IEEE Transactions on Neural Systems and Rehabilitation Engineering

Wearable inertial measurement units (IMUs) provide a practical and objective approach for gait assessment in clinical populations. Although several handcrafted gait features have been proposed, these features may not fully capture the multidimensional signal characteristics associated with different pathological gait patterns. This study proposes a digital biomarker called Embedding-Distance Gait Biomarker (EDGB) based on supervised contrastive representation learning of wearable IMU signals. A compact multi-input convolutional neural network is developed to encode raw acceleration, angular velocity, and their temporal derivatives into a 32-dimensional latent representation. Class-specific prototypes are computed from the training embeddings of healthy, neurological, and orthopedic participants. The proposed EDGB is then derived from the distances between each trial embedding and the learned group prototypes. The proposed architecture is evaluated on the publicly available Voisard clinical gait dataset using a subject-level split, with 20% of participants held out for testing to prevent leakage across repeated trials. On unseen test subjects, the proposed biomarker distinguished healthy from neurological, healthy from orthopedic, and neurological from orthopedic gait patterns with AUCs of 90.59%, 88.47%, and 99.50%, respectively. The biomarker also demonstrated a large group effect, with clinical category explaining 71% of its variance. Reliability analysis showed significant consistency across repeated trials, with an ICC (2,1) of 0.82, indicating that most variability reflected between-subject differences rather than within-subject trial-to-trial fluctuations.

Human-in-the-loop optimization (HILO) is an established method for identifying subject-specific optimal controllers for performance augmentation. For HILO algorithms to be useful in rehabilitation, however, the optimization algorithm may need to account for how the human response changes over time in response to assistance. In this study, we tested a modified version of Bayesian optimization (BO), dynamic Bayesian optimization (DBO), in a three-parameter optimization problem that sought to identify participant-specific optimal solutions for increasing walking speed. As opposed to BO, DBO accounts for the non-stationarity of human responses. Sixteen healthy participants received bilateral hip torque pulses delivered by a hip exoskeleton. The exoskeleton torque parameters were determined using HILO with either DBO or BO. Validation iterations were introduced to objectively compare performance across optimizers at different time points of HILO. The results showed that both DBO and BO significantly increased walking speed compared to baseline. When comparing performance between DBO and BO, DBO emerged as an improvement over BO both in terms of efficacy, modeling accuracy, and personalization. DBO induced changes in walking speed relative to baseline that exceeded those induced by BO at three of the four validation iterations. DBO outperformed BO in modeling accuracy in later validation iterations. DBO personalization induced changes in walking speed that were significantly greater than those induced by previously identified assistive solutions, while this was not the case of BO. Overall, our results indicate that DBO outperformed BO due to its greater ability to account for non-stationary aspects of the human response.

Older adults face an increased risk of falls that may have severe consequences for their well-being. Routine, accessible clinical screening may help mitigate fall risk through early detection of balance impairments. Portable force plates offer a convenient and practical solution for balance assessment in clinical settings. A new force-plate-based balance measure, the intersection-point-height, has shown particularly promising results in its ability to distinguish between healthy and impaired balance behaviors. However, the intersection-point-height measure requires measurement of shear force during standing, which exhibits magnitudes of less than 0.2% of normal forces (body weight), taxing the dynamic range of most sensor technologies. The ability of existing force plates to measure such low-magnitude shear forces observed during quiet standing is currently unknown. This study presents a force-plate performance assessment method to evaluate shear-force measurement errors and quantify the uncertainty of the intersection-point-height measure. The method was applied to test a commonly used laboratory-grade portable force plate. While the device successfully captured sagittal-plane intersection-point-height at the lowest frequencies, low signal strength prevented precise readings in the frontal plane. Thus, the tested device only marginally met the precision required for quiet-standing analysis, underscoring the critical need for systematic performance validation of portable force plates prior to clinical use. Future efforts should focus on evaluating alternative portable force plates and exploring economical design improvements to enhance shear-force measurement precision.

