I
n
d
on
e
s
ian
Jou
r
n
al
of
E
lec
t
r
ical
E
n
gin
e
e
r
in
g
a
n
d
Com
p
u
t
e
r
S
c
ience
Vol.
25
,
No.
2
,
F
e
br
ua
r
y
2022
,
pp.
93
1
~
940
I
S
S
N:
2502
-
4752,
DO
I
:
10
.
11591/i
jee
c
s
.
v
25
.i
2
.
pp
931
-
940
931
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
e
e
c
s
.
iaes
c
or
e
.
c
om
C
la
ssi
f
y
a
r
r
h
y
t
h
m
ia
b
y u
si
n
g
2
D
s
p
e
c
t
r
al
i
m
ag
e
s an
d
d
e
e
p
n
e
u
r
al
n
e
t
w
o
r
k
T
r
an
Anh
Vu
1
,
Hoan
g
Qu
an
g
Hu
y
1
,
P
h
a
m
Duy
Kh
an
h
1
,
Ng
u
ye
n
T
h
i
M
i
n
h
H
u
ye
n
1
,
T
r
in
h
T
h
i
T
h
u
Uye
n
1
,
P
h
am
T
h
i
Vie
t
Hu
on
g
2
1
B
io
me
di
c
a
l
E
ngi
ne
e
r
in
g
D
e
pa
r
tm
e
nt
,
S
c
hool
of
E
le
c
tr
oni
c
s
a
n
d
T
e
le
c
omm
uni
c
a
ti
on
s
,
H
a
noi
U
ni
ve
r
s
it
y
of
S
c
ie
nc
e
a
nd
T
e
c
hn
ol
ogy,
H
a
noi
,
V
ie
tn
a
m
2
I
nt
e
r
na
ti
ona
l
S
c
hool
,
V
ie
tn
a
m
N
a
ti
ona
l
U
ni
ve
r
s
it
y,
H
a
noi
,
V
ie
tn
a
m
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
J
ul
21
,
2021
R
e
vis
e
d
De
c
8
,
2021
Ac
c
e
pted
De
c
1
8
,
2021
E
l
ec
t
ro
car
d
i
o
g
ram
(
E
CG
)
is
t
h
e
mo
s
t
c
o
mmo
n
met
h
o
d
fo
r
mo
n
i
t
o
r
i
n
g
t
h
e
w
o
r
k
i
n
g
of
t
h
e
h
eart
.
E
CG
s
i
g
n
al
is
the
b
a
s
i
s
to
d
et
ermi
n
e
n
o
rma
l
or
ab
n
o
rmal
r
h
y
t
h
m,
t
h
ere
b
y
h
el
p
i
n
g
to
accu
ra
t
el
y
d
i
a
g
n
o
s
e
card
i
o
v
as
c
u
l
ar
d
i
s
eas
e
s
.
T
h
eref
o
re,
an
au
t
o
mat
i
c
al
g
o
r
i
t
h
m
to
d
et
ec
t
an
d
d
i
a
g
n
o
s
e
a
b
n
o
rmal
h
eart
rh
y
t
h
m
s
is
es
s
en
t
i
a
l
.
T
h
ere
are
man
y
met
h
o
d
s
of
cl
as
s
i
f
y
i
n
g
arrh
y
t
h
m
i
as
u
s
i
n
g
mach
i
n
e
l
earn
i
n
g
al
g
o
ri
t
h
m
s
s
u
c
h
as
k
-
n
eares
t
n
ei
g
h
b
o
r
s
(K
N
N
),
s
u
p
p
o
r
t
v
ect
o
r
mach
i
n
es
(SV
M),
b
as
e
d
on
t
h
e
feat
u
res
ex
t
ra
ct
e
d
fro
m
t
h
e
reco
r
d
of
E
CG
s
i
g
n
al
.
A
ct
u
al
l
y
,
d
ee
p
l
ear
n
i
n
g
al
g
o
r
i
t
h
ms
ar
e
ev
o
l
v
i
n
g
an
d
h
i
g
h
l
y
effect
i
v
e
in
i
mag
e
an
a
l
y
s
i
s
an
d
p
ro
ces
s
i
n
g
.
In
t
h
i
s
res
earch
,
a
d
en
s
e
n
e
u
ra
l
n
e
t
w
o
rk
m
o
d
e
l
is
p
ro
p
o
s
ed
to
cl
as
s
i
f
y
n
o
rmal
an
d
ab
n
o
rmal
b
ea
t
s
.
In
p
u
t
E
CG
s
i
g
n
a
l
p
re
s
en
t
i
n
g
a
t
i
me
s
er
i
es
is
co
n
v
er
t
ed
i
n
t
o
2
-
D
s
p
ec
t
ral
i
mag
e
by
a
p
p
l
y
i
n
g
w
av
e
l
et
t
ran
s
fo
rm
.
Our
res
earch
is
e
v
al
u
at
e
d
b
as
e
d
on
u
s
i
n
g
t
h
e
Mas
s
ach
u
s
e
t
t
s
I
n
s
t
i
t
u
t
e
of
T
ec
h
n
o
l
o
g
y
-
Bet
h
Is
rae
l
H
o
s
p
i
t
a
l
(MIT
-
BIH
)
arrh
y
t
h
m
i
a
d
at
a
b
as
e.
The
accu
rac
y
of
the
cl
as
s
i
fi
ca
t
i
o
n
al
g
o
ri
t
h
m
we
emp
l
o
y
is
9
9
.
8
%
,
d
em
o
n
s
t
ra
t
i
n
g
t
h
e
mo
d
e
l
's
v
al
i
d
i
t
y
w
h
e
n
co
mp
are
d
to
o
t
h
er
rep
o
r
t
s
'
f
i
n
d
i
n
g
s
.
T
h
i
s
is
the
f
o
u
n
d
a
t
i
o
n
fo
r
o
u
r
al
g
o
r
i
t
h
m
to
p
r
o
v
e
it
can
be
u
t
i
l
i
ze
d
as
an
effi
ci
e
n
t
m
o
d
e
l
fo
r
cat
eg
o
r
i
zi
n
g
arrh
y
t
h
mi
a
u
s
i
n
g
E
CG
s
i
g
n
al
s
.
K
e
y
w
o
r
d
s
:
2
-
D
s
pe
c
tr
a
l
im
a
ge
C
onti
nuous
w
a
ve
let
tr
a
ns
f
or
m
De
e
p
ne
ur
a
l
ne
twor
k
E
C
G
s
ignal
Th
i
s
is
an
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
the
CC
BY
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
P
ha
m
T
hi
Vie
t
Huong
I
nter
na
ti
ona
l
S
c
hool
,
Vie
tnam
Na
ti
ona
l
Unive
r
s
it
y
144
Xua
n
T
huy
,
C
a
u
Gia
y
,
Ha
noi,
Vie
tnam
E
mail:
huongptv@i
s
vnu.
vn
1.
I
NT
RODU
C
T
I
ON
An
a
r
r
hythm
ia
[
1
]
is
an
e
lec
tr
ica
l
i
r
r
e
gular
it
y
of
the
he
a
r
t,
whic
h
can
be
a
pa
c
ing
or
e
lec
tr
ica
l
c
onduc
ti
on
a
nomaly
in
the
he
a
r
t
c
ha
mber
s
,
in
whic
h
the
he
a
r
tbea
t
is
i
r
r
e
gular
,
too
f
a
s
t
or
too
s
low.
An
a
r
r
hythm
ia
can
be
a
s
ympt
omatic
or
c
a
us
e
s
ympt
oms
s
uc
h
as
pa
lpi
tations
,
a
s
e
ns
e
that
the
he
a
r
t
is
be
a
ti
ng
too
quickly
or
ir
r
e
gular
ly,
or
a
br
e
a
k
be
twe
e
n
he
a
r
tbea
ts
[
2]
.
M
a
ny
c
a
s
e
s
of
s
e
ve
r
e
a
r
r
hythm
ias
c
a
us
e
th
e
pa
ti
e
nt
to
be
c
ome
dizz
y,
f
a
int
,
ha
ve
tr
ouble
br
e
a
thi
ng,
a
n
d
ha
ve
c
he
s
t
pa
in.
C
ompl
ica
ti
ons
can
oc
c
ur
s
uc
h
as
s
tr
oke
,
he
a
r
t
f
a
il
ur
e
,
or
s
udde
n
de
a
th.
Ac
c
or
ding
to
W
H
O
[
3]
,
c
a
r
diovas
c
ular
dis
e
a
s
e
s
a
r
e
the
c
a
us
e
of
th
e
lar
ge
s
t
mor
talit
y
in
the
wo
r
ld
(
mo
r
e
than
30%
)
,
highe
r
th
a
n
de
a
th
f
r
om
c
a
nc
e
r
.
It
is
e
s
ti
mate
d
that
each
ye
a
r
a
bout
17.
9
mi
ll
ion
pe
ople
wor
ldwide
die
f
r
om
c
a
r
diov
a
s
c
ular
dis
e
a
s
e
s
of
whic
h
85%
a
r
e
f
r
om
he
a
r
t
a
tt
a
c
k
a
nd
s
tr
oke
.
E
s
pe
c
ially
in
the
c
u
r
r
e
nt
s
it
ua
ti
on
of
C
OV
I
D
-
19
e
pidemic,
the
r
is
k
of
de
a
th
of
ten
f
oc
us
e
s
mainly
on
the
e
lder
ly
or
pa
ti
e
nts
with
unde
r
lyi
ng
medic
a
l
c
ondit
ions
including
c
a
r
diovas
c
ular
dis
e
a
s
e
.
E
lec
tr
oc
a
r
diogr
a
m
(
E
C
G)
is
a
c
ha
r
t
that
r
e
c
or
ds
the
e
lec
tr
ica
l
im
puls
e
s
ge
ne
r
a
ted
by
c
a
r
diac
mus
c
le
c
e
ll
thr
ough
e
lec
tr
ode
s
plac
e
d
in
the
body
.
T
he
E
C
G
s
ignals
a
r
e
dis
playe
d
in
a
1
-
D
time
s
e
r
ies
t
ha
t
he
lps
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
,
Vol.
25
,
No.
2
,
F
e
br
ua
r
y
20
22
:
931
-
940
932
tr
a
c
k
a
nd
de
tec
t
ir
r
e
gular
it
ies
in
the
he
a
r
t
r
hythm
ba
s
e
d
on
the
wa
ve
f
or
m
a
nd
the
f
r
e
que
nc
y
of
the
h
e
a
r
tbea
t.
E
lec
tr
oc
a
r
diogr
a
ms
can
be
us
e
d
to
diagnos
e
c
a
r
diovas
c
ular
pr
oblems
in
indi
viduals
.
E
lec
tr
oc
a
r
diogr
a
m
r
e
a
ding
is
a
dif
f
icult
tas
k
that
ne
e
ds
e
xpe
r
ienc
e
a
nd
tr
a
ini
ng.
T
he
s
pe
c
ialis
t
can
e
va
luate
if
a
c
li
nica
l
s
ympt
om
of
a
he
a
r
t
pr
oblem
is
pr
e
s
e
nt
ba
s
e
d
on
the
r
e
c
o
r
de
d
da
ta.
As
a
r
e
s
ult
,
identif
ying
c
a
r
diac
a
r
r
hyt
hmi
a
s
is
mos
tl
y
de
pe
nde
nt
o
n
the
knowle
dge
of
the
doc
t
or
,
a
nd
va
r
ious
doc
to
r
s
will
p
r
ovide
d
if
f
e
r
e
nt
o
utcome
s
.
F
ur
ther
mor
e
,
with
a
lengthy
ti
me
int
e
r
va
l
E
C
G
r
e
c
or
d,
young
medic
a
l
p
r
a
c
ti
ti
one
r
s
may
ove
r
l
ook
mild
s
ignals
of
c
a
r
diovas
c
ular
il
lnes
s
.
As
a
r
e
s
ult
,
we
r
e
quir
e
a
too
l
to
a
s
s
is
t
c
li
nicia
ns
in
the
a
na
lys
is
of
E
C
Gs
.
