sensors
Article
Induction of Neural Plasticity Using a Low-Cost Open Source
Brain-Computer Interface and a 3D-Printed Wrist Exoskeleton
Mads Jochumsen 1, * , Taha Al Muhammadee Janjua 1 , Juan Carlos Arceo 2 , Jimmy Lauber 2 ,
Emilie Simoneau Buessinger 2 and Rasmus Leck Kæseler 1
1
2
*
Citation: Jochumsen, M.;
Janjua, T.A.M.; Arceo, J.C.; Lauber, J.;
Buessinger, E.S.; Kæseler, R.L.
Induction of Neural Plasticity Using
a Low-Cost Open Source
Brain-Computer Interface and a
3D-Printed Wrist Exoskeleton. Sensors
2021, 21, 572. https://doi.org/
10.3390/s21020572
Department of Health Science and Technology, Aalborg University, 9220 Aalborg, Denmark;
taha@hst.aau.dk (T.A.M.J.); rlk@hst.aau.dk (R.L.K.)
LAMIH UMR CNRS 8201, INSA Hauts de France, Université Polytechnique Hauts de France,
F-59313 Valenciennes, France; juancarlos.arceo@uphf.fr (J.C.A.); jimmy.lauber@uphf.fr (J.L.);
emilie.simoneau@uphf.fr (E.S.B.)
Correspondence: mj@hst.aau.dk
Abstract: Brain-computer interfaces (BCIs) have been proven to be useful for stroke rehabilitation,
but there are a number of factors that impede the use of this technology in rehabilitation clinics
and in home-use, the major factors including the usability and costs of the BCI system. The aims of
this study were to develop a cheap 3D-printed wrist exoskeleton that can be controlled by a cheap
open source BCI (OpenViBE), and to determine if training with such a setup could induce neural
plasticity. Eleven healthy volunteers imagined wrist extensions, which were detected from single-trial
electroencephalography (EEG), and in response to this, the wrist exoskeleton replicated the intended
movement. Motor-evoked potentials (MEPs) elicited using transcranial magnetic stimulation were
measured before, immediately after, and 30 min after BCI training with the exoskeleton. The BCI
system had a true positive rate of 86 ± 12% with 1.20 ± 0.57 false detections per minute. Compared
to the measurement before the BCI training, the MEPs increased by 35 ± 60% immediately after and
67 ± 60% 30 min after the BCI training. There was no association between the BCI performance
and the induction of plasticity. In conclusion, it is possible to detect imaginary movements using an
open-source BCI setup and control a cheap 3D-printed exoskeleton that when combined with the
BCI can induce neural plasticity. These findings may promote the availability of BCI technology for
rehabilitation clinics and home-use. However, the usability must be improved, and further tests are
needed with stroke patients.
Keywords: brain-computer interface; neural plasticity; neurorehabilitation; motor imagination; exoskeleton
Received: 8 December 2020
Accepted: 12 January 2021
Published: 15 January 2021
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Copyright: © 2021 by the authors. Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1. Introduction
Brain-computer interfaces (BCIs) have over the past years been proposed also as a
tool for motor rehabilitation after neural injuries, such as spinal cord injury or stroke [1–6].
It is well-established that BCIs can be used for inducing neural plasticity [7–11], which is
believed to be the underlying mechanism of motor learning/recovery [12]. These neuroplastic changes are induced in the brain by pairing the movement-related activity of the
brain with the inflow of congruent somatosensory feedback from, e.g., electrical stimulation [7], rehabilitation robots, or exoskeletons [8]. Movement-related cortical potentials [13]
or event-related desynchronization [14,15] have typically been extracted from single-trial
electroencephalography (EEG) recordings and used as the control signals for triggering
the external device that is going to elicit the somatosensory feedback. The concept of
inducing plasticity using a BCI has been shown in several studies; however, this technology
is rarely used in rehabilitation clinics and the patient’s home. This is due to several reasons,
one of them being that it is still a fairly new technology, while some of the translational
issues include the complexity of the systems in terms of setting up (e.g., mounting the
EEG cap and calibrating the system), ensuring a good and stable signal quality, which
Sensors 2021, 21, 572. https://doi.org/10.3390/s21020572
https://www.mdpi.com/journal/sensors
Sensors 2021, 21, 572
2 of 14
may require a skilled operator, the mental fatigue of the user, user compliance, the price
of the technology, and access to the detection algorithms [16,17]. In recent years, several
low-cost commercial EEG systems have become available [18]. Some of these systems
may not be useful for applications where neural plasticity is induced in the motor system,
since they do not record electrical activity from the relevant brain areas [19]. However, it
is possible to record the electrical activity of the motor cortex with some low-cost EEG
systems. Moreover, several research groups have made their detection algorithms publicly available (see, e.g., the OpenViBE project [20]). The feasibility of using such low-cost
systems for detecting movement-related brain activity has been outlined recently [21–23],
thus making the BCI technology available to a wider audience than BCI researchers. To
use the BCI for inducing neural plasticity, besides for neurofeedback applications [24], an
external device is needed to provide congruent somatosensory feedback. This could be a
stimulator that could stimulate the relevant nerves and muscles electrically, or it could be
an exoskeleton. With the current advances made within the design and manufacturing of
exoskeletons through 3D printing [25,26], it has become cheap to create simple exoskeletons
for controlling certain joints such as the wrist or ankle. It is possible to create a simple
exoskeleton that can perform wrist extensions or the dorsiflexion of the ankle joint with a
single actuator [11]. Both of these movement types are important to train during stroke
rehabilitation. It has been shown previously that neural plasticity, when quantified with
transcranial magnetic stimulation (TMS), can be induced using BCI-triggered electrical
stimulation and passive movements from rehabilitation robots/exoskeletons for the cortical
projections of the lower limb muscles [7–11], but this has not been shown for the cortical
projections of the upper limb muscles, although functional improvements in stroke patients
have been reported for the upper limbs (see, e.g., Refs. [4,5,27,28]). Therefore, the aim of
this study is to investigate if a BCI-triggered exoskeleton can induce neural plasticity in the
cortical projections of the forearm muscles that control wrist extension. Moreover, it will
be tested if this is possible using a low-cost EEG amplifier and open source BCI software.
Lastly, a cheap 3D-printed exoskeleton will be developed to replicate wrist extension. The
BCI-triggered exoskeleton will be evaluated in terms of BCI system performance and the
ability to induce neural plasticity.
2. Materials and Methods
2.1. Subjects
Eleven healthy subjects participated (four females, age: 28 ± 3 years). Prior to participation, the subjects provided their written informed consent and filled in a questionnaire
for their eligibility for TMS based on the recommendations in Ref. [29]. All procedures
were approved by the local ethical committee (N-20130081), and were in accordance with
the Helsinki Declaration.
