Thesis

New treatments for induction of motor plasticity after stroke

NGUYEN-DANSE, Dung Anh

Abstract

Stroke is a major health issue. It is one of the main causes of acquired adult disability. A
common deficit after stroke is the hemiparesis of the contralateral upper limb, with more than
80% of stroke inpatients experiencing this condition and in spite of intensive rehabilitation
care, more than 40% of the patients still have this impairment chronically. The overall aim of
this thesis is to understand how we reduce impairment in stroke patients with new treatment
approaches. Previous studies have shown possible targets for new treatments. Patients with
severe motor deficits show severe damage of the cortico-spinal tract (CST) and secondary
degradation of white matter during the weeks after stroke. In these patients, inducing plasticity
at the level of the CST or avoiding secondary degradation may enable the motor recovery
processes, which are usually very limited or absent in these patients. Patients with mild or
moderate motor deficits show high FC between the motor areas and the rest of the brain in
the first weeks after stroke, which associates with better clinical recovery. In these patients,
enhancing FC may improve [...]

Reference
NGUYEN-DANSE, Dung Anh. New treatments for induction of motor plasticity after
stroke. Thèse de doctorat : Univ. Genève, 2022, no. Neur. 326

DOI : 10.13097/archive-ouverte/unige:161342
URN : urn:nbn:ch:unige-1613429

Available at:
http://archive-ouverte.unige.ch/unige:161342

Disclaimer: layout of this document may differ from the published version.
 DOCTORAT EN NEUROSCIENCES
des Universités de Genève
et de Lausanne

UNIVERSITÉ DE GENÈVE FACULTÉ DE MEDECINE

Professeur Adrian GUGGISBERG, directeur de thèse

Professeur Armin SCHNIDER, co-directeur de thèse

TITRE DE LA THESE

NEW TREATMENTS FOR INDUCTION OF MOTOR PLASTICITY

AFTER STROKE

THESE
Présentée à la
Faculté de Médecine

de l’Université de Genève

pour obtenir le grade de

Docteure en Neurosciences

par

Dung Anh NGUYEN-DANSE

de Paris

Thèse N° 326

Genève

Editeur ou imprimeur : Université de Genève

2022
Acknowledgments
In 2014 after a long day in my physiotherapy office, I returned home determined to develop my
profession by entering research. Planning my university journey, and thanks to google I
discovered the Imaging-Assisted Neurorehabilitation lab. I emailed Adrian and was encouraged
to pursue my project. That email meant a lot to me. It gave me strength and a goal to pursue.
Today this goal is about to be achieved and that is thanks to my supervisor that showed openness
and welcomed me warmly in his laboratory.
I would like to thank Prof. Armin Schnider, my co-director, for his availability and input at each
milestone of my thesis.
The academic world can be tough. However, from the discussion I had with my jury committee I
know that scientific rigor can go hand in hand with benevolence and listening. Thank you
professors Friedhelm Hummel, Lucas Spierer, and Dimitri Van de Ville.
A clinical study cannot be performed without patients, I thank all the patients that in spite of their
condition, agreed to participate in our studies.
A special thanks to Ulrike, our LNDS coordinator who showed so much kindness and guidance to
make all the procedures more fluid and brighten up the grey days.
One cannot speak of a laboratory without the (ex-)labmates. Thank you office 4-048 (Dr. Leslie
Allaman, Dr. Léa Chauvigné, Dr. Anaïs Mottaz, Dr. Marija Uscumlic) for all the discussion to
remake the world but especially to break this glass ceiling once and for all. Thank you office “it-
is-always-too-hot-in-this-office” (Dr. Alexandra Adam-Darqué, Dr. Alexia Bourgeois, Dr. Selim Coll,
Naz Doganci, Dr. Selim Habiby, Dr. Giannarity Iannotti, Emilie Marti, Stephan Schneider, Dr.
Domile Tautvydaite) for all the exchanges which allowed me to keep an ear on the news. Special
mention when we did a team work to understand the phenomenon of the red moon eclipse. But
especially for joining me with such enthusiasm in our booth at La Nuit de la science.
To my brother Nguyễn Arthur “Nhâm”, thank you for showing me that I could choose my future
and that obstacles are often those that we put to ourselves. Without him, I would have stayed
where I was 10 years ago, without daring much. You showed me the way, now I wish you all the
same for the end of your thesis and after.
To my parents, Nguyễn Vĩnh Bình and Chung Sophie, I am proud to show that their sacrifices post-
1975 were not in vain. I want them to know and see that indeed I had greater opportunities and
a brighter future. Cám ơn ba má.
To my father-in-law, Dr. Louis Danse, who showed me so much understanding and support.
To my dear husband, Romain Danse, thank you for not only supporting me during this thesis, but
also helping me during my 2 pregnancies. You have proven to be an extraordinarily calm and
patient person. 2 kids and a thesis and keeping a peaceful home, I guess we are ready for any
challenge.
Finally, I have a special thought for my sons, Tom Bình An and Jules Minh Khang, to whom I thrive
to be at the same time a loving mother and a fighter who travels the world hoping to improve it
for you.
Abstract (English)
Stroke is a major health issue. It is one of the main causes of acquired adult disability. A common
deficit after stroke is the hemiparesis of the contralateral upper limb, with more than 80% of
stroke inpatients experiencing this condition and in spite of intensive rehabilitation care, more
than 40% of the patients still have this impairment chronically. The overall aim of this thesis is to
understand how we reduce impairment in stroke patients with new treatment approaches.
Previous studies have shown possible targets for new treatments. Patient with severe motor
deficits show severe damage of the cortico-spinal tract (CST) and secondary degradation of white
matter during the weeks after stroke. In these patients, inducing plasticity at the level of the CST
or avoiding secondary degradation may enable the motor recovery processes, which are usually
very limited or absent in these patients. Patients with mild or moderate motor deficits show high
FC between the motor areas and the rest of the brain in the first weeks after stroke, which
associates with better clinical recovery. In these patients, enhancing FC may improve the motor
outcome above the proportional rule.
In a first study, we recruited patients with mild and moderate motor deficits. We targeted the
neural interactions of the perilesional cortical areas with the application of 4 different transcranial
direct current stimulation (tDCS) montages: conventional anodal, high-definition (HD) anodal, bi-
hemispheric, and SHAM. We aimed to find the optimal montage that would increase FC the most
and lead to the best clinical outcome. After comparison, the HD tDCS montage had the most effect
on motor outcome.The effect only appeared in the Action Research Arm Test and not in our
primary outcome which was the Fugl-Meyer assessment of the upper extremity (FMA-UE). On a
neurophysiological level, the MEP analyses pointed out a tendency of the HD montage to increase
the MEP amplitudes and reduce the rest motor threshold in the contralesional hemisphere rather
than having an effect in the ipsilesional hemisphere. The computation of FC confirmed previous
findings where connectivity was an early marker of good motor recovery, while its late arrival was
associated with worse recovery.
In a second study, we recruited patients with severe motor deficits. Here, a closed-loop setup
combining functional electrical stimulations (FES) contingent with a brain-computer interface
(BCI) aimed to prevent the secondary degeneration of the CST integrity. The motor scores did not
display significant differences between the BCI and a control group consisting of triggering the
FES based on the contingency of patients undergoing the real BCI training. However, BCI-FES
reduced spasticity of the patients as evaluated with the Ashworth score. On a group level, global
FC between the primary motor cortex and the rest of the brain was significantly reduced after
BCI-FES as compared to after SHAM treatment, and this reduction of global FC was correlated
with greater FMA-UE improvement. Regarding CST integrity, the two groups evolved similarly. FA
asymmetry showed that the CST of all patients did not deteriorate during the brief observation
time window. No significant difference was observed in the MEPs of the ipsilesional or the
contralesional side.
In a third study, we explored the feasibility of developing a neurofeedback training targeting FC
for clinical use. We aimed at finding an analysis pipeline that would allow to compute source FC
of a region of interest based on a low-density EEG coverage and a template MRI. We compared

2
several algorithms that would allow an FC reconstruction from only 19 channels. We used
numerical simulations of coherent sources as well as real datasets. Of these analyses, the use of
beamformer inverse solution emerged as the best performing under the constraints imposed. We
finally tested its performance in an independent dataset recorded from a low-density EEG using
dry-gel electrodes. However, we could not reproduce the findings of the previous dataset. This
may be due to the quality of the signal or to the design of the FC reconstruction, which was not
yet efficient enough.
In conclusion, this thesis allowed the observation of innovative treatments on the patients
themselves in the first two projects, and as a proof-of-concept in the third project. Although
promising, the three treatments still need more research to fine-tune their setup for more
efficient interventions. Faced with motor deficits that are still too present, it is important to
pursue research to perfect these interventions until the condition of the patients is improved. To
that end, from our studies the next steps are to 1) go in the main phase to compare HD-tDCS
against SHAM-tDCS, 2) add more BCI sessions to reach the usual amount of training, 3) perform
complementary FC analyses to observe the neural interactions between motor areas, 4)
implement a neurofeedback targeting FC with a high-density EEG using sponge electrodes.

3
Abstract (French)
L’accident vasculaire cérébral (AVC) est un problème de santé majeur. C’est l'une des principales
causes de handicap chez l'adulte. Un déficit courant après un AVC est l'hémiparésie du membre
supérieur controlatéral. Plus de 80 % des patients hospitalisés pour un AVC souffrent de cette
paralysie et, malgré une rééducation intensive, plus de 40 % des patients présentent toujours ce
déficit une fois arrivée à la phrase chronique. L'objectif de cette thèse est de comprendre
comment nous pouvons amplifier les effets de la rééducation pour réduire encore plus ces
déficits.
Des études antérieures ont montré des cibles possibles pour de nouveaux traitements. Les
patients souffrant de déficits moteurs graves présentent des lésions importantes du tractus
cortico-spinal (TSC) et une dégradation secondaire de la substance blanche dans les semaines qui
suivent l'AVC. Chez ces patients, induire une plasticité au niveau du TSC ou éviter la dégradation
secondaire peut permettre les processus de récupération motrice, qui sont généralement très
limités ou absents chez ces patients. Les patients avec des déficits moteurs légers ou modérés
présentent une FC élevée entre les zones motrices et le reste du cerveau dans les premières
semaines suivant l'AVC, ce qui est associé à une meilleure récupération clinique. Chez ces
patients, l'amélioration de la FC peut améliorer le résultat moteur au-delà de la règle
proportionnelle.
Dans la première étude, nous avons recruté des patients présentant des déficits moteurs légers
à modérés. Nous avons ciblé les interactions neurales des zones corticales périlésionnelles avec
l'application de 4 montages différents de stimulation transcrânienne à courant direct (StCD) :
anodale conventionnelle, anodale à haute densité (HD), bi-hémisphérique et SHAM. Nous avons
cherché à trouver le montage optimal qui augmenterait le plus la CF qui conduirait au meilleur
résultat clinique. Après comparaison, le montage anodale HD a eu le plus d'effet sur le résultat
moteur. L'effet n'est apparu que dans l'Action Research Arm Test et non dans notre premier
résultat primaire qui était l'évaluation Fugl-Meyer du membre supérieur (FMA-UE). Au niveau
neurophysiologique, les analyses PEM ont mis en évidence une tendance du montage HD à
augmenter les amplitudes PEM et à réduire le seuil moteur de repos dans l'hémisphère
contralésionnel plutôt que d'avoir un effet dans l'hémisphère ipsilésionnel. Le calcul de la FC a
confirmé les résultats précédents où la connectivité était un marqueur précoce de bonne
récupération motrice, tandis que son arrivée tardive était associée à une moins bonne
récupération.
Dans une seconde étude, nous avons recruté des patients présentant des déficits moteurs
sévères. Ici, une configuration en boucle fermée combinant des stimulations électriques
fonctionnelles (SEF) et une interface cerveau-ordinateur (ICO) visait à prévenir la dégénérescence
secondaire de l'intégrité du TCS. Les évaluations motrices n'ont pas montré de différences
significatives entre le groupe ICO et le groupe-contrôle qui consistait à déclencher la FES en
fonction de la contingence des patients bénéficiant du vrai entraînement ICO. Cependant, notre
montage a réduit la spasticité des patients, évaluée par le score d'Ashworth. Au niveau du groupe,

4
le FC global entre le cortex moteur primaire et le reste du cerveau était significativement réduit
avec notre montage, par rapport au traitement SHAM, et cette réduction de CF global était
corrélée à une plus grande amélioration de la perforomance au FMA-UE. En ce qui concerne
l'intégrité du TCS, les deux groupes ont évolué de manière similaire. L'asymétrie AF a montré que
le TCS de tous les patients ne s'est pas détérioré pendant la brève période d'observation. Aucune
différence significative n'a été observée dans les PEM du côté ipsilésionnel ou contralésionnel.
Dans une troisième étude, nous avons exploré la faisabilité du développement d'un entraînement
de neurofeedback ciblant la CF pour une utilisation clinique. Nous avons cherché à trouver un
algorithme qui permettrait de calculer la CF à la source correspondant à une région d'intérêt à
partir d'une couverture EEG à faible densité et d'un modèle IRM standardisé. Nous avons comparé
plusieurs algorithmes qui permettraient une reconstruction de la CF à partir de 19 canaux. Nous
avons utilisé des simulations numériques de sources qui étaient créées pour être cohérentes ainsi
que des ensembles de données réelles. Parmi ces analyses, l'utilisation de la solution inverse du
beamformer est apparue comme la plus performante sous les contraintes imposées. Nous avons
finalement testé ses performances dans un jeu de données indépendant enregistré à partir d'un
EEG à faible densité utilisant des électrodes à gel sec. Cependant, nous n'avons pas pu reproduire
les résultats du jeu de données précédent. Cela peut être dû à la qualité du signal ou à la
conception de la reconstruction de la CF qui n'était pas encore assez efficace.
En conclusion, cette thèse a permis d'observer des traitements innovants sur les patients eux-
mêmes dans les deux premiers projets, et comme preuve de concept dans le troisième projet.
Bien que prometteurs, les trois traitements nécessitent encore des recherches afin de permettre
leur utilisation en clinique. Face à des déficits moteurs encore trop présents, il est important de
poursuivre les recherches pour perfectionner ces interventions jusqu'à l’amélioration de l'état
des patients. Dans cette optique, les prochaines étapes de nos études sont les suivantes : 1)
passer à la phase principale pour comparer le HD-tDCS au SHAM-tDCS, 2) ajouter plus de sessions
de BCI pour atteindre un nombre plus standard d'entraînement, 3) réaliser des analyses
complémentaires de CF pour observer les interactions neuronales entre les zones motrices, 4)
mettre en place un neurofeedback ciblant la CF avec un EEG à haute-densité utilisant des
électrodes à éponges.

5
List of Abbreviations
ARAT Action Research Arm test
ADL Activities of Daily Living
BCI Brain-Computer Interfaces
BIHEM bi-hemispheric tDCS
CIMT Constraint-Induced Movement Therapy
CMT Cortical Motor Threshold
CONV conventional anodal tDCS
CST Corticospinal Tract
DTI Diffusion-Tensor MRI
DWI-MRI Diffusion-Weighted Magnetic Resonance Imaging
EEE Electroencephalography
EMG Electromyogram
EPI Echo-planar imaging
ES Electrical Stimulation
FA Fractional Anisotropy
FDI First dorsal interosseous muscle
FMA-UE Fugl-Meyer Assessment of the Upper Extremity
FC Functional Connectivity
FES Functional Electrical Stimulation
FIM Functional Independence Measure
fMRI Functional Magnetic Resonance Imaging
HD-tDCS High-Definition tDCS
I-waves Indirect Waves
LTD Long-Term Depression
LTP Long-Term Potentiation
M1 Primary motor cortex
MEG Magnetoencephalography
MGSM Maximal Grip Strength Measurement
MD Mean Diffusivity
MEP Motor Evoked Potential
NMES Neuromuscular Electrical Stimulation
NHPT Nine-Hole Peg Test
POOR Poor Group
M1 Primary Motor Cortex
PROP Proportional Group
rTMS Repetitive Transcranial Magnetic Stimulation
SHAM-tDCS tDCS with no stimulation
SWMT Semmes-Weinstein Monofilament Test
tDCS Transcranial Direct Current Stimulation
T0 Week of assessments pre-intervention
T1 Week of assessments post-intervention
T2 Clinical assessments 1 month after the last tDCS stimulation
T3 Clinical assessments 3 months after stroke onset
6
TMS Transcranial Magnetic Stimulation
VR Virtual reality

7
Table of Contents
Acknowledgments ............................................................................................................................ 1
Abstract (English) ............................................................................................................................. 2
Abstract (French) .............................................................................................................................. 4
List of Abbreviations ......................................................................................................................... 6
Table of Contents ............................................................................................................................. 8
List of Figures.................................................................................................................................. 11
List of Tables ................................................................................................................................... 13
1. Introduction............................................................................................................................ 14
1.1. Stroke ............................................................................................................................... 14
1.1.1. Clinical impact and societal cost .............................................................................. 14
1.1.2. Upper limb impairment: Recovery path .................................................................. 14
1.1.3. Clinical practice for upper extremity motor therapy in stroke patients .................. 15
1.2. Assessment of brain plasticity in humans ....................................................................... 17
1.2.1. Clinical evaluations of motor improvements ........................................................... 17
1.2.1.1. Fugl-Meyer Assessment of the Upper-Extremity (FMA-UE) ............................. 17
1.2.1.2. Action Research Arm test ................................................................................. 17
1.2.1.3. Jamar ................................................................................................................. 18
1.2.1.4. Nine-hole peg test............................................................................................. 18
1.2.1.5. Semmes-Weinstein Monofilament Test ........................................................... 18
1.2.1.6. Modified Ashworth Scale .................................................................................. 19
1.2.1.7. Functional Independence Measure .................................................................. 19
1.2.1. Diffusion-weighted imaging of the cortico-spinal tract ........................................... 19
1.2.2. Brain network connectivity ...................................................................................... 20
1.2.3. Transcranial magnetic stimulation (TMS) ................................................................ 21
1.3. Principles of brain plasticity............................................................................................. 22
1.3.1. Principle 1: Function loss from lesions to specific pathways ................................... 22
1.3.2. Principle 2: Secondary white matter tract degeneration ........................................ 24
1.3.3. Principle 3: Neural Reorganization ........................................................................... 25
1.3.4. Principle 4: Neural network interaction loss ............................................................ 26
1.3.5. Principle 5: Neural network plasticity ...................................................................... 27
1.3.6. Principle 6: Contralesional hemisphere influence ................................................... 27
1.4. New treatment concepts ................................................................................................. 29

8
 1.4.1. Repetitive transcranial magnetic stimulation .......................................................... 29
1.4.2. Transcranial direct current stimulation.................................................................... 30
1.4.3. BCI-FES Close-loop system ....................................................................................... 32
1.4.4. Neurofeedback training of FC .................................................................................. 33
1.5. Objectives of the studies ................................................................................................. 34
2. Results .................................................................................................................................... 35
2.1. Study 1 ............................................................................................................................. 35
2.1.1. Introduction.............................................................................................................. 35
2.1.2. Materials and methods ............................................................................................ 35
2.1.2.1. Participants ....................................................................................................... 35
2.1.2.2. Study design ...................................................................................................... 37
2.1.2.3. MEPs with TMS ................................................................................................. 37
2.1.2.4. Resting-state EEG and FC .................................................................................. 38
2.1.2.5. tDCS ................................................................................................................... 39
2.1.3. Results ...................................................................................................................... 39
2.1.3.1. Clinical effects ................................................................................................... 39
2.1.3.2. Neurophysiological effects................................................................................ 44
2.1.4. Discussion ................................................................................................................. 47
2.2. Study 2 ............................................................................................................................. 50
2.2.1. Introduction.............................................................................................................. 50
2.2.2. Materials and methods ............................................................................................ 50
2.2.2.1. Participants ....................................................................................................... 50
2.2.2.2. Study design ...................................................................................................... 51
2.2.2.3. Clinical assessment ........................................................................................... 52
2.2.2.4. High-density resting-state EEG ......................................................................... 52
2.2.2.5. MEPs with TMS ................................................................................................. 52
2.2.2.6. DTI analyses ...................................................................................................... 52
2.2.2.7. BCI-FES .............................................................................................................. 53
2.2.3. Results ...................................................................................................................... 54
2.2.3.1. Clinical effects ................................................................................................... 54
2.2.3.2. Neurophysiological effect ................................................................................. 60
2.2.4. Discussion ................................................................................................................. 62
2.3. Study 3 ............................................................................................................................. 66

9
3. Discussion ............................................................................................................................... 67
3.1. General synthesis............................................................................................................. 67
3.2. Spontaneous recovery and proportional rule ................................................................. 68
3.2.1. Fitters and non-fitters .............................................................................................. 68
3.2.2. The right time for the intervention .......................................................................... 69
3.3. Plasticity principles .......................................................................................................... 70
3.4. Thymic aspects in stroke survivors and rehabilitation .................................................... 73
3.5. Limitations ....................................................................................................................... 73
3.6. The future of innovative rehabilitation intervention ...................................................... 75
3.7. Future work ..................................................................................................................... 75
4. References .............................................................................................................................. 77
5. Appendix................................................................................................................................. 98

10
List of Figures
Figure 1. Proportional motor recovery in the upper limb ............................................................. 14
Figure 2. TMS applied over the motor cortex……... ....................................................................... 21
Figure 3. Corticospinal tract (CST) asymmetry……….. .................................................................... 23
Figure 4. Comparison between MEP response group MEP non-response group…….. .................. 23
Figure 5. Dynamic changes in the ratios of the fractional anisotropy…………………………………………24
Figure 6. Relationships between rFA change of the CST and the Motricity Index……………………….24
Figure 7. Structural changes associated with LTP and LTD….. ....................................................... 25
Figure 8. Tract-related white matter integrity of corticofugal pathways after stroke………. ......... 25
Figure 9. Mean absolute imaginary coherence between C3/C4 and all other electrodes……… .... 26
Figure 10. Structural and functional signatures of plasticity………….. ............................................ 26
Figure 11. Associations between network interactions and clinical improvement……………………..27
Figure 12. The interhemispheric imbalance model. ...................................................................... 28
Figure 13. Lidocaine anesthesia of undamaged motor cortex………. ............................................. 29
Figure 14. Schematic diagram of changes in neuron potential by tDCS…... .................................. 31
Figure 15. Putative mechanisms for the enhancement of long-term potentiation by tDCS. ........ 31
Figure 16. Illustration of typical BCI systems used in post-stroke motor rehabilitation…….. ........ 32
Figure 17. FMA-UE score at T0, T1, T2, T3….. ................................................................................ 40
Figure 18. FMA-UE change between T1 and T0, T2 and T0, T3 and T0………. ................................ 40
Figure 19. The percentage of maximum recovery in FMA-UE change. ......................................... 41
Figure 20. Distribution of the tDCS patients regarding the proportional recovery rule……….. ..... 41
Figure 21. ARAT score at T0, T1, T2, T3. ......................................................................................... 42
Figure 22. ARAT change between T1 and T0, T2 and T0, T3 and T0….. ......................................... 42
Figure 23. JAMAR change between T1 and T0, T2 and T0, T3 and T0. .......................................... 43
Figure 24. Nine hole peg test velocity change between T1 and T0, T2 and T0, T3 and T0….. ....... 43
Figure 25. FIM score at T0, T1, T2, T3….......................................................................................... 44
Figure 26. Increase of MEP in percentage between T1 and T0. .................................................... 45
Figure 27. FC in the ipsilesional M1/SMA among the 4 groups at T0, T1, between T1 and T0. .... 46
Figure 28. FC and FMA-UE relationship…....................................................................................... 46
Figure 29. FMA-UE score at T0, T1, T2……...................................................................................... 54
Figure 30. FMA-UE change between T1 and T0, T2 and T0…... ..................................................... 54
Figure 31. The percentage of maximum recovery in FMA-UE change………... ............................... 55
Figure 32. Distribution of the BCI patients regarding the proportional recovery rule .................. 55
Figure 33. BCI Fitters and non-fitters……........................................................................................ 56
Figure 34. ARAT change between T1 and T0, T2 and T0……….. ..................................................... 56
Figure 35. JAMAR change between T1 and T0, T2 and T0. ............................................................ 57
Figure 36. ASHWORTH score at T0, T1, T2……... ............................................................................ 57
Figure 37. ASHWORTH change between T1 and T0, T2 and T0…... ............................................... 58
Figure 38. MONOFILAMENT score at T0, T1, T2. ........................................................................... 58
Figure 39. MONOFILAMENT change between T1 and T0, T2 and T0………... ................................. 59
Figure 40. FIM score at T0, T1, T2……............................................................................................. 59
Figure 41. FIM change between T1 and T0, T2 and T0…... ............................................................ 60
Figure 42. FA asymmetry at T0, T1, between T1 and T0. Correlation FA asymmetry and FMA. ... 60
Figure 43. MEP change in the affected and unaffected hand between T1 and T0…….. ................ 61
11
Figure 44. Decrease of RMT in percentage in both hemisphere between T1 and T0……….. ........ 61
Figure 45. FC at T0, T1, between T1 and T0. Correlation between FC change and FMA change. . 62
Figure 46. Parallels between windows of plasticity in development and stroke .......................... 71

12
List of Tables
Table 1. Patient characteristics of study 1……................................................................................ 36
Table 2. Patient characteristics of study 2……….. ........................................................................... 51
Table 3. Patient characteristics of the DROP-OUT group of study 2…………... ............................... 51

13
 1. Introduction

1.1. Stroke

1.1.1. Clinical impact and societal cost

Stroke is one of the major health issues; it is among the top three causes of years of life lost living,
adult disability, and cost per person (Guilbert, 2003; Gustavsson et al., 2011; WHO, 2014). Hence
the need for evidence-based methods to allow better efficiency for the therapists and especially
for innovative treatments (Hatem et al., 2016).

1.1.2. Upper limb impairment: Recovery path

50% of stroke survivors will have chronic disabilities among which hemiplegia of various degrees
of the extremities. More specifically, 80% of the patients will suffer from upper limb impairment
in the acute phase and 40% in the chronic phase (Donkor, 2018; Hatem et al., 2016). Every patient
requires individualized support in their rehabilitation training, which should target specific
expected improvements.
Based on the clinical Fugl-Meyer Assessment of the Upper Extremity (FMA-UE), a subpopulation
of patients follow a proportional rule in their recovery (Prabhakaran et al., 2008; Winters et al.,
2015). This rule predicts that 6 months after stroke onset, patients will recover ≈70% of the
difference between their initial FMA-UE score and the maximum FMA-UE score, which is defined
as the proportional recovery rule:
ΔFMA = 0.7 x (FMA{𝑚𝑎𝑥𝑖𝑚𝑢𝑚} − 𝐹𝑀𝐴{𝑖𝑛𝑖𝑡𝑖𝑎𝑙} )
However, some patients do not follow the proportional rule. They have severe initial impairment
and little to no recovery. This group was defined as the poor recovery group (figure 1).

Figure 1. Proportional motor recovery in the upper limb. Illustration of different

recovery curves of patients with initially severe upper limb impairment who have
recovery as predicted (blue) or poor recovery (red). The recovery curves of these
groups of patients suggests that the factors important for recovery are different
from those responsible for initial severity (Ward, 2017)

14
This rule was later criticized due to mathematical coupling between the predictor and the
outcome, arising because the outcome is a difference measure between the FMA at 6 months
and the initial FMA and is thus not independent from the predictor (initial FMA). This leads to a
spurious overestimation of the model fit (Hawe et al., 2019; Hope et al., 2019). Moreover, plateau
effects of the FMA score also contribute to this overestimation. However, the observation of
proportionality of the recovery to the initial impairment in some patients, and the presence of
two distinct recovery patterns remains valid and is unaffected by these concerns (Bonkhoff et al.,
2021)

1.1.3. Clinical practice for upper extremity motor therapy in stroke patients
Regardless of the degree of the deficit, motor therapy has been found necessary to rehabilitate
stroke survivors as spontaneous plasticity often does not allow them to fully recover from their
initial deficit (Takeuchi & Izumi, 2015). Indeed the more intensive the therapy is, the faster the
maximum recovery potential seems to be obtained, for instance regarding selective movements
(Veerbeek et al., 2014). After an intensive rehabilitation period that occurs during the acute and
sub-acute phase, many patients should still be able to pursue the therapy to obtain further
progress in the recovery or to avoid later complications such as spasticity, pain, or falls (Cramer
& Riley, 2008). The therapy is essentially based on learning and repeating functional exercises in
an appropriate environment (Takeuchi & Izumi, 2013). Motor rehabilitation can be classified into
three parts. The first part of the rehabilitation program is to express the full recovery potential of
the patient. The second part, needed in patients with insufficient recovery, focuses on
compensatory strategies that would enhance the patient’s autonomy in activities of daily living
(ADL). This can be supported with external aids such as orthoses. Adaptations of the home
environment of the patients lead to a reduced rehospitalization rate (Burke et al., 2014). The third
part aims to maintain the gains from the intensive rehabilitation phase (Pollock et al., 2015;
Veerbeek et al., 2014). If done rapidly after stroke onset as soon as the patient is capable of
enduring therapy, and if scheduled at least once a day and 5 days a week, these rehabilitation
efforts have a valuable impact on stroke motor outcome. These outcomes are observed through
the clinical assessment and transferred in the ADL of the patients when they return home(Pollock
et al., 2014).
In parallel with therapy, motor evaluation is key to updating and fitting the learning objectives
with the patient. Early motor evaluation can also help to estimate patients' possible recovery.
As no specific rehabilitation methods have demonstrated superiority compared to other
approaches, and knowing that intensive training increases the chance of motor recovery, the
expertise is left to the therapist to assess and analyze the individual situation of the patient and
tailor a suitable treatment program in order to optimize all the functional abilities of the patient
(de Morand, 2014).
For upper extremity impairment, a first concern is preventing the shoulder deterioration to avoid
later pain, inflammation, diastasis, and complex regional pain syndrome (Crichton et al., 2016; de
Morand, 2014; Fery-Lemonnier, 2009). At later stages, a further goal or therapy is to avoid or

15
reduce joint contracture and spasticity, which typically develops over time in patients with poor
recovery of their upper limb and especially their hand (Winstein et al., 2016).
There exist many sensory-motor training methods which we cannot cover extensively here. I will
however briefly mention a few frequently used approaches and tools.
Constraint-induced movement therapy (CIMT) seems to enable a robust gain in motor abilities,
but it can be applied only in patients with mild to moderate deficits. It consists of constraining the
healthy upper extremity to induce an increased use of the paretic upper extremity in therapy and
ADL. It is a multidisciplinary approach where the therapist, medical staff, and the patient work
together. Multiple versions exist where the intensity, the duration, the frequency, the type of
constraint (bandage for the whole upper limb or just a padded mitt for the hand) can vary (Furlan
et al., 2016; Veerbeek et al., 2014).
Robot-assisted rehabilitation provides to the patient a playful way to train the upper limb either
through an end-effector robotic mechanism or an exoskeleton. It allows training specific
movements and improving muscle strengths. It is thus an interesting option for therapy (Zhang et
al., 2018). It must be stressed, however, that it cannot stand by itself to improve all ADL but needs
to be integrated into an individually tailored multidisciplinary rehabilitation program.
Virtual reality (VR) offers the patients interaction inside a virtual environment either immersively
through VR glasses or non-immersively with an environment projected on a screen. VR can be
specifically designed for rehabilitation purposes or non-specific (i.e., games designed for all public
entertainment purposes). Specific VR implements exercises that are task-oriented, possess
features that monitor body compensations to encourage the paretic arm, provide feedback on
their results and their performance and adjust the difficulty to the individual level of the patient
(Maier et al., 2019). Specific VR following these principles has a significant impact on upper limb
motor recovery both in the arm function and ADL.
Electrotherapy is the application of electrodes on a muscle motor point or belly to deliver
electrical current at specific frequencies, intensities, and pulses width in order to either contract
or relax the muscles. In the case of a goal of motor improvement, three levels of stimulation are
applicable. Electrical stimulation (ES) is preferably used in sports rehabilitation to improve muscle
strength and joint amplitude. Neuromuscular electrical stimulation (NMES) induces muscle tetany
of contraction and is more commonly used with non-athlete patients. ES and NMES are often
used interchangeably in the literature. Nevertheless, ES for muscle training in athletes is more
likely referring to low-frequency electrical stimulation (Babault et al., 2011; Doucet et al., 2012a).
Functional electrical stimulation (FES) is similar to NMES combined with a functional task. It is
more indicated in neurorehabilitation (Doucet et al., 2012b). NMES/FES has mixed effects
depending on the muscles it is stimulating. The stimulation on the leg has a better effect than in
the arm (Veerbeek et al., 2014). NMES reduces the risk for glenohumeral subluxation when
applied on the shoulder, improves the forearm extensor mobility, and enhances specific exercised
movements (Veerbeek et al., 2014).

16
 1.2. Assessment of brain plasticity in humans
Given the limited motor recovery and chronic handicap in approximately 50 % of stroke patients
(Donkor, 2018), there is a great need for new treatment options. In order to be able to propose
more efficient therapies in the future, we first need a detailed understanding of how the brain
reorganizes after stroke to achieve functional gains. There are several methods that we can use
to observe recovery and plasticity, both on the behavioral and the neural level. In this chapter, I
will give a summary of the most common techniques applied in human patients.

