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.
933
~
944
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijee
cs
.v1
3
.i
3
.pp
933
-
944
933
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Identific
atio
n
of
Plasmo
diu
m fa
l
ciparu
m
and
Plasmo
dium viv
ax
on
d
igit
al
i
mage o
f
t
hin
b
lood
f
ilm
s
Ha
n
un
g Adi
Nugro
ho
1
,
Made S
at
ri
a
Wib
awa
2
, N
oor Akhma
d S
e
tiaw
an
3
,
E. El
sa
Herdi
ana
Murhan
d
arwa
ti
4
, Ra
tna Les
ta
ri
Bu
diani Bu
ana
5
1,3
Depa
rtment
of
Elec
tr
ical
Engi
n
ee
ring
and
Infor
m
at
ion
T
ec
hnolo
g
y
,
Facu
lty
of E
ngine
er
ing,
Univer
sita
s Gad
j
ah
Mada
,
Yog
y
a
kar
ta,
Indon
esia.
2
Depa
rtment of I
nform
at
ion
S
y
s
t
em,
Facu
lty
of
E
ngine
er
ing, STI
KO
M,
Bal
i
,
Ind
onesia
.
4
Depa
rtment of
Para
sitol
og
y
,
Fa
cul
t
y
of
Med
ic
in
e,
Univ
ersit
as
Gadja
h
Mada
,
Yo
g
y
ak
arta, Indones
ia
.
5
Diploma
3
of
Medical
R
ec
ord
,
V
oca
t
iona
l
Coll
eg
e,
Univ
ersitas Gadj
ah
Mada
,
Yo
g
y
ak
arta, Indones
ia
.
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Sep
24
, 201
8
Re
vised
N
ov
25
, 2
018
Accepte
d
D
ec
1
7
, 201
8
Obs
erv
ing
pre
senc
e
of
Plasm
odium
par
asit
e
of
stai
ned
th
ic
k
or
thi
n
blood
fil
m
s
through
m
ic
roscopic
ex
aminat
ion
is
a
gold
standa
rd
for
m
al
ari
a
dia
gnosis.
Al
tho
ugh
the
m
ic
rosc
opic
exa
m
inatio
n
has
be
en
exten
sively
used
,
m
iside
nti
ficat
ion
m
ight
occ
ur
ca
u
sed
b
y
hum
an
fa
ct
ors.
In
orde
r
to
over
come
m
iside
nti
ficat
ion
proble
m
,
seve
ra
l
studie
s
have
bee
n
conducte
d
t
o
deve
lop
a
computer
-
ai
d
ed
m
al
ari
a
d
ia
gnosi
s
(CAD
x)
to
ass
ist
par
amedi
cs
i
n
dec
ision
-
m
aki
ng.
Thi
s
st
ud
y
proposes
a
n
appr
oac
h
to
i
dent
if
y
spe
ci
es
and
stage
of
Plasm
o
dium
fal
ci
par
um
and
Plasm
odium
viva
x
o
n
thi
n
blood
film
s
col
le
ct
e
d
from
the
La
bora
tor
y
of
Para
si
tol
og
y
,
Facult
y
of
Med
ic
ine
,
Univer
sita
s
Gadja
h
Mada
.
Adapti
ve
k
-
m
ea
ns
cl
ustering
is
appl
ie
d
to
segm
ent
Plasm
odium
par
asit
es
.
A
total
of
39
fea
ture
s
con
sisting
of
shape
and
te
xtu
r
e
fea
tur
es
are
ext
r
ac
t
ed
and
the
n
s
el
e
ct
ed
b
y
using
wrappe
r
-
base
d
f
orward
and
bac
kw
ard
direct
ions.
Cl
assific
a
t
ion
is
eva
lu
ated
in
two
sche
m
e
s.
The
first
sche
m
e
is
to
class
if
y
th
e
spec
i
es
of
par
asite
i
nto
two
cl
asses.
The
sec
ond
sche
m
e
is
to
cla
ss
if
y
th
e
spe
ci
es
and
stag
e
of
pa
rasit
e
int
o
six
class
es.
Three
cl
assifi
ers
appli
ed
ar
e
k
-
ne
are
st
nei
ghbour
(KN
N),
support
ve
ctor
m
ac
hine
(SV
M)
an
d
m
u
lt
i
-
l
a
y
er
p
erceptron
(MLP).
Furthermore,
to
f
a
ci
litate
th
e
m
ult
ic
la
ss
cl
assi
fic
a
ti
on,
one
-
v
er
sus
-
on
e
(OV
O)
and
one
-
v
ersus
-
all
(OV
A)
m
et
hods
are
implemente
d
.
Th
e
f
i
rst
sche
m
e
a
chi
e
ves
the
accurac
y
of
88.
70%
base
d
on
MLP
cl
assifi
er
using
thre
e
se
lecte
d
f
e
at
ure
s.
W
hi
le
th
e
accur
a
c
y
gai
ned
b
y
th
e
se
cond
sche
m
e
is
95.
16%
bas
ed
o
n
OV
O
and
ML
P
cl
assifi
er
using
29
sele
cte
d
fea
tur
es.
The
s
e
result
s
indica
t
e
tha
t
the
propos
ed
appr
oa
ch
succ
essfull
y
id
e
nti
fie
s
th
e
spec
i
es
and
stage
of
par
asite
on
thi
n
blood
fil
m
s
and
has
pote
nt
i
al
to
be
imple
m
ent
ed
in
the
CAD
x
sy
stem
f
or
assisting
par
amedic
s
in
d
i
agnosing
m
al
ar
i
a.
K
eyw
or
ds:
Ad
a
ptive
K
-
Me
ans
Cl
ust
erin
g
Plasmo
dium
fa
lc
ipa
r
um
Plasmo
dium
vi
vax
Wr
a
pper
Feat
ure Sele
ct
io
n
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
:
Hanu
ng Adi
N
ugr
oho,
Dep
a
rtm
ent o
f El
ect
rical
En
gi
neer
i
ng and
Inf
or
m
at
ion
Tec
hnol
og
y,
Faculty
of E
ngineerin
g, U
nive
rsita
s G
a
djah
Ma
da,
Jl. Grafi
ka 2
K
a
m
pu
s
UG
M
, Yo
gyaka
rta
55281, I
ndonesi
a.
Em
a
il
:
adinu
gr
oho
@ugm
.ac.id
1.
INTROD
U
CTION
Ma
la
ria
is
on
e
of
the
global
diseases
re
po
rted
by
the
W
or
l
d
He
al
th
O
rg
a
nisati
on
[
1]
.
Ba
sed
on
World
Ma
la
ria
Re
po
rt
20
16
,
there
we
re
21
2
m
illi
on
cases
fo
un
d
in
2015
wh
ic
h
cause
d
429,0
00
of
dea
ths
[2]
.
In
I
nd
on
esi
a,
m
al
aria
is
per
sist
s
with
highe
r
end
em
ic
ity
in
east
ern
par
t,
s
uc
h
as
in
Pa
pu
a
wit
h
hi
gh
est
num
ber
of
m
al
aria
annual
par
asi
te
in
c
idence
was
31.
93
m
al
aria
infe
ct
ion
per
1,000
popula
ti
on
in
2015
[3]
.
Ma
la
ria
is
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
:
933
–
944
934
cause
d
by
parasi
te
s
Plas
m
odium
par
asi
te
s
wh
ic
h
are
m
e
diate
d
by
fem
al
e
Ano
ph
el
es
m
os
qu
it
os
.
T
her
e
a
re
sever
al
s
pecies
of
Plasm
od
ium
na
m
e
ly
Pl
asm
od
ium
falci
parum
(P
.
falci
par
um
),
Plasm
od
ium
viv
ax
(P.
viv
a
x),
Plasm
od
i
um
m
a
la
riae
(P
.
m
al
ari
ae
),
Plasm
od
ium
oval
e
(P.
oval
e)
an
d
the
la
st
is
Plasm
o
diu
m
knowle
si
(P
.
knowle
si).
In
2013,
56%
of
m
al
aria
cases
i
n
I
ndonesi
a
w
as
cause
d
by
P
la
s
m
od
ium
fal
ci
parum
and 44%
w
as
c
ause
d by Plas
m
od
iu
m
v
iva
x
[4]
.
Althou
gh
m
al
aria
is
li
fe
threa
te
nin
g,
a
n
earl
y
detect
ion
c
ould
sa
ve
li
ves
.