As
a
r
e
s
ult
,
we
r
e
quir
e
a
tool
to
a
s
s
is
t
c
li
nicia
ns
in
the
a
na
lys
is
of
E
C
Gs
.
In
whic
h
one
of
the
ke
y
f
a
c
tor
s
f
or
pr
ope
r
ly
diagnos
ing
he
a
r
t
-
r
e
late
d
il
lnes
s
e
s
is
the
c
a
tegor
iza
ti
on
of
a
bno
r
mal
he
a
r
t
be
a
ts
.
T
he
E
C
G
s
ignal
is
a
1
-
D
time
s
e
r
ies
that
can
be
pr
oc
e
s
s
e
d
a
nd
a
na
ly
z
e
d
a
utom
a
ti
c
a
ll
y
by
mac
hine
lea
r
ning
a
lgor
it
hms
.
F
ur
ther
mor
e
,
de
e
p
lea
r
ning
a
lgor
it
hms
ha
ve
r
e
c
e
ntl
y
be
e
n
de
mons
tr
a
ted
to
be
e
xtr
e
mely
e
f
f
icie
nt
in
the
pr
oc
e
s
s
ing
a
nd
c
a
tegor
iza
ti
on
of
2D
im
a
ge
s
.
De
e
p
lea
r
ning
a
lgor
it
hms
,
whic
h
a
r
e
a
s
ubs
e
t
of
mac
hine
lea
r
ning
,
r
e
ly
on
da
ta
to
unde
r
s
tand
ho
w
to
s
olve
pr
oblems
.
De
e
p
lea
r
ning
e
mpl
oys
the
ne
ur
a
l
ne
twor
k,
a
mul
ti
-
laye
r
e
d
s
tr
uc
tur
e
of
a
lgor
it
h
ms
.
Ar
ti
f
icia
l
ne
ur
a
l
ne
two
r
ks
of
f
e
r
unique
c
ha
r
a
c
ter
i
s
ti
c
s
that
a
ll
ow
de
e
p
lea
r
ning
models
to
a
c
c
ompl
is
h
tas
ks
that
mac
hine
lea
r
ning
models
ha
ve
li
mi
tations
.
T
he
r
e
ha
ve
be
e
n
s
e
ve
r
a
l
s
tudi
e
s
in
the
s
ubjec
t
of
a
utom
a
ted
c
a
tegor
iza
ti
on
a
r
r
hythm
ias
.
W
a
ve
mor
phologi
c
a
l
c
ha
r
a
c
ter
is
t
ics
[4
]
-
[
6]
,
as
we
ll
as
p
a
r
a
mete
r
s
s
uc
h
as
va
r
ianc
e
a
nd
s
tanda
r
d
de
viation
[
7
]
[
8]
,
ha
ve
be
e
n
e
xtr
a
c
ted
f
r
om
1
-
D
E
C
G
s
ignals
in
the
pa
s
t,
with
the
us
e
of
mac
hine
lea
r
ning
tec
hniques
s
uc
h
as
KNN,
s
uppor
t
ve
c
tor
mac
hine,
de
c
is
ion
tr
e
e
,
a
nd
r
a
ndom
f
or
e
s
t
[9
]
-
[
11]
.
In
or
de
r
to
e
xt
r
a
c
t
the
f
e
a
tur
e
s
or
nor
malize
da
ta
[
7]
mos
t
c
or
r
e
c
tl
y,
thes
e
tec
hniqu
e
s
r
e
quir
e
a
s
ignal
pr
e
p
r
oc
e
s
s
ing
s
tep
to
f
il
ter
noi
s
e
,
f
il
ter
ba
s
e
li
ne
dr
if
t
[1
]
-
[
9]
.
De
e
p
lea
r
n
ing
a
ppr
oa
c
he
s
a
r
e
incr
e
a
s
ingl
y
be
ing
us
e
d
in
im
a
ge
pr
oc
e
s
s
ing
a
nd
a
na
lys
is
with
gr
e
a
t
e
f
f
icie
nc
y
a
nd
a
c
c
ur
a
c
y
[
12
]
-
[
16]
.
T
he
ne
ur
a
l
ne
twor
k
model
may
ope
r
a
te
e
f
f
e
c
ti
ve
ly
with
mul
ti
dim
e
ns
ional
input
s
without
the
f
e
a
tur
e
e
xtr
a
c
ti
on
s
tep.
How
e
ve
r
,
be
c
a
us
e
the
output
of
a
1
-
D
input
s
ignal
is
les
s
r
e
li
a
ble
than
a
2
-
dim
e
ns
ional
input
,
de
e
p
lea
r
ning
models
of
ten
us
e
a
2D
pictur
e
as
their
input
.
A
pr
e
vious
s
tudy
[
17]
uti
l
ize
d
a
p
ictur
e
of
the
E
C
G
s
ignal
that
ha
d
not
be
e
n
t
r
a
ns
f
or
med,
whic
h
obtaine
d
99.
21%
a
c
c
ur
a
c
y
r
a
te.
T
he
input
E
C
G
(1
-
D)
time
s
e
r
ies
s
ignal
may
be
c
onve
r
ted
in
to
a
2
-
D
s
pe
c
tr
a
l
pictur
e
us
ing
tr
a
ns
f
or
mation
tec
hniques
.
S
ome
r
e
c
e
nt
r
e
s
e
a
r
c
h
us
e
d
a
tr
a
ns
f
or
med
2D
s
pe
c
tr
a
l
im
a
ge
as
the
input
of
ne
ur
a
l
ne
twor
k
c
las
s
if
ica
ti
on
model
f
or
3
c
las
s
e
s
c
las
s
if
ica
ti
on
[
15]
with
the
a
c
c
ur
a
c
y
of
98.
7%
,
a
nd
8
c
las
s
e
s
c
las
s
if
ica
ti
on
[
16]
a
c
hieving
an
a
c
c
ur
a
c
y
of
99
.
11
%
.
Our
c
ontr
ibut
ion
f
oc
us
e
s
pr
im
a
r
il
y
on
a
pp
r
oa
c
he
s
f
or
e
xtr
a
c
ti
ng
c
ha
r
a
c
ter
is
ti
c
s
f
r
om
an
E
C
G
s
ignal
a
nd
then
doing
c
las
s
if
ica
ti
on
us
ing
s
tanda
r
d
mac
h
ine
lea
r
ning
models
,
whic
h
yielde
d
e
nc
our
a
ging
r
e
s
ult
s
.
In
a
dd
i
ti
on
to
ob
taining
f
e
a
tur
e
s
in
the
time
domain
[4
]
-
[
6]
,
s
ome
a
ppr
oa
c
he
s
e
mpl
oy
tr
a
ns
f
or
mation
a
l
gor
it
hms
s
uc
h
as
the
F
ou
r
ier
tr
a
ns
f
or
m
a
nd
the
wa
ve
let
tr
a
ns
f
or
m
to
e
xtr
a
c
t
mor
e
c
ha
r
a
c
ter
is
ti
c
s
of
the
s
ignal
in
the
f
r
e
que
nc
y
domain
[7
]
-
[
9]
.
How
e
ve
r
,
if
li
ne
a
r
f
e
a
tur
e
s
a
r
e
pr
e
s
e
nt,
pe
r
f
or
mi
ng
f
e
a
tur
e
e
xtr
a
c
ti
on
is
e
x
tr
e
mely
dif
f
icult
a
nd
mi
ght
r
e
duc
e
the
c
las
s
if
ica
ti
on
mode
l's
e
f
f
e
c
ti
ve
ne
s
s
.
F
ur
ther
mo
r
e
,
if
the
da
taba
s
e
s
ize
is
huge
,
s
tanda
r
d
mac
hine
lea
r
ning
m
e
thods
will
not
a
tt
a
in
the
opti
mum
e
f
f
icie
nc
y.
He
a
r
tbea
t
c
las
s
if
ica
ti
on
a
ppr
oa
c
he
s
ba
s
e
d
on
de
e
p
lea
r
ning
a
lgor
it
hms
ha
ve
r
e
c
e
ntl
y
be
e
n
pr
e
s
e
nted
as
a
s
olut
ion
to
thi
s
c
ha
ll
e
nge
.
T
he
input
pr
oc
e
s
s
ing
of
the
ne
ur
a
l
ne
twor
k
is
a
ls
o
take
n
int
o
c
ons
ider
a
ti
on,
in
a
ddit
ion
to
the
us
a
ge
of
mul
ti
-
laye
r
ne
ur
a
l
ne
two
r
k
models
with
s
upe
r
ior
im
a
ge
c
las
s
if
ica
ti
on
e
f
f
icie
nc
y.
Only
in
f
or
mation
a
bout
the
wa
ve
f
or
ms
is
obtaine
d
whe
n
the
c
las
s
if
ica
ti
on
mo
de
l's
input
is
an
im
a
ge
of
a
1D
E
C
G
s
ignal
[
17
]
,
a
nd
thi
s
inf
or
mation
is
loca
ll
y
r
e
pr
e
s
e
nted
on
the
im
a
ge
,
w
hich
mea
ns
that,
a
s
ide
f
r
om
the
mor
phology
of
th
e
s
ignal,
the
r
e
maining
int
e
r
va
ls
on
the
im
a
ge
c
ontain
no
inf
or
mation.
As
a
r
e
s
ult
,
s
e
ve
r
a
l
a
ppr
oa
c
he
s
ha
ve
tr
a
ns
f
or
med
a
1D
E
C
G
s
ignal
int
o
a
2D
s
pe
c
tr
a
l
pictur
e
us
ing
tr
a
ns
f
or
mation
a
lgor
it
hms
[
15
]
,
[
16]
.
T
he
s
ignal's
tempor
a
l
a
nd
f
r
e
que
nc
y
domain
inf
or
mati
on
a
r
e
both
c
ontaine
d
in
2D
s
pe
c
tr
a
l
im
a
ge
s
.
C
li
p
ping
the
s
ignal
s
e
gments
at
s
pe
c
if
ic
int
e
r
va
ls
f
r
om
the
be
g
inni
ng
to
the
c
onc
lus
ion
of
the
s
ignal,
on
the
oth
e
r
ha
nd,
mi
ght
pr
oduc
e
une
ve
nne
s
s
in
the
c
las
s
if
ica
ti
on
of
t
he
be
a
ts
in
the
2D
pictu
r
e
s
.
We
s
ugge
s
t
a
ne
w
a
ppr
oa
c
h
in
thi
s
s
tudy
that
is
ba
s
e
d
on
the
e
volut
ion
of
e
a
r
li
e
r
methods
,
whic
h
a
r
e
:
a)
E
qua
ll
y
c
ut
the
s
ignal
s
e
gments
by
taking
the
s
a
me
int
e
r
va
l
on
both
s
ides
of
the
R
pe
a
ks
.
b)
Us
ing
the
c
onti
nuous
wa
ve
let
tr
a
n
s
f
or
m
(
C
W
T
)
to
c
onve
r
t
the
s
ignal
s
e
gment
inf
or
mation
int
o
2D
s
pe
c
tr
a
l
pictur
e
s
f
r
om
the
c
li
ppe
d
s
ignal
s
e
gments
.
c)
T
he
de
ns
e
ne
ur
a
l
ne
twor
k
model
is
ut
il
ize
d
to
identif
y
he
a
r
tbea
ts
us
ing
thes
e
im
a
ge
s
as
input
.
d)
Our
r
e
s
e
a
r
c
h
pa
pe
r
is
or
ga
nize
d.
P
a
r
t
II
e
xplains
our
r
e
s
e
a
r
c
h
method.
P
a
r
t
I
I
I
pr
e
s
e
nts
e
xpe
r
im
e
ntal
r
e
s
ult
s
,
a
nd
P
a
r
t
IV
c
onc
ludes
the
pa
pe
r
.
2.
RE
S
E
AR
CH
M
E
T
HO
D
2.
1.
Dat
ab
as
e
s
In
thi
s
s
tudy,
we
e
va
luate
the
e
f
f
e
c
ti
ve
ne
s
s
of
our
a
lgor
it
hm
ba
s
e
d
on
us
ing
the
M
I
T
-
B
I
H
a
r
r
hythm
ia
da
taba
s
e
[
1]
whic
h
is
publi
s
he
d
on
P
h
ys
ionet.
or
g.