1.2.1. Clinical evaluations of motor improvements

1.2.1.1. Fugl-Meyer Assessment of the Upper-Extremity (FMA-UE)

The FMA is a functional scale performed in clinical practice to measure the recovery of the upper
and lower extremity in stroke patients. It is the most frequently used in clinical studies for its
comprehensive items, its low variability between performers, and its sensibility to change. The
FMA-UE evaluates the motor function over 66 points and the sensory function over 24 points.
Each function category contains multiple items scored from 0 to 2 (0 = cannot perform, 1 =
performs partially, 2 = performs fully). These items evaluate either the movement ability, the
reflexes, or the speed of functional gestures of the shoulder, the elbow, the wrist, and the fingers.
The reliability of the UE motor function makes the FMA-UE a gold standard in clinical trials with
an intrarater variability of 0.97, and interrater variability of 0.995-0.996. Even though multiple
grasping items are present, many fine finger movements are missing to properly evaluate the
distal recovery of the upper limb. Moreover, the validity of using the FMA-UE to observe the
evolution of these fine finger movements is not clear, while the Action Research Arm test (ARAT)
shows such evolution calling for its use to complement the FMA-UE. Despite this limitation, FMA
is responsive enough to grab the evolution of the patient. This evolution is found associated with
activities of daily living. Due to this responsiveness and within a span change of -5 to 6.6 points
over 66, inferences can be drawn from FMA points change (Gladstone et al., 2002).
The initial FMA-UE score helps predict whether a patient will follow a so-called proportional
recovery pattern or evolve according to the poor pattern. Indeed, the poor group (POOR) had an
initial FMA-UE score between 0 and 17, whereas the proportional group (PROP) scored between
0 and 65 (Winters et al., 2015). Hence, patients with an initial FMA over 17 have a very high
probability of proportional recovery.

1.2.1.2. Action Research Arm test

The ARAT is a functional assessment of the upper limb impairment in its fine distal movement. It
is standardized for neurological pathologies such as stroke, brain injury, multiple sclerosis, or
Parkinson’s disease (Platz et al., 2005). It contains 4 groups of tests for a total of 19 movements.
The groups cover grasp, grip, pinch, and gross movement. For each item, a score from 0 to 3 can
be given (0 = no movement, 3 = movement performed normally) (Yozbatiran et al., 2008). Clinical
relevance is seen if a difference of 5.7 out of 57 points is scored (Van der Lee et al., 2001). It is
highly correlated with the FMA-UE motor category and has a good reliability with high intra and
17
interrater scores. The ARAT is responsive in stroke patients at all stages. One limitation of this test
is the ceiling and floor effects among stroke patients that have either a good or poor recovery
(Van der Lee et al., 2002). These effects can be overcome with grip strength and the nine-hole
peg test (Heller et al., 1987).

1.2.1.3. Jamar
The Jamar dynamometer allows the maximal grip strength measurement (MGSM). The grip
strength assessment is well suited to stroke populations and can give an interesting predictor
value of upper limb recovery (Bertrand et al., 2015). We can observe a decrease of strength
difference between left and right hands during the first year after stroke onset with a larger
improvement in the first 6 months (Stock et al., 2019). Moreover, in chronic stroke populations,
grip strength has shown its association with arm strength making it a simpler tool to assess and
represent the entire upper extremity muscle weakness (Ekstrand et al., 2016).
The elbow is bent at 90° and can be held by the therapist, while the hand is gripping for several
seconds the dynamometer. Different protocols exist to determine the MGSM, among them is to
perform 3 measures, then keep the best score (Hanten et al., 1999). This assessment allows the
quantitative measure of distal strength of the upper extremity. A minimally clinically important
difference in the affected side is seen for a difference of 5.0 kg if it is the dominant hand and 6.2
kg if it is a non-dominant hand (Lang et al., 2008; Roberts et al., 2011). The interrater and
intrarater, test-retest reliability is high (Bertrand et al., 2007; Boissy et al., 1999).

1.2.1.4. Nine-hole peg test

The nine-hole peg test (NHPT) is a finger dexterity test. It is recommended to assess the speed of
fine movements during the sub-acute and acute phases in stroke patients. Interrater and
intrarater test-retest reliability (Chen et al., 2009; Heller et al., 1987) and responsiveness (Beebe
& Lang, 2009) are high. The test consists of placing 9 pegs into the holes of a pegboard and
removing them again. NHPT times the duration that is taken to perform the task in the healthy
hand than in the affected. Patients perform one practice trial before the timing trial in each hand.
A minimal detectable change in the NHPT is 32.8 seconds (Chen et al., 2009). A modified NHPT
recently emerged where the pegs already stand in a first 9-hole block. The second 9-hole block is
numbered and patients are asked to place the pegs in a specific numerical order (Johansson &
Häger, 2019).

1.2.1.5. Semmes-Weinstein Monofilament Test

The Semmes-Weinstein Monofilament Test (SWMT) addresses the somatosensory impairment by
assessing light touch sensation. It has a high interrater and intrarater reliability (Suda et al., 2021).
Twenty flexible nylon monofilaments of constant length with different diameters allow the
evaluation of light touch with multiple weights mainly on the palm of the hand or the arch of the
foot (Bell-Krotoski et al., 1995). The 20 monofilaments are grouped in 5 levels depending on the

18
nylon thickness ranging as follows: normal sensation, diminished light touch, diminished
protective sensation, loss of protective sensation, deep pressure sensation only (Coast, 2011).

1.2.1.6. Modified Ashworth Scale

The Modified Ashworth Scale assesses the muscle spasticity in neurological pathologies on a 5-
level-scale, scoring the resistance of a muscle following a sudden stretch during a passive
movement (Bohannon & Smith, 1987). Spasticity increases over time as the patients move further
from stroke onset (Kong et al., 2012; Welmer et al., 2010). Its monitoring can help to understand
the difficulties of the patients’ clinical performances. The test-retest in the upper extremity is high
for the elbow and the wrist (Gregson et al., 1999). The intrarater reliability is moderate while the
interrater reliability is poor (Blackburn et al., 2002).

1.2.1.7. Functional Independence Measure

The functional independence measure (FIM) brings together 18 motor and cognitive tasks to
evaluate the level of assistance a patient needs in his or her daily life activity. This measure helps
to understand the degree of disability he or she endures. A scale of 7 grades scores each item
from complete dependence to complete independence (Oczkowski & Barreca, 1993). It is
commonly performed during the stay of stroke patients (Ring et al., 1997). The minimally clinically
important difference is 22 points out of 126 on the FIM total score and more specifically 17 out
of 91 points for the motor subscale (Beninato et al., 2006). Total FIM has a good sensitivity to
change and has no flooring or ceiling effect (Dromerick et al., 2003). The motor FIM presents a
small flooring effect (Hsueh et al., 2002). Caution is to be taken for patients who suffer from
neglect or aphasia as FIM has very few cognitive items that do not allow for a high sensibility of
cognitive deficit (Ring et al., 1997).

1.2.1. Diffusion-weighted imaging of the cortico-spinal tract

Diffusion-weighted magnetic resonance imaging (DWI-MRI) is an MRI sequence that takes into
account the diffusion of water molecules. Changing the strength of the pulsed gradients infers on
the diffusion measurement that generates sequences with specific diffusion coefficients in each
voxel (Merboldt et al., 1985). This then allows the display of the diffusion process in the tissues
showing details about their state (Taylor & Bushell, 1985). Acute stroke lesions lead to cytolysis
which enhances the signal.
It is furthermore possible to model water diffusion with diffusion-tensor MRI (DT-MRI or DTI).
Several parameters can be computed from DTI modelization among which there is the mean
diffusivity (MD) and the fractional anisotropy (FA) (Özarslan et al., 2005). MD is a model-
independent parameter representing the average magnitude of the molecular diffusion in a voxel.
FA however depends on the eigenvalues of the tensors and gives information about the direction
of the diffusivity. The more FA tends toward 0 the more the voxel is isotropic; on the contrary,
the more FA tends toward 1 the more anisotropic is the voxel (Özarslan et al., 2005; Valsasina et
al., 2005). Intact white matter tracts show high FA, while damage leads to a reduction. Thus FA

19
can be used as a microstructural measure of tract integrity, which permits researchers to follow
the progression of neurological diseases group-wise and time-wise (Keihaninejad et al., 2013;
Smith et al., 2006).
We can further extend the analysis of anisotropy to multiple voxels to appreciate the behavior of
the diffusion inside an observed volume. This creates the possibility to see how fiber tracts in the
brain are organized (Basser & Pierpaoli, 2011). From the tensors, fibers can be tracked by
following the orientation of the diffusion. When putting one after another, pathways are thus
made visible (Jeurissen et al., 2019). These algorithms of tractography aim to reconstruct in 3D
white matter tracts (Girard et al., 2020). This visualization can help assess the anatomy,
development, and function of the brain such as the modification in white fiber tracts in many
neurological pathologies (Newton et al., 2006).
Together with FC, assessment of structural connectivity with DTI gives a more holistic grasp on
the brain behavior comprehension. In the case of stroke and motor impairment, FA asymmetry
of the corticospinal tract (CST) predicts upper extremity improvement. This asymmetry ratio could
differentiate stroke patients with severe deficits who recover poorly from those with proportional
recovery (Buch et al., 2016a).
It appears that early assessment of CST integrity with DTI could help tailor an individualized
rehabilitation program that fits best the patient’s needs. Then, CST can become a target to be
treated with the proper therapies (Guggisberg et al., 2017b).

1.2.2. Brain network connectivity

The brain is a network with densely interconnected nodes. Simulations of the stroke brain suggest
that damage to parts of the network leads to widespread changes in activity in the entire network
(Alstott et al., 2009; Honey & Sporns, 2008), which likely influences behavior. Furthermore,
animal research has shown that plasticity leads to new synaptic connections and thus to a
reshaping of the network. It is thus crucial to assess these network processes in humans, if we
want to understand plasticity.
To do this, we investigate the so-called connectivity of the brain. There are three levels of
connectivity that can be distinguished: structural, functional, and effective connectivity.
Structural connectivity studies the anatomical neural fibers connecting neurons and brain areas.
Functional connectivity (FC) computes the statistical dependency of activity patterns at two or
more brain areas. Effective connectivity attempts to understand the influence that one neural
system has on another, by integrating observations of FC into a generative model of neural
activity (Friston, 1994; Horwitz, 2003).
When recording the brain activity at rest, i.e. without demanding any specific task, one can
observe spontaneous activity of the brain through neuroimaging such as functional magnetic
resonance imaging (fMRI), magnetoencephalography (MEG), or electroencephalography (EEG)
(Guggisberg et al., 2011; Westlake et al., 2012). Even at rest, spontaneous brain activity is
organized into coherent network, such that brain areas responsible for the same function also
interact preferentially at rest (Greicius et al., 2003). These resting-state interactions can change

20
after specific tasks or at various timepoints in the different phases of stroke (Dubovik et al.,
2012a). FC can measure these neural network changes. FC can thus track the brain network, in
particular also during motor learning. Offering a system where the patient could target and
directly manipulate FC is an innovative tool in rehabilitation strategies (described in 4.4).
Many different measures of FC exist, depending on the imaging modality (fMRI, EEG, MEG) and
on the particular research question. When using EEG or MEG to compute FC, signal source
reconstruction is needed to estimate the time course of the cerebral signal from what is recorded
on the scalp. Again several algorithms of source reconstruction are available, see (Guggisberg et
al., 2014; Hadjipapas et al., 2005; Sekihara et al., 2002; Steinstrater et al., 2010; Van Veen et al.,
1997).

1.2.3. Transcranial magnetic stimulation (TMS)

TMS is used in clinical practice and research to assess brain excitability and CST integrity in
multiple neurological and psychiatric disorders (Groppa et al., 2012). With a magnetic coil placed
on the scalp of the patient, a changing magnetic field induces an electric current at a determined
brain area to activate cortical neurons (Barker et al., 1985). As seen in figure 2, the induced
current leads to a depolarization of the membrane potential triggering an action potential
(Groppa et al., 2012).

Figure 2. TMS applied over the motor cortex preferentially activates interneurons oriented in a plane parallel to the brain
surface. This placement leads to a transynaptic activation of pyramidal cells evoking descending volleys in the pyramidal
axons projecting on spinal motoneurons, also termed the CST. Motoneuron activation in response to corticospinal volleys
induced by TMS leads to a contraction in the target muscle evoking a MEP on electromyography recorded by using surface
electrodes applied over the muscle belly. Its peak-to-peak amplitude is used to estimate excitability of the CST (Klomjai et
al., 2015).

21
TMS comes with different coil shapes where the figure-of-eight-coil gives a more focal magnetic
field for the brain and a four-leaf coil a more focal field for peripheral nerves (Groppa et al., 2012).
One typical example is stimulating with single pulses of the primary motor cortex (M1) at the
hand-knob (Barker et al., 1985; Groppa et al., 2012). The motor evoked potential (MEP) in the
muscle elicited by TMS permits the observation of the CST integrity and can be recorded with
electromyography electrodes (Groppa et al., 2012; Rossini & Rossi, 2007). A specific measure that
is assessed with MEPs is the cortical motor threshold (CMT) to better understand its excitability
(Groppa et al., 2012). As a prognostic tool, the same parameter could give insights on the level of
motor recovery in stroke patients and becomes complementary to DTI for prognostic evaluation
(Dimyan & Cohen, 2010). When MEPs can be elicited in acute or subacute stroke patients, a good
motor recovery is predictable (Cakar et al., 2016), while the absence of MEPs is associated with
poor motor recovery (Blabe et al., 2015; Hendricks et al., 2002; Jo et al., 2016).
There are more advanced protocols to assess brain plasticity using dual pulses (Lazzaro et al.,
2013), but they will not be of primary interest in this work.

1.3. Principles of brain plasticity

The brain has an intrinsic ability to adjust to new environmental demands, an ability called
plasticity (Fuchs & Flügge, 2014). Plasticity involves both structural and functional changes in
neural circuitry which are mutually dependent. Structural plasticity takes into consideration the
neuroanatomical elements of the brain and their physical connections. Functional plasticity
considers the interaction and adaptation of brain activity.
It is further useful to distinguish between spontaneous and experience-dependent plasticity
(Cramer, 2008). Spontaneous plasticity refers to neural remodeling taking place during a limited
time window after the injury with or without therapy and which usually accounts for a large
proportion of clinical improvement in patients. However, these spontaneous neural changes can
also become maladaptive for recovery. Hence, neurorehabilitation and intensive therapy take a
great opportunity of a short aperture during the first months after stroke to facilitate adaptive
spontaneous plasticity (Carey et al., 2019). Experience-dependent plasticity refers to a continuous
process of neural remodeling influenced by the individual’s experience. Such structural
modification occurs rapidly during the experience in itself and can last if it is sufficiently repeated
(Fu & Zuo, 2011; Rossiter et al., 2014).
The techniques of assessment of brain plasticity in humans introduced in the previous chapter
have enabled increasing detailed insights into how the brain achieved repair and recovery. In this
chapter, I will summarize some influential principles.

1.3.1. Principle 1: Function loss from lesions to specific pathways

Right after stroke onset, the cerebral lesions lead to neuronal death and, thereby, to axonal
degradation. A loss of function due to a specific lesion either in the grey or white matter has an
impact on the clinical outcome of the patients (Fuchs & Flügge, 2014; Rossini et al., 2003).

22
Neuroimaging studies, in particular with stroke patients, showed that integrity loss of specific
white matter tracts can lead to severe, specific behavioral impairment and poor successive
recovery (Lindenberget al., 2010; Rehme et al., 2015). A great example is seen with the disruption
of the CST that plays an important role in motor recovery for stroke patients. The CST links the
motor areas and the spinal cord thanks to axons where M1 provide 40% of the corticospinal fibres
(Rothwell, 2012). The tract alteration reduces the synaptic transmission that creates the deficit.
To counteract this transmission weakening, experience-dependent plasticity as done in
rehabilitation has the potential to strengthening the connectivity between neuronal assemblies
(Dobkin, 2007b; Markram et al. 2011). When assessed with DTI or TMS, patients who present
severe CST disruption turn out to show particularly poor recovery of their motor ability (figure 3
and 4). The CST lesion load can be represented in DTI through FA asymmetry between the ipsi
and the contralesional hemisphere where a more impairment corresponds to a greater
asymmetry. That is how patients with poor recovery were found to have higher CST asymmetry
than patients who recovered proportionally to their impairment (Buch et al., 2016b; Feng et al.,
2015). The severity of the disruption of the CST is thus the main predictor of poor or proportional
types of motor recovery discussed in 1.2 (Buch et al., 2016b; Jo et al., 2016; Plow et al., 2015; Puig
et al., 2017; Zheng & Schlaug, 2015).

Figure 3. Corticospinal tract (CST) asymmetry at 2 weeks
correlated with the severity of motor deficits at 3 months
after stroke (adapted from Buch, 2016).
Figure 4. Comparison between patients showing an MEP response to TMS of
the ipsilesional M1 (MEP response group) and patients having absent MEP
response (MEP non-response group) (Jo et al., 2016). Values are presented
as mean±standard deviation or number (%). MEP, motor evoked potential;
FMA, Fugl-Meyer Assessment scale.

23
The disrupted CST thus becomes a target in finding therapies allowing the enhancement of CST
integrity. Such strategies are already seen in clinical practice through Kabat methods (Pollock et
al., 2014) where the patient is asked to initiate a particular movement while the therapist
accompanied the gesture more or less active depending on the patient’s motor ability. In some
innovative studies, linking the patient cerebral intention to sensory feedback is achieved via
combined brain-computer interface (BCI) systems (described in 4.2). In both approaches, afferent
and efferent inputs between the brain and the upper limb are externally provided in the hope of
mimicking an integrated tract.

1.3.2. Principle 2: Secondary white matter tract degeneration

After the hyperacute phase, secondary degeneration of the ipsilesional white matter fibers after
stroke is observed with neuroimaging based on DTI and is also known as the Wallerian
degeneration (Hermann & Chopp, 2014). This degeneration begins few days after stroke onset
and sometimes continues during the subacute phase affecting the axonal skeleton and the
membrane leading to a myelin degradation and a decay of these fibers (Yu et al., 2009) (figure 5).
Patients showing poor recovery in the chronic phase have extensive initial stroke-induced damage
to the CST and then notable secondary white matter degradation in the lesioned hemisphere
(Feng et al., 2015; Guggisberg et al., 2017b) (figure 6).

Figure 5. Dynamic changes in the ratios of the fractional
anisotropy (rFA) between the affected and unaffected
sides of the CST in individual stroke patients (adapted
from Yu et al., 2009).
Figure 6. Plots of relationships between the changes in
the ratios of fractional anisotropy (rFA) of the CST within
the first 2 weeks and the Motricity Index (MI) of the
affected side (adapted from Yu et al., 2009)

On the other hand, observing the secondary degeneration process in the early stage of stroke can
predict the motor outcome of the patient (Bigourdan et al., 2016). Hence, due to severe damage
of the CST, a delayed decaying process is observed in the affected hemisphere leading to a
degeneration of white fibers, cortical atrophy, and poor motor recovery (Doughty et al., 2016;
Guggisberg et al., 2017b; Koch et al., 2016; Zhang et al., 2014).

24
 1.3.3. Principle 3: Neural Reorganization
While the neural lesion and degradation occur, the recovery processes starts. Indeed, the brain
reacts to stroke-induced lesions and is able to recover some of the lost functions. One important
mechanisms is synaptic plasticity through long-term potentiation (LTP) and long-term depression
(LTD). LTP and LTD occur at the microscopic level, where an activation of the pre-synaptic neuron
during a short time window just before the postsynaptic neuron leads to long-term strengthening
of the synaptic weight and thus to a potentiation of this neural connection (figure 7). The re-
modelization of this circuit can lead to the learning of new skills which once stabilized can be
considered memorized (Nicoll, 2017). Conversely, if the activation of the presynaptic neuron
consistently occurs after the activation of the postsynaptic neuron, the synaptic weight will be
decreased, leading to LTD with a weakening of the synaptic weight linked to a reduction of the
associated activity. The decreased efficiency of the synaptic circuit leads to the loss of the
associated skill (Lüscher & Malenka, 2012). The more an activation sequence is repeated, the
longer the potentiation or depression will last. Together, these synaptic circuit changes have an
impact on the grey matter but also on the white matter structure. Indeed at a more macroscopic
level, white tracts are also receptive to training and reshaping. In an animal study, the motor
outcome of rats that had a stroke was correlated to cortical and CST reshaping (Pekna et al.,
2012). It has been seen that animals who suffered from stroke could gain new connections. Fibers
of the CST that come from other sites than the primary motor cortex, fibers connecting to the
cerebellum or the cortico-rubral tract are all elements that compensate for the loss of the main
CST. Thus, the residual structural connections of a specific network provide possibilities of
compensation for stroke damage and support patients’ recovery (Cunningham, et al., 2015;
Lindenberg et al., 2010; Wessel & Hummel, 2017) (figure 8).

Figure 7. Structural changes associated with LTP and LTD. Figure 8. Tract-related white matter integrity of
(A) Synaptic strength correlates with spine volume and the corticofugal pathways after stroke. Estimated means of
area of the postsynaptic density (orange). (B) LTP can also tract-related fractional anisotropy (FA) as a measure of
lead to the appearance of new spines. Within 30 min of structural integrity with 95% confidence intervals (Schulz
triggering LTP (30 stimuli applied to the presynaptic axon et al., 2017).
at 10-sec intervals paired with depolarizing current
injection into the postsynaptic neuron; black bar) of a
synaptic connection in the hippocampus, new spines
appear (Lüscher & Malenka, 2012).

25
 1.3.4. Principle 4: Neural network interaction loss
Once a brain location is damaged, communication with that particular cerebral area will diminish
or cease. Furthermore, damage to white matter tracts disturbs communication between
structurally intact areas. Thus, not only the damaged area shows modifications, but also
structurally intact, more or less remote areas will have altered communication (Fuchs & Flügge,
2014). This is why stroke is considered a network disease. FC (described in 1.2.2.) is one of the
means to assess such alterations in neural interactions. Reduced or disrupted FC of areas of
interest had specific impacts on neurological deficits as observed with the decrease of FC between
motor areas in stroke patients that were linked to their motor deficits (Westlake et al., 2012). This
was the case no matter whether FC of neural activity was quantified with functional magnetic
resonance imaging (Fan et al., 2015; Grefkes et al., 2008; Zhang et al., 2014) or EEG (Dubovik et
al., 2012a) (figure 9).

Figure 9. Mean absolute imaginary coherence (±standard error of

mean) between the central electrode (C3/C4) of the affected
hemisphere (black line) and all other electrodes was selectively
reduced in the alpha frequency range (8-12 Hz) as compared to
IC between the central electrode of the unaffected hemisphere
(dark grey) (p=0.0018, paired t-test) or to IC of healthy controls
(light grey)(p=0.023, unpaired t-test) (adapted Dubovik et al.,
2012b).

FC changes are in turn dependent on the underlying white matter tract architecture (figure 10).
Reduced FC in motor areas was associated with loss of CST integrity whereas enhanced FC took
place in patients with less damage CST (Guggisberg et al., 2017b; Rosso et al., 2013). Thus,
deteriorations in the communication of structurally intact brain areas with other areas of the
network in stroke patients result in neurological deficits.

GOOD Figure 10. Multimodal assessments. Structural and functional signatures of

plasticity were related to each other. Patients with greater CST integrity had
POOR proportionally greater FC between perilesional areas of the affected
hemisphere and the rest of the brain (Guggisberg et al., 2017a).

26
 1.3.5. Principle 5: Neural network plasticity
The preserved neural tissue shows dynamic adaptations in its interaction with other brain areas
and participates in adaptive brain plasticity. The increase of such communications is related to
clinical improvement during rehabilitation (De Vico Fallani et al., 2013; Nicolo & Rizk, et al., 2015;
Westlake et al., 2012) (figure 11). For instance, an increase of FC between perilesional motor areas
and the rest of the brain occurring in the first weeks after stroke was correlated to better clinical
recovery of upper limb motor scores in stroke patients at 3 months (Nicolo & Rizk et al., 2015).
Furthermore, patients with proportional recovery had enhanced neural interaction (i.e.,
enhanced FC ) of perilesional cortical motor areas compared to the poor group (Guggisberg et al.,
2017a). It is thought that the cortical and axonal remodeling taking place after stroke leads to an
enhancement of synchronized neural oscillations that has a reinforcement effect on their
projections (Carmichael & Chesselet, 2002). Thus, perilesional areas can compensate for lesioned
tissue by recruiting other brain areas through enhanced interactions, leading to improved clinical
outcomes.

Figure 11. Associations between network interactions and clinical improvement were regionally specific. A voxel-wise correlation
between WND and clinical recovery shows that only voxels around motor areas correlated with motor improvement. Functional
maps are thresholded at P < 0.05, uncorrected, to visualize the full extent of network predictors.

1.3.6. Principle 6: Contralesional hemisphere influence

The role of the contralesional hemisphere in post-stroke recovery is controversial. It is connected
to the lesioned hemisphere, and, thus, depending on the time after stroke, it can influence the
recovery process. In the case of motor function, the role of the contralesional hemisphere has
been investigated with MEPs in response to TMS (Kobayashi & Pascual-Leone, 2003; Rossini et
al., 1994; Trompetto et al., 2000). This showed that the contralesional hemisphere is more
excitable at the expense of the ipsilesional hemisphere in some patients (figure 12).

27
 Figure 12. The interhemispheric imbalance model, as represented by a
diagram of the rat brain with unilateral cortical stroke, predicts that
each side of the brain inhibits the other equally. After a stroke, the
inhibition coming from the stroke-affected hemisphere is decreased
(interhemispheric dashed line) alongside a decreased excitability in the
peri-lesional tissue. The unaffected hemisphere therefore becomes more
excitable and exerts a stronger inhibition onto the peri-lesional tissue.
The unaffected hemisphere therefore becomes more excitable and
exerts a stronger inhibition onto the peri-lesional tissue. Modulating this
zone of inhibition in the peri-infarct tissue appears to be a potential
target for stroke therapy (Boddington & Reynolds, 2017).

Moreover, applying a conditioning pulse to the contralesional hemisphere was found to reduce
the amplitude of MEPs obtained from the ipsilesional hemisphere in chronic stroke patients with
mild motor impairment (Murase et al., 2004). This led to the concept of an excitability balance
between two hemispheres with mutual interhemispheric inhibition through transcallosal
pathways stating that, in healthy humans, each hemisphere inhibits the other during its
activation. In stroke patients, it is thought that the overexcitability of the contralesional
hemisphere inhibits the lesioned hemisphere and that this contributes to the neurological deficits
and less recovery (Jo et al., 2016; Kobayashi & Pascual-Leone, 2003; Murase et al., 2004). This
concept of interhemispheric imbalance led to treatment attempts with noninvasive brain
stimulation, where the excitability of the contralesional hemisphere was inhibited (Corti et al.,
2012). However, mixed results came out of these studies which could be explained by differences
in age of the study participants, the volume of the lesion, and the different times of stimulation
after stroke onset (Cunningham et al., 2015; Rossini et al., 2003). Furthermore, the severity of
motor impairment may also lead to the differential implication of the contralesional hemisphere
(Di Pino et al., 2014). Thus, the inhibition of the contralesional hemisphere may be effective in
some, but not all patients.
Eventually, other studies contributed to the understanding of the contralesional hemisphere and
complementarily found that in the case of motor deficits after stroke, this contralesional
hemisphere could positively influence clinical motor recovery in some patients. For instance, in
rats, inhibiting the contralesional M1 led to more motor disabilities (Biernaskie et al., 2005) (figure
13). Thus, the contralesional hemisphere displays structural and functional reshaping which can
have both positive and negative influence on clinical recovery. It is currently unknown which
influence takes place in which patient, which complicates the design of neuromodulation studies
aiming at either enhancing or reducing the excitability of the contralesional hemisphere
(Guggisberg et al., 2019).

28
 Figure 13. Lidocaine anesthesia of undamaged motor
cortex affects reaching performance to a greater degree in
animals with large infarcts. Percent reaching accuracy
showed a similar lidocaine-induced impairment in controls
and in animals with small infarcts. Because large-infarct
animals could not successfully retrieve pellets, their
accuracy score was ‘0’. Performance of control animals
following lidocaine injection did not differ over the duration
of the experiment and so were pooled into a single time
point. Zeros indicate where no pellets were retrieved. *p <
0.05 compared with pre-lidocaine performance (Biernaskie
et al., 2005).

1.4. New treatment concepts

The ambitious goal of neuro-rehabilitation is to induce activity-dependent plasticity to reveal the
maximum potential of recovery in stroke patients. Thanks to various techniques of brain
assessment, we have obtained a better understanding of plasticity mechanisms. An optimal
modulation of neurophysiological targets should lead to better clinical improvements (Beaulieu
& Milot, 2018; Cramer et al., 2011; Maier et al. 2019). Combining neuro-physiological targets with
clinical goals, structural and functional connectivity as seen in the previous chapters can be
reshaped after stroke thanks to the effect of motor learning in specific training and situations
(Johansson, 2011; Takeuchi & Izumi, 2015).
In this chapter, I will introduce promising new techniques that can specifically modulate these
neurophysiological targets and thus should enable better recovery.

1.4.1. Repetitive transcranial magnetic stimulation

Repetitive TMS (rTMS) provides trains of pulses that can have different frequencies. It is
performed as a treatment for pathologies such as depression and has demonstrated beneficial
effects for stroke patients for aphasia and motor improvement. Depending on the stimulation
protocol, one can induce excitatory or inhibitory effects that last beyond the stimulation session
(Rossini & Rossi, 2007; Rubens & Zanto, 2012). rTMS seems to be able to mimic LTP and LTD
(Klomjai et al., 2015). Low-frequency stimulation is expected to have an inhibitory effect on brain
excitability while high-frequency rTMS is thought to enhance cortical excitability and to induce
LTP-like synaptic changes (Kobayashi & Pascual-Leone, 2003). For instance, rTMS was shown to
improve recovery from motor impairment and aphasia after stroke (Lefaucheur et al., 2014).
Indeed high-frequency rTMS on the lesioned hemisphere in stroke patients showed motor

29
improvement regarding dexterity, force, and spasticity (Corti et al., 2012). As for aphasia, rTMS
could improve naming and phrase length (Martin et al., 2010).
Despite these positive effects on function, there is still insufficient evidence for an improvement
of ADL, and we still lack large-scale studies with a bigger sample size (Veerbeek et al., 2014).
Furthermore, there is prominent inter-individual variability in the response to rTMS which
influences the outcomes in clinical trials. Indeed, only a minority of subjects will respond with a
reduction of neural excitability (as measured with MEP amplitudes) after an inhibitory rTMS
protocol and with an increase after excitatory protocols (Hamada et al., 2012; López-Alonso et
al., 2014). The session time, anatomical variability, coil orientation, and genetic variation are
suspected as possible variable causes. However, the main reason is the state-dependency of the
response to rTMS, such that areas with high excitability receiving an excitatory stimulation will
actually respond with a reduction of excitability rather than a further enhancement. Similarly,
areas with low excitability receiving inhibitory simulation will respond with an excitation (Siebner
et al., 2004; Silvanto & Pascual-Leone, 2008). One solution to address this variability to rTMS is to
predict its response through the measure of FC as described in 2.2 (Nicolo et al., 2015). Indeed,
preliminary evidence suggests that the magnitude of FC can predict the behavioral gains to rTMS
(Rizk et al., 2013).

1.4.2. Transcranial direct current stimulation

Transcranial direct current stimulation (tDCS) is a non-invasive neuromodulation method that can
improve upper extremity motor recovery when added to regular motor therapy (Hatem et al.,
2016). This stimulation can modify the neuronal membrane resting potential by inducing a weak
continuous electric current (Roche et al., 2015). Figure 14 shows the excitatory effect of an anodal
tDCS. This stimulation does not generates action potential, but it modulates the membrane
resting potential to be closer to the firing threshold, enabling the firing of action potential with
less depolarization. Indeed in their review of neurobiological mechanisms of tDCS, (Yamada &
Sumiyoshi, 2021) summarized the neural reaction cascade induced by anodal tDCS that seems to
lead to LTP (figure 15). Furthermore, recent studies have suggested that this type of stimulation
can also enhance neural interactions (functional connectivity) of cortical motor areas (Nicolo et
al., 2018). It might thus boost motor improvements.

30
Figure 14. Schematic diagram of changes in neuron potential by tDCS (Yamada & Sumiyoshi, 2021).

Figure 15. Putative mechanisms for the enhancement of long-term potentiation by tDCS. Various neurotransmitters
activate/inhibit transduction cascades bound to G-proteins of ion-channels, leading to phosphorylation of cAMP-responsive
element binding protein (CREB) and activation of genes in the nucleus of neurons. These signal transduction cascades enhance the
synthesis of various proteins, such as neurotransmitter synthases, receptors, ion channels, and intracellular signal proteins.
Facilitative actions of these proteins that regulate efficiency of neurotransmissions in the cerebral cortex circuit may explain the
ability of tDCS to induce LTP (Yamada & Sumiyoshi, 2021).

31
 Several tDCS montages have been described as enhancing motor outcome, including an
ipsilesional excitatory montage (Allman et al., 2016), a bi-hemispheric montage that combined
ipsilesional excitatory and contralesional inhibitory stimulation (Volz et al., 2017), and high-
definition montage (HD-tDCS) protocol with a more focal application (Minhas et al., 2010). It is
unknown which montage is optimal for patients with stroke. Although tDCS seems to be beneficial
as a complementary therapy accompanying motor rehabilitation sessions, it remains unclear
which specific montage would have the highest effects and if stimulating specifically during the
subacute phase tDCS would indeed enhance tDCS impact on behavioral outcomes. Furthermore,
we do not know whether tDCS applied during the subacute phase is able to enhance spontaneous
plasticity (Nicolo et al., 2018). Finally, the concerns with regards to variability in the response
expressed for rTMS above also apply to tDCS.

1.4.3. BCI-FES Close-loop system

BCI are systems allowing to link cerebral activity in real-time to sensory feedback given to the
patient. The recorded neural oscillations are decrypted and translated by a computer into specific
actions. With some training, users and patients can learn to modulate their own activity, and thus
control movements of robotic arms or even have sensory and motor feedback on their paretic
arm (Cervera et al., 2018a) (figure 16). For motor training, the instructions given to the
participants are either motor imagery tasks or movement attempt tasks. For motor imagery, the
participant is asked to project themselves into an imaginary movement without performing any
muscular contractions whereas movement attempts simply allow the participants to naturally do
their best in moving. In the case of stroke populations, movement attempts were found to be
more effective than motor imagery (Bai et al., 2020).

Figure 16. Illustration of typical brain-computer interface (BCI) systems used in post-stroke motor rehabilitation highlighting
sensory feedback modalities. EEG = electroencephalography, NIRS = near-infrared spectroscopy, ECoG = electrocorticography,
SMR = sensorimotor rhythm, MRCP = motor-related cortical potential (Cervera et al., 2018a).

