A cross-modal epidermal sensor enables single-channel fusion of biopotential and biomechanical signals

A cross-modal epidermal sensor enables single-channel fusion of biopotential and biomechanical signals | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article A cross-modal epidermal sensor enables single-channel fusion of biopotential and biomechanical signals Yuxin Liu, Xiaodong Wu, Jinchao Wang, Cheng Zhu, Lifei Zheng, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7300896/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 19 Mar, 2026 Read the published version in Nature Sensors → Version 1 posted 10 You are reading this latest preprint version Abstract Disease monitoring typically requires the acquisition of multiple physiological signals with different modalities. Existing epidermal electronics use separate sensors for each modality, which necessitates a large footprint, high bandwidth and power consumption. We report a wearable electronic system that can fuse physiological signals with multiple modalities into a singlecross-modal biosignal (X-Sig). Leveraging hierarchical device architecture and in-sensor signal fusion strategy, X-Sigsensor concurrently acquires biopotential signals (e.g., electrocardiography and electromyography) and biomechanical signals (e.g., force myography and radial pulse) through a single channel. The single-channel X-Sig sensor is capable of continuous monitoring of multiple dynamic haemodynamics, including heart rate, pulse arrival time, diastolic and systolic blood pressure with high accuracy. In machine-learning-based gesture recognition, the X-Sig sensor reduced the decoding error rate by 7.8-fold compared to conventional electromyography. By fusing complementary modalities at the sensor level, X-Sig sensor provides a versatile platform for designing bandwidth-efficient and low-power wearable electronics. Physical sciences/Engineering/Biomedical engineering Physical sciences/Materials science/Materials for devices/Sensors and biosensors cross-modal epidermal sensor dynamic haemodynamic sensing wearable electronics Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Full Text Human skin serves not only as a barrier but also as a window into the body’s internal physiology. A wide spectrum of vital physiological signals can be detected on the skin, such as electrocardiography (ECG), [ 1-3 ] electromyography (EMG), [ 4 , 5 ] force myography (FMG), [ 6 , 7 ] radial pulse wave (RPW). [ 8 , 9 ] The continuous and simultaneous acquisition of multiple epidermal physiological signals is crucial for the diagnosis and monitoring of various diseases. [ 10-13 ] Currently, epidermal signals with various modalities (e.g., electrical, mechanical, and thermal signals) are recorded separately using multiple sensors with different transduction mechanisms. For instance, biopotential monitoring (e.g., ECG, EMG) requires epidermal electrodes, [ 3 , 4 , 14 ] while detecting epidermal biomechanical signals (e.g., FMG, RPW) relies on flexible pressure or strain sensors. [ 7 , 15 , 16 ] For precise disease diagnosis and management, multiple independent sensors with different modalities are attached to human skin for simultaneous data acquisition ( Fig . 1a ). Each sensor demands its own dedicated analog front end ( Fig . 1b ), resulting in complex circuitry, large printed circuit boards, and high power consumption. [ 17-19 ] Additionally, each modality requires separate signal processing methods and computationally intensive algorithms. [ 19 , 20 ] Moreover, different sensors typically need distinct anatomical sites for placement. [ 13 , 17 , 21 , 22 ] The increased skin coverage causes discomfort and a higher risk of skin irritation during long-term wear, compromising user compliance. To address those challenges, we propose a wearable epidermal sensor that can record and fuse health data from multiple modalities into a single cross-modal biosignal (X-Sig). By performing signal fusion directly at sensor level rather than downstream in software, X-Sig unifies biopotential signals (e.g., ECG, EMG) and biomechanical signals (e.g., FMG, RPW) into one composite waveform that preserves the salient features of both modalities. We demonstrate that complex cardiac and muscular activities can be efficiently and accurately interpreted based on a singular output of a single X-Sig sensor. Design concept of X-Sig sensor The X-Sig sensor is enabled by the combination of hierarchical architecture integration ( Fig. 1 c ) and in-sensor signal fusion strategy ( Fig. 1 d ). Hierarchical architecture spatially integrates different sensing elements into a monolithic form factor, allowing for the simultaneous sensing of multiple biosignals at a single position with minimal skin coverage. The X-Sig sensor comprises two essential functional elements: a heterogeneous conductive, adhesive, recyclable and dry (CARD) electrode for biopotential recording and a perforated piezoelectric sensing layer for biomechanical signal detection. The adhesive polyurethane (PU) and elastic PU layers in CARD electrode can penetrate through the openings of the perforated PU conductive layer and the perforated piezoelectric layer, forming a strong bond directly with the skin surface. Meanwhile, the perforated PU conductive layer and perforated piezoelectric sensing layer can then be tightly pressed onto the skin surface. This heterogeneous design enables us to break the tradeoff between electrical conductivity and mechanical adhesiveness of existing dry electrodes [ 4 , 23-25 ] ( Fig. 1 e , Fig. S1, Note S1). The signal fusion strategy via hybridizing complementary signals at the sensor level allows us to continuously monitor and analyze both biopotential and biomechanical characteristics in a composite waveform ( Fig. 1 d ). A single X-Sig sensor on a single anatomical site allows recording of complex dynamic haemodynamics, including heart rate (HR), pulse arrival time (PAT), diastolic blood pressure (DBP) and systolic blood pressure (SBP), which typically relies on multiple sensors ( Fig. 1 f ). Moreover, in a gesture recognition task, the X-Sig sensor enables us to substantially lower the decoding error rate by 7.8-fold and 4.8-fold compared to single-modal electromyography and force myography ( Fig. 1 g ). Heterogeneous CARD electrodes decoupl e the trade-off between electrical and adhesive properties We rationally design and synthesize three types of polyurethane (PU) materials to construct CARD electrode with heterogeneous configuration ( Fig. 2a , 2 b ). Specifically, a highly elastic PU (EPU, Fig. S2-S3, Note S2) is designed as the top backing layer, a highly adhesive PU (APU, Fig. S4-S5, Note S2) is synthesized as the intermediate adhesion layer, and a non-adhesive PU (NPU, Fig. S6-S7, Note S2) is employed to fabricate the bottom perforated conductive layer via incorporating silver nanoflakes (AgNF). The cross-linked structural design and polybutadiene chain segments in EPU function as molecular springs to enhance strength and elasticity (Fig. S8a). The excellent adhesion property of the APU (Fig. S8b) originates from the following two reasons. Firstly, the APU molecular chain contains a large number of polar groups (urethane and urea groups, Fig. 2 b ), which can form intermolecular interactions and hydrogen bonding with human skin. Secondly, the introduction of polypropylene glycol chain segments in APU, which have a low glass transition temperature (below -40 °C) [ 26 ] , gives APU good skin conformability and increased intermolecular interactions per unit area. NPU uses polytetramethylene glycol chain segments and urea groups as soft and hard segments to improve flexibility and strength. The fabrication of CARD electrodes is illustrated inFig. S9. Benefiting from the heterogeneous configuration design and molecular engineering of PU, the CARD electrodes exhibit high conformability (Fig. S10) and robustness on skin surface (Fig. S11). A high adhesion strength of 1.47 N/cm is achieved on human skin under standard 90 o peeling test ( Fig. 2 c , Fig.12a). On wet or dehydrated skin, the adhesion strength slightly decreases to 0.97 and 1.31 N/cm respectively, yet still much higher than that of commercial Ag/AgCl gel electrodes (0.26 N/cm). Despite high skin adhesivity, the CARD electrodes can be easily removed from skin via applying water to dissolve the APU intermediate layer (Fig. S13, Movie S1). This water-triggerable release mechanism reduces the risk of epidermal injury, making the electrodes especially suitable for fragile, sensitive, neonatal, and geriatric skin. Moreover, no skin irritability is observed after wearing the CARD electrodes for 24 h (Fig. S14). The fabricated CARD electrodes with AgNF loading exceeding 30 wt% exhibit substantially lower impedances than that of commercial Ag/AgCl gel electrodes from 0.1 to 100 Hz ( Fig. 2 d ). We also observed that the skin-electrode impedance decreases gradually and tends to saturate after 0.5 hours after placing the CARD electrodes onto skin (Fig. S12b). This can be attributed to gradual adaptation of CARD electrodes to viscoelastic skin. The orthogonal solubility designs of EPU, APU and NPU layers (i.e., insoluble, water-soluble, ethanol-soluble) not only allow the layer-by-layer fabrication but also enable a unique recyclable process ( Fig. 2 e , S15). Raw materials of EPU, NPU, APU and AgNF can be simply recycled via sequential exposure to water and ethanol. Using the recycled materials, we can fabricate a new batch of CARD electrodes with comparable performance to the original ones ( Fig. 2 f-g , S16). The heterogeneous design of CARD electrodes with vertical stacking of perforated layers enables us to address the intrinsic tradeoff in conventional homogeneous electrodes ( Fig. 2 h ). The CARD electrodes exhibit lower electrode-skin impedance and higher skin adhesion strength compared with literature-reported electrodes ( Fig. 2 i , Table S1). High-fidelity biosignals (e.g., ECG, EMG) can be recorded even when the CARD electrodes are subjected to a high mechanical force ( Fig. 2j ). Hierarchical integration and signal fusion of X-Sig sensor To incorporate multiple sensors without increasing the skin coverage, we vertically integrate an ultrathin (28 µm) and perforated piezoelectric sensing layer beneath the CARD electrodes, forming highly adhesive X-Sig sensor ( Fig. 3a ,S17-S19). The perforated design allows the piezoelectric sensing layer to be tightly and conformally pressed upon the skin surface by CARD electrodes ( Fig. 3b , 3c ). Finite-element simulations show that the perforated sensing layers conform tightly to curved surfaces ( Fig. 3d , 3e ). Non-perforated films, however, form gaps at the curvature center ( Fig. 3c , 3f ), resulting in degradation of signal quality and increased motion artifacts. The perforated design of X-Sig sensor allows high skin adhesiveness and conformability without compromising its biomechanical and bioelectrical sensing performance. High-fidelity piezoelectric responses are detected when the sensor is subjected to different mechanical stimulations ( Fig. 3 g , 3h ). For bioelectrical sensing, X-Sig sensor outputs ECG signals with comparable signal quality as hydrogel-based electrodes ( Fig. 3 i , 3j ) .Notably, the recorded ECG signals remain stable even when X-Sig sensor is subjected to different skin deformations, including compression, twisting and stretching of the skin ( Fig. 3 k , Movie S2). Because both the CARD layer and the piezoelectric layer output potential differences, we can connect the two layers to output a fused electrical signal. The X-Sig sensor can reliably capture both biopotential (ECG, EMG) and biomechanical (RPW, FMG) signals with high fidelity ( Fig. 3 l , 3m ). ECG and RPW signals, which both arise from cardiac activity but carry orthogonal information, can be merged into a composite waveform ( Fig. 3 l ). Similarly, EMG and FMG can be fused into a single channel which measures muscle activity ( Fig. 3 m ). With the bandwidth as a single modality, the composite waveform contains richer physiological insights than either modality alone and permits extraction of key electrical and mechanical features. The biocompatibility of X-Sig sensor is evaluated by attaching the sensor to the skin surface for 24 h. After removing the sensor from the skin, no obvious skin redness and irritation are observed ( Fig. 3 n ). The low cytotoxicity of X-Sig sensor is also validated with a cell viability test ( Fig. 3 o , S20). X-Sig sensor for monitoring multiple haemodynamic parameters Continuous tracking of dynamic blood pressure (BP) is crucial in health monitoring. Pulse wave propagation (PWP) monitoring has emerged as a widely used approach for non-invasive dynamic BP assessment ( Fig. 4a ). [ 13 , 27 , 28 ] Nevertheless, PWP monitoring typically relies on signals collected from at least 2 different anatomical sites with discrete sensors, which leads to large skin coverage, complex circuitry design, redundant data transmission, high power consumption, and complicated signal processing. To overcome these limitations, we designed X-Sig sensor for continuous and precise monitoring of complex dynamic haemodynamics (including HR, PAT, SBP and DBP) with a single sensor and single channel. By placing X-Sig sensor to the radial artery of the forearm ( Fig. 4b ,S21), both ECG and RPW signals can be continuously monitored. Instead of processing these two signals individually, we propose a mechanism and methodology for in-sensor fusing of both ECG and RPW signals into a single-signal output ( Fig. 4c ), from which both vital biopotential and biomechanical features can be simultaneously extracted. The in-sensor signal fusion strategy can simplify data acquisition circuitry and lower the amount of data without sacrificing the measurement accuracy. Only one analog front end is needed to record the fused signal ( Fig. 4d , S22). Both ECG features (e.g., R peak) and RPW features (e.g., systolic peak) are well preserved in the synthesized cross-modal signal. Key vital signs including HR, PAT and variance of PAT (VPAT) can be derived quantitatively using a customized algorithm (Fig. S23). Based on PAT, diastolic blood pressure (DBP) and systolic blood pressure (SBP) can be further estimated with a pre-trained supervised learning model ( Fig. 4e ). Finally, we show that diverse dynamic haemodynamic parameters (e.g., HR, PAT, DBP and SBP), which provide vital information for cardiovascular health, can be continuously monitored, processed and displayed on a portable device ( Fig. 4f , Movie S3). The predicted DBP and SBP values by X-Sig sensor are in good consistency with the values from a commercial sphygmomanometer (Fig. S24, Movie S3). Fig. 4g and Fig. 4h show the Bland-Altman plots of predicted SBP and DBP values, with small mean differences of -0.43±6.11 mmHg for SBP and 0.96±3.51 mmHg for DBP. The accuracy reaches the Class A level as per the IEEE standard [ 29 ] and is comparable or higher than that of previously reported sensing methods with multiple separate sensors ( Fig. 4i, 4j , Table S2). To evaluate real-world performance, healthy volunteers wore the X-Sig sensor while changing body postures, including sitting, standing, performing the Valsalva maneuver, and brief exercise ( Fig. 4k ). During this process, dynamic haemodynamic parameters (including HR, PAT, SBP and DBP) are continuously monitored and analyzed. SBP increases and DBP decreases during Valsalva maneuver, and both SBP and DBP increase during standing up ( Fig. 4l ). In addition, after the subjects take a brief exercise, HR, SBP and DBP show a gradual decrease. The results are consistent with the reported literatures, [ 30 , 31 ] indicating the reliable performance of X-Sig sensor for continuously monitoring complex dynamic haemodynamics. X-Sig sensor for accurate gesture recognition Another promising application of the X-Sig sensor is to resolve difficult-to-detect gestures. Muscle activities give rise to both biomechanical (i.e., force myography, FMG) and biopotential (i.e., electromyography, EMG) signals ( Fig. 5a ). Typically, FMG signals are measured via mechanical sensors (e.g., strain sensors and pressure sensors) while