Deep Learning has shown promise in accelerating MRI by reconstructing high-quality images from under-sampled data. While recent work has leveraged multi contrast information to improve reconstruction performance, these methods rely on supervised learning, which requires fully sampled k-space for training. One method, self-supervised learning via data undersampling (SSDU), enables direct training on under-sampled k-space by partitioning it into two sets, with a network mapping between the two. In this work, we improve MRI self-supervised MRI reconstruction with two modifications. We propose a multi-contrast self-supervised learning framework that jointly trains on multiple under-sampled contrasts without requiring fully sampled k-space data as a reference. Moreover, we learn an optimal self-supervised data partitioning for each contrast in an end-to-end manner, further enhancing reconstruction quality. Specifically, we learn an optimal partitioning probability distribution, which is sampled to generate a mask for partitioning. Experiments on two publicly available multi-contrast MRI datasets demonstrate the improved reconstruction quality of our proposed self-supervised multi-contrast learned partitioning method compared to the current single-contrast self-supervised learning methods. We also demonstrate that learning the partitioning of k-space data further enhances the fidelity of reconstructions. Multi contrast reconstruction combined with learned partitioning improves reconstruction fidelity over single-contrast self-supervised MRI reconstructions. Our method can facilitate higher image fidelity and/or accelerated MRI protocol times compared to previous self-supervised methods, and without requiring fully sampled k-space for training.
Magnetic Resonance Imaging (MRI) is indispensable in clinical diagnosis and biomedical research due to its advantages such as non-ionizing radiation and high soft tissue resolution. As a core component of the MRI system, the performance of radiofrequency (RF) coils directly affects imaging quality. Wired RF coils are the standard configuration in clinical practice, but they have several limitations including cable constraints and high costs. Active wireless coils face technical challenges such as high system complexity, difficult clock signal synchronization and data throughput limitation. As a passive wireless signal transmission solution, the inductively coupled wireless coil (ICWC), serving as a complementary RF component, achieves signal and energy transmission through near-field magnetic coupling with the wired coil. They possess numerous advantages including localized reception unique advantages of non-ionizing radiation, high soft-tissue resolution, multi-parametric analysis, and metabolic information monitoring. Since the 1970s [1, 2], MRI has not only revolutionized the traditional medical understanding of human anatomy and pathological changes but also achieved groundbreaking advancements in numerous specialized subfields, such as neuroscience [3-5] and oncology [6-8]. sensitivity enhancement, cross-tissue/species transplantability, adaptability to different main magnetic fields, reduction of the g-factor, B1 + field shaping, cable-free advantages in special scenarios, design scalability, and cross-manufacturer compatibility. This review elaborates on the inductive coupling mechanism of ICWCs, the derivation of the SNR formula, potential causes of g-factor reduction, and recommendations for fabrication methods. Additionally, it systematically summarizes the application progress in scenarios such as invasive imaging, multi-site human imaging, animal imaging, and applications in special scenarios. Finally, it discusses the development prospects ICWCs in fields including neuroimaging, ultra-high/ultra-low field MRI, and X-nucleus MRI/MRS, as well as ICWCs' advantages and limitations, providing a reference for the innovation of MRI RF coil technology and the clinical translation of ICWCs.
Clinical assessment of disorders of consciousness (DOC) remains challenging because motor impairments and fluctuating vigilance may obscure residual awareness. Sleep electroencephalography (EEG) provides a non-participatory window into residual brain function; however, pathological recordings are scarce and heterogeneous, fine-grained stage annotation is unreliable, andexisting methods rely on multi-channel acquisition with limited cross-subject generalization. This study aims to develop a single-channel EEG framework for sleep staging and sleep-informed consciousness assessment in DOC patients. SleepConFormer integrates a multi-task EEG representation learning (MTERL) backbone pretrained on public sleep datasets, a Stage Confusion Estimation Transformer (SCE-Transformer) for confusion-aware temporal modeling under pathological conditions, and a logit space aggregation strategy for robust Wake-Sleep and Wake-NREM-REM inference. Stage predictions and stage conditioned EEG embeddings are aggregated into subject level features for minimally conscious state (MCS) and unresponsive wakefulness syndrome (UWS) discrimination under subject-independent evaluation. On three public sleep datasets from non-DOC participants, Sleep ConFormer achieves 84.5-87.7% five-class accuracy with strong cross-dataset generalization (average MF1 78.73%). In a clinical DOC cohort (n = 24), coarse Wake-Sleep staging reaches 80.78% accuracy. Sleep-derived features distinguish MCS from UWS with 91.7% accuracy and 0.846 AUC, surpassing single-modal features by 12.5%. SleepConFormer demonstrates the feasibility of transferable single-channel sleep staging and sleep informed consciousness discrimination in DOC under subject-independent internal validation. This work provides a promising proof-of-concept framework for sleep-informed DOC assessment using single-channel EEG and supports future investigation toward objective bedside monitoring in neurocritical care.
