Biomedical Signal Processing and Control

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.

The electrocardiogram (ECG) contains rich nonlinear and non-stationary dynamic information that is only partly captured by conventional ECG interpretation and beat-to-beat metrics, and is increasingly analyzed using black-box artificial intelligence models that often lack interpretability. Here, we introduce the ECG time-frequency "eyeball", an interpretable framework that transforms a brief single-lead ECG recording into a geometric signature and a set of low-dimensional rotational and geometrical features using empirical mode decomposition and Hilbert-based analytic signal mapping. In 30-second lead I-equivalent recordings from 170 healthy subjects and 80 patients with acute myocardial infarction (AMI), the proposed "eyeball" metrics significantly differentiated groups, with AMI associated with higher rotational frequency metrics, lower envelope metrics, and displaced centroid location. Representative examples revealed a coherent morphologic spectrum from normal patterns to geometries consistent with myocardial ischemia, injury, and infarction. The representation remained stable across recording windows from 30 seconds to 5 minutes, and individual "eyeball" features achieved areas under the receiver operating characteristic curve (AUCs) of up to 0.78 for AMI detection. These findings suggest that the ECG time-frequency "eyeball" condenses clinically relevant nonlinear ECG dynamics into an interpretable representation that may reveal hidden AMI signatures, complement conventional ECG interpretation, and provide a foundation for accessible single-lead cardiovascular screening using future smart wearables.

The behavioural activity of freely moving animals is a confounding factor that affects the recording, analysis, and final results of animal EEG experiments. Along with the lack of standardisation in animal in vivo electrophysiology experiments, this could lead to huge inconsistencies, especially in the analysis of centrally acting drugs. Therefore, the main aim of this paper is to investigate the effects of behavioural activity versus inactivity on the multichannel EEG in freely moving rats. In a large sample (n = 116) of waking recordings from 12 cortical electrodes (ECoG) in Wistar rats, we evaluated behavioural activity-related changes in the power spectrum, current source density, and power-based global functional connectivity (GFC) in a 3D rat brain model, according to the TOHOKU Rat Brain Atlas. The main findings were that behavioural activity induced 1) a robust power increase in 6-8 Hz, peaking at 7 Hz with maximum changes over the parietal and temporal cortex, 2) an increase in gamma power (30-80 Hz) across the whole brain, 3) a decrease in delta (1-4 Hz) and beta (12-30 Hz) power across the whole cortex. Changes were also localised in subcortical regions, particularly in the diencephalon/thalamus. The GFC analysis showed a similar pattern of power changes across the 6-8 Hz, delta, and beta bands; however, GFC in the gamma band decreased. Again, the GFC analysis revealed changes in connectivity within subcortical structures, primarily in the thalamus. None of the measures was affected in the alpha band (8-12 Hz). These findings emphasise behavioural state as a critical factor influencing EEG outcomes, with important implications for the standardisation and translational validity of preclinical neurophysiological studies.

Adhesive electrocardiography (ECG) electrodes used in neonatal intensive care units (NICUs) may cause skin injury in premature infants. Although photoplethysmography (PPG)-based ECG reconstruction has been explored, existing studies have mainly focused on adult data and often rely on direct PPG-to-ECG mapping or artificial signal alignment, which may be unsuitable for neonates with highly variable pulse arrival time (PAT). In this study, we propose an alignment-free RoPE-based dual-stream Transformer for reconstructing missing neonatal ECG segments using concurrent PPG signals and bidirectional ECG context. A total of 52,566 10-second ECG-PPG windows were extracted from 159 NICU patients and split at the patient level to prevent data leakage. The model was designed to learn ECG-PPG temporal coupling without forced synchronization by integrating PPG-derived hemodynamic timing information with lead-specific ECG context. Under a 40% random missing condition, the model achieved a Pearson correlation coefficient of 0.96, mean absolute error of 0.04, and root mean square error of 0.07. It also maintained robust performance under 4.0-second continuous block loss and 60% random patch loss, preserving a PCC of at least 0.90. These findings suggest that the proposed framework may serve as a signal imputation module for maintaining ECG monitoring continuity in NICU environments. Prospective validation is required before clinical diagnostic use.

