Clinical decision support system to electrostimulation treatments for muscle rehabilitation in the elderly

Uni e sidade do Minho
Escola de Engenha ia
Tiago Sanches F anco
Clinical Decision Suppo Sys em o
Elec os imula ion T ea men s o
Muscle Rehabili a ion in he Elde ly
Oc obe , 2024
UMinho | 2024 Tiago Sanches F anco Clinical Decision Suppo Sys em o Elec os imula ion
T ea men s o Muscle Rehabili a ion in he Elde ly
Uni e si y o Minho
School o Enginee ing
Tiago Sanches F anco
Clinical Decision Suppo Sys em o
Elec os imula ion T ea men s o
Muscle Rehabili a ion in he Elde ly
PhD Thesis
Doc o a e in In o ma ics
Thesis supe ised by
Ped o Manuel Rangel San os Hen iques
Paulo Alexand e Va a Al es
oc obe 2024
Copy igh and Te ms o Use o Thi d Pa y Wo k
This disse a ion epo s on academic wo k ha can be used by hi d pa ies as long as he in e na ionally
accep ed s anda ds and good p ac ices a e espec ed conce ning copy igh and ela ed igh s.
This wo k can he ea e be used unde he e ms es ablished in he license below.
Reade s needing au ho iza ion condi ions no p o ided o in he indica ed licensing should con ac he
au ho h ough he Reposi ó iUM o he Uni e si y o Minho.
License g an ed o use s o his wo k:
CC BY-NC-SA
h ps://c ea i ecommons.o g/licenses/by-nc-sa/4.0/
i
Acknowledgemen s
Wi h deep g a i ude, I would like o since ely hank e e yone who was pa o his challenging doc o al
jou ney. This wo k esul s om he suppo , dedica ion, and gene osi y o many.
Fi s and o emos , I am inc edibly hank ul o my supe iso , P o esso Ped o Hen iques, and my
co-supe iso , P o esso Paulo Al es. Thei knowledge, pa ience, and consis en guidance led me h ough
each s ep o his jou ney. I am equally g a e ul o P o esso Ma ia João Va anda Pe ei a, who p o ided
weekly suppo alongside my supe iso s. Thank you o he us you placed in me and he challenges
you p esen ed, which ha e shaped me bo h as a esea che and a pe son.
To my pa ne and lo e, Laís Fabiana Se a ini, who was by my side h ough momen s o exhaus ion,
joy, and achie emen . I o e my deepes hanks o you essen ial suppo , unde s anding, and pa ience
h oughou his jou ney.
I am also g a e ul o my a he and mo he , whose a en i eness and ca e ha e been a cons an sou ce
o s eng h, e en om a a . To my sis e , whose encou aging wo ds ha e suppo ed me no only in his
wo k bu h oughou my li e, I am deeply hank ul.
To he pa icipan s in he expe imen al s udies, hank you o gene ously sha ing you ime and com-
mi men o help ad ance his p ojec . Wi hou you in ol emen , his wo k would no ha e been possible.
I would also like o hank P o esso s Paulo Lei ão, Tiago Ped osa, and José Ru ino, who played key
oles in he NanoS im p ojec , along wi h s uden s Leona do Ses em de Oli ei a, Felipe Gimenez da Sil a,
Raul Kaize , and João Gonçal es, who con ibu ed signi ican ly o he p ojec .
I am also hank ul o physio he apis s Nelson Aze edo and Elsa Cos a o hei echnical and scien i ic
suppo h oughou he NanoS im p ojec .
My since e hanks o Ana Ca olina Ca doso De Sousa and he Resea ch Cen e o Biomedical Engi-
ii

nee ing membe s a he Uni e si a Poli ècnica de Ca alunya. I am g a e ul o you wa m welcome and
p o essionalism, which p o ided a s imula ing en i onmen .
To all my colleagues a he Resea ch Cen e in Digi aliza ion and In elligen Robo ics (CeDRI), wi h
whom I sha ed coun less con e sa ions, discussions, and cups o co ee, hank you o you iendship.
Finally, I am g a e ul o he Founda ion o Science and Technology (FCT), Po ugal o he PhD schol-
a ship g an numbe
2020.05704.BD
, which made his esea ch possible and allowed me o ocus en i ely
on my s udies.
This wo k was suppo ed Eu opean Regional De elopmen Fund (ERDF) h ough he Ope a ional P o-
g amme o Compe i i eness and In e na ionaliza ion (COMPETE 2020), unde Po ugal 2020, in he
amewo k o he NanoS im (POCI-01-0247-FEDER-045908) p ojec . This wo k was suppo ed by na-
ional unds h ough FCT/MCTES (PIDDAC): CeDRI, UIDB/05757/2020 and UIDP/05757/2020; SusTEC,
LA/P/0007/2020; Cen o ALGORITMI, UIDB/00319/2020.
This wo k e lec s he con ibu ions o each o you, and I am since ely g a e ul.
iii
S a emen o In eg i y
I he eby decla e ha ing conduc ed his academic wo k wi h in eg i y.
I con i m ha I ha e no used plagia ism o any o m o undue use o in o ma ion o alsi ica ion o esul s
along he p ocess leading o i s elabo a ion.
I u he decla e ha I ha e ully acknowledged he Code o E hical Conduc o he Uni e si y o Minho.
Uni e si y o Minho, B aga, oc obe 2024
Tiago Sanches F anco
i
Abs ac
The aging popula ion poses inc easing challenges o heal hca e sys ems, pa icula ly ega ding long- e m
muscula ehabili a ion se ices o he elde ly. Condi ions such as sa copenia and knee os eoa h i is,
which a e common in his age g oup, comp omise mobili y and quali y o li e, making ehabili a ion es-
sen ial o ensu ing ac i e aging. Neu omuscula elec ical s imula ion has p o en o be an e ec i e in e -
en ion, bu i s p ope applica ion depends on pe sonalized pa ame e iza ion and con inuous moni o ing.
This PhD p ojec p esen s an inno a i e solu ion o his scena io: a Clinical Decision Suppo Sys em
(CDSS) aimed a elec os imula ion ea men s. De eloped wi hin he scope o he NanoS im p ojec ,
CDSS in eg a es Indus y 4.0 echnologies o c ea e an e icien , po able, and cos -e ec i e wea able
sys em, allowing ea men o be conduc ed a home, wi h inc eased accessibili y and educed clinical
isi s. A sys em a chi ec u e wi h sensi i e da a p o ec ion o home ea men scena ios is p oposed.
The sys em p ocesses eal- ime muscle a igue me ics, in e p e s knee ex ension exe cises, and adjus s
elec os imula ion in a pe sonalized manne . Th ee expe imen al s udies alida ed CDSS. In he i s ,
he sys em demons a ed he abili y o synch onize bio eedback da a and emo e a i ac s om he elec-
omyog aphic signal. In he second, he accu acy o he mobile applica ion in p ocessing muscle a igue
me ics in eal- ime was demons a ed. In he hi d, he CDSS was compa ed wi h a gold-s anda d mo ion
cap u e sys em, and esponsi e elec os imula ion modes based on eal- ime bio eedback we e es ed.
Finally, a s a is ical co ela ion analysis was conduc ed be ween he a igue le els p edic ed by he CDSS
and he pa icipan s’ subjec i e eedback. The CDSS ecei ed e hical app o als om di e en ins i u ions
and clea ance o sensi i e da a s o age. The analyzed esul s show ha CDSS has he po en ial o inno-
a e in he ield o emo e muscula ehabili a ion and imp o e he quali y o li e o he elde ly h ough
pe sonalized ea men s.
Keywo ds Bio eedback, Decision Suppo , Elec os imula ion, Remo e Rehabili a ion, Wea able De ice
Resumo
O en elhecimen o da população impõe desa ios c escen es aos sis emas de saúde, especialmen e no
que diz espei o aos se iços de eabili ação muscula de longo p azo pa a idosos. Condições como sa -
copenia e os eoa i e do joelho, comuns nessa aixa e á ia, comp ome em a mobilidade e a qualidade de
ida, o nando a eabili ação essencial pa a ga an i um en elhecimen o a i o. A ele oes imulação neu-
omuscula em se mos ado uma in e enção e icaz, mas a sua aplicação adequada depende de uma
pa ame ização pe sonalizada e acompanhamen o con ínuo. Es a ese de dou o amen o ap esen a uma
solução ino ado a pa a esse cená io: um Sis ema de Supo e à Decisão Clínico (SSDC) pa a a amen os
de ele oes imulação. Desen ol ido no âmbi o do p oje o NanoS im, o SSDC in eg a ecnologias da Indús-
ia 4.0 pa a c ia um sis ema wea able e icien e, po á il e de baixo cus o, pe mi indo que o a amen o
seja ealizado em casa, com maio acessibilidade e edução de consul as clínicas. Assim, é p opos o
uma a qui e u a de sis ema com p o eção de dados sensí eis pa a o cená io de a amen o ao domicilia .
O sis ema p ocessa em empo eal mé icas elacionadas a adiga muscula , in e p e a o exe cício de
ex ensão de joelho e ajus a a ele oes imulação de o ma pe sonalizada. T ês es udos expe imen ais ali-
da am o SSDC. No p imei o, o sis ema demons ou a capacidade de sinc oniza os dados de bio eedback
e emo e a e a os do sinal ele omiog á ico. No segundo, oi e idenciada a p ecisão da aplicação mó el
em p ocessa mé icas de adiga muscula em empo eal. No e cei o, o SSDC oi compa ado com um
sis ema pad ão-ou o de cap u a de mo imen o, o am es ados modos de ele oes imulação esponsi os
ao bio eedback cole ado em empo eal. Po im, oi ealizado uma análise es a ís ica de co elação en-
e o ní el de adiga p e is o pelo SSDC com o eedback subje i os dos pa icipan es. O SSDC ecebeu
ap o ações é icas de di e en es ins i uições e ap o ação pa a a mazenamen o de dados sensí eis. Os e-
sul ados analisados mos am que o SSDC em po encial de ino a a á ea de eabili ação muscula emo a
e melho a a qualidade de ida dos idosos po meio de a amen os pe sonalizados.
Pala as-cha e Bio eedback, Disposi i o Ves í el, Ele oes imulação, Reabili ação Remo a, Supo e a
Decisão
i
55 Example o segmen wi h iden i ied s imula ion a i ac . . . . . . . . . . . . . . . . . 111
56 Example o EMG signal cap u ed du ing he FES expe imen il e ed. . . . . . . . . . . 113
57 Example o segmen wi h iden i ied s imula ion a i ac il ed. . . . . . . . . . . . . . 113
58 Mean muscle ac i i y o subjec s in mV. . . . . . . . . . . . . . . . . . . . . . . . . 114
59 Maximum Volun a y Con ac ion o subjec s. . . . . . . . . . . . . . . . . . . . . . . 114
60 Mean muscle ac i i y o subjec s no malized by MVC. . . . . . . . . . . . . . . . . . 115
61 Va ia ion in Maximum Angle and Du a ion o Knee Ex ension Mo emen - Expe imen 1. . 116
62 Cycles o Knee Ex ension Mo emen . . . . . . . . . . . . . . . . . . . . . . . . . . 117
63 Muscle Ac i i y in Knee Angle Ra io. . . . . . . . . . . . . . . . . . . . . . . . . . . 118
64 Compa ison be ween he Fi s and Thi d Tes . . . . . . . . . . . . . . . . . . . . . . 119
65 Compa ison be ween he Fi s and Las Mo emen . . . . . . . . . . . . . . . . . . . 119
66 T ajec o y o Median F equency and RMS du ing he Fi s and Thi d es s. . . . . . . . . 121
67 Conca ena ed T ajec o y o Median F equency and RMS du ing Tes s 1 and 3. . . . . . 122
68 Va ia ion in Maximum Angle and Du a ion o Knee Ex ension Mo emen - Expe imen 2. . 125
69 Cycle o Knee Ex ension Mo emen - Exp 2. . . . . . . . . . . . . . . . . . . . . . . 126
70 T end o he a gF eq me ic du ing he session. . . . . . . . . . . . . . . . . . . . . 128
71 T end o he medianF eq me ic du ing he session. . . . . . . . . . . . . . . . . . . 128
72 T end o he RMS me ic du ing he session. . . . . . . . . . . . . . . . . . . . . . . 128
73 JASA Me hod apply o Expe imen al Tes . . . . . . . . . . . . . . . . . . . . . . . . 129
74 (A) Occu ence o Fa igue (B) In ensi y o Occu ences by Con igu a ion. . . . . . . . . 131
75 Fa igue Le els Achie ed by Con igu a ion. . . . . . . . . . . . . . . . . . . . . . . . 132
76 Con ac ion numbe a i s a igue le el achie ed. . . . . . . . . . . . . . . . . . . . 132
77 (A) Amoun o Occu ence (B) A e age In ensi y o Fi s and Las Le el pe Pa icipan . . 133
78 (A) Con ac in Numbe o Le els Change (B) A e age In ensi y pe Le el. . . . . . . . . 134
79 Compa ison be ween sys ems o he A gF eq me ic. ................135
80 Compa ison be ween sys ems o he MedF eq me ic.................135
81 Compa ison be ween sys ems o he RMS me ic. ..................135
82 MoCap equipmen : Came a (A) and Body Make s (B). . . . . . . . . . . . . . . . . . 138
83 Expe imen al Se up Showing he Posi ioning o Elec odes, Senso s, and Ma ke s. . . . . 140
84 P e-P ocessing Me hodology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
85 OpenSimModelScaling. ...............................142
86 Va ia ion in Maximum Angle and Du a ion o Knee Ex ension Mo emen - Expe imen 3. . 144
xiii

87 Compa ison o Maximum Angle Va ia ion be ween CDSS and MoCap. . . . . . . . . . . 145
88 Compa ison o Maximum Angle Va ia ion be ween CDSS Adjus ed and MoCap. . . . . . 146
89 Kinema ic Compa ison o Mo emen be ween he sys ems. . . . . . . . . . . . . . . . 147
90 Dynamic Compa ison o Mo emen be ween he sys ems. . . . . . . . . . . . . . . . 148
91 Knee angle a he s a and end o s imula ion modes.. . . . . . . . . . . . . . . . . . 150
92 EMG signal du ing
Raising he Leg
s imula ion. ....................151
93 EMG signal du ing
Range o Angle
s imula ion......................151
94 Sel -Repo ed Feedback o Expe imen 3. . . . . . . . . . . . . . . . . . . . . . . . 152
95 S imula ion Noise in he EMG signal. (A)
Raising he Leg
(B)
Range O Angle
. . . . . . . 154
96 Muscle Ac i i y in Bo h Legs - Expe imen 3. . . . . . . . . . . . . . . . . . . . . . . 155
97 Muscle Ac i i y Va ia ion Du ing Expe imen al Sessions in Bo h Legs - Expe imen 3. . . 155
98 T end o he A e ange F equency Me ic - Expe imen 3. . . . . . . . . . . . . . . . . 156
99 T end o he Median F equency Me ic - Expe imen 3. . . . . . . . . . . . . . . . . . 156
100 T end o he RMS Me ic - Expe imen 3. . . . . . . . . . . . . . . . . . . . . . . . . 157
101 Join Ampli ude and Spec um Analysis (JASA) - Expe imen 3. . . . . . . . . . . . . . 157
102 Fa igue Occu ences by Con igu a ion in Bo h Legs - Expe imen 3. . . . . . . . . . . . 158
103 In ensi y o Occu ences by Con igu a ion in Bo h Legs - Expe imen 3. . . . . . . . . . 159
104 Fa igue Le els Achie ed by Con igu a ion on he Le Leg - Expe imen 3. . . . . . . . . 160
105 Fa igue Le els Achie ed by Con igu a ion on he Righ Leg - Expe imen 3. . . . . . . . 160
106 Con ac ion Numbe a Fi s Fa igue Le el Achie ed on he Le Leg - Expe imen 3. . . . 161
107 Con ac ion Numbe a Fi s Fa igue Le el Achie ed on he Righ Leg - Expe imen 3. . . 161
108 Amoun o Fa igue Occu ences (A) and In ensi y (B) by Subjec s on he Le Leg. . . . . 162
109 E olu ion o Fa igue Pe cep ion in he Sel -Repo on he Le Leg. . . . . . . . . . . . 162
110 Amoun o Fa igue Occu ences (A) and In ensi y (B) by Subjec s on he Righ Leg. . . . 163
111 E olu ion o Fa igue Pe cep ion in he Sel -Repo on he Righ Leg. . . . . . . . . . . . 164
112 Compa ison o sel - epo a igue wi h p edic ed le el by subjec . . . . . . . . . . . . . 166
xi
Lis o Tables
1 InclusionSea chTe ms................................. 24
2 ExclusionSea chTe ms. ............................... 25
3 Resea chQues ions. ................................. 26
4 InclusionC i e ia.................................... 27
5 ExclusionC i e ia.................................... 27
6 Iden i ica ion o Selec ed A icles. . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
7 Da a and Acquisi ion Tools Used in he Selec ed S udies. . . . . . . . . . . . . . . . . 38
8 S udy pa icipan s by ype. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
9 Bio eedback da a and packe olume. . . . . . . . . . . . . . . . . . . . . . . . . . 59
10 Op imized Da a T ansmission F equency. . . . . . . . . . . . . . . . . . . . . . . . 59
11 E alua ion me ics o he applied il e s in deg ees. . . . . . . . . . . . . . . . . . . . 104
12 An h opome ic da a o he subjec s - Expe imen 1 . . . . . . . . . . . . . . . . . . 107
13 Compa a i e Me ics be ween he Fi s and Thi d Tes s. . . . . . . . . . . . . . . . . 120
14 An h opome ic da a o he subjec s - Expe imen 2. . . . . . . . . . . . . . . . . . . 124
15 Assessmen me ics on a igue me ics. . . . . . . . . . . . . . . . . . . . . . . . . 136
16 An h opome ic da a o he subjec s - Expe imen 3. . . . . . . . . . . . . . . . . . . 139
17 Kinema ic E alua ion Me ics Compa ison. . . . . . . . . . . . . . . . . . . . . . . . 147
18 Dynamic E alua ion Me ics Compa a ion. . . . . . . . . . . . . . . . . . . . . . . . 148
19 A e age Va ia ion in Fa igue and Com o Ac oss Expe imen al Sessions. . . . . . . . . 153
20 Co ela ions be ween p edic ed and sel - epo ed a igue le els. . . . . . . . . . . . . . 167
x
Chap e 1
In oduc ion
This chap e opens he disse a ion, ou lining i s con en s. Ini ially, he i s sec ion con ex ualizes he
mo i a ion behind he doc o al p ojec . Subsequen ly, he second sec ion explains he objec i es and
highligh s hei ele ance in he esea ch con ex . The hi d sec ion de ails he esea ch me hodologies
used in he s udy. Finally, he ou h sec ion discusses he gene al s uc u e o his documen .
1.1 Mo i a ion and Rele ance
O e he las ew decades, a subs an ial global inc ease in li e expec ancy has been eco ded, pa icula ly
in egions such as Eu ope and Po ugal. This posi i e end is linked o ad ances in heal hca e echnology,
imp o emen s in li ing s anda ds, and g ea e access o educa ion. No ably, he demog aphic g oup o
indi iduals aged 65 and abo e, cons i u ing he elde ly popula ion, s ands ou as one o he as es -g owing
segmen s, signi ican ly impac ing socie al dynamics and he heal hca e sys ems (Da ani e al., 2023).
Figu e 1: Popula ion g ow h in Po ugal by age g oup (Ri chie e al., 2023).
To unde s and he magni ude o popula ion g ow h, Figu e 1 illus a es popula ion g ow h in Po ugal
di ided by age g oups. Compa ing s a is ics om 1980 o 2020, i is possible o no e ha Po ugal’s
elde ly popula ion g ew om 1.12 million o 2.32 million, ep esen ing an inc ease o mo e han 100%.
1
Simila ly, Eu opean popula ions a e e lec ing his g ow h a a p opo ional a e, wo ldwide his g ow h is
e en bigge , close o 200% o he same pe iod (Ri chie e al., 2023).
As he p opo ion o olde adul s inc eases, he demand o p ima y and long- e m heal h se ices
also g ows, especially hose ela ed o ch onic condi ions and age- ela ed diseases. The g adual decline
in unc ional capaci y due o aging is na u al p og ess common o all li ing beings, equi ing cons an
ca e o p ese e an ac i e and digni ied li e (WHO, 2021). Thus, ehabili a ion ea men s ace a c i ical
accessibili y p oblem ha ends o ge wo se.
A S udy was ca ied ou by Cieza e al. (2020) o assess he demand o ehabili a ion se ices wo ld-
wide and i s impac on he heal hca e sys em. The esul s ob ained e eal ha one in h ee people in he
wo ld, app oxima ely 2.4 billion indi iduals, need hese se ices. These es ima es con adic he common
belie ha he need o ehabili a ion is es ic ed o a limi ed numbe o people. This inding highligh s he
magni ude o he demand o ehabili a ion se ices and emphasizes he impo ance o e ec i e s a egies
o se e his conside able po ion o he global popula ion.
The conce n abou he delay in inding he necessa y ea men is add essed by he Wo ld Heal h
O ganiza ion (WHO) in hei epo on heal hy aging o he decade be ween 2021 and 2030 (WHO, 2021).
The epo highligh s he signi icance o imp o ing unc ional capaci y o p omo e mo e independen aging
and sugges s da a-d i en solu ions o achie e scalable and accessible answe s.
The adi ional ea men o mobili y ehabili a ion is egula physical exe cise, imp o ing he pe o -
mance o s eng h, endu ance, balance, lexibili y, and so on. Howe e , egula aining o he elde ly can
be expensi e since some pa hologies such as ca dio ascula diseases and neu omuscula p oblems can
make he execu ion o some exe cises un easible, gene a ing a need o a end specialized aining cen e s
o hi e specialized agen s (Chodzko-Zajko e al., 2009).
A solu ion posi ioned as an in e es ing and less expensi e al e na i e o he adi ional aining me hods
could be home-based neu omuscula elec ical s imula ion (NMES), also known as Func ional Elec ical
S imula ion (FES). NMES is a echnique ha a i icially p oduces muscle con ac ion h ough he applica ion
o an elec ical cu en wi h su ace elec odes on he muscles. The e is s ill no consensus on all eac ions
ha he applica ion o NMES p o okes in he body and especially in he muscle. Howe e , i is belie ed ha
aining wi h NMES p oduces an inc ease in he muscle’s capaci y o gene a e o ce, igge ing a se ies o
bene i s (Imo o e al., 2013).
NMES is gene ally applied o imp o e unc ional capaci y, p ese a ion and eco e y o muscle mass,
and muscle s eng hening. A sys ema ic e iew p esen ed by Langea d e al. (2017) show ha aining
wi h NMES is sa e and has an e iciency simila o adi ional aining in he elde ly. Fu he mo e, hey
2
show ha aining wi h NMES can b ing conside able bene i s o he physiological s a e, imp o ing gai
and balance pe o mance, especially in less ac i e elde ly people.
Despi e his, he use o NMES as a aining me hod o muscle ehabili a ion in he elde ly is conside ed
ecen . So a , he e is no consensus on which con igu a ions p oduce a mo e adequa e ehabili a ion.
Se e al pa ame e s ha e al eady been es ed o he ea men , howe e , he e is no e idence ha only a
ce ain se o pa ame e s can igge posi i e e ec s in he ea men . (Langea d e al., 2017). Ma iule i
(2010) epo s ha e idence sugges s ha he e ec i eness o elec os imula ion depends mo e on he
pa ien ’s na u al cha ac e is ics han on he adjus able pa ame e s o NMES ea men .
In his con ex , he esea ch a ea Heal h 4.0 can play an essen ial ole in cons uc ing ea men s ha
conside indi idual pa icula i ies. Heal h 4.0 is a ecen a ea ha encompasses how heal h is managed
and deli e ed, conside ing he la es echnological ad ances ha combine concep s de i ed om Indus y
4.0, including he In e ne o Things, Big Da a, Machine Lea ning, and mobile apps (Bause e al., 2019).
Combining hese elemen s p o ides he undamen al equi emen s o c ea e pa ien -cen e ed ap-
p oaches ha we e no a ailable be o e. The key ha enables he design o inno a i e ea men s is
he da a collec ed h ough senso s in eg a ed in o wea ables o sma de ices (Lou e al., 2020). Se e al
senso s can acqui e high-quali y da a ha can be clinically ele an , o example, mo ion iden i ica ion
h ough he Ine ial Measu emen Uni senso (IMU), hea a e measu emen h ough he Elec oca -
diog am senso , muscle con ac ion olume h ough he Elec omyog am senso (EMG), blood p essu e
h ough Oxime e senso , among o he s (Jaya aman e al., 2020).
1.2 Resea ch Objec i es
This doc o al p ojec aims o ad ance he p inciples o Heal h 4.0 by designing and de eloping a Clini-
cal Decision Suppo Sys em (CDSS) ailo ed o neu omuscula elec ical s imula ion (NMES) ea men s
in muscle ehabili a ion o he elde ly. The CDSS is concei ed as an in elligen sys em ha suppo s
heal hca e p o essionals in c ea ing pe sonalized ehabili a ion plans ha adap o each pa ien ’s unique
biological esponses. Addi ionally, he sys em acili a es emo e, home-based ea men s h ough a wea -
able de ice, add essing he g owing demand o scalable, accessible, and e ec i e ehabili a ion solu ions
o manage condi ions such as sa copenia and knee os eoa h i is in he aging popula ion.
3