Computed tomography (CT) images obtained in clinical settings are often acquired with diverse scanner types and acquisition parameters. They may exhibit significant variations in fields of view (FOVs) and levels of contrast enhancement. An automated method for navigating the content in these images is therefore essential for effective dataset curation and downstream analyses. This work introduces a framework called Body- Part-Phase Regression (BPPR) to automatically identify regions of interest and determine the contrast enhancement phase of body CT images. The framework consists of two key components: (1) A two-phase body part regression method for predicting the anatomical location of 2D slices within 3D volumes. (2) A circular regression model for predicting the contrast timing of CT images (i.e., the timing of the scan relative to contrast agent injection) from a continuous perspective, providing a fine-grained understanding of contrast differences, particularly in relation to patient-specific vascular effects. These two components are linked via a positional weighting mechanism which enhances volumelevel phase prediction by leveraging slice-level predictions. By unifying the "part" and "phase" regression models, our framework establishes a cohesive approach to continuous content navigation in CT images. We train and evaluate our models on large-scale datasets consisting of multi-contrast images and compare their performance with alternative approaches pursuing similar goals. The experiments demonstrate improvements in both slice localization and contrast phase prediction. In particular, the two-phase training scheme reduces the slice localization error of previous body part regression methods from 9.2 mm to 6.1 mm. We also discuss the distinctive advantages of BPR over segmentationbased approaches and highlight potential clinical applications that may benefit from the proposed BPPR framework.
Wearable ultrasound bladder monitoring often assumes static body positions, which is unrealistic for all-day usage. This work aims to enable reliable, non-invasive bladder-fullness classification by mitigating performance degradation caused by sleeping position variations. A miniature wearable ultrasound system for bladder monitoring has been developed. It weighs 32.2 gram (inc. battery) and draws an average current of 8.59 mA, supporting long-term operation. To ensure robustness to sleeping position variations, three modeling strategies were investigated: (i) Pooled training on multi-sleeping-position data, (ii) Oracle routing to sleeping-position-specific models based on dual-modal data, and (iii) Adversarial training using a confounder-free network (CF-Net) to achieve sleeping-position-independent feature extraction. The system has been validated on 12 participants across diverse sleeping positions. The baseline model trained on specific sleeping positions achieved only 75.95% average accuracy when tested on unseen position samples. Relative to this baseline, after applying the Pooled, Oracle and CF-Net strategies, classification accuracy improved by 16.48%, 20.03% and 18.56%, respectively. The proposed system enabled wearable ultrasound bladder monitoring with substantially improved robustness to sleeping-position variations. This study is the first to address the challenge of sleeping-position variation in wearable bladder monitoring, demonstrating its feasibility of all-day, home-based monitoring with clinically meaningful pre-void alerts.
Rapid Serial Visual Presentation (RSVP) had been applied to human-computer interactions such as spelling, device control and so on. Traditional single RSVP paradigm detects targets in only one image stream, which may lead to missed or false detection. Dual RSVP paradigm can enhance the classification robustness through specific encoding to increase the number of targets. To enhance the classification performance of dual RSVP, an EEG-EM Self Attention and Cross Attention Network (EESCAN) is proposed. EEG and EM signals undergo a symmetric two-stream backbone. Each stream consists a convolution module and a self-attention module to extract local and global features. Subsequently, an inter-modal bidirectional interaction module is proposed to provide complementary information between EEG and EM modality. Finally, dynamic reweighting and fusion module is employed to dynamically adjust the sample-level weights according to the contribution of EEG and EM features. Moreover, a VR-based dual RSVP virtual robotic arm control system using the proposed algorithm is then designed to achieve gaze-independent device control. EEG and EM data from 21 subjects were collected and analyzed. The proposed network and system achieved better performance than existing decoding methods and single-modal baselines. Ablation experiments and visualization results further verified the effectiveness of each proposed module. EESCAN network is proposed for the dual RSVP paradigm that integrates intra-modal self-attention, inter-modal bidirectional interaction, and dynamic fusion to achieve EEG-EM fusion. EESCAN markedly improves classification performance of dual RSVP-based BCIs. This gaze-independent control system is suitable for patients with restricted gaze shifts.
