senso s
Re iew
Ad anced Bioelec ical Signal P ocessing Me hods: Pas ,
P esen , and Fu u e App oach—Pa III: O he Biosignals
Radek Ma inek 1,*,† , Ma ina Lad o a 1, Michaela Sidiko a 1, Rene Ja os 1, Khos ow Behbehani 2,
Radana Kahanko a 1and Aleksand a Kawala-S e niuk 3,*,†
Ci a ion: Ma inek, R.; Lad o a, M.;
Sidiko a, M.; Ja os, R.; Behbehani, K.;
Kahanko a, R.; Kawala-S e niuk, A.
Ad anced Bioelec ical Signal
P ocessing Me hods: Pas , P esen ,
and Fu u e App oach—Pa III: O he
Biosignals. Senso s 2021,21, 6064.
h ps://doi.o g/10.3390/s21186064
Academic Edi o : Paweł Pławiak
Recei ed: 26 July 2021
Accep ed: 7 Sep embe 2021
Published: 10 Sep embe 2021
Publishe ’s No e: MDPI s ays neu al
wi h ega d o ju isdic ional claims in
published maps and ins i u ional a il-
ia ions.
Copy igh : © 2021 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license (h ps://
c ea i ecommons.o g/licenses/by/
4.0/).
1Depa men o Cybe ne ics and Biomedical Enginee ing, VSB-Technical Uni e si y Os a a, FEECS,
708 00 Os a a, Czech Republic; [email p o ec ed] (M.L.); [email p o ec ed] (M.S.);
ene.ja [email p o ec ed] (R.J.); [email p o ec ed] (R.K.)
2College o Enginee ing, The Uni e si y o Texas in A ling on, A ling on, TX 76019, USA; [email p o ec ed]
3Facul y o Elec ical Enginee ing, Opole Uni e si y o Technology, Au oma ic Con ol and In o ma ics,
45-758 Opole, Poland
*Co espondence: [email p o ec ed] (R.M.); [email p o ec ed] (A.K.-S.)
† These au ho s con ibu ed equally o his wo k.
Abs ac :
Analysis o biomedical signals is a e y challenging ask in ol ing implemen a ion o
a ious ad anced signal p ocessing me hods. This a ea is apidly de eloping. This pape is a Pa
III pape , whe e he mos popula and e icien digi al signal p ocessing me hods a e p esen ed.
This pape co e s he ollowing bioelec ical signals and hei p ocessing me hods: elec omyog-
aphy (EMG), elec oneu og aphy (ENG), elec ogas og aphy (EGG), elec ooculog aphy (EOG),
elec o e inog aphy (ERG), and elec ohys e og aphy (EHG).
Keywo ds:
biomedical signals; signal p ocessing; elec omyog aphy; elec oneu og aphy;
elec ogas og aphy; elec ooculog aphy; elec o e inog aphy; elec ohys e og aphy
1. In oduc ion
This wo k is he las and hi d pa o he e iew pape ega ding he mos ecen
and he mos ad anced p ocessing me hods o bioelec ical signals. Pa I conce ned
hea signals [
1
]; Pa II— b ain signals. This pa (Pa III) is abou analysis me hods o
o he bioelec ical signals such as elec omyog aphy (EMG), elec oneu og aphy (ENG),
elec ogas og aphy (EGG), elec ooculog aphy (EOG), elec o e inog aphy (ERG), and elec-
ohys e og aphy (EHG).
This s udy mainly p esen s he so-called in o ma ic- ela ed ad anced signal p ocessing
me hods, which a e used o digi al, pos -ADC signals. Pu pose o hese me hods is o
ex ac c i ical in o ma ion ega ding he heal h condi ion o he pa ien s, which is con ained
in he signals.
Analysis o bioelec ical signals is a e y challenging ask, as hey a e all p one o
a ious dis u bances and a i ac s occu ence. The use o sophis ica ed signal p ocessing
me hods may imp o e hese signals’ quali y and may make hem mo e sui able o a ious
diagnos ics’ pu poses.
The me hods desc ibed in his wo k a e based on au ho s’ subjec i e decisions. I
was impossible o men ion all he me hods, mos ly due o he apid de elopmen o his
scien i ic a ea.
2. Elec omyog aphy
Elec omyog aphy (EMG) is a diagnos ic me hod, which enables eco ding o bioelec-
ic signals esul ing om he ac i i ies o he skele al muscles [
2
,
3
]. I is o en pe o med
while s imula ing he ele an mo o and pe iphe al ne es. The measu emen may be
ca ied ou ei he in an in asi e o su ace-based way, a he le el o a single muscle ibe ,
Senso s 2021,21, 6064. h ps://doi.o g/10.3390/s21186064 h ps://www.mdpi.com/jou nal/senso s
Senso s 2021,21, 6064 2 o 32
single mo o uni , o he en i e muscle. The p ocessing o in o ma ion om he EMG
enables diagnos ics o muscle and neu omuscula diso de s, o o analyze o use he EMG
o he ehabili a ion o limb p os heses con ol pu poses [2,4,5].
Fo he examina ion pu poses, he monopola o bipola elec odes may be used and,
in some cases, he combina ion o in amuscula and su ace elec odes can also be in ol ed.
Fo eco ding om he su ace, he so-called mul i-elec odes a e used, when he elec odes
a e placed in slo s on he silicone ma , ei he in a ow (s ips) o ma ix (g ids). In o de
o educe he signal in e e ence, he powe supply ol age wi h he igh oo (same as
wi h he ECG) can be applied, when he g ounding elec ode is placed su icien ly a om
he scene o he eco ding (see Figu e 1). The EMG equency anges a y om 0.01 Hz o
10 kHz, depending on he ype o examina ion (in asi e o nonin asi e). The mos use ul
and impo an equency anges a e wi hin he ange om 50 o 150 Hz [6–10].
Ca hode
Anode
S imula o
Sensing elec odes
EMG
Figu e 1. Example o he su ace EMG signal measu emen .
I is also impo an o men ion mode n solu ions in heal hca e, which in ol e mea-
su emen o among he o he s EMG signals—wea able and wi eless body-a ea ne wo ks
(BAN o WBAN), which in eg a e mul iple senso s o mo ion, ine ial, and biosignals
wi h low-powe adios. They usually wo k in eal- ime. In o de o make such sys ems
e icien comp essed sensing, o comp essi e sampling (CS), is applied. The CS is a me hod
o da a acquisi ion, whe e only ew incohe en measu emen s a e equi ed in o de o
comp ess spa se in some domain signals [
3
,
11
–
13
]. In [
11
–
13
], he au ho s showed in e es -
ing solu ions o comp essing bo h ECG (elec oca diog aphy) and EMG da a. In [
3
], he
au ho s ocused on CS o he pu pose o he EMG signals econs uc ion in o de o design
e icien low-powe s WBANs. Thei wo k compa es ou di e en algo i hms applied in
p ac ical implemen a ions.
2.1. EMG Reco dings
The EMG eco ding me hods can be di ided in o he wo main ca ego ies—mo o
uni ac ion po en ial (MUAP) and compound muscle ac ion po en ial (CMAP) [14].
2.1.1. MUAP
Muscles consis o mo o uni s, which a e he smalles possible po ions o muscles,
which can be ac i a ed [
15
]. The sum o he ac ion po en ials o he espec i e muscle
ibe can be measu ed in asi ely wi h needle elec odes placed di ec ly in o he muscle in
he a ea o in e es . The signal is hus ob ained wi h he supe posi ion o he indi idual
MUAP [14].
The e o e, i is necessa y o u he disassemble his signal on he con ibu ions
Senso s 2021,21, 6064 3 o 32
o each mo o uni . Typical MUAP a e wo-phase o h ee-phase, las ing app oxima ely
3–15 ms and eaching an ampli ude be ween 100 and 300 µV (see Figu e 2).
Time (ms)
01 2 3
(a) Example 1.
Time (ms)
01 2 3
(b) Example 2.
Time (ms)
01 2 3
(c) Example 3.
Time (ms)
01 2 3
(d) Example 4.
Figu e 2. Examples o he MUAP po en ials.
They epea 6–30 MUAP pe second. The shape o he MUAP depends on he ype o
he used needle elec ode, i s loca ion in ela ion wi h he mo o uni , and he esolu ion o
he elec ical ac i i y o he elec odes. Fu he mo e, he wa e shape is di e en in case o
some pa icula diseases such as, e.g., neu opa hy o myopa hy. The neu opa hy causes
slow conduc ion o unsynch onized ac i a ion o muscle ibe s wi hin a single mo o uni .
The MUAP has in his case a g ea e ampli ude. The myopa hy is mani es ed wi h loss o
muscle ibe s when neu ons a e non- unc ional. Then, he agmen a ion o he MUAP
occu s due o asynch onous ac i a ion, which leads o he mul i-phase MUAP [6,7,16].
2.1.2. CMAP
Ano he me hod o EMG measu emen is he eco ding o he en i e muscle ac i -
i y, which consis s o he sum o he indi idual MUAP. I is measu ed wi h he su ace
elec odes and i s analysis is a mo e complex, as i uses a equency ange om 10 Hz
o 1 kHz (see Figu e 3). The sampling equency o 2 kHz is usually applied. The signal
eaches up o 10 mV ampli ude. Due o he possible muscle disease, he ampli ude o he
signal is isibly educed. I he ampli ude g adually dec eases, i causes a p oblem wi h
he ansmission a he neu omuscula junc ion. In case o he demyelina ion o ne es,
con ac ion o he muscle ibe s delays wi h a no mal ampli ude o he esponses [17].
