Convolutional neural networks (CNNs) have shown outstanding performance in image recognition, but their application to non-sequential tabular data remains debatable. This study investigates the architectural sensitivity of CNNs when applied to non-sequential medical datasets and compares their performance with multi-layer perceptrons (MLPs) under various structural settings. Three publicly available medical tabular datasets were used: integrative clinical and CT feature dataset (iCTCF), Breast Cancer Wisconsin Diagnostic (BCWD), and UCI Heart Disease (UCI-HD). We systematically varied the number of kernels, kernel sizes, and fully connected (FC) nodes in a 1D-CNN architecture and compared the classification performance with that of MLP models, while conducting 1,000 feature-order permutation experiments to quantify order sensitivity under randomized structural settings. Effect-size statistics were computed to describe class separability; no feature filtering was performed. Across permutations, MLPs demonstrated superior stability with significantly tighter dispersion than CNNs across all datasets. While CNNs achieved peak AUROCs comparable to (BCWD: 0.987 vs. 0.986) or higher than (iCTCF: 0.739 vs. 0.681) MLPs in certain configurations, they exhibited greater performance variability and a distinct negative skew, reflecting high sensitivity to feature ordering. In UCI-HD, the peak AUROC favored MLP (0.878 vs. 0.829). Post-hoc analyses confirmed that CNN performance is highly contingent on structural hyperparameters-particularly kernel size-rather than robust feature learning. CNN performance on tabular data is heavily dependent on arbitrary feature ordering and structural design, posing risks of stochastic degradation. Clinical AI applications using such data must prioritize stability over peak performance and account for the lack of inherent spatial structure in tabular inputs.

Smart wheelchairs enhance mobility and independence for individuals with physical impairments, including older adults. However, challenges remain in sensor reliability, affordability, and real-world applicability. This systematic review (PROSPERO ID: CRD420250642191) analyzed 57 studies (Jan 2016–Jan 2026) from IEEE Xplore, Scopus, PubMed, Embase, and ScienceDirect, following PRISMA guidelines. We examined control modalities (e.g., brain-computer interfaces, eye tracking, voice, gesture, and multimodal), sensor integration (e.g., cameras, LiDAR, ultrasonic, IMU, and EEG), computational platforms, and safety mechanisms. Most studies (45.6%) focused on non-autonomous designs, with fewer on semi- (45.6%) or fully autonomous (8.8%) systems. While healthy participants were commonly used for validation, few studies included individuals with mobility impairments. Evidence indicates smart wheelchairs can reduce user effort, improve safety, and support daily activities, yet adoption is limited by sensor fragility, high costs, lack of standardization, and limited long-term evaluation. Future research should focus on developing robust, interoperable sensors; integrating smart homes and tele-rehabilitation; conducting inclusive testing; fostering cross-disciplinary collaboration; and securing regulatory support to enable scalable, user-centered mobility solutions.

Contactless heart rate (HR) monitoring demonstrates significant potential for mobile health and telemedicine, but current remote photoplethysmography (rPPG) approaches remain vulnerable to various noise sources. While existing research has emphasized signal-level enhancement, correcting erroneous HR estimates remains underexplored. We present a plug-and-play adaptive correction algorithm that leverages cardiac dynamics constraints, adjusting HR estimates based on physiological priors of HR elevation and recovery. By mapping HR frequencies to indices and applying adaptive corrections, our method significantly reduces measurement errors with minimal computational load, even under challenging conditions. Across three public datasets, the algorithm increased the proportion of accurate measurements (mean absolute error ≤ 10 beats per minute) from 46.26% to 84.14% (LGI-PPGI), 48.03% to 69.21% (BUAA-MIHR), and 92.22% to 96.67% (UBFC-rPPG), outperforming existing correction techniques. The lightweight design facilitates seamless edge-side integration, providing a scalable solution for enhancing the reliability of contactless HR monitoring in mobile and remote healthcare settings.

Cuffless blood pressure monitoring is becoming increasingly feasible with the rise of wearable technologies, offering significant promise for preventive cardiovascular care. Yet diverse sensing modalities and modeling strategies have produced fragmented evidence that is difficult to compare. In this systematic review, we introduce a unified multi-axis taxonomy that relates surrogate signals, modeling paradigms, and calibration strategies, providing a coherent structure for the field. Building on this foundation, we analysed current approaches across physiological and wearable sensing and evaluated performance using best machine-learning practices. We highlight key barriers to real-world deployment—including protocol realism, calibration drift, population diversity, and fairness relevant reporting—and translated these findings into an actionable model-card-style reporting framework for cuffless BP studies. Our framework establishes the methodological basis for reliable, equitable, and clinically valid cuffless BP monitoring, a vital step towards democratizing continuous BP monitoring for widespread clinical benefit.