Falls are the leading cause of injuryrelated hospitalization and mortality among older adults. Consequently, mitigating age-related declines in gait stability and reducing fall risk during walking are critical goals for assistive devices. Lower-limb exoskeletons have the potential to support users in maintaining walking stability. However, most exoskeleton controllers are optimized to reduce the energetic cost of walking rather than to improve stability. Existing stability-focused controllers use a fixed, heuristic assistance profile, limiting understanding of how to optimize and personalize assistance strategies. To address this gap, we systematically modulated the magnitude and duration of torque provided by a bilateral hip exoskeleton during slip perturbations in eight healthy adults, quantifying stability using whole-body angular momentum (WBAM). WBAM responses were governed by a significant interaction between assistance magnitude and duration, with duration determining whether exoskeleton assistance was stabilizing or destabilizing relative to not wearing the exoskeleton device. Compared to an existing energy-optimized controller, experimentally identified stability-optimal parameters reduced WBAM range by 27.4% on average. Notably, substantial inter-subject variability was observed in the parameter combinations that minimized WBAM during perturbations. We found that optimizing exoskeleton assistance for energetic outcomes alone is insufficient for improving reactive stability during gait perturbations. Stability-focused exoskeleton control should prioritize temporal assistance parameters and include user-specific personalization. This study represents an important step toward personalized, stability-focused exoskeleton control, characterizing the effects of assistance on reactive stability in young adults and motivating future investigations in aging populations.
Motor imagery electroencephalography (MI-EEG) classification is essential for brain-computer interfaces (BCIs), but achieving high accuracy across different individuals remains challenging due to significant inter-subject variability. Recently, unsupervised domain adaptation (UDA) methods have addressed this problem by adapting models without target labels, often using pseudo-labeling. However, existing pseudo label techniques evaluate each sample in isolation and employ a simple threshold-based strategy, overlooking the relationship among samples and often excluding useful data points. We aim to overcome these limitations for cross-subject MI-EEG classification. We propose the relation-aware progressive pseudo-label generation (RAP2G) method, a novel UDA framework combining Optimal Transport (OT) with structure aware regularization and dynamic pseudo-label selection. RAP2G leverages the inherent structure within the target subject's data by incorporating feature similarity into the OT-based pseudo label generation process. It also adaptively selects pseudo-labeled samples using a progressive schedule based on OT confidence. We evaluated RAP2G on three public benchmarks, BCI Competition IV dataset 2a, BCI Competition IV dataset 2b, and the High Gamma Dataset, using leave-one-subject-out cross-validation. RAP2G consistently outperforms existing state-of-the-art UDA techniques and baseline models. Ablation studies confirm the contribution of the structure-aware component. Visualizations show enhanced feature separability after adaptation, and the learned attention maps are qualitatively consistent with known motor-cortex organization. RAP2G provides an effective approach for robust cross-subject MI-EEG classification. By improving label-free adaptation across subjects, this work supports more reliable and practical BCI systems for biomedical applications.
