
s
ing t
he
followin
g
equ
ation
Mask
E
Seg
~
~
~
(19
)
Accu
racy
Measur
e
ments of Segmen
tation
Res
u
lts
T
o
ob
ta
in
a
ccu
r
a
c
y
th
e se
gme
n
t
a
t
io
n r
e
su
lt
s o
n
F
und
us i
m
ag
e
DRIVE databa
se
[16], it
is n
e
cessa
r
y
to b
e
cal
c
ul
ated the
diff
eren
ce
bet
ween experi
m
ental re
sults and ground
t
r
uth
image data
b
a
s
e by usi
ng
Miss Cla
ssif
i
c
a
t
i
on
E
rro
r
eq
uation
BT
BG
FT
FG
BT
BG
Seg
Seg
Seg
Seg
Seg
Seg
ME
1
(20
)
In this
c
a
se
Seg
BG
and
Se
g
BT
rep
r
e
s
e
n
t backg
ro
un
d and fo
re
ground
gro
und
truth imag
e
s
,
whe
r
ea
s ba
ckgroun
d and
foregro
und
groun
d truth image
s of segme
n
tation re
sults a
r
e
rep
r
e
s
ente
d
by using
Seg
FG
and
Seg
FT
.
3. Results a
nd Analy
s
is
To prove pe
rf
orma
nce of the p
r
opo
se
d
me
thod, the
DRIVE Fu
nd
us im
age
dat
aba
se i
s
use
d
fo
r exp
e
rime
nts [1
6]. It co
nsi
s
ts of
20
imag
es.
It also
sup
port
ed the
segm
entation
re
sul
t
s,
the se
gmenta
t
ion re
sults
are perfo
rme
d
by people
wh
o are
expert
s
in relate
d fiel
ds. In orde
r to
obtain segm
entation a
c
cura
cy,
experi
m
ental re
sult
s are compa
r
ed
with gro
und truth F
u
ndu
s
image DRIVE databa
se. Th
e followin
g
are example
s
o
f
Fundu
s ima
ge DRIVE databa
se.
Figure 2. Two
Fundu
s Imag
es on
DRIVE Datab
a
se [16
]
The
gro
und
truth Fu
ndu
s i
m
age
DRIVE datab
as
es can b
e
see
n
F
i
gure
3
and
F
i
gure
4.
Fundu
s imag
e DRIVE d
a
taba
se
s h
a
ve
two
groun
d
truth
mod
e
l
s
,
which
are
the first
mo
del
(Figu
r
e 3
)
an
d the se
con
d
model (Fi
g
u
r
e 4).
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 63
1 – 638
636
Figure 3. The
First Mod
e
l Grou
nd T
r
uth
DRIVE Data
base [16]
Figure 4. The
Second M
o
d
e
l Grou
nd Truth DRIVE Databa
se [16]
The expe
rim
ental re
sult
s have be
en
comp
ar
ed
wi
th the first g
r
oun
d truth
model of
Fundu
s DRIV
E images d
a
taba
se. The e
x
perime
n
tal result
s of pro
posed metho
d
sho
w
that the
minimum a
c
curacy is 62
.28%, the averag
e of
seg
m
entation a
c
curacy is 89
.48%, and the
maximum se
gmentation a
c
cura
cy is
99.
744 a
s
sh
own in Figure 5.
Figure 5. The
Segmentatio
n
Accu
ra
cy of Propo
sed M
e
thod on Fu
n
dus
DRIVE image
s
databa
se
The bigg
est
error o
c
curs
on the 7
th
image. Error
on image i
s
influenced
by some
para
m
eters,
whi
c
h a
r
e
th
e thre
sh
old v
a
lue
s
of o
b
je
ct is delete
d, wid
e
obj
ect,
rou
n
d
of obj
e
ct,
and ratio bet
wee
n
the are
a
of circumfe
rence of obj
e
c
t. The pre
s
en
ce of bloo
d vessel
s of bra
n
ch
obje
c
t is
re
ga
rded
a
s
n
o
ise
,
but the el
on
gated
sha
pe
of the obj
ect
apart f
r
om it
s bra
n
che
s
so
it
woul
d be del
eted whe
n
the deletion is
con
d
u
c
ted.
