Aug 2026· Topics in Stroke Rehabilitation· pp.
1-17
· 0 citations· 58 references
Medicine
TL;DR
The neurophysiological basis of EEG-BCI and three major rehabilitation paradigms are outlined, including motor imagery with physical feedback, motor imagery with virtual/multisensory feedback, and the steady-state visual evoked potential (SSVEP)-driven paradigm.
Abstract
Background
Post-stroke hand dysfunction severely limits patients' independence, and conventional rehabilitation often fails those without active movement. Noninvasive EEG-based brain-computer interface (BCI) technology addresses this by creating a closed-loop feedback system rooted in Hebbian learning principles. This system decodes rhythmic signals from the sensorimotor cortex during imagined hand movements in real-time. The decoded intention is then translated into commands to drive exoskeletons, functional electrical stimulation (FES), or virtual reality (VR) devices, thereby moving the affected limb. This process strengthens or remodels damaged neural pathways, promoting motor recovery.
Methods
This article systematically outlines the neurophysiological basis of EEG-BCI and three major rehabilitation paradigms: motor imagery with physical feedback, motor imagery with virtual/multisensory feedback, and the steady-state visual evoked potential (SSVEP)-driven paradigm.
Results
Studies confirm these approaches can improve upper limb function, showing significant potential. However, widespread clinical use faces challenges like low signal-to-noise ratios, significant individual variability, and "BCI blindness."
Conclusions
Future work should focus on improving decoding algorithms, developing more user-friendly devices, deepening mechanistic understanding, and establishing standardized clinical assessments. This review aims to offer valuable guidance for subsequent research.
Electroencephalography-based brain-computer interfaces (EEG-based BCIs) provide a non-invasive pathway for incorporating voluntary neural activity into lower-limb rehabilitation robots, exoskeletons, robotic orthoses, and gaittraining systems. This focused review examines the MI-based brain-robot rehabilitation loop, including lower-limb intention decoding, high-level command generation, robot or gait-device interaction, and closed-loop feedback. Evidence is interpreted across four categories: offline lower-limb MI decoding, online BCI demonstrations, lower-limb device integration, and patient-oriented or clinical evaluation. Representative studies support the technical feasibility of decoding lower-limb motor imagery (MI) and using selected BCI outputs in virtual-reality, exoskeleton, and treadmill systems. However, the evidence remains dominated by offline analyses and small proof-of-concept studies, with limited standardized clinical outcomes. Hybrid sensing and multimodal feedback have been explored as complementary strategies, but their value in physical lower-limb rehabilitation requires direct online and patient-oriented validation. The review therefore distinguishes transferable decoding advances from direct rehabilitation evidence and identifies priorities for safe and clinically meaningful system development.
Yong-Kang Li, Aihui Wang, Xin-Yu Liu et al.· International Conference on...· 0 citations
The study stresses the necessity of embedding relational autonomy and neural rights into BCI development, tying technological trajectories to governance demands in order to shape responsible paths for future neurotechnologies.
Yuzhang Wu· Theoretical and Natural Scie...· 0 citations
Background Brain-computer interfaces (BCIs) based on electroencephalography (EEG) signals have been applied to improve active hand motor function rehabilitation of stroke patients. Besides, the bilateral arm rehabilitation training can effectively promote the recovery of the unilateral affected hand. However, existing studies on BCIs for stroke patients mainly focus on decoding the unilateral affected hand without effectively utilizing the unaffected hand. Objective In this paper, we studied the neural signatures and decoding of bimanual and unimanual motor attempts of hemiplegic stroke patients from EEG signals to explore the neural signature differences between bimanual and unimanual movements of patients and whether the differences can improve movement decoding accuracy. Methods: We designed the experimental paradigm of unimanual and bimanual opening and closing motor attempts. Furthermore, we compared the neural signature differences using movement-related cortical potential (MRCP) and event-related desynchronization (ERD). Results Experimental results from hemiplegic stroke patients indicated that greater MRCP activation was found during bimanual movements than unimanual movements. Although the event-related spectral perturbation (ERSP) varied among patients, it mainly appeared in the [8–30] Hz frequency band, and the ERD activation of bimanual movements was greater than that of unimanual movements in specific frequency bands and brain regions for each patient. The average decoding accuracy of bimanual movements was higher than that of unimanual movements by 4% to 12%. Discussion This work can potentially advance BCI-based active rehabilitation of hand motor function of stroke patients.