F equency (Hz)
Powe Spec um (mV)
050 100 150 250
0.8
0.6
0.4
0.2
200
Figu e 3. Example o he su ace EMG equency spec um (based on he wo k in [18]).
When measu ing he EMG, he so-called maximum ee con ac ion o he muscle is
e alua ed. The equi ed s eng h o con ac ion is achie ed wi h he g adual ac i a ion o
mo o uni s. The numbe o ac i a ed uni s equi ed o his con ac ion a ies depending
on he size o he muscles. Fi s , he e a e always ac i a ed mo o uni s inne a ing a
smalle numbe o muscle ibe s, inc easing he need o con ac ion, will hen engage
he d i e inne a ing a la ge numbe o h eads. This allows slowing he escala ion o
Senso s 2021,21, 6064 4 o 32
con ac ion. The muscles a e ne e able o achie e a cons an o ce con ac ion, due o he
ce ain luc ua ions in ac ion po en ial p opaga ion ime o he mo o uni and he di e si y
o eac ions o he muscle ibe s o ipping up. Oscilla ions we e de ec ed on he equency
o con ac ion o 1–2 Hz [6,7,16].
2.2. Clinical Applica ions
The EMG me hod is an impo an ool o assessmen o neu omuscula diso de s,
e en in a clinically no mal muscle. I is a low- isk p ocedu e associa ed wi h a e complica-
ions, such as bleeding, in ec ion, o ne e inju y [
19
–
21
]. In clinical p ac ice, i is possible
o in e p e bo h he appea ance o EMG signal wa e o m and he sound gene a ed by an
audio ampli ie . Du ing he EMG measu emen , a heal hy muscle issue a he es ing s a e
should be (elec ically) silen , wi h he excep ion o he a ea o he neu omuscula junc-
ion [
2
,
4
]. The esponse ela ed o incomple e elaxa ion (due o pa ien ’s inabili y o s ay
elaxed) di e s om pa hological spon aneous ac i i y by i s hy hmici y. The app oach
used in examining he EMG a ies depending on he diso de ype, pa ien ’s age, and
abili y o coope a e. Based on ha , he e a e se e al a iables such as numbe o elec odes,
hei loca ions and placemen , o hei ype. When in e p e ing he EMG esul s, a a ie y
o ac o s mus be aken in o accoun , including pa ien ’s heal h his o y, cu en heal h
condi ion, esul s o o he ele an diagnos ic me hods—such as ne e conduc ion s udies,
imaging me hods (e.g., magne ic esonance imaging o ul asound), muscle and ne e
biopsy, e c. [2].
The EMG me hod can be inco po a ed in a a ie y o applica ions. Fo example, as a
diagnos ic me hod o a esea ch ool o ields ela ed wi h kinesiology, mo o con ol-
ela ed, o neu omuscula diso de s. This me hod is also use ul in p os he ic de ice con ol,
especially p os he ic a ms o lowe limbs [
19
]. Ne e heless, he EMG can only be applied
in o de o assess p ima y myopa hic condi ions ela ed o elec ical ac i i y o he muscles.
The e o e, o he app op ia e e alua ion o mos neu omuscula diso de s, i is necessa y
o also measu e he elec ical ac i i y o he ne es ha gene a e elec ical signals o con ol
he muscle esponse. The e o e, he EMG is usually pe o med simul aneously wi h he
so-called ne e conduc ion s udies (NCSs), which enable measu emen o he ne es
conduc i i y. The needle-based EMG and he NCSs a e es ed in case he pa ien su e s
om neu omuscula diso de symp oms, such as limb pain o weakness, muscle pa alysis,
o spasms [
21
–
23
]. Mo eo e , he examina ion is also bene icial in diagnosing he ne e
comp ession o inju y (e.g., ca pal unnel synd ome o scia ica), and o he neu omuscula
diso de s, such as amyo ophic la e al scle osis (ALS), myas henia g a is, and muscula
dys ophy [22].
The EMG is a low- isk p ocedu e, and complica ions du ing i s pe o mance a e e y
a e. The e is, howe e , a small isk o bleeding, in ec ion, and ne e inju y in a loca ion
whe e a needle elec ode is inse ed. When muscles along he ches wall a e examined wi h
he needle elec ode, he e is a e y small chance ha i could cause ai o leak in o he a ea
be ween he lungs and ches wall, causing a pneumo ho ax (lung o collapse) [19–21].
2.3. EMG Signal P ocessing
The EMG signal can be conside ed o be challenging om he pe spec i e o he la ge
ange o in e e ence emo al (as summa ized in Table 1) due o i s wide equency ange.
The signi ican in e e ences o he su ace EMG signal a e he mo ion a i ac s, which
a e e y di icul o emo e and occu in almos all biomedical signals. They a e caused
by he senso mo emen s in he a ea o in e es which inc ease due o he di ec p ess
on o he senso , as mo emen s o he body pa s whe e he senso is ixed o due o he
changes in he balance o he elec ode–skin in e ace caused by he muscle con ac ion
esul ing in olume changes. These a i ac s mean he issue occu ing du ing measu emen
o dynamic ac i i ies. The mo ion a i ac s ypically ha e he high-ampli ude peak beside
he EMG signal bu hey occu a he low equencies, so i is possible o emo e hem
Senso s 2021,21, 6064 5 o 32
using he high-pass il e which does no a ec he measu ed equency ange o he EMG
signal [24,25].
Ano he ype o a i ac p esen in he EMG da a is he signal con amina ion wi h
he signal o he nea by muscle, which is no in he a ea o in e es , hese a e he so-called
“c oss- alk” signals. The da a is hen dis o ed in bo h ampli ude and ime du a ion o
he EMG signal. This a i ac can be educed using e e ence elec odes which enable
compa ison o he ampli udes o he gene a ed myopo en ials and e alua ion o he signal
cou se in he a ea o in e es . In case o he senso disconnec ion o he signal ampli ude
o e doing, he signal sa u a ion can o igina e which esul s in he dis o ion o he high
ampli udes. The e o e, i is impo an o con ol he con ac o he elec ode and skin
su ace, educe he gain o he ampli ie o ans e he elec ode o ano he place on he
es ed muscle whe e he ampli ude can be educed.
As in he case o he EEG eco dings, he EMG signal is also signi ican ly in luenced
wi h he ECG signal, which is a i s mos isible du ing he measu emen o he muscles
o he uppe pa s o he body. The p oblems a e he high ampli ude o he ECG signal
agains he EMG signal and he o e lapping o hei equency spec a, so i is no possible
o use he o dina y il e ing me hods in o de o emo e hese a i ac s. One o he ways is
o loca e he EMG elec odes as a om he hea as possible [26].
F om he echnical a i ac s, he EMG signal can be co up ed wi h he PLI, elec omag-
ne ic in e e ence o he o he p esen sou ces, and componen s o he eco ding sys em.
Fo he high- equency noise supp ession, he low-pass il e s wi h he cu -o equency
highe han he EMG equency ange can be success ully applied [27–29].
2.4. EMG P ocessing Me hods
In case o he EMG signal noise educ ion, he classic equency-based me hods a e
no e y sui able because he signal is non-s a iona y. Howe e , hese me hods a e o en
used o hei easy implemen a ion. A p oblem begins in pa icula du ing he elimina ion
o noise signals which o e lap wi h he desi ed EMG signal in he equency band [
24
,
26
].
Table 1summa izes he mos common p ocessing me hods o he EMG da a.
Table 1. Summa y o he EMG signal p ocessing me hods.
Me hod O e all SNR Compu a ional Real-Time Implemen a ion
Pe o mance Imp o emen Cos Complexi y
DF Low Low Low Yes Simple
ANC High Medium Medium Yes Medium
WT Low Medium Medium Yes Medium
ICA High High Medium No Medium
EMD High Medium High No Medium
Hyb id Me hods High High Medium No Complex
I is also impo an o men ion, beside he me hods desc ibed below, some in e es ing
s udies, whe e he EMG da a has been analyzed. In Kawala-Janik e al. [
5
], a cus omized
h eshold-based me hod, p e iously es ed on he EEG da a (see in [
30
,
31
]) has been applied
o he pa e n ecogni ion pu poses. In [
14
], implemen a ion o ac ional il e ing as an
al e na i e o he adi ional, in ege -o de il e s in analysis o he EMG da a was in de ail
p esen ed. The ob ained esul s we e in e es ing and p omising.
2.4.1. Digi al Fil e ing
The mos common way o noise emo al in he EMG signals is il e ing. Applica ion
o a ious digi al il e s is qui e a simple and as me hod o he pu pose o p ep ocessing
s ep o he u he analysis (e.g., muscula a igue de ec ion, limb mo emen s, o emo ions
ecogni ion). Usually, he BPF wi h he cu -o equencies o 20 Hz and 400–600 Hz is used
o he dema ca ion o he physiological equency band o he EMG signal and emo ing
Senso s 2021,21, 6064 6 o 32
he bo h high- and low- equency in e e ence. Then, he BSF wi h he cu -o equencies
49–51 Hz is included in he p ep ocessing s ing in o de o emo e he PLI [8,10,32,33].
