Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
13
,
No.
3
,
Ma
rch
201
9
, p
p.
1191
~
1198
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
3
.i
3
.pp
1191
-
1
198
1191
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Detectio
n of k
erato
co
nu
s i
n anteri
or segm
en
t photograph
ed
images u
sing co
rn
ea
l
curvatu
re fe
atures
Ma
ri
z
ua
n
a M
at Daud
1
, Wa
n M
im
i
Diyan
a
W
an Z
ak
i
2
, Aini
Hu
ssa
in
3
,
Ha
li
z
a
A
bdul
M
u
t
alib
4
1
,2,3
Cent
er
for
In
t
egr
ated
S
y
s
te
m
s
Engi
ne
eri
ng
and
Advanc
ed
T
ec
h
nologi
es
(INT
E
GRA
),
Facul
t
y
of Engin
ee
ring
and
Buil
t Envi
ronm
ent
,
N
at
ion
al
Univ
ersity
of
Mal
a
y
s
ia, M
al
a
y
s
ia
4
Optom
et
r
y
and
Vision
Scie
n
ce
s
Program
m
e,
School
of
He
althca
r
e
Sci
ences,
Fa
cu
lty
of
Hea
lt
h
Sc
i
enc
es,
Nati
ona
l
Unive
r
sit
y
of
Ma
lay
si
a, Ma
lay
si
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Sep
1
5
, 201
8
Re
vised
D
ec
2
,
2018
Accepte
d
D
ec
19
, 201
8
Kera
toc
onus
is
a
cor
n
ea
l
ecta
t
ic
disorder
with
c
om
ple
x
ae
t
iol
og
y
and
m
a
y
induc
e
m
il
d
to
seve
re
visual
impairment
and
conse
quent
l
y
dec
re
ase
the
qual
ity
of
life.
T
his
pape
r
pr
ese
nt
s
a
new
ker
at
oco
nus
det
ection
m
e
thod
using
cor
neal
cur
vat
u
r
e
fea
tur
es
to
diffe
ren
t
ia
t
e
norm
al
and
ker
a
toc
o
nus
ca
ses.
In
thi
s
stud
y
,
t
he
e
y
e
i
m
age
s
known
as
anteri
or
segm
ent
ed
p
hotogra
phe
d
images
(AS
PIs
)
are
c
apt
ur
ed
fro
m
side
vie
w
using
a
sm
art
phone’
s
ca
m
era.
For
the
side
-
view
images,
the
c
orne
al
cur
vat
ur
e
is
segm
ent
ed
using
splin
e
func
ti
on
to
m
e
asure
the
cor
n
e
al
cur
va
ture.
A
te
m
pla
te
dis
c
m
et
h
od
is
implemente
d
to
quant
itati
v
ely
m
ea
sure
the
stee
pen
ing
of
the
cor
n
eal
cur
vat
ur
e
of
the
ca
pture
d
AS
PIs
.
Para
m
et
ers
obta
in
ed
from
thre
e
diffe
r
ent
te
m
pla
t
e
disc
m
et
hods,
n
amel
y
,
non
li
n
ea
r,
,
cro
ss
over
poin
t,
,
and
tri
gonom
et
ri
c,
,
are
in
vesti
g
at
ed
to
rep
r
ese
nt
the
m
ost
suita
ble
cu
rva
ture
fea
tur
e.
SV
M
is
the
n
emplo
y
ed
to
class
if
y
nor
m
al
and
ker
at
oc
onus
e
y
es
.
Result
s
rev
eal
th
at
a
standalone
nonli
ne
ar
m
et
hod
give
s
a
rel
ia
b
l
e
par
ameter
with
90%
a
cc
u
racy
in
class
if
ying
the
data.
How
eve
r,
th
e
c
la
ss
ifi
c
at
io
n
per
form
anc
e
has
inc
rea
sed
to
99
.
5%
accura
c
y
w
it
h
the
use
of
all
combined
fea
tur
es
known
as
a
feature
vec
tor
,
=
<
,
,
>
.
Addit
iona
l
l
y
,
cl
assifi
ca
t
ion
wi
th
th
e
proposed
has
succ
essfull
y
disti
nguished
no
rm
al
a
n
d
ker
at
o
con
us
ca
s
es
with
sensiti
vi
t
y
and
spec
ifi
c
ity
ra
te
s
of
99%
and
100%,
respe
ctively
.
Th
e
result
s
port
ra
y
th
e
brigh
t
p
ote
ntial
of
thi
s
m
et
hod
in
assisting
expe
rts
during
ocul
ar
s
cre
en
ing
spec
if
i
ca
l
l
y
to
detec
t
k
era
to
conus
disea
se.
Ke
yw
or
ds:
An
te
rior
s
e
gme
nt
Corneal
cur
vatur
e
Ker
at
ocon
us
Photo
gr
a
phed
i
m
ages
Tem
plate
d
isc
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed.
Corres
pond
in
g
Aut
h
or
:
Wan M
im
i Diy
ana
Wab Za
ki,
Ce
nter fo
r In
te
gr
at
e
d
Syst
em
s
Enginee
rin
g
a
nd Ad
van
ce
d Tec
hnol
og
ie
s
(INT
EGR
A),
Faculty
of E
ngineerin
g
a
nd B
uilt
Environm
ent,
Nati
on
al
U
niv
e
rsity
o
f
Mal
ay
sia
, Mal
ay
sia
.
Em
a
il
: w
m
diyan
a@
ukm
.ed
u.m
y
1.
INTROD
U
CTION
Ker
at
ocon
us
(
KC)
is
a
n
ocul
ar
disease
i
nvolv
in
g
t
he
pro
gr
essi
ve
a
nd
non
-
in
flam
m
at
or
y
co
rn
ea
l
thinn
i
ng
a
nd
ste
epen
i
ng
of
the
c
orneal
curvatu
re.
H
oweve
r,
the
no
n
-
i
nf
la
m
m
a
tory
conditi
on
ha
s
bee
n
qu
e
sti
on
e
d
due
to
the
existe
nc
e
of
in
flam
matory
c
om
po
ne
nts
[
1],
[2
]
.
Th
e
aet
iolog
y
of
KC
is
heter
og
e
ne
ous
and
va
ries
wide
ly
dep
e
nd
i
ng
on
ge
ogra
phic
al
factor,
fam
i
l
y
histor
y
a
nd
r
aces.
T
he
pr
e
va
le
nce
of
t
his
disease
in
the
ge
ogra
phic
al
locat
ion
s
of
a
c
ountry
with
hot
cl
i
m
at
e
is
hig
he
r
than
t
hat
of
a
country
wit
h
coo
l
e
r
cl
i
m
at
e.
KC
is
al
so
a h
ere
ditar
y
disea
se. H
ow
ever,
ey
e
r
ubbi
ng
an
d
al
le
r
gy
are
the
m
os
t
co
ns
ist
ent
fi
nd
i
ngs
in
con
t
rib
ution
s
t
o
the
accel
erati
on of c
orneal c
urvatu
re a
dv
a
nc
e
m
ent [
3]
-
[6
]
.
