A. FAST System Overview
According to the Operational Principle, the proposed FAST device consists of three major components, i.e., the high-spatial-resolution and high-sensitivity flexible pressure sensing array for pulse pressure measurement is mounted onto the miniature high-precision step-motor, to which the controlling hardware with the embedded searching algorithm is connected. The FAST system is designed to be a wearable watch shape, as illustrated in Fig. 1e, which offers an easy means to be worn on the wrist by the patients; and meanwhile, can establish a reliable and stable data acquisition interface for arterial pulse waveforms. In particular, the ultrasensitive pressure sensing array, based on the emerging FITS principle, is comprised of 1×8 units, with a unit size of 0.9×4mm² and 0.1mm spacing (Fig. 1f and Supplementary Fig. 1). Benefiting from its minute dimensions and high lateral resolution, the FAST device can simultaneously acquire the real-time pressure pulse data directly above the radial artery without any alignment. Moreover, the high-precision step-motor enables prompt adjustment of the applanation level and can program the static load from 0mmHg to 300mmHg, which covers the standard blood pressure range with the capability of detecting minute variations of the blood pressure to establish the desired applanation level24.
Moreover, the applanation searching process has been achieved by the closed-loop motion system with the undistorted pulse waveforms measured by each sensing unit as the input signals, leading to the precisely controlled movement by the step-motor, as shown in Fig. 1d. Specifically, the pressure signals acquired by the sensing array provide the spatial distribution of the radial arterial pulsation transversely above the vessel, which can be extracted into both the static and pulsed components after passing an analog filter38. The sensing unit exhibiting the highest amplitude of the pulsed signals is selected as the one sitting right above the artery within half a millimeter range, and the pulsed amplitude is used as the control input to determine the desired applanation level by controlling the step-motor. Subsequently, hemodynamic parameters are calculated by combining the applanation level (the static component) and pulse waveform (the pulsed component) and transmitted to the host computer via Bluetooth.
Finally, the above-mentioned electronics and controlling hardware are connected to and controlled by a microcontroller unit, all of which are packed into a compact 3D printed watch case of 90×23×15mm³ in size (Fig. 1g and Supplementary Fig. 2). In brief, integrating the flexible and highly packed pressure sensing array with the high-precision closed-loop control, the FAST device can provide adaptive applanation levels by applying a uniaxial search algorithm in an accurate and reliable manner, from which the arterial pulse waveforms can be recorded and analyzed to obtain critical hemodynamic information.
B. Device Performance of the FITS Array
In order to accurately measure hemodynamic parameters, it is essential for the pressure sensors to precisely capture the fine characteristics of the pulse wave, enabling extraction of cardiovascular parameters from the human body. Therefore, such a device should possess the sensing properties of high dynamic and static sensitivity, exceptional linearity, rapid response time and low-pressure resolution for the accurate detection of slight pressure variation. The design features of the miniaturized sensing unit, array sensing function and isolated functional materials to locate the vessel precisely, as well as the mechanical properties of high flexibility and bending stability for conformal contact on the skin. To fulfill these requirements, a 1×8 sensing array based on FITS mechanism with a tiny spatial resolution of 1mm is developed, named FITS array, as shown in Fig. 2a. The FITS array possesses a simple structure of three stacked layers: the ionic active membrane layer with isolated material pattern functions as the pressure-sensitive film, the flexible interdigital electrode array, and an adhesion layer in between for bonding purposes, as illustrated in Fig. 2b. To prove the outstanding performance of the FITS array on the high-quality pulse wave recording, the mechanism exploration and properties characterizations of the device will be focused in this section.
B1. Mechanical response principle of the FITS array
The sensing properties of the pressure sensor are primarily determined by its mechanism, structure, and materials employed39. The recently developed FITS mechanism provides an emerging alternative with a structurally deformable layer presenting a distinctive super-capacitive ionic-electronic (iontronic) interface containing unit-area capacitance (UAC) as high as several µF/cm², in response to external mechanical stimuli. Notably, the FITS sensor offers superior sensitivity compared to its counterparts, at least three orders of magnitude greater than that of the conventional capacitive counterparts, in addition to its ultrahigh signal-to-noise ratio and anti-interference performance40. Therefore, the iontronic mechanism is selected as the fundamental approach for the high-precise pulse wave measurement in the sensor. Generally, the pressure-sensitive principle of the iontronic sensor can be categorized into the structural-bending model, material-elasticity model and combined model. The structural-bending model relies on the bending deformation of the sensing structure, which always presents higher resilience and repeatability, but lower sensing range and linearity. The material-elasticity model relies on the elastic compression of the micro-structure at the sensing interface, which exhibits controllable sensing range and linearity via the adjustment of the iontronic functional material. But it always faces the problem of low resilience and repeatability for the irreversible deformation of the interfacial micro-structure upon pressure. While, the combined model combines the deformation modality of the former two models, representing the advantages of high resilience, repeatability, sensing range and linearity in one sensing architecture41. As a result, the pressure-sensitive principle of the FITS array follows the combined model. The simplified sensing model of the FITS array is illustrated in Fig. 2c. The ionic active layer, which is prepared by printing ionic coating with a highly coarse surface morphology on PET film, is simply supported on the electrode through the adhesion layers at both edges of the layer. As the pressure is applied on the sensor, the bending of the ionic active layer and the compression of the micro-protrude on the ionic coating will both contribute to the arise of the contact area between the ionic active layer and the electrode, leading to an increase in the electric double layer (EDL) capacitance of the pressure sensor42.