32
Indeed clinical motor improvement was significantly higher in stroke patients that were trained
with BCI in comparison with groups that either received conventional therapy, performed the
same tasks without BCI-assistance, or in whom the BCI feedback was not contingent upon their
brain activity. This was the case no matter whether the BCIs were combined with motor imagery,
movement attempts with FES or device-controlled feedback (Bai et al., 2020; Biasiucci et al., 2018;
Leeb et al., 2016; Pichiorri et al., 2015). On the neurophysiological level, BCI training enhanced
MEPs on the ipsilesional hemisphere suggesting greater excitability and reduced excitability on
the contralesional primary motor area. However, clinical improvements could not be maintained
in the long term (Bai et al., 2020).
To increase the motor effect of BCI training, various studies have thus tried to combine it with
other innovative therapeutical tools such as exoskeleton assistance or muscular electrical
stimulations. The paired devices were triggered by the features extracted from the recorded EEG
signal. It appears that BCI combined with FES/NMES brought the best result in terms of clinical
improvement (Biasiucci et al., 2018; Corbet et al., 2015; Leeb et al., 2016).
A large lesion load of the CST in stroke patients leads to particularly poor recovery (Buch et al.,
2016b) due to a disruption of the descending pathway of the sensorimotor loop needed to
perform a movement. BCI can analyze the neural activity of the cortical motor area by decoding
in real-time motor intents and by transforming them into defined actions. When accompanied
with the right feedback, patients can change their neural activity which can lead to brain plasticity
(Dobkin, 2007a). BCI can be coupled with an efficient type of feedback, that is, meaningful sensor
inputs that are specific to the targeted area thus following the natural ascending pathways (Sur
& Rubenstein, 2005). FES recruits muscle spindles, thus being the most relevant tool providing
proprioceptive and somatosensory inputs. This stimulation causes antidromic activation of
motoneurons strengthening the relationship between motor movement and sensory afferent
inputs. The spike-timing created by the movement intent for the efferent fibers and the sensory
input for the afferent fibers might consolidate the synaptic connection of the CST (Markram et
al., 2011), stimulating on both ends the remaining fibers of the disrupted sensorimotor loop to
recreate the conduction of nerve impulse.
This contingency between the motor intention and the muscular stimulation has already turned
out to be essential for the induction of plasticity and recovery (Biasiucci et al., 2018).

1.4.4. Neurofeedback training of FC

EEG allows the localization of coherent brain areas, therefore it can be used to compute FC
between brain sublobes of interest (Guggisberg et al., 2011). With this tool, FC was demonstrated
to correlate with motor and cognitive performance in stroke patients (Dubovik et al., 2012b). In
addition, stroke patients were able to modulate their FC through neurofeedback training. Their
clinical scores improved when they learned to increase FC between the motor region and the rest
of the brain with neurofeedback which demonstrated a causal relationship between both
measures (Mottaz et al., 2018). In spite of these promising results, the extraction of FC requires
high-density EEG with long setup times and need for specialized know-how. FC training will only
become feasible for a large scale clinical application if the neurofeedback system is made portable

33
enough for clinical practice and with fewer setup constraints for the patient. Further studies are
needed to develop FC tracking with lower EEG density montage that could be rapidly installed on
the patients.

1.5. Objectives of the studies

Stroke often leads to hemiparesis with an important impact on patients’ independence. Despite
intensive rehabilitation training, recovery is often limited. Focusing on the upper limb and based
on recent advances in our understanding of brain plasticity, we aim to facilitate the translation of
new therapies to clinical practice in order to improve patients’ upper limb functional recovery
and quality of life. In particular, we evaluate three innovative treatment approaches: tDCS, BCI
combined with FES, and neurofeedback of FC by monitoring their impact on brain plasticity and
motor upper-limb rehabilitation. Treatments will be applied to specific subtypes of stroke which
are most likely to benefit.
The first study of my thesis aims to compare in a pilot phase the efficacy of four tDCS setups (high-
density anodal, conventional anodal, and bi-hemispheric montages vs. SHAM stimulation) among
subacute stroke patients who show a moderate motor deficit. Later a main study will then
compare the most optimal montage with SHAM stimulation. The hypothesis is that the
application of tDCS with the right montage in the first 4 weeks after stroke onset can help increase
neural interactions of perilesional cortical areas, increase their excitability and improve clinical
outcome above the proportional rule when compared to SHAM setup.
The second study of the thesis aims to evaluate a BCI protocol in patients with a severe motor
deficit. Our setup brings together a 16-channel EEG to read movement intention with an FES that
allows sensory inputs and contractions of the targeted muscles for motor response. It is
hypothesized that the POOR group initially suffers from severe CST damage and its subsequent
secondary degeneration. If started within the first weeks after stroke onset, motor cortex training
with a BCI and sensory reward with FES can help improve the clinical score of the patients by
preventing secondary white matter degeneration and, possibly, by inducing CST plasticity.
The third study of the thesis consists of making accessible to clinical practice a neurofeedback
training method previously implemented in the laboratory environment, which allows training FC
in the brain. Neurofeedback training of functional connectivity currently needs delicate material
such as high-density EEG which is difficult to set up especially in clinical settings, along with long
data processing time. Hence, the will to test the feasibility of FC training with fewer EEG
electrodes. The goal is to enable FC modulation while reducing the material and methodological
constraints. We hypothesize that, knowing the brain areas of interest, it is possible to retrieve FC
measures using the signal from a limited number of surface electrodes only.
These three studies pursue the objective of amplifying motor improvement and thus stroke
patients’ quality of life. Enhancing their functional recovery with these experimental techniques
would offer new solutions where efficient therapy is still lacking.

34
 2. Results

2.1. Study 1

2.1.1. Introduction
Many stroke patients benefit from spontaneous plasticity that favors their clinical outcome. It
corresponds to spontaneous neural remodeling that takes place during a limited time window
after the injury with or without therapy. Stroke patients with limited damage to the CST
particularly benefit from spontaneous plasticity, which allows them to recovery a fixed proportion
(about 70%) of their initial impairment (Ward, 2017). There is currently no known treatment that
allows enhancing spontaneous plasticity. Instead, current rehabilitation acts upon activity-
dependent plasticity that is based on learning and repeating functional exercises in an
appropriate environment. The more they train during the subacute phase, the faster the
maximum recovery potential is obtained. After this early rehabilitation phase, therapists focus on
the autonomy of the patients (Veerbeek van Wegen, et al., 2014).
However, there is evidence that high beta-band FC between motor areas and the rest of the
brain is associated with greater clinical motor recovery (Nicolo & Rizk et al., 2015). FC is thus a
possible marker of spontaneous plasticity. Enhancing beta-band FC trough treatment may thus
boost spontaneous plasticity and reduce patients’ motor deficit. Such neuromodulation is
possible with the use of a tDCS. It is a neuromodulator that enhances structural changes and
improves motor outcomes when accompanying motor therapy (Allman et al., 2016; Bolognini et
al., 2009; Doppelmayr et al., 2016; Lindenberg et al., 2010; Roche et al., 2015). Despite these
findings, it is still unclear which tDCS parameters at which stroke phase best improve motor
recovery in stroke patients. We hypothesize that in patients with a moderate motor deficit, the
application of tDCS with the optimal montage, in the first 4 weeks after stroke onset can help
increase neural interactions of perilesional cortical areas and improve clinical outcome above the
proportional rule.

2.1.2. Materials and methods

This study includes two phases. A pilot phase to determine an optimal tDCS montage among 4
different setups. Since we only aimed to recruit 6 patients per group for preliminary analyses, we
expected tendencies in our primary and secondary outcomes. Anything significant would not be
necessarily obtained from this sample of data. Once the optimal montage is highlighted, the main
phase of the study can begin including additional patients and testing only the best tDCS montage
against sham. Here we present the preliminary results of the pilot phase.

2.1.2.1. Participants
28 stroke patients participated in the pilot phase of this study. Among them 24 patients (10
women; 65.35 ± 13.77 years old; 21.62 ± 5.65 days since stroke onset) completed the whole
protocol, 6 patients in each of the 4 groups (CONV, BIHEM, HD, SHAM tDCS) (Table 1); and 4
35
patients withdrew from the protocol at their request. Patients were in the subacute state and
recruited at their arrival in the neurorehabilitation unit. The inclusion criteria were (a)
hemorrhagic or ischemic stroke, (b) a stroke onset under 4 weeks, (c) unilateral upper extremity
hemiparesis with a FMA-UE score between 15 and 55, (d) ability to give informed consent, (e)
ability to follow protocol instructions. The exclusion criteria were having a (a) second stroke
during rehabilitation, (b) history of epileptic seizure, (c) skull breach, (d) metallic object in the
brain, (e) pacemaker, (f) delirium or disturbed vigilance, (g) inability to follow treatments sessions,
(h) severe language comprehension deficits, (i) severe dystonia or spasticity, (j) severe co-
morbidity, (k) pregnancy.

Table 1. Patient characteristics. Abbreviations: f = female, m = male, NA = not available. * Time is calculated between the day of
the completed clinical assessments and stroke onset.

PATIENT GENDER AGE LESION HANDEDNESS STROKE TIME SINCE STROKE* FMA-UE
SIDE LOCATION (DAYS) AT T0
1 m 52 right right both 22 28
2 f 74 right right both 31 26
3 f 76 right right subcortical 21 16
4 m 56 left left both 21 44
5 f 81 right NA subcortical 17 47
6 m 75 left right subcortical 27 53
7 f 79 right right both 22 26
8 m 44 right right subcortical 17 57
9 m 83 left right subcortical 21 45
10 m 72 left right both 22 34
11 f 81 left right cortical 32 31
12 f 48 right left cortical 22 34
13 f 74 right right subcortical 19 34
14 f 66 left right both 25 17
15 m 68 right right subcortical 15 20
16 m 61 left right both 24 12
17 m 42 left right subcortical 29 52
18 f 76 right right cortical 9 53
19 m 53 right right cortical 16 50
20 f 82 right right subcortical 31 37
21 m 47 left right NA 22 52
22 m 48 right right cortical 19 49
23 m 77 left right cortical 21 35
24 m 56 right right both 14 30

36
 2.1.2.2. Study design
This is a triple-blind randomized study. The patients, the clinical assessor and the occupational
therapists were kept blind from the patients' group allocation. A stratification algorithm took into
account the initial FMA-UE, the lesion side, the age, the gender to balance the 4 groups on these
criteria before randomly assigning one to the patients. Each group corresponded to one
particular tDCS montage, they were: sham tDCS (SHAM), conventional anodal tDCS (CONV), bi-
hemispheric tDCS (BIHEM), high-definition-tDCS (HD). All patients benefitted from an
individualized neurorehabilitation therapy according to their condition with daily sessions of
physiotherapy, occupational therapy and if necessary neuropsychology and speech therapy. The
tDCS intervention was added to their rehabilitation schedule. The intervention consisted in
applying tDCS to the ipsilesional M1 during their regular occupational therapy sessions. The
assessments included clinical evaluations, eyes-closed-resting-state EEG, and MEPs search. One
blind assessor trained as an occupational therapist and specialized in neurorehabilitation
therapies performed all clinical assessments. A therapist different from the assessor performed
the stimulated occupational therapy sessions. The pre-intervention assessments occurred during
the first week (T0). Patients were then tDCS-stimulated during 6 sessions during the second and
the third week. The assessments post-intervention occurred during the fourth week (T1). One
month after the last stimulation session, the patients underwent a series of clinical assessments
(T2). Three months after their stroke onset, the assessor evaluated a last time the patients clinical
improvement (T3). All patients signed a written informed consent for the study protocol, which
was approved by the Hospital of Geneva Ethics Committee. The clinical assessment included the
FMA-UE, our primary outcome evaluation is. The secondary outcome evaluations are the ARAT,
Jamar, Nine-hole peg test, and the functional independence measure. These clinical assessments
are detailed in section 1.2.1. They are performed few days before the intervention, few days after
the intervention, one month after the intervention, three months after the stroke onset. A
specialized occupational therapist blind to the allocation group of the patients and not being the
patients' therapist for their standard rehabilitation performs all the assessments except the FIM
that is scored by the nurse in the rehabilitation unit.

2.1.2.3. MEPs with TMS

We assessed the cortical excitability and the integrity of the CST of the patients right before and
after the intervention by looking for motor evoked potential with a MagPro X100 stimulator
(MagVenture A/S, Farum, Denmark). A figure-of-eight coil (MCF-B65) with a diameter of 2 x 75
mm placed over the hand motor hotspot in M1 triggered the hand knob. 1cm-diameter
electromyogram (EMG) pads recorded the MEP response in the contralateral first dorsal
interosseous (FDI) muscle. The neuronavigation software LOCALITE (TMS Navigator, Localite,
Bonn, Germany) used individualized T1-weighted images to calculate and display the entry point
corresponding to the presumed hand knob location (Mayka et al., 2006). This allowed keeping
the stimulation location constant between pre and post-sessions. To find the optimal stimulation
site that activates the contralateral FDI muscle, we moved the coil in 0.5 cm steps around the
entry point until evoking steadily the largest MEP responses. From 0, we incremented the TMS
intensity by 10% until triggering these large MEPs that would correspond to suprathreshold
stimuli. This optimal location was then marked and saved on the LOCALITE visualization. We then

37
reduced by step of 1% the TMS intensity until obtaining 50 μV-MEPs in 5 out of 10 consecutive
stimuli which defined the rest motor threshold. Once the hot spot was defined and the RMT
found, we performed 2 series of 10 stimulations at 100%RMT, 110%RMT, 120%RMT. Both
ipsilesional and contralesional MEPs were obtained (Stinear et al., 2015). The Kruskal-Wallis test
assessed the peak-to-peak amplitude and the RMT changes from pre- to post-intervention for
differences between groups. The Chi-square analysis assessed the proportion of patients with an
increase in peak-to-peak MEP amplitude and or with a reduction of the RMT for differences
between groups.

2.1.2.4. Resting-state EEG and FC

To assess functional connectivity, a high-density EEG system recorded 10 minutes of patients’
spontaneous activity in a resting state. Patients’ were asked to close their eyes, remain relaxed,
and avoid falling asleep. This recording occurred right before and after the intervention, using a
128-channel BrainVision Actichamp EEG system (Brain Products GmbH, Germany) at a sampling
rate of 500 Hz. The EEG signal is referenced to Cz. The artifacts were manually removed from the
signal using Cartool (http://sites.google.com/site/cartoolcommunity/home). The signal was
separated into 1 second-epochs. The 300 epochs with the highest alpha power were kept for the
EEG analyses.
T1 MRI images allowed the segmentation of patients’ brain into scalp, skull, grey, and white
matter using MARS toolbox (https://www.parralab.org/mars/) (Huang and Parra 2015) and
NUTMEG (http://nutmeg.berkeley.edu) (Dalal et al. 2011). With diffusion-weighted images, a
mask of the lesion was drawn. From this, we generated the lead field potential. It was composed
of 10mm voxels and calculated with the boundary element model (BEM) computed from patients’
T1-weighted images with the Helsinki BEM library (http://peili.hut.fi/BEM/) (Stenroos et al.,
2007). The lead field potential was the forward solution translating how the EEG sensors captured
neuronal signals in each patient.
EEG data was bandpass filtered between 1 and 20 Hz. We used a scalar minimum variance
beamformer which is an adaptive spatial filter to calculate the inverse solution and obtain the
source signal.
The coherence between one voxel and all other voxels is calculated from the source signal in
Nutmeg for the beta frequency band (13-16 Hz, 17-20 Hz), as previous studies have observed that
FC in this band correlates with future clinical improvements (Nicolo & Rizk et al., Brain 2015).
Functional connectivity (FC) is computed as the absolute imaginary coherence (Nolte et al., 2004).
From the HMAT Atlas (Human Motor Area Template), the selected region of interest (ROI) mask
was the combination of M1 and SMA. With the functional connectivity mapping (FCM) toolbox
(Guggisberg et al., 2011), the average FC of the voxels contained in the ROI computed the region
FC. We then perform a Pearson correlation to compare FC against FMA change between T1 and
T0 and a one-way ANOVA to compare the FC of the 4 groups at T0, T1 and change between T1
and T0.

38
 2.1.2.5. tDCS
The four tDCS and sham montages in the pilot study are:
The conventional anodal montage: the electrodes are 5x5cm. The anode is placed over the
ipsilesional primary motor cortex (M1) and the cathode over the contralateral supraorbital area.
In this configuration, the anodal electrode is expected to enhance the excitability of the
ipsilesional M1.
The bihemispheric montage: the electrodes are 5x5cm. The anode is placed over the ipsilesional
M1 and the cathode over the contralesional M1. This montage is expected to enhance the
excitability of the ipsilesional M1 and to inhibit the contralesional M1 (Lindenberg et al., 2010).
The high-density montage: the electrodes are round with a diameter of 1cm. One anode electrode
is centered on the ipsilesional M1 and four cathodes are placed around the anode (Minhas et al.,
2010) in a cross shape.
the sham montage: the electrodes are 5x5cm. The anode is placed over the ipsilesional M1 and
the cathode over the contralateral supraorbital area.
A NeuroConn DC-Stimulator applies a constant electrical current of 2mA for 20 minutes (Monte-
Silva et al., 2013) with a ramp up and down of 30 seconds. For the sham stimulation, the current
is only a ramped up and down of 30 seconds. Each neuromodulation session is delivered during a
rehabilitation session with 6 applications spread over 2 weeks.

2.1.3. Results

2.1.3.1. Clinical effects

FMA-UE
In figure 17, we can see that before the intervention the FMA scores are similar in the four groups.
All patients treated with the conventional and HD montages have recovered well while 2 patients
in the SHAM group and 2 patients in the bihemispheric group have worsened their condition
because of a retractile capsulitis of the shoulder and/or pain and edema in the impaired upper
limb.

39
Figure 17. Top left figure is the FMA-UE score at T0. Top right figure is the FMA-UE score at T1. Bottom left figure is the FMA-UE
score at T2. Bottom right figure is the FMA-UE score at T3.

Figure 18 shows the change of FMA-UE score for each patient at each timepoint compared to T0.
Only the conventional anodal and HD montage do not contain patients with a reduced score.

Figure 18. Left figure shows the difference in FMA-UE score between T1 and T0. Middle figure shows the difference in FMA-UE
score between T2 and T0. Right figure shows the difference in FMA-UE score between T3 and T0.

40
Figure 19 shows the percentage of maximum recovery. The difference of points between the
maximum score of 66 and the FMA-UE score corresponds to the potential maximum recovery of
the patients. Here, at a specific timepoint, we want to see the obtained score in percentage of
the potential. It appears that only the conventional anodal and the HD group present patients
who all constantly improved their FMA-UE score.

Figure 19. % of max recovery = FMA-UE(t) * 100 / (66 – FMA-UE(T=0)). Left figure shows the percentage of maximum recovery in
FMA-UE score between T1 and T0. Middle figure shows the percentage of maximum recovery in FMA-UE score between T2 and
T0. Right figure shows the percentage of maximum recovery in FMA-UE score between T3 and T0.

Figure 20 shows the distribution of all patients regarding the proportional recovery rule. We can
observe that patients of the HD and CONV group recovered along with the proportional rule
(dashed line) while some patients of the SHAM and BIHEM group recovered poorly.

Figure 20. Distribution of the patients regarding the proportional recovery rule (in dashed-line) where the x-axis is the initial
impairment (66 – FMA-UE(T0)) and the impairment at timepoint t (FMA-UE(t) – FMA-UE(T0)). In the left figure, t = T1. In the middle
figure, t = T2. In the right figure, t = T3.

ARAT
In figure 21, we see a comparable ARAT score at baseline (pre). When analyzing the change of
ARAT between the post-assessments and the pre-assessments we can see at T1 a significant

41
change 𝐹𝑝𝑜𝑠𝑡𝑉𝑆𝑝𝑟𝑒 (3, 20) = 4.62, 𝑝𝑝𝑜𝑠𝑡𝑉𝑆𝑝𝑟𝑒 = 0.013) between the HD and the SHAM group
(figure 22).

Figure 21. Top left figure is the ARAT score at T0. Top right figure is the ARAT score at T1. Bottom left figure is the ARAT score 1
month at T2. Bottom right figure is the ARAT score at T3.

Figure 22. Left figure shows the difference in ARAT score between T1 and T0. Middle figure shows the difference in ARAT score
between T2 and T0. Right figure shows the difference in ARAT score between T3 and T0. The ARAT shows a clinical effect between
the post-intervention score and when comparing the HD-tDCS and the SHAM group.

42
JAMAR
The ratio of grip strength between the affected and the unaffected hands evolves similarly among
the 4 groups. Focusing on the JAMAR change, the HD group displays a wider range in grip strength
during the first 3 months after stroke onset (figure 23), but the difference was not statistically
significant in the small sample.

Figure 23. Left figure shows the difference in grip strength change between T1 and T0. Middle figure shows the difference in grip
strength between T2 and T0. Right figure shows the difference in grip strength between T3 and T0.

Nine-hole peg test

There was no difference in fine movement dexterity as measured with the nine-hole peg test
between groups (figure 24).

Figure 24. Left figure shows the difference in velocity between T1 and T0. Middle figure shows the difference in velocity between
T2 and T0. Right figure shows the difference in velocity between T3 and T0.

43
FIM
In the figure 25, all the groups evolve similarly regarding their FIM score. It is noticeable that the
2 patients with poor FIM recovery, as seen at T3, did not benefit from an active tDCS, being in the
SHAM group.

Figure 25. Top left figure is the FIM at T0. Top right figure is the FIM at T1. Bottom left figure is the FIM at T2. Bottom right figure
is the FIM at T3.

2.1.3.2. Neurophysiological effects

MEP
There were no significant differences in MEP amplitude change between stimulation conditions.
There is a tendency of the HD-tDCS group to have an increase of MEP amplitude after the
intervention on the contralesional side (figure 26). This HD-tDCS stimulation seems to be the
montage that enhances the most cortical excitability of the contralesional hemisphere while it
does not seem to influence the ipsilesional hemisphere.

44
Figure 26. Left figure shows the increase of MEP in percentage in the affected hand between T1 and T0. Right figure shows the
increase of MEP in percentage in the unaffected hand between T1 and T0.

There is no significant difference between the groups in RMT in the ipsi or contralesional
hemisphere. From figure 15, we can notice negative changes in patients of the CONV and HD
group.

Figure 15. Left figure shows the change in RMT in the ipsilesional hemisphere between T1 and T0.
Right figure shows the change in RMT in the contralesional hemisphere between T1 and T0.

EEG FC
Patients of the 4 groups have similar initial FC. At T1, the CONV group shows a lower FC than in
the other groups, but the change from pre to post assessments did not show consistent
differences between groups (figure 27).

45
Figure 27. Left figure shows the FC in the ipsilesional M1 and SMA among the 4 groups at T0. Middle figure shows the FC in the
ipsilesional M1 and SMA among the 4 groups at T1. Right figure shows the difference in FC in the ipsilesional M1 and SMA among
the 4 groups between T1 and T0.

In figure 28, we see a significant correlation between the FC in high-beta of the ipsilesional
motor areas at T0 and the FMA-UE change between T1 and T0 (r=0.51, p=0.0116). Early FC
appears to predicts motor outcome, confirming previous findings (Nicolo & Rizk et al., 2015a)
On the other hand, when looking at the difference between T1 and T0, a lowering of beta-
band FC between ipsilesional M1/SMA and the rest of the brain is significantly correlated with
more FMA-UE improvement (r=-0.53, p=0.0079). Late arrival of a FC in the ipsilesional motor
area is associated with worse recovery which is in line with the results in (Nicolo & Rizk et al.,
2015a).

Figure 28. Top left figure are the selected ROIs M1 and SMA. Top right figure is the FC in the high-beta range at T0 regarding the
FMA change between T1 and T0. Bottom left figure is the FC in the high-beta range at T1 regarding the FMA change between T1
and T0. Bottom right figure is the FC difference in the high-beta range at between T1 and T0, and the FMA change between T1 and
T0.

46
 2.1.4. Discussion
Patients presenting a mild to moderate upper limb impairment show a correlation between their
global FC and their motor outcome (Nicolo & Rizk et al., 2015). To enhance their recovery, we
targeted these neural oscillations in the motor area through neuromodulation with the help of
tDCS. So far, a pilot phase is completed during which different tDCS montages were compared.
As a result, we observed the best clinical effects with the HD montage, visible most clearly in the
ARAT score. However, the neurophysiological mechanisms underlying this improvement are still
unclear.
Clinical outcomes

The pilot phase is underpowered to reveal statistically significant differences between the 4 tDCS
montages. Indeed, the goal of this phase was to determine the montage with the best potential
then to compare it to SHAM in a main phase study, which is currently running.

The FMA-UE, our primary motor outcome did not show a significantly better improvement of one
particular stimulation montage in the pilot phase. Yet, interestingly the conventional and HD
stimulation montages have prevented a worsening of the patients’ upper limb due to pain, as it
has been observed in some patients of the bihemispheric and sham conditions. Indeed (Khedr et
al., 2017) found that when applied over M1, anodal tDCS reduced pain and depression in
fibromyalgia patients. In mice, the application of a bicephalic tDCS succeeded in reducing the paw
sensitivity of mechanical allodynia induced by partial sciatic nerve ligation (Souza et al., 2018). In
a review paper studying the impact of tDCS as a pain reliever, (Lefaucheur et al., 2008) found that
conventional anodal tDCS reduced pain in spinal cord lesion patients. tDCS used as a pain reliever
requires 20 minute-session of 5 to 10 consecutive days. Because the motor area is implicated in
the pain neural pathways, multiple studies already targeted this region for neuromodulation
(Morya et al., 2019). Early results in athletes and a use-case of a multiple sclerosis patient are
encouraging regarding a positive effect on mood (Chalah et al., 2017; Valenzuela et al., 2019).
Referred to stroke patients, the tDCS neuromodulation could diminish the emotional impact of
their subacute condition.
In general, the clinical results exhibited higher motor outcomes in the CONV and the HD group.
In particular, a significant improvement is scored with the ARAT at T1 in the HD group. This can
be explained by the fine distal movement assessment present in the test which is more accurate
than the FMA-UE. It appears that patients who best recovered benefitted from an anodal
stimulation whether it is conventional or HD. Applying the HD-tDCS montage was sometimes
tedious because of the regular loss of impedance. The electrodes being only of a diameter of 1cm,
it created a montage where the impedance was high and corresponded to low conductivity. As a
result, we often had to reduce the wanted intensity of 2mA. (Caparelli-Daquer et al., 2012)
demonstrated in a pilot study an excitability effect on the motor cortex with only a 1mA intensity
in the HD-tDCS montage. With the improvement of the HD setup, we can hope for a higher motor
outcome in this HD anodal montage. It seems that offering simultaneously an excitatory
stimulation on the ipsilesional hemisphere and inhibitory in the contralesional hemisphere as

47
done in the bihemispheric montage had comparable results to the SHAM group. These
observations support the doubt on the interhemispheric imbalance concept (Guggisberg et al.,
2019).
Neurophysiological outcomes

Patients who experienced the greatest clinical effect belonged to the CONV and HD group. Both
are anodal tDCS with an excitatory stimulation over the ipsilesional M1. Although we were
expecting effects in the ipsilesional hemisphere (Caparelli-Daquer et al., 2012), we instead
observed a trend for an excitatory enhancement of the contralesional hemisphere in the MEPs.
It appears that the anodal montage may have some effect on the contralesional hemisphere that
could positively influence motor performance. When assessing chronic stroke patients,
(McCambridge et al., 2018) observed an excitability increase in the motor cortex in both
hemispheres when applying an anodal tDCS over the contralesional M1. As documented in
section 1.3.6., the role of the contralesional hemisphere is still unclear, being maladaptive for
some patients while facilitating the recovery for others. In this case, they suggested that in spite
of not being stimulated, the ipsilesional M1 became more excitable after the tDCS session thanks
to the transcallosal pathway. In the same study, they tested the effect of a cathodal tDCS over
the contralesional M1. Echoing our results regarding the bihemispheric tDCS montage, theirs did
not affect the cortical excitability of any hemisphere or the clinical score.

Regarding FC, we reproduced in our sample previous observations (Hordacre et al., 2020; Nicolo,
Rizk et al., 2015a) regarding the correlation between beta-band FC in the pre-recording and
clinical improvements. However, we did not see a clear effect of any stimulation montage on
beta-band FC. Indeed the effects of tDCS on FC have been inconsistent also in the literature.
(Sehm et al., 2013) observed an after-effect of bihemispheric tDCS in the stimulated M1, but
conventional anodal tDCS did not produce an FC change after the stimulation. In another study
where chronic stroke patients underwent bihemispheric tDCS, an FC increase was still present
one week after a single motor training session (Lefebvre et al., 2017). Although effects are present
in single sessions, the impact of tDCS on motor performance can vary across studies due to the
number of stimulation therapies, the stroke phase of the patients, the timing of the assessment
post-intervention, and patients’ mood (Chang et al., 2021; Jonker et al., 2021; Pavlova et al.,
2020).
Similarly, as in our previous study (Nicolo & Rizk et al., 2015), we also observed that a further
increase of FC during the treatment period was associated with worse recovery. One
interpretation of this could be that the FC increase during the stimulation period appeared too
late to allow motor improvement. Thus, there could be a limited time window, during which beta-
band FC has a favorable effect on recovery. Alternatively, we can speculate that if beta-band FC
is enhanced above a critical level, it represents an overexcitability that hinders recovery. When
the cortical excitability is increased after stroke by applying an excitatory neuromodulation, it
could worsen neural damage by inducing an excessive stimulation (Pavlova et al., 2020). Finally,
a third possible interpretation is that the high global FC with the rest of the brain observed during
48
the pre-recording (and associated with better recovery) is refined such that FC becomes more
selective to other areas of the motor network. In other words, the global FC to the rest of the
brain is replaced by a more selective FC among motor nodes. We plan to test this latter hypothesis
by looking more specifically at the changes of FC among motor areas.
To summarize, we have a clinical effect on motor recovery with the HD-tDCS montage for which
we have so far not been able to reveal the underlying neurophysiological correlates. Indeed we
could not observe an enhanced ipsilesional excitability in the MEPs or a significant change in
global FC. Following this pilot phase, we will compare HD-tDCS to sham in the main phase.

49
 2.2. Study 2

2.2.1. Introduction
Stroke patients with severe upper limb impairments do not benefit from the spontaneous
plasticity and belong to a group that recovers poorly, which is less than 30% of their initial
impairment (Ward, 2017). Despite an intensive rehabilitation program, their progression is
plateauing and therapy switches often to compensatory strategies to focus on patients’
autonomy in daily life activities after working on specific motor improvement. A first explanation
for this poor recovery is the severe damage to the CST (Stinear et al., 2015). Severe CST damage
is associated to lower FC in the primary motor cortex that seems to prevent a satisfactory motor
recovery (Guggisberg et al., 2017b). To preserve the integrity of the CST, a closed-loop setup that
stimulates both the remaining afferent and efferent fibers could recreate the link between an
intended movement and its execution. BCI coupled with FES offers these ascending and
descending inputs where FES stimulation is made contingent on the movement intention
recognized by the BCI to reinforce the closed-loop effect (Mane et al., 2020b). This setup showed
promising results to improve motor deficits in chronic stroke patients (Biasiucci et al., 2018).
Because rehabilitating patients during their subacute phase brings better outcomes, it seems
essential to offer this innovative therapy before their chronic state. Here we hypothesize that in
patients with a severe motor deficit, the application of an FES contingent on the patient’s
activation of the motor cortex recognized by a BCI can maintain the CST integrity, to initiate the
cortical process of neural interactions and eventually improve the clinical outcome.

2.2.2. Materials and methods

2.2.2.1. Participants
The study requires the participation of 32 patients, 16 for the BCI and 16 for the SHAM-BCI. So far
35 patients were recruited. Among them 20 patients (8 women; 68.16 ± 9.64 years old; 24.85 ±
11.16 days since stroke onset) completed the whole protocol, 10 in the BCI group and 10 in the
SHAM-BCI group (Table 2); 9 patients (1 woman; 66.26 ± 15.20 years old; 34.67 ± 13.27 days since
stroke onset) left the protocol at their request before accomplishing 5 BCI sessions but agreed to
perform the clinical assessments at T2, they formed the DROP-OUT group (Table 3); 4 patients
left the protocol at their request and refusing to perform the clinical assessments at T2; 1 patient
was excluded due to severe language comprehension deficits; and 1 patient was excluded due to
technical events extending the time of the intervention. Patients were in the subacute state and
recruited at their arrival in the neurorehabilitation unit. The inclusion criteria were (a)
hemorrhagic or ischemic stroke, (b) lesions affecting the cortico-spinal tract, (c) a stroke onset
under 8 weeks, (d) a severe unilateral upper extremity hemiparesis or hemiplegia with a FMA-UE
score below 15, (e) ability to give informed consent, (f) ability to follow protocol instructions. The
exclusion criteria were having a (a) second stroke during rehabilitation, (b) history of epileptic
seizure, (c) skull breach, (d) metallic object in the brain, (e) pacemaker, (f) delirium or disturbed
vigilance, (g) inability to follow treatments sessions, (h) severe language comprehension deficits,
(i) severe dystonia or spasticity, (j) severe co-morbidity, (k) pregnancy.
50
Table 2. Patient characteristics of the BCI and SHAM group. Abbreviations: f = female, m = male. * Time is calculated between the
day of the completed clinical assessments and stroke onset.

