Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
12
,
No.
3
,
Decem
ber
201
8
, p
p.
1
289
~
1
296
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
2
.i
3
.pp
1
289
-
1
296
1289
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Cardia
c Moti
ons Classi
fication
on S
equential PSA
X
Echocar
diogram
Adam S
hidqu
l Az
i
z
1
, R
iy
an
t
o
Si
git
2
, Ach
m
ad
B
as
u
ki
3
, T
au
fik
Hid
ayat
4
1,2,3
Inform
at
ic
s a
nd
Com
pute
r
En
gine
er
ing,
Polit
e
knik
E
le
ktron
ika Nege
ri
Suraba
ya,
Surab
a
y
a
4
Depa
rtment
of
Pedia
trics,
Airlangga
Univer
si
t
y
,
Suraba
y
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
ul
1
,
201
8
Re
vised
A
ug
3
0
, 2
01
8
Accepte
d
S
ep
2
1
, 201
8
Cardi
a
c
wal
l
m
oti
ons c
l
assific
a
tion on
2
-
dimensional
(2D) echoc
ard
iogra
ph
i
c
images
is
an
im
porta
nt
issue
fo
r
quantitative
di
agnosii
ng
of
he
art
d
isea
se
.
Unfortuna
tel
y
,
t
he
bad
qual
i
t
y
of
ec
hocardiogr
am
ca
use
computat
ion
a
l
l
y
cl
assifi
ca
t
ion
on
ca
rdi
ac
wa
ll
m
o
ti
ons
is
stil
l
b
e
come
a
big
ho
m
ework
for
m
an
y
r
ese
ar
che
r
s
to
provid
e
th
e
best
resul
t.
Ec
ho
ca
rdiogr
am
is
pr
oduce
d
b
y
soundw
ave
s
which
absolu
te
l
y
m
ake
it
s
imag
es
have
spe
ckle
noise
in
diffe
ren
t
int
ensi
t
y
.
The
re
fore
,
this
rese
arc
h
improves
a
set
of
m
et
hodolog
y
to
cl
assif
y
c
ard
iac
wall
m
oti
on
se
m
i
-
aut
om
at
ically
.
R
aw
ec
ho
ca
r
diogra
m
will
be
enha
n
ce
d
an
d
segm
ent
ed
to
ta
ke
th
e
bound
ar
y
o
f
endocard
ium
of
le
ft
vent
ri
cul
ar
in
PSAX
ca
rdia
c
images.
New
improvem
en
t
of
Sem
i
-
aut
om
at
i
ca
l
l
y
m
et
hodolog
y
is
appr
oac
h
on
det
ecti
ng
the
cont
our
of
endoc
ard
ium
and
will
be
input
e
d
as
good
fea
ture
s
in
Luc
as
-
Kana
de
Optical
Flow
in
al
l
seq
uent
i
al
e
choc
arg
rap
hic
imag
es.
On
cl
assif
y
ing
ca
rdi
ac
wal
l
m
oti
ons,
thi
s
rese
arc
h
proposes
two
important
fea
tur
es
inc
luding
le
ngth
of
displa
c
ement
a
nd
flow
dire
c
t
ion.
New
prop
rosed
flow
determ
ina
ti
on
al
gorit
hm
and
E
ucl
id
ea
n
dista
n
c
e
is
used
to
ca
lculate
those
fea
tu
res.
All
the
fea
tur
es
will
b
e
t
rai
ned
b
y
N
eur
a
l
Network
(NN
)
and
valida
te
d
b
y
Le
av
e
On
e
Out
(LOO)
to
g
et
ac
cur
ate
resul
t.
NN
m
et
hod,
which
is
v
al
id
ated
b
y
LOO,
has
the be
st
resu
lt
of
81.
82%
cor
rec
tn
ess t
han
the ot
her
compare
d
m
et
hods.
Ke
yw
or
d
s
:
E
uclidea
n dist
ance
F
low dete
rm
in
at
ion
alg
ori
thm
Lucas
-
Ka
na
de
op
ti
cal
f
l
ow
N
eu
ral
netw
ork
S
em
i
-
autom
at
i
cal
ly
card
ia
c
m
ot
ion
s classi
f
ic
at
ion
Copyright
©
201
8
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
:
Ad
am
Sh
id
qul
Aziz
,
Inform
at
ic
s an
d
C
om
pu
te
r
E
nginee
rin
g,
Po
li
te
kn
i
k
Ele
ktr
on
i
ka Nege
r
i Su
rab
ay
a,
S
urabaya
.
Em
a
il
:
azi
z.add
am
@g
m
a
il
.co
m
1.
INTROD
U
CTION
Hear
t,
wh
ic
h
is
locat
ed
withi
n
th
or
aci
c
cavi
ty
,
is
on
e
of
th
e
m
os
t
i
m
po
rtant
orga
n
in
the
body.
T
he
siz
e
of
ty
pical
hear
t
is
just
li
ke
a
hu
m
an
fist:
12
cm
in
le
ng
th,
8
cm
in
wide
,
an
d
6
cm
in
thickne
ss.
Howev
e
r,
hear
t
has
a
vital
j
ob
of
pum
pin
g
a
nd
s
pr
ea
di
ng
blood
th
rou
ghout
the
body
[1]
.
hear
t
flo
w
s
ox
yge
n
-
co
nta
inin
g
blood
th
rou
ghou
t
the
body.
O
xyge
n
will
dec
om
po
se
glu
c
os
e
i
n
t
he
bl
ood
a
nd
pro
du
ce
A
de
nosin
e
Trip
hosphat
e
(
ATP)
an
d
bec
om
e
a
us
efu
l
energy
f
or
cel
l
ular
res
pirati
on.
T
he
deat
h
s
ta
ti
sti
c
rep
or
t
of
the
card
i
ov
asc
ular
disease
in
2018
has
be
en
al
arm
ing
.
W
or
l
d
Healt
h
Orga
nizat
ion
(
WH
O
)
re
port
ed
t
hat
card
i
ov
asc
ular
d
ise
ase
ca
us
e
d
17,7
m
illi
on
de
at
hs
e
ver
y
ye
ar
,
it
m
eans
31%
of
al
l
gl
obal
de
at
hs
a
re
ca
us
e
d
by
card
i
ov
asc
ular
dise
ase
[2]
.
Ther
e
f
or
e,
ec
ho
ca
r
diogra
phy
app
ea
rs
as
one
of
non
-
inv
asi
ve
an
d
painles
s
te
chnolo
gy
to
create
i
m
age
of
hu
m
an
hear
t.
