Personal Health Monitoring Research Articles

Accuracy of Wrist-Worn Heart Rate Monitors: A Comprehensive Review of Smartwatches in Exercise Monitoring

Wrist-worn heart rate monitors have become integral in personal health monitoring, particularly during exercise and daily activities. This review evaluates the precision of various commercially available smartwatches in assessing heart rate, focusing on their accuracy across different physical activities, such as walking, running, cycling, and elliptical training. A systematic analysis of 29 peer-reviewed studies, encompassing over 900 participants, was conducted to compare devices like the Apple Watch, Fitbit, Garmin, and others. The results demonstrate significant variation in accuracy depending on the device, type of activity, and individual factors such as body mass index and skin tone. The Apple Watch consistently outperformed other devices, showing the lowest mean absolute percentage error (MAPE), particularly during moderate-intensity activities. In contrast, devices such as the Fitbit Blaze exhibited greater error rates, especially during high-intensity exercises or activities involving significant arm movement. Although wrist-worn heart rate monitors provide users with convenient real-time data, limitations such as motion artifacts, device placement, and external factors may affect their reliability.

Wearable Patch Biosensor through Electrothermal Film-Stimulated Sweat Secretion for Continuous Sweat Glucose Analysis at Rest.

Wearable patch biosensors for noninvasive and continuous diabetes management through sweat glucose analysis present a promising prospect. However, how to obtain sweat samples safely and effectively remains a huge challenge, especially in a resting state. In this work, we propose an innovative wearable patch biosensor through a heat-stimulated approach for sweat collection. A silver nanowire-loaded electrothermal film was designed as the heat source to stimulate sweat glands for sweat secretion. Subsequently, the secreted sweat sample was transported and enriched through microfluidic channels, which was continuously and sensitively analyzed by a Prussian blue and glucose oxidase comodified glucose electrochemical biosensor. Under optimal conditions, its sensitivity could achieve 14 μM sweat glucose within 15 min, which was 17 min shorter than that without heating. The specificity, reproducibility, and accuracy were also adequate. To achieve on-body perspiration monitoring of human subjects, the wearable patch biosensor was integrated with a portable electrochemical workstation, a temperature controller, and a power source. The glucose concentration was presented on a smartphone. Results showed that the glucose concentration in sweat detected by the wearable biosensor presented a highly consistent trend with the blood glucose measured by a blood glucose meter throughout the day with normal meals. Compared with other conventional sweat stimulation strategies, the simple device and safe principle made it more suitable for individuals who were sedentary or at rest. This work provides a new approach to realizing wearable patch biosensors for personalized health monitoring.

Streamlined Flow Synthesis of Plasmonic Nanoparticles and SERS Detection of Uremic Toxins with Trace-Level Liquid Volumes in a Microchamber.

Rapid detection of uremic toxins is crucial due to their severe health risks, including oxidative stress, inflammation, neurotoxicity, cardiovascular complications, and progression of chronic kidney disease. Surface-enhanced Raman spectroscopy (SERS) may provide sensitive, fast, and clinical-grade real-time monitoring of these toxins, enabling effective management with timely dialysis treatments. This study introduces a 3D-printed microchamber that integrates the fabrication of plasmonic metal nanoparticles for the in-flow detection of biological toxins and pharmaceutical drugs using SERS, making it ideal for on-site diagnostics in clinical settings. The microchamber supports quantitative and highly reproducible detection with liquid volumes under 100 μL, which is crucial for trace-level biomarker detection and minimizing cross-contamination. It employs a tunable solvent exchange method for the in situ synthesis of silver nanoparticles (AgNPs) on flexible PDMS or rigid Si wafer substrates, avoiding costly nanofabrication techniques. Ultralow detection limits were achieved for two model compounds and three pharmaceutical drugs: 10-11 M for rhodamine 6G, 10-7 M for adenine, and 10-6 M for the pharmaceutical drugs. A total of 13 biological toxins, including three neurotransmitters, one neuromodulator, five amino acids, two polyamines, and two urea cycle metabolites, were detected with quantitative limits ranging from 10-3 to 10-6 M, all below permissible levels and aligning with physiological conditions. SERS detection within microchambers facilitates rapid on-site analysis, proving ideal for personalized health monitoring, point-of-care diagnostics, and environmental pollution assessment.