PATIENT GENDER AGES LESION HANDEDNESS STROKE TIME SINCE FMA-UE

SIDE LOCATION STROKE* (DAYS) AT T0
1 f 76 left right subcortical 26 5
2 m 54 right right subcortical 35 6
3 f 64 right right subcortical 12 11
4 m 60 left right both 22 3
5 m 64 left right subcortical 11 5
6 m 53 right right subcortical 20 7
7 m 82 right right both 15 4
8 f 58 right right subcortical 28 6
9 m 66 right right both 29 5
10 f 77 right right subcortical 9 4
11 m 65 right right subcortical 42 4
12 m 74 right right both 20 11
13 f 79 right right both 19 5
14 m 63 right right cortical 34 4
15 f 55 right right both 37 7
16 m 68 left right subcortical 27 4
17 m 66 right right both 12 2
18 f 76 right right both 22 10
19 f 85 right right both 25 5
20 m 76 right right cortical 52 10

Table 3. Patient characteristics of the DROP-OUT group. Abbreviations: f = female, m = male, NA = not available. * Time is calculated
between the day of the completed clinical assessments and stroke onset.

PATIENT GENDER AGES LESION HANDEDNESS STROKE TIME SINCE FMA-UE

SIDE LOCATION STROKE* (DAYS) AT T0
1 m 70 right right both 35 5
2 m 86 left right subcortical 39 6
3 m 57 left right subcortical 30 3
4 m 45 left right subcortical 53 4
5 m 52 left NA NA 50 4
6 m 84 right right both 30 4
7 m 52 right right both 8 4
8 f 76 right right both 28 4
9 m 74 right right both 39 4

2.2.2.2. Study design

Patients were randomly allocated either to the BCI or the SHAM-BCI group . The randomization
was stratified for initial FMA-UE, lesion side, age, and gender. The patients, the experimenter

51
performing the BCI-based traning sessions, and the therapist performing the clinical assessment
are all blinded to the patient group allocation. All patients benefitted from an individualized
standard neurorehabilitation therapy according to their condition. The study interventions
consisted in providing BCI-FES sessions in addition to their regular therapy sessions. The
assessments included clinical evaluations, eyes-closed-resting-state EEG, and MEPs. A trained
occupational therapist and specialized in neurorehabilitation therapies performed all clinical
assessments. The pre-intervention assessments occurred during the first week (T0). Patients were
then trained during 10 sessions (2 calibration and 8 training sessions) during 3 weeks. The
assessments post-intervention occurred the week after the intervention (T1). Three months after
their stroke onset, the assessor evaluated a last time the patient’s clinical improvement (T2). All
patients signed a written informed consent for the study protocol, which was approved by the
Hospital of Geneva Ethics Committee.
2.2.2.3. Clinical assessment
Cf. study 1
2.2.2.4. High-density resting-state EEG
Cf. study 1
2.2.2.5. MEPs with TMS
Cf. study 1
2.2.2.6. DTI analyses
DTI analyses take as input T1, T2 and DWI images. TORTOISE software
(https://tortoise.nibib.nih.gov/) preprocesses the diffusion images in two steps. The first
step through the DIFFPREP module resamples the image, corrects the head motion, the
eddy current distortion and echo-planar imaging (EPI) distortions. The second step uses
the DIFFCALC module. It applies the robust tensor fitting algorithm RESTORE (Chang et al.,
2005), corrects the estimation of the diffusion anisotropy degree, and offers directionally
encoded color map visualization. DTI-TK toolkit (http://dti-
tk.sourceforge.net/pmwiki/pmwiki.php) allows spatial normalization and atlas
construction across all patients from which is calculated the mean FA of the CST. The
normalization steps include a series of 3 transformations for coregistration with a tensor
template. First, each patient’s images are transformed through a rigid alignment to keep
the size and the shape of the objects. Second, an affine alignment on the data performs a
second linear transformation by allowing this time the alteration of global size and shape
increasing the precision of the first transformation estimation. Third, the application of a
deformable alignment finalizes the normalization procedure by removing the size and
shape differences between local structures. From these tensor volumes, TVtool
(http://dti-tk.sourceforge.net/pmwiki/pmwiki.php?n=Documentation.TVtool) generates
FA maps among other computations for each patient at T1 and T2, in the ipsilesional and
the contralesional CST. These FA are averaged by patient’s CST and by side to finally
determine the asymmetry:
𝐹𝐴𝑐𝑜𝑛𝑡𝑟𝑎𝑙𝑒𝑠𝑖𝑜𝑛𝑎𝑙 −𝐹𝐴𝑖𝑝𝑠𝑖𝑙𝑒𝑠𝑖𝑜𝑛𝑎𝑙
𝐶𝑆𝑇𝑎𝑠𝑦𝑚𝑚𝑒𝑡𝑟𝑦 = 𝐹𝐴𝑐𝑜𝑛𝑡𝑟𝑎𝑙𝑒𝑠𝑖𝑜𝑛𝑎𝑙 +𝐹𝐴𝑖𝑝𝑠𝑖𝑙𝑒𝑠𝑖𝑜𝑛𝑎𝑙
(1)
A t-test analysis compares this CST asymmetry between the BCI and the sham group.

52
 2.2.2.7. BCI-FES
The BCI setup consists of a 16-channel EEG placed over the bilateral sensorimotor cortex, 2 pairs
of surface stimulation electrodes from the FES stimulator for the forearm and the upper arm, and
the real-time BCI software developed by the laboratory of Prof. José del R. Millán (Leeb et al.,
2013). Two movements are trained, hand-to-mouth (flexion of the fingers, wrist, elbow, internal
rotation of the shoulder) and reaching (extension of the shoulder, elbow, wrist, fingers). For the
hand-to-mouth movement, the flexor digitorum superficialis and the biceps muscle are
stimulated. For the reaching movement, the extensor digirotum communis and triceps muscles
are stimulated. For each session, the BCI operator determines a sensory and a motor threshold.
The sensory threshold is the minimum intensity at which patients can feel the electrical current.
The motor threshold is the minimum intensity at which the muscles contract. The FES is limited
to 30mA. If the patients' muscles cannot have a clear contraction, we look for the best surface
electrodes arrangement that can give muscle tremors. In all cases, the intensities are sub-painful
even if the thresholds are not found. The first two sessions are the calibration sessions, one for
the hand-to-mouth movement and the other for the reaching movement. They are meant to
record patients’ EEG during the tasks, movement and rest, in order to generate offline the first
classifier allowing the discrimination of the movement and the rest trials during the online
sessions (i.e. BCI training sessions). For one specific movement, 3 runs of 30 trials are performed.
1 run includes 15 rest instructions and 15 movement instructions. Based on these calibration
sessions, we selected the most discriminant EEG channels and frequencies for separating
between resting and movement trials (Leeb et al., 2013). During the training sessions, we define
the sensory and the motor FES threshold of the patients. During the runs, the sensory FES
feedback is the afferent signal allowing our setup to close the sensori-motor loop of the attempt
movement and is the response that is contingent to the movement intention decoded by the BCI.
The motor FES intensity contracts the targeted muscles, thus rewarding the patient for
attempting the movement. Patients performed a total of 8 training sessions with BCI coupled to
FES (or sham), with 4 sessions per trained movement. In each training session, patients perform
3 to 4 runs of 20 trials with 5 rest instructions and 15 movement instructions.During the training
sessions, the selected discriminant features classify the received intention attempt through a
Gaussian distribution. The discriminant features were updated weekly. Based on these features,
the BCI calculates the probability of being a rest or a motor intent. A sensory feedback is sent as
soon as the BCI recognized a movement intention to encourage the success of a motor attempt
in the movement trials in training sessions and to induce the contigent feedback increasing the
Hebbian plasticity. This feedback corresponds to stimulation above the sensory threshold defined
at the beginning of the session. Then, stimulation above the motor threshold is delivered if and
only if the movement intention read from EEG features exceeds the second threshold.
In the sham group, the FES is triggered based on the EEG replay of previous BCI patients. When a
newly recruited patient is allocated SHAM, he/she is matched with a previous BCI patient. The
calibration sessions remain the same. During the training sessions, for each run, one run from the
matched patient is chosen randomly. The calculated probability of movement and rest of that trial
is kept in the same order and integrated. This integrated probability series is finally used as a
replay for the SHAM patient on which the FES will be triggered.

53
 2.2.3. Results

2.2.3.1. Clinical effects

This protocol comes in addition to the standard therapies schedule for the patients. Its intensity
and setup heaviness explain the number of drop-out patients who completed up to the fifth first
BCI session out of 10 but decided to quit due to their fatigue. Nevertheless, these patients agreed
to perform the clinical assessment at T2 allowing us to compare the performance of the patients
who completed the whole protocol and the ones who did not.
FMA
The analyses of the FMA at T0, T1, T2, its change and recovery percentage do not show a
significant difference among the 3 groups (figure 29-32). In spite of these results, we observed
that the only patients that recovered well belonged to the BCI or drop-out group but none in the
SHAM group. As seen in figure 33, two patients of the BCI group and 1 patient from the drop-out
group, but none of the patients of the sham group benefited from a proportional recovery.

Figure 29. Left figure is the FMA-UE score at T0. Middle figure is the FMA-UE score right at T1. Right figure is the FMA-UE score at
T2.

Figure 30. Left figure shows the difference in FMA-UE score between T1 and T0. Right figure shows the difference in FMA-UE score
between T2 and T0.

54
Figure 31. % of max recovery = FMA-UE(t) * 100 / (66 – FMA-UE(T=0)). Left figure shows the percentage of maximum recovery in
FMA-UE score between T1 and T0. Right figure shows the percentage of maximum recovery in FMA-UE score between T2 and T0.

Figure 32. Distribution of the patients regarding the proportional recovery rule (in dashed-line) where the x-axis is the initial
impairment (66 – FMA-UE(T0)) and the impairment at timepoint t (FMA-UE(t) – FMA-UE(T0)). In the left figure, t = T1. In the right
figure, t = T3

55
Figure 33. Fitters are the patients who follow the proportional recovery rule. Non-fitters are the patients who do not follow the
proportional recovery rule. Left figure shows the histogram of the deviation of observed FMA recovery at 3 months from the
recovery that would have been predicted by the proportional recovery rule. Right figure shows the distribution of fitter patients
per group.

ARAT
The ARAT confirms the improvement of the few patients only in the BCI or DROP-OUT group. We
do not see a significant effect at a group level (figure 34).

Figure 34. Left figure shows the difference in ARAT score between T1 and T0. Right figure shows the difference in ARAT score
between T2 and T0.

56
JAMAR
These same patients also improved in the grip strength assessment without a significant effect
between the 3 groups in JAMAR change (figure 35).

Figure 35. Left figure shows the difference in JAMAR score between T1 and T0. Right figure shows the difference in JAMAR score
between T2 and T0.

ASHWORTH
The Ashworth assessment shows a significant difference between the BCI, the SHAM and the
drop-out group at T1 (𝐻 (2) = 8.43, 𝑝 = 0.0148). Indeed more patients in the SHAM than in the
BCI group and all patients in the drop-out group developed spasticity in their affected upper limb
than in the BCI group at T1. Although the tendency remains in favor of the BCI group at T2, the
difference is not significant anymore (Figure 36).

Figure 36. Left figure is the ASHWORTH score at T0. Middle figure is the ASHWORTH score right at T1. Right figure is the ASHWORTH
score at T2.

57
Patients in the BCI group evolve less significantly toward a spasticity of their impaired upper limb
between T1 and T0 (𝐻 (2) = 6.35, 𝑝 = 0.0418) (figure 37).

Figure 37. Left figure shows the difference in ASHWORTH score between T1 and T0. Right figure shows the difference in ASHWORTH
score between T2 and T0.

SEMMES-WEINSTEIN MONOFILAMENT
The change of light touch sensation is similar between the real BCI and the SHAM group at all
time points (Figure 38). The BCI training did not worsen or improve patients’ sensation when
compared to the SHAM patients (Figure 39).

Figure 38. Left figure is the MONOFILAMENT score at T0. Middle figure is the MONOFILAMENT score right at T1. Right figure is the
MONOFILAMENT score at T2.

58
Figure 39. Left figure shows the difference in MONOFILAMENT score between T1 and T0. Right figure shows the difference in
MONOFILAMENT score between T2 and T0.

FIM
FIM scores and its evolution were also similar among the 3 groups (Figure 40 & 41).

Figure 40. Left figure is the FIM score at T0. Middle figure is the FIM score right at T1. Right figure is the FIM score at T2.

Figure 12. Left figure is the FIM score at T0. Middle figure is the FIM score right at T1. Right figure
is the FIM score at T2.

59
Figure 41. Left figure shows the difference in FIM score between T1 and T0. Right figure shows the difference in FIM score between
T2 and T0.

2.2.3.2. Neurophysiological effect

DTI
We use the diffusion tensor images to assess the CST integrity via the FA asymmetry
characteristics (Figure 42). The FA asymmetry and its evolution of both groups are similar at T0,
T1, and FA asymmetry change. It is noticeable that the difference of FA asymmetry of both groups
between T1 and T0 is close to 0. Finally, there is no significant correlation between the FA
asymmetry change and the FMA change between T1 and T0.
Figure 42. Top left figure is the
FA asymmetry score at T0. Top
right figure is the FA
asymmetry score at T1. Bottom
left figure is the difference in FA
asymmetry score between T1
and T0. Bottom right figure is
the correlation between the FA
asymmetry difference of FMA-
UE score between T1 and T0.

60
MEP
MEPs were assessed as peak-to-peak amplitudes and RMT computation. One patient in each
group recovered MEP in their affected hand. Patients of both group displayed stable peak-to-peak
amplitude in the unaffected hand apart from two patients in the BCI group who displayed extreme
variations (FIgure 43). No significant differences between groups appeared from the RMT
analyses, neither in the affected nor the unaffected hemisphere (Figure 44).

Figure 43. Left figure shows the MEP change in the affected hand between T1 and T0. Right figure shows the MEP change in the
unaffected hand between T1 and T0.

Figure 44. Left figure shows the decrease of RMT in percentage in the ipsilesional hemisphere between T1 and T0. Right figure
shows the decrease of RMT in percentage in the contralesional hemisphere between T1 and T0.

61
FC
The analysis of resting-state FC focused on the alpha-band, because previous work has suggested
that it influences motor performance (Dubovik et al., 2012b; Guggisberg et al., 2017). The primary
motor cortex and dorsol premotor cortex were used as ROIs based on previous results
(Guggisberg et al., 2017). Global alpha-band FC between these motor ROIs and the rest of brain
were similar between both groups at T0 and T1 (Figure 45). However, when comparing the FC
evolution between T1 and T0, patients in the BCI group tended to lower and reduce their global
alpha-band FC between ipsilesional motor ROIs and the rest of the brain (M=-0.2302, SD=0.5452)
compared to SHAM patients (t(18)=-2.2088, p=0.0405). This FC difference was correlated with
FMA-UE change (r=-0.5, p=0.0391) where low FCdiff corresponds to a higher increase of the FMA-
UE score.

Figure 45. The FC is calculated in the ipsilesional dorsal premotor (PMd) area in the alpha-band for the BCI and SHAM group. Top
left figure shows the FC at T0. Top right figure shows the FC at T1. Bottom left figure shows the difference in FC between T1 and
T0. Bottom right figure shows the correlation between the difference in FC and FMA-UE between T1 and T0.

2.2.4. Discussion
The motor benefit of BCI-FES intervention is modest. In spite of non-significant results, we could
observe a good recovery pattern in 3 patients that belonged to the BCI group. Indeed, their FMA-
UE were dragged toward a proportional recovery that was absent in all patients of the SHAM
group. We could suspect that these fortunate patients corresponded to the fitters patients who
started with a poor FMA-UE (Bonkhoff et al., 2020). Because the clinical results are not significant

62
between the BCI and the SHAM group we cannot conclude if the good motor outcome of these
patients is due to chance or an actual effect of the BCI.
Spasticity, present in around 30% of stroke patients, is correlated with negative motor outcomes
(Watkins et al., 2002). It appears essentially during passive movement. Spasticity can develop at
any time after stroke onset and can be due to either early immobilization or overactivity (Ward,
2002). The study protocol could theoretically have aggravated spasticity due to the FES
stimulation in addition to the intensive training of the upper limb. On that matter, our results are
reassuring, showing that not only BCI intervention does not lead to more upper limb spasticity
but even seems to prevent it. On the other hand, this reduction of early spasticity in the upper
limb did not seem to facilitate motor recovery.
Somatosensory perception and motor function are both positively correlated to functional
activities. Although a severe somatosensory deficit does not prevent patients from reducing their
motor impairment, its integrity is needed necessary for maximum recovery of motor functions.
(Zandvliet et al., 2020) Our BCI-FES setup offers such sensory stimulations first as a light touch
sensation during the calibration BCI sessions, second as a muscular contraction reward during the
online BCI sessions. The sensory stimulations encourage patients’ movement execution by leading
their attention to the proper muscles that need to be contracted. Our treatment protocol could
potentially have improved the sensory impairment of patients by applying a sensory sensation
with sensory and muscular stimulations (Biasiucci et al., 2018; Smania et al., 2003), but we found
no improvement in relation the group assignement.
Patients display the same FIM evolution between T0 and T2 in the 3 groups. In all three groups,
patients started with a low FIM score and improved similarly. It is important to note that at least
part of this increase in FIM score could be explained by the orientation of rehabilitation therapies
toward autonomy gain instead of pure motor deficit reduction. During occupational therapy
sessions, activities are taught taking into account specific motor deficits and practicing
compensation strategies with sometimes support tools.
FA asymmetry does not show significant differences between the BCI and the SHAM group. The
asymmetry did not worsen in both groups, from which we deduce that no major secondary white
matter degradation happened. From previous studies, we know that patients with severe
impairment as the ones recruited for this study have extensive secondary white matter
degradation in the affected hemisphere (Guggisberg et al., 2017a). The absence of such
degradation in this study is likely because the pre and post-assessments were separated only by
4 weeks, which may be too short to reveal major changes. Alternatively, the CST may have
stabilized in both the BCI and the SHAM group, possibly because of added therapies and the
motivation of the patients (Lang et al., 2015). However, this should then also have led to better
motor recovery. We also could not elicit new MEPs in the affected hemisphere in both groups
which is consistent with the low motor recovery of the patients. This is in accordance with the
literature, as (Yuan et al., 2020) already experienced the absence CST integrity recovery in their
BCI-guided robot hand training.
As described in chapter 1.3.4., FC changes are dependent on the CST integrity, its restoration
could have helped enhance cortical interactions for motor improvement purposes. Although this

63
was not the case in our study, the maintenance of the remaining white fibers may have allowed
the neural network process to occur. In the BCI group, patients lowered their FC while the
evolution is more uneven with a tendency to raise neural interactions in the SHAM group. BCI-
FES montage has already been proved to increase cortical activation and hemodynamic response
(Wang et al., 2019). However, the patients who improved well with real BCI training showed a
negative FC change which early connectivity seems to have benefitted their recovery. The way
we computed FC was from one ROI in regards to the rest of the brain. This global FC might have
decreased to make room for more selective connectivity for the motor network during
rehabilitation.
BCI therapy is a promising tool that has already demonstrated its positive effect in stroke patients
(Frolov et al., 2017; Mane et al., 2020b). These interfaces can be implemented in various manners
to decode motor intention. (Yuan et al., 2020) took alpha suppression signal to trigger a robot.
(Jin et al., 2019) increased classification accuracy in motor imagery BCI with correlation-based
channel selection. (Tariq et al., 2020) used event-related (de)synchronization with mu and beta
characteristics to classify lower limb movement. (Kevric & Subasi, 2017) compared various signal
decomposition methods and found that wavelet signal decomposition brought the highest
classification accuracy. Some implementations may be more appropriate for certain populations.
Therefore, it would be interesting to look at the characteristics of stroke patients to deduce the
best BCI to offer them.
In many previous studies, the protocol included 15 or even 20 BCI training sessions (Tang et al.,
2015; Yuan et al., 2020). In our case, only 8 training sessions were applied which may have limited
the BCI from offering its full potential to the patients.
As any rehabilitation tool, not all patients respond well to BCI therapy. Up to 30% of users are not
able to interact with the device while the others do not automatically show high performance
(Jeunet et al., 2016). This so-called BCI illiteracy can be overcome by rethinking the performance
threshold. It would help define the population that would best benefit from this type of
intervention. For instance, precise metrics could help screen patients with sufficient performance
then offer more individualized BCI training experiences (Thompson, 2018).
Another component to BCI performance is the psychological availability of the patients, i.e., the
presence or absence of fatigue or frustration towards an exercise accomplishment (Myrden &
Chau, 2015). Indeed, our severely impaired patients suffer from extreme fatigue, which in some
cases led to reduced therapy. For the ones that are strong enough to pursue all therapies, the
frustration of visually under-achievement due to a low success rate of the BCI classifier can impact
their attention which leads to reduced motivation.
Although the usage of a closed-loop BCI-FES showed its efficacy in previous studies (Biasiucci et
al., 2018; Wang et al., 2019), in practice not all patients could tolerate the device as we intended
to use it. In fact, with an authorized maximum of 30mA per FES paired electrode and always
remaining under patients’ discomfort, some patients could not feel the sensory stimulation while
others could not support an intensity high enough to elicit a muscular contraction that would
reward their cerebral effort during online sessions. These differences in sensory perception did

64
not allow us to offer the optimal BCI-FES setup for all patients, and this could have affected our
outcome.
(Leeb et al., 2013) already reported that only 50% of their patients performed well enough to
control the end application of their BCI system since controlling their EEG signal during online
sessions does not guarantee an automatic trigger of the controlled application. This control
difficulty can come from a bad classification of the BCI. Although the classifier does not need to
be perfect, we could compare the motor performance and the accuracy of the classifier.
BCI-FES is a promising setup that still needs some readjustment to better suit the severely
impaired patient on their sensation, fatigue, and BCI technical aspect. Our study is still ongoing.
The protocol now allows the patients to benefit from 12 training BCI sessions instead of 8.

65
 2.3. Study 3

Feasibility of Reconstructing Source Functional Connectivity with Low-Density EEG

Dung A. Nguyen-Danse, Shobana Singaravelu, Léa A. S. Chauvigné1, Anaïs Mottaz,
Leslie Allaman, Adrian G. Guggisberg
Published in: Brain Topography volume 34, pages 709-719 (2021) (see appendix)

Contribution of NDDA: Analyzed the data, wrote the paper

Abstract
Objectives: Functional connectivity (FC) is increasingly used as target for neuromodulation and
enhancement of performance. A reliable assessment of FC with electroencephalography (EEG)
currently requires a laboratory environment with high-density montages and a long preparation
time. This study investigated the feasibility of reconstructing source FC with a low-density EEG
montage towards a usage in real life applications.
Methods: Source FC was reconstructed with inverse solutions and quantified as node degree of
absolute imaginary coherence in alpha frequencies. We used simulated coherent point sources
as well as two real datasets to investigate the impact of electrode density (19 vs. 128 electrodes)
and usage of template vs. individual MRI-based head models on localization accuracy. In addition,
we checked whether low-density EEG is able to capture inter-individual variations in coherence
strength.
Results: In numerical simulations as well as real data, a reduction of the number of electrodes led
to less reliable reconstructions of coherent sources and of coupling strength. Yet, when
comparing different approaches to reconstructing FC from 19 electrodes, source FC obtained with
beamformers outperformed sensor FC, FC computed after independent component analysis, and
source FC obtained with sLORETA. In particular, only source FC based on beamformers was able
to capture neural correlates of motor behavior.
Conclusion: Reconstructions of FC from low-density EEG is challenging, but may be feasible when
using source reconstructions with beamformers.

66
 3. Discussion
Thanks to medical advances and the emergence of stroke centers, motor outcomes in stroke
patients have improved in recent years (Li et al., 2015; Sheth & Jadhav, 2022). Unfortunately, still
too many patients remain with chronic deficits such as motor impairment in the upper limb. In
this thesis, we aimed at testing innovative interventions to reduce motor impairment. In two
clinical studies, we offered personalized approaches to fit two levels of motor deficit severity. In
the third study, we attempted the implementation of a neurofeedback targeting source FC with
a low-density EEG setup.
In this discussion, I will first present a general synthesis of the 3 studies of this thesis. In the second
and third parts, the study results are discussed in relation to spontaneous recovery and plasticity
principles. In the fourth part, the influence of thymic aspects on stroke outcome is described.
Then the limitations of our work are discussed before finishing with a portait of a possible future
of innovative rehabilitation and suggestions for future work.

3.1. General synthesis

The first study targeted mild to moderate upper limb motor impairment in stroke survivors.
Because previous work had suggested that high global FC of motor areas correlates with better
motor outcomes, we used 4 tDCS montages to determine the optimal setup for enhancing cortical
connectivity in order to improve patients’ motor performance. Unfortunately, we could not see
a clear improvement in the FMA-UE, our primary outcome, in our active montages. However,
patients in the HD-tDCS group showed motor improvement in the ARAT score change at T1
compared to SHAM. Anodal tDCS, as applied in the CONV and HD groups, seemed to have
prevented poor recovery patterns which were observed in some patients of the BIHEM and SHAM
groups. Patients in the HD group tended to show an increase of MEP amplitude and reduce RMT
after the intervention on the contralesional hemisphere. HD-tDCS seemed to be the montage that
best enhanced cortical excitability of the contralesional hemisphere while it did not influence the
ipsilesional hemisphere. In this study, we confirmed previous findings regarding FC in the beta
frequency band of the ipsilesional motor areas. First, high early FC appeared to predict better
motor outcome. Second, late arrival of increased connectivity in the ipsilesional motor area was
associated with worse recovery (Nicolo & Rizk et al., 2015b; Westlake et al., 2012).
The second study targeted severe upper limb motor impairment in stroke survivors. It consisted
in providing a BCI-FES closed-loop system where the application of the FES was contingent on the
patient’s activation of the motor cortex recognized by the BCI. We hypothesized that such a
system could help maintaining the CST integrity while initiating a cortical process of neural
interactions and eventually improve the clinical outcome. Although we have not found a
significant difference in our primary motor outcome, the FMA-UE score between the SHAM and
the BCI group, more patients developed spasticity in the SHAM group. BCI intervention seemed
to have a preventive action on spasticity which unfortunately did not transfer into better motor
recovery. When looking at the neurophysiological level, FA asymmetry did not show significant
differences between both groups. On the other hand, since we did not observe a major difference
between T1 and T0, the asymmetry did not worsen. From that, we deduced that no major

67
secondary white matter degradation happened during the 4-week observation period of the
study. As for FC, patients in the BCI group tended to lower their global alpha-band FC between
ipsilesional M1 and the rest of the brain compared to SHAM patients. This lowered FC was
associated with a higher FMA-UE increase.
The third study explored the feasibility of transferring neurofeedback training into clinical
practice. FC was already used as a neuromodulation target to enhance performance (Mottaz et
al., 2018). In a laboratory environment, the connectivity corresponded to a source FC
reconstructed with inverse solutions then quantified as node degree of absolute imaginary
coherence in alpha frequencies. Here, we aimed to reduce the sensor density of the EEG used for
the neurofeedback. We investigated various strategies with simulated signals and real datasets.
With high-density EEG signals, we looked at the differences in localization accuracy between
template vs. individual MRI-based head models, and between beamformer or sLORETA as an
inverse solution. With low-density EEG signals, we compared FC resulting from the use of
beamformer and sLORETA as an inverse solution using template head models, the PCA, and the
sensors. When reducing the sensor density, the implemented strategies reconstructed less
reliable coherent sources and coupling strength. Nevertheless, the algorithm which showed the
best performance was using beamformer to reconstruct the source FC. Moreover, this particular
method was able to capture the neural interactions that correlated with motor performance.

3.2. Spontaneous recovery and proportional rule

3.2.1. Fitters and non-fitters

Stroke patients take two distinct recovery trajectories described as fitters and non-fitters of the
proportional recovery rule (Prabhakaran et al., 2008; Winters et al., 2015). Patients, in particular
those with mild to moderate initial deficits, are assumed to recover along that rule which predicts
that they will recover 70% of their initial impairment. In the tDCS trial, we hoped that patients
benefitting from excitatory tDCS increase their motor recovery levels above these 70%. Instead,
we observed that our patients did not succeed in recovering above the proportional rule.
However, only patients in the CONV and HD group consistently followed it, while some patients
of the other groups showed less than the expected recovery. This can be due to the limitations of
our interventions that is later discuss in chapter 3.5. We must recall that the 2mA was not
achieved in all patients due to impedance reasons. It is questionable whether higher motor
outcomes would have been obtained with the application of 20-minute-sessions with 2mA in the
entire HD group. On another hand, we can challenge the proportional rule itself. The limitations
of this rule were already discussed e.g. in (Bonkhoff et al., 2020) who reported that only a subset
of patients follow this precept where mathematical coupling explained in chapter 1.1.2. led to an
overestimation of the model fit.
Patients with a FMA-UE at T0 below 15 points were labeled as severely impaired and classified in
the non-fitters group. They did not benefit from a proportional recovery and effective therapy is
still lacking for them. With our BCI-FES montage, we wanted to meet the therapy needs of this
group. This specific montage depended on the neural state of the patients and was adapted to

68
their condition. Unfortunately, the effects were limited despite positive results in previous studies
using BCI training. Our outcomes probably differed by the inclusion of much more severely
affected patients. Indeed, the other studies included patients with a FMA-UE baseline above 15
points (Cervera et al., 2018b; Mane et al., 2020b; Stinear et al., 2020). It would have been nice if
the positive results under BCI training would have be maintained in this severely impaired
population who currently has no effective treatment options for enhancing recovery.

3.2.2. The right time for the intervention

We performed both clinical trials on patients in the subacute phase of stroke (first 3 months),
which we considered the ideal time for intensive rehabilitation (Veerbeek et al., 2014). During
this phase, a window of plasticity opens that allows for maximum motor recovery potential (Joy
& Carmichael, 2021). Regarding the application of tDCS, a stimulation during the acute phase (first
days) can be deleterious by causing neuronal toxicity with overstimulation (Pavlova et al., 2020).
During the chronic phase, the deficit in patients is stable, thus a tDCS application during the
chronic phase allows a better visualization of its clinical effect compared to a stable baseline
(Wessel et al., 2015), but the effect are more limited and the baseline assessment could reflect a
loss of capacity rather than a loss of function (Stinear et al., 2020). A compromise is present during
the subacute phase where tDCS raises neural growth accompanied by neural protection (Pavlova
et al., 2020).
In their meta-analysis on BCI training, (Bai et al., 2020) mainly found studies including chronic
patients. Only one study recruited patients between 3 and 6 months after stroke onset, but this
period is already far from the starting point of the subacute phase. Nonetheless, positive results
were still obtained with that population and the authors were optimistic about the use of BCI in
the future of rehabilitation. (Mane et al., 2020a) found 4 studies that included subacute patients
in their review of BCI application in stroke. 2 among these 4 studies showed an improvement of
FMA that was higher than the control group while the other 2 did not have the FMA as a primary
outcome. In our study, only the Ashworth assessment displayed a positive outcome while the BCI
and the SHAM group evolved similarly in the other motor assessments. This is consistent with the
review of (Stinear et al., 2020) where they already noticed these neutral results in subacute stroke
patients. In spite of some studies recruiting acute and early subacute patients, these neutral
results were observed in all types of interventions: interventions, training, pharmacological, and
neuromodulation. Although results from patients in the chronic stage appear to show more
positive results, the authors warned about the confusion between impairment and capacity.
Because in the chronic phase, it is not clear if the pre-intervention scores are due to patients’
actual impairment or due to a reduced capacity coming from a lack of training. Thus, the motor
improvement can be a reconditioning of the patients rather than a reduction of their deficit. As a
solution, the authors advocated for better study designs in order to increase number of recruited
participants and reduce the variability within patients.

69
Focusing back on the use of BCI, this technology requires a certain degree of attention that may
be missing during the acute phase. Conversely, the main recovery potential is during the first 3
months after stroke thus rendering more chronic patients less susceptible to improvements.
Targeting the subacute phase is thus an optimal time window to try new therapies, as abundant
recovery mechanisms are still ongoing. The subacute phase is also the critical time where the
therapists often decide to switch for compensation exercises in patients with severe impairment,
thus abandoning specific motor training. The possibility to induce significant gains could motivate
therapists to stick to restitution training. Although an increase in motor performance is missing,
when observing FC in stroke patients, our results confirmed previous studies that displayed early
enhancement of functional connectivity as a marker for good recovery while late FC was
associated with worse recovery. This finding may support to offer innovative therapies as soon as
possible.

3.3. Plasticity principles

Brain damage occurs during the hyperacute phase where the cells' excitotoxicity leads to
neuronal death within the first hours. The damage process continues with the apoptosis in these
cells during the acute phase, which last few days. During this period, any additional events such
as infections or a cortical excitability enhancement can aggravate the damage. Then, the subacute
phase takes place in the following months. As the inflammation reduces, neural plasticity arises
before stabilizing in the chronic phase. Although less impressive, motor recovery is still possible
during the chronic phase at the cost of more intensive training (Joy & Carmichael, 2021).
When performing TMS assessment, we saw that patients in the tDCS study mostly had MEP
responses unlike patients in the BCI study. These responses indicates a better CST integrity of the
tDCS patients than the BCI patients which is in accordance with their respective motor ability.
Unfortunately, we were unable to reduce ipsilesional RMT or increase MEP amplitude in the
active tDCS montages, nor did we elicit MEP responses in patients undergoing BCI sessions.
Although we could not see enhanced cortical excitability, patients in the active arms of the tDCS
study did not have a reduced MEP amplitude or an increased RMT, and patients trained with BCI
maintained their FA asymmetry between T1 and T0. This suggests that in both studies, patients
did not experience secondary CST degeneration. As a reminder, only 4 weeks separated the post
from the pre-MRI. Secondary degeneration as seen in (Guggisberg et al., 2017a) was observed in
MRI images separated by 2 months, thus it is possible that not seeing the degeneration in our
images resulted from the short time between the MRIs.

