Peripheral magnetic stimulation (PMS) has emerged as a non-contact neuromodulation technique for rehabilitation and neuroprosthetic applications. However, its practical use remains limited by high power consumption. While coil geometry optimization has been explored, the influence of individual pulse waveform phases on EMG-based functional responses remains largely uncharacterized in PMS. In this study, we present a dual-coil PMS architecture designed to enhance the axial electric-field gradient along the nerve and systematically investigate how pulse waveform parameters influence evoked responses and energy efficiency. Using an in vivo rat common peroneal nerve model, we independently varied the rising phase (RP), maintaining phase (MP), and falling phase (FP) of rectangular current pulses and quantified electromyographic (EMG) responses and per-pulse energy consumption. Decreasing MP reduced EMG amplitude, whereas increasing MP enhanced recruitment until saturation. Under the fixed-current condition, an MP of 100-150 μs provided a practical low-energy range while maintaining a robust EMG-based response. Meanwhile, the energy-normalized response decreased monotonically as RP and FP increased, indicating that shorter transition phases are consistently advantageous from an energy-efficiency standpoint. The timing of the EMG-based functional response was most consistent with initiation by the first induced component, whereas the second component appeared to modulate the final response magnitude. Complementary electromagnetic simulations further showed that the nerve-axis electric-field component generated by the dual-coil configuration partially corresponds to that of a monopolar stimulation profile, providing a structural context for interpreting the observed waveform-dependent responses. Together, these results provide experimentally grounded pulse-design guidelines for energy-constrained PMS systems.
For blind or visually impaired (BVI), the acquisition of three-dimensional (3D) spatial relationships highly depends on touch, proprioception, and active movement experience. Existing training methods are mostly based on 2D tactile graphics or static models and therefore have difficulty providing continuous, controllable, and quantifiable 3D body-motion guidance. To address this issue, this study proposes a 3D spatial perception training method based on cable-driven parallel robot (CDPR) position guidance and virtual-constraint feedback. The CDPR serves as a force-feedback interaction platform, integrating visual pose measurement, a virtual-constraint force field, tension mapping, and hybrid force-position control to transform spatial information into continuous and perceivable bodymotion guidance. Device performance experiments showed that the 3D tracking root-mean-square errors (RMSEs) for a circular trajectory with a diameter of 100 mm and a square trajectory with a side length of 100 mm were 4.35 mm and 3.31 mm, respectively. When the desired constraint force was 20 N, the measured force-output RMSE was 1.13 N. Furthermore, 24 BVI subjects were recruited for a two-group parallel controlled experiment to compare the CDPR training group and the control group in direction perception, distance perception, 3D endpoint localization, and trajectory reproduction tasks. The results showed that the CDPR training group achieved significantly greater improvements in all four tasks and maintained lower errors in untrained but related transfer tasks. Trajectory-to-object matching results showed that subjects could use CDPR-demonstrated geometric trajectories to identify corresponding physical objects through haptic exploration.
This study presents a novel semi-locked myoelectric pattern recognition algorithm (Onset-Triggered Vote-Locking or OTVL) to reduce misclassifications that typically occur during transition phases of muscular contractions for prosthetic hand control. It segments the electromyographic signals in real-time by adaptively defining enabling and disabling conditions at the onset of a muscular contraction in order to lock classifier predictions only when they are considered stable. The segmentation is based on signal standard deviation to improve robustness and reduce the need for rest transitions between movements. The proposed algorithm was implemented with a standard Linear Discriminant Analysis (LDA). It was functionally assessed by simulating activities of daily living and compared against a continuous LDA classifier with two conventional post-processing algorithms: Majority Vote (MV) and Confidence Rejection (CR). Eleven able-bodied participants executed ten repetitions of three Southampton Hand Assessment Procedure (SHAP) tasks using a bypass socket equipped with an active wrist and a multi-articulated hand. The OTVL algorithm significantly outperformed CR in terms of success rate (median:IQR; OTVL: 100.0%:0%; CR: 10.0%:60.0%; p < 0.001). Compared to MV, OTVL significantly reduced total completion time (OTVL: 56.4s:20.6s; MV: 62.2s:7.7s; p < 0.05) and the number of incorrect preshapes (OTVL: 0.6:0.4; MV: 1.5:1.3; p < 0.01) that occurred between tasks. The perceived workload, as assessed by the NASA Task Load Index, was also significantly reduced compared to MV and CR (OTVL: 52:33.5; MV: 67:21.5; CR: 95:36.3; p < 0.001). Therefore, the OTVL algorithm represents a viable alternative to traditional continuous classifiers, improving control stability and task execution.
