TELKOM
NIKA
, Vol.14, No
.4, Dece
mbe
r
2016, pp. 14
93~150
1
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i4.3714
1493
Re
cei
v
ed Ap
ril 6, 2016; Re
vised Augu
st
27, 2016; Accepted Septe
m
ber 13, 201
6
Exudate and Blood Vessel Feature Extraction in
Diabetic Retinopathy Patients using Morphology
Operation
Sis
w
o
Ward
o
y
o*, Anggo
ro Sur
y
o
Pramud
y
o
, Erik
a Diana
Riza
nti, Imamul
Muttakin
Dep
a
rtment of Electrical E
ngi
neer
ing,
Un
iver
sitas Sultan A
g
eng T
i
rta
y
as
a,
Jl. Jender
al Su
dirman Km 3 C
ileg
on 4
2
4
35, telp
/fa
x: +
62-25
4-37
671
2 (e
xt 20)/ +
62-25
4-3
954
40
*Corres
p
o
ndi
n
g
autjor, e-ma
il
: sis
w
o
@
u
n
tirta.ac.id
A
b
st
r
a
ct
Diab
e
tic
Retin
opathy
is
on
e
of the
retin
a
co
mpl
i
cati
ons
caus
ed
by
dia
betic
dise
a
s
e w
i
th
observ
abl
e sy
mpto
ms
such
as e
m
er
ge
nce
of exu
date
a
n
d
new
bl
oo
d ves
s
els. T
he to
ol
used t
o
scre
e
n
it is
a fun
dus
ca
me
ra. How
e
ver,
a
naly
z
i
n
g
the
fu
ndus
i
m
a
g
e
sh
oul
d b
e
do
ne
by d
o
ctor w
h
o
is a
n
expert
a
nd
w
ill req
u
ire
a l
o
t of time. The
r
efore, aut
o
m
a
t
ic featur
e
dete
c
tion ca
n ass
i
s
t
doctor i
n
pr
oc
essin
g
the r
e
ti
nal
imag
e i
n
a
nal
y
z
i
ng
di
abetic
retino
pathy
dise
ase.
T
h
e
prop
ose
d
method
has
b
e
en teste
d
o
n
the
mor
p
h
o
lo
gica
l oper
ations of the
fund
us i
m
age fro
m
Cice
ndo Ey
e H
o
sp
ital, Ban
d
u
ng.
T
he calc
ulati
o
n
results on feat
ure extractio
n
exud
ate
are
a
h
a
s a rang
e of 0
pixels for nor
ma
l retin
a
l i
m
a
ge, 17-2
1
2
13 p
i
xe
l
for retin
a
l i
m
a
ge NP
DR,
an
d 1
25-1
2
2
99 r
e
tina
l i
m
age
p
i
xel for
PDR.
T
he calc
ul
atio
n resu
lts o
n
t
h
e
extraction
are
a
of blo
od v
e
ss
els h
a
s a r
ang
e of 13
31
9-4
6
681
pixe
l to th
e nor
mal reti
na
, the retin
a
l i
m
age
743
5-49
93
8 pi
xel for NPDR, and 1
3
.81-
53.8
02 retin
a
l i
m
a
g
e
pixe
l for PDR
.
Ke
y
w
ords
: Di
abetic R
e
tin
o
p
a
thy, Exudate,
Bloo
d Ve
sse
l, Morfolo
g
y Ope
r
ation, Area C
e
ntroid
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
One
of the
ch
roni
c di
se
ase
s
a
m
on
g the
gro
w
ing
nu
m
ber of tod
a
y's so
ciety i
s
a
diabeti
c
with a
hig
h
n
u
mbe
r
of
peo
ple in
the
wo
rld.
Wo
rl
d
He
alth O
r
ga
niza
tion (WHO) reporte
d th
at t
h
e
numbe
r of diabetic
cases i
n
the world
was
abo
ut 171
million peopl
e
in 2000, and
Indonesi
a
was
fourth by the
numbe
r of pe
ople with
dia
betes i
s
8.5
million inha
bitants in 2
000.
This nu
mbe
r
is
expecte
d to reach 21.3 mil
lion by 2030 [
1
].
Diab
etic retin
opathy is on
e of the com
p
licat
io
ns dia
betes that is
the
leading cause
of
blindn
ess in
adults.
Re
sea
r
ch i
n
Ame
r
ica, Australi
a, Europ
e
an
d Asia repo
rted
that the num
be
rs
of diabeti
c
retinopathy p
a
tients
will in
cre
a
se f
r
om
1
0
0
.
8 million in
2
010 to 1
54.9
million in 2
0
3
0
with 30% of them are thre
atened
with blindn
es
s [2]. The Dia
b
Ca
re Asia 2
008
Study involve
d
1,785
diab
etes
patient
s in
18
prim
ary a
nd
se
con
dary
health
cente
r
s in Ind
one
si
a an
d repo
rte
d
that 42% of peopl
e with
diabete
s
ha
d
retinop
athy
compli
catio
n
s, and 6.4%
wa
s proliferat
ive
diabeti
c
retin
opathy [3].
The main
pro
b
lem is the
d
e
lay in treat
ment
of diab
etic retin
opat
hy diagno
si
s
becau
se
most
of the
patients in th
e ea
rly sta
g
e
s
are
not i
m
paired vi
sio
n
[4]. Screen
ing p
r
og
ram
for
diabeti
c
retin
opathy nee
d
comp
uter a
ssi
stan
ce
to
analyze fund
us ima
g
e
s
o
b
tained from
the
fundu
s cam
e
ra. Dete
ction
system req
u
ire
s
a co
m
putational m
odel to tran
sform pixel re
tinal
image i
n
to a
retinal fe
ature of di
abeti
c
retinop
at
hy in
dicate
d by
exudate
and
bl
ood ve
ssel
s t
hat
appe
ar usin
g morp
holo
g
ica
l
ope
ration
s. Thus, with
e
a
r
ly dete
c
tion
of diabeti
c
retinopathy, it can
prom
pt the healing a
c
tion
quickly.