EMG signals are recorded with epidermal electrodes. [ 32-34 ] The reliability and accuracy of gesture recognition using EMG or FMG sensors alone are limited without a large training dataset. Here, we leverage X-Sig sensor to address this challenge and demonstrate high accurate gesture recognition with a small training dataset based on single-channel X-Sig. We attached a X-Sig sensor to the forearm to measure the FMG-EMG fused signal from flexor digitorum superficialis ( Fig. 5b , 5c ,S25). The FMG-EMG fused signal was compared with EMG or FMG signals alone when the subject conducted 10 different hand gestures ( Fig. 5d , S26). It can be observed that both FMG and EMG features are well preserved in the fused signal. A convolution neural network (CNN) model, including three convolutional layers, two max-pooling layers and two fully connected layers ( Fig. 5 c , S27), is employed to analyze these hand gestures. Instead of using large datasets to train the CNN model, we only use 70 sets of data from each hand gesture for training. The prediction accuracies of the 10 hand gestures based on FMG signal, EMG signal and the fused signal are 82.9%, 72.1%, and 96.4%, respectively ( Fig. 5e-g ). The substantial higher prediction accuracy of X-Sig sensor is because the fused FMG-EMG signal contains orthogonal electrophysiological and biomechanical information. Collectively, these findings show that our signal-fusion approach robustly deciphers subtle muscle activity even with minimal training data. Conclusion We have developed a cross-modal epidermal sensor via fusing biopotential and biomechanical signals into a reconstructed composite waveform. The hierarchical architecture integration breaks the intrinsic tradeoff between electrical and adhesive properties in existing dry electrodes, while the in-sensor signal fusion strategies allow recording of high-fidelity epidermal signals of different modalities with lower power consumption and higher bandwidth efficiency. A single-channel X-Sig sensor is capable of continuous monitoring of multiple dynamic haemodynamics, including HR, PAT, VPAT, DBP, and SBP with high accuracy. In addition, X-Sig sensor enables high gesture prediction accuracy with a small training dataset and has 4.8 times and 7.8 times lower prediction error compared with FMG and EMG alone. The X-Sig sensor addresses long-standing bandwidth, accuracy, power computation, and size limitations of multi-sensor wearable systems, and opens a scalable path for the acquisition of multimodal data from a single-channel sensor. Experiments and Methods Materials and reagents Silver nanoflakes (AgNF, average diameter of 5 μm) were provided by Bohuas Nano Technology Co. Ltd. (China). Conductive silver ink (CI-1036) was purchased from Engineered Materials Solutions (EMS) Inc. (USA). Polyvinylidene Fluoride (PVDF) piezoelectric films are provided by TE Connectivity Ltd. (Switzerland). Tegaderm transparent film dressing was purchased from 3M Company (USA). Polytetramethylene glycol (PTMG, Mn = 2000 g/mol) was purchased from Mitsubishi Chemical Industries, Ltd. (Japan). Polypropylene glycol (PPG, Mn = 2000 g/mol), dimethylolpropionic acid (DMPA, AR), lysine (98%), isophorone diisocyanate (IPDI, 99.5%), dihydroxyl terminated polybutadiene (PB, Mn = 3000 g/mol), N, N, N', N'-Tetrakis(2-hydroxypropyl)ethylenediamine (THEA, 98%), xylene (AR) and dibutyltin dilaurate (DBTDL, 95%) were obtained from Shanghai Titan Scientific Co. Ltd. (China). Trimethylolpropane polyethylene glycol monomethyl ether (Ymer N120, reagent grade, Mn = 1000 g/mol) was obtained from Perstorp in Sweden. Triethylamine (TEA, AR), ethylenediamine (EDA, AR), N, N-dimethylformamide (DMF, AR), anhydrous ethanol, and sodium hydroxide (NaOH, AR) were purchased from Chengdu Kelong Chemical Reagent Co. Ltd. (China). Before use, the PTMG, PPG and Ymer N120 were vacuum-dried at 110 °C for 2 h. Medical gypsum was obtained from Henan Jindan Environmental Protection New Materials Co. Ltd. (China). Curable urethane rubber (Vytaflex40) was acquired from Shanghai Zhixin Technology Co. Ltd. (China) to prepare the skin replica. Synthesis of EPU, APU and NPU The synthetic route of EPU is presented in Fig. S2 and described below. Firstly, 36 g (0.018 mol) PTMG, 4 g (0.0013 mol) PB, 12.876 g (0.058 mol) IPDI, and 0.026 g (DBTDL) were added into a three-round bottom flask with an overhead stirrer and reacted at 80 °C for 3 h to obtain prepolymer. Secondly, the prepolymer was cooled to 30 °C, and 4.8 g (0.016 mol) THEA, along with a certain amount of xylene, were added, and the mixture was stirred at 30 °C for 10 min. Finally, the mixture was coated on the plate glass with a common doctor blade and cured at 100 °C for 6 h to obtain the EPU film. The synthetic route of APU is presented in Fig. S4 and described below. Firstly, 30 g (0.015 mol) PPG2000, 1.5 g (0.0015 mol) Ymer N120, 0.6 g (0.045 mol) DMPA, and 6.99 g (0.0315 mol) IPDI were added into a three-round bottom flask with an overhead stirrer and reacted at 80 °C for 15 h to obtain the prepolymer. Secondly, the prepolymer was cooled to 40 °C, and the 2.64 g (0.026 mol) TEA was added to neutralize the carboxyl in DMPA. Afterwards, the deionized water was gradually introduced into the prepolymer under stirring to obtain the waterborne prepolymer dispersion. Finally, the dispersion was cooled to 25 °C. An aqueous solution of lysine and sodium hydroxide, prepared by mixing 1.53 g (0.01 mol) lysine and 0.4 g (0.01 mol) sodium hydroxide with 15 g deionized water, was added dropwise into the dispersion under continuous agitation within 30 minutes. The mixture was then stirred for an additional 1 h to complete the chain extension and obtain the APU solution. The synthetic route of NPU is presented in Fig. S6 and described below. The NPU was synthesized using water as the dispersion agent. Firstly, 100 g (0.05 mol) PTMG, 3.5 g (0.0035 mol) Ymer N120, 3.5 g (0.026 mol) DMPA, 26.5 g (0.119 mol) IPDI, and 30 g DMF were added into a three-round bottom flask with an overhead stirrer and reacted at 80 °C for 12 h to obtain prepolymer. Secondly, the prepolymer was cooled to 40 °C, and then 2.64 g (0.026 mol) TEA was added to neutralize the carboxyl in DMPA. Subsequently, deionized water was gradually introduced into the prepolymer while stirring, resulting in a waterborne prepolymer dispersion. Finally, the dispersion was cooled to 25 °C, and an EDA aqueous solution, prepared by dissolving 2.60 g (0.043 mol) of EDA in 30 g deionized water, was added dropwise to the dispersion under continuous agitation within 30 minutes, followed by further stirring for 1 h to complete the chain extension and obtain the NPU emulsion. Fabrication of CARD electrodes Soft EPU films (about 50 μm) were casted, dried, cured and cut into rectangles of 40 mm × 35 mm using a laser cutting machine (Liaocheng Foster Laser Technology Co. Ltd. China) to serve as the elastic backing layer of the CARD electrodes. The EPU film was subjected to plasma treatment for about 2 minutes to improve the surface hydrophilicity. Then, a specific amount of APU solution (19.4 wt% solid content) was drop-cast and evenly spread onto the surface of EPU film, followed by drying in an oven at 60 °C for 30 minutes. After drying, an adhesive APU layer was coated on the elastic EPU backing layer, resulting in a high adhesive APU@EPU film. The AgNF/NPU composite layer was prepared by incorporating specific amounts of AgNF powder into NPU emulsion (29.3 wt% solid content). The weight ratio of AgNF to NPU was controlled in the range of 10 wt% to 40 wt%. The AgNF and NPU emulsion mixture was mechanically stirred for 10 minutes and then sonicated for 15 minutes to well disperse the AgNF in the NPU emulsion. Subsequently, 4.0 g of AgNF-NPU mixture was cast into a polytetrafluoroethylene (PTFE) mold (144 cm 2 in area) and dried in an oven at 60 °C for 2.5 hours. The dried AgNF/NPU composite film (around 150 μm in thickness) was peeled off from the mold. Then, the conductive AgNF/NPU composite layer was cut into specific sizes and perforated with square openings by using a digitally controlled laser cutting process, resulting in a perforated AgNF/NPU conductive layer. Finally, the perforated AgNF/NPU conductive layer was stacked onto the adhesive APU@EPU film, forming eventual CARD electrodes with both high adhesiveness and conductivity. Ultrasoft silver-plated conductive threads (Dongguan Shengxin Special Rope Co. Ltd. China) were used to connect the CARD electrodes with external circuitry. Recycling of CARD electrodes The CARD electrodes can be fully and facilely recycled via an eco-friendly process. Specifically,used CARD electrodes were first immersed into deionized water. Under strong magnetic stirring for 1.5 h, the water-soluble intermediate APU layer can be completely dissolved, resulting in APU aqueous solution and residual EPU film and AgNF/NPU film. The APU aqueous solution can be vacuum filtered and concentrated to a specific concentration for reuse. The EPU film and AgNF/NPU film are first dried in an oven to remove the residual water. The elastic EPU film was cleaned with ethanol and can be directly reused for constructing new batches of CARD electrodes. The AgNF/NPU film was immersed into anhydrous ethanol and exposed to magnetic stirring for 2 h to fully dissolve the ethanol-soluble NPU matrix. After complete dissolution, the NPU ethanol solution with AgNF particulates was set overnight, allowing the AgNF particulates of higher density to settle at the bottom under gravity. The supernatant NPU ethanol solution was filtered to completely remove the impurity and then concentrated to a specific concentration for reuse. The bottom AgNF particulates were also filtered and washed with anhydrous ethanol several times to remove the NPU residual, followed by drying into AgNF powder. The recycled products (including EPU film, APU aqueous solution, and NPU ethanol solution) could be used to refabricate new batches of CARD electrodes using similar procedures as mentioned above. Fabrication of X-Sig sensors The X-Sig sensors were constructed on the basis of the fabricated CARD electrodes mentioned above and perforated piezoelectric sensing layers. The perforated piezoelectric sensing layers were fabricated from thin PVDF piezoelectric sensing films (28 μm). Conductive Ag electrode patterns were printed onto both sides of the piezoelectric film by screen-printing method using conductive silver ink (CI-1036), followed by drying the patterned Ag electrodes at 50 o C for 10 min. Then, soft, thin and adhesive tapes (3M Tegaderm) were applied to both sides of the piezoelectric sensing film to encapsulate the patterned electrodes. Subsequently, the adjacent areas outside the electrode pattern were precisely cut off using a sharp scalpel blade, forming the perforated piezoelectric sensing layer. For X-Sig sensors used for RPW signal detection, a circular 3M double-sided tape (VHB 4905) with a thickness of ≈0.5 mm and diameter of ≈6 mm was placed at the center of the perforated piezoelectric sensing layer, which acts as a stress enhancer and helps to magnify the RPW signal. Finally, the CARD electrode, stress enhancer, and perforated piezoelectric sensing layer were integrated together to form the final adhesive X-Sig sensors. For X-Sig sensors used for FMG signal detection, the perforated piezoelectric sensing layer was directly placed beneath the CARD electrode and no stress enhancer was used. Preparation of skin replica for morphology observation The skin replica was prepared via a reverse molding method (Fig. S28). Specifically, medical gypsum powder and water (3:1 in weight ratio) were mixed evenly to obtain the molding precursor, which was then applied to the wrist of a subject to form the inverted skin structure. After 10 minutes of solidification, the solidified plaster mold was removed from the wrist. Subsequently, a curable polyurethane precursor (Vytaflex40) was cast onto the plaster mold and then degassed in vacuum for 10 minutes, followed by transferring into an oven at 60 °C for 3 hours. Finally, the fully cured polyurethane model with duplicated skin morphology was demolded and used as the skin replica. Characterization Adhesion property measurements were conducted on a microcomputer-controlled universal testing machine (QLW-5E, Xiamen Qunlong Instrument Co., Ltd, China). Rectangular adhesive electrodes were affixed to the specific substrates (human skin or glass) and then peeled off perpendicularly (at 90°) at a speed of 50 mm/min. The adhesion strength was calculated based on the maximum peeling force in the peel curves divided by the width of the samples. Impedance spectra were recorded with an electrochemical workstation (CHI760E, Shanghai CH Instruments Inc, China). During the tests, three electrodes were affixed to the forearm of a subject. Before the tests, the forearm was cleaned first with 75% medical alcohol followed by deionized water to remove the skin surface contaminants. The distance between the centers of adjacent electrodes was set to 5 cm. Commercial Ag/AgCl gel electrodes (2223CN, 3M) were employed for comparison. The frequency range for impedance measurement was set between 0.1 Hz and 100 Hz. The impedance changes over time were measured with a digital LCR meter (Tonghui TH2817B, China). In this measurement, two electrodes were affixed to the forearm of the subject, with a distance of 5 cm between them. Each measurement was repeated at least three times. ECG signals were recorded by placing two electrodes on the right forearm and left forearm respectively and one electrode on the back of the left hand. These electrodes were connected to the ECG signal acquisition module with ultrasoft silver-plated conductive threads. Commercial Ag/AgCl gel electrodes were utilized for comparison. During the comparison experiments, a commercial gel electrode served as the common reference electrode on the back of left hand, while the electrodes on the right forearm and left forearm were the fabricated X-Sig sensors or the commercial gel electrodes. Optical microscopy images were acquired with an optical microscope (Zhiyuan ZY-H5000, China) in reflection mode. FTIR was employed on an infrared spectrometer (Nicolet iS50, America) to study the chemical structure of synthesized PU materials. The molecular weight of the synthesized PU was measured by a gel permeation chromatography (GPC, HLC-8320) with DMF as eluent and polystyrene as standard. The tensile properties were performed at room temperature on a universal testing machine (Instron 4302) with an extension rate of 50 mm/min. 1 H NMR was performed on NMR spectrometer (JNM-ECZ400S/L1) at 400MHz with the CDCl 3 as solvent to confirm the structure of the synthesized PU materials. To evaluate the cytotoxicity of X-Sig sensors, the sensors were first rinsed three times with phosphate-buffered saline (PBS) and then immersed in a serum-free culture medium for 24 hours (1 mg/mL) to prepare the extracting solution of the sensors. L929 cells (Pricella) were cultured in a culture medium composed of 89% minimum essential medium (MEM), 10% fetal bovine serum and 1% penicillin-streptomycin (P/S), followed by incubating overnight (37°C, 5% CO 2 ) to facilitate cell adhesion. After pretreatment with serum-free medium for 6 h, the cells were allocated into a control group and experimental group. The control group was treated with 1000 μL of serum-free medium. The experimental group was treated with 1000 μL of the sensor extracting solution as prepared above. The cells were incubated and cultured for 12 hours and subsequently stained with a diluted calcein-AM/PI double staining kit for 15 minutes. Finally, fluorescence images of live and dead cells were captured using a confocal laser scanning microscope (LSM900, Germany). Mechanical simulation Mechanical simulation of the X-Sig sensors and counterparts was conducted using ANSYS Workbench (2024R1). The original total length of the simulated sample is 15 mm. The horizontal width of the slightly curved sample is 14.4 mm. The segments of the perforated conductive layer are set as: width of 1.5 mm, thickness of 0.15 mm, density 6713 kg/m 3 , Young’s modulus 0.3 