2.4.2. Adap i e Noise Cancele
The Adap i e Noise Cancele (ANC)-based il e ing can be success ully used o he
pu pose o he PLI o he ECG a i ac s elimina ion. Soedi djo e al. [
34
] used he ANC wi h
he LMS adap i e algo i hm and syn he ic e e ence. The ANC migh no be su icien as
he undamen al equency o he PLI de ia es up o
±
1% om i s nominal equency, so a
syn he ic e e ence wi h a ying equency and phase is p oposed. In compa ison wi h he
o he me hods es ed by hem (such as no ch il e , Kesh ka an and Yang’s adap i e il e ing,
and spec al in e pola ion), he applied ANC esul ed wi h he bes SNR imp o emen .
In case o he ECG a i ac emo al [
35
,
36
], he easies way was o eco d he ECG signal
and o use i as he e e ence inpu o he ANC.
2.4.3. Wa ele T ans o m
The WT il e ing me hod is equen ly used in he EMG signal p ocessing because o
i s non-s a iona y cha ac e and he abili y o he me hod o dis inguish da a well in bo h
equency and ime domain. Hussain e al. [
37
] used many wa ele unc ions in o de o es
he WT o he pu pose o noise educ ion in he su ace EMG signals (Daubechies, symle ,
Meye ). The wa ele Db2 seems o be he mos powe ul ool o he signal denoising using
he WT. Jiang e al. [
38
] compa ed he di e en ypes o h esholding—Uni e sal, SURE,
Hyb id, and Minimax. Thei expe imen s p o ed ha he denoised EMG was insensi i e
o he selec ion o he denoising me hods.
2.4.4. Independen Componen Analysis
Howa d e al. [
39
] p oposed a ool o educ ion o he c oss- alk signals om he
EMG eco dings ia he ICA me hod. The EMG signal is sui able o using he ICA
because i sa is ies all i s c i e ia (componen s mus be s a is ically independen and he
independen componen s mus be non-Gaussian), whe e each muscle can be assumed o
be an independen sou ce, as he mo o uni s in each muscle a e well sepa a ed om he
o he muscles, and he ini e sum o he MUAPs is non-Gaussian. Naik e al. [
40
] used a
mul i- un ICA me hod in o de o elimina e he c oss- alk signals ob ained om he su ace
EMG and analyzed he numbe o sou ces o he signal. The mul i- un ICA is he p ocess
whe e he ICA algo i hm is compu ed many imes—a each ins ance, so as a esul — he
di e en mixing ma ices a e ob ained.
2.4.5. Empi ical Mode Decomposi ion
The EMD me hod was used o he pu pose o decomposi ion o he EMG signal o
he IMFs which we e hen p ocessed wi h he implemen a ion o o he il e ing me hods
and hen, he denoised signal was econs uc ed [
41
]. And ade e al. [
42
] implemen ed he
EMD in combina ion wi h he so h esholding. The EMD was compa ed also wi h he WT
using many ypes o ma e nal wa ele s. Bo h o he p oposed me hods educed he o al
powe o he noise and a he same ime p ese ed he majo i y o he ene gy o he signal.
Mish a e al. [
43
] p oposed a simple echnique wi h he applica ion o he imp o ed
EMD (IEMD) in conjunc ion wi h ou di e en ea u es, which was used o he analysis
o amyo ophic la e al scle osis (ALS) and he no mal EMG signals. The EMD me hod
ollowed wi h he median il e has been employed o emo al o he impulsi e noise
om he IMF componen s gene a ed h ough he EMD. The il e ed IMF componen s a e
summed oge he in o de o gene a e a new signal. The EMD p ocess is u he applied o
he new EMG signal o gene a e imp o ed IMFs called he IEMD me hod. A new echnique
based on he IEMD algo i hm was p oposed o he i s ime, which enabled he choice o
he window size o he applied median il e .
The EMD and he imp o ed EMD, called he Ensemble EMD (EEMD), enable o e -
coming he limi a ion o he mode mixing ou inely induced wi h he egula EMD, and is
Senso s 2021,21, 6064 7 o 32
also used by Zhang e al. [
44
]. I he PLI componen was p esen in he IMF, he no ch
il e was applied o he IMF. In o de o educe he whi e noise, a simila app oach o he
wa ele -based denoising me hods can be implemen ed, including so o ha d h esholding.
The BW componen s in ol ed mainly he highe -o de IMFs, which can be assessed by
applying he LPF o he IMFs. Then, he signal can be econs uc ed om he IMFs.
2.4.6. Hyb id Me hods
Abbaspou e al. [
45
] p oposed a combina ion o he WT and he ICA me hods o
he pu pose o he ECG a i ac s elimina ion. The i s s ep o he me hod was a wa ele
decomposi ion (wi h he Db4 wa ele ) in o de o c ea e 8 le els o he aw signal. A e he
wa ele coe icien s we e calcula ed, he ICA me hod was applied o he mul idimensional
da a p oduced wi h he WT. Then, he independen componen s we e classi ied au oma -
ically as ei he EMG signal o ECG a i ac . The implemen a ion o such il e ing o he
signals p o ided sa is ac o y esul s in he SNR imp o emen . The ma ix o signals was
ob ained om i e senso s— h ee o hem we e a ached in ideal loca ions and wo o he s
we e placed non-ideally in o de o ga he he c oss- alk da a. Th ough a combina ion o he
channels, he h ee inal signals we e p ocessed wi h he applied ICA me hod. The ob ained
esul s we e iden i ied as a success ul dis inc ion be ween indi idual muscle ac i a ion.
Ren e al. [
46
] emo ed he PLI using he abo e men ioned me hod in he con a y
sequence. Fi s , he ICA decomposi ion was done, when he independen componen s
con aining he PLI we e il e ed wi h he WT and he MUAP componen s we e e ained,
which was he aim o he s udy. The whi e noise was hen emo ed also using he WT.
The ha d h esholding was used o he WT and he signal was econs uc ed wi h he
in e se disc e e WT.
3. Elec oneu og aphy
The elec oneu og aphy (ENG) is a me hod used o isualize di ec ly eco ded elec-
ical ac i i y o neu ons in he cen al ne ous sys em (CNS), which consis s o he b ain
and he spinal co d o in he pe iphe al ne ous sys em (PN) consis ing o he ne es and
nodes. The elec oneu og aphy is simila o he elec omyog aphy (EMG), bu i is used in
o de o isualize he muscles ac i i y as i is used o measu e he conduc ion eloci ies
and la encies in pe iphe al ne es by s imula ing a ne e a di e en poin s along i [47].
The i s ENG om a single ne e ibe was eco ded by Edga Ad ian in 1928 using
Lippmann’s elec ici y me e [
48
]. In 1953, he i s I idium eco ding mic o-elec ode was
de eloped [
49
]. The i s simul aneous eco ding wi h he implemen a ion o he mul iple
uni s wi h a use o he mul i-elec ode se , which was pe o med on a pa ien du ing b ain
su ge y, was published by Ma g and Adams in 1967 [50].
The ENG is usually ob ained h ough eco ding wi h he elec odes placed in he
ne ous issue. The elec ical ac i i ies gene a ed wi h he neu ons a e eco ded wi h
he elec odes and a e hen ansmi ed o a collec ion sys em, which usually allows
isualiza ion o he ac i i y o he neu on. Each e ical line in he elec oneu og am
ep esen s only one neu on ac ion po en ial. Depending on he accu acy o he elec ode
applied o he eco ding o he ne e ac i i y, he ob ained elec oneu og am signal
may con ain he ac i i y o one o housands o neu ons. The esea che s adap ed he
accu acy o a ious elec odes ei he by ocusing on he ac i i y o he single neu on o
on he gene al ac i i y o a g oup o neu ons, and bo h s a egies ha e hei ad an ages
and disad an ages.
3.1. ENG Reco ding
Indi idual ne e ibe s conduc exci a ion a di e en paces. I he ac ion po en ial o
he whole ne e is ead as he o al ac ion po en ial, hen he ob ained elec oneu og am is
a cu e wi h se e al cha ac e is ic peaks. Each wa e co esponds wi h he one ype o he
ne e ibe , such as, e.g., ype A ibe s, which a e ibe s media ing he mo emen s o he
skele al muscles and hey a e led a he speed o app oxima ely 70–120 m/s. In con as ,
Senso s 2021,21, 6064 8 o 32
he ype C ibe s a e ibe s ha media e he eeling o hea o pain and a e led a he
speed o only 0.5–1 m/s [
51
]. In Figu e 4, i is possible o see an ENG mixed ne e,
whe e he ca ego y o he A ibe s—myelina ed, 4 subg oups, and he ca ego y he B
ibe s—myelina ed p eganglio ege a i e and he ca ego y o he C ibe s—unmyelina ed
pos ganglionic sympa he ic ibe s (Cs), cen ipe al pain ibe s Cd. —do sal oo s.
Time (ms)
Ampli ude (mV)
100
10
1
3
2
1
A
C
B
α
β
γ
δ
0
Figu e 4. The ENG mixed ne e.
Examina ion o hei conduc i e unc ion (exci a ion conduc ion eloci y oge he
wi h a enua ion cha ac e is ics). In Figu e 5, he ENG measu emen s we e p esen ed.
Re
sponse
S imula ion
?
