Motor impairment following stroke frequently leads to long-term disability, limiting independence and quality of life. Brain-Computer Interface (BCI) systems integrating motor imagery (MI) with virtual reality (VR) offer promising avenues for enhancing neuroplasticity and engagement through immersive, real-time, and proprioceptive feedback. Yet, identifying reliable electroencephalography (EEG)-based biomarkers that reflect or predict recovery remains challenging. This study investigated the relationship between event-related desynchronization (ERD) dynamics during MI-VR training and motor recovery in individuals with chronic stroke. Fourteen participants with stroke (9 experimental, 5 control) completed a 4-week VR-BCI intervention and were compared with a non-stroke reference cohort (N = 35). Linear mixed-effects models assessed ERD modulation across sessions and groups, and a two-stage regression evaluated the predictive value of ERD features for Fugl-Meyer Assessment (FMA) gains. Results showed no significant ERD change across sessions, but stroke participants exhibited significantly reduced ERD compared to controls. Baseline ERD amplitude predicted motor improvement, whereas ERD progression did not. Ipsilateral ERD showed a compensatory trend in ischemic stroke. These findings indicate that baseline ERD may serve as a stronger prognostic biomarker than short-term ERD dynamics, supporting the development of personalized VR-BCI rehabilitation strategies for chronic stroke recovery.

Introduction:The sense of embodiment (SoE), describing the experience of owning, controlling, and being located within a body, underpins virtual reality (VR) interaction, brain-computer interfaces (BCIs), and multisensory body-illusion research. Although SoE is typically assessed through subjective questionnaires, their variability and limited validity have motivated the search for objective neural markers. Electroencephalography (EEG) has become the most widely used technique given its portability and high temporal resolution; however, the existence of a consistent EEG correlate of embodiment remains unclear. Methods:This systematic review summarizes 35 EEG studies (2010-June 2025) identified through structured database searches, examining SoE across immersive and non-immersive VR, augmented reality, and non-VR paradigms. We analyze EEG features including spectral power, event-related desynchronization/synchronization (ERD/ERS), connectivity, and temporal dynamics, and examine methodological variability in illusion induction and SoE assessment. Results:Across studies, the reduction of the alpha-band over central-parietal regions emerges as the most recurrent correlate of embodiment. Beta-band decreases and gamma-band increases appear in several studies but lack consistent replication, while findings in Delta and Theta bands remain sparse and contradictory. Considerable heterogeneity is found in VR paradigms, EEG setups, preprocessing, and psychometric tools, contributing to inconsistent results and limiting cross-study comparability. Discussion:Critically, no EEG feature demonstrates sufficient reproducibility to qualify as a universal biomarker of SoE, and no standardized protocol for EEG-based embodiment assessment currently exists. Overall, this review highlights both the promise and current limitations of EEG-based approaches to measuring embodiment. We conclude by identifying methodological gaps and outlining recommendations to support the development of reliable EEG markers for future applications in VR rehabilitation, MI-BCIs, cognitive neuroscience, and clinical interventions.