MPa, and Poisson’s ratio 0.37. The intact conductive layer for comparison is set as: width of 6 mm, thickness of 0.15 mm, density 6713 kg/m 3 , Young’s modulus 0.3 MPa, and Poisson’s ratio 0.37. The APU@EPU layer is regarded as a single layer, with thickness of 0.15 mm, density 1000 kg/m 3 . The mechanical behaviour of the APU@EPU layer is characterized using a third-order Ogden hyperelastic model, with material constants μ 1 =43438, μ 2 =82.7, μ 3 =-698.5, α 1 =1.3, α 2 =5, and α 3 =-2. The simulated skin layer has a thickness of 0.3 mm, density of 1000 kg/m 3 , Young’s modulus of 6 kPa, and Poisson’s ratio of 0.48. The segments of the perforated piezoelectric layer under the perforated conductive layer are set as: width of 2.1 mm, thickness of 0.15 mm, density of 1000 kg/m 3 , Young’s modulus of 0.4 MPa, and Poisson’s ratio of 0.3. The adhesion between the APU@EPU layer and the skin layer was modeled as an applied distributed pressure on the APU@EPU layer for simplification. The simulative pressure distribution is illustrated in Fig. S29. Single-signal analysis of dynamic haemodynamics A commercial electronic sphygmomanometer (Yuyue Medical Equipment & Supply Co. Ltd, China) was used for blood pressure calibration. All human subjects participating in the tests involving the human body agreed to the informed consent for all tests and the photographs included in the manuscript. The study was conducted based on the protocol approved by the Medical Ethics Committee of Sichuan University, with the approval number of K2024014. RPW- and ECG-signal detection and fusion : The X-Sig sensor was attached to the radial artery position of the subject’s right hand to record both RPW signal and ECG signal (Fig. S21). Another two CARD electrodes were affixed to the subject's left chest and left abdomen for ECG signal recording. The piezoelectric sensing layer in the X-Sig sensor has two terminals of signal output. One terminal of the signal outputs was connected with the CARD electrode fixed at the right radial artery position. The other terminal of piezoelectric outputs was connected with the CARD electrode placed on the left chest. The CARD electrode in the X-Sig sensor was connected to the positive terminal of the ECG circuit board. The CARD electrode on the left chest was connected to the negative terminal of the ECG circuit board. The CARD electrode on the left abdomen was connected to the reference terminal of the ECG circuit board. With such connections of the X-Sig sensor, the RPW signal (potential difference output) from the piezoelectric sensing layer is seamlessly fused with the ECG signal (also potential difference output) from the CARD electrodes, resulting in a fused single-signal output that carries both RPW and ECG information. Single- channel signal acquisition and wireless transmission : A customized ECG circuit board was employed to acquire the fused single-signal output. The original fused signal was first passed through a 2 kHz low-pass filter and then amplified for ADC sampling. The sampled signal was then filtered through a low-pass digital filter to 65 Hz. Finally, the signal was transmitted to a portable user interface via Bluetooth (BLE) for subsequent signal processing and display. Single- channel signal processing and feature extraction: The recorded fused signal was first standardized using Z-score normalization. The signal offset was then extracted through a fast median filtering algorithm, and this offset was subtracted from the original signal to conduct the baseline correction. A 50 Hz digital filter was subsequently applied, and the signal was smoothed using a moving window average method with a window size of 7, which could enhance the quality of the fused signal. The pre-processed signal, as mentioned above, was then subjected to a first-order difference operation and moving average processing with a window size of 5. This was followed by a second-order difference operation and moving average processing with window size of 3. Next, the minimum values of the signal were multiplied by 0.6 to set the extraction threshold for identifying a series of points that meet the conditions. Adjacent points were then merged using the method of means, and these combined points were designated as the systolic peak points. Based on the extracted systolic peak points, two adjacent systolic peak points were used as critical points to divide the windows. Each window contains one R-peak, and only a portion of the window data was extracted—specifically, excluding the first 30% and the last 10%. The maximum value within the extracted window data was considered as the R-peak point. At this point, pulse arrival time (PAT), variance of PAT (VPAT), and heart rate (HR) can be calculated from the extracted systolic peak points and R-peak points. The time interval between adjacent systolic and R-peaks was calculated as the PAT. Approximately 60 seconds of data were selected as the signal segment, and multiple PAT values were calculated for this segment. The average of these PAT values was taken as the final PAT value, while the variance represents the VPAT value. Similarly, the time intervals between adjacent systolic peak points or R-peak points were calculated to derive multiple HR values. The average of the derived HR values is taken as the final HR value for the whole data segment. Real-time analysis of dynamic haemodynamics : The original fused signal was processed based on the above-mentioned method to extract HR, PAT and VPAT. These parameters were then used to predict dynamic blood pressure, including diastolic blood pressure (DBP) and systolic blood pressure (SBP), through a supervised machine learning model (e.g., multiple linear regression, MLR). A commercial electronic sphygmomanometer (Yuyue Medical Equipment & Supply Co. Ltd, China) was used for blood pressure calibration. HR, PAT and VPAT were calculated as mentioned above. To improve the reliability of the training data set, outlier removal was performed where the absolute error between the calculated HR and the actual measured HR exceeds 5 bpm (classified as abnormal data). The remaining data was used to train the supervised machine learning model. PAT and VPAT were used as the input parameters for the SBP predict model. PAT, VPAT, and HR were used as the input parameters for the DBP predict model. Once the model was fully trained, dynamic SBP and DBP values could be continuously predicted by inputting the subsequent HR, PAT and VPAT parameters into the trained model. A signal segment of 5 seconds was used to calculate PAT, VPAT and HR. Finally, all of the haemodynamic parameters were displayed using a customized data visualization interface. Single-signal resolving of elusive gestures FMG- and ECG-signal detection and fusion : The X-Sig sensor was attached to the flexor digitorum superficialis on the subject's right forearm to record both EMG and FMG signals (Fig. S25). Another two CARD electrodes were affixed to the skin next to the X-Sig sensor on the forearm and the elbow joint position separately to assist EMG signal recording. The piezoelectric sensing layer in the X-Sig sensor has two terminals of signal output. One terminal of the signal outputs was connected with the CARD electrode fixed at the flexor digitorum superficialis. The other terminal of piezoelectric outputs was connected with the CARD electrode placed next to the X-Sig sensor. The CARD electrode fixed at the flexor digitorum superficialis was connected to the positive terminal of the EMG circuit board. The other CARD electrode on the forearm was connected to the negative terminal of the EMG circuit board. The CARD electrode on the elbow joint position was connected to the reference terminal of the EMG circuit board. With such connections of the X-Sig sensor, the FMG signal from the piezoelectric sensing layer is seamlessly fused with the EMG signal from the CARD electrodes, resulting in a fused single-signal output that carries both FMG and EMG information. Date acquisition for hand motion recognition : The subjects sat still in a quiet environment. 10 hand motions were selected, including gesture number six, fist clenching, 5 kg grip dynamometer, 15 kg grip dynamometer, cylindrical griping, gesture of ‘OK’, index finger extension, wrist flexion, wrist rotation, and 15 kg grip ball pinching. When collecting the signal of a specific hand motion, the subjects repeatedly conducted the hand motion at an interval of 5 s. During this process, standalone FMG signals, standalone EMG signals, and FMG-EMG fused signals were continuously collected based on the aforementioned setups. The collected signals were cut into segments with a window size of 5 s, finally acquiring 70 groups of signal sets for each hand motion. The time series signals were used for training and testing. CNN model for hand motion recognition :We chose a CNN model as the classification method. The CNN models included three convolutional layers, two max-pooling layers, and two fully connected layers (Fig. S27). The data sets are processed and adjusted to the type of 256*192*1 (i.e., image size of 256*192 pixels and one channel). 80% of the date sets were used for training and 20% were used for testing. The model solver was a stochastic gradient descent with momentum optimizer, and the learning rate is 0.0008. The maximum number of epochs was 50. After training above CNN models, hand motion recognitions based on standalone FMG signals, standalone EMG signals, and FMG-EMG fused signals were conducted, and confusion matrices were used to evaluate the recognition accuracy. Declarations Acknowledgements This study was supported by Sichuan Science and Technology Program (2024YFFK0133), National University of Singapore Presidential Young Professorship Award (22-4974-A0003), MTC Programmatic ‘BLISS’ (M24M9b0013), Wellcome Leap’s Dynamic Resilience Program jointly funded by Temasek Trust, MOE AcRF Tier 1 grant (22-5402-A0001-0), A*STAR Manufacturing, Trade and Connectivity (MTC) MedTech Programmatic Seed grant (M24N9b0121), National Natural Science Foundation of China (52203272), Fundamental Research Funds for the Central Universities of China, Medical Interdisciplinary Research Key Project of Sichuan University (2022). Author contributions X.W. conceptualized the work. X.W., Z.L. and Y.L. developed the methodology. J.W., C.Z., L.Z., Y.S. and X.W. performed experiments. J.W., C.Z., L.Z., Y.S., X.Z., Z.Y., Z.L. and X.W. visualized data. Y.H., Z.W. and Z.J. provide supporting resources, invaluable expertise and insights. X.W., Z.L. and Y.L. acquired funding. X.W., Z.L. and Y.L. supervised the project. J.W., C.Z., L.Z., X.W., Z.L. and Y.L. wrote the manuscript. X.W., Z.L. and Y.L. edited the manuscript. Conflict of Interest X.W., C.Z., L.Z. and Y.H. are inventors on a pending patent application (202410442845.X, disclosed in Aug. 2024) pertaining to this research. The other authors declare no conflict of interest. Data Availability Statement All data needed to evaluate the conclusions in the paper are present in the paper and Supplementary Materials. Other data is available from the corresponding authors upon reasonable request. References Li D , et al. Motion-unrestricted dynamic electrocardiogram system utilizing imperceptible electronics. Nat Commun 16 , 3259 (2025). Chen S , et al. Starfish-inspired wearable bioelectronic systems for physiological signal monitoring during motion and real-time heart disease diagnosis. Sci Adv 11 , eadv2406 (2025). Zhang L , et al. Fully organic compliant dry electrodes self-adhesive to skin for long-term motion-robust epidermal biopotential monitoring. Nat Commun 11 , 4683 (2020). Tan P , et al. Solution-processable, soft, self-adhesive, and conductive polymer composites for soft electronics. Nat Commun 13 , 358 (2022). Yang S , et al. Stretchable surface electromyography electrode array patch for tendon location and muscle injury prevention. Nat Commun 14 , 6494 (2023). Xue J , et al. A patterned mechanical–electrical coupled sensing patch for multimodal muscle function evaluation. InfoMat 7 , e12631 (2025). Wang H , et al. High-Performance Hydrogel Sensors Enabled Multimodal and Accurate Human–Machine Interaction System for Active Rehabilitation. Adv Mater 36 , 2309868 (2024). Meng K , et al. Wearable Pressure Sensors for Pulse Wave Monitoring. Adv Mater 34 , e2109357 (2022). Wang J , et al. Wearable multichannel pulse condition monitoring system based on flexible pressure sensor arrays. Microsyst Nanoeng 8 , 16 (2022). Ates HC , et al. End-to-end design of wearable sensors. Nat Rev Mater 7 , 887-907 (2022). Mahato K, Saha T, Ding S, Sandhu SS, Chang A-Y, Wang J. Hybrid multimodal wearable sensors for comprehensive health monitoring. Nat Electron 7 , 735-750 (2024). Zhao C, Park J, Root SE, Bao Z. Skin-inspired soft bioelectronic materials, devices and systems. Nat Rev Bioeng 2 , 671-690 (2024). Chung HU , et al. Binodal, wireless epidermal electronic systems with in-sensor analytics for neonatal intensive care. Science 363 , eaau0780 (2019). Zhao Y , et al. Ultra-conformal skin electrodes with synergistically enhanced conductivity for long-time and low-motion artifact epidermal electrophysiology. Nat Commun 12 , 4880 (2021). Li. S , et al. Monitoring blood pressure and cardiac function without positioning via a deep learning–assisted strain sensor array. Sci Adv 9 , eadh0615 (2023). Min S , et al. Wearable blood pressure sensors for cardiovascular monitoring and machine learning algorithms for blood pressure estimation. Nat Rev Cardiol , (2025). Cho S , et al. Wireless, multimodal sensors for continuous measurement of pressure, temperature, and hydration of patients in wheelchair. npj Flexible Electron 7 , 8 (2023). Lee S , et al. A fabric-based wearable sensor for continuous monitoring of decubitus ulcer of subjects lying on a bed. Sci Rep 13 , 5773 (2023). Yu Y , et al. All-printed soft human-machine interface for robotic physicochemical sensing. Sci Rob 7 , eabn0495 (2022). Xu C , et al. A physicochemical-sensing electronic skin for stress response monitoring. Nat Electron 7 , 168-179 (2024). Chung HU , et al. Skin-interfaced biosensors for advanced wireless physiological monitoring in neonatal and pediatric intensive-care units. Nat Med 26 , 418-429 (2020). Ryu D , et al. Comprehensive pregnancy monitoring with a network of wireless, soft, and flexible sensors in high- and low-resource health settings. PNAS 118 , e2100466118 (2021). Shin JH, Choi JY, June K, Choi H, Kim Ti. Polymeric Conductive Adhesive‐Based Ultrathin Epidermal Electrodes for Long‐Term Monitoring of Electrophysiological Signals. Adv Mater 36 , (2024). Xiao Y , et al. High-Adhesive Flexible Electrodes and Their Manufacture: A Review. Micromachines 12 , (2021). Derakhshankhah H , et al. Conducting polymer-based electrically conductive adhesive materials: design, fabrication, properties, and applications. J Mater Sci: Mater Electron 31 , 10947-10961 (2020). Wang S, Fang Y, He H, Zhang L, Li Ca, Ouyang J. Wearable Stretchable Dry and Self‐Adhesive Strain Sensors with Conformal Contact to Skin for High‐Quality Motion Monitoring. Adv Funct Mater 31 , 2007495 (2020). Ma Y , et al. Relation between blood pressure and pulse wave velocity for human arteries. PNAS 115 , 11144-11149 (2018). Tian G , et al. Hierarchical Piezoelectric Composites for Noninvasive Continuous Cardiovascular Monitoring. Adv Mater 36 , 2313612 (2024). IEEE Standard for Wearable Cuffless Blood Pressure Measuring Devices. IEEE Std 1708-2014 , 1-38 (2014). Franklin D , et al. Synchronized wearables for the detection of haemodynamic states via electrocardiography and multispectral photoplethysmography. Nat Biomed Eng 7 , 1229-1241 (2023). Li S , et al. Monitoring blood pressure and cardiac function without positioning via a deep learning–assisted strain sensor array. Sci Adv 9 , eadh0615 (2023). Moin A , et al. A wearable biosensing system with in-sensor adaptive machine learning for hand gesture recognition. Nat Electron 4 , 54-63 (2020). Luo Y , et al. Learning human–environment interactions using conformal tactile textiles. Nat Electron 4 , 193-201 (2021). Kim KK , et al. A substrate-less nanomesh receptor with meta-learning