Figu e 5. Sample ENG measu emen .
The esul ing signal and he con ibu ion o indi idual neu ons (i.e., he ampli ude
and mo phology o he ENG signal) a e a ec ed by he ype and loca ion o he neu al
ibe (s) and hei p oximi y o he measu ing elec ode. The shape o he signal and he
associa ed con en is also in luenced by he con igu a ion o he de ice (e.g., bipola o
unipola measu emen , common e e ence o g ound, e c.), as well as size and shape o
he elec ode’s ac i e con ac wi h he skin and i s placemen . The e o e, he eco ded
signal can consis o single peaks o i can be a composi e o se e al ac ion po en ials. I s
ampli ude can hus ange om 1 o 100 mV and i s equency om a ew Hz o 10 kHz [
47
].
3.2. Clinical Applica ions
The ENG usually e e s o he eco dings ob ained om he axon bundles in he
pe iphe al ne es. In clinical p ac ice, he ENG wa e o m can be displayed on a sc een,
con e ed o a sound and played h ough an audio ampli ie , o used as a con ol signal o
a ious neu al p os heses o o he de ices. The ENG is eco ded by means o elec ode(s)
placed in nea p oximi y wi h he neu ons o in e es in o de o co e hei ac i i y [52].
The ENG can be used o a ious pu poses. Fo example, i is a aluable diagnos ic
ool o he assessmen o a ious mo emen - ela ed diso de s [
53
] (e.g., ALS [
54
], mul iple
Senso s 2021,21, 6064 9 o 32
scle osis (MS) [
55
], in bio eedback (e.g., in he closed loop sys ems applied o he end
o gan s imula ion), o he assessmen o muscles condi ions, and o he pu poses o
con ol o he o ho ics and p os heses [56–58].
Bo h ENG and EMG signals consis o a ious signals om se e al sou ces, including
he desi ed ones om indi idual neu ons and o he s, conside ed as noise, om su ound-
ing issue, o gans, de ices, o en i onmen . The ENG signal can hus be conside ed as
challenging in e ms o signal p ocessing and noise emo al mos ly due o i s wide e-
quency ange and he in e e ence accompanying i s acquisi ion. The main in e e ence o
he ENG signal sensed on he skin su ace a e he mo ion a i ac s, which a e di icul o
emo e. They a e caused by he mo emen s o he senso , which a e inc eased by p essing
o o he wise manipula ing wi h he senso , apid mo emen s o he pa s o he body
whe e he senso is moun ed, o changes in he balance o he elec ode–skin in e ace
caused by he muscle con ac ion leading o he changes in olume. I is possible o emo e
hem using in e alia high-pass il e s, which do no a ec he measu ed equency ange
o bo h EMG and ENG signals [59–61].
In gene al, he aw ENG signal ob ained mus be i s p ep ocessed be o e any u he
p ocessing, ex ac ion o analysis akes place. The p ep ocessing phase usually includes
he signal ampli ica ion and basic band pass il e ing de ined acco ding o he signal’s
cha ac e is ics. This s ep is necessa y o emo e he unwan ed signals, such as he EMG
noise, ne e issue backg ound noise, mo ion a i ac s, o conduc ion noise. The ENG
de ice ypically includes an inpu ampli ie which enables he a enua ion o he signal
in he speci ic equency bands, mos equen ly de ined as high pass (0.01–1000 Hz),
a low pass (500–10,000 Hz), and a no ch (50 o 60 Hz. Mo eo e , he eco ding sys em
is cons uc ed in a way o p o ide a low noise (
<
2 mVpp), high no mal mode ejec ion
a io (CMRR
>90 dB),
high inpu impedance and high gain (1000–500,000) bandwid h
di e en ial eco ding capabili y (0.01–10 kHz). The p ep ocessed ENG signals a e hen
digi ized and s o ed o ansmi ed o he compu ing machine o u he analysis and/o
simply displayed on he sc een [52,62].
3.3. ENG P ocessing Me hods
The aw ENG signal con ains e y aluable in o ma ion, al hough he ex ac ion
o his in o ma ion om he eco ded signal usually equi es implemen a ion o a i-
ous online o o line signal p ocessing o machine lea ning p ocedu es [
63
]. Such signal
p ocessing me hods usually include analysis in he ime and equency domains, o he
mixed mode me hods. Fo example, he equency domain me hods include he Laplace,
Fou ie , and Z- ans o ms, o calcula ions o powe spec al densi y and signal phase.
The ime domain app oaches may include he ime-se ies analysis (e.g., mo ing a e ages o
c oss-co ela ions). The mixed mode me hods a e me hods such as he band-pass il e ing,
wa e o m analysis, sho Fou ie ans o m, e c. The machine lea ning p ocedu es, which
ha e become e y popula , a e algo i hms used o he de ec ion o he s a is ical egula i-
ies in he speci ied da a. They include lea ning in o ma ion ob ained om he p e ious
da a o gene a ing (p edic ing) he new da a. The mos common me hods used in he
ENG signals p ocessing include he p incipal componen analysis (PCA), he independen
componen analysis (ICA), he auxilia y ec o machines (SVM), and he a i icial neu al
ne wo ks (ANN) [62,64,65].
Robe e al. [
66
] de eloped a new scheme o inc easing he comp ession and in e -
p e abili y o he mul ichannel bio-po en ials ob ained om he implan ed ne e senso s.
Spa ial and empo al co ela ions be ween he samples we e de e mined o ind he ap-
p op ia e pa e ns associa ed wi h he ex e nal s imula ion a ec ing he scanned issues.
The applied ime analysis included peak de ec ion and so ing. The spa ial p ocessing was
based on he ICA analysis o he obse ed ne e ac i i y. Expe imen s using modeling and
simula ions we e pe o med in o de o es he abili y o he sys em o assign he obse ed
po en ials o he applied ex e nal s imuli [65].
Senso s 2021,21, 6064 16 o 32
5.1. EOG Reco ding
The EOG signal belongs o he g oup o andom (non-s a iona y) signals, as hei
spec um a y o e ime [
93
]. I s equency anges om 0.5 o 15 Hz and is cha ac e ized
wi h a signi ican DC componen . Values o he ampli ude do no exceed mV ange,
bu usually a e lowe han 2 mV (a ound 50–3500
µ
V). The change o he ol age is caused
by he 1-deg ee change in iew angle [
7
]. Samples o he EOG signals a e illus a ed wi h
Figu e 10, while Figu e 11 p esen ed he EOG spec um.
Time (s)
00.2 0.4 0.6
(a) Sample 1.
Time (s)
00.2 0.4 0.6
(b) Sample 2.
Time (s)
00.2 0.4 0.6
(c) Sample 3.
Time (s)
00.2 0.4 0.6
(d) Sample 4.
Time (s)
00.2 0.4 0.6
(e) Sample 5.
Time (s)
00.2 0.4 0.6
( ) Sample 6.
Figu e 10. Sample EOG sisgnals.
F equency (Hz)
Rele ance (%)
0 10 20 30 60
100
80
60
40
20
40 50
Ve ical EOG
Figu e 11. Sample EOG spec um.
5.2. Clinical Applica ions
In some cases, he EOG is a use ul ool applied o he diagnos ics pu poses o inhe i ed
macula diseases [
94
]. Unde s anding he disease s a es ha he a ec ed wi h he disease
EOG helps wi h he in e p e a ion o he esul s. In conjunc ion wi h he ERG, i may be
use ul in he p ocess o diagnosis o a ious p og essi e e inal diso de s.
Typically, he EOG is a me hod used o he pu pose o eye mo emen s eco ding
du ing he ENG, excep o he classical EOG, which uses di ec cu en ampli ica ion (di ec
cu en ), while he ENG in clinical p ac ice o en uses al e na ing cu en ampli ica ion
Senso s 2021,21, 6064 17 o 32
(al e na ing cu en capaci o ) wi h a ime cons an o 5 o 10 seconds, which esul s in a
high- equency il e ed signal whe e slow undamen al shi s a e damped.
One o he signals ha can be de ec ed using he EOG me hod is he eyeblink. In some
cases, his signal is conside ed as he desi ed one, o en used as means o communica-
ion wi h pa alyzed indi iduals [
95
,
96
], whe eas in o he s as an a i ac and hus should
be emo ed.
Finally, among he mos ecen ly discussed applica ions o EOG is a so-called ’eye-
w i ing’, whe e he EOG-based me hod can be used as an al e na i e o con en ional
came a-based eye acke s. The easibili y o he EOG-based eye-w i ing as a new commu-
nica ion app oach o indi iduals wi h amyo ophic la e al scle osis (ALS) was in es iga ed
in [
97
,
98
]. In [
97
], he au ho s de eloped an EOG-based eye-w i ing sys em and es ed i on
21 pa icipan s (18 heal hy and h ee wi h ALS). The sys em achie ed a mean ecogni ion
a e o 95.93% o heal hy pa icipan s and 85% o subjec s wi h ALS. In [
98
], he au ho s
e alua ed he eye-w i ing sys em wi h a symbol se consis ing o symbols o 10 A abic
nume als and ou ma hema ical ope a o s. Expe imen s on 11 olun a y human subjec s
showed ecogni ion a e anging om 50% o 100% wi h di e en symbols.