This study aims to predict human intentions during intense sports activities, specifically in table tennis. Using a publicly available Real World Table Tennis dataset containing simultaneous EEG and video recordings, we developed a series of participant-specific classifiers for nine players (7 males and 2 females; age range 18-30), based on pose features and EEG signals. The pose-based classifier used a stochastic gradient descent model with logistic loss, whereas the EEG-based classifier employed a modified convolutional neural network architecture (EEGNet). Both classifiers successfully predicted left-right attack intentions from the time windows preceding racket-ball impact, with optimal decoding occurring at -100 ms for pose features and -500 ms for EEG signals. EEG-based decoding achieved higher performance than pose-based decoding, and a multi-modal ensemble further improved prediction, reaching a mean macro F1 score of 0.563 (bootstrapped 95% CI: 0.523-0.603), corresponding to gains of +0.03 over pose-only and +0.02 over EEG-only classifiers. Because each classifier is trained independently, the ensemble can be feasibly extended to incorporate additional modalities in the future. These results suggest potential applications in neural prosthetic systems and neurofeedback tools for sports training.

High performance intracortical brain-computer interface (iBCI) control has been demonstrated in research settings, but performance can still vary within and between sessions. One potential source of this variability is the change in attentional load that comes from processing naturally occurring distractors such as thoughts, sounds, fatigue, or pain. To improve the consistency of iBCI performance in real-world environments where this sort of multi-tasking is inevitable, we must understand how shifts in attention can impact performance. Here we examined the effect of attentional load on iBCI performance and movement-related neural activity using a 2D cursor translation + click iBCI task paired with an N-Back working memory task to increase attentional load during dual-task performance. Two participants (P2 and P4) with tetraplegia completed the study while enrolled in a long-term clinical trial of an iBCI device (NCT1894802). Common neural correlates of attention (theta and alpha band power) were measured with simultaneously recorded scalp electroencephalography (EEG). While the EEG recordings and difficulty ratings suggested increased attentional load during dual tasking, iBCI performance was quite robust across the various dual tasking conditions. One participant, P2, experienced a small but significant increase in trial completion time and normalized path length during the mild attentional load condition. Signal quality differences between the two participants may have impacted the results, as P2 had lower signal quality and was therefore likely more vulnerable to attentional load. P4s higher signal quality likely allowed him to accommodate increased attentional load without a drop in performance. Overall, iBCI performance appears to be robust to attentional load, but the complex trends observed here reflect a need for continued investigation of BCI use under different cognitive states to elucidate potential challenges and compensatory mechanisms across participants.

The most prominent signature associated with motor execution and motor imagery is the event-related desynchronisation and synchronisation (ERD/S) in the mu and beta bands (8-30 Hz). In the context of brain-computer interfaces (BCI), this ERD/S signature is helpful for binary decisions, such as left vs. right imagery, but it is not a robust biomarker for continuous prediction, such as precisely decoding different levels of force application. This is essential for developing better BCI applications with precise dynamic force outputs. Recent studies have revealed that sensorimotor beta bursts have a stronger relationship with motor control, even at a single-trial level, than beta band power. We, therefore, investigated whether the transient nature of beta bursts provide an alternative, but robust biomarker for BCI force decoding. Here, we designed an experiment where human participants (N = 16) performed both motor execution (ME) at four force levels (10%, 25%, 50%, and 75% of maximum voluntary contraction) and imagined exerting the same, i.e. a motor imagery (MI) task, as their electroencephalogram was recorded. We observed a clear and classical ERD pattern in the motor cortex during the ME task, whereas it was less pronounced during the MI task. After extracting sensorimotor beta bursts, we observed differences in spectral burst features between motor execution and imagery including burst amplitude, spectral width, and temporal width. Moreover, different force levels were correlated with changes in the burst amplitude and burst spectral width, specifically during motor execution. Interestingly, we found that different beta burst waveforms are associated with the different force levels and conditions. This suggests that the bursts-level features could be driven by changes in the underlying beta burst waveforms. Overall, our study shows that sensorimotor beta burst can be an important piece of the puzzle to implementing precise force control in brain-computer interface-based prosthetics.