The
expert
in
echo
ca
r
diogra
ph
y
dia
gnos
e
hear
t
co
ndit
ion
base
d
on
so
m
e
sy
m
pto
m
s
wh
ic
h
a
ppear
i
n
th
e
im
age.
O
ne
exa
m
ple
of
sy
m
pto
m
is
card
ia
c
wall
m
otion
.
Ca
rd
ia
c
wall
m
otion
ca
n
giv
e
a
n
i
nd
i
cat
ion
wh
et
her
the
hear
t
is
he
al
thy
or
not.
Nev
e
rtheless
,
echo
ca
r
diogra
ph
y
ha
s
so
m
e
lim
it
ation
s
inclu
ding
im
age
qu
al
it
y,
op
erat
or
depend
e
ncy,
an
d
interp
reter
de
pe
nd
e
ncy
[
3]
.
Th
os
e
lim
it
at
ion
aff
e
ct
s
an
im
pact
on
th
e
acc
ur
acy
of
the
doct
or’s
dia
gnos
is.
T
he
accu
racy
of
th
e
doct
or’s
dia
gnos
i
s
dep
e
nds
on
th
e
do
ct
or’
s
knowle
dge
an
d
ex
per
ie
nce.
This
case
m
otivates
so
m
e
of
engi
neer
s
sta
rt
to
create
var
i
ou
s
i
nv
e
nt
ion
s
relat
ed
to
im
age
pr
oc
essing
on
ech
ocardio
gr
am
.
So
m
e
of
the
inv
e
ntions
f
oc
us
on
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.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
289
–
1
296
1290
enh
a
ncem
ent
[4
-
7]
.
The
obj
ect
iv
e
of
e
nh
a
ncem
ent
is
to
rem
ov
e
sp
ec
kle
noi
se
an
d
s
harp
en
in
echo
ca
r
diogra
ph
ic
im
age.
M
os
t
of
m
edical
researc
h
pro
po
se
m
edian
filt
e
r
to
rem
ov
e
spe
ckle
no
ise
in
i
m
age.
The
oth
e
r
i
nventions
fo
c
us
on
cl
assify
in
g
the
ec
hocar
di
ogram
beco
m
e
so
m
e
of
s
pec
ific
co
nd
it
io
n
[8
-
9]
.
Current
resea
r
ch,
not
on
ly
2
-
dim
ension
al
echo
ca
r
diogram
,
bu
t
al
so
3
-
dim
ensio
nal
ech
oc
ard
i
ographic
im
age.
So
,
the i
nv
e
nti
on on dia
gn
os
i
ng ech
oca
rd
i
ographic i
m
age is stil
l op
e
ned.
The
hi
gh
risk
of
m
or
ta
li
ty
th
at
caused
by
card
i
ov
asc
ular
disease
m
otivate
s
researc
her
s
to
dev
el
op
a
com
pu
ta
ti
on
al
diagnosis
with
their
own
a
ppr
oac
h
in
m
any
ty
pes
of
resou
rce.
Espe
ci
al
ly
on
2D
or
3D
echo
ca
r
diogra
ph
ic
im
age,
th
is
ty
pe
of
res
ource
c
onti
nu
e
s
to
be
a
t
rend
for
resea
rc
her
to
dia
gnos
e.
Me
ysam
Siy
ah
Ma
ns
oo
ry,
et
.al
[9]
ev
al
uated
hea
rt
m
ov
e
m
ent
to
diag
nose
hea
rt
disease.
He
use
d
the
I
ndepe
nd
e
nt
Com
pu
te
r
Al
gorithm
(I
CA
)
t
o
ob
ta
in
the
fe
at
ur
es
of
ec
hocard
i
ography.
Me
ysam
us
ed
the
ne
ural
net
work
t
o
cl
assify
an
d
obta
ined
the
hi
gh
e
s
t
accu
racy
of
95.7%
f
or
norm
al
data
and
93.
8%
f
or
abno
rm
al
data.
Sari
na
Ma
ns
or
an
d
J
Aliso
n
N
ob
le
[10]
in
20
08
c
onduct
ed
a
cl
assif
ic
at
i
on
st
ud
y
on
wall
m
ov
em
e
nt
of
echo
ca
r
diogra
ph
ic
im
ages
usi
ng
t
he
Hidd
en
Ma
r
kov
M
od
el
(H
MM
)
m
et
ho
d
a
nd
usi
ng
Leave
O
ne
O
ut
Vali
dation (
LO
O)
m
et
hod
re
s
ulted in
an a
ve
rag
e
accu
racy
of 80%.
In
this
resea
rc
h,
we
would
li
ke
to
inv
e
sti
ga
te
wh
et
he
r
the
flo
w
di
recti
on
and
le
ngth
of
m
ov
e
m
ent
as
featur
e
s
on
ca
rd
ia
c
wall
m
otion
can
cl
ass
ify
hear
t
cond
it
ion
.
Th
os
e
f
eat
ur
es
are
th
e
te
chn
iq
ue
of
the
echo
ca
r
diogra
ph
y
e
xpert
to
diag
nose
car
di
ac
wall
m
otio
n.
We
us
e
d
se
m
i
autom
a
ti
ca
ll
y
go
od
featu
re
f
or
Lucas
-
Ka
na
de
Op
ti
ca
l
Flow
to
ob
ta
in
th
ose
featur
es
.
Finall
y,
arti
fici
al
neu
ral
net
work
is
perf
or
m
ed
to
cl
assify
the ca
r
diac wall
m
otion.
2.
WALL
MOTI
ON ABN
ORMALITIE
S
Ech
ocardio
gr
a
ph
y
pr
oduces
four
ty
pes
of
views:
sho
rt
axis
(SAX
),
lo
ng
a
xis
(L
AX),
apical
tw
o
cham
ber
(
A2
C
),
a
pical
f
our
cham
ber
(
A4
C
).
nev
e
rtheless
,
this
re
searc
h
fo
c
us
on
S
A
X
vie
w.
O
n
P
SAX
echo
ca
r
diogra
m
,
the
exp
e
rt
will
ob
ser
ve
wall
m
otion
s
of
le
ft
ve
ntri
cular
to
deter
m
ine
the
curr
ent
first
j
ud
gem
ent
of
hear
t
c
onditi
on.
a
bnorm
al
i
ties
of
he
art
will
be
in
dicat
ed
by
m
otion
s
of
e
ndoca
r
dium
and
per
ic
ar
diu
m
ti
ssu
e.