for rapid hand task recognition. Nat Electron 6 , 64-75 (2022). Additional Declarations There is NO Competing Interest. Supplementary Files MovieS1WaterassistedpeelingofCARDelectrodefromskin.mp4 Movie S1 MovieS2MonitoringofECGsignalunderskindeformations.mp4 Movie S2 MovieS3MonitoringofHRSBPandDBPwithXSigsensor.mp4 Movie S3 SupplementaryMaterials.docx Supplementary Materials Cite Share Download PDF Status: Published Journal Publication published 19 Mar, 2026 Read the published version in Nature Sensors → Version 1 posted Editorial decision: revise 17 Sep, 2025 Review # 1 received at journal 16 Sep, 2025 Review # 3 received at journal 15 Sep, 2025 Review # 2 received at journal 15 Sep, 2025 Reviewer # 3 agreed at journal 02 Sep, 2025 Reviewer # 2 agreed at journal 02 Sep, 2025 Reviewer # 1 agreed at journal 02 Sep, 2025 Reviewers invited by journal 02 Sep, 2025 Editor assigned by journal 28 Aug, 2025 First submitted to journal 05 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7300896","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":509278675,"identity":"8a3c70c8-fa25-408a-9af2-fb14e2191e58","order_by":0,"name":"Yuxin Liu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYBACPgYGxgcQZgKRWtgYGJgNSNbCJkGiFokcs+rCtjoGfvYcA8Yfv4jSkpZ2e2bbYQbJnjcGzLx9RGlJPnabt+0Ag8GNHANmxh6itCS2FfMCHWYP1ML4kzgtyceYeduYGQwkcgwYeH4Qo4XnWbI0z7nDPBJnnhUc5m0gQgswoAw/85TVyfG3J298+OMPEVoYBBLAFA+IOMDYRowW/gPIPKJsGQWjYBSMgpEGAG4fLnXpvzrNAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0003-0623-9402","institution":"National University of Singapore","correspondingAuthor":true,"prefix":"","firstName":"Yuxin","middleName":"","lastName":"Liu","suffix":""},{"id":509278676,"identity":"5fddc8a7-5651-4f07-8e63-383c574084c9","order_by":1,"name":"Xiaodong Wu","email":"","orcid":"","institution":"National University of Singapore","correspondingAuthor":false,"prefix":"","firstName":"Xiaodong","middleName":"","lastName":"Wu","suffix":""},{"id":509278677,"identity":"686cd6e0-fee5-4963-97ab-d365e58e3a0f","order_by":2,"name":"Jinchao Wang","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Jinchao","middleName":"","lastName":"Wang","suffix":""},{"id":509278678,"identity":"d8521cb6-638f-48c9-9383-a4ba6ef44fa2","order_by":3,"name":"Cheng Zhu","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Cheng","middleName":"","lastName":"Zhu","suffix":""},{"id":509278679,"identity":"98568aa6-9b1b-4a18-8f63-d06734a247ff","order_by":4,"name":"Lifei Zheng","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Lifei","middleName":"","lastName":"Zheng","suffix":""},{"id":509278680,"identity":"3ce683a5-6ac0-403b-b9b6-633d4a331692","order_by":5,"name":"Yangyang Song","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Yangyang","middleName":"","lastName":"Song","suffix":""},{"id":509278681,"identity":"7c6c4173-d7dc-408a-8d19-fe7ec8a0b43a","order_by":6,"name":"Xuyi Zhang","email":"","orcid":"","institution":"Royal Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Xuyi","middleName":"","lastName":"Zhang","suffix":""},{"id":509278682,"identity":"ad4efc8d-061d-48e5-aa99-095b0760acaa","order_by":7,"name":"Zhentao Yao","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Zhentao","middleName":"","lastName":"Yao","suffix":""},{"id":509278683,"identity":"c6893d7b-1ac0-4958-9ee2-cf3638c933bc","order_by":8,"name":"Yangyang Han","email":"","orcid":"https://orcid.org/0000-0001-9869-4538","institution":"Disruptive \u0026 Sustainable Technologies for Agricultural Precision, Singapore-MIT Alliance for Research and Technology Centre","correspondingAuthor":false,"prefix":"","firstName":"Yangyang","middleName":"","lastName":"Han","suffix":""},{"id":509278684,"identity":"cc2299b0-21ef-46b4-925e-6ac1e335476a","order_by":9,"name":"Zhuqing Wang","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Zhuqing","middleName":"","lastName":"Wang","suffix":""},{"id":509278685,"identity":"5d6aba8e-bf76-4567-86dc-3eb21051fcdf","order_by":10,"name":"Zhou Jiang","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Zhou","middleName":"","lastName":"Jiang","suffix":""},{"id":509278686,"identity":"31d70c96-4340-488e-831c-92b6d8346a8d","order_by":11,"name":"Zhimeng Liu","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Zhimeng","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2025-08-05 13:01:04","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7300896/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7300896/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s44460-026-00044-0","type":"published","date":"2026-03-19T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":90940298,"identity":"ad4d773a-ffff-44e5-868f-f73695da272d","added_by":"auto","created_at":"2025-09-09 17:57:18","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":266698,"visible":true,"origin":"","legend":"\u003cp\u003eDesign concept and principle of epidermal cross‐modal biosignal (X-Sig) sensor. (\u003cstrong\u003ea-b\u003c/strong\u003e) Schematic illustration showing the detection of skin physiological signals of different modalities (e.g., ECG, EMG, FMG, RPW) with conventional sensors. Discrete sensors need to be attached to multiple skin sites. Multiplexed signals are recorded, transmitted, and processed independently. (\u003cstrong\u003ec\u003c/strong\u003e) Cross-sectional schematic of the X-Sig sensorattached to a curvilinear skin surface. The perforated electrode layer for biopotential sensing and perforated piezo-sensor layer for biomechanical sensing are tightly and conformally pressed onto the curved surface by the adhesive layer that penetrates through the perforated openings. (\u003cstrong\u003ed\u003c/strong\u003e) Schematic illustration showing the hierarchical architecture design and sensor-end signal fusion of X-Sig sensor. Only one placement site is needed and only single-channel signal is recorded to resolve both biopotential and biomechanical data. (\u003cstrong\u003ee\u003c/strong\u003e) Photo of X-Sig sensor attached to skin surface. The inset picture shows the strong adhesion between X-Sig sensor and skin during the peeling process. (\u003cstrong\u003ef\u003c/strong\u003e) Schematic illustration showing continuous monitoring of complex dynamic haemodynamics (including HR, PAT, DBP, SBP) via in-sensor fusion of ECG and RPW. (\u003cstrong\u003eg\u003c/strong\u003e) Schematic illustration showing low error-rate interpretation of diverse gestures based on in-sensor fusion of EMG and FMG.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/3d4d46c2c415c91fbe32f2af.png"},{"id":90940303,"identity":"dcc21e0a-c04e-48b2-957a-c7b75e73f72a","added_by":"auto","created_at":"2025-09-09 17:57:19","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":391700,"visible":true,"origin":"","legend":"\u003cp\u003eCARD electrodes based on heterogeneous configuration design and polyurethane molecular engineering. (\u003cstrong\u003ea\u003c/strong\u003e) Schematic showing the structural layout of heterogeneously configured CARD electrodes (left) as well as the interactions between different PU layers and skin (right). (\u003cstrong\u003eb\u003c/strong\u003e) Chemical structures of the three molecularly engineered PU for constructing CARD electrodes, including EPU (elastic and insoluble), APU (adhesive and water-soluble), and NPU (non-adhesive and ethanol-soluble). (\u003cstrong\u003ec\u003c/strong\u003e) Peel force of CARD electrodes measured at different skin conditions based on standard 90\u003csup\u003eo\u003c/sup\u003e peeling test. (\u003cstrong\u003ed\u003c/strong\u003e) On-skin electrochemical impedance spectra of CARD electrodes with different AgNF loadings and commercial Ag/AgCl electrodes. (\u003cstrong\u003ee\u003c/strong\u003e) Photos illustrating the full recyclability of CARD electrodes. (\u003cstrong\u003ef-g\u003c/strong\u003e) Peel force curves and electrochemical impedance spectra of pristine and recycled CARD electrodes measured on skin. (\u003cstrong\u003eh\u003c/strong\u003e) Comparison in conductivity and adhesivity of homogeneous APU/AgNF electrodes and heterogeneous CARD electrodes. The intrinsic trade-off between conductivity and adhesivity in homogeneous electrodes can be overcome with our heterogeneous configuration design. (\u003cstrong\u003ei\u003c/strong\u003e) Performance comparison of CARD electrodes and previously reported epidermal electrodes. (\u003cstrong\u003ej\u003c/strong\u003e) Photos showing CARD electrodes stably adhered to forearm skin under 100 g weight, during which high-fidelity ECG and EMG signals were recorded.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/279bd81b9b91eec7bf4b598e.png"},{"id":90940558,"identity":"d372f1e8-9545-42bf-aa5a-76c04df09167","added_by":"auto","created_at":"2025-09-09 18:05:19","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":415981,"visible":true,"origin":"","legend":"\u003cp\u003eHierarchical integration and signal fusion of X-Sig sensor. (\u003cstrong\u003ea\u003c/strong\u003e) Schematic illustration showing the cross-sectional view of X-Sig sensor before (upper) and after (lower) attachment to a curvilinear skin surface. (\u003cstrong\u003eb\u003c/strong\u003e) Photos showing the continuous peeling process of X-Sig sensor from the skin surface. A portion of skin is pulled up by X-Sig sensor due to its high adhesiveness. (\u003cstrong\u003ec\u003c/strong\u003e) Cross-sectional microscope images of X-Sig sensor with perforated (upper) and non-perforated (lower) \u0026nbsp;piezo-sensor on a skin replica. (\u003cstrong\u003ed-f\u003c/strong\u003e) Finite element simulations of CARD electrode (left), X-Sig sensor(middle), and non-perforated piezo-sensor (right) adhered to a curved surface. (\u003cstrong\u003eg-h\u003c/strong\u003e) Representative piezoelectric responses of the X-Sig sensor when subjected to mechanical stimulations of different intensities and frequencies. (\u003cstrong\u003ei\u003c/strong\u003e) ECG signals recorded with X-Sig sensor and commercial Ag/AgCl gel electrode. (\u003cstrong\u003ej\u003c/strong\u003e) Representative ECG signals and power density spectra measured with X-Sig sensor and commercial gel electrode. (\u003cstrong\u003ek\u003c/strong\u003e) ECG signals recorded with X-Sig sensor under different skin deformations. (\u003cstrong\u003el\u003c/strong\u003e) Separately recorded RPW signal, ECG signal, and fused RPW-ECG signal with X-Sig sensor. The fused signal carries both RPW and ECG features. (\u003cstrong\u003em\u003c/strong\u003e) Recorded FMG signal, EMG signals and FMG-EMG fused signal with X-Sig sensor. (\u003cstrong\u003en\u003c/strong\u003e) Skin irritation study of X-Sig sensor over 24 h. (\u003cstrong\u003eo\u003c/strong\u003e) Cytotoxicity study of X-Sig sensor.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/c62d900b068d062340790104.png"},{"id":90940301,"identity":"349f0e94-9bc1-4ab8-a459-a87c25a85e2d","added_by":"auto","created_at":"2025-09-09 17:57:19","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":353682,"visible":true,"origin":"","legend":"\u003cp\u003eSingle-signal monitoring of complex haemodynamicparameters with X-Sig sensor. (\u003cstrong\u003ea\u003c/strong\u003e) Schematic illustration showing pulse arrival time (PAT) derived from the ECG and RPW signal. (\u003cstrong\u003eb\u003c/strong\u003e) Photo of a X-Sig sensor attached to the radial artery for detecting both ECG and RPW signals. (\u003cstrong\u003ec\u003c/strong\u003e) Continuously recorded single output of X-Sig sensor by fusing ECG and RPW signals. Both ECG feature (e.g., R-peak) and RPW feature (e.g., systolic peak) can be identified from the fused signal. Heart rate (HR) and PAT can be extracted from the fused signal. (\u003cstrong\u003ed\u003c/strong\u003e) Schematic diagram showing the single-channel X-Sig recording, wireless data transmission, automatic feature extraction and blood pressure prediction with a supervised learning model. Leveraging the single output of X-Sig sensor, both diastolic blood pressure (DBP) and systolic blood pressure (SBP) can be estimated in real time. (\u003cstrong\u003ee\u003c/strong\u003e) Diagram illustrating the blood pressure prediction flow chart based on single output of X-Sig sensor. (\u003cstrong\u003ef\u003c/strong\u003e) Customized software interface for real-time data visualization (e.g., original X-Sig waveform, HR, DBP and SBP). (\u003cstrong\u003eg-h\u003c/strong\u003e) Bland-Altman plots of SBP and DBP values predicted based on single output of X-Sig sensor. (\u003cstrong\u003ei-j\u003c/strong\u003e) Comparison of reliability and accuracy of SBP and DBP prediction with X-Sig sensor and previously reported methods. (\u003cstrong\u003ek) \u003c/strong\u003eFour different body postures with different dynamic haemodynamicparameters. (\u003cstrong\u003el\u003c/strong\u003e) Variations in HR, PAT, SBP and DBP of the subject when changing body postures.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/e28c99567f2a21db194103c1.png"},{"id":90940300,"identity":"a88e7c1f-8dd5-4b13-9379-3d87ab53f10f","added_by":"auto","created_at":"2025-09-09 17:57:19","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":333952,"visible":true,"origin":"","legend":"\u003cp\u003eX-Sig sensor for accurate gesture recognition. (\u003cstrong\u003ea\u003c/strong\u003e) Schematic illustrationshowing recording of FMG and EMG signals during muscle contraction. (\u003cstrong\u003eb\u003c/strong\u003e) Diagram showing the mechanism of detecting both FMG and EMG signals with X-Sig sensor. The inset shows the photo of a X-Sig sensor attached to the forearm for resolving both FMG and EMG signals. CARD electrodes are placed near the X-Sig sensor to assist EMG signal recording. (\u003cstrong\u003ec\u003c/strong\u003e) Illustration depicting the fusion of FMG and EMG signals into a single-signal output, which carries both biomechanical and biopotential features. This fused signal can be analyzed with a convolutional neural network (CNN) model for accurate recognition of gestures. (\u003cstrong\u003ed\u003c/strong\u003e) Recorded FMG signal, EMG signal, and fused FMG-EMG signal when the subject performs 10 different gestures. (\u003cstrong\u003ee-g\u003c/strong\u003e) Gesture recognition confusion matrices from FMG signal, EMG signal, and fused FMG-MEG signal from X-Sig sensor.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/f785c9024d9f9b0feccc0cae.png"},{"id":105040057,"identity":"89655cc9-e38e-40d8-b7c6-028b8f6ef74b","added_by":"auto","created_at":"2026-03-20 07:48:08","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2794537,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/69e8710a-60b1-428f-8a7e-d51c8f609c6c.pdf"},{"id":90940306,"identity":"b0f10bc7-f942-4774-9877-70f516f95f40","added_by":"auto","created_at":"2025-09-09 17:57:19","extension":"mp4","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":6821800,"visible":true,"origin":"","legend":"Movie S1","description":"","filename":"MovieS1WaterassistedpeelingofCARDelectrodefromskin.mp4","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/863584020fa4433a4d8e0491.mp4"},{"id":90940305,"identity":"f9feb9e9-0360-44de-858b-fa1468b07c7d","added_by":"auto","created_at":"2025-09-09 17:57:19","extension":"mp4","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":3577263,"visible":true,"origin":"","legend":"\u003cp\u003eMovie S2\u003c/p\u003e","description":"","filename":"MovieS2MonitoringofECGsignalunderskindeformations.mp4","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/8d577ed24b093c08e1b659f4.mp4"},{"id":90940559,"identity":"34b96066-9fc5-427e-885e-18e192640b56","added_by":"auto","created_at":"2025-09-09 18:05:19","extension":"mp4","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":892845,"visible":true,"origin":"","legend":"\u003cp\u003eMovie S3\u003c/p\u003e","description":"","filename":"MovieS3MonitoringofHRSBPandDBPwithXSigsensor.mp4","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/d4c2d91a9024e3d446b5b36d.mp4"},{"id":90940307,"identity":"6487a37d-5c28-44b5-b408-5943215715c9","added_by":"auto","created_at":"2025-09-09 17:57:19","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":9765410,"visible":true,"origin":"","legend":"Supplementary Materials","description":"","filename":"SupplementaryMaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-7300896/v1/7be57a5806e683509eb39029.