Resea ch Ques ion
How can a Clinical Decision Suppo Sys em be designed o suppo emo e ehabili a ion
wi h elec os imula ion using a wea able de ice?
This main esea ch ques ion guides he in es iga ion in o how a CDSS can be de eloped o e ec i ely
suppo emo e muscle ehabili a ion by in eg a ing eal- ime bio eedback da a om NMES ea men ses-
sions. The goal is o c ea e a sys em capable o pe o ming au oma ed analyses o muscle beha io ,
p o iding pe sonalized ea men ha adjus s o he pa ien ’s condi ion o e ime. Mo eo e , he CDSS
is expec ed o enhance communica ion be ween heal hca e p o essionals and pa ien s, ensu ing a mo e
in eg a ed, da a-d i en, esponsi e ehabili a ion p ocess.
In o de o add ess he cen al esea ch ques ion e ec i ely, i is necessa y o b eak i down in o speci ic
in es iga ions ha ocus on c i ical elemen s o he CDSS design and unc ionali y. Each esea ch ques ion
a ge s a undamen al pa o he sys em, con ibu ing o a s uc u ed and comp ehensi e unde s anding
o i s de elopmen . These ques ions a e c ucial o guiding he design and implemen a ion o he CDSS in
a s uc u ed, e idence-d i en app oach ha acili a es he achie emen ’s eplicabili y.
Speci ic Resea ch Ques ions
1. How can eal- ime bio eedback om a wea able de ice be p ocessed o moni o and assess muscle
a igue du ing emo e ehabili a ion accu a ely?
2. Wha me hods can be implemen ed o secu ely s o e and ansmi sensi i e pa ien da a wi hin he
sys em a chi ec u e?
3. How can elec os imula ion ea men s be dynamically adjus ed in eal- ime based on bio eedback
da a du ing a ea men session?
Gi en on he esea ch ques ions ha pilo his PhD hesis, he doc o al p ojec es ablishes ou main
objec i es o o e coming he echnological, clinical, and e hical challenges in de eloping he CDSS.
Objec i es
• Design a Sys em A chi ec u e: De elop a sys em a chi ec u e ha suppo s he in eg a ion o ad-
anced algo i hms such as machine lea ning and compu a ional simula ions wi hin he con ex o
muscle ehabili a ion, ensu ing he easibili y o emo e ea men s.
4
• Es ablish a Technological In e ace: C ea e an in e ace ha s eamlines he ehabili a ion p ocess
and enhances communica ion be ween pa ien s and heal hca e p o ide s, p omo ing mo e e icien
ea men moni o ing and pe sonalized eedback.
• Ensu e Da a P o ec ion and P i acy: Implemen obus mechanisms o secu e s o age and ans-
mission o sensi i e pa ien da a ha align wi h e hical and legal s anda ds.
• De elop a Reasoning Me hod: Design a easoning me hod capable o dynamically adjus ing elec-
os imula ion pa ame e s based on eal- ime bio eedback.
In addi ion, he p ojec commi s o con ibu e o he WHO p oposals o he decade o heal hy aging in
wo scien i ic o ms. The i s e e s o collabo a ion wi h inno a i e ways o collec , measu e and analyze
da a on he heal h o he elde ly. The second e e s o lowe he cos o new ea men s and expand access,
op ing o al e na i es ha enable online communica ion.
1.3 Resea ch me hodology
This PhD wo k used wo me hodologies. Conside ing he na u e o his esea ch p ojec , which in ol es he
de elopmen o an In o ma ion Sys em, he me hodology adop ed was Design Science Resea ch, speci i-
cally Design Science Resea ch Me hodology o In o ma ion Sys ems (DSRM-IS).
Since i was also equi ed o de elop models and explo e di e en ypes o da a, a second me hodology
ha i s be e was applied. In his con ex he esea ch was guided by he S anda d C oss-Indus y P ocess
o Da a Mining (CRISP-DM).
1.3.1 Design Science Resea ch Me hodology o IS
The DSRM-IS me hodology consis s o a se o p inciples, p ac ices and p ocedu es ha mus be ollowed
o de elop In o ma ion Sys ems and he necessa y analy ical pe spec i es o de elop IS esea ch. DSRM-IS
is an i e a i e p ocess ha includes six main ac i i ies: Iden i y P oblem and Mo i a e, De ine objec i es
o a solu ion, Design and De elopmen , Demons a ion, E alua ion, Communica ion (Pe e s, 2008). The
DSRM-IS ac i i ies a e p esen ed in Figu e 2.
5
Figu e 2: DSRM P ocess Model (Pe e s, 2008).
B ie ly, he compe encies ha mus be pe o med in each ac i i y a e desc ibed below:
•Iden i y P oblem and Mo i a e: The i s ac i i y consis s o de ining he speci ic esea ch
p oblem and jus i y he alue o he solu ion p esen ed.
•De ine objec i es o a solu ion: his ac i i y aims o in e he solu ion objec i es om he
de ini ion o he p oblem and he possible and a ainable knowledge. Objec i es can be quan i a i e
o quali a i e, as long as hey allow eaching a solu ion o he iden i ied p oblem.
•Design and De elopmen : he hi d ac i i y consis s o c ea ing he a i ac . A design esea ch
a i ac can be any designed objec as long as i has he esea ch con ibu ion buil in o he design.
•Demons a ion: In ac i i y ou i is necessa y o demons a e he use o he a i ac sol ing a
leas he p oblem ins ances. The demons a ion can be made in he o m o expe imen a ion,
simula ion, case s udy, p oo , o o he app op ia e ac i i y.
•E alua ion: In his ac i i y, i is necessa y o obse e and measu e how well he a i ac suppo s a
solu ion o he p oblem. Fo his, knowledge o ele an me ics and analysis echniques is equi ed.
A he end o his ac i i y, i is equi ed o decide whe he o i e a e back o ac i i y 3 o y o imp o e
he a i ac ’s e ec i eness.
•Communica ion: The las ac i i y is ela ed o he communica ion o he p oblem and i s im-
po ance, he a i ac , i s use ulness and no el y, he igo o i s design and i s e ec i eness o
esea che s. Fo his p ojec , he communica ion ac i i y was ca ied ou h ough he publica ion
o esea ch a icles and he publica ion o he doc o al disse a ion.
6
1.3.2 C oss-Indus y S anda d P ocess o Da a Mining
The CRISP-DM me hodology de ines a hie a chical p ocess model in e ms o he execu ion o Machine
Lea ning p ojec s. This me hodology can be desc ibed in ou le els o abs ac ion: phase, gene ic ask,
specialized ask and p ocess ins ance. The pu pose o CRISP-DM is o p o ide an o e iew o he li e cycle
o a p ojec di ided in o 6 phases. The di e en phases a e: business unde s anding, da a unde s anding,
da a p epa a ion, modeling, e alua ing and deploymen . This li e cycle is an i e a i e p ocess, whe e he
ou come o he phase will de e mine he nex s ep o be aken (Wi h and Hipp, 2000). Figu e 3 illus a es
he phases and links be ween he phases o he CRISP-DM me hodology.
The compe encies ha mus be pe o med in each phase a e desc ibed below:
•Business unde s anding: The i s phase is conce ned wi h unde s anding he objec i es and
equi emen s om a business pe spec i e. F om he goals, a da a mining p oblem and a p elimina y
p ojec plan a e de ined.
•Da a unde s anding: The second phase aims o know he da a a e an ini ial collec ion. I is
in ended ha his phase can iden i y da a quali y p oblems, disco e ea u es and de ec impo an
subse s o he da a. The e is a close link wi h he i s phase, as o mula ing he da a mining
p oblem equi es a leas some unde s anding o he a ailable da a.
•Da a p epa a ion: In his phase, all ac i i ies o build a inal da ase a e pe o med, including
he selec ion o da ase s, ows and a ibu es, as well as da a ans o ma ion and cleansing. Thus,
he ou pu o his phase is he inpu se o he modeling ools.
•Modeling: The ou h phase consis s o selec ing and applying se e al modeling echniques o
he da a. The e is a close link be ween he hi d phase as some da a issues can be ound du ing
modeling and need o go back o da a p epa a ion.
•E alua ing: A his s age, i is expec ed ha he p ojec has some models ha demons a e high
quali y om he pe spec i e o da a analysis. The main objec i e is o e alua e using me ics ha
jus i y a solu ion in he business objec i es de ined in he i s phase. A he end, a decision mus
be made abou he implemen a ion o he models.
•Deploymen : The ask assigned o he las phase o he CRISP me hodology depends on he
pu pose o which he models we e c ea ed. As an example, o a heo e ical s udy i is p e e able
o c ea e documen a ion o w i e an a icle. Howe e in indus y i may be mo e in e es ing o
implemen he model in p oduc ion.
7
ampli ude o s eng h o he pulses, and pulse wid h indica es he empo al du a ion o each pulse ain
(Lynch and Popo ic, 2008).
Despi e hese undamen al pa ame e s, he e is a a ie y o s udies and ea men s ha go u he ,
modi ying he shape o he wa e used. Among hese modi ica ions a e ec angula , iangula o e en
i egula wa e con igu a ions. Gene ally, he pulses a e biphasic, consis ing o a posi i e pulse ollowed
by a nega i e pulse. Howe e , he e a e a ia ions, including monophasic (a single pulse) o h ee-phase
( h ee pulses in sequence) con igu a ions.
Figu e 6: Schema ic o biphasic elec os imula ion pulses. Adap ed (Lynch and Popo ic, 2008).
Figu e 6 isualizes hese concep s schema ically. The pa ame e s A in Figu e 6 ep esen he size o
an indi idual pulse. Since he schema ic is o a biphasic elec os imula ion, he nega i e componen is
p esen . The nega i e componen size is ep esen ed by pa ame e B, which gene ally has he same alue
as A. Pa ame e C desc ibes he pe iod, being he amoun o ime be ween he beginning o one pulse and
he beginning o he nex . The D pa ame e indica es he pulse wid h, ep esen ing how long he pulse
ain will emain ac i e.
In a e iew p esen ed by Langea d e al. (2017) o he unc ional e ec s ha NMES has in he elde ly,
epo s ha elec ical s imula ion p og ams a e highly inconsis en . The use o symme ical ec angula
pulses om 100 o 400 µs biphasic was he only common poin o all he esea ched s udies. The
equency anged om 20 Hz o high le els close o 100 Hz, wi h adjus men o he ole ance le el o
each pa ien . In addi ion o he a ia ion in s imula ion cha ac e is ics, s udies also show a ia ions in he
empo al composi ion o he ea men . The du a ion o ea men p og ams can a y om 4 o 16 weeks
be o e e o mula ion. The numbe o weekly sessions also a ies, om 2 o 4 imes a week. Las ly, a
single NMES ea men session can ake 9 o 40 minu es o comple e .
14

Despi e his, se e al p og ams we e success ul in hei pu pose. Langea d epo s any NMES aining
be ween 2 and 4 imes a week o a leas 4 weeks, wi h a equency be ween 20 Hz and 70 Hz and wi h
an in ensi y be ween 30 mA and 128 mA seems o be sa e o igge posi i e e ec s o NMES o muscle
ehabili a ion in he elde ly. Fu he mo e, he use o NMES combined wi h olun a y aining can inc ease
he e ec i eness o he ea men .
The e iew p esen ed by Nishida e al. (2016) in es iga ed he po en ial use o elec os imula ion as
an in e en ion o sa copenia in he elde ly. Al hough he e iew emphasizes he a iabili y o he e ec s
o NMES depending on di e en ea men p o ocol, including s imula ion equency and session du a ion,
he li e a u e indica es ha he applica ion o low equency s imuli (less han 20 Hz), sho sessions (less
han 30 minu es) linked o a igue managemen may ha e be e esul s. No ably, p o ocols wi h highe
equencies (60 Hz) also demons a ed posi i e esul s in heal hy elde ly people, indica ing an inc ease
in he diame e s o bo h ype I and ype II muscle ibe s. Thus, i is no ewo hy ha he de e mina ion o
he ideal pa ame e s o ea sa copenia in he elde ly has no ye been de ini i ely es ablished, equi ing
addi ional clinical s udies wi h ou come measu es ocused on imp o ing muscle s eng h.
Imo o e al. (2013) pe o med a andomized clinical ial looking o e idence o he e ec i eness o
NMES in pa ien s wi h KOA. S udies show ha NMES is e ec i e in imp o ing pain, unc ion and ac i i ies
o daily li ing in pa ien s wi h KOA. The ea men p og am adminis e ed anged om 4 o 12 weeks in
du a ion using a equency o 25 o 50 Hz.
2.3 Decision Suppo Sys em
Belonging o he ield o in o ma ion sys ems (IS), Decision Suppo Sys ems (DSS) a e compu e -based
sys ems de eloped o assis in decision making. The e m DSS s a ed be o e he 1970s and i s ocus
was o imp o e he managemen decision-making p ocess and pay-o (A no and Pe an, 2005). Powe
(2002) de ines DSS as
in e ac i e compu e -based sys ems ha help people use communica ions, da a,
documen s, knowledge and compu e models o sol e p oblems and make decisions
and ag ees wi h
A no and Pe an (2005) ha e e o Al e (1980) as one o he pionee s in he DSS ield.
Al e (1980) indica es ha a DSS is iden i ied h ough he ollowing cha ac e is ics:
1. Speci ically designed o acili a e decision-making p ocesses;
2. Design o suppo , a he han au oma e decision-making;
3. Able o espond quickly o he changing needs o decision make s.
15
F om he beginning, much o he heo e ical s udy in DSS is ocussed in unde s anding how he
decision-making p ocess akes place and how g ea manage s make hei decisions. In p ac ice, he
e olu ion o he DSS shows ha he ans o ma ion o hese heo ies in o he e ec i e use o DSS is a ime-
consuming p ocess and only a e o he sub ields (mainly Business In elligence and business analy ics)
ma u ed and we e inco po a ed o sol e majo p oblems in DSS a ea, he opic began o expand (A no
and Pe an, 2014).
A no and Pe an ha e been wo king o yea s looking o o ganize knowledge abou DSS and de ine
new ques ions abou he discipline. Among hei publica ions, a diag am o he DSS a ea genealogy was
buil om 1960 o 2000 (A no and Pe an, 2005) and upda ed un il 2010 (A no and Pe an, 2014).
Shown in Figu e 7 , he pu pose o his diag am is o unde s and how, and which ields o in o ma ion
echnology (IT) ha e come oge he o c ea e a new sub ield in DSS a ea.
Figu e 7: The genealogy o he DSS ield, 1960–2010 (A no and Pe an, 2014).
In sequence, A no and Pe an (2008) indica e he mos ele an sub ields in he DSS, namely:
1. Pe sonal Decision Suppo Sys ems (PDSS): Sys ems de eloped o one manage o a small g oup
o manage s ocused on assis ing hei decision asks.
2. G oup Suppo Sys ems (GSS): DSS buil o imp o e he e iciency o g oup wo k and communica-
ion.
16
3. Nego ia ion Suppo Sys ems (NSS): Sys ems ha deal wi h he op imiza ion o nego ia ions be ween
opposing pa ies.
4. In elligen Decision Suppo Sys ems (IDSS): he applica ion o a i icial in elligence echniques o
decision suppo .
5. Knowledge Managemen -Based DSS (KMDSS): Sys ems ha use he knowledge o indi idual and
o ganiza ional memo y o decision suppo .
6. Da a Wa ehousing (DW): sys ems ha p o ide he la ge-scale da a in as uc u e o decision sup-
po .
7. En e p ise Repo ing and Analysis Sys ems: en e p ise ocused DSS including execu i e in o ma ion
sys ems (EIS), business in elligence (BI), and co po a e pe o mance managemen sys ems (CPM).
Since his PhD p ojec is ocused on building cus om NMES ea men s using da a-d i en, i is na u al
o ocus on In elligen DSS (IDSS), which in ol es DSS ha inco po a es A i icial In elligence like machine
lea ning echniques and mode n op imiza ions. Acco ding o he genealogy shown in Figu e 7, he doc o al
wo k scope alls unde Knowledge Managemen , which encompasses he objec i es o analyzing da a
collec ed h ough a ious sys ems o gene a e knowledge and apply i o imp o e decisions.
2.3.1 Clinical Decision Suppo Sys ems
Clinical Decision Suppo Sys ems (CDSS) a e ools designed o imp o e heal hca e se ices by u ilizing
clinical knowledge, pa ien in o ma ion, senso da a, and diagnos ics. By p o iding knowledge ha is
in elligen ly il e ed and p esen ed a app op ia e imes, CDSS aims o imp o e heal h and heal hca e.
I suppo s clinical decision making, con ibu ing o sys em e iciency, educing e o s and unnecessa y
expenses, and po en ially imp o ing pa ien well-being (Loya e al., 2014).
Since he incep ion o he e m CDSS in 1970, a wide a ie y o ools and in e en ions, bo h au oma ed
and non-au oma ed, ha e been in oduced o imp o e clinical decision making. The sui e o non-au oma ed
ools includes clinical guidance and digi al esou ces, also known as an Elec onic Heal h Reco d (EHR).
This echnological ad ance plays an essen ial ole in he managemen o medical in o ma ion, ansla ing
he adi ional pape pa ien eco d in o a digi alized e sion, p o iding subs an ial imp o emen s in he
e iciency o he heal hca e sys em (Su on e al., 2020).
Ano he ca ego y, known as basic o simple clinical decision suppo sys ems, co e sys ems ocused
on guiding a en ion and ime managemen . Examples o hese sys ems include labo a o y in o ma ion
17
sys ems (LISs) and pha maceu ical in o ma ion sys ems (PISs) Wasylewicz and Scheepe s-Hoeks (2019).
These pla o ms issue ale s and sugges p ac ical ac ions, such as ecommending new medica ions o
wa nings abou possible d ug in e ac ions.
Finally, he la es p og ess is seen in ad anced CDSS, p o iding pe sonalized ecommenda ions o
each pa ien aking in o accoun hei unique cha ac e is ics. These sys ems ep esen a no able ad ance in
he ield o medical diagnosis, being applied in ex ensi e and in ica e domains (Wasylewicz and Scheepe s-
Hoeks, 2019). CDSSs ha e been widely discussed in he li e a u e, helping doc o s choose app op ia e
ea men s and iden i y pa hologies. Cu en ly, CDSS a e al eady used in p ac ice, such as he use o x- ay
images o ecognize cance , inc easing he accu acy o diagnosis and accele a ing he eco e y p ocess.
Figu e 8: The CDSS Componen s (Shoaip e al., 2019).
Ad anced CDSS ypically consis s o h ee essen ial componen s. The i s componen e e s o s o age
s a egies o clinical knowledge and heal h da a. This componen encompasses conce ns abou he
p o ec ion o sensi i e da a, he a ailabili y o esou ces, in e ope abili y, and among o he esponsibili ies.
The second componen is he In e ence Engine, whe e a ce ain deduc i e p ocess is applied o ans o m
he s o ed in o ma ion in o knowledge ele an o a speci ic decision-making p ocess. The hi d componen
is he Use In e ace, which se es as a communica ion b idge be ween he knowledge gene a ed and he
use s. CDSS ypically embed a Use In e ace h ough mobile applica ions, desk op sys ems o websi es,
signi ican ly in luencing he accep ance and use o hese ools.
The co e o he CDSS lies in he In e ence Engine, which employs easoning me hods. These eason-
ing me hods p o ide powe ul ools and echniques o manipula ing knowledge, making in e ences, and
making decisions o e ec i ely sol e p oblems. The easoning p ocess in a medical diagnosis is complex,
as i mus conside se e al ac s, including he pa ien ’s his o y, cu en symp oms, es esul s, he apies
18
ecei ed, and possible alle gies (Shoaip e al., 2019).
The mos common easoning me hods in he medical ield include Rule-Based Reasoning (RBR), Case-
Based Reasoning (CBR), Machine Lea ning-Based Reasoning (MLBR) and Model-Based Reasoning (MBR).
Each o hese me hods o e s dis inc app oaches o in e ing clinical knowledge, om applying logical
ules o lea ning om la ge clinical da a se s. This di e si y e lec s he need o choose easoning me hods
app op ia e o di e en ep esen a ions o knowledge and a eas o applica ion in medical p ac ice.
Acco ding o Papadopoulos e al. (2022), easoning me hods can be ound di ided in o wo main
ca ego ies. The i s , “Knowledge-Based,” includes app oaches such as Wo k- low d i en, RBR and P oba-
bilis ic easoning. The second ca ego y is “Non-Knowledge Based”, in ol ing di e en AI app oaches such
as Machine Lea ning (ML) algo i hms, A i icial Neu al Ne wo ks (NN), Gene ic Algo i hms (GA), Suppo
Vec o Machines (SVM), among o he s. Despi e being labeled as “Non-Knowledge-Based”, hese me hods
do no imply he absence o he use o da a. In eali y, he e m means ha AI algo i hms ypically do no
ely on p e-exis ing knowledge du ing p ocessing.
2.3.2 Rule-Based Reasoning
Rule-Based Reasoning (RBR) is a me hod ha uses logical ules o in e new in o ma ion om exis ing ac s.
In he con ex o CDSS, he applica ion o RBR allows he o mal ep esen a ion o medical knowledge,
main aining machine in e p e abili y (Papadopoulos e al., 2022).
Rule-based sys ems a e ecognized o hei agili y in dealing wi h cons an ly changing ci cums ances
and a e commonly e e ed as he co e o Expe Sys ems. These sys ems comp ise a base o ules, usually
o ganized in o se s, and an in e ence engine ha ope a es based on hese ules. The ep esen a ion o
ules ollows he IF −THEN o m, ma hema ically A=> B, whe e Aa e he condi ions (an eceden )
leading o he ac ions C(consequen ) (Velicko ski, 2016).
A c ucial ea u e is ha al hough ules collabo a e as pa o he p og am, hey mus be s o ed as da a
o acili a e main enance and enable scalabili y. Thus, ules gene ally eside in he “p oduc ion memo y”
o “ ule knowledge base” o a CDSS. The in e ence engine compa es hese ules wi h he “ ac s” (pa ien
da a) in he “wo king memo y” and, when he condi ions a e me , he ules a e igge ed, being able o
gene a e new ac s, wi hd aw in o ma ion, add an ale , o modi y an exis ing s a e.
Se e al expe sys ems ha e al eady been p esen ed in he li e a u e, some o which a e cu en ly in
he public domain and widely used, such as CLIPS 1and D ools 2
1h ps://clips ules.ne
2h ps://d ools.o g
19