2. Diabetic
Retinopa
th
y
Detec
t
ion
Diab
etic
retin
opathy (DR) i
s
a
disea
s
e t
hat
initiates with
mi
crova
s
cula
r compli
cations
i
n
the retina,
where th
e ph
o
t
orecepto
r
ce
lls, the
ne
uronal el
ement
s respon
sible
for vision,
a
r
e
locate
d. The i
n
itial dise
ase
is cha
r
a
c
teri
zed
by in
crea
sed va
scular
perm
eability and p
r
og
re
ssi
ve
vascular occl
usio
n
and ne
ovascula
rization. Without
medical treat
ment, the reti
nal cell
s/tissu
e
s
become m
a
l
nouri
s
h
ed a
n
d
deg
ene
rati
ve, which lea
d
s to d
a
mag
e
in the
cell
s respon
sibl
e
for
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1493 – 150
1
1494
vision [5].
Di
abetic retino
pathy is g
e
n
e
rally
cla
s
si
fied into
n
onp
roliferative
dia
betic
retin
o
p
a
thy
(NP
D
R), whi
c
h com
p
ri
se
s t
he ea
rly stag
es of the
dise
ase, a
nd p
r
oli
f
erative diab
e
t
ic retino
path
y
(PDR), which
is the most
seri
ou
s and
vision-th
re
ate
n
ing sta
ge [6
]. It can be t
ediou
s an
d time
con
s
umi
ng to deci
phe
r subtle mo
rp
hologi
cal ch
ange
s in o
p
tic disk, microan
eury
s
ms,
hemo
rrh
age,
blood vessel
s, macula, and
exudates
through ma
nual
insp
ectio
n
of fundu
s image
s.
A com
puter a
i
ded di
agn
osi
s
system
ca
n
sig
n
ificant
ly
redu
ce the
bu
rden
on
the o
phthalmol
ogi
sts
and may alleviate the inter and intra
obse
r
ver va
riability. The review in [7] discu
ssed
the
available met
hod
s of vario
u
s retin
a
l f
eature extra
c
tion
s and a
u
toma
ted analysi
s
.
Image recog
n
ition for the
scree
n
ing
of diabeti
c
reti
nopathy
wa
s explore
d
in
[8]. The
pre
s
en
ce
of exudate
s
wit
h
in the ma
cul
a
r regio
n
is a
main hallm
a
r
k of di
abeti
c
macul
a
r
ede
ma
and allo
ws its detectio
n
wit
h
a high
se
nsitivity.
Theref
ore, dete
c
tion
of exudate
s
is an im
porta
nt
diagn
osti
c ta
sk. Exudate
s
are fou
nd u
s
i
ng their
hi
gh
grey level variation, and th
eir contou
rs
are
determi
ned b
y
means of m
o
rph
o
logi
cal
reco
nstructio
n
techniq
u
e
s
. A new alg
o
rit
h
m for dete
c
ti
on
of exudates
wa
s pre
s
e
n
te
d and di
scussed in [9].
In another st
udy [10], a te
chni
que ba
se
d on morp
hol
ogical image
pro
c
e
ssi
ng a
nd fuzzy
logic to dete
c
t hard exu
d
a
tes from
DR retinal im
a
ges
wa
s pro
posed. At th
e initial stag
e, the
exudate
s
were identified
u
s
ing m
a
them
atical mo
rp
ho
logy that incl
ude
s elimin
ation of the o
p
tic
disc. The fu
zzy output for all the pixels in ev
ery ex
udate was
calcul
ated for
a given inp
u
t set
corre
s
p
ondin
g
to red,
gre
en an
d blu
e
cha
nnel
s of
a
pixel in an
e
x
udate. This
fuzzy o
u
tput
wa
s
comp
uted for hard
exudat
es a
c
cording
to the
prop
ortion of the
area
of the hard
exudat
es.
Similarly, report [11] discusse
d a hybrid
fuzzy
image-p
r
o
c
essing sy
ste
m
for situation
asse
ssm
ent of
diabetic re
tinopathy
to sup
port the
early dete
c
tio
n
of diabeti
c
retinop
athy in a
prima
r
y-care environ
ment.
In publication
[12], classifiers
su
ch a
s
t
he Gau
s
sian
Mixture mod
e
l (GMM
), k-nearest
neigh
bor (kNN), su
ppo
rt vector ma
chi
n
e (SVM),
an
d AdaBoost
were analy
z
e
d
for classifying
retinop
athy lesio
n
s from
nonle
s
ion
s
.
An algor
ith
m
to detect
the pre
s
e
n
c
e of exud
a
t
es
automatically wa
s p
r
op
o
s
ed [1
3]. Rese
arch
in
[14] presente
d
a meth
od
for auto
m
a
t
ed
identificatio
n
of exudate
p
a
thologi
es in
retinop
athy i
m
age
s
ba
sed
on
computati
onal i
n
tellige
n
ce
techni
que
s. T
he
colo
r
retin
a
l imag
es a
r
e segm
ented
usi
ng fu
zzy
c-m
ean
s
clu
s
tering
followi
ng
some
p
r
ep
ro
ce
ssi
ng
step
s. A gen
etic-b
ase
d
al
gorith
m
is u
s
ed
to
ran
k
th
e feat
ure
s
and
ide
n
tify
the subset that gives th
e be
st
cl
as
si
f
i
cat
i
on
re
sult
s.
The
s
e
le
cted feature v
e
ctors
are
th
en
cla
ssifie
d
usi
ng a multilayer neu
ral n
e
twork cl
assifie
r
.
Segmentatio
n method
without initia
lizati
on p
r
o
c
ess wa
s p
r
opo
sed in [
15]. The
segm
entation
wa
s
con
d
u
c
ted by u
s
ing
the maximu
m
value sel
e
ction re
sults of
co
nvolutio
n
8
dire
ction
s
. Publication [16] sho
w
ed
a me
thod for vascula
r
pattern eh
nacement a
n
d
segm
entation
.