The followin
g
are the se
gm
entation re
sul
t
s of
prop
osed met
hod a
s
se
en
on Figu
re 6.
0%
20%
40%
60%
80%
100%
1
2
3
4
5
6
7
8
9
1
01
1
1
21
3
1
41
5
1
61
7
1
81
9
2
0
Segmentation
Accuracy
(%)
Ima
g
e
Used
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
9
30
Matrix
M
a
sk Overlapping and
Co
nvoluti
on Eight Directions
for Blood .... (Arif M
untasa)
637
Figure 6. The
segme
n
tatio
n
results, the
1
st
and the 2
nd
Groun
d Trut
h (from the le
ft to
the right)
The
differen
c
e bet
wee
n
b
a
ckgroun
d a
nd bl
ood
vessel
s h
a
s re
su
lted in
bre
a
ki
ng of t
h
e
end
s of the retinal bloo
d vessel
s. Separation bet
wee
n
the end
s of the retinal bl
ood vessel
s
and
the main b
r
a
n
ch
ha
s form
ed small obj
e
c
ts. Th
e obj
e
c
t is
reg
a
rd
e
d
as
noi
se a
n
d
rem
o
ved al
ong
with the a
c
tu
al noise a
s
shown in Figu
re 7. Both
for
parts A a
nd
B of Figure 7
are the pi
eces of
the retinal i
m
age h
a
ve bee
n enla
r
ge
d to
400%. Part
A
of Figure 7
shows that
the
r
e is
a sepa
ra
te
end
s of th
e b
l
ood ve
ssel
s
of the m
a
in
b
r
an
ch. Pa
rt B
of Fig
u
re
7
descri
be
s of
the e
n
d
s
of th
e
severed bl
oo
d vessels i
s
regarded
as n
o
ise a
nd rem
o
ved at the time of the act
ual noi
se rem
o
val
pro
c
e
ss. Th
e error resulted
in sen
s
itivity
and a
c
cura
cy
values
decre
ase
d
.
A
B
Figure 7. Explanation of Image Segm
ent
a
tion Fault
A. Ends of the retinal blo
o
d
vessels that
is
s
e
parate from the main
branc
h
.
B. Pieces tip of retinal bloo
d vessels h
a
s
been re
move
d becau
se it is co
nsi
d
ered
as noi
s
e
Our propo
se
d meth
od
ha
s b
een
comp
ared
to
othe
r metho
d
a
s
see
n
In
Tabl
e 1. T
h
e
comp
ari
a
son
result sho
w
s that our p
r
opo
sed me
th
od is supe
rio
r
to the othe
r two meth
o
d
s.
Ho
wever, it is impo
rtant to improve o
u
r
pro
p
o
s
ed
method to in
cre
a
se the a
c
cura
cy. Co
mmon
errors th
at o
c
cur in
ea
ch
obje
c
t is cau
s
ed
by two t
h
ing
s
. Th
e first e
rro
r, d
e
licate b
r
an
che
s
of
blood
vessel
s
can
not b
e
dete
c
ted, b
e
c
au
se
the
de
licate bra
n
ch
es of
bloo
d
v
e
ssel
s
h
a
s b
een
cut off from
the m
a
in
b
r
an
ch
and
t
he b
r
an
ch
i
s
co
nsi
dered
as
noi
se. T
h
e second
e
r
ror
,
exce
ssive dil
a
tion pro
c
e
ss that causes
blood
ve
ssel bran
ch
es i
s
forme
d
too thick.
Table 1. Co
m
pari
s
on of the
segme
n
tatio
n
results Accura
cy
Method
Accuracy
(%
)
Chaudhu
ri et al. [17]
87.77
Jiang et al. [18]
89.11
Our P
r
oposed M
ethod
89.48
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 63
1 – 638
638
4. Conclusio
n
The propo
se
d method ha
s bee
n able to s
egm
ent the Fundu
s im
age DRIVE databa
se,
the averag
e
error rate 10.
