Multimodal flexible sensor system toward telediagnosis

Abstract

Advancements in living standards, medical technologies, and nutrition have contributed to the global increase in life expectancy. However, the widening gap between life expectancy and healthy life expectancy indicates that a growing population requires ongoing medical care and hospitalization. The development of a home-use telediagnostic system is a promising solution for improving the quality of life for patients and healthcare providers to extend a healthy lifespan and reduce the burden on clinicians. Such a system also holds the potential for rapid health assessments during emergency situations, where timely triage is critical. This study reviews recent progress in noninvasive, multimodal sensor patches capable of continuously monitoring vital signs and biomarkers via the skin. Furthermore, we explore the integration of machine learning for real-time, on-device data analysis as an edge system, enabling autonomous health feedback without reliance on the Internet. Although several technical challenges remain before practical implementation, these innovations may pave the way for a paradigm shift in conventional medical care.

Edge-type smart sensor patch toward telediagnosis.

1. Introduction

The global average life expectancy has been steadily increasing, largely because of improvements in living standards, medical technologies, and nutrition. However, the duration of life spent in good health, which is referred to as “healthy life expectancy”, remains significantly shorter than the overall life expectancy.¹⁾⁾ This disparity indicates that many older individuals live a substantial portion of their extended lives with various chronic illnesses. In 2019, the global gap between life expectancy and healthy life expectancy reached approximately 9.6 years, representing a 13% increase since 2000.¹⁾⁾ This growing disparity can be attributed to multiple interrelated factors, such as the rapid global aging population trend, a rising incidence of chronic illnesses among older adults, deterioration in physical and cognitive functions, and insufficient social care infrastructure. Consequently, the demand for medical care and hospitalization has continued to rise. As a result, although people are living longer, their quality of life and overall well-being during old age may not be improving in tandem with lifespan extension. This growing disparity imposes not only a decline in individual well-being but also a significant socioeconomic burden, including escalating healthcare costs and increased workload for medical professionals, such as physicians and nurses.

It is important to reduce the growing socioeconomic burden associated with aging and chronic diseases to realize a sustainable and health-conscious society. Wearable technologies that enable autonomous health monitoring in daily life without requiring the direct involvement of medical professionals are a promising approach. Watch-type or similar wearable devices can automatically record physiological parameters, such as heart rate and physical activity, by simply wearing the devices. However, these limited parameters are typically insufficient for accurate health diagnosis, and clinical diagnosis is still required. To enable remote health monitoring and pre-screening for assessing the need for medical intervention, next-generation wearable devices, such as conformable multimodal flexible and/or stretchable sensor patches, have emerged as promising platforms for telediagnosis. To fully substitute gold standard clinical diagnostics, these sensors must be capable of simultaneously detecting the five vital physiological signals of body temperature, heart rate, respiratory rate, blood pressure, and blood oxygen saturation (SpO₂)^2)–6)) as well as key biochemical markers, such as lactate, electrolytes, and glucose,^7)–15)) highlighting the necessity of multimodal sensor systems that can provide real-time and comprehensive health assessments.

Several approaches for integrating these sensing functions into a single patch have been proposed.^{2),10),16)–19))} These sensor patches are usually a noninvasive way to monitor these vital signals and biomarkers from the skin surface. Mechanically flexible and/or stretchable thin sensor patches can conformally attach to the skin, resulting in less noise of the sensing outputs caused by environmental change and motion of body.²⁰⁾⁾ However, for the signal processing and wireless system for medical trials and practical applications, the sensor patch still requires conventional rigid circuit chips because of the difficulty in integrating all flexible circuits on a patch at an economically low cost and with stable operation under mechanical stress caused by body motion. Because the sensors are flexible and the circuit is rigid, it is called a “hybrid system”.¹⁰⁾⁾ Another challenge lies in the timely analysis of abnormal physiological data and the immediate feedback to users regarding potential health issues. Continuous monitoring of vital signs and biomarkers generates large daily volumes of data. While cloud-based analysis using high-performance and multiple workstations is a conventional solution because of quick analyses of large amounts of data and complicated machine learning processing, this approach requires constant wireless communications to connect cloud continuously, leading to significant power consumption and frequent battery recharging that compromise user convenience. Reliance on cloud computing sometimes introduces latency owing to network conditions, hindering real-time responsiveness. To address power efficiency and processing delay, the implementation of edge computing powered by machine learning emerges as a key strategy for enabling real-time telediagnosis using wearable sensor patches.

This review summarizes recent progress in the area of multimodal flexible sensor patch with wireless sensor system to continuously monitor physiological vital signs and biomarkers for remote healthcare and telediagnosis. Realizing an edge-computed wearable telediagnostic system requires core technologies of flexible circuits and sensors and advanced technologies about wireless system, machine learning, and edge computing (Fig. 1). The scope covers not only wireless sensor patches but also data analyses in an edge system without using the Internet or cloud computer to enhance the speed of the analysis to provide a prompt feedback to the user.

Figure 1.

Devices and system components for wearable telediagnosis. Flexible circuits and multimodal sensors capable of physical and chemical sensing form core technologies, enabling the development of wireless systems and the acquisition of diverse datasets from demonstrations, which serve as the basis for the development of machine learning algorithms. The integration of all components and data realizes an edge computing–based telediagnosis system. Reproduced with permission from Refs. 2 (CC-BY), 29 and 62. Copyright 2020 American Chemical Society and 2018 Nature Publishing Group.

2. Vital sign monitoring

The five gold standard vital signs—electrocardiogram (ECG), respiration rate, body temperature, blood pressure, and oxygen saturation (SpO₂)—serve as essential indicators for initial health screening and diagnosis. These physiological parameters can be measured noninvasively by placing sensors on the skin. Numerous healthcare and medical devices for monitoring individual vital signs are commercially available at relatively low costs for personal and home use. However, most of these devices are designed to measure only one specific signal and are often bulky, limiting their ability to continuously track dynamic changes in vital signs. Such limitations hinder the early detection of abnormalities and the remote diagnosis of emerging conditions. As a result, real-time, long-term monitoring of multiple vital signs and understanding their interrelationships during daily physiological fluctuations remains a significant challenge. Overcoming this limitation is crucial for developing next-generation healthcare systems capable of disease prediction and remote timely intervention. To address this issue, bandage-type flexible sensor patches have gained increasing attention. Thin and flexible film of the sensors conformably covers the skin roughness to reduce noise, and multimodality simultaneously detects multiple physiological signals from the skin. This section introduces the individual flexible sensors that target each vital sign, as well as the integrated system capable of simultaneous, continuous monitoring.

2.1. ECG sensor.

ECG sensors detect the electrical potential differences generated by cardiac muscle contraction. A minimum of two electrodes is typically sufficient to capture this potential; however, a third reference electrode is often employed to stabilize the baseline signal by compensating for fluctuations in body potential.

One of the major challenges in wearable applications is the presence of motion artifacts, i.e., artificial noise caused by body motion, which can compromise the accuracy and precision of ECG recordings. To address this issue, ultrathin, conformal electrodes have been developed to ensure close contact with the skin.²¹⁾⁾ These electrodes effectively accommodate the skin’s microtopography, significantly minimizing noise even during physical activity. User comfort is another consideration for wearable ECG sensors, particularly as ECG measurements often require a relatively large sensing area to capture the spatial potential differences associated with cardiac activity. It is essential to ensure breathability and mechanical compatibility of the sensor sheet with skin movement. To improve wearability, two main strategies have been pursued: (1) the use of porous and stretchable films²²⁾⁾ and (2) the introduction of perforations or mechanical structures, such as kirigami, to enhance stretchability and air permeability.^3),23)) The first approach resembles traditional bandages and is straightforward to implement. However, the integration of additional sensors can be challenging owing to mechanical deformation during stretching and signal interference from sweat-induced ionic conduction. The sensor using kirigami architectures offers greater control: breathable, stretchable areas can be selectively introduced while keeping the sensing regions mechanically stable. It is possible to achieve high wearability without compromising sensor performance by carefully designing the location and pattern of the kirigami cuts (Fig. 2a–c).³⁾⁾ This section further explores the kirigami-based ECG sensor.