KC
is
a
n
unc
om
m
on
disease
with
1
case
out
of
20
00
in
div
i
du
al
s
[
7];
none
thele
ss,
r
ecent
pr
e
valenc
e
stud
ie
s
ha
ve
s
how
n
m
ajo
r
g
e
ogra
phic
al
v
aria
ti
on
s, wit
h
t
he hig
hest cases
of 3.5
9%
r
e
port
ed
in
Tehra
n
[
8]
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
13
, N
o.
3
,
Ma
rc
h 201
9
:
1
1
9
1
–
1
1
9
8
1192
the
lo
west
pr
e
valence
of
0.0
0017%
i
n
Ja
pa
n
[9
]
.
Des
pite
this
fin
ding,
only
a
few
e
pid
em
iolog
ic
al
stu
dies
hav
e
been
pe
r
form
ed
in
Asia,
es
pecial
ly
in
Ma
la
ysi
a,
be
cause
t
he
fo
c
us
was
f
or
m
erly
on
ot
her
di
seases.
On
ly
on
e
K
C
prevale
nce
resea
rch has
b
ee
n
c
onduct
ed wit
h a re
ported p
reval
ence of
1.2%
[10
]
.
Seve
ral
m
et
ho
ds
an
d
eq
uip
m
ent,
su
c
h
as
pl
aci
do
disc,
co
r
neal
topo
gr
a
phy
and
pe
ntaca
m
,
can
be
us
e
d
to
dia
gnose
KC.
H
owe
ve
r,
detect
ing
th
e
early
phase
of
KC
wh
e
re
t
he
co
rn
ea
lo
oks
relat
ively
heal
thy
is
diff
ic
ult.
T
he
l
at
est
extensi
ve
eq
uip
m
ent
in
cl
ud
es
co
r
neal
tom
og
ra
ph
y,
corneal
bio
m
e
chan
ic
s
an
d
in
vivo
conf
ocal
m
ic
ro
sc
op
y.
T
hese
ty
pes
of
e
qu
i
pm
ent,
wh
ic
h
can
dete
ct
ea
rly
sign
s
of
KC,
pr
ov
i
de
detai
le
d
analy
sis
of
the
corneal
thic
kness
an
d
in
f
or
m
at
ion
on
the
c
orneal
s
ha
pe
of
anterio
r
a
nd
po
ste
rior
segm
ent
[11]
.
Howe
ver,
m
os
t
of
these
ty
pe
s
of
cl
inica
l
eq
uip
m
ent
are
ex
pensi
ve,
heav
y
and
im
m
ob
il
e
and
r
eq
uire
e
xp
e
rts
o
r
well
-
t
raine
d cl
inici
ans.
I
n
a
dd
it
io
n
to u
si
ng these ty
pes
of equ
i
pm
ent, o
ph
t
halm
olo
gists al
so
m
anu
al
ly
r
efer
to
ot
her
cl
inica
l
sign
s
,
s
uc
h
as
in
Ri
zz
uti
sig
n,
V
og
t
stria
e
and
Mu
nson
[
12]
sig
n,
as
guideli
nes
i
n
diag
no
si
ng
KC cases
.
W
it
h
t
he
high
pace
of
te
c
hnology
I
nter
net
of
T
hings,
re
searche
rs
can
integrate
im
age
proce
ssin
g
te
chn
iq
ues
with
sm
artph
ones
via
cl
oud
c
om
pu
ti
ng
.
Tar
e
q
pate
nted
his
work
on
the
s
yst
e
m
and
m
e
t
hod
for
ophth
al
m
olo
gi
cal
i
m
aging
usi
ng
m
ob
il
e
pr
oc
essing
dev
ic
e
[13].
He
de
ve
lop
e
d
a
de
v
ic
e
wh
ic
h
m
i
m
ick
ed
a
topogra
ph
ic
al
con
ce
pt
us
in
g
a
frust
um
cone
segm
ent,
opt
i
m
isa
t
ion
le
ns
an
d
ver
te
x
-
a
ngulati
on
posit
ion
i
ng
li
gh
t.
T
o
t
he
best
of
our
kn
ow
le
dg
e
,
on
ly
fe
w
w
orks
re
ported
on
KC
detect
io
n
a
ppro
ac
h
us
in
g
m
achine
le
arn
in
g
a
nd
i
m
age
proces
sing
m
et
ho
ds
[13]
-
[
16]
.
H
owe
ver,
these
w
or
ks
did
not
in
ve
sti
gate
the
relat
ion
s
hip
of co
rn
eal
c
urv
at
ur
e
of ASP
I a
nd the
presen
ce o
f KC.
Nu
m
erous
stu
di
es
wer
e
c
ondu
ct
ed
on
ot
her
oc
ular
disease
de
te
ct
ion
m
et
ho
ds
[
17]
-
[
20
]
usi
ng
im
age
processi
ng.
Ante
rior
segm
ent
ph
otogra
ph
e
d
i
m
age
(A
SPI)
is
an
i
m
age
t
aken
us
in
g
a
sm
artphon
e
’s
ca
m
era,
a
dig
it
al
cam
era
or
a
ny
ty
pe
of
cam
eras.
Nay
ak
[21]
propos
ed
a
n
a
uto
m
at
i
c
cl
assifi
cat
ion
of
norm
al
,
catar
act
and
post
-
cat
ar
act
of
A
SP
Is
us
in
g
SV
M
cl
assifi
er.
The
a
lgorit
hm
cou
ld
cl
assi
fy
co
rre
ct
ly
with
nea
rl
y
90%
accuracy.
Ra
ih
anah
et
al
.
[
22
]
pro
posed
a
scr
eenin
g
syst
em
wh
ic
h
ca
n
dete
ct
an
d
cl
assify
pterygi
um
and
non
-
pterygi
um
of
AS
P
Is
us
i
ng
i
m
age
processi
ng
te
ch
nique
that
was
com
bin
ed
with
m
ach
ine
le
arn
in
g
al
gorithm
.
The
syst
em
per
for
m
ed
with
accuracy
of
95.
6%
.
Ma
so
ud
et
al
.
[23]
us
ed
th
e
tem
plate
disc
m
et
ho
d
to
cal
culat
e
the
to
rtu
os
it
y
of
reti
nal
bloo
d
vessel
of
fundus
im
ages
f
or
an
a
uto
m
ated
gr
a
ding
of
diabeti
c
reti
nopathy
.
They
i
m
pr
ovi
sed
the
m
et
ho
ds
a
nd
pro
duced
t
he
ot
he
r
two
m
et
h
ods,
nam
ely,
cro
ss
over
point
an
d
trigon
om
et
ric
m
et
ho
ds,
wh
ic
h
a
re a
dap
te
d
i
n
this
work.
A
ne
w
m
et
ho
d
of
KC
detect
i
on
was
de
velo
ped
us
i
ng
t
he
corneal
cu
r
vatur
e
of
ASPI
.
The
c
orneal
curvatu
re
cal
culat
ed
us
i
ng
te
m
pla
te
disc
is
pr
ese
nted
i
n
th
is
pap
e
r.
Q
ualit
at
ively
,
cor
ne
al
cur
vat
ur
e
m
easure
d
from
a
side
view
of
AS
P
I
can
be
us
e
d
as
an
ind
ic
at
io
n
of
KC
se
ver
it
y.