The study of human locomotion has long relied on descriptive frameworks of the gait cycle, which have provided essential insights into the functional phases of walking and their underlying biomechanical demands. While these models remain highly informative, they are largely based on observational analyses and may not fully capture the continuous, global coordination that characterizes human movement. The present study proposes an integrated framework to study whole-body coordination. This framework combines network theory with non-negative matrix factorization (NNMF) to treat gait as a dynamic system of coordinated joint interactions. Using three-dimensional kinematic data from 60 healthy subjects, we constructed a representation of whole-body coordination across time, named "dynamic kinectome". It was then decomposed using NNMF to extract spatial patterns of joint coordination and their corresponding temporal activations, allowing for an interpretable characterization of locomotor organization while preserving physiological meaning. Our analysis extracted six robust, highly consistent, and symmetrical coordination patterns across participants, effectively capturing the primary functional subtasks of locomotion. Rather than challenging classical phase descriptions, these findings enrich them by showing how coordination emerges as a continuous, often proactive process that can extend across conventional phase boundaries and systematically integrates the upper limbs for dynamic stability. Overall, this study provides a holistic, data-driven perspective on human locomotion, offering a promising basis for future investigations in motor control and may contribute to the development of sensitive biomarkers for clinical and rehabilitative applications.

PurposeEEG-based brain-machine interfaces (BMIs) may support assistive technologies for individuals with stroke-related motor impairment by translating neural activity into control commands for external devices. However, post-stroke neural reorganization and interindividual EEG variability challenge reliable decoding. This study characterized motor imagery EEG features in healthy and acute stroke participants and evaluated whether population-trained Q-learning Kernel Temporal Difference (Q-KTD) decoders could improve individual stroke decoding through transfer learning. These analyses assess the feasibility of healthy-to-stroke translation for EEG-based BMI neural decoding. Materials and MethodsPublicly available motor imagery EEG datasets from healthy participants (n = 109) and individuals with acute stroke (n = 50) were analyzed using left- and right-hand motor imagery trials. The datasets were selected because of their relatively large sample sizes and comparable motor imagery tasks. EEG characterization included baseline and motor imagery-period band power, ERD/ERS, hemispheric asymmetry, and time-frequency representations. For Q-learning Kernel Temporal Difference (Q-KTD) decoding, filtered time-domain EEG from 0- 0.5 s after motor imagery onset was used as the neural-state input. A Q-KTD model trained on the healthy population was transferred to individual stroke participants, and repeated Monte Carlo simulations compared decoding performance with and without transfer learning across multiple learning epochs. ResultsHealthy and acute stroke participants showed shared motor imagery-related EEG structure, including post-onset mu-band suppression, while the stroke group exhibited greater interparticipant variability, more diffuse time- frequency modulation, and altered hemispheric asymmetry. No channel-level healthy-stroke differences in windowed band power remained significant after false discovery rate correction. Healthy-source transfer learning improved first-epoch Q-KTD success rates in 29 of 50 stroke participants (58%). Across all participants, mean success rate increased from 49.46% without transfer learning to 51.82% with transfer learning. Among participants showing positive transfer, the mean gain was 7.34% and the maximum gain was 18.75%. However, 21 participants showed negative transfer, demonstrating substantial subject-level variability. ConclusionHealthy-source Q-KTD transfer learning improved first-epoch motor imagery BMI decoding for a majority of acute stroke participants, supporting the offline feasibility of population-informed Q-KTD decoding in stroke. These early performance gains may reduce subject-specific calibration burden, although substantial interparticipant variability and negative transfer indicate the need for individualized transfer-selection or adaptation strategies. Assistive Technology ImplicationsO_LIEEG-based brain-machine interfaces may support assistive technologies for individuals with stroke-related motor impairment by translating motor imagery-related neural activity into control commands for external devices. C_LIO_LIHealthy-to-stroke transfer learning may improve early BMI neural-decoder performance and potentially reduce the amount of subject-specific calibration required. C_LIO_LIThe findings support the offline feasibility of Q-KTD for motor imagery BMI neural decoding in individuals with acute stroke. C_LIO_LISubstantial interparticipant variability and negative transfer suggest that individualized source-model selection or adaptation strategies may be needed for reliable post-stroke BMI implementation. C_LIO_LIPhysiological EEG characteristics, including ERD/ERS and hemispheric asymmetry, may provide candidate markers for future transfer-selection strategies, although their predictive value requires direct validation. C_LI