Figure 2.

Electrocardiogram (ECG) sensors with kirigami structure. (a) Schematic and photo of the ECG sensor. (b) Schematic of each layer of the ECG sensor. (c) System architecture of the sensor, wireless system, and application. (d) Photo of a demonstration. (e) Display of real-time ECG monitoring and heart rate calculated from ECG signals. (f) Compiled results of ECG and activity monitoring during daily life activities. Reproduced with permission from Ref. 3. Copyright 2022 AIP Publishing.

When applying a kirigami structure to a flexible polyethylene terephthalate (PET) film, careful design of the interconnecting electrodes between the ECG sensor and the signal processing circuit, including a wireless module, is essential because the film is physically cut to form the kirigami pattern. To simplify the fabrication process and avoid the need for precise alignment between the kirigami pattern and the electrode formation, the width of the interconnects was designed to be larger than the width of the kirigami slits³⁾⁾; this allows us to first form electrodes and then cut the film for the kirigami pattern without the precise alignment process.

In the chest region, the skin typically undergoes deformation of up to ∼30% with multidirectional stretching. To accommodate this, the kirigami pattern was designed to achieve stretchability up to 35% in multiple directions. Kirigami slits localize strain and enable out-of-plane buckling, thereby decoupling the global stretch from the electrode–skin interface. This preserves the contact impedance between the ECG sensor and the skin during motion and improves the ECG fidelity under large deformations. In addition to mechanical adaptability, the kirigami cuts introduce physical openings in the film, significantly enhancing its breathability. The measured water vapor transmission rate was ∼17.5 kg m⁻² d⁻², which is higher than that of the standard polyurethane film (∼12.6 kg m⁻² d⁻²) commonly used in commercial bandages. To evaluate breathability and sweat management, real-time impedance measurements were conducted using a pair of electrodes integrated across the kirigami structure under 60% relative humidity at 28 ℃. The impedance of the kirigami-integrated patch remained stable over time, indicating effective moisture dissipation. In contrast, a patch without kirigami exhibited a significant drop in impedance because of sweat bridging between the electrodes. These findings indicate that the kirigami design, featuring patterned perforation in the PET film, significantly enhances breathability and mitigates moisture accumulation with mechanical stretchability, making them a promising candidate for ECG monitoring and other wearable sensor applications requiring high conformability and long-term stability.

As a proof-of-concept demonstration, a kirigami-structured ECG sensor patch was integrated with a Bluetooth Low Energy (BLE) wireless module and a signal processing circuit (Fig. 2d–e).³⁾⁾ This system enabled the real-time transmission of ECG signals and physical activity data acquired by a built-in three-axis accelerometer in the hybrid rigid circuit board to a smartphone. Continuous ECG monitoring was performed during various daily activities, including desk work, lunch, and walking. Importantly, heart rate variability, extracted from the R-R interval derived from ECG peak-to-peak measurements, exhibited a clear trend in response to changes in physical activity levels. These variations were corroborated by the simultaneously recorded accelerometer data (Fig. 2f), confirming the system’s dynamic responsiveness. This demonstration highlights the feasibility of a wearable ECG sensor patch with high conformability, comfort, and the ability to continuously track cardiovascular dynamics during daytime. By further optimizing sensor design and integration, such system holds strong potential for seamless, long-term health monitoring.

2.2. Respiration sensor.

Respiration rate is another critical physiological parameter for medical diagnosis. Although it can be easily measured at home without specialized equipment simply by counting breaths through the mouth or nose, routine monitoring and recording of this parameter in daily life remains uncommon. This is primarily because, despite the simplicity of the method, manual timing and self-monitoring are still required, which are considered time consuming and inconvenient. However, for remote health monitoring and telediagnosis applications, real-time and simultaneous measurement of respiration and other vital signs is essential.

To address this, various respiration sensor patches have been proposed.^19),24),25)) One approach involves detecting abdominal motion,^2),19)) whereas another focuses on changes in humidity near the nostrils.²⁴⁾⁾ Both types of signals correlate with respiration and can serve as breathing activity indicators. A straightforward technique for abdominal motion sensing is to attach a stretchable strain sensor to the abdominal region to detect the inflation and deflation associated with breathing. Then, the respiration rate can be determined by measuring the time interval between the signal peaks.

An example of such a stretchable strain sensor is based on laser-induced graphene (LIG).^19),26),27)) LIG is fabricated by irradiating a polyimide (PI) film with a laser, which induces the photothermal conversion of the PI into porous, multilayered graphene with defects, as confirmed by Raman spectroscopy. Aromatic polyimide absorbs CO₂ laser irradiation and undergoes photothermal pyrolysis to form a porous, conductive graphene-like network. The imide groups also promote nitrogen doping, improving conductivity. The high thermal stability of PI prevents melting and preserves pattern fidelity, resulting in continuous, low-resistance tracks. In contrast, PET/PEN tends to soften or incompletely carbonize under the same influence, producing discontinuous or less conductive residues. A dimethylpolysiloxane (PDMS) solution was then cast onto the LIG/PI film and cured at 90 ℃. After curing, the PDMS layer is peeled off, transferring the LIG onto the stretchable PDMS film due to the porous and solution-based process. Finally, silver (Ag) electrodes were printed onto the LIG/PDMS surface to enable electrical interconnection. The three-dimensional porous LIG structures embedded in the PDMS film lead to variable contact areas between LIG domains under applied tensile strain, thereby inducing electrical resistance changes. Thus, this LIG/PDMS composite functions as a resistive stretchable strain sensor. Its sensitivity, referred to as the gauge factor, varies depending on the LIG formation conditions, with typical values of around 20 for standard fabrication methods¹⁹⁾⁾ and up to ∼1200 using specialized photothermal catalyst techniques.²⁸⁾⁾ These LIG-based strain sensors enable real-time continuous monitoring of respiration rate in a wearable and noninvasive manner.^2),19))

Monitoring airflow or humidity near the nostrils is a direct and effective method for assessing respiration rate. However, placing sensors in the nasal region often poses esthetic concerns, making it less suitable for daily use, particularly in emergency or inpatient scenarios, despite its clinical use. A promising alternative is the integration of humidity sensors into face masks.^24),25)) This design eliminates the need for direct skin contact and allows for discreet monitoring because the sensor is embedded within the mask and remains invisible to others. Importantly, exhaled breath contains chemicals as NH₄⁺ and NO₂⁻, which can serve as indicators of medical diagnosis. By converting breath into exhaled breath condensate, these chemical biomarkers can potentially be monitored continuously.²⁵⁾⁾ Therefore, the face mask–based approach is a promising platform for multimodal sensing in telemedicine applications. A representative demonstration of this concept for respiration rate monitoring was conducted using a flexible humidity sensor embedded in a face mask equipped with a wireless module.²⁴⁾⁾ The sensor was fabricated from a film comprising stacked ZnIn₂S₄ (ZIS) nanosheets. Its sensing mechanism is based on proton hopping on the ZIS surface, which causes the electrical resistance to decrease as the humidity level increases.²⁹⁾⁾ The sensor exhibited a humidity sensitivity of approximately 1.1% per %RH change and remained stable over 150 h of continuous operation.²⁴⁾⁾ To validate its utility, the sensor-integrated face mask was tested on three volunteers to monitor respiration during sleep. Among the participants, one had been diagnosed with sleep apnea syndrome (SAS). The humidity-based respiration monitoring clearly identified frequent apneas in the patient with SAS, whereas no such events were observed in the healthy subjects. Additionally, for the healthy volunteers, variations in respiration rate corresponding to different sleep stages were recorded, suggesting the potential for sleep quality assessment. However, the requirement for continuous wearing of the face mask to enable ongoing monitoring may restrict its practical use to specific scenarios.