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"A cross-modal epidermal sensor enables single-channel fusion of biopotential and biomechanical signals","fulltext":[{"header":"Full Text","content":"\u003cp\u003eHuman skin serves not only as a barrier but also as a window into the body\u0026rsquo;s internal physiology. A wide spectrum of vital physiological signals can be detected on the skin, such as electrocardiography (ECG),\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e1-3\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e electromyography (EMG),\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e4\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e5\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e force myography (FMG),\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e6\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e7\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e radial pulse wave (RPW).\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e8\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e9\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e The continuous and simultaneous acquisition of multiple epidermal physiological signals is crucial for the diagnosis and monitoring of various diseases.\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e10-13\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e Currently, epidermal signals with various modalities (e.g., electrical, mechanical, and thermal signals) are recorded separately using multiple sensors with different transduction mechanisms. For instance, biopotential monitoring (e.g., ECG, EMG) requires epidermal electrodes,\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e3\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e4\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e14\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e while detecting epidermal biomechanical signals (e.g., FMG, RPW) relies on flexible pressure or strain sensors.\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e7\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e15\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e16\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eFor precise disease diagnosis and management, multiple independent sensors with different modalities are attached to human skin for simultaneous data acquisition (\u003cstrong\u003eFig\u003c/strong\u003e\u003cstrong\u003e.\u003c/strong\u003e\u003cstrong\u003e 1a\u003c/strong\u003e). Each sensor demands its own dedicated analog front end (\u003cstrong\u003eFig\u003c/strong\u003e\u003cstrong\u003e. 1b\u003c/strong\u003e), resulting in complex circuitry, large printed circuit boards, and high power consumption.\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e17-19\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e Additionally, each modality requires separate signal processing methods and computationally intensive algorithms.\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e19\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e20\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e Moreover, different sensors typically need distinct anatomical sites for placement.\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e13\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e17\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e21\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e22\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e The increased skin coverage causes discomfort and a higher risk of skin irritation during long-term wear, compromising user compliance.\u003c/p\u003e\n\u003cp\u003eTo address those challenges, we propose a wearable epidermal sensor that can record and fuse health data from multiple modalities into a single cross-modal biosignal (X-Sig). By performing signal fusion directly at sensor level rather than downstream in software, X-Sig unifies biopotential signals (e.g., ECG, EMG) and biomechanical signals (e.g., FMG, RPW) into one composite waveform that preserves the salient features of both modalities. We demonstrate that complex cardiac and muscular activities can be efficiently and accurately interpreted based on a singular output of a single X-Sig sensor.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDesign concept of X-Sig sensor\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe X-Sig sensor is enabled by the combination of hierarchical architecture integration (\u003cstrong\u003eFig. 1\u003c/strong\u003e\u003cstrong\u003ec\u003c/strong\u003e) and in-sensor signal fusion strategy (\u003cstrong\u003eFig. 1\u003c/strong\u003e\u003cstrong\u003ed\u003c/strong\u003e). Hierarchical architecture spatially integrates different sensing elements into a monolithic form factor, allowing for the simultaneous sensing of multiple biosignals at a single position with minimal skin coverage. The X-Sig sensor comprises two essential functional elements: a heterogeneous conductive, adhesive, recyclable and dry (CARD) electrode for biopotential recording and a perforated piezoelectric sensing layer for biomechanical signal detection. The adhesive polyurethane (PU) and elastic PU layers in CARD electrode can penetrate through the openings of the perforated PU conductive layer and the perforated piezoelectric layer, forming a strong bond directly with the skin surface. Meanwhile, the perforated PU conductive layer and perforated piezoelectric sensing layer can then be tightly pressed onto the skin surface. This heterogeneous design enables us to break the tradeoff between electrical conductivity and mechanical adhesiveness of existing dry electrodes\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e4\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e23-25\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e (\u003cstrong\u003eFig. 1\u003c/strong\u003e\u003cstrong\u003ee\u003c/strong\u003e, Fig. S1, Note S1).\u003c/p\u003e\n\u003cp\u003eThe signal fusion strategy via hybridizing complementary signals at the sensor level allows us to continuously monitor and analyze both biopotential and biomechanical characteristics in a composite waveform (\u003cstrong\u003eFig. 1\u003c/strong\u003e\u003cstrong\u003ed\u003c/strong\u003e). A single X-Sig sensor on a single anatomical site allows recording of complex dynamic haemodynamics, including heart rate (HR), pulse arrival time (PAT), diastolic blood pressure (DBP) and systolic blood pressure (SBP), which typically relies on multiple sensors (\u003cstrong\u003eFig. 1\u003c/strong\u003e\u003cstrong\u003ef\u003c/strong\u003e). Moreover, in a gesture recognition task, the X-Sig sensor enables us to substantially lower the decoding error rate by 7.8-fold and 4.8-fold compared to single-modal electromyography and force myography (\u003cstrong\u003eFig. 1\u003c/strong\u003e\u003cstrong\u003eg\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHeterogeneous CARD electrodes decoupl\u003c/strong\u003e\u003cstrong\u003ee\u003c/strong\u003e\u003cstrong\u003ethe trade-off between \u003c/strong\u003e\u003cstrong\u003eelectrical and adhesive properties\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe rationally design and synthesize three types of polyurethane (PU) materials to construct CARD electrode with heterogeneous configuration (\u003cstrong\u003eFig.\u003c/strong\u003e \u003cstrong\u003e2a\u003c/strong\u003e, \u003cstrong\u003e2\u003c/strong\u003e\u003cstrong\u003eb\u003c/strong\u003e). Specifically, a highly elastic PU (EPU, Fig. S2-S3, Note S2) is designed as the top backing layer, a highly adhesive PU (APU, Fig. S4-S5, Note S2) is synthesized as the intermediate adhesion layer, and a non-adhesive PU (NPU, Fig. S6-S7, Note S2) is employed to fabricate the bottom perforated conductive layer via incorporating silver nanoflakes (AgNF). The cross-linked structural design and polybutadiene chain segments in EPU function as molecular springs to enhance strength and elasticity (Fig. S8a). The excellent adhesion property of the APU (Fig. S8b) originates from the following two reasons. Firstly, the APU molecular chain contains a large number of polar groups (urethane and urea groups, \u003cstrong\u003eFig.\u003c/strong\u003e \u003cstrong\u003e2\u003c/strong\u003e\u003cstrong\u003eb\u003c/strong\u003e), which can form intermolecular interactions and hydrogen bonding with human skin. Secondly, the introduction of polypropylene glycol chain segments in APU, which have a low glass transition temperature (below -40 \u0026deg;C)\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e26\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e, gives APU good skin conformability and increased intermolecular interactions per unit area. NPU uses polytetramethylene glycol chain segments and urea groups as soft and hard segments to improve flexibility and strength. The fabrication of CARD electrodes is illustrated inFig. S9.\u003c/p\u003e\n\u003cp\u003eBenefiting from the heterogeneous configuration design and molecular engineering of PU, the CARD electrodes exhibit high conformability (Fig. S10) and robustness on skin surface (Fig. S11). A high adhesion strength of 1.47 N/cm is achieved on human skin under standard 90\u003csup\u003eo\u003c/sup\u003e peeling test (\u003cstrong\u003eFig. 2\u003c/strong\u003e\u003cstrong\u003ec\u003c/strong\u003e, Fig.12a). On wet or dehydrated skin, the adhesion strength slightly decreases to 0.97 and 1.31 N/cm respectively, yet still much higher than that of commercial Ag/AgCl gel electrodes (0.26 N/cm). Despite high skin adhesivity, the CARD electrodes can be easily removed from skin via applying water to dissolve the APU intermediate layer (Fig. S13, Movie S1). This water-triggerable release mechanism reduces the risk of epidermal injury, making the electrodes especially suitable for fragile, sensitive, neonatal, and geriatric skin. Moreover, no skin irritability is observed after wearing the CARD electrodes for 24 h (Fig. S14).\u003c/p\u003e\n\u003cp\u003eThe fabricated CARD electrodes with AgNF loading exceeding 30 wt% exhibit substantially lower impedances than that of commercial Ag/AgCl gel electrodes from 0.1 to 100 Hz (\u003cstrong\u003eFig. 2\u003c/strong\u003e\u003cstrong\u003ed\u003c/strong\u003e). We also observed that the skin-electrode impedance decreases gradually and tends to saturate after 0.5 hours after placing the CARD electrodes onto skin (Fig. S12b). This can be attributed to gradual adaptation of CARD electrodes to viscoelastic skin. \u003c/p\u003e\n\u003cp\u003eThe orthogonal solubility designs of EPU, APU and NPU layers (i.e., insoluble, water-soluble, ethanol-soluble) not only allow the layer-by-layer fabrication but also enable a unique recyclable process (\u003cstrong\u003eFig.\u003c/strong\u003e \u003cstrong\u003e2\u003c/strong\u003e\u003cstrong\u003ee\u003c/strong\u003e, S15). Raw materials of EPU, NPU, APU and AgNF can be simply recycled via sequential exposure to water and ethanol. Using the recycled materials, we can fabricate a new batch of CARD electrodes with comparable performance to the original ones (\u003cstrong\u003eFig.\u003c/strong\u003e \u003cstrong\u003e2\u003c/strong\u003e\u003cstrong\u003ef-g\u003c/strong\u003e, S16).\u003c/p\u003e\n\u003cp\u003eThe heterogeneous design of CARD electrodes with vertical stacking of perforated layers enables us to address the intrinsic tradeoff in conventional homogeneous electrodes (\u003cstrong\u003eFig. 2\u003c/strong\u003e\u003cstrong\u003eh\u003c/strong\u003e). The CARD electrodes exhibit lower electrode-skin impedance and higher skin adhesion strength compared with literature-reported electrodes (\u003cstrong\u003eFig. 2\u003c/strong\u003e\u003cstrong\u003ei\u003c/strong\u003e, Table S1). High-fidelity biosignals (e.g., ECG, EMG) can be recorded even when the CARD electrodes are subjected to a high mechanical force (\u003cstrong\u003eFig. 2j\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHierarchical integration and signal fusion of X-Sig sensor\u003c/strong\u003e\u003c/p\u003e\n\n\u003cp\u003eTo incorporate multiple sensors without increasing the skin coverage, we vertically integrate an ultrathin (28 \u0026micro;m) and perforated piezoelectric sensing layer beneath the CARD electrodes, forming highly adhesive X-Sig sensor (\u003cstrong\u003eFig. 3a\u003c/strong\u003e,S17-S19). The perforated design allows the piezoelectric sensing layer to be tightly and conformally pressed upon the skin surface by CARD electrodes (\u003cstrong\u003eFig. 3b\u003c/strong\u003e, \u003cstrong\u003e3c\u003c/strong\u003e). Finite-element simulations show that the perforated sensing layers conform tightly to curved surfaces (\u003cstrong\u003eFig. 3d\u003c/strong\u003e, \u003cstrong\u003e3e\u003c/strong\u003e). Non-perforated films, however, form gaps at the curvature center (\u003cstrong\u003eFig. 3c\u003c/strong\u003e, \u003cstrong\u003e3f\u003c/strong\u003e), resulting in degradation of signal quality and increased motion artifacts. \u003c/p\u003e\n\u003cp\u003eThe perforated design of X-Sig sensor allows high skin adhesiveness and conformability without compromising its biomechanical and bioelectrical sensing performance. High-fidelity piezoelectric responses are detected when the sensor is subjected to different mechanical stimulations (\u003cstrong\u003eFig. 3\u003c/strong\u003e\u003cstrong\u003eg\u003c/strong\u003e, \u003cstrong\u003e3h\u003c/strong\u003e). For bioelectrical sensing, X-Sig sensor outputs ECG signals with comparable signal quality as hydrogel-based electrodes (\u003cstrong\u003eFig. 3\u003c/strong\u003e\u003cstrong\u003ei\u003c/strong\u003e\u003cstrong\u003e, \u003c/strong\u003e\u003cstrong\u003e3j\u003c/strong\u003e\u003cstrong\u003e)\u003c/strong\u003e.Notably, the recorded ECG signals remain stable even when X-Sig sensor is subjected to different skin deformations, including compression, twisting and stretching of the skin (\u003cstrong\u003eFig. 3\u003c/strong\u003e\u003cstrong\u003ek\u003c/strong\u003e, Movie S2).\u003c/p\u003e\n\u003cp\u003eBecause both the CARD layer and the piezoelectric layer output potential differences, we can connect the two layers to output a fused electrical signal. The X-Sig sensor can reliably capture both biopotential (ECG, EMG) and biomechanical (RPW, FMG) signals with high fidelity (\u003cstrong\u003eFig. 3\u003c/strong\u003e\u003cstrong\u003el\u003c/strong\u003e,\u003cstrong\u003e 3m\u003c/strong\u003e). ECG and RPW signals, which both arise from cardiac activity but carry orthogonal information, can be merged into a composite waveform (\u003cstrong\u003eFig. 3\u003c/strong\u003e\u003cstrong\u003el\u003c/strong\u003e). Similarly, EMG and FMG can be fused into a single channel which measures muscle activity (\u003cstrong\u003eFig. 3\u003c/strong\u003e\u003cstrong\u003em\u003c/strong\u003e). With the bandwidth as a single modality, the composite waveform contains richer physiological insights than either modality alone and permits extraction of key electrical and mechanical features. \u003c/p\u003e\n\u003cp\u003eThe biocompatibility of X-Sig sensor is evaluated by attaching the sensor to the skin surface for 24 h. After removing the sensor from the skin, no obvious skin redness and irritation are observed (\u003cstrong\u003eFig. 3\u003c/strong\u003e\u003cstrong\u003en\u003c/strong\u003e). The low cytotoxicity of X-Sig sensor is also validated with a cell viability test (\u003cstrong\u003eFig. 3\u003c/strong\u003e\u003cstrong\u003eo\u003c/strong\u003e, S20). \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eX-Sig \u003c/strong\u003e\u003cstrong\u003esensor \u003c/strong\u003e\u003cstrong\u003efor monitoring multiple haemodynamic parameters \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eContinuous tracking of dynamic blood pressure (BP) is crucial in health monitoring. Pulse wave propagation (PWP) monitoring has emerged as a widely used approach for non-invasive dynamic BP assessment (\u003cstrong\u003eFig. 4a\u003c/strong\u003e).