Electrode arrays along peripheral nerves can record sensory, motor and autonomic signals propagating in afferent and efferent direction at different conduction velocities. To extract this temporal information in a delay-and-sum approach, velocity-selective recording (VSR) utilizes summation to amplify time-aligned action potentials and average out noise. This improved contrast subsequently improves the automated detection and classification of action potentials - an essential prerequisite to interface with the peripheral nervous system. In this study, we replaced the summation with multiplication and investigated resulting signal changes. We focused on two derived methods: Lock-IN-Dispersion-Analysis (LINDA) multiplies all recorded signals; Delay-Combinatory-Multiply-And-Sum (DCMAS) multiplies all combinations of an adjustable number of signals and sums the products. Processing simulated neural array recordings, we compared the effects of LINDA and DCMAS on contrast-to-noise ratio (CNR), signal-to-noise-and distortion ratio (SINAD) and AP detectability after thresholding. Compared to VSR, both methods increased CNR, SINAD, and AP detectability but also introduced noise intermodulation distortion which limited SINAD gains. Additionally, LINDA formed a logical AND-gate between signals, suppressing interfering signal peaks. Overall, by increasing CNR, signal multiplication may improve decoding of array recordings, despite increased signal distortion. It may thereby advance neural interfaces for fundamental research, diagnostics or applications in bioelectronic medicine and neuroprostheses.
Data-driven approaches are increasingly being adopted in human motion analysis research. Joint moment estimation plays a crucial role in providing valuable insights for robotic design and rehabilitation-related applications. Estimating joint moments using surface electromyography (sEMG) offers the advantages of being non-invasive and cost-effective. However, the end-to-end design of deep learning models for joint moment estimation faces several challenges, such as the requirement for large datasets, model generalizability, and interpretability. To address these challenges, this paper presents a physics-informed neural network (PINN)-based framework for knee moment prediction, designed using the sit-to-stand process of fifteen healthy individuals. The proposed model integrates Hill-type muscle dynamics and enforces constraints on model outputs through a physics-informed composite loss function. Using five-channel lower-limb sEMG signals as input, the model predicts knee joint moments and provides additional information on muscle activations and forces, offering biomechanical insights. The primary contribution of this design strategy lies in its capability to predict knee joint moments from sEMG signals while providing physiologically interpretable intermediate outputs. Our model achieved an average coefficient of determination (R²) of 0.919 ± 0.059 and an average root mean square error (RMSE) of 4.438 ± 3.136 Nm in leave-one-subject-out cross-validation across fifteen subjects. This work introduces a PINN design strategy that improves knee moment estimation accuracy while maintaining biomechanical plausibility within the evaluated sit-to-stand task.