Optimization of electrical stimulation patterns can enhance the performance of visual neuroprosthetics by replicating the selective and temporally precise neural responses characteristic of natural vision. An in silico approach employing the Artificial Bee Colony (ABC) algorithm was developed to optimize electrical stimulus waveforms that evoke physiologically realistic responses in retinal ganglion cells (RGCs). The optimization was guided by a computational model of ON and OFF RGCs, with the goal of matching recorded neural responses from the rat dorsolateral geniculate nucleus (dLGN). The ABC algorithm iteratively modified the amplitude envelope of charge-balanced, biphasic pulse trains to maximize similarity between simulated and recorded responses. The ABC algorithm consistently achieved $\geq$80% cross-correlation with target neural responses within 2000 iterations. It also identified single optimized waveforms capable of eliciting distinct responses in neighboring neurons, indicating potential for simultaneous physiological responses with electrical stimulations. Under constrained computational conditions, ABC demonstrated superior accuracy compared to alternative optimization algorithms. The proposed ABC-based optimization approach effectively generates neural activation patterns closely resembling natural responses, supporting its use for real-time and adaptive stimulation control in visual neuroprosthetic systems. This study introduces a biologically inspired optimization framework that advances the development of intelligent stimulation strategies to enhance the functionality of next-generation visual neuroprostheses.
Asymmetric pulse waveforms represent an optimized treatment protocol for high-frequency irreversible electroporation (H-FIRE), offering a lower ablation threshold compared to symmetric waveforms. However, the impact of waveform asymmetry has not been considered in current numerical models aimed at predicting tissue ablation zones. This study aims to develop an asymmetric waveform ablation area prediction model incorporating dynamic conductivity and dielectric dispersion effects (AW-DCDE) to predict ablation zones under varying asymmetric conditions. A linear scaling parameter relating pulse asymmetry to the ablation threshold was integrated into a modified Heaviside conductivity function. Dielectric dispersion was captured via a fourth-order Debye model. Variations in electric field intensity and dielectric dispersion were analyzed at pulse widths of 5 μs and 10 μs across different asymmetry levels. Finally, potato tissue experiments were conducted to validate the AW-DCDE model. Increasing waveform asymmetry leads to a reduction in electric field intensity and polarization current, but a significant expansion of the ablation area. Statistical analysis confirmed no significant difference (p > 0.05) between AW-DCDE predictions and experimental results. Dynamic conductivity was found to have a greater influence on the predicted ablation area than dielectric dispersion. Waveform asymmetry significantly affects ablation area and electrical parameters. The AW DCDE model provides a theoretical framework for optimizing clinical H-FIRE treatment protocols.
Current intraoperative hepatocellular carcinoma (HCC) detection primarily relies on surgeon visual inspection, palpation, and intraoperative ultrasound, which struggle to precisely localize tumor 3D position and morphology for surgical guidance. Fluorescence molecular to mography (FMT) offers a promising solution by visualizing the 3D quantitative distribution of fluorescent biomarkers. However, existing FMT techniques remain confined to preclinical research due to challenges in dynamically acquiring anatomical structures, obtaining accurate human optical parameters, and developing robust reconstruction algorithms, hindering clinical translation. To address these challenges, we proposed a novel free-structure based efficient NIR-II FMT solution. This solution features three key innovations: 1) A Free-Structural Multimodal Fusion Imaging (FSMFI) system utilizing a 3D scanner to dynamically acquire intraoperative anatomy; 2) Calculation of accurate human HCC optical parameters for precise light transport modeling; 3) An Adaptive Gaussian Weighted Smoothing (AGWS) method incorporating an energy-intensity difference prior and a smooth-solving strategy to ensure stable and accurate source reconstruction. Simulation experiments confirmed the accuracy of the AGWS method. Porcine liver phantom experiments validated the overall efficacy of the solution. Ex vivo and in vivo human HCC experiments demonstrated its clinical efficacy and translational potential. This work achieves the first breakthrough in intraoperative FMT reconstruction by overcoming preoperative structural constraints, advancing 3D surgical navigation strategies, and accelerating FMT's clinical translation.