Accurate detection of P300 event-related potentials from electroencephalography (EEG) re-mains challenging for small numbers of trials due to low signal-to-noise ratios and substantial inter-subject variability. This study presents a systematic comparison of data aggregation strate-gies for improving P300 classification, evaluated on a 10-subject dataset using two convolutional neural network architectures (EEGNet and BaseCNN) and a support vector machine (SVM). We compared: (1) subject-specific and pooled general models for single trials; (2) epoch aver-aging with 5 and 10 stimuli repetitions; (3) multi-channel models where subjects corresponded to different input channels; (4) cross-subject averaging; (5) mixed (uncontrolled) averaging; (6) a combined approach with K trials per subject across all participants; and (7) time-shifted channels from extended single-trial epochs. Decoding performance was quantified using the Information Transfer Rate (ITR), computed for binary classification accuracy. We found that single-trial ITR was unpractical (0.15-0.64 bits/trial), whereas controlled aggregation improved the performance. The combined cross-subject approach with K = 3 trials per participant (30 channels) achieves the highest ITR with multi-channel EEGNet: 0.95 bits/aggregated decision in the no-aperture recordings and 0.97 bits/aggregated decision on Aperture data, approaching the theoretical binary-classification limit for the aggregated decision. Controlled cross-subject averaging consistently outperformed random trial mixing, and multi-channel architectures out-performed simple averaging when inter-subject structure was preserved. These findings con-tribute to improving P300 decoding and implementing multi-subject brain-computer interfaces (BCIs).

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.

Cervical cancer represents a pressing global health challenge, emphasizing the critical need for accurate and timely diagnostic methods to facilitate effective treatment and improve survival rates. In response to this challenge, the study presents CerViX-Net, an innovative classification framework designed to advance cervical cancer detection through enhanced computational efficiency and diagnostic accuracy. The development of CerViX-Net is motivated by the limitations of traditional diagnostic models, particularly in handling the computational and memory demands of large-scale data, while ensuring precise feature extraction and classification. CerViX-Net employs a hybrid deep learning architecture that combines the capabilities of ResNet50, EfficientNet-B0, and a Modified Vision Transformer (ViT) module. The ResNet50 branch extracts hierarchical features through stacked convolutional and identity blocks. In another path, the modified ViT module transforms image patches via linear projection, augments them with positional and class embeddings, and processes them using Parallel Transformer Encoder layers to model contextual relationships. Concurrently, EfficientNet-B0 utilizes MBConv blocks to extract multi-scale representations. The feature outputs from all three branches are integrated and passed through a classification head consisting of dropout layers and dense layers to ensure robust and accurate predictions. The proposed framework is rigorously evaluated on the Mendeley LBC dataset, achieving exceptional performance metrics with an accuracy of 99.69%, precision of 99.28%, recall of 99.48%, and an F1-score of 99.52%. The robustness of CerViX-Net is further validated on the SIPaKMeD and Herlev Pap Smear datasets, where it demonstrates comparable excellence, underscoring its efficacy and adaptability across diverse cytology datasets. Statistical validation using Friedman's test further reinforces its superiority over competing methods.