Ensuring fairness in clinical machine learning is a major concern, yet the dominant driver of unequal performance across sex groups remains unclear: is it the dataset or the algorithm. We conducted a systematic fairness evaluation across three healthcare domains-wearable physiology (MHEALTH), cardiac risk prediction (UCI Heart Disease), and stroke assessment-using ten widely used classifiers and three controlled sex-ratio sampling scenarios (50/50, 90/10, 10/90) under an identical analytical pipeline. Gender accuracy gaps varied markedly across datasets and exhibited dataset-specific patterns that did not generalize across clinical domains. Mixed-effects interaction modelling showed that the same algorithm could display negligible bias in one dataset and substantial bias in another. Variance contribution decomposition of the absolute Gender Accuracy Gap (∣GAG∣) indicated that dataset identity accounted for most of the observed variability (63.4%), with additional contribution from dataset-algorithm interactions (17.2%); algorithm choice alone explained 9.7%, whereas sampling scenario contributed negligibly (0.2%). Balanced sampling reduced disparities but did not eliminate them, consistent with residual sex-associated signal/feature structure beyond representation imbalance. These findings demonstrate that fairness in healthcare machine learning is primarily dataset-dependent, motivating dataset- and context-specific auditing before clinical deployment.

Wearable technology has become increasingly important for health monitoring, sports performance, and ergonomic assessments because it enables continuous, non-invasive, and real-time tracking of physiological and biomechanical signals in real-world environments, overcoming limitations of laboratory-based assessments. This paper presents the development, testing, and initial study of a sensor-embedded loose garment designed for motion analysis using conductive ink. Sensors were strategically placed across key areas of the T-shirt to capture comprehensive motion data from the torso. Positioned on the chest, shoulders, ribcage, and lower torso, these sensors detect detailed movements. The study evaluates various sensor combinations with four classifiers-XGBoost, RandomForest, SVM, and K-Nearest Neighbors-using data from ten sensor locations analyzed with three holdout methods (20-80%, 30-70%, and 50-50%). Results underscore the impact of specific sensor placements, with combinations on the shoulder, ribcage, and abdomen yielding the highest accuracy. This work advances textile-based motion recognition, showing the potential for wearable technology to distinguish among eight movements in a loose garment.

Remote photoplethysmography (rPPG) can evaluate real-time changes in blood flow volume by capturing facial videos and analyzing the color changes. Although the rPPG technique enables contactless heart rate (HR) monitoring, it remains highly susceptible to ambient lighting variations. Previous studies have demonstrated the critical influence of facial skin detection on the extracted rPPG signals. In this research, we defined 15 facial regions of interest (ROIs) based on anatomical criteria and evaluated their HR measurement performance on two public datasets (BUAA-MIHR and MMPD) using four representative rPPG algorithms (CHROM, LGI, OMIT, and POS). The experimental results confirmed that the glabella, nasal dorsum, and malar regions consistently serve as robust physiological signal sources under complex illumination environments. Furthermore, we revealed the significant advantage of multi-ROI combinations compared to both single-ROI and holistic-face strategies. Our work provides data-supported insights for establishing novel HR measurement pipelines with enhanced accuracy and robustness.

Anxiety disorders affect hundreds of millions of people worldwide, yet objective and continuous assessment remains limited in clinical practice. To our knowledge, this is the first modality-specific, translational synthesis focusing on wearable ECG and PPG for anxiety detection. Wearable electrocardiography (ECG) and photoplethysmography (PPG), combined with data-driven analytics, have emerged as promising tools for anxiety monitoring, but translation into routine care has been slow. Here, we present a PRISMA-guided systematic review of 38 studies (2015-2025) investigating wearable ECG- and PPG-based anxiety detection. We analyze anxiety induction paradigms, sensor configurations, signal acquisition strategies, and analytical approaches, including statistical, machine learning, and hybrid methods. While autonomic markers derived from ECG and PPG consistently reflect anxiety-related physiological changes, substantial heterogeneity in study design, limited population diversity, and laboratory-centric validation constrain clinical generalizability. Critically, most studies lack evaluation in real-world settings and do not demonstrate clinical utility or impact on patient outcomes. We identify key translational barriers and propose a digital medicine roadmap emphasizing standardized protocols, robust validation across diverse populations, workflow integration, and outcome-driven evaluation to enable clinically actionable, real-world anxiety monitoring.

