Background
Stroke is a major cause of long-term disability worldwide [
1,
2], and one of the most frequent impairments is hemiparesis, which is characterized by weakness, lack of control, increased muscle tone on the contralesional upper limb (UL) and lower limb or hemibody, and deteriorating independence in activities of daily life (ADL), especially walking, dressing or eating [
3]. Critically, other impairments (e.g., somatosensory, visual, and cognitive), whether isolated or combined, also significantly deteriorate ADL. Most ADL require skilled bimanual coordination that can be impaired by a stroke, thus leading to a loss of independence that may in turn lead to a 50% reduction in quality of life [
4]. Despite rehabilitative care provided during the acute phase of stroke, 30% of patients still suffer from participation restrictions after four years [
5]. It has been suggested that neurorehabilitation should not focus exclusively on impairments of the paretic arm or hand and should instead consider more bimanual actions and activities [
6‐
8]. After a unilateral stroke, impairments of the contralesional UL can deteriorate bimanual actions [
9], thus supporting the importance of training both ULs to achieve better functional recovery in (bimanual) ADL [
8,
10]. Interestingly, during bilateral cooperative movements (e.g., opening a bottle), neural coupling from the ipsilesional to the contralesional (impaired) UL is preserved in most patients with stroke, suggesting the relevance of bilateral training that supports cooperative hand movements for ADL [
11]. During bimanual training, various tasks can be utilized to promote intensive and repetitive coordinated movement of the ULs. A classification of bimanual tasks has been proposed for different bimanual actions. Grossly, two types of tasks can be distinguished: those with symmetrical movements that engage homologous muscles (e.g., picking up a box simultaneously with both hands) and those with asymmetrical movements that engage nonhomologous muscles nonsimultaneously (e.g., cutting a piece of steak). Similarly, two types of task goals can be distinguished: independent goals (e.g., one hand lifting a cup and the other hand lifting a glass simultaneously) versus common goals (e.g., both hands working together to accomplish a common task) [
8]. Many bilateral actions, such as arms swinging during bipedal locomotion, seem to depend on “default-mode” neural coupling. However, in most skilled ADL, bimanual actions are accomplished through asymmetrical movements that cooperate to achieve a common goal, e.g., buttoning a skirt or changing the gear while steering a car. Such complex bimanual, cooperative, asymmetrical skills have to be learned.
After a stroke, motor skill learning (MSkL) plays a key role in recovery by compensating for activity limitations and participation restrictions. MSkL is a fundamental ability that allows for the acquisition of unimanual or bimanual skills (i.e., writing, playing the piano) and adaptation of these skills to changing environments. It has been suggested that procedural learning, including MSkL, proceeds over three phases [
12]: (i) an early “strategic/cognitive” phase, which presents rapid performance improvement, especially in the dorsolateral prefrontal cortex and posterior parietal cortex (PPC); (ii) a consolidation phase, which involves the stabilization of the learned skill based principally through a corticostriatal loop (striatum and supplementary motor area); and (iii) a retention phase, which is also called the “automatization phase”, during which the performance of the skill is optimized due to the increased activity in the primary motor cortex (M1), the premotor cortex and PPC [
13]. Improvement of a skill is linked to practice-dependent training: the more we practice a skill, the better we perform it, with smoother movements and reduced variability [
14,
15]. Sensorimotor skill acquisition (or MSkL) represents the ability to select and refine the movements needed to attain a goal in which the sensory stimuli for selecting and correcting our actions are considered and then the skill is executed consistently with both speed and accuracy (i.e., with motor acuity). Once learned, a motor skill can be retained for long periods of time, thus leading to lasting performance improvements, which is the aim of neurorehabilitation [
16].
Robotic devices have long been expected to enhance recovery after a brain injury, such as a stroke [
17‐
20], because they offer the possibility of providing intensive task-specific training to regulate task parameters, quantify and monitor improvements, and continuously adapt the task’s difficulty [
21,
22]. E.g., Keeling and al. [
20] showed that the use of the bimanual robotic tasks for rehabilitation in subacute stroke is feasible and suggested that the use of robotic devices added to standard of care therapy could augment recovery. Moreover, the proprioceptive feedback during active movements, delivered through a robotic therapy improved sensorimotor function in chronic stroke patients [
23]. In fact, the proprioceptive training could enhance somatosensory and motor functions and induce cortical reorganization [
24]. Interestingly, robotics has the potential to formally implement the principles of motor learning in neurorehabilitation. Cuppone et al. [
25] found that somatosensory learning is linked to motor learning because these processes similar features of memory formation. Robotic devices can provide four main training modalities: (i) active mode (where the subject fully performs the task), (ii) active-assisted mode (where the robot provides assistance either at a fixed rate or “as needed”), (iii) passive mode (where the robot fully performs the task), and (iv) resistive mode (where the robot perturbs the subject’s attempts); these modalities allow for valuable interactions with patients [
26]. In a meta-analysis, Kwakkel et al. showed significant improvement in UL impairment (but not in ADL) with robot-assisted training (RAT) [
27]. However, a recent Cochrane review showed that RAT enhances both UL impairments and ADL in stroke survivors [
18], and another team showed that RAT improved motor coordination compared to unilateral training in patients with stroke with severe impairments [
28].