. Al hough no speci ically designed o medical da a, hey a e widely used in CDSS applica ions.
These ools, highligh ed by s abili y and suppo om a as communi y, suppo he de elopmen o
expe sys ems and CDSSs (Papadopoulos e al., 2022).
RBR p esen s bene i s such as modula i y, ease o explana ion, and simila i y wi h human hinking and
he cogni i e p ocess. Howe e , i aces challenges when dealing wi h ule excep ions, lack o in o ma ion,
unexpec ed alues in he da a, and some asks can be e y speci ic o a condi ion o p ocess, which makes
i ha d o pa ame e he ac ions (Shoaip e al., 2019).
2.3.3 Case-Based Reasoning
Case-Based Reasoning (CBR) eme ges as an inno a i e app oach o sol ing compu a ional p oblems in-
spi ed by he human cogni i e p ocess. The undamen al p emise o CBR lies in he euse o solu ions
de i ed om pas cases o add ess cu en challenges, p o iding an e icien and adap i e app oach.
The bene i s o CBR become e iden in he con inuous imp o emen o sys em pe o mance, which
becomes mo e e ec i e by emembe ing and adap ing p e ious solu ions o simila p oblems. This me hod
op imizes esolu ion ime and con ibu es o con inuous lea ning o e ime (Shen e al., 2015).
CBR ope a es h ough a cycle o ou undamen al s eps. Fi s ly, he e ie al o ele an pas cases
occu s, whe e case memo y s o es in o ma ion abou p e ious diagnoses and ea men s. Then, he
pas solu ion is adap ed o sui he cu en clinical si ua ion, conside ing indi idual pa ien cha ac e is ics.
Applica ion o he adap ed solu ion is ollowed by e alua ion o esul s, o ming he basis o con inuous
upda ing o he case memo y. Finally, he las s ep is o decide whe he he new case will be main ained
o no (Ve ma, 2022). The main challenge in a CBR sys em is he simila i y ma ching algo i hm o ex ac
om he CB he p e ious cases ’mos simila ’ o he p esen one.
The CBR me hod in eg a es quali a i e and quan i a i e aspec s o s o ing and e ie ing cases, esem-
bling p oblem-sol ing h ough compa ison wi h p e ious indexed cases. Seman ic dis ances om di e en
app oaches, such as s uc u al simila i y algo i hms and s a is ical lea ning, a e gene ally used o ob ain
p e ious cases.
CBR s ands ou o i s in ui i e na u e, lack o knowledge elici a ion o c ea e ules, and con inuous
lea ning h ough use. Howe e , i p esen s challenges such as conside able s o age space equi ed and
he la ge amoun o p ocessing ime needed o ind simila cases. The adap a ion o he cases can also
be a challenging p ocess (Shoaip e al., 2019).
20
2.3.4 Model-Based Reasoning
Model-Based Reasoning (MBR) is a echnique ha employs compu a ional models o simula e complex eal-
wo ld sys ems. These models can be based on heo e ical o empi ical knowledge, and aim o accu a ely
ep oduce he eal beha io o he chosen p ocess. In he con ex o CDSS, MBR is gene ally used o c ea e
simula ions o p edic and diagnose medical condi ions.
MBR is pa icula ly e ec i e in si ua ions whe e a comp ehensi e unde s anding o he unde lying
mechanisms is essen ial. In he medical ield, his in ol es c ea ing models o physiological p ocesses,
disease p og ession, and ea men ou comes. These models enable CDSS o gene a e hypo heses, es
scena ios, and p o ide insigh s ha suppo clinical decision-making. Fo example, models o ca dio as-
cula dynamics can simula e he apeu ic in e en ions and p edic pa ien ou comes (Shoaip e al., 2019).
A no able example o MBR in p ac ice is he use o musculoskele al models, such as OpenSim 3, o
simula e human body dynamics. OpenSim is an open-sou ce ool ha allows he cons uc ion and analysis
o musculoskele al models o he human body. These models can simula e mo emen s and dynamics in
a ious si ua ions, p o iding aluable insigh s o muscle ehabili a ion (Delp e al., 2007a).
MBR suppo s con inuous lea ning and adap a ion. As new da a is collec ed, he models can be up-
da ed o e lec he la es knowledge and ends, ensu ing ha he CDSS emains upda ed and e ec i e.
This adap abili y is c ucial in he medical ield, whe e new ea men s and disco e ies a e cons an ly eme g-
ing. Howe e , building and main aining hese models equi es high-quali y da a and can be complex and
cos ly. Model alida ion is essen ial o ensu e hey accu a ely e lec eal pa ien condi ions (Papadopoulos
e al., 2022).
2.3.5 Machine Lea ning-Based Reasoning
Machine lea ning-Based easoning (MLBR) in ol es using machine lea ning algo i hms o analyze la ge
da a se s and iden i y pa e ns use ul in decision-making. Unlike adi ional me hods ha ely on explici
p og amming, MLR uses algo i hms ha lea n om da a, iden i y pa e ns, and make decisions wi h less
human in e en ion.
MLBR u ilizes la ge da ase s o ain models ha can p edic ou comes, classi y medical condi ions,
and sugges ea men s based on his o ical da a. This app oach is pa icula ly aluable in heal hca e,
whe e as amoun s o pa ien da a, including EHR, medical images, and genomic da a, can be ha nessed
o imp o e clinical decision-making (Su on e al., 2020).
3h ps://sim k.o g/p ojec s/opensim
21
A signi ican ad an age o MLBR is i s abili y o manage complex and non-linea ela ionships wi hin
he da a. Machine lea ning models, such as neu al ne wo ks (NN) and suppo ec o machines (SVM),
a e p o icien a cap u ing hese in ica e pa e ns, which migh be challenging o adi ional s a is ical
me hods. Fo example, deep lea ning algo i hms ha e shown ema kable esul s in image ecogni ion
asks, such as de ec ing abno mali ies in adiog aphs and Magne ic Resonance Imaging (RMI) (Wasylewicz
and Scheepe s-Hoeks, 2019). An exempla y case is Enli ic 4, a company ha de elops CDSS speci ically
designed o analyze medical images like X- ays and MRI. Thei AI-d i en solu ions assis adiologis s by
accu a ely de ec ing anomalies, hus speeding up diagnos ic p ocesses and enhancing clinical ou comes.
This app oach also suppo s pe sonalized medicine by conside ing indi idual pa ien cha ac e is ics
and ailo ing ecommenda ions acco dingly. This pe sonalized app oach enhances ea men e icacy and
educes he isk o ad e se e ec s. Howe e , ob aining and managing he ex ensi e and di e se da a
equi ed o e ec i e MLBR poses a signi ican challenge (Papadopoulos e al., 2022).
4h ps://enli ic.com
22
Chap e 3
S a e o he A
To unde s and he s a e o he a on CDSS o muscula ehabili a ion o he lowe limbs wi h elec ical
s imula ion, an app oach was adop ed ha in eg a es bibliome ic analysis and sys ema ic li e a u e e iew.
The bibliome ic analysis o e s an o e iew o esea ch ends o e ime, while he sys ema ic e iew
ocuses on p o iding a de ailed assessmen o he mos ecen and ele an s udies in he ield.
3.1 Me hodology
The me hodology used o de elop he s a e o he a can be di ided in o wo main s ages. Ini ially,
an i e a i e sea ch p ocess was ca ied ou o iden i y keywo ds ha should be included and excluded
om he sea ch que y. A his s age, he R S udio ool and he Bibliome ix lib a y (A ia and Cuccu ullo,
2017) we e used o analyze he me a in o ma ion o he a icles. In he second s age, i was applied
he me hodology de eloped by Ki chenham and Cha e s (2007) o c ea e a Sys ema ic Li e a u e Re iew
conside ing publica ions om he las 5 yea s, ha is, om he beginning o 2019 o he end o 2023.
3.1.1 Keywo ds Iden i ica ion
The p ope iden i ica ion o keywo ds is a c ucial s ep in conduc ing a li e a u e e iew, especially when
dealing wi h a mul idisciplina y subjec such as he de elopmen o CDSS o muscula ehabili a ion o he
lowe limbs wi h elec ical s imula ion. Nowadays online li e a y da abases p o ide a se o ools ha help
c ea e ad anced sea ches, such as he use o Boolean logic and ield es ic ion, bu o some subjec s i
s ill common o ind a huge numbe o s udies un ela ed o he main opic.
In ou case, he high numbe o un ela ed s udies is mainly due o wo ac o s. The i s ac o is
ela ed o he as li e a u e a ailable on muscula ehabili a ion ea men s wi hou men ioning he use
o any sys em. The second ac o is linked o he high applicabili y o elec ical s imula ion in he human
body, no es ic ed only o he lowe limbs. Fu he mo e, he e m elec ical s imula ion can also be ound
23
Figu e 11: Top 10 big am e ms o e ime.
The analysis o he big ams p esen ed in Figu e 11 e ealed in e es ing pa e ns in he e olu ion o
he mos equen e ms o e ime. The e m “con ol sys em” eme ges as he main and mos consis en
e m o e he yea s, wi h he i s publica ions da ing back o he ea ly 1980s and a con inuous p esence
un il he p esen day. This sugges s ha esea che s ha e consis en ly sough o con ol mo emen since
he beginning o in es iga ions.
Secondly, “muscle a igue” s ands ou , especially om 2010 onwa ds, indica ing a g owing in e es
in his complex opic. This ise sugges s an e olu ion in esea che s’ unde s anding o e ime, expanding
he scope o in es iga ions o conside no only mo emen con ol bu also he e ec s o muscle a igue in
his con ex . This shi in ocus is suppo ed by he e olu ion o he e m “muscle model”, which ollows a
simila bu less in ense end.
The hi d mos used e m, “neu al ne wo k”, demons a es a cons an ise o e he yea s, wi h pe iods
o s agna ion in e spe sed. This pa e n sugges s compu a ional challenges ha may limi he e ec i e use
o neu al ne wo ks a ce ain imes, highligh ing he need o con inued de elopmen in his a ea.
O he no able e ms include “FES Cycling” and “ELECTRIC MOTOR”, which show a simila inc ease
in equency and a e co ela ed. This indica es ha esea ch on muscula ehabili a ion wi h elec ical
s imula ion, using bicycles and exoskele ons, began o gain p ominence in 2015 and con inues o be one
o he main esea ch opics oday.
Con inuing wi h he analysis o he big ams ex ac ed om he abs ac s, a Co-occu ence ne wo k
30

g aph was gene a ed. This g aph was cons uc ed based on he e ms ha co-occu ed mos equen ly in
he abs ac s, e ealing he ela ionships be ween he e ms and iden i ying signi ican hema ic g oupings.
The Lou ain algo i hm was applied o iden i y clus e s, wi h he numbe o nodes se o 100 o ensu e a
comp ehensi e analysis. This algo i hm u ilizes a hie a chical clus e ing me hod ha ecu si ely combines
communi ies in o a single node and pe o ms modula i y clus e ing on condensed g aphs. The goal is o
maximize a modula i y sco e o each communi y.
Modula i y quan i ies he quali y o he assignmen o nodes o communi ies. I assesses how densely
connec ed he nodes a e wi hin a communi y compa ed o how connec ed hey would be in a andom ne -
wo k. This esul s in g oups o e ms ha co-occu mo e equen ly and a e mo e s ongly in e connec ed,
wi h no o e lap Lu e al. (2015).
Figu e 12: Co-occu ence Ne wo k o big am e ms.
Fi e dis inc clus e s we e iden i ied, as illus a ed in Figu e 12, each ep esen ing a speci ic b anch
o s udy applied o sys ems wi h elec ical s imula ion o lowe limbs ehabili a ion. The cen al heme o
each clus e was de e mined empi ically based on he ela ed e ms included in he clus e .
•Sys em De elopmen : This clus e is cen e ed a ound e ms ela ed o he de elopmen o con ol
and s imulus sys ems. Includes wo ds such as “con ol sys em”, “s imula ion pa ame e s”, “closed
loop”, “s imula ion pa e ns”, “mo o con ol”, “powe consump ion”. This sugges s a ocus on
c ea ing and imp o ing sys ems o con olling and adminis e ing elec ical s imula ion.
31
•Con ol Me hod: Se ela ed o s a egies and con ol me hods o pa ame e ize elec ical s im-
ula ion gi en a ce ain objec i e. I includes e ms such as “con ol s a egy”, “con ol scheme”,
“adap i e con ol”, “con ol law”, “sliding mode” and “ acking pe o mance”. This indica es a con-
ce n wi h he de elopmen o adap i e and obus con ol echniques o op imize he e ec i eness
o elec ical s imula ion.
•A i icial In elligence: This clus e ocuses on he use o a i icial in elligence in esea ch ela ed
o muscle ehabili a ion wi h elec ical s imula ion. I includes e ms such as “neu al ne wo k”,
“machine lea ning”, “a i icial neu al”, “musculoskele al model”, and “clinical applica ion”. This
sugges s a g owing end o using machine lea ning algo i hms and ad anced compu a ional models
o imp o e muscle ehabili a ion app oaches.
•Bio eedback: This g oup is ela ed o he physiological esponse and moni o ing o biological sig-
nals du ing he muscula ehabili a ion p ocess. Rele an e ms include “muscle a igue,” “muscle
con ac ion,” “muscle model,” “signal p ocessing,” “ eal ime,” “muscle ac i i y,” and “EMG sig-
nal.” These e ms e lec he in e es in unde s anding muscle a igue and muscle ac i i y du ing
ea men .
•FES Cycling: This clus e ocuses on esea ch explo ing he use o elec ical s imula ion using
bicycles and exoskele ons. Includes e ms such as “ es cycling”, “elec ic mo o ”, “ acking e o ”,
“s abili y analysis”, “cadence acking”, “desi ed cadence”, “cycling sys em” and “hyb id sys em”.
This clus e collabo a es wi h he idea ha his opic is inc easingly on he agenda and can b ing
ele an bene i s o muscula ehabili a ion.
Analyzing big am e ms and iden i ying clus e s p o ides in o ma ion abou he main opics su ound-
ing his Ph.D. esea ch opic and how hey ela e o each o he . The clus e s highligh di e en aspec s
o esea ch, om he de elopmen o con ol sys ems and hei me hods o he applica ion o a i icial
in elligence and he use o bio eedback. The in e sec ion be ween hese clus e s sugges s p omising op-
po uni ies o in e disciplina y esea ch and he in eg a ion o inno a i e app oaches.
Finally, his bibliome ic analysis p o ides a comp ehensi e o e iew o he leading esea ch ends
in CDSS o lowe limb muscle ehabili a ion h ough elec ical s imula ion. This de ailed unde s anding
o ends o e ime can guide u u e in es iga ions and con ibu e o he con inuous ad ancemen o he
a ea, allowing he iden i ica ion o knowledge gaps and oppo uni ies o inno a ion.
32
3.3 Sys ema ic Li e a u e Re iew
This sec ion p esen s an analysis o he 29 selec ed a icles, ollowing he p e iously desc ibed sys ema ic
e iew me hodology (see Figu e 9), co e ing he pe iod om 2019 o 2023 inclusi e. As said abo e, he
sea ch was o ind a icles desc ibing decision suppo sys ems de eloped o assis physicians in elec-
os imula ion ea men . Howe e , none o he analyzed a icles ac ually p esen ed a decision suppo
sys em. Ins ead, all selec ed a icles desc ibe elec os imula ion con ol sys ems o lowe limb ehabili-
a ion, wi hou he e ec i e pa icipa ion o he physician du ing ea men o moni o ing o he pa ien ’s
clinical p og ess h oughou he sessions.
The selec ed a icles mainly add ess h ee main hemes: 17 ocused on FES Cycling, 8 on Exoskele on
and 4 on Knee Con ol. Rega ding he ype o publica ion, 8 a e om con e ences and 21 a e jou nal
a icles. The co esponding au ho s come om a ious coun ies, wi h he Uni ed S a es leading wi h 12
a icles, ollowed by B azil wi h 3. I aly, F ance, I an and China con ibu ed 2 a icles each, while Japan,
Tu key, Malaysia, Se bia, Ge many and he Uni ed Kingdom con ibu ed wi h 1 a icle each. This de ailed
in o ma ion is a ailable in Table 6, along wi h he e e ences and au ho s o he a icles.
ID Au ho (Yea ) Aim Coun y Type
A1 Wa anabe and Tadano (2019) FES Cycling Japan Con e ence
A2 Obuz e al. (2019) Exoskele on Tu key Jou nal
A3 Es ay e al. (2019) FES Cycling USA Con e ence
A4 Ghanba i e al. (2019) FES Cycling USA Jou nal
A5 Rahim e al. (2019) Exoskele on Malaysia Con e ence
A6 Ce one e al. (2019) FES Cycling I aly Con e ence
A7 Sijobe e al. (2019) FES Cycling F ance Jou nal
A8 Duenas e al. (2020) FES Cycling USA Jou nal
A9 Bao e al. (2020) Exoskele on USA Jou nal
A10 Teodo o e al. (2020) FES Cycling B azil Jou nal
A11 Rica e e al. (2020) FES Cycling B azil Con e ence
A12 Zhang e al. (2020b) Knee Con ol China Jou nal
33
A13 Zhang e al. (2020a) Knee Con ol China Con e ence
A14 A colezi e al. (2021) Knee Con ol F ance Jou nal
A15 Isaly e al. (2021) FES Cycling USA Jou nal
A16 Cousin e al. (2021) FES Cycling USA Jou nal
A17 Ald ich and Cousin (2021) FES Cycling USA Jou nal
A18 Molazadeh e al. (2021) Exoskele on USA Jou nal
A19 Rouse e al. (2021) FES Cycling USA Jou nal
A20 Allen e al. (2022) FES Cycling USA Jou nal
A21 Chang e al. (2022) Exoskele on USA Jou nal
A22 Nekouka (2021) Knee Con ol I an Jou nal
A23 Ja a i and E anian (2022) FES Cycling I an Jou nal
A24 Coelho-Magalhães e al. (2022) FES Cycling B azil Jou nal
A25 Popo ić-Maneski and Ma eo (2022) FES Cycling Se bia Jou nal
A26 Lyu e al. (2023) Exoskele on Ge many Jou nal
A27 Wannawas and Faisal (2023) FES Cycling England Con e ence
A28 Sun e al. (2023) Exoskele on USA Jou nal
A29 Fe a i e al. (2023) Exoskele on I aly Con e ence
Table 6: Iden i ica ion o Selec ed A icles.
FES-Cycling
FES-Cycling is a he apeu ic app oach ha combines he use o s a iona y bicycles wi h elec ical s imula-
ion o he muscles o he lowe limbs. Gene ally, he quad iceps, hams ings and glu eal muscle g oups
a e s imula ed in a coo dina ed sequence o gene a e a posi i e c ank cycle. This me hod p esen s im-
p o emen s in he ca dio espi a o y, neu omuscula and skele al sys em, such as inc eased muscle mass,
imp o ed blood ci cula ion and educed bone loss. Howe e , hese imp o emen s may be mi iga ed due
o he possibili y o accele a ed a igue o physical exhaus ion ( an de Schee e al., 2021).
Gene ally, bicycles o icycles ha e a mo o ha also assis s he mo emen , allowing pa ien s wi h
34
li le o no capaci y o olun a y mo emen in hei legs o pe o m he exe cise. This p ac ice is ca ied
ou indoo s, gene ally in a s a iona y sys em wi h a ocus on ehabili a ion, howe e , i is cu en ly possible
o ind compe i ions ha use his sys em and demons a e new echnologies, such as he Bike Race -
Cyba hlon 1.
The main challenge o FES-Cycling is o coo dina e elec ical s imula ion wi h he bicycle mo o o main-
ain an e ec i e and com o able cadence o he pa ien wi hou excessi e a igue. This synch oniza ion is
o en done h ough complex ma hema ical models and he use o senso s o moni o cadence and adjus
he elec ical s imulus and mo o acco dingly. In he Uni ed S a es, he g oup led by esea che Wa en
Dixon was esponsible o all 8 publica ions on FES-Cycling ha we e selec ed o his sys ema ic e iew.
The wo k o his g oup has con ibu ed signi ican ly o he ad ancemen o knowledge abou FES-Cycling
and i s po en ial in lowe limb ehabili a ion.
Exoskele on
Exoskele ons ep esen an impo an inno a ion in he a ea o ehabili a ion, designed o help people
wi h mo o disabili ies egain mobili y and independence. These de ices consis o ex e nal mechanical
s uc u es equipped wi h mo o s, senso s and con ol sys ems, designed o be wo n a ound he join s
o he human body, p o iding suppo and assis ance du ing mo emen . I s main objec i e is o o e
addi ional help o people wi h physical disabili ies, such as spinal co d inju ies, s okes o musculoskele al
diso de s (Anaya e al., 2018).
The selec ed s udies add ess an exoskele on model ha inco po a es he applica ion o elec ical
s imula ion o muscles o aid mo emen . This echnique, also known as hyb id exoskele ons o FES
exoskele ons, aims o p o ide a mo e e ec i e, sa e and obus ehabili a ion he apy. I compensa es o
he lack o s eng h in muscles s imula ed wi h he mo o o p oduce mo emen , simila o FES-cycling.
Hyb id exoskele ons o lowe limb ehabili a ion gene ally ha e mo o s in he hip and knee join s,
ocusing on mo emen s such as li ing he leg while si ing in a chai , ge ing up om a chai o walking.
Howe e , as in he case o FES-cycling, he bigges challenge is coo dina ing he s imula o and he mo o s
oge he o p oduce he necessa y mo emen , wi hou causing excessi e a igue o discom o o he
pa ien . Fo mo e complex mo emen s, such as walking, he challenge can be e en g ea e , equi ing a
mo e sophis ica ed con ol sys em.
1h ps://cyba hlon.e hz.ch/en/e en /disciplines/ es
35

