Wearable flexible electronics is one of the development trends in medical-health monitoring, particularly cardiovascular-disease monitoring. Pulse wave is an important source of information for assessing cardiovascular health; however, it is a non-stationary weak signal and imposes high requirements on the sensitivity and stability of detection. To solve the key technical problems of wearable health monitoring, a graphene-based flexible pressure sensor with a multilevel branched microstructure is designed and developed in this study, which significantly improves the sensing performance of pulse waves and forms the foundation for a wearable flexible pressure sensor. The sensing health-monitoring system is developed using a sensing-like cuffless blood-pressure-monitoring algorithm based on single-point radial artery pulse waves. The prediction errors of the system for human systolic blood pressure (SBP) and diastolic blood pressure (DBP) are (0.7±10.5) mmHg and (0.5±6.1) mmHg, respectively. The findings of this study can provide important technical support for cardiovascular health-monitoring systems and application research, as well as for wearable precision medical-health monitoring.

Flexible pressure sensors based on array structures exhibit key technical issues such as difficulty in achieving high sensitivity and a wide pressure-detection range, as well as limited usage. Hence, a hierarchical branch (HB) structure pressure-sensor design scheme is proposed in this study to improve the performance of array-microstructure pressure sensors. First, we completed the design of the HB structure via finite-element analysis. Results of the finite-element analysis reveal the unique effect of the HB structure: it not only includes the elastic-modulus- reduction effect caused by the superposition of multiple elastic layers but also integrates the pressure-diffusion effect caused by the HB structure, thus realizing the gradual activation and further strengthening of the active-layer conduction path. As such, the deformation range of the elastic layer (sensor pressure-detection range) and the deformation sensitivity to pressure (sensor sensitivity) can be improved. To solve the problem wherein the existing blood-pressure-detection equipment requires cuff pressurization and continuous blood-pressure monitoring cannot be achieved easily without pressurization interference, we construct a cuffless blood-pressure detection model, the class-aware model based on the Moens-Korteweg (M-K) equation and Transformers (the CAMKformer model). This model incorporates the idea of ​​cascade learning, uses the basic formula for blood-pressure calculation based on pulse-wave conduction velocity as the principle, and applies the Transformer model to classify the input pulse wave for blood pressure, thus forming a two-stage cuffless blood-pressure detection model. Compared with conventional machine-learning algorithms based only on formulas or tree models, this model combines formula-related features with original pulse-wave data, where the complex feature-extraction capabilities of deep-learning models and the strong interpretability of theoretical models are fully utilized. In addition to affording high robustness, it integrates multimodal blood-pressure related information (discrete pulse-wave characteristics and continuous pulse-wave data), thus significantly reducing the blood-pressure detection errors inherent in conventional research methods.

Experimental results show that the HB structure enables flexible pressure sensors based on array microstructures to simultaneously improve sensitivity (an increase by 14 times, which is more significant than that of previously published single-layer structure strategies) and the linear range (Fig.6). Additionally, the HB structural strategy based on template and multilayer superposition methods offers significant advantages in terms of structural uniformity, adjustability, and scalability. For example, molds can be fabricated via highly controllable processes (such as photolithography), thus allowing parameters such as structural shape and size to be adjusted. We believe that the HB strategy can be used as a general strategy to adjust mechanical-stress transfer and optimize sensor performance, as well as exhibits broad application prospects in sensor design. A diverse database can further demonstrate the robustness and generalizability of the CAMKformer model. The results show that the wearable system and CAMKformer model constructed in this study can adapt promptly to the pulse-wave characteristics of different individuals and accurately detect human SBP and DBP [with errors of (0.7±10.5) mmHg and (0.5±6.1) mmHg, respectively, as shown in Table 3]. Different from the pressurized blood-pressure monitoring method of conventional electronic sphygmomanometers and commercial blood-pressure measurement smart watches, the abovementioned system does not require pressure application to the user’s radial artery to detect blood pressure; hence, it is suitable for continuously measuring the user’s blood pressure during daily activities or at night, as blood pressure changes during sleep. In addition, this model uses a single-cycle pulse wave as input, presents a simple system configuration, and is highly flexible for use.

In this study, wearable flexible electronic technology and medical-health monitoring are adopted as the research background. The requirements of wearable cardiovascular health monitoring are identified, and the associated principles, devices, systems, and application levels are investigated systematically. First, a design scheme for a HB structure flexible pressure sensor is proposed, and a graphene HB structure flexible pressure sensor is reconstructed simultaneously with a pulse-wave measurement system. Experimental results show that the bionic HB structure strategy enhances the sensitivity (an increase by 14 times) and linear range (4.6 times expansion) of the array-based microstructure pressure sensor, thus enabling distortion-free and accurate measurements of pulse waves. Subsequently, a class-aware cuffless blood-pressure-detection model is established. This model, which is based on the abovementioned flexible pressure sensor, obtains single-point pulse waves of the radial artery. Additionally, it uses a deep-learning model based on the Transformer for blood-pressure classification and a theoretical model based on the M-K equation for blood-pressure prediction. Compared with conventional machine-learning algorithms based only on formulas or tree models, this algorithm combines commonly used pulse features with original pulse-wave data, where the complex feature-extraction capabilities of deep-learning models and the strong interpretability and robustness of theoretical models are fully exploited. Owing to its high stickiness, it realizes the fusion of multimodal pulse-related information and significantly reduces the error of cuffless blood-pressure detection [the errors are (0.7±10.5) mmHg and (0.5±6.1) mmHg for SBP and DBP, respectively], thus satisfying the international standards for non-invasive blood-pressure monitors. The cuffless blood-pressure monitoring system proposed herein is devoid of external pressure interference. Additionally, it is expected to transform the blood-pressure dynamic detection mode from “single-point, high-dispersion transient detection” to “multipoint, low-dispersion online monitoring,” thus facilitating users or doctors in dynamically monitoring blood-pressure changes to achieve early proactive screening of hypertension and improve hypertension early-warning and monitoring capabilities.