Recei ed Janua y 21, 2021, accep ed Feb ua y 7, 2021, da e o publica ion Feb ua y 11, 2021, da e o cu en e sion Feb ua y 23, 2021.
Digi al Objec Iden i ie 10.1109/ACCESS.2021.3058733
Non-In asi e Fe al Elec oca diog am Ex ac ion
Based on No el Hyb id Me hod o
In apa um ST Segmen Analysis
RADEK MARTINEK 1, (Membe , IEEE), RADANA KAHANKOVA 1, RENE JAROS 1,
KATERINA BARNOVA 1, ADAM MATONIA 2, MICHAL JEZEWSKI 3,
ROBERT CZABANSKI 3, KRZYSZTOF HOROBA2,
AND JANUSZ JEZEWSKI2, (Senio Membe , IEEE)
1Depa men o Cybe ne ics and Biomedical Enginee ing, Facul y o Elec ical Enginee ing and Compu e Science, VSB—Technical Uni e si y o Os a a,
708 33 Os a a, Czech Republic
2Łukasiewicz Resea ch Ne wo k—Ins i u e o Medical Technology and Equipmen , 41800 Zab ze, Poland
3Depa men o Cybe ne ics, Nano echnology and Da a P ocessing, Silesian Uni e si y o Technology, 44100 Gliwice, Poland
Co esponding au ho : Rene Ja os ( ene.ja os@ sb.cz)
This wo k was suppo ed in pa by he Minis y o Educa ion o he Czech Republic unde P ojec SP2021/32, and in pa by he Eu opean
Regional De elopmen Fund in he Resea ch Cen e o Ad anced Mecha onic Sys ems p ojec h ough he Ope a ional P og amme
Resea ch, De elopmen and Educa ion unde P ojec CZ.02.1.01/0.0/0.0/16_019/0000867.
ABSTRACT This s udy ocuses on non-in asi e e al elec oca diog am ex ac ion based on a no el hyb id
me hod, which combines he ad an ages o non-adap i e and adap i e app oaches o non-in asi e e al
elec oca diog am mo phological analysis. Besides es ima ing e al hea a e, which is he main pa ame e
used in he clinical p ac ice, his s udy p o ides non-in asi e ST segmen analysis on da a om Abdominal
and Di ec Fe al Elec oca diog am Da abase consis ing o simul aneous adi ional - gold s anda d in asi e
e al scalp elec ode and non-in asi e e al elec oca diog am eco ded du ing deli e y. This inno a i e
app oach u ilizing he combina ion o independen componen analysis and ecu si e leas squa es algo i hms
has he po en ial o ex ac aluable in o ma ion om non-in asi e e al elec oca diog am in o de o
iden i y e en ual sign o e al dis ess. This was a p ospec i e obse a ional s udy o non-in asi e e al
elec oca diog am, using 4 abdominally si ed elec odes, agains he adi ional e al scalp elec ode on
8 pa ien s. In e ms o e al hea a e es ima ion, he accu acy was high o all 8 es ed pa ien s wi h a e age
alue equaled 0.20 bea s pe minu e and a e age alue o 1.96 s anda d de ia ion equaled 5.80 bea s pe
minu e. In 7 pa ien s, i was possible o pe o m he ST segmen analysis wi h high accu acy in de e mining
T/QRS in compa ison wi h he e e ence e al scalp elec ode signal wi h a e age alues and 1.96 s anda d
de ia ion equaled 0.008 and 0.031 espec i ely. This s udy hus demons a es ha ST segmen analysis is
easible using non-in asi e ECG using he p oposed hyb id me hod.
INDEX TERMS Non-in asi e e al elec oca diog aphy (NI- ECG), e al hea a e ( HR), ST segmen
analysis (ST-analysis), hyb id me hod (HM), independen componen analysis and ecu si e leas squa es
(ICA-RLS), elec onic e al moni o ing (EFM), e al dis ess (FD), e al scalp elec ode (FSE).
I. INTRODUCTION
Elec onic e al moni o ing (EFM) is an essen ial pa o
mode n obs e ics, se ing mainly o diagnose e al dis-
ess (FD). Con en ional EFM me hods, such as ca dio ocog-
aphy (CTG), ha e been used o e he las decades in clinical
The associa e edi o coo dina ing he e iew o his manusc ip and
app o ing i o publica ion was Filbe Juwono .
p ac ice mainly o con inuous e al hea a e ( HR) moni-
o ing du ing labo bu also o he in e mi en assessmen
du ing p egnancy [1], [2]. Howe e , se e al s udies [3]–[5]
show ha CTG is no su icien ly accu a e and conclusi e
and is bu dened wi h a la ge in e - and ex a-obse e dis-
ag eemen [6], [7]. This is demons ably one o he easons
o inapp op ia e diagnosis o e al dis ess and consequen ly,
high numbe o unnecessa ily pe o med caesa ean sec ions.
28608 This wo k is licensed unde a C ea i e Commons A ibu ion 4.0 License. Fo mo e in o ma ion, see h ps://c ea i ecommons.o g/licenses/by/4.0/ VOLUME 9, 2021
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
Fo hese easons, physicians a e demanding concep ually
new solu ions o non-in asi e diagnos ic me hods [8].
Fe al elec oca diog aphy ( ECG) is among he mos
p omising EFM me hods. The main eason is ha he ECG
signal ca ies aluable in o ma ion as changes in he mo -
phology o he ECG wa e o m which a e associa ed wi h
dys unc ion induced by FD [9]–[11]. The e o e, ST segmen
analysis o he ECG has been de eloped o p o ide objec i e
in o ma ion abou he e al condi ion as an adjunc o HR
moni o ing [1], [12]. Two la ge andomized clinical ials
showed a signi ican ly lowe a e o me abolic acidosis a
bi h and ewe ope a i e deli e ies o FD when CTG and
ST analysis we e used simul aneously [13], [14]. Me abolic
dys unc ion induced by FD migh be e lec ed in ECG wa e-
o m as ST segmen changes, such as an inc ease in T wa e,
which can be quan i ied by he a io o he T wa e o he QRS
ampli ude (T/QRS a io) [15]. Simul aneous moni o ing o
HR and ST segmen analysis can hus help o educe unce -
ain y o FD diagnosis and hus a numbe o unnecessa ily
pe o med caesa eans o pa ien s wi h suspec ed FD [1].
In non-in asi e e al elec oca diog aphy (NI- ECG) based
moni o ing, e al heal h s a e is assessed using he in o ma-
ion ex ac ed om he elec ical po en ials p oduced by he
e al hea , which a e eco ded by means o elec odes placed
on he ma e nal abdomen. Howe e , in hese eco dings,
e al elec oca diog am ( ECG) is accompanied by ma e -
nal elec oca diog am (mECG) and a signi ican amoun o
noise [16]. Un o una ely, he magni ude o he e al compo-
nen is low compa ed o ma e nal one. Mo eo e , he signals
o e lap in ime as well as equency domain making he
accu a e ex ac ion o mo phological analysis o he ECG
wa e o m a challenging ask [16]–[18]. The esul ing quali y
o ECG ex ac ion has a majo impac on bo h he accu acy o
HR es ima ion and ex ac ion o he PQRST wa es. In his
s udy, a no el me hod o ST segmen analysis is p esen ed,
which makes also possible u he mo phological analysis o
o he ECG ea u es, such as QT in e al [19]. Abno mali ies
in he e al QT in e al indica e elec ophysiological changes
in he myoca dium. Long QT synd ome is a condi ion in
which epola iza ion o he hea is a ec ed. I esul s in an
inc eased isk o an a hy hmia which can esul in sudden
in an dea h synd ome and e al hypoxia [19]–[22].
A numbe o NI- ECG ex ac ion me hods ha e been
in oduced in he pas , including p incipal componen
analysis (PCA) [23], [24], independen componen analy-
sis (ICA) [23], [25], wa ele ans o m (WT) [26], adap-
i e neu o- uzzy in e ence sys em (ANFIS) [27] o leas
mean squa es (LMS) and ecu si e leas squa es (RLS)
algo i hms [28]. Recen s udies [29]–[32] ha e shown ha
hyb id me hods (HM), which combine he ad an ages o
non-adap i e and adap i e app oach o ECG ex ac ion,
achie e g ea e accu acy in ECG ex ac ion han when using
he me hods indi idually. Howe e , mos algo i hms p e-
sen ed in hese s udies as well as he p e ious esea ch o
he au ho s’ eam p esen ed in [23], [28]–[30], we e able o
ob ain only a small po ion o he la ge in o ma ion po en ial
o NI- ECG; hei main aim was o de e mine e al R posi ions
and use hem o es ima e he HR.
Recen s udies [12], [33], [34] show he possibili y
o ex ac addi ional clinically ele an in o ma ion om
NI- ECG. Howe e , o ob ain a ECG signal o a su icien
quali y o pe o m mo phological analysis, i is necessa y
o selec ex ac ion me hods mo e ca e ully so ha he p o-
cess does no de o m he signal’s mo phology. In ou p e-
ious wo ks [29], [30], we achie ed he bes esul s using
he ICA-RLS-WT algo i hm, combining h ee di e en algo-
i hms (ICA, RLS, and WT). Howe e , in he las s ep he
WT was applied o highligh he R peaks o mo e accu a e
HR es ima ion which caused dis o ion o he ECG signal
mo phology and hus loss o impo an diagnos ic in o ma-
ion. Fo his eason, he s udy in oduced he ein pe o ms
he ST analysis on NI- ECG signals ex ac ed wi h a help o
ICA-RLS algo i hm comp ising he ICA and RLS me hods.
In addi ion o he selec ion o sui able ex ac ion me hod,
i is necessa y o pay a en ion o algo i hm se ings. Au ho s
o [35] ocus on op imiza ion app oaches o di e en il e -
ing me hods, howe e , hei esea ch ela es mainly o he
R-R in e al de ec ion. Such se ing is hus applicable o
u he mo phological analysis and will be a subjec o ou
esea ch.
Finally, i is necessa y o choose he app op ia e da abase
o es he me hod’s e icacy. Un o una ely, lack o publicly
a ailable da abases wi h high quali y abdominal eco dings
makes he NI- ECG esea ch di icul . Cu en da abases
p o ide eco dings ha a e ei he o insu icien leng h o
quali y [36]. Mo eo e , each o he da abase di e s in he
elec ode placemen and acquisi ion sys em con igu a ion,
as he loca ion and numbe o elec odes is no s anda dized
as is he case wi h classical ECG [36]. The e ec o elec-
ode placemen and da a acquisi ion quali y on he e icacy
o ECG ex ac ion ha e been demons a ed in [28], [37].
Thus, o ensu e accu a e mo phological analysis, he eco d-
ings should no con ain a signi ican amoun o noise, he
e al/ma e nal componen a io should be high enough, and
he pola i y o he signals should be uni ied.
Some au ho s [38] in oduced syn he ic signal gene a o s
o p oduce da a o hei expe imen s, howe e , he esul s
ob ained using a i icial es signals o en di e om hose
pe o med on eal signals. In his s udy, he Abdominal and
Di ec Fe al ECG Da abase (ADFECGDB) [39], [40] was
selec ed o es he p oposed HM. This da abase is sui able
o he objec i es es ablished since i includes bo h abdominal
and e e ence scalp signals, eco ded in asi ely du ing he
labo by means o e al scalp elec ode. The e e ence signal
can be conside ed as a gold s anda d because i allows us o
ob ain e e ence PQRST wa es, de e mine T/QRS a io and
hus e alua e he accu acy o he NI- ST-analysis.
II. STATE OF THE ART
Mo phological analysis o he ECG signal, acqui ed
by in e nal moni o ing, can be pe o med by means o
STAN (Neo en a Medical AB, Mölndal, Sweden). STAN
VOLUME 9, 2021 28609
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
FIGURE 1. Example o he STAN moni o ou pu : (a) HR ace, (b) u e ine
con ac ions ( oco), (c) T/QRS ace, and (d) example o he a e aged
T/QRS.
is an analysis ool o e al moni o ing, which combines
ST-Analysis and CTG, p o iding ex ended and mo e accu a e
in o ma ion abou he e us du ing labo han he CTG alone.
An example o he STAN moni o ou pu is shown in Fig. 1.
The ECG is an in asi e p ocedu e in which he HR is
con inuously calcula ed using he de ec ed QRS complexes,
and ST segmen analysis is pe o med by moni o ing changes
in he T/QRS a io [36]. Ne e heless, he ECG can be
pe o med only du ing he labo (a e up u e o memb anes)
since i equi es he FSE o be a ached on he e al head
o hip dec easing com o o he pa ien . The in asi e e al
moni o ing is also wa an ed in se e al isks ha include
in ec ion o b uising o he e us [41]. Fo hese easons, he
bene i s o his me hod a e o en ques ioned [42], [43].