The
cu
rv
at
ure
of
a
side
view
of
a
patie
nt’s
ey
e
im
age
will
be
m
easur
ed
befo
re
validat
in
g
th
e
ey
e’s
conditi
on
us
in
g
a
to
pogra
ph
y
m
achi
ne
with
an
ex
per
t
’s
ad
vice.
T
he
pro
pose
d
m
e
thod
is
exp
la
ine
d
in
t
he
ne
xt
sect
ion,
w
her
eas
t
he
r
esults
an
d
disc
us
si
on
are
discuss
e
d
i
n
Sect
io
n 3. La
stl
y, the concl
usi
on su
m
m
aris
es the e
ntire
w
ork
i
n
Sect
io
n 4.
2.
PROP
OSE
D
METHO
D
The
pr
opos
e
d
m
et
ho
d
c
on
sis
ts
of
f
our
par
t
s;
pr
e
processi
ng,
segm
entat
ion,
featu
re
e
xtracti
on
a
nd
cl
assifi
cat
ion
a
s in
Fi
gure
1.
Figure
1.
Flo
w
ch
a
rt of
ker
at
oc
onus detec
ti
on syst
em
2.1.
Preproces
sing
The
ASPIs
a
r
e
captur
e
d
f
rom
the
side
vie
w
of
patie
nts’
ey
es
us
ing
s
m
artphon
e
ca
m
eras.
In
this
stud
y,
H
uaw
ei
P9
a
nd
iP
hone
SE
ar
e
use
d
to
capt
ur
e
th
e
ey
e
i
m
ages
colle
ct
ed
f
ro
m
the
O
phthalm
ology
Dep
a
rtm
ent,
Ku
al
a
Lum
pu
r
Ho
s
pital
.
A
total
of
106
norm
al
i
m
ages
and
112
KC
im
age
s
are
colle
ct
ed,
and
al
l
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Detect
ion
of ke
ra
toc
onus i
n a
nterior
seg
me
nt
p
hoto
gra
ph
e
d i
mages
us
i
ng
cor
nea
l
.
..
(
M
ari
zu
an
a
M
at Da
ud
)
1193
i
m
ages
are
validat
ed
by
a
colla
borati
ve
opto
m
et
rist
us
ing
Corneal
To
pogr
aph
y
(CT
).
CT
gen
erates
a
dif
fer
e
nt
topogra
ph
y
m
ap
in
dif
fer
e
nt
ey
e
conditi
on
s,
su
c
h
a
s
on
Fig
ur
e
2
(a)
an
d
(c
)
f
or
KC
an
d
norm
al
ey
e,
resp
ect
ively
.
T
opogra
phy
m
a
p
is
a
colour
-
ba
sed
m
ap,
wh
e
re
co
ol
colo
ur
s
(cyan
to
blu
e
)
denote
flat
cur
ves
,
m
il
d
colo
ur
s
(
gr
ee
n
t
o
or
a
nge)
sp
e
ci
fy
m
edium
curvatu
re
a
nd
wa
rm
colo
ur
s
(
re
d
t
o
black
)
pr
es
e
nt
hi
gh
curvatu
re.
T
he
range
of
colo
ur
is
dep
en
den
t
on
the
ty
pe
of
e
qu
i
pm
ent.
Howev
e
r,
we
on
l
y
fo
cus
on
p
r
oc
essing
the co
ll
ect
ed
ASPIs
u
si
ng im
a
ge pr
o
cessi
ng
appr
oach.
Corneal
is
the
transpare
nt
fron
t
str
uctu
re
of
the
ey
e.
I
n
norm
al
cor
ne
as,
the
li
gh
t
s
cat
te
ring
is
m
ini
m
al
because
of
the
tran
s
par
e
ncy
char
ac
te
r;
ho
w
eve
r,
f
or
ab
norm
al
cor
neas
,
the
li
gh
t
scat
te
ring
incr
eases,
and
c
onseq
ue
nt
loss
of
c
orne
al
transp
a
renc
y
occu
r
s
[
24
]
.
Give
n
this
co
ndit
ion
,
ASPIs
hav
e
noise
s
suc
h
as
ref
le
ct
io
n,
un
e
ven
il
lum
inatio
n
an
d
lum
inosi
ty
.
Thu
s
,
ga
m
m
a
cor
recto
r
te
chn
i
qu
e
is
app
li
ed
to
t
he
i
m
age
durin
g
pr
e
proc
essing
t
o
co
ntr
ol
the
lum
ino
sit
y
of
A
SPI
s
,
th
ereb
y
in
directl
y
e
nh
a
ncin
g
th
e
edg
e
s
of
the co
r
ne
a
and re
duci
ng t
he reflect
io
n.
(a)
(b)
(c)
(d)
Figure
2. The
s
ide
-
view
im
age of
(a
) KC ey
e w
it
h (
b)
t
opogra
phic
al
m
ap,
and
(c)
norm
al
ey
e w
it
h
(d)
to
pogra
ph
i
cal
m
ap
2.
2.
Segmenta
tio
n
Im
age
segm
entat
ion
is
a
pa
rtit
ion
of
pix
el
s
in
to
s
ubreg
i
on
to
sim
plify
the
i
m
age
into
so
m
et
hin
g
that
is
rem
ark
able.
In
this
w
ork
,
a
s
em
i
-
autom
ated
segm
entat
ion
a
ppr
oac
h
is
perform
ed
by
sel
ect
ing
t
he
points
(
,
)
ar
ound the
cor
neal cu
r
ve,
as
s
how
n
in
Fig
ure
3
.
Figure
3. Co
rneal
cu
r
vatu
re se
gm
ented
us
i
ng S
pline
f
un
ct
i
on
Sp
li
ne
li
nea
r
i
nter
po
la
ti
on
is
then
us
e
d
to
connect
the
po
ints
with
li
nes
,
wh
ic
h
are
the
red
li
ne
as
sh
ow
n
in
Fi
gur
e 4
.
T
he
sel
ect
ion m
us
t be
on
the edge
of the
curve
w
it
h m
or
e tha
n five
co
ordinates.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
13
, N
o.
3
,
Ma
rc
h 201
9
:
1
1
9
1
–
1
1
9
8
1194
(a
)
Gr
ay
scal
e
Im
a
ge
b) Bi
na
ry im
a
ge
Figure
4. Co
rneal
cu
r
vatu
re e
xtracted
into
b
i
nar
y i
m
age
Our
ex
pe
rim
en
t
sh
ows
that
th
e
best
num
ber
of
p
oi
nt
sel
ect
ion
is 10
points
;
nev
e
rtheless, m
or
e
po
int
s
chosen
will
yield
m
or
e
accurate
curvatu
re.
Sp
li
ne
is
a
c
on
ti
nu
ous
f
unct
ion
w
hich
i
nter
po
la
te
s
the
data
b
et
wee
n
tw
o
points
w
he
re
it
i
s
const
ru
ct
e
d
f
or
m
li
near
fun
ct
ion
s
inter
pola
ti
ng
poly
nom
i
al
s.