Ba
sed
on
guid
el
ines
f
or
t
he
treatm
ent
of
m
al
aria,
m
ic
ro
scop
ic
e
xam
inatio
n
or
rap
i
d
diagnostic
te
st
(R
DT)
is
re
quire
d
to
co
nf
i
rm
m
al
a
ria
infecti
on
[
5]
. I
n Ind
on
esi
a,
m
ic
ro
sc
op
ic
e
xa
m
inati
on
is c
onside
red as
gold sta
nd
a
rd for
m
al
aria diagno
sis.
Ma
nu
al
ly
,
the
m
ic
ro
sco
pic
exam
inati
on
co
ns
ist
s
of
three
ste
ps
i.e.
by:
1)
obser
ving
the
pr
ese
nce
of
m
al
aria
par
asi
te
/Pl
as
m
od
ium
,
2)
ide
ntifyi
ng
the
Plasm
odium
sp
eci
es
a
nd
3)
i
den
ti
fy
ing
the
Plasm
od
i
um
sta
ge.
Dete
rm
i
nation
of
t
he
infect
ing
Pl
asm
od
i
um
spe
ci
es
and
sta
ges
are
i
m
po
rtant
f
or
decidin
g
a
ppr
opriat
e
thera
py
[
6]
as
well
as
pr
e
dicti
ng the
pro
gnosi
s.
Me
dical
la
bor
at
or
y
te
ch
no
l
ogist
s
in
ei
the
r
public
healt
h
centres
or
cl
inics
or
ho
s
pital
s
carry
ou
t
m
ic
ro
sco
pic
e
xam
inati
on
.
Bl
ood
from
finger
pri
ck
is
us
e
d
an
d
sm
eared
on
obj
ect
glas
s/
sli
des,
f
ollo
wed
by
Giem
sa
sta
ining
process
a
nd
finall
y,
visu
al
isa
ti
on
under
th
e
m
ic
ro
sco
pe
.
I
n
te
rm
s
of
diag
nosti
c
accuracy,
m
ic
ro
sco
pic
exam
inati
on
has
po
te
ntial
we
akn
e
sses
that
m
igh
t
be
influe
nced
by
the
l
aborato
ry
per
s
onnel
,
abili
ty
,
extern
a
l
distract
ion
an
d
ex
per
ie
nce
be
sides
sm
ear
qu
al
it
y,
reag
ent o
r
eq
uip
m
ent
us
ed
[
7].
So
m
e
i
s
su
es
su
c
h
as
obser
va
ti
on
tim
e
and
low
-
reli
abili
ty
resu
lt
ha
ve
al
s
o
em
erg
ed
in
s
ever
al
la
borat
ori
es
in
Ce
ntral
Java,
Ind
on
esi
a.
D
ue
to
these
pro
ble
m
s,
the
exam
i
nation
accu
rac
y
var
ie
s
with
a
range
from
50
to
95%.
a
sit
ua
ti
on
that m
a
y l
ead to
a
w
ron
g
treat
m
en
t and fatal
it
y
[7]
.
The
us
e
of
c
om
pu
te
r
ai
ded
diag
nosis
(CA
Dx)
syst
em
ca
n
assist
the
pa
ram
edics
in
ob
ser
ving
the
pr
ese
nce
of
Pl
asm
od
ium
parasi
te
s
to
gen
e
r
at
e
the
qu
a
ntit
at
ive
analy
sis.
It
m
ay
inv
ol
ve
the
di
gital
i
m
age
processi
ng,
art
ific
ia
l
intel
l
igence
an
d
data
m
ining
.
Mo
re
over
,
the
syst
em
is
able
to
lear
n
f
r
om
kn
owle
dg
e
of
par
am
edics in
analy
sing t
he
c
aptu
red im
age to
inc
rease t
he a
ccur
acy
of d
ia
gnos
is
[8]
.
2.
LIT
ERATUR
E REVIE
W
The
early
sy
m
pto
m
of
m
al
ar
ia
is
no
t
sp
eci
fic
and
m
igh
t
be
m
i
m
ic
kin
g
oth
e
r
vir
us
inf
ect
ion
e.g.
head
ac
he,
fati
gue,
bo
dy
aches
and
fe
ver.
Mi
sd
ia
gn
os
is,
la
te
or
inap
pro
pr
i
at
e
anti
m
al
ari
a
therap
y
co
ul
d
le
ad
to
sever
it
y
-
com
a,
anae
m
ia
,
hypoglyc
aem
i
a,
kidney
fail
ur
e,
brai
n
dam
age
an
d
m
e
ta
boli
c
aci
do
sis
-
or
eve
n
death
[
9]
.
T
he
li
fe
cy
cl
e
of
m
al
aria
par
asi
te
s
de
velo
ps
i
n
two
dif
fere
nt
hosts,
i
n
hum
ans
an
d
i
n
f
e
m
al
e
Ano
ph
el
es
m
os
qu
it
oes
as
de
picte
d
in
Fi
gure
1.
I
n
the
hum
an
body,
the
pa
rasit
es
unde
r
go
tw
o
c
yc
le
s
,
exo
e
ryt
hrocyt
ic
and
eryt
hroc
yt
ic
cy
cl
es.
Mer
oz
oites
as
the p
r
oduct o
f
e
xo
eryt
hrocyt
ic
cy
le
,
inv
a
de
hu
m
an
re
d
blood cel
ls, tu
r
n
int
o
tr
ophozo
it
e,
m
at
ur
e an
d bec
om
e schizon
t st
a
ge.
Re
d blo
od cell
s con
ta
ining
schiz
onts are
finall
y
burst
a
nd
re
le
ase
d
ne
w
m
ero
zoite
s
tha
t
are
goin
g
t
o
i
nv
a
de
t
he
new
r
ed
bloo
d
cel
ls
.
A
fter
s
om
e
cy
cl
es
,
so
m
e
par
asi
te
s
tur
n
into
gam
et
ocyt
e
sta
ge.
Wh
e
n
a
m
os
qu
it
o
bites
hum
a
n
ha
ving
gam
e
tocy
te
s
in
their
blood,
nex
t
tra
ns
m
issio
n
occ
ur
s
a
fter
the
gam
et
ocyte
com
pleti
ng
sp
or
ogonic
cy
cl
e
and
ge
ner
at
e
s
spor
ozo
it
es
i
n
the
sal
ivary fl
uid
s
of the m
os
quit
o,
w
hich
a
re
re
ady to
be
i
nject
ed
to
anothe
r h
um
an
blood
[
10]
.
Figure
1. The
li
fe cycl
e of m
a
la
ria p
ar
asi
te
[10
]
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
Id
e
ntif
ic
ation
of
Plasmo
dium f
alciparu
m a
nd
Plasmo
dium vi
vax
on
d
i
gital
i
mage
…
(
H
anung A
di Nu
groho
)
935
Var
i
ou
s
re
sear
ch
w
orks
ha
ve
been
co
nduct
ed
to
identify
m
al
aria
par
asi
te
based
on
C
AD
syst
em
.
Me
hrjou
et
al
.
[11]
detect
ed
the
red
bloo
d
cel
ls
infected
by
m
a
la
ria
par
asi
te
on
th
e
thin
blood
f
ilm
s.
Segm
entat
ion
process
was
c
onduct
ed
on
HSV
colo
ur
m
od
el
based
on
a
dap
ti
ve
est
i
m
at
ed
thresh
ol
d
t
o
separ
at
e
pa
rasi
te
fr
om
the
back
gr
ound.
K
han
e
t
al
.
[12
]
us
ed
b
-
ba
nd
fr
om
CIE
L*a*
b
c
olour
m
od
el
to
facil
it
at
e
identific
at
ion
proces
s
f
or
the
pr
e
se
nce
of
pa
rasit
e.
K
-
m
eans
cl
ust
eri
ng
was
ap
pl
ie
d
in
se
gm
entat
ion
process
.
H
ow
e
ver,
k
value
was
m
anu
al
ly
determ
ined.
More
ov
e
r,
this
stud
y
lim
it
ed
on
ly
to
ide
nt
ify
the
pr
ese
nce
of
parasi
te
w
hile t
he
par
asi
te
s
pecie
s w
as
not acc
om
pl
ished
ye
t.
Savkare
et
al
.
[
13
]
c
onduct
ed
a
stud
y
to
i
de
nt
ify
the
sp
eci
es
and
sta
ge
of
P
.
falci
pa
ru
m
and
P.
vi
vax.
So
m
e
m
et
ho
ds
wer
e
ap
plied
i
n
pr
e
-
proc
ess
s
uch
as
m
edian
filt
er,
La
placi
an
filt
er
a
nd
c
ontrast
e
nh
a
nce
m
ent.