A. MORPHOLOGICAL ANALYSIS
Al hough he NI- ECG esea ch has been signi ican ly e ol -
ing in he pas decade and no el ex ac ion algo i hms a e
cons an ly eme ging, he e iciency o he il a ion p ocess
emains assessed solely on he basis o HR es ima es [36].
Mos o he con ibu ions claim o ob ain excellen esul s in
HR de e mina ion, howe e , his e alua ion does no e lec
he ex ac ion e iciency in e ms o signal mo phology. Ne -
e heless, se e al a emp s we e made o ex ac he mo pho-
logical ea u es om he NI-FECG [44]. Thei esul s can be
summa ized as ollows:
•In [1], he au ho s e iewed he de elopmen o a
h ee-s age me hodology. In he i s s age, he HR was
ex ac ed om he abdominal ECG signals (aECGs)
using a nonlinea analysis. In he second s age, a blind
sou ce sepa a ion echnique was applied o ob ain he
ECG. Finally, moni o ing o he e us was implemen ed
using ea u es ex ac ed om bo h he HR and ECG
mo phology ( he T/QRS a io and he e al ST wa e-
o ms cha ac e is ics). Syn he ic eco dings we e used
o he expe imen s.
•The au ho s o [45] used an algo i hm based on
he op imal-sh inkage unde he wa e-shape mani old
model o ex ac ECG. Bo h HR and signal mo phol-
ogy (PR, QT and ST in e als) we e analyzed du ing he
expe imen s on a da ase including eal and simula ed
signals indica ing he physiological and pa hological
condi ion o e uses (e.g. e al a hy hmia).
•In s udy p esen ed in [46], he au ho s deal wi h he
de ec ion o e al a hy hmias. The ype o a hy hmia
was de e mined based on he es ima ed P-wa e mo -
phology. Blocked no mal P wa e can be associa ed wi h
he p esence o second-deg ee AV block. The analysis
was pe o med on 500 eal eco dings.
•The au ho s o he s udy in oduced in [44] es ed h ee
classes o NI-FECG ex ac ion algo i hms: blind sou ce
sepa a ion, empla e sub ac ion and adap i e me hods.
In addi ion o he de ec ion o QRS complexes, he
p oblem o de e mina ion o QT in e al leng h and
T/QRS a io was conside ed. The expe imen s we e pe -
o med only on syn he ic da a.
•In a s udy p esen ed in [47], he au ho s in oduced
Bayesian il e ing amewo k based on he ex ended
Kalman il e . Syn he ic da a was used o he e alua ion
based on de e mining he leng h o he QT in e al.
•In [48] he ECG signal was ex ac ed using he Kalman
il e amewo k. The HR es ima ion and ST segmen
analysis we e pe o med based on eal eco dings.
•In [20] he au ho s pe o med QT in e al analysis on
eal eco dings using model-based es ima ion me hod.
Fi s , R peaks we e de ec ed based on h eshold alue
and RR in e als we e de e mined. Subsequen ly, he
in e al, in which he T wa e should occu , was cal-
cula ed. The end o he T wa e was calcula ed as he
median o his in e al. The beginnings o he Q peaks
we e de e mined manually. The di e ence be ween QT
in e als ob ained by means o e e ence me hods (scalp
ECG and Dopple Ul asound) was less han 5%. The
e ec o QT p olonga ion in b adyca dia and long QT
synd ome has been demons a ed.
•In [19], he au ho s pe o med QT in e al analysis using
a STAN S21 de ice on 68 e uses ha showed signs
o me abolic acidosis a bi h (pH <7.05). App ox-
ima ely he same numbe o pa ien s was used as a
con ol g oup. The measu emen s we e aken a he
beginning o he eco ding (on he HR baseline), du ing
decele a ions and a he end o he eco ding. The QT
pa ame e was calcula ed by Baze ’s o mula and he
de e mined in e als we e compa ed using Wilcoxon
es . The esul s con i med ha he e is a signi ican
sho ening o he QT in e al du ing se e e in apa um
hypoxia and me abolic acidosis and hus p o ed ha
in apa um QT in e al moni o ing may p o ide addi-
ional in o ma ion on he condi ion o he e us.
•Fe al QT in e al analysis on non-in asi e abdominal
eco dings was also pa o he Challenge 2013 call [49].
Fo his pu pose, a da ase was c ea ed con aining
eco dings om di e en sou ces, wi h a iable ges a-
ional age and elec ode placemen . The accu acy o
he e al QT in e al es ima ion was e alua ed using he
oo mean squa e di e ence be ween he e e ence and
he calcula ed QT in e als. In he e e ence eco dings,
28610 VOLUME 9, 2021
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
e al QT in e als we e de e mined by an expe [49].
The bes esul s we e achie ed by au ho s o [50].
The esul s o all hese s udies indica e ha mo phological
analysis using NI- ECG is possible and p o ides esul s as
ob ained by in asi e a ian o ECG. Howe e , quali y aECG
eco dings mus be used, il e ing me hods and se ings ha
do no cause signal dis o ion mus be app op ia ely selec ed,
and sui able me hods mus be chosen o he co ec de ec ion
o indi idual wa es and oscilla ions in he ECG signal.
B. METHODS FOR ECG SIGNAL EXTRACTION
The challenge in p ocessing and analyzing ECG is o ex ac
he high quali y e al componen om aECG signals. In he
pas , many au ho s [51]–[56] designed and es ed me hods
ha could supp ess he in e e ence con ained in he aECG
signal (especially he mECG componen ) and highligh he
ECG componen as much as possible.
•Wa ele ans o m is one o he mos commonly used
me hods o ECG ex ac ion, mainly due o he possi-
bili y o signal analysis in he ime- equency domain.
To ob ain op imal esul s, i is necessa y o pay a en ion
o he sys em se ing, namely o he selec ion o he
mo he wa ele , i s scale, and he numbe o decom-
posi ion le els. Fo he pu poses o ECG ex ac ion,
he me hod was es ed in [51], [57], whe e Daubechies
wa ele was selec ed; in [58], he Symle wa ele was
ound as sui able o ECG ex ac ion; in [59] he au ho s
concluded he Bio hogonal wa ele as he mos e ec-
i e wa ele base ype o he ECG ex ac ion.
•Empi ical mode decomposi ion based me hods - he
p inciple o EMD based me hods is based on he decom-
posi ion o he inpu signal in o simple signals, which
a e called in insic mode unc ions (IMFs). By selec -
ing one o mo e IMFs and summing hem, i is pos-
sible o c ea e he il e ed ECG signal. The basic
EMD me hod achie ed a ela i ely high-quali y ex ac-
ion o ECG in he [60], [61]. The ex ac ion was
u he imp o ed by using imp o ed a ian s o his
me hod, such as ensemble empi ical mode decompo-
si ion (EEMD) [60], [61], complemen a y ensemble
empi ical mode decomposi ion (CEEMD) [60], [61],
o complemen a y ensemble empi ical mode decompo-
si ion wi h adap i e noise (CEEMDAN) [62].
•Kalman il e ing - his me hod es ima es he use ul sig-
nal om noisy da a based on he mos ecen ly mea-
su ed da a, he sys em model, bu also using da a on he
p e ious s a e o he sys em. The disad an age o he
basic e sion o he il e is ha i can be used only o
linea sys ems. Fo p ac ical use, an ex ended Kalman
il e (EKF) and ex ended Kalman smoo he (EKS) ha e
been de eloped, which can also be used o nonlinea
sys ems. Bo h il e a ian s, EKF and EKS, demon-
s a ed hei e ec i eness in mECG componen sup-
p ession and ECG signal ex ac ion in [53], [63], [64]
and [54], espec i ely.
•A i icial neu al ne wo ks a e used o pa allel da a
p ocessing based on mimicking he beha io o bio-
logical s uc u es. The use o con olu ional neu al ne -
wo ks (CNN) o emo e noise om he ECG signal has
been es ed in [65] wi h e y good esul s. High-quali y
ECG il a ion was also achie ed in [66], whe e
dynamic neu al ne wo ks (DNN) wi h FIR synapses
we e es ed and analyzed. The adap i e neu o- uzzy
in e ence sys em (ANFIS), combining neu al ne wo ks
and uzzy logic p inciples, was es ed in [67]. The echo
s a e ne wo ks (ESN) me hod was used o ECG il e -
ing in [68] and achie ed e y high quali y esul s.
•Blind sou ce sepa a ion me hods - hese me hods ha e
ecei ed a g ea deal o a en ion in he pas , as hey
ha e been success ul in ECG ex ac ion. The p inci-
ple is based on he decomposi ion o he signal mix-
u e (aECG) in o he o iginal sou ce componen s ( ECG,
mECG, noise). These me hods assume ha he sou ces
a e s a is ically independen . The basic ep esen a i es
include he ICA [23], [55], [69]–[72], PCA [23], [73]
o he singula alue decomposi ion (SVD) [74], [75].
F om his g oup o me hods, he ICA [23] me hod
p o ed o be he mos sui able o ECG ex ac ion.
Se e al ex ended a ian s o ICA ha e been in o-
duced, such as join app oxima e diagonaliza ion o
eigen-ma ices (JADE) [76], [77], he e y e icien
Fas ICA algo i hm [78]–[80] o he mul idimensional
ICA (MICA) me hod [81]. A compa ison o he pe -
o mance o se e al ICA based me hods was pe o med
in [82]. The au ho s analyzed he Fas ICA algo i hm,
JADE, he e icien e sion o he Fas ICA (EFICA)
and he second o de blind iden i ica ion (SOBI), whe e
EFICA p o ed o be mos e icien o ECG ex ac ion.
Simila ly, in [83], he au ho s compa ed di e en ICA
based algo i hms: Fas ICA based on ku osis, Fas ICA
based on negen opy and JADE me hod. The mos accu-
a e ex ac ion o ECG was ob ained using he JADE
me hod, bu he Fas ICA me hod was able o ex ac
ECG as e . The esul ing pe o mance o ICA based
algo i hms is a ec ed by he numbe o inpu signals.
A he same ime, he highe he numbe o inpu sig-
nals, he mo e accu a ely i is possible o ob ain sou ce
signals. Mos o he p esen ed ICA based me hods a e
mul ichannel, bu he e is also a single-channel a ian
p esen ed in [45], [84], achie ing e y p omising esul s
in he ex ac ion o ECG. Ne e heless, a la ge numbe
o inpu signals is associa ed wi h g ea e compu a ional
complexi y and lowe com o o he p egnan subjec
when eco ded.
•Adap i e il e s - hese il e s a e based on minimizing
he e o signal by au oma ically adjus ing he il e
coe icien s. The e o signal is de ined as he di e ence
be ween he desi ed ou pu and he ac ual ou pu o
he algo i hm. Fo ECG ex ac ion, he mECG compo-
nen (mECG), which can be eco ded om he mo he ’s
ches o ex ac ed om a mix u e o aECG signals using
VOLUME 9, 2021 28611
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
a blind sou ce sepa a ion me hod (e.g. ICA), is mod-
i ied o o m he mECG componen con ained in he
aECG signal. This modi ied mECG componen is hen
sub ac ed om he aECG signal and ECG is ob ained.
The adap i e LMS algo i hm uses mean squa e e o
minimiza ion be ween he desi ed and ac ual ou pu .
The LMS il e in combina ion wi h WT was used o
ex ac he ECG e icien ly in [59], [85]. The combina-
ion o LMS and ICA also achie ed p omising esul s
in [70]. O he ex ended a ian s o his il e no mal-
ized LMS (NLMS), delayed LMS (DLMS) and block
LMS (BLMS) we e compa ed in [86], whe e he bes
esul s we e achie ed using he BLMS algo i hm. The
adap i e RLS me hod, which uses he minimiza ion o
he o al squa ed e o be ween he desi ed and ac ual
ou pu , was es ed in [87] o ECG il e ing. The pe o -
mance o he me hod was compa ed wi h LMS bu be e
esul s we e achie ed wi h RLS algo i hm. In [88], he
pe o mance o he me hod in ECG ex ac ion was com-
pa ed wi h NLMS, and e en in his case be e esul s
we e ob ained wi h RLS. A o al o nine combina ions
o cascading RLS, LMS and NLMS il e s we e es ed
in [56]. The mos p omising esul s we e achie ed by he
combina ion o LMS and RLS me hods.
A compa ison di e en ex ac ion me hods o HR de e -
mina ion, mo phology bias and compu a ional complexi y is
summa ized in Table 1. The e alua ion o indi idual pa am-
e e s was pe o med as ollows:
•Accu acy o HR de e mina ion - his pa ame e was
e alua ed using high, medium and low ca ego ies,
de ined as:
–High - he me hod was able o supp ess all noise
e icien ly and in he s a is ical e alua ion o HR
de e mina ion achie ed he accu acy o ≥95%
(based on he ACC, SE, PPV and F1 pa ame e s).
