Open research questions in Non-Invasive Vital Sign Monitoring

What the literature leaves open

• Several limitations of the present study should be ac- knowledged. First, the intervention duration was relatively short (six weeks); although significant effects were observed, longer intervention periods are required to determine the sus- tainability and long-term transfer of the observed cognitive and physiological improvements. Second, the sample consist- ed exclusively of male athletes, which limits the generalizabil- ity of the findings to female athletes and mixed-gender popu- lations. Third, despite using validated devices, optical heart rate (OHR) sensors are susceptible to motion artifacts during highly dynamic movements, which may affect signal preci- sion, particularly in explosive and intermittent sport contexts, even though mitigation algorithms were applied. Fourth, the study focused on performance and physiological outcomes and did not include direct psychological measures such as per- ceived stress, attentional control, or working memory, which would have allowed a more comprehensive explanation of the mechanisms underlying decision-making improvements. Fi- nally, although athletes from different sport types (precision, explosive, intermittent) were included to enhance ecological validity, the study did not conduct sport-specific differential analyses, thereby limiting conclusions regarding whether the magnitude of effects differed across sport categories.

  The Synergy of Wearable Optical Heart Rate Sensor Technology and Box Breathing in Real-Time Decision Making in Precision, Explosive, and Intermittent Sports · 2026 · DOI

• The current study was a pilot, proof-of-concept study conducted in the ED ambulatory wings. The patients enrolled in the study were triaged to low acuity, and therefore most participants (though not all) had vital signs within the normal range. A follow-up study is currently underway in ED zones of higher acuity and in other settings to assessthe systems’ accuracy in assessing vital signs outside of the normal range. Patients in this study were monitored for 150 s and required to sit still for the duration of the measurement. However, this was done primarily to obtain a large number of data points for comparison, not as a technical requirement of the system. The system can display measurements within a few seconds of the patient sitting in front of the sensor. On the other hand, a follow-up study is currently being planned to assess the system’s ability to perform prolonged patient monitoring, rather than point measurements. To these ends, the system’s software will be upgraded to either filter out patient movements or, at a minimum, flag measurements acquired during patient movements as less reliable, so that the clinician may take this information into account. Researchers in the current study were unblinded topatients’ vital signs, which were displayed in real time on the monitor screen. However, since the data were automatically recorded from both systems,later assessed for completeness by unrelated technicians, andthen analyzed by a third-party statistician, we believe this did not affect the results. The system in our study currently lacks the ability to assess some traditional vital signs, such as Blood Oxygen Saturation (SaO2), Blood Pressure (BP), and body temperature, despite importance as a central component of vital signs their assessment at the triage station; however, this feature of body tempearture is currently under development. While BP is not part of commonly used triage systems such as the Emergency Severity Index (ESI) (2), and O2 Sat is not part of other commonly used scores such as the Modified Early Warning Score (MEWS) (79) or the Quick Sequential Organ Failure Assessment (qSOFA) Score (80), research is underway to assess for the ability of the systems to provide these measurements, and/or to provide other respiratory and circulatory information that may prove equally effective in triage (such as assessment of Tidal Volume, Cardiac Output, Compensatory Reserve (81, 82), and fetal heart monitoring. Previous reports have described the use of remote monitoring based on radar technology in clinical settings. Hoang Thi Yen et al. (61) reported only 10 individuals, and Fabian Michler et al. (62) reported on a device located under the hospitalized patient’s bed. Our report is the first one to describe the mobile use of Radar technology in an ED setting with a large volume of patients in a multi-center study. The system is still under development, with planned additional features to better support real-world clinical workflows and triage needs. The present study was intentionally scoped as a technical and clinical validation study, focusing on measurement accuracy and feasibility. While the anticipated benefits—such as infection control, improved workflow efficiency—are highly relevant, formal assessment of: clinical outcomes, patient-reported outcomes and healtheconomic impact requires prospective implementation studies and regulatory-grade clinical trials, which are planned as future phases of development.

  Development of remote radar based vital sign acquisition for emergency department patient triage · 2026 · DOI

• For the RR, Bland-Altman analysis demonstrated a mean bias of −0.43 ± 1.17 breaths/min, indicating minimal systematic difference between the two measurement methods. The 95% LOA ranged from −2.74 to 1.87 breaths/min, with no evidence of proportional bias across the measurement range. (Table 2) intervals: Mean bias 95% CI: −0.51 to −0.35 Confidence breaths/min, Lower LOA 95% CI: −2.95 to −2.53 breaths/min, Upper LOA 95% CI: 1.66 to 2.08 breaths/min. A strong linear association was observed between Mindray and TAMAR RR measurements (Pearson’s r = 0.967, p < 0.001). Lin’s concordance correlation coefficient (CCC) demonstrated excellent agreement (CCC = 0.965). With n = 191, the 95% CI the CCC = 0.965 (95% CI: 0.952–0.975). This demonstrate “excellent agreement” range (>0.90). Agreement within predefined clinical thresholds: ≥99% of measurements within ±4 breaths/min and ∼96%–98% within ±10% of reference values. For HR, Paired Bland-Altman analysis demonstrated a small mean bias of 0.50 ± 2.09 bpm, indicating minimal systematic difference between the two measurement systems. The 95% LOA ranged from −3.58 to +4.59 bpm, reflecting a relatively narrow dispersion of differences. (Table 3) The 95% confidence intervals (CI) for the limits of agreement were: Lower LOA: −4.01 to −3.15 bpm and Upper LOA: +4.16 FIGURE 2 Excluded patient during data analysis. HR, heart rate; RR, respiratory rate.