In
se
gm
entat
ion
pr
ocess,
wat
ersh
e
d
was
use
d
to
i
den
ti
fy
sta
ge
of
par
asi
te
wh
il
e
Ots
u
thres
ho
l
ding
use
d
to
identify
it
s
sp
eci
es.
Two
kind
s
of
extracte
d
featur
e
s
especi
al
ly
con
tour
-
ba
sed
an
d
histo
gr
am
-
base
d
fe
at
ur
es
wer
e
cl
assifi
e
d
by
us
in
g
SVM
cl
assifi
er.
Si
m
i
la
r
stud
y
co
nducted
by
Pur
nam
a
et
al
.
[1
4]
,
bu
t
they
co
nducte
d
the
exp
e
rim
ent
on
the
thick
bl
ood
fil
m
s.
The
par
asi
te
obj
e
ct
was
no
t
ob
t
ai
ned
by
segm
entat
ion
pr
oce
ss
bu
t
m
anu
al
l
y
crop
ping
in
the
re
gi
on
of
i
nterest
(RoI).
The
n,
hi
stog
ram
-
base
d
featur
e
s
we
re
extracte
d
from
RoI
fo
ll
owe
d by cl
assifi
cat
ion
pro
cess b
a
sed
on geneti
cs al
gorithm
.
Fu
rt
her
m
or
e,
a
schem
e
to
identify
Plasm
od
ium
viv
ax
i
n
thin
blood
s
m
ear
was
al
so
pro
po
se
d
by
Akba
r
et
al
.
[
15
]
.
T
his
stu
dy
was
co
nduct
ed
with
t
he
L
aborato
ry
of
P
arasit
ology,
F
acult
y
of
Me
di
ci
ne,
Un
i
ver
sit
as
G
a
dj
a
h
Ma
da.
A
total
of
60
i
m
ages
us
e
d
c
on
sist
of
20
t
r
ophozoite
s,
20
gam
et
ocyt
es
and
20
schizo
nts.
Se
gm
entat
ion
proc
ess
was
done
by
app
ly
ing
k
-
m
eans
cl
us
te
ri
ng.
H
oweve
r,
t
he
pa
ram
et
er
of
k
was
m
anu
al
ly
d
et
er
m
ined.
A
total
of sev
e
n hist
og
ram
-
based
feat
ur
es
w
e
re
u
se
d i
n
this
stu
dy.
3.
METHO
DOL
OGY
Ba
sed
on
the
afo
rem
entione
d
li
te
ratur
es,
there
are
som
e
chall
eng
es
in
or
de
r
to
increase
the
perform
ance
fo
r
ide
ntifyi
ng
the
sp
eci
es
an
d
sta
ges
of
m
a
la
ria.
The
im
pr
ov
em
ent
of
im
age
qual
it
y
is
sta
rted
by
ta
kin
g
im
a
ges
in
loss
le
s
s
fo
rm
at
with
hig
he
r
res
olut
ion
durin
g
ac
qu
isi
ti
on
proc
ess.
The
n,
co
ntrast
enh
a
ncem
ent
is
co
nducte
d
in
pr
e
-
processi
ng.
T
he
k
pa
ra
m
et
er
of
k
-
m
eans
cl
us
te
rin
g
us
e
d
in
segm
entat
ion
process
can
b
e
autom
at
ic
ally
determ
ined
by
consi
der
i
ng the
i
m
age h
ist
og
r
a
m
.
More
ov
e
r,
m
or
e
num
ber
of
f
eat
ur
es
us
ed
f
ollow
e
d
by
fe
at
ur
e
sel
ect
ion
process
w
ere
est
i
m
at
ed
to
increase
the
pe
rfor
m
ance
of
cl
assifi
cat
ion
.
The
cl
assi
ficat
ion
of
s
pecies
and
sta
ge
s
of
m
al
aria
par
asi
t
e
is
a
m
ul
ti
cl
ass
te
r
m
.
Most
of
t
he
cl
assifi
ers
we
r
e
dev
e
l
op
e
d
to o
ve
rc
om
e
bin
ary
cl
ass
te
rm
.
Ther
e
f
or
e, OV
O
(
on
e
-
ver
s
us
-
one
)
an
d
O
V
A
(
on
e
-
ve
rsu
s
-
al
l)
ca
n
be
ap
plied
for
op
ti
m
isi
ng
the
cl
assifi
cat
ion
resu
lt
.
T
he
fi
ve
m
ai
n
ste
ps
c
onduct
ed
in
this st
ud
y
are
descr
i
bed in F
i
gure
2.
Figure
2. Bl
oc
k diag
ram
o
f
th
e pro
posed
app
ro
ac
h
T
h
i
n
b
l
o
o
d
d
i
g
i
t
a
l
i
m
a
g
e
P
r
e
-
p
r
o
c
e
s
s
i
n
g
S
e
g
m
e
n
t
a
t
i
o
n
F
e
a
t
u
r
e
s
e
x
t
r
a
c
t
i
o
n
C
l
a
s
s
i
f
i
c
a
t
i
o
n
A
c
c
u
r
a
c
y
,
s
e
n
s
i
t
i
v
i
t
y
,
s
p
e
c
i
f
i
c
i
t
y
F
e
a
t
u
r
e
s
e
l
e
c
t
i
o
n
P
l
a
s
m
o
d
i
u
m
F
a
l
c
i
p
a
r
u
m
P
l
a
s
m
o
d
i
u
m
V
i
v
a
x
G
a
m
e
t
o
c
y
t
e
s
S
c
h
i
z
o
n
t
T
r
o
p
h
o
z
o
i
t
e
s
G
a
m
e
t
o
c
y
t
e
s
S
c
h
i
z
o
n
t
T
r
o
p
h
o
z
o
i
t
e
s
S
c
h
e
m
e
-
1
S
c
h
e
m
e
-
2
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
:
933
–
944
936
3.1.
Material
a
n
d Tools
This
stu
dy
us
e
s
124
dig
it
al
pa
rasit
e
i
m
ages
of
t
hin
blood
fi
l
m
s
including
61
P.
falci
paru
m
and
63
P.
viv
a
x.
Plasm
od
ium
falci
par
um
con
sist
ing
of
24
gam
et
ocy
te
s,
16
tr
opho
zoite
s
an
d
21
schizo
nts.
M
oreov
e
r,
Plasm
o
diu
m
vi
vax
c
om
pr
ise
s
of
23
gam
et
ocyt
es,
23
tr
opho
zoite
s
an
d
17
s
chizo
nts
as
il
lustrate
d
in
Fi
gure
3.
The
im
ages
wer
e
colle
ct
ed
f
ro
m
the
Labor
at
or
y
of
Parasi
tolog
y,
Facult
y
of
Me
dicine,
Un
ive
rsita
s
G
adj
a
h
Ma
da.
T
he
groun
d
tr
uth
im
a
ges
wer
e
m
an
ually
sk
et
che
d
by
m
edical
e
xp
e
rts
us
in
g
a
phot
o
-
e
diti
ng
too
l.
A
new
la
ye
r
is
ad
ded
a
bove
ori
gi
nal
im
age,
the
n
m
edical
ex
pe
rt
m
ark
s
the
bounda
ry
of
pa
ra
sit
e
ob
j
ect
bas
ed
on
the appea
ran
ce
of
or
igi
nal im
age. Fi
nally
, ma
rk
e
d
im
age is
conve
rted
t
o binar
y sca
le
.
(a)
(b)
(c)
(d)
(e)
(f)
Figure
3. The
im
ages s
am
ple o
f
P. f
alcip
ar
um
(
a)
gam
et
ocyt
es (
b) sc
hizo
nts (
c
)
tr
ophoz
oites;
The
im
ages of
P. vivax
(a
) ga
m
et
ocyt
es (
b) s
chizo
nts (
c
)
tr
ophoz
oite
3.2.
Data A
c
quisi
tion
The
ac
qu
isi
ti
on
pr
ocess
in
volves
bin
ocu
la
r
l
igh
t
m
ic
ro
sco
pe
and
O
ptil
ab
ca
m
era.
The
m
agn
i
ficat
ion
us
e
d
in
obje
ct
ive
le
ns
is
100
tim
es
wh
ic
h
is
10
ti
m
es
big
ger
than
t
hat
of
the
ocu
la
r
le
ns.
The
re
so
l
ution
of
or
i
gin
al
im
age
ob
ta
ine
d
by
O
ptil
ab
cam
era
i
s
1600
x1200
pi
xels
with
los
sle
ss
(
bit
m
ap)
f
orm
at
.
Each
sli
de
of
blood fil
m
s can gen
e
rate se
ve
ral p
a
rasit
e im
ages d
e
pe
nd
i
ng on t
he paras
it
e d
ensity
cont
ai
ned
.