2.2. Experimental Setup
Initially, the subjects were seated in a comfortable chair, where the procedures were
explained, and they were familiarized with TMS. See Figure 1 for a timeline of the experiment. Afterwards, they were instructed on how to perform motor imagination, and they
spent ~5 min training this. After the motor imagination training, the subjects imagined
30 wrist extensions of the right wrist while continuous EEG was recorded. A visual cue
was generated by the “Motor Imagery BCI” in OpenViBE; the visual cue was modified such
that 30 idle/rest trials (“REST” was displayed on the screen) and 30 motor imagination
trials (a red arrow pointing to the right was displayed on the screen) were performed. The
imaginary movement was maintained for four seconds. These trials were used to calibrate
the asynchronous BCI for controlling the wrist exoskeleton. During the actual BCI training,
the wrist exoskeleton was mounted on the subject on the right forearm and hand. The
forearm and hand rested on the armrest of the chair during the training. The subjects were
asked to trigger the exoskeleton by imagining an extension of the right wrist; the training
was complete when 50 correct pairings of motor imagination and the movement of the
Sensors 2021, 21, 572
3 of 14
exoskeleton were obtained. The subjects had to keep imagining the movement while the
exoskeleton performed the movement. Before, immediately after, and 30 min after the
BCI training, TMS measurements were performed, whereby 30 motor-evoked potentials
(MEPs) were obtained.
Setup EEG
MI training
TMS familiarization
(15 min)
BCI calibration
30 × MI
30 × Rest
(10 min)
TMS hotspotting
Find Resting
Threshold (RTh)
(10 min)
Pre-intervention
TMS
30 × 120% RTh
(5 min)
Post 30-min
intervention TMS
30 × 120% RTh
(5 min)
25 min
Post-intervention
TMS
30 × 120% RTh
(5 min)
Find BCI threshold
BCI intervention
50 × True Positives
(5 min + 15 min)
Figure 1. Timeline of the experiment; the approximate duration of each block is indicated in parentheses. First, the subject
was familiarized with transcranial magnetic stimulation (TMS) and motor imagination (MI), and the EEG cap was mounted.
Next, the brain-computer interface (BCI) was calibrated, followed by the identification of the optimal stimulation site
(hotspot) and intensity (RTh). The pre-intervention TMS, post-intervention TMS, and post-30 min intervention TMS were
identical. After the pre-intervention TMS, the threshold for each subject was tested with an online BCI and changed if
needed. Afterwards, the intervention started, and it was stopped when the subject reached 50 correct pairings between
motor imagination (MI) and movement of the exoskeleton. The post-30 min intervention TMS started 30 min after the BCI
intervention ended.
2.3. Recordings
2.3.1. EEG
Seven channels of continuous EEG were recorded (Cyton Biosensing Board, OpenBCI,
Brooklyn, NY, USA) from F1, F2, C3, Cz, C4, P1, and P2 with respect to the International
10-20 System using sintered ring electrodes placed in an EASYCAP EEG cap (EASYCAP
GmbH, Herrsching, Germany). The signals were sampled at 250 Hz. The ground electrode
was placed at AFz, and the reference electrode was placed on the mastoid bone behind the
right ear. The subjects were asked to sit still and avoid the contraction of facial muscles
and blinking.
2.3.2. EMG
MEPs were recorded using surface EMG electrodes (Neuroline 720, Ambu A/S,
Ballerup, Denmark) placed on the extensor digitorum muscle in a bipolar derivation.
Two electrodes were placed on the belly of the muscle, which was identified through
palpation, and a ground electrode was placed on the distal head of the Humerus bone. The
signals were amplified with a gain of 5000 using a customized amplifier (Jan Stavnshøj,
Aalborg University), and the signals were sampled at 4000 Hz using the Mr. Kick software
(Knud Larsen, Aalborg University).
2.4. Transcranial Magnetic Stimulation
MEPs (Figure 2) were elicited with a single-pulse TMS (Magstim 200, Magstim Company, Dyfed, UK) using a figure-of-eight coil with a posterior–anterior current direction.
Sensors 2021, 21, 572
4 of 14
First, the optimal stimulation site was determined. This was defined as the location where
the largest MEP peak-peak amplitudes were obtained. Next, the resting threshold was
determined. It was defined as the lowest stimulation intensity that would elicit an MEP of
at least 50 µV peak-peak amplitude in five out of ten simulations. In the measurements
before, immediately after and 30 min after the intervention, 30 stimuli were given at 120%
of the resting threshold. A random break of 5–7 s separated two consecutive stimuli.
Motor-Evoked Potential (MEP)
2
1.5
Peak-Peak Amplitude
1
Amplitude (mV)
0.5
0
-0.5
-1
-1.5
-2
0
25
50
75
100
125
150
175
200
225
250
Time (ms)
Figure 2. Motor-evoked potential (MEP) from a representative subject (post-intervention transcranial magnetic stimulation measurement for subject 1). The peak around 25 milliseconds is the stimulation artefact from the transcranial
magnetic stimulation.
2.5. Brain-Computer Interface
The “Motor Imagery BCI” from OpenViBE was used for the detection of the imaginary
wrist extensions. The data were bandpass filtered from 8 to 30 Hz using a 5th order
Butterworth filter, and a common spatial pattern spatial filter was calculated from the
calibration data, which maximized the differences in spectral power between the idle and
motor imagery classes. The trials were divided into 1 s windows with a shift of 1/16 s,
and the logarithmic band powers were calculated from each window and used as features.
Based on the features extracted from the training data, a linear discriminant analysis
classifier was trained using 5-fold cross-validation. In the online test, the classification
of the imagined movement was altered compared to the original OpenViBE scenario.
An imagined movement was detected when eight consecutive windows (8/16 s = 0.5 s)
exceeded a threshold that was determined for each subject individually. The determination
of the threshold took less than five minutes (see Figure 1), and it was done to obtain a
trade-off between true positive and false-positive detections. When the BCI detected an
imagined movement, a trigger was sent through a transmission control protocol to an
Arduino MKR1000 that activated the wrist exoskeleton (see Figure 3). The performance
metrics of the BCI were the true positive rate, the number of false negatives per minute,
and the number of false positive detections per minute. The subjects indicated verbally if
the trial was correct (true positive) or if it was incorrect (false positive or false negative).