Background:The high incidence and prevalence of upper limb impairment post stroke highlights the need for advancements in rehabilitation. Brain-computer interfaces (BCIs) represent a promising technology by directly training the central nervous system. The integration of motor imagery (MI) and motor observation through virtual reality (VR) using BCIs provides valuable opportunities for rehabilitation. However, the diversity in intervention designs demonstrates the lack of guiding recommendations integrating neurorehabilitation principles for BCIs. Objective:This study aims to develop recommendations for BCI interventions using task specificity and ecological validity through simulated VR tasks for upper limb stroke survivors by gathering tacit knowledge from neurorehabilitation experts, patients' experiences, and engineers' expertise to ensure a comprehensive approach. Methods:A multiperspective qualitative study was conducted through collaborative design workshops involving stroke survivors (n=17), neurorehabilitation experts (n=13), and biomedical engineers (n=3), totaling 33 participants. This innovative approach aimed to actively engage stakeholders in developing multifaceted solutions for complex health interventions. Results:Six themes emerged from the thematic analysis: (1) importance of patient-centered approach, (2) clinical evaluation and patient selection, (3) recommendations for task design, (4) guidelines for structuring BCI intervention, (5) key factors influencing motivation, and (6) technology features. From these themes, the following recommendations (R) are established: (R1) MI-based VR-BCI interventions must be conducted through a patient-centered approach, based on individualized preferences, needs, and goals of the user, by an interdisciplinary team; (R2) selection criteria must include upper limb impairment, cognitive and communication assessment, and clinical traits, such as MI capacity, neglect, and depression must be assessed since they might influence intervention outcomes; (R3) tasks to perform should preferably be based on daily living activities, including unilateral and bilateral tasks, and a variety of tasks must be available for selection to ensure meaningfulness for the user and suitability to clinical traits; (R4) intervention must be structured by different progressing levels starting with simple, gross movements and adding complexity through additional movement features, cognitive demand, or MI difficulty; (R5) optimal levels of motivation must be sustained through task variability, gamification elements, and task demand adequacy; and (R6) multisensorial potential of MI-based VR-BCI must be effectively harnessed through the adequate adjustment of visual, haptic, and proprioceptive feedback modalities to the patient. Conclusions:Current results contribute to establishing clear guidelines on patient selection, task design, intervention structuring, motivation factors, and tailoring of sensory feedback. This framework presents a foundation for optimal implementation of VR-BCI-based interventions that associate MI and motor observation, optimizing cortical activity during the intervention, patients' engagement, and clinical outcomes. Future research should explore the application of these guidelines for validation and investigate BCIs' efficacy according to different combinations of patients' profiles, task characteristics, and technology features.

BackgroundVirtual Reality (VR) feedback is increasingly integrated into Brain-Computer Interface (BCI) applications, enhancing the Sense of Embodiment (SoE) toward virtual avatars and fostering more vivid motor imagery (MI). VR-based MI-BCIs hold promise for motor rehabilitation, but their effectiveness depends on neurofeedback quality. Although SoE may enhance MI training, its role as a priming strategy prior to VR-BCI has not been systematically examined, as prior work assessed embodiment only after interaction. This study investigates whether embodiment priming influences MI-BCI outcomes, focusing on event-related desynchronization (ERD) and BCI performance.MethodsUsing a within-subject design, we combined data from a pilot study with an extended experiment, yielding 39 participants. Each completed an embodiment induction phase followed by MI training with EEG recordings. ERD and lateralization indices were analyzed across conditions to test the effect of prior embodiment.ResultsEmbodiment induction reliably increased SoE, yet no significant ERD differences were found between embodied and control conditions. However, lateralization indices showed greater variability in the embodied condition, suggesting individual differences in integrating embodied feedback.ConclusionOverall, findings indicate that real-time VR-based feedback during training, rather than prior embodiment, is the main driver of MI-BCI performance improvements. These results corroborate earlier findings that real-time rendering of embodied feedback during MI-BCI training constitutes the primary mechanism supporting performance gains, while highlighting the complex role of embodiment in VR-based MI-BCIs.

Brain-Computer Interfaces (BCIs) can provide a non-muscular communication channel for individuals with motor impairments. When integrated with virtual reality (VR) and haptic feedback, motor imagery (MI)-based BCIs can augment the rehabilitation outcome for patients with severe motor impairments. However, the physiological impact of these protocols beyond brain-related signals, that reflect autonomic nervous system (ANS) activity, remains underexplored. This study aims to investigate variations in a broader range of physiological signals besides electroencephalography (EEG) - including electrocardiography (ECG), photoplethysmography (PPG), and respiration - across different experimental conditions and to identify the factors driving these changes. 19 healthy subjects underwent MI training across five combinations of feedback conditions: abstract vs. realistic feedback, head-mounted display (HMD) vs. monitor, and the presence or absence of haptic feedback, compared with motor execution data. PPG results were compared with ECG results to assess the reliability of the finger-clip PPG sensor regarding its ability to replace ECG in cases where ease of use and unobtrusiveness in heart monitoring are required. Current findings show that VR-based MI with haptic feedback, results in increased modulation of Beta and Gamma bands, while all conditions may impose a greater mental burden than motor execution, as indicated by the increased respiration rate and decreased heart-rate variability.