In [
99
], he au ho s p oposed a con inuous eye-w i ing ecogni ion sys em ha ecei es
eye-w i en cha ac e s con inuously and hus achie es a high inpu a e. This sys em de ec s
eye mo emen s by EOG and hen applies a hidden Ma ko model (HMM) o model he
EOG signals and ecognize he eye-w i en cha ac e s. Expe imen s wi h six pa icipan s
showed an a e age inpu speed o 27.9 cha ac e /min.
5.3. EOG P ocessing Me hods
In he EOG signal, he mos common a i ac s a e (excep he PLI o o he noise
coming om he elec omagne ic ields) baseline d i , neck mo emen ( esul ing in he low-
equency changes in he signal), muscles po en ials a ound he eyes (ha ing he maximum
powe in he equencies abo e 20 Hz), eye blinking ( andom po en ial di e ences in
equency ange o 0.5 o 3 Hz), and saccades. Mo eo e , he simul aneous ac i i y may
cause he elec odes o lose con ac o mo e on he skin su ace, which dec eases he da a
quali y in a signi ican way [100,101].
The EOG signals can be e icien ly applied in a ious Human–Machine and/o
Human–Compu e In e aces, as hey a e usually based on use s’ eye mo emen in en ions
o ac ual mo emen s, hus such sys ems use limi ed numbe o commands based on basic
eye mo emen s (looking up, down, le , o igh ) [102–105].
He e al. [
105
] p oposed a no el single-channel EOG-based solu ion, which allows
use s o spell wi h only blinking. In his sys em, o y bu ons co espond wi h he 40 cha -
ac e s displayed o he use in a andom o de , whe e he use mus blink when he a ge
bu on lashes. The au ho s used wo di e en signal p ocessing me hods— he suppo
ec o machine (SVM) classi ica ion and he wa e o m de ec ion, which we e combined o
he pu pose o he eye blinks de ec ion. The ob ained expe imen al esul s demons a ed
he e ec i eness o he p oposed solu ion wi h an a e age accu acy o 94.4% and a esponse
ime o 4.14 s.
Typical signal p ocessing algo i hms equen ly su e om la ge execu ion delays,
which is challenging o makes i e en impossible o he eal- ime analysis— he e o e,
implemen a ion o high-speed algo i hms is needed. Aga wal e al. [
92
] implemen ed a
mul iplie Sa i zky–Golay smoo hing il e (SGSF) based on dis ibu ed a i hme ic (DA)
o he pu pose o he EOG signals p ep ocessing, so ha he p ocessing speed has been
inc eased along wi h he necessi y o he chip a ea educ ion. The il e applied o
his pu pose should be e icien enough in o de o emo e he a i ac s along wi h leas
de o ma ion o a ec ing o he ac ual signal in a nega i e way. The Sa i zky–Golay (SG)
il e is widely applied in analysis o biomedical signals [
88
]; howe e , despi e i being as
and e icien and easily implemen able, i has no been applied o he pu pose o he EOG
analysis so a . The SGSF is some imes chosen o he pu pose o accu a e diagnosis using
saccade de ec ion o he EOG signals. The e iciency o he p oposed il e was es ed in
Senso s 2021,21, 6064 18 o 32
e ms o he signal- o-signal-plus-noise a io (SSNR) and he eal- ime compu a ions. I
was obse ed based on he pe o med analysis ha he DA based a chi ec u e inc eased
he p ocessing speed, educed he chip a ea and he o iginal ea u es o he il e ed signal
we e p ese ed. In Table 4, a summa y o EOG signal p ocessing me hod was p esen ed.
Table 4. Summa y o he EOG signal p ocessing me hods.
Me hod O e all SNR Compu a ional Real-Time Implemen a ion
Pe o mance Imp o emen Cos Complexi y
SVM Medium Medium Low Yes Medium
SGFM High High Medium Yes Medium
DA Low Medium Low Yes Low
As o he de ec ion o ex ac ion o he eye blink, a ious me hods ha e been used
in he li e a u e, such as WT [
106
,
107
], e alua ion wi h h esholds [
108
,
109
], de i a ion o
he signals [
110
,
111
], o analysis o EOG eloci y based on expe ules [
112
]. Mo eo e ,
addi ional ea u es o he blinks, such as s a ime, speed, o du a ion o he eye closu e,
can be ob ained using ad anced signal p ocessing and classi ica ion
me hods [113–115].
In con as , he e a e a eas whe e he eye blink mus be elimina ed, mos ly om he EEG
signal [
116
–
119
]. Fo example, in [
120
], he au ho s in oduced a no el me hod o EEG
il e ing based on signal modeling, ime a ian co a iance ma ices, and Kalman il e .
The eye blink model was c ea ed using a single channel EOG.
6. Elec o e inog aphy
Go ch in 1903 [
121
] was he i s o s a e ha he eye’s esponse o a lash o ligh
consis ed o he wo wa es; he i s one was nega i e, while he second one—posi i e
(wi h a g ea e ampli ude). La e , Ein ho en e al. in 1908 [
122
] di ided he ERG esponse
in o he h ee wa es: The i s wa e, which appea ed immedia ely a e u ning on he ligh
s imulus, was nega i e on he co nea and was ollowed wi h a posi i e wa e. The las wa e
was slowe , bu also posi i e. Ein ho en e al. sugges ed ha he ligh s imulus igge ed a
chain o eac ions leading o he o ma ion o he p oduc s A, B, and C, and ha each elec ic
wa e indica ed a change in he “ ele an ” p oduc . The wo k o hese au ho s was he
basis o s a ing he esea ch on he ERG signals’ p ocessing, analysis and implemen a ion,
and hese a e used o his day. The wa es a e called a-, b- and c-wa es. Ano he posi i e
co nea wa e, which is less equen ly eco ded a he end o a lash o ligh , is called he
d-wa e. In Figu e 12 he biphasic wa e o m o a ypical heal hy pa ien was p esen ed.
Time (ms)
Ampli ude (μV)
0
25 50 100
100
50
0
−50
75
b-wa e
a-wa e
−100
Figu e 12.
The biphasic wa e o m o a heal hy pa ien — he nega i e wa e (
a
) and he posi i e
wa e (b).
Senso s 2021,21, 6064 19 o 32
The elec o e inog aphy (ERG) is a diagnos ic es , which enables measu emen o
he elec ical ac i i y o he e ina in esponse o a ligh s imulus. I is based on cu en s
gene a ed di ec ly by e inal neu ons in combina ion wi h he con ibu ions om he e inal
glia. The ERG is an objec i e measu e o he e inal unc ion, which can be eco ded
nonin asi ely unde physiological condi ions. The ERGs a e o en eco ded using a hin
ibe elec ode which is placed in con ac wi h he co nea o an elec ode which is buil in o
he con ac lens o he co nea. These elec odes enable eco ding o he elec ical ac i i y
gene a ed by he e ina on he su ace o he co nea. The ERG can be induced wi h di use
lashes o pa e ned s imuli [123].
6.1. ERG Reco ding
Fo he ERG eco ding pu poses a ious loca ions o he eco ding elec odes place-
men : in con ac wi h co nea, bulba conjunc i a, o skin below lowe eyelid. I is possible
o dis inguish he ollowing ypes o he eco ding elec odes [123,124]:
•
Bu ian–Allen (BA): made o a s ainless-s eel annula ing su ounding he co e o he
polyme hyl me hac yla e (PMMA) con ac lens. The BA elec odes con ain a lid mi o
in o de o help o minimize blinking. They a e eusable.
•
Dawson–T ick–Li zkow (DTL): low-weigh elec odes made o conduc i e sil e o
nylon ibe . They a e disposable and a e usually mo e con enien o pa ien s.
• Je : disposable plas ic lens wi h gilded pe iphe al ci cum e ence.
•
Skin elec ode: can be used as a eplacemen o co neal elec odes by placing he
elec ode on he skin o e an in a ed comb nea he lowe lid. The ERG ampli udes
end o be small and p one o noise occu ence, bu hey a e be e ole a ed in
pedia ic popula ions.
• Myla elec ode: Alla ized o gold-pla ed Myla , no commonly used.
•
Co on-wick: A Bu ian–Allen elec ode shea h equipped wi h a co on wick, which is
use ul o minimizing ligh -induced a i ac s, no commonly used.
•
Hawlin–End Elec ode: Te lon insula ed hin me al wi e (sil e , gold o pla inum)
wi h h ee cen al windows, leng h 3 mm, shaped o i in he lowe conjunc i al sac,
a e no commonly used.
6.2. Clinical Applica ions
The ERG signals a e mainly used by oph halmologis s and op ome is s, and a e used
o he diagnos ics pu poses o a ious e ina diseases. The ERG has impo an clinical
applicabili y in p o iding diagnos ic in o ma ion o a numbe o inhe i ed and acqui ed
e inal diso de s. In addi ion, he ERG can be used o moni o ing disease p og ession and
e alua ion o he e inal oxici y due o usage o a ious d ugs o p esence o
o eign bodies.
O he ERG assays, such as he pho opic nega i e esponse (PhNR) and he ERG
model (PERG), may be use ul in assessing e inal ganglion cell unc ion in diseases such
as glaucoma. The mul i ocal ERG is used o eco ding o sepa a e esponses o di e en
e inal si es.
