IEEE Access

This paper presents a two-stage pipeline for implicit feature engineering in time series-based physiological stress detection using electrodermal activity (EDA) signals. In the first stage, we forecast three descriptive statistics of future EDA signals over short horizons (3, 5, and 10 seconds) based on a 60-second context window. In the second stage, a lightweight linear classifier detects stress from these predicted statistics. We evaluate three forecasting architectures spanning the domain expertise spectrum: a domain-specific bidirectional long short-term memory (BiLSTM) recurrent neural network, zero-shot and fine-tuned variants of Amazon Chronos T5 time series foundation model, and the Tabular Prior-data Fitted Network (TabPFN) applied to engineered physiological features. Experiments on the publicly available Wearable Stress and Affect Detection (WESAD) dataset, comprising chest-worn multimodal physiological signals from 15 subjects under baseline and stress conditions, demonstrate that the domain-specific BiLSTM achieves the highest classification performance, with area under the receiver operating characteristic curve (AUC) values ranging from 0.913 to 0.962. TabPFN follows with AUC values of 0.853-0.869, while Chronos variants yield 0.528-0.744. Notably, models using predicted features consistently outperform those using oracle features derived from the true future signals--the theoretical upper bound--suggesting effective noise filtering through learned sequence representations. Chronos models quickly reach performance saturation regardless of training depth, highlighting challenges in tokenizing continuous physiological time series. The proposed approach advances implicit feature engineering for wearable stress monitoring by leveraging forecasting as a powerful inductive bias, thereby improving robustness and providing insights into the limitations of the foundation model for physiological signals.

Breast cancer survivors frequently experience upper-limb impairments, making continuous monitoring essential for effective rehabilitation. We propose REINA (Recognize-Then-Infer Wearable-to-App AI Framework), a two-stage deep-learning approach for remote monitoring of motor function during breast cancer rehabilitation using wearable-device data. Inertial measurement unit (IMU) signals from wearable devices are first used to recognize physical activities via supervised learning, followed by an activity-specific recurrent neural network (RNN) to infer corresponding electromyography (EMG) signals. REINA establishes reliable inference of neuromuscular activity from wearable IMU data, enabling real-time, cost-effective assessment of motor function recovery in real-world settings.

Background and ObjectiveAccurately parameterizing dynamic, time-varying interactions in physiological systems is a methodological challenge, as global causal discovery methods may obscure transient, local fluctuations. This study introduces tempord, an open-source Python library designed to estimate local temporal orders and evaluate the short-term stability, directionality, and strength of causal links in non-stationary biological signals. MethodsThe algorithm estimates temporal relationships by keeping one signal stationary while iteratively shifting another one within a sliding window. To parameterize optimal inter-signal shifts (causal vector, CV), the framework utilizes linear modeling or time series distance metrics. The methodology was validated through a simulation study on synthetic bivariate signals with mathematically imposed dynamic phase delays, under both deterministic and noisy conditions. Furthermore, in-vivo capabilities were demonstrated by evaluating cardiorespiratory coupling dynamics across spontaneous and music-induced relaxation breathing states. ResultsThe simulation study demonstrated that the extracted CV trajectories precisely aligned with ground-truth temporal delays, assessed using mean absolute error and root mean square error for both noise-free and noisy synthetic data. In-vivo application demonstrated dynamic temporal stability and the detection of minor step changes during autonomic nervous system state transitions. ConclusionsThe tempord Python package bridges the gap between global causal discovery and local beat-by-beat statistical parameterization. It provides a robust "bottom-up" analytical instrument for investigating the transient mechanisms governing complex biological networks.