Non-invasive brain-computer interfaces (BCIs) have substantially advanced in the field of continuous cursor control over the past decade. Yet, current methods lack key control aspects such as initiation and termination of cursor movements as well as evaluation in real-world applications. In this study, we introduce a framework for continuous, electroencephalography-based cursor control that supports both active movement and no-movement states, thereby allowing for inactive periods of the user when no control input is desired. We demonstrate its applicability in healthy participants and show its performance in real-world application through the selection of targets on a screen. This demonstrates that participants can leverage the continuous control cursor control and the intentional starting and stopping of motions to effectively select targets on a screen through dwell-time selection. On average, 7.1 out of 40 targets were correctly selected (level of significant performance: 4.5 targets), while experienced BCI users achieved an average of 12.8 targets. The proposed framework additionally demonstrates compatibility with motor-impaired people without residual hand motions since it does not rely on observable movements for model training.

Brain-computer interfaces face a fundamental trade-off between the signal fidelity and stimulation precision of noninvasive systems and the surgical burden and scalability of invasive systems. Non-invasive BCIs suffer from low signal quality and poor stimulation accuracy due to the skull barrier and the variability introduced by the scalp and skull. Existing invasive BCIs rely on traumatic surgical procedures or brain-penetrating electrodes, which limits their spatial extensibility, application, and patient acceptance. Here, we introduce a minimally invasive hybrid BCI architecture that uses the skull as a distributed interface layer rather than treating it solely as a barrier. The hybrid BCI comprises four integrated components: (1) the safe and smart micro-hole craniotomy; (2) distributed microelectrodes subcutaneously implanted in micro-holes in the skull with the distal end in contact with the dura; (3) an external bi-directional wearable headset for coupling, recording, stimulation, and channel selection; and (4) an AI-assisted planning and control agent. Animal studies have shown that micro-holes with a diameter of 300-800 m can be safely and conveniently prepared at any predefined locations across the skull without impairing the dura. In vivo experiments on rats demonstrate that the hybrid BCI with skull-implanted microelectrodes evidently increases resting-state spectral power and improves the signal-to-noise ratio of somatosensory and steady-state visual evoked responses compared to the scalp EEG; the computational modelling shows that distributed skull-dura microelectrodes can increase the intracranial electrical field strength and steer focused temporal-interference fields towards predefined deep brain targets. These findings will lay a solid foundation for future endeavors in wireless integration, safety evaluation and clinical benefits of the hybrid BCI. In summary, we propose the hybrid BCI as a distinct minimally invasive BCI paradigm with the great potential as a distributed, scalable, and upgradable neural interface that can expand the clinical application of minimally invasive BCI techniques.

Wearable exoskeletons are a promising tool for augmenting balance and reducing fall risk. Recent work suggests that active ankle exoskeletons need to act faster than the human to improve reactive balance control. However, the magnitude of exoskeleton torque that is best for improving reactive balance remains unknown. Drawing from the optimal torque for minimizing metabolic expenditure, we hypothesized that reactive balance would improve with increased exoskeleton torque. Participants wearing bilateral ankle exoskeletons were instructed to maintain standing balance during 15cm backward support-surface perturbations. Three exoskeleton plantarflexion torque conditions were tested: NO (Off), LOW (15Nm), or HIGH (30Nm). LOW torque improved balance performance compared to NO torque (p<0.001), with a 7{+/-}3% decrease in peak center of mass (CoM) displacement. Although HIGH torque caused a 9{+/-}11% decrease in peak CoM displacement compared to NO torque (p=0.12), it was not significant due to high intersubject variability. Whereas LOW torque decreased peak CoM displacement in all (range: -0.2 to -1.6cm), HIGH torque only decreased it in some (range = 1.2 to -2.6cm). The change in CoM displacement from LOW to HIGH torque was associated with balance ability, quantified by the narrowing beam test (R2=0.29, p=0.06), while this relationship didnt meet conventional statistical significance, likely due to the small sample size, it suggests that higher levels of exoskeleton torque may hinder balance performance in individuals with better balance ability. Taken together, more exoskeleton torque is not always better for balance, highlighting a potential need to personalize exoskeleton torque for balance augmentation.