This study evaluated dyadic endocrine interdependence among cancer patients and their spousal caregivers using a machine learning (ML) framework that assesses how partner-level information improves prediction of individual stress biomarker responses. A multilevel ML framework was applied to 149 patient-caregiver dyads. Predictive performance was evaluated for mean and variability of diurnal slopes for cortisol, alpha-amylase, and DHEAS. Five regression models were compared across eight configurations incorporating PCA, correlation-based filtering, and data augmentation via leave-one-out cross-validation. Incremental dyadic signals were quantified via partial Spearman correlation, and predictor importance redistribution was characterized using SHAP-based decomposition. Nonlinear methods (Extra Trees, Random Forest, Gradient Boosted Regression) yielded top-performing models for 11 of 12 outcomes with at least one significant model at 95% confidence ($\rho$ = 21.1% to 50.6%). For slope mean outcomes, targeted preprocessing converted negative or near-zero baseline correlations into significant predictions. Partner signal significance was biomarker-specific: at 90% confidence, mean outcome partner signals were significant for cortisol and DHEAS (8 of 11 models) but negligible for alpha-amylase (1 of 10), whereas variability partner signals were distributed across all three biomarkers (8 of 22 models). SHAP analysis revealed two structural signatures: mean outcomes showed an incremental magnitude pattern (AUC-ROC = 0.91), indicating a unidirectional predictive partner signal, whereas variability outcomes exhibited bilateral convergence (AUC-ROC = 0.90), indicating patterns consistent with interpersonal processes. Dyadic interdependence in endocrine regulation manifests through distinct structural signatures for mean versus variability measures. This analytic framework offers a hypothesis-generating approach to identifying psychobiological pathways for improving health in oncology populations.
EMG-based state estimation and prediction in human-machine interaction,biomechanics, and robotics applications is an emerging approach offering potential improvements in control and user intent prediction. Koopman operator theory (KOT) is a powerful method for analyzing and transforming nonlinear dynamical systems into a higher-dimensional space so that they can be described using linear operations. In this study, first, we utilize the power of neural networks to capture the nonlinear relationships between surface electromyography (EMG) signals and lower-limb joint states. We use these relationships to estimate the current joint states purely from sEMG signals. The second stage of this study is to capture wearers' intentions in gaits with Koopman operators' powerful nonlinear system representation capabilities and use this framework to learn the temporal relationships between the current and near-future joint states. We start simply by implementing the EMG-based Koopman operator to estimate knee and ankle angles during different gaits, as a pioneer study. For intra-subject prediction, we achieved an RMSE of $3.61^{\circ }$ for the knee and $1.78^{\circ }$ for the ankle under same-gait conditions, and $3.78^{\circ }$ for the knee and $1.43^{\circ }$ for the ankle under transient-gait conditions. Cross-subject leave-one-subject-out (LOSO) generalisation yielded a mean RMSE of $7.79^{\circ }$ for the knee and $2.12^{\circ }$ for the ankle under same-gait conditions, and $8.77^{\circ }$ for the knee and $3.09^{\circ }$ for the ankle under transient-gait conditions.
Wrist function is essential in performing activities of daily living (ADLs). However, there is limited experimental evidence on the functional impact of wrist Abduction-Adduction (Ab-Ad) joint assistance in upper limb exoskeletons (ULEs) during ADLs. This study provides the first implementation and demonstration of a clock spring-based wrist Ab-Ad joint into a five degree of freedom (DoF) ULE, EXOTIC2 exoskeleton and evaluates its effect, to support individuals with severe motor impairments. A compact, lightweight wrist module with tendon-driven abduction and spring-driven adduction was integrated into the EXOTIC exoskeleton. Eight adults with no motor disabilities completed drinking and scratching tasks under randomized wrist-enabled and wrist-locked conditions along with a preliminary feasibility test in one individual with Amyotrophic lateral sclerosis (ALS). Kinematic and task performance metrics including wrist range of motion, task completion time, spillage and leveling metrics were assessed. Implementing the wrist Ab-Ad DoF improved task success metrics. Spill incidence during the drinking task decreased from 56% to 3%, and leveling success for scratching task improved from 28% to 75%. Integrating wrist Ab-Ad assistance improved key functional task outcomes without increasing execution time. The study provides the experimental evidence that active wrist Ab-Ad control enhances task-level performance in exoskeleton-assisted ADLs and supports the inclusion of wrist deviation in future assistive exoskeletons.