–Medium - he me hod was able o supp ess mos
in e e ence, bu some esidues dec eased he alues
o ACC, SE, PPV and F1 eaching he accu acy
o ≥80% in he s a is ical e alua ion o HR
de e mina ion.
–Low - he me hod was no able o su icien ly
emo e he in e e ence and in he s a is ical e al-
ua ion o HR achie ed he accu acy o <80% was
achie ed using he pa ame e s ACC, SE, PPV and
F1.
•Dis o ion o mo phology - de e mines mo phology o
he signal is dis o ed by he gi en me hod. I so, i is
no longe possible o pe o m mo phological analysis o
he signal, i no , mo phological analysis o he signal is
possible.
•Compu a ional complexi y - he pa ame e was e alua ed
using he ca ego ies high, medium and low de ined as
ollows:
–High - he me hod is compu a ionally in ensi e and
canno be used in eal- ime applica ions.
TABLE 1. Compa ison o me hods o ECG ex ac ion.
–Medium - he me hod is compu a ionally sligh ly
mo e demanding and can be used in eal- ime appli-
ca ions a e op imiza ion.
–Low - he me hod is no compu a ionally demanding
and can hus be used in eal- ime applica ions.
C. FETAL HEART RATE TRACE AND ST ANALYSIS
INTERPRETATION
Co ec ea u e ex ac ion is impo an o he subsequen
e al heal h assessmen . This subsec ion will ou line he clin-
ical me hodology o e alua ing HR aces and di ec ECG
eco dings. These ecommenda ions a e used by he clini-
cians as well as in he e al moni o ing sys ems and de ices
based on ECG (e.g. STAN S31 and STAN S41).
In 2015, he FIGO published a new consensus guideline
on in apa um e al moni o ing (FIGO2015) [89] ha mod-
i ied he ea lie e sion published in 1987 (FIGO1987) [90].
These new ecommenda ions cons i u ed he i s wide-scale
ag eemen on essen ial aspec s o CTG moni o ing. The
pu pose o he FIGO guidelines is o assis in he use and
in e p e a ion o CTG acings [2], as well as in he clinical
managemen o speci ic CTG pa e ns, such as baseline HR,
a iabili y o decele a ions [6], [7] (see Table 2). Mo eo e ,
he Swedish Socie y o Obs e ics and Gynecology (SSOG),
issued hei la es ecommenda ions ela ed o he same ma -
e in 2017 [91] (SSOG2017).
The assessmen is usually pe o med by obs e icians
o midwi es who classi y he acings in o one o h ee
classes: no mal, suspicious o pa hological acco ding o he
c i e ia summa ized in Table 2[89]. The acings should
be ee alua ed wi hin a easonable ime ame (a leas
e e y 30 minu es) due o changing na u e o CTG signals
du ing labo [89], [92].
In addi ion o he abo e-men ioned e alua ion pa am-
e e s, u he mo phological analysis o he ECG wa e-
o m, especially ST-segmen analysis (ST-analysis), has been
p oposed. I s objec i e was o imp o e he clinical use o
EFM since a ious s udies [13], [93]–[95] ha e demons a ed
ha i can p o ide in o ma ion on oxygena ion o he e al
myoca dium [12]. S udies also linked ad enaline su ge and
he appea ance o high T wa es in he e al ECG [96].
28612 VOLUME 9, 2021

R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
TABLE 2. CTG classi ica ion c i e ia, in e p e a ion and ecommended managemen acco ding o FIGO consensus guidelines on in apa um e al
moni o ing.
TABLE 3. Sugges ed Cou se o Ac ion based on ST E en s no ed and Classi ica ion o CTG based on STAN Guidelines.
P olonga ion o he QRS in e al was ound o be associ-
a ed wi h baseline achyca dia, sho ening o he P-R in e -
al wi h CTG decele a ions, and T wa e in e sion wi h
signs o placen al dys unc ion [97]. Addi ionally, Pa di e al.
in [98] showed changes in ECG con igu a ion du ing pe iods
o e al hypoxia, wi h P wa e and P-Q in e al modi ica-
ions appea ing consis en ly du ing la e decele a ions; ST
dep ession/ele a ion/in e sion o inc eased T wa e ampli-
ude occu ed in 48% o cases.
Based on hose indings and he ecommenda ions o ST
analysis e alua ion published in [12], which we e la e aken
o e by Neo en a Medical o c ea e hei e alua ion guide
used in STAN analyze , new ca ego ies o ST ela ed e en s
we e iden i ied including:
•Th ee ca ego ies o biphasic ST segmen s acco ding o
he ela ionship be ween he baseline and hei slope:
–1 (slope abo e baseline),
–2 (c ossing he baseline),
–3 (below baseline).
•Th ee ypes o ele an ST e en s, associa ed wi h acido-
sis and inco po a ed in o signal p ocessing algo i hms:
–episodic T/QRS ise (T/QRS ise >0.10 in 2 con-
secu i e T/QRSs),
–baseline T/QRS ise (T/QRS ise >0.05 in mo e
han 10 minu es),
1Decele a ions a e epe i i e when associa ed wi h >50% con ac ions.
Absence o accele a ions in labou is o unce ain signi icance.
–ST in e al dep essions: ca ego ies 2 and 3 biphasic
ST e en s.
Cu en ly, he clinical use o ST analysis equi es i o be
combined wi h CTG analysis [95], see Table 3. Homogenei y
and ag eemen s a is ics be ween he CTG classi ica ions
SSOG2017, FIGO2015, and FIGO1987 we e pe o med and
published in [91]. The s udy aimed o e eal homogenei y
and ag eemen be ween he sys ems in classi ying CTG and
ST e en s, and ela e hem o ma e nal and pe ina al ou -
comes. The esul s showed disc epancies in he classi ica ion
be ween he old and new sys ems.
III. MATERIAL AND METHODS
The aim o his s udy is o es and s a is ically e alua e he
e ec i eness o he hyb id ICA-RLS ex ac ion me hod o
ECG. The ICA me hod allows o sepa a ing he s a is ically
independen signals and i has been al eady success ully used
o he ECG ex ac ion [25]. This algo i hm is ela i ely
compu a ionally demanding; he e o e, se e al as e a ian s
o his me hod ha e been in oduced [29], [99]. The mos
commonly used a ian , he so-called Fas ICA algo i hm, was
success ully applied o ex ac he ECG om aECGs [72].
The RLS me hod is also an e ec i e ool o he ECG
ex ac ion, as con i med by [28], [100], bu i s main disad an-
age is he high compu a ional complexi y. The backg ound
heo y o ICA and RLS me hods is well desc ibed in he
li e a u e [23], [28], [29], [78], [101], [102], and hei de ailed
p esen a ion will he e o e no be included in his s udy.
VOLUME 9, 2021 28613
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
FIGURE 2. Examples o he omi ed abdominal ECG signals ( eco dings 04, 07, 11, and 12) and es ima ed ECG signals.
This sec ion will in oduce only he hyb id ICA-RLS based
sys em, he es ing da ase , and he e alua ion pa ame e s and
p ocess.
A. DATASET
The hyb id me hod was es ed using eal da a, as he assess-
men o ex ac ion e iciency on syn he ic signals is o en
misleading. In addi ion, i was necessa y o selec a da abase
ha includes a con inuous e e ence scalp eco dings in
addi ion o abdominal ones so ha he accu acy o he
NI- ECG can be e alua ed. Only one publicly a ailable
da abase (ADFECGDB) mee s hese c i e ia, hus we used i
in ou expe imen s [23], [39], [103]. These eco dings we e
ob ained om 12 subjec s be ween he 38 h and 41s week
o p egnancy. Each includes ou abdominal and one di ec
signal along wi h he anno a ions indica ing he loca ion o
he e al R peaks. These anno a ions we e c ea ed by on-line
analysis in he KOMPOREL sys em and e i ied by a g oup
o ca diologis s [39], [103]. The sampling equency was
1000 Hz ( o i e eco dings) and 500 Hz ( o se en eco d-
ings), he signal esolu ion was 16 bi s. The abdominal signals
we e eco ded using he Ag-AgCl elec odes, while he di ec
signals we e eco ded by means o spi al e al scalp elec ode.
The con igu a ion o he abdominal elec odes comp ised
ou elec odes placed on he abdomen, an abdominal e e -
ence elec ode placed abo e he pubic symphysis and a com-
mon mode e e ence elec ode (wi h ac i e-g ound signal)
placed on he le leg [39], [103].
Al hough he da abase includes a o al o 12 eco dings,
we only used 8 o hem o he expe imen s. The eason is ha
he quali y o some o he eco dings ( 04, 07, 11, and 12)
is low o he needs o mo phological analysis. Fig. 2shows
he examples o he omi ed abdominal signals along wi h he
es ima ed ECG signals. I can be no iced ha he ex ac ion
sys em was no able o supp ess he unwan ed signals and
hus he esul ing ECG signals a e oo noisy o he needs
o any u he analysis. In ac , e en accu acy o he HR
de e mina ion is qui e low in compa ison wi h he es o he
da ase as demons a ed by he HR aces displayed in Fig. 3
and o he in es iga ions p esen ed in [30], [104].
Re e ence anno a ions a e a ailable a ADFECGDB o
compa e he accu acy o de e mining he R peak posi ions
28614 VOLUME 9, 2021
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
FIGURE 3. Examples o he HR aces es ima ed using he omi ed eco dings ( 04, 07, 11, and 12) o demons a e
hei low quali y o he pu pose o mo phological analysis.
FIGURE 4. Block diag am o ST analysis using e e ence and ex ac ed ECG signal.
in he ex ac ed signal wi h espec o he e e ence. In his
s udy, howe e , we ocus on mo phological analysis, o
which no e e ence alues o anno a ions a e a ailable
o possible compa ison. Fo his eason, we had o c e-
a e e e ence anno a ions o each eco d. Fig. 4illus a es
he p ocess o gene a ing ST analysis e e ence alues
om he ECG signal measu ed by he scalp elec ode
(di ec ECG) and es ima ed ST analysis alues using he
ex ac ed ECG signal. These anno a ions a e a ailable a
h ps://dx.doi.o g/10.21227/70cd-bw64 [105].
B. HYBRID SYSTEM DESIGN
The hyb id sys em is designed o pe o m ST segmen anal-
ysis om ex ac ed ECG signals using a combina ion o
Fas ICA and RLS algo i hms, which allows us o combine
he ad an ages o bo h me hods and achie e mo e accu a e
ECG ex ac ion. Fig. 5shows schema ic diag am o he
expe imen al se up wi h examples o he ou pu s. The sys em
is based on ou p e ious esea ch ocused on HR es ima ion
in oduced in [29], [30]. The p ocess comp ises o ollowing
s eps:
•Selec ion o sui able aECG signals - Table 4shows
he RLS il e o de se ings o each eco d as well as
selec ed elec ode combina ion used as he ICA-RLS
hyb id sys em inpu . Fil e o de se ings and selec ed
elec ode combina ion was based on he p e ious
s udy [29] ocused on he selec ion o sui able elec odes
and RLS il e o de o each eco ding.
•P ep ocessing - bandpass ini e impulse esponse il-
e (FIR) wi h cu -o equencies 3 and 150 Hz, il e
o de o 500.
•Decomposi ion o he signal using Fas ICA - he
algo i hm decomposes he inpu signals o mul iple
independen componen s including mECG* which is a
componen co esponding o he ma e nal mECG, and
aECG*, which is a componen co esponding o he
aECG inpu s wi h an enhanced e al componen . The
Fas ICA algo i hm is se o a leas 20 i e a ions and
3 ou pu componen s.
•Adap i e il e ing - he es ima ed mECG* and aECG*
signals a e used as e e ence and p ima y inpu s o he
RLS algo i hm, espec i ely. Fo he RLS algo i hm,
he o ge ing ac o was se o 1 and he il e o de
a ied in he ange om 1 o 100. Using his adap i e
algo i hm, he ECG signal was ex ac ed. Table 4shows
he RLS il e o de se ings o each eco ding as well
as selec ed elec ode combina ion used as he ICA-RLS
hyb id sys em inpu .
The es ima ed ECG signal en e s he R peak de ec-
o based on con inuous wa ele ans o m (CWT)
[29], [106]–[108]. This de ec o es ima es he posi ions o
R peaks ha will be subsequen ly compa ed wi h he e e -
ence anno a ions ob ained using he di ec signal egis e ed
wi h a help o FSE (see Fig. 5). This signal is p e-p ocessed
by he FIR il e desc ibed abo e. Fo he eco dings om
ADFECGDB da abase (accessible a Physione ) he anno a-
ions we e a ailable, o he emaining signals he R peak
posi ions we e de e mined using he CWT-based de ec o .