Many people which are affected by drop foot syndrome, have to face difficulty while walking which leads to pathological gait. This type of syndrome is treated by means of an external artificial stimulation known as functional electric stimulator (FES). In this paper we are designing an online feedback control system which optimize the strength of a FES given to paretic muscle which results in correction of pathological gait of the patient in a tolerable domain. Different phases of gait are identified using inertial measurement unit (IMU) as a feedback sensor mounted on the foot. Data is collected form 8 different healthy subjects and average of collected data is used as a reference template. Different trajectories of drop foot patients are simulated (due to unavailability of patients) and corrected according to the reference template.

Motor unit (MU) decomposition using high-density surface electromyography (HD-sEMG) has been widely used to characterize MU behavior in neurophysiology and to build neural-machine interfaces for wearable robots. Recently, many open-source software tools for MU decomposition have been made available on GitHub, which could reduce the effort of researchers in the field. However, the consistency among these open-source tools has never been studied, making researchers hesitate to use them. In this study, we collected 7 open-source software tools on GitHub and applied them to decompose MUs from an open-source HD-sEMG dataset (including 11 isometric contraction trials) to investigate the consistency among these tools. To create a comprehensive MU pool for reference, we combined all unique MUs identified by seven tools, visually inspected and removed bad MUs, and manually edited all remaining MU spike trains. Across 7 tools for 11 trials, the number of identified MUs ranges from 167 to 736. The number of valid MUs after expert inspection ranges from 29 to 210, which is 10% to 72% of the reference pool. The rate of agreement between the raw MUSTs and the manually edited MUSTs ranges from 0.86 to 0.94, and the averaged number of edits per MU to correct misalignments ranges from 14 to 39. The results show inconsistency in the implementation and procedures of each tool, which results in an inconsistent number of identified MUs and valid MUs (29 vs 210). In general, a substantial amount of effort is required to process the raw MUSTs from each tool to conduct further research analysis. This study provided a guideline for using open-source software tools for MU decomposition and indicated that it would be beneficial to develop tools to automatically edit the MUSTs.

Tissue-level mechanical stimuli are primary drivers of tissue adaptation and can be optimised during conservative treatments to improve treatment outcomes for many highly prevalent musculoskeletal conditions. Current laboratory-based technologies limit our ability to connect conservative interventions such as exercise and movement modification with muscle, joint, and tissue-level mechanics, in natural environments. We introduce a physics-informed neural network (PINN) to estimate clinically relevant biomechanics from smart garments. By accounting for physiological dynamics of neural activation and muscle contraction, the PINN accurately predicted hip joint angles (RMSE <6 degrees), moments (RMSE 0.12 N*m/kg to 0.30 N*m/kg), and joint forces (RMSE 6 to 16%) from three inertial measurement units and four electromyographic sensors. We demonstrated that the trained PINN can be combined with a smart garment to estimate hip biomechanics, in real-time, during a gait retraining intervention aimed at modifying joint loading to treat hip osteoarthritis. The developed PINN and smart garment system may be adapted and generalised for personalised management or rehabilitation of a broad range of musculoskeletal diseases and injuries, in clinical, home, workplace, and sporting environments.