Conventional crutch-based control for lowerlimb exoskeletons often imposes a considerable physical burden and limits usability. To address these challenges, we developed a hybrid brain-computer interface (BCI) combining a steady-state visual evoked potential (SSVEP)- based BCI with asynchronous biosignal-based switches triggered by a wink and teeth clench. Practical usability was improved by implementing a wearable headband-type biosignal-recording device to acquire electroencephalography, electromyography, and electrooculography signals. Augmented reality glasses were used to present visual stimuli and gait guidance information. To support robust exoskeleton control in a wearable BCI environment, we proposed an asynchronous operational framework in which SSVEP responses were used for movement-mode selection, whereas wink- and clench-based switches were assigned to command execution and cancellation, respectively. Ten participants completed real-time walking experiments while wearing a custom lower-limb exoskeleton using both the conventional crutch-based and proposed control methods. The performance of the proposed system was evaluated using BCI classification accuracy and F1-scores for two asynchronous switches, whereas usability and workload were assessed using the system usability scale (SUS) and NASA task load index (NASA-TLX), respectively. Despite gross body movement during exoskeleton-assisted walking, the proposed hybrid control framework demonstrated robust mode selection, execution, and cancellation with an average SSVEP classification accuracy of 95.06%, F1-scores of 99.80% and 99.22% for the wink- and clench-based switches, respectively. Notably, only two false positive events were observed per switch across all participants. Furthermore, the proposed method exhibited a significantly higher SUS score than the crutch-based control method (78.25 vs. 53.50; p < 0.01) and a significantly lower physical demand in the NASATLX (3.15 vs. 7.85; p < 0.05), confirming its potential as a practical alternative. To the best of our knowledge, this is among the first studies in which a wearable hybrid SSVEPbased BCI for lower-limb exoskeleton operation applicable to real-world walking tasks was systematically demonstrated. Our findings suggest that the proposed hybrid BCI is a robust and less physically demanding alternative to a conventional control method, offering strong potential for daily assistance and gait rehabilitation.
Characterizing human shoulder joint stiffness provides insight into how the shoulder contributes to arm stability under varying task demands. This study characterizes shoulder joint stiffness under two different task conditions—dynamic movement and static posture—in 3D space, and examines differences in stiffness between male and female participants. During dynamic movement trials, participants performed flexion–extension movements while wearing a parallel-actuated shoulder exoskeleton robot, whereas during static posture trials, they maintained fixed arm postures. In both conditions, the robot applied controlled position perturbations at the shoulder joint in the horizontal flexion–extension direction, and the resulting output torques were measured. Data collected from 20 participants (10 males and 10 females) showed that task condition significantly affected shoulder joint stiffness. Specifically, across three different horizontal flexion– extension arm postures, shoulder joint stiffness during dynamic movement was, on average, 23.1% lower than stiffness during static posture. Furthermore, a significant sex-based difference in shoulder joint stiffness was observed during static postures, whereas no significant difference was found during dynamic movement. These findings enhance the understanding of task-dependent modulation of shoulder joint mechanics and have important implications for quantifying altered shoulder stiffness following neurological impairments and for developing task-specific, individualized rehabilitation and assistive strategies.
Prosthetic ankle-foot design is a rapidly evolving field encompassing three principal device categories: passive, active, and quasi-passive. Although each category has generated substantial published literature, this body of work remains fragmented, leaving clinicians unaware of emerging research developments and engineers disconnected from clinical practice. This disconnect ultimately impedes the translation of insights between disciplines. Here, we synthesize principles from modern prosthetic ankle-foot design, establishing connections among biomechanical objectives, clinical evaluation methodologies, and engineering design fundamentals across devices used in both clinical practice and research settings.
Amyotrophic Lateral Sclerosis (ALS) is an incurable neurodegenerative disease characterized by the selective loss of spinal motor neurons (MN). Trans-spinal direct current stimulation (tsDCS) has emerged as a promising noninvasive neuromodulation technique that could provide neuroprotection and slow down disease progression. However, the mechanisms by which tsDCS affects individual MNs remain poorly understood. This study uses computational modeling to explore how low-intensity extracellular electric fields (EEFs), generated by tsDCS, influence the electrophysiological behavior of MNs. Morphologically realistic, multi-compartment models of neonate mouse alpha-MNs were developed in the NEURON simulation environment. Simulations were conducted under different EEF magnitudes, polarities and orientations. At the population level, consistent directional effects on excitability-related properties were limited and depended on stimulation condition. However, analyses of response magnitude showed that EEFs could modulate several properties, including resting membrane potential and rheobase, even when the direction of change varied across neurons. This heterogeneity was associated with neuronal morphology and its alignment with the applied field, with dendritic length and asymmetry affecting sensitivity. Overall, the results suggest that low-intensity EEFs produce modest, morphology-dependent modulation of MN electrophysiological properties, and that variability in neuronal structure and orientation may help explain discrepancies across previous experimental and modeling studies.