The ins an aneous HR was es ima ed using he de e -
mined in e als be ween consecu i e R peaks (RR in e als).
Subsequen ly, he HR aces o bo h e e ence and es i-
ma ed signals we e de e mined and compa ed using he
VOLUME 9, 2021 28615
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
FIGURE 5. Example o he STAN moni o ou pu : (a) HR ace, (b) u e ine con ac ions ( oco), (c) T/QRS ace, and (d) example o he
a e aged T/QRS.
TABLE 4. ICA-RLS algo i hm se ings.
Bland-Al man plo . The Bland-Al man plo shows he mean
alue µand ±1.96σ, which e lec he di e ence be ween he
es ima ed and e e ence HR aces. Mo eo e , he R peak
posi ions we e used o e alua e he quali y o he ex ac-
ion based one he ACC, SE, PPV and F1 indices de ined
by (1), (2), (3), and (5), espec i ely. The ue posi i e (TP),
alse posi i e (FP), and alse nega i e (FN) alues we e de e -
mined using he es ima ed and e e ence signals. The ime
in e al o he TP de e mina ion was selec ed as ±50 ms
om he e e ence R peak [29], [32].
ACC =TP
TP +FP +FN ·100 (%),(1)
Se =TP
TP +FN ·100 (%),(2)
PPV =TP
TP +FP ·100 (%),(3)
F1=2·Se ·PPV
Se +PPV =2·TP
2·TP +FP +FN ·100 (%).
(4)
Finally, he main pa ame e o e alua e he quali y o he
ex ac ion, being he main goal o his s udy, was based on
he applicabili y o he ST segmen analysis on he es ima ed
ECG signal. The indi idual QRS complexes and he de e -
mined T/QRS a e displayed in he g aph o compa e he
e icacy o isual e alua ion. The objec i e e alua ion o he
de e mined T/QRS is pe o med using he ACCT/QRS pa am-
e e . Fo de ailed desc ip ion o he ST segmen analysis,
please e e o he nex Sec ion.
C. ST SEGMENT ANALYSIS
In he clinical p ac ice, he STAN machine calcula es he
no mal T/QRS a io o each e us and hus es ablishes he
‘baseline alue’ o e he i s 4–5 minu es [109]. The de ice
hen analyses e e y 30 ECG complexes and compa es hem
wi h his ‘baseline alue’. Each analysis is ma ked on he
HR ace wi h a c oss ‘X’ (see Fig. 1) o s a ‘*’ (see Fig. 6).
I he analyzed ECG complexes di e signi ican ly om he
‘baseline alue’, hey will be lagged up as an ‘ST e en ’.
In he p esence o an ‘ST e en ’, i is necessa y o classi y he
CTG ace acco ding o STAN guidelines (see Table 3) and
hen o de e mine whe he i is signi ican and equi es any
ac ion [109].
In ou s udy, we i s applied he R peak de ec ion and
hen ca ied ou he analysis o 30 a e aged consecu i e
ECG cycles. Fo a e aging, we de e mined a window wi h a
28616 VOLUME 9, 2021
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
FIGURE 12. G aphical illus a ion o he HR aces and ST analysis as displayed by he STAN moni o .
FIGURE 13. Compa ison o a e aged QRS complexes ob ained using hyb id ex ac ion sys em ICA-RLS, ICA-RLS-EMD and
ICA-RLS-WT on he eco ding 05 illus a ing he d awback o WT me hod used in pos p ocessing o he esul ing ECG
signal.
ame. I can be no iced ha he quali y o he ex ac ed ECG
is high and hus he ST analysis could be pe o med. The
es o he examples, deno ed as (d), (e), and ( ), co espond
o he sec ions o he signal whe e he de e mina ion o he
ST analysis was no success ul ( he es ima ed and e e ence
T:QRS a io di e ). The quali y o he aECG and conse-
quen ly he ex ac ed ECG signals is poo compa ed o he
p e ious cases. The signal con ains ma e nal esidua and i s
mo phology is de o med, hence, he ST segmen analysis was
no success ul.
Finally, we compa ed he esul s ob ained wi h he esul s
o a e y ew s udies ha ha e deal wi h he mo pholog-
ical analysis o he ECG signal ( hese a e summa ized in
Table 8). Objec i e compa ison o esul s is e y di icul
since he au ho s use di e en da abases o es ing ( he
Non-In asi e Fe al ECG A hy hmia Da abase [46], he Fe al
ECG Syn he ic Gene a o [38], use hei own syn he ically
gene a ed signals [1] o used hei own eal signals [20]. I is
clea ha mos au ho s use syn he ic da a o expe imen s,
on which hey achie e e y accu a e esul s. Un o una ely,
subsequen es ing o algo i hms on eal da a usually achie es
signi ican ly wo se esul s.
He ein, we used eal eco ds o ou expe imen s con-
aining a signal om he scalp elec ode, which we used o
c ea e anno a ions o de e mine he accu acy o ST analysis.
Fo u u e esea ch in his a ea, i would be app op ia e es
he algo i hms on mo e ex ensi e da ase wi h con inuously
eco ded abdominal di ec ECG eco dings. Ano he limi a-
ion in objec i e compa ison is he use o a ious pa ame e s
(coe icien o de e mina ion [44] o he oo mean squa e
e o [48]) o e alua e he e ec i eness o mo phological
analysis. Fo hese easons, we ied o make a combina ion
o bo h objec i e and subjec i e compa ison o he achie ed
esul s.
VOLUME 9, 2021 28623

R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
FIGURE 14. Example o he ST segmen analysis pe o med on he ECG signal ob ained by he ICA-RLS hyb id me hod. Sub igu es (a), (b),
and (c) show 3 sec ions whe e he accu acy o ST analysis was high, while (d), (e) and ( ) show 3 sec ions o inaccu a e ST segmen analysis.
•In [45], op imal sh inkage was used o de e mine HR
wi h accu acy o 79.25±31.75% o (semi)- eal eco ds
and wi h 93.21 ±14.31% o syn he ic eco ds. In bo h
cases, hese a e wo se esul s han we ha e achie ed.
In case o mo phological analysis, he au ho s did no
deal wi h ST analysis, bu wi h P wa e and T wa e
de ec ion, in which hey eached alues o 4.85 ±8.33
and 0.22±0.34 acco ding o he no malized mean ampli-
ude e o . The au ho s s a e ha he me hod was also
e ec i e o eco dings con aining a hy hmias, bu less
e ec i e o signals wi h signi ican mECG ampli ude.
•The au ho s in [46] deal wi h he de ec ion o a hy h-
mias by analyzing he mo phology o he P wa e.
The au ho s used hei own eal eco ds acqui ed
on 500 women (ges a ional age: 22–41 weeks) o es -
ing. The s udy does no epo s a is ical esul s o he
de e mina ion o HR o o he analysis o P wa e mo -
phology. Howe e , he au ho s s a e ha he algo i hm
was able o de ec all cases o a hy hmias and in only
one case was he a hy hmia inco ec ly iden i ied due
o he low esolu ion o he P wa e.
•Th ee ypes o me hods (BSS me hods, TS and adap i e
me hods) we e es ed in [44]. When de e mining HR,
he algo i hms achie ed an accu acy o 86.40–99.90%,
77.40–96.00%, and 87.1–97.90%, espec i ely. In all
cases, hese a e sligh ly wo se esul s han hose
ob ained using he algo i hm p esen ed by us. The
au ho s also deal wi h de e mining he QT in e al
and he T/QRS a io. The e alua ion was pe o med
using he coe icien o de e mina ion, which eached
he alues o 0.189, 0.846, and 0.574, espec i ely, when
e alua ing he accu acy o he es ima ed QT in e al, and
0.098, 0.916, and 0.812, espec i ely, when de e mining
he T/QRS a io. The leas accu a e esul s we e ob ained
wi h eco ds con aining ec opic bea s.
•In [47], he au ho s did no deal wi h he de e mina ion
o HR, bu only wi h he de e mina ion o he accu-
acy o he es ima ed leng h o he QT in e al using
ex ended Kalman il e . They managed o achie e an
accu acy o 4.00 ms e alua ed by he median absolu e
e o . The ad an age o his me hod is ha i equi es
only single-channel as an inpu .
28624 VOLUME 9, 2021
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
TABLE 8. Summa y o s a e-o - he-a me hods o NI- ECG mo phological analysis.
•The au ho s in [48] es ed he KF amewo k o ex ac
he signal o he ST analysis. The au ho s used hei own
eal eco ds acqui ed om 32 women (ges a ional age:
35–41 weeks) o es ing. The accu acy was de e mined
using he oo mean squa e e o , which eached a alue
o 3.20%. The au ho s no e a limi a ion o he s udy
which is he use o epidu al in almos all women du ing
he signal acquisi ion, which could lead o a supp ession
o pa ien ac i i y and made i easie o ex ac he ECG
signal. In his s udy, eal eco ds we e used o es ing,
howe e , hese a e di e en eco ds han we used and an
objec i e compa ison is no possible.
•The combina ion o h ee me hods (Fas ICA, WT, FFT)
was es ed in [72] o de e mine he leng h o he QT
in e al and he T/QRS a io. The au ho s did no p esen
any s a is ical e alua ion, bu acco ding o hem, his
me hod is also sui able o he analysis o eco ds
wi h low ampli ude o he ECG componen . How-
e e , he me hod was es ed only on wo syn he ic
eco ds.
VOLUME 9, 2021 28625
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
FIGURE 15. De ec ion and analysis o he es ima ed ECG signal ea u es in compa ison wi h e e ence signal ob ained
wi h e al scalp elec ode ( he eco ding 08).
FIGURE 16. Example o he ongoing se ies o measu emen s o he esea ch eam om VSB–Technical Uni e si y o
Os a a.
•The e icien a ian o Fas ICA was es ed in [1]. The
au ho s p esen ed he s a is ical e alua ion o he de e -
mina ion o HR, hey achie ed an a e age accu acy
o 94.79%, which is a sligh ly wo se esul han ha
achie ed by ou p oposed me hod. When e alua ing
he T/QRS a io, an a e age accu acy o 92.49% was
achie ed, which is a signi ican ly be e esul han we
achie ed. I should be no ed ha he e icien a ian
o Fas ICA was es ed only on syn he ic eco ds and
ou p oposed ICA-RLS on eal eco ds. In addi ion, he
au ho s deal wi h he ST wa e o m classi ica ion, which
achie ed an a e age accu acy o 79.87%.
•The au ho s in he [50] es ed he empla e sub ac ion
me hod and deal wi h he leng h o he QT in e al in
he mo phological analysis. No s a is ical e alua ion was
p esen ed in he s udy, bu acco ding o he au ho s he
me hod was e icien especially o pa ly noisy signals,
bu good esul s we e also achie ed o e y noisy da a.
•The model-based es ima ion was used in [20], whe e he
au ho s p ima ily deal wi h he analysis o he leng h o
he QT in e al. The au ho s used hei own eal eco ds
eco ded on 58 women (ges a ional age: 20–41 weeks)
o es ing. The me hod was es ed on no mal (physi-
ological) eco ds, bu also on eco ds wi h abno mal-
i ies (e.g. b adyca dia, achyca dia, hea anomalies,
hea ailu e, placen al dys unc ion). A di e ence o
<5.00% was achie ed be ween he e e ence and es i-
ma ed alues.
One o he challenges ha u u e esea ch should ocus on
is he analysis o o he mo phological elemen s ha can help
o u he e ine he e al dis ess de e mina ion. Fig. 15 shows
an example o de ec ion and analysis o he ex ac ed signal
o a high quali y. These signals could be used o pe o m mo -
phological analysis o any pa o he ECG cycle, especially
he QT in e al, which is signi ican ly sho ened due o e al
dis ess.
In o de o e i y he unc ionali y o he p oposed hyb id
sys em in a la ge numbe o pa ien s, i will be necessa y
o c ea e a quali y da abase o eal eco dings. Ideally, he
da abase should con ain a su icien numbe o eco dings
o adequa e leng h co e ing possible a ia ions in he e al
posi ion and he p egnancy s age. Fu he esea ch should be
also ocused on he e ec s o he elec ode placemen and
sys em con igu a ion. This would help in de eloping a new
diagnos ic sys em based on he simul aneous moni o ing o
HR and NI-ST-analysis wi h he bene i o main aining he
non-in asi eness o he examina ion while e ining he e al
dis ess diagnos ics.
