ORIGINAL ARTICLE

Brain–Machine Interface in Chronic

Stroke Rehabilitation: A Controlled Study
Ander Ramos-Murguialday, PhD,1,2 Doris Broetz, PT,1 Massimiliano Rea, PhD,1
Leonhard L€ €
aer, MD,1 Ozge Yilmaz, MSc,1 Fabricio L. Brasil, MSc,1
Giulia Liberati, PhD,1 Marco R. Curado, MSc,1 Eliana Garcia-Cossio, MSc,1
Alexandros Vyziotis, MD,1 Woosang Cho, MSc,1 Manuel Agostini, MSc,1
Ernesto Soares, PhD,1 Surjo Soekadar, MD,3,4 Andrea Caria, PhD,1
Leonardo G. Cohen, MD,3 and Niels Birbaumer, PhD1,5

Objective: Chronic stroke patients with severe hand weakness respond poorly to rehabilitation efforts. Here, we eval-
uated efficacy of daily brain–machine interface (BMI) training to increase the hypothesized beneficial effects of physi-
otherapy alone in patients with severe paresis in a double-blind sham-controlled design proof of concept study.
Methods: Thirty-two chronic stroke patients with severe hand weakness were randomly assigned to 2 matched
groups and participated in 17.8 6 1.4 days of training rewarding desynchronization of ipsilesional oscillatory sensori-
motor rhythms with contingent online movements of hand and arm orthoses (experimental group, n 5 16). In the
control group (sham group, n 5 16), movements of the orthoses occurred randomly. Both groups received identical
behavioral physiotherapy immediately following BMI training or the control intervention. Upper limb motor function
scores, electromyography from arm and hand muscles, placebo–expectancy effects, and functional magnetic reso-
nance imaging (fMRI) blood oxygenation level–dependent activity were assessed before and after intervention.
Results: A significant group 3 time interaction in upper limb (combined hand and modified arm) Fugl–Meyer assess-
ment (cFMA) motor scores was found. cFMA scores improved more in the experimental than in the control group,
presenting a significant improvement of cFMA scores (3.41 6 0.563-point difference, p 5 0.018) reflecting a clinically
meaningful change from no activity to some in paretic muscles. cFMA improvements in the experimental group cor-
related with changes in fMRI laterality index and with paretic hand electromyography activity. Placebo–expectancy
scores were comparable for both groups.
Interpretation: The addition of BMI training to behaviorally oriented physiotherapy can be used to induce functional
improvements in motor function in chronic stroke patients without residual finger movements and may open a new
door in stroke neurorehabilitation.
ANN NEUROL 2013;74:100–108

P aralysis after stroke or neurotrauma is among the

leading causes of long-term disability in adults. Up
to 30% of all stroke survivors experience very limited
evidence that techniques like constraint-induced move-
ment therapy (CIMT) or bilateral arm training represent
useful strategies to improve motor function in chronic
motor recovery and depend on assistance to manage their stroke patients,3–5 but such options are not applicable for
daily living activities.1,2 Recent studies have provided stroke patients with severe limb weakness, because

View this article online at wileyonlinelibrary.com. DOI: 10.1002/ana.23879

Received Nov 4, 2012, and in revised form Feb 12, 2013. Accepted for publication Mar 1, 2013.

Address correspondence to Dr Ramos-Murguialday or Dr Birbaumer, Institute of Medical Psychology and Behavioral Neurobiology and MEG Center,
University of Tubingen, Garten str 29, 72074 T€ubingen, Germany. E-mail: ander.ramos@med.uni-tuebingen.de or
E-mail: niels.birbaumer@uni-tuebingen.de

From the 1Institute of Medical Psychology and Behavioral Neurobiology and Magnetoencephalography Center, University of T€ ubingen, T€ ubingen, Ger-
many; 2Health Technologies Department, Tecnalia, San Sebastian, Spain; 3Neurorehabilitation Section, Human Cortical Physiology and Stroke, National
4
Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD; Department of Psychiatry and Psychotherapy, University
Hospital T€ ubingen, Germany; 5San Camillo Hospital, Institute for Hospitalization and Scientific Care, Venice Lido, Italy.
ubingen, T€

Additional Supporting Information may be found in the online version of this article.