Photoplethysmography (PPG, optical measurement of cardiac blood volume changes) is the foundation of wearable cardiac monitoring, but systematically fails on dark skin due to melanin absorption. We present the Melanin Absorption Invariance (MAI) framework: a label-free method that substantially reduces cross-skin-tone bias in cardiac feature extraction by preserving topological rather than geometric signal structure. We prove two theorems: Theorem 1 bounds attractor bias to O(SNR_eff^-1) under Z-normalization; Theorem 2 reduces residual bias to O(SNR_eff^-2) via SNR-adaptive correction. Empirical validation confirms these theoretical predictions on real dark-skin PPG signals. Comprehensive empirical validation on the complete MMPD dataset (Fitzpatrick III-VI, n = 656 recordings, 33 subjects, spanning all 4 lighting conditions and 5 motion types, Samsung Galaxy mobile phone) demonstrates MAI generalization across real-world deployment conditions. Results show substantial attractor bias reduction across all skin tone groups, with largest effects for Fitzpatrick IV and VI populations most affected by current systems. This work demonstrates a theoretically grounded, label-free, skin-tone-invariant cardiac monitoring framework.

Scientific image sequences often feature recurring structures with unknown appearance and dynamics. While traditional methods using predefined templates, noise models, or motion classes work well when targets are known, such assumptions can hinder exploratory analysis where discovering that knowledge is the primary goal. This work presents a self-supervised framework that maps noisy video into a discrete vocabulary of local states and their spatiotemporal relationships. The method operates without clean targets, pretrained representations, semantic labels, feature templates, or prescribed trajectories, assuming only that informative structures recur for a finite duration within a bounded spatial neighborhood. A temporally masked vector-quantized network predicts the missing central frame from its neighbors, while a masked Transformer refines token assignments by weighing encoder evidence against contextual compatibility. Project-specific token sets define the feature family, while a support model quantifies the temporal evidence for individual occurrences, facilitating optional conservative suppression. The framework is first demonstrated on synthetic videos of moving particles. Without access to clean frames or particle coordinates during learning and selection, the learned states recover compact, point-spread-function-like structures. Tests on low-signal particle-tracking benchmarks yield frame-wise component precision between 87.0% and 94.5%, with recall decreasing as particle density rises. An experimental example using E. coli further shows that discrete states can separate biological structures from illumination and acquisition artifacts. In all cases, the primary outputs are interpretable state, activity, support, and component maps; rendered images serve as diagnostics rather than optimization targets.

Parkinsons Disease (PD) is a progressive neurodegenerative disorder which significantly affects motor function, daily coordination and verbal communication. Speech-based biomarkers provide a non-invasive and scalable approach to early detection, as dysphonia is one of the earliest and most consistent clinical markers of PD. The dataset used in this study is publicly available and consists of 756 voice recordings from 252 subjects (188 with PD and 64 neurologically healthy controls) with a wide range of acoustic parameters such as Mel-Frequency Cepstral Coefficients (MFCCs), energy-based parameters, and higher-order statistical derivatives. After systematic preprocessing and z-score normalisation, five machine learning classifiers were tested: K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Support Vector Machine (SVM), and Naive Bayes (NB) under a subject-independent, GroupKFold cross-validation protocol. The KNN classifier performed best overall with an accuracy of 92.10%, F1 score of 94.50%, and a precision rate of 98.09%, reducing the number of false positive diagnoses. To overcome the lack of interpretability of black-box predictive models, SHapley Additive exPlanations (SHAP) were used to explain the contribution of each feature to the prediction of an individual. The most diagnostically salient acoustic biomarkers were identified as features from the SHAP analysis: std delta delta log energy, the first Mel-Frequency Cepstral Coefficient, and Tunable Q-Factor Wavelet Transform (TQWT). This work introduces a machine learning framework that is both reproducible and clinically interpretable, combining high predictive accuracy with transparent, physiologically grounded decision logic.

Parkinson disease (PD) affects speech production, motivating classification from repeated voice measurements. Record-wise evaluation can place recordings from the same participant in both training and test sets. We propose VoxNeuro, a novel subject-level framework combining Grassmann representations of repeated recordings with Euclidean mean-dispersion summaries through a provably positive-semidefinite kernel for support vector classification. Features undergo training-fold Gaussian rank normalization, and a supervised partial least squares subspace ensemble is activated when the feature count exceeds the number of training recordings. Two public cohorts were evaluated, UCI-489 (80 subjects) and PD-252 (252 subjects), each with three sustained-vowel recordings per subject. At the reference five-fold subject-disjoint partition, balanced accuracy was 87.5% and 81.6%, respectively, exceeding all five standard comparators under identical folds and normalization. PD-252 sensitivity was 89.9% and specificity was 73.4%. Across 20 additional subject partitions, PD-252 balanced accuracy averaged 78.4%, with a mean paired advantage of 3.2 percentage points over the highest-scoring standard comparator. VoxNeuro exceeded each standard comparator in at least 19 partitions for both balanced accuracy and macro-F1. Ablations favored rank normalization and the unaugmented supervised-subspace configuration on PD-252. VoxNeuro thus offers an effective and reproducible framework for subject-level analysis of repeated speech measurements in PD screening research.