6.3. ERG Signal P ocessing
The ERG signals ha e wo impo an ampli udes used o a ious disease diagnos ics
by medical p o essionals. These a e he nega i e a wa e and he posi i e b wa e. The im-
plici wa e imes a and b a e also use ul o diagnos ics pu poses. The ERG signals ha e
small ampli udes (abou
µ
V). Fo his eason, i is impo an o sepa a e he signal om he
noise and om he in e e ence, which may occu due o mo emen .
An a i ac may appea in he ERG, which may in e e e wi h he eco ding and
in e p e a ion o he ERG b wa e. This a i ac is a pho omyoclonic e lex (PMR) and was
s udied by Johnson e al. [
125
], who du ing hei expe imen s co e ed he eye con aining
he eco ding elec ode and s imula ed he o he eye a he same ime. The eco dings
ob ained wi h he implemen a ion o his echnique be o e and a e adminis a ion o he
modi ied Van Lin eyelid block showed ha mos PMRs a e caused by e lex con ac ion o
Senso s 2021,21, 6064 20 o 32
he o bicula is muscle. The emainde o he PMR was de ec ed h ough eco ding he eye
mo emen 1.5 o 3.5 deg ees down and wi h he obse a ion o he medial eye mo emen .
The line noise e e s o he elec ical in e e ence induced in he cable connec ing
he elec ode wi h he ampli ie . Because he inpu impedance o he ampli ie is high,
noise can be gene a ed in he cable wi h he capaci i e o magne ic couplings coming
om he su oundings. These e ec s can be limi ed o some ex en wi h he usage o
shielded o wis ed cables. The main ype o he noise in e e ence comes om 50 o
60 Hz
gene a ed wi h he powe lines and elec ical ou le s. This lies wi hin he bandwid h
anges
(1–300 Hz)
o he elec omagne ic signal ( he ERG), and he e o e se e ely impai s
he quali y o he eco ded da a. Many elec odiagnos ic eco ding sys ems use ac i e
elec odes in o de o o e come he line noise. This in ol es connec ing he i s ampli ie
as close as possible o he eco ding elec ode. The elec ical noise can be educed wi h he
impedance ans o ma ion because he low ou pu impedance o he ampli ie is almos
impe meable o he elec ical o magne ic in e e ence [126].
La i oglu e al. [
127
] p oposed o use empi ical mode decomposi ion in o de o
denoise he ERG esponses. The ERG signals a e he non-s a iona y signals ha a e
decomposed in o a numbe o in insic mode unc ions, so hen he noise and in e e ence
can be elimina ed. Finally, he ERG signals ha ing hei signal- o-noise a io o less han
o equal o 10 dB a e econs uc ed, which enables hem o ob ain he denoised ERG
signals [128].
San iago e al. [
129
] used a me hod o p ocessing he mul i ocal elec o e inog am
(m ERG) eco dings in o de o imp o e he abili y o diagnosing he MS. They examined
he m ERG eco dings ob ained om 15 pa ien s wi h ea ly-s age MS wi hou a his o y o
op ic neu i is and om 6 heal hy pa icipan s (con ol subjec s). The m ERG eco dings
we e il e ed using he EMD. Co ela ion wi h he signals in he no ma i e da abase was
used as a classi ica ion unc ion.
The Disc e e Wa ele T ans o m (DWT) is a as and e icien analysis me hod o
he ERG, which e eals ime and equency in o ma ion ega ding he examined signal.
The choice o pa en ipple is impo an o he bes ex ac ion o he desi ed componen s.
In [
130
], op imiza ion o he selec ion o he Daubechie Wa ele o he ERG collec ed
wi h he Pho opic Nega i e Response (PhNR) s imulus h ough a iable e alua ion and
selec ion o unc ions in he classi ica ion o glaucoma ous and non-glaucoma ous eyes is
p esen ed.
Visual unc ion es ing using he ERG signals is used in o de o de ec e inal ab-
no mali ies. This is achie ed by measu ing, cha ac e izing and analyzing biopo en ial
esponses om a ious e inal cells gene a ed by isual s imula ion. The aim o his s udy
was o imp o e he al eady exis ing models o he ERG signal cha ac e is ics wi h he
iden i ica ion and inclusion o he key componen s called i-wa es in he pho opic esponse.
In o de o e i y he p oposed cha ac e is ics model—Adi hya e al. [
131
] de eloped a
signal analysis and p ocessing algo i hm based on he mul i- esolu ion analysis, which
eliably sepa a ed a ious basic componen s o hese signals. Finally, he accu acy o
his sepa a ion was assessed quan i a i ely and quali a i ely by calcula ing he Pea son
co ela ion coe icien and he co esponding sca e plo s be ween he composi e and he
econs uc ed ERG signal.
Re ini is Pigmen osa (RP) is one o he degene a i e diseases o he e ina a ec ing
he eye signals. The ERG is a signal, which plays an impo an ole in he diagnosis and
ea men o he RP. This signal con ains use ul in o ma ion, which canno be de ec ed
only in he ime domain. Ebdali e al. [
132
] in es iga ed he in luence o he RP on he
ime, equency and ime- equency pa ame e s o he ERG using he Fou ie and wa ele
ans o m me hods. In Table 5, a summa y o signal p ocessing me hods o he ERG da a
is p esen ed.
Senso s 2021,21, 6064 21 o 32
Table 5. Summa y o he ERG signal p ocessing me hods.
Me hod O e all SNR Compu a ional Real-Time Implemen a ion
Pe o mance Imp o emen Cos Complexi y
DF Medium Low Low Yes Simple
ANC High Medium Medium Yes Medium
WT Medium Medium High Yes Medium
EMD High Medium High No Medium
7. Elec ohys e og aphy
Moni o ing o he u e ine con ac ions is commonly used in o de o de e mine
whe he childbi h is coming. In he beginning, he u e ine ac i i y is weak and local-
ized, bu wi h inc easing p egnancy du a ion, he con ac ing g adually becomes s onge
and s onge , hy hmical, and well p opaga ed. Nowadays, he in au e ine p essu e
ca he e (IUCP) is a golden s anda d o he moni o ing o he u e ine con ac ions, bu i
equi es memb ane up u ing, so i can be used only du ing labo and i ca ies a isk o he
in apa um in ec ion. Fo he nonin asi e moni o ing o he u e ine con ac ions, a oco-
dynamome y is commonly used in clinical p ac ice in o de o de e mine bo h equency
and du a ion o he con ac ions. Un o una ely, his app oach is inaccu a e, uncom o -
able, and is pa icula ly dependen on he subjec i e e alua o ’s (medical p o essional)
assessmen [133,134].
Elec ohys e og aphy (EHG) is a nonin asi e me hod o sensing he elec ical ac i i y
o he u e ine con ac ions eco ded om he elec odes placed on he ma e nal abdomen.
This me hod has been known o mo e han six y yea s and p o ides aluable in o ma ion
o e alua ion o he con ac ion in ensi y and s eng h. No e ha he u e ine elec ical
ac i i y changes du ing p egnancy and when he bi h is coming. This is e lec ed in he
empo al and spec al cha ac e is ics o he EHG signals. Some pape s discuss ha bo h
he eloci y and he di ec ion a e associa ed wi h he con ac ion e iciency. Howe e ,
he signal esul ing om he EHG con ains a lo o a i ac s, so i is di icul o es ima e he
use ul in o ma ion. Mo eo e , he e is s ill no s anda dized app oach o he EHG signal
p ocessing and acquisi ion [135–138].
7.1. EHG Wa e o m
The EHG signal is eco ded wi h he implemen a ion o he abdominal elec odes and
could be desc ibed as a slow elec ical wa e wi h he equency wi hin he anges o 0.03
o 0.1 Hz and ampli ude o 1–5 mV. The as -elec ical ac i i y is supe imposed on his
slow elec ical wa e wi h he equency o 0.3–2 Hz and he ampli ude
50 µV–1 mV.
In
he spec al domain, he EHG signal lies in he in e al be ween 0.1 and 3–5 Hz [
139
,
140
].
Figu es 13 and 14 show plo s o he IUCP and he EHG eco ded signals and Figu e 15
shows he powe spec um o he i s con ac ion om he Figu e 14.
P essu e (mmHg)
0
100
50
Time (s)
150 300 450 6000
Figu e 13. Plo o a eco ded IUCP signal.
Senso s 2021,21, 6064 22 o 32
Ampli ude (mV)
Time (s)
−0.2
0.2
0
150 300 450 6000
Figu e 14. Plo o a eco ded EHG signal.
F equency (Hz)
Powe Spec um (mV)
0
0.25 0.50 1
0.2
0.15
0.1
0.05
0.75
Figu e 15. Powe spec um o he i s con ac ion om he Figu e 14.
Gond y e al. in 1993 [
141
] showed ha he measu emen o he EHG signals could be
pe o med as ea ly as a he 19 weeks o p egnancy. Fo he nonin asi e EHG measu e-
men s he e a e commonly used Ag/AgCl elec odes (8 mm diame e ). An example o he
app op ia e elec ode placemen o he EHG is shown in Figu e 16 [
139
,
142
]. Con igu a ion
and placemen o he elec ode in he egion immedia ely below he umbilicus p o ides
he bes SNR. Fo he EHG measu emen pu poses he 25 mm in e elec ode dis ance is
commonly used, he e e ence elec ode is placed on he igh hip and he g ound elec ode
is placed on he le hip [139,140,143–145].
Figu e 16. Con igu a ion o loca ion o he elec odes o he EHG acquisi ion.