Following structural lesions, the brain first loses neural network interactions before showing
plasticity thanks to the perilesional tissues (Westlake et al., 2012). This plasticity can be observed
in multiple ways with FC assessment. The study of FC reveals that cortical connectivity appears in
the different frequency bands depending on the person’s clinical state and time since stroke
onset. In resting-state FC analyses, healthy participants show spontaneous activity in the alpha-
band. FC shifts from the alpha to the beta-band in acute stroke patients (Nicolo & Rizk, et al.,
70
2015b). This shift corresponds to the post-stroke period where a GABAergic environment is
present in the brain (Joy & Carmichael, 2021) and matches the activity we saw in our tDCS
patients. Starting 1 month after stroke onset, as neural interactions modify, late beta-band
connectivity is negatively correlated with motor performance while alpha-band connectivity
returns. In fact, we can definitively see this resurgence where alpha is again a predictor of motor
improvement in chronic patient who continued regular rehabilitation (Westlake et al., 2012). Our
BCI study included patients up to 8 weeks after their stroke onset. This extended time window
for inclusion can explain the neural interactions in the alpha band unlike patients of the tDCS
study whose connectivity appears in the beta band and who are included during the first 4 weeks
after stroke. In both studies, late FC was associated with a lower motor outcome. To assess
connectivity we use the FC of a ROI with regard to the rest of the brain. This global FC plays an
important role if it comes early enough. As seen in figure 46, the subacute phase is comparable
to the developmental phase that we experienced between birth and adulthood. Similar to
learning, global FC may decrease in favor of a more selective connectivity between dedicated
areas. To assess connectivity of specific tasks in stroke patients, (Vinehout et al., 2019) followed
global connectivity with mean fMRI time series and local connectivity with voxel time series of
selected motor ROIs. They found that during a taping task of the paretic foot, local connectivity
increased while global connectivity remained the same. It hints that for a good recovery, global
FC from M1 to the rest of the brain is needed at the early stage of stroke before becoming a
specialized connectivity between the motor areas of both hemispheres during rehabilitation.

We discussed previously about how late FC was related to worse motor outcome but also about
how alpha-band connectivity returns in the chronic phase. (Liu et al., 2016) already observed a
change of FC in the sensorimotor regions after stroke before evolving back to a connectivity level
comparable to healthy participants. In our second study, global alpha-band FC was more reduced
in patients undergoing BCI sessions than SHAM. It seems that our BCI training enabled the refining
of FC to motor areas. Future analyses will test the hypothesis of local FC between motor areas.

Figure 46. Parallels between windows of plasticity in development and stroke (Joy & Carmichael, 2021)

71
As seen above and in chapter 1.2.2., various calculations allow the examination of FC. In our third
study, finding the right methodology to assess connectivity is key to enabling a technology
transfer toward clinical practice. Inpatients during the subacute phase and patients at home could
benefit from a mobile neurofeedback system targeting FC. Our first strategy was to find a way to
compute source FC (in opposition to sensor FC) based on a low-density EEG recording and without
the use of an individualized head model. In numerical simulations and real datasets, using the
MVBF inverse solution offered the best trade-off between FC error localization, FC reproduction
fidelity, and FC correlation with motor performance. Then, when applying this solution to a newly
acquired dataset recorded from a low-density-dry-electrode montage, we could not replicate the
findings of behavioral association and FC. It is difficult to distinguish if this is due to the signal
quality of the dry electrodes, the low-density algorithm used to calculate FC, or both. Indeed the
simplification in setup introduced errors that affected the accuracy of the calculated FC, which
highlighted the difficulties of computing source FC with low-density EEG. Yet, neuromodulation
of cortical connectivity seems to be a great target for motor learning. That is why further studies
should be done to investigate how FC neuromodulation can become a therapeutic routine. In our
approaches, it turns out that for better neurofeedback performance, FC computed from the
source should be preferred over FC calculated from the sensors. There, we focused on the
connectivity of the alpha-band in the right precentral area for healthy participants. As discussed
earlier, in our stroke patients, FC appeared in different frequency bands and ROIs depending on
their phase from stroke onset. We cannot exclude that with better signals, and an intermediate
EEG density coverage, MVBF could obtain a better reconstruction of cortical connectivity, thus
bringing different outcome. To sum up, although our approach of FC reconstruction with low-
density did not end up in an implemented turnkey system, our exploration offers prospects for
improvement.

Another key factor for motor outcome is CST integrity. CST restoration or at least the non-
degeneration can influence the motor outcome (Guggisberg et al., 2017c). Unfortunately, our BCI
study that focused on the CST did not demonstrate an effect on CST integrity.

For the time being, 2 patients responded well to the BCI treatment. They had small initial CST
damage accompanied by a high FC in the first recording. Thus, we can suspect that they would
have showed a better outcome also spontaneously, independently of the usage of BCI. For severe
patients, there are still no proven solutions. Previous studies showing a positive impact of BCI in
stroke patients often recruited patients with less impairment (Stinear et al., 2020).
Finally, regarding the contralesional hemisphere, its influence remains unclear. In the tDCS study,
the anodal montages regularly stood out from the others, in particular anodal HD. Patients with
that stimulation tended to have an increased MEP in their contralesional hemisphere and
lowereded their RMT. These changes translated into motor behavior seen in the ARAT change
between T1 and T0. Although we targeted the ipsilesional motor area, it seems that some
influence occurred in the contralesional side. In contrast, no influence on the contralesional
hemisphere was observed in the SHAM-tDCS group where no stimulation was delivered or in the

72
BIHEM-tDCS where the contralesional side was expected to be inhibited with the cathode. (Dodd
et al., 2017) previously described that after a stroke, the interhemispheric imbalance leads to an
increase of cortical excitability of the contralesional hemisphere. This illustrates that the acutally
observed effects of tDCS often differs from the ones that we would expect based on our simple
models. This underlines the importance of imaging study as the ones conducted here to examine
the effects of neuromodulation at different time points after stroke.
For the BCI study, the lack of MEPs with the affected hemisphere and the absence of evolution in
the unaffected hemisphere do not allow any conclusion regarding the role of the contralesional
hemisphere. As a reminder, our training used ipsilesional motor EEG channels as features for the
classifier of the BCI which theoretically should not influence the contralesional hemisphere. It
should be noted that in previous studies, BCI setup could improve motor performance by
targeting either the contralesional or the ipsilesional side, although, the contralesional BCI design
recruited more severely impaired patients.

3.4. Thymic aspects in stroke survivors and rehabilitation

Stroke is a sudden accident that changes one’s body in a few hours, leaving the patients stunned
about their new situation. Motivation becomes key to taking full advantage of the intensive
rehabilitation care offered to them where reward training can reduce the impairment (Widmer
et al., 2022). At 6 months after stroke onset, (López-Espuela et al., 2020) found more than 40%
of their patients suffering from depression. Depression was in turn associated with less recovery
after stroke. In a more general way, mood, cognition, and fatigue affect at least 20% of stroke
survivors, influencing their attention during the crucial period of therapy. Early assessments and
psychological monitoring can reduce such obstacles (Lanctôt et al., 2020). These effects are also
relevant in clinical studies. To evaluate their mood, patients can answer psychological
questionnaires assessing well-being or motivation (Calabrò et al., 2015; Fujioka et al., 2018;
White et al., 2012). In spite of their consent to participate in our studies, patients exhibited
apprehension toward the tDCS, TMS, and EEG. In the tDCS study, once we explained and showed
the stimulation, patients underwent the therapy as if it was their regular sessions. On the other
hand, in the BCI study patients look forward to attending the BCI training. After a long week of
pre-intervention assessments and 2 BCI calibration sessions, they can feel disappointed when
they cannot trigger the FES for a muscular contraction reward. If that is the case, session after
session, patients could lose interests and were no longer concentrated in their task. Added to a
poor condition, their lack of attention may have lowered their performance.

3.5. Limitations
Our studies have some limitations that could be improved for future work.
For our clinical studies, the number of patients is low. In the tDCS pilot phase, only 24 patients
were recruited to observe a tendency before proceeding to the comparison of one active tDCS
and SHAM. The number of mildly and moderately impaired patients was smaller than projected.

73
In the BCI study, out of 36 recruited individuals, only 20 patients completed the protocol when 32
were needed to analyze the effect of our intervention. As described in chapter 2.2.3., a group of
drop-out patients was created. Unlike in the tDCS study, a significant number of patients decided
to quit or were excluded from the study. These drop-outs were mainly due to their fatigue which
went hand in hand with their severe impairment. It is interesting to see that in the BCI study, we
were able to involve 26 patients between 2018 and 2019, then only 9 patients between 2020 and
2021. We cannot help but think that the sanitary crisis due to COVID-19 influenced our ability to
include stroke patients. For the tDCS study, 19 patients were involved before the crisis and 9 after.
Regarding material difficulties, we note the difficulty of getting good impedance from the bi-
hemispheric and the HD-tDCS. In the bi-hemispheric montage, because the electrodes were on
both sides of the head, often they would slip upwards not allowing a good maintenance of the
set to obtain a good impedance. The only consequence of that issue was that this stimulation
started at the same time the therapy began and not 5 to 10 minutes before the therapy. The HD
montage, on its side, could not always reach 2mA. Sometimes, the intensity was reduced to
perform the needed 20 minutes stimulation. It is possible that we might have observed a better
effect if the 2mA could have been delivered in all patients. In the assessment, the TMS showed
some challenges. The targetted muscle was the FDI which is rather small and distal. Many times
we could see a response from the wrist but could not count it as an MEP. The use of recordings
from more proximal muscles might have brought more insights into the CST integrity, allowing us
multiple channels to assess cortical excitability.
With the BCI study, the number of sessions in our protocol might have been insufficient (Cervera
et al., 2018b). Moreover, the training protocol could be improved both to gain more data for a
better performance of the classifier and to reduce patient fatigue during the sessions. Our
patients performed a series of flexion movements during one session and extension movements
during the next session. Each session contained 3 runs with 20 trials each. The trials were
separated by pauses. Another design could have included alternate and consecutive trials of
flexion and extension with a pause period after a series. This would have had the advantage of
alternating the movement in the same session and having 8 similar sessions. Instead, we have 4
sessions for flexion movement and 4 for the extension movement which might be too few to see
an effect in patients. The number of training would then be increased and patients would have
been less confused between the pause moment and the rest trial.
In the tDCS trial, therapists could choose among a list of active movement exercises. The hope
was to see if tDCS could become a routine in rehabilitation, and for that purpose, we did not want
to constrain the therapy into an analytical movement that is not so meaningful to be reproduced
in a real rehabilitation situation. The downside of letting the therapist choose the exercises is that
it is not possible to track the effect of a particular activity under tDCS stimulation. Balancing
between specific movements for easy monitoring and activities that are closest to what is being
done in therapy is a constant trade-off. Here, we chose to be closer to the therapeutic reality.
Both studies offered a particular training. The intervention lasted 2 weeks with the tDCS
stimulation and 3 weeks with the BCI. Even if patients were motivated when enrolling in the
protocol, we cannot rule out the possibility that their motivation decreases with each passing day
as their stay in rehabilitation is prolonged, but also that motivation may vary according to the
74
degree of their recovery. For that purpose, the BCI operator performed a motivational
questionnaire after the first and the last BCI training to follow-up the patients’ adhesion to the
protocol. In the end, although the answers collected were all positive, we considered that first,
the operator might have biased the participants’ answers, second, patients should have answered
this questionnaire at each sessions either alone or with a third party, third, the questionnaire
should have also assessed patients’ concentration.

3.6. The future of innovative rehabilitation intervention

For the time being, intensive rehabilitation focuses on restoring the body's functions (Veerbeek
et al., 2014). Although therapists strive to individualize their care as closely as possible to the
patient, it is still difficult to predict at the initial clinical assessment which therapeutical technique
would best reduce their deficits. Innovative rehabilitation will start by better defining the
rehabilitation journey according to the profile of the patients. As done in our assessments,
therapists could add TMS and DTI analyses in their diagnostic assessment before beginning their
therapy. MEPs and FA asymmetry could become clear indicators to help discriminate those who
need intensive therapy from those who need to follow a compensation training program focusing
on maximizing independence in daily living despite the impairment. These indications can better
hint the therapist on the tools and techniques to use. Likewise, tDCS is an easy tool that can
integrate the therapists’ arsenal. Since our first study revealed the potential of anodal tDCS, we
could imagine its application by therapist at each sessions where patients train to recover specific
functions. In parallel, to help therapists put patients in the most functional situation possible,
therapies like virtual reality, exoskeleton and BCIs offer an additional dimension to regular
therapies. As presented in chapter 1.4.3. and discussed above, BCI can give multiple outputs and
designs to target in various ways cortical activity.

3.7. Future work

The next step of study 1 is to enter the main phase of the study by comparing HD-tDCS and SHAM-
tDCS. In study 2, additional BCI training is provided to future patients to be consistent with
previous studies (Cervera et al., 2018b).
These projects show the relevance and feasibility of the implementation of innovative
treatments. To go even further, as we neuronavigate to perform MEPs with the TMS, we could
also neuronavigate to better localize M1 before applying the anodal electrode of tDCS. Now, this
stimulation might have come too late for our patients. Consequently, for a next clinical study it
would be interesting to start this complementary stimulation no later than 2 weeks after stroke
onset. In addition, after observing the association of early FC and motor recovery, the
combination of the tDCS action and the contingent close-loop BCI-FES should be considered to
maximize the recovery. Indeed, stimulating M1 while offering BCI training could enhance cortical
connectivity and restoring the CST integrity needed for higher motor performance.
Since patients in the chronic phase could still gain motor function, it would be interesting for them
to follow this therapy as soon as their fatigue allows it, which might be after 8 weeks or even after

75
3 months after stroke. We also discussed previously the influence of the contralesional
hemisphere in BCI studies. Our setup targets the ipsilesional area. During the training, the
operator strives to guide the patients to perform a movement without too much compensation.
These compensations are already taught to be avoided in regular therapy. However, since the
contralesional side did show some positive influence on motor recovery, the BCI could target the
contralesional motor area instead of the ipsilesional one. Another next step for this study is to
perform more BCI training sessions, thus harvesting the full potential of this technique. BCI in
general includes a wide range of developed algorithm that takes into consideration the
decomposition of EEG signal, the features selection method, the targeted area, the EEG density,
or the design of the task (Bundy et al., 2017; Kevric & Subasi, 2017; Meng et al., 2018; Shi et al.,
2021; Yuan et al., 2020). All these parameters offer many opportunities to improve the BCI
system.
In our third study, our goal to reduce the constraint of a source reconstruction connectivity with
a high-density EEG might have been too ambitious. Indeed going from a high-density EEG with gel
electrodes and an individualized MRI to low-density dry electrodes and a template head model
could benefit from more intermediary steps. For instance, the use of other EEG devices could be
a concession between a better accuracy in localization error and speed setup for clinical practice.
For instance, high-density EEG systems with sponge electrodes allow for a relatively quick setup
time while still providing good head coverage.
Our projects focused on FC of one area in regards to the rest of the brain. In their learning process
of recovering their functions, the connectivity of the patients could become more specialized.
Hence, the analyses of neural interactions between motor areas could reveal other dynamics in
the brain as patients train in therapy. In our analyses of FC, we only looked at the coupling of
regions of interest but not at the direction of the neural interaction. In this sense, effective
connectivity takes into account the influence of one neural system on another (Friston, 1994;
Horwitz, 2003). An analysis of effective connectivity could reveal additional insights on plasticity
and the effects of novel therapies.
Although our studies aimed to improve motor recovery of the upper limb in stroke patients, the
obtained results could be extrapolated to other deficits. For instance, aphasia and the lower limb
impairment are two other major deficits stroke patients can experience that could very much
benefit from the findings of upper limb studies.

76
 4. References
Allman, C., Amadi, U., Winkler, A. M., Wilkins, L., Filippini, N., Kischka, U., Stagg, C. J., & Johansen-
Berg, H. (2016). Ipsilesional anodal {tDCS} enhances the functional benefits of rehabilitation
in patients after stroke. Science Translational Medicine, 8(330), 330re1--330re1.
http://stm.sciencemag.org/content/8/330/330re1.abstract
Alstott, J., Breakspear, M., Hagmann, P., Cammoun, L., & Sporns, O. (2009). Modeling the impact
of lesions in the human brain. PLoS Computational Biology, 5(6).
https://doi.org/10.1371/journal.pcbi.1000408
Babault, N., Cometti, C., Maffiuletti, N. A., & Deley, G. (2011). Does electrical stimulation enhance
post-exercise performance recovery? European Journal of Applied Physiology, 111(10),
2501–2507. https://doi.org/10.1007/s00421-011-2117-7
Bai, Z., Fong, K. N. K., Zhang, J. J., Chan, J., & Ting, K. H. (2020). Immediate and long-term effects
of BCI-based rehabilitation of the upper extremity after stroke: A systematic review and
meta-analysis. Journal of NeuroEngineering and Rehabilitation, 17(1), 1–20.
https://doi.org/10.1186/s12984-020-00686-2
Barker, A. T., Jalinous, R., & Freeston, I. L. (1985). Non-invasive magnetic stimulation of human
motor cortex. The Lancet, 325(8437), 1106–1107. https://doi.org/10.1016/s0140-
6736(85)92413-4
Basser, P. J., & Pierpaoli, C. (2011). Microstructural and physiological features of tissues
elucidated by quantitative-diffusion-tensor MRI. Journal of Magnetic Resonance, 213(2),
560–570. https://doi.org/10.1016/j.jmr.2011.09.022
Beaulieu, L. D., & Milot, M. H. (2018). Changes in transcranial magnetic stimulation outcome
measures in response to upper-limb physical training in stroke: A systematic review of
randomized controlled trials. Annals of Physical and Rehabilitation Medicine, 61(4), 224–234.
https://doi.org/10.1016/j.rehab.2017.04.003
Beebe, J. A., & Lang, C. E. (2009). Relationships and responsiveness of six upper extremity function
tests during the first six months of recovery after stroke. Journal of Neurologic Physical
Therapy, 33(2), 96–103. https://doi.org/10.1097/NPT.0b013e3181a33638
Bell-Krotoski, J. A., Fess, E. E., Figarola, J. H., & Hiltz, D. (1995). Threshold Detection and Semmes-
Weinstein Monofilaments. Journal of Hand Therapy, 8(2), 155–162.
https://doi.org/10.1016/S0894-1130(12)80314-0
Beninato, M., Gill-Body, K. M., Salles, S., Stark, P. C., Black-Schaffer, R. M., & Stein, J. (2006).
Determination of the minimal clinically important difference in the FIM instrument in
patients with stroke. Archives of Physical Medicine and Rehabilitation, 87(1), 32–39.
https://doi.org/10.1016/j.apmr.2005.08.130
Bertrand, A. M., Fournier, K., Wick Brasey, M. G., Kaiser, M. L., Frischknecht, R., & Diserens, K.
(2015). Reliability of maximal grip strength measurements and grip strength recovery
following a stroke. Journal of Hand Therapy, 28(4), 356–363.

77
 https://doi.org/10.1016/j.jht.2015.04.004
Bertrand, A. M., Mercier, C., Bourbonnais, D., Desrosiers, J., & Grave, D. (2007). Reliability of
maximal static strength measurements of the arms in subjects with hemiparesis. Clinical
Rehabilitation, 21(3), 248–257. https://doi.org/10.1177/0269215506070792
Biasiucci, A., Leeb, R., Iturrate, I., Perdikis, S., Al-Khodairy, A., Corbet, T., Schnider, A., Schmidlin,
T., Zhang, H., Bassolino, M., Viceic, D., Vuadens, P., Guggisberg, A. G., & Millán, J. D. R. (2018).
Brain-actuated functional electrical stimulation elicits lasting arm motor recovery after
stroke. Nature Communications. https://doi.org/10.1038/s41467-018-04673-z
Biernaskie, J., Szymanska, A., Windle, V., & Corbett, D. (2005). Bi-hemispheric contribution to
functional motor recovery of the affected forelimb following focal ischemic brain injury in
rats. European Journal of Neuroscience, 21(4), 989–999. https://doi.org/10.1111/j.1460-
9568.2005.03899.x
Bigourdan, A., Munsch, F., Coupé, P., Guttmann, C. R. G., Sagnier, S., Renou, P., Debruxelles, S.,
Poli, M., Dousset, V., Sibon, I., & Tourdias, T. (2016). Early Fiber Number Ratio Is a Surrogate
of Corticospinal Tract Integrity and Predicts Motor Recovery After Stroke. Stroke, 47(4),
1053–1059. https://doi.org/10.1161/STROKEAHA.115.011576
Blabe, C. H., Gilja, V., Chestek, C. A., Shenoy, K. V, Anderson, K. D., & Henderson, J. M. (2015).
Assessment of brain–machine interfaces from the perspective of people with paralysis.
Journal of Neural Engineering, 12(4), 043002. https://doi.org/10.1088/1741-
2560/12/4/043002
Blackburn, M., Van Vliet, P., & Mockett, S. P. (2002). Reliability of measurements obtained with
the Modified Ashworth Scale in the lower extremities of people with stroke. Physical
Therapy, 82(1), 25–34. https://doi.org/10.1093/ptj/82.1.25
Boddington, L. J., & Reynolds, J. N. J. (2017). Targeting interhemispheric inhibition with
neuromodulation to enhance stroke rehabilitation. Brain Stimulation, 10(2), 214–222.
https://doi.org/10.1016/J.BRS.2017.01.006
Bohannon, R. W., & Smith, M. B. (1987). Interrater reliability of a modified ashworth scale of
muscle spasticity. Physical Therapy, 67(2), 206–207. https://doi.org/10.1007/978-1-4471-
5451-8_105
Boissy, P., Bourbonnais, D., Carlotti, M. M., Gravel, D., & Arsenault, B. A. (1999). Maximal grip
force in chronic stroke subjects and its relationship to global upper extremity function.
Clinical Rehabilitation, 13(4), 354–362. https://doi.org/10.1191/026921599676433080
Bolognini, N., Pascual-Leone, A., & Fregni, F. (2009). Using non-invasive brain stimulation to
augment motor training-induced plasticity. https://doi.org/10.1186/1743-0003-6-8
Bonkhoff, A. K., Hope, T., Bzdok, D., Guggisberg, A. G., Hawe, R. L., Dukelow, S. P., Chollet, F., Lin,
D. J., Grefkes, C., Bowman, H., & Philip, J. (2021). Recovery after stroke: the severely impaired
are a distinct group. Journal of Neurology, Neurosurgery & Psychiatry, 0, jnnp-2021-327211.
https://doi.org/10.1136/JNNP-2021-327211

78
Bonkhoff, A. K., Hope, T., Bzdok, D., Guggisberg, A. G., Hawe, R. L., Dukelow, S. P., Rehme, A. K.,
Fink, G. R., Grefkes, C., & Bowman, H. (2020). Bringing proportional recovery into proportion:
Bayesian modelling of post-stroke motor impairment. Brain, 143(7), 2189–2206.
https://doi.org/10.1093/brain/awaa146
Buch, E. R., Rizk, S., Nicolo, P., Cohen, L. G., Schnider, A., & Guggisberg, A. G. (2016a). Predicting
motor improvement after stroke with clinical assessment and diffusion tensor imaging. TL -
86. In Neurology: Vols. 86 VN-r (Issue 20). https://doi.org/10.1212/WNL.0000000000002675
Buch, E. R., Rizk, S., Nicolo, P., Cohen, L. G., Schnider, A., & Guggisberg, A. G. (2016b). Predicting
motor improvement after stroke with clinical assessment and diffusion tensor imaging.
Neurology, 86(20), 1924–1925. https://doi.org/10.1212/WNL.0000000000002675
Bundy, D. T., Souders, L., Baranyai, K., Leonard, L., Schalk, G., Coker, R., Moran, D. W., Huskey, T.,
& Leuthardt, E. C. (2017). Contralesional Brain-Computer Interface Control of a Powered
Exoskeleton for Motor Recovery in Chronic Stroke Survivors. Stroke, 48(7), 1908–1915.
https://doi.org/10.1161/STROKEAHA.116.016304
Burke, J. F., Skolarus, L. E., Adelman, E. E., Reeves, M. J., & Brown, D. L. (2014). Influence of
hospital-level practices on readmission after ischemic stroke. Neurology, 82(24), 2196–2204.
https://doi.org/10.1212/WNL.0000000000000514
Cakar, E., Akyuz, G., Durmus, O., & Bayman, L. (2016). The relationships of motor-evoked
potentials to hand dexterity , motor function , and spasticity in chronic stroke patients : a
transcranial magnetic stimulation study. Acta Neurologica Belgica, 116(4), 481–487.
https://doi.org/10.1007/s13760-016-0633-2
Calabrò, R. S., De Cola, M. C., Leo, A., Reitano, S., Balletta, T., Trombetta, G., Naro, A., Russo, M.,
Bertè, F., De Luca, R., & Bramanti, P. (2015). Robotic neurorehabilitation in patients with
chronic stroke: Psychological well-being beyond motor improvement. International Journal
of Rehabilitation Research, 38(3), 219–225.
https://doi.org/10.1097/MRR.0000000000000114
Caparelli-Daquer, E. M., Zimmermann, T. J., Mooshagian, E., Parra, L. C., Rice, J. K., Datta, A.,
Bikson, M., & Wassermann, E. M. (2012). A Pilot Study on Effects of 4×1 High-Definition tDCS
on Motor Cortex Excitability. Conf Proc IEEE Eng Med Biol Soc, 735–738.
https://doi.org/10.1109/EMBC.2012.6346036.A
Carey, L., Walsh, A., Adikari, A., Goodin, P., Alahakoon, D., De Silva, D., Ong, K. L., Nilsson, M., &
Boyd, L. (2019). Finding the Intersection of Neuroplasticity, Stroke Recovery, and Learning:
Scope and Contributions to Stroke Rehabilitation. Neural Plasticity, 2019.
https://doi.org/10.1155/2019/5232374
Carmichael, S. T., & Chesselet, M.-F. (2002). Synchronous neuronal activity is a signal for axonal
sprouting after cortical lesions in the adult. Journal of Neuroscience, 22(14), 6062–6070.
http://www.jneurosci.org/content/22/14/6062.short
Cervera, M. A., Soekadar, S. R., Ushiba, J., Millán, J. del R., Liu, M., Birbaumer, N., & Garipelli, G.

79
 (2018a). Brain-computer interfaces for post-stroke motor rehabilitation: a meta-analysis.
Annals of Clinical and Translational Neurology, 5(5), 651–663.
https://doi.org/10.1002/acn3.544
Cervera, M. A., Soekadar, S. R., Ushiba, J., Millán, J. del R., Liu, M., Birbaumer, N., & Garipelli, G.
(2018b). Brain-computer interfaces for post-stroke motor rehabilitation: a meta-analysis.
Annals of Clinical and Translational Neurology, 5(5), 651–663.
https://doi.org/10.1002/acn3.544
Chalah, M. A., Lefaucheur, J. P., & Ayache, S. S. (2017). Long-term effects of tDCS on fatigue, mood
and cognition in multiple sclerosis. Clinical Neurophysiology, 128(11), 2179–2180.
https://doi.org/10.1016/j.clinph.2017.08.004
Chang, K.-Y., Mizutani-Tiebel, Y., Soldini, A., Padberg, F., & Keeser, D. (2021). tDCS and Functional
Connectivity. Transcranial Direct Current Stimulation in Neuropsychiatric Disorders, 159–
172. https://doi.org/10.1007/978-3-030-76136-3_9
Chang, L., Jones, D. K., & Pierpaoli, C. (2005). RESTORE : Robust Estimation of Tensors by Outlier
Rejection. 1095, 1088–1095. https://doi.org/10.1002/mrm.20426
Chen, H.-M., Chen, C. C., Hsueh, I.-P., Huang, S.-L., & Hsieh, C.-L. (2009). Test-Retest
Reproducibility and Smallest Real Difference of 5 Hand Function Tests in Patients With
Stroke. Neurorehabilitation and Neural Repair, 23, 435–440.
https://doi.org/10.1177/1545968308331146
Coast, N. (2011). Touch-Test Sensory Evaluator. 1–2. https://www.ncmedical.com/wp-
content/uploads/2011/07/Touch-Test-Sensory-Evaluator_11_web.pdf
Corbet, T., Leeb, R., Biasiucci, A., Zhang, H., Perdikis, S., & Millán, J. del R. (2015). BCI-NMES
therapy enhances effective connectivity in the damaged hemisphere in stroke patients.
International Brain-Computer Interface Meeting, 109. https://doi.org/10.3217/978-3-
85125-467-9-109
Corti, M., Patten, C., & Triggs, W. (2012). Repetitive Transcranial Magnetic Stimulation of motor
cortex after stroke: A focused review. American Journal of Physical Medicine and
Rehabilitation, 91(3), 254–270. https://doi.org/10.1097/PHM.0b013e318228bf0c
Cramer, S. C. (2008). Repairing the Human Brain after Stroke: I. Mechanisms of Spontaneous
Recovery. Ann Neurol, 63, 272–287. https://doi.org/10.1002/ana.21393
Cramer, S. C., & Riley, J. D. (2008). Neuroplasticity and brain repair after stroke. Current Opinion
in Neurology, 21(1), 76–82. https://doi.org/10.1097/WCO.0b013e3282f36cb6
Cramer, S. C., Sur, M., Dobkin, B. H., O’Brien, C., Sanger, T. D., Trojanowski, J. Q., Rumsey, J. M.,
Hicks, R., Cameron, J., Chen, D., Chen, W. G., Cohen, L. G., Decharms, C., Duffy, C. J., Eden, G.
F., Fetz, E. E., Filart, R., Freund, M., Grant, S. J., … Vinogradov, S. (2011). Harnessing
neuroplasticity for clinical applications. Brain, 134(6), 1591–1609.
https://doi.org/10.1093/brain/awr039

80
Crichton, S. L., Bray, B. D., & Mckevitt, C. (2016). Patient outcomes up to 15 years after stroke:
survival, disability, quality of life, cognition and mental health. J Neurol Neurosurg Psychiatry,
87, 1091–1098. https://doi.org/10.1136/jnnp-2016-313361
Cunningham, D. A., Machado, A., Janini, D., Varnerin, N., Bonnett, C., Yue, G., Jones, S., Lowe, M.,
Beall, E., Sakaie, K., & Plow, E. B. (2015). The Assessment of Inter-Hemispheric Imbalance
using Imaging and Non-Invasive Brain Stimulation in Patients with Chronic Stroke HHS Public
Access. Arch Phys Med Rehabil, 96(0), 94–103. https://doi.org/10.1016/j.apmr.2014.07.419
Cunningham, D. A., Varnerin, N., Machado, A., Bonnett, C., Janini, D., Roelle, S., Potter-Baker, K.,
Sankarasubramanian, V., Wang, X., Yue, G., & Plow, E. B. (2015). Stimulation targeting higher
motor areas in stroke rehabilitation: A proof-of-concept, randomized, double-blinded
placebo-controlled study of effectiveness and underlying mechanisms. Restorative
Neurology and Neuroscience, 33(6), 911–926. https://doi.org/10.3233/RNN-150574
de Morand, A. (2014). Le patient hémiplégique. In Pratique de rééducation neurologique (Elsevier
M, pp. 1–100). https://doi.org/10.1016/B978-2-294-71091-9.50001-8
De Vico Fallani, F., Richiardi, J., Chavez, M., & Achard, S. (2013). Graph analysis of functional brain
networks: practical issues in translational neuroscience. Philosophical Transactions of the
Royal Society, 369. https://doi.org/10.1098/rstb.2013.0521
Di Pino, G., Pellegrino, G., Assenza, G., Capone, F., Ferreri, F., Formica, D., Ranieri, F., Tombini, M.,
Ziemann, U., Rothwell, J. C., & Di Lazzaro, V. (2014). Modulation of brain plasticity in stroke:
a novel model for neurorehabilitation. Nature Reviews Neurology, 10(10), 597–608.
https://doi.org/10.1038/nrneurol.2014.162
Dimyan, M. A., & Cohen, L. G. (2010). Contribution of transcranial magnetic stimulation to the
understanding of Mechanisms of Functional Recovery After Stroke. Neurorehabilitation and
Neural Repair, 24(2), 125–135. https://doi.org/10.1177/1545968309345270.Contribution
Dobkin, B. H. (2007a). Brain-computer interface technology as a tool to augment plasticity and
outcomes for neurological rehabilitation. Journal of Physiology, 579(3), 637–642.
https://doi.org/10.1113/jphysiol.2006.123067
Dobkin, B. H. (2007b). Brain-computer interface technology as a tool to augment plasticity and
outcomes for neurological rehabilitation. The Journal of Physiology, 579(3), 637–642.
https://doi.org/10.1113/jphysiol.2006.123067
Dodd, K. C., Nair, V. A., & Prabhakaran, V. (2017). Role of the Contralesional vs . Ipsilesional
Hemisphere in Stroke Recovery. 11(September), 1–9.
https://doi.org/10.3389/fnhum.2017.00469
Donkor, E. S. (2018). Stroke in the 21st Century: A Snapshot of the Burden, Epidemiology, and
Quality of Life. Stroke Research and Treatment, 2018.
https://doi.org/10.1155/2018/3238165
Doppelmayr, M., Pixa, N. H., & Steinberg, F. (2016). Cerebellar, but not Motor or Parietal, High-
Density Anodal Transcranial Direct Current Stimulation Facilitates Motor Adaptation. Journal