100 V
C 2013 American Neurological Association
 Ramos-Murguialday et al: BMI in Chronic Stroke

residual active movement is necessary for CIMT.6 For physiotherapy allows generalization of relearned motor
this patient population, brain–machine interface (BMI) skills to meaningful real life activities.
may play a crucial role.
However, severely weakened stroke patients are still Patients and Methods
able to imagine movements of the paretic hand and can In this study, 2 patient groups underwent physiotherapy follow-
attempt to move even in the absence of actual move- ing BMI or sham-BMI training sessions. Whereas the control
ments.7–11 These imagery and intent-to-move strategies group received BMI training in which online reaching and
have been reported to be useful in patients with mild to grasping movements of the orthosis occurred randomly, orthosis
moderate motor deficits.12 In line with this previous in- movements in the experimental group were contingent on
desynchronization of ipsilesional SMR brain oscillations.
formation, it was proposed that BMI systems allowing
online classification of neuroelectric or metabolic brain Study Design
activity (eg, that associated with planning and intended This study involved 32 chronic stroke patients, with combined
execution of grasping movements) and its translation hand and arm scores (motor part) from the modified upper
into control of external devices such as orthoses driving limb cFMA of 12.15 6 8.8 (maximal score is 54 points; Sup-
motions of an extremely weakened hand/arm might have plementary Information, Section 2.1), unable to extend their
a beneficial role in neurorehabilitation.13–15 fingers. The study was conducted at the University of T€ ubingen
Previous studies showed that learning to control in Germany. Informed consent was obtained from all patients
desynchronization of ipsilesional sensorimotor rolandic involved. The study was approved by the ethics committee of
(sensorimotor rhythm [SMR]) brain oscillations after the Faculty of Medicine of the University of T€ ubingen. In the
stroke can be translated into grasping movements of an experimental group patients’ successful control of ipsilesional
SMR brain oscillatory activity was translated concurrently into
orthosis attached to the paralyzed limb.11,16 Furthermore,
movement of the orthosis attached to the paralyzed limb,
simultaneous contingent association between brain oscil-
whereas in the control group patients’ movements of the ortho-
lations and grasping movements of an orthosis has been sis occurred randomly, unrelated to SMR control. Thus,
proven to elicit motor learning in healthy participants.16 hypothesized group effects on motor function would reflect the
In accordance with basic animal and human single-cell contribution of learning to control SMR oscillatory brain activ-
experimental evidence,17–22 we hypothesized superior ity immediately preceding physiotherapy. Both groups received
associative learning in severely brain damaged stroke continuous assessments of subjective expectancies of treatment
patients if a close contingent connection between the success and credibility of differential placebo effects.
neural correlate of an intention to move and the conse-
quent feedback of the movement (visual and propriocep- Patients
tive) is established via a BMI. The extent to which this Patients were recruited via public information (stroke associations,
rehabilitation centers, hospitals) all over Germany from December
approach is useful adjuvant to behavioral physiotherapy
2007 to March 2011. A total of 504 were assessed as potentially
or the generalization of improvements in control of brain
eligible and were contacted; 32 were allocated to intervention.
oscillatory activity to clinically meaningful improvements
Exclusion criteria, number of excluded patients, and rea-
in motor function has not been tested. Our proof of sons for exclusion are described in Supplementary Information,
concept controlled randomized double-blind study tested Section 1.1.
this hypothesis in chronic stroke patients without residual All participants fulfilled the following criteria: (1) paralysis
finger movements, comparing improvements in motor of 1 hand with no active finger extension; (2) time since stroke of
function between an experimental group receiving BMI at least 10 months; (3) age between 18 and 80 years; (4) no psy-
training adjuvant to behaviorally oriented physiotherapy chiatric or neurological condition other than stroke; (5) no cere-
and a control group receiving sham-BMI adjuvant to bellar lesion or bilateral motor deficit; (6) no pregnancy; (7) no
behaviorally oriented physiotherapy, comparing the claustrophobia; (8) no epilepsy or medication for epilepsy during
improvement in combined hand and arm scores (motor the past 6 months; (9) eligibility to undergo magnetic resonance
part) from the modified upper limb combined Fugl– imaging (MRI); and (10) ability to understand and follow
instructions. Summaries of patient group demographic and func-
Meyer assessment (cFMA; excluding coordination, speed
tional data and individual lesion localization are presented in the
and reflexes scores). Furthermore, we tested whether
Table 1 and Supplementary Information, Section 1.2, respec-
BMI training immediately preceding the relevant period
tively. Patients were randomly assigned to the experimental or the
of physiotherapy could prime the effects of our rehabili- control group. An investigator blind to the study design assigned
tation treatment as was demonstrated in healthy partici- patients in a pairwise fashion. Groups were matched for age, gen-
pants,16 that is, we speculated that learning to control der, paretic side, and motor impairment scores (cFMA) at the
oscillatory brain activity through this BMI approach con- time of inclusion, each of them being assigned with a different
stitutes the necessary therapeutic ingredient and that weight from 1 to 4, respectively. Once the matching was

July 2013 101

ANNALS of Neurology

TABLE 1. Means and Standard Deviations of Demographic Data and Functional Scores for the 2 Patient
Groups at the Time of Enrollment in the Study

Group Gender Age, yr Months Lesion Side cFMA Scores GAS Training
Since Stroke Duration, Runs
C1 9 M/7 F 49.3 6 12.5 66 6 45 8 R/8 L 11.15 6 6.92 0.88 6 0.67 275 6 25
S 9 M/5 F 50.3 6 12.2 71 6 72 8 R/6 L 13.28 6 10.71 0.63 6 0.51 291 6 17
In the experimental group (C1), brain activity moved the orthoses; the control group (S) received random orthosis movements
not linked to control of oscillatory brain activity. Lesion side indicates damaged hemisphere. Motor part of the modified upper
limb cFMA (hand and arm parts combined having a maximum score of 54 points), that is, the primary outcome measure, and
GAS total scores are presented for both groups. Training duration indicates the amount of runs during the training. One run con-
tains 11 trials of 5 seconds in which the patients were able to move the orthosis using the brain–machine interface system. None
of the differences of baseline measures between the experimental and control groups was significant (see Supplementary Informa-
tion, Section 3).
F 5 female; GAS 5 goal attainment scale; L 5 left; M 5 male; R 5 right.

performed, the MATLAB (MathWorks, Natick, MA) random

function was used to randomly assign 1 patient of each pair of
patients to 1 of the 2 groups with a 50% probability. Group
assignment was blinded for all participants and for the scientific–
clinical personnel; none of the patients or therapists was able to
identify group assignment reflected in the results of placebo and
motor function scales below. None of the patients could elicit
active finger or wrist extension. The mean 6 standard deviation
(SD) scores for wrist stability at 15 extension, and elbow at 90
and at 0 , were 0.27 6 0.64 and 0.30 6 0.65 respectively. Only 8
and 7 patients presented scores different from zero before inter-
vention from the experimental and control group, respectively.
Two patients in the control group were excluded due to equip-
ment malfunction during BMI training (n 5 1) and faking func-
tional deficit during baseline measurements to be included in the
study (n 5 1).