3.3.
Pre
-
pr
oc
essing
In
it
ia
ll
y,
the
r
e
gion
of
intere
s
t
(RoI)
with
re
so
luti
on
of
250x250
pi
xels
i
s
obta
ine
d
by
crop
ping
the
area
co
ntainin
g
pa
rasit
es
as
dep
ic
te
d
i
n
Fi
gure
4.
T
he
siz
e
is
con
si
der
e
d
to
be
co
ver
i
ng
t
he
w
hole
area
of
par
asi
te
s.
Dur
i
ng this ste
p,
t
he
h
aem
at
olo
gis
ts vali
dated
the
RoI yi
el
ded
.
(a)
(b)
Figure
4. (a
) O
rigin
al
im
age an
d (
b) Ro
I
im
a
ge
Since the Ro
I
im
age
has
a low
contrast as p
resen
te
d
in Figure 4 (
b),
hen
c
e the b
ounda
ry o
f
re
d
bl
oo
d
cel
ls
and
it
s
backg
rou
nd
te
nd
s
to
be
ha
r
d
to
sepa
rate.
This
issue
is
so
lve
d
by
app
ly
in
g
pow
er
-
la
w
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
Id
e
ntif
ic
ation
of
Plasmo
dium f
alciparu
m a
nd
Plasmo
dium vi
vax
on
d
i
gital
i
mage
…
(
H
anung A
di Nu
groho
)
937
trans
form
ation
as
f
or
m
ulate
d
i
n
(1)
[
16
]
.
He
re,
an
d
are
posit
ive
c
on
sta
nts,
w
hile
(g
am
m
a)
is
a
n
expo
nen
ti
al
va
lue.
The
powe
r
-
la
w
tra
nsfo
r
m
at
ion
change
s
the
range
of
i
m
age
con
trast
to
be
m
or
e
ex
te
ns
ive
known
as
gam
m
a cor
recti
on. After
wa
rd
s
, the
green
ch
a
nne
l i
s ex
t
racte
d
to
f
aci
li
ta
te
the s
egm
entat
ion
step.
=
(1)
3.4.
Se
gme
ntati
on
Segm
entat
ion
is
an
i
m
po
rtant
ste
p
befo
re
co
nductin
g
featu
r
e
extracti
on
process.
Se
gm
entat
ion
aim
s
to
se
par
at
e
pa
r
asi
te
s
from
othe
r
obj
ect
s,
su
c
h
as
re
d
bloo
d
cel
ls,
bac
kgr
o
und
an
d
artefact
s.
T
he
num
ber
of
k
in k
-
m
eans clu
ste
ring is
deter
m
ined
by
fin
di
ng the
sig
nific
ant local m
axim
a o
n
t
he gree
n
c
hannel
histo
gr
am
.
Histo
gr
am
is
a
gr
ap
hical
repr
esentat
ion
of
the
intensit
y
va
lues
distrib
utio
n
[17]
.
The
lo
cal
m
axi
m
a
(p
ea
k
of
inten
s
it
y)
represent
t
he
ce
ntre
of
ar
ea,
w
hich
ha
s
si
m
il
ar
char
act
erist
ic
of
inten
sit
y
value.
O
pe
ning
m
or
phologica
l
operati
on
wi
th
str
uctu
rin
g
el
em
ent
siz
e
of
5x5
is
th
en
c
onduct
ed
f
or
obta
inin
g
bette
r
segm
entat
ion
re
su
lt
foll
ow
e
d by cl
os
i
ng ope
rati
on w
it
h t
he st
ru
ct
uri
ng ele
m
ent size
o
f
7x7.
3.5.
Fe
ature
Extr
act
i
on
Thr
ee
ki
nds
of
featu
res
c
onsist
ing
of
sha
pe
,
te
xture
of
hist
ogram
in
first
order
an
d
seco
nd
ord
er
a
re
extra
ct
ed
.
The
twel
ve
sh
a
pe
featur
e
s
are
th
e
obj
ect
area,
conve
x
area,
c
onve
xity
,
obj
e
ct
per
im
e
te
r,
conve
x
per
im
et
er,
so
li
dity
,
com
pactness,
r
oundne
ss,
tri
m
ness,
first
inv
a
riant
m
o
m
ent,
sec
ond
in
va
riant
m
o
m
ent
s
an
d
third
in
var
ia
nt
m
o
m
ents.
Textu
re
featur
e
s
of
h
ist
og
ram
i
n
first
order
c
on
sist
of
m
ean,
sta
nda
rd
de
vi
at
ion,
entr
op
y,
sk
e
w
ness
a
nd
ku
rtosis
wh
il
e
tha
t
of
i
n
sec
ond
orde
r
incl
ud
e
s
co
ntrast,
c
or
relat
ion
,
ene
rgy
an
d
ho
m
og
e
neity
.
These
te
xt
ur
e
f
eat
ur
es
a
re
ext
racted
f
r
om
diff
ere
nt
colo
ur
c
hannels,
par
ti
c
ul
arly
gr
ey
scal
e
and
gr
e
en
channel
an
d
sa
turati
on
cha
nnel
of
HSV
col
our
m
od
el
.
Th
ese
channels
a
re
sel
ect
ed
bec
ause
the
m
os
t
need
e
d
inf
or
m
at
ion
re
la
te
d
to
the
c
har
act
erist
ic
s
of
t
he
pa
rasit
e
can
be
m
or
e
cl
early
seen.
Ther
e
f
or
e,
as
of
nin
e
featur
e
s
are
e
xtract
ed
f
r
om
ea
ch
cha
nnel
,
the
re
are
a
total
of
27
e
xtracte
d
te
xture
featu
res
from
three
dif
fer
e
nt
channels.
3.6.
Fe
ature
Sele
ction
The
w
rappe
r
needs
a
long
tim
e
of
com
pu
ta
ti
on
to
proce
ss;
howev
e
r,
i
t
can
yi
el
d
m
or
e
acc
ur
at
e
resu
lt
[18]
.
Wrapp
e
r
is
c
hose
n
base
d
on
a
J
anecek'
stu
dy
[
19
]
.
So
m
e
featur
e
sel
ect
ion
m
et
ho
ds
are
c
onduct
e
d
to
increase
t
he
cl
assifi
cat
ion
resu
lt
.
I
n
this
s
tud
y,
wr
a
pp
e
r
is
able
to
gain
the
releva
nt
f
eat
ur
es
a
nd
ac
hieve
s
the
best
cl
assif
ic
at
ion
res
ult.
Be
st
first
sear
ch
(BF
S)
with
forw
a
r
d
an
d
backwa
rd
dire
ct
ion
s
ar
e
co
nsi
der
e
d
durin
g
t
he
sel
e
ct
ion
process
.
3.7.
Clas
si
ficat
i
on
The
t
wo
sche
m
es
of
cl
assi
ficat
ion
a
re
e
valuated
i
n
this
s
tud
y.
The
first
schem
e
is
to
cl
assify
the
sp
eci
es
of
par
a
sit
e
into
two
cl
asses,
na
m
ely
P.
falci
pa
ru
m
and
Plasm
od
iu
m
viv
ax.
Wh
il
e
the
seco
nd
s
chem
e
is
to
cl
assify
th
e
sta
ge
a
nd
sp
e
ci
es
of
pa
rasit
e
at
on
ce
into
six
cl
asses,
i
.e.
P.
falci
pa
r
um
i
n
ga
m
et
ocyt
es
sta
ge
(F
G
),
P
.
falci
pa
ru
m
in
schizo
nt
sta
ge
(FS)
,
P.
falci
pa
ru
m
i
n
tro
phoz
oite
s
ta
ge
(F
T
),
P
.
vi
vax
in
gam
et
ocyt
es
sta
ge
(
VG),
P.
viv
a
x
in
schiz
on
t
sta
ge
(VS)
and
P
.
vi
vax
in
tro
phoz
oite
sta
ge
(
VT).
Th
e
second
sc
he
m
e
is
cat
egorised
as
m
u
l
ti
cl
ass
cl
as
sific
at
ion
i
n
w
hich
t
he
num
ber
of
cl
ass
m
or
e
than
tw
o
cl
as
ses.
T
her
e
f
or
e,
O
V
O
and OV
A
m
et
ho
ds are
used
to
sim
plify t
he
cl
assifi
cat
ion
pro
cess.