The
re
d
c
urve
is
then
e
xtracte
d
to
obta
in
t
he
bi
nar
y
pr
e
sentat
ion
as
s
how
n
i
n
Fig
ur
e
4
b.
C
urvatu
re
m
easur
em
ent
is
cal
culat
ed
in the ne
xt sect
ion
.
2.3.
Feature
E
xt
r
ac
tion
–
Estim
ati
on Cur
vat
ur
e
C
alcula
tion
Curvat
ure
is
an
ind
ic
at
io
n
of
local
twist
ed
ness
of
a
cu
r
ve
[
23
]
-
[
25]
.
Seve
ral
m
et
ho
ds
exist
f
or
m
easur
in
g
c
urvatu
re
of
a
c
urve.
T
he
c
urvat
ur
e
is
im
plied
as
an
a
bsolute value o
f
cu
rv
at
ur
e b
y
not
c
onsideri
ng
the
r
otati
on
of
the
ta
ng
e
nt.
The
pa
ram
et
ri
c
re
pr
ese
ntati
on
when
co
ord
inate
s
=
(
)
an
d
=
(
)
a
re
giv
e
n,
[
23
]
=
′
"
+
"
′
[
(
′
)
2
+
(
′
)
2
]
3
/
2
(1)
This
ap
proac
h
known
as
a
te
m
pla
te
disc,
w
her
e
f(
x)
is
a
c
urve,
i
n
this
ca
se,
refe
rs
to
a
c
orneas
c
urve.
A
te
m
plate
disc
of
r
adi
us
,
d
i
s
create
d
at
the
centre
of
t
he
c
urve
as
in
Fig
ure
5
to
cal
culat
e
the
c
urvatu
re
at
a
po
i
nt
(
x, y)
.
Figure
5. Tem
plate
d
isc
The
fun
dam
ent
al
idea
of
this
m
et
ho
d
is
t
he
r
el
at
ion
sh
i
p
between
the
a
reas
of
the
c
urve
. Th
e
te
m
plate
disc
of
a
su
it
a
ble
rad
i
us
is
pl
aced
with
it
s
centre
at
the
pa
rtic
ular
point
of
the
c
urve.
Ra
diu
s
d
sho
ul
d
be
sm
a
ll
er th
an ra
diu
s
r
. Mat
hem
at
ic
al
ly
, th
e p
os
it
ion
of
t
he
te
m
pla
te
d
isc
and c
urvatu
re im
ply t
he follo
wi
ng
:
=
1
2
2
+
(
3
)
(2)
wh
e
re
c
is
cu
r
vatu
re
at
or
i
gin
an
d
(
3
)
denotes
higher
order
t
erm
.
The
norm
al
isa
ti
on
in
pol
ar
co
ordinates
(r,
)
can
b
e
writ
te
n
as:
sin
=
1
2
2
cos
+
(
3
3
)
(3)
wh
e
re
=
⁄
an
d
=
⁄
. T
hen, ass
um
e that
≈
0
(
)
=
sin
−
1
[
1
2
+
(
2
)
]
(4)
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Detect
ion
of ke
ra
toc
onus i
n a
nterior
seg
me
nt
p
hoto
gra
ph
e
d i
mages
us
i
ng
cor
nea
l
.
..
(
M
ari
zu
an
a
M
at Da
ud
)
1195
The
a
ngle
can
now
be
de
note
d
as
th
e
a
ngle
betwee
n
th
e
ta
ngent
li
ne
passing
t
hro
ugh
th
e
or
ig
i
n
wh
il
e a
rea
A
ca
n be ass
um
ed
as the a
rea
betw
een th
e
cor
neal cu
rv
e
an
d
t
he disc as
=
∫
1
0
∫
−
(
)
(
)
(5)
wh
e
re
=
−
2
. Replac
ed
17164
(
4) in
to (5) w
hile i
gnori
ng t
he hig
h o
rd
e
r
te
rm
an
d i
n
te
rm
s o
f
c
y
ie
ld
≈
3
3
−
3
2
(6)
wh
e
re
is t
he
c
om
ple
m
ent o
f
. H
e
nce
17
164
c
∝
and no
nlinear
est
i
m
ation
of a
curve
is d
e
fine
d
as
≜
(7)
The
hypothe
sis
of
this
w
ork
is
that
the
m
or
e conve
x
of
the
co
r
nea,
the
m
or
e
se
ver
e
the
K
C
cornea
is.
Qu
a
ntit
at
ively
,
high
,
ind
ic
at
es
the
hi
gh
se
ver
it
y
of
KC.
On
t
he
basis
of
[
23
]
,
Ma
s
oud
et
al
.
cl
aim
ed
that
this
m
et
ho
d
be
com
es
inaccur
at
e
wh
e
n
the
con
ce
rn
was
high
est
i
m
at
io
n
cu
rv
at
ur
e.
T
her
e
fore,
th
rou
gh
it
s
si
m
plici
t
y,
Ma
so
ud
m
od
ifie
d
the
m
et
ho
d
usi
ng
t
he
c
ro
ss
ov
er
point,
at
and
t
rig
onotrim
et
rical
m
et
ho
d,
. T
he
a
ppro
xim
at
ion
s
of cu
r
va
ture
a
re as
foll
ow
s
≜
1
2
(8)
≜
2
sin
c
os
2
(9)
We
a
da
pt
the
three
feat
ur
e
s
,
a
nd
in
to
on
e
feat
ure
vect
or
,
=
<
,
,
>
.
The
feat
ur
e
of
an
d
is
a
form
of
inte
gr
at
io
n
wh
ic
h
only
rely
on
area
cal
cu
la
ti
on
;
th
us
,
t
he
se
featur
e
s
a
r
e
rob
us
t an
d
yi
el
d
sm
oo
th
r
es
ul
ts.
The
co
m
pari
so
n i
s d
isc
us
s
ed
in
the
ne
xt s
ect
ion
.
2.4.
Clas
sific
at
i
on
Be
fore
the
e
va
luati
on
of
th
e
pro
po
s
ed
m
et
hods
,
one
-
w
ay
analy
sis
of
va
riance
(
A
NOV
A)
is
e
m
plo
ye
d
to
determ
ine
the
sta
ti
sti
cal
l
y
sign
ific
a
nt
val
ue
s
bet
ween
norm
al
and
K
C
corneal
c
urvatu
res.
ANO
VA
is
a
sta
ti
sti
ca
l
m
et
ho
d
us
e
d
to
fin
d
the
opti
m
u
m
featur
e
an
d
th
e
existe
nce
of
sign
ific
a
nt
dif
f
eren
ce
betwee
n
set
s.
The
res
ults
s
how
t
hat
on
ly
1
=
<
>
f
or
the
le
ft
ey
e
is
no
t
sig
nificant
(
p
=
0.8
84),
as
bolde
d
in
Table
1,
whereas
that
f
or
r
igh
t
ey
e
is
sig
nificant
(
p
<
0.0
5)
for
al
l
featu
res.
T
hus,
al
l
f
eat
ur
es
a
re
fe
d
into
Suppor
t
V
ect
or Mac
hin
e
(
S
V
M)
cl
assifi
er.