T
he
da
taba
s
e
include
s
48
E
C
G
r
e
c
or
ds
,
each
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
I
S
S
N:
2502
-
4752
C
las
s
if
y
A
r
r
hy
thmia
by
us
ing
2D
Spe
c
tr
al
I
mage
s
and
De
e
p
N
e
ur
al
N
e
tw
or
k
(
T
r
an
A
nh
V
u
)
933
s
li
ghtl
y
mor
e
than
30
m
inut
e
s
long.
I
ns
ide
the
da
taba
s
e
,
each
r
e
c
or
d
of
each
dif
f
e
r
e
nt
pa
ti
e
nt
is
ba
ndwidth
f
il
ter
e
d
at
the
f
r
e
que
nc
y
r
a
nge
of
0.
1
-
100
Hz
a
nd
digi
ti
z
e
d
at
a
f
r
e
que
nc
y
of
360
Hz
.
T
he
r
e
c
or
ds
we
r
e
labe
led
with
the
R
pe
a
ks
a
nd
the
pos
it
ion
of
the
pe
a
ks
that
a
ppe
a
r
e
d
to
be
an
a
r
r
hythm
ia.
T
he
r
e
f
or
e
,
the
e
f
f
e
c
ti
ve
ne
s
s
of
our
c
las
s
if
ica
ti
on
model
can
be
a
s
s
e
s
s
e
d.
T
he
thr
e
e
c
omponents
of
an
E
C
G
a
r
e
de
picte
d
in
F
ig
ur
e
1.
T
he
P
wa
ve
r
e
pr
e
s
e
nts
a
tr
ia
de
polar
i
z
a
ti
on;
the
QR
S
c
ompl
e
x,
whic
h
r
e
pr
e
s
e
nts
ve
ntr
icula
r
de
polar
iza
ti
on;
a
nd
the
T
wa
ve
,
whic
h
s
hows
ve
ntr
icle
r
e
polar
iza
ti
on.
F
igur
e
1.
E
C
G
of
a
he
a
r
t
in
nor
mal
s
inus
r
hythm
2.
2.
B
lock
d
iagram
F
ig
ur
e
2
s
hows
the
im
pleme
ntation
of
the
a
lgor
it
h
m.
T
he
E
C
G
s
ignal
is
c
las
s
if
ied
int
o
two
c
las
s
e
s
:
nor
mal
a
nd
a
bnor
mal
.
F
ir
s
tl
y,
we
us
e
the
pa
c
k
a
ge
wa
ve
f
or
m
-
da
taba
s
e
(
W
F
DB
)
f
o
r
loading
E
C
G
a
nd
a
nnotations
f
r
om
the
da
tas
e
t
in
Ke
r
a
s
f
r
a
mew
o
r
k
in
P
ython
.
E
a
c
h
type
is
identif
ied
by
a
s
y
mbol
that
c
or
r
e
s
ponds
to
the
number
of
be
a
ts
.
T
he
s
ignal's
pe
a
k
R
may
be
de
ter
mi
ne
d
us
ing
the
pe
a
k's
c
ha
r
a
c
ter
is
ti
c
s
.
How
e
ve
r
,
we
u
ti
li
z
e
R
pe
a
k
va
lue
labe
led
in
the
d
a
tas
e
t
f
or
s
ignal
pr
oc
e
s
s
ing
s
im
pli
c
it
y.
T
he
s
e
c
ond
s
tage
is
s
ignal
s
e
gmenta
ti
on,
whic
h
invol
ve
s
taking
an
e
qua
l
time
-
s
e
r
ies
s
ignal
be
f
or
e
a
nd
a
f
ter
R
pe
a
ks
.
W
a
ve
let
tr
a
ns
f
or
mation
is
us
e
d
to
c
onve
r
t
thes
e
s
e
gmente
d
E
C
G
s
ignal
int
e
r
va
ls
f
r
om
time
s
e
r
ies
to
2
-
dim
e
ns
ional
s
pe
c
tr
a
l
pictur
e
s
.
In
thi
s
s
tep,
we
us
e
the
s
c
a
logr
a
m
tool
to
r
e
p
r
e
s
e
nt
the
tr
a
ns
f
or
med
im
a
ge
s
.
T
h
i
s
im
a
ge
da
tas
e
t
is
then
us
e
d
as
the
input
of
the
c
las
s
if
ica
ti
on
de
e
p
ne
ur
a
l
ne
twor
k
model.
T
he
im
a
ge
is
divi
de
d
int
o
tr
a
ini
ng
s
e
ts
a
nd
va
li
da
ti
on
s
e
ts
to
pe
r
f
or
m
c
las
s
if
ica
ti
on
a
nd
e
va
luate
our
model.
F
igur
e
2.
Ar
r
hythm
ia
de
tec
ti
on
d
iagr
a
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
,
Vol.
25
,
No.
2
,
F
e
br
ua
r
y
20
22
:
931
-
940
934
2.
3.
P
r
op
os
e
d
m
e
t
h
od
2.
3.
1.
P
r
e
-
p
r
oc
e
s
s
in
g
To
c
onve
r
t
the
s
ignal
E
C
G
int
o
im
a
ge
s
,
we
f
ir
s
t
ne
e
d
to
output
the
R
pe
a
k
in
the
s
ignal
r
e
c
or
ds
,
whic
h
is
us
e
d
to
r
e
pr
e
s
e
nt
a
he
a
r
tbea
t
.
We
de
tec
t
R
pe
a
ks
by
s
ignal's
de
tec
ti
ng
pe
a
ks
a
lgor
it
hm
f
r
om
the
S
c
ipy
pa
c
ka
ge
ba
s
e
d
on
pr
ope
r
ti
e
s
of
E
C
G
s
ignal's
pe
a
k.
T
he
a
lgor
it
hm
pe
r
f
o
r
ms
f
indi
ng
a
ll
loca
l
m
a
xim
a
in
the
da
ta
s
e
r
ies
by
s
im
ply
c
ompar
ing
ne
ighbor
ing
va
lues
.
T
he
n
s
e
lec
t
R
pe
a
k
s
as
a
s
ubs
e
t
of
thes
e
pe
a
ks
ba
s
e
d
on
the
c
ondit
ions
of
the
pe
a
k's
pr
ope
r
ti
e
s
.
T
he
E
C
G
s
ignal
in
the
da
ta
s
e
t
ha
s
a
s
a
mpl
ing
f
r
e
que
nc
y
of
360
Hz
,
so
the
s
ignal
will
be
r
e
pr
e
s
e
nted
in
the
time
domain
by
time
index
unit
with
each
time
index
e
qua
l
to
1/360
s.
In
ou
r
c
a
s
e
,
we
de
tec
t
R
pe
a
ks
by
c
hoos
i
ng
the
maximal
va
lue
of
the
E
C
G
s
ignal
in
the
m
ini
mum
hor
izonta
l
dis
tanc
e
of
150
indexe
s
be
twe
e
n
ne
ighbor
ing
pe
a
ks
.
F
igu
r
e
3
pr
e
s
e
nts
the
R
pe
a
ks
of
the
E
C
G
s
ignal.
F
igur
e
3.
R
pe
a
ks
de
tec
ted
by
the
s
c
ipy
pa
c
ka
ge
In
thi
s
s
tudy,
f
o
r
s
im
pli
c
it
y,
we
uti
li
z
e
R
pe
a
ks
va
lue,
whic
h
ha
s
a
lr
e
a
dy
be
e
n
pos
it
ioned
in
the
da
taba
s
e
.
F
r
om
the
da
ta
s
e
t,
9000
he
a
r
tbea
ts
a
r
e
r
a
ndomi
z
e
d
with
e
qua
l
number
s
of
nor
mal
a
nd
ir
r
e
gular
be
a
ts
,
f
or
a
tot
a
l
of
4500
be
a
ts
.
F
r
om
the
pos
it
ion
of
R
pe
a
k
of
each
be
a
t,
it
will
go
ba
c
kwa
r
d
a
nd
f
or
wa
r
d
to
each
s
ide
a
s
ignal
int
e
r
va
l
of
the
length
e
qua
l
200
i
nde
xe
s
.
2.
3.
2.
Gener
at
ion
of
2
-
D
s
p
e
c
t
r
al
im
age
s
T
he
or
e
ti
c
a
ll
y,
a
ny
s
ignal
can
be
de
c
ompos
e
d
int
o
its
c
omponent
s
ignals
in
both
the
tempor
a
l
a
nd
the
f
r
e
que
nc
y
domain.
T
he
r
e
f
or
e
,
the
E
C
G
s
ignal
can
be
a
na
lyze
d
int
o
c
omponent
s
ignals
to
de
ter
mi
ne
whe
n
a
nd
at
wha
t
f
r
e
que
nc
y
the
a
r
r
hy
thm
ia
oc
c
ur
s
[
18]
.
T
he
wa
ve
let
tr
a
ns
f
or
m
f
ul
f
il
ls
thes
e
two
r
e
quir
e
ments
.
It
make
s
the
c
onti
nuous
s
ignal
x(
t)
f
r
om
one
dim
e
ns
i
on
in
two
a
2D
s
pa
c
e
de
f
ined
as
(
1)
,
S
(
a
,
b)
=
1
√
∫
(
)
(
−
)
+
∞
−
∞
(
1)
whe
r
e
a
a
nd
b
a
r
e
the
s
c
a
le
f
a
c
tor
a
nd
s
hif
t
tr
a
ns
lation
a
ppli
e
d
in
the
c
onti
nuous
pa
r
e
nt
wa
ve
let
Φ
(
t
)
.
In
thi
s
s
tep,
the
c
onti
nuous
wa
ve
let
tr
a
ns
f
or
m
(
C
W
T
)
is
a
ppli
e
d
to
ge
ne
r
a
te
an
E
C
G
s
ignal
int
o
a
2D
s
pe
c
tr
um.
De
pe
nding
on
the
s
tudy
goa
l,
s
e
ve
r
a
l
types
of
wa
ve
let
tr
a
ns
f
or
ms
can
be
e
mpl
oye
d
to
a
na
lyze
E
C
G
s
ignals
.
F
or
e
xa
mpl
e
,
to
r
e
move
E
C
G
ba
s
e
li
ne
,
we
us
e
f
i
ve
wa
ve
let
tr
a
ns
f
or
m
f
a
mi
li
e
s
with
a
tot
a
l
of
14
wa
ve
let
c
onf
igur
a
ti
ons
:
Da
ube
c
hies
,
C
oif
lets
,
S
yml
e
ts
,
F
e
jer
-
Kor
ovkin,
a
nd
M
e
ye
r
[
19]
.
T
he
va
r
iatio
n
of
the
a
bnor
mal
he
a
r
tbea
t
is
a
non
-
s
tationar
y
s
ignal
so
it
is
s
uit
a
ble
to
c
hoos
e
M
or
let
as
the
mot
he
r
wa
ve
let
be
c
a
us
e
of
it
s
a
na
lys
is
a
ppli
c
a
ti
on
on
dis
c
r
im
inate
a
r
r
hyth
mi
a
s
in
the
E
C
G
s
ignal
[
20]
.
In
theor
y,
t
he
M
or
let
wa
ve
let
is
the
mos
t
popular
c
ompl
e
x
wa
ve
let
us
e
d
in
pr
a
c
ti
c
e
[
21]
a
nd
is
de
f
ined
as
(
2)
,
(
)
=
1
√
4
(
0
−
−
0
2
2
)
−
2
2
(
2
)
whe
r
e
0
is
the
c
e
ntr
a
l
f
r
e
que
nc
y
of
the
mot
he
r
wa
v
e
let
a
nd
s
e
t
by
de
f
a
ult
f
or
each
wa
ve
let
with
r
e
s
pe
c
ti
ve
va
lue.
T
he
s
e
c
ond
ter
m
in
the
b
r
a
c
ke
t
is
c
or
r
e
c
t
f
or
the
non
-
z
e
r
o
mea
n
of
the
c
ompl
e
x
s
inus
oid
of
the
f
ir
s
t
ter
m
a
nd
can
be
ne
gli
gibl
e
if
0
>
5.