  Development of remote radar based vital sign acquisition for emergency department patient triage · 2026 · DOI

• However, given the reliance on a single, relatively small dataset acquired at one clinical center, broader generalizability across diverse populations, devices, skin tones, comorbidities, and clinical environments remains to be established through larger, ideally multi-center external validation studies.

  An approach to machine learning-based non-invasive hemoglobin estimation using multi-wavelength PPG signal features · 2026 · DOI

• The absence of RMSE and R2 indices in prior classical algorithms (KNN, Weighted KNN, Bagged Trees) limits a complete assessment of generalization and error dispersion, precluding holistic comparison.

  Hospital Triage Optimization: Evaluation of Machine Learning Models for Blood Pressure Estimation to Enhance Emergency Response in Colombia · 2026 · DOI

• The research showed potential for detecting episodes in biomedical time series signals, but model sensitivity requires careful window parameter configuration for real-time signal classification applications.

  Classification of Episodes in the Breathing Signal Using Machine Learning · 2026 · DOI

• The binning strategy employed when computing local/adaptive reliability metrics can significantly affect metric values, as demonstrated by ENCE scaling with the number of bins.

  A systematic evaluation of uncertainty quantification techniques in deep learning: a case study in photoplethysmography signal analysis · 2026 · DOI

• ACE is more sensitive to the chosen number of ranges R, with ranges becoming very large if there are sparsely populated regions in model predictions' distribution, and more adaptive binning schemes lose interpretability across different models.

  A systematic evaluation of uncertainty quantification techniques in deep learning: a case study in photoplethysmography signal analysis · 2026 · DOI

• Choosing a larger number of bins in ECE reduces bias but increases variance of discrepancy in uncertainty magnitude and prediction error measurement per bin as more bins become sparsely populated.

  A systematic evaluation of uncertainty quantification techniques in deep learning: a case study in photoplethysmography signal analysis · 2026 · DOI

• ECE suffers from inability to detect variation in calibration in different subsets of the data and can be minimized by models constantly predicting the marginal distribution, failing to capture behavior of non-predicted class probabilities.

  A systematic evaluation of uncertainty quantification techniques in deep learning: a case study in photoplethysmography signal analysis · 2026 · DOI

• Tissue layer thicknesses (epidermis 0.2mm, dermis 1.5mm, subcutaneous 18.3mm) and scattering properties are fixed across individuals, despite most tissue properties varying from individual to individual.

  Inferring optical tissue properties from photoplethysmography using hybrid amortized inference · 2026 · DOI

• Smartwatches equipped with photoplethysmography (PPG) sensors are increasingly used for arrhythmia detection, yet their diagnostic accuracy in asymptomatic patients presenting to the emergency department (ED) remains underexplored.

  Evaluating Arrhythmia Alerts from Smartwatches in Asymptomatic Emergency Patients · 2025 · DOI

• Conclusion: The current evidence base consists of small, biased, and low-quality studies which are insufficient to advise clinicians on the true value of PPG devices for AH detection.

  Photoplethysmography technology use in smart devices for early diagnosis of arterial hypertension: a systematic review · 2023 · DOI

• Despite promises for future healthcare implementation, the lack of validation studies for clinical grade measurements presently still precludes the use of smartwatches for clinical decision making.

  Smartwatches in the digital health ecosystem for health monitoring in patients with chronic kidney disease and kidney failure: opportunities and challenges · 2026 · DOI

• The gap between this worst-case bound and typical operation under the BLE ARQ scheme remains to be characterized experimentally.

  Battery-aware approximate wireless telemetry framework for resilient wearable ECG monitoring under extreme power constraints · 2026 · DOI

• 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.

  Validation of non-contact sensor quantification of heart rate and respiratory rate dynamics using real-world pretraining and label-efficient fine-tuning on polysomnograms · 2026 · DOI

• However, different definitions of pulse transit time are used in the literature, and their statistical behavior when measured locally at the wrist using pressure sensors has not been systematically examined.