We report a fully textile-based interdigitated capacitive (IDC) strain sensor for wearable posture recognition. The sensor is lightweight, comfortable, and washable, and maintained stable electromechanical performance during 6,500 strain cycles, with absolute value of the gauge factor ranging from 0.67 to 0.81 across the tested strain conditions. Integrated into clothing, the sensor captured multidimensional electrical responses, including capacitance-, resistance-, and phase-related signals, during four biomechanically distinct yoga postures: High Lunge, Lunge Forward, Squat, and Tree. Using a supervised machine-learning pipeline combining handcrafted sensor-channel features and MiniROCKET time-series features, record-level classification achieved an accuracy of 94.4% and a macro-averaged F1 score of 94.2%. Record-level five-fold cross-validation further showed stable performance across folds, supporting the robustness of the classification framework within the present controlled cohort. One-vs-rest receiver operating characteristic analysis yielded area under the curve values of 90%, 94%, 99%, and 99% for High Lunge, Lunge Forward, Squat, and Tree, respectively. Channel ablation analysis identified capacitance-derived features as the most informative predictors. Statistical and temporal analyses further showed posture-dependent differences in signal magnitude, normalized response, and waveform dynamics, while supervised low-dimensional projection revealed distinct class structure. These results demonstrate the feasibility of textile-based IDC sensing for wearable movement monitoring applications such as such digital fitness.

BACKGROUND:There is growing interest in using biosignals from wearable devices to assess anxiety disorders. Among these, electrocardiography is the most widely used due to its ability to monitor cardiovascular activity. Other signals, such as respiratory, electrodermal activity, and photoplethysmography, also show promise. This review aims to evaluate how these signals, individually and in combination, have been used for anxiety detection. METHODS:We systematically reviewed 26 studies published between 2014 and 2024 that used wearable devices to collect signals for anxiety detection. Extracted information included study design, signal types, features, classification methods, and accuracy outcomes. Pooled accuracies were calculated to compare single-signal and multi-signal approaches. RESULTS:Here we show that approaches combining multiple signals outperform those using a single signal, with a pooled accuracy of 81.94% compared to 76.85%. Electrocardiography was the most reliable individual signal, with a pooled accuracy of 80.34% across 12 studies. However, the limited number of single-sensor studies and methodological variability limit conclusions about the superiority of any one modality. The most common features included mean heart rate and heart rate variability for electrocardiography, the mean inspiratory-to-expiratory time ratio for respiratory signals, mean skin conductance for electrodermal activity, and the mean heart rate for photoplethysmography. Support vector machine was the predominant classifier. CONCLUSIONS:This review underscores the clinical potential of wearable devices for anxiety detection, emphasizing the value of multimodal approaches. Future research should focus on refining algorithms, expanding sample sizes, and exploring diverse contexts to improve the accuracy and generalizability of these methods.

Remote photoplethysmography (rPPG) enables non-contact heart rate (HR) and waveform measurement from facial video, offering advantages over contact methods while contending with motion, illumination changes, compression, and frame-rate variability. This roadmap integrates a systematic review, technical synthesis, and a clinic-acquired demonstration to accelerate rPPG research and translation. We outline the measurement pipeline and synthesize progress across eight domains: datasets, device factors, data scarcity, skin/ROI detection, model- and data-driven algorithms, filtering, and evaluation metrics. Strategies for improved robustness include targeted dataset expansion, augmentation, optimized ROI policies, motion and photometric normalization, and hardware configuration. A proof-of-feasibility using data collected in a routine clinical setting alongside reference monitors demonstrates high signal-level agreement under a multi-metric framework (correlation/concordance, absolute errors, and Bland-Altman bias/dispersion). We conclude with recommendations for accuracy, efficiency, and deployment in real-world and clinical contexts, providing a consolidated resource for researchers and clinicians and a pragmatic path toward reliable clinical adoption.