Flexible Thermoelectric BiSbTe/Carbon Paper/BiSbTe Sandwiches for Bimode Temperature‐Pressure Sensors

AbstractBimode temperature‐pressure sensors hold significant promise in personal health monitoring, wearables and robotic signal detection. Traditional bimode sensors typically combine two independent sensors, leading to fabrication complexity. This study develops a bimode temperature‐pressure sensor by using a facile electrodeposition method to create sandwiched BiSbTe/Carbon Paper/BiSbTe thin films and stacking them to a vertical structure. It demonstrates high sensitivity for temperature sensing, capable of detecting temperature difference as low as 1 K, and a rapid response time of 0.92 s due to a vertical structure. Utilizing its thermoelectric mechanism, the sensor achieves self‐powered sensing for finger touch and respiration states. Furthermore, its island‐like contact surface ensures high sensitivity with an extremely fast response time of 0.17 s, by rapidly changing contact resistance under pressure, allowing it to detect various human behaviors, including body movements and micro‐expressions. Beyond its sensing capabilities, the film excels in flexibility, electromagnetic interference shielding, and stability, presenting significant potential for integration into self‐powered electronic skin systems for health monitoring, wearables, artificial intelligence, and other electronic skin applications.

CMOS Pulse Oximeter

Abstract: This paper presents the design and implementation of a non-invasive pulse oximeter based on Complementary MetalOxide-Semiconductor (CMOS) technology, aimed at enhancing the accuracy and accessibility of blood oxygen saturation measurements. Pulse oximetry is a critical tool in clinical and personal health monitoring, providing vital information about respiratory function. The integration of CMOS technology allows for miniaturization and energy efficiency, facilitating the development of compact, wearable devices capable of continuous monitoring. The proposed system utilizes advanced signal processing techniques, including lock-in detection and dual-wavelength illumination, to improve the reliability of photoplethysmographic (PPG) signal acquisition under various conditions. Experimental validation demonstrates the device's capability to deliver accurate SpO2 readings even in the presence of motion artifacts and ambient light interference. Furthermore, the calibration methods employed ensure that measurements are consistent and precise across different users. By leveraging the advantages of CMOS technology, this study not only highlights the potential for improved patient comfort and compliance but also opens new avenues for the application of pulse oximetry in diverse environments. The findings underscore the effectiveness of CMOS-based pulse oximeters in advancing non-invasive health monitoring solutions, ultimately contributing to enhanced patient care and outcomes.

From Lab to Life: Self-Powered Sweat Sensors and Their Future in Personal Health Monitoring.

The rapid development of wearable sweat sensors has demonstrated their potential for continuous, non-invasive disease diagnosis and health monitoring. Emerging energy harvesters capable of converting various environmental energy sources-biomechanical, thermal, biochemical, and solar-into electrical energy are revolutionizing power solutions for wearable devices. Based on self-powered technology, the integration of the energy harvesters with wearable sweat sensors can drive the device for biosensing, signal processing, and data transmission. As a result, self-powered sweat sensors are able to operate continuously without external power or charging, greatly facilitating the development of wearable electronics and personalized healthcare. This review focuses on the recent advances in self-powered sweat sensors for personalized healthcare, covering sweat sensors, energy harvesters, energy management, and applications. The review begins with the foundations of wearable sweat sensors, providing an overview of their detection methods, materials, and wearable devices. Then, the working mechanism, structure, and a characteristic of different types of energy harvesters are discussed. The features and challenges of different energy harvesters in energy supply and energy management of sweat sensors are emphasized. The review concludes with a look at the future prospects of self-powered sweat sensors, outlining the trajectory of the field and its potential to flourish.

The Rise of Biochemical Sensors: Technology for Real-Time Health Monitoring (Applications and Future Scope)

Biochemical sensors integrated into wearable health technology represent one of the most transformative innovations of the 21st century. This review delves into the evolution, principles, and real-time applications of biochemical sensors and their role in personalized health monitoring. Wearable biochemical sensors, capable of continuously measuring various biomarkers like glucose, lactate, cortisol, and electrolytes, are revolutionizing healthcare by enabling proactive management of chronic diseases, including diabetes, cardiovascular disorders, and mental health issues. Advances in biosensing technologies, coupled with the use of AI and machine learning algorithms, have enhanced the sensitivity and accuracy of these devices, ensuring that critical health data is available in real-time. From glucose monitoring devices like Abbott's FreeStyle Libre to the latest nanomaterial-based sensors, these innovations are reshaping healthcare delivery by shifting the focus from hospital-centered treatments to patient-centric, continuous monitoring systems. This review provides an in-depth analysis of the technological advancements, challenges, and future directions in biochemical sensors, focusing on key technologies such as electrochemical, optical, and enzymatic sensors. It also highlights the critical role of AI in interpreting the complex data generated by these sensors, paving the way for more efficient diagnostics and predictive healthcare models. Furthermore, the paper explores how these sensors have been applied in infectious disease detection, particularly during the COVID-19 pandemic, and discusses their potential to enhance global health surveillance systems. In conclusion, wearable biochemical sensors represent a significant leap forward in the pursuit of personalized medicine, offering real-time diagnostics and timely interventions for disease management. The future of healthcare is closely tied to the ongoing innovations in sensor technology, with the promise of even more advanced and multifunctional devices on the horizon.