An automated syste
m
which u
s
e
s
wa
velets to enh
ance the vascula
r
pattern
wa
s
prop
osed an
d
then su
b
s
eq
uently
a
pplied a
pi
e
c
e
w
ise threshold p
r
o
b
ing
and
ada
ptive
thresholdi
ng f
o
r ve
ssel lo
calizatio
n an
d
segm
entation
re
spe
c
tively. In this
arti
cle
[17], a meth
od
to improve th
e qu
ality of in
put retinal im
age
was
p
r
e
s
ente
d
a
nd consi
dered as a
p
r
ep
ro
ce
ssi
ng
step i
n
a
u
tom
a
ted di
agn
osi
s
of
diab
etic retinopat
hy. T
he p
r
e
p
ro
ce
ssing
con
s
ist
s
of ba
ckgroun
d
estimation a
n
d
noise rem
o
val from retin
a
l image by a
pplying coarse and fine se
gmentation.
3. Featur
e Extra
c
tion Me
thod
Flowcha
r
t of the re
se
arch
pro
c
e
ss
sh
o
w
n in
Fig
u
re
1. In this re
search, the in
put image
is an
imag
e o
f
the retina
of
diabeti
c
retin
opathy
patie
n
t
s obtain
ed
b
y
using
a fun
dus
ca
mera
or
Zeiss visuca
m non
med
r
i
a
tic p
r
o
cam
4726
with
a
capture
5.0 M
P
sen
s
o
r
p
r
o
duces imag
e
files
in the format
of the Joint P
hotographi
c
Grou
p (J
PG
)
with a resoluti
on of 244
8 x 3696 pixel
s
from
Eye Hospital
Bandun
g Ci
cen
do that h
a
ve been
val
i
dated. Imag
es
were obta
i
ned 75
pie
c
es,
divided into t
h
ree
cla
s
se
s, i.e. 25 pie
c
es of n
o
rm
al
eyes, 25
ey
es
Non P
r
olif
erative
Diabe
tic
Retino
pathy (NPDR), an
d 25 eyes Prolifer
ative diab
etic retino
pathy
(PDR).
The o
r
iginal i
m
age
sized
2448 x 36
96
pixels is
co
nverted to im
age si
ze
d 57
6 x 720
pixels. Chan
ging ima
ge
size is inte
n
ded to r
edu
ce the
wo
rkl
oad of the
comp
uter
so
the
comp
utation
can
be
don
e
more
qui
ckly. Initial proc
e
s
s
(p
re
-
p
r
o
c
e
ss
in
g)
is
c
o
ndu
c
te
d to
ob
tain
the cha
r
a
c
teristics of exud
ate
and bloo
d vessels. Hence, the
de
sire
d obje
c
t can b
e
obtai
ned
with maximu
m results. Retinal image
data that
have been
re
sized later
cha
n
ged into a g
r
ay
scale imag
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Exudate a
nd
Blood Vessel
Feature Extra
c
tion in Di
abe
tic Retino
path
y
… (Si
s
wo Wardo
y
o
)
1495
Figure 1. Flowchart of the
resea
r
ch pro
c
ess
A feature ex
traction
meth
od u
s
ed i
n
this
resea
r
ch
is the m
o
rp
hologi
cal
op
eration
s
.
Morp
holo
g
ica
l
operation
s
as a m
e
thod
for extr
a
c
tin
g
image
com
pone
nts a
r
e
useful i
n
ima
g
e
rep
r
e
s
entatio
n and d
e
scri
ption of the area
of
exud
ate and bl
oo
d vessels. In
exudate feat
ure
extraction
ph
ase, the in
pu
t image data
that has b
e
e
n
co
nverted i
n
to a gray scale imag
e, then
carrie
d on m
o
rph
o
logi
cal
operation
s
which
applie
d
morp
holo
g
ica
l
clo
s
ing o
p
e
r
ation. It serve
s
to
increa
se th
e
area
of exu
d
a
te an
d remo
ve bloo
d ve
ssel
s. The
nex
t pro
c
e
s
s i
s
t
he
colum
n
wi
se
neigh
borhoo
d
.
In this proce
ss, the
imag
e
is reset
into
colum
n
s th
us forming
a m
a
trix while
usi
ng
the function "
c
olfilt" in matlab whi
c
h
i
s
u
s
eful for ma
rking exudate a
r
ea.
The re
sults
of the proce
ss a
r
e then
pro
c
ee
ding t
h
rou
gh thre
sholdin
g
. The
thresh
old
value applie
d
is 0.9, which if the image is co
nverted i
n
to a binary i
m
age, ca
n be
written:
If x <0.9 then x = 0, otherwise x
= 1
The next sta
ge is a morp
hologi
cal op
e
n
ing ope
ratio
n
. It applies dilation followed by an
ero
s
ion
p
r
o
c
e
ss,
whi
c
h
aim
s
to
fill a
hole
or ga
p
of ex
udate. O
p
tic
disc
contai
ns
the hig
h
e
s
t pi
xel
values i
n
the
image of
retin
a
; therefo
r
e, removing
the
optical
disk i
s
by sea
r
chin
g
for the hi
ghe
st
pixel values o
f
a gray scale
image. The
size of
the op
tic disc define
d
at 576 x 72
0 colo
r imag
e
s
have a size o
f
optical disk defined by 80
pixels
gre
a
te
st, so it will be cre
a
ted ma
sk to rem
o
ve the
area
of the optic disc. The
n
the optic di
sc i
s
re
m
o
ve
d togethe
r wit
h
the border.
Re
sulting d
a
ta is
carrie
d ba
ck
with mo
rphol
ogical ope
rati
ons
namely e
r
osi
on, which
aims to rem
o
ve noi
se tha
t
is
not exudate. Thus, the feat
ure extr
a
c
tion
of exudate is available.