\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e13\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e27\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e28\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e Nevertheless, PWP monitoring typically relies on signals collected from at least 2 different anatomical sites with discrete sensors, which leads to large skin coverage, complex circuitry design, redundant data transmission, high power consumption, and complicated signal processing. To overcome these limitations, we designed X-Sig sensor for continuous and precise monitoring of complex dynamic haemodynamics (including HR, PAT, SBP and DBP) with a single sensor and single channel. \u003c/p\u003e\n\u003cp\u003eBy placing X-Sig sensor to the radial artery of the forearm (\u003cstrong\u003eFig. 4b\u003c/strong\u003e,S21), both ECG and RPW signals can be continuously monitored. Instead of processing these two signals individually, we propose a mechanism and methodology for in-sensor fusing of both ECG and RPW signals into a single-signal output (\u003cstrong\u003eFig. 4c\u003c/strong\u003e), from which both vital biopotential and biomechanical features can be simultaneously extracted. \u003c/p\u003e\n\u003cp\u003eThe in-sensor signal fusion strategy can simplify data acquisition circuitry and lower the amount of data without sacrificing the measurement accuracy. Only one analog front end is needed to record the fused signal (\u003cstrong\u003eFig. 4d\u003c/strong\u003e, S22). Both ECG features (e.g., R peak) and RPW features (e.g., systolic peak) are well preserved in the synthesized cross-modal signal. Key vital signs including HR, PAT and variance of PAT (VPAT) can be derived quantitatively using a customized algorithm (Fig. S23). Based on PAT, diastolic blood pressure (DBP) and systolic blood pressure (SBP) can be further estimated with a pre-trained supervised learning model (\u003cstrong\u003eFig. 4e\u003c/strong\u003e). Finally, we show that diverse dynamic haemodynamic parameters (e.g., HR, PAT, DBP and SBP), which provide vital information for cardiovascular health, can be continuously monitored, processed and displayed on a portable device (\u003cstrong\u003eFig. 4f\u003c/strong\u003e, Movie S3). \u003c/p\u003e\n\u003cp\u003eThe predicted DBP and SBP values by X-Sig sensor are in good consistency with the values from a commercial sphygmomanometer (Fig. S24, Movie S3). \u003cstrong\u003eFig. 4g\u003c/strong\u003eand \u003cstrong\u003eFig. 4h \u003c/strong\u003eshow the Bland-Altman plots of predicted SBP and DBP values, with small mean differences of -0.43\u0026plusmn;6.11 mmHg for SBP and 0.96\u0026plusmn;3.51 mmHg for DBP. The accuracy reaches the Class A level as per the IEEE standard\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e29\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e and is comparable or higher than that of previously reported sensing methods with multiple separate sensors (\u003cstrong\u003eFig. 4i, 4j\u003c/strong\u003e, Table S2).\u003c/p\u003e\n\u003cp\u003eTo evaluate real-world performance, healthy volunteers wore the X-Sig sensor while changing body postures, including sitting, standing, performing the Valsalva maneuver, and brief exercise (\u003cstrong\u003eFig. 4k\u003c/strong\u003e). During this process, dynamic haemodynamic parameters (including HR, PAT, SBP and DBP) are continuously monitored and analyzed. SBP increases and DBP decreases during Valsalva maneuver, and both SBP and DBP increase during standing up (\u003cstrong\u003eFig. 4l\u003c/strong\u003e). In addition, after the subjects take a brief exercise, HR, SBP and DBP show a gradual decrease. The results are consistent with the reported literatures,\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e30\u003c/sup\u003e\u003csup\u003e, \u003c/sup\u003e\u003csup\u003e31\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e indicating the reliable performance of X-Sig sensor for continuously monitoring complex dynamic haemodynamics. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eX-Sig sensor for accurate gesture recognition \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAnother promising application of the X-Sig sensor is to resolve difficult-to-detect gestures. Muscle activities give rise to both biomechanical (i.e., force myography, FMG) and biopotential (i.e., electromyography, EMG) signals (\u003cstrong\u003eFig. 5a\u003c/strong\u003e). Typically, FMG signals are measured via mechanical sensors (e.g., strain sensors and pressure sensors) while EMG signals are recorded with epidermal electrodes.\u003csup\u003e[\u003c/sup\u003e\u003csup\u003e32-34\u003c/sup\u003e\u003csup\u003e]\u003c/sup\u003e The reliability and accuracy of gesture recognition using EMG or FMG sensors alone are limited without a large training dataset. Here, we leverage X-Sig sensor to address this challenge and demonstrate high accurate gesture recognition with a small training dataset based on single-channel X-Sig.\u003c/p\u003e\n\u003cp\u003eWe attached a X-Sig sensor to the forearm to measure the FMG-EMG fused signal from flexor digitorum superficialis (\u003cstrong\u003eFig. 5b\u003c/strong\u003e,\u003cstrong\u003e 5c\u003c/strong\u003e,S25). The FMG-EMG fused signal was compared with EMG or FMG signals alone when the subject conducted 10 different hand gestures (\u003cstrong\u003eFig. 5d\u003c/strong\u003e, S26). It can be observed that both FMG and EMG features are well preserved in the fused signal. A convolution neural network (CNN) model, including three convolutional layers, two max-pooling layers and two fully connected layers (\u003cstrong\u003eFig. 5\u003c/strong\u003e\u003cstrong\u003ec\u003c/strong\u003e, S27), is employed to analyze these hand gestures. Instead of using large datasets to train the CNN model, we only use 70 sets of data from each hand gesture for training. The prediction accuracies of the 10 hand gestures based on FMG signal, EMG signal and the fused signal are 82.9%, 72.1%, and 96.4%, respectively (\u003cstrong\u003eFig. 5e-g\u003c/strong\u003e). The substantial higher prediction accuracy of X-Sig sensor is because the fused FMG-EMG signal contains orthogonal electrophysiological and biomechanical information. Collectively, these findings show that our signal-fusion approach robustly deciphers subtle muscle activity even with minimal training data.\u003c/p\u003e"},{"header":"Conclusion ","content":"\u003cp\u003eWe have developed a cross-modal epidermal sensor via fusing biopotential and biomechanical signals into a reconstructed composite waveform. The hierarchical architecture integration breaks the intrinsic tradeoff between electrical and adhesive properties in existing dry electrodes, while the in-sensor signal fusion strategies allow recording of high-fidelity epidermal signals of different modalities with lower power consumption and higher bandwidth efficiency. A single-channel X-Sig sensor is capable of continuous monitoring of multiple dynamic haemodynamics, including HR, PAT, VPAT, DBP, and SBP with high accuracy. In addition, X-Sig sensor enables high gesture prediction accuracy with a small training dataset and has 4.8 times and 7.8 times lower prediction error compared with FMG and EMG alone. The X-Sig sensor addresses long-standing bandwidth, accuracy, power computation, and size limitations of multi-sensor wearable systems, and opens a scalable path for the acquisition of multimodal data from a single-channel sensor. \u0026nbsp;\u0026nbsp;\u003c/p\u003e"},{"header":"Experiments and Methods","content":"\u003cp\u003e\u003cstrong\u003eMaterials and reagents\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSilver nanoflakes (AgNF, average diameter of 5 μm) were provided by Bohuas Nano Technology Co. Ltd. (China). Conductive silver ink (CI-1036) was purchased from Engineered Materials Solutions (EMS) Inc. (USA). Polyvinylidene Fluoride (PVDF) piezoelectric films are provided by TE Connectivity Ltd. (Switzerland). Tegaderm transparent film dressing was purchased from 3M Company (USA). Polytetramethylene glycol (PTMG, Mn = 2000 g/mol) was purchased from Mitsubishi Chemical Industries, Ltd. (Japan). Polypropylene glycol (PPG, Mn = 2000 g/mol), dimethylolpropionic acid (DMPA, AR), lysine (98%), isophorone diisocyanate (IPDI, 99.5%), dihydroxyl terminated polybutadiene (PB, Mn = 3000 g/mol), N, N, N', N'-Tetrakis(2-hydroxypropyl)ethylenediamine (THEA, 98%), xylene (AR) and dibutyltin dilaurate (DBTDL, 95%) were obtained from Shanghai Titan Scientific Co. Ltd. (China). Trimethylolpropane polyethylene glycol monomethyl ether (Ymer N120, reagent grade, Mn = 1000 g/mol) was obtained from Perstorp in Sweden. Triethylamine (TEA, AR), ethylenediamine (EDA, AR), N, N-dimethylformamide (DMF, AR), anhydrous ethanol, and sodium hydroxide (NaOH, AR) were purchased from Chengdu Kelong Chemical Reagent Co. Ltd. (China). Before use, the PTMG, PPG and Ymer N120 were vacuum-dried at 110 °C for 2 h. Medical gypsum was obtained from Henan Jindan Environmental Protection New Materials Co. Ltd. (China). Curable urethane rubber (Vytaflex40) was acquired from Shanghai Zhixin Technology Co. Ltd. (China) to prepare the skin replica.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSynthesis of EPU, APU and NPU\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe synthetic route of EPU is presented in Fig. S2 and described below. Firstly, 36 g (0.018 mol) PTMG, 4 g (0.0013 mol) PB, 12.876 g (0.058 mol) IPDI, and 0.026 g (DBTDL) were added into a three-round bottom flask with an overhead stirrer and reacted at 80 °C for 3 h to obtain prepolymer. Secondly, the prepolymer was cooled to 30 °C, and 4.8 g (0.016 mol) THEA, along with a certain amount of xylene, were added, and the mixture was stirred at 30 °C for 10 min. Finally, the mixture was coated on the plate glass with a common doctor blade and cured at 100 °C for 6 h to obtain the EPU film.\u003c/p\u003e\n\u003cp\u003eThe synthetic route of APU is presented in Fig. S4 and described below. Firstly, 30 g (0.015 mol) PPG2000, 1.5 g (0.0015 mol) Ymer N120, 0.6 g (0.045 mol) DMPA, and 6.99 g (0.0315 mol) IPDI were added into a three-round bottom flask with an overhead stirrer and reacted at 80 °C for 15 h to obtain the prepolymer. Secondly, the prepolymer was cooled to 40 °C, and the 2.64 g (0.026 mol) TEA was added to neutralize the carboxyl in DMPA. Afterwards, the deionized water was gradually introduced into the prepolymer under stirring to obtain the waterborne prepolymer dispersion. Finally, the dispersion was cooled to 25 °C. An aqueous solution of lysine and sodium hydroxide, prepared by mixing 1.53 g (0.01 mol) lysine and 0.4 g (0.01 mol) sodium hydroxide with 15 g deionized water, was added dropwise into the dispersion under continuous agitation within 30 minutes. The mixture was then stirred for an additional 1 h to complete the chain extension and obtain the APU solution.\u003c/p\u003e\n\u003cp\u003eThe synthetic route of NPU is presented in Fig. S6 and described below. The NPU was synthesized using water as the dispersion agent. Firstly, 100 g (0.05 mol) PTMG, 3.5 g (0.0035 mol) Ymer N120, 3.5 g (0.026 mol) DMPA, 26.5 g (0.119 mol) IPDI, and 30 g DMF were added into a three-round bottom flask with an overhead stirrer and reacted at 80 °C for 12 h to obtain prepolymer. Secondly, the prepolymer was cooled to 40 °C, and then 2.64 g (0.026 mol) TEA was added to neutralize the carboxyl in DMPA. Subsequently, deionized water was gradually introduced into the prepolymer while stirring, resulting in a waterborne prepolymer dispersion. Finally, the dispersion was cooled to 25 °C, and an EDA aqueous solution, prepared by dissolving 2.60 g (0.043 mol) of EDA in 30 g deionized water, was added dropwise to the dispersion under continuous agitation within 30 minutes, followed by further stirring for 1 h to complete the chain extension and obtain the NPU emulsion.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFabrication of CARD electrodes\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSoft EPU films (about 50 μm) were casted, dried, cured and cut into rectangles of 40 mm × 35 mm using a laser cutting machine (Liaocheng Foster Laser Technology Co. Ltd. China) to serve as the elastic backing layer of the CARD electrodes. The EPU film was subjected to plasma treatment for about 2 minutes to improve the surface hydrophilicity. Then, a specific amount of APU solution (19.4 wt% solid content) was drop-cast and evenly spread onto the surface of EPU film, followed by drying in an oven at 60 °C for 30 minutes. After drying, an adhesive APU layer was coated on the elastic EPU backing layer, resulting in a high adhesive APU@EPU film.\u003c/p\u003e\n\u003cp\u003eThe AgNF/NPU composite layer was prepared by incorporating specific amounts of AgNF powder into NPU emulsion (29.3 wt% solid content). The weight ratio of AgNF to NPU was controlled in the range of 10 wt% to 40 wt%. The AgNF and NPU emulsion mixture was mechanically stirred for 10 minutes and then sonicated for 15 minutes to well disperse the AgNF in the NPU emulsion. Subsequently, 4.0 g of AgNF-NPU mixture was cast into a polytetrafluoroethylene (PTFE) mold (144 cm\u003csup\u003e2\u0026nbsp;\u003c/sup\u003ein area) and dried in an oven at 60 °C for 2.5 hours. The dried AgNF/NPU composite film (around 150 μm in thickness) was peeled off from the mold. Then, the conductive AgNF/NPU composite layer was cut into specific sizes and perforated with square openings by using a digitally controlled laser cutting process, resulting in a perforated AgNF/NPU conductive layer.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFinally, the perforated AgNF/NPU conductive layer was stacked onto the adhesive APU@EPU film, forming eventual CARD electrodes with both high adhesiveness and conductivity. Ultrasoft silver-plated conductive threads (Dongguan Shengxin Special Rope Co. Ltd. China) were used to connect the CARD electrodes with external circuitry.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRecycling of CARD electrodes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe CARD electrodes can be fully and facilely recycled via an eco-friendly process. Specifically,used CARD electrodes were first immersed into deionized water. Under strong magnetic stirring for 1.5 h, the water-soluble intermediate APU layer can be completely dissolved, resulting in APU aqueous solution and residual EPU film and AgNF/NPU film. The APU aqueous solution can be vacuum filtered and concentrated to a specific concentration for reuse.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe EPU film and AgNF/NPU film are first dried in an oven to remove the residual water. The elastic EPU film was cleaned with ethanol and can be directly reused for constructing new batches of CARD electrodes. The AgNF/NPU film was immersed into anhydrous ethanol and exposed to magnetic stirring for 2 h to fully dissolve the ethanol-soluble NPU matrix. After complete dissolution, the NPU ethanol solution with AgNF particulates was set overnight, allowing the AgNF particulates of higher density to settle at the bottom under gravity. The supernatant NPU ethanol solution was filtered to completely remove the impurity and then concentrated to a specific concentration for reuse. The bottom AgNF particulates were also filtered and washed with anhydrous ethanol several times to remove the NPU residual, followed by drying into AgNF powder. The recycled products (including EPU film, APU aqueous solution, and NPU ethanol solution) could be used to refabricate new batches of CARD electrodes using similar procedures as mentioned above.