Objective quantification of motor abnormality in Parkinson's disease (PD) remains challenging because conventional clinical assessments are episodic, observer-dependent, and based on coarse ordinal ratings. Wrist-worn wearable sensors provide a scalable opportunity to capture task-specific motor patterns; however, most existing approaches focus on discrete classification rather than deriving continuous latent markers of motor abnormality. In this paper, a multi-stream deep representation network is introduced to derive a latent motor abnormality score using wrist-worn inertial sensor signals. The proposed network includes a local feature extractor based on one-dimensional convolutional layers, a global feature extractor based on Transformer layers, and an embedding layer that maps each task-specific signal into a 64-dimensional embedding vector. A new training procedure is proposed based on a combination of supervised contrastive learning and center loss to cluster the embeddings of PD patients and healthy control (HC) subjects, and to build a latent motor abnormality score based on the average embedding distance from the centroids of PD and HC training embeddings. Across five-fold subject-level cross-validation, the proposed method was assessed for PD/HC classification and achieved an accuracy of 83.10% +/- 3.15%, balanced accuracy of 85.89% +/- 3.93%, precision of 97.02% +/- 2.33%, and ROC-AUC of 0.917. The motor abnormality score showed clear group separation, with healthy controls generally obtaining negative scores and PD subjects obtaining positive scores. Statistical analysis confirmed a significant difference between groups (F = 222.99, p < 0.001), with a substantial effect size (eta-squared = 0.3871). Moreover, exploratory analyses showed that the score tended to increase with higher non-motor symptom burden and longer disease duration, suggesting that the learned latent representation captured disease-related motor abnormalities.

Multiview data analysis has become increasingly important for integrating heterogeneous information from multiple data sources. Cooperative learning provides a flexible framework for multiview integration by bridging early and late fusion strategies. However, its use of squared loss makes it sensitive to outliers. In this study, we propose robust cooperative learning based on the soft-IPOD framework. The proposed method introduces an observation-specific outlier absorption vector into the cooperative learning objective, thereby reducing the influence of outlying observations while preserving information sharing across views. The resulting optimization problem can be formulated as a weighted Lasso-type problem using an augmented design matrix and can be naturally extended to more than two data views. Numerical simulations and a real-data application to labor onset prediction demonstrate the predictive performance of the proposed method in the presence of outliers.

Background: The disconnect between episodic in-clinic assessments and patients goals to improve activity in daily life presents a challenge in rehabilitation. Wearable sensors are promising for addressing this disconnect but are not part of routine practice. The purpose of this study was to develop and test a digital health solution that integrated wearable sensor technology into outpatient rehabilitation. Methods: Using an iterative user-centered design approach, a digital solution was built that integrated data from a wrist sensor into the electronic health record (EHR). The feasibility, usability, and resources required of the solution were evaluated in a 9-month proof-of-concept study in outpatient rehabilitation. Results: 17 clinicians and 38 patients (median monitoring duration: 42 days) participated. Patient adherence to daily data syncing was a median of 68.6% and was not significantly correlated with digital technology self-efficacy (r = 0.03, p = 0.86). Recent sensor data ([&le;] 3 days old) were available in the EHR for 90.2% (193/214) of clinic visits, and clinicians viewed these data in 81.9% (158/193) of these visits. Overall user-friendliness on the System Usability Scale received a median rating of "good" from both patients and clinicians. Resources included costs to build the data transfer pipeline, a 24-month build timeline, and ongoing technical support. Conclusions: By combining user-centered design principles with dedicated resources, we built a feasible and usable digital health solution to bridge the gap between brief clinic visits and daily life in outpatient rehabilitation. These results and future refinements will facilitate broader-scale implementation of sensor technology in rehabilitation.

