Skin-adaptive focused flexible micromachined ultrasound transducers for wearable cardiovascular health monitoring

Abstract

Continuous monitoring of cardiovascular vital signs can reduce the incidence and mortality of cardiovascular diseases, yet cannot be implemented by current technologies because of device bulkiness and rigidity. Here, we report self-adhesive and skin-conformal ultrasonic transducer arrays that enable wearable monitoring of multiple hemodynamic parameters without interfering with daily activities. A skin-adaptive focused ultrasound method with rational array design is proposed to implement measurement under wide ranges of skin curvatures and depths with improved sensing performances. We introduce a self-adhesive hydrogel layer to enhance the acoustic impedance matching, biocompatibility and contact reliability at device-skin interfaces, and a microelectromechanical systems-compatible process for batch-to-batch and high-consistency manufacturing. The resultant transducer array, with a frequency of 2.4 megacycles per second and a –6 dB bandwidth of 47%, implements accurate and continuous detection of cardiovascular signs, including blood pressure (BP) waveforms of various arteries, heart rate and BP during exercises, and arterial stiffness, validated clinically.

Skin-adaptive focused flexible ultrasonic transducers with MEMS-compatible process enable daily cardiovascular disease assessment.

INTRODUCTION

Cardiovascular diseases (CVDs) have become the leading cause of human death over the world in recent years because of their high incidence rate, disability, and mortality. According to the prediction of the World Health Organization, 31% of global deaths, about 17 million people, result from CVDs each year, including hypertension, ischemic heart disease, congestive heart failure, arrhythmias, and coronary atherosclerotic heart disease (1). The high mortality rate stems from two features as follows: (i) as the core parts of the human body, the failure of heart and cardiac vessels can disable the systemic function of the entire body; (ii) CVDs can cause various fatal complications, such as cerebral infarction, myocardial infarction, cerebral hemorrhage, and stroke, which are highly secretive in their early stage but can lead to death in a short time when they break out (2–4). Hence, there are no effective treatment methods to decrease the disability and mortality clinically to date. However, it is demonstrated that most CVDs can be effectively prevented and controlled in their early stage (5). Therefore, there is an increasing requirement for continued, real-time, and long-term monitoring of cardiovascular vital signs, which is substantially important to achieve accurate and timely diagnosis, thus implementing early prevention and treatment of CVDs and reducing mortality.

Diagnosis of CVDs predominantly relies on three approaches: measurement of hemodynamic parameters (6, 7), electrocardiography monitoring, and nuclear magnetic resonance imaging (8). Among these, hemodynamic parameters contain abundant cardiovascular vital signals, such as blood pressure (BP) waveform, heart rate, and arterial stiffness, which can comprehensively reflect the characteristics of CVDs by real-time, continuous, and long-term monitoring (9, 10). For instance, the long-term continuous measurement of heart rate can distinguish heart failure. The combined monitoring of BP waveform and heart rate can provide a reference for stroke, acute myocardial infarction, and shock (11, 12). Continuous monitoring of the cardiovascular dynamic function, such as continuous BP waveform, can reflect orthostatic hypotension, abnormal hypertension at night and in the morning, etc. (13, 14). However, existing technologies fail to implement real-time, continuous, and long-term monitoring, which disables the early prevention and timely treatment. For instance, auscultatory and oscillometer can detect just the diastolic blood pressure (DBP) and systolic blood pressure (SBP) (14). In addition, these methods need a cuff for air inflation to generate the required pressure to compress arms during the measurement process, which causes discomfort and inconvenience for continuous and long-term monitoring, especially nighttime monitoring during sleep (which contains important information of CVDs) (15). Applanation tonometry methods can detect continuous BP waveform, but the testing effect entirely depends on the experience of doctors (16). Although photoplethysmography shows more possibility in continuous measurement of BP and heart rate because of its cuff-less sensing feature, it also suffers technical drawbacks, such as limited penetration in human tissues; it mainly concentrates on peripheral and venous vessel detection, which caused some deficiency in detecting deep arterial vessels (17), susceptible to environmental conditions, such as light, temperature and relative humidity, and low resolution for lack of directivity of light (14). Compared to the aforementioned methods, ultrasonic technology exhibits immense potential in continuous and long-term cardiovascular dynamic parameter measurement due to its unique features, such as noninvasive, cuff less, sensation less, unsusceptible to ambient conditions, and deep penetration in human tissue. In particular, the deep penetration of ultrasonic technology allows accurate detection of the middle and large arteries buried in deep tissues, the hemodynamic parameters of which have a better correlation with CVDs (18, 19). However, currently available ultrasonic BP probes are not suited for long-term usage on human skin owing to their structural rigidity and bulkiness (20, 21). As the rigid probes cannot seamlessly contact with human skin, the air gaps between the device and skin can cause substantial acoustic energy attenuation and waveform distortion, thus decreasing the measurement accuracy (22, 23). These drawbacks degrade the feasibility and accuracy of current ultrasonic detection technologies for continuous and long-term monitoring of hemodynamic parameters.

In recent years, emerging flexible ultrasonic transducers shine a light on aforementioned issues, which enable seamless integration on human skin because of their lightweight, thin, and flexible structure, thus facilitating convenient wearing and long-term and accurate monitoring (24–28). In view of their appealing merits, a diversity of flexible and stretchable ultrasonic transducers has been invented, in which 1-3 piezocomposite or PZT-5H piezoelectric blocks were used as the transducer cells and serpentine metal electrodes were used to connect all the cells to endow the assembly with flexibility and stretchability (15, 23). Successful application of flexible and stretchable ultrasonic transducers on the detection of central vessel BP (29), cardiac functions (30), elastic modulus of human tissue (31), etc., demonstrated their unprecedented potentiality in physiological parameter monitoring below the human epidermis. These works have pioneered the development of flexible ultrasonic transducers and their application in wearable ultrasound medical monitoring. However, the developed fabrication processes for the transducers largely depended on hand operation. That is, the top and bottom electrodes were manually attached to the piezoelectric cells through conductive silver paste or low-temperature solder, which generally caused large performance deviation and was not suitable for batch and high-consistency manufacturing, hindering the commercial application. Besides, the stable monitoring of daily hemodynamic parameters places demands on the reliable adhesion and excellent acoustic performance of the flexible transducer, posing a challenge to current devices that require additional coupling agents or coupling pads. Previous research had rarely reported a specific investigation on cardiovascular health metric evaluation using flexible ultrasonic transducers through multicardiovascular dynamic parameter monitoring under complex living conditions, as well as deep analysis and comparison of cardiovascular status from a clinical medical view.

Here, we demonstrated skin-adaptive focused flexible ultrasonic transducers with microelectromechanical systems (MEMS)-compatible processes for daily cardiovascular health evaluation based on continuous, real-time, and long-term multicardiovascular vital sign monitoring. A full-array skin-adaptive focused ultrasound (SAFU) method was proposed to improve the sensing performances, such as signal-to-noise ratio (SNR), acoustic energy density, penetration depth, and directionality. A self-adhesive hydrogel layer was introduced to enhance the acoustic impedance matching, biocompatibility, and contact reliability at skin-device interfaces. Rational design of the self-adhesive SAFU transducer array was conducted by establishing the theoretical and simulation models for the spatial sound field, contributing to an SNR of 30.1 dB, tissue penetration depth larger than 30 mm, and a beam width varying from 2.1 to 4.6 mm with a beam depth varying from 3.3 to 53 mm, almost covering the curvatures and depths of various blood vessels of the human body. A MEMS-compatible fabrication process was developed to enable batch-to-batch, high-consistency, and large-area manufacturing, exhibiting a small SD of 0.0236 MHz at 2.4 MHz for the fabricated 6 × 6 transducer array, providing a promising approach for the commercial processing of flexible ultrasonic transducers. The SAFU transducer array can implement a highly accurate assessment of various cardiovascular vital signs, including continuous BP waveform monitoring of various blood vessels, 24-hour BP measurement, comprehensive evaluation of heart rate and BP during exercises, and arterial stiffness of both healthy and hypertensive individuals, validated by clinical medical testing. This study provided a promising technology for noninvasive, real-time, continuous, and long-term monitoring of cardiovascular health.