2.3. Skin temperature sensor.

Body temperature is one of the most commonly used vital signs for daily health monitoring at home. However, accurately measuring core body temperature from the skin surface remains challenging, even when using specialized medical and healthcare-grade sensors. While the absolute value of body temperature is a critical diagnostic indicator, the trend of temperature change is equally important. This is because the baseline of body temperature varies among individuals depending on physiological and environmental factors. Continuous monitoring enables the observation of temporal fluctuations that conventional spot-measurement thermometers cannot capture. To address this gap, attachable flexible temperature sensors have emerged as a promising solution for continuous and noninvasive temperature tracking.

Despite their potential, several challenges must be addressed to enable the practical use of sheet-type flexible temperature sensors in clinical and personal healthcare applications. First, the influence of the environmental temperature must be minimized. Because the temperature sensor is a thin-film device, the skin and ambient temperatures can affect its outputs because of thermal conduction through the substrate and sensor materials.³⁰⁾⁾ Second, estimating the core body temperature from the skin surface temperature is inherently complex. The human body regulates core temperature by dissipating heat through the skin, especially under conditions of elevated ambient temperature or physical activity. Consequently, the skin temperature may significantly diverge from the actual core temperature. Conversely, under cold ambient conditions, skin temperature can drop without a corresponding change in core temperature.

To address the issue of ambient temperature interference, an air gap layer can be incorporated on the outer side of the sensor as a thermal insulator.³⁰⁾⁾ In this study, a PDMS structure with an air chamber was constructed between the sensor and the PDMS layer. Owing to the low thermal conductivity of air, this configuration helps isolate the skin temperature from environmental fluctuations. For estimating the body temperature, multiple sensors can be integrated to measure the skin surface temperature, radiative heat from the body, and ambient conditions simultaneously.³¹⁾⁾ By applying heat transfer models and considering the thermal conductivities of sensor materials, skin tissues, and surrounding air, core body temperature can be estimated with good accuracy. These advancements indicate that a flexible, sheet-type temperature sensor combined with wireless modules, optimized thermal insulation, and appropriate computational models holds strong potential for real-time and continuous body temperature tracking in future medical and healthcare systems.

2.4. Blood pressure (BP) sensor.

BP is a critical physiological indicator associated with various life-threatening conditions, including myocardial infarction, stroke, heart failure, and kidney failure. Owing to its diagnostic importance, BP is commonly measured using cuff-based portable devices, which are among the most prevalent home-use health monitors alongside body temperature sensors. Although these commercial devices provide accurate readings, they are inherently unsuitable for continuous and real-time monitoring, particularly in wearable applications, owing to the bulky cuff structure and inability to integrate with other vital sign sensors. To address these limitations and enhance wearability, several cuffless, flexible BP sensor patches have been developed for seamless integration into wearable systems.^32)–39)) One approach involves measuring pulse waves at two points on the wrist and analyzing the rhythmic waveform and pulse wave velocity to estimate BP levels.³²⁾⁾ By applying machine learning algorithms to the continuous pulse data, systolic and diastolic BP can be predicted with high correlation to clinical standards, achieving estimation errors below 10%. Another method uses the pulse transit time, which is calculated from the interval between the R-peak of an ECG signal and the peripheral pulse measured at the wrist.³³⁾⁾ Compared with the validated cuff-based devices, this approach shows a mean deviation of approximately 6%, which is considered acceptable for wearable health monitoring. A more recent and promising technique involves the use of ultrasound-based sensors embedded in flexible patches.^36),37)) When combined with a machine learning algorithm, these sensors directly monitor the arterial wall movement beneath the skin, enabling accurate and continuous tracking of BP. Unlike indirect methods, ultrasound-based systems offer precision and are capable of long-term, stable monitoring. However, the device cost must be reduced for wearable applications. All these approaches aim to provide a reliable estimation of systolic and diastolic BP. Although regulatory approvals, such as clearance from the U.S. Food and Drug Administration (FDA), remain pending, the concept of continuous, real-time, and noninvasive BP monitoring represents a significant advancement toward next-generation telediagnosis and early disease detection systems.

2.5. SpO₂ sensor.

SpO₂ sensors are essential tools for the noninvasive measurement of blood oxygen levels and are crucial for diagnosing and monitoring respiratory and cardiovascular conditions.⁴⁰⁾⁾ During the COVID-19 pandemic, SpO₂ was recognized as a vital indicator for the early detection of hypoxemia, a condition directly associated with life-threatening risks.^41),42)) Traditional sensors primarily use photoplethysmography (PPG) technology, with finger-clip oximeters being the most common design.⁴³⁾⁾ However, their clip-based structure causes discomfort during extended use, and their susceptibility to motion artifacts impairs accuracy. Skin tone bias also significantly affects measurement precision, often leading to overestimated values in individuals with darker skin.^42),44),45)) Additionally, high power consumption and limited suitability for continuous monitoring reduce their effectiveness in home or remote healthcare settings.

To overcome these limitations, flexible SpO₂ sensors are being developed. These conformal devices reduce motion artifacts, support prolonged use, and enable integration with multimodal sensing systems. A representative example is a wireless, low-power epidermal system designed to monitor vital signs, including SpO₂, in neonatal intensive care units.⁴⁶⁾⁾ Sensors placed on the chest and foot capture ECG and PPG signals, enabling the synchronized derivation of heart rate, variability, respiration rate, SpO₂, and systolic BP using pulse arrival time. Clinical evaluations confirmed its improved performance over commercial clip sensors across diverse skin tones with Bland–Altman analysis validating its accuracy. Real-time analytics also support anomaly detection and signal correction, thereby enhancing its clinical application.

Another promising approach involves a vertically stacked, all-organic, ring-structured ultra-low-power pulse oximeter.⁴⁷⁾⁾ This device integrates ring-shaped organic light-emitting diodes (OLEDs) emitting red and green light with a concentrically positioned organic photodiode (OPD). The design increases the photon efficiency and expands the effective area of the OPD, enabling stable PPG signal acquisition at low luminance (∼25 cd/m²) and power consumption (<1 µW). The OLED and OPD were independently fabricated on separate PET layers and bonded with a PDMS interlayer to maintain optical performance and flexibility. The device endured over 10,000 bending cycles, and its thimble-type structure shows promise for home healthcare use.

Together, these developments show advances in SpO₂ sensing for hospital-grade monitoring and wearable health platforms. Future directions may involve integrating ultra-low-power optical sensors with multimodal wireless systems to create next-generation SpO₂ technologies. Furthermore, combining these platforms with machine learning–based analytics will be instrumental for early detection and remote diagnostic applications.

2.6. Integrated multimodal sensor.

A significant advantage of multimodal wearable sensor patches lies in their ability to continuously and simultaneously monitor multiple vital signs in real time. This capability enables the early detection of health anomalies and supports autonomous telediagnosis by analyzing correlations among vital signals similar to the decision-making process of hospital clinicians. An increasing number of studies have reported multimodal flexible sensors for healthcare and medical diagnostics, robotics, and the Internet of Things applications. For multimodal sensor patches to function as practical systems, developing individual flexible sensors that meet the accuracy and reliability requirements of medical use, integrating them seamlessly, and analyzing the output data are crucial. Because the relationships between different physiological parameters can be complex and often involve subtle variations, machine learning techniques are commonly employed to extract meaningful patterns indicative of disease or abnormal conditions.