Fetal and maternal heart rates (FHR and MHR) are routinely monitored during labor and delivery as critical biomarkers of fetal-maternal well-being. Assessing their directed coupling can reveal interactions that are not apparent from either signal alone. This coupling is often nonstationary, changing rapidly with uterine contractions and the stage of labor. Therefore, classical tools for detecting coupling often fail in this context. We introduce a Kalman filter (KF)-based, time-varying autoregressive (TVAR) framework for nonstationary Granger causality (GC) assessment and evaluate it for MHR-FHR coupling analysis. The method models autoregressive (AR) coefficients of univariate and bivariate predictors as latent states in a linear state-space model and tracks them recursively using an optimized KF, yielding continuous coefficient trajectories with confidence intervals. Time-varying GC is obtained by comparing one-step-ahead prediction errors (innovations) computed with and without including the other signal's history. In the stationary limit, the proposed method is shown to reduce to classical fixed-parameter AR estimation, ensuring consistency with conventional GC. The framework is generic, naturally supports online/streaming analysis, enables multivariate extensions (e.g., incorporating uterine activity or respiration to probe common-input effects), and is generalizable to nonlinear models via extended/unscented Kalman filtering or particle methods. Evaluated on a public fetal ECG dataset (ten 20-min pregnancy records and twelve 5-min labor records), KF-TVAR GC identifies brief, statistically significant MHR-FHR coupling episodes that stationary GC fails to localize. Across subjects, coupling is generally stronger from maternal-to-fetal during late pregnancy and shifts toward fetal-to-maternal during labor.
Brain source imaging (BSI), also known as source localization, from magneto- and electroencephalographic (M/EEG) data, is a challenging ill-posed inverse problem. Accurate source estimation is sensitive to multiple modeling and experimental parameters, such as regularization strength and noise level, where misconfigurations can lead to under- or overfitting. Different BSI methods, however, may vary in their robustness to suboptimal parameter choices. Here we conducted extensive simulations of brain sources superimposed by varying degrees of sensor noise to study the ranges of noise misspecification within which different BSI approaches can still localize well. Using the Earth Mover's Distance (EMD) and other metrics, we compare the performance of smooth linear inverse solutions with that of sparse non-linear Bayesian learning solutions. Additionally, we assess the effectiveness of various noise estimation and crossvalidation techniques to select hyperparameters close to those achieving optimal localization. Our results characterize the robustness of commonly used BSI methods to noise and regularization misspecification. Across methods, moderate underfitting to noise generally yielded better performance than overfitting. Spatial cross-validation was effective in identifying hyperparameters that achieved near-optimal localization performance. Methods and experiments are made available within the BSI-Zoo Python package. The choice of BSI method and hyperparameter estimation strategy substantially influences localization accuracy and robustness under noise misspecification. These findings provide practical guidance for selecting and tuning BSI methods, improving the reliability and reproducibility of M/EEG source localization.
Surface electromyography (sEMG) pattern recognition prosthetic systems are unintuitive and cannot control many movements reliably. Intramuscular electrodes and regenerative peripheral nerve interfaces (RPNIs) can improve signal quality and provide additional control signals for controlling more degrees of freedom. This case study compared functional, biomechanical, and cognitive outcomes between control approaches incorporating active wrist rotation using sEMG and implanted intramuscular EMG (imEMG) signals from RPNIs and residual muscles. We also explored whether combining the EMG sources for wrist rotation could provide further advantages due to the lack of available implanted wrist rotation signals. One female with unilateral transradial amputation completed functional assessments using a myoelectric prosthesis with five control approaches. Pattern recognition classifiers were trained to decode sEMG and/or imEMG from residual muscles and RPNIs into functional grips with or without wrist rotation. The participant improved performance of the Clothespin Relocation Test and Coffee Task when using imEMG compared to sEMG, regardless of whether wrist rotation was enabled. Adding wrist rotation only modestly reduced trunk compensations ($\Delta$2-7$^{\circ }$). imEMG with wrist had substantially lower cognitive workload than sEMG ($\Delta$58 pts). A combined classifier was associated with improved performance and the lowest cognitive workload. In this participant, imEMG supported the control of multiple grips and active wrist rotation to achieve better functional performance than with sEMG, without added cognitive burden. These findings highlight the potential of implanted intramuscular EMG signals from RPNIs and residual muscles for improving prosthetic control during daily tasks.