For featu
r
e e
x
traction of bl
ood vessel
s, t
he input ima
ge data which has b
een
converted
into a gray scale ima
ge is then compl
e
mented o
r
in
verted. In this pro
c
e
s
s, the previou
s
re
tinal
image h
a
s a
black on
wh
ite backg
rou
nd, so that
o
b
ject
s whi
c
h
form bloo
d vessel
s wo
uld
b
e
more vi
sible.
Next pro
c
e
ss is known a
s
the C
ont
ra
st Limited Ada
p
tive Histo
g
ram Equali
z
ati
on
(CLA
HE).
In this
process t
he com
p
lem
ented retinal
image is
averag
ed ima
g
e
histog
ram
to
improve the
contra
st of the image in o
r
de
r to make
the
hidde
n feature can b
e
se
e
n
more
clea
rl
y.
After the co
n
t
rast of the
retinal imag
e
is
flattened, t
hen the m
e
d
i
an filter process is
carrie
d out. T
h
is p
r
o
c
e
s
s ai
ms to
elimina
t
e unne
ce
ssa
r
y noi
se in
th
e extra
c
tion o
f
blood ve
ssel
s.
The n
e
xt stag
e is a m
o
rp
ho
logical op
eni
ng o
perat
ion;
the impl
eme
n
tation of
ero
s
ion
followed
by
a dilation
op
eration
aim
s
to prote
c
t
small bloo
d v
e
ssel
s while
dilation aim
s
to increa
se
the
details
of larg
e bloo
d vessels a
nd the
n
remo
ve
d. Subse
que
ntly, bound
ary extraction
co
ndu
cted
by su
btra
ctin
g the
re
sult
of t
he m
edia
n
filter
with t
he
re
sults of
morphol
ogi
cal dilatio
n
, which
aims to sepa
rate the blo
o
d
vessels o
u
t
of bac
kgro
u
nd. The resu
lts of the blo
od vessel
s a
r
e
pro
c
ee
ded
int
o
thre
sh
oldin
g
, with th
re
sh
old valu
e of 0
.
1. If the ima
ge i
s
conve
r
ted into
a
bin
a
ry
image, it can
be written:
If x <0.1 then x = 0, othe
rwi
s
e x
= 1
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1493 – 150
1
1496
The threshol
ding p
r
o
c
e
ss is follo
wed
by image
co
mpleme
nt, so the ima
g
e
of the
previou
s
ly wi
th black b
a
ckgroun
d whe
r
ea
s the
obje
c
t in the fo
rm of white
bl
ood ve
ssels
are
cha
ngin
g
into
the im
age
with white
ba
ckgroun
d a
n
d
the obj
ect
be
come
s bla
c
k. The
result
o
f
blood vessel
s extraction
still cont
ains
noise. Theref
ore, the next
process i
s
noise elimination
with median filter. It will provide the final
result of blood vessels extraction.
After obtainin
g
the results
of
exudates e
x
traction an
d blood ve
ssels, value of area whi
c
h
is extensive
or exudate
s
i
s
cal
c
ul
ated. Centroid is u
s
eful to dete
r
mine the po
sition of object
s
.
This value i
s
obtaine
d from
the following
equatio
n:
∑∑
,
(
1
)
,
∑∑
.
,
;
∑∑
.
,
(
2
)
Whe
r
e:
f(i.j)
=1 if
(i,j)
is an obje
c
t
pixel
.
Feature extraction p
r
o
c
e
ss em
ploys
Fast Fo
u
r
ie
r Tran
sform (FTT) ba
si
c prin
ciple,
decompo
sitio
n
calculation
of Di
scret Fourie
r Tra
n
sf
orm (DFT
) from length N into a serie
s
of
s
m
aller DFT
respec
tively. For s
i
mplic
i
ty purpose,
it is assumed that the samples N in row X (n)
is the
re
sult
of powers of
2;
otherwi
se i
t
is ne
ce
ssary to add 0. S
o
, being th
e
numbe
r n
earest
results of the
powe
r
s. The
next proce
ss is data
norm
a
lizatio
n. Only magnitude
parts of Fo
uri
e
r
trans
form result are tak
e
n into acc
o
unt, omi
tting the imaginary parts
. The maximum magnitude
value is u
s
e
d
as divisor fa
ctor fo
r the rest val
ue of
magnitud
e
s,
so that the m
a
ximum valu
e of
each ch
ara
c
t
e
risti
c
pattern
of exudates i
s
wo
rth 1.
4. Results a
nd Analy
s
is
Exudates fea
t
ure
extractio
n
is do
ne
by
usi
ng
morp
hologi
cal
op
e
r
ation
s
. At th
e final
stage, gen
erated can
d
idat
e
of
exud
at
es is
ca
rrie
d
in
the o
peration
pro
c
e
s
s of e
r
osio
n. It aims to
r
e
mo
ve
no
ise. R
e
s
u
lts
a
r
e
s
h
ow
n
in
F
i
gu
r
e
2
.
(a)
(b)
Figure 2. (a)
Origin
al imag
e; (b) Exudat
e extraction
result
After obtainin
g
the re
sult
s, the ce
ntroid
a
r
ea can be calcul
ated
a
n
d
the
re
sults o
b
tained
from the exu
dates. It ca
n
be de
scrib
e
d
as
sha
pe of
vertical
sam
p
ling in eve
r
y 15 pixel
s
of 7
2
0
pixels to
obta
i
n 48
sa
mplin
gs i
n
a ve
rtical; while
ho
ri
zontal
sampli
ng p
r
o
c
e
s
s in
every 1
4
pix
e
ls
of 576 pixels.
So it obtaine
d 39 numbe
rs of hori
z
onta
l
samplin
g of exudate
s
usi
ng a fast Fou
r
ier
transform. Fi
gure 3 illustra
tes sampling
results.
Figure 3. Sampling result in vertical an
d hori
z
ontal o
f
Exudates
0
5
10
15
20
25
30
35
40
45
50
0
0.
2
0.
4
0.
6
0.
8
1
V
e
r
t
i
c
al
E
x
u
dat
e S
a
m
p
l
i
n
g R
e
s
u
l
t
s
Sa
m
p
l
i
n
g C
o
u
n
t
Ma
g
n
i
t
u
d
e
0
5
10
15
20
25
30
35
40
0
0.