Conventional neuroimaging tools for post-stroke motor function evaluation (e.g., EEG, fMRI) have some constraints. Conversely, functional near-infrared spectroscopy (fNIRS) offers a viable compromise. Nevertheless, few studies have yet quantitatively assessed the current motor function scores based on fNIRS data. This study proposed a Graph Convolutional Network (GCN) and Support Vector Regression (SVR) fusion model to fit Fugl-Meyer Assessment (FMA) scores by leveraging a multi-state integration of fNIRS metrics and clinical indicators. After preprocessing, brain network features and GCN features were extracted from the fNIRS data. Then a modified forward search method was used for SVR model training and feature selection from three feature sets: resting-state/task-state fNIRS feature sets and clinical feature set. Finally, the SVR model was employed to estimate the FMA scores. The coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE)were utilized to evaluate the models. The GCN-SVR fusion model demonstrated good goodness-of-fit, and exhibited limited variability. The multi-state model demonstrated better performance than both single-state models (P<0.001). For cortical cases, Aggregate measures over the nine sparsity levels confirmed both high accuracy and stable fitting (R²=0.8397±0.0487, RMSE=6.75±1.06, MAE=4.91±0.94), with R² consistently above 0.76 across sparsity levels. In subcortical patients, the multi-state model achieved a mean R² of 0.7562±0.0185, with RMSE=10.51±0.40 and MAE=7.74±0.53. The proposed GCN‑SVR fusion algorithm based on fNIRS data achieved high accuracy and stable performance in fitting FMA scores, while subset‑based sequential forward selection enhances multi-dataset feature selection.
The soft tissue envelope within the residual limb influences prosthetic fit, function, and user comfort. The purpose of this study was to examine the influence of residual limb tissue composition and skeletal alignment on its mechanical properties during axial loading. We hypothesized that deviations in the adipose and muscle volume of the limb, as well as frontal plane alignment, will alter the mechanical stiffness and energy loss. We constructed multi-material synthetic residual limb models with varying tissue compositions and mechanically tested them under cyclical axial compression. Stiffness was quantified as the change in force divided by the change in displacement over the entire loading response, and energy loss was quantified as the mechanical hysteresis of the force-displacement loop. A large (~35%) reduction in soft tissue volume, representative of the change from pre- to post-limb recontouring surgery, resulted in an increase in stiffness of 57.6% out-of-socket and 17.3% in-socket relative to the pre-surgery model. Small (±2.5%) changes in tissue composition produced up to an 18.6% change in stiffness and up to a 59.2% change in hysteresis relative to the nominal post-surgery model. Finally, alignment in the frontal plane (±4° and ±8°) had significant effects on stiffness and hysteresis both pre- and post-surgery. We expect that clinical and technical approaches that optimize these factors could potentially contribute to improvements in whole-body outcomes such as metabolic cost and user comfort. In addition, these testing and verification methods offer a non-invasive and powerful technique to test patient-specific residual limb properties.