52%. However, the erro
r t
hat occu
rs in
some F
undu
s imag
e ca
n be
improve
d
by con
s
id
erin
g several p
a
ra
m
e
ters, e
s
p
e
ci
ally the determinati
on of t
he thre
sh
old
of
wide
sp
rea
d
o
b
ject
s that are con
s
id
ere
d
noise. Th
e d
e
termin
ation
of the object
threshold val
u
e
need
s to
be
ca
rri
ed o
u
t
automatically with the
obj
ect la
beling
Fundu
s i
m
ag
e segme
n
tation
results. T
he
ratio of the
ci
rcu
m
fere
nce
and
are
a
of
the o
b
je
ct is
an imp
o
rta
n
t
para
m
eter th
at
sho
u
ld be a
d
ded to avoid
deletion of ob
jects that
a
r
e
not noise but are con
s
ide
r
e
d
as noi
se.
Referen
ces
[1]
Akara So
ph
ara
k
,
Bun
y
arit U
y
ya
no
nvara, S
a
rah
Barm
an.
“Automatic Micr
oan
eur
ysm D
e
tection from
Non-
dil
a
ted
Di
abetic
Reti
no
p
a
th
y R
e
tina
l I
m
ages
Usi
ng Mathematic
al Morph
o
lo
g
y
M
e
thods.
IAENG
Internatio
na
l Journ
a
l of Co
mputer Scie
nce
.
201
1; 38(3): 29
5-30
1.
[2]
Bock R, Meier J, N
y
úl LG, Ho
rneg
ger J, Mi
chels
on G. Glaucoma Risk In
de
x: Automate
d Glaucom
a
Detectio
n From Color Fun
d
u
s
.
Medical Ima
ge Ana
l
ysis
. 20
10; 14(3): 4
71-
481.
[3]
Akara Sop
har
ak, Khine T
het N
w
e, Yi
n A
y
e Mo
e, Matthe
w
N. Da
il
e
y
, Bun
y
a
r
it U
yya
n
onvar
a.
Auto
matic Ex
u
date
Detectio
n
w
i
th a N
a
ive
Bayes C
l
ass
i
fi
er.
T
he 200
8 I
n
ternati
o
n
a
l C
onfere
n
ce
on
Embed
de
d S
y
s
t
ems and Intell
i
gent T
e
chnol
o
g
y
, Ba
ngkok, T
hail
a
n
d
. 200
8.
[4]
Rivero
n
EF
., Guimeras NG. Extractio
n
of blo
od ve
sse
ls in o
phtha
lmic col
o
r
images of hu
man retin
a
s.
Lecture N
o
tes i
n
Co
mp
uter Scienc
e.
200
6; 4225: 11
8–
12
6.
[5]
Ra
w
i
MA., Qutaish
a
t M., Arrar M. An impro
v
ed ma
tch
ed fi
lter for bl
ood v
e
ssel
detectio
n
of di
git
a
l
retina
l imag
es.
Co
mp
uters in
Biol
ogy an
d Medici
ne.
2
007;3
7
(2): 262
–2
67.
[6]
Chan
w
i
maluang T
.
, Fan G.
An efficient b
l
o
o
d
vessel
detect
i
on a
l
g
o
rith
m for retina
l i
m
a
g
e
s usin
g loc
a
l
entropy T
h
res
hol
din
g
. Proceedings of IEEE International
S
y
mpos
ium
on Circuits and S
y
stems,
Bangk
ok, T
hailand. 20
03; 5: 2
1–2
4.
[7]
Yong Ya
ng,
Shu
y
i
ng H
uan
g, Nini R
ao.
An Au
tomatic
H
y
brid M
e
tho
d
F
o
r Retinal
Bloo
d Vesse
l
Ex
traction.
Inte
rnatio
nal J
ourn
a
l of App
licati
o
n Matheta
m
ati
cs and Co
mpu
t
er Science.
2
008; 18(
3):
399
–4
07.
[8]
Staal, MD. Abr`amoff,
M. Niemeijer, MA. Viergever,
B.
van Ginneken. Ridge based vesse
l
segmentation in color
ima
g
e
s
of th
e reti
na
.
IEEE Transactions
on Medical Imaging.
200
4; 2
3
(4)
:
501
–5
09.