RESULTS

Working principle and structure design of self-adhesive SAFU transducer array

The working principle for hemodynamic parameter detection predominantly relies on the detection of the constriction and dilation of blood vessels induced by the rhythmic heart contraction using the time of flight (TOF) of ultrasonic waves (Fig. 1A) (15). The measured changes in the blood vessel diameter can be used to derive the hemodynamic parameters, such as the BP waveform, heart rate, and vascular stiffness, etc., according to the theoretical calculation formulas given in note S1. However, different from the phased array and single transducer cell-based detection approaches (15, 23), we proposed a full-array SAFU method. As shown in Fig. 1A, in this method, all transducer cells within the array were simultaneously excited to generate ultrasound waves, which then focused into an ultrasound wave beam by exploiting the curvature of human skin itself. The ultrasound wave beam contained most of the acoustic energy transmitted by the transducer array, contributing to substantial enhancement in the sensing performance, such as high SNR TOF echoes, high acoustic energy density, large penetration depth, excellent directionality, etc. Figure 1C presented the tested acoustic field distributions of our developed ultrasonic transducer array under bending and nonbending states measured by the acoustic testing system (fig. S1). Compared to the unbending state, the transducer array with a curvature radius exhibited a distinct focused ultrasound beam, demonstrating its ultrasound wave focusing capability with structure bending. fig. S2 displayed the simulated SAFU beam of the flexible ultrasonic patch on cylindrical and sphere surfaces with different curvature radii, further validating the feasibility of the SAFU on various curved surfaces. Compared to the single transducer detection approach (15), the full-array SAFU beam featured 10.8 times higher acoustic pressure (Fig. 1D), excellent directionality (Fig. 1E) [the directivity index was 12.1 and 2.97 dB for SAFU array and single cell, respectively (32)], and a more than 30-mm tissue penetration (Fig. 2F), which were enough for almost all the vascular hemodynamic testing of the human body (note S2). Moreover, as all the transmitted ultrasound waves were focused together, the SAFU array had an extremely higher SNR, 19.5 dB higher than that of the single transducer (30.1 dB versus 10.6 dB) (Fig. 1F), which could substantially improve the accuracy of TOF detection. Compared to the phased array approach (23), the SAFU method eliminated the need for the complex phase-control circuit because all the transducer cells were excited using the same electrical signal, which substantially simplified the complexity of the sensing system and improved its stability. Furthermore, because the phase-control system was not required, the electrical connections with individual transducer cells were substantially simplified. Meanwhile, the 0.5 to 1.0 times wavelength pitch requirement in phased array systems was also eliminated (33). These unique features notable degraded the complexity and difficulty of the fabrication process of the transducer array and electrodes, thus reducing the cost. In all, the full-array SAFU method showed great potential for TOF-based ultrasonic health monitoring applications.

Fig. 1. Design and working principle of the flexible ultrasonic transducer array.

(A) Working principle schematic of ultrasonic cardiovascular dynamic parameter measurement. (B) Structure schematic of the piezoelectric transducer cell. (C) Acoustic field of the transducer with different curvature radii. (D) Comparison of the normalized acoustic pressure between the ultrasonic array and a single cell. R10, R25, R40 and R60 mean the radii of curvature of 10, 25, 40 and 60 mm, respectively. (E) Comparison of directivity between the ultrasonic array and a single cell. The, theory; Mea, measurement. (F) Normalized SNR of the fabricated ultrasonic array and single cell under different curvature radii. (G) Beam width and beam depth of the fabricated ultrasonic array and single cell under different curvature radii. (H) Theoretical modeling and analyses for the flexible ultrasonic array: (a) Theoretical model for flexible array design; (b) beam width variations under different cell numbers and curvature radii; (c) beam depth variations under different cell numbers and curvature radii; and (d) acoustic pressure variations under different cell numbers and curvature radii.

Fig. 2. The MEMS process–compatible fabrication and performance characterization of the flexible ultrasonic transducer array.

(A) Main fabrication process. (B) Digital image of the fabricated flexible ultrasonic array. (a) The fabricated flexible ultrasonic transducer patch. (b) The stretched flexible transducer. (c) The bent flexible transducer. (d) The twisted flexible transducer. (C) Resonant frequencies of multichannels. (D) Impedances of three different flexible transducers. (E) Bandwidths. (F) Acoustic pressure variations with meat thickness. (G) Schematic of the ultrasound transducer with or without the hydrogel patch. (H) Comparison of pulse-echo–received voltages with and without a hydrogel patch. (I) Peeling force testing. h, hours. (J) The process of transducer stripping. (K) Attachment to the skin after 24 hours.

The entire structure of the transducer cell was shown in Fig. 1B. The PZT-5H with a piezoelectric coefficient of 593 pC/N was chosen as the piezoelectric material to achieve a high electromechanical coupling coefficient and enough transmitting acoustic energy (34). The electrode was designed with a serpentine structure to maintain both enough electrical conductivity under high frequency and structure stretchability. The combination of the rigid PZT-5H cells and structured electrodes endowed the assembled transducers with both flexibility and great acoustic performance. The polyimide and polydimethylsiloxane (PDMS) were used as the substrate and device encapsulation layer, respectively. Notably, a hydrogel layer with self-adhesive ability was used as the interlayer between the transducer and human skin, which had the following prominent functions: (i) to facilitate convenient, reliable, and tight attachment of the ultrasonic transducer on human skin for long-term and reliable monitoring; (ii) to implement perfect acoustic impedance matching between the device and human skin because of its similar acoustic impedance [1.5 to 1.8 × 10⁶ Pa·s/m (35)] to human tissue [1.58 to 1.70 × 10⁶ Pa·s/m (36)], which substantially increases the acoustic energy transmission, penetration depth, and SNR; and (iii) to endow the device with biocompatibility to avoid immune response–induced skin damage during long-term wearing. This unique design can contribute to the flexible ultrasonic transducer array with high structure flexibility, ultrasonic performance, excellent device-skin coupling capability, and long-term wearability, providing a promising solution to long-term cardiovascular health monitoring.

To achieve optimal performance of the ultrasound transducer array, we conducted a systematic study on the dependences of the measurement accuracy, acoustic performance, and device stretchability on the key parameters of the PZT-5H transducer cell, including the working frequency, cell dimension, electrode dimension, and thickness of the PDMS encapsulation layer. First, the working frequency and structure parameters of the PZT-5H cell and the thickness of the PDMS encapsulation layer were determined by constructing mechanical-electro-acoustic finite element method (FEM) models (figs. S4 and S5). As shown in fig. S4C, the measurement accuracies of the blood vessel diameters increased rapidly with increasing frequency and tended to flatten out when the frequency was more than 2.4 MHz. However, the acoustic pressure suffered substantial attenuation with increasing frequency (fig. S4C). As a trade-off between the measurement accuracy and penetration depth, the working frequency was selected as 2.4 MHz, maintaining a measurement error of less than 3% and an acoustic pressure attenuation rate of less than 40% for typical arteries of the human body (note S3). On the basis of this frequency, the PZT-5H cell with a side length of 0.7 mm and a thickness of 0.18 mm was used by comprehensively considering the transducer volume and structure robustness (note S4). The PDMS layer on the surface of the PZT cell affected the final acoustic transmission power. As shown in fig. S5D, the normalized transmission power only increased by 2.5% when the PDMS thickness decreased from 50 to 30 μm but decreased rapidly by 82% when it increased to 120 μm. This meant that reducing the PDMS thickness below 50 μm would not further substantially improve the acoustic performance. However, the stress under the same strain increased by 20.4% with the PDMS thickness decreasing from 50 to 30 μm, making it more prone to fatigue failure (37). Therefore, 50 μm was selected as the thickness of the PDMS layer to make a trade-off between the mechanical strength and acoustic performance. Furthermore, to ensure more than 60% stretching ability of the ultrasonic array in reality, the key parameters of the serpentine electrode were optimized using a three-dimensional (3D) mechanical FEM model (fig. S6A). As indicated by fig. S6, the stretching rate of the serpentine electrode increased with the decrease in the electrode width while decreasing with the increases in the electrode radius and electrode angle. To achieve a stretchability of more than 60%, we determined the width, radius, and electrode angle of electrodes as 0.10 mm, 0.375 mm, and 45°, respectively, which could completely surpass the tensile range of human skin of 30% (note S5) (15).