A Robust Core-Shell Nanofabric with Personal Protection, Health Monitoring and Physical Comfort for Smart Sportswear.

Smart textiles with a high level of personal protection, health monitoring, physical comfort, and wearing durability are highly demanded in clothing for harsh application scenarios, such as modern sportswear. However, seamlessly integrating such a smart clothing system has been a long-sought but challenging goal. Herein, based on coaxial electrospinning techniques, a smart non-woven textile (Smart-NT) integrated with high impact resistance is developed, multisensory functions, and radiative cooling effects. This Smart-NT is comprised of core-shell nanofibers with an ionic conductive polymer sheath and an impact-stiffening polymer core. The soft smart textile, with a thickness of only 800µm, can attenuate over 60% of impact force, sense pressure stimulus with sensitivity up to 201.5 kPa-1, achieve temperature sensing resolution of 0.1°C, and reduce skin temperature by ≈17°C under a solar intensity of 1kWm-2. In addition, the stretchable Smart-NT is highly durable and robust, retaining its multifunction features over 10000 bending and multiple washing cycles. Finally, application scenarios are demonstrated for real-time health monitoring, body protection, and physical comfort of smart sportswear based on Smart-NT for outdoor sports. The strategy opens a new avenue for seamless integration of smart clothing systems.

Portable Electrochemical System and Platform with Point-of-Care Determination of Urine Albumin-to-Creatinine Ratio to Evaluate Chronic Kidney Disease and Cardiorenal Syndrome.

The urine albumin (Alb)-to-creatinine (Crn) ratio (UACR) is a sensitive and early indicator of chronic kidney disease (CKD) and cardiorenal syndrome. This study developed a portable and wireless electrochemical-sensing platform for the sensitive and accurate determination of UACR. The developed platform consists of a carbon nanotube (CNT)-2,2'-azino-bis(3-ethylbenzothiazoline-6-sulphonic acid)(ABTS)-based modified UACR sensor, a miniaturised potentiostat, a cup holder embedded with a magnetic stirrer and a smartphone app. The UACR sensing electrode is composed of two screen-printed carbon working electrodes, one screen-printed carbon counter electrode and a screen-printed AgCl reference electrode. The miniaturised potentiostat, which is controlled by the developed app, performs cyclic voltammetry and amperometry to detect Alb and Crn, respectively. Clinical trials of the proposed system by using spot urine samples from 30 diabetic patients indicate that it can accurately classify all three CKD risk statuses within 30 min. The high accuracy of our proposed sensing system exhibits satisfactory agreement with the commercial biochemical analyser TBA-25FR (Y = 0.999X, R2 = 0.995). The proposed UACR sensing system offers a convenient, reliable and affordable solution for personal mobile health monitoring and point-of-care urinalysis.

A Multiplexed Quantitative Proteomics Approach to the Human Plasma Protein Signature.

Despite the plasma proteome being able to provide a unique insight into the health and disease status of individuals, holding singular promise as a source of protein biomarkers that could be pivotal in the context of personalized medicine, only around 100 proteins covering a few human conditions have been approved as biomarkers by the US Food and Drug Administration (FDA) so far. Mass spectrometry (MS) currently has enormous potential for high-throughput analysis in clinical research; however, plasma proteomics remains challenging mainly due to the wide dynamic range of plasma protein abundances and the time-consuming procedures required. We applied a new MS-based multiplexed proteomics workflow to quantitate proteins, encompassing 67 FDA-approved biomarkers, in >1300 human plasma samples from a clinical cohort. Our results indicate that this workflow is suitable for large-scale clinical studies, showing good accuracy and reproducibility (coefficient of variation (CV) < 20 for 90% of the proteins). Furthermore, we identified plasma signature proteins (stable in time on an individual basis), stable proteins (exhibiting low biological variability and high temporal stability), and highly variable proteins (with low temporal stability) that can be used for personalized health monitoring and medicine.