Var
i
ou
s
cl
assi
f
ie
rs
invol
ved
c
om
pr
ise
of
m
ulti
-
la
ye
r
per
ce
pt
ron
(MLP
),
k
-
near
est
neig
hb
our
(
K
NN)
an
d su
pport
ve
ct
or
m
achine (SVM)
. T
he
M
LP used
p
a
ram
et
ers
are lea
rn
i
ng r
at
e
of 0.3
, m
o
m
entum
o
f 0.
2,
t
he
trai
ning
num
ber
of
50
0
a
nd
on
e
hidde
n
la
ye
r.
T
he
a
dj
ace
ncy
pa
ram
et
er
s
of
K
NN
us
e
d
f
or
t
he
fi
rst
and
t
he
seco
nd
schem
es
are
six
a
nd
t
wo
ad
j
ace
ncie
s,
res
pe
ct
ively
.
Me
an
wh
il
e,
t
he
S
VM
cl
assi
fier
us
es
ra
dial
basis
functi
on
(RBF
)
ke
rn
el
.
Cl
assifi
cat
ion
is
co
nducted
in
bo
t
h
of
the
f
ull
f
eat
ur
es
a
nd
th
e
sel
ect
ed
feat
ur
es
.
More
ov
e
r, 1
0
-
f
old
s
cr
os
s
validat
ion i
s c
on
si
der
e
d d
ur
in
g
t
he
pro
ces
s.
3.8.
E
va
lu
at
i
on
an
d
Va
li
d
ati
on
So
m
e
par
am
eter
s
a
re
us
ed
to
m
easur
e
the
pe
rfor
m
ance
of
the
pr
opos
e
d
a
ppr
oach.
T
he
s
egm
entat
ion
process
is
vali
dated
by
com
par
i
ng
the
s
eg
m
ented
pa
rasit
es
with
that
of
the
gro
und
tr
uth
im
age.
Th
en,
the
po
sit
ive
pre
dicti
ve
val
ue
(P
P
V)
as
form
ula
t
ed
i
n
(2)
is
use
d
t
o
e
valuate
the
pe
rfor
m
ance
of
se
gm
entat
ion
m
et
ho
d
[
20,
21]
.
T
r
ue
posit
ive
(TP)
is
the
nu
m
ber
of
pix
el
s
re
pr
e
se
nted
as
pa
rasi
te
that
are
cor
rectl
y
segm
ented
as
pa
rasit
e.
Wh
il
e f
al
se posit
ive (F
P)
is t
he nu
m
ber
of
non
-
par
as
it
e p
ixels se
gme
nted
a
s
par
asi
te
.
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
:
933
–
944
938
=
+
100%
(2)
Fu
rt
her
m
or
e, t
he
pe
rfor
m
ance o
f
cl
assifi
cat
ion
process is m
easur
ed by i
nv
ol
ving so
m
e
par
am
et
ers
as
pr
ese
nted
i
n
T
able
1.
Tr
ue
posit
ive
(T
P)
is
po
sit
ive
cl
ass
correct
ly
cl
assifi
ed
as
posit
ive,
w
hile
true
ne
gative
(TN)
i
s
ne
gative
cl
ass
co
rr
e
ct
ly
cl
assifi
ed
as
ne
gative.
F
al
se
po
sit
ive
(
FP)
is
neg
at
iv
e
cl
ass
cl
assifi
ed
a
s
po
sit
ive
,
w
hil
e
false
neg
at
i
ve
(
FN)
is
ne
gative
cl
ass
cl
assifi
ed
as
ne
gative.
These
par
am
et
ers
are
then
involve
d
to
cal
culat
e the acc
uracy
, s
e
ns
it
ivit
y a
nd s
pecifici
ty
as for
m
ulate
d from
(
3)
t
o (
5).
Table
1.
C
onf
usi
on Mat
rix
Actu
al Class
Predicted
as po
siti
v
e
Predicted
as neg
ati
v
e
Po
sitiv
e
TP
FN
Neg
ativ
e
FP
TN
=
+
+
+
+
100%
(3)
=
+
100%
(4)
=
+
100%
(5)
4.
RESU
LT
S
A
ND AN
ALYSIS
4.1.
Pre
-
pr
oc
essing
The
c
ontrast
of
Ro
I
im
age
is
en
ha
nced
ba
sed
on
po
wer
-
la
w
tra
ns
f
or
m
at
ion
to
m
ake
the
par
asi
te
m
or
e
cl
early
a
s
sh
ow
n
in
Figure
5.
Extrac
te
d
gr
ee
n
cha
nnel
is
then
sel
ect
ed
to
facil
it
at
e
the
segm
e
ntati
on
process
, s
ince
the c
on
t
rast b
et
ween pa
rasit
e
and b
ac
kgr
ound is
bette
r
t
han o
the
rs
a
s s
how
n
in
Fig
ure
6.
Figure
5. The
im
ages r
es
ult o
f
pow
e
r
-
la
w
tra
ns
f
or
m
at
ion
(a)
(b)
(c)
Figure
6. The
e
xtracted
im
age
s of (a)
re
d
c
ha
nn
el
(
b)
green
channel a
nd (
c
) blue c
ha
nn
el
4.2.
Se
gme
ntati
on
Segm
entat
ion
is
condu
ct
e
d
ba
sed
on
k
-
m
ea
ns
cl
us
te
ri
ng.
The
num
ber
of
cl
us
te
r
is
det
erm
ined
by
consi
der
i
ng
t
he
local
m
axi
ma
of
histo
gr
am
distrib
utio
n.
As
il
lustrate
d
i
n
Fig
ure
7,
the
histo
gr
am
distrib
ution
of
e
xtracted
gr
een
cha
nnel
ha
s
the
six
pe
a
ks
of
local
m
a
xim
a.
Thu
s,
th
e
intensit
y
values
that
ha
ve
si
m
il
ar
char
act
e
risti
c are in o
ne
cl
us
te
r.
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
Id
e
ntif
ic
ation
of
Plasmo
dium f
alciparu
m a
nd
Plasmo
dium vi
vax
on
d
i
gital
i
mage
…
(
H
anung A
di Nu
groho
)
939
Figure
7. The
s
a
m
ple o
f
h
ist
ogram
d
ist
rib
ution
Af
te
r
wa
rd
s
,
th
e
resu
lt
ed
im
a
ge
of
cl
us
te
ri
ng
process
is
conver
te
d
to
bin
ary
scal
e
fo
ll
owe
d
by
m
or
phologica
l
op
e
rati
on.
T
he
se
res
ults
are
de
picte
d
f
r
om
F
igure
8
(a)
t
o
F
igure
8
(c).
T
he
segm
ented
pa
rasit
e
is
then
validat
ed
to
the
gr
ound
tr
uth
par
asi
te
as
sh
own
in
Figure
8
(
d).
The
se
gm
ented
par
asi
te
s
obta
ined
by
the pr
opos
e
d
a
ppr
oach are c
overe
d
i
n
re
d
li
ne
whil
e the
gr
e
en
li
ne
is
m
ark
ed
as
the
gro
und
tr
uth.
(a)
(
b)
(c)
(d)
Figure
8. The
im
ages r
es
ult o
f
(
a)
k
-
m
eans cl
us
te
rin
g (
b)
bi
nar
y sca
le
(c)
m
or
phologica
l (d)
validat
ed
p
a
ras
it
e
The
pe
rfo
rm
ance
of
se
gm
entat
ion
proce
ss
is
evaluated
by
cal
culat
ing
PP
V
as
prese
nted
in
Table
2.
The
hi
gh
e
r
th
e
PPV
val
ue,
the
cl
os
er
the
segm
ented
parasi
te
to
the
groun
d
trut
h.
Ba
sed
on
Ta
ble
2,
the
segm
entat
ion
m
et
ho
d
s
ucces
sfu
ll
y
segm
ents
par
asi
te
s
with
the
ave
rag
e
PPV
of
96.
68
%.
The
sm
al
les
t
of
PP
V
is
obta
ined
by
P.
falci
parum
i
n
tr
ophoz
oite
s
ta
ge
(
FT)
since
the
par
a
sit
e
ap
pear
s
in
sm
all
siz
e
an
d
thi
n.
Th
us,
the p
a
rasit
e is
qu
it
e
diff
ic
ult t
o be se
gm
ented.
Table
2.
T
he
E
valuati
on Res
ul
t of
Segm
entat
ion
Process
Clas
s
PPV (
%
)
Stan
d
ard Devia
tio
n
of
PPV
FG
98
.1
8
2.
56
FS
97
.3
7
2.
83
FT
91
.0
5
12
.
8
7
VG
99
.2
2
1.
80
VS
99
.5
9
1.
09
VT
94
.6
8
6.
21
4.3.