The
cl
assifi
cat
ion
of
co
r
neal
curvatu
re
is
te
ste
d
us
i
ng
S
VM
an
d
decisi
on
tree
s.
S
VM
is
a
ty
pe
of
patte
rn
cl
assifi
er
ba
sed
on
a
novel
sta
ti
sti
cal
le
arn
in
g
te
chn
i
qu
e
[
21]
.
This
ste
p
is
c
ru
ci
al
to
analy
se
the
curvatu
re
ei
ther
norm
al
or
KC.
More
over
,
this
ste
p
will
evaluate
the
disti
nctivenes
s
of
the
ap
proache
d
al
gorithm
.
Table
1.
A
NOVA (
p
-
value
)
Re
su
lt
s
f
ro
m
Each F
eat
ur
e
Featu
res
Rig
h
t
Left
1
<
>
1
.64
E
-
02
8
.84
E
-
01
2
<
>
3
.60
E
-
07
3
.40
E
-
07
3
<
>
4
.80
E
-
08
1
.30
E
-
06
Thr
ee
sta
ndal
one
feat
ur
es
1
=
<
>
,
2
=
<
>
,
3
=
<
>
and
f
our
c
om
bi
ned
featu
re
s
vecto
r
12
=
<
,
>
,
13
=
<
,
>
,
23
=
<
,
>
,
=
<
,
,
>
are
use
d
t
o
cl
assif
y
the
corneal
cu
rv
at
ur
e.
T
raini
n
g
im
ages
are
fed
int
o
the
SV
M
cl
assifi
e
r
for
trai
ni
ng.
The
trai
ne
d
cl
assifi
er
pr
e
dicts
the
te
sti
ng
database
without
kn
ow
i
ng
t
he
cl
ass
of
the
i
m
age.
Ke
rn
el
-
base
d
f
un
ct
ion
s
incl
ud
e
li
near
kernel functi
on
, p
olyn
om
ia
l
k
ern
el
fun
ct
io
n,
Gau
s
sia
n
Ke
rnel
an
d
Ra
dia
l B
asi
s Fu
nctio
n (RB
F)
.
A
fter sev
eral
te
sti
ng
an
d
ex
per
im
ental
on
the
trai
ning
data
us
in
g
po
l
ynom
ia
l
ker
ne
l
fu
nctio
n,
c
ubic
kernel
f
un
ct
ion
is
sel
ect
ed
in
S
VM
for
cl
assi
ficat
ion
w
he
re
the
input
dat
a
are
trans
for
m
ed
into
hi
gh
dim
ension
al
sp
ace
to
beco
m
e d
is
cret
e com
par
ed
w
i
th the o
rigin
al
sp
ace.
Κ
(
,
)
=
(
.
+
1
)
(10)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
13
, N
o.
3
,
Ma
rc
h 201
9
:
1
1
9
1
–
1
1
9
8
1196
wh
e
re
p
is a
tu
nab
le
pa
ram
et
e
r; for
cubic
kernel,
p
=3
.
A
decisi
on
tre
e
is
an
arr
ay
of
“i
f
-
the
n
-
el
se”
ru
le
s
f
or
al
loc
at
ion
cl
ass
la
be
l
to
exem
plif
y
the
data
set
in
the
f
or
m
of
t
ree
str
ucture
[26
]
-
[
27]
.
A
decisi
on
nod
e
c
ons
ist
s
of
tw
o
or
m
or
e
br
anc
hes
,
wh
e
reas
a
le
af
node
represe
nts
a
de
ci
sion
.
T
he
proces
s
of
de
ve
lop
in
g
a
tree
involve
s
th
ree
ste
ps
:
(a
)
s
plit
ti
ng
t
he
data
s
et
int
o
su
bse
ts,
wh
e
re
4
nu
m
ber
of
sp
li
ts
are
m
ade;
(b
)
pru
ning
is
trun
c
at
ing
t
he
bra
nch
e
s
an
d
(c
)
tree
sel
ec
ti
ng
is
fin
ding
the
sm
al
le
st
tree
that
yi
el
ds
the
lowe
st
cro
ss
-
valid
at
ed
error.
T
he
deep
e
r
the
tre
e,
the
fitt
er
the
m
od
el
with
m
or
e
co
m
plex
of
decisi
on
ru
le
s
is
.
H
ow
e
ve
r,
the
tre
e
m
us
t
pr
une
t
he
data
well
;
ot
herwise,
it
m
i
gh
t
be
ov
e
r
-
fitt
ing
,
w
hich
le
ad
s
into
false
cl
assifi
cat
ion
.
We
us
e
si
m
ple
decisi
on
tree,
a
nd
the
perform
ance
r
esults
are
discuss
e
d
i
n
the
n
e
xt secti
on.
3.
RESU
LT
S
AND DI
SCUS
S
ION
The
local
data
set
con
sist
in
g
of
218
im
ages
is
us
ed
to
te
st
the
reli
abili
ty
o
f
the
pro
pose
d
al
gorith
m
us
in
g
f
our
m
ain
ste
ps
as
e
xp
l
ai
ned
in
Me
th
odology
sect
io
n.
T
he
norm
al
and
KC
A
SPI
s
include
t
he
le
ft
a
nd
rig
ht
ey
es
capt
ur
e
d
from
the
side
vie
w.
For
these
side
-
vie
w
im
ages,
the
corneal
c
urvatur
e
is
segm
ented
us
in
g
sp
li
ne
functi
on
to
m
easur
e
th
e
corneal
c
urv
at
ur
e
before
si
gn
i
ficant
feat
ures
are
ext
racted.
Fig
ur
es
6
and
7
pr
ese
nt
t
he
re
s
ults
of
218
im
ages,
both
ey
es,
cl
assifi
e
d
usi
ng
SV
M
an
d
decisi
on
tree
.
Both
cl
assi
fier
s
are
app
li
ed
in
this
work
us
i
ng
the
sta
nd
al
on
e
an
d
com
bin
at
io
ns
featu
res,
<
,
,
>
:
nonlinear
e
stim
ation,
cro
ss
over
poi
nt
an
d
tri
gono
m
et
rical
m
et
h
od
s
.
T
he
S
V
M
cl
assifi
er
pe
rfor
m
s
well
with
t
he
tw
o
-
com
bin
ed
featur
e
s
12
=
<
,
>
fo
r
bo
t
h
ey
es
wit
h
ave
ra
ge
acc
ur
acy
of
98.
4%
(F
ig
ure
6)
com
par
ed
wi
th
sta
nd
al
on
e
feat
ur
e
1
=
<
>
an
d
2
=
<
>
wh
ic
h
is
only
belo
w
tha
n
90%
.
T
he
wea
kest
c
om
bin
at
ion
is
23
=
<
,
>
f
or
rig
ht
ey
e
with
72.
6%
and
f
or
le
ft
ey
e
is
65.
4%.
O
ve
rall
,
the
SV
M
achiev
es
99.5%
accuracy
f
or
both
ey
es;
w
he
reas
the
decisi
on
tree
only
at
ta
ins
96.
3%
a
nd
83.