Unilateral spatial neglect (USN) is a complex spatial attentional disorder consisting of a failure to attend to the contralesional side of space, frequently seen after a stroke. However, the majority of cases go undiagnosed due to the lack of a valid and reliable tool that is able to assess USN and its many variants. Recent technological advances in virtual reality (VR) and physiological sensors, allow for the study of this disorder under controlled, and ecologically-valid environments, which hold the promise of reliable and early detection. This proof of concept study aims to evaluate the feasibility of a system for discriminating different attentional states using a multimodal dataset derived from a spatial attention task conducted in VR. Nine healthy young adults underwent two experimental conditions: a Control condition and a Left Occlusion condition. Participants performed a visual search task while their behavioral data, including performance metrics, eye-gaze, head, and controller movement data, were recorded. Additionally, electroencephalography data was synchroniously collected to capture neural correlates of attentional processing. Analysis of results of this within-subjects study found worse performance (higher RT), changes in behavior (right-ward gaze bias, left-ward bias in head and controller movement) in the Left Occlusion condition. Neural differences were found (parieto-occipital mean alpha band power and event related potentials) between the two conditions. If validated, this system could be utilized as a diagnostic VR tool, while it holds the potential to facilitate the participation of stroke patients with USN in VR-driven rehabilitation.

The classification of EEG signals during motor imagery (MI) tasks is a key element of several brain-computer interfaces (BCIs), especially those used for motor rehabilitation of stroke patients. Despite a large body of literature, the classification of MI EEG signals remains challenging, with deep learning approaches showing improved performance but only limited success in real-life applications. This study focuses on two convolutional neural networks (CNNs) developed for classifying EEG signals, EEGNet and EDPNet, and aims to investigate the impact of different aspects of data degradation in real-life acquisitions on the accuracy of left vs. right MI classification. For this purpose, we consider a large public dataset with a well-defined MI task, as well as a private dataset with fewer trials across different types of MI tasks. We further test the impact of reducing the number of channels and time samples used, and compare the performance of subject-specific and group models. High accuracy was achieved using both CNNs in both datasets, with slightly better performance for subject-specific models compared to group models, highlighting high inter-subject variability. Interestingly, reducing the number of trials in the public dataset to match those in the private dataset yielded only a small decrease in performance, consistently with the similar performance obtained between datasets despite the difference in the number of trials. Both reducing the number of channels (from the total of 18 or 32 to only C3, C4 and Cz) and time samples, or trial duration (from 5.5 to 1.0 seconds) decreased performance, with the latter producing the greatest degradation. Nevertheless, even when using only 1.0 seconds of data from C3, C4 and Cz channels, model accuracy remained always above chance level, with a median of approximately 0.8 across subjects. In conclusion, we show that EEGNet and EDPNet are appropriate for MI EEG BCIs, yielding high left vs. right classification accuracies within a 1-second interval and with a limited number of channels.

A growing interest has developed in the problem of training models of EEG features to predict brain activity measured using fMRI, i.e. the problem of EEG-to-fMRI synthesis. Despite some reported success, the statistical significance and generalizability of EEG-to-fMRI predictions remains to be fully demonstrated. Here, we investigate the predictive power of EEG for both task-evoked and spontaneous activity of the somatomotor network measured by fMRI, based on data collected from healthy subjects in two different sessions. We trained subject-specific distributed-lag linear models of time-varying, multi-channel EEG spectral power using Sparse Group LASSO regularization, and we showed that learned models outperformed conventional EEG somatomotor rhythm predictors as well as massive univariate correlation models. Furthermore, we showed that learned models were statistically significantly better than appropriate null models in most subjects and conditions, although less frequently for spontaneous compared to task-evoked activity. Critically, predictions improved significantly when training and testing on data acquired in the same session relative to across sessions, highlighting the importance of temporally separating the collection of train and test data to avoid data leakage and optimistic bias in model generalization. In sum, while we demonstrate that EEG models can provide fMRI predictions with statistical significance, we also show that predictive power is impaired for spontaneous fluctuations in brain activity and for models trained on data acquired in a different session. Our findings highlight the need to explicitly consider these often overlooked issues in the growing literature of EEG-to-fMRI synthesis.