Fe
ature
Extr
act
i
on
and Sele
cti
on
A
total
of
39
f
eat
ur
es
is
extra
ct
ed
from
featur
e
extracti
on
pr
ocess.
Howe
ve
r,
not
al
l
the
featur
es
m
ay
con
t
rib
ute
to
th
e
cl
assifi
cat
ion
process
e
ven
i
t
is
po
te
ntial
to
raise
the
com
pu
ta
ti
on
ti
m
e.
T
hu
s
,
this
stu
dy
us
es
wr
a
pper
featu
r
e
sel
ect
ion
to
gen
e
rate
the
r
el
evan
t
featu
re
s
in
orde
r
t
o
i
ncr
ease
the
cl
assifi
cat
ion
re
su
lt
s.
Red
l
i
n
e
G
reen
l
i
n
e
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
:
933
–
944
940
Wr
a
pper
m
et
h
od
is
ap
plied
usi
ng
f
orwa
rd
a
nd
backwa
rd
di
recti
on
s
to
fi
nd
the
m
os
t
rele
van
t
subset
fea
tures
.
Wr
a
pper
searc
hes
for
t
he
bes
t
subset
of
a
f
eat
ur
e
base
d
on
the
perform
a
nce
resu
lt
of
t
he
a
pp
li
e
d
cl
as
sifie
r.
Ther
e
f
or
e,
d
i
fference
classi
fie
rs gen
e
rate
different selec
te
d
f
eat
ur
es.
In
t
he
first
sc
hem
e,
the
sel
ect
ed
feat
ur
es
ge
ner
at
e
d
by
wr
a
pp
e
r
-
base
d
f
or
ward
directi
on
with
S
VM,
KNN
a
nd
ML
P classi
fiers are 2,
2
an
d 3, re
sp
ect
ively
. T
he
se num
ber
s ar
e
v
ery sm
al
l co
m
par
ed
to the num
ber
of
t
he
39
f
ull
f
eat
ur
es.
H
ow
e
ver,
the
backw
ard
one
is
not
able
to
re
duce
the
num
ber
of
featur
e
s
sig
nifi
cantl
y
as
descr
ib
ed
in
Table
3.
The
nu
m
ber
of
sel
ect
ed
featur
e
s
f
or
the
sec
ond
s
chem
e
wh
ic
h
cl
assifi
es
sp
eci
es
and
sta
ge
of
par
asi
te
is
pr
ese
nted
in
Table
4.
T
he
w
ra
pp
e
r
ba
sed
f
orwa
rd
di
recti
on
i
n
a
m
ulti
cl
ass
m
et
ho
ds
of
OVO
an
d
O
V
A
gains
the
si
gn
i
ficantl
y
le
s
s
nu
m
ber
of
f
eat
ur
es
com
par
ed
to
the
bac
kw
a
r
d
on
e
.
F
or
th
e
wr
a
pper
base
d
forw
a
r
d
directi
on,
the
sm
a
ll
es
t
nu
m
ber
of
se
le
ct
ed
featur
es
is
ob
ta
ine
d
by
the
OVO
an
d
MLP
cl
assifi
er
with
two
featu
res.
Wh
il
st
the
hi
ghest
nu
m
ber
of
sel
ect
ed
featur
e
is
ob
ta
i
ne
d
by
OVO
an
d
K
NN
cl
assifi
er
with
5
featu
res
of
39
feat
ur
es
.
I
n
w
rapper
bas
ed
bac
kw
a
r
d
directi
on,
a
to
ta
l
of
38
featur
es
is
sel
ect
ed
by
O
VA an
d
ML
P c
la
ssifie
r
w
hile
the o
t
her
s
obta
in 29 se
le
ct
ed f
eat
ur
es.
Table
3
.
T
he
N
um
ber
of S
el
ec
te
d
Feat
ures i
n t
he
First
Sche
m
e
W
rapp
er
Dir
ectio
n
Clas
sif
ier
Selecte
d
Fe
atu
re
Fo
r
w
a
r
d
S
V
M
2
KNN
3
MLP
3
B
a
c
k
w
ar
d
S
V
M
36
KNN
36
MLP
38
Table
4.
T
he
N
um
ber
of S
el
ec
te
d
Feat
ures i
n t
he
Sec
ond Sc
hem
e
W
rapp
er
directio
n
Multicl
ass
m
eth
o
d
Clas
sif
ier
Selecte
d
Fe
atu
re
Fo
rwar
d
OVO
S
V
M
3
KNN
5
MLP
2
OVA
S
V
M
3
KNN
3
MLP
5
B
a
c
k
w
a
r
d
OVO
S
V
M
29
KNN
29
MLP
29
OVA
S
V
M
29
KNN
29
MLP
38
4.4.
Clas
si
ficat
i
on
In
t
his
stu
dy,
t
wo
schem
es
of
cl
assifi
cat
ion
are
e
valuate
d.
The
first
sc
he
m
e
cl
assifi
es
the
par
asi
te
into
two
cl
asse
s
based
on
it
s
sp
eci
es
w
hile
the
second
sche
m
e
cl
assifi
es
t
he
pa
rasit
e
into
six
cl
assed
ba
sed
on
it
s
sp
eci
es
a
nd
sta
ge.
D
ur
in
g
the
process
,
the
10
-
fo
l
ds
c
ross
is
in
vo
l
ved.
T
he
data
is
div
i
ded
into
te
n
se
ct
ion
s
with
the
e
qu
al
com
po
sit
ion
;
one
sect
io
n
as
tr
ai
nin
g
data
an
d
the
oth
e
r
sec
ti
on
s
as
te
sti
ng
data.
T
his
pro
cess
is
rep
eat
e
d 10 ti
m
es u
ntil
all
se
ct
ion
s
ha
ve under
gone
as trai
ning a
nd test
in
g data.
4.4.1. Schem
e
-
1: Pa
r
as
ite
Spe
c
ie
s C
las
sific
at
i
on
The
proces
s
of
sp
eci
es
cl
assifi
cat
ion
is
carried
out
in
the
fu
ll
featur
es
and
al
so
in
the
sel
ect
e
d
featur
e
s.
The
OVO
an
d
O
V
A
m
e
tho
ds
a
r
e
no
t
co
nduct
ed
beca
us
e
th
e
sp
eci
es
cl
assifi
cat
ion
is
bi
nar
y
cl
assifi
cat
ion
with
tw
o
cl
ass
es
co
ns
ist
in
g
of
P.
falci
paru
m
and
P.
viv
a
x.
Ta
ble
5
pr
e
sents
the
cl
ass
ific
at
ion
resu
lt
s
of
the
first
schem
e
on
the
39
fu
ll
featur
e
s
inclu
di
ng
of
accu
rac
y,
sensiti
vity
and
s
pecifici
ty
.
The
sensiti
vity
and
sp
eci
fici
ty
ar
e
cal
culat
ed
separ
at
el
y
betw
een
P.
falci
pa
ru
m
and
P.
viv
ax
to
fi
nd
out
the
perform
ance o
f
classi
ficat
ion
i
n reco
gnisi
ng tru
e
posit
ive a
nd tr
ue ne
gative
of eac
h
s
pecie
s.
As
show
n
in
T
able
5,
the
lo
w
est
accuracy
is
ob
ta
ine
d
by
th
e
SV
M
cl
assifi
er.
It
m
eans
th
at
the
SV
M
cl
assifi
er
is
not
pr
oper
in
rec
ognisi
ng
the
pa
rasit
e
sp
eci
es.
KNN
cl
assifi
er
gen
e
rates
bet
te
r
resu
lt
s
than
SV
M
cl
assifi
er,
bu
t
the
accu
racy
rate
is
sti
ll
l
ow
e
r
tha
n
tha
t
of
MLP
cl
a
ssifie
r.
KNN
cl
assifi
er
has
bette
r
perform
ance
in
rec
ognisin
g
P.
falci
pa
ru
m
with
the
se
ns
i
ti
vity
of
90.
16%
than
t
hat
of
P.
viv
a
x
with
the
sensiti
vity
of 49.
21%.
M
os
t o
f
the
s
pecies
m
isc
la
ssific
at
ions
are
com
ing
f
r
om
P.
vi
vax
w
hich
is
rec
ogni
sed
as
P.
falci
parum
.
Ge
ner
al
ly
,
t
he
best
cl
assifi
cat
ion
res
ults
us
in
g
t
he
39
f
ull
featu
res
ar
e
obta
ine
d
by
MLP
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
Id
e
ntif
ic
ation
of
Plasmo
dium f
alciparu
m a
nd
Plasmo
dium vi
vax
on
d
i
gital
i
mage
…
(
H
anung A
di Nu
groho
)
941
cl
assifi
er
with
the
accu
racy,
s
ensiti
vity
of
P.
falci
parum
and
se
ns
it
ivit
y
of
P.
viv
a
x
at
80.00%,
83.