81
 of the International Neuropsychological Society, 22, 928–936.
https://doi.org/10.1017/S1355617716000345
Doucet, B. M., Lam, A., & Griffin, L. (2012a). Neuromuscular Electrical Stimulation for Skeletal
Muscle Function. In YALE JoUrNAL oF BIoLoGY AND MEDICINE (Vol. 85).
Doucet, B. M., Lam, A., & Griffin, L. (2012b). Neuromuscular electrical stimulation for skeletal
muscle function. The Yale Journal of Biology and Medicine, 85(2), 201–215.
http://www.ncbi.nlm.nih.gov/pubmed/22737049
Doughty, C., Wang, J., Feng, W., Hackney, D., Pani, E., & Schlaug, G. (2016). Detection and
Predictive Value of Fractional Anisotropy Changes of the Corticospinal Tract in the Acute
Phase of a Stroke. Stroke, 47(6), 1520–1526.
https://doi.org/10.1161/STROKEAHA.115.012088
Dromerick, A. W., Edwards, D. F., & Diringer, M. N. (2003). Sensitivity to changes in disability after
stroke: A comparison of four scales useful in clinical trials. Journal of Rehabilitation Research
and Development, 40(1), 1–8. https://doi.org/10.1682/JRRD.2003.01.0001
Dubovik, S., Pignat, J.-M., Ptak, R., Aboulafia, T., Allet, L., Gillabert, N., Magnin, C., Albert, F.,
Momjian-Mayor, I., Nahum, L., Lascano, A. M., Michel, C. M., Schnider, A., & Guggisberg, A.
G. (2012a). The behavioral significance of coherent resting-state oscillations after stroke.
NeuroImage, 61(1), 249–257. https://doi.org/10.1016/j.neuroimage.2012.03.024
Dubovik, S., Pignat, J.-M., Ptak, R., Aboulafia, T., Allet, L., Gillabert, N., Magnin, C., Albert, F.,
Momjian-Mayor, I., Nahum, L., Lascano, A. M., Michel, C. M., Schnider, A., & Guggisberg, A.
G. (2012b). The behavioral significance of coherent resting-state oscillations after stroke.
NeuroImage, 61(1), 249–257. https://doi.org/10.1016/J.NEUROIMAGE.2012.03.024
Ekstrand, E., Lexell, J., & Brogårdh, C. (2016). Grip strength is a representative measure of muscle
weakness in the upper extremity after stroke. Topics in Stroke Rehabilitation, 23(6), 400–
405. https://doi.org/10.1080/10749357.2016.1168591
Fan, Y., Wu, C., Liu, H., Lin, K., Wai, Y., & Chen, Y. (2015). Neuroplastic changes in resting-state
functional connectivity after stroke rehabilitation. Frontiers in Human Neuroscience, 9, 546.
https://doi.org/10.3389/fnhum.2015.00546
Feng, W., Wang, J., Chhatbar, P. Y., Doughty, C., Landsittel, D., Lioutas, V.-A., Kautz, S. A., &
Schlaug, G. (2015). Corticospinal tract lesion load: An imaging biomarker for stroke motor
outcomes. Annals of Neurology, 78(6), 860–870. https://doi.org/10.1002/ana.24510
La prévention et la prise en charge des accidents vasculaires cérébraux en France, (2009)
(testimony of Elisabeth Fery-Lemonnier).
Friston, K. J. (1994). Functional and effective connectivity in neuroimaging: A synthesis. Human
Brain Mapping, 2(1–2), 56–78. https://doi.org/10.1002/hbm.460020107
Frolov, A. A., Mokienko, O., Lyukmanov, R., Biryukova, E., Kotov, S., Turbina, L., Nadareyshvily, G.,
& Bushkova, Y. (2017). Post-stroke rehabilitation training with a motor-imagery-based brain-

82
 computer interface (BCI)-controlled hand exoskeleton: A randomized controlled multicenter
trial. Frontiers in Neuroscience, 11(JUL), 400.
https://doi.org/10.3389/FNINS.2017.00400/BIBTEX
Fu, M., & Zuo, Y. (2011). Experience-dependent structural plasticity in the cortex. Trends in
Neurosciences, 34(4), 177–187. https://doi.org/10.1016/j.tins.2011.02.001
Fuchs, E., & Flügge, G. (2014). Adult neuroplasticity: More than 40 years of research. Neural
Plasticity, 2014. https://doi.org/10.1155/2014/541870
Fujioka, T., Dawson, D. R., Wright, R., Honjo, K., Chen, J. L., Chen, J. J., Black, S. E., Stuss, D. T., &
Ross, B. (2018). The effects of music-supported therapy on motor, cognitive, and
psychosocial functions in chronic stroke. Annals of the New York Academy of Sciences,
1423(1), 264–274. https://doi.org/10.1111/nyas.13706
Furlan, L., Conforto, A. B., Cohen, L. G., & Sterr, A. (2016). Upper limb immobilisation: A neural
plasticity model with relevance to poststroke motor rehabilitation. Neural Plasticity, 2016,
8176217.
http://europepmc.org/search?query=(DOI:10.1155/2016/8176217)%0Ahttp://www.hinda
wi.com/journals/np/2016/8176217/
Girard, G., Caminiti, R., Battaglia-Mayer, A., St-Onge, E., Ambrosen, K. S., Eskildsen, S. F., Krug, K.,
Dyrby, T. B., Descoteaux, M., Thiran, J. P., & Innocenti, G. M. (2020). On the cortical
connectivity in the macaque brain: A comparison of diffusion tractography and histological
tracing data. NeuroImage, 221, 117201. https://doi.org/10.1016/j.neuroimage.2020.117201
Gladstone, D. J., Danells, C. J., & Black, S. E. (2002). The Fugl-Meyer Assessment of Motor Recovery
after Stroke: A Critical Review of Its Measurement Properties. In Neurorehabilitation and
Neural Repair (Vol. 16, Issue 3, pp. 232–240). Sage PublicationsSage CA: Thousand Oaks, CA.
https://doi.org/10.1177/154596802401105171
Grefkes, C., Nowak, D. A., Eickhoff, S. B., Dafotakis, M., Küst, J., Karbe, H., & Fink, G. R. (2008).
Cortical Connectivity after Subcortical Stroke Assessed with Functional Magnetic Resonance
Imaging. Ann Neurol, 63, 236–246. https://doi.org/10.1002/ana.21228
Gregson, J. M., Leathley, M., Moore, A. P., Sharma, A. K., Smith, T. L., & Watkins, C. L. (1999).
Reliability of the tone assessment scale and the modified Ashworth scale as clinical tools for
assessing poststroke spasticity. Archives of Physical Medicine and Rehabilitation, 80(9),
1013–1016. https://doi.org/10.1016/S0003-9993(99)90053-9
Greicius, M. D., Krasnow, B., Reiss, A. L., & Menon, V. (2003). Functional connectivity in the resting
brain: A network analysis of the default mode hypothesis. Proceedings of the National
Academy of Sciences of the United States of America, 100(1), 253–258.
https://doi.org/10.1073/pnas.0135058100
Groppa, S., Oliviero, A., Eisen, A., Quartarone, A., Cohen, L. G., Mall, V., Kaelin-Lang, A., Mima, T.,
Rossi, S., Thickbroom, G. W., Rossini, P. M., Ziemann, U., Valls-Solé, J., & Siebner, H. R. (2012).
A practical guide to diagnostic transcranial magnetic stimulation: Report of an IFCN

83
 committee. Clin. Neurophysiol., 123(5), 858–882.
https://doi.org/10.1016/j.clinph.2012.01.010.A
Guggisberg, A. G., Dalal, S. S., Zumer, J. M., Wong, D. D., Dubovik, S., Michel, C. M., & Schnider, A.
(2011). Localization of cortico-peripheral coherence with electroencephalography.
NeuroImage, 57(4), 1348–1357. https://doi.org/10.1016/J.NEUROIMAGE.2011.05.076
Guggisberg, A. G., Koch, P. J., Hummel, F. C., & Buetefisch, C. M. (2019). Brain networks and their
relevance for stroke rehabilitation. In Clinical Neurophysiology (Vol. 130, Issue 7, pp. 1098–
1124). Elsevier Ireland Ltd. https://doi.org/10.1016/j.clinph.2019.04.004
Guggisberg, A. G., Nicolo, P., Cohen, L. G., Schnider, A., & Buch, E. R. (2017a). Longitudinal
Structural and Functional Differences Between Proportional and Poor Motor Recovery After
Stroke. Neurorehabilitation and Neural Repair. https://doi.org/10.1177/1545968317740634
Guggisberg, A. G., Nicolo, P., Cohen, L. G., Schnider, A., & Buch, E. R. (2017b). Longitudinal
Structural and Functional Differences Between Proportional and Poor Motor Recovery After
Stroke. Neurorehabilitation and Neural Repair, 31(12), 1029–1041.
https://doi.org/10.1177/1545968317740634
Guggisberg, A. G., Nicolo, P., Cohen, L. G., Schnider, A., & Buch, E. R. (2017c). Longitudinal
Structural and Functional Differences Between Proportional and Poor Motor Recovery After
Stroke. Neurorehabilitation and Neural Repair, 31(12), 1029–1041.
https://doi.org/10.1177/1545968317740634
Guggisberg, A. G., Rizk, S., Ptak, R., Di Pietro, M., Saj, A., Lazeyras, F., Lovblad, K. O., Schnider, A.,
& Pignat, J. M. (2014). Two Intrinsic Coupling Types for Resting-State Integration in the
Human Brain. Brain Topography, 28(2), 318–329. https://doi.org/10.1007/s10548-014-
0394-2
Guilbert, J. J. (2003). The world health report 2002 - reducing risks, promoting healthy life. In
Education for health (Abingdon, England) (Vol. 16, Issue 2).
https://doi.org/10.1080/1357628031000116808
Gustavsson, A., Svensson, M., Jacobi, F., Allgulander, C., Alonso, J., Beghi, E., Dodel, R., Ekman,
M., Faravelli, C., Fratiglioni, L., Gannon, B., Jones, D. H., Jennum, P., Jordanova, A., Jönsson,
L., Karampampa, K., Knapp, M., Kobelt, G., Kurth, T., … Olesen, J. (2011). Cost of disorders of
the brain in Europe 2010. European Neuropsychopharmacology, 21, 718–779.
https://doi.org/10.1016/j.euroneuro.2011.08.008
Hadjipapas, A., Hillebrand, A., Holliday, I. E., Singh, K. D., & Barnes, G. R. (2005). Assessing
interactions of linear and nonlinear neuronal sources using MEG beamformers: A proof of
concept. Clinical Neurophysiology, 116(6), 1300–1313.
https://doi.org/10.1016/j.clinph.2005.01.014
Hamada, M., Murase, N., Hasan, A., Balaratnam, M., & Rothwell, J. C. (2012). The Role of
Interneuron Networks in Driving Human Motor Cortical Plasticity.
https://doi.org/10.1093/cercor/bhs147

84
Hanten, W. P., Chen, W.-Y., Austin, A. A., Brooks, R. E., Carter, H. C., Carol, P. T., Law, A., Morgan,
M. K., Sanders, D. J., Christe, P. T., Swan, A., & Vanderslice, A. L. (1999). Maximum Grip
Strength in Normal Subjects from 20 to 64 Years of Age. Journal of Hand Therapy, 12(3), 193–
200.
Hatem, S. M., Saussez, G., della Faille, M., Prist, V., Zhang, X., Dispa, D., & Bleyenheuft, Y. (2016).
Rehabilitation of Motor Function after Stroke: A Multiple Systematic Review Focused on
Techniques to Stimulate Upper Extremity Recovery. Frontiers in Human Neuroscience, 10,
442. https://doi.org/10.3389/fnhum.2016.00442
Hawe, R. L., Scott, S. H., & Dukelow, S. P. (2019). Taking Proportional out of Stroke Recovery.
Stroke, 50(1), 204–211. https://doi.org/10.1161/STROKEAHA.118.023006
Heller, A., Wade, D. T., Wood, V. A., Sunderland, A., Hewer, R. L., & Ward, E. (1987). Arm function
after stroke: Measurement and recovery over the first three months. Journal of Neurology,
Neurosurgery and Psychiatry, 50(6), 714–719. https://doi.org/10.1136/jnnp.50.6.714
Hendricks, H. T., Zwarts, M. J., Plat, E. F., & Van Limbeek, J. (2002). Systematic review for the early
prediction of motor and functional outcome after stroke by using motor-evoked potentials.
Archives of Physical Medicine and Rehabilitation, 83(9), 1303–1308.
https://doi.org/10.1053/apmr.2002.34284
Hermann, D. M., & Chopp, M. (2014). Therapeutic Promise and Potential Pitfalls of Clinical
Translation. Lancet Neurology, 11(4), 369–380. https://doi.org/10.1016/S1474-
4422(12)70039-X.Promoting
Honey, C. J., & Sporns, O. (2008). Dynamical consequences of lesions in cortical networks. Human
Brain Mapping, 29(7), 802–809. https://doi.org/10.1002/hbm.20579
Hope, T. M. H., Friston, K., Price, C. J., Leff, A. P., Rotshtein, P., & Bowman, H. (2019). Recovery
after stroke: not so proportional after all? Brain, 142, 15–22.
https://doi.org/10.1093/brain/awy302
Hordacre, B., Goldsworthy, M. R., Welsby, E., Graetz, L., Ballinger, S., & Hillier, S. (2020). Resting
State Functional Connectivity Is Associated With Motor Pathway Integrity and Upper-Limb
Behavior in Chronic Stroke. Neurorehabilitation and Neural Repair, 34(6), 547–557.
https://doi.org/10.1177/1545968320921824
Horwitz, B. (2003). The elusive concept of brain connectivity. NeuroImage, 19(2), 466–470.
https://doi.org/10.1016/S1053-8119(03)00112-5
Hsueh, I. P., Lin, J. H., Jeng, J. S., & Hsieh, C. L. (2002). Comparison of the psychometric
characteristics of the functional independence measure, 5 item Barthel index, and 10 item
Barthel index in patients with stroke. Journal of Neurology Neurosurgery and Psychiatry,
73(2), 188–190. https://doi.org/10.1136/jnnp.73.2.188
Jeunet, C., Jahanpour, E., & Lotte, F. (2016). Why standard brain-computer interface (BCI) training
protocols should be changed: An experimental study. Journal of Neural Engineering, 13(3).
https://doi.org/10.1088/1741-2560/13/3/036024

85
Jeurissen, B., Descoteaux, M., Mori, S., & Leemans, A. (2019). Diffusion MRI fiber tractography of
the brain. NMR in Biomedicine, 32(4), 1–22. https://doi.org/10.1002/nbm.3785
Jin, J., Miao, Y., Daly, I., Zuo, C., Hu, D., & Cichocki, A. (2019). Correlation-based channel selection
and regularized feature optimization for MI-based BCI. Neural Networks, 118, 262–270.
https://doi.org/10.1016/J.NEUNET.2019.07.008
Jo, J. Y., Lee, A., Kim, M. S., Park, E., Chang, W. H., Shin, Y. Il, & Kim, Y. H. (2016). Prediction of
motor recovery using quantitative parameters of motor evoked potential in patients with
stroke. Annals of Rehabilitation Medicine, 40(5), 806–815.
https://doi.org/10.5535/arm.2016.40.5.806
Johansson, B. B. (2011). Current trends in stroke rehabilitation. A review with focus on brain
plasticity. Acta Neurologica Scandinavica, 123(3), 147–159. https://doi.org/10.1111/j.1600-
0404.2010.01417.x
Johansson, G. M., & Häger, C. K. (2019). A modified standardized nine hole peg test for valid and
reliable kinematic assessment of dexterity post-stroke. Journal of NeuroEngineering and
Rehabilitation, 16(1), 1–11. https://doi.org/10.1186/s12984-019-0479-y
Jonker, Z. D., Gaiser, C., Tulen, J. H. M., Ribbers, G. M., Frens, M. A., & Selles, R. W. (2021). No
effect of anodal tDCS on motor cortical excitability and no evidence for responders in a large
double-blind placebo-controlled trial. Brain Stimulation, 14(1), 100–109.
https://doi.org/10.1016/j.brs.2020.11.005
Joy, M. T., & Carmichael, S. T. (2021). Encouraging an excitable brain state: mechanisms of brain
repair in stroke. Nature Reviews Neuroscience, 22(1), 38–53.
https://doi.org/10.1038/s41583-020-00396-7
Keihaninejad, S., Zhang, H., Ryan, N. S., Malone, I. B., Modat, M., Cardoso, M. J., Cash, D. M., Fox,
N. C., & Ourselin, S. (2013). An unbiased longitudinal analysis framework for tracking white
matter changes using diffusion tensor imaging with application to Alzheimer’s disease.
NeuroImage, 72, 153–163. https://doi.org/10.1016/J.NEUROIMAGE.2013.01.044
Kevric, J., & Subasi, A. (2017). Comparison of signal decomposition methods in classification of
EEG signals for motor-imagery BCI system. Biomedical Signal Processing and Control, 31,
398–406. https://doi.org/10.1016/J.BSPC.2016.09.007
Khedr, E. M., Omran, E. A. H., Ismail, N. M., El-Hammady, D. H., Goma, S. H., Kotb, H., Galal, H.,
Osman, A. M., Farghaly, H. S. M., Karim, A. A., & Ahmed, G. A. (2017). Effects of transcranial
direct current stimulation on pain, mood and serum endorphin level in the treatment of
fibromyalgia: A double blinded, randomized clinical trial. Brain Stimulation, 10(5), 893–901.
https://doi.org/10.1016/j.brs.2017.06.006
Klomjai, W., Katz, R., & Lackmy-Vallée, A. (2015). Basic principles of transcranial magnetic
stimulation (TMS) and repetitive TMS (rTMS). Annals of Physical and Rehabilitation Medicine,
58(4), 208–213. https://doi.org/10.1016/j.rehab.2015.05.005
Kobayashi, M., & Pascual-Leone, A. (2003). Basic principles of magnetic stimulation. The Lancet,

86
 2, 145–156. https://doi.org/10.1016/S1474-4422(03)00321-1
Koch, P., Schulz, R., & Hummel, F. C. (2016). Structural connectivity analyses in motor recovery
research after stroke. Annals of Clinical and Translational Neurology, 3(3), 233–244.
https://doi.org/10.1002/acn3.278
Kong, K. H., Lee, J., & Chua, K. S. (2012). Occurrence and temporal evolution of upper limb
spasticity in stroke patients admitted to a rehabilitation unit. Archives of Physical Medicine
and Rehabilitation, 93(1), 143–148. https://doi.org/10.1016/j.apmr.2011.06.027
Lanctôt, K. L., Lindsay, M. P., Smith, E. E., Sahlas, D. J., Foley, N., Gubitz, G., Austin, M., Ball, K.,
Bhogal, S., Blake, T., Herrmann, N., Hogan, D., Khan, A., Longman, S., King, A., Leonard, C.,
Shoniker, T., Taylor, T., Teed, M., … Swartz, R. H. (2020). Canadian Stroke Best Practice
Recommendations: Mood, Cognition and Fatigue following Stroke, 6th edition update 2019.
International Journal of Stroke, 15(6), 668–688.
https://doi.org/10.1177/1747493019847334
Lang, C. E., Edwards, D. F., Birkenmeier, R. L., & Dromerick, A. W. (2008). Estimating minimal
clinically important differences of upper extremity measures early after stroke. Arch Phys
Med Rehabil, 89(9), 1693–1700. https://doi.org/10.1016/j.apmr.2008.02.022.Estimating
Lang, C. E., Lohse, K. R., & Birkenmeier, R. L. (2015). Dose and timing in neurorehabilitation:
Prescribing motor therapy after stroke. In Current Opinion in Neurology (Vol. 28, Issue 6, pp.
549–555). https://doi.org/10.1097/WCO.0000000000000256
Lazzaro, V. Di, Ziemann, U., Shepherd, G. M. G., Brecht, M., & Kalmar, J. (2013). The contribution
of transcranial magnetic stimulation in the functional evaluation of microcircuits in human
motor cortex. https://doi.org/10.3389/fncir.2013.00018
Leeb, R, Biasucci, A., Schmidlin, T., Corbet, T., Vuadens, P., & Millán, J. del R. (2016). BCI controlled
neuromuscular electrical stimulation enables sustained motor recovery in chronic stroke
victims. Proceedings of the 6th International Brain-Computer Interface Meeting, Organized
by the BCI Society, 74(1), 108. https://doi.org/10.3217/978-3-85125-467-9-108
Leeb, R, Perdikis, S., Tonin, L., Biasiucci, A., Tavella, M., Creatura, M., Molina, A., Al-Khodairy, A.,
Carlson, T., & Millán, J. d. R. (2013). Transferring brain-computer interfaces beyond the
laboratory: successful application control for motor-disabled users. Artif Intell Med, 59(2),
121–132. https://doi.org/10.1016/j.artmed.2013.08.004
Leeb, Robert, Perdikis, S., Tonin, L., Biasiucci, A., Tavella, M., Creatura, M., Molina, A., Al-Khodairy,
A., Carlson, T., & Millán, J. D. R. (2013). Transferring brain-computer interfaces beyond the
laboratory: Successful application control for motor-disabled users. Artificial Intelligence in
Medicine, 59, 121–132. https://doi.org/10.1016/j.artmed.2013.08.004
Lefaucheur, J. P., André-Obadia, N., Antal, A., Ayache, S. S., Baeken, C., Benninger, D. H., Cantello,
R. M., Cincotta, M., de Carvalho, M., De Ridder, D., Devanne, H., Di Lazzaro, V., Filipović, S.
R., Hummel, F. C., Jääskeläinen, S. K., Kimiskidis, V. K., Koch, G., Langguth, B., Nyffeler, T., …
Garcia-Larrea, L. (2014). Evidence-based guidelines on the therapeutic use of repetitive

87
 transcranial magnetic stimulation (rTMS). Clinical Neurophysiology, 125(11), 2150–2206.
https://doi.org/10.1016/j.clinph.2014.05.021
Lefaucheur, J. P., Antal, A., Ahdab, R., Ciampi de Andrade, D., Fregni, F., Khedr, E. M., Nitsche, M.,
& Paulus, W. (2008). The use of repetitive transcranial magnetic stimulation (rTMS) and
transcranial direct current stimulation (tDCS) to relieve pain. Brain Stimulation, 1(4), 337–
344. https://doi.org/10.1016/j.brs.2008.07.003
Lefebvre, S., Dricot, L., Laloux, P., Desfontaines, P., Evrard, F., Peeters, A., Jamart, J., &
Vandermeeren, Y. (2017). Increased functional connectivity one week after motor learning
and tDCS in stroke patients. Neuroscience, 340, 424–435.
https://doi.org/10.1016/j.neuroscience.2016.10.066
Li, L. M., Uehara, K., & Hanakawa, T. (2015). The contribution of interindividual factors to
variability of response in transcranial direct current stimulation studies. Frontiers in Cellular
Neuroscience, 9. https://doi.org/10.3389/fncel.2015.00181
Lindenberg, R., Renga, V., Zhu, L. L., Betzler, F., Alsop, D., & Schlaug, G. (2010). Structural integrity
of corticospinal motor fibers predicts motor impairment in chronic stroke. Neurology, 74(4),
280–287. https://doi.org/10.1212/WNL.0b013e3181ccc6d9
Lindenberg, R., Renga, V., Zhu, L. L., Nair, D., & Schlaug, G. (2010). Bihemispheric brain stimulation
facilitates motor recovery in chronic stroke patients. Neurology, 75(24), 2176–2184.
https://doi.org/10.1212/WNL.0b013e318202013a
Liu, H., Tian, T., Qin, W., Li, K., & Yu, C. (2016). Contrasting evolutionary patterns of functional
connectivity in sensorimotor and cognitive regions after stroke. Frontiers in Behavioral
Neuroscience, 10(APRIL), 72. https://doi.org/10.3389/FNBEH.2016.00072/BIBTEX
López-Alonso, V., Cheeran, B., Río-Rodríguez, D., & Fernández-Del-Olmo, M. (2014). Inter-
individual variability in response to non-invasive brain stimulation paradigms. Brain
Stimulation, 7(3), 372–380. https://doi.org/10.1016/j.brs.2014.02.004
López-Espuela, F., Roncero-Martín, R., Canal-Macías, M. de la L., Moran, J. M., Vera, V., Gomez-
Luque, A., Lendinez-Mesa, A., Pedrera-Zamorano, J. D., Casado-Naranjo, I., & Lavado-García,
J. (2020). Depressed mood after stroke: Predictive factors at six months follow-up.
International Journal of Environmental Research and Public Health, 17(24), 1–11.
https://doi.org/10.3390/ijerph17249542
Lüscher, C., & Malenka, R. C. (2012). NMDA Receptor-Dependent Long-Term Potentiation and
Long-Term Depression (LTP / LTD). Cold Spring Harbor Perspectives in Biology, 4(6), 1–16.
Maier, M., Ballester, B. R., Duff, A., Duarte Oller, E., & Verschure, P. F. M. J. (2019). Effect of
Specific Over Nonspecific VR-Based Rehabilitation on Poststroke Motor Recovery: A
Systematic Meta-analysis. Neurorehabilitation and Neural Repair, 33(2), 112–129.
https://doi.org/10.1177/1545968318820169
Maier, M., Ballester, B. R., & Verschure, P. F. M. J. (2019). Principles of Neurorehabilitation After
Stroke Based on Motor Learning and Brain Plasticity Mechanisms. Frontiers in Systems

88
 Neuroscience, 13(December), 1–18. https://doi.org/10.3389/fnsys.2019.00074
Mane, R., Chouhan, T., & Guan, C. (2020a). BCI for stroke rehabilitation: Motor and beyond.
Journal of Neural Engineering, 17(4). https://doi.org/10.1088/1741-2552/aba162
Mane, R., Chouhan, T., & Guan, C. (2020b). BCI for stroke rehabilitation: motor and beyond.
Journal of Neural Engineering, 17(4), 041001. https://doi.org/10.1088/1741-2552/ABA162
Markram, H., Gerstner, W., & Sjöström, P. J. (2011). A history of spike-timing-dependent
plasticity. Frontiers in Synaptic Neuroscience, 3(AUG), 1–24.
https://doi.org/10.3389/fnsyn.2011.00004
Markram, H., Gerstner, W., Sjöström, P. J., & Cho, K. (2011). SYNAPTIC NEUROSCIENCE A history
of spike-timing-dependent plasticity. https://doi.org/10.3389/fnsyn.2011.00004
Martin, P. I., Naeser, M. A., Ho, M., Treglia, E., Kaplan, E., Baker, E. H., & Pascual-Leone, A. (2010).
Research with Transcranial Magnetic Stimulation in the Treatment of Aphasia. Curr Neurol
Neurosci Rep., 9(6), 451–458.
Mayka, M. A., Corcos, D. M., Leurgans, S. E., & Vaillancourt, D. E. (2006). Three-dimensional
locations and boundaries of motor and premotor cortices as defined by functional brain
imaging: A meta-analysis. NeuroImage, 31(4), 1453–1474.
https://doi.org/10.1016/j.neuroimage.2006.02.004
McCambridge, A. B., Stinear, J. W., & Byblow, W. D. (2018). Revisiting interhemispheric imbalance
in chronic stroke: A tDCS study. Clinical Neurophysiology, 129(1), 42–50.
https://doi.org/10.1016/j.clinph.2017.10.016
Meng, J., Edelman, B. J., Olsoe, J., Jacobs, G., Zhang, S., Beyko, A., & He, B. (2018). A study of the
effects of electrode number and decoding algorithm on online EEG-based BCI Behavioral
Performance. Frontiers in Neuroscience, 12(APR), 1–14.
https://doi.org/10.3389/fnins.2018.00227
Merboldt, K. D., Hanicke, W., & Frahm, J. (1985). Self-diffusion NMR imaging using stimulated
echoes. Journal of Magnetic Resonance (1969), 64(3), 479–486.
https://doi.org/10.1016/0022-2364(85)90111-8
Minhas, P., Bansal, V., Patel, J., Ho, J. S., Diaz, J., Datta, A., & Bikson, M. (2010). Electrodes for
high-definition transcutaneous DC stimulation for applications in drug delivery and
electrotherapy, including tDCS. Journal of Neuroscience Methods, 190, 188–197.
https://doi.org/10.1016/j.jneumeth.2010.05.007
Monte-Silva, K., Kuo, M.-F., Hessenthaler, S., Fresnoza, S., Liebetanz, D., Paulus, W., & Nitsche, M.
A. (2013). Induction of Late LTP-Like Plasticity in the Human Motor Cortex by Repeated Non-
Invasive Brain Stimulation. Brain Stimulation, 6, 424–432.
https://doi.org/10.1016/j.brs.2012.04.011
Morya, E., Monte-Silva, K., Bikson, M., Esmaeilpour, Z., Biazoli, C. E., Fonseca, A., Bocci, T., Farzan,
F., Chatterjee, R., Hausdorff, J. M., Da Silva Machado, D. G., Brunoni, A. R., Mezger, E.,

89
 Moscaleski, L. A., Pegado, R., Sato, J. R., Caetano, M. S., Sá, K. N., Tanaka, C., … Okano, A. H.
(2019). Beyond the target area: an integrative view of tDCS-induced motor cortex
modulation in patients and athletes. Journal of NeuroEngineering and Rehabilitation, 16(1),
1–29. https://doi.org/10.1186/s12984-019-0581-1
Mottaz, A., Corbet, T., Doganci, N., Magnin, C., Nicolo, P., Schnider, A., & Guggisberg, A. G. (2018).
Modulating functional connectivity after stroke with neurofeedback: Effect on motor deficits
in a controlled cross-over study. NeuroImage: Clinical.
https://doi.org/10.1016/j.nicl.2018.07.029
Murase, N., Duque, J., Mazzocchio, R., & Cohen, L. G. (2004). Influence of Interhemispheric
Interactions on Motor Function in Chronic Stroke. In Ann Neurol (Vol. 55).
https://onlinelibrary.wiley.com/doi/pdf/10.1002/ana.10848
Myrden, A., & Chau, T. (2015). Effects of user mental state on EEG-BCI performance. Frontiers in
Human Neuroscience, 9(JUNE), 308. https://doi.org/10.3389/FNHUM.2015.00308/BIBTEX
Newton, J. M., Ward, N. S., Parker, G. J. M., Deichmann, R., Alexander, D. C., Friston, K. J., &
Frackowiak, R. S. J. (2006). Non-invasive mapping of corticofugal fibres from multiple motor
areas - Relevance to stroke recovery. Brain, 129(7), 1844–1858.
https://doi.org/10.1093/brain/awl106
Nicoll, R. A. (2017). A Brief History of Long-Term Potentiation. Neuron, 93(2), 281–290.
https://doi.org/10.1016/j.neuron.2016.12.015
Nicolo, P., Magnin, C., Pedrazzini, E., Plomp, G., Mottaz, A., Schnider, A., & Guggisberg, A. G.
(2018). Comparison of Neuroplastic Responses to Cathodal Transcranial Direct Current
Stimulation and Continuous Theta Burst Stimulation in Subacute Stroke. Archives of Physical
Medicine and Rehabilitation, 99(5), 862-872.e1.
https://doi.org/10.1016/j.apmr.2017.10.026
Nicolo, P., Ptak, R., & Guggisberg, A. G. (2015). Variability of behavioural responses to transcranial
magnetic stimulation: Origins and predictors. Neuropsychologia, 74, 137–144.
https://doi.org/10.1016/j.neuropsychologia.2015.01.033
Nicolo, P., Rizk, S., Magnin, C., Pietro, M. Di, Schnider, A., & Guggisberg, A. G. (2015a). Coherent
neural oscillations predict future motor and language improvement after stroke. Brain,
138(10), 3048–3060. https://doi.org/10.1093/brain/awv200
Nicolo, P., Rizk, S., Magnin, C., Pietro, M. Di, Schnider, A., & Guggisberg, A. G. (2015b). Coherent
neural oscillations predict future motor and language improvement after stroke. In Brain
(Vol. 138, Issue 10). https://doi.org/10.1093/brain/awv200
Nolte, G., Bai, O., Wheaton, L., Mari, Z., Vorbach, S., & Hallett, M. (2004). Identifying true brain
interaction from EEG data using the imaginary part of coherency. Clinical Neurophysiology,
115(10), 2292–2307. https://doi.org/10.1016/J.CLINPH.2004.04.029
Oczkowski, W. J., & Barreca, S. (1993). The functional independence measure: Its use to identify
rehabilitation needs in stroke survivors. Archives of Physical Medicine and Rehabilitation,

90
 74(12), 1291–1294. https://doi.org/10.1016/0003-9993(93)90081-K
Özarslan, E., Vemuri, B. C., & Mareci, T. H. (2005). Generalized scalar measures for diffusion MRI
using trace, variance, and entropy. Magnetic Resonance in Medicine, 53(4), 866–876.
https://doi.org/10.1002/mrm.20411
Pavlova, E. L., Semenov, R. V., & Guekht, A. B. (2020). Effect of tDCS on Fine Motor Control of
Patients in Subacute and Chronic Post-Stroke Stages. Journal of Motor Behavior, 52(4), 383–
395. https://doi.org/10.1080/00222895.2019.1639608
Pekna, M., Pekny, M., & Nilsson, M. (2012). Modulation of neural plasticity as a basis for stroke
rehabilitation. Stroke, 43(10), 2819–2828. https://doi.org/10.1161/STROKEAHA.112.654228
Pichiorri, F., Morone, G., Petti, M., Toppi, J., Pisotta, I., Molinari, M., Paolucci, S., Inghilleri, M.,
Astolfi, L., Cincotti, F., & Mattia, D. (2015). Brain-computer interface boosts motor imagery
practice during stroke recovery. Annals of Neurology, 77(5), 851–865.
https://doi.org/10.1002/ana.24390
Platz, T., Pinkowski, C., van Wijck, F., Kim, I. H., di Bella, P., & Johnson, G. (2005). Reliability and
validity of arm function assessment with standardized guidelines for the Fugl-Meyer Test,
Action Research Arm Test and Box and Block Test: A multicentre study. Clinical
Rehabilitation, 19(4), 404–411. https://doi.org/10.1191/0269215505cr832oa
Plow, E. B., Cunningham, D., Varnerin, N., & Machado, A. (2015). Rethinking stimulation of brain
in stroke rehabilitation: Why higher-motor areas might be better alternatives for patients
with greater impairments HHS Public Access. Neuroscientist, 21(3), 225–240.
https://doi.org/10.1177/1073858414537381
Pollock, A., Baer, G., Campbell, P., Choo, P. L., Forster, A., Morris, J., Pomeroy, V. M., & Langhorne,
P. (2014). Physical rehabilitation approaches for the recovery of function and mobility
following stroke (Review). Cochrane Library, 399.
https://doi.org/10.1002/14651858.CD001920.pub3.www.cochranelibrary.com
Pollock, Alex, Farmer, S. E., Brady, M. C., Langhorne, P., Mead, G. E., Mehrholz, J., & Van Wijck, F.
(2015). Cochrane overview: Interventions for improving upper limb function after stroke. In
Stroke (Vol. 46, Issue 3, pp. e57–e58). Lippincott Williams and Wilkins.
https://doi.org/10.1161/STROKEAHA.114.008295
Prabhakaran, S., Zarahn, E., Riley, C., Speizer, A., Chong, J. Y., Lazar, R. M., Marshall, R. S., &
Krakauer, J. W. (2008). Inter-individual Variability in the Capacity for Motor Recovery After
Ischemic Stroke. https://doi.org/10.1177/1545968307305302
Puig, J., Blasco, G., Schlaug, G., Stinear, C. M., Daunis-I-Estadella, P., Biarnes, C., Figueras, J.,
Serena, J., Hernández-Pérez, M., Alberich-Bayarri, A., Castellanos, M., Liebeskind, D. S.,
Andrew, &, Demchuk, M., Bijoy, &, Menon, K., Thomalla, G., Nael, K., Wintermark, M., &
Pedraza, S. (2017). Diffusion tensor imaging as a prognostic biomarker for motor recovery
and rehabilitation after stroke. Neuroradiology, 59, 343–351.
https://doi.org/10.1007/s00234-017-1816-0