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFabrication of X-Sig sensors\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe X-Sig sensors were constructed on the basis of the fabricated CARD electrodes mentioned above and perforated piezoelectric sensing layers. The perforated piezoelectric sensing layers were fabricated from thin PVDF piezoelectric sensing films (28 μm). Conductive Ag electrode patterns were printed onto both sides of the piezoelectric film by screen-printing method using conductive silver ink (CI-1036), followed by drying the patterned Ag electrodes at 50\u003csup\u003e\u0026nbsp;o\u003c/sup\u003eC for 10 min. Then, soft, thin and adhesive tapes (3M Tegaderm) were applied to both sides of the piezoelectric sensing film to encapsulate the patterned electrodes. Subsequently, the adjacent areas outside the electrode pattern were precisely cut off using a sharp scalpel blade, forming the perforated piezoelectric sensing layer. For X-Sig sensors used for RPW signal detection, a circular 3M double-sided tape (VHB 4905) with a thickness of ≈0.5 mm and diameter of ≈6 mm was placed at the center of the perforated piezoelectric sensing layer, which acts as a stress enhancer and helps to magnify the RPW signal. Finally, the CARD electrode, stress enhancer, and perforated piezoelectric sensing layer were integrated together to form the final adhesive X-Sig sensors. For X-Sig sensors used for FMG signal detection, the perforated piezoelectric sensing layer was directly placed beneath the CARD electrode and no stress enhancer was used. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePreparation of skin replica for morphology observation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe skin replica was prepared via a reverse molding method (Fig. S28). Specifically, medical gypsum powder and water (3:1 in weight ratio) were mixed evenly to obtain the molding precursor, which was then applied to the wrist of a subject to form the inverted skin structure. After 10 minutes of solidification, the solidified plaster mold was removed from the wrist. Subsequently, a curable polyurethane precursor (Vytaflex40) was cast onto the plaster mold and then degassed in vacuum for 10 minutes, followed by transferring into an oven at 60 °C for 3 hours. Finally, the fully cured polyurethane model with duplicated skin morphology was demolded and used as the skin replica.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCharacterization\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAdhesion property measurements were conducted on a microcomputer-controlled universal testing machine (QLW-5E, Xiamen Qunlong Instrument Co., Ltd, China). Rectangular adhesive electrodes were affixed to the specific substrates (human skin or glass) and then peeled off perpendicularly (at 90°) at a speed of 50 mm/min. The adhesion strength was calculated based on the maximum peeling force in the peel curves divided by the width of the samples.\u003c/p\u003e\n\u003cp\u003eImpedance spectra were recorded with an electrochemical workstation (CHI760E, Shanghai CH Instruments Inc, China). During the tests, three electrodes were affixed to the forearm of a subject. Before the tests, the forearm was cleaned first with 75% medical alcohol followed by deionized water to remove the skin surface contaminants. The distance between the centers of adjacent electrodes was set to 5 cm. Commercial Ag/AgCl gel electrodes (2223CN, 3M) were employed for comparison. The frequency range for impedance measurement was set between 0.1 Hz and 100 Hz. The impedance changes over time were measured with a digital LCR meter (Tonghui TH2817B, China). In this measurement, two electrodes were affixed to the forearm of the subject, with a distance of 5 cm between them. Each measurement was repeated at least three times.\u003c/p\u003e\n\u003cp\u003eECG signals were recorded by placing two electrodes on the right forearm and left forearm respectively and one electrode on the back of the left hand. These electrodes were connected to the ECG signal acquisition module with ultrasoft silver-plated conductive threads. Commercial Ag/AgCl gel electrodes were utilized for comparison. During the comparison experiments, a commercial gel electrode served as the common reference electrode on the back of left hand, while the electrodes on the right forearm and left forearm were the fabricated X-Sig sensors or the commercial gel electrodes.\u003c/p\u003e\n\u003cp\u003eOptical microscopy images were acquired with an optical microscope (Zhiyuan ZY-H5000, China) in reflection mode. FTIR was employed on an infrared spectrometer (Nicolet iS50, America) to study the chemical structure of synthesized PU materials. The molecular weight of the synthesized PU was measured by a gel permeation chromatography (GPC, HLC-8320) with DMF as eluent and polystyrene as standard. The tensile properties were performed at room temperature on a universal testing machine (Instron 4302) with an extension rate of 50 mm/min. \u003csup\u003e1\u003c/sup\u003eH NMR was performed on NMR spectrometer (JNM-ECZ400S/L1) at 400MHz with the CDCl\u003csub\u003e3\u003c/sub\u003e as solvent to confirm the structure of the synthesized PU materials.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo evaluate the cytotoxicity of X-Sig sensors, the sensors were first rinsed three times with phosphate-buffered saline (PBS) and then immersed in a serum-free culture medium for 24 hours (1 mg/mL) to prepare the extracting solution of the sensors. L929 cells (Pricella)\u0026nbsp;were cultured in a culture medium composed of 89% minimum essential medium (MEM), 10% fetal bovine serum and 1% penicillin-streptomycin (P/S), followed by incubating overnight (37°C, 5% CO\u003csub\u003e2\u003c/sub\u003e) to facilitate cell adhesion. After pretreatment with serum-free medium for 6 h, the cells were allocated into a control group and experimental group. The control group was treated with 1000 μL of serum-free medium. The experimental group was treated with 1000 μL of the sensor extracting solution as prepared above. The cells were incubated and cultured for 12 hours and subsequently stained with a diluted calcein-AM/PI double staining kit for 15 minutes. Finally, fluorescence images of live and dead cells were captured using a confocal laser scanning microscope (LSM900, Germany).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMechanical simulation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMechanical simulation of the X-Sig sensors and counterparts was conducted using ANSYS Workbench (2024R1). The original total length of the simulated sample is 15 mm. The horizontal width of the slightly curved sample is 14.4 mm. The segments of the perforated conductive layer are set as: width of 1.5 mm, thickness of 0.15 mm, density 6713 kg/m\u003csup\u003e3\u003c/sup\u003e, Young’s modulus 0.3 MPa, and Poisson’s ratio 0.37. The intact conductive layer for comparison is set as: width of 6 mm, thickness of 0.15 mm, density 6713 kg/m\u003csup\u003e3\u003c/sup\u003e, Young’s modulus 0.3 MPa, and Poisson’s ratio 0.37. The APU@EPU layer is regarded as a single layer, with thickness of 0.15 mm, density 1000 kg/m\u003csup\u003e3\u003c/sup\u003e. The mechanical behaviour of the APU@EPU layer is characterized using a third-order Ogden hyperelastic model, with material constants μ\u003csub\u003e1\u003c/sub\u003e=43438, μ\u003csub\u003e2\u003c/sub\u003e=82.7, μ\u003csub\u003e3\u003c/sub\u003e=-698.5, α\u003csub\u003e1\u003c/sub\u003e=1.3, α\u003csub\u003e2\u003c/sub\u003e=5, and α\u003csub\u003e3\u003c/sub\u003e=-2. The simulated skin layer has a thickness of 0.3 mm, density of 1000 kg/m\u003csup\u003e3\u003c/sup\u003e, Young’s modulus of 6 kPa, and Poisson’s ratio of 0.48. The segments of the perforated piezoelectric layer under the perforated conductive layer are set as: width of 2.1 mm, thickness of 0.15 mm, density of 1000 kg/m\u003csup\u003e3\u003c/sup\u003e, Young’s modulus of 0.4 MPa, and Poisson’s ratio of 0.3. The adhesion between the APU@EPU layer and the skin layer was modeled as an applied distributed pressure on the APU@EPU layer for simplification. The simulative pressure distribution is illustrated in Fig. S29.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSingle-signal analysis of dynamic haemodynamics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA commercial electronic sphygmomanometer (Yuyue Medical Equipment \u0026amp; Supply Co. Ltd, China) was used for blood pressure calibration. All human subjects participating in the tests involving the human body agreed to the informed consent for all tests and the photographs included in the manuscript. The study was conducted based on the protocol approved by the Medical Ethics Committee of Sichuan University, with the approval number of K2024014.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eRPW- and ECG-signal detection and fusion\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e:\u003c/em\u003e The X-Sig sensor was attached to the radial artery position of the subject’s right hand to record both RPW signal and ECG signal (Fig. S21). Another two CARD electrodes were affixed to the subject's left chest and left abdomen for ECG signal recording.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;The piezoelectric sensing layer in the X-Sig sensor has two terminals of signal output. One terminal of the signal outputs was connected with the CARD electrode fixed at the right radial artery position. The other terminal of piezoelectric outputs was connected with the CARD electrode placed on the left chest. The CARD electrode in the X-Sig sensor was connected to the positive terminal of the ECG circuit board. The CARD electrode on the left chest was connected to the negative terminal of the ECG circuit board. The CARD electrode on the left abdomen was connected to the reference terminal of the ECG circuit board. With such connections of the X-Sig sensor, the RPW signal (potential difference output) from the piezoelectric sensing layer is seamlessly fused with the ECG signal (also potential difference output) from the CARD electrodes, resulting in a fused single-signal output that carries both RPW and ECG information.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSingle-\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003echannel\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003esignal acquisition and wireless transmission\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e:\u0026nbsp;\u003c/em\u003eA customized ECG circuit board was employed to acquire the fused single-signal output. The original fused signal was first passed through a 2 kHz low-pass filter and then amplified for ADC sampling. The sampled signal was then filtered through a low-pass digital filter to 65 Hz. Finally, the signal was transmitted to a portable user interface via Bluetooth (BLE) for subsequent signal processing and display.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSingle-\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003echannel\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003esignal processing and feature extraction:\u0026nbsp;\u003c/em\u003e\u003c/strong\u003eThe recorded fused signal was first standardized using Z-score normalization. The signal offset was then extracted through a fast median filtering algorithm, and this offset was subtracted from the original signal to conduct the baseline correction. A 50 Hz digital filter was subsequently applied, and the signal was smoothed using a moving window average method with a window size of 7, which could enhance the quality of the fused signal.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe pre-processed signal, as mentioned above, was then subjected to a first-order difference operation and moving average processing with a window size of 5. This was followed by a second-order difference operation and moving average processing with window size of 3. Next, the minimum values of the signal were multiplied by 0.6 to set the extraction threshold for identifying a series of points that meet the conditions. Adjacent points were then merged using the method of means, and these combined points were designated as the systolic peak points.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBased on the extracted systolic peak points, two adjacent systolic peak points were used as critical points to divide the windows. Each window contains one R-peak, and only a portion of the window data was extracted—specifically, excluding the first 30% and the last 10%. The maximum value within the extracted window data was considered as the R-peak point.\u003c/p\u003e\n\u003cp\u003eAt this point, pulse arrival time (PAT), variance of PAT (VPAT), and heart rate (HR) can be calculated from the extracted systolic peak points and R-peak points. The time interval between adjacent systolic and R-peaks was calculated as the PAT. Approximately 60 seconds of data were selected as the signal segment, and multiple PAT values were calculated for this segment. The average of these PAT values was taken as the final PAT value, while the variance represents the VPAT value. Similarly, the time intervals between adjacent systolic peak points or R-peak points were calculated to derive multiple HR values. The average of the derived HR values is taken as the final HR value for the whole data segment.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eReal-time analysis of dynamic haemodynamics\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e:\u003c/em\u003e The original fused signal was processed based on the above-mentioned method to extract HR, PAT and VPAT. These parameters were then used to predict dynamic blood pressure, including diastolic blood pressure (DBP) and systolic blood pressure (SBP), through a supervised machine learning model (e.g., multiple linear regression, MLR). A commercial electronic sphygmomanometer (Yuyue Medical Equipment \u0026amp; Supply Co. Ltd, China) was used for blood pressure calibration. HR, PAT and VPAT were calculated as mentioned above. To improve the reliability of the training data set, outlier removal was performed where the absolute error between the calculated HR and the actual measured HR exceeds 5 bpm (classified as abnormal data). The remaining data was used to train the supervised machine learning model. PAT and VPAT were used as the input parameters for the SBP predict model. PAT, VPAT, and HR were used as the input parameters for the DBP predict model. Once the model was fully trained, dynamic SBP and DBP values could be continuously predicted by inputting the subsequent HR, PAT and VPAT parameters into the trained model. A signal segment of 5 seconds was used to calculate PAT, VPAT and HR. Finally, all of the haemodynamic parameters were displayed using a customized data visualization interface.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSingle-signal resolving of elusive\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003egestures\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eFMG- and ECG-signal detection and fusion\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e:\u003c/em\u003e The X-Sig sensor was attached to the flexor digitorum superficialis on the subject's right forearm to record both EMG and FMG signals (Fig. S25). Another two CARD electrodes were affixed to the skin next to the X-Sig sensor on the forearm and the elbow joint position separately to assist EMG signal recording.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe piezoelectric sensing layer in the X-Sig sensor has two terminals of signal output. One terminal of the signal outputs was connected with the CARD electrode fixed at the flexor digitorum superficialis. The other terminal of piezoelectric outputs was connected with the CARD electrode placed next to the X-Sig sensor. The CARD electrode fixed at the flexor digitorum superficialis was connected to the positive terminal of the EMG circuit board. The other CARD electrode on the forearm was connected to the negative terminal of the EMG circuit board. The CARD electrode on the elbow joint position was connected to the reference terminal of the EMG circuit board. With such connections of the X-Sig sensor, the FMG signal from the piezoelectric sensing layer is seamlessly fused with the EMG signal from the CARD electrodes, resulting in a fused single-signal output that carries both FMG and EMG information.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eDate acquisition for hand motion recognition\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e:\u003c/em\u003e The subjects sat still in a quiet environment. 