Knee Con ol
Knee con olle s ep esen a simple app oach compa ed o o he ehabili a ion echnologies such as FES
Cycling and exoskele ons. In his app oach, no mo o s a e used, only FES s imula ion is used o con ol he
knee join . Gene ally, elec ical s imula ion is applied only o he quad iceps, aiming a he mo emen o
li ing he leg. Thus, he sys em needs o calcula e he s imula ion pa ame e s o pe o m he mo emen ,
aiming o inc ease he ange o mo emen o sus ain epe i ions wi hou excessi e a igue.
In addi ion o o e ing a mo e adi ional ehabili a ion solu ion, knee con ol can be seen as a p oo o
concep o o he applica ions, such as exoskele ons. This simpli ied app oach allows he implemen a ion
o new echnologies in a basic e sion, which can hen be scaled o mo e ad anced applica ions. An
example is he addi ion o elec omyog aphy senso s o moni o muscle a igue in eal- ime.
3.3.1 RQ_01: Wha a e he main cha ac e is ics o he sys em a chi ec-
u e?
Mos o he e iewed s udies do no p o ide a p ecise desc ip ion o he adop ed sys em a chi ec u e, men-
ioning mul iple de ices and sys ems wi hou clea ly explaining hei in e ac ions. Ins ead, hey ypically
men ion a se ies o de ices connec ed by cables o a compu e , whe e p ocessing is pe o med p ima ily
in MATLAB/Simulink o in-house de eloped so wa e, wi h ew addi ional de ails. These de ices ange
om encode s o measu e bicycle cadence, powe me e s and acquisi ion boa ds, especially in s udies
ela ed o FES-cycling. Fu he mo e, i is no able ha he mos commonly used s imula ion machine was
he Hasomed RehaS im.
Howe e , h ee s udies s and ou o p oposing sys ems ha a e no limi ed o cable connec i i y and
p esen a mo e uni ied app oach.
S udy A6 p oposes a modula sys ems a chi ec u e composed o wo se s o wi eless modules. One
module is esponsible o signal acquisi ion, while he o he ep esen s a s imula o machine. These
modules communica e ia WiFi o a mul ipla o m so wa e de eloped wi h he Q amewo k in C++.
While his app oach allows lexibili y as he modules can be apply o mul iple egions, i also b ings wi h i
signi ican echnological challenges o he con ol sys em.
S udy A11 desc ibes a modula a chi ec u e based on he Robo Ope a ing Sys em (ROS), ep esen ing
a mo e s uc u ed app oach compa ed o o he s udies. The modula pa o he a chi ec u e allows he
acquisi ion o signals and he use o di e en ypes o senso s. In he case analyzed, i was implemen ed
wi h an IMU mo ion senso . The es o he a chi ec u e consis s o a po able compu e and a s imula o ,
36
bo h connec ed by cables.
S udy A24 dese es a en ion as i p o ides a mo e de ailed desc ip ion o i s sys em a chi ec u e.
Using an elec onic boa d, his s udy es ablishes communica ion wi h an And oid mobile applica ion ia
Blue oo h. Howe e , i emains unclea how he da a is subsequen ly ans e ed o o line p ocessing.
Addi ionally, he s udy inco po a es a powe me e ha ansmi s da a ia Blue oo h o a compu e , bu
wi hou de ails abou i s comple e in eg a ion in o he o e all sys em.
In summa y, while mos s udies do no p o ide a de ailed desc ip ion o he sys em a chi ec u es
used, he e is a end owa ds wi ed connec i i y o a compu e o da a p ocessing, wi h some men ion o
wi eless communica ion o da a ansmission. Mo eo e , he lack o de ails on how colle ed da a is s o ed
o line and in eg a ed in o he global sys em emains a common gap in he s udies e iewed.
3.3.2 RQ_02: Wha da a is used and how is i acqui ed?
A di e se se o senso s and ools we e used in he selec ed a icles o collec c ucial da a o con olling
elec os imula ion. In some cases addi ional da a o alida ion we e also collec ed ha we e no used in
he con ol. Table 7 seeks o summa ize he in o ma ion ound.
ID Da a Used Acquisi ion Tools
A1
C ank angle (EMG da a we e used o de ined
s imula ion pa e ns in p e iously s udy)
2 IMUs senso s
A2 Knee angle Encode
A3 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A4 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A5 Knee angle ( o alida ion only) Came as
A6 Knee angle Elec onic goniome e
A7 Thigh angle 2 IMU senso s
A8 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A9 Knee angle Encode
A10 C ank angle, eloci y and accele a ion Elec onic goniome e and IMU senso
A11 C ank angle 1 IMU senso
37
A12 Knee angle Elec onic goniome e
A13 Ankle and knee angle and ec us emo is EMG Angle senso and myoele onic senso
A14 Knee angle Elec onic goniome e and IMU senso
A15 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A16 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A17 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A18 Knee and hip angle 2 Encode s
A19 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A20 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A21
Knee and hip angle, leg o ce and angula
displacemen
2 Encode s, o ce senso and eadmill
A22 Knee angle 2 IMUs senso s
A23 C ank angle Encode
A24 C ank angle, eloci y and accele a ion and powe Encode and powe me e
A25 C ank angle Encode
A26 Body mo ion cap u e, mo o o que To que senso and came as
A27 C ank angle Encode
A28 Knee angle, o ce, mo o o que Encode , dynamome e and o que senso
A29 Mo o o que, knee angle, ec us emo is EMG Encode and EMG senso
Table 7: Da a and Acquisi ion Tools Used in he Selec ed S udies.
As can be seen om Table 7, mos s udies ely on join angles o guide elec os imula ion con ol, wi h
he knee join being he mos common among hem, al hough he high, ankle, and hip angles we e also
ound o be used. Fo FES-Cycling s udies, excep o a icles A6 and A7, all s udies use c ank angle. I
can also be obse ed ha all s udies we e using some ype o join angle.
Gi en ha he majo i y o s udies ocus on FES-Cycling and employ c ank angle, he mos common
acquisi ion ool is he encode , which also measu es bike speed and accele a ion. Second, i is no iceable
38
ha IMU senso s a e inc easingly gaining space in his esea ch a ea, as well as elec onic goniome e s,
which can be seen as p o o ypes o IMUs in his use case. This is due o he ac ha many o hese IMU
senso s, despi e ha ing accele ome e s and gy oscopes inside, a e used o con e hei da a in o angles,
a unc ion ul illed wi h p ecision by elec onic goniome e s.
Finally, EMG senso s we e also seen, which a e impo an o assessing muscle ac i i y and a e ela ed
o a igue. Howe e , only 2 a icles use his da a, bo h collec ing om he ec us emo is muscle.
3.3.3 RQ_03: How is sensi i e da a p o ec ed?
None o he selec ed s udies add essed o a leas commen ed on da a s o age o sensi i e da a p o ec ion,
e ealing a signi ican and c i ical gap in he ield o esea ch. E en he a icles ha men ion he use o
o line da a o aining machine lea ning algo i hms do no discuss de ails on how his da a is s o ed o
made a ailable, lea ing an impo an ques ion unanswe ed.
This lack o a en ion o da a secu i y highligh s he imma u i y o sys ems ocused on s imula ion
con ol in e ms o in as uc u e and sys em a chi ec u e. I also sugges s a lack o in e ope abili y be ween
hese sys ems and a e y low comme cial a ailabili y. These issues a e c ucial no only o academic
esea ch bu also o he p ac ical implemen a ion o hese echnologies in clinical en i onmen s.
3.3.4 RQ_04: Wha a e he easoning me hods used in hese sys ems?
As p e iously men ioned, he selec ed s udies do no explici ly desc ibe a decision suppo sys em bu
a he con ol sys ems o s imula ion. Howe e , each sys em inco po a es a ype o easoning me hod
o adjus s imula ion pa ame e s, al hough hese me hods a e no explici ly desc ibed using hese e ms.
Figu e 13 p esen s a cha quan i ying he easoning me hods ound.
The majo i y o he selec ed a icles u ilize he model-based me hod as hei easoning me hod, em-
ploying ma hema ical models o simula e scena io beha io and de e mine he necessa y s imula ion
pa ame e s o main ain mo emen as planned. This me hod is ypically applied in a closed-loop con ol
sys em, whe e encode and powe me e da a a e used o eed he model.
Six s udies employed he ule-based easoning me hod o con ol s imula ion. This me hod is simple
han o he s, ypically ac i a ing and deac i a ing s imula ion based on a a iable o se o a iables. A sim-
ple example is con olling he iming o muscle s imula ion ac i a ion and deac i a ion based on p ede ined
c ank angles in FES-cycling.
The emaining six s udies u ilized machine lea ning algo i hms o con ol s imula ion. Some s udies
used hese algo i hms only o ac i a e and deac i a e s imula ion, wi hou con olling pa ame e s such
39
pe o med, ensu ing he de ice is op imized o he ea men en i onmen .
The mobile app will guide he pa ien h ough a use - iendly in e ace, o e ing ins uc ions and session
p og ess in o ma ion. The applica ion will eplica e he s imula ion mode pe o med in he clinic and will
also allow dynamic adjus men s du ing he session. I he sys em de ec s changes in muscle a igue
me ics ha exceed he limi s es ablished by he doc o , he app will au oma ically adjus he ea men .
A he end o he session, he pa ien will comple e a b ie ques ionnai e abou he expe ience and
hei physical condi ion. All collec ed da a, including senso in o ma ion, s imula ion pa ame e s, and
ques ionnai e esponses, will be uploaded o he cloud. The physician is immedia ely no i ied ha he
session has been comple ed.
This app oach allows he physician o access bio eedback da a online, enabling he o she o assess
he pa ien ’s esponse o ea men and c ea e a de ailed e alua ion epo . I necessa y, he physician
can modi y he session se ings and schedule a new session o he pa ien . This con inuous ea men
cycle epea s un il he ehabili a ion p ocess is ully comple ed.
4.1.2 Sys em A chi ec u e
The p oposed sys em a chi ec u e o muscula ehabili a ion ea men s using elec ical s imula ion was
designed o inco po a e he echnological ad ances o heal h 4.0. This includes he use o mic ose ices,
cloud compu ing, in e ne o hings, mobile applica ions, among o he s. This app oach aims o o e a
comp ehensi e and e ec i e solu ion, capable o mee ing he equi emen s desc ibed in he ea men
scena ios. A de ailed o e iew o sys ems ha inspi ed his design and he ini ial implemen a ion is p o-
ided in he a icle (F anco e al., 2022b).
I is impo an o emphasize ha he in o ma ion o be s o ed is highly con iden ial. In addi ion o
pe sonal in o ma ion, such as he names and add esses o sys em use s, heal h- ela ed da a such as
ac i e diseases, clinical obse a ions, and mo e will also be managed and p ocessed. To ensu e he
secu i y o his sensi i e da a, a s a egy has been de eloped o adhe e o bes p ac ices o da a p o ec ion,
segmen ing s o ed da a and mi iga ing po en ial cybe a acks.
The p oposed sys em is based on a h ee-laye a chi ec u e, using he sma phone as an in e media y
de ice. Despi e his, we spli he in e io o he secu e cloud laye in o wo laye s, making i possible o
conside his a chi ec u e as ou laye s. This di ision gua an ees ha only he componen s ha need
ex e nal communica ion a e a ailable wi h public IP. Componen s ha need o be mo e secu e, such as
da abases and he logging se ice, a e sepa a ed and only a ailable on he p i a e ne wo k. The sys em
a chi ec u e shown in Figu e 16 has se en main componen s. The cha ac e is ics and unc ions o each
46

componen a e as ollows:
Figu e 16: Sys em A chi ec u e.
Wea able De ice: Componen ha includes a sys em, senso s o bio eedback da a acquisi ion and
he elec ical s imula o . This componen is he ac ua o de ice ha pe o m he emo e ehabili a ion and
communica e wi h mobile app.
Mobile Applica ion: In e media e applica ion be ween he wea able de ice and he cloud. I s main
unc ions will be: ecei e he s imula ion p o ocol om he cloud and p o iding i o he wea able de ice;
o ganize he da a ecei ed by he ea men session and send i o he cloud; p o ide and collec he pa ien
in o ma ion.
Sys em Panel: Adminis a i e panel o p o essionals o manage pa ien s and ea men s. On his
websi e, he physician will be able o isualize and adap he ea men plan. In addi ion, i will be possible
o iew he bio eedback acqui ed by he wea able de ice and espond o messages om pa ien s.
Clinical Se ice: Se ice ha s o es and p ocesses he clinical da a o pa ien s. This in o ma ion
a e: pa ien cha ac e is ics and medical his o y, clinical epo s and bio eedback collec ed by wea able
de ice du ing sessions.
Managemen API: A module capable o managing and p o iding he necessa y esou ces o he
ope a ion o he mobile app and he adminis a i e panel. Fo example, managing pa ien s, p o essionals,
messages, equipmen , and o he s.
Single Sign-On (SSO): Componen esponsible o gene a ing he access and e esh okens. The
47
pu pose o SSO is o p o ide a single poin o au hen ica ion wi hin he a chi ec u e, hus ensu ing ha
access o he mul iple se ices is secu e and anspa en .
Log Se ice: Componen equi ed o main ain log in eg i y ac oss he en i e a chi ec u e. The in en
is ha all componen s p o ide logs pe iodically o his se ice, c ea ing a unique audi able and secu e
access poin .
The communica ion be ween he componen s is a ied, depending on hei pu pose wi hin he a chi-
ec u e. To op imize ba e y li e, he wea able de ice communica es wi h he mobile app ia Blue oo h
Low Ene gy. Meanwhile, in e ac ions be ween he secu e cloud, he mobile app, and he sys em panel
adhe e o he HTTP p o ocol wi h Secu e Socke s Laye (SSL) ce i ica e and JSON Web Token (JWT). The
Clinical Se ice, SSO, and Managemen API sys ems communica e wi h he Log Se ice also using HTTP
wi h SSL, bu wi h public and p i a e key au hen ica ion.
4.1.3 Sensi i e Da a P o ec ion
The a chi ec u e is based on mic o se ices. This app oach enables da a o be spli and s o ed in di e en
pa s acco ding o he ope a ional needs o each se ice. Thus, he pa ien s’ pe sonal da a, such as
add esses, phone numbe s, esponsible doc o , among o he s, a e s o ed unde he Managemen API.
The clinical da a, such as ea men plans, ecommended ea men s, disease his o y, diagnoses, among
o he s, a e s o ed in he Clinical Se ice.
To main ain da a p i acy, he iden i ie s o each en i y mus be di e en in each da abase. This ensu es
ha i bo h da abase sys ems a e hacked, he a acke would no be able o link which diagnosis belongs
o which pa ien , minimizing he impac .
To main ain he da a ela ionships be ween he sys ems, he pseudonymiza ion echnique was applied.
This echnique allows he iden i y o subjec s o be hidden om any ou sou ced se ices by assigning
pseudo-iden i ie s. Thus, he hi d-pa y se ices a e no able o ela e he pseudonyms o he eal iden i i-
ca ions, so hey do no ecognize he au ho o he da a p o ided. I a clien applica ion needs o ecognize
he iden i y o he da a, a g an o access is eques ed. I he eques is accep ed, he igge ed se ices
e u n he da a ha can be que ied by he clien app ha eques ed i (as can be seen in Figu e 17).
48
Figu e 17: Rep esen a ion o pseudonymiza ion echnique.
To ensu e he in eg i y o eques s JSON Web Signa u e (JWS) was used du ing communica ion. Each
se ice was con igu ed wi h asymme ic keys and has access o public keys o o he se ices. In his way,
he da a in a message is signed by a sending se ice using i s p i a e key, so o he se ices can e i y
he in o ma ion using he public key o he sending se ice. Figu e 18 ep esen s a simpli ied p ocess o
signing and checking con en using asymme ic keys.
Figu e 18: Signa u e and con en e i ica ion p ocess using asymme ic keys.
Using signed messages p e en epudia ion o se ices. The public key can be used o e eal he eal
sende o a message, and modi ica ions du ing ansmission (in eg i y). All mic o se ices ha e a speci ic
key, and he check p ocess only wo ks wi h he co ec public key. The us o in o ma ion is es ablished
when he gene a o o a message is ecognized as an in e nal se ice.
The ela ionship o he Clinical Se ice and he Managemen API can be unde s ood as ou sou ced
se ices o each o he . Thus, when a mobile app wan s o access ei he o he wo se ices, i mus send
in he heade o each eques he access oken p o ided by SSO se ice. Inside his oken, h ee main
pieces o in o ma ion will be s o ed, he ole he use belongs o, and wo enc yp ed packe s. The i s
packe is enc yp ed by he public key o he Clinical Se ice wi h he pseudo-anonymized iden i ie o he
Clinical Se ice. The second packe ollows he same logic bu enc yp ed and pseudo anonymized o he
Managemen API.
When a eques is made, each se ice can open he packe wi h i s p i a e keys and sea ch in hei da a
49
o in o ma ion ha co ela es wi h he sen iden i ie s. This ensu es ha he only ones who could disco e
he ela ionship be ween he wo da abases a e he owne s o he in o ma ion, ha is, he ones who should
eally ha e access. In his Mas e ’s wo k, de eloped in conjunc ion wi h his doc o al p ojec , (Sil a, 2021)
has explo ed he in o ma ion secu i y and scalabili y app oaches o be used in his a chi ec u e.
4.1.4 Da abase Modeling
To maximize he e iciency o each module o he sys em a chi ec u e, di e en Da abase Managemen
Sys ems (DBMS) we e employed. Taking in o accoun he dis inc cha ac e is ics o he s o ed da a, Pos -
g es1(SQL) was chosen o he Managemen API da abase and he Single Sign-On (SSO) se ice, while
MongoDB 2(NoSQL) was selec ed o he Clinical Se ice da abase.
To ensu e sys em in e ope abili y, a se o elemen s was adap ed by adding a ibu es and modi ying
nomencla u e o acili a e he exchange o in o ma ion wi h sys ems ha ollow he HL7 FHIR3heal h
da a exchange s anda d. HL7 FHIR (Fas Heal hca e In e ope abili y Resou ces) is a c ucial s anda d ha
p omo es e icien clinical da a in eg a ion be ween di e en sys ems, mode nizing heal hca e p ocesses.
Figu e 19: En i y-Rela ionship (ER) Model o he Managemen API Da abase.
1h ps://pos g esql.o g
2h ps://mongodb.com
3h ps:// hi .o g
50
Figu e 19 illus a es he en i y- ela ionship model used o design he necessa y da a s uc u e o he
Managemen API. This da abase was se up o handle pe sonal da a, including in o ma ion on p ac i ione s,
pa ien s, appoin men s, and en al de ices. The En i y People was c ea ed o agg ega e ela ed a ibu es,
se ing as he p ima y en i y in he hie a chy. Each eco d in he En i y People mus ha e an SSO_ID o
link he use o he SSO se ice and may ha e mul iple add esses and con ac s.
En i y Appoin men s o e he s a us, s a ime, and end ime, and can include in o ma ion abou
he loca ion. Howe e , hey do no s o e da a ela ed o he pa hology being ea ed. Al hough i has no
ye been implemen ed, he sys em is also designed o suppo he bo owing o wea able de ices, wi h he
expec a ion ha hese de ices may come in di e en e sions.
The NoSQL da a model equi ed o he unc ioning o he clinical se ice is depic ed in Figu e 20.
This model s o es da a ela ed o pa ien heal h and applied ea men s. Since MongoDB suppo s embed-
ded documen s, he da abase consis s in only ou main collec ions: Condi ions, Obse a ions,
Ca e-Plans and Sessions. The ano he “collec ions” ha appea in Figu e 20 a e embedded doc-
umen s o he ou main collec ions men ioned abo e.
The collec ion Condi ions pe ains o speci ic illnesses and hei ini ial s ages. The collec ion
Obse a ions s o es upda ed in o ma ion on he pa ien ’s condi ion, wi h a new obse a ion documen
gene a ed o each appoin men .
The collec ion Sessions collec ion, he mos complex in he da a model, is esponsible o s o ing
s imula ion pa ame e s, bio eedback da a, and calib a ion da a, among o he in o ma ion. Addi ionally,
he e is a con ac ions documen ha eco ds me ics ela ed o muscle a igue du ing ea men sessions,
which a e calcula ed by he mobile app. All aw senso da a is s o ed comp essed in bina y o m.
Fo home-based ea men sessions, a ca e-plan documen needs o be gene a ed, speci ying he s a
da e and objec i es. In he da a model, home sessions ex end he s anda d clinic sessions by inco po a ing
addi ional in o ma ion and ea u es gi ing ise o he Ca e-Plans collec ion. The design includes an
in elligen ules sys em ha adap s sessions in eal- ime. The au oma ic session se ings documen s o es
igge s o s imula ion adjus men s p oposed by he physician, ensu ing pe sonalized and esponsi e
ea men .
This app oach ensu es ha he sys em can handle wi h he complexi y and sensi i i y o pa ien s’
heal h da a, o e ing a secu e, e icien , and scalable da a model o muscle ehabili a ion ea men s.
The inal da abase model was epea edly e ac o ed h oughou he de elopmen p ocess o inco po a e
new echnologies.
51