A representative example is a wireless multimodal sensor patch for remote infant monitoring, which includes a feedback mechanism to alert caregivers when abnormal conditions are detected (Fig. 3a).¹⁹⁾⁾ This flexible sensor sheet, installed on a diaper, integrates a respiration sensor, a humidity (moisture) sensor to monitor wetness, and a tilt sensor to assess the infant’s posture. The patch transmits real-time data via a BLE module to a smartphone, where the data are analyzed and immediate alerts are issued in case of concern. Despite being mounted externally on a diaper, mechanical flexibility and thin film of the sensor sheet ensure minimal discomfort (Fig. 3b). The circuit board containing the BLE module was placed as a reusable component outside the diaper (Fig. 3c). The respiration and moisture sensors were fabricated using LIG and ZIS nanomaterials, respectively, both of which function as resistive-type sensors. Infant sleep posture is a major concern because prone positioning is associated with an increased risk of sudden infant death syndrome, although the underlying mechanism remains unclear. A tilt sensor was developed using a liquid metal (eutectic gallium–indium, EGaIn) at room temperature to address this. A sealed chamber contains an EGaIn droplet whose gravity-driven motion bridges the predefined electrodes. Surface treatments and channel geometry control wetting behavior and hysteresis, enabling stable posture monitoring. The device exploits the motion of EGaIn droplets inside a small chamber with non-wettable textured LIG/PDMS electrodes. Depending on the tilt angle and direction, the droplet moves because of gravity, creating short circuits between paired electrodes. The system can determine the sleep posture by monitoring the ON/OFF states across multiple electrode pairs (Fig. 3d).

Figure 3.

Multimodal sensor sheet with feedback system. (a) Schematic and photo of a wireless multimodal hybrid sensor system installed in a diaper to monitor posture, respiration, and wet conditions. (b) Photo of the multimodal sensor sheet installed on a diaper. (c) Wireless and signal processing circuit board. (d) Posture detection by monitoring the ON and OFF states of each electrode. (e) Application display with feedback functions. (f) Photo of the test. (g) Results of real-time and continuous monitoring of breathing, wettability, and posture. Reproduced with permission from Ref. 19. Copyright 2021 Wiley-VCH GmbH.

Following the individual development and characterization of each sensor, the components were integrated on a PI film and coupled with a smartphone application (Fig. 3e). As a proof-of-concept, the integrated sensor sheet was tested on an adult volunteer, successfully monitoring the respiration rate, moisture level, and body posture in real time during sleep (Fig. 3f). Crucially, the feedback alarm system functioned correctly when respiration ceased, humidity increased, or prone posture was detected (Fig. 3g). Although correlation analysis among multiple vital signs was not conducted in this study and only respiration was monitored as a physiological signal, the study shows a promising platform for remote health monitoring. Further research is warranted to expand the number of integrated vital signs and to explore the correlation of signals for comprehensive telediagnosis applications.

3. Biomarker monitoring

In addition to the five physiological vital signs, blood tests represent a global gold standard for assessing health and diagnosing medical conditions. They provide detailed insights into specific diseases and biological abnormalities. For instance, patients with diabetes routinely monitor their blood glucose levels through self-administered blood sampling. However, such tests are invasive, necessitate skin puncture, and carry risks of infection during collection and handling. Moreover, for applications in remote healthcare and telediagnosis, the analysis of a broader range of biomarkers beyond glucose is essential for early-stage disease detection. In emergency or when assessing seemingly healthy individuals, biomarkers, such as lactic acid and various metabolites, can play a crucial role in determining the urgency of medical intervention. Therefore, biomarker selection must be tailored to the specific application and ensure the simultaneous collection of multiple metabolites and physiological parameters to support a wider range of diagnostic scenarios. As with physiological vital signs, analyzing correlations between multiple biomarkers may enhance diagnostic accuracy and enable presymptomatic disease identification. Simultaneous monitoring of vital signs and biomarkers could offer a similar diagnostic capability compared with initial clinical evaluation in hospitals.

To address this need, bandage-type, noninvasive multimodal flexible sweat sensor patches have attracted interest as a means of detecting diverse metabolites in real time. Enzyme-based sensing materials are often employed to achieve high selectivity and sensitivity toward specific analytes. However, collecting sufficient volumes of sweat is a key challenge in sweat-based sensing, especially in the absence of physical activity or elevated ambient temperature. To overcome this limitation, iontophoresis was introduced onto flexible substrates.⁴⁸⁾⁾ This approach uses a small current (typically sub-milliamperes) applied through the skin to actively stimulate sweat secretion, eliminating the need for exercise or heat exposure.

This section highlights individual flexible chemical sensors designed for sweat-based detection as well as fully integrated multimodal systems that combine the continuous and simultaneous monitoring of multiple biomarkers and vital signs; these emerging technologies pave the way toward truly personalized and data-driven telediagnosis solutions.

3.1. Glucose and lactate sensors.

Glucose levels are a critical biomarker for diabetes diagnosis and management. Patients with diabetes, particularly those with type 1 diabetes, must regularly monitor their glucose levels and administer insulin as needed to maintain glycemic control. Type 1 diabetes is characterized by autoimmune-mediated destruction of pancreatic β-cells, resulting in an absolute deficiency of insulin production.⁴⁹⁾⁾ Consequently, patients require lifelong external insulin therapy, and immediate insulin administration is often necessary when blood glucose levels rise to prevent acute complications such as diabetic ketoacidosis. Therefore, continuous glucose monitoring is essential for maintaining glucose within a physiological range and preventing hyperglycemia and hypoglycemia.⁵⁰⁾⁾

To support continuous glucose monitoring, minimally invasive epidermal monitoring systems using microneedles have been commercialized. In parallel, flexible glucose sensors designed for long-term, continuous wearable devices have received considerable attention, particularly for their potential integration into multimodal wearable platforms. The concept is to detect biomarkers from sweat on the skin as a noninvasive monitoring system. These sensors often utilize glucose oxidase (GOx) as a sensitive enzyme material, which operates via an amperometric detection mechanism.⁷⁾⁾ In this mechanism, glucose undergoes enzymatic oxidation by GOx in the presence of oxygen, producing gluconic acid and hydrogen peroxide (H₂O₂). The H₂O₂ is subsequently decomposed at the electrode interface, often facilitated by electron mediators such as Prussian blue, generating protons and electrons (2H⁺ + 2e⁻), which produce a current flow proportional to the glucose concentration. Wang et al. demonstrated a proof-of-concept wearable glucose sensor using this enzymatic-amperometric approach in a tattoo-based sensor platform, incorporating GOx and Prussian blue-modified carbon electrodes.⁹⁾⁾ The sensor exhibited a sensitivity of ∼23 nA/µM and a limit of detection of ∼3 µM. Since this pioneering work was reported, numerous flexible and skin-conformal glucose sensors have been reported, focusing on improved long-term stability and demonstrations for personalized healthcare applications.

Lactic acid is another important parameter for emergency situations, causing various diseases such as septic shock, cardiogenic shock, hemorrhagic shock, acute respiratory failure, and multiple organ failure.⁵¹⁾⁾ Therefore, this lactic acid biomarker is often measured for patients in emergency situations and to determine priority for medical treatment as triage.⁵²⁾⁾ Although this is an important indicator for telediagnosis in emergency and triage whether the patients need to see a doctor, a wearable, continuous, real-time lactic acid monitoring system has yet to be widely available. However, recently, enzyme-based flexible lactate sensors have been studied to continuously monitor lactic acid from sweat. The sensing mechanism is almost the same as that of the glucose sensor discussed before, and only the enzyme of L-lactate oxidase is different.⁵³⁾⁾ As exemplified by the lactate sensor, enzyme-based chemical sensors can be readily fabricated when the appropriate enzyme is available because of the general uniformity of the underlying sensor platform across different targets. This modularity facilitates the seamless integration of multiple sensors for the simultaneous monitoring of diverse biomarkers. Although practical challenges remain, such as ensuring long-term stability and managing material costs, the underlying materials and device platforms exhibit strong potential for the development of future telediagnostic systems.