10 hand motions were selected, including gesture number six, fist clenching, 5 kg grip dynamometer, 15 kg grip dynamometer, cylindrical griping, gesture of ‘OK’, index finger extension, wrist flexion, wrist rotation, and 15 kg grip ball pinching. When collecting the signal of a specific hand motion, the subjects repeatedly conducted the hand motion at an interval of 5 s. During this process, standalone FMG signals, standalone EMG signals, and FMG-EMG fused signals were continuously collected based on the aforementioned setups. The collected signals were cut into segments with a window size of 5 s, finally acquiring 70 groups of signal sets for each hand motion. The time series signals were used for training and testing.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eCNN model for hand motion recognition\u003c/em\u003e\u003c/strong\u003e:We chose a CNN model as the classification method. The CNN models included three convolutional layers, two max-pooling layers, and two fully connected layers (Fig. S27). The data sets are processed and adjusted to the type of 256*192*1 (i.e., image size of 256*192 pixels and one channel). 80% of the date sets were used for training and 20% were used for testing. The model solver was a stochastic gradient descent with momentum optimizer, and the learning rate is 0.0008. The maximum number of epochs was 50. After training above CNN models, hand motion recognitions based on standalone FMG signals, standalone EMG signals, and FMG-EMG fused signals were conducted, and confusion matrices were used to evaluate the recognition accuracy.\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by Sichuan Science and Technology Program (2024YFFK0133), National University of Singapore Presidential Young Professorship Award (22-4974-A0003), MTC Programmatic ‘BLISS’ (M24M9b0013), Wellcome Leap’s Dynamic Resilience Program jointly funded by Temasek Trust, MOE AcRF Tier 1 grant (22-5402-A0001-0), A*STAR Manufacturing, Trade and Connectivity (MTC) MedTech Programmatic Seed grant (M24N9b0121), National Natural Science Foundation of China (52203272), Fundamental Research Funds for the Central Universities of China, Medical Interdisciplinary Research Key Project of Sichuan University (2022).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eX.W. conceptualized the work. X.W., Z.L. and Y.L. developed the methodology. J.W., C.Z., L.Z., Y.S. and X.W. performed experiments. J.W., C.Z., L.Z., Y.S., X.Z., Z.Y., Z.L. and X.W. visualized data. Y.H.,\u0026nbsp;Z.W. and Z.J. provide supporting resources, invaluable expertise and insights. X.W., Z.L. and Y.L. acquired funding. X.W., Z.L. and Y.L. supervised the project. J.W., C.Z., L.Z., X.W., Z.L. and Y.L. wrote the manuscript. X.W., Z.L. and Y.L. edited the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eX.W., C.Z., L.Z. and Y.H. are inventors on a pending patent application (202410442845.X, disclosed in Aug. 2024) pertaining to this research. The other authors declare no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data needed to evaluate the conclusions in the paper are present in the paper and Supplementary Materials. Other data is available from the corresponding authors upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eLi D\u003cem\u003e, et al.\u003c/em\u003e Motion-unrestricted dynamic electrocardiogram system utilizing imperceptible electronics. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 3259 (2025).\u003c/li\u003e\n \u003cli\u003eChen S\u003cem\u003e, et al.\u003c/em\u003e Starfish-inspired wearable bioelectronic systems for physiological signal monitoring during motion and real-time heart disease diagnosis. \u003cem\u003eSci Adv\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, eadv2406 (2025).\u003c/li\u003e\n \u003cli\u003eZhang L\u003cem\u003e, et al.\u003c/em\u003e Fully organic compliant dry electrodes self-adhesive to skin for long-term motion-robust epidermal biopotential monitoring. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 4683 (2020).\u003c/li\u003e\n \u003cli\u003eTan P\u003cem\u003e, et al.\u003c/em\u003e Solution-processable, soft, self-adhesive, and conductive polymer composites for soft electronics. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 358 (2022).\u003c/li\u003e\n \u003cli\u003eYang S\u003cem\u003e, et al.\u003c/em\u003e Stretchable surface electromyography electrode array patch for tendon location and muscle injury prevention. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 6494 (2023).\u003c/li\u003e\n \u003cli\u003eXue J\u003cem\u003e, et al.\u003c/em\u003e A patterned mechanical\u0026ndash;electrical coupled sensing patch for multimodal muscle function evaluation. \u003cem\u003eInfoMat\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, e12631 (2025).\u003c/li\u003e\n \u003cli\u003eWang H\u003cem\u003e, et al.\u003c/em\u003e High-Performance Hydrogel Sensors Enabled Multimodal and Accurate Human\u0026ndash;Machine Interaction System for Active Rehabilitation. \u003cem\u003eAdv Mater\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 2309868 (2024).\u003c/li\u003e\n \u003cli\u003eMeng K\u003cem\u003e, et al.\u003c/em\u003e Wearable Pressure Sensors for Pulse Wave Monitoring. \u003cem\u003eAdv Mater\u003c/em\u003e \u003cstrong\u003e34\u003c/strong\u003e, e2109357 (2022).\u003c/li\u003e\n \u003cli\u003eWang J\u003cem\u003e, et al.\u003c/em\u003e Wearable multichannel pulse condition monitoring system based on flexible pressure sensor arrays. \u003cem\u003eMicrosyst Nanoeng\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 16 (2022).\u003c/li\u003e\n \u003cli\u003eAtes HC\u003cem\u003e, et al.\u003c/em\u003e End-to-end design of wearable sensors. \u003cem\u003eNat Rev Mater\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 887-907 (2022).\u003c/li\u003e\n \u003cli\u003eMahato K, Saha T, Ding S, Sandhu SS, Chang A-Y, Wang J. Hybrid multimodal wearable sensors for comprehensive health monitoring. \u003cem\u003eNat Electron\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 735-750 (2024).\u003c/li\u003e\n \u003cli\u003eZhao C, Park J, Root SE, Bao Z. Skin-inspired soft bioelectronic materials, devices and systems. \u003cem\u003eNat Rev Bioeng\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 671-690 (2024).\u003c/li\u003e\n \u003cli\u003eChung HU\u003cem\u003e, et al.\u003c/em\u003e Binodal, wireless epidermal electronic systems with in-sensor analytics for neonatal intensive care. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e363\u003c/strong\u003e, eaau0780 (2019).\u003c/li\u003e\n \u003cli\u003eZhao Y\u003cem\u003e, et al.\u003c/em\u003e Ultra-conformal skin electrodes with synergistically enhanced conductivity for long-time and low-motion artifact epidermal electrophysiology. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 4880 (2021).\u003c/li\u003e\n \u003cli\u003eLi. S\u003cem\u003e, et al.\u003c/em\u003e Monitoring blood pressure and cardiac function without positioning via a deep learning\u0026ndash;assisted strain sensor array.\u003cem\u003e\u0026nbsp;Sci Adv\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, eadh0615 (2023).\u003c/li\u003e\n \u003cli\u003eMin S\u003cem\u003e, et al.\u003c/em\u003e Wearable blood pressure sensors for cardiovascular monitoring and machine learning algorithms for blood pressure estimation. \u003cem\u003eNat Rev Cardiol\u003c/em\u003e, (2025).\u003c/li\u003e\n \u003cli\u003eCho S\u003cem\u003e, et al.\u003c/em\u003e Wireless, multimodal sensors for continuous measurement of pressure, temperature, and hydration of patients in wheelchair. \u003cem\u003enpj Flexible Electron\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 8 (2023).\u003c/li\u003e\n \u003cli\u003eLee S\u003cem\u003e, et al.\u003c/em\u003e A fabric-based wearable sensor for continuous monitoring of decubitus ulcer of subjects lying on a bed. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 5773 (2023).\u003c/li\u003e\n \u003cli\u003eYu Y\u003cem\u003e, et al.\u003c/em\u003e All-printed soft human-machine interface for robotic physicochemical sensing. \u003cem\u003eSci Rob\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, eabn0495 (2022).\u003c/li\u003e\n \u003cli\u003eXu C\u003cem\u003e, et al.\u003c/em\u003e A physicochemical-sensing electronic skin for stress response monitoring. \u003cem\u003eNat Electron\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 168-179 (2024).\u003c/li\u003e\n \u003cli\u003eChung HU\u003cem\u003e, et al.\u003c/em\u003e Skin-interfaced biosensors for advanced wireless physiological monitoring in neonatal and pediatric intensive-care units. \u003cem\u003eNat Med\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 418-429 (2020).\u003c/li\u003e\n \u003cli\u003eRyu D\u003cem\u003e, et al.\u003c/em\u003e Comprehensive pregnancy monitoring with a network of wireless, soft, and flexible sensors in high- and low-resource health settings. \u003cem\u003ePNAS\u003c/em\u003e \u003cstrong\u003e118\u003c/strong\u003e, e2100466118 (2021).\u003c/li\u003e\n \u003cli\u003eShin JH, Choi JY, June K, Choi H, Kim Ti. Polymeric Conductive Adhesive‐Based Ultrathin Epidermal Electrodes for Long‐Term Monitoring of Electrophysiological Signals. \u003cem\u003eAdv Mater\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, (2024).\u003c/li\u003e\n \u003cli\u003eXiao Y\u003cem\u003e, et al.\u003c/em\u003e High-Adhesive Flexible Electrodes and Their Manufacture: A Review. \u003cem\u003eMicromachines\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, (2021).\u003c/li\u003e\n \u003cli\u003eDerakhshankhah H\u003cem\u003e, et al.\u003c/em\u003e Conducting polymer-based electrically conductive adhesive materials: design, fabrication, properties, and applications. \u003cem\u003eJ Mater Sci: Mater Electron\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 10947-10961 (2020).\u003c/li\u003e\n \u003cli\u003eWang S, Fang Y, He H, Zhang L, Li Ca, Ouyang J. Wearable Stretchable Dry and Self‐Adhesive Strain Sensors with Conformal Contact to Skin for High‐Quality Motion Monitoring. \u003cem\u003eAdv Funct Mater\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 2007495 (2020).\u003c/li\u003e\n \u003cli\u003eMa Y\u003cem\u003e, et al.\u003c/em\u003e Relation between blood pressure and pulse wave velocity for human arteries. \u003cem\u003ePNAS\u003c/em\u003e \u003cstrong\u003e115\u003c/strong\u003e, 11144-11149 (2018).\u003c/li\u003e\n \u003cli\u003eTian G\u003cem\u003e, et al.\u003c/em\u003e Hierarchical Piezoelectric Composites for Noninvasive Continuous Cardiovascular Monitoring. \u003cem\u003eAdv Mater\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 2313612 (2024).\u003c/li\u003e\n \u003cli\u003eIEEE Standard for Wearable Cuffless Blood Pressure Measuring Devices. \u003cem\u003eIEEE Std 1708-2014\u003c/em\u003e, 1-38 (2014).\u003c/li\u003e\n \u003cli\u003eFranklin D\u003cem\u003e, et al.\u003c/em\u003e Synchronized wearables for the detection of haemodynamic states via electrocardiography and multispectral photoplethysmography. \u003cem\u003eNat Biomed Eng\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 1229-1241 (2023).\u003c/li\u003e\n \u003cli\u003eLi S\u003cem\u003e, et al.\u003c/em\u003e Monitoring blood pressure and cardiac function without positioning via a deep learning\u0026ndash;assisted strain sensor array. \u003cem\u003eSci Adv\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, eadh0615 (2023).\u003c/li\u003e\n \u003cli\u003eMoin A\u003cem\u003e, et al.\u003c/em\u003e A wearable biosensing system with in-sensor adaptive machine learning for hand gesture recognition. \u003cem\u003eNat Electron\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 54-63 (2020).\u003c/li\u003e\n \u003cli\u003eLuo Y\u003cem\u003e, et al.\u003c/em\u003e Learning human\u0026ndash;environment interactions using conformal tactile textiles. \u003cem\u003eNat Electron\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 193-201 (2021).\u003c/li\u003e\n \u003cli\u003eKim KK\u003cem\u003e, et al.\u003c/em\u003e A substrate-less nanomesh receptor with meta-learning for rapid hand task recognition. \u003cem\u003eNat Electron\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 64-75 (2022).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"natsensors","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"natsensors","sideBox":"Learn more about [Nature Sensors](https://www.nature.com/natsensors/)","snPcode":"44460","submissionUrl":"https://mts-natsensors.nature.com","title":"Nature Sensors","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"cross-modal, epidermal sensor, dynamic haemodynamic sensing, wearable electronics ","lastPublishedDoi":"10.21203/rs.3.rs-7300896/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7300896/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDisease monitoring typically requires the acquisition of multiple physiological signals with different modalities. Existing epidermal electronics use separate sensors for each modality, which necessitates a large footprint, high bandwidth and power consumption. We report a wearable electronic system that can fuse physiological signals with multiple modalities into a singlecross-modal biosignal (X-Sig). Leveraging hierarchical device architecture and in-sensor signal fusion strategy, X-Sigsensor concurrently acquires biopotential signals (e.g., electrocardiography and electromyography) and biomechanical signals (e.g., force myography and radial pulse) through a single channel. The single-channel X-Sig sensor is capable of continuous monitoring of multiple dynamic haemodynamics, including heart rate, pulse arrival time, diastolic and systolic blood pressure with high accuracy. In machine-learning-based gesture recognition, the X-Sig sensor reduced the decoding error rate by 7.8-fold compared to conventional electromyography. By fusing complementary modalities at the sensor level, X-Sig sensor provides a versatile platform for designing bandwidth-efficient and low-power wearable electronics.\u003c/p\u003e","manuscriptTitle":"A cross-modal epidermal sensor enables single-channel fusion of biopotential and biomechanical signals","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-09 17:57:14","doi":"10.21203/rs.3.rs-7300896/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"revise","date":"2025-09-17T22:28:29+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"This content is not available.","date":"2025-09-17T03:14:44+00:00","index":1,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2025-09-15T12:59:26+00:00","index":3,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2025-09-15T08:44:04+00:00","index":2,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2025-09-03T00:21:55+00:00","index":3,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2025-09-03T00:09:47+00:00","index":2,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2025-09-02T19:05:23+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewersInvited","content":"","date":"2025-09-02T18:55:22+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-28T15:10:00+00:00","index":"","fulltext":""},{"type":"submitted","content":"Nature Sensors","date":"2025-08-05T12:58:50+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"natsensors","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"natsensors","sideBox":"Learn more about [Nature Sensors](https://www.nature.com/natsensors/)","snPcode":"44460","submissionUrl":"https://mts-natsensors.nature.com","title":"Nature Sensors","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"31cc25a9-a8e0-403f-a365-6863f9422cd0","owner":[],"postedDate":"September 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":54090320,"name":"Physical sciences/Engineering/Biomedical engineering"},{"id":54090321,"name":"Physical sciences/Materials science/Materials for devices/Sensors and biosensors"}],"tags":[],"updatedAt":"2026-03-20T07:44:35+00:00","versionOfRecord":{"articleIdentity":"rs-7300896","link":"https://doi.org/10.1038/s44460-026-00044-0","journal":{"identity":"natsensors","isVorOnly":false,"title":"Nature Sensors"},"publishedOn":"2026-03-19 04:00:00","publishedOnDateReadable":"March 19th, 2026"},"versionCreatedAt":"2025-09-09 17:57:14","video":"","vorDoi":"10.1038/s44460-026-00044-0","vorDoiUrl":"https://doi.org/10.1038/s44460-026-00044-0","workflowStages":[]},"version":"v1","identity":"rs-7300896","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7300896","identity":"rs-7300896","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}