2
0.
4
0.
6
0.
8
1
Ho
r
i
z
o
n
t
a
l
E
x
u
d
a
t
e
S
a
m
p
lin
g
Re
su
lt
s
S
a
m
p
lin
g
C
o
u
n
t
M
agn
i
t
ude
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Exudate a
nd
Blood Vessel
Feature Extra
c
tion in Di
abe
tic Retino
path
y
… (Si
s
wo Wardo
y
o
)
1497
It can
be
see
n
from
Ta
ble
1 that th
e
ch
ara
c
teri
stic of
exudate
s
ca
n be
extra
c
te
d u
s
in
g
morp
holo
g
ica
l
operation m
e
thod. In no
rmal eyes
, th
ere a
r
e
no e
x
udates,
so t
he re
sult
s of
the
cal
c
ulatio
n wi
ll
pro
d
u
c
e a cha
r
a
c
teri
stic
with a
va
lu
e
0 in
the
are
a
. In the
eye
s
of
patient
s
with
diabeti
c
retin
opathy di
sea
s
e in
NP
DR
class, there wil
l
be mo
re exu
dates with ch
ara
c
teri
stic are
a
that is p
r
o
d
u
c
ing a
r
ou
nd 1
7
-
21.21
3 pixel
s
, an
d in
P
D
R
cla
ss
of
t
h
e
re
sult
ing
c
h
a
r
act
e
ri
st
ic
ran
ge
is 12
5-12.299
pixels. Th
e result of
ce
ntroid val
ue i
ndi
cate
s the
po
sition of exuda
tes in th
e reti
nal
image.
Table 1. The
results of feat
ure extra
c
tion
area an
d ce
n
t
roid of exuda
te
Inpu
t
image
(Nor
mal)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
Inpu
t
image
(NPDR
)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
Inpu
t
image
(PDR)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
1-1
0
x : 0
y : 0
2-1
580
x :223,5
744
y :356,0
625
3-1
1392
x :395,6
530
y :389,2
270
1-2
0
x : 0
y : 0
2-2
85
x :352,9
091
y :192,4
818
3-2
2806
x :329,4
180
y :320,8
001
1-3
0
x : 0
y : 0
2-3
84
x : 300,3
200
y :183,0
640
3-3
521
x :355,6
948
y :201,4
549
1-4
0
x : 0
y : 0
2-4
69
x :338,8
000
y :178,9
130
3-4
3216
x :337,2
243
y :353,9
678
1-5
0
x : 0
y : 0
2-5
84
x :400,2
991
y :445,2
710
3-5
1237
x :339,1
827
y :327,7
227
1-6
0
x : 0
y : 0
2-6
46
x :280,4
189
y :184,1
892
3-6
3252
x :383,7
952
y :353,5
723
1-7
0
x : 0
y : 0
2-7
47
x :482,8
108
y :298,2
973
3-7
901
x :268,4
861
y :292,2
020
1-8
0
x : 0
y : 0
2-8
56
x :402,4
878
y :316,8
659
3-8
2761
x :392,9
510
y :321,2
228
1-9
0
x : 0
y : 0
2-9
2841
x :338,8
323
y :310,3
687
3-9
293
x :525,7
571
y :315,2
145
1-10
0
x : 0
y : 0
2-10
4673
x :311,1
095
y :293,7
997
3-10
3422
x :210,0
383
y :294,6
838
1-11
0
x : 0
y : 0
2-11
2330
x :183,9
336
y :300,2
533
3-11
3419
x :240,1
269
y :163,3
103
1-12
0
x : 0
y : 0
2-12
794
x :516,1
103
y :252,4
598
3-12
3711
x :243,5
640
y :179,6
696
1-13
0
x : 0
y : 0
2-13
200
x :281,9
672
y :318,4
270
3-13
886
x :453,0
835
y :292,0
316
1-14
0
x : 0
y : 0
2-14
44
x :542,2
581
y :313,7
742
3-14
308
x :373,2
662
y :355,5
130
1-15
0
x : 0
y : 0
2-15
1773
x :258,3
420
y :283,3
845
3-15
249
x :402,0
442
y :285,8
032
1-16
0
x : 0
y : 0
2-16
1255
x :241,5
995
y :216,1
544
3-16
3235
x :336,8
475
y :353,0
247
1-17
0
x : 0
y : 0
2-17
677
x :452,7
440
y :294,9
463
3-17
277
x :443,4
007
y :484,2
563
1-18
0
x : 0
y : 0
2-18
922
x :177,7
050
y :359,2
722
3-18
21
x :381,3
810
y :278,2
381
1-19
0
x : 0
y : 0
2-19
1003
x :192,1
745
y :359,0
638
3-19
1492
x :306,2
692
y :206,0
562
1-20
0
x : 0
y : 0
2-20
847
x :204,1
747
y :387,8
158
3-20
246
x :476,5
081
y :353,1
707
1-21
0
x : 0
y : 0
2-21
631
x :155,4
152
y :366,1
616
3-21
1592
x :210,8
568
y :350,8
970
1-22
0
x : 0
y : 0
2-22
793
x :239,4
288
y :376,0
858
3-22
1451
x :509,7
560
y :351,0
887
1-23
0
x : 0
y : 0
2-23
1483
x :415,2
232
y :190,1
807
3-23
2856
x :447,7
606
y :346,7
714
1-24
0
x : 0
y : 0
2-24
140
x :529,9
571
y :196,1
500
3-24
2667
x :405,6
547
y :377,6
258
1-25
0
x : 0
y : 0
2-25
185
x :517,7
892
y :206,6
108
3-25
2122
x :367,4
449
y :262,5
608
The
re
sult of
sampli
ng i
n
v
e
rtical
an
d h
o
r
izo
n
tal u
s
e
s
the FFT
in ex
udate
s
. It sh
o
w
s that
there i
s
a
si
milarity pattern between th
e re
sult
s
of image extract
ed that
conta
i
n exudate
s
with
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1493 – 150
1
1498
those
that
do
not
co
ntain
exudate
s
. So
, it will
be
difficult to
di
stin
guish im
age
with the
exu
d
a
tes
or witho
u
t the exudates.