More recently, we used a custom system with computer mice and showed that patients with stroke were able to learn, retain and generalize a complex bimanual skill after a single session of real and sham transcranial direct current stimulation (tDCS) [
29]. Determining whether patients with stroke could achieve bimanual MSkL (bim-MSkL) and identifying the underlying mechanisms and extent of learning are crucial for the development of efficient neurorehabilitation approaches targeting independence in (bimanual) ADL.
To explore how patients (re)learn to coordinate their hands after a stroke, we developed a complex asymmetrical bimanual coordination task (CIRCUIT) that was implemented as a serious game in the bimanual version of the REAplan® robot (AXINESIS, Wavre, Belgium). With a common cursor controlled by coordinated movements of the ULs interacting with robotic handles, one hand exclusively controlled lateral displacements of the common cursor while the other hand exclusively controlled the sagittal displacements. It has been suggested that stroke recovery studies should include quantitative measures, such as speed, accuracy, path length metrics and smoothness of movement [
26]. By analyzing kinematic parameters and providing such real-time quantitative measures of movement, REAplan® can be used for training UL movements [
30,
31]. Our hypotheses were that (i) patients in the chronic phase of stroke would show improvements in a new complex bimanual coordination skill and be able to retain and generalize this skill, i.e., they would be able to achieve complex bim-MSkL and/or improve on other bimanual or unimanual performances; (ii) patients would show similar improvements in bim-MSkL as healthy individuals; and (iii) poorer baseline clinical scales in patients would correlate with poorer bim-MSkL indices.
Discussion
When training with a serious game on a neurorehabilitation robot, patients in the chronic phase of stroke were able to learn and retain a complex bimanual skill and to generalize performance improvements to other bimanual or unimanual tasks. The HIs performed better than the patients with more severe impairment (Group 2, FMA-UE: 28–65), who showed large interindividual variability in both the magnitude and trajectory of bim-MSkL. The patients with minimal impairment (Group 1, FMA-UE: 66) showed intermediate progression between the HIs and Group 2.
Bimanual motor skill learning
Across sessions, the HIs showed changes in the biSAT, biCO and biFOP as well as retention and generalization, and they achieved typical bim-MSkL. Overall, chronic patients with supratentorial stroke were able to achieve complex bim-MSkL involving a new control policy and to generalize performance improvements to a new, untrained, complex bimanual task. The first hypothesis of this study was thus confirmed. These data expand the results from a previous study in which patients in the chronic phase of stroke achieved bim-MSkL over a single training session under real and sham tDCS, and an additional effect of noninvasive brain stimulation was not observed [
29]. In the current study, changes in the three outcomes (biSAT, biCO, and biFOP) were observed across three consecutive days. Compared to the last block on the previous day, a slight performance drop was observed for the first block of D2 and D3, although overnight retention remained consistent.
Given that 9 patients in the chronic phase of stroke had a normal FMA-UE score, we decided to split the patient pool into two groups. In Group 1 (FMA-UE = 66, n = 9), the overall progression of the biSAT was not significantly different compared to that of the HIs. In Group 2 (FMA-UE < 66, n = 15), the overall progression of the biSAT was significantly inferior to that of the HIs. Nevertheless, patients from Group 2 achieved bim-MSkL, including retention and generalization, and did not seem to reach a ceiling; however, whether their ability could eventually match that of Group 1 after further training was not determined.
The overall biCO progression was not significantly different between the three groups, although Group 2 had poorer baseline bimanual coordination than the HIs. Interestingly, the biCO appeared to plateau at D3 in the HIs and Group 1, whereas this was not observed in Group 2. In a previous study in younger HIs, the biCO was correlated with the biSAT [
29], suggesting that the biSAT and biCO reflect either the same process or overlapping processes. It is possible that within their range of potential biCO improvement, the patients (Groups 1 and 2) did not perform significantly worse than the His; however, this did not translate into similar improvements on the primary outcome (biSAT), for which feedback was provided.