W
it
h
r
e
ga
r
ds
to
the
s
c
a
le
f
a
c
tor
,
the
s
ize
of
the
pict
ur
e
's
he
ight
ha
s
an
im
pa
c
t
on
the
r
e
s
olut
ion
of
2D
s
pe
c
tr
a
l
pictur
e
s
.
On
the
other
s
ide,
the
s
ignal
le
ngth
c
or
r
e
s
ponds
to
the
wid
th
s
ize
of
s
pe
c
tr
a
l
im
a
ge
s
.
We
c
hos
e
the
f
e
a
tur
e
s
c
ondit
ion
that
o
f
f
e
r
the
output
pi
c
tur
e
s
s
ize
with
the
be
s
t
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
by
r
unning
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
I
S
S
N:
2502
-
4752
C
las
s
if
y
A
r
r
hy
thmia
by
us
ing
2D
Spe
c
tr
al
I
mage
s
and
De
e
p
N
e
ur
al
N
e
tw
or
k
(
T
r
an
A
nh
V
u
)
935
the
e
xpe
r
im
e
nt
m
ult
ipl
e
ti
mes
with
va
r
ied
s
ize
s
of
2D
output
im
a
ge
s
a
nd
c
ompar
ing
the
c
las
s
if
ica
ti
on
r
e
s
ult
s
in
pa
r
t
3.
T
he
s
c
a
le
f
a
c
tor
is
s
e
lec
ted
to
t
r
a
ns
f
or
m
li
ne
a
r
it
y
f
r
om
1
to
150
.
T
he
s
c
a
le
r
e
pr
e
s
e
nts
the
number
of
ti
mes
the
wa
ve
let
is
s
tr
e
tche
d.
T
he
lar
ge
r
s
c
a
le,
the
mor
e
s
tr
e
tche
d
wa
ve
let
is
,
a
nd
the
mor
e
s
e
ns
it
ive
it
is
to
lowe
r
s
ignal
f
r
e
que
nc
ies
.
F
or
be
tt
e
r
vis
ua
li
z
a
ti
o
n,
the
s
c
a
logr
a
m
is
us
e
d
to
ge
ne
r
a
te
a
nd
s
how
the
2D
s
pe
c
tr
um
f
or
the
C
W
T
.
T
he
C
W
T
c
oe
f
f
icie
nts
of
a
s
ignal
a
r
e
take
n
in
a
bs
olut
e
va
lue
a
nd
its
gr
a
ph
is
plot
ted.
F
ig
ur
e
4
p
r
e
s
e
nts
an
E
C
G
s
ignal
a
nd
i
ts
s
c
a
logr
a
m.
F
igur
e
4.
An
E
C
G
s
ignal
a
nd
its
s
c
a
logr
a
m
In
the
s
c
a
logr
a
m
outpu
t,
the
pe
r
iod
in
the
ve
r
ti
c
a
l
is
de
f
ined
by
(
3
)
,
P
e
r
iod
=
(
3)
whe
r
e
s
is
the
s
c
a
le,
a
nd
b
(
0
)
is
the
c
e
ntr
a
l
f
r
e
qu
e
nc
y
us
e
d
to
buil
d
the
c
hos
e
n
wa
ve
let.
E
a
c
h
hor
i
z
ontal
f
e
a
tur
e
may
be
r
e
ga
r
de
d
as
a
f
r
e
que
nc
y
of
the
tot
a
l
s
ignal,
a
nd
ther
e
is
no
c
onti
nuous
li
ne
in
the
outp
ut
im
a
ge
to
indi
c
a
te
that
the
f
r
e
que
nc
ies
a
r
e
not
time
-
c
o
ns
is
tent.
T
he
s
c
a
logr
a
m
is
a
two
-
dim
e
n
s
ional
pi
c
tur
e
of
150x401
pixels
,
wi
th
150
r
e
pr
e
s
e
nti
ng
the
number
of
s
c
a
les
us
e
d
in
the
wa
ve
let
tr
a
ns
f
or
m
a
nd
401
b
e
ing
the
number
of
indi
c
e
s
in
the
E
C
G
s
ignal
da
ta.
2.
3.
3
.
De
e
p
n
e
u
r
al
n
e
t
wor
k
c
on
s
t
r
u
c
t
ion
Data
s
e
tup:
Our
pr
e
pa
r
e
d
da
tas
e
t
now
is
divi
de
d
int
o
the
tr
a
ini
ng
s
e
t
a
nd
tes
t
s
e
t
with
a
3/1
r
a
ti
o;
the
tr
a
in/
tes
t
s
pli
ts
a
r
e
ge
ne
r
a
ted
to
e
ns
ur
e
that
the
r
e
is
no
ove
r
lap
be
twe
e
n
the
two
s
e
ts
.
M
ode
l
c
ons
tr
uc
ti
on:
Af
ter
the
E
C
G
s
ignal
is
c
on
ve
r
ted
int
o
a
s
pe
c
tr
a
l
im
a
ge
,
each
im
a
ge
will
be
a
s
s
igned
a
va
lue
of
0
(
nor
mal
pe
a
k)
or
1
(
a
bn
or
mal
pe
a
k)
.
To
c
las
s
if
y
thes
e
im
a
ge
s
,
we
us
e
a
ba
s
ic
c
onvolut
ional
ne
ur
a
l
ne
twor
k
.
T
he
s
ize
of
the
model
input
is
a
x
150
x
401
whe
r
e
a
is
the
number
of
150x401
pixel
im
a
ge
s
e
nter
e
d
int
o
the
model.
T
he
hidden
l
a
ye
r
is
c
ons
tr
uc
ted
by
two
de
ns
e
laye
r
s
with
500
a
nd
100
node
s
s
e
pa
r
a
tely.
T
he
output
laye
r
ha
s
two
ne
ur
on
s
f
or
the
f
inal
c
las
s
if
ier
that
a
r
e
e
it
he
r
0
(
nor
mal
p
e
a
k)
or
1
(
a
bnor
mal
pe
a
k)
.
Ac
ti
va
ti
on
f
unc
ti
on:
T
he
a
c
ti
va
ti
on
f
unc
ti
on
may
be
us
e
d
to
c
omput
e
the
output
of
each
node
in
an
a
r
ti
f
icia
l
ne
u
r
a
l
ne
twor
k
given
a
c
oll
e
c
ti
on
of
im
a
g
e
input
s
.
T
he
r
e
c
ti
f
ied
li
ne
a
r
a
c
ti
va
ti
on
f
unc
t
ion
(
R
e
L
U)
is
us
e
d
in
the
hidden
laye
r
a
nd
is
s
ugge
s
ted
as
the
d
e
f
a
ult
f
or
mul
ti
laye
r
pe
r
c
e
ptr
on
(
M
L
P
)
a
nd
c
onvo
lut
ional
ne
ur
a
l
ne
twor
ks
(
C
NN
s
)
[
22]
.
If
the
input
is
pos
it
i
ve
,
it
wil
l
be
di
r
e
c
tl
y
output
;
other
wis
e
,
it
will
be
z
e
r
o.
We
us
e
the
S
o
f
tm
a
x
a
c
ti
va
ti
on
f
unc
ti
on
in
the
output
laye
r
to
ge
ne
r
a
te
a
ve
c
tor
of
c
las
s
if
ica
ti
on
p
r
ob
a
bil
it
ies
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
,
Vol.
25
,
No.
2
,
F
e
br
ua
r
y
20
22
:
931
-
940
936
with
the
pr
oba
bil
it
ies
of
each
va
lue
pr
opor
ti
ona
l
to
the
r
e
lative
s
c
a
le
of
each
va
lue
(0
a
nd
1)
in
th
e
ve
c
tor
us
ing
(
4)
,
(
)
=
∑
=
1
(
4
)
whe
r
e
K
is
the
numbe
r
of
c
las
s
e
s
.
By
a
pplyi
ng
the
s
tanda
r
d
e
xpone
nti
a
l
f
unc
ti
on
to
each
e
leme
nt
a
nd
no
r
malize
s
thes
e
va
lues
by
divi
ding
by
the
s
um
of
a
ll
thes
e
e
xpone
nti
a
ls
,
it
e
ns
ur
e
s
that
each
c
omponent
will
be
in
the
int
e
r
va
l
[
0,
1]
a
nd
the
s
um
of
c
omponents
output
ve
c
tor
is
1.
Af
t
e
r
that,
the
c
las
s
with
higher
pr
oba
bil
it
ies
va
lue
will
be
labe
led
as
the
im
a
ge
’
s
type.
F
ig
ur
e
5
pr
e
s
e
nts
the
d
e
ns
e
ne
ur
a
l
ne
twor
k
(
DN
N)
model
a
r
c
hit
e
c
tu
r
e
.
F
igur
e
5.
T
he
de
ns
e
ne
ur
a
l
ne
twor
k
(
DN
N)
model
a
r
c
hit
e
c
tur
e
C
os
t
function
:
T
he
goa
l
of
the
c
os
t
f
unc
ti
on
is
to
c
ompr
omi
s
e
the
a
c
c
ur
a
c
y
of
the
a
lgor
it
hm
,
by
taking
the
a
ve
r
a
ge
e
r
r
or
be
twe
e
n
the
p
r
e
diction
r
e
s
ult
a
nd
the
pe
r
f
or
manc
e
r
e
s
ult
.
In
theor
y,
the
r
e
a
r
e
a
va
r
iety
of
c
os
t
f
unc
ti
ons
that
can
be
us
e
d.
In
our
pa
pe
r
,
we
c
hoos
e
s
pa
r
s
e
c
a
t
e
gor
ica
l
c
r
os
s
-
e
ntr
opy
as
c
os
t
f
unc
ti
on
be
c
a
us
e
it
s
a
ve
s
memor
y
a
nd
c
omput
a
ti
on
ti
me.
I
ns
tea
d
of
us
ing
an
e
nti
r
e
ve
c
tor
,
it
jus
t
uti
li
z
e
s
a
s
ingl
e
int
e
ge
r
.
T
he
c
r
os
s
-
e
ntr
opy
los
s
be
twe
e
n
the
labe
ls
a
nd
our
r
e
s
ult
s
is
c
a
lcula
ted
with
the
(
5
)
,
C
=
−
1
∑
(
[
(
∗
(
)
+
(
1
−
)
(
1
−
)
]
)
=
1
(
5)
whe
r
e
C
is
the
c
os
t
to
be
mi
n
im
ize
d,
n
is
the
num
be
r
of
t
r
a
ini
ng
point
s
,
y
is
the
tar
ge
t
va
lue,
N
is
th
e
number
of
the
c
las
s
e
s
,
c
is
the
index
of
the
c
las
s
,
a
nd
a
is
the
a
c
tual
va
lue.
We
us
e
the
s
tocha
s
ti
c
gr
a
dient
de
s
c
e
nt
(
S
GD
)
opti
m
ize
r
f
o
r
tr
a
ini
ng
our
model.
It
e
va
luat
e
s
the
e
r
r
or
g
r
a
dient
f
o
r
the
c
ur
r
e
nt
s
tate
of
the
mo
de
l
us
ing
the
tr
a
ini
ng
da
tas
e
t,
then
upda
tes
the
we
ight
s
of
ou
r
model
via
ba
c
kpr
opa
ga
ti
on.
3.
RE
S
UL
T
S
AND
DI
S
CU
S
S
I
ON
3.
1.
Clas
s
if
icat
ion
r
e
s
u
lt
T
he
two
pa
r
a
mete
r
s
in
ou
r
method
that
a
f
f
e
c
t
the
r
e
s
ult
dir
e
c
tl
y
a
r
e
s
c
a
le
of
wa
ve
let
tr
a
ns
f
or
m
a
nd
the
int
e
r
va
l
of
s
ignal
f
or
each
s
pe
c
tr
a
l
im
a
ge
.