Hemoglobin (Hb) concentration is a fundamental physiological marker widely used in the diagnosis of anemia and the assessment of cardiovascular health. Although invasive blood testing provides high accuracy, its reliance on laboratory infrastructure limits scalability and real-time applicability. Here, we present Hb-PPG, a four-wavelength photoplethysmography (PPG) dataset designed to support research on non-invasive hemoglobin assessment and cardiovascular monitoring. The dataset comprises 1008 PPG signal segments acquired at 660, 730, 850, and 940 nm from 252 adult subjects, alongside reference measurements of hemoglobin, fasting blood glucose, and brachial artery systolic and diastolic blood pressure. Hb-PPG enables systematic investigation of wavelength-dependent PPG signal characteristics and their relationships with hematological and hemodynamic parameters. By providing high-quality, multi-wavelength optical signals with clinically grounded reference data, this dataset facilitates the development, validation, and benchmarking of non-invasive approaches for hemoglobin estimation and related vascular health applications. The dataset is intended to support algorithm development, benchmarking, and methodological studies in non-invasive hemoglobin estimation, rather than direct clinical diagnosis.

Capacitive pressure sensors (CPSs) are increasingly important for wearable and textile-based health monitoring due to their high sensitivity, low power consumption, and structural flexibility. Building on a structured literature review (2015–2025), we conducted a systematic analysis of 34 CPS designs across five categories—microstructuring, foams, ionic liquids/gels/metals, bioinspired architectures, and multisensing/strong bonding—using a newly developed Textile Suitability Score (TSS) that integrates nine performance and integration-relevant attributes. Trade-off analysis revealed that ionic metal-based sensors dominate in raw performance, combining very high sensitivity with broad pressure ranges, but face scalability and textile-compatibility challenges. By contrast, microstructuring offers the closest balance between sensitivity and integration, while multisensing and strong bonding form a consistent cluster with the highest TSS values. Foams and bioinspired designs ranked lower due to reproducibility and stability issues. Together, these results expose clear performance-integration trade-offs and uncover unexplored design pathways, charting a roadmap for next-generation textile-integrated health monitoring systems.

The revolutionary remote photoplethysmography (rPPG) technique has enabled intelligent devices to estimate physiological parameters with remarkable accuracy. However, the continuous and surreptitious recording of individuals by these devices and the collecting of sensitive health data without users' knowledge or consent raise serious privacy concerns. Here we explore frugal methods for modifying facial videos to conceal physiological signals while maintaining image quality. Eleven lightweight modification methods, including blurring operations, additive noises, and time-averaging techniques, were evaluated using five different rPPG techniques across four activities: rest, talking, head rotation, and gym. These rPPG methods require minimal computational resources, enabling real-time implementation on low-compute devices. Our results indicate that the time-averaging sliding frame method achieved the greatest balance between preserving the information within the frame and inducing a heart rate error, with an average error of 22 beats per minute (bpm). Further, the facial region of interest was found to be the most effective and to offer the best trade-off between bpm errors and information loss.

BackgroundIncreasing demands, such as the COVID-19 pandemic, have presented substantial challenges to global healthcare systems, resulting in staff shortages and overcrowded emergency rooms. Health kiosks have emerged as a promising solution to improve overall efficiency and healthcare accessibility. However, although kiosks are commonly used worldwide for access to information and financial services, the health kiosk industry, valued at $800 million, accounts for just 1.9% of the $42 billion global kiosk market. This review aims to bridge the research-to-practice gap by examining the development of health kiosk technology from 2013 to 2023.MethodsWe conducted a systematic search across PubMed, IEEE Xplore, and Google Scholar databases, identifying 5,537 articles, with 36 studies meeting inclusion criteria for detailed analysis. We evaluated each study based on kiosk purpose, targeted diseases, measured vital signs, and user demographics, along with an assessment of limitations in participant selection and data reporting.ResultsThe findings reveal that blood pressure is the most frequently measured vital sign, utilized in 34% of the studies. Furthermore, cardiovascular disease detection emerges as the primary motivation in 56% of the included studies. The United States, India, and the United Kingdom are notable contributors, accounting for 43% of the reviewed articles. Our assessment reveals considerable limitations in participant selection and data reporting in many studies. Additionally, several research gaps remain, including a lack of performance testing, user experience evaluation, clinical intervention, development standardization, and inadequate sanitization protocols.ConclusionsThis review highlights health kiosks' potential to ease the burden on healthcare system and expand accessibility. However, widespread adoption is hindered by technical, regulatory, and financial challenges. Addressing these barriers could enable health kiosks to play a greater role in early disease detection and healthcare delivery.