In ou u u e esea ch, ou eam in ends o con inue his
wo k and ocus p ima ily on c ea ing ou own aECG da ase
wi h a la ge numbe o eco ds o he pu pose o es ing
and alida ing a ious app oaches o ECG signal p ocessing
and ex ac ion. Ou da ase will con ain eco ds wi h a i-
ous ges a ional age and di e en e al posi ions. In addi ion
o physiological eco ds, pa hological eco ds will also be
p esen . Thanks o his, a ious si ua ions and e ec s can be
es ed and hus in es iga e which signal p ocessing me hod
is sui able o gi en pu pose in clinical p ac ice. Mo eo e ,
all eco ds will con ain in o ma ion abou he ype o he
signal (physiological o pa hological), ges a ional age, e al
posi ion and so on. Abdominal signals will be mul ichannel,
a he same ime, con inuous measu emen o di ec ECG
will be pe o med using a ans aginal e al scalp elec ode
wi h simul aneous scanning using CTG. This will p o ide
alid e e ences o alida ing whe he he ex ac ion o ECG
signals was success ul in e ms o bo h HR moni o ing and
mo phological analysis. In addi ion, e e ence anno a ions
om he di ec ECG signal will be c ea ed o each eco d.
The anno a ions will con ain he exac posi ions o e al R
peaks o possible es ing o he accu acy o he HR de e mi-
na ion, which will also be possible o compa e wi h he HR
ace p o ided by CTG de ice. Fu he mo e, he anno a ions
28626 VOLUME 9, 2021
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
will con ain e e ence ma ke s o he ST analysis and hus he
mo phological analysis o he ex ac ed ECG cu e can be
es ed and e alua ed. Fig. 16 shows a se ies o measu emen s
al eady pe o med on se e al p egnan olun ee s. The igu e
shows he use o CTG o con inuous moni o ing o HR.
VI. CONCLUSION
This s udy in es iga ed he e ec i eness o he hyb id
ICA-RLS me hod o he mo phological analysis o
he NI- ECG signal. Tes s on eal eco dings om he
ADFECGDB da abase ha e shown ha in addi ion o
accu a e HR es ima ion, i is also possible o pe o m
non-in asi e mo phological analysis o he ECG signals.
The p ima y objec i e o he s udy was o pe o m he ST seg-
men analysis in o de o inc ease he accu acy o e al dis ess
diagnosis. The abili y o accu a ely de e mine HR and pe -
o m ST segmen analysis was assessed using objec i e qual-
i y indices such as ACC, SE, PPV, and F1, bu also wi h a help
o g aphical e alua ion me hods (Bland-Al man plo s and
HR aces). High accu acy o HR es ima ion, e lec ed in he
alues o he mean µwi hin he ange om 0.01 o 1.39 bpm
and he mean ±1.96σ om 1.76 o 14.08 bpm, was achie ed
o all eigh pa ien s examined. Based on he compa ison wi h
he esul s achie ed using in asi e means o e al moni o ing,
he p oposed algo i hm has p o en also i s abili y o pe o m
he high quali y ST segmen analysis. Fo 7 ou o 8 pa ien s
we we e able o de e mine T/QRS a io p ecisely wi h he
alues o he mean |µ| anging om 0.0025 o 0.0307 and
±1.96σ om 0.0043 o 0.1942. The esul s o his s udy
demons a e ha mo phological analysis can be pe o med on
NI- ECG signal while achie ing simila esul s as when di ec
ECG signal is used. The u u e esea ch will aim o ex end
his s udy o he analysis o o he ECG signal ea u es, such
as QT in e al.
ETHICS STATEMENT
The s udy p o ocol was app o ed by he E hical Commi ee o
he Silesian Medical Uni e si y, Ka owice, Poland (NN-013-
345/02). Subjec s ead he app o ed consen o m and ga e
w i en in o med consen o pa icipa e in he s udy.
REFERENCES
[1] E. C. Ka ounis, M. G. Tsipou as, C. Papaloukas, D. G. Tsalikakis,
K. K. Naka, and D. I. Fo iadis, ‘‘A non-in asi e me hodology o e al
moni o ing du ing p egnancy,’’ Me hods In . Med., ol. 49, no. 03,
pp. 238–253, 2010.
[2] J. Jezewski, A. Pawlak, K. Ho oba, J. W obel, R. Czabanski, and
M. Jezewski, ‘‘Selec ed design issues o he medical cybe -physical sys-
em o elemoni o ing p egnancy a home,’’ Mic op ocesso s Mic osys .,
ol. 46, pp. 35–43, Oc . 2016.
[3] T. Y. Euliano, M. T. Nguyen, S. Da manjian, S. P. McGo ay, N. Euliano,
A. Onkala, and A. R. G egg, ‘‘Moni o ing u e ine ac i i y du ing labo : A
compa ison o 3 me hods,’’ Ame . J. Obs e ics Gynecol., ol. 208, no. 1,
pp. 66.e1–66.e6, Jan. 2013.
[4] H. Y. Chen, S. P. Chauhan, C. V. Anan h, A. M. Vin zileos, and
A. Z. Abuhamad, ‘‘Elec onic e al hea a e moni o ing and i s ela-
ionship o neona al and in an mo ali y in he Uni ed S a es,’’ Ame . J.
Obs e ics Gynecol., ol. 204, no. 6, pp. 491.e1–491.e10, 2011.
[5] T. P. Sa welle, ‘‘Elec onic e al moni o ing: A b idge oo a ,’’ J. Legal
Med., ol. 33, no. 3, pp. 313–379, Jul. 2012.
[6] J. W obel, D. Roj, J. Jezewski, K. Ho oba, T. Kupka, and M. Jezewski,
‘‘E alua ion o he obus ness o e al hea a e a iabili y measu es
o low signal quali y,’’ J. Med. Imag. Heal h In o ma ., ol. 5, no. 6,
pp. 1311–1318, No . 2015.
[7] J. Jezewski, K. Ho oba, D. Roj, J. W obel, T. Kupka, and
A. Ma onia, ‘‘E alua ing he e al hea a e baseline es ima ion
algo i hms by hei in luence on de ec ion o clinically impo an
pa e ns,’’ Biocybe ne ics Biomed. Eng., ol. 36, no. 4, pp. 562–573,
2016.
[8] R. Czabanski, M. Jezewski, J. W obel, K. Ho oba, and J. Jezewski,
‘‘A neu o- uzzy app oach o he classi ica ion o e al ca dio ocog ams,’’
in P oc. 14 h No dic-Bal ic Con . Biomed. Eng. Med. Phys., ol. 20,
R. Magja e ic, J. H. Nagel, A. Ka ashe , Y. Dekh ya , and J. Spigulis,
Eds. Be lin, Ge many: Sp inge , 2008, pp. 446–449.
[9] J. Reinha d, B. R. Hayes-Gill, Q. Yi, H. Ha zmann, and S. Schie meie ,
‘‘Compa ison o non-in asi e e al elec oca diog am o Dopple ca -
dio ocog am du ing he 1s s age o labo ,’’ J. Pe ina al Med., ol. 38,
no. 2, Jan. 2010.
[10] W. R. Cohen, S. Ommani, S. Hassan, F. G. Mi za, M. Solomon, R. B own,
B. S. Schi in, J. M. Himswo h, and B. R. Hayes-Gill, ‘‘Accu acy and
eliabili y o e al hea a e moni o ing using ma e nal abdominal su ace
elec odes: Ma e nal su ace elec ode e al moni o ing,’’ Ac a Obs e-
icia e Gynecologica Scandina ica, ol. 91, no. 11, pp. 1306–1313,
No . 2012.
[11] J. Jezewski, J. W obel, A. Ma onia, K. Ho oba, R. Ma inek, T. Kupka,
and M. Jezewski, ‘‘Is abdominal e al elec oca diog aphy an al e na i e
o Dopple ul asound o FHR a iabili y e alua ion?’’ F on ie s Phys-
iol., ol. 8, p. 305, May 2017.
[12] I. Ame -Wahlin, S. A ulkuma an, H. Hagbe g, K. Ma šál, and
G. Visse , ‘‘Fe al elec oca diog am: ST wa e o m analysis in in a-
pa um su eillance,’’ BJOG, In . J. Obs e ics Gynaecology, ol. 114,
no. 10, pp. 1191–1193, Sep. 2007.
[13] I. Ame -Wåhlin, C. Hells en, H. No én, H. Hagbe g, A. He bs ,
I. Kjellme , H. Lilja, C. Lindo , M. Månsson, L. Må ensson,
P. Olo sson, A.-K. Sunds öm, and K. Ma šál, ‘‘Ca dio ocog aphy only
e sus ca dio ocog aphy plus ST analysis o e al elec oca diog am o
in apa um e al moni o ing: A swedish andomised con olled ial,’’
Lance , ol. 358, no. 9281, pp. 534–538, Aug. 2001.
[14] J. Wes ga e, M. Ha is, J. S. H. Cu now, and K. R. G eene, ‘‘Plymou h
andomized ial o ca dio ocog am only e sus ST wa e o m plus ca -
dio ocog am o in apa um moni o ing in 2400 cases,’’ Ame . J. Obs e -
ics Gynecology, ol. 169, no. 5, pp. 1151–1160, No . 1993.
[15] J. P. Neilson, ‘‘Fe al elec oca diog am (ECG) o e al moni o ing du ing
labou ,’’ Coch ane Da abase Sys ema ic Re ., ol. 25, no. 1, Dec. 2015.
[16] Sameni, ‘‘A e iew o e al ECG signal p ocessing issues and p omising
di ec ions,’’ Open Pacing, Elec ophysiology The apy J., Jan. 2010.
[17] G. D. Cli o d, I. Sil a, J. Beha , and G. B. Moody, ‘‘Non-in asi e
e al ECG analysis,’’ Physiological Meas., ol. 35, no. 8, pp. 1521–1536,
Aug. 2014.
[18] J. Jezewski, K. Ho oba, A. Ma onia, A. Gacek, and M. Be nys,
‘‘A new app oach o ca dio ocog aphic e al moni o ing based on analysis
o bioelec ical signals,’’ in P oc. 25 h Annu. In . Con . IEEE Eng. Med.
Biol. Soc., Sep. 2003, pp. 3145–3148.
[19] M. A. Oudijk, A. Kwee, G. H. A. Visse , S. Blad, E. J. Meijboom, and
K. G. Rosen, ‘‘The e ec s o in apa um hypoxia on he e al QT in e -
al,’’ BJOG, In . J. Obs e ics Gynaecology, ol. 111, no. 7, pp. 656–660,
Jul. 2004.
[20] N. Wida alla, Y. Kasaha a, Y. Kimu a, and A. Khandoke , ‘‘Model based
es ima ion o QT in e als in non-in asi e e al ECG signals,’’ PLoS
ONE, ol. 15, no. 5, May 2020, A . no. e0232769.
[21] B. F. Cuneo, J. F. S asbu ge , and R. T. Wakai, ‘‘The na u al his-
o y o e al long QT synd ome,’’ J. Elec oca diology, ol. 49, no. 6,
pp. 807–813, No . 2016.
[22] J. Beha , T. Zhu, J. Os e , A. Niksch, D. Y. Mah, T. Chun,
J. G eenbe g, C. Tanne , J. Ha op, R. Sameni, J. Wa d, A. J. Wol be g,
and G. D. Cli o d, ‘‘E alua ion o he e al QT in e al using non-
in asi e e al ECG echnology,’’ Physiological Meas., ol. 37, no. 9,
pp. 1392–1403, Sep. 2016.
[23] R. Ma inek, R. Kahanko a, J. Jezewski, R. Ja os, J. Mohylo a,
M. Fajkus, J. Nedoma, P. Janku, and H. Naze an, ‘‘Compa a i e e ec i e-
ness o ICA and PCA in ex ac ion o e al ECG om abdominal signals:
Towa d non-in asi e e al moni o ing,’’ F on ie s Physiol., ol. 9, p. 648,
May 2018.
VOLUME 9, 2021 28627
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
[24] R. Pe olis, V. Gin au as, and A. K isciukai is, ‘‘Mul is age p incipal
componen analysis based me hod o abdominal ECG decomposi ion,’’
Physiological Meas., ol. 36, no. 2, pp. 329–340, Feb. 2015.
[25] Q. Yu, H. Yan, L. Song, W. Guo, H. Liu, J. Si, and Y. Zhao, ‘‘Au oma ic
iden i ying o ma e nal ECG sou ce when applying ICA in e al ECG
ex ac ion,’’ Biocybe ne ics Biomed. Eng., ol. 38, no. 3, pp. 448–455,
2018.
[26] K. D. Desai and M. S. Sankhe, ‘‘A eal- ime e al ECG ea u e ex ac ion
using mul iscale disc e e wa ele ans o m,’’ in P oc. 5 h In . Con .
Biomed. Eng. In o ma ., Chongqing, China, Oc . 2012, pp. 407–412.
[27] K. Assaleh, ‘‘Ex ac ion o e al elec oca diog am using adap i e neu o-
uzzy in e ence sys ems,’’ IEEE T ans. Biomed. Eng., ol. 54, no. 1,
pp. 59–68, Jan. 2007.