Feature extra
c
tion of blood
vessel
s
pe
rforme
d
by usi
ng morpholo
g
ical op
eratio
n. At
the
final stage, g
enerated can
d
idate
s
of blood vessels
i
s
done throug
h
median filter pro
c
e
ss. It aims
to remove un
necessa
ry no
ise. Re
sult
s a
r
e sh
own in F
i
gure 4.
(a)
(b)
Figure 4. (a)
Origin
al imag
e; (b) Extracti
on re
sult of blood vessel
s
After obtainin
g
the re
sult
s, the ce
ntroid
a
r
ea can be calcul
ated
a
n
d
the
re
sults o
b
tained
from the
bloo
d vessel
s. Th
en, it ca
n b
e
t
r
an
slated
into
form
with the
vertical
sam
p
ling p
r
o
c
e
ss in
every 15
pixels of
720
pi
xels to o
b
tai
n
48
sam
p
li
ng in
a verti
c
al; an
d h
o
ri
zontal
sa
mpli
ng
pro
c
e
s
s in
every 14
pixel
s
of 57
6 pixel
s
to
obtain
3
9
ho
rizontal
sampli
ng
of
blood
vessel
s by
using a fast F
ourier tr
ansform. Figure 5 illustra
tes sam
p
ling
result
s.
Figure 5. Sampling result in vertical an
d hori
z
ontal o
f
Blood Vessels
It can be se
en from Tabl
e 2 that the
cha
r
a
c
teri
stic of exudates
can be extracted b
y
usin
g mo
rph
o
logi
cal op
eration. The
calcul
ation
r
e
sult
s m
a
y
in
dicat
e
t
h
at
t
he mo
re
sev
e
re
diabeti
c
retin
opathy di
sea
s
e, the
more
ch
ara
c
te
ri
sti
c
s value
are
cou
n
t in bl
oo
d vessel
s a
r
e
a
s.
Espe
cially in
PDR cl
ass,
the ra
nge
ch
ara
c
teri
st
ic is 13.31
9-4
6
.6
81pixel
s
on
a no
rmal
cla
ss,
7.435-49.93
8
pixels on NP
DR
cl
as
s, an
d 13.8
91-53.
802
pixels
on
PDR cl
ass.
Becau
s
e
in P
D
R
cla
ss
ne
w bl
ood ve
ssels
emerge, it
ca
use
s
ma
ny a
r
ea
s that cou
n
t
ed. The
re
sulting of cent
roid
value indi
cat
e
s the
po
sition of the bl
o
od vessel
s in
retinal im
ag
es. Th
e re
sul
t
of samplin
g
in
vertically a
n
d
hori
z
o
n
tally
by usi
ng the
FFT of
th
e bl
ood ve
ssel
s
sho
w
s
similarity betwee
n
t
he
results
of extracted image pattern
of each
class.
T
herefore, it
will be
difficult to distinguish
norm
a
l eyes i
m
age, NP
DR, and PDR.
0
5
10
15
20
25
30
35
40
45
50
0
0.
2
0.
4
0.
6
0.
8
1
B
l
ood
V
e
s
s
e
l
s
V
e
r
t
i
c
al
S
a
m
p
l
i
n
g
R
e
s
u
l
t
s
S
a
m
p
l
i
ng
C
o
unt
M
a
gn
i
t
ud
e
0
5
10
15
20
25
30
35
40
0
0.
2
0.
4
0.
6
0.
8
1
B
l
ood
V
e
s
s
e
l
s
H
o
r
i
z
o
n
t
al
S
a
m
p
l
i
n
g
R
e
s
u
l
t
s
S
a
m
p
lin
g
Co
un
t
M
agn
i
t
u
d
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Exudate a
nd
Blood Vessel
Feature Extra
c
tion in Di
abe
tic Retino
path
y
… (Si
s
wo Wardo
y
o
)
1499
Table 2. The
results of feat
ure extra
c
tion
area an
d ce
n
t
roid of blood
vessel
Inpu
t
image
(Nor
mal)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
Inpu
t
image
(NPDR
)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
Inpu
t
image
(PDR)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
1-1
36906
x
:405,3196
y
:244,7290
2-1
10837
x
:424,1475
y
:265,2343
3-1
26858
x :407,0
451
y :280,0
814
1-2
20719
x
:372,5667
y
:248,4841
2-2
34175
x
:402,7450
y
:258,4805
3-2
32279
x :410,7
888
y :282,9
373
1-3
13514
x
:360,9906
y
:280,5420
2-3
25850
x :361,7
148
y
:250,4610
3-3
19807
x :412,5
977
y :273,4
354
1-4
24206
x
:369,9941
y
:260,1239
2-4
22185
x
:360,5881
y
:259,8978
3-4
37408
x :404,3
489
y :297,0
878
1-5
41902
x
:396,5987
y
:284,8464
2-5
49338
x
:344,9586
y
:290,8033
3-5
13778
x :297,4
377
y :314,0
909
1-6
37586
x
:302,4817
y
:255,4885
2-6
19981
x
:359,0540
y
:263,8233
3-6
38343
x :317,2
044
y :296,8
401
1-7
46681
x
:325,7313
y
:255,7115
2-7
7435
x
:417,6603
y
:311,7155
3-7
53506
x :379,5
169
y :257,6
657
1-8
20467
x
:412,8072
y
:253,6530
2-8
41878
x
:332,7163
y
:267,7692
3-8
32485
x :311,3
984
y :283,1
197
1-9
14084
x
:376,3872
y
:248,6167
2-9
46033
x
:313,9823
y
:260,7691
3-9
13891
x :326,3
494
y :291,8
200
1-10
14075
x
:383,6021
y
:233,9169
2-10
35486
x
:322,5687
y
:242,7940
3-10
29823
x :384,4
850
y :285,2
604
1-11
23564
x
:416,5386
y
:238,4146
2-11
26369
x
:356,8534
y
:284,8482
3-11