Quantitative gait analysis could enhance personalized treatment for many movement-related conditions; however, it is not routinely integrated into clinical care. Advances in mobile sensing, such as smartphone-based motion capture, enable rapid clinical gait assessment, but extracting actionable insights remains challenging. Although machine learning models can support clinical decisions from gait data, they typically require costly task- and condition-specific datasets, which limits progress across various gait-related conditions. Here we present a generative foundation model of walking kinematics that enables various downstream clinical tasks across diverse patient populations using clinically accessible smartphone video-based gait analysis. We aggregated eight gait datasets comprising 657 individuals across seven unique pathologies. Using weakly-supervised learning, we trained a variational autoencoder to distill high-dimensional gait kinematics into a 16-dimensional learned latent representation. We demonstrate generalizability across four downstream clinical tasks spanning pathologies both seen and unseen during training, with and without model fine-tuning, including: 1) classification of neuromuscular disorders unseen during training, 2) predicting clinical severity scores for individuals with Parkinson's disease, 3) tracking of subacute recovery post-stroke, and 4) generating patient-specific kinematic changes following total hip arthroplasty. Our model also computes a deviation from mean unimpaired (DMU) score, an interpretable scalar metric that captures an individual's deviation from typical unimpaired gait, providing rapid, holistic quantification of impairment. This generalizable model provides a foundation for clinically actionable tools that translate mobile sensing-derived gait data into precise biomechanical insights for clinical research and decision-making. The open-source model is deployed in the cloud for automated smartphone video-based gait analysis on our freely available OpenCap platform.

Classification of motor imagery (MI) tasks through EEG is valuable in brain-computer interfacing and rehabilitation engineering. EEG channels selection for MI task classification is well discussed problem and is challenging due to its combinatorial nature. Most of the existing methods are subject and task-dependent. This paper introduces a subject-independent EEG channel selection. The proposed approach consists of two stages. First, we rank channels based on their divergence from a reference channel Cz. We hypothesize that channels less divergent from Cz are more relevant for MI task classification. In the second stage, we employ a three-stage feature selection and classification model to evaluate the selected channels. It consists of a bandpass filter, followed by common spatial pattern (CSP) filter and three classifiers viz. SVM, 1-NN and 5-NN. Two publicly available datasets viz. PhysioNet and BCI Competition III IVa datasets have been used to assess the method. It performs 15.21\% more than 3Cs and just 2.91\% less than all-channels accuracy with as few as 20/118 channels on BCI Competition data and 19.64\% more than 3Cs on the PhysioNet dataset with 16/64 channels. Empirical comparison implies that the method performs better than classical models such as CSP Rank, fishers rank, and normalized mutual information, significantly. Results support that our hypothesis that divergence between channels and a reference channel Cz can be used as a ranking measure for channel selection.

Understanding the neuromuscular properties that allow dexterous manipulation of objects remains a major challenge in neurorehabilitation, largely due to the difficulty of characterizing intrinsic hand muscle activity. These muscles are small, densely packed, and anatomically complex, making selective recordings with intramuscular electromyography (EMG) technically demanding and impractical for comprehensive studies. In this work, we present a custom, high-density (HD) surface EMG grid designed to non-invasively capture activity from intrinsic hand muscles from both dorsal and palmar surfaces. We evaluated the quality and spatial selectivity of the recordings by directly comparing them with intramuscular EMG signals obtained from the dorsal and palmar interossei. Surface EMG signals corresponded closely to the intramuscular recordings, with high correlation values for all subjects and tasks. Double differential spatial filtering significantly improved selectivity, although some residual volume conduction remained. The dorsal grid primarily captured dorsal interossei activity, while the palmar grid was more sensitive to lumbrical activation. The palmar interossei recordings were spatially more varied, with the second palmar interosseous predominantly detected on the dorsal grid and the third and fourth on the palmar grid. Together, these results demonstrate that non-invasive HD surface EMG will allow more complete measurement of intrinsic muscle activity, to provide a better understanding of the complex relation between the intrinsic and extrinsic hand muscles during dexterous movements. This basic information will allow refinement of biomechanical hand models and prosthetic devices, and the development of biomimetic brain computer interfaces aimed at restoring natural hand function after neurological injury.