[28] R. Ma inek, R. Kahanko a, H. Naze an, J. Konecny, J. Jezewski,
P. Janku, P. Bilik, J. Zidek, J. Nedoma, and M. Fajkus, ‘‘Non-in asi e e al
moni o ing: A ma e nal su ace ECG elec ode placemen -based no el
app oach o op imiza ion o adap i e il e con ol pa ame e s using he
LMS and RLS algo i hms,’’ Senso s, ol. 17, no. 5, p. 1154, May 2017.
[29] R. Ja os, R. Ma inek, R. Kahanko a, and J. Kozio ek, ‘‘No el
hyb id ex ac ion sys ems o e al hea a e a iabili y moni o ing
based on non-in asi e e al elec oca diog am,’’ IEEE Access, ol. 7,
pp. 131758–131784, 2019.
[30] K. Ba no a, R. Ma inek, R. Ja os, and R. Kahanko a, ‘‘Hyb id me hods
based on empi ical mode decomposi ion o non-in asi e e al hea a e
moni o ing,’’ IEEE Access, ol. 8, pp. 51200–51218, 2020.
[31] M. Lukoše ičius and V. Ma ozas, ‘‘Nonin asi e e al QRS de ec ion
using an echo s a e ne wo k and dynamic p og amming,’’ Physiological
Meas., ol. 35, no. 8, pp. 1685–1697, Aug. 2014.
[32] L. Billeci and M. Va anini, ‘‘A combined independen sou ce sepa a ion
and quali y index op imiza ion me hod o e al ECG ex ac ion om
abdominal ma e nal leads,’’ Senso s, ol. 17, no. 5, p. 1135, May 2017.
[33] T. Kazmi, F. Rad e , and S. Khan, ‘‘ST analysis o he e al ECG, as an
adjunc o e al hea a e moni o ing in labou : A e iew,’’ Oman Med.
J., ol. 26, no. 6, pp. 459–460, No . 2011.
[34] A. Sacco, J. Muglu, R. Na a a na ajah, and M. Hogg, ‘‘ST analysis o
in apa um e al moni o ing,’’ Obs e ician Gynaecologis , ol. 17, no. 1,
pp. 5–12, Jan. 2015.
[35] A. Daamouche, L. Hamami, N. Alajlan, and F. Melgani, ‘‘A wa ele
op imiza ion app oach o ECG signal classi ica ion,’’ Biomed. Signal
P ocess. Con ol, ol. 7, no. 4, pp. 342–349, Jul. 2012.
[36] R. Kahanko a, R. Ma inek, R. Ja os, K. Behbehani, A. Ma onia,
M. Jezewski, and J. A. Beha , ‘‘A e iew o signal p ocessing echniques
o non-in asi e e al elec oca diog aphy,’’ IEEE Re . Biomed. Eng.,
ol. 13, pp. 51–73, 2020.
[37] M. J. Rooijakke s, S. Song, C. Rabo i, S. G. Oei, J. W. M. Be gmans,
E. Can a o e, and M. Mischi, ‘‘In luence o elec ode placemen on signal
quali y o ambula o y p egnancy moni o ing,’’ Compu . Ma h. Me hods
Med., ol. 2014, pp. 1–12, Feb. 2014.
[38] J. Beha , F. And eo i, S. Zaunsede , Q. Li, J. Os e , and
G. D. Cli o d, ‘‘An ECG simula o o gene a ing ma e nal- oe al
ac i i y mix u es on abdominal ECG eco dings,’’ Physiological Meas.,
ol. 35, no. 8, pp. 1537–1550, Aug. 2014.
[39] J. Jezewski, A. Ma onia, T. Kupka, D. Roj, and R. Czabanski, ‘‘De e mi-
na ion o e al hea a e om abdominal signals: E alua ion o bea - o-
bea accu acy in ela ion o he di ec e al elec oca diog am,’’ Biomedi-
zinische Technik/Biomed. Eng., ol. 57, no. 5, pp. 383–394, Jan. 2012.
[40] A. L. Goldbe ge , L. A. N. Ama al, L. Glass, J. M. Hausdo ,
P. C. I ano , R. G. Ma k, J. E. Mie us, G. B. Moody, C.-K. Peng, and
H. E. S anley, ‘‘PhysioBank, PhysioToolki , and PhysioNe : Componen s
o a new esea ch esou ce o complex physiologic signals,’’ Ci cula ion,
ol. 101, no. 23, pp. e215–e220, Jun. 2000.
[41] L. Co de o and E. H. Hon, ‘‘Scalp abscess: A a e complica ion o e al
moni o ing,’’ J. Pedia ics, ol. 78, no. 3, pp. 533–537, Ma . 1971.
[42] D. M. Okada, A. W. Chow, and V. T. B uce, ‘‘Neona al scalp abscess and
e al moni o ing: Fac o s associa ed wi h in ec ion,’’ Ame . J. Obs e ics
Gynecology, ol. 129, no. 2, pp. 185–189, Sep. 1977.
[43] A. Salmelin, I. Wiklund, R. Bo inga, B. B o sson,
G. Ekman-O debe g, E. E. G im o s, U. Hanson, M. Blom, and
E. Pe sson, ‘‘Fe al moni o ing wi h compu e ized ST analysis du ing
labo : A sys ema ic e iew and me a-analysis: Compu e ized ST analysis
du ing labo ,’’ Ac a Obs e icia e Gynecologica Scandina ica, ol. 92,
no. 1, pp. 28–39, Jan. 2013.
[44] F. And eo i, J. Beha , S. Zaunsede , J. Os e , and G. D. Cli o d,
‘‘An open-sou ce amewo k o s ess- es ing non-in asi e oe al ECG
ex ac ion algo i hms,’’ Physiological Meas., ol. 37, no. 5, pp. 627–648,
May 2016.
[45] P.-C. Su, S. Mille , S. Id iss, P. Ba ke , and H.-T. Wu, ‘‘Reco e y o
he e al elec oca diog am o mo phological analysis om wo ans-
abdominal channels ia op imal sh inkage,’’ Physiological Meas., ol. 40,
no. 11, Dec. 2019, A . no. 115005.
[46] J. A. Beha , L. Bonnemains, V. Shulgin, J. Os e , O. Os as, and
I. Lakhno, ‘‘Nonin asi e e al elec oca diog aphy o he de ec ion o
e al a hy hmias,’’ P ena al Diagnosis, ol. 39, no. 3, pp. 178–187,
Feb. 2019.
[47] J. Beha , F. And eo i, J. Os e , and G. D. Cli o d, ‘‘A Bayesian
il e ing amewo k o accu a e ex ac ing o he non-in asi e
FECG mo phology,’’ in P oc. Compu . Ca diol., Sep. 2014,
pp. 53–56.
[48] G. Cli o d, R. Sameni, J. Wa d, J. Robinson, and A. J. Wol be g, ‘‘Clin-
ically accu a e e al ECG pa ame e s acqui ed om ma e nal abdominal
senso s,’’ Ame . J. Obs e ics Gynecol., ol. 205, no. 1, pp. 47.e1–47.e5,
Jul. 2011.
[49] I. Sil a, J. Beha , R. Sameni, T. Zhu, J. Os e , G. D. Cli o d, and
G. B. Moody, ‘‘Nonin asi e e al ECG: The physione /compu ing in
ca diology challenge 2013,’’ in P oc. Compu . Ca diol., Sep. 2013,
pp. 149–152.
[50] P. Podziemski and J. Gie al owski, ‘‘Fe al hea a e disco e y: Algo i hm
o de ec ion o e al hea a e om noisy, nonin asi e e al ECG eco d-
ings,’’ in P oc. Compu . Ca diol., Sep. 2013, pp. 333–336.
[51] E. Cas illo, D. P. Mo ales, G. Bo ella, A. Ga cía, L. Pa illa, and
A. J. Palma, ‘‘E icien wa ele -based ECG p ocessing o single-lead
FHR ex ac ion,’’ Digi . Signal P ocess., ol. 23, no. 6, pp. 1897–1909,
Dec. 2013.
[52] G. Liu and Y. Luan, ‘‘An adap i e in eg a ed algo i hm o non-
in asi e e al ECG sepa a ion and noise educ ion based on ICA-
EEMD-WS,’’ Med. Biol. Eng. Compu ., ol. 53, no. 11, pp. 1113–1127,
No . 2015.
[53] R. Sameni, M. B. Shamsollahi, and C. Ju en, ‘‘Fil e ing elec oca -
diog am signals using he ex ended Kalman il e ,’’ in P oc. IEEE
Eng. Med. Biol. 27 h Annu. Con ., Shanghai, China, Jan. 2005,
pp. 5639–5642.
[54] D. Panig ahy and P. K. Sahu, ‘‘Ex ac ion o e al ECG signal by an
imp o ed me hod using ex ended Kalman smoo he amewo k om
single channel abdominal ECG signal,’’ Aus alas. Phys. Eng. Sci. Med.,
ol. 40, no. 1, pp. 191–207, Ma . 2017.
[55] R. Sameni, C. Ju en, and M. B. Shamsollahi, ‘‘Wha ICA p o ides
o ECG p ocessing: Applica ion o nonin asi e e al ECG ex ac ion,’’
in P oc. IEEE In . Symp. Signal P ocess. In . Technol., Aug. 2006,
pp. 656–661.
[56] R. Swa nala h and D. V. P asad, ‘‘A no el echnique o ex ac ion o
FECG using mul i s age adap i e il e ing,’’ J. Appl. Sci., ol. 10, no. 4,
pp. 319–324, Feb. 2010.
[57] H. Hassanpou and A. Pa saei, ‘‘Fe al ECG ex ac ion using wa ele
ans o m,’’ in P oc. In . Con . Compu . In eligence Modeling Con ol
Au om. In . Con . In ell. Agen s Web Technol. In . Comme ce (CIMCA),
Sydney, NSW, Aus alia, No . 2006, p. 179.
[58] R. Ma inek, R. Kahanko a, J. Nedoma, M. Fajkus, and K. Chole o a,
‘‘Fe al ECG p ep ocessing using wa ele ans o m,’’ in P oc. 10 h In .
Con . Compu . Modeling Simula ion (ICCMS), Sydney, NSW, Aus alia,
2018, pp. 39–43.
[59] S. Wu, Y. Shen, Z. Zhou, L. Lin, Y. Zeng, and X. Gao, ‘‘Resea ch
o e al ECG ex ac ion using wa ele analysis and adap i e
il e ing,’’ Compu . Biol. Med., ol. 43, no. 10, pp. 1622–1627,
Oc . 2013.
[60] D. D. Ta alunga, I. Gussi, and R. S unga u, ‘‘A new me hod o
e al elec oca diog am denoising using blind sou ce sepa a ion and
empi ical mode decomposi ion,’’ Re ue Roumaine des Sci. Techn.,
se ie Élec o echnique e Éne ge ique, ol. 61, no. 1, pp. 94–98,
2016.
[61] P. Ghobadi Azba i, M. Abdolgha a , S. Mohaqeqi, M. Pooyan,
A. Ahmadian, and N. Ghanba zadeh Gash i, ‘‘A no el app oach o he
ex ac ion o e al elec oca diog am based on empi ical mode decompo-
si ion and co ela ion analysis,’’ Aus alas. Phys. Eng. Sci. Med., ol. 40,
no. 3, pp. 565–574, Sep. 2017.
28628 VOLUME 9, 2021

R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
[62] A. Bin Queyam, S. Kuma Pahuja, and D. Singh, ‘‘Quan i ica ion o e o-
ma e nal hea a e om abdominal ECG signal using empi ical mode
decomposi ion o hea a e a iabili y analysis,’’ Technologies, ol. 5,
no. 4, p. 68, Oc . 2017.
[63] M. Niknaza , B. Ri e , and C. Ju en, ‘‘Fe al ECG ex ac ion by
ex ended s a e Kalman il e ing based on single-channel eco d-
ings,’’ IEEE T ans. Biomed. Eng., ol. 60, no. 5, pp. 1345–1352,
May 2013.
[64] M. Sugan hy and S. Manjula, ‘‘Enhancemen o SNR in e al ECG signal
ex ac ion using combined SWT and WLSR in pa allel EKF,’’ Clus e
Compu ., ol. 22, no. S2, pp. 3875–3881, Ma . 2019.
[65] E. Fo iadou and R. Vullings, ‘‘Mul i-channel e al ECG denoising wi h
deep con olu ional neu al ne wo ks,’’ F on ie s Pedia ics, ol. 8, p. 508,
Aug. 2020.
[66] G. Camps-Valls, M. Ma ínez-Sobe , E. So ia-Oli as, R. Magdalena-
Benedi o, J. Calpe-Ma a illa, and J. Gue e o-Ma ínez, ‘‘Foe al ECG
eco e y using dynamic neu al ne wo ks,’’ A i . In ell. Med., ol. 31,
no. 3, pp. 197–209, Jul. 2004.