Hear
t
m
ove
m
ent
and
siz
e
of
t
he
hear
t
se
gm
ent
wall
when
c
on
tract
in
g
and
relaxati
on
have
an
in
dicat
ion o
f
hea
rt
disease.
Pandian
et al
[
11
]
e
xp
la
i
ns
th
at
m
ov
em
ent o
f
the le
ft v
e
ntri
cular
wall
that
is not
sy
m
m
e
try
and
thickeni
ng
of
t
he
hea
rt
wall
in
ge
ner
al
gi
ve
s
a
sign
of
m
yo
car
dial
isc
he
m
ia
or
infar
ct
i
on.
He
con
cl
ud
e
d
t
hat
the
thicke
ning
an
d
m
ov
em
ent
of
the
hea
rt
wall
pro
vide
s
an
in
dicat
io
n
of
the
hea
rt's
wall
abno
rm
aliti
es.
Figure
1. Ty
pe
s of
Wall
Moti
on
s
Ab
norm
al
i
ti
es
Figure
1
s
how
s
there
a
re
f
our
ty
pes
of
w
al
l
m
otion
s
ab
norm
al
i
ti
es
su
ch
as
norm
al
,
dysk
i
nesia
,
hypoki
nesia,
a
kin
esi
a.
Li
ke
exp
la
ine
d
by
Ca
therine
Otto
[
12
]
,
norm
al
co
nd
it
io
n
is
endo
card
i
um
m
ov
ing
into
deep
to
wa
rd
t
he
center
nor
m
al
l
y
du
rin
g
s
yst
ole,
dysk
ine
sia
is
end
oca
r
diu
m
m
ov
ing
in
dif
fer
e
nt
dir
ect
ion
s,
hypoki
nesia
is
endoca
rd
i
um
m
ov
ing
into
de
ep
tow
a
r
d
the
center
slo
wly
durin
g
syst
ole
(am
plit
ud
e
le
ss
than
5
m
m
),
and
akin
esi
a
is
end
oca
r
diu
m
no
t
m
ov
ing
at
al
l.
In
thi
s
researc
h,
Dy
sk
inesia
,
hypo
kin
esi
a,
a
nd
a
ki
nesia
will
b
e cla
ssifi
ed
as
abn
or
m
al
hea
rt con
diti
on
.
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
Card
i
ac
M
otio
ns
Cl
assi
fi
cation o
n Se
qu
e
ntial P
SAX
E
ch
oc
ar
di
ogram
(
A
dam
Shid
qul Azi
z
)
1291
3.
RESEA
R
CH MET
HO
D
Ma
in
idea
i
n
t
his
res
e
arc
h
is
how
t
o
cl
assify
card
ia
c
m
otion
s
on
ech
oca
rdi
ographic
im
ages
by
us
in
g
flo
w
di
recti
on
and
le
ngth
of
m
ov
e
m
ent
as
f
eat
ur
es.
Bo
unda
ry
of
e
ndocardium
in
this
r
esearch
is
obta
ined
sem
i
-
autom
at
i
cal
ly
. I
t m
eans th
at
syst
e
m
j
ust
n
eed
a ce
nter po
i
nt coo
rd
i
na
te
of e
ndocardi
um
in
first ru
nnin
g.
This
resea
rc
h
us
in
g
set
of
va
rio
us
m
et
ho
ds
befor
e
getti
ng
final
dia
gnos
e
of
wall
m
otion.
Fig
ur
e
2
sh
ows
t
her
e
are
f
our
ste
ps
in
the
propose
d
al
gorith
m
including
enh
a
ncem
ent,
segm
entat
ion
,
feature
extracti
on,
an
d
cl
assifi
cat
ion.
En
ha
ncem
ent
an
d
se
gm
entat
ion
has
bee
n
com
plete
ly
exp
la
ined
by
Ri
ya
nto
Sigit
et.al
[
13
]
in h
is
p
a
per.
Figure
2. Pro
pose
d
Al
gorith
m
3.1
.
Enh
an
c
e
ment
Sp
ec
kle
noise
is
a
m
a
in
pro
bl
e
m
in
the
m
os
t
of
ech
ocardi
ogra
ph
ic
im
age.
The
hi
gh
e
r
intensit
y
of
sp
ec
kle
no
ise
,
the
m
or
e
diff
ic
ult
to
be
obse
r
ved
for
the
do
c
tor.
S
o,
e
nhanc
e
m
ent
is
need
e
d
to
rem
ov
e
s
pe
ckle
no
ise
an
d
t
he
f
inal
res
ult
to
c
onve
rt
ob
j
ect
of
im
age
into
e
dg
e
.
E
nhance
m
ent
is
di
vid
e
d
int
o
fou
r
ste
ps
.
The
exp
la
natio
n
is
as foll
ows:
a
)
Me
dian
Hi
gh
B
oost Fi
lte
r
M
edian
Hi
gh
Boo
st
filt
er
is
a
new
m
od
ifie
d
of
ba
nd
pass
filt
er.
The
hi
gh
-
boos
t
filt
er
c
an
be
us
ed
t
o
enh
a
nce
high
f
reque
ncy
com
pone
nt
w
hile
sti
ll
keep
ing
the
low
f
re
qu
e
ncy
com
po
ne
nts.
Usu
al
ly
,
Hi
gh
Boo
s
t
filt
er
us
e
s
m
ean
fitl
er
as
l
ow
pass
filt
er
in
it
s
co
m
bin
at
ion.
H
ow
e
ve
r,
this
resea
rch
us
es
m
edian
filt
er
a
s
lo
w
pass
filt
er
in
c
om
po
sin
g
Hi
gh
Bo
os
t
Fil
te
r.
Me
dian
filt
er
ha
s
good
perf
orm
ance
on
re
du
ci
ng
s
pec
kle
noise
in
i
m
age.
The
fu
ndam
ental
eq
uat
ion
of
high
boost
f
il
te
r
is m
entione
d
as
foll
ows:
ℎ
=
0
+
∗
ℎ
(1)
Eq
uation
1
ex
plains
t
hat
hi
gh
boos
t
im
age
(Ih
b)
is
re
su
lt
ed
by
a
pp
e
nding
pi
xel
value
of
or
i
gin
al
i
m
age (
I
0) w
it
h hig
h pass im
age
(Ihp) a
nd
m
ul
ti
pled
with
determ
ined
co
ns
ta
nta
(c).