91
Rehme, A. K., Volz, L. J., Feis, D.-L., Eickhoff, S. B., Fink, G. R., Grefkes, C., & Rehme, A. K. (2015).
Individual prediction of chronic motor outcome in the acute post-stroke stage: Behavioral
parameters versus functional imaging. Hum Brain Mapp, 36(11), 4553–4565.
https://doi.org/10.1002/hbm.22936
Ring, H., Feder, M., Schwartz, J., & Samuels, G. (1997). Functional measures of first-stroke
rehabilitation inpatients: Usefulness of the functional independence measure total score
with a clinical rationale. Archives of Physical Medicine and Rehabilitation, 78(6), 630–635.
https://doi.org/10.1016/S0003-9993(97)90429-9
Rizk, S., Ptak, R., Nyffeler, T., Schnider, A., & Guggisberg, A. G. (2013). Network mechanisms of
responsiveness to continuous theta-burst stimulation. https://doi.org/10.1111/ejn.12334
Roberts, H. C., Denison, H. J., Martin, H. J., Patel, H. P., Syddall, H., Cooper, C., & Sayer, A. A. (2011).
A review of the measurement of grip strength in clinical and epidemiological studies:
Towards a standardised approach. Age and Ageing, 40(4), 423–429.
https://doi.org/10.1093/ageing/afr051
Roche, N., Geiger, M., & Bussel, B. (2015). Mechanisms underlying transcranial direct current
stimulation in rehabilitation. Annals of Physical and Rehabilitation Medicine, 58, 214–219.
https://doi.org/10.1016/j.rehab.2015.04.009
Rossini, P.M., Barker, A. T., Berardelli, A., Caramia, M. D., Caruso, G., Cracco, R. Q., Dimitrijevic,
M. R., Hallett, M., Katayama, Y., Lucking, C. H., Maertens de Noordhout, A. L., Marsden, C.
D., Murray, N. M. F., Rothwell, J. C., Swash, M., & Tomberg, C. (1994). Non-invasive electrical
and magnetic stimulation of the brain, spinal cord, roots and peripheral nerves: Basic
principles and procedures for routine clinical and research application: An updated report
from an I.F.C.N. Committee. Clinical Neurophysiology, 91(2), 79–92.
Rossini, Paolo M., & Rossi, S. (2007). Transcranial magnetic stimulation: Diagnostic, therapeutic,
and research potential. Neurology, 68(7), 484–488.
https://doi.org/10.1212/01.wnl.0000250268.13789.b2
Rossini, Paolo M, Calautti, C., Pauri, F., & Baron, J. (2003). Post-stroke plastic reorganisation. THE
LANCET Neurology, 2, 493–502. http://neurology.thelancet.com
Rossiter, H. E., Boudrias, M.-H., & Ward, N. S. (2014). Do movement-related beta oscillations
change after stroke? Journal of Neurophysiology, 112(9), 2053–2058.
https://doi.org/10.1152/jn.00345.2014
Rosso, C., Valabregue, R., Attal, Y., Vargas, P., Gaudron, M., Baronnet, F., Bertasi, E., Humbert, F.,
Peskine, A., Perlbarg, V., Benali, H., Lehéricy, S., & Samson, Y. (2013). Contribution of
Corticospinal Tract and Functional Connectivity in Hand Motor Impairment after Stroke. PLoS
ONE, 8(9), e73164. https://doi.org/10.1371/journal.pone.0073164
Rothwell, J. C. (2012). Overview of neurophysiology of movement control. Clinical Neurology and
Neurosurgery, 114(5), 432–435. https://doi.org/10.1016/j.clineuro.2011.12.053
Rubens, M. T., & Zanto, T. P. (2012). Parameterization of transcranial magnetic stimulation. J

92
 Neurophysiol, 107, 1257–1259. https://doi.org/10.1152/jn.00716.2011.-A
Sehm, B., Kipping, J., Schaefer, A., Villringer, A., & Ragert, P. (2013). A comparison between uni-
and bilateral tDCS effects on functional connectivity of the human motor cortex. Frontiers in
Human Neuroscience, 7(APR 2013), 1–7. https://doi.org/10.3389/fnhum.2013.00183
Sekihara, K., Nagarajan, S. S., Poeppel, D., & Marantz, A. (2002). Performance of an MEG adaptive-
beamformer technique in the presence of correlated neural activities: Effects on signal
intensity and time-course estimates. IEEE Transactions on Biomedical Engineering, 49(12 I),
1534–1546. https://doi.org/10.1109/TBME.2002.805485
Sheth, S. A., & Jadhav, A. P. (2022). Neurointerventional Advances in 2021. Stroke: Vascular and
Interventional Neurology, 2(1), 1–4. https://doi.org/10.1161/svin.121.000277
Shi, B., Wang, Q., Yin, S., Yue, Z., Huai, Y., & Wang, J. (2021). A binary harmony search algorithm
as channel selection method for motor imagery-based BCI. Neurocomputing, 443, 12–25.
https://doi.org/10.1016/j.neucom.2021.02.051
Siebner, H. R., Lang, N., Rizzo, V., Nitsche, M. A., Paulus, W., Lemon, R. N., & Rothwell, J. C. (2004).
Preconditioning of Low-Frequency Repetitive Transcranial Magnetic Stimulation with
Transcranial Direct Current Stimulation: Evidence for Homeostatic Plasticity in the Human
Motor Cortex. Journal of Neuroscience, 24(13), 3379–3385.
https://doi.org/10.1523/JNEUROSCI.5316-03.2004
Silvanto, J., & Pascual-Leone, A. (2008). State-dependency of transcranial magnetic stimulation.
Brain Topography, 21(1), 1–10. https://doi.org/10.1007/s10548-008-0067-0
Smania, N., Montagnana, B., Faccioli, S., Fiaschi, A., & Aglioti, S. M. (2003). Rehabilitation of
Somatic Sensation and Related Deficit of Motor Control in Patients with Pure Sensory Stroke.
Archives of Physical Medicine and Rehabilitation, 84(11), 1692–1702.
https://doi.org/10.1053/S0003-9993(03)00277-6
Smith, S. M., Jenkinson, M., Johansen-Berg, H., Rueckert, D., Nichols, T. E., Mackay, C. E., Watkins,
K. E., Ciccarelli, O., Cader, M. Z., Matthews, P. M., & Behrens, T. E. J. (2006). Tract-based
spatial statistics: Voxelwise analysis of multi-subject diffusion data. NeuroImage, 31(4),
1487–1505. https://doi.org/10.1016/J.NEUROIMAGE.2006.02.024
Souza, A., Martins, D. F., Medeiros, L. F., Nucci-Martins, C., Martins, T. C., Siteneski, A., Caumo,
W., dos Santos, A. R. S., & Torres, I. L. S. (2018). Neurobiological mechanisms of antiallodynic
effect of transcranial direct current stimulation (tDCS) in a mice model of neuropathic pain.
Brain Research, 1682, 14–23. https://doi.org/10.1016/j.brainres.2017.12.005
Steinstrater, O., Sillekens, S., Junghoefer, M., Burger, M., & Wolters, C. H. (2010). Sensitivity of
beamformer source analysis to deficiencies in forward modeling. Hum Brain Mapp, 31(12),
1907–1927.
Stenroos, M., Mantynen, V., & Nenonen, J. (2007). A Matlab library for solving quasi-static volume
conduction problems using the boundary element method. Comput. Methods Programs
Biomed., 88(3), 256–263. https://doi.org/10.1016/j.cmpb.2007.09.004

93
Stinear, C. M., Lang, C. E., Zeiler, S., & Byblow, W. D. (2020). Advances and challenges in stroke
rehabilitation. The Lancet Neurology, 19(4), 348–360. https://doi.org/10.1016/S1474-
4422(19)30415-6
Stinear, C. M., Petoe, M. A., & Byblow, W. D. (2015). Primary motor cortex excitability during
recovery after stroke: Implications for neuromodulation. Brain Stimulation, 8(6), 1183–1190.
https://doi.org/10.1016/j.brs.2015.06.015
Stock, R., Thrane, G., Askim, T., Anke, A., & Mork, P. J. (2019). Development of grip strength during
the first year after stroke. Journal of Rehabilitation Medicine, 51(4), 248–256.
https://doi.org/10.2340/16501977-2530
Suda, M., Kawakami, M., Okuyama, K., Ishii, R., Oshima, O., Hijikata, N., Nakamura, T., Oka, A.,
Kondo, K., & Liu, M. (2021). Validity and Reliability of the Semmes-Weinstein Monofilament
Test and the Thumb Localizing Test in Patients With Stroke. Frontiers in Neurology,
11(January), 1–10. https://doi.org/10.3389/fneur.2020.625917
Sur, M., & Rubenstein, J. L. R. (2005). Patterning and Plasticity of the Cerebral Cortex. Science,
310(5749), 805–810. https://doi.org/10.1126/science.1112070
Takeuchi, N., & Izumi, S.-I. (2013). Rehabilitation with Poststroke Motor Recovery: A Review with
a Focus on Neural Plasticity. Stroke Research and Treatment, 2013, 13.
https://doi.org/10.1155/2013/128641
Takeuchi, N., & Izumi, S. I. (2015). Combinations of stroke neurorehabilitation to facilitate motor
recovery: Perspectives on Hebbian plasticity and homeostatic metaplasticity. Frontiers in
Human Neuroscience, 9(JUNE). https://doi.org/10.3389/fnhum.2015.00349
Tang, Q., Li, G., Liu, T., Wang, A., Feng, S., Liao, X., Jin, Y., Guo, Z., He, B., McClure, M. A., Xing, G.,
& Mu, Q. (2015). Modulation of interhemispheric activation balance in motor-related areas
of stroke patients with motor recovery: Systematic review and meta-analysis of fMRI studies.
Neuroscience & Biobehavioral Reviews, 57, 392–400.
https://doi.org/10.1016/J.NEUBIOREV.2015.09.003
Tariq, M., Trivailo, P. M., & Simic Id, M. (2020). Mu-Beta event-related (de)synchronization and
EEG classification of left-right foot dorsiflexion kinaesthetic motor imagery for BCI.
https://doi.org/10.1371/journal.pone.0230184
Taylor, D. G., & Bushell, M. C. (1985). The spatial mapping of translational diffusion coefficients
by the NMR imaging technique. Physics in Medicine and Biology, 30(4), 345–349.
https://doi.org/10.1088/0031-9155/30/4/009
Thompson, M. C. (2018). Critiquing the Concept of BCI Illiteracy. Science and Engineering Ethics
2018 25:4, 25(4), 1217–1233. https://doi.org/10.1007/S11948-018-0061-1
Trompetto, C., Assini, A., Buccolieri, A., Marchese, R., & Abbruzzese, G. (2000). Motor recovery
following stroke: A transcranial magnetic stimulation study. Clinical Neurophysiology,
111(10), 1860–1867. https://doi.org/10.1016/S1388-2457(00)00419-3

94
Valenzuela, P. L., Amo, C., Sánchez-Martínez, G., Torrontegi, E., Vázquez-Carrión, J., Montalvo, Z.,
Lucia, A., & de la Villa, P. (2019). Enhancement of Mood but not Performance in Elite Athletes
With Transcranial Direct-Current Stimulation. International Journal of Sports Physiology and
Performance, 14(3), 310–316. https://doi.org/10.1123/IJSPP.2018-0473
Valsasina, P., Rocca, M. A., Agosta, F., Benedetti, B., Horsfield, M. A., Gallo, A., Rovaris, M., Comi,
G., & Filippi, M. (2005). Mean diffusivity and fractional anisotropy histogram analysis of the
cervical cord in MS patients. NeuroImage, 26(3), 822–828.
https://doi.org/10.1016/j.neuroimage.2005.02.033
Van der Lee, J. H., De Groot, V., Beckerman, H., Wagenaar, R. C., Lankhorst, G. J., & Bouter, L. M.
(2001). The intra- and interrater reliability of the action research arm test: A practical test of
upper extremity function in patients with stroke. Archives of Physical Medicine and
Rehabilitation, 82(1), 14–19. https://doi.org/10.1053/apmr.2001.18668
van der Lee, J. H., Roorda, L. D., Beckerman, H., Lankhorst, G. J., & Bouter, L. M. (2002). Improving
the action research arm test: A unidimensional hierarchical scale. Clinical Rehabilitation,
16(6), 646–653. https://doi.org/10.1191/0269215502cr534oa
Van Veen, B. D., Van Drongelen, W., Yuchtman, M., & Suzuki, A. (1997). Localization of Brain
Electrical Activity via Linearly Constrained Minimum Variance Spatial Filtering. In IEEE
TRANSACTIONS ON BIOMEDICAL ENGINEERING (Vol. 44, Issue 9).
Veerbeek, J. M., van Wegen, E. E. H., van Peppen, R. P. S., Hendriks, H. J. M., & M.B., R. (2014).
KNGF Clinical Practice Guideline for Physical Therapy in patients with stroke.
Veerbeek, J. M., Van Wegen, E., Van Peppen, R., Van Der Wees, P. J., Hendriks, E., Rietberg, M., &
Kwakkel, G. (2014). What is the evidence for physical therapy poststroke? A systematic
review and meta-analysis. PLoS ONE, 9(2). https://doi.org/10.1371/journal.pone.0087987
Vinehout, K., Schmit, B. D., & Schindler-Ivens, S. (2019). Lower Limb Task-Based Functional
Connectivity Is Altered in Stroke. Brain Connectivity, 9(4), 365–377.
https://doi.org/10.1089/brain.2018.0640
Volz, L. J., Vollmer, M., Michely, J., Fink, G. R., Rothwell, J. C., & Grefkes, C. (2017). Time-
dependent functional role of the contralesional motor cortex after stroke. NeuroImage:
Clinical, 16, 165–174. https://doi.org/10.1016/J.NICL.2017.07.024
Wang, Z., Zhou, Y., Chen, L., Gu, B., Yi, W., Liu, S., Xu, M., Qi, H., He, F., & Ming, D. (2019). BCI
Monitor Enhances Electroencephalographic and Cerebral Hemodynamic Activations during
Motor Training. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 27(4),
780–787. https://doi.org/10.1109/TNSRE.2019.2903685
Ward, A. B. (2002). A literature review of the pathophysiology and onset of post-stroke spasticity.
https://doi.org/10.1111/j.1468-1331.2011.03448.x
Ward, N. S. (2017). Restoring brain function after stroke — bridging the gap between animals and
humans. Nature Reviews Neurology, 13(4), 244–255.
https://doi.org/10.1038/nrneurol.2017.34

95
Watkins, C. L., Leathley, M. J., Gregson, J. M., Moore, A. P., Smith, T. L., & Sharma, A. K. (2002).
Prevalence of spasticity post stroke. Clinical Rehabilitation, 16(5), 515–522.
https://doi.org/10.1191/0269215502cr512oa
Welmer, A. K., Widén Holmqvist, L., & Sommerfeld, D. K. (2010). Location and severity of spasticity
in the first 1-2 weeks and at 3 and 18 months after stroke. European Journal of Neurology,
17(5), 720–725. https://doi.org/10.1111/j.1468-1331.2009.02915.x
Wessel, M. J., & Hummel, F. C. (n.d.). Non-invasive Cerebellar Stimulation: a Promising Approach
for Stroke Recovery? https://doi.org/10.1007/s12311-017-0906-1
Wessel, M. J., & Hummel, F. C. (2018). Non-invasive Cerebellar Stimulation: a Promising Approach
for Stroke Recovery? The Cerebellum 2018 17:3, 17(3), 359–371.
https://doi.org/10.1007/S12311-017-0906-1
Wessel, M. J., Zimerman, M., & Hummel, F. C. (2015). Non-Invasive Brain Stimulation: An
Interventional Tool for Enhancing Behavioral Training after Stroke. Frontiers in Human
Neuroscience, 9(June 2016). https://doi.org/10.3389/fnhum.2015.00265
Westlake, K. P., Hinkley, L. B., Bucci, M., Guggisberg, A. G., Findlay, A. M., Henry, R. G., Nagarajan,
S. S., & Byl, N. (2012). Resting state alpha-band functional connectivity and recovery after
stroke. Experimental Neurology, 237(1), 160–169.
https://doi.org/10.1016/j.expneurol.2012.06.020
White, G. N., Cordato, D. J., & Mbbs, F. O. R. (2012). Validation of the Stroke Rehabilitation
Motivation Scale : a pilot study. 7(2).
WHO. (2014). Global status report on noncommunicable diseases 2014. In World Health.
https://doi.org/ISBN 9789241564854
Widmer, M., Held, J. P. O., Wittmann, F., Valladares, B., Lambercy, O., Sturzenegger, C., Palla, A.,
Lutz, K., & Luft, A. R. (2022). Reward During Arm Training Improves Impairment and Activity
After Stroke: A Randomized Controlled Trial. Neurorehabilitation and Neural Repair, 36(2),
140–150. https://doi.org/10.1177/15459683211062898
Winstein, C. J., Wolf, S. L., Dromerick, A. W., Lane, C. J., Nelsen, M. A., Lewthwaite, R., Cen, S. Y.,
Azen, S. P., & Interdisciplinary Comprehensive Arm Rehabilitation Evaluation (ICARE)
Investigative Team. (2016). Effect of a Task-Oriented Rehabilitation Program on Upper
Extremity Recovery Following Motor Stroke: The ICARE Randomized Clinical Trial. JAMA,
315(6), 571–581. https://doi.org/10.1001/jama.2016.0276
Winters, C., Van Wegen, E. E. H., Daffertshofer, A., & Kwakkel, G. (2015). Generalizability of the
Proportional Recovery Model for the Upper Extremity After an Ischemic Stroke.
Neurorehabilitation and Neural Repair. https://doi.org/10.1177/1545968314562115
Yamada, Y., & Sumiyoshi, T. (2021). Neurobiological Mechanisms of Transcranial Direct Current
Stimulation for Psychiatric Disorders; Neurophysiological, Chemical, and Anatomical
Considerations. Frontiers in Human Neuroscience, 15(February), 1–10.
https://doi.org/10.3389/fnhum.2021.631838

96
Yozbatiran, N., Der-Yeghiaian, L., & Cramer, S. C. (2008). A standardized approach to performing
the action research arm test. https://doi.org/10.1177/1545968307305353
Yu, C., Zhu, C., Zhang, Y., Chen, H., Qin, W., Wang, M., & Li, K. (2009). A longitudinal diffusion
tensor imaging study on Wallerian degeneration of corticospinal tract after motor pathway
stroke. NeuroImage, 47(2), 451–458. https://doi.org/10.1016/j.neuroimage.2009.04.066
Yuan, K., Wang, X., Chen, C., Lau, C. C. Y., Chu, W. C. W., & Tong, R. K. Y. (2020). Interhemispheric
functional reorganization and its structural base after BCI-guided upper-limb training in
chronic stroke. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 28(11),
2525–2536. https://doi.org/10.1109/TNSRE.2020.3027955
Zandvliet, S. B., Kwakkel, G., Nijland, R. H. M., van Wegen, E. E. H., & Meskers, C. G. M. (2020). Is
Recovery of Somatosensory Impairment Conditional for Upper-Limb Motor Recovery Early
After Stroke? Neurorehabilitation and Neural Repair, 34(5), 403–416.
https://doi.org/10.1177/1545968320907075
Zhang, J. M., Meng, L. M., PhD, Q. W., Liu, N. M., Shi, F.-D. M., & Yu, C. M. (2014). Structural
Damage and Functional Reorganization in Ipsilesional M1 in Well-Recovered Patients With
Subcortical Stroke. Stroke, 45(3), 788–793.
https://doi.org/10.1161/STROKEAHA.113.003425
Zhang, K., Chen, X., Liu, F., Tang, H., Wang, J., & Wen, W. (2018). System framework of robotics in
upper limb rehabilitation on poststroke motor recovery. Behavioural Neurology,
2018(December 2017). https://doi.org/10.1155/2018/6737056
Zheng, X., & Schlaug, G. (2015). Structural white matter changes in descending motor tracts
correlate with improvements in motor impairment after undergoing a treatment course of
tDCS and physical therapy. Frontiers in Human Neuroscience, 9, 229.
https://doi.org/10.3389/fnhum.2015.00229

97
 5. Appendix

Study 3
Nguyen-Danse, D.A., Singaravelu, S., Chauvigné, L.A.S. et al. Feasibility of Reconstructing Source
Functional Connectivity with Low-Density EEG. Brain Topogr 34, 709–719 (2021).
https://doi.org/10.1007/s10548-021-00866-w

98
Brain Topography (2021) 34:709–719
https://doi.org/10.1007/s10548-021-00866-w

ORIGINAL PAPER

Feasibility of Reconstructing Source Functional Connectivity

with Low‑Density EEG
Dung A. Nguyen‑Danse1 · Shobana Singaravelu1 · Léa A. S. Chauvigné1 · Anaïs Mottaz1 · Leslie Allaman1 ·
Adrian G. Guggisberg1 

Received: 11 March 2021 / Accepted: 2 August 2021 / Published online: 20 August 2021
© The Author(s) 2021

Abstract
Objectives  Functional connectivity (FC) is increasingly used as target for neuromodulation and enhancement of perfor-
mance. A reliable assessment of FC with electroencephalography (EEG) currently requires a laboratory environment with
high-density montages and a long preparation time. This study investigated the feasibility of reconstructing source FC with
a low-density EEG montage towards a usage in real life applications.
Methods  Source FC was reconstructed with inverse solutions and quantified as node degree of absolute imaginary coher-
ence in alpha frequencies. We used simulated coherent point sources as well as two real datasets to investigate the impact of
electrode density (19 vs. 128 electrodes) and usage of template vs. individual MRI-based head models on localization accu-
racy. In addition, we checked whether low-density EEG is able to capture inter-individual variations in coherence strength.
Results  In numerical simulations as well as real data, a reduction of the number of electrodes led to less reliable reconstruc-
tions of coherent sources and of coupling strength. Yet, when comparing different approaches to reconstructing FC from
19 electrodes, source FC obtained with beamformers outperformed sensor FC, FC computed after independent component
analysis, and source FC obtained with sLORETA. In particular, only source FC based on beamformers was able to capture
neural correlates of motor behavior.
Conclusion  Reconstructions of FC from low-density EEG is challenging, but may be feasible when using source reconstruc-
tions with beamformers.

Keywords  Functional connectivity · Low–density · Electroencephalography · Alpha oscillations · Neurofeedback

Introduction et al. 2001; Nolte et al. 2004). FC magnitude computed

Interregional neural communication is thought to be accom-
panied by a synchronization of oscillations between differ-
ent brain regions (Aertsen et al. 1989; Varela et al. 2001).
This interregional synchronization can be quantified with
the concept of “functional connectivity” (FC) which is a
measure of statistical dependency between activity in dif-
ferent brain regions and is therefore considered to be an
index of functional interaction (Nunez et al. 1997; Varela
Handling Editor: Christoph M Michel .
* Adrian G. Guggisberg
aguggis@gmail.com
1
Imaging‑Assisted Neurorehabilitation Lab, Department
of Clinical Neurophysiology, University of Geneva, Av. de
Beau‑Séjour 26, 1211 Geneva, Switzerland
from fMRI was found to correlate with performance, e.g.,
with reading competency, executive function, and episodic
memory capacity (Wang et al. 2010; Koyama et al. 2011;
Reineberg et al. 2015). fMRI-FC was also associated with
motor performance in a sit-to-stand-to-sit task in low back
pain patients (Pijnenburg et al. 2015). In electroencepha-
lography (EEG), neural assemblies spontaneously produce
prominent oscillations at a frequency of about 8 to 12 cycles
per second even when the recorded subjects are at rest (Hin-
driks et al. 2014), which is also the preferred frequency for
interregional communication (Chapeton et al. 2019). Alpha-
band FC between a brain area and the rest of the brain (i.e.,
the node degree of a brain area) correlates with behavio-
ral performance in healthy participants (Guggisberg et al.
2015), and with neurological deficits in patients (Dubovik
et al. 2012). For instance, the more spontaneous alpha activ-
ity in Broca’s area is coherent with the rest of the brain, the

13
Vol.:(0123456789)

710
better subjects are able to produce words. Recent findings
have demonstrated that spontaneous FC in the alpha band
enables particularly high performance and is thus a better
predictor of performance than classical task-induced activa-
tions (Allaman et al. 2020).
Thus, FC is increasingly used as marker to probe for novel
disease biomarkers (Fox and Greicius 2010) or predictors of
outcome (Westlake et al. 2012; Nicolo et al. 2015). Further-
more, it has become a target for new treatment approaches
aiming at enhancing performance or reducing neurological
deficits. Patients were able to improve their motor function
when asked to enhance the FC in their motor cortex; which
was not the case when a non-related brain area was trained
(Mottaz et al. 2018). However, the setup of a neurofeed-
back as performed in a laboratory is time-consuming and
tiring for the patients. Not only it requires individual head
models based on magnetic resonance imaging (MRI), but
also high-density electroencephalography (hd-EEG). Such
installation is hardly feasible in routine clinical practice or
at a larger scale.
For end-users and patients to use EEG in clinical settings,
the montage must be user-friendly and easily set up. One
solution is to reduce the number of channels, which reduces
the setup time and makes the entire EEG installation more
mobile. Motor-disabled patients have successfully used low-
density EEG for brain-computer interface (BCI) tasks (Leeb
et al. 2013). In clinical practice, low-density EEG is already
part of standard practice. For diagnosis purposes, 9 to 21
electrodes are used in epileptology (Koutroumanidis et al.
2017; Rossetti et al. 2012; Wilmshurst et al. 2015).
However, reducing the number of electrodes is more
delicate when the aim is to reconstruct FC. The electro-
magnetic potential spreads throughout the brain because of
volume conduction and is thus received to some degree by
all sensors. This leads to a mix of signals from multiple
brain areas at each electrode, which makes the targeting of
a given brain lobe of interest difficult. Furthermore, volume
conduction induces a massive overestimation and distortion
of FC (Srinivasan 2007; Schoffelen and Gross 2009). The
computation of inverse solutions allows to partially revert
volume conduction and obtain the signal at the source. This
gives access to signals at regions of interest and reduces
volume conduction issues in the computation of FC. Yet,
the usage of inverse solutions usually requires a good spatial
sampling and thus hd-EEG recordings (Michel et al. 2004).
This study aimed to evaluate the feasibility of using
a low-density EEG (with 19 channels) for reconstructing
FC markers of performance and disability in the exam-
ple of motor performance. Reducing the number of chan-
nels would reduce the time spent on the installation, de-
installation, maintenance, and care of the participant and
the whole equipment. If feasible, we could then make
FC markers and treatment targets accessible to clinical
13
Brain Topography (2021) 34:709–719
practice. To this end, we will need to satisfy two main
requirements. First, we need to be able to capture vari-
ance in FC magnitude, such that periods or subjects having
lower neural coupling show proportionally lower values of
reconstructed FC. Second, we need to have some fidelity in
localization accuracy such that neural coupling at a target
brain area is correctly captured.
We compared different technical approaches to the prob-
lem. Mixed signals captured by average and Laplacian re-
referenced (McFarland et al. 1997) electrodes were used
to compute sensor FC. To obtain unmixed signals we used
inverse solutions (Pascual-Marqui 2002; Sekihara et  al.
2004) and independent component analyses (ICA), which
allowed computing FC between sources or independent
components, respectively. Each of these approaches was
then evaluated with regards to localization error and with
regards to the correlation of reconstructed FC magnitude
with behavior. This was first done with numerical simula-
tions and then tested in different real datasets.
Materials and Methods
Datasets
The study comprised 3 datasets.
Dataset 1: 70 healthy subjects (45 women,
37.9 ± 17.6  years old) underwent resting-state EEG
recording with a 128 channel Biosemi ActiveTwo EEG-
system (Biosemi B.V., Amsterdam, Netherlands).
Dataset 2: 20 healthy subjects (13 women;
28.7 ± 5.6 years old) underwent resting-state EEG record-
ing with a 128 channel Biosemi ActiveTwo EEG-system
and behavioral assessment of motor performance with a
finger-tapping task. All had a normal or corrected-to-nor-
mal vision and no history of neurological or psychiatric
disorders and were paid for their participation.
Dataset 3: 20 healthy subjects (17 women,
25.5 ± 5.4 years old) underwent resting-state EEG record-
ing with a 20 channel Enobio system with dry-gel elec-
trodes (neuroelectrics, Barcelona) and behavioral assess-
ment of motor performance with a finger-tapping task.
All had a normal or corrected-to-normal vision and no
history of neurological or psychiatric disorders and were
paid for their participation.
All procedures were approved by the ethical committee of
the canton of Geneva and performed according to the dec-
laration of Helsinki. All participants gave written informed
consent after receiving an explanation on the nature of the
experiments.
Brain Topography (2021) 34:709–719 711
Behavioral Assessments
Participants of dataset 2 and 3 performed a sequential finger-
tapping task (FTT) (Zhang et al. 2012) immediately after
the resting-state EEG recording. The task was designed
using E-Prime 2.0 software (Psychology Software Tools,
Pittsburgh, PA). Participants were instructed to repeat a
given five-item sequence with their left hand (little finger to
index) on four horizontally arranged buttons numbered left
to right on a Chronos box (Psychology Software Tools, Pitts-
burgh, PA; https://​pstnet.​com/​produ​cts/​chron​os/). The same
sequence was used throughout the whole experiment (1-4-
2-3-1). It was continuously presented to participants while
they had to perform it. They had to repeat the sequence as
fast and accurately as possible during two blocks of 30 s. No
feedback was given. Motor performance was quantified as
the average number of correct motor sequences per minute.
Recordings
EEG was recorded with a 128-channel Biosemi ActiveTwo
EEG-system using active gel electrodes (Biosemi B.V.,
Amsterdam, Netherlands) at a sampling rate of 512 Hz in
datasets 1 and 2, or with a 20-channel Enobio system using
dry-gel electrodes (neuroelectrics, Barcelona) at 500 Hz
in dataset 4 in an awake, resting condition during which
subjects kept their eyes closed. Artifacts and data segments
with signs of drowsiness were excluded by visual inspec-
tion of the data. Electrodes containing artifacts persistent
across multiple epochs were excluded. These electrodes
were interpolated from neighboring electrodes using a 3D
Fig. 1  Schematic overview of analysis setups. High-density montages
with 128 electrodes were compared to low-density subsets or sys-
tems with 19 electrodes, according to the international 10–20 stand-
ards. Unmixing of electrode signals was performed with either source
localization or independent component analysis (ICA) and compared
spline interpolation (< 5% interpolated electrodes) (Perrin
et al. 1987) for analyses of sensor FC and for source recon-
structions with standardized low-resolution electromagnetic
tomography (sLORETA), but not for source reconstruction
with beamformers.
The 19 electrodes followed the positions of the interna-
tional 10–20 system. In datasets 1 and 2, they were selected
from the full 128 montage. For comparison, sensor FC was
also computed from sensor data with an average reference
and a small Laplacian reference of each electrode.
The EEG was segmented into 300 non-overlapping,
artifact-free epochs of 1 s duration and bandpass-filtered
between 1 and 20 Hz.
Figure  1 gives a schematic overview of the different
analysis steps that were taken in order to reconstruct FC in
real data from low-density recordings, as compared to the
high-density montages as gold standard.
Source Localization
For individual MRI-based head models, the MRI proto-
col contained a high-resolution T1-weighted, 3-D spoiled
gradient-recalled echo in a steady state sequence cover-
ing the whole skull (192 coronal slices, 1.1 mm thickness,
TR = 2500 ms, TE = 3 ms, flip angle = 8°).
Each subject’s brain was segmented into scalp, skull, grey
and white matter with NUTMEG (http://​nutmeg.​berke​ley.​
edu) (Dalal et al. 2011) and the toolbox MARS (https://w
​ ww.​
parra​lab.​org/​mars/) (Huang and Parra 2015).
We computed the lead-potential with 10 mm grid spac-
ing (~ 1200 solution points) using a boundary element head
to the performance obtained without unmixing. MVBF minimum
variance beamformer, sLORETA standardized low-resolution elec-
tromagnetic tomography, ICA independent component analysis, FC
functional connectivity
13