T
he
lar
ge
s
c
a
le
can
o
f
f
e
r
the
model
h
igh
s
e
ns
it
ivi
ty,
bu
t
it
take
s
a
long
time
to
c
las
s
if
y
the
da
ta
a
nd
the
s
pe
e
d
of
t
he
pr
oc
e
s
s
is
ve
r
y
s
low.
T
he
s
ignal
s
e
gment
int
e
r
va
ls
a
r
e
s
im
il
a
r
.
L
ong
int
e
r
va
l
c
a
r
r
ies
mor
e
inf
o
r
mation
of
E
C
G
s
ignal,
but
it
a
ls
o
take
s
mor
e
time
a
nd
de
c
r
e
a
s
e
s
the
s
pe
e
c
h
of
the
p
r
oc
e
s
s
.
W
it
h
the
c
a
pa
c
it
y
of
our
s
e
tup
s
ys
tem,
we
ha
ve
to
t
r
a
de
of
f
be
twe
e
n
the
s
c
a
les
a
nd
the
int
e
r
va
l
va
lues
.
T
a
ble
1
r
e
pr
e
s
e
nts
the
pa
r
a
mete
r
s
a
nd
the
c
or
r
e
s
ponding
tes
ti
ng
ti
me
.
We
ha
ve
the
be
s
t
a
c
c
ur
a
c
y
with
a
C
W
T
s
c
a
le
of
150
a
nd
an
E
C
G
s
ignal
s
e
gment
int
e
r
va
l
of
401
indi
c
e
s
.
T
he
high
s
e
ns
it
ivi
ty
of
the
tr
a
ns
f
or
mation
model
is
s
hown
by
the
s
c
a
le
va
lue
of
150.
T
he
s
ignal
int
e
r
va
l
f
r
om
R
pe
a
ks
is
401
indexe
s
be
f
or
e
a
nd
a
f
ter
that
the
r
e
is
a
de
qua
te
in
f
or
mation
in
the
pr
e
s
e
nt
pe
r
iod
a
nd
c
ompar
ing
the
c
ur
r
e
nt
p
e
r
iod
to
the
be
f
or
e
a
nd
a
f
ter
pe
r
iods
.
T
he
a
c
c
ur
a
c
y
of
the
model
is
99.
8
%
a
f
ter
r
unn
ing
10
e
poc
hs
us
ing
an
opti
mi
z
e
r
of
s
tocha
s
ti
c
gr
a
dient
de
s
c
e
nt
a
nd
c
omput
ing
the
los
s
with
a
s
pa
r
s
e
c
a
tegor
ica
l
c
r
os
s
-
e
ntr
opy.
In
the
las
t
s
tep,
we
us
e
the
c
onf
us
ion
matr
ix
to
e
va
luate
our
model.
T
he
c
onf
us
ion
matr
ix
a
ll
ows
us
to
e
va
luate
the
c
la
s
s
if
ica
ti
on
model
vis
ua
ll
y.
E
a
c
h
r
ow
r
e
pr
e
s
e
nts
the
a
c
tual
or
tr
ue
v
a
lue,
a
nd
each
c
olum
n
r
e
pr
e
s
e
nts
the
pr
e
dicte
d
v
a
lue;
we
ne
xt
c
ompar
e
the
a
c
tual
a
nd
pr
e
dicte
d
va
lues
f
or
each
c
las
s
.
T
he
diagona
l
va
lues
a
r
e
the
bigg
e
s
t
in
an
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
I
S
S
N:
2502
-
4752
C
las
s
if
y
A
r
r
hy
thmia
by
us
ing
2D
Spe
c
tr
al
I
mage
s
and
De
e
p
N
e
ur
al
N
e
tw
or
k
(
T
r
an
A
nh
V
u
)
937
e
f
f
icie
nt
model,
e
quivale
nt
to
the
number
of
p
r
e
dicte
d
va
lues
e
qua
l
to
the
a
c
tu
a
l
va
lue.
T
he
va
lues
in
the
matr
ix
a
r
e
then
no
r
malize
d
to
a
r
a
nge
of
0
to
1,
wi
th
1
be
ing
the
de
s
ir
e
d
va
lue
in
the
diagona
l
c
e
ll
s
.
F
ig
ur
e
6
pr
e
s
e
nts
the
c
onf
us
ion
matr
ix
f
or
the
pr
opos
e
d
c
las
s
if
ica
ti
on
model.
T
a
ble
1
.
T
a
ble
of
pa
r
a
mete
r
s
a
nd
r
e
s
ult
s
in
tes
ti
ng
ti
mes
S
c
a
le
I
nt
e
r
va
l
S
a
mpl
e
s
A
c
c
ur
a
c
y
50
401
9
,
000
99.16%
100
201
9
,
000
99.51%
100
201
9
,
000
99.73%
100
601
9
,
000
99.69%
150
201
9
,
000
99.20%
150
401
9
,
000
99.81%
F
igur
e
6.
C
onf
us
ion
matr
ix
f
o
r
the
pr
opos
e
d
c
las
s
if
ica
ti
on
model
Our
model
ha
s
a
c
onf
us
ion
matr
ix
with
c
e
ll
s
in
diagona
l
e
qua
l
1,
that
s
how
the
a
mount
s
of
a
c
tual
a
bnor
mal
be
a
ts
a
nd
pr
e
dicte
d
a
bnor
mal
be
a
ts
a
lm
os
t
s
im
il
a
r
.
T
he
lea
r
ning
c
ur
ve
in
F
ig
ur
e
7
s
h
ows
the
e
f
f
e
c
ti
ve
ne
s
s
of
thi
s
c
las
s
if
ica
ti
on
model.
T
he
lea
r
ning
c
ur
ve
s
hows
the
gr
a
phs
of
the
va
lue
of
los
s
f
unc
ti
on
a
nd
a
c
c
ur
a
c
y
of
t
r
a
ini
ng
s
e
t
a
nd
va
li
da
ti
on
s
e
t
dur
ing
the
c
las
s
if
ica
ti
on
ti
me
.
In
ou
r
model
,
we
r
e
c
e
i
ve
good
r
e
s
ult
s
in
both
tr
a
ini
ng
s
e
t
a
nd
va
li
da
ti
on
s
e
t.
A
f
ter
the
f
i
r
s
t
e
poc
h,
the
a
c
c
ur
a
c
y
of
the
t
r
a
ini
ng
s
e
t
a
nd
va
li
da
ti
on
s
e
t
a
r
e
ve
r
y
high
a
nd
s
table
,
a
ppr
oxi
mate
ly
1.
C
ontr
a
r
y
to
the
a
c
c
ur
a
c
y,
the
va
lues
of
the
los
s
f
unc
ti
on
of
two
s
e
ts
a
r
e
ve
r
y
low
a
nd
s
table
a
f
ter
the
f
ir
s
t
e
poc
h,
a
pp
r
oxim
a
tely
0.
T
ha
t
pr
oo
f
ou
r
model
is
not
ove
r
f
it
t
ing
or
unde
r
f
it
ti
ng.
F
igur
e
7.
L
e
a
r
ning
c
ur
ve
s
3.
2
.
Dis
c
u
s
s
ion
To
r
e
c
ognize
the
e
f
f
e
c
ti
ve
ne
s
s
of
the
method
whi
c
h
us
e
s
wa
ve
let
tr
a
ns
f
or
m
to
c
onve
r
t
E
C
G
s
ignal
to
2
-
D
s
pe
c
tr
a
l
im
a
ge
s
then
c
las
s
if
ica
ti
on
by
de
ns
e
ne
ur
a
l
ne
twor
k
model,
we
c
ompar
e
it
with
other
methods
a
ls
o
de
tec
ti
ng
a
r
r
hythm
ias
a
nd
us
ing
M
I
T
-
B
I
H
d
a
taba
s
e
.
We
c
ompar
e
d
the
model
of
a
utom
a
ti
c
a
r
r
hythm
ia
c
las
s
if
ica
ti
on
ba
s
e
d
on
E
C
G
s
ignaling
with
othe
r
r
e
c
e
nt
models
a
s
s
hown
in
T
a
ble
2
.
Our
a
ve
r
a
ge
a
c
c
ur
a
c
y,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
,
Vol.
25
,
No.
2
,
F
e
br
ua
r
y
20
22
:
931
-
940
938
s
e
ns
it
ivi
ty,
s
pe
c
if
icity,
pr
e
c
is
ion
va
lue,
in
tu
r
n,
r
e
a
c
he
d
99.
8
%
,
99.
7%
,
99
.
8%
,
99
.
8%
r
e
s
pe
c
ti
ve
ly,
de
mons
tr
a
ted
s
upe
r
ior
pe
r
f
or
manc
e
whe
n
c
ompar
e
d
to
pr
e
vious
a
lgor
it
hms
that
pr
oduc
e
d
two
c
las
s
e
s
in
the
f
ive
ini
ti
a
l
models
.
Ou
r
model
ha
s
the
highes
t
a
ve
r
a
ge
a
c
c
ur
a
c
y
of
the
a
lgor
it
hms
c
ompar
e
d.
T
his
s
hows
the
s
upe
r
ior
it
y
of
de
e
p
lea
r
ning
al
gor
i
thm
s
c
ompar
e
d
to
mac
hine
lea
r
ning
a
lgor
i
thm
s
[
9]
,
[
23]
,
[2
4
].
W
i
th
other
models
us
ing
C
NN
or
long
s
hor
t
ter
m
memo
r
y
(
L
S
T
M
)
[2
5
],
ge
ne
r
a
l
s
pa
r
s
e
d
ne
ur
a
l
ne
twor
k
(
GSN
N)
[
26]
,
r
a
dial
ba
s
is
f
unc
ti
on
(
R
B
F
)
[
27]
,
ou
r
model
ha
s
be
tt
e
r
r
e
s
ult
s
,
as
our
input
to
ne
ur
a
l
ne
twor
k
mo
de
l
is
2D
s
pe
c
tr
a
l
im
a
ge
s
whic
h
ha
ve
inf
or
mation
in
both
time
a
nd
f
r
e
que
nc
y
domain
.
One
of
the
pos
s
ibl
e
f
a
c
tor
s
a
f
f
e
c
ti
ng
the
f
inal
r
e
s
ult
is
the
les
s
e
r
number
of
he
a
r
tbea
ts
we
uti
li
z
e
f
or
tr
a
ini
ng
a
nd
tes
ti
ng
ou
r
de
e
p
ne
ur
a
l
ne
tw
or
k
model.
W
he
n
we
de
ve
lop
ou
r
model
to
be
a
mul
ti
c
las
s
c
las
s
if
ica
ti
on
model,
the
r
e
s
ult
s
of
s
e
ve
n
models
with
output
of
mo
r
e
than
two
c
las
s
e
s
s
how
that
our
model
may
be
a
good
im
pr
ove
ment.
T
he
de
tec
ti
on
of
a
r
r
hythm
ias
on
hour
ly
long
E
C
G
r
e
c
or
ds
is
time
-
c
ons
umi
ng
a
nd
r
e
quir
e
s
the
e
xa
mi
ne
r
to
pa
y
c
los
e
a
tt
e
nti
on
.
It
is
f
e
a
s
ibl
e
to
im
pr
ove
the
pe
r
f
or
manc
e
of
medic
a
l
pr
of
e
s
s
ionals
by
guidi
ng
the
obs
e
r
ve
r
to
a
na
lyze
noti
c
e
a
ble
a
nomalies
us
ing
a
utom
a
ted
c
a
tegor
iza
ti
on
methods
.
As
a
r
e
s
ult
,
the
diagnos
is
a
nd
tr
e
a
tm
e
nt
of
c
a
r
d
iovas
c
ular
dis
or
de
r
s
in
the
c
li
nic
may
be
done
f
a
s
ter
a
nd
mo
r
e
e
f
f
icie
ntl
y.
T
a
ble
2.