The Sense of Embodiment (SoE) refers to the subjective experience of perceiving a non-biological body part as one's own. Virtual Reality (VR) provides a powerful platform to manipulate SoE, making it a crucial factor in immersive human-computer interaction. This becomes particularly relevant in Electroencephalography (EEG)-based Brain-Computer Interfaces (BCIs), especially motor imagery (MI)-BCIs, which harness brain activity to enable users to control virtual avatars in a self-paced manner. In such systems, a strong SoE can significantly enhance user engagement, control accuracy, and the overall effectiveness of the interface. However, SoE assessment remains largely subjective, relying on questionnaires, as no definitive EEG biomarkers have been established. Additionally, methodological inconsistencies across studies introduce biases that hinder biomarker identification. This study aimed to identify EEG-based SoE biomarkers by analyzing frequency band changes in a combined dataset of 41 participants under standardized experimental conditions. Participants underwent virtual SoE induction and disruption using multisensory triggers, with a validated questionnaire confirming the illusion. Results revealed a significant increase in Beta and Gamma power over the occipital lobe, suggesting these as potential EEG biomarkers for SoE. The findings underscore the occipital lobe's role in multisensory integration and sensorimotor synchronization, supporting the theoretical framework of SoE. However, no single frequency band or brain region fully explains SoE. Instead, it emerges as a complex, dynamic process evolving across time, frequency, and spatial domains, necessitating a comprehensive approach that considers interactions across multiple neural networks.

Brain-Computer Interfaces (BCIs) enable direct communication between the brain and external devices, offering significant potential for rehabilitation and assistive technologies. A major challenge is the performance gap between BCI training and real-time control due to the variability of the underlying brain signals measured using EEG and user mental strategies. Virtual Reality (VR) plays a crucial role in BCI training by providing immersive, embodied feedback, enhancing user engagement, and potentially improving performance. This study investigates how closed-loop BCI systems, based on motor imagery (MI) of left and right-hand movements, with different VR feedback modalities-comparing true vs. positive feedback-impact EEG features like Event-Related Desynchronization (ERD) and BCI performance. Fifteen participants performed MI BCI training and control, in a VR environment, designed to induce increased sense of embodiment. Our results show consistent classification accuracy and ERD levels across both feedback conditions, indicating that VR feedback, when embodying a virtual body, whether true or positive, supports stable BCI performance. Additionally, key discriminative features emerged outside conventional ERD regions, highlighting the value of exploring non-traditional EEG features for MI task differentiation. This study underscores the importance of VR in optimizing closed-loop BCI systems and improving understanding of neurophysiological responses during MI tasks.

Simultaneous EEG-fMRI acquisitions leverage the complementary strengths of the two functional neuroimaging modalities, with promising applications in the development of neurofeedback (NF) brain-computer interfaces (BCIs). While fMRI provides superior mapping of brain activity, EEG is more accessible for NF-BCI interventions. Previous work has attempted to identify the EEG features that best predict fMRI activity patterns, but the performance achieved is still poor. In this work, we leverage a well-established deep learning network for the classification of EEG signals, EEGNet, and propose an extension to the regression task of predicting the fMRI signal at a specific time point from concurrent EEG data (R-EEGNet). We target the activity of the somatomotor network (SMN) during the execution of two motor imagery (MI) tasks used in NF-BCIs for motor rehabilitation in stroke patients. For this purpose, we use a simultaneous EEG-fMRI dataset collected from 15 healthy subjects while executing the tasks in two separate sessions. The fMRI data are analyzed to extract a time series of MI activity for each subject and task, and the R-EEGNet model is trained to predict each fMRI time sample from a $\mathbf{1 5}$-seconds segment of multi-channel EEG data. We evaluated the proposed R-EEGNet model performance in comparison with a conventional machine learning model (Group Lasso) trained on EEG spectral features, as well as with a Naïve model based on the EEG somatomotor rhythm. We found that R-EEGNet achieved a similar performance to Group Lasso, both being significantly superior to the Naïve model. Our results provide the first demonstration of the ability of a subject-specific deep learning model to predict fMRI motor signals based directly on the EEG signal, without the need to extract spectral features. Future work should improve model performance through further hyperparameter optimization and the exploitation of data augmentation to cope with the typically small size of EEG-fMRI datasets.