Understanding brain-behavior relationships requires models capturing the distributed, interactive, and multiscale nature of neural systems. Traditional univariate approaches and single-biomarker models are inherently limited in this context, as they fail to represent dependencies across regions and the hierarchical organization of brain networks. In this study, we propose a graph-based multivariable framework for brain imaging analysis that integrates key organizational principles of brain function-including segregation, integration, modularity, and temporal dynamics-within a unified graph neural network architecture. The framework represents brain data as hierarchical graphs, where node features encode regional activation and temporal variability, and graph structure captures interactions within and between functional modules. The proposed approach is evaluated using functional near-infrared spectroscopy (fNIRS) data as a case study, where subject-specific brain graphs are constructed from task-based recordings acquired shortly after cochlear implant activation to predict speech understanding outcomes one year later. Under leave-one-subject-out validation, the model demonstrates strong predictive performance (R = 0.73, p < 0.001), outperforming previously reported single-biomarker approaches. Perturbation-based analyses further show that predictions are driven by distributed patterns of activity and interaction across regions and modalities, rather than isolated features. These results illustrate the capability of the proposed framework to capture complex brain organization and highlight its potential as a generalizable platform for multivariable analysis and prediction in neuroimaging applications beyond the specific clinical use case considered here.

Transformer-based deep learning models have shown great potential for decoding visual EEG signals. However, their internal attention mechanisms are often evaluated primarily on optimization objectives, leaving their alignment with biological brain connectivity an open question. This study empirically evaluates how variations in EEG preprocessing strategies affect these attention representations using the Adaptive Thinking Mapper (ATM) model as a framework. We compared a baseline pipeline (MVNN only) against a comprehensive cleaning pipeline integrating ICA and notch filtering. The models were evaluated through cross-generalization, noise robustness, and spectral-temporal ablation analyses. Furthermore, we investigated the structural correspondence between the model's data-driven attention weights and neurophysiological reference networks (GPDC, PDC, and DTF) using Node Strength Correlation and Representational Similarity Analysis (RSA). The results show that the comprehensive preprocessing successfully suppresses non-neural artifacts, such as frontal noise and electrical interference, while maintaining comparable decoding accuracy and baseline robustness. Alignment analyses revealed that the broad spatial organization of the learned attention patterns remains highly stable across pipelines, capturing key directed connectivity dynamics with subtle, metric-dependent variations in global representational geometry. This work provides an empirical exploration into bridging data-driven attention weights with neurophysiological consistency, offering insights toward more transparent brain-computer interfaces.

Ventricular tachycardia (VT) and ventricular fibrillation (VF) are the leading electrical causes of sudden cardiac death, but automated detection is limited by strong class imbalance, where lethal arrhythmias account for fewer than 22% of ECG segments. In this setting, standard classifiers can achieve high accuracy by predicting normal rhythm in most cases while missing many lethal events, a failure mode referred to as rare-class collapse. We evaluated six imbalance-handling approaches: naive logistic regression, inverse-frequency reweighting, label-distribution-aware margin loss (LDAM), cost-sensitive training, two-stage cascade classification, and anomaly detection on 15,614 ECG segments from three PhysioNet databases (VTaC, VFDB, CUDB), with an overall normal-to-lethal ratio of 3.6:1. All methods were assessed at a fixed operating point of 95% specificity using recall, area under the precision-recall curve (AUPRC), and missed-lethal-event rate (MLER). The naive model achieved 45.1% recall (MLER = 0.549), missing 564 of 1,027 lethal events despite 84.1% accuracy. The two-stage cascade performed best, with 65.2% recall, AUPRC of 0.821, and MLER of 0.348, reducing missed events by 37% and achieving the highest decision-curve net benefit. Per-source analysis showed near-complete VF detection (recall up to 0.975) but much lower VT detection (recall 0.183), suggesting a feature-space limitation due to spectral similarity between organized VT and rapid sinus rhythm. Overall, the results show that evaluation metrics strongly influence the visibility of rare-class failure, and that cascade-based methods outperform simpler reweighting approaches for detecting lethal arrhythmias.