Although largely preventable, cardiovascular disease (CVD) is the leading cause of death worldwide. Synthetic data generation (SDG) of health data can facilitate data sharing for research and development of personalised cardiovascular disease prevention tools. Current state-of-the-art SDG methods are limited to cross-sectional data or focused on short-term or in-patient EHR applications lacking benchmarks for generalisability. Since prevention is a long-term longitudinal objective, these limitations must be addressed. We evaluated two tabular generative adversarial networks, TGAN and CTGAN, for privacy and fidelity on a structured 6-year longitudinal trial dataset within CVD prevention. We compared performance with a rudimentary perturbation method. In addition, we performed a variable-wise evaluation and an age-stratified analysis of the higher performing model to assess preservation of clinically relevant characteristics. Overall, CTGAN showed better fidelity than TGAN. The models performed similarly within privacy, showing better results than the perturbation method. CTGAN had lower performance in comparison to the perturbation method within numerical fidelity of continuous variables and overall temporal fidelity. Furthermore, CTGAN's performance was lower for younger age groups than older. CTGAN shows better than or equal performance within fidelity and privacy than TGAN and the perturbation method on a long-term longitudinal dataset within CVD prevention. To the best of our knowledge, this is the first study evaluating CTGAN on a longitudinal dataset and our findings suggest that handling of temporal characteristics and imbalanced subgroups need further refinement.

Discourse analysis can reliably predict real-world communicative success, yet is rarely implemented clinically, due to challenges in manually generating stimulus-specific main concept inventories (MCIs) and scoring patient narratives. We evaluated whether automatically derived macrolinguistic (main concept, sentiment) and microlinguistic (fluency, lexical, syntactic) features extracted from movie clip narrations can augment the clinical assessment of discourse in aphasia. The MCIs generated directly from each clip's video, audio and subtitles using a vision-language model approximated the main concepts healthy controls produced. Group classification (control: n=51; aphasia: n=54) based on the discourse features was excellent (AUC=0.95, 95% CI [0.90, 0.99]), with macrolinguistic features showing the largest group differences (main concept completeness: d=-1.84, 95% CI [-2.30, -1.38]) and strongest associations with aphasia severity (semantic distance to inventory: r=-0.79, 95% CI [-0.87, -0.66]). Findings support the feasibility and scalability of automated criterion-based discourse assessment using ecological stimuli, and highlight its potential for clinical implementation.

Noncontact dynamic optical coherence elastography (OCE) can measure the elastic properties of anterior soft tissues. However, applying OCE to skin requires tracking mechanical waves propagating in multiple directions to reconstruct mechanical anisotropy. Robotic scanning is well suited to this task but requires accurate probe positioning, alignment to the surface normal, and precise, repeatable rotation. To ensure phase and polarization stability, polarization-maintaining (PM) fibers can be used, although they introduce stripe and ghost artifacts that can destabilize surface detection. We present DINOCT, a DINO-based method to localize the skin surface from optical coherence tomography (OCT) images that combines self-supervised pretraining with DINO (self-DIstillation with NO labels) and iBOT (image BERT pre-training with Online Tokenizer) objectives, low-rank adaptation, and a lightweight curve decoder. Compared with classical and supervised learning baselines, DINOCT achieved competitive localization accuracy on clean test data with a near-zero spike rate. Among the learned methods, DINOCT achieved the lowest average and maximum values of mean absolute error (MAE) for severe synthetic stress conditions and the lowest catastrophic failure rates on a separate recording with strong PM-fiber artifacts.