1D Textile Yarn Battery with MoS2@Si Anode and NCM Cathode

AbstractWearable electronics are surging for various applications ranging from critical functions like personal health monitoring to communication and entertainment. To power these electronic devices, advanced high‐performing textile‐based batteries are reckoned. In this work, a 1D textile yarn battery is designed using silicon (Si) nanoparticles wrapped in molybdenum disulfide (MoS2) as an anode and layered Ni‐rich material Li[Ni0.8Co0.1Mn0.1]O2 (NCM) as a cathode. The anode materials design is selected to ensure the use of Si due to its high specific capacity but suppressing its known issue of volume expansion by layered MoS2 nanosheets and, at the same time, MoS2 providing channels for lithium‐ion (Li‐ion) transport during electrochemical cycles. The NCM cathode, on the other hand, is adopted as it has higher energy density and improved cycle life. The full yarn battery (FYB) delivered an excellent electrochemical performance (areal capacity of 3.13 mAh cm−2, power density of 421 mW cm−3, and energy density of 78.9 mWh cm−3) with a capacity retention of 86% at 0.1 C and coulombic efficiency of 91.3%. This work pointed out a new way to design and fabricate textile‐based batteries with high‐performance materials using simple, cost‐effective, and scalable approaches targeting to be used as energy sources for future wearable electronics.

Silk Fabric Functionalized by Nanosilver Enabling the Wearable Sensing for Biomechanics and Biomolecules.

Integrating biomechanical and biomolecular sensing mechanisms into wearable devices is a formidable challenge and key to acquiring personalized health management. To address this, we have developed an innovative multifunctional sensor enabled by plasma functionalized silk fabric, which possesses multimodal sensing capabilities for biomechanics and biomolecules. A seed-mediated in situ growth method was employed to coat silver nanoparticles (AgNPs) onto silk fibers, resulting in silk fibers functionalized with AgNPs (SFs@Ag) that exhibit both piezoresistive response and localized surface plasmon resonance effects. The SFs@Ag membrane enables accurate detection of mechanical pressure and specific biomolecules during wearable sensing, offering a versatile solution for comprehensive personalized health monitoring. Additionally, a machine learning algorithm has been established to specifically recognize muscle strain signals, potentially extending to the diagnosis and monitoring of neuromuscular disorders such as amyotrophic lateral sclerosis (ALS). Unlike electromyography, which detects large muscles in clinical medicine, sensing data for tiny muscles enhance our understanding of muscle coordination using the SFs@Ag sensor. This detection model provides feasibility for the early detection and prevention of neuromuscular diseases. Beyond muscle stress and strain sensing, biomolecular detection is a critical addition to achieving effective health management. In this study, we developed highly sensitive surface-enhanced Raman scattering (SERS) detection for wearable health monitoring. Finite-difference time-domain numerical simulations ware utilized to analyze the efficacy of the SFs@Ag sensor for wearable SERS sensing of biomolecules. Based on the specific SERS spectra, automatic extraction of signals of sweat molecules was also achieved. In summary, the SFs@Ag sensor bridges the gap between biomechanical and biomolecular sensing in wearable applications, providing significant value for personalized health management.

Plasmonic Nanostructures Grown from Reacting Droplet-In-Microwell Array on Flexible Films for Quantitative Surface-Enhanced Raman Spectroscopy in Plant Wearable In Situ Detection.

Plant wearable detection has garnered significant interest in advancing agricultural intelligence and promoting sustainable food production amidst the challenges of climate change. Accurately monitoring plant health and agrochemical residue levels necessitates qualities such as precision, affordability, simplicity, and noninvasiveness. Here, a novel attachable plasmonic film is introduced and designed for on-site detection of agrochemical residues utilizing surface-enhanced Raman spectroscopy (SERS). By functionalizing a thin polydimethylsiloxane film with silver nanoparticles via controlled droplet reactions in micro-well arrays, a plasmonic film is achieved that not only maintains optical transparency for precise analyte localization but also conforms closely to the plant surface, facilitating highly sensitive SERS measurements. The reliability of this film enables accurate identification and quantification of individual compounds and their mixtures, boasting an ultra-low detection limit ranging from 10-16 to 10-13 m, with mini mal relative standard deviation. To showcase its potential, on-field detection of pesticide residues on fruit surfaces is conducted using a handheld Raman spectrometer. This advancement in fabricating plasmonic nanostructures on flexible films holds promise for expanding SERS applications beyond plant monitoring, including personalized health monitoring, point-of-care diagnosis, wearable devices for human-machine interface, and on-site monitoring of environmentalpollutants.