Assessment
A comprehensive battery of assessment instruments was given FIGURE 1: Brain–machine interface (BMI) in stroke. (A) Exper-
twice before (8 weeks and 1 day before the first training ses- imental time course of the online BMI for paralyzed chronic
stroke patients’ rehabilitation. (B) User wearing the 16-chan-
sion) and once immediately after treatment (Fig 1A).
nel electroencephalographic (EEG) system with the hand
PRIMARY BEHAVIORAL OUTCOME MEASURES. We attached to the orthosis to drive finger-extending (hand
opening) motions muscles as indicated by the illustration dur-
used the combined hand and arm scores (motor part) from the
ing the second part of the BMI training. The sensorimotor
modified upper limb cFMA (see Supplementary Information, rhythm (SMR) power recorded from the ipsilesional electro-
Section 2.1; maximal score 5 54 points) as primary behavioral des (gray line) is translated into movement of the orthosis. A
outcome measures.23 We excluded upper limb cFMA scores threshold (dashed line) calculated as the point of equal dis-
related to (1) coordination and speed and (2) reflexes because: tance to the mean of the power distribution during rest (red
line) and motor intention (blue line) calculated over the last
(1) patients in this study could not touch their noses with the
15 seconds defines rest (red shading) and motor intention
index finger fully extended and had no remaining finger exten- (blue shading) classification areas. If the SMR power is contin-
sion (inclusion criteria) and (2) reflex scores add uncertainty to uously in the motor intention classification area (blue shad-
the measurement.24 We used these scores as primary outcome ing) for 200 milliseconds the orthosis moves; the orthosis
measure because they are related to the 2 body parts trained stops if it returns to the rest classification area (red shading)
for 200 milliseconds, and maintains the previous state other-
during the BMI (hand and arm) and reflect motor recovery
wise. The same BMI principle was applied when training
and measures of motor aspects that may limit but are not reaching movements with the arm orthosis (Supplementary
related to task accomplishment (eg, joint motion). Fig 6). Finger extension and flexion when using hand orthosis
(grasping) and upper arm extension and flexion when using
SECONDARY BEHAVIORAL OUTCOME MEASURES. arm orthosis (reaching) were part of the training task while
Secondary behavioral outcome measures included the Ashworth the wrist was immobilized and fixed to the orthoses.

102 Volume 74, No. 1

 Ramos-Murguialday et al: BMI in Chronic Stroke

Scale, the Motor Activity Log (MAL),25 and a goal attainment vs control) as between factor and session (before vs after) as
scale (GAS).26 More information about assessment instruments within factor was performed on LI values. Subsequently, sepa-
can be found in Supplementary Information, Section 2. rate paired-samples t tests were carried out as post hoc analy-
Two expectancy–placebo questionnaires were collected ses to compare the dependent variables in the before and after
from each patient: (1) after each fifth treatment session and (2) sessions for each group.
at the end of treatment. The first questionnaire contained 15
questions (scale 5 1–6) concerning professional behavior of the Interventions
therapists, mood, and expectations of improvement. The second Intervention involved daily training for 4 weeks (excluding
contained 12 questions (scale 5 1–6) concerning comfortable weekends), and there was no difference in time of training
and proper functioning of the BMI–orthosis system. Examples (BMI 1 physiotherapy) between groups.
can be found in Supplementary Information, Section 4).