Sensors 2021, 21, 572
5 of 14
Open BCI
PC for EEG computation
Exoskeleton controlled by Acutonix servo motor
Arduino MKR1000
Acutonix Linear Actuator Control board
12V
Figure 3. Overview of the hardware setup. The Arduino and Linear Actuator Control board were mounted on the
exoskeleton. The EEG electrodes were connected through wires to the Open BCI board from which the signals were
transmitted through wireless communication to the PC running OpenViBE. Once an imagined wrist extension was detected
a trigger was sent through wireless communication to the Arduino on the exoskeleton. The Arduino was connected to the
Linear Actuator Control board with a wire. The Linear Actuator Control board was powered with a 12 V power supply. The
motor was connected to the Linear Actuator Control board with a wire.
2.6. Exoskeleton
The exoskeleton was 3D-printed and developed specifically for this study (see Figure 4).
The purpose of the exoskeleton was to control the wrist angle position (denoted by Ψ) to
replicate an extension of the wrist (see Figure 3). The exoskeleton was 3D-printed using
the material PLAMAX, and it was actuated by a model L16-P linear piston (Actuonix,
Motion Devices Inc® , Victoria, BC, Canada) which was connected to the X2 input of a linear
actuator control (LAC) board (Actuonix, Motion Devices Inc® , Victoria, BC, Canada) with
default settings. A 12 V power supply was connected to the piston and LAC board in the
(±) X6 inputs. The analog output (A0) and ground reference of the Arduino MKR1000 were
connected to the (VC) and (−) X6 inputs of the LAC board, respectively; the Arduino board
was programmed and powered via a USB connection (5 V power supply). Finally, the
design includes an LED bar that indicated the wrist angle position, which was connected
to the (±) X3 inputs and (P) X4 input of the LAC board. The exoskeleton received an
activation signal from the BCI through serial communication to the Arduino board with a
baud rate of 9600. The Arduino sends a reference position signal for the linear piston and
the LAC board for compensating for the position error signal of the position in order to
follow the desired trajectory. The predefined trajectory Ψd (t) begins at the initial position
Ψd (0) = 180◦ , then the wrist is extended in 1.8 s to Ψd (1.8) = 112.36◦ and stays in this
position for 0.5 s before returning to the initial position, which also takes 1.8 s. The average
movement speed while moving is 37.58◦ /s.
Sensors 2021, 21, 572
6 of 14
A – Strap holes for securing the exoskeleton
B – Linear Actuator
C – LAC Board
D – Potentiometer
E – Arduino board
F – LED indicator of arm position
A
A
B
C
D
E
F
A
A
Figure 4. View of the 3D-printed exoskeleton. The illustration is not drawn to scale. The surfaces that were in contact with
the forearm and hand were padded with foam to improve the comfort. The exoskeleton was fixated to the subject’s hand
and forearm with Velcro straps (A). ‘LAC’: Linear Actuator Control.
2.7. Statistics
A one-way repeated measure analysis of variance (ANOVA) with time as a factor
(3 levels: pre-, post-, and post-30 intervention) was performed on the median MEP peakpeak amplitudes to investigate if there was a difference between the MEP amplitudes at
the three different time points. A significant test statistic was followed up with a posthoc
analysis using Bonferroni correction to avoid multiple comparisons. Moreover, Spearman correlation was calculated between the BCI performance metrics (true positive rate,
the number of false negatives per minute, and the number of false positive detections
per minute) including the duration of the training and the changes in MEP amplitude.
Moreover, the correlation between the MEP changes and BCI performance metrics was calculated with respect to age (Spearman correlation) and gender (Point Biserial correlation).
A significant test was assumed when p < 0.05.
3. Results
The results are summarized in Table 1 and Figure 5. The presented p-values for the
posthoc test have been Bonferroni corrected. On average, 86 ± 12% of the imaginary wrist
extensions were correctly detected by the asynchronous BCI, while there were 1.20 ± 0.57
false positive detections per minute and 0.63±0.58 false negatives per minute (see Table 1).
It should be noted that there is a large standard deviation, especially for the true positive
rate, and especially subject 2 had difficulties in activating the exoskeleton through the BCI.
The true positive rate was higher for subjects 5 and 6, but they had a large number of false
positive detections per minute.
The results of the intervention are presented in Figure 5. There was an increase in
the MEP from before intervention to immediately after, and 30 min after, the intervention
in both absolute units (mV) and relative units (percent). The statistical analysis showed
a significant effect of time (F (2,20) = 4.63; p = 0.022). The posthoc analysis revealed a
significant increase in the MEP amplitude from the measurement before the intervention
to the measurement 30 min after the intervention (p = 0.028). There was no difference
between the MEP from the measurement before the intervention and that immediately
after (p = 0.73), or the MEPs in the measurements after the intervention (p = 0.34).
There was no correlation between the true positive rate (correlation coefficient: 0.36;
p = 0.28), the number of false positive detections per minute (correlation coefficient: −0.35;
p = 0.30), the number of false negatives per minute (correlation coefficient: −0.39; p = 0.24),
or duration (correlation coefficient: 0.31; p = 0.36), and the changes in MEP amplitude from
Sensors 2021, 21, 572
7 of 14
before to 30 min after the intervention. Age and gender did not correlate with any of the
other measures.
Table 1. Brain-computer interface performance.
Subject
True Positive Rate
(%)
False Negatives per Minute
False Positive Detections per Minute
Duration of the Training
(Minutes)
1
2
3
4
5
6
7
8
9
10
11
Mean ± std
93
56
98
79
81
83
100
94
86
89
94
86 ± 12
0.36
2.11
0.10
1.08
1.09
0.67
0
0.23
0.53
0.43
0.33
0.63 ± 0.58
0.55
0.78
1.00
0.50
2.10
1.93
1.81
1.77
1.10
0.57
1.11
1.20 ± 0.57
11
18
11
12
11
15
16
13
15
14
9
13 ± 3
9
a
140
*
b
8
120
Changes in MEP Amplitude (%)
MEP Amplitude (mV)
7
6
5
4
3
100
80
60
40
2
20
1
0
0
Pre
Post
Post 30
Measurement
Pre-Post
Pre-Post 30
Measurement
Figure 5. Summary of the MEP results. (a) Averaged MEP amplitudes (in mV) across the subjects, the vertical black line
represents the standard deviation across subjects. The MEPs from the measurement 30 min after the intervention (Post 30)
were significantly higher (denoted by *) than those from the measurement before the intervention (Pre). (b) MEP changes
(in percent) from the measurement before the intervention to the measurement immediately after the intervention (Pre-Post)
and 30 min after the intervention (Pre-Post 30).
4. Discussion
It was possible to detect imaginary wrist movements with a low-cost BCI with a true
positive rate of 86 ± 12%, with 1.20 ± 0.57 false detections and 0.63 ± 0.58 false negatives
per minute. The BCI training with the exoskeleton led to increased MEPs after the training
with respect to the pre-intervention measurement. There was a non-significant increase
from pre- to post-intervention measurements of 35 ± 60%, and a significant increase from
pre- to post-30 min intervention measurements of 67 ± 60%.