8%
f
or
rig
ht
an
d
le
ft
ey
es,
resp
ect
ively
,
w
it
h
the
c
om
bin
at
ion
of
al
l
fea
tures
=
<
,
,
>
.
Hen
ce
,
the
c
orneal
cu
r
vatu
re
m
et
hod
is reli
able a
nd
com
petent to c
la
ssify t
he KC
and no
n
-
KC ey
e u
si
ng S
VM c
la
ssifie
r.
Figure
6. Perce
ntage o
f
acc
ur
a
cy
f
or eac
h
a
nd co
m
bin
at
ion f
eat
ur
es cla
ssi
fied usin
g SVM
Figure
7. Perce
ntage o
f
acc
ur
a
cy
f
or eac
h
a
nd co
m
bin
at
ion f
eat
ur
es cla
ssi
fied usin
g decisi
on tree
90
7
3
.
7
7
3
.
2
9
8
.
4
9
8
.
4
7
2
.
6
9
9
.
5
8
1
.
5
6
3
.
2
6
6
.
5
9
8
.
4
9
5
.
7
6
5
.
4
9
9
.
5
0
20
40
60
80
1
0
0
1
2
0
Cnl
Cc
p
Ct
r
Cnl
+
Ccp
Cnl
+
Ct
r
Cc
p + Ctr
Cnl
+
Ccp +
Ct
r
Rig
ht
Left
9
1
.
1
6
6
.
3
6
5
.
3
9
6
.
3
9
5
.
3
6
3
.
7
9
6
.
3
7
4
.
6
6
5
.
9
6
4
.
9
8
2
.
7
80
6
3
.
2
8
3
.
8
0
20
40
60
80
1
0
0
1
2
0
Cnl
Cc
p
Ct
r
Cnl
+
Ccp
Cnl
+
Ct
r
Cc
p + Ctr
Cnl
+
Ccp +
Ct
r
Rig
ht
Left
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Detect
ion
of ke
ra
toc
onus i
n a
nterior
seg
me
nt
p
hoto
gra
ph
e
d i
mages
us
i
ng
cor
nea
l
.
..
(
M
ari
zu
an
a
M
at Da
ud
)
1197
The
sen
sit
ivit
y
of
the
appr
oach
m
et
ho
d
is
99
%,
a
nd
t
he
sp
eci
fici
ty
is
10
0%
.
I
n
this
co
ntext,
sensiti
vity
ind
i
cat
es
that
the
al
gorithm
can
cl
assify
the
K
C
disease
in
KC
gr
oup;
w
he
reas
the
sp
ec
ific
it
y
sp
eci
fies
t
hat
t
he
syst
em
is
com
petent
to
cl
assify
the
no
n
-
KC
disease
in
norm
al
gr
ou
p.
Howe
ver,
on
ly
on
e
case
of
KC
cl
a
ssifie
d
in
norm
al
group
is
i
de
ntifie
d
as
false
ne
gative.
T
he
ov
e
rall
pe
rform
ance
of
this
syst
e
m
is reli
able in
d
e
te
ct
ing
K
C
eye
.
4.
CONCL
US
I
O
N
A
new
detect
io
n
m
et
ho
d
f
or
KC
wa
s
propo
sed
us
in
g
l
ocal
databa
se
a
nd
f
our
ne
w
m
od
ul
es,
nam
ely,
pr
e
processi
ng,
featur
e
e
xtracti
on,
segm
entat
ion
a
nd
cl
assifi
cat
ion
.
AS
P
I
was
ca
ptu
re
d
f
ro
m
side
view
of
a
n
ey
e
i
m
age,
and
the
co
r
neal
curvatu
re
was
cal
culat
ed
us
i
ng
the
th
ree
pa
ram
et
ers
of
te
m
pla
te
disc
m
et
hod,
nam
ely,
nonlin
ear,
c
ro
s
sover
po
i
nt
an
d
tri
gon
om
et
rical
m
eth
od.
The
feat
ure
vect
or,
=
<
,
,
>
was
then
fed
i
nto
S
VM
to
cl
assify
the
ey
e
i
m
ages
with
acc
ur
ac
y
of
99.
5%
,
se
ns
it
ivit
y
of
99
%
an
d
sp
eci
fici
ty
of
100%
.
This
syst
e
m
m
a
y
pr
ov
ide
a
scree
ning
platf
orm
fo
r
KC
detect
io
n
cases
nam
el
y
people
li
ving
i
n
r
ur
al
area,
w
he
re
op
hth
al
m
olo
gists
are
hard
to
be
reache
d.
I
n
li
ne
with
that
pur
pose,
a
n
aut
om
at
ed
seg
m
e
ntati
on
appr
oach
sho
ul
d
be
de
velo
pe
d
to
pro
duce
a
us
e
r
-
fr
ie
nd
l
y
syst
e
m
in
assist
ing
e
xp
e
rt
s
duri
ng
t
he
c
orneal
exam
inati
on
.
ACKN
OWL
E
DGE
MENTS
The
aut
hors
wi
sh
to
tha
nk
the
oth
er
m
e
m
ber
s
of
the
Ce
nte
r
fo
r
In
te
gr
at
e
d
Syst
e
m
s
Eng
in
eerin
g
an
d
Adva
nced
Tec
hnologies
(
I
N
TEGR
A)
UKM
fo
r
t
heir
s
uppo
rt.
Fun
ding
:
This
w
ork
was
s
upported
by
the
Mi
nistry of Hi
gh
e
r
E
ducat
io
n, Ma
la
ysi
a w
it
h gr
a
nt
no.
FR
GS
/1/
2016/IC
T0
1/UKM
/02
/
4.
REFERE
NCE
S
[1]
Galvi
s V
.
,
e
t
a
l.
,
“
Kera
toc
onus:
a
n
inflam
m
at
or
y
disorder
?
”
E
ye
,
vol/
issue:
29(7)
,
pp.
843
–
59
,
201
5
.
[2]
Gatz
iouf
as
Z
.
,
et
al
.
,
“
Kera
to
c
onus:
i
s
it
a
N
on
-
infl
amm
at
or
y
Disea
se
?
”
M
e
dic
al
Hypo
the
si
s,
Discov
ery
&
Innov
ati
on
Opht
halmology
Journal
,
vo
l/
issue:
6(1
)
,
pp
.
6
–
7
,
2017
.
[3]
Xu L
.
,
et
al
.
,
“
Preva
l
enc
e and
associations
of
steep c
orne
a/
Ke
rat
o
conus
in
Grea
te
r
Bei
ji
ng
.
The
Be
ij
ing
E
y
e
Stud
y
,”
PLoS
One
,
vol/is
sue:
7(7)
,
2012
.
[4]
Davidson
A
.
E
.
,
et
a
l.
,
“
The path
ogene
s
is of
ker
a
t
oconus
,”
Eye
(
Lond)
,
vol/
issue:
28(2)
,
pp
.
189
–
95
,
2014
.
[5]
Sharif
R
.
,
et
al
.
,
“
Pathoge
nesis
of
Kera
toc
onus:
T
he
int
rigu
ing
therape
ut
ic
pot
ent
i
a
l
of
Prolac
t
in
-
in
duci
bl
e
protein
,”
Prog
Retin Eye
Re
s
.