Abnormal states such as erroneous behaviors are generally difficult to represent from neural data. However, such states are also known to have specific spatiotemporal features, indicating a feasibility of developing a method to focus on them. If a method can highlight these spatiotemporal features, it may effectively represent such abnormal states, helping evaluate abnormal brain functions. In the present study, we proposed the hierarchy of supported modules (HSM) to highlight spatiotemporal features that can represent abnormal states. HSM spatiotemporally transforms multidimensional neural time-series based on their spatiotemporal context. We evaluated HSM through decoding and similarity analyses using multiple publicly available datasets. In the HSM results, decoding accuracies were higher for erroneous behaviors than for normal behaviors, and similarities were lower between erroneous behaviors and normal behaviors than between normal behaviors, demonstrating the ability of HSM to capture the spatiotemporal features of erroneous behaviors. Surprisingly, many parts of these results were also present even before HSM learning, showing the virtue of HSM as a simple-to-use method. The proposed HSM method may help elucidate the mechanisms underlying erroneous behaviors.

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.

Pneumonia is a leading cause of infectious disease mortality worldwide, accounting for approximately 2.5 million deaths annually and 15% of deaths in children under five. Chest X-ray imaging remains the primary diagnostic tool, but accurate interpretation requires radiological expertise that is disproportionately concentrated in high-income settings, creating a diagnostic gap where disease burden is highest. Automated deep learning offers a scalable complement to specialist-dependent diagnosis, yet clinical adoption requires both high accuracy and transparent, interpretable reasoning. Convolutional neural networks (CNNs) have shown strong potential for pneumonia detection from chest X-rays, but two barriers impede clinical translation: the interpretability of black-box models and the computational feasibility of large architectures in resource-constrained settings. Explainable AI (XAI) methods such as Grad-CAM, Grad-CAM++, and Score-CAM address the interpretability barrier, yet systematic quantitative comparisons across multiple CNN architectures remain scarce. Furthermore, CNN architectures widely used for medical image classification carry high parameter counts that limit feasibility in resource-constrained settings, motivating architectures that achieve competitive accuracy with substantially fewer parameters. Here we propose a parameter-efficient deep learning framework for pneumonia detection based on transfer learning, evaluated across three CNN architectures representing distinct architectural families: EfficientNet-B0 with fine-tuning (proposed method), ResNet50, and DenseNet121, trained under identical conditions on the Kaggle chest X-ray dataset (5,863 images). Our method achieved 90% classification accuracy, outperforming both baselines while requiring 4.8x fewer parameters than ResNet50. To evaluate explainability, Grad-CAM, Grad-CAM++, and Score-CAM were applied across all three architectures and compared quantitatively using Intersection over Union against manually annotated lung segmentation masks, Insertion score, and Deletion score, with pairwise statistical validation via Wilcoxon signed-rank tests and Bonferroni correction. Findings show that classification accuracy and XAI explanation quality must be evaluated independently, and that the proposed parameter-efficient architecture offers a favorable trade-off for resource-constrained clinical deployment.

Minimally invasive endovascular procedures offer reduced surgical trauma, shorter recovery times, and improved outcomes, but rely on 2D fluoroscopic X-ray imaging, which provides limited depth perception and exposes patients and clinicians to ionizing radiation. Here we present an augmented reality (AR) system that fuses intravascular ultrasound (IVUS) and electromagnetic (EM) position tracking with preoperative computed tomography (CT) to produce an anatomically accurate, deformation-corrected navigational reference. A robotic device performs ECG-gated pullback of the IVUS probe, capturing 4D aortic motion across the cardiac cycle. We introduce a deep learning architecture for extracting vascular lumen boundaries and side-branch orifices from artifact-prone IVUS streams, and a semantically driven non-rigid CT-IVUS fusion pipeline robust to false positive landmarks. We evaluate the platform with trained surgeons in benchtop phantom studies and in-vivo ovine models, and demonstrate its application to fenestrated endovascular aneurysm repair (FEVAR). Compared to fluoroscopy alone, AR guidance significantly reduces cannulation time, radiation exposure, and cognitive workload, while improving procedural efficiency and safety. Our IVUS-EM and CT aortic datasets are released open source.