Figu e 20: NoSQL da a model o he clinical se ice.
52
4.2 P o o yping
4.2.1 Cloud In as uc u e
The cloud in as uc u e de eloped dis ibu ed he componen s o he sys em a chi ec u e ac oss se en
i ual machines (VMs) based on Ubun u Se e 20, hos ed wi hin he Resea ch Cen e in Digi aliza ion
and In elligen Robo ics (CeDRI) clus e a he Ins i u o Poli écnico de B agança (IPB) building. Ou o he
se en VMs, only h ee ha e IPs on he public ne wo k, each dedica ed o a di e en module o he sys em
being Clinical Se ice, Managemen API, and SSO.
The Clinical Se ice and Managemen API we e de eloped om sc a ch using Py hon4p og amming
language and Flask5 amewo k, ecognized o i s lexibili y and ease o implemen a ion. Bo h se ices
we e designed ollowing he Model-View-Con olle (MVC) design pa e n (T yg e, 2003), ensu ing clea
sepa a ion o conce ns o enhanced main ainabili y and scalabili y. Fu he mo e, he se ice endpoin s
we e documen ed using Swagge 6, a ool ha o e s in e ac i e and dynamic documen a ion o APIs,
simpli ying unde s anding and used by de elope s o s anda dizing inpu s and ou pu s.
Fo p o iding he SSO se ice, he open-sou ce solu ion Keycloak7was adop ed. Ins alled ia Docke 8,
his choice o e s a use - iendly in e ace o implemen ing secu e and complex au hen ica ion s a egies,
such as wo- ac o au hen ica ion and OAu h 29.
The emaining ou VMs we e alloca ed in an in e nal ne wo k, allowing access only o he h ee se ices
wi h public IPs. Thus, one VM was o he MongoDB da abase, and wo we e o he Pos g es da abase o
supply Managemen API and SSO. The las VM is des ined o Logs se ice and i s da abase we e ins alled
on he same VM, as his module does no equi e a public IP. This con igu a ion aims o enhance he
secu i y o access o each componen o he in as uc u e.
4.2.2 Wea able De ice
Suppo ed by he Po ugal 2020 p og am and in e na ional pa ne ships, he NanoS im p ojec 10 uni ed a
conso ium ocused on ad ancing emo e ehabili a ion. The eam included six Po uguese o ganiza ions:
Ins i u o Poli écnico de B agança, Uni e sidade do Minho, NATG, INOVA+, Impe us, and TeandM, and he
4h ps://www.py hon.o g
5h ps:// lask.palle sp ojec s.com
6h ps://swagge .io
7h ps://www.keycloak.o g
8h ps://docke .com
9h ps://oau h.ne /2
10 h ps://nanos im.p
53
Uni e si y o Texas, USA. In his documen , he e m
wea able de ice
e e s p ima ily o he ha dwa e
componen de eloped du ing he NanoS im p ojec . The e m
wea able sys em
is used o e e o he
sys em unning inside he
wea able de ice
ha is pa o he p oposed sys em a chi ec u e.
Figu e 21 illus a es he design o he wea able de ice, which in eg a es essen ial componen s chosen
o hei low cos and abili y o p o ide e ec i e muscle ehabili a ion ea men .
Figu e 21: Wea able De ice Design.
The p ima y componen is he low-cos mic ocon olle , a small compu e on a single chip ha in e-
g a es a p ocesso , memo y, and pe iphe als. This mic ocon olle manages all he senso s and commu-
nica es wi h he mobile applica ion. The ESP32 mic ocon olle , speci ically he WROOM-32E11 e sion,
known o i s e sa ili y, cos -e ec i eness, and buil -in Wi-Fi and Blue oo h ea u es, was u ilized. The
ESP32 is ideal o wea able de ices and IoT applica ions due o i s low powe consump ion, suppo o
mul iple communica ion in e aces, and e icien p ocessing capabili ies, such as Blue oo h Low Ene gy
4.212. I s key speci ica ions include a dual-co e 32-bi Tensilica X ensa LX6 CPU ope a ing a up o 240
MHz, 448 KB o SRAM, 520 KB o ROM, and pe iphe al in e aces wi h a 12-bi ADC.
The wea able de ice includes wo ypes o senso s: Elec omyog aphy (EMG) senso s and Ine ial
Measu emen Uni (IMU) senso s. The EMG senso measu es he elec ical ac i i y o muscles h ough
elec odes posi ioned on he skin. The measu ed elec ical ac i i y is e u ned as an analog signal wi h a
equency band o in e es be ween 0 and 500 Hz, which is hen con e ed o digi al by he mic ocon olle .
The signal condi ioning ci cui , essen ial o p ocessing EMG signals, was de eloped om sc a ch and mo e
11 h ps://www.esp essi .com/en/p oduc s/socs/esp32
12 h ps://www.blue oo h.com/speci ica ions/specs/co e-speci ica ion-amended-4-2
54
de ails can be ound in a icle (Ses em. e al., 2022).
IMU senso s measu e accele a ion and o a ion ac oss mul iple axes, ypically consis ing o a 3-axis
accele ome e and a 3-axis gy oscope, known as a 6-axis IMU. The accele ome e measu es linea accel-
e a ion, while he gy oscope measu es angula eloci y. In he ield o ehabili a ion, his da a is gene ally
used o moni o mo emen s and o ien a ions in h ee-dimensional space. The MPU-605013 IMU senso
was chosen o i s eliabili y and e iciency in I2C communica ion14.
An essen ial componen in he p oposed applica ion is he elec ical s imula ion ci cui , which gen-
e a es elec ical impulses o s imula e ne es con olling he muscles, p omo ing con ac ions and mo e-
men s. This componen is he ac ua o o es o ing o imp o ing muscle unc ion in pa ien s wi h inju ies
o diseases in ou con ex . The s imula ion ci cui was also de eloped om sc a ch and a Mas e ’s hesis
(Gonçal es, 2023) was de eloped es ing a ious low-cos componen s o eplica e he elec ical s imula-
ion o comme cial de ices.
All men ioned componen s a e in eg a ed in o a P in ed Ci cui Boa d (PCB), as illus a ed in Figu e
22. The PCB is enclosed wi hin a 3D-p in ed case wi h dimensions o 10 cm (leng h) × 12.5 cm (wid h)
× 3.5 cm (heigh ), excluding he USB cable. This case also houses he sys em’s ba e y, which p o ides
an ope a ional li e o app oxima ely 6 hou s pe cha ge. The po able design allows he wea able de ice
o be con enien ly cha ged ia USB and easily used in a ious en i onmen s. Addi ionally, he o al cos o
p oduc ion o each uni emains below 100 eu os, making i a cos -e ec i e solu ion.
Figu e 22: Wea able De ice Componen s in PCB.
13 h ps://in ensense. dk.com/wp-con en /uploads/2015/02/MPU-6000-Da ashee 1.pd
14 h ps://docs.a duino.cc/lea n/communica ion/wi e
55
such as esea ch ins i u ions, clinics and hospi als. The main ea u es a ailable o he adminis a o a e:
1. Manage O ganiza ions: The adminis a o can c ea e, edi and dele e o ganiza ions. To egis e an
o ganiza ion i is necessa y o p o ide he add ess and con ac de ails o someone esponsible o
he ins i u ion.
2. Manage Physicians: The adminis a o can add, edi , and emo e physicians by associa ing hem
wi h speci ic o ganiza ions. The physician’s egis a ion includes c eden ials, quali ica ions, con ac
and add ess.
Physician: The physician is esponsible o managing pa ien s, appoin men s and ca e plans. The
unc ionali ies a ailable o he physician includes managing:
1. Pa ien s: Regis e and main ain pa ien in o ma ion, including pe sonal da a, c eden ials, con ac
and add ess.
2. Appoin men s: Schedule, edi , and cancel appoin men s, allowing e icien managemen o he
appoin men calenda .
2.1 Obse a ions: Add obse a ions and no es o appoin men s, eco ding impo an in o ma ion
abou he pa ien ’s condi ion and ea men p og ess.
3. Medical Reco ds: Main ain pa ien s’ medical eco ds, including:
3.1 Pa hologies: Reco d and upda e pa ien pa hologies, documen ing ele an medical condi-
ions.
3.2 T ea men Plans: De ine and moni o ea men plans, speci ying he apeu ic goals and me h-
ods o be used.
3.3 T ea men Sessions: Reco d and ack sessions pe o med, including iewing bio eedback
da a eco ded by he wea able sys em, eco ding an au oma ic ea men session and simu-
la ing a igue le els based on a p e iously pe o med session.
The Use Case diag am in Figu e 24 desc ibes in de ail he in e ac ions o he ac o s, enhancing he
ac ions each ole can pe o m. Be o e accessing hese unc ionali ies, use s mus i s log in using a SSO
sys em wi h OAu h 2.0 au hen ica ion, ensu ing secu e access o he sys em.
The de elopmen o he web-based adminis a i e panel enhances use con enience, allowing access
o he sys em a any ime and om any loca ion. This app oach p o ides lexibili y and acili a es he
62

e icien and accessible managemen o ea men s and clinical esea ch. Addi ionally, he managemen
o sessions enables de ailed analysis o he bio eedback da a collec ed by he wea able sys em, as well as
he applied s imula ion pa ame e s, o e ing aluable insigh s o moni o ing and op imizing ea men s.
Figu e 24: Use Case Diag am o Sys em Panel.
63
Chap e 5
In elligen Fea u es o T ea men Cus omiza ion
In his chap e , i e essen ial ea u es a e p esen ed ha o m he ounda ion o a CDSS designed o
make muscle ehabili a ion ea men s mo e esponsi e o bio eedback da a collec ed om pa ien s du ing
elec os imula ion sessions. These ea u es no only enhance he in o ma iza ion o he ea men bu
also signi ican ly imp o e esponse ime, making he sys em mo e in elligen and adap able compa ed o
adi ional ehabili a ion ea men s using con en ional ac ua o s.
The i e ea u es can be di ided in o h ee main ca ego ies. The i s ca ego y encompasses mo emen
ecogni ion and a igue pa ame e iza ion, desc ibing he me hods used o ans o m aw da a in o aluable
in o ma ion abou he physiological s a e o he muscle unde ea men . The second ca ego y ocuses on
he implemen ed elec os imula ion ea men modes and how hey can be adap ed based on bio eedback.
Finally, he las sec ion ou lines he u ili ies de eloped o enhance he heal hca e p o essional’s expe ience
when using he CDSS.
5.1 Mo emen Recogni ion
Unde s anding he mo emen s pe o med by he pa ien du ing ea men sessions is c ucial o op imizing
ehabili a ion p o ocols. Ini ially, he NanoS im p ojec planned o use a single EMG senso o ex ac he
pa ame e s ela ed o muscle s a e and e o . Howe e , his app oach may be limi ed, as wi h jus he
EMG signal, i would be challenging o di e en ia e he ampli ude o mo emen s, making i di icul o
assess he e ec i eness o he ea men o e ime.
To add ess his limi a ion, a esea ch was conduc ed o iden i y he mos sui able senso s and da a
p ocessing me hods needed o accu a ely cap u e leg mo emen s du ing ea men . A li e a u e e iew on
biomechanical app oaches o classi ying knee os eoa h i is F anco e al. (2021) highligh ed he ex ensi e
use o IMU senso s. These senso s a e o en used o measu e he angle o he knee join h ough wo IMU
senso s, one a ached o he high and he o he o he shin.
64
Guided by hese indings, he decision was made o in eg a e wo IMU senso s in o he wea able sys em
o moni o he knee join angle in eal ime. The e o e, his session desc ibes how aw da a ex ac ed om
IMU senso s is used o ecognize and classi y mo emen s pe o med du ing ehabili a ion ea men .
5.1.1 Ma hema ical Model
The ma hema ical model used o compu e knee angle o he subjec is based on he heo y o mul ibody
sys em dynamics, as p esen ed by (Olinski e al., 2017). In his app oach, he o ien a ion o a body in
space is gi en by he o ien a ion o a local ame a ached o he body wi h espec o a e e ence coo dina e
sys em as illus a ed in Figu e 25.
Figu e 25: IMU senso wi h he e e ence coo dina e sys em.
Conside ing he pa icula case in which he o ien a ion changes occu in a speci ic plane, as p esen ed
in Figu e 26a, he mapping o ames wi h espec o a e e ence ame can be ep esen ed by a single
o a ion om a e e ence o ano he (Figu e 26b).
Figu e 26: (a) F ame ep esen a ion ela i e o a e e ence. (b) Linea mapping o a ame o ano he .
65
To map a ame wo h in espec o ano he in he case o o a ions in he YZ-plane o an angle Φj
a ound he x-axis, as p esen ed in Figu e 26, he Eule angles a e de e mined by applying he linea
mapping successi ely om o a ion ma ices (RΦ1 and RΦ2) in each o hese spaces. The o a ion ma ix
is gi en by Equa ion (5.1), whe e j= 1 o j= 2:
Rj=




1 0 0
0cos(Φj)−sen(Φj)
0sen(Φj)cos(Φj).





(5.1)
Conside ing he sagi al plane as he plane o e e ence o he mo emen s, once he abduc ion/ad-
duc ion angles a e neglec ed (Figu e 27), he o ien a ion o he leg’s ame and he high’s ame, bo h
conce ning he ine ial coo dina e sys em, a e ep esen ed as o a ions in he e e ed plane.
When he moni o ed mo emen occu s in he sagi al plane, he knee angle is gi en by he di e ence
be ween he high and leg angles Φknee = Φ2−Φ1.
Figu e 27: Senso s wi h he angles ep esen a ion. Adap ed om (Olinski e al., 2017).
IMU da a is measu ed ela i e o he ine ial coo dina e sys em. By a aching an IMU o bo h he
high and he shin, hei espec i e o ien a ions can be de e mined in ela ion o he ine ial sys em. The
di e ence in hese o ien a ions p o ides he knee angle. This me hod allows o accu a e and con inuous
moni o ing o knee mo emen s, ensu ing p ecise da a o ehabili a ion pu poses.
66
5.1.2 Knee Ex ension Exe cise
The knee ex ension exe cise is undamen al in muscle ehabili a ion, especially o he quad iceps muscle
g oup. This exe cise in ol es ex ending he knee join and ac i a ing he quad iceps emo is, he la ges
muscle g oup in he high. S eng hening he quad iceps is essen ial o main aining knee join s abili y
and p omo ing p ope gai pa e ns (McGin y e al., 2000).
To pe o m he knee ex ension exe cise, he pa ien si s in a chai o on a knee ex ension machine wi h
knees ben a a 90-deg ee angle and ee la on he loo o oo es s. The exe cise begins by ex ending
he knees and aising he legs un il hey a e pa allel o he loo o ully ex ended. The mo emen is
comple ed when he pa ien ’s leg e u ns o he s a ing posi ion. Figu e 28 illus a es he da a collec ed
by he bio eedback sys em du ing a knee ex ension exe cise.
Figu e 28: Da a collec ed du ing knee ex ension exe cise.
The knee angle illus a ed in Figu e 28 passes h ough se e al pa abolas wi h he conca i y acing
upwa ds, ep esen ing each ime he knee ex ension mo emen was pe o med. I is also possible o
isualize he EMG signal synch onized wi h he mo emen , wi h an inc ease in he signal ampli ude when
he leg begins o ise and a decline in ampli ude when he leg begins o descend, ep esen ing he muscle
ac i a ion equi ed by he as us medialis muscle o pe o m he mo emen .
In his me hod, he knee ex ension mo emen was segmen ed in o ou dis inc phases, sys ema ically
de ined by wo h eshold lines. The i s h eshold line de e mines he minimum angle he knee angle
mus go h ough o be conside ed he s a o he leg ise. The alue o he i s h eshold was de e mined
as 20°. The second h eshold line ep esen s he minimum angle he knee angle mus go h ough o
67

he knee ex ension mo emen o be conside ed su icien . The physician should adjus his alue o each
pa ien since he ull knee ex ension ange can be educed in a muscle ehabili a ion con ex . Howe e , in
his s udy, i was de ined as 60º.
F om he h eshold lines, he ou phases a e de e mined om he ollowing in e als: The i s phase
ep esen s he upwa d mo emen , s a ing wi h alues abo e he i s h eshold and ending a he second
h eshold; o example, he leg le 20º and eached 60º. The second phase is when he mo emen is
al eady conside ed su icien . I can be ca ego ized when he knee angle is abo e he alue o he second
h eshold, ha is, alues abo e 60º. The hi d phase ep esen s he downwa d mo emen o he leg, which
can only happen i phase 1 and phase 2 ha e occu ed p e iously. I can be ca ego ized as descending
when he alues a e be ween he i s and second h eshold, equal o phase 1. Finally, phase 4 ep esen s
he leg a es , classi ied when he knee angle alues a e below he i s h eshold, o example, alues
below 20º.
Figu e 29: Classi ica ion o mo emen be ween he ou phases.
In his way, as he mobile applica ion ecei es he knee angle da a, he cu en s a e o mo emen
is classi ied based on his sys em o ules, as illus a ed in Figu e 29. Simila o a s a e machine, i
is conside ed ha he subjec managed o pe o m he knee ex ension mo emen co ec ly once when
he ou phases we e pe o med in sequence. When he sequence is comple e, and he mo emen is
conside ed co ec , i is conside ed in his s udy ha he subjec pe o med a con ac ion. Thus, i is
possible o isualize ou ecognized con ac ions in Figu e 29.
68
The EMG signal used o analyze each con ac ion in his sys em is ob ained by cu ing he co e-
sponding EMG signal be ween phases 1, 2, and 3 o he knee ex ension mo emen , excluding phase 4.
This cu ing me hod ocuses on cap u ing only he mos signi ican pa mo emen , omi ing he es ime
be ween con ac ions. The decision o exclude phase 4 is based on he unde s anding ha his phase,
cha ac e ized by leg es , con ains less ep esen a i e in o ma ion o analysis. Addi ionally, i is likely
o in oduce posi ioning noise and p olonged pe iods wi h minimal muscle ac i i y, which can impai he
accu acy o he analysis.
When he knee ex ension mo emen ails o go h ough all ou phases, i is classi ied as inco ec
con ac ion. Fo example, he subjec s a ed o li he leg, ailed o each he second h eshold, and
e u ned o he es ing posi ion, keeping phases 2 and 3 missing. In hese cases, he co esponding EMG
signal is dis ega ded o analysis.
Algo i hm 1 ma e ializes he desc ibed me hod esponsible o upda ing he mo emen o he knee
ex ension phase. This code uns con inuously in he backg ound du ing ehabili a ion ea men in he
wea able sys em and he mobile app.
In o de o p o ide a use - iendly in e ace, a sc een in he mobile applica ion was de eloped o supply
eal- ime isualiza ion o knee angle mo emen . The sc een upda es a a equency o 10 imes pe second,
aligning wi h he da a ansmission a e o he wea able sys em. This dynamic in e ace allows use s o
con inuously moni o knee ex ension mo emen s as hey occu .
The in e ace displays he cu en phase o con ac ions, he desc ip ion o he cu en angle, and
he o al numbe o con ac ions ecognized. Addi ionally, he physicians can easily adjus he minimum
angle limi using use - iendly bu ons on he sc een, as depic ed in Figu e 30. This unc ionali y aims
o imp o e he adap a ion o he ehabili a ion p ocess o he needs o each pa ien wi h accu a e and
upda ed in o ma ion o he physician.
Figu e 30: Phases o con ac ion ecogni ion in e aces.
69
Algo i hm 1 Knee Ex ension Mo emen Recogni ion Algo i hm
1: kneeEx ensionPhase ←“ es ing”
2: angleS a Con ac ion ←20
3: angleRecognizeCon ac ion ←60
4: cu en Con ac ionS a us ←False
5: while isRecogni ionAc i e do
6: angleKnee ←upda eKneeAngle()
7: i angleKnee >0 hen
8: i angleKnee <angleS a Con ac ion hen
9: i kneeEx ensionPhase == “mo ing down” hen #Con ac ion Recognize
10: p ocessCon ac ion()
11: else #Rese wi hou con ac ion
12: kneeEx ensionPhase ←“ es ing”
13: cu en Con ac ionS a us ←False
14: end i
15: else i cu en Con ac ionS a us == False hen #S a Con ac ion
16: cu en Con ac ionS a us ←T ue
17: kneeEx ensionPhase ←“mo ing up”
18: else i angleKnee >angleRecognizeCon ac ion hen #Recogni ion Angle Achie ed
19: kneeEx ensionPhase ←“up”
20: else i kneeEx ensionPhase == “up” hen #Below he achie ed ecogni ion angle
21: kneeEx ensionPhase ←“mo ing down”
22: end i
23: end i
24: delay(10)
25: end while
70
5.2 Muscle Fa igue Moni o
Moni o ing muscle a igue is c ucial o muscle ehabili a ion ea men s, allowing physician o adjus
ea men p o ocols and p e en o e exe ion o inju y. Muscle a igue is he inabili y o a muscle o
gene a e o sus ain a gi en le el o o ce, leading o dec eased pe o mance, inc eased isk o inju y, and
delayed eco e y (Enoka and Ducha eau, 2008). By moni o ing muscle a igue, physician can assess
pa ien p og ess, de e mine i he cu en ea men plan is e ec i e, and adjus ea men in ensi y and
equency acco ding o he pa ien ’s muscle esponse.
In muscle ehabili a ion ea men s, pa ien s usually pe o m exe cises a ge ing speci ic muscles,
such as he knee ex ension exe cise, which ocuses on eco e ing quad iceps muscle, especially he as us
medialis and as us la e alis. These exe cises a e usually epea ed se e al imes, and he in ensi y and
du a ion can g adually inc ease o e ime, leading o muscle a igue (Hassanlouei e al., 2012).
In adi ional clinical se ings, physician closely moni o pa ien s and adjus exe cises as needed. How-
e e , in emo e ehabili a ion, his le el o in e ac ion is mo e di icul , despi e ad ances in emo e ehabil-
i a ion, making i c ucial o ha e a sys em ha can moni o muscle a igue in eal- ime du ing a ea men
session.
5.2.1 Fa igue Me ics
The as us medialis muscle is one o he high muscles esponsible o ex ending he knee. When his
muscle is heal hy, mo e s abili y is gene a ed in he knee, a oiding inapp op ia e mo emen s. Howe e ,
wi h aging, he as us medialis can become weake , esul ing in dec eased knee s abili y and an inc eased
isk o inju y (Moznuzzaman e al., 2021). Figu e 31 shows an example o da a om he EMG senso
posi ioned in he as us medialis muscle and applied he me hod o ecognize knee ex ension mo emen .
This sec ion desc ibes he ope a ions pe o med on EMG signal clipped o calcula e h ee ele an
me ics in he con ex o muscle a igue. Two me ics a e ela ed o he signal spec um: he a e age and
median equencies. These me ics p o ide in o ma ion abou he equency cha ac e is ics o he EMG
signal and can indica e changes in muscle ibe ac i a ion. Speci ically, a shi in ac i a ion om as - wi ch
( ype 2) ibe s o slow- wi ch ( ype 1) ibe s indica e muscle a igue. This shi is e lec ed in he signal
spec um by a le wa d skew o he mean and median equencies, esul ing in lowe equency alues
(Ci ek e al., 2009).
In addi ion, Roo Mean Squa e (RMS) is calcula ed o quan i y he ampli ude o he EMG signal. The
RMS alue indica es he le el o muscle ac i a ion. Highe RMS alues co espond o g ea e muscle
71
o pa ien s, p o iding g ea e managemen capabili ies and esponsi eness o eal- ime mo emen . This
sec ion desc ibes h ee dis inc s imula ion modes, de ailing hei gene al and speci ic pa ame e s, as well
as p ac ical implemen a ion ia he wea able sys em and mobile applica ion.
All s imula ion modes sha e h ee main pa ame e s common o muscle s imula o s: in ensi y, pulse
wid h, and equency, as de ailed in Sec ion 2.2. In addi ion o hese, each mode has speci ic pa ame e s
o manage i s unique cha ac e is ics.
5.3.1 Comme cial S imula o
The i s mode eplica es he unc ionali y o he adi ional comme cial s imula o . In his mode, he s im-
ula o ope a es in a cyclical pa e n, whe e i is ac i a ed o a p ede e mined pe iod and hen deac i a ed
o ano he pe iod. This cyclical pa e n is usually desc ibed as a du y cycle a io, such as 1:2, meaning
one second wi h he s imula o ac i e and wo seconds inac i e.
Figu e 33: Sequence Diag am o he Comme cial S imula o .
Figu e 33 shows he sequence diag am o he in eg a ion o he mode 1 in CDSS. The physician
78