To enable the SAFU transducer array to meet the beam width and depth requirements for the detection of various blood vessels in the human body, we established both theoretical and FEM simulation models for the spatial sound field analysis [Fig. 1H (a); see Materials and Methods]. The SAFU beam range should cover the main blood vessels of the human body when the transducer array bends within a specific curvature range of 10 to 60 mm, which was validated by the measured depths and diameters of the fingertip artery, radial artery, brachial artery, and carotid artery of 10 volunteers using the Verasonics Vantage system (Vantage 256, Verasonics), the measured curvature of the skin overlying the blood vessels using the arc-chord method (38), and the results reported previously (note S2) (23, 39–42). Specifically, the beam depth (–6-dB acoustic pressure range in the depth direction) should almost cover the varying vessel area in the depth direction, and the beam width (the average value of –6-dB acoustic pressure range in the transverse direction) should almost coincide with the vessel diameter. In this content, the average acoustic pressure of the SAFU beam within the vessel area should be kept at maximum. For these, the cell pitch, array size, cell number, and bending and stretching states of the SAFU array on its main performance parameters, such as beam width, beam depth, and acoustic pressure, were investigated systematically.

As shown in fig. S7 (A to C), for different array sizes, the beam width and beam depth changed slightly with the cell pitch varying from 0.6 to 1.4 mm, while the acoustic pressure showed a distinct decrease by 58% with the varied cell pitch. Hence, the cell pitch was designed as 0.6 mm to achieve a maximum acoustic pressure while maintaining the size of the serpentine electrode unchanged. Under this condition, we further studied the influence of the cell number and curvature radius on the aforementioned three parameters. Figure 1Hb showed the variation of the SAFU beam width with the array curvature and cell numbers. For the curvature radius varying from 10 to 60 mm, the beam width exhibited a rapid decrease and then a slight increase with the cell number increasing from 1 × 1 to 8 × 8. However, for a given cell number, the beam width increased with the increase in array curvature radius, reflecting the abatement of the focusing effect. Figure 1Hc showed the dependence of the beam depth on the cell number and array curvature radius. Both theoretical and numerical results indicated that the SAFU beam depth increased with both the curvature radius and cell number. The variation regulation of the width and depth of the focused ultrasound beam with the array curvature radius was highly consistent with the inherent correlation between the blood vessel and skin curvature, that is, the curvature radii of the skin increased correspondingly with the depth and diameter of blood vessels beneath it (fig. S3 and note S2). Therefore, with rational design of the transducer array, the depth and width (–6 dB) of the focused ultrasound beam could adaptively cover or focus on the blood vessel area in different parts of the human body with the curvature variation of the skin. The ultrasound beam shapes under different curvatures, tested by an acoustic testing system (fig. S1A), further demonstrated the aforementioned variation regulation of the beam width and depth (–6 dB) with the skin curvature (fig. S8).

Figure 1Hd presented the variation of the acoustic pressure with the cell number and curvature radius. The acoustic pressure increased with the increased cell number while decreasing with the increased curvature radii. However, for all the studied curvature radii, the acoustic pressure reached a plateau when the cell number was larger than 6 × 6. The summaries of the SAFU beam width, beam depth, and acoustic pressure with the cell number and curvature radius were given in tables S1 to S3. On the basis of these regulations, the transducer array with cell numbers of 6 × 6 and a cell pitch of 0.6 mm was chosen. Figure 1G showed the experimental results of the designed transducer array under different bending states. The transducer array featured a beam width varying from 2.1 to 4.6 mm and a beam depth varying from 3.3 to 53.0 mm when its curvature radius increased from 10 to 60 mm, which could well cover the blood vessel area under variable skin curvature radius (fig. S3 and note S2). In contrast, the beam width and beam depth of the single cell were substantially different from the average width and depth of the vascular area, which would lead to low SNR and echo amplitude, further verifying the advantages of the SAFU array. In addition, the effects of the stretch rate on the beam width and depth were investigated (fig. S7, D to F). As the array became sparser with the increased stretch rate by 30% [the tensile range of human skin (15)], the beam width and depth increased, while the acoustic pressure slightly decreased by 17%, which were acceptable for the TOF detection. For all the studied performances, the simulated numerical results agreed well with the theoretical results by our established spatial sound field analysis theories [Fig. 1H, (b) to (d)]. The structure parameters of the SAFU transducer array can be facilely tuned to increase the depth and width of the focused ultrasound beam if needed (fig. S7 and tables S1 and S2).

In summary, with rational design of the ultrasonic transducer array, the SAFU method can enable high-energy ultrasonic beams to adaptively focus on target vessel regions based on the inherent curvature of the skin, without additional phase-control circuit system as required for the phased-array method (note S6). This innovative approach substantially reduced the complexity in the sensing system and device fabrication process while maintaining high sensing performances, offering an advantageous technical solution for the TOF-based BP waveform monitoring. Table S4 provided a detailed comparison between the SAFU method, phased-array method, and single cell-based method, further demonstrating the distinct advantages of our proposed SAFU method in TOF-based biomonitoring (note S7). The realistic skin surface may not be a perfect arc surface and would affect the focusing of the ultrasound beam. However, a comparison between the experimental results shown in figs. S9B and S8A indicated that the acoustic fields from the realistic skin surface and the perfect arc surface featured similar beam shapes. Compared to the perfect arc surfaces, the realistic skin surface did not cause severe beam defocusing. The maximum beam depth (–6 dB) and beam width (–6 dB) deviations between them were as small as 8.0 and 8.2%, respectively (note S8). To further reduce the effect, array structure optimization and an impedance matching layer with special curved surfaces may be possible approaches.

MEMS process–compatible fabrication and device characterization

Currently, the fabrication of flexible ultrasonic transducers still heavily relies on manual processes, such as bonding piezoelectric cells and assembling and transferring the electrodes, which substantially hinders the batch-to-batch and commercial application of the device. Therefore, we proposed a MEMS-compatible process to eliminate complex manual steps, enabling batch-to-batch production and ensuring processing consistency (a comparison between the two kinds of processes is shown in table S5). To enable their batch-to-batch, high-consistency, and large-area manufacturing, we developed a MEMS-compatible process for the proposed flexible ultrasonic transducer array (see Materials and Methods).

The fabrication process started from spinning a layer of PDMS on a glass slide as the substrate and laminating a layer of polyimide/Cu foil on the PDMS layer by oxygen plasma to enhance its surface adhesion [Fig. 2A (a)] (43, 44). Then, a PZT-5H piezoelectric film with a nickel layer was bonded on the surface of the Cu layer after coating conductive silver adhesive (05001-AB, SPI), which was further baked at 150°C for 2 hours to increase the bonding strength [Fig. 2A (b)]. Next, the femtosecond laser etching technology was used to etch the PZT-5H film into transducer cells in an array (the processing parameters were optimized to minimize the depolarization phenomenon; note S9) and pattern the polyimide/Cu foil to form the bottom serpentine electrode [Fig. 2A (c); see figs. S11 to S13 for the detailed process]. Sequentially, the assembly was carried out by spin coating another layer of PDMS to isolate the bottom electrode [Fig. 2A (d)]. Then, the femtosecond laser etching technology was used to etch a 250 μm–by–250 μm opening at the center of the PZT cell to remove the PDMS [Fig. 2A (e) and fig. S14] and expose the PZT with a nickel layer. Then, the magnetron sputtering process and femtosecond laser technology were used to successively deposit and etch the Au/Ti metal layer to form the structured top electrode [Fig. 2A (f)].

After encapsulating the whole device with PDMS, a self-adhesive hydrogel layer was chemically bonded on the surface of the encapsulation PDMS layer under ultraviolet (UV) irradiation [Fig. 2A (g)]. Before bonding, the PDMS layer was treated with benzophenone solution to activate hydrophilic groups at the PDMS surface, which would form covalent cross-linking with the hydrogel layer under UV irradiation (44). Figure 2A (h) showed the fabricated flexible ultrasonic patch. Figure 2B (a) showed the digital picture of the fabricated self-adhesive flexible ultrasonic array. As shown in Fig. 2B (b) to (d), the flexible ultrasonic array could be facilely stretched, conformably attached to a cylindrical plastic pipe with a diameter of 5 mm, and twisted to an angle of more than 180°, demonstrating its excellent flexibility and stretchability.