4.1. Induction of Plasticity
The BCI-triggered exoskeleton movements increased the excitability of the cortical
projections to the forearm extensor muscles. The increase in MEP size was similar to what
has been reported previously for the electrical stimulation of the radial nerve based on an
Sensors 2021, 21, 572
8 of 14
associative BCI, which was approximately 50% compared to baseline MEPs [30]. In addition, the changes in excitability are in a similar range of what has been reported previously
for BCI-triggered electrical stimulation of the common peroneal nerve and exoskeleton
movement of the ankle joint. The BCI intervention in these studies has consistently reported increases in corticospinal excitability in the range of 40–100% [7–11,31]. In this study,
the increase from pre- to post-intervention measurement was not significant, however, it
could be attributed to the large standard deviation of approximately 60%, or to the fact
that the effect of the intervention takes some time to consolidate. Large variability has
commonly been reported in neuromodulation studies where the effect of the intervention
was quantified using MEPs elicited through TMS [7–11,32]. The variability in the MEP size
is affected by multiple factors, such as attention and time of day (reviewed in Refs. [33,34]),
but it may also be due to the variable response of neuromodulation interventions [35–38].
It should also be noted that there exist other types of techniques that have been used for
the induction of neural plasticity and proposed for stroke rehabilitation, and that activate
the cortical brain areas [39]. One of these techniques is repetitive TMS, whereby the cortical
excitability of specific brain areas can be upregulated, which has led to increased amplitudes of motor-evoked potentials [40]. TMS has also been paired with afferent inflow from
the electrical stimulation of a peripheral nerve (paired associative stimulation). This protocol has been used to consistently induce neural plasticity when the correct interstimulus
interval between the magnetic and electrical stimulation has been selected [33]. However,
the use of TMS may not be tolerated well by some stroke patients [41], and there would be
safety precautions to consider [29]. Another way to activate the motor cortex is by the use
of transcranial direct current stimulation, which has been used to increase the excitability
in the motor cortex [42–44]. Moreover, this technique has been used for priming before BCI
training, but there is no clear indication of an additive effect [45,46]. However, it has been
shown that transcranial direct current stimulation can improve the BCI performance [45],
possible through the modulation of the mu event-related desynchronization [47].
The induction of plasticity in this study is expected to be due to the combination of motor imagery and afferent feedback that was temporally correlated. Two control experiments
could have been performed to investigate the effect of motor imagery alone on the MEP and
the effect of passive movement alone on the MEP amplitude. However, the control experiment for motor imagery has been conducted three times, wherein 50 imaginary movements
have been performed, and no change in MEP amplitudes has been reported [7,8,48]. For
the afferent feedback alone (i.e., passive movement in this study), it has been reported
that 50 passive movements do not change the MEP amplitudes [8], and when delivering
afferent feedback through electrical stimulation alone (50 stimuli), no change in MEP
amplitudes has been reported [7,49]. The changes in plasticity in this study are expected
to be mediated through long-term potentiation (LTP)-like changes, as has been suggested
in several previous studies using a similar methodology [8,9,48]. The criteria of LTP-like
plasticity include associativity (pairing between motor imagery and afferent feedback from
the passive movement), rapid onset (indicated by the post-intervention measurement),
and lasting effects (at least 30 min, as indicated by the post-30 min intervention measurement) [50]. There was only a measurement 30 min after the intervention, but the changes
associated with this intervention have been reported to last at least 60 min [51]. It is possible
that the effects last longer, but probably not longer than 24 h. It has been reported that there
was no difference in MEP amplitude between two pre-intervention measurements before
two similar plasticity-inducing protocols, when separated by 24 h [49]. Another criterion
for LTP-like plasticity is specificity, which was not tested in this study, but it has been
reported that this type of intervention is specific [48]. The changes in the neural plasticity
that were observed could happen throughout the nervous system, but it has been suggested
in several studies using the stretch reflex that the changes are supraspinal [7,8,32,48].
Sensors 2021, 21, 572
9 of 14
4.2. Brain-Computer Interface System Performance
The BCI system that was used in this study performed well in terms of the true positive
rate and number of false positive detections per minute. The performance is comparable
with other asynchronous BCI studies that have been used for inducing plasticity, which
have reported true positive rates in the range of 67–85% and a number of false positive
detections per minute in the range of 0.5–2.8 [5,7–11,31]. The approaches to movement
intention detection in those studies have primarily relied on movement-related cortical
potentials, but the results of the current study show that a BCI based on sensorimotor
rhythms is just as effective in terms of movement intention detection, and it has also been
used successfully for BCI training in stroke patients [4,5,52,53]. The BCI performance of
the participants in this study was variable, but there was only a single participant that
experienced low control (true positive rate of 56%). It would be possible to reduce the
detection threshold to allow a higher true positive rate, but that also increases the number
of false-positive detections per minute. However, the number of false-positive detections
could potentially be controlled using a paradigm whereby the BCI only accepts inputs
in predefined periods, instead of always being active, as in the asynchronous paradigm,
or if the number of windows is increased when the detection is exceeded; the latter approach would increase the detection latency. It has been shown previously that the afferent
feedback should coincide with the movement intention (i.e., short detection latency) [48],
but recent findings have suggested that plasticity can be induced with less strict detection latencies [9]. This may allow the use of residual EMG, from which it is possible to
decode multiple movement types [54], which could introduce some task variability in
the training [55], and it may be easier for the stroke patients to control the exoskeleton.
The correlation analysis showed that there was no correlation between the induction of
plasticity in terms of peak-peak amplitudes in the MEP and the performance metrics of
the BCI system. This may suggest that the current level of BCI system performance is
sufficient for inducing plasticity, and that it may not be needed to optimize the system
further from the movement detection point of view, although it should be pointed out that
correlation analyses were performed on a limited sample wherein all subjects (except one)
had good performance. It has been reported previously that the true positive rate, number
of false positive detections and total time of the intervention explain little of the variance
in the peak-peak amplitude of the MEP, with the duration explaining more than the other
two measures [9,11]. However, in a similar BCI study, Niazi et al. reported a statistically
significant correlation of 0.8 between the BCI system’s performance and changes in MEP
amplitude in eight healthy subjects [7]. The BCI system performance in that study was
calculated as the ratio between the true positive rate and false positive detections; when
performing the same calculation in the current study, a similar significant correlation is
observed (correlation coefficient: 0.64; p=0.034). This indicates that there is an incentive to
improve the BCI performance.