ℎ
=
+
∗
ℎ
(2)
Eq
uation
2
de
scr
ibes
the
high
boos
t
c
onvo
luti
on
ke
rn
el
.
Kernel
is
ob
ta
ined
by
a
pped
ing
al
l
pas
s
kernel (
Wa
p) a
nd contanta
(c
)
m
ult
ipled
by hi
gh
pass
k
e
r
nel (Wh
p).
ℎ
=
0
−
(3)
Othe
rside, hi
gh pass im
age (
I
hp)
ca
n be pr
oduce
d by ori
gi
nal im
age (
I0) m
inu
s lo
w pas
s i
m
age (
Il
p)
.
Lo
w
pas
s
filt
er
is
us
e
d
to
re
m
ov
e
or
reduc
e
no
ise
.
I
n
thi
s
researc
h,
hi
gh
bo
os
t
filt
er
us
es
m
edian
fi
lt
er
to
rem
ov
e sp
ec
kl
e noise c
on
ta
in
ing
i
n
ec
ho
c
ar
diogr
am
.
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.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
289
–
1
296
1292
b)
M
orpholo
gy
M
orpholo
gy
is
a
br
oad
set
of
im
age
proce
ssing
ope
rati
ons
t
hat
proc
es
s
im
ages
base
d
on
s
hap
es
.
Mor
phologica
l
is
div
i
ded
int
o
dilat
ion
a
nd
erosi
on.
C
ombinati
on
of
dilat
ion
a
nd
er
osi
on
m
akes
tw
o
ne
w
al
gorithm
s o
f o
pen
i
ng and cl
osi
ng.
∘
=
(
⊝
)
⊕
(4)
∘
=
(
⊕
)
⊝
(5)
Eq
uation
4
is
an
openi
ng
m
orp
ho
l
og
ic
al
.
Op
e
ning
is
th
e
dilat
ion
of
t
he
er
os
io
n
of
a
set
A
by
a
structu
rin
g
el
e
m
ent
B.
Equ
a
ti
on
5
is
a
cl
os
in
g
m
or
phol
og
ic
al
.
Cl
os
i
ng
is
a
set
(b
i
nar
y
im
age)
A
by
a
s
tructu
rin
g
el
e
m
ent B is t
he
e
ro
si
on of t
he dil
at
ion
of t
hat s
et
.
c
)
Thre
s
ho
ldi
ng
Thr
e
sholdi
ng
processi
ng
will
conver
t
im
age
into
bin
a
ry
im
age.
This
m
et
hod
will
create
a
bla
c
k
wh
it
e
i
m
age.
To
co
nvert
i
m
age
to
bin
a
ry,
i
m
age
m
us
t
be
conver
t
into
gray
scal
e
i
m
age
and
co
nverti
it
into
bin
a
ry w
it
h det
erm
ined
thres
hold
value
. T
he fo
rm
ula o
f bin
ary pi
xel d
et
e
r
m
inati
on
is e
xpla
ined
as
fo
ll
ow
s:
(
,
)
=
{
0
,
(
,
)
<
ℎ
1
,
(
,
)
≥
ℎ
(6)
d
)
Cann
y
Fil
t
er
Ca
nn
y
Fil
te
r
is
an
e
dge
detect
ion
op
e
rato
r
t
hat
use
s
al
m
ulti
-
sta
ge
al
gori
thm
to
detect
a
wide
ra
nge
of ed
ges
i
n
im
a
ge.
The
Proces
s of Ca
nn
y e
dg
e d
et
ect
io
n
al
gorithm
can be
bro
ken do
wn to 5
diff
e
re
nt steps:
1)
Apply Ga
us
sia
n fil
te
r
to sm
ooth the im
age in o
rd
e
r
to
r
em
ov
e the
noise
2)
Find the i
ntens
it
y gr
adie
nts
of the im
age
3)
Apply n
on
-
m
axim
u
m
su
ppres
sion t
o get ri
d of sp
ur
i
ous
res
pons
e
to
e
dge
detect
ion
4)
Apply d
ouble t
hr
es
hold t
o determ
ine p
otenti
al
ed
ge
s
5)
Track
ed
ge
by
hyste
resis:
Fi
naliz
e
the
dete
ct
ion
of
e
dges
by
sup
pr
es
sin
g
al
l
the
oth
er
edges
that
a
re
w
eak
and
not c
onnected
to st
r
ong
e
dges.
3.
2
.
Se
gme
ntati
on
Segm
entat
ion
is
a
ne
xt
ste
p
af
te
r
en
han
c
em
e
nt
to
ta
ke
only
the
co
ntour
of
endoca
rd
i
um
.
t
he
c
on
t
our
will
beco
m
e
go
od
feat
ur
e
f
or
Lucas
-
Ka
nad
e
Op
ti
cal
Flow
to
est
i
m
at
e
m
o
ti
on
of
en
do
ca
rd
i
um
in
sequ
e
nt
ia
l
echo
ca
r
diogra
ph
ic
im
ages.
T
her
e
are
t
hr
ee
s
te
ps
in
seg
m
entat
ion
m
et
ho
d.
The
e
xpla
natio
n
is as
foll
ows:
a
)
Re
gion Fil
t
er
Re
gion
Fil
te
r
will
rem
ov
e
al
l
unnecessa
ry
con
t
our
a
nd
e
dg
e
outsi
de
of
determ
ined
ra
ng
e
.
Ra
nge
value
is
deter
m
ined
m
anu
al
ly
fr
om
center
po
i
nt
of
en
doc
ard
i
um
to
est
i
m
at
ion
of
t
he
ou
te
st
of
e
ndoc
ard
iu
m
bounda
ry.
b)
Co
ll
ine
ar
Coll
inear
will
rem
ov
e
con
t
our
an
d
e
dg
e
inside
dete
r
m
ined
ra
ng
e
.
The
eq
ua
ti
on
of
c
olli
nea
r
descr
i
bed in
E
qu
at
io
n 7.
1
(
2
−
3
)
+
2
(
3
−
1
)
+
3
(
1
−
2
)
=
0
(7)
Eq
uation
7
ex
plains
ab
out
three
po
i
nts
in
on
e
strai
ght
li
ne.