Compared with previously reported manual operation–based fabrication technologies, our fabrication process exhibited prominent advantages. First, the entire process was MEMS compatible, which contributed to substantial improvement in the device structure and thus performance consistencies (25). The device consistency was well validated by the experimental testing of the impedance and resonant frequencies of the transducer array. As shown in Fig. 2C, the resonant frequencies of the 36 cells in one transducer array were all around 2.4 MHz with a small SD of 0.0236 MHz. Figure 2D further showed a comparison of the impedance-amplitude curves among different transducer arrays. The impedance-frequency curves of different transducer arrays overlapped well with each other, demonstrating high device-to-device performance consistency and batch-to-batch fabrication capacity. Second, the MEMS-compatible process could contribute to low-cost and large-area batch-to-batch fabrication, exhibiting desirable advantages such as traditional MEMS ultrasonic transducers for commercial applications (note S11 and table S6) (45, 46). The acoustic performance of the fabricated ultrasound transducer array was experimentally characterized using an acoustic pressure measurement system (fig. S1). As shown in Fig. 2 (E and G), the maximum acoustic pressure sensitivity reached a high level of 2.2 kPa/V with a focal point beyond 30 mm, which surpassed that of the previously reported flexible patch (23, 43). An experimental test on the pork tissue showed that the flexible patch could generate an acoustic pressure of more than 21.5 kPa with a –6-dB fractional bandwidth of 47% (Fig. 2E) after a penetrating depth of 30 mm (Fig. 2F), implying a superior capability for detecting deep vessels such as the carotid artery.

Besides the advantages in the device fabrication process, the introduction of the self-adhesive hydrogel layer could not only enable convenient adhesion of the transducer array on human skin but also function as an acoustic impedance layer to improve the transmission efficiency of the ultrasonic waves from the device to human skin. This point was experimentally evaluated by testing the received pulse-echo voltages of the ultrasonic arrays with and without the hydrogel layer. As indicated in Fig. 2H, the received voltage amplitude by the transducer array with the hydrogel layer was 181% higher than that of the one without the hydrogel layer, which demonstrated the improved acoustic impedance matching capability of the self-adhesive ultrasonic transducer array with human skin. To evaluate the adhesion performance of the transducer array, we conducted a 180° peeling test to measure the peeling force of the self-adhesive hydrogel layer from the porcine skin because of its resemblance to human skin (note S12) (46, 47). As shown in Fig. 2I, the peeling force was 1.6 N/cm (71-fold higher than the case without self-adhesive hydrogels) after 0 hours of attachment and 1.1 N/cm after 24 hours of attachment, respectively, demonstrating a tough and long-lasting adhesion performance. The removal of the self-adhesive transducer array with tweezers indicated its excellent adhesion performances on human skin (Fig. 2J). In addition, as shown in Fig. 2K, no immune response was observed after 24 hours of attachment on human skin, which verified the excellent biocompatibility of the self-adhesive transducer patch. Overall, the developed flexible transducer array had excellent advantages in both structure design and fabrication process, exhibiting immense promise in wearable ultrasonic medical monitoring applications.

Continuous BP waveform monitoring of blood vessels with various curvatures and depths

Continuous BP waveform monitoring can be used to predict CVDs such as instantaneous BP increase and blood flow plaque formation (48). The experimental testing of the BP waveform was conducted by attaching the flexible ultrasonic transducer array to blood vessels at different parts of the human body. The details on the data acquisition system and the volunteers were shown in fig. S17 and table S8, respectively. A mercury sphygmomanometer (Sph) (XJ-A, Yuanyan) was used to verify the BP measurement accuracy of the transducer array. Obvious pulse-echo ultrasonic waves reflected from the anterior and posterior walls of the radial artery vessel were observed. Their periodic changes during continuous testing of 4.2 s indicated the rhythmic changes of the radial artery vessel (Fig. 3A). Figure 3B showed experimental results of the radial artery BP tested on a healthy volunteer (volunteer 1), which was calculated using eqs. S1 to S3 based on the received ultrasound waves. Continuous BP waveform of the radial artery vessel was obtained, exhibiting DBP and SBP of around 79.3 and 125.9 mmHg, respectively, which was within the normal range of healthy people. As shown in Fig. 3C, the tiny changes, such as the tidal notch, tidal peak, dicrotic notch, and dicrotic peak, and key feature points, such as systolic peaks and diastolic troughs, were distinguished, demonstrating the superior resolution of our flexible transducer array. These tiny peaks and troughs played an important role in the detection and prevention of CVDs. For instance, a low tidal peak often indicated abnormal peripheral vascular resistance, and the amplitude and time of the dicrotic notch and dicrotic peak reflected the vasoconstriction and blood flow reflux after aortic valve closure, which implied diseases such as aortic valve insufficiency (49–51). The accuracy and reliability of our ultrasonic transducer array were verified by repeatable measurement and the results from the mercury Sph. The measurement errors of the flexible ultrasonic transducer in SBP and DBP were as small as 1.8 mmHg (an SD of 2.98 mmHg) and 1.1 mmHg (an SD of 2.43 mmHg), respectively, demonstrating high accuracy (Fig. 3B). A deep insight into the cycling testing of the BP showed that our devices had a small uncertainty of 2.99 mmHg, which was comparable to that of commercial Sphs, and demonstrated the high reliability (note S13). In practical usage, the relative position variations between the transducer array (or focused ultrasound beam) and blood vessels would cause inaccuracies in the vessel diameter estimation and thus BP, which included the nonperpendicularity of the blood vessel to the ultrasound beam direction (fig. S19, A and D) and the misalignment of its radial direction with the ultrasound beam direction (fig. S19F). However, as shown in fig. S19 (B and E), for the studied typical arterial vessels, the maximum measurement errors caused by the nonperpendicularity situations were as small as 0.09 and 0.56 mmHg in the vessel’s longitudinal and transverse cross sections, respectively, which were negligible for the general measurement. For the misalignment situation, as indicated in figs. S19J and S20C, within a certain range of misalignment distance (2.22 to 6.04 mm), the received signal amplitude remained above –6 dB, and the measurement error was less than 3 mmHg, which was comparable to the accuracy of the commercially available Sph, such as Omron RS8 (note S14) (52). These results demonstrated the performance robustness of our proposed SAFU transducer array. For high-precision measurement, the nonperpendicularity-induced measurement error could be compensated through eq. S9 or S10 after the calibration of the angle between the ultrasound beam direction and the radial direction of the blood vessel (note S14). In addition, the echo amplitude–based locating method could be used to align the transducer to the perfect alignment position with the target blood vessel, thus ensuring measurement accuracy (note S15).

Fig. 3. BP waveform testing on various parts of the human body and long-term BP monitoring by the SAFU transducer array.

(A) Pulse-echo signals from the anterior and posterior walls of blood vessels. (B) BP waveforms tested by our work and Sph. (C) BP waveform analysis. (D) BP testing on volunteers with different body mass index (BMI). (E) Repeatability testing. (F) BP detection range testing of our SAFU transducer array. (G) BP waveform testing on various blood vessels at different depths. (H) Twenty-four–hour continuous BP testing. Error bars represent ±SD (N = 5). h, hours.

To demonstrate the robustness performance of the fabricated flexible transducer array, we used three volunteers with body mass index (BMI) values of 30.2 (volunteer 1), 28.7 (volunteer 2), and 25.8 (volunteer 3) for BP measurement testing (table S8). Figure 3D showed the experiment results of artery BP tested on volunteers 1, 2, and 3. Continuous BP waveform of radial artery vessels was obtained for all these tested volunteers. The average values of SBP (DBP) of volunteers 1, 2, and 3 were 130.4 (84.3), 125.3 (86.7), and 115.1 (81.6) mmHg, respectively, which were almost the same as results from the Sph, i.e., 131.1 (85.1), 126.7 (86.0), and 116.0 (80.6) mmHg, verifying the high accuracy for volunteers with different BMI (Fig. 3E). An analysis of the BP waveforms of the three volunteers indicated that volunteer 1 had a higher SBP and a larger pulse pressure (the difference between SBP and DBP), which reflected the higher arterial stiffness of the blood vessel. The BP measurement range of the ultrasonic transducer was characterized by a pressure gauge system (fig. S18B). As shown in Fig. 3F, the results indicated that the ultrasonic transducer could accurately capture the pressure waveform within a pressure range of 10 to 350 mmHg, which was enough for BP pressure monitoring, even under extreme health conditions such as hemorrhagic shock, hyperoxia, and acute hypertension (53, 54).

Furthermore, experimental tests of BP waveforms on the fingertip artery, radial artery, brachial artery, and carotid artery were conducted to verify the detection capability of our flexible transducer array under different curved surfaces and depths. Figure 3G showed the tested results of the BP waveforms of the aforementioned arteries, in which their B-mode ultrasound images were measured by the Verasonics Vantage system (Vantage 256, Verasonics). For all the tested parts, distinct BP waveforms were measured, validating the excellent sensing performance of the flexible transducer patch on various curved surfaces and depths, covering all the main arteries of the human body. A comparison of SBP variations between peripheral vessels and central vessels indicated that the SBP of peripheral arteries (fingertip artery and radial artery) was about 14 mmHg higher than that of central arteries (carotid artery). This is because the enhanced blood flow reflex in the peripheral blood vessels would cause superimposition of the forward- and backward-traveling waveforms (50).