4.3. Limitations and Future Perspectives
In this study, it was shown that young healthy participants could control the BCI, and
it could be used for inducing neural plasticity. These findings should be validated in future
studies with the intended end-users, which are stroke survivors with motor impairment
who often are more than 65 years old. The motor cortex excitability decreases with age [34],
but it has been shown that the MEP amplitudes can increase 100% in stroke patients using
an associative BCI protocol [56]. It is likely that the reported BCI system’s performance will
be slightly lower for stroke patients [13,57,58], and if the same experimental protocol is used,
it must be considered that some of the stroke patients may not be able to communicate due
to, e.g., aphasia, and hence will not be able to indicate what they intended to do (i.e., true
positive, false positive or false negative). An alternative approach to verbal feedback to
the experimenter could be to use error-related potentials as a check to indicate if the trial
was a true positive or false positive; however, this approach will also be prone to the
uncertainty related to the decoding algorithm used for these potentials [59]. The lower BCI
Sensors 2021, 21, 572
10 of 14
system performance for stroke patients may cause frustration, and the performance can be
affected by fatigue and inattention. To avoid this, it could be a possibility to implement
the BCI training in a game to make the training more engaging. Using game mechanics, it
would be possible to bias the classifier to improve the performance and conceal it for the
user. The proposed system may be used for motor training in this patient group, but it
is important to note that an increase in MEP size does not equal improvement in motor
function, although increases in MEP size have been reported alongside skill acquisition in
healthy participants [12] and motor recovery in stroke patients [56,60]. The BCI training
may be used as a training intervention in itself, but it could also be possible to utilize the
lasting increase in the MEP size (more than 30 min) in a rehabilitation scenario where the
BCI training is used to prime the nervous system before other types of training, such as
physiotherapy or occupational therapy. For the BCI to be used in rehabilitation clinics
or the home of the patient, the usability should be improved in terms of various aspects,
such as the hardware setup, which should be simplified, and the safety of BCI use in acute
patients should be assessed [61]. The BCI system should be coded on the Arduino to reduce
the amount of hardware, eliminate potential communication problems and delays, which
could allow a faster response of the exoskeleton, and the calibration time should be reduced
or removed using, e.g., a subject-independent movement intention detector [62,63]. As
regards the communication problems, the robustness of the EEG recordings and usability
testing is where the expensive systems potentially differ most from the current cheaper
alternatives (this is just a speculation); these are important factors to consider for the
technology to be adopted in a clinical setting. It should be investigated if the cost of the BCI
(i.e., the EEG amplifier) and exoskeleton can be reduced further. In total, the price for the
BCI system (including EEG amplifier, cap, electrodes and cables) and the exoskeleton (all
parts including motor, control board and Arduino) was less than USD 1000. Additionally,
the design of the exoskeleton should be improved so it will be easier to put on and take off
by the users, and the comfort should be increased as well. This could be done by adapting
the exoskeleton to the individual, which would be possible when the components are
3D-printed. Another option could be to use soft exoskeletons, such as a glove or sleeve,
that can perform the intended movements [64].
In this study, the exoskeleton was used to execute the intended movement and provide
afferent feedback, but it would be possible to use electrical stimulation as well to provide
the afferent feedback. In two recent studies, it has been shown that there is no difference
between the afferent feedback from electrical stimulation and passive movements from an
exoskeleton/robot in terms of the induction of plasticity [11,32]. This gives the patient and
therapist freedom to choose the modality that works best for the patient. Some patients
may not be able to tolerate electrical stimulation well, or may have problems in placing the
stimulation electrodes correctly, so the usability of the exoskeleton may be better compared
to electrical stimulation; however, this has not been tested, and it should be validated in
future studies with the end-users.
To summarize, for this technology to be used in a home setting the BCI and exoskeleton
have to fulfill some requirements, as follows: (1) it must be easy to take on and off the
exoskeleton; (2) the software must be easy to use, since the user may not be used to working
with technology and may suffer from cognitive impairments to some degree; (3) the system
calibration should be done automatically; (4) the hardware setup of the BCI must be simple,
and (5) the patient should be able to place EEG electrodes over the motor cortex. The latter
may be difficult for patients with severe motor impairments; in a recent study it was shown
that half of the stroke participants were able to mount EEG headsets that covered the motor
cortex while the other half was unable to mount the EEG headsets [17]. Those that could
mount the headset spent roughly 10 min. On the contrary, relatives to the patients and
therapists could quickly mount the EEG headset with little prior instruction (5 min). The
setup times for them were between 3 and 5 min. Thus, it would be important to have a
headset that is easy to mount with one hand for the most affected patients, unless they
have someone to help them with the setup.
Sensors 2021, 21, 572
11 of 14
5. Conclusions
In this study, it was shown that the movement intentions associated with imagined
wrist extensions could be detected with good performance (true positive rate: 86 ± 12%;
number of false positive detections per minute: 1.20 ± 0.57) using a cheap amplifier and
open source BCI system. A cheap 3D-printed wrist exoskeleton was developed, wrist
extensions were successfully replicated, and afferent feedback was provided. The BCItriggered exoskeleton movements increased the excitability of the cortical projections to
the extensor muscles in the forearm. These findings may have implications for the transfer
of BCI technology to rehabilitation clinics and home training, by making the technology
affordable to more rehabilitation clinics and patients.
Author Contributions: Conceptualization, all authors; methodology, all authors; software, M.J.;
formal analysis, M.J.; data curation, M.J., T.A.M.J., R.L.K.; writing—original draft preparation, M.J.;
writing—review and editing, all authors; funding acquisition, M.J. All authors have read and agreed
to the published version of the manuscript.
Funding: This research was funded by VELUX FONDEN, grant number 22357 and by the European
Community through the ERASMUS+ student mobility programme.
Institutional Review Board Statement: The study was conducted according to the guidelines of
the Declaration of Helsinki, and approved by The North Denmark Region Committee on Health
Research Ethics (protocol number N-20130081, approved the 15th of January 2019).
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author.
Acknowledgments: The Authors would like to thank Gilliane Johanna Costa for setting up and
testing the BCI system in the laboratory and helping with interfacing the BCI with the exoskeleton.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
Daly, J.J.; Wolpaw, J.R. Brain-computer interfaces in neurological rehabilitation. Lancet Neurol. 2008, 7, 1032–1043.
Grosse-Wentrup, M.; Mattia, D.; Oweiss, K. Using brain-computer interfaces to induce neural plasticity and restore function.
J. Neural Eng. 2011, 8, 025004.
Ramos-Murguialday, A.; Broetz, D.; Rea, M.; Läer, L.; Yilmaz, Ö.; Brasil, F.; Liberati, G.; Curado, M.; Garcia-Cossio, E.; Vyziotis, A.
Brain–machine interface in chronic stroke rehabilitation: A controlled study. Ann. Neurol. 2013, 74, 100–108.