3.2. Electrolyte and sweat rate sensor.

Excessive loss of electrolytes and perspiration can precipitate dehydration as well as cardiac and neurophysiological dysfunction, ultimately posing life-threatening risks.⁵⁴⁾⁾ Continuous, real-time tracking of sweat composition therefore requires the simultaneous determination of ion concentration and sweat rate, given their strong physiological interdependence. Such capability is indispensable not only for older adults who often exhibit impaired thermoregulation and respiratory control⁵⁵⁾⁾ but also for children, athletes, and laborers operating in hot, humid, or otherwise extreme environments.

To meet this need, flexible ion-selective sensors, such as Na⁺, K⁺, and Cl⁻, have been developed on a flexible film, where ion-selective membranes transduce concentration changes into open-circuit potentials according to the Nernst equation.^{10),12),56),57))} While analyte-specific platforms offer precise clinical insight, total electrolyte monitoring has also been explored as an initial screening tool.^58),59)) In this approach, the electrical conductivity of sweat serves as a proxy for the aggregate ionic content, enabling simpler architectures and lower cost than ion-sensitive sensors.

In addition to monitoring electrolyte concentrations, quantifying the sweat rate is essential for accurately assessing ionic loss and overall hydration status.⁶⁰⁾⁾ Several techniques have been proposed for flexible sensor platforms to evaluate sweat rate. A commonly used approach involves placing a humidity sensor within a confined cavity between the skin and sensor.⁶¹⁾⁾ This method allows for accurate estimation of sweat rate by monitoring changes in relative humidity. However, during periods of high perspiration, such as intense exercise, the enclosed space can quickly reach 100% humidity saturation, making it difficult to effectively measure the high sweat rate.

Another technique involves monitoring conductance changes within a fluidic channel.¹⁸⁾⁾ In this method, interdigitated electrodes are patterned along a long fluidic pathway, and the conductance increases in separate steps as sweat gradually fills the channel. While this approach offers simplicity and compatibility with flexible substrates, its limitation lies in its inability to track ongoing perspiration once the channel is fully filled by sweat, unless a mechanism for fluid removal is incorporated. This constraint renders continuous monitoring during heavy sweating scenarios, such as sports or heatstroke conditions, less suitable.

A droplet-based sensing strategy has been developed to address these limitations, enabling long-term and continuous monitoring of sweat rate even under high perspiration.⁵⁹⁾⁾ This system uses a specially designed chamber where sweat accumulates and forms a droplet that intermittently bridges the gap between two electrodes in the channels. Each droplet connection results in a spike-like increase in the electrical conductance, and the sweat rate can be calculated by measuring the time interval between these spikes. As the formation and release of sweat droplets are influenced by gravity, the orientation of the sensor affects its output, which is critical for wearable applications. To overcome this, the system incorporates an inertial measurement unit consisting of an accelerometer and gyroscope along with machine learning algorithms to compensate for the tilt angle, allowing accurate sweat rate detection regardless of body position or movement.

To evaluate the long-term monitoring of sweat rate and electrolyte concentration under high perspiration conditions, integrated sweat sensor patches were applied to 10 subjects during a home-use sauna, with and without the intake of a sports drink. In all cases, the sweat rate was successfully recorded without performance degradation, even under high sweat rates and prolonged measurements. In addition to sweat rate tracking, an intriguing trend was observed: the sweat impedance, which correlates with electrolyte concentration, exhibited a noticeable decrease of approximately 5–15 minutes after intake of the sports drink for 7 out of 10 volunteers. In contrast, no significant impedance change was observed in the absence of drink intake. While direct comparison with blood electrolyte data obtained from clinical devices was not conducted, limiting definitive conclusions about hydration status, these findings highlight the potential of this method for continuous, noninvasive sweat-based monitoring of physiological conditions.

3.3. Strategy to improve chemical sensing performance.

Sweat sensors offer significant potential as noninvasive platforms for the continuous monitoring of physiological conditions by detecting and analyzing biomarkers in sweat. However, two primary challenges limit the practical application of sweat-based chemical sensor patches: the low sensitivity to detect low biomarker concentrations and the need for signal calibration owing to instabilities caused by external factors, such as temperature fluctuations. The sensitivity of such chemical sensors is often governed by the Nernst equation; for example, in the case of pH sensing, the theoretical sensitivity is approximately 60 mV per unit pH at room temperature.

To detect subtle changes in analytes, such as glucose, which is typically present in low concentrations in sweat, higher sensitivity is essential. While signal amplification can be implemented via an external circuit board in a hybrid system, this configuration introduces the risk of noise generation at the sensor–amplifier interconnection. For reliable telediagnosis applications where the accurate detection of small signal variations is crucial, the direct integration of the amplifier with the sensor is ideal.

To address this need, a flexible charge-coupled device (CCD) chemical sensing platform was demonstrated, integrating flexible transistors on a PI film for signal amplification (Fig. 4a–b).⁶²⁾⁾ The sensing mechanism transfers the electron charges injected from the input voltage (V_input) into a capacitor, where they are read as an output voltage. The charge transfer is initiated by two sequential gate inputs (V_ICG and V_TG), and the amount of transferred charge is determined by the potential well depth in the semiconductor channel, which varies with the pH of the solution. By repeating this transfer process, electrons accumulate in the capacitor in proportion to the pH value. After the output voltage is read, the accumulated charge in the capacitor is discharged by applying a reset voltage (V_RST), enabling the measurement cycle to restart and allowing continuous pH monitoring within this CCD-based platform. Although this study focused on high-sensitivity pH sensing, the underlying potentiometric mechanism is broadly applicable to other ion- or biomarker-specific sensors. In this architecture, the electrical potential generated by ionic concentration changes modulates the amount of charge transferred to a capacitor, which is then accumulated over multiple CCD transfer cycles. Unlike conventional analog amplifiers, which are often difficult to implement on flexible substrates owing to variations in device parameters, such as threshold voltage, the CCD-based amplification mimics a digital process, generating stepwise output voltages that are robust against such variability (Fig. 4c). This concept enabled the realization of a sensitivity of 240 mV/pH, approximately four times higher than the theoretical Nernst limit, with further enhancement achievable by optimizing the transfer cycles, operating voltages, and target detection range. The CCD-based pH sensor patch was successfully applied to the human skin as a proof-of-concept, allowing for continuous sweat pH monitoring (Fig. 4d–e). A temperature sensor was integrated with the CCD-based pH sensor to correct for temperature-dependent variations in transistor performance and chemical sensing behavior, ensuring greater reliability of the measured data. In parallel with the CCD platform, recent electrochemical microfluidic systems provide complementary pathways tailored to metabolite and hormone targets. Representative wearable platforms integrate on-patch sampling and sensing for metabolites and nutrients (Fig. 4f),⁶³⁾⁾ aptamer-based estradiol sensing with on-patch calibration (Fig. 4g),¹¹⁾⁾ and multiplexed profiling of stress hormones via iontophoresis and valved microfluidics (Fig. 4h).⁶⁴⁾⁾ Direct performance comparisons are not appropriate because these approaches target different analytes and employ distinct transduction mechanisms; instead, they highlight orthogonal design trade-offs in sampling, calibration, and system integration.

Figure 4.