Furthermore, we performed 24-hour continuous BP monitoring by adhering the flexible transducer array to the radial artery of a healthy volunteer, which could reveal abnormal BP fluctuations and thus identify potential CVDs such as myocardial infarction and ischemia (16). Figure 3H showed the measured SBP and DBP during 24 hours using both our flexible ultrasonic transducer array and the commercially available Sph. Both the SBP and DBP measured by our fabricated device showed high consistency with those detected by the Sph during the whole testing period. The average relative errors of the SBP and DBP were ±2.48 and ±1.99 mmHg, respectively, indicating the high measurement accuracy and performance reliability of our device for long-term monitoring. A deep insight into the tested results revealed that the DBP showed slight fluctuation during a day, while the SBP presented a higher value at 0 and 23 hours and a lower value at 13 hours. This phenomenon stemmed from the abnormal excitement of the sympathetic nerve at 0 and 23 hours of a day, leading to enhanced contractility of the human heart (55). Besides, long-term BP data can also predict many CVDs. For instance, morning hypertension substantially increased the risk of cardiovascular and cerebrovascular diseases such as stroke. Persistent nighttime BP above normal levels during sleep may indicate heart failure, coronary heart disease, and cardiomyopathy, which is difficult to detect by traditional BP monitoring techniques. CVDs such as myocardial ischemia and arrhythmia can also lead to abnormal BP fluctuations (56, 57). Therefore, the successful long-term continuous and high-accuracy BP monitoring achieved by our flexible ultrasonic transducer array demonstrated its capacity for clinically relevant diagnosis and early prevention of CVDs.

Real-time heart rate and BP waveform monitoring during exercises

Continuous heart rate measurement can detect hidden arrhythmias, such as bradycardia or ventricular tachycardia, thus screening high-risk individuals for sudden death and fainting (53). To demonstrate the accuracy of the fabricated ultrasonic transducer array for heart rate measurement, we measured the heart rates of volunteer 1, volunteer 2, and volunteer 3 and compared their results with those measured by a pulse oximeter (accuracy of 1%; 82A1, Haier). Figure S22 (A to C) showed the heart rate testing results of the three volunteers in the resting state. The heart rates of the three volunteers tested by the flexible ultrasonic transducer array showed high consistency with those detected by the pulse oximeter, with a maximum error of ±2 bpm (beats per minute), indicating the excellent accuracy of our device in heart rate measurement.

Heart rate and BP monitoring during exercise can help diagnose hidden CVDs that generally were unable to be found at resting states, such as arrhythmia, hypertension, and fainting (58). For instance, the increase and decrease in heart rate and abnormality in BP waveforms before and after exercise can reflect the cardiac chronotropic response and abnormal heart functions (58, 59). For this, we conducted the heart rate and BP waveform monitoring under different exercises using our SAFU transducer array. As shown in Fig. 4A, three types of exercises were tested, i.e., squat, one-arm dumbbell, and spot running. In the squat and one-arm dumbbell exercises, volunteers performed exercises with a duration of 60 s/cycle. Both types of exercise were conducted at a frequency of 30 repetitions/min. The spot running was performed at a frequency of 140 steps/min for 2 min. After each cycle of exercise, the volunteers rested for 5 min to restore their heart rate and BP. A sedentary layperson (volunteer 1) and a long-term trained athlete (volunteer 4) participated in the aforementioned three types of exercises for comparison. The experimental results of the heart rate and BP waveforms were shown in Fig. 4.

Fig. 4. Continuous and real-time detection of both heart rate and BP waveform during exercise.

(A) Schematic of different exercises, such as spot running, squat exercise, and one-arm dumbbell exercise with a barbell weight of 2 kg. (B) Heart rate variations during spot running, squat exercise, and one-arm dumbbell exercise. (C) BP waveform variations during spot running, squat, and one-arm dumbbell exercises. (D) Comparison of RSI during spot running, squat, and one-arm dumbbell exercises. Error bars represent ±SD (N = 5). *P < 0.05; the difference between layperson and athlete before exercise versus the difference during exercise.

Figure 4B showed the heart rate variations of both the athlete and the layperson during spot running, one-arm dumbbell, and squat exercises from a resting state to 3 min after exercises (note S16). For both volunteers, their heart rate exhibited a distinct increase during the exercise and gradually recovered to their value at the resting status after exercise. This heart rate change was completely captured by our SAFU transducer array, demonstrating its capability of monitoring heart rate under exercise status. An insight into the heart rate difference between the two volunteers revealed that the athlete had a smaller heart rate increase than that of the layperson, which could be attributed to the higher cardiac stroke volume of the athlete [fig. S23; clinically measured by a color Doppler ultrasound system (DC-8, Mindray)]. Figure 4C showed the BP waveform variations of the two volunteers under the aforementioned three types of exercises. Similar to the variation of heart rate, the BP waveforms exhibited a rapid increase at the beginning of the exercise, decreased gradually, and ultimately returned to baseline levels within 3 min after exercise. Complete BP waveforms during the whole exercise process were detected, indicating the excellent capability of our device in BP waveform monitoring under exercise conditions. For all the conducted exercises, the athlete presented smaller SBP and heart rate fluctuations than those of the layperson. The lower SBP and heart rate fluctuation of the athlete under the same strength exercise could be attributed to the better vascular compliance, which was further verified by the vascular stiffness coefficient testing of the two volunteers (i.e., 9.31 for the layperson and 6.84 for the athlete given in the next section). Compared with one-arm dumbbell and spot running exercises, the squat exercise caused larger increases in both the heart rate and SBP and a longer recovery time, which was because the squat exercise was a higher intensity exercise than the other two types of exercise (60). The recovery times of heart rate and BP waveform after exercise can be used to evaluate the cardiovascular function, and the longer recovery time may be related to cardiovascular risks, such as hypertension, high cholesterol, diabetes, and heart failure, etc. (61). In this study, the heart rate and BP recovery times of the athlete for all the exercises were much smaller than those of the layperson. Specifically, the heart rate of the athlete almost returned to normal within 3 min. These results revealed the stronger cardiovascular function of the athlete, which was clinically verified by the echocardiographic testing (fig. S23). However, as the heart rates of both layperson and athlete doped more than 18 bpm within 1 min after exercises, their cardiovascular capacities were within the healthy or normal status (61). Besides, the heart rate and BP waveform variations of the two volunteers were also tested under low intensity for the three types of exercises, the results of which further validated the aforementioned conclusions (note S17).

A combined detection of BP and heart rate can be used to diagnose many specific CVDs, such as stroke, arrhythmia, and shock (11). For instance, a simultaneous increase in the SBP and heart rate could reflect the sympathetic nervous system’s excitement and suggest the risk of stroke, acute myocardial infarction, and other diseases (11, 12). An increase in heart rate but a decrease in SBP can indicate heart failure or malignant arrhythmia (62). If both heart rate and SBP do not increase evidently during exercise, then it may imply an increase in vagal nerve tension, which could cause fainting and shock (63).

Therefore, by combining the tested BP and heart rate results, we used a reverse shock index (RSI) to quantitatively analyze cardiovascular function during exercise. The RSI was defined as the ratio of SBP to heart rate. A higher RSI value meant a stronger cardiac contractile function (64). Figure 4D showed the comparison of the RSI before and after exercise. For all the tested exercises, the RSIs of the athlete were higher than those of the layperson, indicating a healthier cardiovascular system. It could also be found that the difference in RSI between the layperson and athlete during exercise was much higher than that before exercise. For instance, in the high-intensity squat exercise, the RSI difference between the layperson (1.46 mmHg/bpm) and athlete (1.96 mmHg/bpm) was 0.50 mmHg/bpm before the exercise, while it increased to 0.64 mmHg/bpm (difference between 0.90 and 1.54 mmHg/bpm) during exercise. The larger difference in RSI under exercise could better reflect the difference in their cardiovascular function (P < 0.05). These results indicated that monitoring BP and heart rate during exercise can be used to assess cardiovascular function and prevention of CVDs; however, this was difficult to achieve by ordinary Sphs.