ASSESSMENTS ASSOCIATED WITH THE PRIMARY BMI TRAINING. During BMI training, patients were
BEHAVIORAL OUTCOME MEASURE. We employed instructed to desynchronize SMR rhythms measured by electro-
electromyography (EMG) to document muscle activity and encephalography (EEG) electrodes overlying the ipsilesional
muscle innervation,27 and blood oxygenation level–dependent motor cortex by intending to move their severely impaired
signal functional MRI (fMRI) to identify possible changes in upper limb. Successful SMR control resulted in concurrent
brain function with the interventions.28 movements of the arm and hand orthoses in the experimental
group only, whereas in the control group patients received
EMG. We recorded EMG during attempted performance of sham feedback, which means random movements of the robotic
arm movements (opening and closing the hand and arm exten- orthoses not linked to the patient’s ipsilesional SMR oscillations
sion) to quantify the patients’ ability to generate EMG activity (Supplementary Video 1). The training using the arm orthosis
as a function of time and intervention. The EMG data were targeted the patient’s ability to move the upper arm and reach
preprocessed, and the cumulative amplitude changes for the rel- forward. Upon hearing the corresponding auditory cue, the
evant frequency bins of the signal were extracted serving as a patient was instructed to try to reach (even if the arm does not
measure of muscle control. This was quantified by calculating follow their intention), grasp, and bring an imaginary apple to
the waveform length, providing indicators for EMG signal their lap, thus involving finger extension during the reach and
amplitude and frequency (see Supplementary Information, grasp movement. This movement was chosen because of its
Section 6.1). functional value and following the findings of Tyc and Boyad-
jian33 indicating that proximal (upper arm) training induces
fMRI. Inside the scanner, patients were asked to perform 3
distal (hand) recovery but distal training does not produce
different tasks: (1) to perform (try to perform) hand closing
proximal recovery unless it uses coordination movements,
and opening, (2) to imagine hand closing and opening, and
implying distal and proximal joint control.9 Concurrently, the
(3) to remain motionless; all conditions were cued by audi-
reach and grasp attempt supposedly generates brain activity
tory–visual signals every 1.5 seconds. A lateralization index
assisting BMI intention detection and influencing not only
(LI) was calculated to assess changes in cortical lateralization
proximal but also distal muscles. The training using the hand
between session before and after BMI and physiotherapy inter-
orthosis targeted the patient’s ability to open and close the
vention.29,30 In healthy subjects, cortical activity is lateralized
hand.
to sensorimotor areas contralateral to the moving hand.28 Ac-
None of the patients in the control or experimental
tivity associated with affected hand motions in well-recovered
groups reported any perception of inconsistency during train-
stroke patients resemble patterns identified in healthy individ-
ing. Patients were instructed to avoid blinking, coughing, chew-
uals, being mainly contralateral during movement and move-
ing, head movement, and body compensation movements.
ment preparation.28,31 The LI, computed as the normalized
They were told that these actions could affect the training. By
difference between the number of all active voxels in the ipsi-
asking the patients to produce these artifacts before training,
lesional and contralesional areas (anatomically defined regions
the credibility of the measurement was enhanced for both
of interest conforming to Montreal Neurologic Institute
groups; the placebo questionnaires showed no differences in
space), was assessed separately for motor and premotor corti-
perception of the BMI system in both groups. After calibration
ces, and for somatosensory cortex for the paretic and healthy
(see Supplementary Information, Section 5.1), the BMI training
hand in the pre- and post-training sessions.32 All patients
began.
underwent fMRI, but only those with subcortical lesions (ex-
perimental group, n 5 14; control group, n 5 7) not involving
sensorimotor and premotor areas were considered for LI PHYSIOTHERAPY. Immediately following a BMI training
assessment. More information about fMRI data acquisition session, patients in both groups received 1 hour of behavioral
and processing can be found in Supplementary Information physiotherapy focused on transferring arm reaching and hand
Section 6.2. The differences of LI calculated individually were movements to real life situations such as grasping a toothpaste
assessed across sessions and groups. A 2 3 2 repeated meas- tube, eating, relaxation in case of spasticity, and reaching
ures analysis of variance (ANOVA) with group (experimental and grasping while standing and with social distractions (see

July 2013 103

ANNALS of Neurology

Supplementary Information, Section 5.3 and Supplementary independent measures for group and repeated measures
Video 2).34,35 for time) was conducted to explore the impact of BMI
training and time on hope for improvement, as measured
Results by BMI–placebo questionnaire, and did not show any
Primary Behavioral Outcome Measure: significant effect. Furthermore, Mann–Whitney U tests
Combined Hand and Arm Scores (Motor Part) comparing the experimental group and control group for
from the Modified Upper Limb cFMA professional competence for every training week did not
We performed the statistical analysis on the cFMA scores. reach statistical significance either. Placebo scores
For the pre- to postintervention comparison, the average remained high during and after training, with no signifi-
of the 2 baseline measurements was used as a single pre- cant difference between groups (Supplementary Table 7),
measurement, reducing test variability effects, as used demonstrating stable positive expectancies, hope for
before in other studies of stroke rehabilitation.36 A 2-way improvement, and no recognition of group assignment,
mixed model ANOVA (with independent measures on which would have resulted in lower scores for the control
group and repeated measures on time) showed a signifi- group. More information about these analyses and statis-
cant time (before and after) 3 group (F1,28 5 6.294, tics can be found in Supplementary Information,
p 5 0.018) interaction and a significant effect of time Section 7.1.
(F1, 28 5 9.588, p 5 0.004) on cFMA scores. There was
no main effect of group (F1,28 5 0.034, p 5 0.855). BMI Control, EMG, and fMRI
Post hoc comparisons using 2-tailed paired-samples
BMI CONTROL. The movements of the arm/hand were
t test revealed a significant improvement in cFMA scores
directly dependent upon sensorimotor oscillations of 8 to
for the experimental group comparing between before
13Hz recorded over the ipsilesional sensorimotor cortex
and after BMI training (t1,15 5 26.049, p < 0.001). Spe-
and were used as a measure of BMI performance. The
cifically, average cFMA score 6 standard error (SE)
patients observed and felt their arm/hand moving during
increased from 11.16 6 1.73 before training to
a successful trial in BMI training. The statistical analysis
14.56 6 1.95 after training. By contrast, a 2-tailed
performed on BMI performance (moving the arm/hand
paired-samples t test comparison did not reveal signifi-
with brain oscillations) showed that only the experimen-
cant improvement from before (13.29 6 2.86) to after
tal group was able to improve BMI control significantly.
(13.64 6 2.91) BMI training in the control group
More information about the BMI performance measures,
(t1,13 5 20.316, p 5 0.757; Supplementary Fig 5). Raw
results, and analysis can be found in Supplementary In-
data post-training was significantly different from pre-
formation, Section 7.3.3. Learning self-regulation of
training in the absence of averaging pre1 and pre2 meas-
BMI control followed a monotonic positive course over
urements in the experimental group only, that is, when
time in the experimental group, similar to other reports
comparing 1 of each premeasurements separately with
of BMI learning indicating procedural memory mecha-
the postmeasurement (see Supplementary Information,
nisms for training periods as used here.7,16,37
Section 7.2). Change in the range of 3.4 points on
cFMA motor activity–related scores reflects a change EMG. We analyzed the muscle activity related to
from no activity to some in muscles involved in, for grasping movements before and after training. A Wil-
example, lifting and stretching the arm, turning the fore- coxon signed ranks test (EMG data were not normally
arm, and extending the wrist and/or fingers (Supplemen- distributed) on the amplitude and frequency of the
tary Video 3). Eleven of 16 patients in the experimental muscle activity as reflected by the waveform length of
group and 7 of 14 in the control group improved their the extensor digitorum EMG signal (see Supplementary
hand FMA scores. Fifteen of 16 patients in the experi- Information, Section 6.1) during opening and closing
mental group and 7 of 14 in the control group improved of the hand elicited a statistically significant change in
their modified arm FMA scores. Fifteen of 16 patients in the experimental group (z 5 22.327, p 5 0.020). EMG
the experimental group and 8 and 14 in the control waveform length (6SE) increased from 2.42 6 0.46
group improved their cFMA scores. before training to 3.69 6 0.71 after treatment in the ex-
perimental group, and in the control group values
Secondary Outcome Measures: GAS, MAL, increased from 1.95 6 0.45 to 3.58 6 0.97, although
Ashworth, Placebo Questionnaires not significantly (z 5 21.601, p 5 0.109). Overall the
We found no significant differences in Ashworth values results suggest an improvement in the ability to volun-
but significant improvements in GAS and MAL in both tarily engage muscle activity in the paretic hand.
groups. A 2-way mixed model ANOVA (with Mann–Whitney U tests comparing experimental and