ini ia es mode 1 on he mobile applica ion, con igu es he s imula ion de aul pa ame e s, de ines he du y
cycle pa ame e s, and s a s he ea men . A his s age, he wea able sys em s a s wo loops: he i s
manages he o e all ea men ime, while he second con ols he du y cycle. The second loop uns wi h
a pe iod equal o he o al du a ion o he s imula ion and es phases combined. Du ing each i e a ion
o his loop, he sys em checks he cu en ime wi hin he cycle and adjus s he s imula ion acco dingly,
ensu ing ha he ac i a ion and es in e als ollow he p ede ined pa ame e s.
5.3.2 Raising he Leg S imula o
The second mode in oduces a mo e dynamic app oach o muscle s imula ion. In his mode, he s imula o
is ac i a ed only when he leg is in phase one (mo ing upwa d) o he knee ex ension mo emen . This
ensu es ha he elec ical s imula ion is synch onized wi h he ac i e muscle con ac ion phase, p o iding
suppo p ecisely when he muscle is wo king mos in ensi ely. This me hod can help imp o e muscle
coo dina ion and educe he isk o a igue by op imizing he iming o s imula ion.
Figu e 34: Sequence Diag am o he Raising he Leg S imula o .
Figu e 34 p esen s he sequence diag am o he second mode. The physician ini ia es mode 2 on he
79
mobile applica ion, con igu es he s anda d s imula ion pa ame e s, se s he limi angle o ecognize he
knee ex ension mo emen , and s a s he ea men . In his mode, du ing each i e a ion o he ea men
cycle, he con ac ion phase is upda ed acco ding o Algo i hm 1. Nex , i is checked whe he he mo emen
phase co esponds o he ac i e phase o s imula ion.
5.3.3 Range o Angle S imula o
The hi d mode aims a a mo e speci ic cus omiza ion, allowing he physician o selec a speci ic ange
o knee angles du ing which he s imula o should be ac i a ed. This a ge ed app oach ensu es ha he
muscle ecei es s imula ion du ing he momen o g ea es ac i a ion, p o iding assis ance in sus aining
he mo emen . Fu he mo e, his s imula ion mode is mo e adap able o pa ien s’ muscle ac i a ions,
which may di e om he expec ed pa e n o knee ex ension mo emen .
Figu e 35: Sequence Diag am o he Range o Angle S imula o .
Figu e 35 illus a es he sequence diag am o he ange o angle s imula ion mode. The physician
s a s mode 3 in he mobile applica ion, con igu es he s anda d s imula ion pa ame e s, de ines he ange
o angles, ha is, he minimum angle o s a s imula ion and he maximum o s op s imula ion, and s a s
he ea men . E e y 10 ms, he wea able sys em upda es he knee angle and checks whe he i is wi hin
he de ined ange. I he knee angle is wi hin he ange, he s imulus is ac i a ed; o he wise, i is disabled.
80
5.4 Rule-based S imula ion
One o he key in elligen ea u es in es iga ed in his hesis o ea men cus omiza ion is ule-based s im-
ula ion. This app oach enables he adjus men o elec os imula ion pa ame e s in esponse o muscle
a igue by u ilizing eal- ime bio eedback. I also allows physicians o in e p e he pa ien ’s muscle e-
sponse du ing he session, which is no commonly possible in adi ional clinical physio he apy se ings.
This in e ac ion allows o he pa ame iza ion and cus omiza ion o ea men acco ding o he indi idual
cha ac e is ics o each pa ien .
Rule-Based Reasoning (RBR) sys ems a e a well-es ablished app oach in CDSS, as discussed in he
backg ound sec ion. The implemen a ion o his sys em was mo i a ed by wo main ac o s. Fi s , he
ease o in e p e a ion o ule-based me hods by physicians esul s in a sho e lea ning cu e compa ed
o mo e complex me hods. Ad anced echniques can cause p oblems o lead o inadequa e ea men s
i physician a e no ully amilia wi h hem. The e o e, he simplici y o RBR ensu es ha physicians can
adop and use he sys em wi h con idence.
Second, c ea ing consis en da ase s o applying mo e complex me hods is a signi ican challenge.
Ob aining la ge olumes o consis en , high-quali y da a is especially di icul in p ojec s ha s a ed om
sc a ch, as i equi es mul iple ini ial alida ions and e hics commi ee app o als. The need o obus da a
limi s he e ec i eness o a i icial in elligence-based app oaches, making RBR a p ac ical and e icien
al e na i e.
5.4.1 Fa igue S ages
The ule-based sys em de eloped o he CDSS elies on he de ini ion o a igue s ages as one o i s key
ope a ional mechanisms. These a igue s ages a e de e mined using me ics ela ed o muscula a igue,
being Roo Mean Squa e and Median F equency, calcula ed o each con ac ion pe o med by he pa ien .
To in e p e hese me ics and iden i y a igue occu ences, he CDSS uses he Join Analysis o Spec um
and Ampli ude (JASA) me hod, widely used in he li e a u e o di e en ia e be ween no mal and a igued
muscle esponses du ing exe cise.
The JASA me hod was o iginally p oposed o dis inguish he e ec s o muscle a igue om a ia ions
in muscle s eng h h ough he simul aneous analysis o changes in he equency spec um and he
ampli ude o he EMG signal. S udies like Lu mann e al. (2000), Con o o e al. (2014) and Du aug e al.
(2020) help o alida ed he e ec i eness o JASA in assessing changes in muscle beha io , especially in
ac i i ies in ol ing epe i i e o p olonged e o s.
81
JASA is e ec i e in g aphically ep esen ing he ela ionship be ween Median F equency, which e lec s
he conduc ion eloci y o muscle ibe s, and RMS, which indica es signal ampli ude and is associa ed
wi h muscle s eng h. When obse ed on a Ca esian plane, hese wo me ics allow he ca ego iza ion o
muscle s a e in o ou main physiological scena ios:
Q1: An inc ease in RMS and a shi in MDF o he igh , sugges ing an inc ease in muscle s eng h.
Q2: A dec ease in RMS and a shi in MDF o he igh , indica ing a igue eco e y.
Q3: A dec ease in RMS and a shi in MDF o he le , indica ing a educ ion in muscle s eng h.
Q4: An inc ease in RMS and a shi in MDF o he le , cha ac e izing muscle a igue.
This app oach has been widely alida ed, and s udies ha e shown a s ong co ela ion be ween he
d op in equency o he EMG signal and he inc ease in ampli ude as ma ke s o a igue. The JASA me hod
e ec i ely iden i ies he poin a which he muscle begins o show signs o a igue, allowing ea men s,
such as elec os imula ion, o be adjus ed based on he pa ien ’s muscle condi ion.
In simple e ms, a a igue occu ence is in e p e ed when he muscle needs o gene a e mo e o ce o
pe o m a ask bu does so less e icien ly by using slowe ibe s. This pa e n is iden i ied when he e is
a simul aneous inc ease in RMS (indica ing g ea e muscle e o ) and a d op in MDF (indica ing he use
o slowe ibe s). The CDSS calcula es hese me ics o each con ac ion, and when bo h condi ions a e
me , he sys em iden i ies a a igue occu ence.
In addi ion o de ec ing hese occu ences, he CDSS e alua es he in ensi y o each a igue e en . This
in ensi y is calcula ed using he Euclidean no m equa ion px2+y2, whe e xis he pe cen age inc ease
in RMS and yis he pe cen age dec ease in MDF. This allows he sys em no only o de ec a igue bu
also o quan i y i s se e i y, dis inguishing be ween mild and in ense episodes.
Con ol Pa ame e s
The ule-based sys em o he CDSS uses h ee main pa ame e s o de ine he p og ession o a igue s ages:
La e S a (S): De ines he numbe o ini ial con ac ions ha a e igno ed when moni o ing a igue
occu ences, elimina ing e en s ela ed o he muscle’s ini ial adap a ion o he exe cise.
Fa igue Occu ences (F): Speci ies he numbe o a igue occu ences equi ed o he sys em o
egis e a change in s age, adjus ing he ea men sensi i i y acco ding o a igue p og ession.
Minimum In ensi y (M): De e mines he minimum in ensi y a a igue occu ence mus each o
be conside ed alid, ensu ing ha only signi ican e en s a e accoun ed o .
These pa ame e s allow he physician o cus omize he ea men acco ding o he pa ien ’s muscle
esponse. When a a igue s age is eached, he sys em can igge speci ic ules ha adjus elec os imu-
82
la ion pa ame e s as needed.
To acili a e he adjus men o hese pa ame e s and p o ide a clinical alida ion ool, a a igue s age
simula o was de eloped and in eg a ed in o he CDSS adminis a i e panel. This simula o allows physi-
cian o es di e en combina ions o S,F, and Mand isualize how he sys em esponds o he pa ien ’s
muscula a igue p og ession. This ea u e is pa icula ly use ul o op imizing ea men and ensu ing he
sys em app op ia ely esponds o he needs o each pa ien .
Figu e 36 demons a es he simula o in a es session wi h 100 con ac ions. This ool p o ides a clea
isualiza ion o a igue s age p og ession, helping physicians p ecisely and op imally adjus he pa ame e s
o inc ease ea men e ec i eness.
Figu e 36: Muscle Fa igue S ages Simula o .
As shown in Figu e 36, he ajec o y o he a igue me ics does no ollow a linea pa e n, p esen ing
oscilla ions h oughou he es session. These a ia ions e lec he dynamic na u e o muscle ac i i y
and luc ua ions in he e o equi ed o pe o m he mo emen . In his example, he a igue occu ence
pa ame e was con igu ed o de ec h ee occu ences o igge a s age change, wi h a minimum in en-
si y o 1 and he la e s a se o 10, ensu ing ha he coun o occu ences only begins a e he en h
con ac ion, a oiding in luences om he ini ial phase o muscle adap a ion.
The o ange poin s in Figu e 36 ep esen con ac ions iden i ied by he CDSS as a igue occu ences,
83

signaling momen s when he e was an inc ease in RMS and a dec ease in MDF, bo h wi h in ensi ies abo e
1, acco ding o he de ined pa ame e s. In his example, he pa icipan expe ienced 15 a igue occu ences
h oughou he session, and since he occu ence pa ame e was se o 3, he sys em ecognized 5 a igue
s ages. This indica es ha he clinician would ha e i e oppo uni ies o adjus he s imula ion pa ame e s,
such as pulse wid h and s imula ion in ensi y, as muscle a igue p og esses.
5.4.2 Au oma ic Session
To di e en ia e sessions conduc ed in he clinic and manually adjus ed by he physician om p e-p og ammed
sessions using he ule-based sys em, he e minology o assis ed sessions and au oma ic sessions was
es ablished. As desc ibed in he p oposed ea men scena io, ini ial ea men sessions a e conduc ed in
he clinic o amilia ize bo h he physician and he pa ien wi h he new ea men and de ices. Addi ionally,
hese sessions aim o c ea e an ini ial baseline o he pa ien ’s muscle esponse.
Wi h he ini ial da a collec ed, he a igue s age simula o is used o analyze muscle esponses and
de e mine he op imal alues ha allow o cohe en s age ansi ions as pe sonally assessed by he physi-
cian. Once he a igue pa ame e s a e adjus ed, he physician can egis e an au oma ic session. In
hese sessions, he s imula ion pa ame e s a e au oma ically adjus ed by he sys em when a a igue s age
change occu s, as de ined by he ules.
Figu e 37 shows he o m ha needs o be illed ou o c ea e an au oma ic session. Ini ially, he
o m eques s he ea men plan, he ecommended da e, and a session desc ip ion. Nex , he physician
can adjus he ini ial s imula ion pa ame e s, including s anda d pa ame e s and hose speci ic o he
s imula ion mode. A e ha , i is possible o adjus he adap i e pa ame e s o he pa ien ’s a igue s age,
such as he numbe o a igue occu ences, minimum in ensi y, la e s a , and he minimum angle o
con ac ion ecogni ion. Subsequen ly, he physician mus de ine how much o inc ease he in ensi y and
pulse wid h a each a igue s age change.
Finally, he s op session pa ame e s a e de ined, which include he maximum s age a chi ed, he o al
du a ion o he session, and he maximum numbe o con ac ions. The p ima y objec i e is o comple e
all he con ac ions planned by he physician be o e one o he o he wo s op condi ions is eached.
Howe e , i he e is an indica ion o excessi e a igue, i is expec ed ha one o he o he wo condi ions
will be ac i a ed.
84
Figu e 37: Au oma ic Session Fo m.
Figu e 38 displays he main in e aces o he ea men session wi h he ule-based sys em ac i a ed.
As shown in he i s sc een, he e is an indica ion o he a igue s age as 0, ep esen ing he ini ial s a e
o he session. When he subjec eaches he i s a igue s age, he second sc een appea s, eques ing
he pa ien ’s pe mission o inc ease he s imula ion in ensi y as p e iously se by he physician. Following
his, he ea men in e ace is upda ed wi h he new a igue s age, and he session con inues.
85
Figu e 38: Au oma ic Session Adap a ion Sc eens.
5.5 Da a isualiza ion
Da a isualiza ion plays a undamen al ole in he CDSS, con e ing complex da a se s in o comp ehensi-
ble and ac ionable in o ma ion. In muscle ehabili a ion, eal- ime isualiza ion helps physicians moni o
p og ess, adjus pa ame e s, and ensu e he accu acy o he adminis e ed ea men . In addi ion o p o-
iding specialized analysis, hese isual ools a e designed o enhance he in e ac ion be ween he sys em
and i s use s, making he ehabili a ion p ocess mo e anspa en and manageable.
The inal sec ion o his chap e explo es he da a isualiza ion componen s implemen ed in he CDSS.
In addi ion o he in e aces al eady desc ibed in p e ious sec ions, ou mo e in e aces ha e been im-
plemen ed, wo in he mobile applica ion and wo in he adminis a i e panel. These in e aces aim o
acili a e physician’ decision-making by ans o ming cap u ed bio eedback in o in e p e a i e and in ui i e
isual componen s.
5.5.1 Mobile App
Two main sc eens we e de eloped in he mobile applica ion ha ensu e accu a e cap u e o bio eedback
da a and help he physician unde s and he muscle esponse in eal ime.
86
Figu e 39: Calib a ion O e iew Sc een.
The calib a ion o e iew sc een, as shown in Figu e 39, displays he esul s o a success ul calib a ion
pe o med wi h he myHeal h Sys em. This sc een p esen s p ocessed da a o he h ee con ac ions
ecognized du ing he calib a ion phase. The da a, including he maximum angle eached, he median
equency alue calcula ed in Hz, and he du a ion o each con ac ion, a e no jus numbe s. They a e
p ac ical ools ha can be used o compa e wi h p e ious ea men sessions and check o p og ess in
he pa ien ’s ange o mo ion.
A cha showing he a e age olume o he EMG signal cap u ed du ing he h ee con ac ions is
p esen ed a he end o he in e ace. This cha allows o analyzing he knee angles a which he muscle
shows g ea e ac i a ion du ing he knee ex ension mo emen . In his way, his cha is signi ican o wo
s imula ion modes: he aising he leg s imula o and he ange o angles s imula o , which ely on he angle
o ac i a e he s imula ion. Thus, besides ensu ing ha he elec odes a e well-posi ioned and cap u ing
he co ec muscle ac i i y, i also enables a mo e p ecise adjus men o he s imula ion, conside ing he
pa ien ’s cu en esponse.
The second da a isualiza ion in e ace o he mobile app is illus a ed in Figu e 40. This in e ace
p o ides a eal- ime analysis o he con ac ions ecognized du ing he ea men session. Each ime a
con ac ion occu s, he mobile app calcula es me ics ela ed o muscle a igue and upda es he g aphs.
87
wi h he objec i es o p omo ing pe sonalized ea men s. Th ough a igo ous me hodology and de ailed
empi ical alida ions, i is in ended o demons a e ha he de eloped CDSS is an inno a i e and aluable
ool o physicians and pa ien s, signi ican ly con ibu ing o he ad ancemen o muscle ehabili a ion wi h
elec os imula ion.
This chap e is di ided in o i e main sec ions. The i s sec ion ou lines he e alua ion Me ics used
du ing he de elopmen and es ing phases. The second sec ion desc ibes he p elimina y alida ion
p ocedu es, including ha dwa e e i ica ion, secu i y analysis and p elimina y sessions. The subsequen
sec ions p esen he expe imen al s udies, each add essing a di e en alida ion o he sys em. The i s
expe imen al s udy examines he quali y and synch oniza ion o he da a acquisi ion and p ocess EMG
signals wi h s imula ion a i ac s gene a ed by comme cial de ice. The second s udy ocuses on eal- ime
muscle a igue moni o ing and analyzes he sys em’s abili y o p ocess da a du ing ea men sessions.
The inal s udy p o ides an in-dep h analysis o he CDSS’s accu acy by u ilizing a gold-s anda d mo ion
cap u e sys em, compa ing di e en s imula ion modes, and examining he s a is ical co ela ion be ween
he a igue pe cei ed by pa icipan s and he le els o a igue p edic ed by he CDSS.
6.1 E alua ion Me ics
To e alua e he pe o mance o he de eloped CDSS, h ee key me ics we e employed: Mean Absolu e
E o (MAE), Mean Squa ed E o (MSE), and Roo Mean Squa ed E o (RMSE). In ou con ex , hese
me ics will be applied o compa a i e analysis be ween he CDSS wi h o he sys ems and me hodologies,
such as compa ing eal- ime p ocessing done by he mobile app wi h p ocessing done on a compu e
using aw da a.
To elucida e hese e alua ion me ics, conside wo da a se s we aim o analyze o simila i y: he
da a poin s ec o om he CDSS (A) and he da a poin s ec o om a compa able echnology (F).
I all elemen s o ec o Aa e sub ac ed om ec o F, he esul ing e o ec o E ep esen s he
disc epancies be ween he wo da a se s. An ideal sys em would yield an e o ec o wi h a sum o ze o,
indica ing no e o .
The MAE me ic is calcula ed as he a e age o he absolu e e o s, dis ega ding he e o signs. This
means ha e o s abo e and below he ac ual alue a e ea ed equally. MAE p o ides a s aigh o wa d
in e p e a ion o he a e age magni ude o he e o s, making i a use ul me ic o gene al e o analysis.
The equa ion o MAE is gi en by:
94

MAE =1
n
n
X
=1 |A −F |(6.1)
The MSE me ic is he a e age o he squa ed di e ences be ween he ac ual and sys em ou pu alues.
By squa ing he e o s, his me ic places a la ge penal y on la ge e o s, hus highligh ing signi ican
disc epancies mo e e ec i ely han MAE. The MSE is calcula ed using he ollowing equa ion:
MSE =1
n
n
X
=1
(A −F )2(6.2)
The RMSE me ic is he squa e oo o he MSE, p o iding an e o me ic in he same uni s as he
o iginal da a, which makes i mo e in e p e able. RMSE is pa icula ly sensi i e o la ge e o s (ou lie s)
and hus is a obus measu e o he sys em’s pe o mance. The equa ion o RMSE is:
RMSE =
u
u
1
n
n
X
=1
(A −F )2=√MSE (6.3)
6.2 P elimina y Valida ion
This sec ion co e s he ini ial expe imen s and undamen al alida ions conduc ed o ensu e he unc ion-
ali y and obus ness o he p oposed sys em. These s eps we e essen ial ea ly in he de elopmen p ocess
o ensu e ha he sys em ope a es as expec ed and mee s he igo ous sa e y and e iciency s anda ds
equi ed in clinical se ings.
P elimina y es ing includes a a ie y o analyses, such as da a ansmission es s, sa e y analysis,
ini ial ha dwa e and so wa e es s, and alida ion o he IMU senso s. The p ima y goal was o es ablish
a solid ounda ion o subsequen de elopmen . Each o hese subsec ions p o ides a de ailed o e iew
o he me hods and esul s ob ained, demons a ing how each componen o he sys em was es ed.
6.2.1 T ansmission Tes
Ini ially, he i s p o o yping o he p esen ed sys em included he implemen a ion o all componen s de-
sc ibed in he sys em a chi ec u e, albei wi h basic unc ionali ies such as use and session egis a ion.
Be o e ha dwa e implemen a ion, he capabili y o he ESP32 mic ocon olle was es ed o handle EMG
senso acquisi ion, manage s imula ion and pa ame e changes, and communica e ia BLE wi h he mobile
app simul aneously. Addi ionally, he sys em’s abili y o manage a ea men session, including s a ing,
pausing, canceling, and sa ing da a o he cloud, was e i ied.
95
The i s es simula ed a ea men las ing 1 minu e, wi h s imula ion pa ame e changes sen e e y
8 seconds. To e i y he ecep ion o EMG senso da a, a da a packe was sen e e y 200 ms om
he wea able sys em o he mobile app wi h simula ed da a. Mo e de ails abou his es and pe inen
discussions a he beginning o he sys em’s de elopmen we e p esen ed in he a icle (F anco e al.,
2022b).
Figu e 44: Packe exchange ime be ween he wea able sys em and he mobile app.
As shown in Figu e 44, he low-cos , low-powe ESP32 mic ocon olle p o ed capable o e icien ly
main aining bo h he p ocessing o elec os imula ion and da a ansmission. The analysis o packe ex-
change imes showed ha he a e age packe ansmission ime is 206 ms, indica ing an a e age delay
o only 6 ms o assemble each p ocessed packe . This esul sugges s ha he addi ional la ency in o-
duced by da a ansmission is minimal and does no signi ican ly a ec he sys em’s ope a ion. The abili y
o pe o m mul iple unc ions simul aneously wi hou no iceable pe o mance deg ada ion is a p omising
indica o o he p oposed sys em and he chosen ha dwa e.
6.2.2 Secu i y Analysis
To ensu e he in eg i y, p i acy, and a ailabili y o clinical da a managed by he CDSS, a mas e ’s hesis
was de eloped in conjunc ion wi h his doc o al hesis wi hin he NanoS im p ojec (Sil a, 2021). This
hesis was mo i a ed by he need o scien i ic igo in alida ing he p o ec ion o sensi i e da a in clinical
en i onmen s. The main s a egies o p o ec ing sensi i e da a implemen ed in he CDSS we e co e ed
in Sec ion 4.1.3. This subsec ion desc ibes he key es s conduc ed and hei conclusions as de ailed in
Sil a (2021)p, elucida ing he obus ness o he p esen ed sys em a chi ec u e.
The secu i y analysis comp ised h ee main componen s: a ailabili y es ing, unc ional es ing, and
h ea modeling. To simula e a ealis ic scena io o he sys em’s usage, he CDSS implemen a ion used
o secu i y analysis included Kube ne es o con aine o ches a ion, Clus e Con ol o da abase man-
96
agemen , and HAP oxy as a load balance . This se up aimed o ensu e au oma ed ailo e , ease o main-
enance, scalabili y, and audi abili y o he a chi ec u e.
The i s es , a ailabili y es ing, aimed o e alua e he CDSS’s beha io when one o i s se ices ails
and measu e he esponse ime un il he sys em s abilizes again. The esul s showed ha he Manage-
men API and Clinical Se ice ook 2 o 4 seconds o become a ailable again, while he SSO se ice wi h
Keycloak ook abou 50 seconds o ees ablish. Using Kube ne es, Clus e Con ol, and HAP oxy minimizes
a ailabili y issues, as edundancy ensu es he sys em emains a ailable e en i a se ice o da abase is
emo ed, wi h only a b ie educ ion in p ocessing powe .
The unc ional es aimed o alida e he communica ion s a egies be ween he sys em’s componen s
and he CDSS beha io unde peak eques loads on each o he APIs, p e en ing Dis ibu ed Denial o
Se ice (DDoS) a acks. Fo his, endpoin s om each se ice (Managemen API, Clinical Se ice, and
SSO) we e selec ed o es ing. Each endpoin in ol ed a speci ic ou ine o p omo e au hen ica ion and
he exchange o pseudo-anonymized in o ma ion. A Py hon sc ip was used o simula e he in o ma ion
low, epea edly sending eques s o iden i y i he se e ’s esou ce usage inc eased. The numbe o
eques s inc emen ed by 100, up o a maximum o 3500 simul aneous eques s. The esul s showed no
signi ican changes in he a e age esponse ime, emaining consis en ac oss all eques le els.
Finally, STRIDE h ea modeling was conduc ed. This analysis desc ibes po en ial h ea s, unde s ands
hei causes, and c ea es mi iga ion s a egies. STRIDE modeling add esses aspec s such as spoo ing,
ampe ing, epudia ion, in o ma ion disclosu e, denial o se ice, and ele a ion o p i ileges. The analysis
e ealed ha all sys em asse s could be a ge ed, indica ing a b oad a ack su ace. The asse s in insecu e
en i onmen s, such as he mobile app, web adminis a i e panel, and wea able de ice, a e po en ial
en y poin s o a acks. Iden i ying and educing he likelihood o hese h ea s is essen ial o enhancing
CDSS secu i y. Es ablishing obus me hods o ecognizing se ices, de ices, use s, and in o ma ion
complica es he malicious exploi a ion o CDSS unc ionali ies, posi i ely impac ing p i acy.
The secu i y analysis conduc ed suppo s he hesis ha i is possible o c ea e a CDSS ha handles
sensi i e da a. The desc ibed s a egies, along wi h he es esul s, demons a e he abili y o ensu e da a
p o ec ion in a CDSS ha collec s senso da a h ough a wea able sys em. Addi ionally, he implemen a-
ion used in he es s wi h Kube ne es, Clus e Con ol, and HAP oxy so wa e demons a es he CDSS’s
capabili y in a mo e ealis ic en i onmen , wi h scalabili y and edundancy.
97
6.2.3 A Fi s Expe imen
The i s p o o ype o he ha dwa e o he wea able sys em de eloped in he NanoS im p ojec was assem-
bled on a b eadboa d, connec ing he componen s o he ESP32 De Ki mic ocon olle , a comme cial
signal acquisi ion boa d (EBZ-AD8233), and a cus om-made elec ical s imula ion ci cui . This ini ial p o o-
ype was desc ibed in (Ses em. e al., 2022), which explo es in g ea e de ail he elec onic componen s
included in he wea able sys em and he EMG signal acquisi ion es wi h we and d y elec odes.
Wi h his ha dwa e, i was possible o implemen a i s e sion o he mobile applica ion, desc ibed
in (F anco e al., 2022a). This e sion o he applica ion included all he ac i i y lows desc ibed in he
ea men scena ios, including ecei ing s imula ion pa ame e s egis e ed in he cloud by he physician,
a ea men sc een ha displayed he acqui ed EMG senso signal in eal- ime, and a sel - epo eedback
in e ace a e he comple ion o he ea men session.
To alida e his e sion o he applica ion, a es ea men session was con igu ed h ough he adminis-
a i e panel, las ing 1 minu e and a ying he s imula ion con igu a ions. The es session was conduc ed
wi h ou heal hy olun ee s, wi h he EMG senso posi ioned on he
as us medialis muscle
wi h he help
o a physician. When he es session began, he olun ee , sea ed on a bench, li ed hei leg i e imes
in one minu e wi hou being s imula ed.
Figu e 45: Tes ea men session on a olun ee wi hou elec os imula ion. (A) Subjec ’s leg elaxed. (B)
Subjec ’s knee is ully ex ended.
The es conduc ed o alida e ha he applica ion can ollow he necessa y s eps o NMES muscle
ehabili a ion ea men , as illus a ed in Figu e 45. In Figu e 45A, he olun ee ’s leg is elaxed, ep e-
98
sen ing he es pe iod. A his momen , muscle ac i i y is low, esul ing in a sligh al e a ion in he EMG
signal displayed by he applica ion. When he olun ee li s hei leg, a olun a y muscle con ac ion is
gene a ed. The e o e, in Figu e 45B, i is possible o see how he EMG signal displayed by he applica ion
signi ican ly changed i s wa e ampli ude.
To e i y i he NMES con igu a ions we e being ansmi ed co ec ly, a esis o was placed a he
NMES ac ua o ou pu o he wea able sys em o simula e human skin. A mul ime e was connec ed o
his load o e i y he cu en alue as well as an oscilloscope o measu e he equency and wa e o m, as
shown in Figu e 46. The signal indica ed by he de ices ollowed he pa ame e s p o ided by he mobile
applica ion, con i ming he accu acy o he ansmission o he s imula ion con igu a ions.
Figu e 46: Oscilloscope wi h s imula ion pulses applied o he esis o .
6.2.4 Valida ion o IMU Senso s
Following he implemen a ions, he nex s ep was o alida e he IMU senso s. Fo his, he s udy desc ibed
in (F anco e al., 2022c) was conduc ed. In his s udy, a new wea able sys em was c ea ed, composed
o wo iden ical modules. Each module included an ESP32, an MPU-6050 IMU senso , and a ba e y, all
solde ed on o a pe o a ed boa d and housed in a 3D-p in ed case. The case was designed o be placed
on he high and shin, and hold by an elas ic band. The pu pose o his s udy was h ee old.
The i s pu pose was o e i y ha he ESP32 could handle wo addi ional asks: eading om wo
IMU senso s using I2C communica ion and sending he da a ia BLE. The second pu pose was o alida e
he ma hema ical model p esen ed in Sec ion 5.1.1 o knee angle ecogni ion and he eal- ime upda e
99