The
thir
d
one
will
be
detect
ed
as
unnecessa
ry
point an
d wil
l be
rem
ov
ed.
c
)
Tr
ian
gle E
quation
This
m
et
ho
d
is
propose
d
s
pec
ia
ll
y
by
Riyanto
Sigit
et
.al
[13]
to
co
nn
ect
t
wo
unco
nnect
ed
li
ne.
T
w
o
determ
ined
po
ints
will
be
c
hoos
e
n
base
d
on
the
cl
os
est
an
gle
f
ro
m
tho
s
e
points
.
T
he
e
xp
la
nation
is
as
fo
ll
ows:
1)
Syst
e
m
has
to
determ
ine
the
ap
prox
im
at
e
center
of
the
enclose
d
reg
i
on
that
will
be
co
nn
ect
e
d
by
cal
culat
in
g
t
he c
entr
oid
.
2)
The
e
ndpoints
of
t
he
bo
unda
r
y
are
m
ark
ed
as
two
c
ouples
po
i
nt
that
wil
l
be
co
nnect
ed
.
Th
os
e
points
can
be
il
lustrat
ed
as
first
po
i
nt
is
B
an
d
t
he
seco
nd
one
is
C.
T
hen,
the
center
point
be
com
e
the
third
po
i
nt.
S
o, BA
C wil
l be
form
ed
as
tri
an
gle.
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
Card
i
ac
M
otio
ns
Cl
assi
fi
cation o
n Se
qu
e
ntial P
SAX
E
ch
oc
ar
di
ogram
(
A
dam
Shid
qul Azi
z
)
1293
3)
The
ga
p
in
the
boun
dar
y
is
m
ini
m
iz
ed
by
fixing
one
of
the
e
nd
points
m
su
ch
as
po
i
nt
B.
point
C
i
s
al
lowed
to
tra
nsvers
e
in
ward along
the
bo
undar
y un
ti
l
the point
at
w
hic
h
the
a
ng
le
B
AC is
m
ini
m
iz
ed
is
fou
nd.
4)
The o
pen bo
undar
y i
s t
hen co
nn
ect
e
d by a
li
ne fr
om
p
oin
t
B t
o po
i
nt C.
3.
3
.
Fe
ature
Extr
act
i
on
Lucas
-
Ka
na
de
Op
ti
cal
Flo
w
is
perf
or
m
ed
to
est
i
m
at
e
the
boun
dar
y
of
endoca
rd
i
um
i
n
se
qu
e
ntial
echo
ca
r
diogra
ph
ic
im
ages.
Op
ti
cal
Flow
us
es
inten
sit
y
of
im
age
to
determ
ine
identic
al
pix
el
in
a
no
t
he
r
i
m
a
ge.
I
n
Op
ti
cal
Flow
(
,
,
)
is
the
inte
ns
it
y
of
pix
el
s
with
(
,
)
express
l
ocati
on
of
t
he
pix
el
an
d
t
express
tim
e. Basi
cal
ly
, O
ptica
l Flow
can
be desc
ribe
d by
E
quat
ion 8
[
3]
:
(
,
,
)
=
(
+
∆
,
+
∆
,
+
)
(8)
∆
an
d
∆
are
deter
m
ined
as
belo
w:
{
∆
=
∆
=
(9)
(
,
)
are
vel
ocity
com
po
ne
nts
in
horizo
ntal
an
d
ver
ti
cal
direct
ion
s
at
the
po
i
nt
(
,
)
.
∆
,
∆
are
m
ov
e
m
ent
at
m
entioned
dir
ect
ion
res
pect
ively
.
More
over,
δt
is
the
sm
a
ll
-
tim
e
int
erv
al
bet
ween
two
seq
uen
ti
al
fra
m
es.
This
researc
h
s
plit
s
vid
e
o
of
e
cho
ca
r
diogram
bec
om
e
te
n
f
r
a
m
es.
All
f
ram
es
are
a
repres
entat
ion
of
on
e
hear
t
cy
cl
e
fr
om
dias
tole
to
syst
ole.
B
y
us
ing
O
ptica
l
Flow
,
determ
ining
bounda
r
y
of
endoca
rd
i
um
can
perform
ed
in a
ll
seq
ue
ntial
f
r
a
m
es w
it
hout repeat
ing t
he
in
it
ia
l pr
ocess
of
enh
a
ncem
ent an
d se
gm
entat
io
n.
Figure
3. Luca
s
-
Ka
na
de Op
ti
cal
Flow Esti
m
at
ion
Boun
dar
y
of
endoca
rd
i
um
that
has
been
obta
ined
from
se
gm
entat
ion
m
e
thod
will
be
good
feat
ur
e
s
in
Op
ti
cal
Flo
w.
O
ptica
l
Flo
w
of
Lucas
-
Ka
nad
e
will
find
identic
al
pix
el
in
nex
t
im
age
base
d
on
the
c
losest
intensit
y i
n
two
pix
el
s.
Fi
gur
e 3
sh
ows Opti
cal
Flow
es
tim
at
ion
can
fin
d m
os
t
o
f
pix
el
bet
ween
tw
o
i
m
ages in
on
e
set
se
qu
e
nt
ia
l
i
m
ages.
H
oweve
r,
the
s
uc
cess
rate
of
O
pt
ic
al
Flow
est
im
at
ion
dep
e
nd
s
on
th
e
qual
it
y
of
all
fr
am
es.
If
the
ne
xt
i
m
age
has
bad
q
ualit
y,
then
O
ptica
l
Flow
will
be
diff
i
cult
to
find
ide
nt
ic
al
pix
el
and
surel
y
m
any
of
nex
t
good
feat
ur
es
will
be
m
issed.
This
resea
rc
h
us
es
11
set
of
echo
ca
r
diogra
m
s
wh
ic
h
are
div
ide
d
beco
m
e
two
gro
up
cl
ass
of
norm
al
and
a
bnorm
al
.
Sp
eci
f
ic
al
ly
,
there
a
re
fi
ve
norm
al
an
d
si
x
a
bnorm
a
l
echo
ca
r
diogra
m
s.
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.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
289
–
1
296
1294
Ever
y
go
od
f
eat
ur
es
f
ound
ed
in
eve
ry
im
ages
will
be
extracte
d
in
to
necessa
ry
featur
e
for
cl
assifi
cat
ion
.
Ther
e
are
tw
o
featur
e
s
that
is
le
ng
t
h
of
m
ov
e
m
ent
and
fl
ow
of
directi
on
s.