Multifunctional hydrogel sensor with Tough, self-healing capabilities and highly sensitive for motion monitoring and wound healing

Flexible epidermal sensors based on conductive hydrogels hold great potential for wearable devices and personal medical monitoring. Integrating real-time monitoring of injury and motion activities can essentially enhance treatment outcomes by providing detailed data to guide clinical practice. However, the application of conductive hydrogel epidermal sensors in diverse remains challenging. In this study, a multifunctional PVA/AM/AAPBA/Graphene Oxide (referred to as MHPBA-GO hydrogel) was designed as an epidermal sensor for combined functions of accelerating wound healing and motion sensing. This sensor possesses excellent biocompatibility and multifunctional therapeutic properties, including flexibility, self-healing features, and tunable photothermal conversion capability. It can sensitively monitor human motion and small electrophysiological signals to diagnose related activities and diseases. The corresponding gauge factor (GF) values under 0–200% strain are 1.8 (0–90%), 2.1 (90–160%), and 8.5 (160–200%), respectively. Furthermore, using a diabetic rat mode demonstrated that the composite hydrogel sensor can be used as an efficient wound dressing to promote wound healing. This study provides a valuable reference and guidance for the development of flexible epidermal sensors with multifunctionality for personal health monitoring.

Flexible sensor based on conformable, sticky, transparent elastomers for electronic skin

As 5G technology advances and enhances the interconnection of science and technology, the demand for advanced sensors has increased. Conventional rigid sensors are inadequate to meet these new requirements, thus sparking significant interest in flexible sensors. However, taking into account the biocompatibility, skin compliance, sensing performance, mechanical properties and optical properties of flexible sensors is still the direction of efforts. Therefore, in this study, the flexible sensor with a sandwich structure was prepared using AgNWs as a conductive filler and P32-PDMS elastomer as a flexible substrate. The PDMS elastomer is modified by blending PEG-PPG-PEG and PDMS, giving the P32-PDMS elastomer low Young’s modulus (32.35 kPa), high elongation at break (over 450 %) and good adhesion (18.94 N/m), and the P32-PDMS elastomer still retains a certain light transmittance (82.22 %) and biocompatibility (RGR=89.524). These properties simultaneously endow the flexible sensor with good sensitivity (GF=17.27), wide detection range (0–100 % strain), and excellent cycling stability (5000 cycles). When the flexible sensor is used as e-skin to monitor human body movements, stable signals are generated regardless of the low-strain movements of the human body (pulse and sound) or the high-strain movements of the human body (pressing and joint bending). Therefore, the prepared flexible sensor has great potential as e-skin in many fields such as personal health monitoring, human motion detection, human–computer interaction, flexible display, etc.

A Dye-Sensitized Sensor for Oxygen Detection under Visible Light.

Sensors that can accurately assess oxygen (O2) concentrations in real time are crucial for a wide range of applications spanning personal health monitoring, environmental protection, and industrial process development. Here a high-performance chemiresistive sensor that allows for the rapid detection of O2 at room temperature under visible light illumination is described. Inspired by the operating principles of dye-sensitized solar cells, the chemiresistor is based on a single-walled carbon nanotube-titania hybrid (SWCNT-TiO2) bearing a molecular Re-based photosensitizer [(Pbpy)(CO)3ReBr] (Pbpy = 4,4'-[P(O)(OH)2]2-2,2'-bipyridine). The resulting SWCNT-TiO2-Re composite undergoes photoinduced charge transfer that is sensitive to ppb levels of O2, thereby yielding a rapid and reversible chemiresistive response. Owing to its unique mode of operation and robust components, the sensor shows a high degree of selectivity for O2 over a range of interferants, humidity tolerance, and multimonth benchtop stability. The approach presented herein demonstrates the translatability of concepts in light harvesting to the development of robust, rapid, and low-power sensing technologies.

Wearable, Battery-Free, Self-Powered Electrochemical Sensors for Personalized Health Monitoring

Recent advances in materials, mechanics, and device architectures form the foundations for rapidly emerging classes of sensors with mechanical characteristics that allow for conformal interfaces with the soft, curvilinear surfaces of the human body. While the field of tissue-integrated biophysical sensors exhibits commendable progress in capturing clinical quality thermal, kinematic, and electrophysiological information, the advancement of complementary biochemical sensors severely lags due to unique challenges associated with seamless integration of delicate biochemical receptors, transducing components, and suitable packaging materials. In this talk, I will discuss non-traditional approaches to address some of these grand challenges. Specifically, I will discuss three technologies that my lab is developing -1) wearable, biofuel cell-inspired sensors for sweat monitoring 2) wearable, soft microfluidics with in-built valving and pumping capabilities for precise, on-demand sensing, and 3) AI-powered, wearable wound monitors for early prediction of chronic wounds.

