Differentiated Effects of Robot Hand Training With and Without Neural Guidance on Neuroplasticity Patterns in Chronic Stroke

Xin Wang, Wan-wa Wong, Rui Sun, Chiu‐Wing Winnie Chu, Raymond Kai‐Yu TongFrontiers in Neurology 2018 · DOI

TL;DR

This study investigated whether adding neural guidance (EEG-based motor imagery detection) to robot-assisted hand training improves long-term motor recovery and neuroplasticity in chronic stroke patients. The findings suggest that guided training leads to sustained motor improvements and measurable changes in brain activity, unlike non-guided training.

What Problem Does This Paper Address?

Stroke survivors often experience persistent motor impairments, particularly in their upper limbs, significantly impacting their independence and quality of life. While robot-assisted therapy has shown promise in aiding motor recovery, its long-term effectiveness and the underlying neuroplastic changes are not fully understood. Specifically, it's unclear if adding "neural guidance"—using brain signals to control the therapy—enhances these outcomes compared to standard robot-assisted training. This research aims to bridge this gap by comparing a guided robot-hand training approach with a non-guided one, examining both clinical motor function and neurophysiological changes.

Methods

The study employed a randomized controlled trial design with 24 chronic stroke participants. Participants were divided into two groups:

  • RobotEEG_AO Group (n=13): Received robot-hand training combined with Action Observation (AO) and real-time EEG guidance. The robot hand was activated only when specific EEG signals (mu suppression, indicating motor imagery) were detected.
  • Robotnon-EEG_Text Group (n=11): Received robot-hand training with text-based instructions for motor imagery, but the robot hand was triggered randomly, independent of their EEG signals.

Both groups underwent 20 sessions of robot-assisted hand training. Motor function was assessed using the Fugl-Meyer Assessment for upper-extremity (FMA-UE) at three time points: before, immediately after, and 6 months after the intervention.

To investigate neuroplasticity, two neuroimaging techniques were used:

  • Electroencephalography (EEG): Used to calculate the "brain discriminant ability" (DR) between motor imagery and resting states. This measures how well brain activity patterns can be distinguished between these two states, reflecting motor imagery performance.
  • Functional Magnetic Resonance Imaging (fMRI): Used to analyze "brain network variability" in resting-state fMRI data. This measures the dynamic fluctuations in functional connectivity within different brain networks over time, providing insights into brain reorganization.

Datasets

The study utilized data from 24 chronic stroke participants. For neuroimaging, 16 participants (8 from each group) who had no MRI contraindications underwent MRI scans before and after the intervention. The EEG data was collected during the training sessions, and resting-state fMRI data was acquired before and after the intervention. No new public datasets were introduced; the study relied on data collected from its participants.

Key Insights and Ideas

  • Neural Guidance Matters for Long-Term Recovery: The core insight is that actively using brain signals (EEG) to guide robot-assisted therapy leads to more sustained and significant improvements in motor function compared to training without this guidance.
  • Neuroplasticity is Enhanced by Guidance: The study provides evidence that guided training not only improves motor skills but also induces measurable changes in brain activity patterns, suggesting a deeper level of neuroplastic reorganization.
  • Brain Discriminant Ability as a Biomarker: The EEG-based "brain discriminant ability" emerged as a sensitive measure of improved motor imagery performance and neuroplasticity, particularly in the ipsilesional hemisphere, after guided training.
  • Dynamic Brain Network Changes: fMRI revealed that guided training increased the temporal variability of brain networks, indicating greater flexibility and adaptability in brain regions involved in sensory-motor control, attention, auditory processing, and the default mode network. This suggests a more widespread and dynamic brain reorganization.
  • Synchronization of Motor Pathways: The authors hypothesize that the guided approach, by synchronizing peripheral stretching, motor imagery, and action observation, might more effectively activate and retrain motor pathways.