[67] M. Nasi i, K. Faez, and A. M. Nas abadi, ‘‘A new me hod o ex ac ion
o e al elec oca diog am signal based on adap i e ne o- uzzy in e ence
sys em,’’ in P oc. IEEE In . Con . Signal Image P ocess. Appl. (ICSIPA),
Kuala Lumpu , Malaysia, No . 2011, pp. 456–461.
[68] J. Beha , A. E. Johnson, J. Os e , and G. Cli o d, ‘‘An echo s a e neu al
ne wo k o oe al ECG ex ac ion op imised by andom sea ch,’’ in P oc.
Ad . Neu al In . P ocess. Sys , 2013, pp. 1–5.
[69] V. Za zoso and A. K. Nandi, ‘‘Nonin asi e e al elec oca diog am
ex ac ion: Blind sepa a ion e sus adap i e noise cancella ion,’’ IEEE
T ans. Biomed. Eng., ol. 48, no. 1, pp. 12–18, 2001.
[70] A. Gup a, M. C. S i as a a, V. Khandelwal, and A. Gup a, ‘‘A no el
app oach o e al ECG ex ac ion and enhancemen using blind sou ce
sepa a ion (BSS-ICA) and adap i e e al ECG enhance (AFE),’’ in
P oc. 6 h In . Con . In ., Commun. Signal P ocess., Singapo e, 2007,
pp. 1–4.
[71] M. Ko as, ‘‘Combined applica ion o independen componen analysis
and p ojec i e il e ing o e al ECG ex ac ion,’’ Biocybe n. Biomed.
Eng., ol. 28, no. 1, p. 75, 2008.
[72] V. Ionescu, ‘‘Fe al ECG ex ac ion om mul ichannel abdominal ECG
eco dings o heal h moni o ing du ing labo ,’’ P ocedia Technol.,
ol. 22, pp. 682–689, 2016.
[73] R. Ma ín-Clemen e, J. L. Cama go-Oli a es, S. Ho nillo-Mellado,
M. Elena, and I. Román, ‘‘Fas echnique o nonin asi e e al ECG
ex ac ion,’’ IEEE T ans. Biomed. Eng., ol. 58, no. 2, pp. 227–230,
Feb. 2011.
[74] P. P. Kanjilal, S. Pali , and G. Saha, ‘‘Fe al ECG ex ac ion om single-
channel ma e nal ECG using singula alue decomposi ion,’’ IEEE T ans.
Biomed. Eng., ol. 44, no. 1, pp. 51–59, 1997.
[75] M. Va anini, G. Ta a isco, L. Billeci, A. Mace a a, G. Pioggia, and
R. Balocchi, ‘‘A mul i-s ep app oach o non-in asi e e al ECG anal-
ysis,’’ in P oc. Compu . Ca diol., Sep. 2013, pp. 281–284.
[76] M. Ko as, J. Gi aldo, S. H. Con e as-O iz, and G. I. B. Lasp illa,
‘‘Fe al ECG ex ac ion using independen componen analysis by jade
app oach,’’ in P oc. 13 h In . Con . Med. In . P ocess. Anal., J. B ie a,
J. D. Ga cía, N. Lepo e, and E. Rome o, Eds. San And es Island,
Colombia: SPIE, No . 2017, p. 55.
[77] K. Su ya, K. K. Abdul Majeed, and R. V. Ra i, ‘‘Diagnosis o e al
a hy hmia using JADE algo i hm,’’ in P oc. 7 h In . Con . Sma S uc .
Sys . (ICSSS), Chennai, India, Jul. 2020, pp. 1–5.
[78] L. Yuan, Z. Zhou, Y. Yuan, and S. Wu, ‘‘An imp o ed Fas ICA me hod o
e al ECG ex ac ion,’’ Compu . Ma h. Me hods Med., ol. 2018, pp. 1–7,
May 2018.
[79] Y. Wang, Y. Fu, and Z. He, ‘‘Fe al elec oca diog am ex ac ion based on
as ICA and wa ele denoising,’’ in P oc. 2nd IEEE Ad . In . Manage.,
Communica es, Elec on. Au om. Con ol Con . (IMCEC), Xi’an, China,
May 2018, pp. 466–469.
[80] J. J. R. Immanuel, V. P abhu, V. J. Ch is ophe aj, D. Suguma , and
P. T. Vana hi, ‘‘Sepa a ion o ma e nal and e al ECG signals om
he mixed sou ce signal using FASTICA,’’ P ocedia Eng., ol. 30,
pp. 356–363, Jan. 2012.
[81] J. L. Cama go-Oli a es, R. Ma ín-Clemen e, S. Ho nillo-Mellado,
M. M. Elena, and I. Román, ‘‘The ma e nal abdominal ECG as inpu o
MICA in he e al ECG ex ac ion p oblem,’’ IEEE Signal P ocess. Le .,
ol. 18, no. 3, pp. 161–164, Ma . 2011.
[82] D. Ta alunga, M. Ungu eanu, R. S unga u, and W. Wol , ‘‘Pe o mance
compa ison o ou ICA algo i hms applied o ECG ex ac ion om
ansabdominal eco dings,’’ in P oc. In . Symp. Signals, Ci cui s Sys .
(ISSCS), Iasi, Romania, Jun. 2011, pp. 1–4.
[83] D. A. Ramli, Y. H. Shiong, and N. Hassan, ‘‘Blind sou ce sepa a ion
(BSS) o mixed ma e nal and e al elec oca diog am (ECG) signal:
A compa a i e s udy,’’ P ocedia Compu . Sci., ol. 176, pp. 582–591,
Jan. 2020.
[84] P. J. He, X. M. Chen, Y. Liang, and H. Z. Zeng, ‘‘Ex ac ion o e al ECG
using single channel blind sou ce sepa a ion algo i hm based on mul i-
algo i hm usion,’’ in P oc. MATEC Web Con ., ol. 44, p. 01026, 2016.
[85] S. Ziani and Y. El Hassouani, ‘‘Fe al elec oca diog am analysis based
on LMS adap i e il e ing and complex con inuous wa ele 1-D,’’ in
Big Da a and Ne wo ks Technologies, ol. 81, Y. Fa haoui, Ed. Cham,
Swi ze land: Sp inge , 2020, pp. 360–366.
[86] R. Ma inek, R. Kahanko a, H. Sku o a, P. Koudelka, J. Zidek, and
J. Kozio ek, ‘‘Adap i e signal p ocessing echniques o ex ac ing
abdominal e al elec oca diog am,’’ in P oc. 10 h In . Symp. Commun.
Sys ., Ne w. Digi . Signal P ocess. (CSNDSP), P ague, Czech Republic,
Jul. 2016, pp. 1–6.
[87] J. Beha , A. Johnson, G. D. Cli o d, and J. Os e , ‘‘A compa ison o single
channel e al ECG ex ac ion me hods,’’ Ann. Biomed. Eng., ol. 42, no. 6,
pp. 1340–1353, Jun. 2014.
[88] S.-J. Liu, D.-L. Liu, J.-Q. Zhang, and Y.-J. Zeng, ‘‘Ex ac ion o e al
elec oca diog am using ecu si e leas squa es and no malized leas
mean squa es algo i hms,’’ in P oc. 3 d In . Con . Ad . Compu . Con ol,
Ha bin, China, Jan. 2011, pp. 333–336.
[89] D. Ay es-de-Campos, C. Y. Spong, E. Chand aha an, ‘‘FIGO consensus
guidelines on in apa um e al moni o ing: Ca dio ocog aphy,’’ In . J.
Gynecology Obs e ics, ol. 131, no. 1, pp. 13–24, Oc . 2015.
[90] F. News, ‘‘Guidelines o he use o e al moni o ing,’’ In . J. Gynaecol
Obs e , ol. 25, no. 1, pp. 67–159, 1987.
[91] P. Olo sson, H. No én, and A. Ca lsson, ‘‘New FIGO and swedish
in apa um ca dio ocog aphy classi ica ion sys ems inco po a ed in he
e al ECG ST analysis (STAN) in e p e a ion algo i hm: Ag eemen s
and disc epancies in ca dio ocog aphy classi ica ion and e alua ion o
signi ican ST e en s,’’ Ac a Obs e icia e Gynecologica Scandina ica,
ol. 97, no. 2, pp. 219–228, Feb. 2018.
[92] D. M. F. Gibb and S. A ulkuma an, Fe al Moni o ing in P ac ice, 4 h ed.
London, U.K.: Chu chill Li ings one, 2017.
[93] I. Ame -Wahlin, A. Ugwumadu, B. M. Yli, A. Kwee, S. Timonen,
V. Cole, D. Ay es-de-Campos, G.-E. Ro h, C. Schwa z, L. A. Ramenghi,
T. Tod os, V. Ehlinge , and C. Vayssie e, ‘‘Fe al elec oca diog aphy
ST-segmen analysis o in apa um moni o ing: A c i ical app aisal o
con lic ing e idence and a way o wa d,’’ Ame . J. Obs e ics Gynecol.,
ol. 221, no. 6, pp. e11.577–e11.601, Dec. 2019.
[94] H. Lilja, K. R. G eene, K. Ka lsson, and K. G. Rosen, ‘‘ST wa e o m
changes o he e al elec oca diog am du ing labou —A clinical s udy,’’
BJOG, In . J. Obs e ics Gynaecol., ol. 92, no. 6, pp. 611–617, Jun. 1985.
[95] S. A ulkuma an, H. Lilja, K. Lindec an z, S. S. Ra nam, A. S. Tha a asah,
and K. G. Rosén, ‘‘Fe al ECG wa e o m analysis should imp o e e al
su eillance in labou ,’’ J. Pe ina al Med., ol. 18, no. 1, pp. 13–22,
Jan. 1990.
[96] K. G. Rosen, ‘‘Al e a ions in he e al elec oca diog am as a sign o e al
asphyxia–expe imen al da a wi h a clinical implemen a ion,’’ J. Pe ina al
Med., ol. 14, no. 6, pp. 355–363, 1986.
[97] P. Ch is ian and B. Da idsen, ‘‘The signi icance o he oe al elec oca -
diog am du ing labou wi h de ailed epo o one case,’’ Ac a Obs e icia
e Gynecologica Scandina ica, ol. 50, no. 1, pp. 45–49, Jan. 1971.
[98] G. Pa di, E. Tucci, A. Ude zo, and D. Zanini, ‘‘Fe al elec oca diog am
changes in ela ion o e al hea a e pa e ns du ing labo ,’’ Ame . J.
Obs e ics Gynecology, ol. 118, no. 2, pp. 243–250, Jan. 1974.
[99] L.-Y. Chen and C.-J. Lu, ‘‘An imp o ed independen componen analysis
algo i hm based on a i icial immune sys em,’’ In . J. Mach. Lea n.
Compu ., ol. 3, no. 1, pp. 93–97, 2013.
[100] E. Sulas, M. U u, R. Tumba ello, L. Ra o, and D. Pani, ‘‘Sys em-
a ic analysis o single- and mul i- e e ence adap i e il e s o non-
in asi e e al elec oca diog aphy,’’ Ma h. Biosci. Eng., ol. 17, no. 1,
pp. 286–308, 2020.
[101] A. Hy ä inen and E. Oja, ‘‘Independen componen analysis: Algo-
i hms and applica ions,’’ Neu al Ne w., ol. 13, nos. 4–5, pp. 411–430,
Jun. 2000.
VOLUME 9, 2021 28629
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
[102] J. Benes y and T. Gänsle , ‘‘New insigh s in o he RLS algo i hm,’’
EURASIP J. Ad . Signal P ocess., ol. 2004, no. 3, Dec. 2004,
A . no. 243857.
[103] A. Ma onia, J. Jezewski, T. Kupka, M. Jezewski, K. Ho oba, J. W obel,
R. Czabanski, and R. Kahankowa, ‘‘Fe al elec oca diog ams, di ec and
abdominal wi h e e ence hea bea anno a ions,’’ Sci. Da a, ol. 7, no. 1,
p. 200, Dec. 2020.
[104] R. Kahanko a, ‘‘Ad anced signal p ocessing me hods o non-in asi e
e al elec oca diog am ex ac ion,’’ Ph.D. disse a ion, VSB Tech. Uni .
Os a a, Os a a, Czech Republic, 2019.
[105] R. Ja os, R. Ma inek, R. Kahanko a, and K. Ba no a. (Jan. 2021). ST
Anno a ions o ADFECGDB Da abase. [Online]. A ailable: h ps://ieee-
da apo .o g/documen s/s -anno a ions-ad ecgdb-da abase
[106] C. Li, C. Zheng, and C. Tai, ‘‘De ec ion o ECG cha ac e is ic poin s using
wa ele ans o ms,’’ IEEE T ans. Biomed. Eng., ol. 42, no. 1, pp. 21–28,
1995.