,
pp.
1
–
17
,
2
018
.
[6]
Carl
son A
.
N.
,
“
Kera
toc
onus
,”
O
phthal
mology
,
v
ol/
issue:
1
16(10)
,
pp
.
2036
–
7
,
20
09
.
[7]
Kenne
d
y
R
.
H
.
,
et
al
.
,
“
A
48
-
y
e
ar
clinical
and
e
pide
m
iol
ogi
c
stu
d
y
of
k
era
to
con
us
,”
Am
J
Opht
halmol
,
vol
/i
ss
ue:
101(3)
,
pp
.
267
–
73
,
1986
.
[8]
Hashem
i
H
.
,
et
al
.
,
“
High
pre
va
le
nc
e
and
famil
i
al
aggr
egation
o
f
ker
a
toc
onus
in
an
Ir
ani
an
rura
l
popul
a
ti
on:
A
popula
ti
on
-
b
ase
d
stud
y
,”
Ophth
almic
Phy
siol
O
pt
.
,
pp.
1
–
9
,
201
8
.
[9]
U.
Ta
nab
e,
et
a
l.
,
“
Preva
l
ence
o
f
ker
at
oconus
pat
i
ent
s
in
Japa
n
,”
Nihon
Gank
a
Gakk
ai
Zas
shi
,
vol/
issue:
85(3
)
,
pp.
407
,
1985
.
[10]
Mohd
Ali
B
.
,
et
al.
,
“
Cli
ni
ca
l
ch
a
rac
t
eri
sti
cs
of
ker
at
oconus
pat
i
en
ts
in
Malay
si
a:
A
rev
ie
w
from
a
cor
nea
spec
i
al
is
t
ce
ntr
e
,”
J
Optom
,
vol
/i
ss
ue:
5(1)
,
pp.
38
–
42
,
2012
.
[11]
Gokul
A
.
,
et
al.
,
“
Advanc
ed
ant
e
rior
segm
ent
i
ma
ging
in
ker
a
toconus
:
a
rev
ie
w
,”
Cli
n
Ex
p
Ophthalmol
,
vol/
issue
:
46(2)
,
pp
.
122
–
32
,
2018
.
[12]
Kok Y
.
O
.
,
e
t
a
l.
,
“
Review:
ke
ratoconus i
n
As
ia
,”
Cornea
,
vol/is
sue:
31(5)
,
pp
.
58
1
–
93
,
2012
.
[13]
Nabha
n
T
.
I.
,
“
S
y
stem
and
m
et
hod
for
ophtha
l
m
ologi
ca
l
imagi
ng
ada
pte
d
to
a
m
obil
e
proc
essing
devi
c
e
,”
US
2018/0092534
(
Pate
nt)
,
2018
.
[14]
Us
m
an
M
.
,
et
a
l.
,
“
Com
pute
r
Vision
Te
chn
iq
ues
Applie
d
for
Diagnosti
c
An
aly
s
is
of
Retina
l
OCT
Im
age
s:
A
Revi
ew
,”
Arch Com
put
Me
thod
s E
ng
.
,
pp
.
1
–
17
,
2016
.
[15]
Tout
ounchian
F
.
,
et
al.
,
“
Dete
cti
on
o
f
Kera
toc
on
us
and
Su
spec
t
Kera
toc
onus
b
y
Mac
hine
Vision
,”
Inte
rnational
Mult
ic
on
fe
ren
ce
of
Eng
ine
ers an
d
Computer
Sc
ientists.
Hong Kon
g
,
vol
.
1
,
pp
.
14
–
6
,
2012
.
[16]
Arbel
aez
M
.
C
.
,
et
al.
,
“
Us
e
of
a
support
vec
tor
m
ac
hine
for
ker
at
oconus
and
su
bcl
inica
l
ker
a
toconus
det
ec
t
ion
b
y
t
opogra
phi
c
a
nd
tomographic
d
ata
,”
Ophthalmolo
gy
,
vo
l/
issue:
11
9(11)
,
pp
.
2231
–
8
,
2012
.
[17]
Mathe
w
M
.
K
.
,
et
al.
,
“
Vari
ous
Cat
ar
ac
t
De
te
c
tion
Methods
-
a
Surve
y
,”
In
t
Re
s
J
Eng
Technol
.
,
vol/
issue:
4(1)
,
pp.
1517
–
9
,
201
7
.
[18]
Fuadah
Y
.
N
.
,
e
t
al.
,
“
Mobile
Cat
ar
ac
t
De
tecti
o
n
using
Optimal
Com
bina
ti
on
of
Stat
isti
ca
l
Te
xt
ure
Anal
y
s
is
,”
I
n
t
Conf
Instrum
entation, Commun
,
Inf
Techno
l Bi
o
med
Eng. B
andu
ng
,
pp
.
232
–
6
,
2
015
.
[19]
Li
H
.
,
et
al
.
,
“
I
m
age
base
d
gra
ding
of
nuclea
r
ca
t
aract
b
y
SV
M
reg
ression
,”
SPIE
Me
d
Imaging
,
vol
.
6915
,
pp.
691536
-
691
536
–
8
,
2008
.
[20]
Patwar
i
A
.
U.
,
“
Dete
c
ti
on,
C
at
eg
oriz
a
ti
on,
and
As
sessment
of
E
y
e
Cat
ar
acts
Us
ing
Digit
al
Im
age
Proce
ss
ing
,”
The
Fi
rs
t
Int
ernati
on
al
Conf
ere
nce o
n
Inte
rdisc
ipl
ina
ry
Re
search
and
Dev
el
opmen
t. Thail
and
,
pp.
1
–
5
,
2011
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
13
, N
o.
3
,
Ma
rc
h 201
9
:
1
1
9
1
–
1
1
9
8
1198
[21]
Na
y
ak
J.
,
“
Autom
at
ed
class
ifi
c
ation
of
norm
al
,
c
at
ar
ac
t
and
post
ca
t
aract
opt
ical
e
y
e
images
using
SV
M
cl
assifie
r
,”
Proc
World
Con
gr E
ng
.
,
vol
.
1
,
p
p.
23
–
5
,
2013
.
[22]
W
an
Za
ki
W
.
M
.
D
.
,
et
al
.
,
“
Autom
at
ed
pte
r
y
gium
detec
t
ion
m
et
hod
of
ant
e
r
ior
segm
ent
photogra
phed
ima
ges
,”
Comput
Me
thod
s P
rogr
ams
Bi
omed
,
vol
.
154
,
p
p.
71
–
8
,
2018
.
[23]
Agham
ohamadia
n
S
.
M
.
,
et
a
l.
,
“
A
Novel
Curvat
ure
B
ase
d
Alg
orit
hm
for
Autom
at
ic
Gradi
ng
o
f
Ret
in
al
Blood
Vess
el
Tort
uosi
t
y
,”
I
EEE
J
B
iomed
Heal
Informa
ti
cs
,
vol/is
sue:
2
0(2)
,
pp
.
586
-
59
5
,
2015
.
[24]
Mee
k
K
.
M
.
and
Knupp C.
,
“
Cor
nea
l
struct
ur
e an
d
tra
nspar
ency
,”
Prog
Retin Eye
Re
s
.