[9]
Niemeijer, Staal, B. van Gin
neken, M. Loog, MD. Abr`amoff. Com
parative study
of retinal vessel
segme
n
tatio
n
methods on a
ne
w
pu
blic
l
y
avai
lab
l
e dat
abas
e.
in
SPI
E Medic
a
l I
m
agi
ng
, J. M.
F
i
tzpatrick and
M. Sonka, Ed
s. 2004; 5
370: 6
48–
65
6.
[10]
HF. Jelinek, RM. Cesar-Jr.
Segmentation
of re
tin
a
l fu
nd
us vasc
ulatur
e
in
no
n-m
y
dri
a
tic camer
a
imag
es usi
ng
w
a
vel
e
ts.
Angi
ogra
phy a
nd P
l
aq
ue I
m
ag
in
g:
Advanc
ed S
e
gmentati
on T
e
chni
ques
, J.
Suri an
d T
.
Laxminara
y
a
n
, Ed
s. CRC Press. 200
3: 193
–2
24
.
[11]
F
engsh
ou Yi
n,
Jian
g Li
u, Si
m Heng O
ng,
Yi
ng S
un, Da
mon W
K
. W
ong, Nga
n
Men
g
T
an, Carol
Che
ung, Ma
ni
Baskara
n, T
i
n Aung, T
i
en Yin
W
ong.
Mod
e
l-
base
d
Optic N
e
rve He
ad S
e
gmentati
o
n
on R
e
tina
l F
u
ndus I
m
a
ges.
33rd A
n
nual Internat
ional Confer
ence of
t
he IEEE EMBS, Boston,
Massach
usetts USA, August 30 - Septemb
e
r
3. 2011.
[12]
Handay
ani T
j
andras
a, Ari Wija
y
a
nti, Nanik Suciati. Optic
Nerv
e Head Segmenta
tion Using Hough
T
r
ansform and Active Conto
u
r
s
.
TELKOMNIKA.
2012; 10(
3
)
: 531~
536.
[13]
Lal
ond
e M, B
eau
lie
u M, G
agn
on
L. F
a
s
t
and
ro
bust
optic
d
i
sc d
e
tection
us
ing
p
y
ram
i
da
l
decom
positi
o
n
and
Ha
usdorf
f
-based tem
p
l
a
te matchin
g
.
IEEE Transaction
Medical Imaging
, 20
01;
20(1
1
): 119
3–1
200.
[14]
Yu
H, Barri
ga
ES, Agurto
C, Echeg
ara
y
S,
Patti
chis MS,
Bauma
n
W,
Soliz P. Fast
lo
calizati
o
n
an
d
segme
n
tatio
n
of optic disk
in
retinal
imag
es
using directi
o
nal
matc
hed
fi
l
t
ering and leve
l
sets,
IEEE
T
r
ans Inf T
e
chnol Bi
o
m
ed
. 2
0
12; 16(4): 6
44-
657.
[15]
Ibaa Jama
l, M. Usman Akra
m, Anam
T
a
riq. Retina
l Image Prepr
ocessi
ng: Backgro
un
d and N
o
is
e
Segmentation.
TEL
K
OMNIKA
, 2012; 1
0
(3): 5
37~
54
4.
[16]
DRIVE. Dig
ital
Retin
a
l Ima
g
e
s
for Vesse
l E
x
trac
tio
n
, Imag
e Scie
nces Ins
t
itute. Avail
abl
e on:
URL
:
http://
w
w
w
.
isi.
u
u
.nl/Res
earch/
D
atab
ases/DRI
VE, accesed 2
011.
[17]
S. Chau
dh
uri, S. Chatterje
e
,
N. Katz, M. Nelson, M. Gold
b
aum. De
tecti
o
n
of blo
od vess
els in r
e
tin
a
l
imag
es usin
g tw
o
-
dim
ens
ion
a
l
matched filter
s.
IEEE
Transactions on Medical Im
aging.
20
02;8: 26
3.
[18]
Jiang. J. Staal, MD. Abramo
ff, M. Niemeijer,
MA. Viergever, B.
van Ginneken. Ridg
e based vess
el
segme
n
tatio
n
i
n
color im
ages
of the retina.
IEEE Transactions on Medical I
m
aging,
20
04;
23: 501.
Evaluation Warning : The document was created with Spire.PDF for Python.