o he mobile app in e ace. The hi d pu pose was he cha ac e iza ion and compa ison o low-cos
compu a ional il e s o imp o e he accu acy o he IMU senso o measu e angles.
Compu a ional Powe P oblem
The ESP32 is an MCU wi h su icien compu a ional powe o acqui e senso da a a high equencies,
as he clock o a s anda d model, such as he ESP32-WROOM-32D, exceeds 150 MHz. Howe e , due
o he numbe o simul aneous asks, i was obse ed ha he ime be ween each sample om he IMU
senso a ied. To add ess his, he IMU modules we e p og ammed o execu e a so wa e eplica ing he
ea men p o ocol, wi h he addi ion o a dedica ed h ead designed o collec a IMU sample e e y 8 ms.
This se up ideally esul s in a sampling equency o 125 Hz.
Howe e , due o he sha ed p ocessing powe wi h o he asks, he collec ed da a showed acquisi ion
equencies o 99–100 Hz. Addi ionally, he a e age ime in e al be ween each sample was 9.8 ms,
wi h some peaks abo e 24 ms. The g aph in Figu e 47 shows he ime di e ence be ween each sample
collec ed by he wea able sys em in a sample acquisi ion.
Figu e 47: ∆Time o IMU samples collec ed by he wea able sys em.
In con as , he eading o he EMG senso does no expe ience hese delays. This is because he
EMG senso eading is execu ed h ough an in e up unc ion, igge ed p ecisely e e y 1 ms. This le el
o p ecision canno be achie ed wi h he IMU senso s because hei communica ion occu s ia I2C. On
he ESP32 MPU, i is no possible o collec da a om he I2C in e ace wi hin an in e up unc ion. As a
100
esul , he IMU eading h ead compe es wi h o he h eads, such as da a ansmission ia BLE, leading
o he delays obse ed in Figu e 47.
IMUs Senso Valida ion Scena io
To in es iga e he pe o mance o IMU senso s o knee angle ecogni ion, he UR3 obo ic a m was used.
The UR3 collabo a i e indus ial obo om Uni e sal Robo ics1is sui able o assembly and sc ewd i e
ac i i ies, usually posi ioned on he op o benches. This obo ic a m was chosen o wo main easons:
i s ly, he a ailabili y o he obo in he labo a o y whe e he esea ch was ca ied ou ; secondly, due
o he ac ha he UR3 has a ce i ica e alida ing he p ecision o he join ’s mo emen s. Thus, i is
possible o con igu e he UR3 o pe o m a gi en mo emen wi h a minimum e o in ajec o ies, making
he alida ion mo e cohesi e.
Each wea able module was a ached wi h clo hing elas ics o di e en pa s in one o he join s o he
UR3 obo , ep esen ing he knee join , as illus a ed in Figu e 48.
Figu e 48: Wea able modules a ached o he UR3 obo .
The UR3 obo can be p og ammed o ollow a sequence o posi ions de e mined by he join angle.
Thus, o all es s pe o med, he obo was p og ammed o mo e he selec ed join in he ollowing posi ions
epea edly: [0◦,90◦,75◦,90◦,60◦,45◦]. Each ime he join eached one o he chosen angles, he UR3
emained s a ic o wo seconds be o e mo ing o he nex posi ion.
To compa e he mo emen pe o med by he UR3 obo wi h he wea able sys em, a Py hon applica ion
1h ps://uni e sal- obo s.com/p oduc s/u 3- obo
101
was de eloped o communica e wi h he UR3 ia Wi-Fi and collec he angle o he chosen join . I is
impo an o no e ha he UR3 upda es he egis e ha s o es he join angle a a equency o 30 Hz. To
align wi h he amoun o da a gene a ed by he wea able sys em, he sampling equency o he so wa e
was se o 100 Hz, allowing o a highe densi y o compa able da a.
To analyze he da a acqui ed by he wea able sys em, a ea men session wi hou s imula ion was
eco ded o each es pe o med. The acqui ed da a we e s o ed in a MongoDB da abase, om whe e hey
could be impo ed in o a Jupy e No ebook o analysis be o e implemen ing he knee angle ecogni ion
ea u e in he mobile app. In he end, wo ime-se ies a ays we e gene a ed o each es , con aining he
angles eco ded by he UR3 and he wea able sys em.
The es s consis ed o powe ing bo h sys ems and acqui ing da a con inuously o 100 seconds. This
p ocedu e was epea ed se e al imes o e i y he consis ency o he acqui ed da a, and no signi ican
di e ences we e ound be ween he uns. The esul s p esen ed in his sec ion ep esen a snapsho o
he mo emen pe o med by he sys ems h ee imes du ing one o he acquisi ions.
Al hough bo h sys ems we e con igu ed wi h he same sampling equency, i was no possible o au-
oma ically synch onize da a acquisi ion om bo h sys ems simul aneously. The e o e, manual synch o-
niza ion o he da a was necessa y. Figu e 49 shows he aw da a om he wea able sys em synch onized
wi h he da a acqui ed by he UR3 obo .
Figu e 49: Wea able sys em aw da a synch onized wi h UR3 da a.
As can be seen in Figu e 49, bo h senso s, gy oscope, and accele ome e can ep oduce he mo e-
men pe o med by he UR3, bu hey a e inaccu a e. As he accele ome e can be e measu e slow
102
mo emen s, he e o is small du ing con inuous mo emen s, as shown in Figu e 49 in he ange o mo-
ion om 0° o 90°. Howe e , when he UR3 s ops, he accele ome e akes ime o s abilize, gene a ing
noise peaks.
Con a ily, he gy oscope can be e measu e sudden speed changes, p esen ing less noise. Howe e ,
in con inuous mo emen s, a small e o is accumula ed in e e y sample, esul ing in a signi ican di e ence
compa ed o he desi ed deg ee a he end o he mo emen .
Compa ing he wo senso s using he e alua ion me ics, he gy oscope has a conside ably wo se esul
han he accele ome e . Fo he MAE me ic, he accele ome e has a alue o 1.27 and he gy oscope
5.37, ou imes highe . Fo he RMSE, he accele ome e has 2.09 and he gy oscope 6.18, h ee imes
highe . Las ly, he MSE o he accele ome e is 4.38, and o he gy oscope i is 38.27, eigh imes highe .
Fil e s wi h Algo i hmic Complexi y O(1)
Wi hin he a ea o Signal P ocessing, se e al il e s ha e been s udied o imp o e he accu acy o IMU
senso s. As commen ed in he li e a u e e iew o he a icle F anco e al. (2022c), he implemen a ion o
il e s such as Kalman and Madgwick can conside ably educe e o s in he measu emen o join angles.
Howe e , he implemen a ion o he bes il e s equi es a e y high compu a ional cos , which is a esou ce
o en limi ed in embedded sys ems.
The Big O no a ion is one o he mos used no a ions o desc ibe he compu a ional cos o a gi en
algo i hm. This no a ion akes in o accoun he size o he inpu and coun s he numbe o ins uc ions
used o execu e a gi en sequence o code. Fo example, o an algo i hm ha calcula es whe he he gi en
inpu is e en o odd, only one ins uc ion will be used, esul ing in an algo i hmic complexi y o O(1). Fo
an algo i hm ha needs o a e se a ec o o size n, he algo i hmic complexi y is O(n), since a leas
n ins uc ions will be execu ed o comple e he ask (Chi e s e al., 2015).
The algo i hm complexi y p esen ed by Valade e al. (2017) o he Kalman il e is O(10n3); o he
ex ended Kalman il e , i is O(4n3). A s udy on he algo i hmic complexi y o he Madgwick il e was no
ound, bu as calcula ions wi h ma ices we e used, he algo i hmic complexi y was o be a leas O(n).
Fu he mo e, hese algo i hms equi e memo y esou ces o s o e he in e media e ma ices needed in
e e y calcula ion, ha is no also abundan in embedded sys ems.
Gi en his scena io, h ee il e s wi h lowe compu a ional cos and algo i hmic complexi y o O(1)
we e es ed o de e mine he bes one o implemen a ion in he wea able embedded sys em. The il e s
es ed we e:
Simple Mo ing A e age
(SMA),
Exponen ial Mo ing A e age
(EMA), and
Complemen a y Fil e
(CF) o he accele ome e and gy oscope.
103
a ious clea ly isible peaks can be obse ed, indica ing he s imula ion a i ac s.
Despi e hese peaks, he olun a y muscle esponse can also be seen occu ing o e he s imula ion
a i ac s. When he s imula ion is u ned o , he peaks disappea , and he signal e u ns o no mal ac i i y,
ee om he in e e ence o he a i ac s. This sec ion desc ibes he me hodology applied o emo e hese
a i ac s, esul ing in clean, il e ed EMG signals sui able o subsequen analysis.
The i s s ep was o calcula e he exac pe iod be ween he s imula ion pulses. Gi en ha he com-
me cial de ice was se o 1 Hz, he expec ed in e al be ween pulses was 1000 ms. To con i m his, he
signal om es s 1 and 2 was c opped o a segmen ha de ini ely con ained s imula ion o all pa icipan s,
speci ically om seconds 100 o 130.
Nex , he ind_peaks unc ion om he SciPy lib a y was used o de ec peaks in he ec i ied EMG
signal. These peaks we e de ined as alues exceeding he signal’s mean plus h ee imes he s anda d
de ia ion. Wi h his, he a e age ime be ween he peaks was calcula ed, and he median in e al among
all pa icipan s was de e mined, de ining he ime in e al be ween each s imula ion pulse. The esul was
an in e al o 970.33 ms, di e ging om he expec ed alue. Figu e 54 shows an example o he ec i ied
EMG signal wi h i s iden i ied peaks, as desc ibed.
Figu e 54: Example o peak de ec ion in he EMG signal.
A i ac Segmen a ion
The logic o iden i ying he s imula ion segmen is based on he equency o occu ence o he s imula ion
pulses. Knowing his equency, he i s s ep is o de e mine he s a o he s imula ion window. To achie e
his, an algo i hm was de eloped o segmen he in e -pulse in e als (IPIs).
The algo i hm wo ks by analyzing wo consecu i e windows o da a. Fo each window, he index wi h
110

he lowes alue wi hin ha window is ound. I he di e ence be ween hese indices is less han a h eshold
o 30 poin s, i indica es ha s imula ion has likely s a ed, as he minimum poin s a e e y close, signaling
he p esence o a s imula ion a i ac . To con i m ha s imula ion has indeed s a ed, he nex i e windows
a e es ed. I he di e ence be ween he indices emains below he h eshold, he ec o index indica ing
he s a o s imula ion is iden i ied.
To segmen he windows, he index is mo ed back 50 poin s o ensu e a sa e y ma gin, and segmen-
a ion is pe o med e e y 970 poin s. Segmen a ion o he a i ac s s ops when he di e ence be ween he
minimum indices o he windows exceeds he h eshold.
A e iden i ying he s a o s imula ion, he nex s ep is o sepa a e he s imula ion a i ac , he M
wa e, and he olun a y signal. The s a o he a i ac is de ined as 30 poin s be o e he local minimum
wi hin he i s 70 poin s o he window. Nex , wi hin he ollowing 200 poin s o he ec o , i is calcula ed
whe e he signal c osses ze o (i.e., shi s om nega i e o posi i e o om posi i e o nega i e). The i s
ze o-c ossing poin is de ined as he end o he s imula ion a i ac . The M wa e is de ined as he segmen
be ween he end o he s imula ion a i ac and he ou h subsequen ze o-c ossing poin .
Figu e 55: Example o segmen wi h iden i ied s imula ion a i ac .
111
In Figu e 55A, an example o an IPI is shown wi h he ma king o he s a and end o he a i ac and
he M wa e. Figu es 55B and 55C display, espec i ely, he a e age and s anda d de ia ion o he a i ac s
and M wa es o all expe imen pa icipan s. The a e age ime o he s imula ion a i ac s and M wa e
among pa icipan s was 160 ms ou o 970 possible, co esponding o abou 16% o he IPI. This segmen
o he signal needs o be il e ed o achie e a clean EMG signal ha closely ep esen s he pa icipan ’s
ue esponse.
Cubic In e pola ion Fil e
To ensu e he con inui y o he EMG signals and emo e s imula ion a i ac s, a il e based on cubic
in e pola ion o he da a adjacen o he a i ac segmen s was applied. Cubic in e pola ion is a ma h-
ema ical echnique used o es ima e new da a poin s wi hin he ange o a disc e e se o known da a
poin s. Unlike linea in e pola ion, which connec s adjacen poin s wi h s aigh lines, cubic in e pola ion
uses hi d-deg ee polynomials o connec he poin s, esul ing in a smoo h cu e ha passes h ough he
o iginal da a poin s.
Cubic in e pola ion is p e e ed in many applica ions because, unlike linea in e pola ion, i no only
gua an ees he con inui y o he in e pola ed unc ion bu also he con inui y o i s i s and second de i a-
i es. This makes cubic in e pola ion especially use ul o biological signals, such as EMG, whe e main-
aining he signal’s p ope ies is pa icula ly impo an .
To apply he il e o each iden i ied a i ac segmen , he o al size o he segmen , o he numbe o
poin s occupied by he a i ac , is de e mined. Then, a se o da a poin s equal o he size o he a i ac
is selec ed be o e and a e he a i ac . These segmen s a e used o cons uc he cubic in e pola ion
unc ion. The ‘CubicSpline‘ unc ion om he SciPy lib a y is used o gene a e he in e pola ion.
The cubic in e pola ion unc ion is applied o he in e al co esponding o he a i ac segmen , gen-
e a ing an in e pola ed segmen ha eplaces he a i ac in he o iginal signal. This p ocedu e ensu es
ha he ansi ion be ween he eal da a and he in e pola ed da a is smoo h and con inuous, main aining
he in eg i y o he EMG signal.
In Figu e 56, an example o an EMG signal a e he applica ion o he il e can be seen. No e ha
his is he same signal shown in Figu e 53, bu wi h he s imula ion a i ac s emo ed. Figu e 57 shows
an example o a segmen wi h he s imula ion a i ac and he espec i e il e ed signal. I can be obse ed
ha he applica ion o he cubic in e pola ion il e e ec i ely emo ed he s imula ion a i ac s, esul ing
in a clean and con inuous signal.
112
Figu e 56: Example o EMG signal cap u ed du ing he FES expe imen il e ed.
Figu e 57: Example o segmen wi h iden i ied s imula ion a i ac il ed.
6.3.3 Maximum Volun a y Con ac ion (MVC)
MVC is a c ucial concep in elec omyog aphy s udies, used o measu e he mos signi ican o ce ha a
muscle o muscle g oup can gene a e du ing olun a y con ac ion. P ac ically, MVC se es as a e e ence
poin o no malizing EMG signals, allowing muscle ac i i ies o be exp essed as a pe cen age o he indi id-
ual’s maximum e o . This no maliza ion enables clea compa isons be ween subjec s and expe imen al
condi ions, as i accoun s o he inhe en a ia ions in absolu e muscle s eng h ac oss subjec s.
In Figu e 58, he a e age muscle ac i i y o each subjec du ing he i s es is displayed in milli ol s
(mV). The isual analysis sugges s ha mos olun ee s exhibi ed simila muscle ac i a ion le els, wi h
113
some no able a ia ions in h ee subjec s. This obse a ion implies ha he subjec s exe cised compa able
e o du ing he es .
Figu e 58: Mean muscle ac i i y o subjec s in mV.
Fo a p ope no maliza ion o he da a, EMG signals collec ed du ing he second es (STS es ) we e
used o iden i y he maximum muscle ac i i y du ing a dynamic ask. The inal 15 seconds o he es we e
selec ed based on he assump ion ha subjec s had pe o med hei maximum e o du ing his pe iod.
The ex ac ed EMG signals we e p ocessed in ou s eps: (1) ec i ica ion o he signal, (2) smoo hing
using a mo ing a e age window o size 300, (3) iden i ica ion o he peak ampli ude, and (4) calcula ion
o he a e age o 20 adjacen poin s a ound he iden i ied peak. The esul s o his p ocess a e shown in
Figu e 59, which p esen s he MVC alues o each subjec .
Figu e 59: Maximum Volun a y Con ac ion o subjec s.
114
Following he MVC calcula ion, all subjec s’ EMG da a we e no malized, exp essing muscle ac i i y as
a pe cen age o he MVC, as illus a ed in Figu e 60.
Figu e 60: Mean muscle ac i i y o subjec s no malized by MVC.
A compa ison be ween Figu e 58 and Figu e 60 highligh s he no maliza ion alue. In Figu e 58,
muscle ac i a ion alues appea ela i ely simila ac oss subjec s; meanwhile, in Figu e 60, a bigge
a iabili y is obse ed, wi h some subjec s using up o 40% o hei o al muscle capaci y, while o he s
u ilized less han 10%. This indica es ha some subjec s had mo e ease in pe o ming he mo emen
han o he s, o in o he wo ds, some subjec s needed o demand mo e o hei as us medialis muscle o
pe o m he same ask compa ed o o he subjec s.
I is impo an o emphasize ha he mean muscle ac i a ion no malized by he MVC is a weak indica o
o he subjec s’ muscle a igue. Al hough i is possible o assess he pe cen age o muscle equi ed o
pe o m he mo emen , i is no possible o say how much longe he subjec would be able o main ain
he mo emen wi h he same pe cen age o muscle use. I is possible ha some subjec s, despi e ha ing
high mean muscle ac i a ion alues, can sus ain he mo emen longe han a subjec wi h a lowe a e age.
This happens because ou muscles ha e a ied in insic cha ac e is ics, such as ibe ype, body a , and
muscle olume, which in luence he muscle’s abili y o sus ain he mo emen o a longe o sho e ime,
which would mo e accu a ely indica e muscle a igue.
6.3.4 Knee Ex ension Mo emen Recogni ion
Wi h he EMG signals p ope ly p ocessed and no malized, he nex s ep in he expe imen al s udy was o
apply he knee ex ension mo emen ecogni ion algo i hm, as desc ibed in Sec ion 5.1. This phase o he
115