C
ompar
is
on
be
twe
e
n
the
pr
opos
e
d
model
a
nd
other
s
tate
-
of
-
the
-
a
r
t
E
C
G
c
las
s
if
ica
ti
on
tec
hniques
Y
e
a
r
s
M
ode
l
C
la
s
s
A
c
c
ur
a
c
y
%
S
pe
c
if
ic
it
y
%
S
e
ns
it
iv
it
y
%
P
r
e
c
is
io
n
%
F1
s
c
or
e
2018
S
V
M
[
23]
2
96%
-
-
-
-
2018
KNN
[
9]
2
97.5%
-
-
-
-
2019
C
N
N
[2
5
]
2
97.2%
98.7%
93.8%
96.8%
-
2019
L
S
T
M
[2
5
]
2
71.4%
50.1%
93.6%
64.2%
-
2019
S
V
M
[2
4
]
2
98.3%
97.5%
99.1%
-
98.3%
2021
P
r
opos
e
d
mode
l
(
D
N
N
)
2
99.8%
99.8%
99.7%
99.8%
-
2020
G
S
N
N
[
26
]
5
98%
-
-
98%
98%
2016
S
V
M
-
RBF
[
27]
5
98.91%
97.85%
98.91%
-
-
2019
F
a
s
te
r
R
-
C
N
N
[
17]
5
99.21%
99.45%
98.06%
-
-
2020
C
N
N
[
28]
5
98.33%
99.09%
98.33%
98.34%
-
2020
L
S
T
M
[
29]
5
99.37%
99.14%
94.89%
96.73%
95.77%
2019
C
N
N
[
30]
5
99%
-
-
-
-
2020
C
N
N
[
16]
8
99.11%
99.61%
97.91%
98.58%
98%
2021
C
N
N
[
15]
3
98.7%
-
-
-
-
4.
CONC
L
USI
ON
In
thi
s
pa
pe
r
,
we
s
how
how
to
us
e
a
de
ns
e
ne
ur
a
l
ne
twor
k
model
to
de
tec
t
a
r
r
hythm
ias
f
r
om
E
C
G
da
ta
r
e
c
or
dings
.
An
a
c
c
ur
a
te
taxonomy
of
E
C
G
s
ignals
pr
ovides
an
e
xc
e
ll
e
nt
f
ounda
ti
on
f
o
r
c
a
r
dio
va
s
c
ular
dis
e
a
s
e
diagnos
is
a
nd
pr
ognos
is
.
Our
a
ppr
oa
c
h
is
unique
in
that
it
us
e
s
the
W
a
ve
let
tr
a
ns
f
o
r
m
to
tu
r
n
a
one
-
dim
e
ns
ional
E
C
G
s
ignal
int
o
two
-
dim
e
ns
ional
s
pe
c
tr
a
l
pictur
e
s
,
whic
h
a
r
e
then
us
e
d
as
input
to
a
c
las
s
if
ica
ti
on
model.
W
he
n
c
ompar
e
d
to
methods
that
int
e
gr
a
te
f
e
a
tur
e
e
xtr
a
c
ti
on
a
nd
c
ur
r
e
nt
mac
hine
lea
r
ning
tec
hnologi
e
s
,
the
ne
ur
a
l
ne
two
r
k
model
h
a
s
s
hown
be
ne
f
icia
l
in
e
nha
nc
ing
the
a
c
c
ur
a
c
y
of
he
a
r
tbea
t
diagnos
e
s
.
RE
F
E
RE
NC
E
S
[
1]
C
.
A
nt
z
e
le
vi
tc
h
a
nd
A
.
B
ur
a
s
hni
kov,
“
O
ve
r
vi
e
w
of
ba
s
i
c
me
c
ha
ni
s
ms
of
C
a
r
di
a
c
A
r
r
hyt
hmi
a
,”
in
C
ar
di
a
c
E
le
c
tr
ophy
s
io
lo
gy
C
li
ni
c
s
, vol
. 3, no. 1, pp. 23
-
45, 2011, doi:
10.1016/j
.c
c
e
p.2010
.10.012
.
[
2]
G.
B.
M
oody
a
nd
R.
G.
M
a
r
k,
"
T
he
im
pa
c
t
of
th
e
M
I
T
-
B
I
H
A
r
r
hyt
hmi
a
D
a
ta
ba
s
e
,"
in
I
E
E
E
E
ngi
ne
e
r
in
g
in
M
e
di
c
in
e
and
B
io
l
ogy
M
agaz
in
e
,
vol
.
20,
no.
3,
pp.
45
-
50,
M
a
y
-
J
une
2001,
doi
:
10.11
09/
51.932724.
[
3]
W
H
O
,
“
C
a
r
di
ova
s
c
ul
a
r
di
s
e
a
s
e
s
(
C
V
D
s
)
,”
w
ho.i
nt
,
ht
tp
s
:/
/ww
w
.w
ho.i
nt
/e
n/
ne
w
s
-
r
oom/
f
a
c
t
-
s
he
e
ts
/d
e
t
a
il
/
c
a
r
di
ova
s
c
ul
a
r
-
di
s
e
a
s
e
s
-
(
c
vds
)
(
a
c
c
e
s
s
e
d
D
e
c
.
2021
)
[
4]
M
.
H
a
mm
a
d,
A.
M
a
he
r
,
K.
W
a
ng,
F.
J
ia
ng,
a
nd
M.
A
mr
a
ni
,
"
D
e
te
c
ti
on
of
a
bnor
ma
l
he
a
r
t
c
ondi
ti
ons
ba
s
e
d
on
c
ha
r
a
c
te
r
is
ti
c
s
of
E
C
G
s
ig
na
ls
,"
M
e
as
ur
e
m
e
nt
,
v
ol
.
125,
pp
.
634
-
644,
2018
,
doi
:
10.1016/j
.me
a
s
ur
e
me
nt
.2018.05.033
.
[
5]
M.
K.
S
e
na
pa
ti
,
M.
S
e
n
a
pa
ti
,
a
nd
S.
M
a
k
a
,
"
C
a
r
di
a
c
a
r
r
hyt
hmi
a
c
la
s
s
if
ic
a
ti
on
of
E
C
G
s
ig
na
l
us
in
g
mor
phol
ogy
a
nd
he
a
r
t
be
a
t
r
a
te
,"
2014
F
our
th
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
A
dv
anc
e
s
in
C
om
put
in
g
and
C
om
m
uni
c
at
io
ns
,
2014,
pp.
60
-
63,
doi
:
10.1109/I
C
A
C
C
.2014.20.
[
6]
P.
C
ha
z
a
l,
M.
O
'
D
w
ye
r
,
a
nd
R.
B.
R
e
il
ly
,
"
A
ut
oma
ti
c
c
la
s
s
if
i
c
a
ti
on
of
he
a
r
tb
e
a
ts
u
s
in
g
E
C
G
mor
phol
ogy
a
nd
he
a
r
tb
e
a
t
in
te
r
va
l
f
e
a
tu
r
e
s
,"
in
I
E
E
E
T
r
ans
ac
ti
ons
on
B
io
m
e
di
c
al
E
ngi
ne
e
r
in
g
,
vol
.
51
,
no.
7,
pp.
1196
-
1206,
J
ul
y
2004
,
doi
:
10.1109/T
B
M
E
.2004.827359
.
[
7]
H
.
M.
R
a
i,
A
.
T
r
iv
e
di
,
a
nd
S
.
S
hukl
a
,
"
E
C
G
s
ig
na
l
pr
oc
e
s
s
in
g
f
or
a
bnor
ma
li
ti
e
s
de
te
c
ti
on
us
in
g
mul
ti
-
r
e
s
ol
ut
io
n
w
a
ve
le
t
tr
a
ns
f
or
m
a
nd
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
k
c
la
s
s
if
ie
r
,
"
M
e
as
ur
e
m
e
nt
,
vol
.
46
,
no.
9
,
pp.
3238
-
3246,
2013
,
doi
:
10.1016/j
.me
a
s
ur
e
me
nt
.2013.05.021.
[
8]
A
.
D
ik
e
r
,
D
.
A
vc
i,
E
.
A
vc
i,
a
nd
M
.
G
e
di
kpi
na
r
,
"A
ne
w
te
c
hni
que
f
or
E
C
G
s
ig
na
l
c
la
s
s
if
ic
a
ti
on
ge
ne
ti
c
a
lg
or
it
hm
W
a
ve
le
t
K
e
r
ne
l
e
xt
r
e
me
le
a
r
ni
ng
ma
c
hi
ne
,
"
O
pt
ik
,
v
ol
.
180,
pp
.
46
-
55,
2019
,
doi
:
10.1016/j
.i
jl
e
o.2018.11.065
.
[
9]
C.
V
e
nka
te
s
a
n,
P.
K
a
r
th
ig
a
ik
uma
r
,
a
nd
R.
V
a
r
a
th
a
r
a
ja
n,
"A
nove
l
L
M
S
a
lg
or
it
hm
f
or
E
C
G
s
ig
na
l
pr
e
pr
oc
e
s
s
in
g
a
nd
KNN
c
la
s
s
if
ie
r
-
ba
s
e
d
a
bnor
ma
li
ty
de
te
c
ti
on,"
M
ul
ti
m
e
di
a
T
ool
s
and
A
ppl
ic
at
io
ns
,
vol
.
77,
pp
.
10365
-
10374
,
2018
,
doi
:
10.1007/s
11042
-
018
-
5762
-
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
I
S
S
N:
2502
-
4752
C
las
s
if
y
A
r
r
hy
thmia
by
us
ing
2D
Spe
c
tr
al
I
mage
s
and
De
e
p
N
e
ur
al
N
e
tw
or
k
(
T
r
an
A
nh
V
u
)
939
[
10]
C.
V
e
nka
te
s
a
n,
P.
K
a
r
th
ig
a
ik
uma
r
,
A.
P
a
ul
,
S.
S
a
th
e
e
s
kuma
r
a
n
,
a
nd
R.
K
um
a
r
,
"
E
C
G
S
ig
na
l
P
r
e
pr
oc
e
s
s
in
g
a
nd
S
V
M
C
la
s
s
if
ie
r
-
B
a
s
e
d
A
bnor
ma
li
ty
D
e
te
c
ti
on
in
R
e
mot
e
H
e
a
lt
hc
a
r
e
A
ppl
ic
a
ti
ons
,"
in
I
E
E
E
A
c
c
e
s
s
,
vol
.
6,
pp.
9767
-
9773,
2018,
doi
:
10.1109/AC
C
E
S
S
.2018.2794346.
[
11]
P.
S
hi
mpi
,
S.
S
ha
h,
M.
S
hr
o
f
f
,
a
nd
A.
G
odbole
,
"A
ma
c
hi
ne
le
a
r
ni
ng
a
ppr
oa
c
h
f
or
th
e
c
la
s
s
if
ic
a
ti
on
of
c
a
r
di
a
c
a
r
r
hyt
hmi
a
,"
2
017
I
nt
e
r
nat
io
nal
C
onf
e
r
e
n
c
e
on
C
om
put
in
g
M
e
th
odol
ogi
e
s
and
C
om
m
uni
c
at
io
n
(
I
C
C
M
C
)
,
2017,
pp.
603
-
607,
doi
:
10.1109/I
C
C
M
C
.2017.8282537.
[
12]
E.
I
z
c
i,
M.
A.
O
z
de
mi
r
,
M.
D
e
gi
r
me
nc
i
,
a
nd
A.
A
ka
n,
"
C
a
r
d
ia
c
A
r
r
hyt
hmi
a
D
e
t
e
c
ti
on
f
r
om
2D
E
C
G
I
ma
ge
s
by
U
s
in
g
D
e
e
p
L
e
a
r
ni
ng
T
e
c
hni
que
,"
2019
M
e
di
c
al
T
e
c
hnol
ogi
e
s
C
ongr
e
s
s
(
T
I
P
T
E
K
N
O
)
,
2019,
pp.
1
-
4,
doi
:
10.1109/T
I
P
T
E
K
N
O
.2019.8895011.
[
13]
G.
S
w
a
pna
,
P.
S
oma
n
k,
a
nd
R.
V
in
a
ya
kuma
r
,
"
A
ut
oma
te
d
D
e
te
c
ti
on
of
C
a
r
di
a
c
A
r
r
hyt
hmi
a
us
in
g
D
e
e
p
le
a
r
ni
ng
T
e
c
hni
que
s
,
"
P
r
oc
e
di
a
C
om
put
e
r
S
c
ie
nc
e
,
vol
.