Charge-coupled device (CCD)-based sweat pH sensing and complementary electrochemical/microfluidic exemplars for sweat biomarkers. (a) Photo of the CCD-based pH sensor integrated with a temperature sensor. (b) Equivalent circuit of the pH sensor. (c) Input, reset, and output voltages of the CCD operation. (d) Photo of a demonstration to monitor sweat pH sensor. (e) Output results of sweat pH and skin temperature with control results measured by commercial sensors. Reproduced with permission from Ref. 62. Copyright 2018 Nature Publishing Group. (f) Wearable electrochemical microfluidic platform with on-patch sweat stimulation electrodes, temperature sensor, and sweat biosensing interface. Reproduced with permission from Ref. 63. Copyright 2022 Nature Publishing Group. (g) On-skin aptamer-based estradiol patch (finger-mounted demonstration). Reproduced with permission from Ref. 11. Copyright 2024 Nature Publishing Group. (h) Multiplexed stress-hormone biosensor integrating on-board electronics and microfluidics. Reproduced from Ref. 64 (CC-BY).

Although the above platforms enhance sensitivity and analyte coverage, robust calibration under dynamic wearable conditions remains critical to address challenges such as temperature and pH variations. Such dependencies are particularly problematic in wearable applications because of the dynamic fluctuation with activity, environment, and individual physiology. Although body temperature is a critical parameter for telediagnosis and should be monitored in wearable systems, compensating for its effects on chemical sensor outputs requires additional signal processing and power consumption. To overcome this issue, a promising approach known as the “time-derivative of open-circuit potential (dOCP/dt)” method has been proposed.⁶⁵⁾⁾ This technique effectively minimizes external interferences, including temperature and pH variations. Furthermore, this approach does not affect the sensor size, suggesting that the size variation in the sensor fabrication can be ignored to precisely measure the sensor output. The dOCP/dt method offers a robust pathway toward more stable and reliable chemical sensing in practical wearable applications by minimizing the impact of external interferences.

3.4. Integrated multimodal sensor.

Monitoring a single specific biomarker provides only limited insight for remote diagnosis and thus has a relatively small impact on the application of wearable sensors. To enable accurate and meaningful telediagnosis comparable with assessments performed by clinicians in close proximity to patients, correlations among multiple vital signs and biomarkers must be analyzed. In this context, multimodal sensor patches are discussed as a next-generation solution for healthcare and medical applications. Recent developments include multimodal patches for vital sign monitoring,⁴⁾⁾ biomarker detection,^{10),14),66),67))} and hybrid physicochemical sensing platforms.^68),69))

While the optimal integration of various sensors, such as those for body temperature, ECG, respiration, BP, SpO₂, lactate, and other key biomarkers, is still under investigation, research on the practical realization of multimodal sensor systems is making especially rapid progress. One example involves the integration of ECG, skin temperature, and glucose sensors with a fluidic platform, which effectively reduces signal distortion caused by glucose accumulation on the sensor surface, thereby enhancing measurement stability and accuracy during continuous monitoring (Fig. 5a–b).⁶⁹⁾⁾

Figure 5.

Chemical and physical sensor patch. (a) Schematic and (b) photo of an integrated glucose sensor patch with a fluidic channel, an electrocardiogram (ECG) sensor, and a temperature sensor. (c) Skin-interfaced wireless neonatal intensive care unit platform showing chest and limb units for clinical-grade monitoring; additional motion/orientation/acoustic channels are supported. Reproduced with permission from Ref. 5. Copyright 2020 Nature Publishing Group. (d) Chip-less, battery-free, and stretchable passive-tag network featuring a soft, stretchable on-skin sensor and a flexible readout circuit integrated into clothing for mechanically decoupled, multisite sensing. Reproduced with permission from Ref. 70. Copyright 2019 Nature Publishing Group. (e) All-solution-processed ultra-flexible optoelectronics demonstrating an organic photovoltaic module on the hand and a photoplethysmography sensor on the fingertip, indicating self-powered, scalable on-skin implementations. Reproduced from Ref. 71 (CC-BY). Real-time and continuous monitoring results measured by a volunteer who conducted the bike-pedaling exercise: (f) pedaling speed, (g) skin temperature difference, (h) ECG signals, (i) heart rate (extracted from the ECG signals, and (j) glucose level in sweat. (a, b, f–j) Reproduced with permission from Ref. 69. Copyright 2021 American Chemical Society.

The fluidic channel plays a key role by continuously refreshing sweat at the glucose sensor interface. As a result, the glucose signal remains stable over extended periods, unlike the nonchannelized counterpart, which showed a gradual increase in signal output even under a constant glucose concentration, likely due to local accumulation effects. Representative skin-interfaced wireless platforms have achieved clinical-grade monitoring in neonatal intensive care settings, incorporating additional motion, orientation, and acoustic channels (Fig. 5c).⁵⁾⁾ In complementary approaches, chip-less, battery-free stretchable passive-tag networks have demonstrated mechanically decoupled, multisite sensing (Fig. 5d),⁷⁰⁾⁾ whereas all-solution-processed ultra-flexible optoelectronics with integrated organic photovoltaics indicate self-powered, scalable on-skin implementations (Fig. 5e).⁷¹⁾⁾

Returning to the fluidic-integrated system, multi-parameter monitoring during cycling was demonstrated. Once the stable operation of the glucose sensor in sweat was validated, the integrated system was applied for continuous, real-time monitoring of multiple physiological and biomarker parameters during a cycling exercise (Fig. 5f). As expected, the ECG-extracted vital signs of skin temperature and heart rate increased during physical exercise, reflecting normal physiological responses (Fig. 5g–i). Despite the use of a hydrogel-based electrode sheet, the ECG signal remained stable even under profuse perspiration (Fig. 5h). However, signal amplitude began to attenuate after approximately 1300 s, likely because of the increased impedance caused by the excessive swelling of the hydrogel and the weakened contact at the hydrogel-skin interface. For glucose monitoring, no signal was initially observed at the onset of exercise owing to the absence of sweat in the fluidic channel (Fig. 5j). Continuous glucose detection was successfully achieved once the channel was filled with sweat, indicating the efficacy of fluidic-assisted sensing for long-term wearable chemical monitoring.

4. Edge system

Multimodal sensor patches hold great promise for wearable telediagnosis by enabling the continuous monitoring of multiple physiological parameters and facilitating the correlation-based analysis of these datasets. Real-time, continuous acquisition of diverse biosignals represents a transformative shift in healthcare, offering a path toward proactive and individualized medical intervention. However, the vast volume of data generated from such multimodal systems presents a significant challenge: extracting meaningful insights from complex datasets in real-time remains a bottleneck, especially when timely feedback to the user is required to alert abnormal physiological conditions. Moreover, in emergency scenarios, such as natural disasters, access to cloud infrastructure may be disrupted, rendering cloud-based computation infeasible. To overcome this limitation, edge computing where data analysis is conducted locally on a smartphone or microcontroller, becomes a critical enabler for real-time, Internet-independent telediagnosis. Among various computational strategies, reservoir computing (RC), a type of recurrent neural network, has emerged as a viable approach for efficiently processing time-series physiological data and capturing correlations across multiple modalities. RC maps an input time series into a high-dimensional, dynamic reservoir formed by fixed, nonlinear recurrent connections while training is applied only to the readout layer. The computational cost is significantly reduced because the reservoir weights remain untrained, facilitating streaming inference and bounded-latency updates on edge devices. These features make RC particularly well suited for implementation in wearable sensing patches.