Key Results

  • Motor Function: The RobotEEG_AO group showed a statistically significant improvement in paretic upper-limb motor functions that was sustained at the 6-month follow-up. In contrast, the Robotnon-EEG_Text group did not show significant long-term improvements.
  • Clinically Important Difference (MCID): A higher proportion of participants in the RobotEEG_AO group exceeded the minimal clinically important difference (MCID) for FMA-UE scores, suggesting a more meaningful functional recovery.
  • EEG Discriminant Rate: The RobotEEG_AO group exhibited a significant increase in ipsilesional EEG discriminant rate after training, indicating improved motor imagery performance. The Robotnon-EEG_Text group showed no significant change.
  • Brain Network Variability (fMRI): The RobotEEG_AO group showed significant increases in temporal variability in four out of six functional brain subnetworks (sensory-motor, attention, auditory, and default mode networks). No significant changes were observed in the non-guided group.

Limitations and Future Work

  • Sample Size: The study involved a relatively small sample size (24 participants, with only 16 undergoing MRI), which limits the generalizability of the findings.
  • Heterogeneity: Variations in lesion site, size, time since stroke, and stroke type among participants could influence the results.
  • Ambiguous Training Factors: The study design involved multiple influencing factors (AO vs. text, EEG-guided vs. random triggering), making it difficult to isolate the precise contribution of each component. Future studies could benefit from more granular designs to disentangle these effects.
  • MCID Variability: The authors note that different methods for determining MCID can lead to variations, posing a challenge for consistent interpretation.

Future work should focus on larger, more diverse cohorts to validate these findings. Investigating the specific contributions of action observation and EEG feedback independently would also be valuable. Further research into the dynamic functional connectivity patterns and their relationship to motor skill acquisition is also recommended.

Summary updated

Abstract

Robot-assisted training combined with neural guided strategy has been increasingly applied to stroke rehabilitation. However, the induced neuroplasticity is seldom characterized. It is still uncertain whether this kind of guidance could enhance the long-term training effect for stroke motor recovery. This study was conducted to explore the clinical improvement and the neurological changes after 20-session guided or non-guided robot hand training using two measures: changes in brain discriminant ability between motor-imagery and resting states revealed from electroencephalography (EEG) signals and changes in brain network variability revealed from resting-state functional magnetic resonance imaging (fMRI) data in 24 chronic stroke subjects. The subjects were randomly assigned to receive either combined action observation (AO) with EEG-guided robot-hand training (Robot<sub>EEG_AO</sub>, <i>n</i> = 13) or robot-hand training without AO and EEG guidance (Robot<sub>non-EEG_Text</sub>, <i>n</i> = 11). The robot hand in Robot<sub>EEG_AO</sub> group was activated only when significant mu suppression (8-12 Hz) was detected from subjects' EEG signals in ipsilesional hemisphere, while the robot hand in Robot<sub>non-EEG_Text</sub> group was randomly activated regardless of their EEG signals. Paretic upper-limb motor functions were evaluated at three time-points: before, immediately after and 6 months after the interventions. Only Robot<sub>EEG_AO</sub> group showed a long-term significant improvement in their upper-limb motor functions while no significant and long-lasting training effect on the paretic motor functions was shown in Robot<sub>non-EEG_Text</sub> group. Significant neuroplasticity changes were only observed in Robot<sub>EEG_AO</sub> group as well. The brain discriminant ability based on the ipsilesional EEG signals significantly improved after intervention. For brain network variability, the whole brain was first divided into six functional subnetworks, and significant increase in the temporal variability was found in four out of the six subnetworks, including sensory-motor areas, attention network, auditory network, and default mode network after intervention. Our results revealed the differences in the long-term training effect and the neuroplasticity changes following the two interventional strategies: with and without neural guidance. The findings might imply that sustainable motor function improvement could be achieved through proper neural guidance, which might provide insights into strategies for effective stroke rehabilitation. Furthermore, neuroplasticity could be promoted more profoundly by the intervention with proper neurofeedback, and might be shaped in relation to better motor skill acquisition.
BibTeX
@inproceedings{wang2018differentiated,
  author = {Xin Wang and Wan-wa Wong and Rui Sun and Chiu‐Wing Winnie Chu and Raymond Kai‐Yu Tong},
  title = {{Differentiated Effects of Robot Hand Training With and Without Neural Guidance on Neuroplasticity Patterns in Chronic Stroke}},
  booktitle = {Frontiers in Neurology},
  year = {2018},
  doi = {10.3389/fneur.2018.00810},
}
Figure