Objective: Precise localization of motor hotspots is critical for the therapeutic efficacy of transcranial magnetic stimulation (TMS). However, traditional manual methods are highly subjective and time-consuming, while existing automated approaches often struggle to balance efficiency and accuracy. This study aims to develop and validate an Automated Hotspot Search (AHS) algorithm to address these limitations. Methods: The AHS algorithm integrates robotic TMS with a Gaussian Process-based Bayesian Optimization (GP-BO) framework. Unlike exhaustive search approaches, AHS employs novel heuristics and a dynamic acquisition function to efficiently model cortical excitability, ensuring rapid convergence with minimal pulses. In an intra-subject study (n=11), we compared AHS with manual hotspot search (MHS) and semi-automatic hotspot search (SAHS) in terms of localization error, operational time, and physiological outcome (resting motor threshold, RMT). Results: AHS demonstrated significantly reduced localization error (6.59 ± 2.06 mm, p < 0.01) compared to MHS and SAHS. AHS search time (226.55 ± 47.52 s) was reduced by 37% and 61% relative to MHS and SAHS, respectively, exhibiting superior operational stability (p < 0.001). Furthermore, the RMT determined by AHS (57.00 ± 9.10% MSO) was significantly lower than that of MHS (61.36 ± 9.06% MSO, p = 0.009). A significant correlation was also found between localization error and RMT. Conclusion: The AHS algorithm provides a more precise, efficient, and physiologically valid solution for motor hotspot localization. Significance: This work provides an objective, data-driven tool that overcomes the subjectivity and inefficiency of manual operations, thereby enhancing the rigor and reproducibility of TMS applications.
Facial paralysis assessment and rehabilitation monitoring require not only accurate analysis of facial biomarkers but also a clinically deployable workflow that spans standardized data acquisition, quantitative modeling, and interpretable reporting. We present a clinically implemented facial paralysis assessment system that integrates four modules: a Data Acquisition module for standardized 4K/120 fps video recording, a Key Point Extraction and Indicator Calculation module that derives 313 fine-grained indicators from 11 standardized facial actions, an Analysis module built on the Hierarchical Dynamic Attention Patient-Adaptive Network (HiDAPA), and a Result Visualization Report Generation module for interpretable clinical reporting. At the core of the system, HiDAPA models biomarker importance in a hierarchical and personalized manner. It organizes the 313 indicators into a three-level representation to capture both structured biomarker relationships and individual heterogeneity. Using Sunnybrook scores as a clinical reference, experiments on 200 subjects show that HiDAPA achieves a Pearson correlation of 0.980 and an Acc@10 of 90.5%. The learned importance patterns are clinically interpretable, highlighting the dominance of symmetry-related indicators and spontaneous blink asymmetry. The system enables an objective and interpretable assessment of facial paralysis and supports long-term monitoring of rehabilitation progress.
Although movement efficiency and smoothness are fundamental optimality principles in human motor control, few human-robot interfaces effectively integrate these two principles during human-robot interaction. To address this, we proposed a biomimetic human-robot interface (BHRI) by cascading an EMG-driven neuromechanical model with a minimum-jerk motor behavior model. To decode continuous motion intentions, a neuromechanical model was established to estimate voluntary ankle torque based on EMG signals. The voluntary torque then served as the input for a motor behavior model, which generated robotic motion based on the minimum-jerk principle. We conducted robot voluntary control experiments among five healthy and five post-stroke subjects to validate the BHRI in an ankle rehabilitation robot. For comparison with the BHRI, an interaction torque-based minimum-jerk controller (IMC) and an EMG-driven admittance controller (EAM) were also implemented. Compared to IMC, the results demonstrated that our BHRI significantly reduced the interaction torque (by 40.7% and 25.0%), EPUD (by 55.9% and 51.1%), and averaged muscular effort (by 8.4% and 13.3%) for healthy and post-stroke subjects, respectively. Furthermore, when compared with the EAM, our BHRI achieved a significant reduction in the DSJ value of the generated desired trajectory (by 32.7% and 40.0%) for healthy and post-stroke subjects, respectively. For the DSJ of actual trajectory, the BHRI significantly reduced the DSJ value by 34.3% for post-stroke subjects compared to EAM. Therefore, our BHRI offers a biomimetic solution for natural human-robot cooperation by jointly optimizing movement efficiency and smoothness.