s udy was c ucial o alida ing he e ec i eness o he de eloped algo i hm in accu a ely iden i ying and
segmen ing he mo emen s pe o med by he olun ee s using eal da a collec ed du ing he expe imen al
sessions. The main goal was o e i y whe he he p oposed me hod could p ecisely ecognize mo emen
cycles, an essen ial ask o he CDSS’s e ec i e applica ion in clinical en i onmen s.
The ecogni ion algo i hm was hen applied o he da a om all pa icipan s, esul ing in 15 ecognized
mo emen s in he i s es and 15 mo e in he hi d es , as expec ed o each pa icipan . The i s
in es iga ion on his opic ocused on he a ia ion in he maximum angle achie ed and he du a ion o
each con ac ion.
Figu e 61: Va ia ion in Maximum Angle and Du a ion o Knee Ex ension Mo emen - Expe imen 1.
The cha in Figu e 61 indica ed ha he e was a a ia ion in he du a ion o con ac ions ecognized
among he olun ee s, wi h an a e age o app oxima ely 8 seconds pe con ac ion. Mos subjec s com-
ple ed he mo emen by aking 1 second o aise he leg, main aining he leg a he maximum posi ion o
6 seconds, and aking 1 second o lowe he leg.
Howe e , he mos no able obse a ion was ha all subjec s displayed a maximum angle exceeding
90 deg ees, which is ana omically imp obable. This angula de ia ion was a ibu ed o he imp ecise
posi ioning o he IMU modules, which had al eady been iden i ied in p elimina y s udies. The con i ma ion
o his de ia ion in his expe imen al s udy sugges ed he need o a sys em calib a ion p ocess be o e
each session o educe angula e o and ensu e a mo e accu a e ep esen a ion o knee mo emen .
A no maliza ion p ocess was pe o med o co ec he knee angle collec ed in his expe imen al s udy,
adjus ing he maximum alue o each es o 90 deg ees, espec ing human ana omical limi s.
116
Figu e 62: Cycles o Knee Ex ension Mo emen .
Figu e 62 p esen s he mean and s anda d de ia ion o he ecognized con ac ions, wi h he c opped
knee angle and he EMG signal. The da a we e in e pola ed so ha he du a ion o he mo emen cycle
was ans o med in o a pe cen age o he o al cycle, allowing o compa isons dis ega ding di e ences
in con ac ion du a ion. The cha in Figu e 62 e idence ha he algo i hm wo ked as expec ed since he
da a om he IMU and EMG senso s we e synch onized. I he algo i hm had segmen ed he mo emen s
inco ec ly, diso ganized peaks would be expec ed in he EMG signal. The cha shows an inc ease in
muscle ac i i y a he beginning o he mo emen , speci ically du ing leg ele a ion, ollowed by a s able
ac i a ion le el du ing he holding phase and a g adual dec ease du ing he descen phase.
The analysis o he esul s indica es sligh a ia ion in knee angle du ing he mo emen s pe o med
by he olun ee s, sugges ing ha he mo emen s we e ca ied ou consis en ly. Howe e , muscle ac i i y
showed mo e signi ican a ia ion among pa icipan s, indica ing ha some indi iduals needed o exe
mo e muscle o ce o comple e he mo emen compa ed o o he s. In some cases, muscle ac i i y was
up o wice as high in ce ain indi iduals.
Fo a mo e de ailed analysis o muscle ac i i y du ing knee ex ension mo emen , he EMG signal da a
we e ans o med in o a a io o he knee angle o each con ac ion. The mean and s anda d de ia ion o
hese a ios we e hen calcula ed, gene a ing he cha shown in Figu e 63.
117
Figu e 63: Muscle Ac i i y in Knee Angle Ra io.
The analysis e eals ha he olun ee s’ muscle ac i i y g adually inc eased as he knee angle in-
c eased, eaching i s peak in he las 20 deg ees o ex ension. This esul sugges s ha he as us medialis
muscle emains highly ac i e du ing he ini ial phase o he mo emen , playing a c ucial ole in he ull
ex ension o he knee. Howe e , muscle ac i i y du ing he holding phase was less in ense, indica ing ha
muscle e o is signi ican ly highe in he ea ly s ages o he mo emen .
In o de o ob ain a mo e p ecise ep esen a ion o his hypo hesis, a line in Figu e 63 was added o
e idence he muscle ac i i y o a speci ic pa o he mo emen cycle, being be ween 15% and 35% o he
o al cycle, whe e he peak o ac i i y was obse ed in Figu e 63. The cha e ealed ha muscle ac i i y
du ing his pe iod was conside ably highe han du ing he holding phase, ein o cing he impo ance o
he as us medialis muscle in he las 20 deg ees o he knee ex ension mo emen .
Fi s Tes s Thi d Tes
Following he objec i es, a compa a i e analysis was conduc ed be ween he da a ob ained in he i s
and hi d es s, wi h he p ima y goal o iden i ying and quan i ying possible di e ences be ween he wo
assessmen momen s. This allowed he iden i ica ion o signs o muscle a igue o changes in mo emen
execu ion o e ime.
118
Figu e 64: Compa ison be ween he Fi s and Thi d Tes .
Figu e 64 p esen s a g aphical compa ison be ween he mo emen s ecognized in he i s and hi d
es s. In Figu e 64A, he g aph shows he ajec o y o he knee angle o bo h es s, while pa B p esen s
he muscle ac i i y. As seen in Figu e 64A, he mo emen ecognized in bo h es s is qui e simila , indica ing
ha he olun ee s consis en ly pe o med he knee ex ension mo emen . Howe e , i can be obse ed
ha , in he hi d es , he knee angle du ing he sus aining phase is sligh ly lowe compa ed o he i s
es . This sugges s ha he subjec s had some di icul y main aining he maximum angle as he muscles
became mo e a igued.
In Figu e 64B, which illus a es he muscle ac i i y, a mo e no iceable di e ence be ween he wo es s
is obse ed. The muscle ac i i y in he hi d es is isibly highe compa ed o he i s es , indica ing ha
he olun ee s’ muscles had o wo k ha de o pe o m he mo emen in he las epe i ion. This inc ease
in muscle ac i i y may be indica i e o accumula ed a igue h oughou he exe cises, equi ing addi ional
e o om he muscles o comple e he knee ex ension mo emen .
Figu e 65: Compa ison be ween he Fi s and Las Mo emen .
A second analysis was conduc ed o in es iga e his di e ence u he , p esen ed in Figu e 65. In
his analysis, he i s and las mo emen s pe o med h oughou he en i e expe imen al session we e
selec ed. In Figu e 65A, i is obse ed ha he e was no signi ican di e ence in he sus aining phase
119
be ween subjec s, wi h an app oxima e a e age o 5 seconds o each con ac ion, wi h an a e age o 1
seconds o aise he leg, 3 second emaining a he op and 1 seconds on he way down.
The maximum angle eached was 83° on a e age ac oss all subjec s. I was also no ed ha he ma-
jo i y o subjec s inc eased he maximum angle eached wi h each con ac ion, esul ing in he maximum
peak be ween con ac ions 30 and 40.
Figu e 69: Cycle o Knee Ex ension Mo emen - Exp 2.
Simila o he i s expe imen al s udy, he g aph in Figu e 69 p esen s he a e age and s anda d
de ia ion o he segmen s om he EMG signal and knee angle, bo h no malized o he comple e mo emen
cycle. The da a analysis shows ha he signals a e well synch onized, as expec ed, demons a ing he
accu acy o he mo emen ecogni ion algo i hm. Howe e , a no able di e ence compa ed o he i s
s udy is he iming o he peak muscle ac i i y. In he second expe imen al s udy, his peak occu s close
o he middle o he con ac ion cycle.
This a ia ion can be explained by he di e ence in he a e age du a ion o con ac ions be ween
he wo s udies. In he i s expe imen al s udy, he con ac ions las ed an a e age o 8 seconds, wi h 6
seconds dedica ed o he holding phase. In con as , in he second s udy, he holding phase was educed
o an a e age o 2 seconds, esul ing in a edis ibu ion o muscle ac i i y h oughou he mo emen cycle.
This change was in en ional in he p o ocol o eplica e a con ac ion mo e simila o wha is ypically
pe o med in ehabili a ion clinics.
126

6.4.3 Fa igue Me ics E alua ion
This sec ion p esen s a s a is ical analysis o h ee me ics ela ed o muscle a igue: A gF eq, MedF eq,
and RMS. These me ics we e ob ained om he EMG signal de i ed om he segmen iden i ied by
ecognizing con ac ions based on knee angle and subsequen ly calcula ed as desc ibed in sec ion 4.2.
The calib a ion p ocess desc ibed con e s hese me ics in o pe cen ages ela i e o he ini ial s a e
o he es , which is c ucial o moni o ing he session’s p og ession. Howe e , o compa e ac oss mul iple
subjec s, he da a s ill needs o be s anda dized due o a ia ions in EMG signal s eng h. Thus, he da a
we e no malized as s anda d de ia ions om he mean.
Addi ionally, due o some ailu es in eading he EMG signal om he wea able sys em, i was necessa y
o emo e up o 3 con ac ions om some subjec s. Consequen ly, he epo ed alues ep esen a mo ing
a e age wi h a window size o i e ac oss 45 con ac ions. Fo subjec s wi hou signal disc epancies,
he ini ial 5 da a poin s we e excluded. This me hodology o e s insigh s in o he end o me ic alues
h oughou he session.
Figu e 70 and Figu e 71 illus a e he ends o me ics ela ed o signal equency, A gF eq and Med-
F eq o each subjec , and he a e age end among he en i e g oup. The a e age end dec eases as
con ac ions p og ess. This decline in equency indica es a dec ease in muscle ibe conduc ion eloc-
i y, which is consis en wi h he expec ed physiological esponse o a igue. As he pa icipan becomes
a igued, he muscle’s abili y o gene a e o ce dec eases, esul ing in a slowe conduc ion eloci y o he
muscle ibe , shi ing he signal powe spec um o lowe equencies Ci ek e al. (2009).
In some cases, he e was a cons ancy o e en an inc ease in he A gF eq and MedF eq me ics un il
he hal o he session. This can be unde s ood as a pe iod o adap a ion o he mo emen ha he subjec
wen h ough. These esul s demons a e simila i ies wi h hose p esen ed in Liu e al. (2019), whe e he
median equency wen h ough se e al wa es wi h a nega i e slope un il he end o he exe cise.
In con as o he equency me ics, he a e age end o he RMS me ic illus a ed in Figu e 72 does
no exhibi a sha p dec ease. In gene al, he p edominan beha io was an inc ease un il hal way h ough
he session and hen a dec ease. This obse a ion sugges s ha he subjec added s eng h un il hal way
h ough he session, po en ially adap ing o he mo emen and compensa ing o he onse o a igue.
A e wa d, he e a e wo mos likely possibili ies: The i s is ha he RMS dec eases, indica ing a igue
wi h a lack o abili y o gene a e mo e o ce o sus ain he mo emen , as demons a ed by he au ho s in
Shaw e al. (2020). The second op ion is when he RMS s abilizes o inc eases, indica ing ha he subjec
calmly endu ed he exe cise un il he end.
127
Figu e 70: T end o he a gF eq me ic du ing he session.
Figu e 71: T end o he medianF eq me ic du ing he session.
Figu e 72: T end o he RMS me ic du ing he session.
128
The Join Analysis o Spec um and Ampli ude (JASA) is a me hod ha seeks o unde s and muscula
esponses by simul aneously conside ing he MedF eq and RMS me ics in a Ca esian plane. The JASA
me hod aims o dis inguish he changes obse ed in he EMG signal be ween he e ec s o muscle a igue
and a ia ions in muscle o ce ha occu ed du ing exe cise (Lu mann e al., 2000). These quad an s o
he Ca esian plane ep esen di e en physiological scena ios: he inc ease in RMS and he displacemen
o MedF eq o he igh indica e a possible inc ease in muscula o ce (Q1); he dec ease in RMS and he
displacemen o MedF eq o he le sugges a p obable educ ion in muscula o ce (Q3); he inc ease in
RMS and he shi o MedF eq o he le indica e muscle a igue (Q4); he dec ease in RMS and he shi
o MedF eq o he igh indica e eco e y om p e ious a igue (Q2).
Figu e 73: JASA Me hod apply o Expe imen al Tes .
Figu e 73 illus a es he applica ion o he JASA me hod o he con ac ions ecognized o he en i e
g oup. The do s we e colo ed in a g adien om ed o g een, indica ing he i s con ac ions as edde o
he las ones as g eene . Thus, i is possible o see ha despi e he a ied s a , he a e age p og ession
o he muscula esponse du ing he session was a dec ease in o ce and some cases o a igue.
This is an expec ed muscula esponse o heal hy subjec s who p ac ice physical ac i i y egula ly.
129
Con ibu ing again o an in e p e a ion ha he e was some challenge a he beginning o he session o
adap o he mo emen , and hen he e was a con ol o o ce un il he end o he session.
Using he chosen a igue me ics allows a comp ehensi e assessmen o muscle a igue. While e-
quency me ics e lec physiological changes associa ed wi h a igue, RMS p o ides in o ma ion abou
pa icipan exe ion le els. The in eg a ion o hese me ics imp o es he unde s anding o he p og ession
o a igue and acili a es he iden i ica ion o di e en s ages o muscle a igue, as exempli ied using he
JASA me hod. Fu he mo e, o he pa ame e s ela ed o muscle a igue ex ac ed om he EMG senso
can be used, as desc ibed by Ci ek e al. (2009), once he da a has been p ope ly cu and p ocessed wi h
mo ion ecogni ion by he IMUs.
I is essen ial o emphasize ha he in e p e a ion o hese indings mus be he esponsibili y o he
physician, who has he necessa y knowledge abou he pa ien ’s pa hology. I is essen ial o ecognize ha
di e en pa hologies may p esen di e en a igue beha io s compa ed o heal hy indi iduals. The e o e,
he physician plays a c ucial ole in analyzing hese me ics and making in o med decisions based on
unde s anding he pa ien ’s condi ion.
6.4.4 Muscle Fa igue Le el Simula o
The inal analysis conduc ed wi h he da a om he second expe imen al s udy ocused on he beha io o
he muscle a igue le el simula o , applied o he a igue me ics calcula ed by he CDSS. This subsec ion
aims o exempli y how he adjus able pa ame e s o he simula o can in luence he cha ac e iza ion o
each pa icipan ’s a igue, allowing he s imula ion ea men o be ailo ed acco ding o he eal- ime
bio eedback collec ed.
As p e iously explained in Sec ion 5.4 o his documen , he a igue le el simula o has h ee main
pa ame e s ha de e mine i s adap abili y. The i s pa ame e , e e ed o as S, ep esen s he numbe
o con ac ions a e which he simula o begins o ope a e, helping o exclude a igue episodes caused
by ini ial muscle adap a ion o he exe cise. The second pa ame e , F, indica es he numbe o a igue
occu ences equi ed o igge a change in a igue le el. Las ly, he hi d pa ame e , M, co esponds o
he minimum in ensi y ha a a igue occu ence mus each o be conside ed alid.
To demons a e he simula o ’s lexibili y, di e en se s o alues we e es ed o each o hese pa am-
e e s: S= [5,10,20],M= [1,2,3], and F= [2,3,4]. These pa ame e combina ions esul ed in
27 di e en con igu a ions, which we e applied indi idually o each pa icipan ’s session, gene a ing he
esul s o his sec ion.
130
Figu e 74: (A) Occu ence o Fa igue (B) In ensi y o Occu ences by Con igu a ion.
Figu e 74A p esen s a box plo showing he numbe o a igue occu ences when a ying he Sand M
pa ame e s. I is e iden ha he a e age numbe o a igue occu ences among pa icipan s signi ican ly
dec eased as he minimum in ensi y (M) inc eased, wi h M= 3 showing he ewes occu ences. The
a ia ion o he Spa ame e e ealed li le change when s a ing he moni o ing a e he 5 h o 10 h
con ac ion. Howe e , when he simula o began ope a ing a e he 20 h con ac ion, bo h he a e age
numbe o a igue occu ences and he s anda d de ia ion dec eased subs an ially.
Figu e 74B shows he a e age in ensi y o he eco ded a igue occu ences o each con igu a ion,
wi h a ia ions in he Sand Mpa ame e s. As expec ed, he a e age in ensi ies o he con ac ions
inc eased as he minimum h eshold (M) was aised, excluding lowe -in ensi y alues and pushing he
a e age upwa d. Addi ionally, some a igue occu ences we e iden i ied as ou lie s because hei alues
we e signi ican ly abo e he no m. I is no ed ha hese con ac ions wi h g ea e in ensi y p edominan ly
occu ed la e in he session since only one o he ou line occu ences was il e ed when he S= 20
con igu a ion was applied.
Following he analysis, a compa ison was made be ween he a e age le els o a igue eco ded by he
CDSS, conside ing di e en numbe s o occu ences equi ed o a le el change. The inal le el eached
is a c ucial me ic in he CDSS’s ule-based me hod, as i indica es how o en can he physician adjus he
pa ame e s o he elec os imula ion ea men du ing a session.
As shown in Figu e 75, pa icipan s expe ienced a ange o ze o o i e a igue le els, depending on
he simula o pa ame e s. Fo con igu a ions ha equi ed only wo occu ences o a igue, a leas one
le el change occu ed o all pa icipan s, wi h he a e age being a ound h ee le el changes. In con as ,
when ou occu ences we e equi ed o a le el change, some pa icipan s did no each he i s a igue
le el. In he las scena io, mos pa icipan s did no each le el 1, and only a ew managed o do so.
In o de o complemen Figu e 75, Figu e 76 shows he epe i ion a which he i s le el change
occu ed in each con igu a ion. I can be seen ha con igu a ions equi ing wo occu ences o a igue
led o ela i ely ea ly le el changes in he session, wi h mos pa icipan s eaching he i s le el be ween
131

con ac ions 20 and 30, excep o con igu a ions wi h a la e s a de ined a con ac ion 20.
Figu e 75: Fa igue Le els Achie ed by Con igu a ion.
Figu e 76: Con ac ion numbe a i s a igue le el achie ed.
The con igu a ions ha mos accu a ely cap u ed he pa icipan s’ muscle esponses du ing he ex-
pe imen 2 sessions we e hose using he pa ame e s F= 3,S= 10, and M= 1. This con igu a ion
conside s a highe numbe o a igue occu ences han M= 2, wi h all pa icipan s eaching a leas one
le el, and o mos , his le el occu ed a ound con ac ion 30.
Addi ionally, he hypo hesis de eloped in he p e ious subsec ion sugges s ha , while pa icipan s
expe ienced some le el o a igue, especially a he beginning o he session, hey did no eel excessi ely
132
a igued. The e o e, a igue le els be ween one and h ee a e expec ed o be no mal o heal hy indi iduals
pe o ming his ype o exe cise. This in o ma ion p o ides aluable da a o he physician o make in o med
adjus men s o he ea men acco ding o he expec ed e ec s.
Subjec s Fa igue E alua ion
To u he analyze he esul s, he chosen con igu a ion o he subsequen da a was F= 3,S= 10, and
M= 1 o he a igue le el simula o , as i p esen ed an op imal balance be ween sensi i i y o muscle
e o and he abili y o ack pa icipan s’ a igue p og ession h oughou he expe imen .
Figu e 77: (A) Amoun o Occu ence (B) A e age In ensi y o Fi s and Las Le el pe Pa icipan .
Figu e 77A illus a es he numbe o a igue occu ences o each subjec du ing he session o expe -
imen al s udy 2. I is no ed ha pa icipan s exhibi ed a a ied numbe o a igue occu ences, suppo ing
he hypo hesis ha only some subjec s el mo e a igued. Th ee subjec s eco ded eigh o mo e occu -
ences, while, on he o he hand, subjec s 4 and 6 had only 4 and 3 occu ences, espec i ely. This
indica es ha hese subjec s we e be e p epa ed o he exe cise.
In Figu e 77B, he a e age in ensi y o he i s and las a igue le els eached by he pa icipan s is
compa ed. I is obse ed ha , o h ee subjec s, he ini ial in ensi ies we e highe han he inal ones,
highligh ing g ea e di icul y in adap ing o he exe cise. Addi ionally, subjec s 7 and 2 showed he la ges
inc eases in a e age in ensi y and also had high numbe s o a igue occu ences. Fo he de ined con-
igu a ions, he simula o indica ed ha subjec s 2, 3, and 7 expe ienced mo e di icul y pe o ming he
exe cise.
Finally, Figu e 78 aims o cla i y a wha poin du ing he expe imen al session he a igue le els we e
eached on a e age and hei espec i e in ensi ies. As p e iously no ed, o mos pa icipan s, he i s
a igue le el was eached a ound he 30 h epe i ion, which is mo e han hal way h ough he p oposed
exe cise. The second and hi d le els, on a e age, we e eached du ing he las 10 epe i ions. The
in ensi y be ween he a igue le els was qui e simila , al hough o le el 2, a g ea e a ia ion was obse ed
133
compa ed o le el 1.
Figu e 78: (A) Con ac in Numbe o Le els Change (B) A e age In ensi y pe Le el.
6.4.5 Mobile Da a s. Cloud Da a
In o de o e alua e he consis ency and accu acy o he Fa igue Moni o ing Me hod in eal- ime, his
sec ion compa es he da a p ocessed in he mobile app wi h he da a p ocessed on he compu e om
he aw da a. Wi h his, i is possible o ob ain in o ma ion abou he pe o mance and eliabili y o da a
p ocessing algo i hms.
Two me ics we e used o pe o m he compa a i e analysis be ween he sys ems, namely: Mean
Absolu e E o (MAE) and Mean Mean Squa e E o (RMSE). These me ics seek o pa ame e ize he
di e ence be ween he da a ob ained in each sys em, in e p e ed as ou e o .
Figu es 79, 80 and 81 p esen compa a i e cha s showing he a e age e o be ween eal- ime p o-
cessing pe o med by he mobile applica ion and cloud p ocessing o calcula ing a igue me ics o all
subjec s. These cha s also p o ide insigh s in o he a e age pos -applica ion e o esul ing om he im-
plemen a ion o a mo ing a e age wi h a window size o 5, as p e iously discussed. The e ical lines
associa ed wi h each da a poin in he cha s ep esen he s anda d de ia ion calcula ed ac oss all sub-
jec s, allowing a deepe unde s anding o he da a dis ibu ion.
F om hese cha s, i is e iden ha he me ics ela ed o equency (A gF eq and MedF eq) had mo e
e o s han he RMS me ic. The mos plausible explana ion o he di e gence be ween he sys ems lies
in he calcula ions ela ed o he ans o ma ion o he signal in he equency spec um and no in how he
senso da a is cu and synch onized. O he wise, he RMS me ic would also p esen a e age e o s simila
o A gF eq and MedF eq. Also, hese calcula ions we e one o he ew pa s wi hin he mobile sys em ha
was no p og ammed om sc a ch, especially he Fou ie T ans o m Algo i hm. Gi en he use o di e en
p og amming languages in implemen ing he me hod, a ia ions such as di e ing a iable p ecision may
con ibu e o his e ec .
134
Figu e 79: Compa ison be ween sys ems o he A gF eq me ic.
Figu e 80: Compa ison be ween sys ems o he MedF eq me ic.
Figu e 81: Compa ison be ween sys ems o he RMS me ic.
135