61%
an
d
76.19%
, r
e
sp
ec
ti
vely
.
Table
5.
T
w
o
Cl
ass Cl
assifi
cat
ion
without F
eat
ur
e
Sele
ct
ion
Clas
sif
ier
Accurac
y
(%)
Sen
sitiv
ity
of
P.
fa
lcip
a
ru
m
(%)
Sp
ecif
icity
of
P.
fa
lc
ip
a
ru
m
(%)
Sen
sitiv
ity
of
P.
viva
x
(%)
Sp
ecif
icity
of
P.
viva
x
(%)
SVM
4
7
.
6
3
3
9
.
3
4
4
8
.
6
1
5
5
.
5
6
3
9
.34
KNN
6
9
.
6
9
9
0
.
1
6
8
3
.
7
8
4
9
.
2
1
9
0
.16
MLP
8
0
.
0
0
8
3
.
6
1
8
2
.
7
6
7
6
.
1
9
8
3
.61
Table
6
prese
nts
the
pe
rform
ance
of
cl
as
sific
at
ion
proc
ess
by
us
in
g
sel
ect
ed
feat
ur
es.
In
both
forw
a
r
d
an
d
ba
ckw
a
r
d
dire
ct
ion
s
,
the
S
VM
cl
assifi
er
on
t
he
sel
ect
ed
fea
tures
ob
ta
in
s
be
tt
er
accuracy
than
that
of
t
he
39
f
ull
featu
res.
H
ow
e
ve
r,
the
se
ns
it
ivit
y
rate
of
P.
falci
parum
and
P.
viv
a
x
i
s
not
prom
isi
ng
ye
t.
Fo
r
the
KNN
cl
assifi
er,
bette
r
accuracy
is
ob
ta
ine
d
by
f
orwa
rd
directi
on
at
81
.
45%.
Howe
ver,
m
os
t
of
P.
viv
a
x
is
cl
assifi
ed
as
P.
falc
ipar
um
ind
ic
at
ed
by
t
he
lo
w
sensiti
vity
rate
of
P.
viv
a
x
at
65
.
08%.
Th
e
best
perform
ance
of
cl
assifi
cat
io
n
is
achie
ve
d
by
MLP
cl
ass
ifie
r
us
in
g
for
ward
directi
on
with
the
a
cc
ur
acy
,
sensiti
vity
of
P
.
falci
parum
and
sensiti
vity
of
P.
viv
a
x
at
88
.
70%,
93.
44%
and
84.13%
,
re
sp
ect
ively
.
The
re
is
a
sig
nificant
i
ncr
ease
of
cl
a
ssific
at
ion
res
ults
com
par
ed
to
that
of
the
39
fu
ll
feat
ures.
T
hese
res
ul
ts
al
so
in
dicat
e that t
he
MLP cla
ssifi
er is a
ble to rec
ognise t
he
P
. fal
ci
par
um
an
d
P. viva
x.
Table
6.
T
w
o
Cl
ass Cl
assifi
cat
ion
with Fe
at
ur
e
Select
ion
W
rapp
er
d
irection
Clas
sif
ier
Accurac
y
(%)
Sen
s. of
P.
falcip
a
ru
m
(%)
Sp
ec.
o
f
P.
fa
lcip
a
ru
m
(%)
Sen
s. of
P.
viva
x
(%)
Sp
ec.
o
f
P.
viva
x
(%)
F
o
r
w
a
r
d
S
V
M
7
8
.
2
2
9
1
.
8
0
8
9
.
1
3
6
5
.
0
8
9
1
.
8
0
KNN
8
1
.
4
5
9
8
.
3
6
9
7
.
6
2
6
5
.
0
8
9
8
.
3
6
MLP
8
8
.
7
0
9
3
.
4
4
9
2
.
9
8
8
4
.
1
3
9
3
.
4
4
B
a
c
k
w
a
r
d
S
V
M
7
6
.
6
1
6
5
.
5
7
7
2
.
3
7
8
7
.
3
0
6
5
.
5
7
KNN
7
2
.
5
8
9
1
.
8
0
8
7
.
1
8
5
3
.
9
7
9
1
.
8
0
MLP
8
7
.
0
9
9
1
.
8
0
9
1
.
2
3
8
2
.
5
4
9
1
.
8
0
*S
e
ns
: se
ns
it
iv
it
y;
Sp
ec:
sp
eci
fici
ty
4.4.2.
Sc
hem
e
-
2: P
ar
as
ite
Specie
s
an
d
Stag
e
C
l
as
sific
ati
on
The
sec
ond
s
chem
e
is
condu
ct
e
d
to
cl
as
sify
the
s
pecies
an
d
sta
ge
of
pa
rasit
e
at
once.
T
he
perform
ance
of
cl
assifi
cat
io
n
on
t
he
39
f
ull
featur
e
s
is
dis
pl
ay
ed
in
Table
7.
The
a
ver
a
ge
val
ues
of
acc
uracy
,
sensiti
vity
and
sp
eci
fici
ty
resu
lt
ed
by
S
VM
cl
assifi
er
are
73.39%
,
17.
48
%
an
d
83.
51
%,
res
pecti
vel
y.
The
lowest
ave
rage
of
sen
sit
ivit
y
sta
te
s
that
SV
M
cl
assifi
er
is
no
t
able
to
rec
ognise
th
e
true
po
sit
ive
cl
ass.
Ba
sic
al
ly
, th
e SV
M cl
assifi
e
r
is m
erely
ap
propriat
e f
or b
i
na
ry cla
ssific
at
ion.
Table
7.
Six
Cl
ass Cl
assifi
cat
ion wit
ho
ut Fea
ture Sel
ect
ion
Clas
sif
ier
Av
erage of
Accurac
y
(
%)
Av
erage of
Sen
sitiv
ity
(
%
)
Av
erage of
Sp
ecif
icity
(
%
)
SVM
73
.
3
9
17
.
4
8
83
.5
1
KNN
87
.
0
9
60
.
6
8
92
.2
2
MLP
94
.
6
2
82
.
9
2
96
.7
8
Fo
r
the
KNN
cl
assifi
er,
the
aver
a
ge
of
accuracy
an
d
the
aver
a
ge
of
sp
e
ci
fici
ty
are
res
pecti
vely
at
87.09%
an
d
92
.22%.
H
oweve
r,
t
he
a
ver
a
ge
of
sen
sit
ivit
y
is
not
sat
isfie
d
si
nce
the
rate
m
erely
at
60.
68%.
The
MLP
cl
assifi
er
achieve
d
t
he
best
pe
r
form
ance
of
cl
assifi
ca
ti
on
with
the
a
ver
a
ge
of
accu
racy,
se
ns
it
ivit
y
and
sp
eci
fici
ty
o
f
94.62%
, 8
2.92% an
d 9
6.78
%
.
The
cl
assi
ficat
ion
res
ults
usi
ng
sel
ect
ed
fea
tures
a
re
prese
nted
i
n
Ta
ble
8.
The
m
ulti
class
m
et
ho
ds
especial
ly
OVO
an
d
O
VA
ar
e
con
si
der
e
d
duri
ng
the
proc
ess
to
i
m
pr
ove
the
cl
assifi
cat
i
on
perform
ance.
The
best
pe
rfor
m
ance
of
cl
assifi
cat
ion
is
yi
el
ded
by
us
i
ng
t
he
sel
ect
ed
fe
at
ur
es
of
w
ra
pp
e
r
-
base
d
ba
ckw
a
r
d
directi
on
fo
ll
owed
by
OVO
and
MLP
cl
assi
fier.
Th
e
aver
a
ge
of
acc
ur
acy
,
sensiti
vity
and
sp
eci
fici
ty
achieve
d
are
95.16%
,
84.
02%
a
nd
97.
09%,
res
pecti
ve
ly
.
Wh
il
st
th
e
pe
rfor
m
ance
res
ults
obta
in
ed
by
oth
er
s
a
re
no
t
sign
ific
a
ntly
d
i
ff
e
ren
t.
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
:
933
–
944
942
Table
8.
Six
Cl
ass Cl
assifi
cat
ion wit
h Feat
ur
e Selec
ti
on a
nd
Mult
ic
la
ss Sc
hem
e
W
rapp
er
Dir
ectio
n
Multiclas
s
m
e
th
o
d
Clas
sif
ier
(%)
Av
erage of
Accurac
y
(
%)
Av
erage of
Sen
sitiv
ity
(
%
)
Av
erage of
Sp
ecif
icity
(
%
)
F
or
w
ar
d
OVO
SVM
9
2
.