104 Volume 74, No. 1

 Ramos-Murguialday et al: BMI in Chronic Stroke

control group EMG waveform length delta (before–after hand movements and cFMA scores after training was found
difference) did not reach statistical significance in patients with subcortical lesions in the experimental
(U 5 107, p 5 0.835). group (Pearson r12 5 0.55, p 5 0.05, 2-tailed). More infor-
To control for changes in muscle activation in the mation regarding fMRI statistical analysis can be found in
upper arm, EMG data were analyzed using paired t test Supplementary Information, Section 7.3.2.
between before and after intervention. The experimental
group showed a significant increase in paretic side activ- Discussion
ity during upper arm and elbow extension at the deltoid The results of this study indicate that contingent online
location from 1.35 6 0.08 to 1.47 6 0.1 (t 5 2.246, orthosis–BMI training adjuvant to physiotherapy results
p 5 0.040) and at triceps from 1.17 6 0.08 to in more prominent improvement in cFMA in chronic
1.38 6 0.13 (t 5 2.253, p 5 0.040) toward normal EMG stroke without residual movement capacity of the
activity, whereas the control group did not show any sig- affected hand than control BMI 1 physiotherapy. They
nificant EMG waveform length change at deltoid, show that BMI training, involving proprioceptive positive
increasing from 1.53 6 0.14 to 1.84 6 1.03 (t 5 1.739, feedback and reward that is time-contingent upon con-
p 5 0.106) and decreasing at triceps from 1.66 6 1.18 to trol of ipsilesional sensorimotor brain oscillations, may
1.51 6 0.76 (t 5 0.667, p 5 0.517). prime and thus improve the beneficial effects of physio-
Independent-sample t test comparing experimental therapy on motor function.38 Significant improvement
and control group EMG waveform length delta (before– on cFMA motor activity–related scores reflected a clini-
after difference) during upper arm and elbow extension cally meaningful change from no activity to some in
did not reach statistical significance at the deltoid loca- muscles involved in, for example, lifting and stretching
tion (t1,28 5 21. 014, p 5 0.319) or at the triceps loca- the arm, turning the forearm, and extending the wrist
tion (t1,28 5 1.589, p 5 0.123). and/or fingers. Immediate and correct feedback and
No significant paretic side EMG activity change reward in the framework of reinforcement learning of
during supination and wrist extension was found in any control of brain oscillatory activity translated into a
of the groups (see Supplementary Information, Section reaching and grasping movement of the paretic limb con-
7.3.1). Neither of the 2 groups of patients showed signif- stitutes the critical ingredient.39–41
icant changes in EMG at the electrodes placed over the The finding of significant differences between the
healthy side. experimental group and the control group receiving ran-
dom feedback indicates that this contingency is critical to
FMRI. The repeated measures ANOVA of group 3 ses- improve a physiotherapy-based neurorehabilitative inter-
sion (before or after) on LI of activity in the motor and vention. Placebo effects could not explain the results. It
premotor cortices during the actual movement condition is conceivable that the random noncontingent feedback
revealed a significant interaction effect (F1,19 5 10.22, in the sham group resulted in a diminished positive
p 5 0.005; experimental group, before 5 20.04 6 0.37, effect of physiotherapy or motor learning. Thus, contin-
after 5 20.27 6 0.48; sham group, before 5 20.12 6 0.39, gent BMI may produce a better outcome because it
after 5 0.27 6 0.42 [mean 6 SD]). After training, a signifi- avoids this negative effect. A crossover design trial could
cant difference of the LI in the motor and premotor corti- possibly help sort out this issue, but crossover designs
ces only during the actual movement condition was produce sequence effects that are difficult to separate.
measured in the experimental group for all 14 patients Furthermore, the questionnaires used to uncover placebo
(t13 5 2.61 p 5 0.02 paired-samples t test), whereas the effects may not be sensitive enough to reflect such non-
control group showed no significant changes either for conscious deleterious effects. Because of the subconscious
motor and premotor cortices or for somatosensory cortex nature of such an effect, it is difficult if not impossible
during executed (attempted) and imagined hand move- to rule out such unconscious placebos. However, if such
ments (Fig 2 and Supplementary Information, Section a nonconscious learning impairment occurred in the con-
7.3.2.). Eleven of 14 patients and 0 of 7 in the experimen- trol group, it should surface in a more negative attitude
tal and control groups, respectively, showed a shift of motor and treatment evaluation in the controls, which was not
and premotor activity from the contralesional hemisphere the case; controls rated their treatment and therapists as
toward the ipsilesional hemisphere, that is, toward normal equally efficient and competent.
activity, when movements were performed with the paretic It is conceivable that BMI training immediately
hand. Moreover, a significant correlation between the differ- preceding the relevant period of physiotherapy operates
ence of lateralization of brain activity (LIpre–LIpost) for as proposed by cortical stimulation,42–44 priming the
motor and premotor cortices during executed (attempted) effects of customary rehabilitation treatments,45 as shown