Frolov, A.A.; Mokienko, O.; Lyukmanov, R.; Biryukova, E.; Kotov, S.; Turbina, L.; Nadareyshvily, G.; Bushkova, Y. Post-stroke
rehabilitation training with a motor-imagery-based brain-computer interface (BCI)-controlled hand exoskeleton: A randomized
controlled multicenter trial. Front. Neurosci. 2017, 11. [CrossRef]
Biasiucci, A.; Leeb, R.; Iturrate, I.; Perdikis, S.; Al-Khodairy, A.; Corbet, T.; Schnider, A.; Schmidlin, T.; Zhang, H.; Bassolino, M.
Brain-actuated functional electrical stimulation elicits lasting arm motor recovery after stroke. Nat. Commun. 2018, 9, 2421.
Cervera, M.A.; Soekadar, S.R.; Ushiba, J.; Millán, J.d.R.; Liu, M.; Birbaumer, N.; Garipelli, G. Brain-computer interfaces for
post-stroke motor rehabilitation: A meta-analysis. Ann. Clin. Transl. Neurol. 2018, 5, 651–663. [PubMed]
Niazi, I.K.; Kersting, N.M.; Jiang, N.; Dremstrup, K.; Farina, D. Peripheral electrical stimulation triggered by self-paced detection
of motor intention enhances motor evoked potentials. IEEE Trans. Neural Syst. Rehabil. Eng. 2012, 20, 595–604. [PubMed]
Xu, R.; Jiang, N.; Mrachacz-Kersting, N.; Lin, C.; Asin, G.; Moreno, J.; Pons, J.; Dremstrup, K.; Farina, D. A closed-loop braincomputer interface triggering an active ankle-foot orthosis for inducing cortical neural plasticity. IEEE Trans. Biomed. Eng. 2014,
20, 2092–2101.
Jochumsen, M.; Navid, M.S.; Rashid, U.; Haavik, H.; Niazi, I.K. EMG-versus EEG-triggered electrical stimulation for inducing
corticospinal plasticity. IEEE Trans. Neural Syst. Rehabil. Eng. 2019, 27, 1901–1908.
Jochumsen, M.; Navid, M.S.; Nedergaard, R.W.; Signal, N.; Rashid, U.; Hassan, A.; Haavik, H.; Taylor, D.; Niazi, I.K. Self-paced
online vs. cue-based offline brain—Computer interfaces for inducing neural plasticity. Brain Sci. 2019, 9, 127.
Jochumsen, M.; Cremoux, S.; Robinault, L.; Lauber, J.; Arceo, J.; Navid, M.; Nedergaard, R.; Rashid, U.; Haavik, H.; Niazi, I.
Investigation of optimal afferent feedback modality for inducing neural plasticity with a self-paced brain-computer interface.
Sensors 2018, 18, 3761. [CrossRef] [PubMed]
Pascual-Leone, A.; Dang, N.; Cohen, L.G.; Brasil-Neto, J.P.; Cammarota, A.; Hallett, M. Modulation of muscle responses evoked by
transcranial magnetic stimulation during the acquisition of new fine motor skills. J. Neurophysiol. 1995, 74, 1037–1045. [CrossRef]
[PubMed]
Sensors 2021, 21, 572
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
12 of 14
Jochumsen, M.; Niazi, I.K.; Mrachacz-Kersting, N.; Jiang, N.; Farina, D.; Dremstrup, K. Comparison of spatial filters and features for
the detection and classification of movement-related cortical potentials in healthy individuals and stroke patients. J. Neural Eng.
2015, 12, 056003. [CrossRef] [PubMed]
Müller-Putz, G.R.; Kaiser, V.; Solis-Escalante, T.; Pfurtscheller, G. Fast set-up asynchronous brain-switch based on detection of
foot motor imagery in 1-channel EEG. Med. Biol. Eng. Comput. 2010, 48, 229–233. [CrossRef] [PubMed]
Ibáñez, J.; Serrano, J.I.; del Castillo, M.D.; Monge-Pereira, E.; Molina-Rueda, F.; Alguacil-Diego, I.; Pons, J.L. Detection of the onset
of upper-limb movements based on the combined analysis of changes in the sensorimotor rhythms and slow cortical potentials.
J. Neural Eng. 2014, 11, 056009. [CrossRef]
Leeb, R.; Perdikis, S.; Tonin, L.; Biasiucci, A.; Tavella, M.; Creatura, M.; Molina, A.; Al-Khodairy, A.; Carlson, T.; Millán, J.D.R. Transferring brain-computer interfaces beyond the laboratory: Successful application control for motor-disabled users. Artif. Intell. Med.
2013, 59, 121–132. [CrossRef]
Jochumsen, M.; Knoche, H.; Kidmose, P.; Kjær, T.W.; Dinesen, B.I. Evaluation of EEG headset mounting for brain-computer
interface-based stroke rehabilitation by patients, therapists, and relatives. Front. Human Neurosci. 2020, 14. [CrossRef]
Hairston, W.D.; Whitaker, K.W.; Ries, A.J.; Vettel, J.M.; Bradford, J.C.; Kerick, S.E.; McDowell, K. Usability of four commerciallyoriented EEG systems. J. Neural Eng. 2014, 11, 046018. [CrossRef]
Jochumsen, M.; Knoche, H.; Kjaer, T.W.; Dinesen, B.; Kidmose, P. EEG headset evaluation for detection of single-trial movement
intention for brain-computer interfaces. Sensors 2020, 20, 2804. [CrossRef]
Renard, Y.; Lotte, F.; Gibert, G.; Congedo, M.; Maby, E.; Delannoy, V.; Bertrand, O.; Lécuyer, A. Openvibe: An open-source
software platform to design, test, and use brain-computer interfaces in real and virtual environments. Presence Teleoperators Virtual
Environ. 2010, 19, 35–53. [CrossRef]
McCrimmon, C.M.; Fu, J.L.; Wang, M.; Lopes, L.S.; Wang, P.T.; Karimi-Bidhendi, A.; Liu, C.Y.; Heydari, P.; Nenadic, Z.; Do, A.H.
Performance assessment of a custom, portable, and low-cost brain-computer interface platform. IEEE Trans. Biomed. Eng. 2017,
64, 2313–2320. [CrossRef] [PubMed]
Rashid, U.; Niazi, I.; Signal, N.; Taylor, D. An EEG experimental study evaluating the performance of texas instruments ADS1299.