20350
x :404,6
880
y :254,8
547
1-12
26488
x
:418,8456
y
:229,3821
2-12
27378
x
:354,0234
y
:306,7256
3-12
20803
x :414,7
340
y :278,0
944
1-13
33852
x
:374,1127
y
:267,8125
2-13
32927
x
:399,5387
y
:265,7998
3-13
53802
x :341,5
023
y :257,0
063
1-14
20610
x
:309,1946
y
:286,9788
2-14
40259
x
:368,3261
y
:275,4682
3-14
21170
x :406,4
070
y :203,8
502
1-15
13319
x
:360,2306
y
:283,2986
2-15
23552
x
:386,2624
y
:258,8777
3-15
14110
x :446,0
674
y :228,9
484
1-16
40691
x
:343,3053
y :2702
422
2-16
20736
x
:404,7231
y
:265,4687
3-16
37273
x :404,1
395
y :297,5
277
1-17
26634
x
:336,0672
y
2-17
35798
x
:326,8952
y
3-17
19087
x :419,9
178
y :219,6
311
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1493 – 150
1
1500
Inpu
t
image
(Nor
mal)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
Inpu
t
image
(NPDR
)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
Inpu
t
image
(PDR)
Ar
e
a
(pixel
)
Centr
o
id
(pixel
)
:239,9709
:260,6726
1-18
23674
x
:406,1873
y
:233,8725
2-18
12344
x
:403,0478
y
:226,8983
3-18
23747
x :339,7
121
y :244,1
759
1-19
6527
x
:476,7973
y
:346,0488
2-19
10829
x
:411,6681
y
:215,5729
3-19
44031
x :370,5
633
y :260,0
265
1-20
15835
x
:425,5202
y
:337,9812
2-20
11251
x
:409,3723
y
:227,4543
3-20
11997
x :288,6
491
y :260,4
625
1-21
14261
x
:442,4966
y
:333,6138
2-21
9951
x
:393,7897
y
:226,9817
3-21
39953
x :356,6
404
y :257,4
752
1-22
14846
x
:426,0496
y
:359,3261
2-22
12039
x
:421,5169
y
:224,1961
3-22
40321
x :364,6
118
y :257,8
295
1-23
15676
x
:307,4101
y
:347,1737
2-23
23740
x
:318,3868
y
:252,8161
3-23
18727
x :414,7
949
y :265,2
459
1-24
16820
x
:328,2629
y
:330,1377
2-24
11579
x
:315,2440
y
:228,4444
3-24
35499
x :363,5
906
y :250,6
627
1-25
39148
x
:372,0514
y
:317,0558
2-25
11890
x
:328,0332
y
:235,7320
3-25
18766
x :345,7
100
y :248,2
891
5. Conclusio
n
Morp
holo
g
ica
l
operation h
a
s
be
en u
s
e
d
as featu
r
e ext
r
actio
n
meth
o
d
. It can be
u
s
ed to
extract the
ch
ara
c
teri
stic of
diabetic
retin
opathy
disea
s
e that is exu
dates a
nd bl
o
od vessel
s. The
results
obtain
ed fro
m
the
extraction
of
featur
e
area
exudate
s
ha
s a rang
e 0
p
i
xel for n
o
rm
al
eyes, 1
7
-21.
213
pixels fo
r retinal i
m
a
ge
with
NPDR
cla
ss,
and
125
-1
2.299
pixels fo
r
reti
nal
image with
PDR cla
ss. Ce
ntroid cal
c
ula
t
ion
re
su
lt obt
ained in th
e e
x
udates of fe
ature extra
c
ti
on
has a range
of x=0; y=0 for normal eyes, x=
15
0.97
15-5
68.95
65;
y=167.013
3-445.87
84 pixel
s
for retinal im
age
with
NP
DR cl
ass, a
nd x=
187.1
0
98-5
35.23
28
pixels; y
=
17
6.0333
-46
8
.7
908
pixels for retinal image
with PDR
class.
The re
sults obtaine
d
fro
m
the extra
c
tion
of
fe
ature a
r
ea
blo
o
d
vessel
ha
s
a rang
e
13.319
- 46.6
81 pixel for n
o
rmal eye
s
,
7.435 - 4
9
.9
3
8
pixel for ret
i
nal image
wi
th NPDR cla
ss,
and
13.891
-
53.802
pixel f
o
r
retinal i
m
a
ge
with PD
R cla
ss. Ce
ntroi
d
calculation result
obtai
ne
d
in the bl
ood
vessel of fea
t
ure extractio
n
ha
s a
ra
ng
e of x = 29
5,5133
-
489,0
853
pixel ; y
=
222,45
09
- 3
65,122
6 pixel
for n
o
rm
al e
y
es, x = 30
9
,
8893
- 4
54,
2538
pixel
;
y =
167,1
1
8
-
317,15
32 pix
e
l for retinal
image with
NPDR
clas
s, a
nd x = 302,4
443 - 44
3,22
36 pixel ; y
=
202,88
27 - 3
15,600
3 pixel
;
y=176.033
3
-
468.7
908 pix
e
l for retinal i
m
age with P
DR
cla
ss.
Referen
ces
[1]
W
ild S,
Rog
lic
G, Green A, S
i
cree
R, Ki
ng
H
.
Globa
l Pr
eval
ence
Of Di
abet
es: esstimates
for the
year
200
0 an
d proj
e
c
tion for 203
0. 200
4; 27: 53-1
047.
[2]
W
ong T
Y
, Yau
J, Rog
e
rs S,
Ka
w
a
s
a
ki
R, L
a
mour
eu
x E
L
,
Ko
w
a
lski J.
Gl
oba
l pr
eval
enc
e of
dia
betic
retino
pathy: P
ool
ed d
a
ta from th
e U
n
ite
d
States, Aus
t
ralia E
u
rop
e
and As
ia
. Pr
osidi
ng T
he
Associati
on for
Researc
h
in Vi
sion a
nd Optha
l
molo
g
y
Ann
u
a
l
Meetin
g. 201
1.