To demonstrate this concept, a hybrid multimodal sensor patch integrating ECG, skin temperature, perspiration, and respiration sensors was developed along with a BLE module, microprocessor, and accelerometer (Fig. 6a–c). These physiological signals were wirelessly transmitted to a smartphone via BLE, where an RC algorithm was implemented in a dedicated smartphone app (Fig. 6d–e). Data acquisition and lightweight preprocessing were performed on the patch, and the detailed data analyses, including feature extraction and RC-based classification, were executed locally on the smartphone, which provided real-time feedback without cloud processing. In the human-subject testing, arrhythmia, coughs, and falls were detected on the phone (Fig. 7a). Participants performed various physical activities, and several individuals with known arrhythmia participated in the evaluation to validate the real-time analytical capability of the system. Although the smartphone-based system (Google Pixel 5a) exhibited a slight processing delay of approximately 5 s likely owing to the limited computational power compared with a standard laptop computer, the classification performance remained comparable with that achieved using a conventional computer, including laptops (Fig. 7b–c). For instance, arrhythmia and coughing events were detected with high accuracy, achieving recognition rates of 0.966 and 0.843, respectively. Importantly, the system demonstrated strong generalization, maintaining performance across different individuals without retraining. In summary, although challenges remain for fully real-time processing on mobile platforms as telediagnosis, this demonstration highlights the feasibility and effectiveness of the edge system toward wearable telediagnosis. Such edge computing–enabled sensor systems represent a promising step toward scalable, infrastructure-independent remote healthcare solutions.

Figure 6.

Multimodal sensor patch integrated with wireless and signal processing system powered by machine learning. Photos of the patch (a) on a body, (b) before attaching with a support film, and (c) while delaminating the supporting film. (d) Detailed system diagram. (e) Machine learning algorithms to detect multiple information. Reproduced from Ref. 2 (CC-BY).

Figure 7.

Edge-computing wearable sensor. (a) Photo of a demonstration of the edge-computing sensor. (b) Real-time results analyzed by a machine learning algorithm installed in a smartphone. (c) Display of real-time edge-computing smartphone. Reproduced from Ref. 2 (CC-BY).

5. Translation to practical applications

5.1. Regulatory approvals for wearable sensor patch.

Translating wearable sensor patches from laboratory prototypes to clinically approved devices requires compliance with national regulatory frameworks, such as those in the U.S. FDA and Japan’s Pharmaceuticals and Medical Devices Agency (PMDA). In the U.S., medical devices are categorized into Class I, II, or III under the Quality System Regulations (21CFR Part 820) and Good Laboratory Practice (21 CFR Part 58), depending on their intended diagnostic or therapeutic use. Wearable devices for physiological monitoring and drug delivery are typically classified as Class II or III, necessitating either Premarket Notification (510(k)) or Premarket Approval (PMA) depending on their risk level. In Japan, the PMDA enforces equivalent regulatory procedures through a quality management system aligned with ISO 13485 and ISO 14971.

The approval process for wireless wearable devices generally involves several critical stages. Initially, measurement repeatability, accuracy, and robustness must be clinically validated under diverse conditions, such as movement, perspiration, and temperature and humidity variations. Safety and quality assurance were confirmed through GMP/ISO 13485-certified fabrication processes, evaluation of biological risks in accordance with ISO 10993, and verification of mechanical reliability through accelerated aging and sterilization testing. Because physiological data in wearable systems are transmitted wirelessly, cybersecurity and privacy protection are also essential regulatory components. To ensure data safety, the system and its software are evaluated for compliance with international and domestic standards including IEC 60601-1 for electrical safety, UL 2900 for software cybersecurity, and the FDA Cybersecurity in Medical Devices guidance. Measures should include end-to-end encryption for data in transit (e.g., TLS 1.3) and encryption at rest (e.g., AES-256), supported by hardware-backed key management and tamper-resistant on-device storage (e.g., secure elements/trusted execution environment (TEE)), together with strong authentication and authorization to mitigate risks during remote access. Ethical challenges also arise because sensitive data are vulnerable to unauthorized access or re-identification, and studies have reported that even limited attributes can enable 99.98% re-identification accuracy.⁷²⁾⁾ Accordingly, to avoid ethical dilemmas, informed consent procedures must be transparent, particularly for vulnerable populations. Moreover, existing frameworks, such as the Health Insurance Portability and Accountability Act, often do not fully cover consumer-generated wearable data, creating regulatory gaps and inconsistencies between federal and state laws. International differences, for example, the stringent General Data Protection Regulation of the European Union versus the greater reliance on self-regulation of the United States can increase compliance costs and produce uneven consumer protection.⁷³⁾⁾ Finally, algorithmic bias in artificial intelligence (AI) analyses may exclude minority groups and undermine health equity, whereas a balanced benefit–risk assessment that weighs privacy risks against preventive benefits is required. After these evaluations, human-factor studies and clinical trials are conducted under the approval of the Institutional Review Board to assess the system’s usability and benefit–risk balance. Finally, postmarket surveillance is required to continuously monitor safety and performance after commercialization. Future regulations should explicitly address the sensor-to-sensor variability that affects data quality and require real-time bias detection within AI-integrated systems. Improvements in privacy-preserving distributed architectures, user-centric consent interfaces, and AI transparency are recommended to help bridge these gaps. Establishing such a comprehensive regulatory framework is indispensable for accelerating the clinical translation and widespread adoption of wearable medical devices.

5.2. Long-term user compliance and durability.

Because wearable sensor patches are designed for continuous and long-term use, their stability, comfort, and user adherence are as crucial as sensing accuracy. Continuous attachment of a sensor patch to the skin exposes it to various mechanical and environmental stresses, including stretching, perspiration, sebum, and temperature fluctuations, which can degrade the signal quality or cause skin irritation. Several strategies focusing on material design and adhesion control have been proposed to overcome these challenges. Breathable and vapor-permeable films incorporating porous or kirigami structures have proven effective in mitigating sweat accumulation by providing physical vapor pathways from the skin to the external environment. Another promising approach involves fabricating ultrathin sensor sheets that closely conform to the skin’s microtexture, thereby achieving excellent adhesion and comfort without the need for additional adhesives. However, such an ultrathin structure can be fragile and difficult to handle in practical applications.

An alternative strategy is to optimize the adhesive strength to balance durability and comfort; excessive adhesion can cause irritation, whereas insufficient adhesion leads to delamination due to sweat or motion. To further enhance stability, a protective encapsulation layer that blocks moisture, ultraviolet light exposure, and temperature fluctuations has been employed to ensure reliable operation in various environments. When sensor outputs exhibit drift or motion-induced artifacts, calibration algorithms must be integrated into the system to compensate for such variations and maintain accurate physiological monitoring.

5.3. Standardization of medical information systems.

For the effective integration of wearable sensor patches into clinical practice, their physiological data and diagnostic outputs must conform to a standardized medical information system. The Health Level 7 (HL7) standard framework provides an essential foundation for the exchange, integration, and secure management of electronic health information. Standardization requires not only technical compatibility but also semantic harmonization of health data across different systems to ensure that information can be universally interpreted and used. Furthermore, establishing interoperability among devices involves the unification of wireless communication protocols, synchronization of data timestamps across multimodal sensing units, and verification of analytical accuracy obtained by edge computing compared with clinical gold standard instruments and physician diagnosis. Implementing these standardization efforts not only streamlines regulatory audits but also enables more accurate and reliable diagnostic support through machine learning–based data analysis, ultimately facilitating the seamless incorporation of wearable systems into existing medical infrastructures.

5.4. Social impacts.

Edge-enabled wearable sensor patches for telediagnosis can significantly transform healthcare systems by reducing medical costs, improving accessibility, and enabling an early disease detection. According to the U.S. Centers for Disease Control and Prevention, approximately 90% of annual healthcare expenditures are spent on chronic and health conditions. Continuous remote physiological monitoring offers a shift from reactive treatment to preventive medicine, allowing earlier interventions and reducing hospitalizations. Compared with conventional diagnostic instruments, these lightweight and low-cost wearable patches can be widely distributed, enabling individuals to easily monitor their physiological status and identify abnormalities before symptoms become severe.

Such wearable systems can play an essential role during emergencies or disasters, such as earthquakes, pandemics, or extreme weather events, when medical infrastructure may be disrupted. These patches can facilitate rapid triage and remote medical support by transmitting physiological data wirelessly to clinicians. When combined with edge computing, they can autonomously analyze data and provide immediate feedback, thereby contributing to resilient, efficient, and equitable healthcare systems that support individual well-being and societal sustainability.