Inappropriate antibiotic use presents a major global health challenge, particularly in low-resource settings where access to quality care is limited but antibiotics remain relatively unrestricted. This study estimates the causal effect of frontline primary care quality on inappropriate community antibiotic use, combining detailed community-based data from approximately 100 rural villages in rural China with an instrumental variable (IV) approach embedded within a double/debiased machine learning (DML) framework. We linked objective measures of village doctor clinical practice quality, measured through unannounced standardized patient visits, to household-level antibiotic use data collected from the same villages. To identify the causal effect, we constructed multiple candidate instruments from extensive provider characteristics and used an ensemble of machine learning algorithms within a flexible DML-IV framework to approximate an optimal instrument, addressing a many-weak-instruments problem. We found that improving village provider clinical practice quality reduced both antibiotic receipt during healthcare encounters for common diseases and household antibiotic storage for future self-medication. Our findings suggest that strengthening frontline primary care quality can meaningfully reduce inappropriate community antibiotic use without restricting access to essential treatment. More broadly, this study illustrates how causal machine learning can strengthen conventional causal estimation in complex observational settings in global health economics research.

Tabular synthetic data generation (SDG) can facilitate research and development in the health domain, as access to health data is restricted. To accommodate for its structural complexity, such as multimodal distributions and fragmentation, transformer models provides an alternative to conventional SDG methods since they are better equipped to infer global spatial coherence due to textual context of tabular records. However, methods for optimising their adaption to tabular SDG are still largely unexplored. We contribute to the understanding of transformer-based SDG by evaluating fidelity and privacy of three prompting regimes, zero shot, one shot and few shot training, for tabular health SDG using small transformer models. We evaluated llama 3.2 (1B and 3B parameters) and Phi (1.3B and 2.7B parameters), on a real clinical dataset. Our results showed that zero shot learning was less prone to hallucinations and performed overall better for fidelity, but at the loss of privacy in comparison to the other regimes. Our results showed no differences between model sizes and families. Future studies should extend the analysis to other models and datasets, and explore strategies to reduce hallucinations.

FeePredict is a three-stage random forest machine learning framework to simul-taneously predict whether Medicare reimbursement rates for specific procedures will change, in which direction they will change, and by how much. FeePredict was ap-plied to the four major Medicare fee schedules: the Clinical Laboratory Fee Schedule (CLFS), the Physician Fee Schedule (PFS), the Ambulance Fee Schedule (AFS), and the Durable Medical Equipment, Prosthetics, Orthotics, and Supplies (DMEPOS) fee schedule. Each of these fee schedules contains publicly available data from the Centers for Medicare & Medicaid Services (CMS) for the years 2024, 2025, and 2026, with the number of procedures represented in the data ranging from 3,264 to 2,952,842 observations.FeePredict utilizes lag-1 feature engineering and train-only preprocessing steps to ensure that there is no data leakage into the model. Chronological out-of-time valida-tion was performed on three of the four fee schedules to determine the generalizability of the model over time. FeePredict significantly outperformed the assumption that there would be no changes to Medicare reimbursement rates for procedures (p < 0.001), achieving concordance indices between 0.815 and 0.998, and reducing the mean abso-lute error for predicting changes to reimbursement rates by 29% to 85%. Permutation testing of the model with shuffled reimbursement rate labels indi-cates that there is no evidence of data leakage (AUC values: 0.467-0.515). The model achieved concordance indices of 0.854 and 0.972 for the CLFS and DMEPOS fee sched-ules, respectively, outside of its training period, but performed less well outside of its training period for the PFS, indicating that it generalizes less well to changes to the Medicare policy regime that existed after its training period. Overall, though, these re-sults indicate that it is possible to accurately predict whether Medicare reimbursement rates for medical procedures will change using only data from the historical versions of those fee schedules.

Conventional electrocardiography is highly effective for waveform and rhythm diagnosis, but it is less suited to showing how the internal shape of hundreds or thousands of consecutive heartbeats changes over time. We introduce FOXTAIL, a complementary view that represents each cardiac cycle as an ordered sequence of changes in signal direction. Overlaying these sequences in a fixed visual field makes beat-to-beat organization visible and allows the density, size, stability, and scale persistence of those changes to be measured. We evaluated the representation in recordings containing normal sinus rhythm, paroxysmal atrial fibrillation, severe heart failure, ventricular tachyarrhythmia, and controlled electrode-motion noise. Paired recordings showed that FOXTAIL descriptors can reveal within-person state changes that are not conveyed by a single average beat. The noise and pre-fibrillation analyses also showed that a dense event pattern is not automatically equivalent to physiological complexity, measurement artifact, or impending disease. FOXTAIL is therefore not proposed as a replacement for the diagnostic ECG or as a new classifier, but as an observation and measurement domain for asking a more basic question: how is the electrical organization of the heart changing from one beat to the next, and which of those changes persist across scale?