Senso s 2021,21, 6064 23 o 32
7.2. Clinical Applica ions
Cu en ly, his p omising nonin asi e me hod o sensing he u e ine con ac ions
is no used in clinical p ac ice due o he di icul ies in ol ed in in e p e a ion o he
in o ma ion con ained in he EHG signal. This signal con ains a lo o in e e ences coming
om bo h mo he and e us. Va ious ex ac ion echniques we e used in o de o imp o e
he quali y o he EHG signals and he au oma ic de ec ion o he bi h o p egnancy ela ed
con ac ions. Some s udies [
146
–
148
] show ha he EHG seems o be a mo e accu a e
al e na i e me hod o he IUCP han he ocodynamome y, especially o women wi h a
highe BMI. Howe e , al hough he EHG accu a ely de ec s he u e ine con ac ions and
he complexi y o he con ac ion’s cha ac e is ics ob ained wi h he EHG, which canno be
compa ed o he IUCP [149].
Pa ame e s o he EHG signals ob ained du ing con ac ions could be used o he
pu pose o diagnosis o he p e e m labo coming and o p edic p e e m deli e y. G ea
a en ion has been ecen ly paid on p edic ing labo and disc imina ing p e e m con ac-
ions based on he in o ma ion ecei ed om he EHG signals. Some p e e m con ac ion
will lead o he p e e m deli e y, bu some o he s will no , which makes hei app op ia e
in e p e a ion a e y challenging ask. The pa ame e s de i ed om he EHG signals
and ela ed wi h hei equency a e expec ed o be mo e compa able om one subjec o
ano he han he pa ame e s ela ed wi h he ampli ude and less sensi i e o he elec ode
placemen [133].
7.3. EHG Signal P ocessing
The EHG is a e y p oblema ic signal because i has a e y low equency and a e y
low ampli ude o en leading o i s con usion wi h di e en noise and in e e ences. This
signal con ains he u e ine elec ical ac i i y (con ac ions) bu also a lo o in e e ence,
such as hose caused by he abdominal muscle ac i i y, he baseline luc ua ions, o he
mo ion a i ac s [
139
,
148
,
150
]. P esence o hese a i ac s could lead o dis o ion o he
signal’s spec al powe densi y and o he un eliable (ine icien ) au oma ic iden i ica ion o
he con ac ions ob ained om he EHG signal. The mo ion a i ac s a e he main p oblem
in he analysis o he EHG signals. Howe e , also a g ea a en ion is needed o o be pu
on he ECG and he mECG signals. I needs o be ealized and aken in o accoun ha
he noise p esen in he monopola EHG is non-s a iona y and has usually much highe
ampli ude han he signal o in e es . In addi ion, he abo e-men ioned in e e ences ha e
hei main equency componen e y close o he equency o he EHG signal.
The in e e ence signals a e usually o e lapping he EHG signal o in e es in bo h
ime and equency domain, he e o e i is necessa y o educe o comple ely elimina e
hem. Nowadays, a ious ad anced signal p ocessing me hods a e used in o de o ex ac
he EHG signal om signals measu ed wi h he elec odes placed on he ma e nal abdomen.
Ve y o en he a ious digi al il e s—EMD, ICA, and WT—a e used o he pu pose o
in e e ence componen s elimina ion om he measu ed signals and in o de o es ima e
he app op ia e EHG signal. Ve y p omising esul s a e gi en by some mode n hyb id
me hods, especially based on he combina ion o he EMD wi h some o he me hods.
Table 6shows a summa y o me hods o he EHG signal p ocessing.
Table 6. Summa y o he EHG signal p ocessing me hods.
Me hod O e all SNR Compu a ional Real-Time Implemen a ion
Pe o mance Imp o emen Cos Complexi y
DF Low Medium Low Yes Simple
DWT Low High Medium Yes Medium
EMD High Medium High No Medium
Hyb id me hods High High High No Complex
Senso s 2021,21, 6064 24 o 32
7.3.1. Linea Fil e ing
Ye-Lin e al. in 2014 [
139
] used a 5 h-o de Bu e wo h bandpass digi al il e wi h
i s cu -o equencies a 0.1 and 4 Hz o he pu pose o he da a denoising. They also
downsampled he analyzed signals o only 20 Hz in o de o dec ease he compu a ional
cos . They concluded ha such a low sampling equency is su icien o he compu a ion
o he spec al pa ame e s and ha he applied Bu e wo h bandpass digi al il e is
sui able o he pu pose o he da a p ep ocessing. Ga cia-Gonzalez e al. in 2013 [
151
] used
a bandpass digi al il e wi h i s cu -o equencies a 0.3 and 4 Hz o denoise he EHG
signals, whe e Acha ya e al. in 2017 [
152
] used a 4-pole bandpass Bu e wo h il e wi h
he cu -o equencies a 0.3 and 3 Hz o ca y ou he p ep ocessing o he EHG signals.
7.3.2. Wa ele T ans o m
Leman and Ma que, in 2000 [
153
], ied o elimina e he ECG signal om he measu ed
EHG signals on he ma e nal abdomen wi h he implemen a ion o he WT. Fo hei
expe imen s hey applied simul aneously eco ded mECG signals wi h he EHG signals.
Thei p oposed algo i hm was based on he local minima de ec ion in he his og am. They
concluded ha he ECG signal was e y e icien ly emo ed om he EHG da a and
imp o ed i s o e all quali y and eligibili y o u he p ocessing.
Bei an and e al., in 2017 [
154
], used he DWT in o de o ex ac some ea u es om
he EHG signals. A e pe o ming he DWT, he suppo ing ec o machine echnique was
used in o de o classi y he analyzed signals. Real da a om an open access da abase, The
Te m-P e e m EHG, was used in hei s udy, whe e hey chose 26 eco dings om he e m
deli e y and 26 eco dings om he p e e m deli e y. Fo he e alua ion pu poses he
calcula ion o ela i e wa ele ene gy, oo mean squa e, accu acy, sensi i i y, and speci ici y
we e applied. They came o he esul s ha he bes accu acy was achie ed wi h he use o
he 4 h le el DWT decomposi ion and wi h he ma e nal wa ele db2.
7.3.3. Empi ical Mode Decomposi ion
Ta alunga e al., in 2015 [
155
], p oposed he EMD me hod o denoising he EHG
signals. The main ad an age o hei me hod is ha no a p io i knowledge ega ding he
signal is equi ed, because i is ully da a-d i en. Fo he expe imen pu poses, he syn he ic
da a and he calcula ion o he SNR imp o emen we e used. They concluded ha he
EHG signal has a signi ican p edic i e alue in case o a p e e m labo . The au ho s also
men ioned ha when he SNR o he inpu signals is low, hen he EMD me hod could be
combined wi h ano he me hod in o de o imp o e he o e all pe o mance.
Ren e al. in 2015 [
156
] used he EMD me hod o he isk o p e e m deli e y as-
sessmen based on he in o ma ion ob ained om he EHG signals. The eal da a used
in hei s udies was acqui ed om he open access o he Te m-P e e m EHG da abase
(262 e m and 38 p e e m). A e pe o ming he EMD o he pu pose o he ex ac ion
o he IMF, he ins an aneous ampli ude and he equency o each IMF componen we e
compu ed. A ea unde he cu e (AUC) alues was used as an e alua ion pa ame e in
ha s udy. They came o he conclusion ha hei app oach imp o es p edic ion accu acy
o he p e e m deli e y isk compa ed wi h some p e ious app oaches based on he high
alue o he AUC.
7.3.4. Hyb id Me hods
Chkei e al., in 2010 [
157
], p oposed he EHG enhancemen me hod based on he
combina ion o he EMD-based me hod wi h he SG il e , which enabled o emo e he
high- equency noises and he baseline wande wi h a minimal signal dis o ion. They
used he eal da a p o ided by he Landspi ali Uni e si y hospi al in Iceland ( hey selec ed
10,000 andom
con ac ions) and a ious a i icially gene a ed signals in o de o e alua e
he pe o mance o he p oposed me hod. Fo he pu pose o e alua ion, a SNR imp o e-
men calcula ion was pe o med. The au ho s concluded ha hei app oach p o ided
Senso s 2021,21, 6064 25 o 32
good esul s compa ed wi h he WT me hod, wi h he ad an age o being adap i e and
wi h no need o p e-de ini ion o he pa ame e s.
Hassan e al., in 2011 [
158
], used a combina ion o he canonical co ela ion analysis
(CCA) and he EMD o denoise he monopola EHG. The CCA me hod belongs o he
g oup o he BSS me hods. Thei me hod was compa ed wi h he o he BSS me hods
(ICA, PCA, e c.), while hei app oach sol es he main BSS p oblem by o cing he sou ces
o be maximally au oco ela ed and mu ually unco ela ed, while he mixing ma ix is
assumed o be squa e. Fi s , hey used he CCA in o de o ex ac he u e ine bu s s,
and hen he EMD was used o emo al o he bigges pa o any esidual noise om
bu s s. Thei app oach was compa ed wi h he ICA and wi h he WT. Fo he s udy
pu poses, he eal da a om he Landspi ali Uni e si y hospi al in Iceland we e used
and he me hod’s accu acy was e alua ed wi h he SNR imp o emen calcula ions. They
concluded ha he p oposed me hod success ully emo ed a i ac s om he signal wi hou
al e ing he unde lying u e ine ac i i y. O he me hods analyzed in ha pape did no
achie e compa able accu acy o he one ob ained wi h he implemen a ion o he p oposed
app oach.