Accurate lower-limb joint moment estimation is essential for clinical gait analysis in children with cerebral palsy (CP), yet conventional methods require force plates for ground reaction force (GRF) measurement, limiting assessment to specialized laboratories. Although data-driven methods can bypass force plates, learning from limited, heterogeneous CP data is challenging because spasticity and impaired selective motor control alter neuromuscular coordination. We developed a biomechanics-embedded framework to estimate sagittal-plane hip, knee, and ankle moments from three joint angles and four surface electromyography (sEMG) channels. Joint-angle sequences are encoded by a Transformer and sEMG signals by a graph encoder. The fused representation is decoded by a learnable Hill-type pathway and a data-driven residual head. Under 23-fold leave-one-subject-out cross-validation, the final model achieved an overall Pearson correlation coefficient (PCC) of 0.826 and root mean square error (RMSE) of 0.156 N $\cdot $ m/kg. Before self-distillation, the model achieved a PCC of 0.824, compared with 0.771 for the best of 12 capacity-matched baselines. Adding the Hill-type pathway to the data-driven backbone increased PCC from 0.773 to 0.824. Two data-side strategies were examined under limited CP data, namely same-domain self-distillation and healthy-to-CP transfer. Self-distillation yielded a small PCC increase, whereas both healthy-to-CP transfer protocols reduced PCC relative to direct CP training. These findings support further evaluation of biomechanics-embedded models for GRF-free joint moment estimation in CP.
We characterised central nervous system reorganisation in rotator cuff injury (RCI) during three-plane antagonistic shoulder tasks using electroencephalography (EEG) connectomics, and tested whether graph metrics discriminate RCI from healthy controls (HC). We enrolled 51 RCI patients and 50 HCs performing three-plane tasks. EEG (64 channels) was preprocessed and source-localised; source-space connectivity was computed in five bands ( $\theta $ , $\alpha $ , $\beta $ , low- $\gamma $ , high- $\gamma $ ) using the weighted phase-lag index (WPLI). Whole-brain graph metrics (global efficiency (GE), characteristic path length (CPL), clustering coefficient (CC), modularity (Mod) and small-worldness (SW)) were derived, and network-based statistics (NBS) identified differential subnetworks from which mean subnetwork strength (WPLINBS) and related measures were extracted. Per action-band, single-feature linear support vector machines (SVMs) were evaluated with nested cross-validation for RCI-HC classification. Group differences were frequency-specific: $\beta $ (13-30 Hz) and low- $\gamma $ (30-50 Hz) showed RCI "over-integration/under-segregation" (higher GE and CC), whereas CPL, Mod and SW were higher in HC; $\theta $ / $\alpha $ effects were weak, whereas high- $\gamma $ effects were less robust and less consistent across shoulder tasks than the $\beta $ and low- $\gamma $ effects. NBS detected significant components only in $\beta $ /low- $\gamma $ , characterised by stronger coupling between occipital visual cortex and frontoparietal/cerebellar regions; within these components, WPLINBS, nodal strength and local efficiency were higher in RCI. CPL, GE or WPLINBS in $\beta $ /low- $\gamma $ provided robust classification (ACC $\approx 0.86$ ; AUC $\approx 0.90$ ). Overall, RCI is associated with $\beta $ /low- $\gamma $ -centred network reorganisation and enhanced occipito-parietal-frontal coupling, suggesting greater engagement of visual-related visuomotor networks during shoulder task execution. These EEG graph metrics may serve as candidate CNS biomarkers and potential targets for personalised rehabilitation.
Electroencephalography (EEG) is a common technique to measure field potentials of various brain regions, and event-related potentials (ERPs) reflect stimulus-or response-locked brain responses that are spatially distributed across the scalp due to volume conduction. However, accurately capturing stimulus specific ERPs from EEG remains a major challenge due to complex mixtures of ERPs from multiple stimulus sources. This study addresses the challenge of separating nonlinearly mixed ERPs in EEG data using synthetic and experimental datasets. We simulated synthetic ERP datasets with known ground truth, by projecting source signals through a linear forward model, followed by various nonlinear mixing functions. We implemented different models, including tree-based deep forests, convolutional neural network (CNN), long short-term memory (LSTM), Transformer, and hybrid CNN-LSTM architectures to recover the individual source-specific scalp ERP projections. Performance was assessed via 2D Pearson correlation between the predicted ERPs and the ground truth as well as the source localization accuracy. Our results showed that model architectures integrating spatial and temporal modeling, particularly CNN-LSTM combined models, achieved superior signal reconstruction and anatomical localization accuracy. These findings highlight the potential of deep learning methods for nonlinear source-projection separation in EEG data, and provide a framework for future applications under realistic neural constraints.