In patients with epilepsy, seizures are associated with pathological neural synchronization. However, the preictal period preceding a seizure often exhibits reduced spatial synchronization compared to normal cognition. This observation aligns with the concept of the brain as a complex dynamical system, where a reduction in dimensionality and resilience can precede a phase transition. The Critical Brain Hypothesis suggests a connection between the loss of healthy scale-free behavior and various disorders, including epilepsy. Our study investigates preictal changes by utilizing network features, such as mean node degree and mean clustering coefficient, derived from thresholded correlation matrices of patient intracranial electrocorticographic electrode data. We observed a suppression of intermittent high-synchronization periods within the feature space during the minutes leading up to seizure onset. This constriction of the explored hypervolume in the preictal state indicates a breakdown in the brains ability to maintain normal coherence. We use these preictal changes to predict the probability of seizure onset using a Support Vector Machine algorithm. These discrete predictions can then be combined into real-time continuous seizure risk forecasts via Bayesian updating. This innovative and computationally lightweight approach has the potential to significantly improve upon static predictions, providing opportunities for more adaptable, quantitative, and interpretable tools for managing seizures.

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

Accurate measurement of intra-cardiac blood oxygen (O2) saturation is essential for cardiovascular assessment, yet current methods require invasive catheterization. T2-based cardiac magnetic resonance imaging (CMRI) enables non-invasive O2 quantification, but deep learning automation is constrained by scarce annotated data. We propose a unified self-supervised learning (SSL) framework integrating cine CMRI and T2 oximetry CMRI to learn generalizable representations without labels. Our approach pre-trains ResNet and vision transformer encoders using contrastive learning and masked image modeling on over 48,000 cardiac images. Pre-trained encoders are fine-tuned for O2 saturation regression with uncertainty quantification to enhance clinical trustworthiness. Our SSL framework significantly outperforms traditional radiomics and supervised baselines, with SimCLR pre-trained ResNet achieving a mean absolute error of 3.70, representing over 15\% improvement. These findings demonstrate SSL's potential to address annotation bottlenecks in non-invasive cardiac diagnostics.

Non-contact mechanical bed sensors can passively and longitudinally monitor the dynamics of cardiopulmonary physiology to detect changes from patient-specific baselines and facilitate care. This requires accurate longitudinal quantification of established metrics like respiratory rate (RR) and heart rate (HR) from the underlying raw waveforms, validated against ground truth labeled datasets like simultaneous polysomnography (PSG). Whereas head-to-head labeled datasets are scarce and costly to collect, unlabeled real-world datasets are often abundant. Here, we show that non-optimized heuristic algorithms can be used to soft-label large real-world data (>40M minutes across >50,000 nights) for pretraining models. This enables label-efficient fine-tuning on small numbers of head-to-head PSG-labeled datasets while maximizing generalizability and robustness to hyperparameters. The result is a highly performant validated model with mean absolute errors (MAE) of 0.6 brpm for RR and 1.1 bpm for HR across 1-minute windows. Although demonstrated in the context of bed sensor cardiopulmonary quantification, these methods are applicable to development of sensor algorithms whenever labeled ground truth data is scarce and unlabeled real-world data is abundant.

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.

Background To build a clinically translatable neonatal seizure detection algorithm using amplitude-integrated electroencephalography (aEEG) and compressed spectral array (CSA). Methods Using a public dataset of annotated neonatal EEGs, features of the aEEG and CSA were extracted from the left and right centroparietal electrodes. These features were then used to train and test three machine learning classifiers, Random Forest (RF), Support Vector Machines (SVM), and Artificial Neural Networks (ANN). Results The trained RF, SVM, and ANN classifiers had areas under the curve (AUC) of 0.80, 0.69, and 0.79 for capturing seizure time periods and an average accuracy of 0.91, 0.90, and 0.92 respectively for capturing seizure and non-seizure time periods. Median accuracy scores were higher among patients without hypoxic-ischemic encephalopathy (HIE; median = 1 for all three classifiers) than HIE patients (median = 0.92, 0.93, 0.93). Conclusion A clinically interpretable aEEG-CSA algorithm is feasible for neonatal seizure detection by extracting standard EEG features and coupling these features with a supervised ML classifier.