5.5. Long-term power management and energy harvesting.

Continuous wireless transmission and on-device machine learning in wearable sensor patches require sustained power for autonomous operation. Energy-harvesting strategies and ultra-low-power circuit design are essential for ensuring long-term monitoring without the need for frequent recharging. In this regard, thermoelectric generators (TEGs) that convert body heat into electricity, including flexible devices based on PDMS matrices or organic materials, have shown potential as wearable applications.⁷⁴⁾⁾ Under typical skin–ambient temperature gradients, TEGs generally deliver electrical power densities of a few to tens of µW·cm⁻² (with ∼35 µW·cm⁻² reported under favorable conditions), whereas skin heat flux can reach the mW·cm⁻² range depending on conditions.^74),75)) These harvesters supplement rechargeable batteries, reducing reliance on external power sources and thereby enhancing system sustainability. Additionally, ultra-low-power circuit optimizations, including event-driven processing and complementary metal–oxide–semiconductor subthreshold operation, can reduce the average consumption of optical biosensors (e.g., organic pulse oximeters) to from a few to tens of µW, with sub-µW operations reported in specific designs.^40),47)) Integrating harvesters and sensing front ends with power-management integrated circuits (PMICs), preferably with maximum power point tracking; safe charging; and medically robust packaging that withstands exposure to sweat, washing, and temperature variations, supports reliable performance in resource-constrained edge environments and helps address the challenges of prolonged use in telediagnosis.⁷⁶⁾⁾

5.6. Hardware implementation of machine learning on edge devices.

While RC enables efficient on-device analysis for wearable sensors, smartphone-based demonstrations have reported delays in the order of seconds (∼5 s), which can limit true real-time telediagnosis.²⁾⁾ To address this limitation, next-generation hardware architectures have been recommended, including application-specific integrated circuits for low-latency RC inference, field-programmable gate arrays for flexible reconfiguration, and neuromorphic computing platforms that mimic neural dynamics for energy-efficient processing.⁷⁷⁾⁾ For instance, memristor- or ferroelectric-based neuromorphic reservoirs have demonstrated millisecond-level delays with device- or array-level power in the nanowatt to microwatt range; further, in end-to-end systems, total power includes sensing, input/output, and PMIC overhead.^78),79)) In practice, fixed-point microcontroller unit pipelines with circular buffers and sliding windows, combined with 8-bit quantization, pruning, operator fusion, and interrupt-driven front ends, can bound latency and energy per decision while preserving accuracy.⁸⁰⁾⁾ Targeting <1 s end-to-end latency under battery-constrained energy budgets via hardware software co-design enables seamless integration with edge systems for autonomous health feedback.

6. Conclusion and outlook

This review summarized recent advancements in noninvasive, continuous, flexible sensor patches designed for detecting physiological vital signs and sweat biomarkers, focusing on enabling telediagnosis for remote healthcare applications at home and in emergency scenarios such as natural disasters. Beginning with the core technologies of flexible circuits and sensors, we reviewed noninvasive monitoring of vital signs such as ECG with kirigami structures for enhanced wearability, respiration via LIG-based strain sensors, perspiration-related skin humidity, skin temperature trends, cuffless BP patches using piezoelectric or ultrasound methods, and SpO₂ via organic optoelectronics along with biomarker sensing (glucose, lactate, electrolytes, and sweat rate). Subsequently, we discussed integrated and hybrid systems that pair these modalities with wireless transmission and machine learning for edge-computing wearable devices performing on-device analysis without cloud dependence, thereby reducing latency and power consumption. The review also highlighted the importance of analyzing correlations among various physiological parameters to facilitate real-time, continuous diagnosis similar to clinical evaluations performed in hospitals. In addition, this review discussed the role of edge-computing systems in identifying abnormal health conditions and providing timely feedback, demonstrating their significance for future telediagnosis platforms.

A wide range of innovations, spanning novel sensor architecture, transducers, and system-level developments such as data analysis and edge AI integration, have been introduced. These advancements mark a significant step toward sustainable healthcare by enabling home-based telediagnosis and earlier interventions in emergencies, thereby helping to narrow the growing gap between life expectancy and healthy life expectancy. Nevertheless, several challenges remain, including motion artifacts, signal noise, and material durability, as evidenced by ongoing needs for breathable designs, robust encapsulation, and motion-robust signal chains. Extensive clinical validation is essential for transitioning from laboratory-scale demonstrations to practical medical applications. While many existing studies on flexible sensors focus primarily on discovering new sensing modalities and improvements in sensitivity and selectivity, these efforts must be complemented by sensor integration strategies and practical data acquisition through clinical collaborations. Furthermore, it is imperative to develop fully integrated systems that seamlessly combine edge-computing architectures with flexible materials, sensors, transducers, and algorithms with long-term operational stability, reliability, and low power consumption. Looking ahead, further optimization of sensor integration is needed to achieve seamless, long-term wearability and accelerate clinical translation. Guided by practical considerations, regulatory approvals (e.g., FDA/PMDA in alignment with ISO standards and cybersecurity protocols) and standardization (e.g., HL7/Fast Healthcare Interoperability Resources for electronic health record interoperability) can accelerate adoption. Enhancing durability against sweat and motion along with energy harvesting (e.g., body-heat TEGs delivering power in the order of µW·cm⁻²) and ultra-low-power circuits (e.g., sub-µW organic sensors) will enable prolonged autonomous operations. Advanced hardware for edge machine learning such as neuromorphic reservoirs with millisecond-scale latencies has the potential to deliver real-time feedback. Ultimately, expanded clinical trials and interdisciplinary collaborations with healthcare professionals will be necessary to establish future telediagnosis.

Biographies

Profile

Do Hoon Lee was born in Daegu, Republic of Korea, in 1993. He graduated from Soongsil University with a degree in Mechanical Engineering in 2015 and earned his Ph.D. in Mechanical Engineering from Korea University, Seoul, in 2022. During his doctoral studies, he conducted research on nanomaterial-based biosensors and strain, gas, and other functional sensors for advanced detection technologies. Since 2023, he has been a postdoctoral researcher at Graduate School of Information Science and Technology, Hokkaido University, where he continues his work on laser-induced graphene, photothermal materials, and wearable sensors for healthcare. His research aims to advance ultra-high-sensitivity mechanical sensors, multimodal biosignal monitoring systems, and intelligent diagnostic technologies. Lee’s work focuses on bridging nanotechnology with practical biomedical applications, contributing to innovations in real-time health monitoring and personalized medicine.

Kuniharu Takei was born in 1980. He received his B.Eng. degree from Toyohashi University of Technology in 2005 and his Ph.D. in 2009 for research on the development of microneedle arrays integrated with circuits for neural signal monitoring. From 2009 to 2013, he was a postdoctoral fellow at the University of California, Berkeley. Subsequently, he served as an assistant, associate, and full professor at Osaka Prefecture University and later as a full professor at Osaka Metropolitan University. Since 2023, he has been a full professor at Graduate School of Information Science and Technology, Hokkaido University. He was also appointed as a JST PRESTO researcher. His research focuses on developing platforms for multimodal, high-performance flexible electronics that integrate inorganic and organic materials. He has also explored machine learning approaches for automated, real-time data analysis in multimodal flexible sensor systems. His achievements have been recognized with numerous awards, including the JSPS Prize (2025), Marubun Prize (2021), Funai Prize (2020), Young Scientist’s Prize by MEXT (2018), NISTEP Researcher by MEXT (2015), and 35 Innovators under 35 by MIT Technology Review (2013).

Conflict of interest

The authors declare no conflicts of interest.