T
ho
s
e
feat
ures
ar
e
basic sym
pto
m
for
t
he do
ct
or
to d
ia
gnos
e
h
e
art co
ndit
ion
t
hro
ugho
ut
wall
m
ot
ion
in
S
AX ech
ocardio
gra
m
.
3.
3
.
1
Le
n
gth
of
Movemen
t o
r D
ispl
acemen
t
Len
gth
of
m
ove
m
ent
is
a
le
ngth
betwee
n
t
w
o
points
from
i
m
age
A
to
im
a
ge
B
w
her
e
im
age
B
is
the
nex
t
good f
eat
ur
e im
age o
n
s
equ
e
ntial
ech
oc
ard
i
ogram
. P
oin
t i
n
i
m
age B is g
ood feat
ure
estim
at
ed
by O
ptica
l
Flow. T
he
e
qu
at
ion
of len
gth
of d
is
placem
ent is d
esc
ribe
d b
el
ow
:
=
√
(
+
1
−
)
2
+
(
+
1
−
)
2
(10)
Eq
uation
10
e
xpla
ins
dif
fer
e
nc
e
(
d)
is
obta
in
ed
from
sq
uar
e
d
root
of
su
m
of
squa
red
dif
f
eren
ce
of
x
and y.
3.
3
.
2
Fl
ow
Dir
ection
D
e
term
inat
i
on
Flow
Directi
on
is
a
featur
e
that
descr
ibes
a
directi
on
of
m
ov
em
ent
betwe
en
to
pix
el
.
T
he
re
are
two
kind
of
directi
on
s
that
is
un
c
hange
d,
i
nw
a
r
d
a
nd
outwa
rd.
U
nch
a
nged
pi
xel
is
tw
o
pix
e
ls
sta
y
in
one
place.
Inward
is
go
od
feat
ur
es
m
o
ving
insi
de
to
wards
the
ce
nt
er
point
of
e
ndoca
r
diu
m
.
Outwar
d
is
good
featur
e
s
m
ov
ing
ou
tsi
de
tow
a
r
d
f
r
om
the cente
r po
i
nt
o
f
end
ocardiu
m
.
(1)
(2)
(3)
(4)
(5)
(6)
Figure
4. Al
gorithm
o
f
Dete
r
m
inati
on
of
Flow Dir
ect
ion
Figure 4
s
hows
so
m
e
al
go
rith
m
s
to
determ
ine
kind o
f
flo
w
directi
on
be
tw
een
tw
o
se
que
ntial
i
m
ages.
flo
w
di
recti
on
determ
inati
on
us
es
first
bo
unda
ry
of
e
ndoca
r
diu
m
as
an
i
niti
al
co
ntour
.
T
her
e
are
six
al
gorithm
s d
escribe
d belo
w:
a)
Image
part
(
1
)
:
an
uncha
nge
d
pix
el
is
a
con
d
it
io
n
w
her
e
the
nex
t
est
i
m
at
ed
-
pix
el
(C)
do
e
s
no
t
cha
ng
e
against
pr
e
vi
ous p
i
xel (
B)
.
b)
Image
pa
rt
(
2):
a
n
ou
t
ward
pi
xel
is
a
c
ondit
ion
that
occurs
w
hen
the
ne
xt
est
i
m
at
ed
-
pixe
l
(C)
is
outsi
de
the init
ia
l co
nto
ur
wh
il
e t
he p
rev
i
ou
s
p
i
xel (B
)
is i
ns
i
de
the
init
ia
l con
t
our (A
).
c)
Image
part
(
3)
:
ano
ther
out
ward
pi
xel
is
a
conditi
on
th
at
al
so
occurs
wh
e
n
previ
ous
(B)
an
d
ne
xt
est
i
m
at
ed
-
pix
e
l
(C)
are
inside
the
init
ia
l
con
t
our
(
A)
,
bu
t
the
ne
xt
est
i
m
at
ed
-
pix
e
d
has
a
short
er
disp
la
cem
ent than t
he p
rev
i
ous
pix
el
.
d)
I
m
age
part
(
4
)
:
an
inwa
r
d
pix
el
is
a
con
diti
on
that
occ
urs
wh
e
n
previ
ous
(B)
and
nex
t
est
i
m
at
ed
-
pix
e
l
(C)
a
re
in
side
the
init
ia
l
co
nto
ur
(A),
but
th
e
ne
xt
est
im
a
t
ed
-
pix
e
d
has
a
longe
r
disp
la
c
e
m
ent
than
th
e
pr
e
vious
pix
el
.
e)
Image
part
(
5):
ano
t
her
in
wa
rd
pi
x
el
is
a
con
diti
on
that
al
so
occ
ur
s
wh
e
n
pr
evi
ous
pix
el
(B)
is
ou
tsi
de
the init
ia
l co
nto
ur a
nd the
n
e
xt esti
m
a
te
d
-
pi
xel (
C
)
are
insi
de
the
init
ia
l con
t
our (A
).
f)
Image
part
(
6
)
:
an
inwa
r
d
pix
el
is
a
con
diti
on
that
occ
urs
wh
e
n
previ
ous
(B)
and
nex
t
est
i
m
at
ed
-
pix
e
l
(C)
are
outsi
de
the
init
ia
l
cont
our
(
A)
,
but
the
ne
xt
est
i
m
ated
-
pix
e
d
ha
s
a
sh
ort
er
disp
la
ce
m
ent
than
th
e
pr
e
vious
pix
el
.
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
Card
i
ac
M
otio
ns
Cl
assi
fi
cation o
n Se
qu
e
ntial P
SAX
E
ch
oc
ar
di
ogram
(
A
dam
Shid
qul Azi
z
)
1295
Figure
5. Re
su
l
t of Flo
w Direc
ti
on
Determ
inati
on
Figure
5
s
ho
ws
the
res
ult
of
fl
ow
direct
ion
bet
wee
n
two
se
quentia
l
i
m
ag
es.
Every
directi
on
is
represe
nted
by
diff
ere
nt
col
or.
The
blue
pix
el
s
re
pr
ese
nt
inward
dir
e
ct
ion
an
d
the
red
pix
el
s
re
pr
ese
nt
ou
t
ward
dire
ct
ion.
The
re
su
lt
sh
owe
d
that
bi
t
of
pix
el
s
ca
n
be
tracke
d
co
rrec
tl
y.
Ho
we
ve
r,
the
res
ult
can
sti
ll
be use
d
as t
he ne
xt
good
feat
ur
es
.
3.
4
.