Despite the use of technological countermeasures, studies show that the sensory conflict, and thus symptoms of motion sickness such as nausea, cannot be completely prevented while traveling as a car passenger. Accordingly, reliable prediction of motion sickness will represent a possible solution for intelligent vehicles to avoid the abortion of the drive, e.g., through adapted driving strategies or route planning. In previous works, the authors were able to demonstrate an application-oriented parameter set for the popular 6DOF-SVC model to estimate motion sickness incidence (MSI) in passengers with a lowered gaze on a non-driving-related task. In the current study, these parameters were validated and further optimized using five previous conducted motion sickness driving experiments. For the first time, it is also shown that the expected natural MSI fluctuations based on the driving dynamic profile, due to route characteristics or driving style, as well as the variation in experimental determined MSI is approximately 10%. Moreover, findings indicate that the participants appear to have an individual MSI threshold that correlates with their susceptibility to motion sickness, however, this threshold fluctuates too strongly across experiments to be suitable for practical application. As a result, the model can successfully estimate motion sickness risk across the test tracks, however, inaccuracies in the predicted MSI trajectories remain.

Background: System One Model is designed for structured decision-making and can provide probabilistic outputs, but their performance in domain-specific physical therapy tasks has not been established. Objective: To evaluate the performance and potential utility of Jev, a System One Model, for professional knowledge-based selection tasks in physical therapy by comparing it with GPT-4o. Methods: A total of 380 multiple-choice questions from the 2024 and 2025 Korean Physical Therapist Licensing Examinations were independently submitted to Jev and GPT-4o. The outcomes included answer accuracy, subgroup performance, output validity, API response latency, token usage, estimated API cost, and Jev's prediction uncertainty. A retrospective probability-based model-cascading simulation was also performed. Results: Jev correctly answered 282 questions (74.2%), whereas GPT-4o answered 326 (85.8%), a difference of -11.6 percentage points. Jev produced valid structured responses for all questions, whereas GPT-4o produced three invalid responses. Jev's selected-answer probabilities discriminated correct from incorrect answers (AUROC = 0.866), and accuracy reached 99.3% when probabilities were greater than or equal to 0.90. The median API response latency was 0.94 s for Jev and 1.06 s for GPT-4o, with estimated costs of US$0.008 and US$0.161, respectively. At a retrospective cascading threshold of 0.70, accuracy was 86.6% with 175 simulated GPT-4o calls. Conclusion: Jev showed lower overall accuracy than GPT-4o but provided reliable structured outputs, informative prediction probabilities, lower latency, and lower estimated cost. Its probabilistic outputs warrant further prospective evaluation for selective model escalation and decision-support workflows in physical therapy.

The -Bonacci weighted sequence space {ell}(,) is introduced for [&ge;] 2, 1 [&le;] < {infty}, and > 0). It is proved to be a Banach space which is isometrically isomorphic to {ell}, with an unconditional Schauder basis. The space is also reflexive and uniformly convex. Its strict inclusion chain, continuous dual, and asymptotic truncation estimate of order - / are also established. Based on this framework, Scaled &#119898-BonacciNet is proposed as a lightweight CNN that rescales Fibonacci-Net channels using the dominant &#119898-Bonacci root and employs depth wise separable convolutions. Tribonacci-Net achieved 96.09% accuracy and a 0.9581 macro F1-score on the 15-class Brain MRI dataset using 0.85 million parameters and 0.366 GFLOPs. Pentanacci-Net attained 95.87% accuracy on the 44-class dataset while reducing parameters and FLOPs by 47.10% and 44.48%, respectively. On the CKD ultrasound dataset, Hexanacci-Net achieved 83.75% accuracy and a 0.7941 macro F1-score, with corresponding reductions of 50.37% and 45.10%. These findings support dominant-root scaling as a structured approach to lightweight CNN design.