712
model (BEM). The BEM was created with the Helsinki BEM
library (http://​peili.​hut.​fi/​BEM/) (Stenroos et al. 2007) and
the NUTEEG plugin of NUTMEG, based on the individual
T1 MRI of each subject as well as based on the Montreal
Neurological Institute template brain.
Source FC was calculated in Matlab (The MathWorks
Inc., Natick, USA) with NUTMEG (http://​nutmeg.​berke​ley.​
edu) (Dalal et al. 2011) and its functional connectivity map-
ping (FCM) toolbox (Guggisberg et al. 2011).
Most previous studies reporting correlations between
alpha-band FC and behavior, and previous trials using neu-
rofeedback to train alpha-band FC were based on source
reconstructions of hd-EEG arrays with beamformers (Dubo-
vik et al. 2012; Guggisberg et al. 2015; Mottaz et al. 2015,
2018). Thus, we used the same approach here as a reference
for comparison with other approaches. A scalar minimum
variance beamformer (MVBF) was used to project preproc-
essed hd-EEG data to source space (Sekihara et al. 2004).
The MVBF uses the temporal covariance of the EEG
data (in addition to the sensor geometry) to create a custom
spatial filter depending on the signal characteristics. This
enables more precise and focal source localization (Seki-
hara et al. 2005). However, beamformers are sensitive to the
accuracy of the head model; measured data that is incon-
sistent with the head model is liable to be rejected as noise
(Steinsträter et al. 2010). We thus compared beamformers
to standardized low-resolution electromagnetic tomography
(sLORETA) (Pascual-Marqui 2002) as a widely used inverse
solution that does not have these limitations of beamformers.
On the downside, it enables less focal reconstructions (Seki-
hara et al. 2005) and thus is likely to induce more spatial
leakage of the reconstructed sources.
Dipole Orientation
For the localization of FC in the brain, we require an estima-
tion of neural network oscillations at each solution point.
Vector weights obtained from the inverse solution allow only
reconstructing squared power values; the reconstruction of
neural oscillations requires a scalar weight matrix. In order
to scalarize the lead-potential as input to scalar weights com-
putation, we need to determine the dipole orientation at each
grid location. We computed the optimal dipole orientation
at each solution point as the orientation yielding maximum
output signal-to-noise ratio (SNR). For the MVBF, the opti-
mum orientation at each solution point v is given by (Seki-
hara et al. 2004):
{[ ]−1 [ T −2 ]}
𝜼v = vmin LTv R−1 Lv Lv R Lv (1)
where vmin is the eigenvector corresponding to the minimum
eigenvalue of the matrix in {}, L is the vector lead-potential,
13
Brain Topography (2021) 34:709–719
R the sensor covariance matrix, and superscript T denotes
the matrix transpose.
For sLORETA, it is obtained as (Pascual-Marqui et al.
2009):
𝜼v = vmax LTv G−1 RG−1 Lv , LTv G−1 Lv (2)
{ }
where vmax is the eigenvector corresponding to the maximum
generalized eigenvalue, and G the gram matrix defined as
LLT.
The scalar lead-potential was then calculated as:
lv,𝜂 = lv 𝜼v (3)
Independent Component Analysis (ICA)
EEG electrodes carry information from multiple brain
sources that are mixed together. One approach to unmix
these signals is independent component analysis (Hyvarinen
1999; Reineberg et al. 2015). Average referenced sensor time
series with 19 channels of all subjects were normalized to
common amplitude and variance using z-scores, concat-
enated in the time dimension, and subjected to a FastICA
algorithm (Hyvarinen 1999) with the FastICA package for
MATLAB (https://​resea​rch.​ics.​aalto.​fi/​ica/​fasti​ca/). Default
parameters were used. This led to 19 independent compo-
nents. The unmixing matrix obtained from all subjects was
then applied to the sensor data of each subject.
Functional Connectivity (FC)
We used the absolute imaginary component of coherence
(IC) as an index of FC (Nolte et al. 2004; Sekihara et al.
2011). One FC value was obtained from all 300 epochs of
data. For sensor FC, it was computed directly between the
preprocessed and filtered EEG sensor data. For ICA data, IC
was computed between independent components. For source
FC, we used the source time series estimated with inverse
solutions. From this, we calculated the weighted node degree
(WND) for each voxel, component, or electrode as the mean
of its coherence with all other voxels/components/electrodes
(Newman 2004). The WND can been as index of the overall
importance of a brain area in the network.
FC values can be influenced by the signal-to-noise ratio
of the EEG. To minimize this potential confound, we nor-
malized WND values at each voxel by subtracting the mean
WND across all voxels and then dividing by the standard
deviation of all voxels, thus obtaining z-scores (Dubovik
et al. 2012; Mottaz et al. 2015).
In case of individual head models, normalized WND val-
ues were spatially normalized to canonical Montreal Neu-
rological Institute space with functions from the Matlab
Brain Topography (2021) 34:709–719 713
toolbox SPM8 (https://​www.​fil.​ion.​ucl.​ac.​uk/​spm/​softw​are/​
spm8/).
Regions, Independent Components, and Electrodes
of Interest
For correlations between source FC and visuo-motor behav-
ior, we defined the right (i.e. contralateral to the moved
hand) Precentral gyrus as region of interest (ROI) using
the automated anatomical labeling (AAL) atlas (Tzourio-
Mazoyer et al. 2002). ROI values were obtained as the aver-
age WND across its containing voxels. For sensor FC, elec-
trode C4 was analyzed. For ICA processed data, we defined
an independent component of interest. For this, we corre-
lated source time series obtained with the full EEG array,
individual head models, and MVBF with the time series of
each independent component. The independent component
whose time series correlated positively with source time
series at the right sensorimotor cortex was used for further
analysis.
Statistical Analyses
To investigate the fidelity of source reconstructions with
template head models and with low-density EEG arrays,
we correlated normalized WND values obtained with indi-
vidual MRI-based head models and 128 channel data (gold
standard) to normalized WND values obtained with template
head models using 128 or 19 electrodes. Pearson correlation
coefficients were computed over all 82 cortical ROIs of the
AAL atlas for each of the 70 subjects of dataset 1, i.e., the
normalized WND values of the gold standard at all ROIs
were correlated with the values of the test setup at all ROIs.
Fisher-Z transformed correlation coefficients of all subjects
were then fed to a one-way ANOVA with the analysis setup
as a dependent factor. Pairwise post-hoc comparisons were
done with the Tukey–Kramer HSD correction.
For comparison between sensor and source FC, we
matched each of the 19 electrodes of the international 10–20
system to the ROI that had the shortest Euclidean distance
from the electrode’s position. Source normalized WND
obtained from 19 electrodes at these ROIs as well as sen-
sor normalized WND obtained from 19 average-referenced
or Laplacian-referenced electrodes was then correlated to
the normalized WND values obtained with individual MRI-
based head models and 128 channel data.
We also correlated source FC at the right precentral gyrus
obtained with template head models and 19 electrodes to the
source FC obtained with the gold standard, since this was
our ROI for motor behavior. This time, the correlation was
done over subjects.
Next, we investigated the ability of low-density EEG
arrays to capture associations with behavioral performance.
Pearson correlation coefficients were computed for associa-
tions between alpha-band FC and FTT performance in sub-
jects of datasets 2 and 3. This was done for the precentral
ROI, C4 electrode, and the independent component of inter-
est. The correlation obtained with the gold standard (indi-
vidual MRI-based head models, 128 channel data, MVBF)
was juxtaposed to the correlation coefficients obtained with
19 channels using either source localization, sensor FC, or
FC between independent components.
To check the feasibility of obtaining FC using even
more convenient dry-gel electrodes, we then used data
from patients in dataset 3 to correlate source FC with FTT
performance.
Numerical Simulations
For 38 randomly selected subjects of dataset 1, we simulated
3 cortical point sources with a 10 Hz sinusoidal rhythm. The
main source of interest was placed to the center of the right
primary motor cortex. Two additional sources with 10 Hz
oscillation were defined at the left primary motor cortex and
the right premotor area. They had a radial phase lag of π/2
(= 25 ms) or − π/2, respectively, relative to the first point
source. This phase difference leads to maximal values in
the imaginary part of coherence, while the lag of π between
sources 2 and 3 produce 0 imaginary coherence. The dipole
orientations were fixed to point in a random orientation at
each location. In addition, to test whether our settings are
able to capture variance in coupling strength across subjects,
we additionally simulated variance in coupling strength
between point source 1 and point sources 2/3. This was
achieved by simulating alpha activity in only 60% of the 300
epochs. The number of epochs with alpha activity at both
source 1 and sources 2/3 at the same time (i.e., with coher-
ent alpha activity) was then varied across subjects between
a minimum of 57% and maximum of 100% of alpha epochs.
Thus, the remaining epochs contained alpha activity either
only at source 1 or only at sources 2/3. The cortical sources
were then projected to the EEG sensors by using a scalar
lead-potential calculated with a BEM head model based on
individual MRIs. Four different levels of Gaussian random
noise were added to the sensors (SNRs of 1, 2, 3, or 5). A
total of 300 epochs of 1 s were created in this way to obtain
the same data size as in real data. The simulated sensor data
was then processed as the real data: it was bandpass filtered
between 1 and 20 Hz, and projected back to all gray matter
grid locations through a spatial filter matrix calculated with
the MVBF and sLORETA inverse solutions described above.
The WND of IC was computed at all cortical grid locations
for the alpha frequency band as in the real data.
The Euclidean distance between the WND peak and the
coordinate of source 1 was then calculated to determine
the localization error. Errors were subjected to a three-way
13

714
ANOVA with the setup (head model and number of chan-
nels), inverse solution, and SNR as dependent factors.
The coupling strength that was simulated was correlated
with the magnitude of WND. Correlation coefficients were
tested for difference between setups using Meng’s test for
correlated correlations (Meng et al. 1992).
Results
Numerical Simulations
In simulations, we observed that the localization accuracy
was significantly better when using MVBF than sLORETA
at all SNRs ­(F1,911 = 351, p < 0.0001), in accordance with
previous findings from high-density settings (Guggisberg
et al. 2011). Furthermore, localization error was significantly
influenced by the montage and head model (­F2,911 = 85,
p < 0.0001) with a rapid drop in localization accuracy
when using a template head model and low density mon-
tages in MVBF. When using sLORETA, the accuracy was
low already with MRI head models and 128 channels, but
remained stable when using 19 electrodes and a template
head model (Fig. 2A). The SNR of simulations did not sig-
nificantly influence localization accuracy (p = 0.36).
All setups were able to capture the simulated variation
in coupling strength, as indicated by a significant correla-
tion between the simulated coupling and the computed FC
(Fig. 2B). There was significant variation of the correlation
Fig. 2  Results from simulations of 3 coherent point sources. A
MVBF enabled better localization accuracy than sLORETA (sLOR),
but was vulnerable to the usage of template head models and low-
density montages. When using sLORETA, the accuracy remained
stable when using only 19 electrodes. Horizontal bars indicate sig-
nificant differences between setups (p < 0.05, Tukey Kramer HSD).
13
Brain Topography (2021) 34:709–719
coefficient between setups (χ2 = 64, p < 0.0001), but the cor-
relation coefficient obtained with sLORETA and 19 elec-
trodes did not significantly differ from the one obtained with
MVBF and 19 electrodes (z = − 1.4, p = 0.92).
Real Data
When using MVBF to compute source FC, the usage of tem-
plate head models led to only minor changes in the recon-
structed FC across 82 cortical ROIs as indicated by high
correlations to FC values obtained with individual MRI-
based head models in individual subjects (r = 0.85 ± 0.03),
see Fig. 3A. Conversely, MVBF was vulnerable to the usage
of low density montages as indicated by lower correlations
to the gold standard (r = 0.30 ± 0.38). Source FC obtained
with sLORETA differed to those obtained with MVBF,
even when using 128 electrodes and individual headmodels.
Pearson correlations of ROI WND values in each subject
revealed relatively low correlation coefficients on average
(r = 0.35 ± 0.33). On the other hand, sLORETA was robust to
reduction of the number of electrodes. Source FC obtained
with sLORETA and 19 channels was similar to source FC
obtained with an individual MRI-based headmodel and
sLORETA (r = 0.73 ± 0.09), while the correlation with the
MVBF gold standard (r = 0.30 ± 0.31) was similar as when
obtained with 128 electrodes.
When focusing on the right Precentral ROI which
was of primary interest here in the search for FC corre-
lates of motor performance, we also observed a drop of
B The magnitude of functional connectivity (FC) computed with all
setups correlated with the simulated variations in coupling strength,
thus confirming their ability to capture these variations. FC computed
after source localization with sLORETA covaried better with the sim-
ulated coupling strength than FC obtained from MVBF, but showed
vulnerability to usage of low-density montages
Brain Topography (2021) 34:709–719 715

Fig. 3  Fidelity of reproduction of source FC in real data. FC recon-
structions obtained with 128 electrodes, individual MRI head mod-
els, and MVBF were used as gold standard and correlated to source
FC obtained with other setups. A Box plot of correlation coefficients
obtained in each subject across at all 82 cortical ROIs of the atlas.
A drop in the average correlation coefficient occurred when using
19 electrodes or sLORETA (sLOR). B The same drop in reconstruc-
tion fidelity could be observed when focusing on the right Precentral
gyrus. Bars indicate the correlation coefficient across subjects. C
Using inverse solutions with 19 electrodes allowed better reliability
of reconstructing FC than using sensor FC without source localiza-
tion, no matter the reference. Box plots indicate the correlation coef-
ficents obtained in each subject across electrodes for sensor FC (grey
bars) or across the 19 cortical ROIs closest to each electrode for
source FC (green bars)

FC reconstruction fidelity when using MVBF with low-

density montages, while sLORETA had a generally poor
performance (Fig. 2B).
Sensor WND values correlated poorly with the source
WND values at the 19 ROIs that were closest to the posi-
tions of the electrodes. Using source reconstruction with
MVBF or sLORETA based on the same 19 electrodes yields
significantly better fidelity of FC (Fig. 3C).
Next, we investigated the ability of low-density EEG to
capture FC correlates of behavioral performance by com-
puting the correlation between the reconstructed WND and
motor performance in healthy subjects of dataset 2. Source
WND reconstructed from 128 channel data with individual
MRI-based head models and MVBF showed high correla-
tions with motor performance at the precentral gyrus, as
expected (Figs. 4 and 5). Using a subset of only 19 elec- Fig. 4  Correlation between FC and motor performance as quantified
trodes and a template head model did not reduce the ability by the number of completed sequences per minute in a finger-tapping
to capture FC correlates of motor performance when per- task (FTT) by 20 healthy subjects. Source FC derived after source
forming source localization with MVBF. Conversely, this localization with MVBF revealed significant FC correlates of motor
behavior, even when using only 19 electrodes. This was not the case
correlation was lost when doing source reconstruction with for source FC obtained from sLORETA (sLOR), sensor FC, or FC
sLORETA. Similarly, sensor FC obtained without unmixing obtained after ICA
or FC computed between ICA were unable to reveal signifi-
cant correlates of motor performance.
Finally, we intended to reproduce the findings of dataset 2 Discussion
in independent dataset 3, this time using a low-density EEG
system with dry-gel electrodes which are more convenient in Our findings show that it is challenging to reconstruct reli-
setup than regular gel electrodes. Data from 5 subjects had able estimates of FC with low-density EEG. The low den-
to be excluded due to abundant artifacts and only data from sity setup leads to a considerable localization error in the
the remaining 15 subjects were further analyzed. Figure 5 reconstructed FC when source localization is performed.
demonstrates that the correlations obtained in dataset 2 were Furthermore, analyzing sensor FC without unmixing is
not present in dataset 3, no matter the inverse solution used.

13

716 Brain Topography (2021) 34:709–719

Fig. 5  FC correlates of motor

performance. Source FC
reconstructed with MVBF from
data recorded with active gel
electrodes correlated with FTT
performance, even when using
template head models and 19
electrodes. Conversely, source
FC computed with sLORETA
or from 19 dry-gel electrodes
was unable to capture correlates
of motor performance

unable to track FC correlates of motor performance. Yet,
when high quality recordings are available and source FC
is computed with MVBF, correlations with motor perfor-
mance remained present even when using template head
models and 19 electrodes, such suggesting that low-den-
sity montages may suffice in some cases for tracking FC
as a neural correlate of behavior performance.
In accordance with the literature (Sekihara et al. 2005;
Hadjipapas et al. 2005; Guggisberg et al. 2011), we observed
superior localization accuracy of MVBF compared to sLO-
RETA in both simulations (Fig. 2A) and real data (Fig. 3A,
B). This can be explained with the fact that it adapts to the
data and with its superior handling of noise, leading to more
precise and focal reconstructions (Sekihara et al. 2005; Had-
jipapas et al. 2005). In the case of FC calculations, a lack of
focality leads to a greater susceptibility of inducing “ghost
interactions”, i.e., spurious interactions reconstructed in the
vicinity of true interactions due to signal mixing which is
incompletely removed by the inverse solutions (Palva et al.
2018). Together, this leads to issues when reconstructing
FC with non-adaptive inverse solutions such as sLORETA.
It is, however, important to note that the reconstructions
obtained with low-density sLORETA remain more similar
to high-resolution MVBF than when using sensor-space FC
without source localization (Fig. 3C).
The advantage of MVBF over other approaches was
reduced when using low-density montages (Figs. 2 and 3).
This is because the noise handling of MVBF depends on
a sufficiently large number of sensors. Thus, the localiza-
tion error of MVBF is proportional to the sensor coverage
(Steinsträter et al. 2010). Furthermore MVBF depends
on head model accuracy. If the head model is not precise

13
Brain Topography (2021) 34:709–719 717
enough, the real source is simply considered as noise and
ignored (Van Veen et al. 1997; Steinsträter et al. 2010).
Conversely, sLORETA is a non-adaptive filter where the
weights do not depend on the data. Hence, sLORETA’s
location accuracy is less electrode-density-dependent. Yet,
reducing the number of sensors reduces the number of
available data points for reconstructing the source activity,
leading to a redundancy of the estimated FC among the
source regions even when using sLORETA.
Our simulations showed that MVBF led to lower cor-
relations between the simulated coupling strength and
the reconstructed FC magnitude than sLORETA, at least
in case of high density montages (Fig. 2B). MVBF has
known difficulties in reconstructing the time course of the
source signal (Huang et al. 2014). Beamformers presume
that the reconstructed sources are uncorrelated in time.
Source activities are seen as orthogonal to one another in
the time domain making source interdependencies non-
existent (Van Veen et al. 1997; Sekihara et al. 2002; Had-
jipapas et al. 2005). Previous studies have demonstrated
that beamformers only show acceptable accuracy in recon-
structing correlated sources when the correlation was tran-
sient, lasting less than 30–40% of the analysis duration.
When exceeding 40%, temporal bias and signal cancel-
lation appears (Hadjipapas et al. 2005). As the signals in
our simulations did show some correlation, this may have
introduced distortions of the reconstructed time series, and
thus led to a reduced ability to quantify coupling strength.
This is probably also the case in real recordings where
some areas may show longer lasting temporal correlations.
Despite the disadvantages of MVBF, we observe that
even when using template head models and low density
montages, source FC reconstructed with MVBF gave best
overall ability to capture FC correlates of motor perfor-
mance (Fig. 4). In particular, FC reconstructions obtained
in source space with MVBF were clearly superior in find-
ing correlates of motor performance than sensor FC or FC
obtained after ICA. Unlike sensor FC or ICA, MVBF uses
information about head geometry to unmix the underly-
ing neural signals, which gives it a precious advantage for
applications on diagnosis and treatment of FC changes.
Yet, we have to bear in mind that we were unable to find
such correlates of motor performance in a second, inde-
pendent dataset which was recorded with a low-density
EEG system using more convenient dry-gel electrodes
(Fig. 5). This is most likely due to more noisy recordings
obtained with this system, which is evident already by
the fact that 5 out of 20 subjects had to be excluded for
this reason. The impedance between skin and electrodes
is larger when using dry or dry-gel electrodes, which dete-
riorates signal quality and further complicates the already
difficult task of reconstructing FC from low-density EEG.
This also generally raises a caveat with regards to the
robustness of FC reconstructions when using low-density
montages.
Nevertheless, our results suggest that low-density EEG
may in some instances be sufficient for applications for
motor training to improve the participant’s performance.
Both patients and healthy subjects in sport and music could
benefit from neuromodulation of alpha-band FC, using neu-
rofeedback. A low-density EEG then enables more comfort
and ease-of-use for the end-user.
It is important to stress that our study was performed
with the aim of potential clinical applications with a par-
ticular neural target. The assessment of other frequency
bands, which often have lower SNR than the alpha band,
and the usage of other indices of FC, may lead to different
results. Furthermore, our study focused on FC correlates
of motor performance. As FC correlates with performance
also in other domains, it would be intersting to extend this
to other functions. In these cases, the ability of low-density
EEG to capture these correlates would need to be evaluated
separately. Our analyses across all cortical ROIs suggests
that many of the observations made for the motor cortex
and many of the observed difficulties apply also to other
brain aras.
Conclusion
Our study illustrates the difficulties in reconstructing FC
from low-density EEG. It is common practice to compute
sensor FC among electrodes, and to attribute the resulting
findings to certain brain areas. This study shows that the
mixed signals recorded by each electrode precludes an esti-
mation of FC at a given brain area. Furthermore, FC cor-
relates of behavior which are robustly observed at the source
level are lost when analyzing sensor FC. FC reconstruction
from low-density EEG is more feasible when obtaining
source localization with adaptive inverse solutions such as
MVBF, although these reconstructions are less robust than
the ones obtain from high-density EEG. These insights may
enable new possibilities for training and learning in clinical
practice and public usage.
Author Contributions  Conceived and designed the experiments: AGG.
Performed the experiments: AM, LA, SS. Analyzed the data: DAND,
AGG. Contributed materials/analysis tools: AM, AGG. Wrote the
paper: DAND, AGG.
Funding  Open Access funding provided by Université de Genève. This
work was supported by the Swiss National Science Foundation, Grant
Nos. CRSII5-170985 and 320030-169275 to AGG.
Data Availability  The datasets analyzed during the current study are
not publicly available because the consent from the participants did not
13

718
include public dissemination, but are available from the corresponding
author upon reasonable request.
Code Availability  Analyses have been done with open source Matlab
toolboxes that can be downloaded freely as indicated in the manuscript.
Declarations  Hindriks R, van Putten MJAM, Deco G (2014) Intra-cortical propaga-
Conflict of interest  The authors have no conflict of interest to declare
that are relevant to the content of this article.
Consent to Participate  Written informed consent was obtained from
all individual participants included in the study.
Consent for Publication  Patients signed informed consent regarding
publishing their data.
Ethical Approval  All procedures were approved by the University Hos-
pital of Geneva Ethics Committee.
Open Access  This article is licensed under a Creative Commons Attri-
bution 4.0 International License, which permits use, sharing, adapta-
tion, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source,
provide a link to the Creative Commons licence, and indicate if changes
were made. The images or other third party material in this article are
included in the article's Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in
the article's Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a
copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.
References
Aertsen AMHJ, Gerstein GL, Habib MK, Palm G (1989) Dynamics of
neuronal firing correlation: modulation of “effective connectivity.”
J Neurophysiol 61:900. https://​doi.​org/​10.​1152/​jn.​1989.​61.5.​900
Allaman L, Mottaz A, Kleinschmidt A, Guggisberg AG (2020) Spon-
taneous network coupling enables efficient task performance
without local task-induced activations. J Neurosci 40:9663–9675.
https://​doi.​org/​10.​1523/​JNEUR​OSCI.​1166-​20.​2020
Chapeton JI, Haque R, Wittig JH et al (2019) Large-scale communica-
tion in the human brain is rhythmically modulated through alpha
coherence. Curr Biol 29:2801-2811.e5. https://​doi.​org/​10.​1016/j.​
cub.​2019.​07.​014
Dalal SS, Zumer JM, Guggisberg AG et al (2011) MEG/EEG source
reconstruction, statistical evaluation, and visualization with NUT-
MEG. Comput Intell Neurosci. https://​doi.​org/​10.​1155/​2011/​
758973
Dubovik S, Pignat JM, Ptak R et al (2012) The behavioral significance
of coherent resting-state oscillations after stroke. Neuroimage
61:249–257. https://​doi.​org/​10.​1016/j.​neuro​image.​2012.​03.​024
Fox MD, Greicius M (2010) Clinical applications of resting state func-
tional connectivity. Front Syst Neurosci. https://​doi.​org/​10.​3389/​
fnsys.​2010.​00019
Guggisberg AG, Dalal SS, Zumer JM et al (2011) Localization of
cortico-peripheral coherence with electroencephalography. Neu-
roimage 57:1348–1357. https://​doi.​org/​10.​1016/j.​neuro​image.​
2011.​05.​076
13
Brain Topography (2021) 34:709–719
Guggisberg AG, Rizk S, Ptak R et al (2015) Two intrinsic coupling
types for resting-state integration in the human brain. Brain
Topogr 28:318–329. https://​doi.​org/​10.​1007/​s10548-​014-​0394-2
Hadjipapas A, Hillebrand A, Holliday IE et al (2005) Assessing inter-
actions of linear and nonlinear neuronal sources using MEG
beamformers: a proof of concept. Clin Neurophysiol 116:1300–
1313. https://​doi.​org/​10.​1016/j.​clinph.​2005.​01.​014
tion of EEG alpha oscillations. Neuroimage 103:444–453. https://​
doi.​org/​10.​1016/J.​NEURO​IMAGE.​2014.​08.​027
Huang Y, Parra LC (2015) Fully automated whole-head segmentation
with improved smoothness and continuity, with theory reviewed.
PLoS ONE 10:1–34. https://​doi.​org/​10.​1371/​journ​al.​pone.​01254​
77
Huang MX, Huang CW, Robb A et al (2014) MEG source imaging
method using fast L1 minimum-norm and its applications to sig-
nals with brain noise and human resting-state source amplitude
images. Neuroimage 84:585–604. https://​doi.​org/​10.​1016/j.​neuro​
image.​2013.​09.​022
Hyvarinen A (1999) Fast and robust fixed-point algorithms for inde-
pendent component analysis. IEEE Trans Neural Netw 10:626–
634. https://​doi.​org/​10.​1109/​72.​761722
Koutroumanidis M, Arzimanoglou A, Caraballo R et al (2017) The
role of EEG in the diagnosis and classification of the epilepsy
syndromes: a tool for clinical practice by the ILAE Neurophysiol-
ogy Task Force (Part 1). Epileptic Disord 19:233–298. https://d​ oi.​
org/​10.​1684/​epd.​2017.​0935
Koyama MS, Di Martino A, Zuo XN et al (2011) Resting-state func-
tional connectivity indexes reading competence in children and
adults. J Neurosci 31:8617–8624. https://​doi.​org/​10.​1523/​jneur​
osci.​4865-​10.​2011
Leeb R, Perdikis S, Tonin L et al (2013) Transferring brain-computer
interfaces beyond the laboratory: successful application control
for motor-disabled users. Artif Intell Med 59:121–132. https://d​ oi.​
org/​10.​1016/j.​artmed.​2013.​08.​004
McFarland DJ, McCane LM, David SV, Wolpaw JR (1997) Spatial fil-
ter selection for EEG-based communication. Electroencephalogr
Clin Neurophysiol 103:386–394. https://​doi.​org/​10.​1016/​S0013-​
4694(97)​00022-2
Meng XL, Rosenthal R, Rubin DB (1992) Comparing correlated cor-
relation coefficients. Psychol Bull 111:172–175. https://​doi.​org/​
10.​1037/​0033-​2909.​111.1.​172
Michel CM, Murray MM, Lantz G et al (2004) EEG source imaging.
Clin Neurophysiol 115:2195–2222
Mottaz A, Solcà M, Magnin C et al (2015) Neurofeedback training of
alpha-band coherence enhances motor performance. Clin Neu-
rophysiol 126:1754–1760. https://​doi.​org/​10.​1016/j.​clinph.​2014.​
11.​023
Mottaz A, Corbet T, Doganci N et al (2018) Modulating functional con-
nectivity after stroke with neurofeedback: effect on motor deficits
in a controlled cross-over study. Neuroimage Clin 20:336–346.
https://​doi.​org/​10.​1016/j.​nicl.​2018.​07.​029
Newman MEJ (2004) Analysis of weighted networks. Phys Rev E Stat
Nonlin Soft Matter Phys 70:56131. https://d​ oi.o​ rg/1​ 0.1​ 103/P
​ hysR​
evE.​70.​056131
Nicolo P, Rizk S, Magnin C et al (2015) Coherent neural oscillations
predict future motor and language improvement after stroke. Brain
138:3048–3060. https://​doi.​org/​10.​1093/​brain/​awv200
Nolte G, Bai O, Wheaton L et al (2004) Identifying true brain interac-
tion from EEG data using the imaginary part of coherency. Clin
Neurophysiol 115:2292–2307. https://d​ oi.o​ rg/1​ 0.1​ 016/J.C​ LINPH.​
2004.​04.​029
Nunez PL, Srinivasan R, Westdorp AF et al (1997) EEG coherency
I: statistics, reference electrode, volume conduction, Lapla-
cians, cortical imaging, and interpretation at multiple scales.
Brain Topography (2021) 34:709–719 719
Electroencephalogr Clin Neurophysiol 103:499–515. https://​doi.​
org/​10.​1016/​S0013-​4694(97)​00066-7
Palva JM, Wang SH, Palva S et al (2018) Ghost interactions in MEG/
EEG source space: a note of caution on inter-areal coupling meas-
ures. Neuroimage 173:632–643. https://​doi.​org/​10.​1016/j.​neuro​
image.​2018.​02.​032
Pascual-Marqui RD (2002) Standardized low-resolution brain electro-
magnetic tomography (sLORETA): technical details. Methods
Find Exp Clin Pharmacol 24(Suppl D):5–12
Pascual-Marqui RD, Sekihara K, Brandeis D, Michel CM (2009) Imag-
ing the electric neuronal generators of EEG/MEG. In: Michel CM,
Koenig T, Brandeis D et al (eds) Electrical neuroimaging. Cam-
bridge University Press, Cambridge, pp 49–77
Perrin F, Pernier J, Bertrand O et al (1987) Mapping of scalp potentials
by surface spline interpolation. Electroencephalogr Clin Neuro-
physiol 66:75–81. https://d​ oi.o​ rg/1​ 0.1​ 016/0​ 013-4​ 694(87)9​ 0141-6
Pijnenburg M, Brumagne S, Caeyenberghs K et al (2015) Resting-state
functional connectivity of the sensorimotor network in individu-
als with nonspecific low back pain and the association with the
sit-to-stand-to-sit task. Brain Connect 5:303–311. https://​doi.​org/​
10.​1089/​brain.​2014.​0309
Reineberg AE, Andrews-Hanna JR, Depue BE et al (2015) Resting-
state networks predict individual differences in common and spe-
cific aspects of executive function. Neuroimage 1:69–78. https://​
doi.​org/​10.​1016/j.​neuro​image.​2014.​09.​045.​Resti​ng-​state
Rossetti AO, Carrera E, Oddo M (2012) Early EEG correlates of neu-
ronal injury after brain anoxia. Neurology 79:2010. https://​doi.​
org/​10.​1212/​WNL.​0b013​e3182​768eaf
Schoffelen JM, Gross J (2009) Source connectivity analysis with MEG
and EEG. Hum Brain Mapp 30:1857–1865
Sekihara K, Nagarajan SS, Poeppel D, Marantz A (2002) Performance
of an MEG adaptive-beamformer technique in the presence of
correlated neural activities: effects on signal intensity and time-
course estimates. IEEE Trans Biomed Eng 49:1534–1546. https://​
doi.​org/​10.​1109/​TBME.​2002.​805485
Sekihara K, Nagarajan SS, Poeppel D, Marantz A (2004) Asymptotic
SNR of scalar and vector minimum-variance beamformers for
neuromagnetic source reconstruction. IEEE Trans Biomed Eng
51:1726–1734. https://​doi.​org/​10.​1109/​TBME.​2004.​827926
Sekihara K, Sahani M, Nagarajan SS (2005) Localization bias and
spatial resolution of adaptive and non-adaptive spatial filters for
MEG source reconstruction. Neuroimage 25:1056–1067
Sekihara K, Owen JP, Trisno S, Nagarajan SS (2011) Removal of spu-
rious coherence in MEG source-space coherence analysis. IEEE
Trans Biomed Eng 58:3121–3129. https://d​ oi.o​ rg/1​ 0.1​ 109/T ​ BME.​
2011.​21625​14
Srinivasan N (2007) Cognitive neuroscience of creativity: EEG based
approaches. Methods 42:109–116. https://​doi.​org/​10.​1016/j.​
ymeth.​2006.​12.​008
Steinsträter O, Sillekens S, Junghoefer M et al (2010) Sensitivity of
beamformer source analysis to deficiencies in forward modeling.
Hum Brain Mapp 31:1907–1927. https://​doi.​org/​10.​1002/​hbm.​
20986
Stenroos M, Mantynen V, Nenonen J (2007) A Matlab library for solv-
ing quasi-static volume conduction problems using the boundary
element method. Comput Methods Programs Biomed 88:256–263.
https://​doi.​org/​10.​1016/j.​cmpb.​2007.​09.​004
Tzourio-Mazoyer N, Landeau B, Papathanassiou D et al (2002) Auto-
mated anatomical labeling of activations in SPM using a macro-
scopic anatomical parcellation of the MNI MRI single-subject
brain. Neuroimage 15:273–289. https://​doi.​org/​10.​1006/​nimg.​
2001.​0978
Van Veen BD, Van Drongelen W, Yuchtman M, Suzuki A (1997)
Localization of brain electrical activity via linearly constrained
minimum variance spatial filtering. IEEE Trans Biomed Eng
44:867–880. https://​doi.​org/​10.​1109/​10.​623056
Varela F, Lachaux JP, Rodriguez E, Martinerie J (2001) The brainweb:
phase synchronization and large-scale integration. Nat Rev Neu-
rosci 2:229–239. https://​doi.​org/​10.​1038/​35067​550
Wang L, Negreira A, Laviolette P et al (2010) Intrinsic interhemi-
spheric hippocampal functional connectivity predicts individual
differences in memory performance ability NIH Public Access.
Hippocampus. https://​doi.​org/​10.​1002/​hipo.​20771
Westlake KP, Hinkley LB, Bucci M et al (2012) Resting state alpha-
band functional connectivity and recovery after stroke. Exp Neu-
rol 237:160–169. https://d​ oi.o​ rg/1​ 0.1​ 016/j.e​ xpneu​ rol.2​ 012.0​ 6.0​ 20
Wilmshurst JM, Gaillard WD, Vinayan KP et al (2015) Summary of
recommendations for the management of infantile seizures: Task
Force Report for the ILAE Commission of Pediatrics. Epilepsia
56:1185–1197. https://​doi.​org/​10.​1111/​epi.​13057
Zhang J, Hughes LE, Rowe JB (2012) Selection and inhibition
mechanisms for human voluntary action decisions. Neuroimage
63:392–402. https://d​ oi.o​ rg/1​ 0.1​ 016/j.n​ euroi​ mage.2​ 012.0​ 6.0​ 58S1​
053-​8119(12)​00674-​X[pii]
Publisher's Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
13