Clas
si
ficat
i
on
On
e
cy
cl
e
of
hear
t
sta
rted
f
r
om
diastole
to
syst
ole
is
rep
r
esented
by
10
fr
am
es.
So
,
th
ere
are
nin
e
changes
betwe
en
tw
o
c
onsec
utive
im
ages
duri
ng
one
cy
cl
e.
The
re
a
re
t
wo
featu
res
t
ha
t
will
be
ext
r
act
ed
in
on
e
cha
nge
tha
t
is
flow
direct
ion
(F)
an
d
le
ngth
of
dis
placem
ent
(L)
.
Th
er
efore,
i
n
one
c
yc
le
of
hear
t,
t
here
are
18
featu
res
that
will
be
ob
ta
ined
a
nd
will
be
us
ed
as
f
eat
ur
es
in
cl
as
sific
at
ion
m
et
h
od.
Ar
ti
fici
al
Neu
ral
Netw
ork
is
pro
po
s
ed
for
cl
ass
ify
ing
the
feat
ur
es
bec
om
e
t
wo
cl
asse
s
that
is
norm
al
and
abno
rm
al
.
Ab
norm
al
can
be divid
ed
beco
m
e thr
ee
he
art co
ndit
ion
t
hat is
dysk
ine
s
ia
, ak
ine
sia
, hypokine
sia
.
Figure
6. Pro
pose
d
Cl
assifi
ca
ti
on
Met
ho
d
4.
RESU
LT
A
N
D DIS
CUSSI
ON
Ther
e
a
re
11
s
et
s
of
f
ram
es
t
hat
su
c
ces
sf
ul
to
segm
ent
in
this
researc
h.
Th
os
e
dataset
s
are
div
i
de
d
into
tw
o
cl
asse
s
of
norm
al
and
ab
norm
al
.
Sp
eci
fical
ly
,
there
are
5
norm
al
dataset
s
and
6
ab
norm
al
dataset
s.
Pr
op
os
e
d
m
et
h
od
i
n
F
ig
ure
6
has
been
a
ppli
ed.
Fl
ow
d
irect
ion
h
as b
een obtai
ned
a
nd
t
he
sta
ti
sti
c
resu
lt
of
t
he
featur
e
d
e
scri
be
d
i
n Ta
ble 1.
Table
1.
Stat
ist
ic
o
f
Flo
w Dire
ct
ion
Co
n
d
itio
n
Flo
w I
n
(
in
wa
rd)
Flo
w Out (ou
tward
)
No
r
m
al
57%
43%
Ab
n
o
r
m
al
35%
65%
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.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
289
–
1
296
1296
Table
1
ex
plain
flo
w
directi
on
of
norm
al
and
ab
norm
al
hav
e
dif
fer
e
nt
re
su
lt
.
Norm
al
c
l
ass
has
the
per
ce
ntage
value
of
inwa
rd
f
low
hi
gh
e
r
tha
n
outwa
rd
fl
ow.
Ot
herwise,
abno
rm
al
cl
ass
has
the
pe
rce
ntage
value o
f
in
ward f
l
ow lo
wer t
ha
n ou
t
ward
flo
w.
T
hese
r
es
ults co
rr
es
pond
w
it
h
the e
xp
la
na
ti
on
in
secti
on
2
that
the
char
act
eris
ti
c
of
no
rm
al
hear
t
is
al
l
of
tissu
es
m
ov
ing
i
ns
ide
.
Table
1
sh
ows
di
ff
e
rence
per
centa
ge
valu
e
betwee
n
in
ward
flo
w
an
d
out
ward
flo
w
in
norm
al
c
la
ss
i
s
s
m
al
le
r
than
diff
e
re
nce
valu
e
in
ab
norm
al
cl
ass.
This is ca
us
e
d by the
re ar
e
sti
ll
w
r
ong deter
m
inati
on
in
f
lo
w direct
io
n.
Cl
assifi
cat
ion
i
s
pe
rfor
m
ed
by
us
i
ng
Ra
pi
dMiner
with
fe
w
of
cl
assifi
cat
io
n
m
et
ho
ds.
Ta
ble
2
s
how
s
the
com
par
iso
n
res
ult
of
ev
ery
m
et
ho
d.
P
ercenta
ge
is
ob
ta
ined
with
Lea
ve
O
ne
O
ut
(LOO)
validat
io
n
m
et
ho
d.
T
his
validat
io
n
m
e
t
hod
is
com
m
on
ly
us
ed
to
va
li
date
featur
e
in
m
edical
c
om
pu
ti
ng
case
.
It
is
ob
s
er
ved
that
neural
netw
ork
with
le
arn
in
g
rate
0.01
has
the
highest
pe
r
form
ance
with
value
81.
82%
.
The
seco
nd p
la
ce is
suppo
rt
vect
or m
achine w
it
h per
form
ance val
ue
is
72.73%
.
Table
2.
Cl
assi
ficat
ion
Res
ults
Metho
d
Res
u
lt (
%)
Near
est
Neig
h
b
o
r
(
N =
3
)
5
4
.55
%
Neu
ral
N
etwo
rk (
L
R = 0.0
1
)
8
1
.82
%
Naïv
e Bay
es
6
3
.64
%
Su
p
p
o
rt
Vector M
achi
n
e
7
2
.73
%
5.
CONCL
US
I
O
N
Diag
nose
of
ca
rd
ia
c
m
otion
s
by
ech
oca
rd
i
ogra
phy
ex
per
t
i
s
base
d
on
flo
w
of
directi
on
and
le
ngth
of
disp
la
cem
ent
of
ec
hocar
diogram
.
Lucas
-
Kan
a
de
O
ptica
l
Flow
is
ap
plied
in
feat
ure
extracti
on
t
o
est
i
m
at
e
good
feature
of
al
l
co
ns
ec
utive
ec
hocar
diogram
.
Ex
per
im
ent
in
F
ig
ur
e
3
s
h
ow
s
that
Lucas
-
Ka
na
de
Op
ti
cal
Flow
has
good
perform
ance
to
est
im
a
te
go
od
feat
ur
e
i
n
th
e
nex
t
im
age.
Flow
directi
on
can
be
deter
m
ined
alm
os
t
pr
eci
sel
y
with
pr
op
ose
d
al
gorithm
.
Len
gth
of
dis
pl
ace
m
ent
can
be
cal
culat
ed
by
us
in
g
eq
uat
ion
8.