  A statistical analysis of pulse transit time captured using pressure sensors at the human radial artery of the wrist · 2026 · DOI

• Future Works Building upon the results of this study, future research will focus on the following trajectories: • • Lightweight Architectures: We aim to explore the transition from ViT-Small to a highly optimized ViT-Tiny or MobileViT backbone.

  Remote Photoplethysmography Using Triple-Head Spatio-Temporal Transformer with Reaction-Driven Gating and Illumination Separation · 2026 · DOI

• The study's applicability is limited to isolated regions with specific demographic and comorbidity characteristics; generalization to other populations requires validation.

  Hospital Triage Optimization: Evaluation of Machine Learning Models for Blood Pressure Estimation to Enhance Emergency Response in Colombia · 2026 · DOI

• CNN-based methods leveraging Korotkoff sounds have shown clinical applicability but often lack detailed quantitative metrics for direct benchmarking.

  Hospital Triage Optimization: Evaluation of Machine Learning Models for Blood Pressure Estimation to Enhance Emergency Response in Colombia · 2026 · DOI

• Entropy scales with the number of classes K, preventing direct comparison of model uncertainty or calibration across different datasets without normalization.

  A systematic evaluation of uncertainty quantification techniques in deep learning: a case study in photoplethysmography signal analysis · 2026 · DOI

• Infinitesimal softmax outputs introduce large biases in sparsely populated bins and the last bin containing extreme confidence values in calibration error calculations.

  A systematic evaluation of uncertainty quantification techniques in deep learning: a case study in photoplethysmography signal analysis · 2026 · DOI

• The refractive index is assumed constant at 1.4 across all tissue layers, and scattering properties are assumed identical across layers, which may not reflect biological heterogeneity.

  Inferring optical tissue properties from photoplethysmography using hybrid amortized inference · 2026 · DOI

• Scattering coefficient µs is modeled empirically rather than from first principles, as the relationship between biological factors (cell density and size) and scattering is less direct.

  Inferring optical tissue properties from photoplethysmography using hybrid amortized inference · 2026 · DOI

• The wavelength-independence assumption for light propagation may be violated in practice if the sensor material has wavelength-dependent reflection properties, though this assumption reduces computational burden.

  Inferring optical tissue properties from photoplethysmography using hybrid amortized inference · 2026 · DOI

Most-cited papers in Non-Invasive Vital Sign Monitoring

• Assessing Physical Activity Using Wearable Monitors · Medicine & Science in Sports & Exercise · 2011 · 270 citations
• PP-Net: A Deep Learning Framework for PPG-Based Blood Pressure and Heart Rate Estimation · IEEE Sensors Journal · 2020 · 223 citations
• Advancements in optical fiber-based wearable sensors for smart health monitoring · Biosensors and Bioelectronics · 2024 · 190 citations
• Contactless Methods For Measuring Respiratory Rate: A Review · IEEE Sensors Journal · 2020 · 159 citations
• Stressing the accuracy: Wrist‐worn wearable sensor validation over different conditions · Psychophysiology · 2019 · 141 citations
• In-depth correlation analysis between tear glucose and blood glucose using a wireless smart contact lens · Nature Communications · 2024 · 113 citations
• Traditional Chinese Medicine (TCM)‐Inspired Fully Printed Soft Pressure Sensor Array with Self‐Adaptive Pressurization for Highly Reliable Individualized Long‐Term Pulse Diagnostics · Advanced Materials · 2024 · 86 citations
• Wearable Devices for Respiratory Monitoring · Advanced Functional Materials · 2024 · 75 citations
• Monitoring long-term cardiac activity with contactless radio frequency signals · Nature Communications · 2024 · 66 citations
• DeepHeart · ACM Transactions on Sensor Networks · 2021 · 50 citations

Most recent work

• A Review of Next-Generation Sensor Technologies for Human–Machine Interaction · IEEE Sensors Journal · 2026
• Passive heart-rate monitoring during smartphone use in everyday life · Nature · 2026
• Wearable health-tracking technologies in pregnancy: capabilities, clinical evidence, and implementation challenges · Pharmacia · 2026
• AI-enabled wireless wearable breathing sensor for breathing pattern recognition · Scientific Reports · 2026
• Design and implementation of a hybrid machine learning framework for predicting heart rate status · Scientific Reports · 2026
• Wearable Textile Microwave Sensor for Wrist Pulse Wave and Heart Rate Monitoring · IEEE Sensors Journal · 2026
• Inferring optical tissue properties from photoplethysmography using hybrid amortized inference · Machine Learning: Health · 2026
• A systematic evaluation of uncertainty quantification techniques in deep learning: a case study in photoplethysmography signal analysis · Machine Learning: Health · 2026
• Classification of Episodes in the Breathing Signal Using Machine Learning · 2026
• Hospital Triage Optimization: Evaluation of Machine Learning Models for Blood Pressure Estimation to Enhance Emergency Response in Colombia · Journal of Hunan University Natural Sciences · 2026

Related topics in Engineering