In the current study, the biFOP increased across sessions in both patients and HIs, which was inconsistent with the decrease observed previously in younger HIs [
41]. The biFOP quantifies the forces exerted in nondesired directions by each hand against virtual walls. Theoretically, training should result in improvement, and the biFOP should thus
decrease, which would reflect less force “wasted” in the wrong direction [
41]. Here, instruction about the force was not provided and a penalty was not assigned for pushing against the virtual walls. Therefore, the HIs and patients might have simply not paid attention to this aspect and remained focused on the biSAT (for which feedback was provided) at the cost of some increase in the biFOP.
To summarize, the HIs and patients from Groups 1 and 2 achieved bim-MSkL over three days, including overnight retention. The progression of Group 1 was intermediate between that of the HIs and Group 2, in which robotic outcome improvements remained globally inferior to that of the HIs, suggesting that there was still room for improvement in more impaired patients.
Generalization
The HIs achieved a larger biSAT generalization and biFOP increase compared to the patients from Groups 1 and 2, whereas the biCO generalization was similar across the groups. Thus, both the HIs and patients were able to generalize the newly learned bimanual control policy, which is a hallmark of MSkL [
43].
Furthermore, the improvements driven by the bimanual CIRCUIT training transferred to bimanual REACHING. The HIs and patients were thus able to use the newly learned bimanual control policy to achieve a different task within the same robotic environment. Finally, in both the HIs and patients, there was also a trend for a (transfer of) performance improvement to the unimanual BBT, whereas the unimanual GF remained unchanged. Of course, we cannot rule out that this improved trend is actually due to the short interval between test and retest. Although the BBT improvements remained modest in patients (see Table
3), this finding is encouraging for neurorehabilitation but it remains to be confirmed in future experiments.
Correlations between robotic outcomes and clinical scales
The FMA-UE did not correlate with the baseline biSAT, biCO or biFOP, suggesting that the degree of unilateral motor impairment was not an accurate predictor of how well chronic patients could coordinate bimanual movements. Interestingly, the biSAT evolution correlated positively with the FMA-UE, BBT and SIS, suggesting that patients with less baseline impairment and participation restriction could achieve larger bim-MSkL after training on a robotic device. Although the lack of correlation between the bimanual robotic outcomes and the ABILHAND is surprising at first glance, it may reflect a discrepancy between the bimanual ADL the patients believe they are able to achieve and the tasks that can be objectively quantified with a bimanual robotic system. Furthermore, the ABILHAND questionnaire does not consider compensation while performing these ADL, whereas compensation is limited during evaluations with the REAplan®.
Therapeutic implications
Our data demonstrate that bim-MSkL with the REAplan® robot may help improve bimanual coordination in patients with chronic stroke with either minimal or mild-moderate impairment, which may indicate interesting prospects for neurorehabilitation [
44]. Previous rehabilitation studies have shown improvements in bilateral limb functions after stroke, such as bilateral arm training with rhythmic auditory cueing (BATRAC) and robotic mirror image movement enabler (MIME) [
45]. It might be interesting to perform more randomized control trials that implement cooperation between the ULs, such as training to perform asymmetrical bimanual actions that sharing a common goal, and to compare bimanual with unimanual interventions [
44,
46].
In this study, we used an active mode (i.e., no robotic assistance), which requires a minimal level of residual function of the paretic UL to perform the bimanual tasks. It would be interesting to investigate other RAT modes, e.g., passive and/or active-assisted mode with an adaptive algorithm [
31], in patients with more severe motor or cognitive impairments.
Limitations
This study has several limitations. The sample size was relatively small and heterogeneous (e.g., strokes with different sizes, residual impairments, time since stroke, etc.). Furthermore, the patients had mostly mild to moderate UL impairment (FMA-UE: 56.6 ± 13.5, range 28–66). It is unknown whether similar results would be found in patients with more severe impairments. Moreover, on D3, the duration of the robotic session was longer than on the previous days. Indeed, the patients performed the Generalization task (three 1-min runs) and the REACHING task (16 trials, back and forth) in addition to the training. It is possible that this led to a decrease in performance at the end of on D3, due to either physical or cognitive fatigue. Large interindividual variability was observed, and it would be interesting to correlate bim-MSkL with the size and localization of the stroke on brain imaging. One method of confirming and expanding these results would be to recruit more patients. Next, the improvement in bim-MSkL was large over three consecutive days of training and the biCO seemed to plateau in Group 1 on D3, such as in the HIs. It is unknown whether expanding the number of training sessions would further strengthen the improvements. However, combining several unimanual and bimanual serious games on a robotic device might enhance recovery further. Finally, it would be interesting to combine bimanual RAT with “classical” neurorehabilitation.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.