[3]
Soe
w
o
n
d
o
P,
Soeg
on
do S,
Suastika K, Pr
anot
o A, S
oeat
madji
DW
, T
j
o
k
ropra
w
i
r
o A.
T
he DiabC
are
Asia 20
08 Stu
d
y
- Out - Co
mes on Co
ntrol an
d Co
mp
li
cations of T
y
p
e
2 Dia
betic
Patients in
Indon
esi
a
.
Med J Indon
esia
.
201
0; 19(4): 43
-235.
[4]
Sitompul,
Ratna.
Retino
pati D
i
ab
etik.
J Indon
Med Assoc
. 2011; 61(
8): 377
-341.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Exudate a
nd
Blood Vessel
Feature Extra
c
tion in Di
abe
tic Retino
path
y
… (Si
s
wo Wardo
y
o
)
1501
[5]
June C
hu, Yus
u
f Ali. Diab
e
tic Retin
opat
h
y
: A
Revie
w
.
Dr
ug Devel
o
p
m
ent
Rese
arch
. 20
0
8
; 69: 1-14.
[6]
Jenn
ifer L, W
i
l
k
inso
n-Berka,
Antoni
a G Mi
ll
er. Upd
a
te
on t
he T
r
eatment
of Dia
betic
Ret
i
no
path
y
.
The
Scientific W
o
rl
d Journ
a
l
. 20
0
8
; 8: 98-12
0.
[7]
Muthu R
a
ma
Krishn
an Mo
ok
iah, U R
a
j
endr
a Achar
y
a
,
Ch
ua Ku
an
g Ch
u
a
, Cho
o
Min
Li
m, EYK Ng,
Augusti
nus L
a
ude. Com
puter
-aid
ed di
ag
nosi
s
of diabetic ret
i
no
path
y
: A rev
i
e
w
.
Co
mputer
s in Biol
ogy
and Me
dici
ne.
201
3; 43(1
2
): 2136-
215
5.
[8]
R Pires,
HF Je
linek,
J Wai
ner
, S Goldenstein, E Valle, A Rocha. As
se
ssi
ng
th
e N
e
e
d
for Referral
in
Automatic Diabetic Re
tinopathy
Detection
.
IEEE Transactions on Biom
edic
al E
ngineering
. 2
013;
60(1
2
): 339
1-3
398.
[9]
T
W
a
lter, JC
Klei
n, P Massin, A Ergina
y. A c
ontributi
on
of image pr
oc
essin
g
to the dia
gnos
is of
dia
betic r
e
tino
path
y
-d
etectio
n
of
e
x
ud
ates
in c
o
lor
fun
d
u
s ima
ges
of
the
huma
n
r
e
tina
.
IEEE
T
r
ansactio
n
s o
n
Medic
a
l Imag
ing
. 20
02; 21(
1
0
): 1236-
12
43.
[10]
NG Ranam
uka
,
RGN Meega
ma.
Detection
of hard exu
dat
es from di
ab
eti
c
retinop
athy i
m
a
ges us
ing
fu
zz
y
l
ogic.
In IET
Image Processin
g
. 201
3; 7(2): 121-1
30.
[11]
G Z
ahlmann,
et al. H
y
bri
d
fuzz
y
imag
e pr
oce
ssi
ng for si
tuation
assess
ment [dia
betic
retinop
ath
y
].
IEEE Engin
eeri
ng in Me
dici
ne
and Bi
olo
g
y Maga
z
i
ne
. 2
000;
19(1): 76-8
3
.
[12]
S Ro
y
c
ho
w
d
h
u
r
y
, D Ko
ozek
ana
ni, K Parhi.
DREAM: Diab
e
tic Retin
opat
h
y
Analysis Usi
ng Mach
in
e
Lear
nin
g
.
IEEE Journal of Biom
edic
al
and Health Infor
m
atics
. 2014; 18(
5)
: 1717-1
7
2
8
.
[13]
SW
F
r
anklin, S
E
Raj
an.
Diagnosis
of diabetic re
tinopathy
by em
ploying image
proc
essing technique
to detect exud
ates in retin
a
l i
m
a
ges.
IET
Image Proc
essin
g
. 2014; 8(
10): 601-
609.
[14]
A Osareh, B Shad
gar, R Mar
k
ham. A Com
putatio
na
l-Intell
ige
n
ce-Bas
ed
Appro
a
ch for Detectio
n of
Exu
dates i
n
Diab
e
tic Reti
n
opath
y
Imag
e
s
.
IEEE
Transactions on Info
rmation Technology in
Bi
om
ed
i
c
in
e
. 2
009; 13(
4): 535
-545.
[15]
A Muntasa, IA
Siraju
di
n, MK
Soph
an. Matri
x
Mask Overla
p
p
in
g a
nd C
onv
oluti
on Ei
ght D
i
rectio
ns for
Bloo
d Vess
el S
egme
n
tatio
n
o
n
F
u
n
dus
Retin
a
l Imag
e.
TEL
K
OMNIKA Teleco
mmu
n
icati
o
n Co
mputi
n
g
Electron
ics an
d Contro
l.
201
4; 12(3): 63
1-6
38.
[16]
MU Akram, I J
a
mal, A T
a
riq.
Bloo
d Vess
el E
nha
nceme
n
t a
nd S
egme
n
tati
on for Scr
e
e
n
in
g of D
i
a
beti
c
Retin
opat
h
y
.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mp
uti
ng Electr
onics
and C
ontrol.
201
2;
10(2):
327-
334.
[17]
I Jamal, M
U
A
k
ram, A T
a
riq. Reti
nal
Ima
g
e
Prepr
ocessi
ng
: Backgro
un
d
and
No
ise
Se
gmentati
on.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol.
20
12; 10(3): 5
37-
544.
Evaluation Warning : The document was created with Spire.PDF for Python.