Freezing of gait (FoG) in Parkinson's disease is a brief but hazardous gait failure that often precedes falls. For wearable cueing or other closed-loop assistance, a detector that reacts only after FoG onset is usually too late; the more useful task is to recognize the pre-freezing transition from physiological signals. This study presents PreFoGNet, a dual time-frequency deep learning framework for early FoG prediction using plantar pressure signals. The temporal stream combines a multi-scale Inception encoder with a bidirectional Mamba module to capture both short contact-related transients and several-second gait deterioration without the quadratic cost of attention. In parallel, the frequency stream uses band-wise spectral modeling and attention-based gating to emphasize physiologically meaningful changes in the locomotion, freeze-related, and high-frequency bands. On the WearGait-PD dataset, with a 2 s prediction horizon and subject-wise evaluation, PreFoGNet achieved a sensitivity of 93.94%, a specificity of 89.76%, a G-Mean of 0.9183, and an AUC-ROC of 0.9607. It outperformed classical machine-learning and deep learning baselines, and retained usable performance under moderate noise and single-channel loss. Additional horizon analysis showed that plantar pressure contains a stable pre-freezing signature within 0-3 s before onset, with a practical prediction boundary of approximately 6-7 s. These findings suggest that time-frequency modeling of plantar pressure is a promising signal-processing route for wearable FoG early-warning systems.

Millimeter-wave radar can quietly monitor health and behavior at home, which is vital for supporting people living with dementia. Most studies, however, remain limited to short-term testing in controlled spaces. Real-world deployment requires robust activity classification as a prerequisite: vital-sign and behavioral sensing require fundamentally different processing pipelines, and absent periods need to be reliably distinguished from stationary states. Bridging the critical gap between controlled laboratory demonstrations and continuous home monitoring, this paper introduces a self-adapting radar framework that extracts meaningful behavioral segments from massive, unconstrained real-world data. The system performs continuous real-time activity classification (stationary, walking, and absent) and target localization, selectively directing downstream processing to the most informative segments. It addresses key real-world deployment challenges including adaptive thresholding across subjects and environments, and walking detection under naturalistic activity conditions. Prior to integration with the Minder platform, the system was validated in a fully instrumented studio apartment against ground truth. Across 12 subjects, the system achieved an overall classification accuracy of 0.98, with F1 scores of 0.99 for absence and stationary states, and 0.95 for walking. Event-based evaluation yielded a per-subject walking sensitivity of 0.916{+/-} 0.058 and F1 score of 0.935 {+/-}0.030. Localization root mean square error during movement was 0.40 m. The results demonstrate reliable performance suitable for transitioning to long-term real-world home deployment.

BackgroundEar-EEG-based brain-computer interfaces (BCIs) provide improved wearability and comfort compared to traditional scalp-EEG systems. However, their performance is constrained by low signal-to-noise ratios (SNRs) and high rates of BCI illiteracy under conventional luminance-modulated steady-state visual evoked potential (SSVEP) paradigms. MethodsThis study introduces a text-sequence stimulation paradigm to address these limitations by leveraging ventral visual pathway responses that are more accessible to electrodes near the ear. Using offline frequency-sweeping experiments across 4-8 Hz, we identified optimal stimulus parameters (4.6-6.8 Hz with 0.25{pi} phase shifts) and integrated them into a 12-target BCI system. We further conducted online experiments to compare the response characteristics and real-time spelling performance between the proposed text-sequence paradigm and conventional luminance stimulation. ResultsComparative experiments with 14 participants demonstrate that text sequence stimuli achieve an average information transfer rate (ITR) of 44.59 {+/-} 10.50 bits/min, outperforming luminance modulation by 76.18% in ITR. Notably, text sequence stimulation effectively mitigated BCI illiteracy, with all participants achieving near or above 70% accuracy (mean: 86.37 {+/-} 9.61%). This represents a significant improvement over luminance modulation, where 50% of users fell below 70% accuracy. ConclusionsBy reducing the flicker area by 14% and mimicking the natural luminance variations that occur during reading, the proposed method enhanced visual comfort. The online results further validate text-sequence stimulation as a high-performance and user-friendly paradigm for ear-EEG BCIs, supporting their practicality for assistive applications.