132,
pp.
1192
-
1201,
2018
,
do
i:
10.1016/j
.pr
oc
s
.2018.05.034.
[
14]
R
.
R
ohma
nt
r
i
an
d
N
.
S
ur
a
nt
ha
,
"
A
r
r
hyt
hmi
a
C
la
s
s
if
ic
a
ti
on
u
s
in
g
2D
C
onvolut
io
na
l
N
e
ur
a
l
N
e
twor
k
,
"
I
nt
e
r
nat
io
nal
J
our
n
al
of
A
dv
anc
e
d
C
om
put
e
r
S
c
ie
nc
e
and
A
ppl
ic
at
io
ns
(
I
J
A
C
SA
)
,
vol
.
11
, no.
4,
pp.
201
-
208,
2020
,
doi
:
10.14569/I
J
A
C
S
A
.2020.0110427
.
[
15]
R
.
F
.
O
la
nr
e
w
a
ju
,
S.
N
.
I
br
a
hi
m,
A
.
L
.
A
s
na
w
i
,
a
nd
H
.
A
lt
a
f
,
“
C
la
s
s
if
ic
a
ti
on
of
E
C
G
s
ig
na
ls
f
or
de
te
c
ti
on
of
a
r
r
hyt
hmi
a
a
nd
c
onge
s
ti
ve
h
e
a
r
t
f
a
il
ur
e
ba
s
e
d
on
c
ont
in
uou
s
w
a
v
e
le
t
tr
a
ns
f
o
r
m
a
nd
de
e
p
ne
ur
a
l
ne
twor
ks
,
”
I
ndone
s
ia
n
J
ou
r
nal
of
E
le
c
tr
ic
al
E
ngi
ne
e
r
in
g
and
C
om
put
e
r
S
c
ie
nc
e
(
I
J
E
E
C
S)
,
v
ol
.
22
, no.
3,
J
une
2021,
pp.
1520
-
1528
,
doi
:
10.11591/i
je
e
c
s
.v22.i3.pp1520
-
1528
.
[
16]
A.
U
ll
a
h,
S.
A
nw
a
r
,
M.
B
il
a
l
,
a
nd
R.
M.
M
e
hmood
,
“
C
la
s
s
if
ic
a
ti
on
of
A
r
r
hyt
hmi
a
by
u
s
in
g
D
eep
L
e
a
r
ni
ng
w
it
h
2D
E
C
G
sp
e
c
tr
a
l
i
ma
ge
r
e
pr
e
s
e
nt
a
ti
on
,
”
R
e
m
ot
e
Se
n
s
in
g
,
v
ol
.
12,
1685
,
2020,
doi
:
10.3390/r
s
12101685.
[
17]
Y.
J
i,
S.
Z
ha
ng
,
a
nd
W.
X
ia
o,
"
E
le
c
tr
oc
a
r
di
ogr
a
m
C
la
s
s
if
ic
a
ti
on
B
a
s
e
d
on
F
a
s
te
r
R
e
gi
ons
w
it
h
C
onvolut
io
na
l
N
e
ur
a
l
N
e
two
r
k
,
"
Se
ns
or
s
(
B
as
e
l)
,
vol
.
19
, no.
11
, p.
2558,
2019
,
doi
:
10.3390/s
19112558.
[
18]
C.
H
e
r
f
f
a
nd
D.
J.
K
r
us
i
e
ns
ki
,
“
E
xt
r
a
c
ti
ng
F
e
a
tu
r
e
s
f
r
om
T
im
e
S
e
r
ie
s
,”
I
n:
K
ubbe
n
P
.,
D
umont
ie
r
M
.,
D
e
kke
r
A.
(
e
ds
)
F
undame
nt
al
s
of
C
li
ni
c
al
D
at
a
Sc
ie
nc
e
.
Spr
in
g
e
r
,
C
ham
.
,
pp.
85
-
100,
2019,
doi
:
10.1007/978
-
3
-
319
-
99713
-
1_7.
[
19]
C.
C
he
n
a
nd
F.
R.
T
s
ui
,
“
C
ompa
r
in
g
di
f
f
e
r
e
nt
w
a
ve
le
t
tr
a
ns
f
o
r
ms
on
r
e
movi
ng
e
le
c
tr
oc
a
r
di
ogr
a
m
ba
s
e
li
ne
w
a
nde
r
s
a
nd
s
pe
c
ia
l
tr
e
nds
,
”
B
M
C
M
e
di
c
al
I
nf
or
m
at
ic
s
and
D
e
c
is
io
n
M
ak
in
g
,
vol
.
20
, no.
S
11
,
p.
343
,
2020
,
doi
:
10.1186/s
12911
-
020
-
01349
-
x
.
[
20]
O.
P.
N
e
to
e
t
al
.,
“
M
or
le
t
w
a
v
e
le
t
tr
a
ns
f
or
ms
of
h
e
a
r
t
r
a
te
va
r
ia
bi
li
ty
f
or
a
ut
onomi
c
ne
r
vous
s
y
s
te
m
a
c
ti
vi
ty
,”
A
ppl
ie
d
and
C
om
put
at
io
nal
H
ar
m
oni
c
A
nal
y
s
is
,
vol
40
, no.
1,
pp.
200
-
206,
2016,
doi
:
10.1016/j
.a
c
ha
.2015.07.002.
[
21]
P.
S.
A
ddi
s
on
,
“
W
a
ve
le
t
tr
a
ns
f
or
ms
a
nd
th
e
E
C
G
:
A
r
e
vi
e
w
,”
P
hy
s
io
lo
gi
c
al
m
e
as
ur
e
m
e
nt
,
vol
.
26
,
no.
5,
pp.
R
155
–
R
199
,
2
005,
doi
:
10.1088/0967
-
3334/26/
5/
R
01
.
[
22]
J
.
B
r
ow
nl
e
e
,
“A
G
e
nt
le
I
nt
r
oduc
ti
on
to
th
e
R
e
c
ti
f
ie
d
L
in
e
a
r
U
ni
t
(
R
e
L
U
)
,
”
J
anuar
y
9,
2019,
D
e
e
p
L
e
ar
ni
ng
P
e
r
fo
r
m
a
nc
e
.
ht
tp
s
:/
/m
a
c
hi
ne
le
a
r
ni
ngma
s
te
r
y.c
om/
r
e
c
ti
f
ie
d
-
li
ne
a
r
-
a
c
ti
va
ti
on
-
f
unc
ti
on
-
f
or
-
de
e
p
-
le
a
r
ni
ng
-
ne
ur
a
l
-
ne
twor
ks
(
a
c
c
e
s
s
e
d
D
e
c
.
15,
2021)
.
[
23]
C.
V
e
nka
te
s
a
n,
P.
K
a
r
th
ig
a
ik
uma
r
,
A.
P
a
ul
,
S.
S
a
th
e
e
s
kuma
r
a
n
,
a
nd
R.
K
um
a
r
,
"
E
C
G
S
ig
na
l
P
r
e
pr
oc
e
s
s
in
g
a
nd
S
V
M
C
la
s
s
if
ie
r
-
B
a
s
e
d
A
bnor
ma
li
ty
D
e
te
c
ti
on
in
R
e
mot
e
H
e
a
lt
hc
a
r
e
A
ppl
ic
a
ti
ons
,"
in
I
E
E
E
A
c
c
e
s
s
,
vol
.
6,
pp.
9767
-
9773,
2018,
doi
:
10.1109/AC
C
E
S
S
.2018.2794346.
[
24]
M.
K.
M
or
id
a
ni
,
M.
A
.
Z
a
de
h
,
a
nd
Z.
S
.
M
a
z
r
a
e
h,
“
A
n
E
f
f
ic
ie
nt
A
ut
oma
te
d
A
lg
or
it
hm
f
or
D
is
ti
ngui
s
hi
ng
N
or
ma
l
a
nd
A
bnor
ma
l
E
C
G
S
ig
na
l,
”
I
R
B
M
,
vo
l
.
40,
no.
6,
pp.
332
-
340,
2019
,
doi
:
10.
1016/j
.i
r
bm.2019.09.002
.
[
25]
A
.
L
ong,
“
D
e
te
c
ti
ng
H
e
a
r
t
A
r
r
hyt
hmi
a
s
w
it
h
D
e
e
p
L
e
a
r
ni
ng
in
K
e
r
a
s
w
it
h
D
e
ns
e
,
C
N
N
,
a
nd
L
S
T
M
,
”
ht
tp
s
:/
/t
ow
a
r
ds
da
ta
s
c
ie
n
c
e
.c
om/
de
te
c
ti
ng
-
he
a
r
t
-
a
r
r
hyt
hmi
a
s
-
w
it
h
-
de
e
p
-
le
a
r
ni
ng
-
in
-
ke
r
a
s
-
w
it
h
-
de
ns
e
-
c
nn
-
a
nd
-
ls
tm
-
a
dd337d9e
41f
(
a
c
c
e
s
s
e
d
N
ov
26,
20
21)
.
[
26]
S.
T.
S
a
na
mdi
ka
r
,
S.
T.
H
a
mde
,
a
nd
V.
G.
A
s
ut
ka
r
,
"
A
na
ly
s
is
a
nd
c
la
s
s
if
ic
a
ti
on
of
c
a
r
di
a
c
a
r
r
hyt
hmi
a
ba
s
e
d
on
ge
ne
r
a
l
s
pa
r
s
e
d
ne
ur
a
l
ne
twor
k
of
E
C
G
s
ig
na
ls
,”
SN
A
pp
li
e
d
Sc
i
e
nc
e
s
2,
p.
124
4
,
2020
,
doi
:
10.1007/s
42452
-
020
-
3058
-
8
.
[
27]
F.
A.
E
lh
a
j,
N.
S
a
li
m,
A.
R.
H
a
r
r
is
,
T.
T.
S
w
e
e
,
a
nd
T.
A
hm
e
d
,
"
A
r
r
hyt
hmi
a
r
e
c
ogni
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on
us
in
g
c
ombi
ne
d
li
n
e
a
r
a
nd
nonl
in
e
a
r
f
e
a
tu
r
e
s
of
E
C
G
s
ig
na
ls
,
"
C
om
put
e
r
M
e
th
ods
and
P
r
ogr
am
s
in
B
io
m
e
di
c
in
e
,
vol
.
127,
pp
.
52
-
63,
2016
,
doi
:
10.1016/j
.c
mpb.2015.12.024.
[
28]
A.
A.
B
ona
b
,
M.
C
.
A
mi
r
a
ni
,
a
nd
A.
M
e
hr
i
,
"
S
pe
c
tr
a
l
e
nt
r
opy
a
nd
de
e
p
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
f
or
E
C
G
b
e
a
t
c
la
s
s
if
ic
a
ti
on
,
"
B
io
c
y
be
r
n
e
ti
c
s
and
B
io
m
e
di
c
al
E
ngi
ne
e
r
in
g
,
vo
l.
40,
no.
2,
pp
.
691
-
700,
2020
,
doi
:
10.1016/j
.bbe
.2020.02.004.
[
29]
S.
P
a
nde
y
a
nd
R.
R.
J
a
nghe
l,
“
A
ut
oma
ti
c
a
r
r
hyt
hmi
a
r
e
c
ogni
ti
on
f
r
om
e
le
c
tr
oc
a
r
di
ogr
a
m
s
ig
na
ls
us
in
g
di
f
f
e
r
e
nt
f
e
a
tu
r
e
me
th
ods
w
it
h
lo
ng
s
hor
t
-
te
r
m
me
mor
y
ne
twor
k
mode
l,
”
Si
gnal
,
I
m
age
and
V
id
e
o
P
r
oc
e
s
s
in
g
,
vol
.
14
,
no.
4
,
pp.
1255
–
1263
,
2020
,
doi
:
10.1007/s
11760
-
020
-
01666
-
8.
[
30]
J.
H
ua
ng,
B.
C
he
n,
B.
Y
a
o
,
a
nd
W.