Sensors 2018, 18, 3721. [CrossRef] [PubMed]
Peterson, V.; Galván, C.; Hernández, H.; Spies, R. A feasibility study of a complete low-cost consumer-grade brain-computer
interface system. Heliyon 2020, 6, e03425. [CrossRef] [PubMed]
Pichiorri, F.; Morone, G.; Petti, M.; Toppi, J.; Pisotta, I.; Molinari, M.; Paolucci, S.; Inghilleri, M.; Astolfi, L.; Cincotti, F. Braincomputer interface boosts motor imagery practice during stroke recovery. Ann. Neurol. 2015, 77, 851–865. [CrossRef]
Dudley, D.R.; Knarr, B.A.; Siu, K.; Peck, J.; Ricks, B.; Zuniga, J.M. Testing of a 3D printed hand exoskeleton for an individual with
stroke: A case study. Disabil. Rehabil. Assist. Technol. 2019, 1–5. [CrossRef]
Yoo, H.; Lee, S.; Kim, J.; Park, C.; Lee, B. Development of 3D-printed myoelectric hand orthosis for patients with spinal cord
injury. J. Neuroeng. Rehabil. 2019, 16, 162. [CrossRef]
Ang, K.K.; Guan, C.; Phua, K.S.; Wang, C.; Zhou, L.; Tang, K.Y.; Ephraim Joseph, G.J.; Kuah, C.W.K.; Chua, K.S.G. Brain-computer
interface-based robotic end effector system for wrist and hand rehabilitation: Results of a three-armed randomized controlled
trial for chronic stroke. Front. Neuroeng. 2014, 7, 30. [CrossRef]
Sebastián-Romagosa, M.; Cho, W.; Ortner, R.; Murovec, N.; von Oertzen, T.; Kamada, K.; Allison, B.Z.; Guger, C. Brain computer
interface treatment for motor rehabilitation of upper extremity of stroke patients—A feasibility study. Front. Neurosci. 2020, 14.
[CrossRef]
Rossi, S.; Hallett, M.; Rossini, P.M.; Pascual-Leone, A. Safety, ethical considerations, and application guidelines for the use of
transcranial magnetic stimulation in clinical practice and research. Clin. Neurophysiol. 2009, 120, 2008–2039. [CrossRef]
Mrachacz-Kersting, N.; Dosen, S.; Aliakbaryhosseinabadi, S.; Pereira, E.M.; Stevenson, A.J.T.; Jiang, N.; Farina, D. Brain-state
dependent peripheral nerve stimulation for plasticity induction targeting upper-limb. In Proceedings of the International
Conference on Neurorehabilitation, Pisa, Italy, 16–20 October 2018; pp. 1061–1065.
Mrachacz-Kersting, N.; Aliakbaryhosseinabadi, S. Comparison of the efficacy of a real-time and offline associative brain-computerinterface. Front. Neurosci. 2018, 12, 455. [CrossRef]
Mrachacz-Kersting, N.; Voigt, M.; Stevenson, A.J.T.; Aliakbaryhosseinabadi, S.; Jiang, N.; Dremstrup, K.; Farina, D. The effect
of type of afferent feedback timed with motor imagery on the induction of cortical plasticity. Brain Res. 2017, 1674, 91–100.
[CrossRef] [PubMed]
Alder, G.; Signal, N.; Olsen, S.; Taylor, D. A systematic review of paired associative stimulation (PAS) to modulate lower limb
corticomotor excitability: Implications for stimulation parameter selection and experimental design. Front. Neurosci. 2019, 13, 895.
[PubMed]
Ridding, M.C.; Ziemann, U. Determinants of the induction of cortical plasticity by non-invasive brain stimulation in healthy
subjects. J. Physiol. 2010, 588, 2291–2304. [PubMed]
Wiethoff, S.; Hamada, M.; Rothwell, J.C. Variability in response to transcranial direct current stimulation of the motor cortex.
Brain Stimul. 2014, 7, 468–475. [CrossRef]
Hamada, M.; Murase, N.; Hasan, A.; Balaratnam, M.; Rothwell, J.C. The role of interneuron networks in driving human motor
cortical plasticity. Cereb. Cortex 2013, 23, 1593–1605. [CrossRef]
Sensors 2021, 21, 572
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
13 of 14
López-Alonso, V.; Cheeran, B.; Río-Rodríguez, D.; Fernández-del-Olmo, M. Inter-individual variability in response to non-invasive
brain stimulation paradigms. Brain Stimul. 2014, 7, 372–380. [CrossRef]
Lahr, J.; Paßmann, S.; List, J.; Vach, W.; Flöel, A.; Klöppel, S. Effects of different analysis strategies on paired associative stimulation.
A pooled data analysis from three research labs. PLoS ONE 2016, 11, e0154880. [CrossRef]
Ziemann, U.; Paulus, W.; Nitsche, M.A.; Pascual-Leone, A.; Byblow, W.D.; Berardelli, A.; Siebner, H.R.; Classen, J.; Cohen, L.G.;
Rothwell, J.C. Consensus: Motor cortex plasticity protocols. Brain Stimul. 2008, 1, 164–182. [CrossRef]
Tung, Y.; Lai, C.; Liao, C.; Huang, S.; Liou, T.; Chen, H. Repetitive transcranial magnetic stimulation of lower limb motor function
in patients with stroke: A systematic review and meta-analysis of randomized controlled trials. Clin. Rehabil. 2019, 33, 1102–1112.
[CrossRef]
Heald, A.; Bates, D.; Cartlidge, N.; French, J.M.; Miller, S. Longitudinal study of central motor conduction time following stroke:
2. Central motor conduction measured within 72 H after stroke as a predictor of functional outcome at 12 months. Brain 1993,
116, 1371–1385. [CrossRef]
Bastani, A.; Jaberzadeh, S. Does anodal transcranial direct current stimulation enhance excitability of the motor cortex and motor
function in healthy individuals and subjects with stroke: A systematic review and meta-analysis. Clin. Neurophysiol. 2012, 123,
644–657. [CrossRef] [PubMed]
Edwards, D.J.; Krebs, H.I.; Rykman, A.; Zipse, J.; Thickbroom, G.W.; Mastaglia, F.L.; Pascual-Leone, A.; Volpe, B.T. Raised
corticomotor excitability of M1 forearm area following anodal tDCS is sustained during robotic wrist therapy in chronic stroke.
Restor. Neurol. Neurosci. 2009, 27, 199–207. [CrossRef] [PubMed]
Hummel, F.; Celnik, P.; Giraux, P.; Floel, A.; Wu, W.; Gerloff, C.; Cohen, L.G. Effects of non-invasive cortical stimulation on skilled
motor function in chronic stroke. Brain 2005, 128, 490–499. [CrossRef] [PubMed]
Ang, K.K.; Guan, C.; Phua, K.S.; Wang, C.; Zhao, L.; Teo, W.P.; Chen, C.; Ng, Y.S.; Chew, E. Facilitating effects of transcranial direct current stimulation on motor imagery brain-computer interface with robotic feedback for stroke rehabilitation.