Traditional Mirror Therapy (MT) uses the reflection of an unaffected limb to promote neural recovery in the affected limb post-stroke. Robotic Mirror Therapy (RMT) has since emerged with robotics development, but current devices are often unsuitable for Point-Of-Care (POC) due to their cost, complexity, and lack of user-friendliness. To address these issues, we developed a portable, affordable, and user-friendly hand-finger rehabilitation system integrating a 21-Degree-Of-Freedom (DoF) parallel manipulator (5 active DoF), a motion sensor, and a Virtual Reality (VR) piano task. The VR tasks, inspired by RMT, Robotic Therapy (RT), and MT, leverage Discrete Sequence Progress Tasks (DSPTs) to engage brain regions responsible for sequential actions, critical for Activities of Daily Living (ADL). Operating in a student-teacher configuration, the system mirrors movements of the unaffected hand onto the affected hand, forming a neurological loop. In preliminary tests with 16 healthy participants, our system demonstrated increased motor skill acquisition compared to RT.

IntroductionRestorative Brain-Computer Interfaces (BCIs) provide an alternative non-muscular channel for stroke patients lacking volitional movement by enhancing sensorimotor rhythm modulation through motor imagery (MI). MI practice can be augmented with embodied feedback via virtual reality (VR). However, the clinical impact of embodied VR-BCI training remains under-explored.MethodsThis study examines the effects of embodied VR-BCI training on brain activity patterns (measured through EEG and fMRI) and clinical outcomes (assessed by the Fugl-Meyer Assessment, FMA) in four chronic stroke patients. Over a 3-week MI-BCI intervention, patients performed a bimanual rowing task (NeuRow) in VR. EEG data were used to extract Event-related desynchronization (ERD) and lateralization indices, while fMRI data focused on the primary motor (M1) and supplementary motor (SMA) regions of interest.ResultsResults indicated that all patients significantly induced ERD power, though the affected side exhibited reduced ERD compared to baseline during contralateral MI. Post-intervention, significant ERD differences from both hemispheres were observed, with decreased ERD correlating with no clinical improvement. Patients showing decreased ERD lateralization had no FMA score improvement. Activity within ipsilesional M1 and SMA correlated with FMA scores.ConclusionsThe findings suggest a relationship between brain activity and clinical outcomes, highlighting that increased ERD lateralization is associated with clinical improvement.

Objective: To examine the clinical effects of combining motor imagery-based neurofeedback training with bilateral repetitive transcranial magnetic stimulation for upper limb motor function in subacute and chronic stroke. Design: Clinical trial following an AB/BA crossover design with counterbalanced assignment. Subjects: Twenty individuals with subacute (n = 4) or chronic stroke (n = 16). Methods: Ten consecutive sessions of bilateral repetitive transcranial magnetic stimulation alone (therapy A) were compared vs a combination of10 consecutive sessions of bilateral repetitive transcranial magnetic stimulation with 12 non-consecutive sessions of motor imagery-based neurofeedback training (therapy B). Patients received both therapies (1-month washout period), in sequence AB or BA. Participants were assessed before and after each therapy and at 15-days follow-up, using the Fugl-Meyer Assessment-upper limb, hand-grip strength, and the Nottingham Sensory Assessment as primary outcome measures. Results: Both therapies resulted in improved functionality and sensory function. Therapy B consistently exhibited superior effects compared with therapy A, according to Fugl-Meyer Assessment and tactile and kinaesthetic sensory function across multiple time-points, irrespective of treatment sequence. No statistically significant differences between therapies were found for hand-grip strength. Conclusion: Following subacute and chronic stroke, integrating bilateral repetitive transcranial magnetic stimulation and motor imagery-based neurofeedback training has the potential to enhance functional performance compared with using bilateral repetitive transcranial magnetic stimulation alone in upper limb recovery.