[107] A. Gha a i, H. Golbayani, and M. Ghasemi, ‘‘A new ma hema ical based
QRS de ec o using con inuous wa ele ans o m,’’ Compu . Elec . Eng.,
ol. 34, no. 2, pp. 81–91, Ma . 2008.
[108] P. Du, W. A. Kibbe, and S. M. Lin, ‘‘Imp o ed peak de ec ion in
mass spec um by inco po a ing con inuous wa ele ans o m-based
pa e n ma ching,’’ Bioin o ma ics, ol. 22, no. 17, pp. 2059–2065,
Sep. 2006.
[109] E. Chand aha an, Ed., Handbook CTG In e p e a ion: F om Pa e ns o
Physiology. Camb idge, U.K.: Camb idge Uni . P ess, 2017.
[110] M. Elgendi, B. Esko ie , and D. Abbo , ‘‘Fas wa e de ec ion calib a ed
by clinical knowledge wi h anno a ion o p and wa es,’’ Senso s, ol. 15,
no. 7, pp. 17693–17714, Jul. 2015.
[111] J. Pan and W. J. Tompkins, ‘‘A eal- ime QRS de ec ion algo-
i hm,’’ IEEE T ans. Biomed. Eng., ols. BME–32, no. 3, pp. 230–236,
Ma . 1985.
[112] S. Ve ma and R. Vashis ha, ‘‘E icien RR-in e al ime se ies o mula ion
o hea a e de ec ion,’’ in Mul imedia, Signal P ocessing and Com-
munica ion Technologies (IMPACT), 2013 In e na ional Con e ence on,
Aliga h, India, No . 2013, pp. 84–87.
[113] G. B. Moody and R. G. Ma k, ‘‘The impac o he MIT-BIH a hy h-
mia da abase,’’ IEEE Eng. Med. Biol. Mag., ol. 20, no. 3, pp. 45–50,
May/Jun. 2001.
[114] J. A eaga-Falconi, H. Al Osman, and A. El Saddik, ‘‘R-peak de ec ion
algo i hm based on di e en ia ion,’’ in P oc. IEEE 9 h In . Symp. In ell.
Signal P ocess. (WISP), Siena, I aly, May 2015, pp. 1–4.
[115] S. K a a, K. Ouni, and N. Ellouze, ‘‘ECG signal maxima de ec ion using
wa ele ans o m,’’ in P oc. IEEE In . Symp. Ind. Elec on., Mon eal,
QC, Canada, Jul. 2006, pp. 700–703.
[116] R. Haddadi, E. Abdelmounim, M. El Hanine, and A. Belaguid, ‘‘Disc e e
wa ele ans o m based algo i hm o ecogni ion o QRS complexes,’’
in P oc. In . Con . Mul imedia Compu . Sys . (ICMCS), Ma akech,
Mo occo, Ap . 2014, pp. 375–379.
[117] J. P. Ma inez, R. Almeida, S. Olmos, A. P. Rocha, and P. Laguna,
‘‘A wa ele -based ECG delinea o : E alua ion on s anda d
da abases,’’ IEEE T ans. Biomed. Eng., ol. 51, no. 4, pp. 570–581,
Ap . 2004.
[118] K. F. Tan, ‘‘De ec ion o he QRS complex, p wa e and wa e in
elec oca diog am,’’ in P oc. 1s In . Con . Ad . Med. Signal In . P ocess.,
B is ol, U.K., 2000, pp. 41–47.
[119] P. Laguna, R. Ma k, A. Goldbe g, and G. Moody, ‘‘A da abase o
e alua ion o algo i hms o measu emen o QT and o he wa e o m
in e als in he ECG,’’ in P oc. Compu . Ca diol., Lund, Sweden, 1997,
pp. 673–676.
[120] S. Lixin, W. Yuhong, and G. Lili, ‘‘A p-wa e de ec ion me hod in ECG
based on mul i- ea u e and wa ele -ampli ude h eshold,’’ in P oc. In .
Con . In . Commun. Technol. (ICT). Nanjing, China: Ins i u ion o Engi-
nee ing and Technology, 2014, p. 3.093–3.093.
[121] M. Rakshi , D. Panig ahy, and P. K. Sahu, ‘‘EKF wi h PSO echnique
o delinea ion o p and wa e in elec oca diog am(ECG) signal,’’ in
P oc. 2nd In . Con . Signal P ocess. In eg . Ne w. (SPIN), Noida, India,
Feb. 2015, pp. 696–701.
[122] C. Bock, ‘‘ECG signal analysis based on he wa ele ans o m,’’
Ph.D. disse a ion, JKU Linz, Linz, Aus ia, 2015.
[123] M. Ch obak and J. Kozumplik, ‘‘Compa ison o he mos widely used
algo i hms o he de ec ion o QT in e al,’’ Elek o e ue, ol. 15, no. 4,
pp. 233–237, 2013.
RADEK MARTINEK (Membe , IEEE) was bo n
in Czech Republic, in 1984. He ecei ed he mas-
e ’s deg ee in in o ma ion and communica ion
echnology om he VSB—Technical Uni e si y
o Os a a, in 2009. In 2014, he success ully
de ended his disse a ion hesis i led ‘‘The Use o
Complex Adap i e Me hods o Signal P ocessing
o Re ining he Diagnos ic Quali y o he Abdom-
inal Fe al Elec oca diog am.’’ Since 2012 he has
been a Resea ch Fellow wi h he VSB—Technical
Uni e si y o Os a a. In 2017, he became an Associa e P o esso in echnical
cybe ne ics a e de ending he habili a ion hesis i led ‘‘Design and Op i-
miza ion o Adap i e Sys ems o Applica ions o Technical Cybe ne ics and
Biomedical Enginee ing Based on Vi ual Ins umen a ion.’’ He has been an
Associa e P o esso wi h he VSB—Technical Uni e si y o Os a a, since
2017. His cu en esea ch in e es s include digi al signal p ocessing (linea
and adap i e il e ing, so compu ing a i icial in elligence and adap i e
uzzy sys ems, non-adap i e me hods, biological signal p ocessing, and
digi al p ocessing o speech signals); wi eless communica ions (so wa e-
de ined adio); and powe quali y imp o emen . He has mo e han 200 jou -
nal and con e ence pape s in his esea ch a eas.
RADANA KAHANKOVA was bo n in Opa a,
Czech Republic, in 1991. She ecei ed he
bachelo ’s deg ee and he mas e ’s deg ee
in biomedical enginee ing om he Depa -
men o Cybe ne ics and Biomedical Enginee -
ing, VSB—Technical Uni e si y o Os a a,
in 2014 and 2016, espec i ely, and he Ph.D.
deg ee in echnical cybe ne ics, in 2019. He
cu en esea ch in e es includes imp o ing he
quali y o elec onic e al moni o ing.
RENE JAROS was bo n in Os a a, Czech Repub-
lic, in 1992. He ecei ed he bachelo ’s deg ee
and he mas e ’s deg ee in biomedical enginee ing
om he Depa men o Cybe ne ics and Biomed-
ical Enginee ing, VSB—Technical Uni e si y o
Os a a, in 2015 and 2017, espec i ely, and he
Ph.D. deg ee in echnical cybe ne ics, in 2019. His
esea ch in e es includes e al elec oca diog a-
phy ( ECG) ex ac ion by using hyb id me hods.
KATERINA BARNOVA was bo n in Os a a,
Czech Republic, in 1993. She ecei ed he mas e ’s
deg ee om he Depa men o Cybe ne ics and
Biomedical Enginee ing, VSB—Technical Uni-
e si y o Os a a, in 2019, whe e she is cu en ly
pu suing he Ph.D. deg ee in echnical cybe ne -
ics. He esea ch in e es includes ad anced signal
p ocessing me hods, especially o e al elec oca -
diog aphy ( ECG) ex ac ion.
ADAM MATONIA was bo n in Poland, in 1975.
He ecei ed he M.Sc. deg ee in elec onic engi-
nee ing and he Ph.D. deg ee in biocybe ne ics and
biomedical enginee ing om he Silesian Uni e -
si y o Technology, Gliwice, Poland, in 2000 and
2018, espec i ely. He is cu en ly he P ojec
Leade wi h he Biomedical Signal P ocessing
Depa men , Łukasiewicz Resea ch Ne wo k—
Ins i u e o Medical Technology and Equipmen ,
Zab ze, Poland. His esea ch in e es s include e al
elec oca diog aphy, elec ohys e og aphy, mobile biomedical ins umen a-
ions, and so wa e de elopmen o compu e ized e al moni o ing sys ems.
He is a membe o he Polish Socie y o Biomedical Enginee ing. He was a
ellowship holde o he Founda ion o Polish Science in 2005.
28630 VOLUME 9, 2021
R. Ma inek e al.: Non-In asi e ECG Ex ac ion Based on No el Hyb id Me hod o In apa um ST Segmen Analysis
MICHAL JEZEWSKI was bo n in Zab ze, Poland,
in 1982. He ecei ed he M.Sc. deg ee in com-
pu e science and he Ph.D. deg ee in elec on-
ics om he Silesian Uni e si y o Technology,
Gliwice, Poland, in 2006 and 2011, espec i ely.
He is cu en ly wi h he Depa men o Cybe ne -
ics, Nano echnology and Da a P ocessing, Silesian
Uni e si y o Technology, Gliwice, Poland. His
esea ch in e es s include biomedical signal p o-
cessing and compu a ional in elligence me hods
wi h emphasis on uzzy clus e ing and uzzy classi ie s. He is a membe o
he Polish Socie y o Theo e ical and Applied Elec o echnics.
ROBERT CZABANSKI was bo n in Tychy,
Poland. He ecei ed he M.Sc. and Ph.D. deg ees
in elec onics and he D.Sc. deg ee in biocybe -
ne ics and biomedical enginee ing om he Sile-
sian Uni e si y o Technology, Gliwice, Poland,
in 1997, 2003, and 2018, espec i ely. He is
cu en ly wi h he Depa men o Cybe ne ics,
Nano echnology and Da a P ocessing, Silesian
Uni e si y o Technology, Gliwice, Poland. His
esea ch in e es s include uzzy and neu o- uzzy
modeling, lea ning heo y, and biomedical signal p ocessing. He is a membe
o he Polish Socie y o Theo e ical and Applied Elec o echnics.
KRZYSZTOF HOROBA was bo n in Poland,
in 1968. He ecei ed he M.S. deg ee in elec-
onic enginee ing om he Silesian Uni e si y o
Technology, Gliwice, Poland, in 1993, he Ph.D.
deg ee in medical science om he Uni e si y o
Medical Sciences, Poznan, Poland, in 2001, and
he D.Sc. deg ee in biocybe ne ics and biomed-
ical enginee ing om he Silesian Uni e si y o
Technology, in 2017. He is cu en ly he Head
o he Biomedical Signals P ocessing Depa men ,
Łukasiewicz Resea ch Ne wo k—Ins i u e o Medical Technology and
Equipmen , Zab ze, Poland. His esea ch in e es s include e al elec oca -
diog aphy, elec ohys e og aphy, as well as he so wa e de elopmen o he
compu e ized e al moni o ing sys ems. He is a membe o he Polish Socie y
o Biomedical Enginee ing. He was a ellowship holde o he Founda ion o
Polish Science in 1997.
JANUSZ JEZEWSKI (Senio Membe , IEEE) was
bo n in Zab ze, Poland. He ecei ed he M.Sc.
deg ee in elec onics om he Silesian Uni e -
si y o Technology, Gliwice, Poland, he Ph.D.
deg ee in biological sciences om he Uni e -
si y o Medical Sciences, Poznan, Poland, and
he D.Sc. deg ee in biocybe ne ics and biomedi-
cal enginee ing om he Ins i u e o Biocybe ne -
ics and Biomedical Enginee ing, Polish Academy
o Sciences, Wa saw, Poland. He is cu en ly he
Di ec o o Science o he Łukasiewicz Resea ch Ne wo k—Ins i u e o
Medical Technology and Equipmen , Zab ze. He has au ho ed o coau ho ed
mo e han 300 in e na ional jou nal and con e ence pape s. His esea ch
in e es s include biomedical ins umen a ion, digi al signal p ocessing, and
applica ion o compu a ional in elligence in medical cybe -physical sys ems.
He is a membe o he Commi ee o Biocybe ne ics and Biomedical Engi-
nee ing o he Polish Academy o Sciences, he Polish Socie y o Biomedical
Enginee ing, he Ins i u e o Physics and Enginee ing in Medicine, and he
Eu opean Socie y o Enginee ing and Medicine. He ecei ed he i le o
P o esso con e ed by he P esiden o he Republic o Poland.
VOLUME 9, 2021 28631