July 2013 105

ANNALS of Neurology

FIGURE 2: Lateralization index of blood oxygenation level–dependent activity (1 5 entirely contralesional, 21 5 entirely ipsile-
sional) was calculated for pre- and post-training functional magnetic resonance imaging (fMRI) sessions during hand-opening
attempts by patients with the paretic and with the healthy hand in the experimental or contingent positive group (C1) and
control or sham group (S). Top images show brain activations during paretic hand movements versus rest before and after
brain–machine interface (BMI) training (p < 0.001 uncorrected for visualization). fMRI maps were obtained from mixed effect
analysis on the experimental group with subcortical lesion only (n 5 14; maps of patients with lesion on the left hemisphere
were flipped to the right hemisphere). The data for the control group are not shown, as no significant changes were observed
between pre- and post-training sessions. Bottom graph shows lateralization index of active voxels in the ipsilesional and con-
tralesional motor and premotor areas during the actual movement condition for the paretic and healthy hand in the experi-
mental and control group before and after BMI training (only for patients with subcortical lesions). *p < 0.05. L 5 left; R 5 right;
t 5 t value.

in healthy participants.16 We believe a contingent link We proved that altering a brain signal (increase in
between brain activity (intention to move) and paretic SMR desynchronization) that is linked to prosthesis
limb movements (orthoses) influences the specific neural movements in time leads to motor learning and induces
network activity of the visuomotor loop involved in a neural plasticity or neural compensation that induces
motor task. This contingency could be interpreted as an motor function improvement.37 Conversely, a difference
instrumental motor learning task strengthening the asso- between BMI training and cortical stimulation is that
ciative (and neural) connection between movement BMI training engages a group of ecologically relevant
attempt and the consequence consisting of an actual brain regions related to the intention to perform a move-
arm/hand movement,46 following principles of Hebbian ment that these patients could not execute (eg, paretic
plasticity. The neuronal consequence of such a plastic finger motions), whereas cortical stimulation is com-
procedure may consist of an incremental excitability of monly applied over 1 target region like the primary
motor pools that represent these movements to the level motor cortex (but also the vicinity structures depending
that this neuronal activity is high enough to produce a on the invasiveness of the stimulation) and is not related
voluntary action potential in latently functional, spared to volitional brain signals. It is conceivable that BMI
descending corticospinal fibers. The best timing between training engaging a crucial network of brain regions
BMI training and physiotherapy to elicit the beneficial related to intent of the lost function could have contrib-
effects on cFMA scores remains to be determined. uted to improving the effects of physiotherapy evidenced