The accurate assessment of hemiparetic gait after stroke was essential for understanding the extent of motor impairment and for guiding rehabilitation efforts. In this study, a kinematic based deep learning classifier was proposed to eliminate inter-rater variability in lower-limb Brunnstrom Recovery Stage (BRS) assessment and enable quantitative longitudinal tracking of post-stroke motor recovery. Forty healthy adults and fifty-one hemiparetic stroke patients (BRS III-VI) were recruited, and video recorded kinematic data were collected during their standardized walking tasks. Anatomical keypoint coordinates were extracted from the video recordings via advanced pose estimation algorithms for fine-grained kinematic analysis. A novel hybrid Skeleton-Attention long short-term memory (LSTM)-Inception architecture was then developed for BRS stage classification using three-dimensional keypoint coordinate data. The architecture synergistically integrated LSTM layers for temporal sequence modeling with Inception modules for multi-scale spatial feature extraction. The proposed model's performance was systematically compared against conventional deep learning benchmarks, including convolutional neural networks (CNNs) and coupled LSTM-CNN hybrid models. Our experimental results revealed that the proposed framework achieved superior classification accuracy compared to alternatives (97.3% vs. 87.7-95.2%). These findings demonstrated the clinical potential of deep learning-driven motion analytics to eliminate assessment subjectivity in BRS staging. This approach is expected to facilitate quantitative longitudinal monitoring of neurorehabilitation progress and support the development of data-driven personalized therapeutic strategies, potentially reducing reliance on clinician expertise.
Implantable Medical Devices (IMDs) are evolving into collaborative networks, demanding robust and ultra-low power communication in dynamic and often pathological biolog ical environments. Multi-chamber leadless pacemakers (LCPs) represent a particularly challenging case, because the intrac ardiac channel exhibits not only periodic fluctuations during healthy rhythms but also abrupt and aperiodic fades under pathological conditions such as atrioventricular (AV) block. To address this challenge, an adaptive receiver based on a Super-Regenerative Receiver (SRR) and a hybrid analog-digital Automatic Gain Control (AGC) system is proposed. At its core is a beat-synchronous two-degree-of-freedom (2-DOF) control strategy that uses peak RSSI as an electrocardiogram-free rhythm surrogate, enabling active signal stabilization through predictive feedforward and incremental Proportional-Integral Derivative (PID) control. The proposed architecture was vali dated through a three-layer framework comprising Hardware in-the-Loop (HIL) transient characterization, programmable ex vivo porcine-heart testing under a representative Mobitz Type II AV block model, and complementary ATP-provoked acute in-vivo porcine validation. Experimental results show that the proposed 2-DOF controller stabilizes the input signal to within 1% of its target under channel variations exceeding 15dB, reducing the Bit Error Rate (BER) by one to two orders of magnitude and achieving an improvement factor of more than 60 at 5kbps. In the living heart, the adaptive loop maintained stable gain control and preserved digital demodulation during ATP-provoked transient AV-block-like intervals, while supplementary in-vivo BER measurements provided additional quantitative support for end-to-end communication robustness. These results support the proposed architecture as a promising solution for robust intrac ardiac communication in future multi-chamber LCP systems.
We investigated the effect of posture on the link between cerebral circulation and cortical activity without applying cognitive tasks. We computed the zero-lag mutual information (MI) between spontaneous variations of mean cerebral blood velocity (MCBv) and the series of spectral powers computed over electroencephalographic (EEG) channels in traditional frequency bands. Estimation of MI was performed according to a fully linear approach and to a technique able to describe nonlinear components of the relationship as well. Time-shifting surrogate approach was utilized to reject the null hypothesis of uncoupling. Analysis was carried out in 27 healthy young individuals (age: 33±8 yrs; 13 males; 14 females) at rest in supine position (REST) and during active standing (STAND). Percentage of rejection of the null hypothesis of uncoupling peaked to 52% and did not vary across the MI estimates and experimental conditions. STAND did not affect MI regardless of the method utilized to its estimate. Results were consistent across brain areas and EEG frequency bands. In healthy young subjects in the absence of task-related activity, neurovascular coupling (NVC) can be assessed from spontaneous fluctuations of MCBv and EEG spectral powers, posture is not a confounding factor, and nonlinear components negligibly contribute to the information exchange between MCBv and resting-state brain activity. The significant values of MI suggest that NVC can be assessed from spontaneous variability of MCBv and EEG spectral power series and was not affected by orthostatic position.