Acha ya e al., in 2017 [
152
], p oposed he combina ion o he EMD and he WT o
he es ima ion o he p ema u e deli e y based on he in o ma ion ob ained om he EHG
signals. Real da a om he open access Te m-P e e m EHG da abase we e used in ha
s udy (262 e m and 38 p e e m). Calcula ion o accu acy, sensi i i y, and speci ici y was
used o he e alua ion pu poses. They concluded ha he p oposed me hod eached a e y
good accu acy o 96.25% and could be used in hospi al gynecology depa men s in o de
o p edic he p e e m o no mal deli e y. Simila app oach was used by Hoseinzadeh and
Ami ani in 2018 [
159
]. They also used he same da abase and e alua ion pa ame e s and
concluded ha his app oach achie ed a e y high accu acy.
8. Discussion and Conclusions
This pape is a Pa III wo k and ocuses on a e iew o signal p ocessing me hods o
o he ( emaining) bioelec ical signals, which do no belong o he g oup o signals ela ed
wi h hea o b ain elec ical ac i i y.
The choice o he mos op imal p ocessing me hods was based on he signal ype,
and he signals desc ibed in his wo k a ied, which means hey had among he o he
di e en equency anges, ampli ude spec um, e c.
As i was men ioned in o he pa s o his pape , ypical, classical digi al il e ing
me hods can be applied when bo h he in e e ence and desi ed biological signals ha e
a ious equency anges. The e o e, using adap i e noise canceling echniques can be a
good al e na i e o he adi ional il e ing me hods.
The second g oup o he ad anced signal p ocessing me hods does no equi e he
e e ence signals and seems o be mo e sensi i e o p ese ing he analyzed signal’s
in o ma ion bo h in he ime and he equency domain. One o he me hods p o iding
good ime- equency dis inguish-abili y is he WT. Howe e , he op imal selec ion o he
wa ele decomposi ion pa ame e s and p ope ies o he used ma e nal wa ele is c ucial.
In many s udies, hese p ope ies a e chosen empi ically, because o a ious ime and
equency cha ac e is ics o he signals eco ded om di e en subjec s. The choice o
he app op ia e WT pa ame e s depends also on he ype o biological signal. The e o e,
he in oduc ion o his me hod in o medical p ac ice is no ye well es ablished, especially
as he o he me hods such as ICA and PCA ha e p o ided be e pe o mance in elimina ing
a i ac s o e lapping he equency spec um. Thei disad an age is he necessi y o he
mul ichannel eco dings, as he algo i hms wo k wi h ma ices o he signals. The ICA
me hod has he ad an age o enabling sepa a ion o he highe numbe o componen s han
he numbe o channels (compa ed o he PCA me hod, whe e he numbe o channels gi es
he maximum numbe o componen s). On he o he hand, he independen componen s
a e o di e en ampli ude compa ed wi h he ex ac ed signal, and hei o de is andom
o each calcula ion. Signal decomposi ion using he EMD me hod seems o be use ul o
Senso s 2021,21, 6064 32 o 32
144.
Hassan, M.M.; Te ien, J.; Muszynski, C.; Alexande sson, A.; Ma que, C.; Ka lsson, B. Be e P egnancy Moni o ing Using
Nonlinea Co ela ion Analysis o Ex e nal U e ine Elec omyog aphy. IEEE T ans. Biomed. Eng.
2013
,60, 1160–1166. [C ossRe ]
[PubMed]
145.
Ma que, C.K.; Te ien, J.; Rihana, S.; Ge main, G. P e e m Labou De ec ion by Use o a Biophysical Ma ke : The U e ine Elec ical
Ac i i y. BMC P egnancy Childbi h 2007,7, S5. [C ossRe ] [PubMed]
146.
Euliano, T.Y.; Nguyen, M.T.; Da manjian, S.; McGo ay, S.P.; Euliano, N.; Onkala, A.; G egg, A.R. Moni o ing U e ine Ac i i y
du ing Labo : A Compa ison o 3 Me hods. Am. J. Obs e . Gynecol. 2013,208, 66.e1–66.e6. [C ossRe ]
147.
Euliano, T.Y.; Nguyen, M.T.; Ma osse o, D.; Edwa ds, R.K. Moni o ing Con ac ions in Obese Pa u ien s: Elec ohys e og aphy
Compa ed Wi h T adi ional Moni o ing. Obs e . Gynecol. 2007,109, 1136–1140. [C ossRe ] [PubMed]
148.
Jezewski, J.; Ho oba, K.; Ma onia, A.; W obel, J. Quan i a i e Analysis o Con ac ion Pa e ns in Elec ical Ac i i y Signal o
P egnan U e us as an Al e na i e o Mechanical App oach. Physiol. Meas. 2005,26, 753–767. [C ossRe ]
149.
Jacod, B.C.; G aa sma, E.M.; Van Hagen, E.; Visse , G.H.A. A Valida ion o Elec ohys e og aphy o U e ine Ac i i y Moni o ing
du ing Labou . J. Ma e n. Fe al Neona al Med. 2010,23, 17–22. [C ossRe ]
150.
Schlembach, D.; Mane , W.L.; Ga ield, R.E.; Maul, H. Moni o ing he P og ess o P egnancy and Labo Using Elec omyog aphy.
Eu . J. Obs e . Gynecol. Rep od. Biol. 2009,144, S33–S39. [C ossRe ] [PubMed]
151.
Ga cia-Gonzalez, M.T.; Cha les on-Villalobos, S.; Va gas-Ga cia, C.; Gonzalez-Cama ena, R.; Aljama-Co ales, T. Cha ac e iza ion
o EHG Con ac ions a Te m Labo by Nonlinea Analysis. In P oceedings o he 2013 35 h Annual In e na ional Con e ence o
he IEEE Enginee ing in Medicine and Biology Socie y (EMBC), Osaka, Japan, 3–7 July 2013; IEEE: New Yo k, NY, USA, 2013;
pp. 7432–7435. [C ossRe ]
152.
Acha ya, U.R.; Suda shan, V.K.; Rong, S.Q.; Tan, Z.; Lim, C.M.; Koh, J.E.; Nayak, S.; Bhanda y, S.V. Au oma ed De ec ion o
P ema u e Deli e y Using Empi ical Mode and Wa ele Packe Decomposi ion Techniques wi h U e ine Elec omyog am Signals.
Compu . Biol. Med. 2017,85, 33–42. [C ossRe ]
153.
Leman, H.; Ma que, C. Rejec ion o he Ma e nal Elec oca diog am in he Elec ohys e og am Signal. IEEE T ans. Biomed. Eng.
2000,47, 1010–1017. [C ossRe ]
154.
Bei an and, M.; Shahbakh i, M.; Eslamizadeh, M.; Ba i, M.; Mohammadi a , S. In es iga ing Wa ele Ene gy Vec o o P e-Te m
Labo De ec ion Using EHG Signals. In P oceedings o he 2017 Signal P ocessing: Algo i hms, A chi ec u es, A angemen s,
and Applica ions (SPA), Poznan, Poland, 22–24 Sep embe 2017; IEEE: New Yo k, NY, USA, 2017; pp. 269–274. [C ossRe ]
155.
Ta alunga, D.D.; Ungu eanu, M.; Hu ezeanu, B.; Gussi, I.; S unga u, R. Empi ical Mode Decomposi ion Applied o Non-
In asi e Elec ohys e og ahic Signals Denoising. In P oceedings o he 2015 37 h Annual In e na ional Con e ence o he
IEEE Enginee ing in Medicine and Biology Socie y (EMBC), Milan, I aly, 25–29 Augus 2015; IEEE: New Yo k, NY, USA, 2015;
pp. 4134–4137. [C ossRe ]
156.
Ren, P.; Yao, S.; Li, J.; Valdes-Sosa, P.A.; Kend ick, K.M. Imp o ed P edic ion o P e e m Deli e y Using Empi ical Mode
Decomposi ion Analysis o U e ine Elec omyog aphy Signals. PLoS ONE 2015,10, e0132116. [C ossRe ] [PubMed]
157.
Chkei , A.; Ma que, C.; Te ien, J.; Ka lsson, B. Denoising Elec ohys e og am ia Empi ical Mode Decomposi ion. In P oceedings
o he ISSNIP Biosignals. And Bio obo ics Con e ence, Vi o ia, B azil, 4–6 Janua y 2010; pp. 32–35.
158.
Hassan, M.; Boudaoud, S.; Te ien, J.; Ka lsson, B.; Ma que, C. Combina ion o Canonical Co ela ion Analysis and Empi ical
Mode Decomposi ion Applied o Denoising he Labo Elec ohys e og am. IEEE T ans. Biomed. Eng.
2011
,58, 2441–2447.
[C ossRe ] [PubMed]
159.
Hoseinzadeh, S.; Ami ani, M.C. Use o Elec o Hys e og am (EHG) Signal o Diagnose P e e m Bi h. In P oceedings o he
Elec ical Enginee ing (ICEE), I anian Con e ence, Mashhad, I an, 8–10 May 2018; IEEE: New Yo k, NY, USA, 2018; pp. 1477–1481.
[C ossRe ]