H
e
,
"
E
C
G
A
r
r
hyt
hmi
a
C
la
s
s
if
ic
a
ti
on
U
s
in
g
S
T
F
T
-
B
a
s
e
d
S
pe
c
tr
ogr
a
m
a
nd
C
onvolut
io
na
l
N
e
ur
a
l
N
e
twor
k,"
I
E
E
E
A
c
c
e
s
s
,
vol
.
7,
pp.
92871
-
92880,
2019,
doi
:
10.1109/AC
C
E
S
S
.2019.2928017.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Dr
.
T
r
an
-
A
nh
Vu
is
a
s
en
i
o
r
l
ec
t
u
rer
an
d
V
i
ce
H
e
ad
of
Bi
o
me
d
i
ca
l
E
n
g
i
n
eeri
n
g
D
ep
ar
t
men
t
,
Sch
o
o
l
of
E
l
e
ct
r
o
n
i
cs
a
n
d
T
el
eco
mm
u
n
i
c
at
i
o
n
s
,
H
an
o
i
U
n
i
v
ers
i
t
y
of
Sci
e
n
ce
an
d
T
ech
n
o
l
o
g
y
,
H
an
o
i
,
V
i
et
n
am.
He
earn
ed
h
i
s
Ph
.
D
.
d
e
g
ree
in
E
l
ect
r
i
cal
E
n
g
i
n
eer
i
n
g
fro
m
the
U
n
i
v
er
s
i
t
y
of
Mas
s
ach
u
s
e
t
t
s
at
L
o
w
e
l
l
(U
S
A
)
in
2
0
1
4
,
MS
d
eg
ree
in
Bi
o
med
i
cal
E
n
g
i
n
eer
i
n
g
fro
m
T
u
ft
s
U
n
i
v
er
s
i
t
y
(
U
SA
)
in
2
0
1
0
,
MS
d
e
g
ree
in
E
l
ect
r
o
n
i
cs
a
n
d
T
e
l
eco
mm
u
n
i
cat
i
o
n
fro
m
H
an
o
i
U
n
i
v
ers
i
t
y
of
Sc
i
en
ce
a
n
d
T
ech
n
o
l
o
g
y
(V
i
et
n
am
)
in
2
0
0
2
,
an
d
BS
d
eg
ree
in
E
l
ect
r
o
n
i
c
s
an
d
T
e
l
ec
o
mmu
n
i
ca
t
i
o
n
s
fro
m
H
a
n
o
i
U
n
i
v
ers
i
t
y
of
Sc
i
e
n
ce
an
d
T
ech
n
o
l
o
g
y
(
V
i
e
t
n
am)
in
2
0
0
0
.
His
ma
i
n
res
earc
h
i
n
t
eres
t
s
i
n
cl
u
d
e
t
h
e
med
i
ca
l
d
at
a
an
al
y
s
i
s
an
d
cl
as
s
i
f
i
cat
i
o
n
,
res
earch
an
d
d
ev
e
l
o
p
men
t
ap
p
l
i
ca
t
i
o
n
s
fo
r
reh
ab
i
l
i
t
a
t
i
o
n
,
s
mart
h
eal
t
h
.
He
can
be
co
n
t
ac
t
ed
at
emai
l
:
Vu
.
t
ra
n
an
h
@
h
u
s
t
.
e
d
u
.
v
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
ian
J
E
lec
E
ng
&
C
omp
S
c
i
,
Vol.
25
,
No.
2
,
F
e
br
ua
r
y
20
22
:
931
-
940
940
Ho
a
ng
-
Qu
a
ng
Huy
is
a
l
ect
u
rer
of
B
i
o
me
d
i
cal
E
n
g
i
n
eeri
n
g
D
ep
ar
t
men
t
,
Sch
o
o
l
of
E
l
ec
t
ro
n
i
c
s
an
d
T
e
l
eco
mm
u
n
i
cat
i
o
n
s
,
H
an
o
i
U
n
i
v
er
s
i
t
y
of
Sci
e
n
ce
an
d
T
ec
h
n
o
l
o
g
y
,
H
a
n
o
i
,
V
i
e
t
n
am.
He
earn
e
d
h
i
s
MS
d
e
g
ree
in
E
l
ec
t
ro
n
i
c
s
an
d
T
el
e
co
mm
u
n
i
cat
i
o
n
fro
m
H
a
n
o
i
U
n
i
v
er
s
i
t
y
of
Sci
en
ce
an
d
T
ech
n
o
l
o
g
y
(V
i
et
n
am)
in
2
0
0
6
,
an
d
BS
d
eg
ree
in
E
l
ec
t
ro
n
i
c
s
an
d
T
el
ec
o
mmu
n
i
ca
t
i
o
n
s
fro
m
H
a
n
o
i
U
n
i
v
er
s
i
t
y
of
Sc
i
en
ce
an
d
T
ec
h
n
o
l
o
g
y
(V
i
e
t
n
am)
in
2
0
0
2
.
H
i
s
mai
n
re
s
earch
i
n
t
eres
t
s
i
n
cl
u
d
e
the
me
d
i
cal
d
at
a
a
n
al
y
s
i
s
an
d
cl
a
s
s
i
fi
ca
t
i
o
n
,
re
s
earch
an
d
d
ev
e
l
o
p
men
t
ap
p
l
i
ca
t
i
o
n
s
fo
r
reh
ab
i
l
i
t
a
t
i
o
n
,
s
mar
t
h
eal
t
h
,
h
o
s
p
i
t
al
i
n
fo
rma
t
i
o
n
s
y
s
t
em
s
.
H
e
can
be
co
n
t
act
e
d
at
emai
l
:
H
u
y
.
h
o
a
n
g
q
u
an
g
@
mai
l
.
h
u
s
t
.
ed
u
.
v
n
.
P
ha
m
D
uy
Kh
a
nh
o
b
t
a
i
n
e
d
h
i
s
B.
Sc
in
Bi
o
me
d
i
ca
l
E
n
g
i
n
eer
i
n
g
D
e
p
art
me
n
t
,
Sch
o
o
l
of
E
l
ec
t
ro
n
i
cs
an
d
T
e
l
ec
o
mmu
n
i
ca
t
i
o
n
s
,
H
an
o
i
U
n
i
v
er
s
i
t
y
of
Sci
e
n
ce
an
d
T
ec
h
n
o
l
o
g
y
,
H
an
o
i
,
V
i
et
n
am
in
2
0
2
1
.
H
i
s
ma
i
n
re
s
earch
i
n
t
eres
t
s
i
n
c
l
u
d
e
t
h
e
med
i
cal
d
at
a
a
n
al
y
s
i
s
a
n
d
cl
as
s
i
f
i
cat
i
o
n
,
res
earc
h
an
d
d
ev
e
l
o
p
men
t
ap
p
l
i
cat
i
o
n
s
in
B
i
o
me
d
i
ca
l
E
n
g
i
n
ee
ri
n
g
.
H
e
can
be
co
n
t
act
e
d
at
emai
l
:
d
u
y
k
h
a
n
h
0
3
0
7
9
9
b
k
a@
g
mai
l
.
co
m
.
N
g
uy
en
Thi
M
i
nh
Huy
en
o
b
t
ai
n
ed
h
er
B.
Sc,
in
Bi
o
med
i
cal
E
n
g
i
n
eeri
n
g
D
ep
ar
t
men
t
,
Sch
o
o
l
of
E
l
e
ct
r
o
n
i
cs
a
n
d
T
el
eco
mm
u
n
i
c
at
i
o
n
s
,
H
an
o
i
U
n
i
v
ers
i
t
y
of
Sci
e
n
ce
an
d
T
ech
n
o
l
o
g
y
,
H
an
o
i
,
V
i
et
n
am
in
2
0
2
1
,
an
d
n
o
w
is
a
mas
t
er
s
t
u
d
en
t
.
H
er
mai
n
res
earch
i
n
t
eres
t
s
i
n
c
l
u
d
e
the
me
d
i
ca
l
d
a
t
a
an
al
y
s
i
s
a
n
d
cl
as
s
i
fi
cat
i
o
n
,
res
earch
an
d
d
e
v
el
o
p
me
n
t
a
p
p
l
i
ca
t
i
o
n
s
in
Bi
o
me
d
i
ca
l
E
n
g
i
n
eeri
n
g
.
Sh
e
ca
n
be
co
n
t
ac
t
ed
at
emai
l
:
mi
n
h
h
u
y
e
n
n
g
u
y
en
1
2
1
1
9
9
@
g
ma
i
l
.
co
m
.
Tri
nh
Thi
Thu
U
y
en
o
b
t
a
i
n
e
d
h
er
B.
Sc,
in
Bi
o
me
d
i
ca
l
E
n
g
i
n
eeri
n
g
D
e
p
art
me
n
t
,
Sch
o
o
l
of
E
l
ec
t
ro
n
i
cs
an
d
T
e
l
ec
o
mmu
n
i
ca
t
i
o
n
s
,
H
an
o
i
U
n
i
v
er
s
i
t
y
of
Sci
e
n
ce
an
d
T
ec
h
n
o
l
o
g
y
,
H
an
o
i
,
V
i
et
n
am
in
2
0
2
1
,
a
n
d
n
o
w
is
a
ma
s
t
er
s
t
u
d
e
n
t
.
H
er
mai
n
re
s
earch
i
n
t
eres
t
s
i
n
c
l
u
d
e
t
h
e
med
i
ca
l
d
a
t
a
an
a
l
y
s
i
s
an
d
cl
a
s
s
i
fi
c
at
i
o
n
,
res
earch
an
d
d
ev
e
l
o
p
men
t
ap
p
l
i
cat
i
o
n
s
in
B
i
o
me
d
i
ca
l
E
n
g
i
n
eeri
n
g
.
Sh
e
can
be
co
n
t
act
ed
at
ema
i
l
:
t
ri
n
h
t
h
u
u
y
e
n
3
2
6
@
g
ma
i
l
.
co
m
.
D
r.
P
ha
m
-
Thi
-
V
i
et
Huo
ng
o
b
t
ai
n
ed
h
er
B.
Sc
in
E
l
ect
ri
ca
l
E
n
g
i
n
eer
i
n
g
fro
m
H
an
o
i
U
n
i
v
er
s
i
t
y
of
Sc
i
en
ce
a
n
d
T
ec
h
n
o
l
o
g
y
in
2
0
0
7
.
Sh
e
g
o
t
h
er
MSc
an
d
Ph
D
,
b
o
t
h
in
E
l
ec
t
ri
ca
l
E
n
g
i
n
eeri
n
g
,
fro
m
U
n
i
v
er
s
i
t
y
of
Mas
s
ac
h
u
s
et
t
s
L
o
w
el
l
in
t
h
e
U
n
i
t
ed
S
t
at
e
s
,
in
2
0
1
0
an
d
2
0
1
2
.
Fro
m
2
0
1
2
to
2
0
1
5
,
s
h
e
w
as
a
res
earc
h
er
in
the
Man
n
i
n
g
Sch
o
o
l
of
Bu
s
i
n
es
s
,
L
o
w
el
l
,
Mas
s
ac
h
u
s
et
t
s
.
Fro
m
2
0
1
7
-
2
0
2
0
,
she
w
a
s
the
fac
u
l
t
y
of
V
N
U
U
n
i
v
er
s
i
t
y
of
E
n
g
i
n
eeri
n
g
an
d
T
ech
n
o
l
o
g
y
,
V
i
e
t
n
am
(V
N
U
-
U
E
T
).
Si
n
ce
2
0
2
0
,
she
w
o
rk
s
in
In
t
er
n
at
i
o
n
al
Sch
o
o
l
-
V
N
U
.
Sh
e
is
i
n
t
eres
t
ed
in
d
a
t
a
mi
n
i
n
g
an
d
an
a
l
y
t
i
c
s
,
mach
i
n
e
l
ear
n
i
n
g
met
h
o
d
o
l
o
g
i
e
s
,
w
i
t
h
a
p
p
l
i
ca
t
i
o
n
s
in
Bi
o
me
d
i
ca
l
E
n
g
i
n
eeri
n
g
.
Sh
e
ca
n
be
co
n
t
ac
t
ed
at
emai
l
:
H
u
o
n
g
p
t
v
@
i
s
v
n
u
.
v
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.