Powered knee prosthesis can enhance mobility by allowing voluntary control of knee flexion, thereby improving foot clearance for crossing obstacle. However, prior controllers have typically focused on crossing obstacle with the prosthesis serving as the trailing leg, while the scenario where the prosthesis leads has been overlooked. This paper introduces a self-adaptive volitional control method utilizing mechanical signals, enabling amputee to cross an obstacle with the prosthesis either as the leading or trailing leg, and even negotiate multiple continuous obstacles. In volitional controller, we partition the swing flexion phase into four conditions and calculate the desired maximum knee angle in real-time accordingly. We recruited amputees to participate in experiments, which included three distinct tasks: single obstacle crossing with the prosthesis leading, single obstacle crossing with the prosthesis trailing, and dual obstacles crossing. This work offers amputees capability to cross obstacle freely and naturally, a functionality that is currently unavailable.
Proprioceptive input is essential for motor control, yet clinical assessment of proprioceptive deficits relies largely on psychophysical tests that depend on self-report, are vulnerable to cognitive interference and bias, and probe conscious aspects of sensation. In this proof-of-concept study, we propose the automatic muscular response to passive limb displacement—the placing response—as a complementary, more objective marker of proprioceptive adequacy. Using surface electromyography (EMG) from the anterior and posterior deltoid during shoulder movements, we first confirm that placing is a genuine response in young and older healthy controls and in persons with stroke (PwS). We then use these data to derive and test a simple biomechanical model that predicts temporal synergy between these muscle heads under intact versus impaired proprioception. Group-level analyses show that an EMG-based metric differentiates between paretic and non-paretic arms in PwS and between PwS and controls. At the individual level, a classifier based on this physiological metric performs comparably to a psychophysical classifier but appears less susceptible to extreme “guessing-like” behavior. These findings suggest that automatic motor reactions to passive limb displacement offer a promising avenue for more comprehensive assessment of proprioceptive adequacy after neurological injury.
Motor imagery-based brain-computer interface (MI-BCI) is used in stroke rehabilitation to match brain activity with contingent feedback to establish closed-loop pathways and provide a measure of neuroplasticity changes in patients. However, most studies assessed neural function only at pre- and post-train, thereby longitudinal trends of neural patterns and mechanisms during full-process of intervention remain unclear. Fourteen stroke patients were recruited to receive a total of 8-session (2-week) MI-BCI-controlled “sixth-finger” intervention. Resting-state electroencephalography (EEG) and clinical scales, including the Fugl-Meyer Assessment (FMA-UE) and Barthel Index (BI), were evaluated pre- and post-train. Furthermore, MI tasks EEG signals throughout the full-process of intervention were tracked to reflect the longitudinal continuous trends of neural activity. EEG longitudinal trend shows two phases over full-process of intervention: event-related desynchronization (ERD) gradually increased in the first week of training, weakened and focused on the contralateral sensorimotor area in the second week, and showed a significant correlation over sessions. And resting-state functional connectivity increased after intervention. Motor function improved significantly from pre- to post-train by clinical metrics, with + 7.9 in FMA-UE and + 7.1 in BI. More than half of patients (9/14) reached the minimally clinically important difference (MCID) of 6.6 points change for FMA-UE after therapy. Meanwhile, the improvement of motor function is associated with the enhancement of resting-state functional connectivity. This work reveals longitudinal trend of neural patterns over full-process of intervention and its correlation with motor recovery, providing crucial evidence for understanding the mechanisms of neuroplasticity in stroke rehabilitation.