Arch. Phys. Med. Rehabil. 2015, 96, S79–S87. [CrossRef]
Chew, E.; Teo, W.; Tang, N.; Ang, K.K.; Ng, Y.S.; Zhou, J.H.; Teh, I.; Phua, K.S.; Zhao, L.; Guan, C. Using transcranial direct current
stimulation to augment the effect of motor imagery-assisted brain-computer interface training in chronic stroke patients—Cortical
reorganization considerations. Front. Neurol. 2020, 11. [CrossRef]
Kasashima, Y.; Fujiwara, T.; Matsushika, Y.; Tsuji, T.; Hase, K.; Ushiyama, J.; Ushiba, J.; Liu, M. Modulation of event-related
desynchronization during motor imagery with transcranial direct current stimulation (tDCS) in patients with chronic hemiparetic
stroke. Exp. Brain Res. 2012, 221, 263–268. [CrossRef]
Mrachacz-Kersting, N.; Kristensen, S.R.; Niazi, I.K.; Farina, D. Precise temporal association between cortical potentials evoked by
motor imagination and afference induces cortical plasticity. J. Physiol. 2012, 590, 1669–1682. [CrossRef]
Jochumsen, M.; Niazi, I.K.; Signal, N.; Nedergaard, R.W.; Holt, K.; Haavik, H.; Taylor, D. Pairing voluntary movement and
muscle-located electrical stimulation increases cortical excitability. Front. Human Neurosci. 2016, 10, 482. [CrossRef]
Nitsche, M.A.; Roth, A.; Kuo, M.; Fischer, A.K.; Liebetanz, D.; Lang, N.; Tergau, F.; Paulus, W. Timing-dependent modulation of
associative plasticity by general network excitability in the human motor cortex. J. Neurosci. 2007, 27, 3807–3812. [CrossRef]
Olsen, S.; Signal, N.; Niazi, I.K.; Christensen, T.; Jochumsen, M.; Taylor, D. Paired associative stimulation delivered by pairing
movement-related cortical potentials with peripheral electrical stimulation: An investigation of the duration of neuromodulatory
effects. Neuromodulation Technol. Neural Interface 2017, 21, 362–367. [CrossRef]
Ang, K.K.; Chua, K.S.G.; Phua, K.S.; Wang, C.; Chin, Z.Y.; Kuah, C.W.K.; Low, W.; Guan, C. A randomized controlled trial of
EEG-based motor imagery brain-computer interface robotic rehabilitation for stroke. Clin. EEG Neurosci. 2015, 46, 310–320.
[CrossRef] [PubMed]
Kawakami, M.; Fujiwara, T.; Ushiba, J.; Nishimoto, A.; Abe, K.; Honaga, K.; Nishimura, A.; Mizuno, K.; Kodama, M.; Masakado,
Y. A new therapeutic application of brain-machine interface (BMI) training followed by hybrid assistive neuromuscular dynamic
stimulation (HANDS) therapy for patients with severe hemiparetic stroke: A proof of concept study. Restor. Neurol. Neurosci.
2016, 34, 789–797. [CrossRef] [PubMed]
Jochumsen, M.; Niazi, I.K.; Zia ur Rehman, M.; Amjad, I.; Shafique, M.; Gilani, S.O.; Waris, A. Decoding attempted hand
movements in stroke patients using surface electromyography. Sensors 2020, 20, 6763. [CrossRef] [PubMed]
Krakauer, J.W. Motor learning: Its relevance to stroke recovery and neurorehabilitation. Curr. Opin. Neurol. 2006, 19, 84–90.
[CrossRef]
Mrachacz-Kersting, M.; Jiang, N.; Stevenson, A.J.T.; Niazi, I.K.; Kostic, V.; Pavlovic, A.; Radovanović, S.; Djuric-Jovicic, M.;
Agosta, F.; Dremstrup, K.; et al. Efficient neuroplasticity induction in chronic stroke patients by an associative brain-computer
interface. J. Neurophysiol. 2016, 115, 1410–1421. [PubMed]
Jochumsen, M.; Niazi, I.K.; Navid, M.S.; Anwar, M.N.; Farina, D.; Dremstrup, K. Online multi-class brain-computer interface for
detection and classification of lower limb movement intentions and kinetics for stroke rehabilitation. Brain Comput. Interfaces
2015, 2, 202–210.
Jochumsen, M.; Niazi, I.K.; Taylor, D.; Farina, D.; Dremstrup, K. Detecting and classifying movement-related cortical potentials
associated with hand movements in healthy subjects and stroke patients from single-electrode, single-trial EEG. J. Neural Eng.
2015, 12, 056013.
Sensors 2021, 21, 572
59.
60.
61.
62.
63.
64.
14 of 14
Usama, N.; Leerskov, K.K.; Niazi, I.K.; Dremstrup, K.; Jochumsen, M. Classification of error-related potentials from single-trial
EEG in association with executed and imagined movements: A feature and classifier investigation. Med. Biol. Eng. Comput. 2020,
58, 2699–2710.
Mrachacz-Kersting, N.; Stevenson, A.J.; Jørgensen, H.R.; Severinsen, K.E.; Aliakbaryhosseinabadi, S.; Jiang, N.; Farina, D. Brain
state–dependent stimulation boosts functional recovery following stroke. Ann. Neurol. 2019, 85, 84–95. [CrossRef]
Hashimoto, Y.; Kakui, T.; Ushiba, J.; Liu, M.; Kamada, K.; Ota, T. Portable rehabilitation system with brain-computer interface for
inpatients with acute and subacute stroke: A feasibility study. Assist. Technol. 2020. [CrossRef]
Jochumsen, M.; Niazi, I.K.; Nedergaard, R.W.; Navid, M.S.; Dremstrup, K. Effect of subject training on a movement-related
cortical potential-based brain-computer interface. Biomed. Signal Process. Control 2018, 41, 63–68.
Niazi, I.K.; Jiang, N.; Jochumsen, M.; Nielsen, J.F.; Dremstrup, K.; Farina, D. Detection of movement-related cortical potentials
based on subject-independent training. Med. Biol. Eng. Comput. 2013, 51, 507–512. [PubMed]
Cheng, N.; Phua, K.S.; Lai, H.S.; Tam, P.K.; Tang, K.Y.; Cheng, K.K.; Yeow, R.C.; Ang, K.K.; Guan, C.; Lim, J.H. Brain-computer
interface-based soft robotic glove rehabilitation for stroke. IEEE Trans. Biomed. Eng. 2020, 67, 3339–3351. [PubMed]