Arterial stiffness assessment by the self-adhesive SAFU transducer array

Arterial stiffness can reflect the elasticity and compliance of arterial vessels. The increase in arterial stiffness will cause a series of diseases such as atherosclerosis, coronary heart disease, peripheral arterial disease, etc. (65). To demonstrate the capability of our SAFU transducer array in arterial stiffness evaluation, we conducted experimental tests on three groups of volunteers, i.e., a patient with hypertension (volunteer 5) and a normotensive volunteer (volunteer 7) (group I), an elderly with age older than 55 years old (volunteer 6) and a young volunteer with ages younger than 30 years old (volunteer 3) (group II), and a layperson (volunteer 1) and an athlete (volunteer 4) (group III) (table S8). The arterial stiffness parameters, β (beta stiffness index) and PWV (pulse wave velocity), were calculated using eqs. S4 and S5 based on the detected vessel diameter variations.

Figure 5 (A to C) showed the arterial stiffness parameters, β and PWV, and the corresponding BP waveforms of the aforementioned three groups. As shown in Fig. 5A, the SBP of the patient with hypertension reached 152 mmHg, which severely exceeded the normal range (90 to 140 mmHg) of a healthy person (66). The β and PWV of the patient with hypertension were 13.09 and 13.04 m/s, which were much higher than those (i.e., 9.90 and 9.52 m/s) of the normotensive volunteer, implying that his arterial vessel elasticity and compliance suffered serious deterioration [Fig. 5A (b)]. Figure 5B showed the tested results of the β and PWV of the elderly and young volunteer groups. The elderly volunteer had both β and PWV of 10.37 and 11.20 m/s, respectively, which exhibited a distinct increase compared to those of the young volunteer [Fig. 5B (b)]. This was reasonable because the arterial stiffness of human beings increased with their age (65). Figure 5C showed the experimental results of the β and PWV of the layperson and athlete groups. As shown in Fig. 5C, the layperson exhibited both higher β and PWV (9.31 and 9.18 m/s, respectively) than those of the athlete, implying that the athlete had better vascular elasticity and compliance [Fig. 5C (b)]. This also suggested that long-term exercise had a positive effect on the human blood vessels. These results demonstrated the effectiveness of the SAFU transducer array in arterial stiffness evaluation.

Fig. 5. Vessel stiffness analysis for individuals with different ages and health conditions.

(A) Measured results of the hypertensive and normotensive individuals. (a) BP waveforms. (b) β and PWV. Error bars represent ±SD (N = 5). *P < 0.05; hypertensive versus normotensive. (B) Experimental results of the elderly and young individuals. (a) BP waveforms. (b) β and PWV. Error bars represent ±SD (N = 5). *P < 0.05; elderly versus young. (C) Tested results of the layperson and the athlete individuals. (a) BP waveforms. (b) β and PWV. Error bars represent ±SD (N = 5). *P < 0.05; layperson versus athlete. (D) Schematic of the vessel stiffness of the elderly and young individuals. (E) Comparison of β and PI between elderly and young individuals. Error bars represent ±SD (N = 5). *P < 0.05; young subjects versus elderly subjects. (F) Comparison of BP waveforms between the typical elderly and young individuals.

To further validate its accuracy, we compared the PWV results measured by our ultrasonic transducer array with those tested by the arteriosclerosis detector (VBP-9, Chjoy), a recognized high-precision measurement method for vascular stiffness evaluation in clinical practice. As shown in fig. S25, both aforementioned methods could effectively distinguish PWVs of the hypertensive and normotensive, the elderly and young, and the athlete and layperson. Compared to the arteriosclerosis detector, the relative measurement errors of our ultrasonic transducer were less than 25% for all the tested groups mentioned above, demonstrating the high accuracy of our device. The small deviation between them could be attributed to their different measurement positions (note S18). That is, the arteriosclerosis detector measured the average PWV across the ankle-brachial pathway, whereas the ultrasonic transducer array assessed localized PWV at the radial site. In addition, the estimation of pulse wave propagation distance and fluctuations in BP will also lead to an increase in errors.

Although the β and PWV in Fig. 5 (A to C) exhibited similar trends in their changes for the aforementioned tests, their changes in magnitude were not identical with each other. This is because the PWV is more sensitive to pulse pressure than β, which embodies the combined influence of BP and arterial stiffness. Consequently, associating the PWV with hypertension resulting from arteriosclerosis is more reasonable, while the β exhibits a stronger correlation with CVDs such as coronary heart disease. Therefore, a combined analysis of both β and PWV was essential for the comprehensive assessment of arterial stiffness and its related diseases (66).

As demonstrated by the aforementioned results, elderly individuals often exhibited larger arterial stiffness compared to younger ones, which mainly caused age-related CVDs (Fig. 5D). Furthermore, clinical tests based on a color Doppler ultrasound system (DC-8, Mindray) were implemented to obtain the pulsatility index (PI; representing vessel compliance, obtained using eq. S6) to verify the results given by our SAFU transducer array. Figure 5E showed the β and PI results of two groups, i.e., a group of elderly individuals with an average age of 68 years and a group of young individuals with an average age of 25 years. As shown in Fig. 5E, all the elderly individuals exhibited higher β than those of the young group (P < 0.05), indicating the higher blood vessel stiffness in the elderly. These results were highly validated with the PI results and clinical imaging of arteries measured by the clinical instrument (fig. S26). Moreover, BP waveforms can also provide evidence for arterial stiffness evaluation. Figure 5F showed the comparison of the BP waveform between the typical elderly and young individuals within a single cycle. As shown in Fig. 5F, the T, SP, BP, TP, and DP in the BP waveform represented the rise times, systolic peak, blood pressure, tidal peak, and notch peak, respectively. Compared to the T₂ (value of 0.20) of the young volunteer, the smaller T₁ (value of 0.12) in the BP waveform of the elderly volunteer meant a fast increase in the SBP. This is because the pervasive arteriosclerosis in the elderly individuals would exacerbate the heart ejection function and aorta compliance, thus decreasing the time from diastolic trough to systolic peak (51). Besides, the elderly individual also exhibited a smaller amplitude of normalized DP₂ (value of 0.25) than the DP₁ (value of 0.43) of the young individual. This could be attributed to the increase in the elastic modulus of blood vessels for elderly individuals, which resulted in a smaller tidal wave reflected back by peripheral blood vessels (67). These results demonstrated that our SAFU transducer array could effectively assess the characteristics of arterial stiffness, thus achieving the diagnosis of CVDs.

DISCUSSION

We have demonstrated the development of a skin-adaptive focused skin-conformal ultrasonic transducer patch with a MEMS-compatible process for noninvasive, long-term, and real-time monitoring of cardiovascular vital signs. With rational design of the ultrasonic transducer array, the introduced SAFU transducer array can enable high-energy ultrasonic beams to adaptively focus on target vessel regions based on the inherent curvature of the skin, without an additional phase-control circuit system as required for the phased-array method. This method can contribute to substantial improvement in SNR, acoustic pressure, penetration depth, and directionality while simplifying the controlling circuits. The rational design of the SAFU transducer array based on the established theoretical and simulation models for the spatial sound field enabled a wide range of beam widths and depths, which completely covered the diameters and depths of various blood vessels of the human body. The introduced self-adhesive hydrogel layer effectively enhanced the acoustic coupling, biocompatibility, and skin attachment reliability between the transducer array and skin interfaces. The developed MEMS-compatible processes enabled high-consistency and scalable manufacturing, overcoming the limitation of traditional manual methods and exhibiting a small SD of 1.0% in resonant frequencies of the fabricated 6 × 6 transducer array. The self-adhesive SAFU transducer array with a thickness of 300 μm could conformably adhere to the skin of human subjects and detect various cardiovascular vital signs with high accuracy, including continuous BP waveforms for various vessels, long-term BP over 24 hours, simultaneous heart rate and BP during exercises, and arterial stiffness for individuals with different ages and health conditions, validated by clinical medical testing.

Further research can concentrate on the following several aspects to promote the self-adhesive SAFU transducer patch into practical usage. First, the resonant frequency and bandwidth of the device can be optimized through strategic approaches including dimensional tuning of piezoelectric cells, introduction of the gradient matching layer, and use of composite piezoelectric materials. Second, replacing the current piezoelectric bulk material with high-performance piezoelectric thin films that can be sputtered or deposited by a MEMS process is vital to further enhance the batch-to-batch and high-consistency manufacturing capability of the device and also lower the cost. Furthermore, improving the moisture retention of the self-adhesive hydrogel layer will extend the device duration and prevent dehydration and skin irritation. Moreover, incorporating an automatic calibration mechanism that adjusts the parameters for BP calculation during measurements would further enhance accuracy and eliminate the need for manual recalibration. Last, integrating advanced algorithms, flexible signal processing circuits, wireless communication systems, and low-power designs would ultimately reduce the system dimension and implement real wearability, promoting the noninvasive, continuous, real-time, and long-term cardiovascular health assessment to reality.