Innovative Wearable Sweat Sensor Array for Real-Time Volatile Organic Compound Detection in Noninvasive Diabetes Monitoring.

Wearable sweat sensors are reshaping healthcare monitoring, providing real-time data on hydration and electrolyte levels with user-friendly, noninvasive devices. This paper introduces a highly portable two-channel microfluidic device for simultaneous sweat sampling and the real-time detection of volatile organic compound (VOC) biomarkers. This innovative wearable microfluidic system is tailored for monitoring diabetes through the continuous and noninvasive tracking of acetone and ammonia VOCs, and it seamlessly integrates with smartphones for easy data management. The core of this system lies in the utilization of carbon polymer dots (CPDs) and carbon dots (CDs) derived from monomers such as catechol, resorcinol, o-phenylenediamine, urea, and citric acid. These dots are seamlessly integrated into hydrogels made from gelatin and poly(vinyl alcohol), resulting in an advanced solid-state fluorometric sensor coating on a cellulose paper substrate. These sensors exhibit exceptional performance, offering linear detection ranges of 0.05-0.15 ppm for acetone and 0.25-0.37 ppm for ammonia, with notably low detection limits of 0.01 and 0.08 ppm, respectively. Rigorous optimization of operational parameters, encompassing the temperature, sample volume, and assay time, has been undertaken to maximize device performance. Furthermore, these sensors demonstrate impressive selectivity, effectively discerning between biologically similar substances and other potential compounds commonly present in sweat. As this field matures, the prospect of cost-effective, continuous, personalized health monitoring through wearable VOC sensors holds significant potential for overcoming barriers to comprehensive medical care in underserved regions. This highlights the transformative capacity of wearable VOC sweat sensing in ensuring equitable access to advanced healthcare diagnostics, particularly in remote or geographically isolated areas.

Nanoarchitectonics of Fe single-atom catalyst modified with flexible wearable patch for sweat biomarker analysis

In personalized medical diagnostics and health monitoring, the analysis of biomarkers in sweat, such as uric acid (UA) and dopamine (DA), offers valuable insights into metabolic states. This study introduces a flexible wearable patch based on Fe single-atom catalysts (FeSAs) to simultaneously detect UA and DA in sweat for comprehensive health assessment. Our wearable patch demonstrated linear response ranges of 1–800 μM for UA and 1–500 μM for DA, with detection limits of 2.68 μM and 3.76 μM, and sensitivities of 0.015 μA μM−1 cm−2 and 0.0176 μA μM−1 cm−2, respectively. Additionally, we integrated agarose hydrogel into the wearable patches, optimizing sweat collection efficiency at agarose concentrations ranging from 2 % to 3.8 %. This innovation enables effective biomarker enrichment directly from the skin surface. In conclusion, our study presents an innovative FeSAs-based electrocatalyst capable of simultaneous detection of low-concentration UA and DA biomarkers.

On-Device Semi-Supervised Activity Detection: A New Privacy-Aware Personalized Health Monitoring Approach.

This paper presents an on-device semi-supervised human activity detection system that can learn and predict human activity patterns in real time. The clinical objective is to monitor and detect the unhealthy sedentary lifestyle of a user. The proposed semi-supervised learning (SSL) framework uses sparsely labelled user activity events acquired from Inertial Measurement Unit sensors installed as wearable devices. The proposed cluster-based learning model in this approach is trained with data from the same target user, thus preserving data privacy while providing personalized activity detection services. Two different cluster labelling strategies, namely, population-based and distance-based strategies, are employed to achieve the desired classification performance. The proposed system is shown to be highly accurate and computationally efficient for different algorithmic parameters, which is relevant in the context of limited computing resources on typical wearable devices. Extensive experimentation and simulation study have been conducted on multi-user human activity data from the public domain in order to analyze the trade-off between classification accuracy and computation complexity of the proposed learning paradigm with different algorithmic hyper-parameters. With 4.17 h of training time for 8000 activity episodes, the proposed SSL approach consumes at most 20 KB of CPU memory space, while providing a maximum accuracy of 90% and 100% classification rates.