7
4
7
6
.
0
7
9
5
.
6
3
KNN
9
3
.
2
8
7
7
.
7
1
9
5
.
9
7
MLP
9
4
.
0
9
8
0
.
9
4
9
6
.
4
8
OVA
SVM
91.
67
7
1
.
8
4
9
4
.
9
1
KNN
9
4
.
0
9
8
0
.
9
9
9
6
.
4
8
MLP
9
4
.
6
2
8
3
.
1
9
9
6
.
8
1
B
a
c
k
w
a
r
d
OVO
SVM
8
4
.
1
3
4
9
.
2
2
9
0
.
2
5
KNN
8
7
.
0
9
6
0
.
1
4
9
2
.
2
0
MLP
9
5
.
1
6
8
4
.
0
2
9
7
.
0
9
OVA
SVM
9
0
.
3
2
6
8
.
7
6
9
4
.
0
9
KNN
8
7
.
3
6
6
1
.
8
8
9
2
.
4
0
MLP
9
4
.
8
9
8
3
.
2
3
9
6
.
9
3
5.
CONCL
US
I
O
N
An
ide
ntific
at
ion
a
ppr
oach
of
Plasm
od
ium
par
asi
te
on
digi
ta
l
i
m
age
of
thin
bl
ood
film
s
has
bee
n
pro
po
se
d.
Fi
rs
tl
y,
the
par
asi
te
is
segm
ente
d
by
us
i
ng
a
da
ptive
k
-
m
eans
cl
us
te
rin
g
on
the
ext
racted
gree
n
channel
w
hich
has
enh
a
nce
d
by
app
ly
in
g
th
e
power
-
la
w
t
r
ansfo
rm
ation
.
The
segm
entat
ion
m
et
ho
d
is
able
to
segm
ent p
arasi
te
p
r
operly
ass
ur
e
d by a
ver
a
ge
PPV
of
96.
86%.
Af
te
r
wa
rd
s
,
two
schem
es
of
cl
assifi
cat
ion
are
eval
uated
on
seve
ral
re
le
van
t
feat
ur
es
sel
ect
ed
by
wr
a
pper
m
et
hod.
T
he
first
sc
hem
e
cl
assifi
e
s
the
pa
rasit
e
s
pecies
into
tw
o
cl
asses,
nam
el
y
P.
falci
parum
and
P.
viv
a
x.
T
he
best
cl
assifi
cat
ion
pe
rfo
rm
ance
of
the
fir
st
schem
e
is
achieved
by
us
in
g
three
sel
ect
ed
f
eat
ur
es
of
wr
a
pper
-
ba
sed
f
orwa
rd
directi
on
f
ollo
wed
by
MLP
cl
assifi
er
with
the
acc
ur
ac
y,
sens
it
ivit
y
of
P
.
falci
parum
an
d sensit
ivit
y o
f P.
viv
a
x
at
88.
70%,
93.
44
%
a
nd 84.1
3%, res
pecti
vely
.
Fu
rt
her
m
or
e,
t
he
seco
nd
sche
m
e
cl
assifi
es
t
he
sp
eci
es
a
nd
sta
ge
of
par
a
sit
e
into
six
cl
ass
es.
The
tw
o
m
et
ho
ds
of
m
ulti
cl
ass
cl
assifi
cat
ion
es
peci
al
ly
OV
O
an
d
O
VA
a
re
i
nvolv
e
d
durin
g
t
he
process.
I
n
this
schem
e,
the
be
st
pe
rfor
m
ance
of
cl
assifi
c
at
ion
is
gai
ne
d
by
us
in
g
29
sel
ect
ed
featu
res
of
w
ra
pp
e
r
-
base
d
backwa
rd
dire
ct
ion
f
ollo
wed
by
O
VO
a
nd
MLP
cl
assi
fier.
T
he
a
verage
of
acc
ur
a
cy
,
sensiti
vity
an
d
sp
eci
fici
ty
achi
eved are
95.1
6%
, 84.0
2%
a
nd
97.09%
, res
pe
ct
ively
.
Gen
e
rall
y,
the
identific
at
ion
of
pa
rasit
e
is
su
ccess
fu
ll
y
co
nducted
in
bo
t
h
the
first
an
d
the
second
schem
es.
Thes
e
achiev
em
ent
s
in
dicat
e
that
the
pr
opos
e
d
a
ppr
oach
has
a
po
te
ntial
to
be
i
m
ple
m
ented
as
pa
rt
of
t
he
de
velo
pm
ent
a
com
pu
t
erised
ai
ded
m
al
aria
diag
nosi
s
syst
em
fo
r
as
sist
ing
the
pa
r
a
m
edics.
F
or
f
ur
t
her
stud
y,
the
e
xpansio
n
of
the
pro
po
se
d
ap
pr
oach
will
be
cond
ucted
to
re
cognise
the
ot
her
obj
ect
s
su
c
h
as
the
wh
it
e
bl
ood
ce
ll
s,
artefact
s
or
oth
e
r
pa
rasit
es
in
the
bl
ood
cel
ls.
More
over
,
the
aut
hor
s
will
con
sid
er
m
or
e
featur
e
s to
ide
ntify m
or
e sp
e
ci
es an
d
sta
ge
s
of the
Plasm
od
ium
p
arasit
es.
ACKN
OWLE
DGE
MENTS
This
st
ud
y
is
f
unde
d
by
Di
re
ct
or
at
e
Gen
e
ra
l
of
Hi
gh
e
r
E
ducat
ion,
Mi
nist
ry
of
Re
sea
rch,
Tec
hn
ology
and
Hi
gher
Ed
ucati
on,
Re
public
of
Ind
on
e
sia
.
Th
e
a
utho
rs
would
al
so
li
ke
to
ac
knowle
dge
t
he
I
nt
el
li
gen
t
Syst
e
m
s
research
gro
up
m
e
mb
ers
in
t
he
De
par
tm
ent
of
Ele
ct
rical
Eng
ine
erin
g
an
d
I
nform
at
ion
Tech
nology
,
UG
M,
for g
rea
t shar
i
ng and
di
scussion.
REFERE
NCE
S
[1]
W
.
H.
Organi
zat
ion,
"H
andbook:
Inte
gra
te
d
Man
age
m
ent
of
Chil
dhood
Ill
ness
[IMCI]
,
"
Gene
va:
W
HO
,
pp.
239
-
50,
2005
.
[2]
W
.
H.
Organ
isation,
"W
orld
Ma
l
ari
a
Report 2016
,
" Gene
v
a, Switzerl
and
,
2016
.
[3]
P.
K.
K.
R.
Indo
nesia
,
"InfoDA
TIN
Mala
ri
a,
"
Pus
dat
in
,
Indon
esia
2442
-
7659,
201
6.
[4]
W
.
H.
Organ
isation,
"W
orld
Ma
l
ari
a
Report 2015
,
" Gene
v
a, Switzerl
and
,
2015
.
[5]
W
.
H.
Organ
isation,
"G
uid
el
in
es
for
the t
r
ea
tmen
t
of
m
alari
a
,
" 3rd
ed.
G
ene
wa
,
Sw
i
tz
er
la
nd,
2015
.
[6]
F.
B.
Te
k,
A.
G.
Dem
pster,
and
I.
Kale
,
"P
ara
sit
e
det
e
c
ti
on
and
id
ent
ifica
ti
on
for
a
utomate
d
thi
n
bl
ood
fil
m
m
al
ari
a
dia
gnosis," Computer
v
ision and im
age
under
st
an
ding,
vo
l. 114, p
p.
21
-
32
,
2010
.
[7]
S.
Tut
i
and
W
.
W
orowija
t,
"S
it
u
asi
Mikroskopis
Mala
ri
a
di
Berbagai
La
bor
at
oriu
m
Dae
rah
Ende
m
ik,
Ta
hun
2006,
"
Bule
t
in
Pene
li
t
ian Siste
m
Keseha
ta
n,
vol
.
12
,
200
9.
[8]
G.
Doughert
y
,
Medic
a
l
imag
e
p
roc
essing: te
c
hni
ques
and applications:
Spring
er
S
ci
en
ce &
Busin
e
ss
Media
,
2011.
[9]
W
.
H.
Organ
isation,
"W
orld
Ma
l
ari
a
Report 2014
,
" 3rd edi
t
ion ed. Gene
wa
,
Sw
it
z
e
rla
nd,
,
2014.
Evaluation Warning : The document was created with Spire.PDF for Python.