Jiarong Wang, Luzheng Bi, Weijie Fei et al.· Frontiers in Human Neuroscie...· 0 citations
Stroke recovery is increasingly understood as a process shaped by disrupted interactions among motor intention, descending motor output, peripheral movement, and sensory feedback, rather than by motor weakness alone. After stroke, residual motor intention may not be effectively translated into spinal motor output, peripheral movement may be too limited to provide sufficient sensory feedback, and compensatory network recruitment may not always support efficient motor control. Closed-loop brain-computer interfaces offer a mechanistically motivated approach to this problem by detecting motor imagery, motor attempt, or sensorimotor rhythms in real time and translating these signals into contingent functional electrical stimulation, robotic assistance, virtual reality, multisensory feedback, or neuromodulation. By partially restoring the temporal coupling between detected motor intention, assisted or stimulated movement, and sensory feedback, these systems are hypothesized to promote activity-dependent plasticity, reinforce sensorimotor circuits, and modulate distributed motor, sensory, attentional, and interhemispheric networks. In this review, we synthesize the neurobiological basis, recovery mechanisms, implementation strategies, and emerging predictive biomarkers of closed-loop brain-computer interfaces for post-stroke sensorimotor rehabilitation. Although closed-loop brain-computer interfaces show promise for selected patients, treatment effects remain heterogeneous and depend on patient characteristics, decodability of motor-related brain signals, intervention dose, feedback modality, and trial design. Future progress will require mechanism-informed patient stratification, adaptive decoding, individualized feedback, standardized outcomes, and real-world validation.
Ran Tao, Can Duan, Yangpu Zhang· Frontiers in Neuroscience· 0 citations
Brain-computer interface (BCI)-enhanced lower-limb rehabilitation exoskeletons can translate movement-related brain signals into assisted walking, thereby linking motor intention, robotic assistance, and sensory feedback. Existing evaluations of these systems mainly focus on decoding accuracy, gait control, safety, and clinical rehabilitation outcomes, but these indicators do not fully explain whether patients experience assisted walking as self-initiated and related to their own effort. This paper approaches this issue through the concept of sense of agency and proposes a loop-based analytical framework for BCI-controlled exoskeleton rehabilitation. It suggests that sense of agency depends not only on whether motor intention is accurately decoded, but also on whether the decoded command is triggered within a plausible temporal window, whether system responses remain stable across repeated training, whether robotic assistance preserves the patient’s perceived contribution, and whether sensory and contextual feedback can be meaningfully interpreted. On this basis, the paper argues that assisted walking may feel machine-driven when patients cannot integrate their intention, device response, limb movement, and feedback into a coherent process of self-attribution. Conversely, agency may be better supported when the system makes intention perceptible, assistance attributable, and shared control understandable. This perspective reframes the design significance of low-latency decoding, assist-as-needed control, and multimodal feedback: these features are not only technical or clinical indicators, but also important design conditions for supporting patients’ sense of agency during assisted movement.
Brain–computer interface (BCI) technology has shown potential for future rehabilitation-related and assistive control applications. Nevertheless, single-modality electroencephalography-based motor imagery (EEG-MI) signals are susceptible to interference, whereas existing algorithmic models suffer from limited classification accuracy and insufficient actionable control commands for interactive devices, thereby impeding their practical deployment. To tackle these limitations, this study presents a multimodal human–computer interaction control scheme that integrates eye-movement command encoding with EEG motor imagery decoding. Self-collected EEG-MI and eye-movement datasets were established to support the proposed multimodal control framework. In this framework, eye movements are not used merely as auxiliary inputs, but are encoded as discrete commands for start, stop, grasp, and release, thereby reducing the command burden of EEG-MI decoding. The EEG-TransNet model is enhanced by integrating a time–frequency feature branch and replacing the original convolutional encoder with an adaptive multi-branch EEG feature gating module, strengthening the representation and fusion of multi-domain features. The model yields average classification accuracies of 86.96% and 88.73% on the BCI IV-2a dataset and the self-collected EEG dataset, respectively. Four independent SVM binary classifiers are adopted to identify four eye movement patterns. The EEG and eye movement classification results are binary-encoded to generate hardware-compatible control commands. Robotic-arm grasping experiments with healthy trained participants showed an average task completion time of 17 s, and the repeated grasping success-rate results further provide preliminary evidence for the real-time feasibility of the multimodal control framework under controlled laboratory conditions.
Xiangyang Sun, Wenjun Zhang, Jiahua Wu et al.· Journal of Eye Movement Rese...· 0 citations