106 Volume 74, No. 1

 Ramos-Murguialday et al: BMI in Chronic Stroke

in cFMA scores and EMG activity. The use of only ipsi- National Institute of Neurological Disorders and Stroke
lesional brain oscillations could be a limitation in our (Bethesda, MD), Center for Neuroscience and Regenera-
study, because after stroke there is a shift of activity to- tive Medicine, Uniformed Services University of Health
ward the contralesional hemisphere, and engagement of Sciences (Bethesda, MD), Werner Reichardt Center for
activity in these regions could have improved BMI per- Integrative Neuroscience, University of T€ubingen, and
formance. However, as presented in previous work, func- Tecnalia. Several authors were supported by the
tional improvements were associated with changes in LI Deutscher Akademischer Austauschdienst (O.Y., E.G.,
toward ipsilesional motor regions, that is, toward normal F.L.B., and M.R.C.), Brazilian National Counsel of
LI in healthy individuals.41,44 This effect is in line with Technological and Scientific Development (F.L.B.), and
the view that training results in increased recruitment of Coordination for the Improvement of Higher Education
brain networks located in the vicinity of the lesion Personnel (Brazil) (M.R.C.), and a Humbolt Award to
accompanied by a decrease of contralesional activity in L.G.C. The funders had no role in study design, data
the healthy hemisphere.47,48 collection and analysis, decision to publish, or prepara-
Unbalanced bilateral brain activity toward the tion of the manuscript.
nonlesioned hemisphere in the chronic stage might We thank Dr A. Luft for his advice in the protocol
indicate a failure of compensatory mechanisms to design; S. Halder, J. Mellinger, and J. Dax for their help
restore normal, predominantly lateralized motor activa- in the development of the orthosis–BMI platform; B.
tion. Therefore, although the redundancy of an unaf- Benkner and M. Grammer for their help during the
fected cortex and the potential functional role of BMI training; T. Oesterle for his help during physiother-
ipsilateral pathways seem advantageous and might help apy; R. Veit and S. Lee for their assistance during the
during the acute phase, in the chronic phase the abnor- fMRI measurements; and J. Jesser for her help creating
mally increased inhibitory influence of the healthy the lesion mask images.
hemisphere upon the ipsilesional hemisphere may play
a maladaptive role.36,44 The neuroplastic processes that Potential Conflicts of Interest
characterize early brain reorganization after stroke Nothing to report.
change with time.45 The direct physiological regulation
of these networks using behavioral principles of rein-
forcement learning and procedural memory for skill ac- References
quisition may be responsible for such a lasting and 1. Langhorne P, Coupar F, Pollock A. Motor recovery after stroke: a
widespread cortical reorganization accompanied by the systematic review. Lancet Neurol 2009;8:741–754.

positive clinical modifications.36 2. Young J, Forster A. Review of stroke rehabilitation. BMJ

2007;334:86–90.
In summary and despite of the limited number of
3. Wolf SL, Winstein CJ, Miller JP, et al. Effect of constraint-induced
patients involved, our proof of concept study demon-
movement therapy on upper extremity function 3 to 9 months af-
strates that BMI training can successfully prime behavior- ter stroke: the EXCITE randomized clinical trial. JAMA
ally oriented physiotherapy to induce more clinically 2006;296:2095–2104.

significant improvements in motor function in chronic 4. Luft AR, McCombe-Waller S, Whitall J, et al. Repetitive bilateral
arm training and motor cortex activation in chronic stroke. JAMA
stroke patients with substantially restricted residual finger 2004;292:1853–1861.
movements and that these improvements are accompa- 5. Langhorne P, Bernhardt J, Kwakkel G. Stroke rehabilitation. Lancet
nied by a pattern of cortical reorganization previously 2011;377:1693–1702.
associated with spontaneous recovery of function and by 6. Thrasher TA, Zivanovic V, Mcllroy W, et al. Rehabilitation of reach-
an increase in EMG activity in muscles of the paretic ing and grasping function in sever hemiplegic patients using func-
tional electrical stimulation therapy. Neurorehabil Neural Repair
hand. 2008;22:706–714.

7. Buch E, Weber C, Cohen LG, et al. Think to move: a neuromag-

netic brain-computer interface (BCI) system for chronic stroke.
Acknowledgment Stroke 2008;39:910–917.

8. Ietswaart M, Johnston M, Dijkerman HC, et al. Mental practice

This work was supported by the German Federal Minis- with motor imagery in stroke recovery: randomized controlled trial
try of Education and Research (F€orderzeichen of efficacy. Brain 2011;134:1373–1386.
01GQ0831) as well as the Deutsche Forschungsgemein- 9. Page SJ, Levine P, Leonard AC. Effects of mental practice on
schaft, European Research Council (ERC 227632), Euro- affected limb use and function in chronic stroke. Arch Phys Med
Rehabil 2005;86:399–402.
pean Union Information and Communication
10. Johnson SH, Sprehn G, Saykin AJ. Intact motor imagery in chronic
Technologies Framework Program 7 (HUMOUR upper limb hemiplegics: evidence for activity-independent action
231724), Intramural Research Program of the NIH representations. J Cogn Neurosci 2002;14:841–852.