MATERIALS AND METHODS

Materials

The PDMS (184 silicone elastomer) was acquired from SYLGARD. The copper foil polyimide film (17-μm Cu with 25-μm polyimide) was supplied by Zhejiang Dongjin New Materials Co. Ltd. The conductive silver adhesive (05001-AB) was acquired from SPI. The PZT-5H was acquired from Changzhou Ultrasonic Electronics Co. Ltd. Aladdin. The Ti and Au target material was supplied by Zhongnuo New Materials Co. Ltd. The acrylamide [Analytical Reagent (AR), 99%], acrylic acid (AR, 98%), and gelatin (type A) were acquired from Aladdin. The Irgacure 2959 photoinitiator (AR, 98%) was acquired from Sigma-Aldrich.

Fabrication of the self-adhesive SAFU transducer array

Figure S10 showed the detailed fabrication process of the ultrasonic array. First, a glass slide was spin coated with PDMS (SYLGARD 184 silicone elastomer; 20:1). This process was performed at 1000 rpm for 1 min. The glass slide was cured in a vacuum oven at 80°C for 1 hour. Then, the PDMS layer was subjected to oxygen plasma treatment for 2 min to enhance surface adhesion. Afterward, the glass slide served as the substrate to laminate the copper foil polyimide film (17-μm Cu with 25-μm polyimide) in contact with the PDMS. Then, the conductive silver adhesive (05001-AB, SPI) mixed with diluent (05004, SPI) was spin coated on Cu at 1000 rpm for 1 min, and the whole PZT-5H material with a 2-μm nickel-plated layer was bonded to the copper foil polyimide film at 150°C for 120 min. A laser ablation system (1.65-W power, 25-kHz pulse repetition frequency, 400 mm/s of laser cutting speed, 1000-ns pulse width, 15-μm etch line spacing, and 350 etching times) was then used to create the PZT-5H cell array (fig. S11). Then, the bottom serpentine circuit pattern was also created by the laser etching system (0.92-W power, 25-kHz pulse repetition frequency, 400 mm/s of laser cutting speed, 1000-ns pulse width, 15-μm etch line spacing, and 180 etching times) with the width of 0.15 mm (fig. S13). The device was then coated with a PDMS layer with the thickness of 200 μm (500 rpm for 1.5 min) to isolate the bottom electrode. The femtosecond laser etching technology (10.02-W power, 50-kHz pulse repetition frequency, 400 mm/s of laser cutting speed, 1000-ns pulse width, 15-μm etch line spacing, and 10 etching times) was used to etch a 250 μm–by–250 μm opening at the center of the PZT cell to remove the PDMS (fig. S14). Then, the magnetron sputtering process was used to form the top electrode with 30-nm Ti and 200-nm Au and well bond them with the nickel layer on the exposed PZT. The laser ablation system created the top circuit pattern (0.92-W power, 25-kHz pulse repetition frequency, 400 mm/s of laser cutting speed, 1000-ns pulse width, 15-μm etch line spacing, and 10 etching times). As the strain is primarily localized in the PDMS between the PZT blocks, the PDMS and top electrode at the center of each cell remain nearly undeformed (note S10), and thus, these processes can substantially reduce the local strain on the top electrode at the soft-hard material junction and ensure a stable electrical connection. A top layer of PDMS was coated on the top surface of the device, and thus, the top electrode can be protected. Last, the self-adhesive hydrogel patch was formed on the device. The chemicals for synthesizing the bioadhesive hydrogel include acrylamide, acrylic acid, gelatin, and photoinitiator (Irgacure 2959). To prepare the hydrogel, we dissolved 20% (w/w) acrylamide, 15% (w/w) acrylic acid, 15% (w/w) gelatin, and 1% (w/w) photoinitiator in deionized water. The mixture was then centrifuged at 500 rpm to remove air bubbles. Then, the prepolymer solution was poured on the transducer array treated with benzophenone solution for 5 min with a PDMS mold with a thickness of 1 mm. The solution was cured in the UV light chamber (50 mW·cm² of power) for 5 min to get the bioadhesive hydrogel patch bonded to the transducer array by the covalent cross-linking. Figure S16 showed the excellent flexibility and attachability with the skin of the hydrogel patch.

Measurement of the acoustic field, sensitivity, and bandwidth

The acoustic testing platform diagram was shown in fig. S1A, which was composed of a signal generator (DG1032Z, Rigol), a voltage amplifier (ATA-3000, Aigtek), a needle hydrophone (NH2000, Precision Acoustics), a radio frequency (RF) amplifier (BR-640A, Ritec), an oscilloscope (MSO44, Tektronix), a DC-coupling power supply, a 3D motion platform, and the fabricated flexible transducer. The signal generator was used to provide AC excitation voltages for the device. The voltage amplifier was used to amplify the excitation voltage. The flexible transducer transmitted ultrasonic signals, and the needle hydrophone was used to receive ultrasonic signals. The DC-coupling power supply was used to power the hydrophone. The RF amplifier was used to filter and amplify received signals, and the oscilloscope was harnessed to display and record received signals. The 3D motion platform was used to control the position of the hydrophone in the acoustic field.

The sound field testing range was 35 mm by 10 mm, and the spatial resolution was 0.2 mm. To detect the sound field of flexible devices, the ultrasonic excitation frequency was set to 2.4 MHz, and the excitation voltage was set to 40 Vpp (peak-to-peak voltage) by the voltage amplifier. The amplification factor of the RF amplifier was set to 24 dB. Because the sensitivity of the hydrophone is 0.308 V/MPa, the acoustic pressure at point (x_p, y_p, z_p) in the acoustic field is as follows

$$P(x_{\mathrm{p}},y_{\mathrm{p}},z_{\mathrm{p}})=V_{\mathrm{r}}(x_{\mathrm{p}},y_{\mathrm{p}},z_{\mathrm{p}})/(M_{\mathrm{r}}\times n_r)$$ (1)

where V_r is the received voltage, M_r is the sensitivity of the needle hydrophone, and n_r is the gain of the rf amplifier. The acoustic pressure sensitivity is as follows

$$\mathrm{\eta }=P(x_{\mathrm{p}},y{}_{\mathrm{p}},z_p)/V_{\text{ac}}$$ (2)

where V_ac is the device excitation voltage. The frequency bandwidth (–6 dB) is determined as follows

$$\text{BW}=\frac{f_{\mathrm{u}}-f_{\mathrm{l}}}{f_{\mathrm{c}}}\times 100\%$$ (3)

where f_u is the upper frequency, f_l is the lower frequency, and f_c is the central frequency.

Analytical model for acoustic beam performance of the SAFU transducer array

We proposed a theoretical model for analyzing the variations in acoustic beam performance of flexible arrays under bending and stretching states, as shown in Fig. 2H (a). The m is the column cell index; n is the row cell index; M is the number of columns; N is the number of rows; d′_m is the pitch between column cells; d′_n is the pitch between row cells; l_m and l_n are the length and width of a cell, respectively; point C (0, 0, R) is the center of curvature; point P (x_p, y_p, z_p) is the position where the acoustic pressure is calculated; α_m and α_n are the stretching rates in the length and width directions, respectively; and R is the radius of curvature. First, with the array stretching and bending states, the coordinates of (m, n) cell can be represented as

$$x_{\mathrm{o}}(m,n)=R\left\{1-\text{cos}\left[\frac{(l_{\mathrm{m}}+d_{\mathrm{m}}^{\prime })(m-M/2)}{R}\right]\right\}$$ (4)

$$y_{\mathrm{o}}(m,n)=(l_{\mathrm{n}}+d_{\mathrm{n}}^{\prime })(n-N/2)$$ (5)

$$z_{\mathrm{o}}(m,n)=R\text{sin}\left[\frac{(l_{\mathrm{m}}+d_{\mathrm{m}}^{\prime })(m-M/2)}{R}\right]$$ (6)

where

$$d_{\mathrm{m}}^{\prime }=\frac{m}{m+1}\cdot l_{\mathrm{m}}\cdot {\mathrm{\alpha }}_{\mathrm{m}}+(1+{\mathrm{\alpha }}_{\mathrm{m}})d_{\mathrm{m}}$$ (7)

$$d_{\mathrm{n}}^{\prime }=\frac{n}{n+1}\cdot l_{\mathrm{n}}\cdot {\mathrm{\alpha }}_{\mathrm{n}}+(1+{\mathrm{\alpha }}_{\mathrm{n}})d_{\mathrm{n}}$$ (8)

where d_m and d_n are the pitch between column cells and row cells without stretching and bending, respectively.