Flow
Directi
on
an
d
le
ngth
of
dis
placem
ent
featur
e
s
has
good
e
nough
pe
rfor
m
ance
by
us
in
g
neural
ne
twork
.
Howe
ver,
op
ti
m
iz
at
ion
is
ne
eded
to
im
pr
ov
e
the
re
su
lt
.
This
researc
h
pro
ves
th
os
e
two
feat
ur
es
ca
n
be
use
d
to d
ia
gnos
e
car
diac d
ise
ase
c
om
pu
ta
ti
on
al
ly
.
REFERE
NCE
S
[1]
J.
G.
B
et
ts
et al
.
,
“
Anatom
y
and P
h
y
siolog
y
”
.
Ho
uston:
OpenStax
,
2016
.
[2]
W
orld
Hea
lt
h
Or
gani
z
at
ion
,
“
On
W
orld
Hea
rt
Da
y
W
HO
c
al
ls f
o
r
accelerate
d
a
ct
i
on
to
pr
eve
n
t
th
e
world’s l
e
adi
ng
globa
l
killer”,
20
18.
[Onlin
e]
.
Av
ai
l
abl
e
:
ht
tp:
/
/www
.
who.i
nt/
c
ard
iova
scul
ar_
dis
eas
es/e
n/. [Ac
c
essed:
06
-
Jul
-
2018]
.
[3]
P.
Torka
shv
and,
H.
Behna
m
,
and
Z.
San
i, “M
odifi
ed
Opti
ca
l
Flow
Te
chn
ique
for
C
ard
iac
Motions
Anal
y
sis
in
Ec
hoc
ard
iogr
ap
h
y
Im
age
s,”
vol
.
2,
no.
Jul
y
,
pp
.
1
–
8,
2012
.
[4]
Maity
,
A.
Pa
tt
an
ai
k,
S.
Sagn
ika, and
S.
Pani, “A
compara
ti
v
e
stu
d
y
on
appr
o
ac
h
e
s to
spec
k
le noi
s
e
red
u
ct
ion
in
images,
”
Proc.
-
1st I
nt. Conf. Co
m
put.
Intell
.
Ne
t
works
,
CINE
20
15,
no
.
Mar
ch, p
p.
148
–
155
,
201
5.
[5]
Y.
-
W
.
Song a
nd
S.
S.
Udpa
,
“
A
New Morphologi
cal
Approa
ch
fo
r
Reducing
Spe
c
kle
Noise
in
Ultr
asonic
Im
ag
es,
”
pp.
1397
–
1400
,
1997.
[6]
N.
Rajal
akshm
i,
K.
Nara
y
a
nan
,
a
nd
P.
Am
udhavalli
,
“
W
ave
let
-
bas
ed
weigh
te
d
m
edi
an
f
il
t
er
for
im
age
d
enoi
sing
of
MRI bra
in im
ag
es,
”
Indone
s.
J
.
El
e
ct
r.
Eng. Co
mput.
Sc
i.
,
vo
l. 1
0,
no
.
1
,
pp
.
201
–
206,
2018
.
[7]
S.
A.
Sa
le
em
an
d
T. A.
Raz
ak, “A
n
eff
e
ct
iv
e
noi
se
ada
p
ti
ve
m
edian
fi
lt
er
for
removing
high
densi
t
y
impuls
e
no
ise
s
in
co
lor
imag
es,
”
Int
.
J. E
l
ectr. Com
put.
Eng
.
,
v
ol.
6
,
no
.
2
,
pp
.
6
11
–
620,
2016
.
[8]
O.
Bobkova
,
V.
V.
Z
y
uz
in, S.
V
.
Pors
hnev,
and V.
V.
Bobkov
,
“
Expe
ri
enc
e
of
using of
m
a
chi
ne
l
ea
rning
m
et
hods
to
id
ent
if
y
th
e
l
e
ft
ven
tri
c
le re
gi
o
n
in ec
ho
ca
rdiog
rap
hic records,”
2016
IEEE
10
th Int.
Con
f. A
ppl
.
I
nf.
Comm
un.
Technol
.
,
pp
.
1
–
5,
2016
.
[9]
M.
S.
Mansoor
y
,
M.
As
htiy
a
ni, and H. T. N., “Ca
rdia
c
Motion
Ev
al
ua
ti
on
for
Dise
ase
Diagnosis
U
sing I
CA Ba
sis
Neura
l
N
et
work,
”
2009
In
t. A
ss
oc.
Comput
.
S
ci.
I
nf.
Te
chnol.
-
Sp
ring Conf
.
,
no
.
1
,
pp
.
496
–
500
,
2
009.
[10]
S.
Mansor a
nd
J.
A.
Nobl
e, “LOCA
L
W
ALL
M
OTION
CLASS
I
FICA
TION OF
STRESS
ECHOCA
RDIOGRAP
H
Y US
IN
G A HIDDEN MA
RKOV
MODEL
AP
PR
OA
C
H Sari
na
M
ansor
,
J Al
ison Nobl
e,
”
pp.
1295
–
1298
,
2008.
[11]
N.
G.
Pand
ia
n
,
D.
J.
Skorton
,
S.
M.
Col
li
ns,
H. L.
Fa
lset
t
i, E.
R
.
Burke, a
nd
R.
E
.
Kerb
er,
“
Hetr
o
gene
i
t
y
of
le
f
t
vent
ri
cul
ar
segm
ent
a
l
wal
l thic
k
e
ning
and
ex
cur
si
on
in two di
m
en
sional
ec
hoc
ard
i
ogra
m
s of
norm
al
hum
an
subjec
ts,
”
vo
l. Am
.
J.
Car
,
no
.
51
,
pp
.
1667
–
1673
,
1983.
[12]
C.
Otto
,
R
.
Sch
wae
gle
r
,
and
R
.
Free
m
a,
“
Te
x
tbo
ok
of
Cl
ini
c
al E
c
hoca
rdiogr
aph
y
”
.
2013
.
[13]
R.
Sigi
t, A.
R
.
B
ara
kbah
,
I
.
A
.
Su
li
stij
ono
,
and
A.
S.
Aziz,
“
Autom
at
i
c
c
ard
i
ac seg
m
ent
at
ion
using
tri
angle and
opti
c
al
f
low,
”
In
dones.
J
.
Elec
tr
.
Eng.
Comput
.
S
c
i.
,
vol
.
8
,
no
.
2
,
p
p.
315
–
326
,
201
7.
Evaluation Warning : The document was created with Spire.PDF for Python.