Sleep disorders affect millions globally, leading to serious health issues. Accurate sleep-wake classification is essential for diagnosis and management. While polysomnography is the gold standard, it is costly and invasive; photoplethysmography (PPG) offers a viable alternative. Using the Cyclic Alternating Pattern Sleep Database (84 participants, 85,542 epochs), we extracted 330 features and reduced dimensionality via statistical tests and the SelectFromModel method. To address class imbalance, we applied Adaptive Synthetic (ADASYN) sampling. A Random Forest model, validated with 20-fold cross-validation on the unbalanced dataset (75 features), achieved an F1 score of 89.05% but struggled with wake detection. With ADASYN balancing and 35 features, it achieved 88.57% sensitivity (sleep) and 71.31% specificity (wake), with an F1 score of 81.40%. This feature-based approach improves PPG-based sleep classification, supporting clinical adoption and integration into wearable devices for remote sleep monitoring.

Remote photoplethysmography (rPPG) is gaining traction for non-contact heart rate estimation, yet most publicly available datasets are demographically biased. In this study, we analyze 100 rPPG studies, providing the first quantitative cross-model audit of demographic bias in rPPG and demonstrating significant underrepresentation of darker skin tones and gender imbalance. Our findings reveal how this bias limits model fairness and accuracy and propose steps to improve dataset inclusivity and algorithmic robustness.

Wearable EEG sleep monitoring devices (wEEGs) are increasingly popular in both clinical and consumer applications. However, their performance compared to polysomnography (PSG), the gold standard, remains under study. This meta-analysis of 43 validation studies assessed wEEGs against PSG, analyzing the influence of study design and device characteristics. The results revealed moderate to substantial agreement between wEEGs and PSG, with performance varying across sleep stages. The N1 stage posed significant classification challenges, while N3 (Deep Sleep) was most reliably detected. Manually scored wEEG data outperformed automatic scoring for N1 detection, and a higher electrode count was associated with improved N3 classification. This study proposes a standardized framework with balanced metrics like MCC and κ to address stage-specific performance variabilities, enhancing device comparability. The findings highlight the strengths and weaknesses of wEEGs and guide future research to refine automatic staging, contributing to their optimization for clinical and consumer applications.

The ability to infer demographic patterns from passive inertial sensing raises important privacy considerations and has potential applications in user-aware systems. In this paper, we investigated whether natural walking patterns can provide indicative demographic insights. Building upon previous work in this field, we constructed an evaluation study to compare the performance of previous methods in IMU soft biometrics, random convolution models, and InceptionTime-based models to estimate subjects’ age and gender under a controlled and fair experimental setup on the OU-ISIR Gait Database. InceptionTime and H-InceptionTime obtained the best performance among all three tasks. The InceptionTime model achieved a mean absolute error of 6.60 years in age value regression, suggesting sufficient accuracy for coarse-grained demographicaware adaptation. It also yielded a correct classification rate of 88.88% in gender classification. H-InceptionTime achieved a correct classification rate of 59.46% in the challenging fourcategory age group classification (notably above the 25% chance level). The experimental results demonstrate the potential of using random convolution models like HYDRA, and advanced InceptionTime-based models, to effectively extract and identify subtle demographic signatures embedded in human movement patterns measured by inertial sensors.

Remote photoplethysmography (rPPG) is a technique that extracts physiological signals, such as heart rate, from facial videos using standard Red-Green-Blue cameras. While rPPG offers valuable health insights, it also exposes individuals to potential misuse, as sensitive information can be inferred without consent. This paper introduces a reversible video modification framework for removing, encrypting, transmitting, and restoring rPPG signals in facial videos, using frame-wise sinusoidal modulation applied to specific rPPG-rich facial regions, with a focus on maintaining perceptual quality and concealing the true heart rate. Our approach is contrasted with prior methods using seven rPPG techniques on the LGI-PPGI dataset, encompassing various activities. Evaluation metrics include PSNR, SSIM, correlation, dynamic time warping, and a composite score reflecting both signal suppression and visual fidelity. Here we show that our method achieves an overall score above 0.75 across all rPPG methods, approximately 50