July 2013 107

ANNALS of Neurology

11. Sirigu A, Cohen L, Duhamel JR, et al. Congruent unilateral impair- 29. Stinear CM. Prediction of recovery of motor function after stroke.
ments for real and imagined hand movements. Neuroreport Lancet Neurol 2010;9:1228–1232.
1995;6:997–1001.
30. Caria A, Weber C, Br€otz D, et al. Chronic stroke recovery after
12. De Vries S, Mulder T. Motor imagery and stroke rehabilitation: a combined BCI training and physiotherapy: a case report. Psycho-
critical discussion. J Rehabil Med 2007;39:5–13. physiology 2011;48:578–582.
13. Birbaumer N, Cohen LG. Brain-computer interfaces: communica- 31. Ward NS. Mechanisms underlying recovery of motor function after
tion and restoration of movement in paralysis. J Physiol stroke. Postgrad Med J 2005;81:510–514.
2007;579:621–636.
32. Wilke M, Lidzba K. LI-tool: a new toolbox to assess lateralization
14. Daly JJ, Wolpaw JR. Brain-computer interfaces in neurological in functional MR-data. J Neurosci Methods 2007;163:128–136.
rehabilitation. Lancet Neurol 2008;7:1032–1034.
33. Tyc F, Boyadjian A. Plasticity of motor cortex induced by coordi-
15. Buch ER, Schanechi AM, Fourkas AD, et al. Parietofrontal integrity nation and training. Clin Neurophysiol 2011;122:153–162.
determines neural modulation associated with grasping imagery
34. Mastos M. Goal-directed training: linking theories of treatment to
after stroke. Brain 2012;135:596–614.
clinical practice for improved functional activities in daily life. Clin
16. Ramos-Murguialday A, Sch€ urholz M, Caggiano V, et al. Proprio- Rehabil 2007;21:47–55.
ceptive feedback and brain computer interface (BCI) based neuro-
35. Br€
otz D, Braun C, Weber C, et al. Combination of brain com-
prostheses. PLoS ONE 2012;7:e47048.
puter-interface training and goal directed physical therapy in
17. Moritz CT, Perlmutter SI, Fetz EE. Direct control of paralysed chronic stroke: a case report. Neurorehabil Neural Repair
muscles by cortical neurons. Nature 2008;456:639–642. 2010;24:674–679.
18. Jackson A, Mavoori J, Fetz EE. Long-term motor cortex 36. Whitall J, Waller SM, Sorkin JD, et al. Bilateral and unilateral arm
plasticity induced by an electronic neural implant. Nature training improve motor function through differing neuroplastic
2006;444:56–60. mechanisms: a single-blinded randomized controlled trial. Neuro-
rehabil Neural Repair 2011;25:118–129.
19. Edwardson MA, Lucas TH, Carey JR, Fetz EE. New modalities of
brain stimulation for stroke rehabilitation. Exp Brain Res 37. Dayan E, Cohen LG. Neuroplasticity subserving motor skill learn-
2013;224:335–358. ing. Neuron 2011;72:443–454.
20. Legenstein R, Chase SM, Schwartz AB, Maas W. A reward-modu- 38. Abe M, Schambra H, Wassermann EM, et al. Reward improves
lated Hebbian learning rule can explain experimentally observed long-term retention of a motor memory through induction of off-
network reorganization in a brain control task. J Neurosci line memory gains. Curr Biol 2011;21:557–562.
2010;30:8400–8410.
39. Lee D, Seo H, Jung MW. Neural basis of reinforcement learning
21. Collinger JL, Wodlinger B, Downey JE, et al. High-performance and decision making. Annu Rev Neurosci 2012;35:287–308.
neuroprosthetic control by an individual with tetraplegia. Lancet
40. Abe M, Schambra H, Wassermann EM, et al. Reward improves
2012;381:347–422.
long-term retention of a motor memory through induction of off-
22. Hochberg LR, Bacher D, Jarosiewicz B, et al. Reach and grasp by line memory gains. Curr Biol 2011;21:557–562.
people with tetraplegia using a neurally controlled robotic arm.
41. Koralek AC, Jin X, Long JD, et al. Corticostriatal plasticity is nec-
Nature 2012;485:372–375.
essary for learning intentional neuroprosthetic skills. Nature
23. Fugl-Meyer AR, J€
aa€sk€
o L, Leyman I, et al. The post-stroke patient: 2012;483:331–335.
1. A method for evaluation of physical performance. Scand J
42. Hummel FC, Cohen LG. Non-invasive brain stimulation: a new
Rehabil Med 1975;7:13–31.
strategy to improve neurorehabilitation after stroke? Lancet Neu-
24. Crow JL, Harmeling-van der Wel BC. Hierarchical properties of rol 2006;5:708–712.
the motor function sections of the Fugl-Meyer assessment scale
43. Fritsch B, Reis J, Martinowich K, et al. Direct current stimulation
for people after stroke: a retrospective study. Phys Ther
promotes BDNF-dependent synaptic plasticity: potential implica-
2008;88:1554–1567.
tions for motor learning. Neuron 2010;66:198–204.
25. Uswatte G, Taub E, Morris D, et al. Reliability and validity of the
44. Cramer SC, Riley JD. Neuroplasticity and brain repair after stroke.
upper-extremity Motor Activity Log-14 for measuring real-world
Curr Opin Neurol 2008;21:76–82.
arm use. Stroke 2005;36:2493–2496.
45. Stinear CM, Barber PA, Coxon JP, et al. Priming the motor system
26. Hurn J, Kneebone I, Cropley M. Goal setting as an outcome mea-
enhances the effects of upper limb therapy in chronic stroke. Brain
sure: a systematic review. Clin Rehabil 2006;20:756–772.
2008;131:1381–1390.
27. Lee SW, Wilson KM, Lock BA, Kamper DG. Subject-specific myo-
46. Fetz EE. Volitional control of neural activity: implications for brain-
electric pattern classification of functional hand movements for
computer interfaces. J Physiol 2007;579:571–579.
stroke survivors. IEEE Trans Neural Syst Rehabil Eng 2011;19:558–
566. 47. Rossini PM, Calautti C, Pauri F, et al. Post-stroke plastic reorgan-
isation in the adult brain. Lancet Neurol 2003;2:493–502.
28. Sehm B, Perez MA, Xu B, et al. Functional neuroanatomy of mir-
roring during a unimanual force generation task. Cereb Cortex 48. Murphy TH, Corbett D. Plasticity during stroke recovery: from syn-
2010;20:34–45. apse to behavior. Nat Rev Neurosci 2009;10:861–872.

108 Volume 74, No. 1