,
vol
.
49
,
pp
.
1
–
16
,
2015
.
[25]
T
.
F.
Ban
chof
f
and
S
.
T.
Lov
ett
,
“
Diffe
r
ent
i
al
geometr
y
of
cur
ves
and
surf
ac
es
,”
First
Edition
,
New
York:
A
K
Pete
rs/CRC Pres
s
,
2010.
[26]
Twa
M
.
D
.
,
et
al
.,
“
Autom
at
ed
dec
ision
tr
ee
class
ifi
c
at
ion
of
cor
n
ea
l
shape
,”
Opto
metry
and
vi
sion
scie
nce
:
of
fi
c
ial
publi
cation
o
f
th
e
Ame
ri
can Ac
a
demy
of
Optome
try
,
vo
l/
issue:
82
(12)
,
pp
.
1038
,
2
005
.
[27]
Twa
M
.
D
.
,
et
al
.,
“
Dec
ision
tr
ee
cl
assifi
cation
of
spat
ia
l
da
ta
pat
t
ern
s
from
vide
oker
at
og
rap
h
y
using
Z
ern
ik
e
pol
y
nom
ials
,”
P
roce
edi
ngs In
te
r
nati
onal
Con
fe
re
nce
on
Data
M
in
ing.
San
Franc
isco
,
pp
.
3
-
12
,
200
3
.
BIOGR
AP
HI
ES OF
A
UTH
ORS
Mariz
uan
a
Mat
Daud
is
a
PhD
ca
ndid
at
e
at
Univer
siti
Keb
a
ngsaa
n
Mal
a
y
s
i
a.
Her
rese
arc
h
int
er
ests
include
image
pro
ce
ss
i
ng
with
m
ac
h
in
e
l
ea
rning
foc
us
ing
on
biomed
ical
f
ie
ld
.
More
spec
ifica
l
l
y
,
h
er
work
exa
m
ine
s
how
image
proc
essing
with
m
ac
hine
learni
ng
te
c
hnique
s
coul
d
cont
ribute
towa
rds
bet
te
r
and
m
ore
eff
ic
i
ent
wa
y
s
of
an
aly
z
i
ng
and
under
st
andi
ng
m
edi
c
al
images.
Apar
t
fr
om
tha
t,
her
rese
arc
h
int
er
est
al
so
inc
lud
es
assistiv
e
t
ec
hnolog
ie
s
f
or
people
with
disabi
litie
s
.
W
an
Mim
i
Di
y
a
na
W
an
Za
k
i
ob
ta
in
ed
her
B
ac
h
el
or
degr
ee
(
El
e
ct
roni
cs
Engi
ne
e
ring)
in
2000,
Master
degr
ee
(
Engr.
Sc.
)
in
2005
and
PhD
degr
ee
in
2012,
al
l
f
rom
the
Multi
me
dia
Univer
si
t
y
(MM
U),
C
y
ber
j
a
y
a,
Malay
si
a.
She
is
cu
rre
nt
l
y
a
rese
ar
che
r
and
senior
l
ec
tur
er
a
t
the
Centre
of
Inte
gra
te
d
S
y
st
e
m
s
Engi
ne
eri
ng
and
Advanc
ed
T
ec
hnolog
y
(INT
EGRA),
Univer
siti
Keba
ngsa
an
Malay
s
ia
(UK
M),
which
she
joi
ned
in
200
8.
Her
rese
a
rch
spec
ialisation
i
s
in
biomedic
al
engi
ne
eri
ng,
an
d
her
rese
arc
h
i
nte
rests
include
int
el
l
ige
n
t
s
y
st
ems
,
image
proc
essing
and
IoT
rel
a
te
d
h
eal
thcar
e
t
ec
hnolog
y
.
Aini
Hus
sain
o
bta
in
ed
her
B.
Sc.
in
Elec
tr
ical
Engi
ne
eri
ng
fro
m
Loui
siana
Stat
e
Univer
si
t
y
(LSU),
Bat
on
R
ouge,
US
A;
M.
Sc.
in
S
y
st
ems
and
Contro
l
fro
m
the
Unive
rsit
y
o
f
Manc
h
este
r
Instit
ute
of
Science
and
Techno
log
y
(UM
IST),
Manc
heste
r
,
U.
K.,
and
Ph.D.
in
El
ec
t
rical
and
El
e
ct
roni
c
Eng
i
nee
ring
from
th
e
Nati
on
al
Uni
ver
sit
y
of
Ma
lay
sia
in
1985
,
1
991
and
1997
,
respe
ctively
.
Sh
e
is
a
Profess
or
and
cur
r
entl
y
,
th
e
Cha
ir
of
the
I
NTEGRA
rese
ar
ch
c
enter
a
lso
known
as
“
Cent
re
for
Int
egr
at
ed
S
y
s
te
m
s
Engi
ne
eri
ng
an
d
Advanc
ed
T
ec
hnolo
g
ie
s”.
Her
rese
ar
ch,
fo
r
which
she
has
rec
e
ive
d
fund
i
ng,
foc
uses
on
Inte
lligen
t
S
y
st
e
m
s
and
Im
ag
e
Proce
ss
ing.
Her
cur
ren
t
r
ese
arc
h
int
er
ests
are
in
m
ac
hine
learni
n
g,
pat
t
ern
rec
og
nit
ion
and
vid
eo
&
image
proc
essing.
Hali
z
a
Abdul
M
uta
li
b
is
an
As
sociate
Profess
or
in
Cen
tre
for
Com
m
unit
y
Hea
lt
h,
Facult
y
of
Hea
lt
h
Sc
ie
n
ce
s,
Nati
ona
l
Unive
rsit
y
of
Mal
a
y
s
i
a
since
y
e
ar
20
08.
She
spec
iali
ze
s
in
Con
ta
c
t
Le
nses
&
Corne
al
Morpholog
y
.
Her
cur
ren
t
r
ese
arc
h
intere
sts
ar
e
on
m
orphologi
ca
l
ch
ange
s
in
c
ontact
l
ens
wea
r,
observa
t
ion
of
ph
y
siolog
ic
a
l
c
orne
al
ce
l
l
cha
n
ges
using
conf
oc
al
m
ic
roscop
y
,
the
rap
eut
i
c
contac
t
l
ense
s,
contac
t
l
ens
and
cl
e
ani
ng
reg
ime
ef
fec
t
to
ocu
la
r
st
at
us
and
othe
r
topi
cs
relate
d
to
cont
act
le
nses.
She
was
conf
err
ed
MS
c
in
Ophthal
m
olog
y
&
Vi
sion
Scie
nce
s
(1997)
and
a
P
hD
(Optom
et
r
y
)
from
UM
IST,
Unite
d
Kingdo
m
in
2000.
She
m
ana
ged
to
complet
e
her
PhD
in
23
m
onths
and
was
awa
rde
d
with
VD
C
Pet
er
Abel
Aw
ard
(
Germ
an
y
)
fo
r
Best
Th
esis
in
C
onta
c
t
L
ens
Res
ea
rch
.
Evaluation Warning : The document was created with Spire.PDF for Python.