Motor-imagery brain-computer interfaces (MI-BCIs) have the potential to improve motor function in individuals with neurological disorders. Their effectiveness relies on patients’ ability to generate reliable MI-related electroencephalography (EEG) patterns, which can be influenced by the quality of neurofeedback. Virtual Reality (VR) has emerged as a promising tool for enhancing proprioceptive feedback due to its ability to induce a sense of embodiment (SoE), where individuals perceive a virtual body as their own. Although prior research has highlighted the importance of SoE in enhancing MI skills and BCI performance, to date, no study has successfully isolated nor manipulated the SoE in VR before MI training, creating a gap in our understanding of the precise role of the priming effect of embodiment in MI-BCIs. In this study, we aimed to examine whether the virtual SoE when induced, as priming of avatar embodiment, and assessed before MI training, could enhance MI-induced EEG patterns. To achieve this, we divided 26 healthy participants into two groups: the embodied group, which experienced SoE with an avatar before undergoing VR-based MI training, and the non-embodied group, which underwent the same MI training without a prior embodiment phase, serving as a control. We analyzed subjective measures of embodiment, the event-related desynchronization (ERD) power of the sensorimotor rhythms, lateralization of ERD, and offline classification BCI accuracy. Although the embodiment phase effectively induced SoE in the embodied group, both groups exhibited similar MI-induced ERD patterns and BCI classification accuracy. This suggests that the induction of SoE prior to MI training may not significantly influence the training outcomes. Instead, it appears that the integration of embodied VR feedback during MI training itself is sufficient to induce appropriate ERD, as evidenced by previous research.

Brain-computer interfaces (BCIs) can provide a non-muscular channel of control to stroke patients for motor rehabilitation. This can be achieved through the use of motor imagery (MI) training, involving the modulation of sensorimotor rhythms. The practice of MI has been shown to be able to strengthen key motor pathways when reinforced with rewarding feedback. Recently, there has been a growing evidence of the positive impact of embodied virtual reality (VR) and vibrotactile feedback in MI training. Nonetheless, it is not yet clear what the optimal MI-BCI setup is for evoking stronger sensorimotor rhythms in VR. In this study, we investigate the impact of head-mounted VR, and vibrotactile feedback during MI-BCI training in the induced sensorimotor rhythms. To achieve this, 19 healthy subjects performed MI training with embodied VR between four conditions: head-mounted vs. screen VR, with and without vibrotactile feedback; and two control conditions: abstract MI without embodied feedback, and motor execution. The event-related desynchronization (ERD) and the lateralization indices (LI) of the Alpha and Beta EEG rhythms were analyzed in a within-subject design. Results show that the combination of vibrotactile feedback and embodied VR can induce stronger and more lateralized Alpha ERD; nonetheless, LI was not significantly different across conditions.

Open hardware and the need for ecologically valid measurements drive the Electroencephalography (EEG) democratization movement-EEG has been steadily transcending the boundaries of clinical research, making its way into interdisciplinary fields. In Human-Computer Interaction (HCI), EEG is used to measure cognitive workload and infer cognitive processes for building cognition-aware systems. We describe and evaluate our BCIglass prototype where EEG electrodes are embedded in the frame of a mainstream Head-Mounted Display (HMD) to create a skull-peripheral topology. We devised a lab study with 34 participants who completed seven established cognitive tasks. Then, we conducted a pilot field study with one participant to test BCIglass in everyday-life settings. Our findings demonstrate that BCIglass captures EEG activity in a manner comparable to a research-grade EEG-cap system. Our topology infers the cognitive task at hand, and the underlying cognitive process(es) by proxy, with an accuracy of similar to 80% and only three electrodes at the skull periphery. Embedding EEG electrodes in lightweight HMDs represents a promising approach in the quest to achieve ubiquitous brain-computer interfacing in real-world settings.

As robots become integral to various sectors, improving human-robot collaboration is crucial, particularly in anticipating human actions to enhance safety and efficiency. Electroencephalographic (EEG) signals offer a promising solution, as they can detect brain activity preceding movement by over a second, enabling predictive capabilities in robots. This study explores how EEG can be used for action anticipation in human-robot interaction (HRI), leveraging its high temporal resolution and modern deep learning techniques. We evaluated multiple Deep Learning classification models on a motor imagery (MI) dataset, achieving up to 80.90% accuracy. These results were further validated in a pilot experiment, where actions were accurately predicted several hundred milliseconds before execution. This research demonstrates the potential of combining EEG with deep learning to enhance real-time collaborative tasks, paving the way for safer and more efficient human-robot interactions.