Figure 2H (a) showed the schematic diagram of the orientation of the (m, n) cell in the spatial coordinate system. $\mathbf{a}$ is the normal vector of (m, n) cell, $\mathbf{b}$ is the plane vector of (m, n) cell along the x direction, $\mathbf{c}$ is a vector from the point (x₀, y₀, and z₀) to the point P, and $\mathbf{d}$ is the projection of $\mathbf{c}$ on the (m, n) cell. These vectors can be derived on the basis of vector calculation as

$$\mathbf{a}(m,n)=\left[-x_0(m,n),0,R\right]$$ (9)

$$\mathbf{b}(m,n)=\left[1,0,\frac{x_{\mathrm{o}}(m,n)}{R}\right]$$ (10)

$$\mathbf{c}(m,n)=\left[x_{\mathrm{p}}-x_{\mathrm{o}}(m,n),y_{\mathrm{p}}-y_{\mathrm{o}}(m,n),z_{\mathrm{p}}-z_{\mathrm{o}}(m,n)\right]$$ (11)

$$\mathbf{d}(m,n)=\mathbf{b}-(\mathbf{b}\cdot \mathbf{c}/{\Vert \mathbf{c}\Vert }^2)\cdot \mathbf{c}$$ (12)

Subsequently, the acoustic pressure at point P (x_p, y_p, z_p) can be calculated by summing the acoustic pressure contributions from each cell in the flexible array, the result of which is given as follows (68)

$$\begin{array}{c}P(x_{\mathrm{p}},y_{\mathrm{p}},z_{\mathrm{p}})={\sum }_{n=1}^N{\sum }_{m=1}^{M}P(m,n)=\\ {\sum }_{n=1}^N{\sum }_{m=1}^M\frac{j\mathrm{\rho }{\mathit{kc}}_{\mathrm{l}}{\mathit{vl}}_{\mathrm{m}}{l}_{\mathrm{n}}}{2{\mathrm{\pi \beta }}^{f\Vert \mathbf{c}\Vert }}\frac{\text{sin}\left[({\mathit{kl}}_{\mathrm{m}}/2)\text{sin}\mathrm{\theta }(m,n)\text{cos}\mathrm{\varphi }(m,n)\right]}{({\mathit{kl}}_{\mathrm{m}}/2)\text{sin}\mathrm{\theta }(m,n)\text{cos}\mathrm{\varphi }(m,n)}\frac{\text{sin}\left[({\mathit{kl}}_{\mathrm{n}}/2)\text{sin}\mathrm{\theta }(m,n)\text{cos}\mathrm{\varphi }(m,n)\right]}{({\mathit{kl}}_{\mathrm{n}}/2)\text{sin}\mathrm{\theta }(m,n)\text{cos}\mathrm{\varphi }(m,n)}\end{array}$$ (13)

where ρ is the medium density, k is the wave number, c_l is the medium sound velocity, v is the vibration velocity, β is the sound attenuation coefficient, f  is the ultrasonic frequency, θ is the angle between $\mathbf{a}$ and $\mathbf{c}$ , and Φ is the angle between $\mathbf{b}$ and $\mathbf{d}$ , which can be written as

$$\mathrm{\theta }(m,n)=\mathrm{a}\text{cos}[\mathbf{a}\cdot \mathbf{c}/\left(\Vert \mathbf{a}\Vert \Vert \mathbf{c}\Vert \right)]$$ (14)

$$\mathrm{\varphi }(m,n)=\mathrm{a}\text{cos}[\mathbf{b}\cdot \mathbf{d}/\left(\Vert \mathbf{b}\Vert \Vert \mathbf{d}\Vert \right)]$$ (15)

After obtaining the acoustic pressure at any spatial point, the focal point, z_f, can be calculated by solving the equation by the numerical method as

$$z_{\mathrm{f}}=\text{max}\{z_{\mathrm{p}}\mid P^{\prime }(0,0,z_{\mathrm{p}})=0\}$$ (16)

Then, the beam depth range can be expressed as z_l to z_u, where z_l and z_u are the lower and upper coordinate values corresponding to the –6-dB maximum acoustic pressure on the z axis, respectively, which can be given as

$$z_{\mathrm{u}}=\text{max}\{z_{\mathrm{p}}\mid P(0,0,z_{\mathrm{p}})=P(0,0,z_{\mathrm{f}})/2\}$$ (17)

$$z_{\mathrm{l}}=\text{max}\{z_{\mathrm{p}}/z_{\mathrm{u}}\mid P(0,0,z_{\mathrm{p}})=P(0,0,z_{\mathrm{f}})/2\}$$ (18)

The beam width, d_f, can be given as

$$d_{\mathrm{f}}=x_{\mathrm{u}}-x_{\mathrm{l}}$$ (19)

$$x_{\mathrm{u}}=\text{max}\{x_{\mathrm{p}}\mid P(x_{\mathrm{p}},0,z_{\mathrm{f}})=P(0,0,z_{\mathrm{f}})/2\}$$ (20)

$$x_{\mathrm{l}}=\text{min}\{x_{\mathrm{p}}\mid P(x_{\mathrm{p}},0,z_{\mathrm{f}})=P(0,0,z_{\mathrm{f}})/2\}$$ (21)

where x_l and x_u are the lower and upper coordinate values corresponding to the –6-dB maximum acoustic pressure on the x axis under the depth of z_f, respectively. To validate the theoretical model, a field II simulation model was established, which was used to simulate the beam. The vibration velocity of the ultrasonic cell was calculated through the model shown in fig. S5A and assigned. The resolution of the acoustic field was set to 0.2 mm. The density and sound velocity of the medium were 1060 kg/m³ and 1540 m/s. The beam width, beam depth, and acoustic pressure were all extracted from the simulated acoustic field.

Experiment method for cardiovascular vital sign measurement

During the experiment, the volunteers were seated in a quiet, disturbance-free room. Throughout the resting-state measurement process, the volunteers maintained a seated posture. For BP waveform monitoring, data were alternately collected every 2 min using both the Sph and the flexible transducer to eliminate the influence on BP measurements caused by the compression of blood vessels by the Sph cuff. For heart rate monitoring, data were collected simultaneously using both the pulse oximeter and the flexible transducer. For arterial stiffness assessment, the Sph was first used to measure SBP and DBP, followed by the flexible transducer to measure changes in vessel diameter. The average values from multiple tests were then used in eq. S4 and to calculate arterial stiffness.

Measurement and data analysis of the BP

The measurement and data analysis platform diagram were shown in fig. S17. The device was activated by square wave excitation (CTS-8077PR, STNDT) of 100 V with 100-Hz pulse frequency. The echo signal was amplified by 50 dB and then received by a signal acquisition board (PXIe 5172, National Instruments) with a sampling rate of 250 MHz. After that, MATLAB R2020b software was used to digitally filter the signals, thus improving the signal quality. Interpolation functions were applied to upsample the data to 10 times the original resolution. The echo signals at different times were processed using a cross-correlation function to calculate the TOF changes, which were further converted into size parameters of the blood vessels.

The measurement range of the ultrasonic transducer was then characterized by an experimental system consisting of a peristaltic pump (BR8000), digital pressure gauge, and latex tube (inner diameter of 3 mm and outer diameter of 5 mm) (the experimental system diagram was shown in fig. S18B). The pressure in the latex tube changed with the flow rate of the peristaltic pump, causing a variation in the tube diameter, thus simulating the pulsation of human blood vessels. There was a silicone layer between the ultrasonic transducer array and the latex tube to simulate human tissue. Before the experiment, the pressure gauge was used to calibrate the relationship between flow rate and pressure in the latex tube.

First Affiliated Hospital of Xi’an Jiao Tong University approval for human volunteer testing

The conducted human subject experiments were performed in compliance with the protocols that have been approved by the ethics committee of the First Affiliated Hospital of Xi’an Jiao Tong University (XJTU1AF2023LSYY-067). All subjects gave written informed consent before participation in the study. For all demonstrations on human skin, signed consent was obtained from the volunteer.

Statistical analysis

Statistical analysis was performed using SPSS 23.0 and Origin 8.0. Quantitative data were expressed as means ± SDs. Statistical differences in quantitative data between the groups were tested with t-tests. P < 0.05 was considered statistically significant.

Supplementary Materials

This PDF file includes:

Notes S1 to S18

Figs. S1 to S26

Tables S1 to S8

References