B2. Specifications of the FITS array
The capacitance-pressure (C-P) response of the sensing unit of the FITS array is essential for the evaluation of the sensing performance of the device, for instance the sensitivity, linearity and sensing range. As illustrated in Fig. 2d, through the analysis of the C-P response from 0mmHg to 300mmHg, the sensing unit of the FITS array demonstrates a static sensitivity of 1.19nF/mmHg/cm² with high linearity (R2 = 0.97) within the whole sensing range43. Such sensitivity value is able to neglect the capacitive noise of dozens of pF from the environment and human body, enabling high resolution to distinguish the slight pressure variation of 0.177mmHg, which satisfies the requirement of recording the fine characteristics of the pulse wave44, as shown in Fig. 2e. Notably, through the analysis of the 4 cycles of the successive C-P response test in Fig. 2d, the FITS array demonstrates the extremely high repeatability of 0.44%, which has achieved the repeatability standard of a load cell according to GB/T 7551 − 2008, ensuring the accurate detection of the pressure applied to the sensor for medical-grade hemodynamic monitoring.
The dynamic sensitivity ensures that the sensor is able to reproduce the individual frequency components of the dynamic pulse wave precisely and accurately during hemodynamic measurements, which is highly important for high-fidelity pulse wave recording45. Commonly, the major energy of a pulse wave is less than 5 Hz46,47. In order to assess the dynamic sensitivity of the FITS array for pulse wave detection, the physical stimulus with different frequencies varied from 1 Hz to 10 Hz generated using a piezoelectric actuator are applied to the sensor. The results shown in Fig. 2f reveal that less than 1% output differences are found under different stimulus frequencies, indicating the output and sensitivity of the sensor are the same under static and dynamic (< 10Hz) load. This can be explained by the low response time of the FITS device. As shown in Fig. 2g, benefiting from the high reliance on the combined model, the response and reset times of the FITS array are obtained as 11.50ms and 11.75ms, respectively, which have far exceeded the requirements on the response time under 10Hz stimulus. As a result, the FITS array with high static and dynamic sensitivity, high-pressure resolution and low response time has met the requirements of precisely and stably capturing the fine characteristics of the pulse wave from the sensor characterization results.
Additional characterization tests, for instance the long-term stability of the FITS array (Supplementary Fig. 3), the C-P responses using different ionic coating (Supplementary Fig. 4), the capacitance-voltage relationship curve (Supplementary Fig. 5), the micro-morphologies (Supplementary Fig. 6) and AFM surface profile images (Supplementary Fig. 7) of the ionic coating, as well as the stickiness of the ionic coating under pressure (Supplementary Fig. 8) are shown and discussed in Supplementary Materials.
B3. Bending stability of the FITS array
Bending stability is of the essence for the FITS array because of the uneven surface of the wrist. To evaluate the bending stability of the flexible sensor, the FITS array is wrapped on various molds with different curvatures that resemble those of the human wrist (Fig. 2h). Consequently, the bending stability of the sensor can be assessed by comparing the capacitance-to-pressure curves measured on different molds. The results are illustrated in Fig. 2i, as pressure gradually increases to 250 mmHg, there are only slight changes in capacitive outputs between bending and non-bending states. Notably, less than 3% of relative differences are observed when the sensor is bent with curvatures ranging from 1.5mm to 5mm compared with the non-bending state, as shown in Fig. 2j, revealing the exceptional bending stability of the FITS array sensor. This property of bending-insensitivity can be derived from the specific materials selection and design optimization of the FITS array. On one hand, the utilization of membranous building materials with high flexibility and low thickness effectively mitigates the generated bending stress. On the other hand, the pressure-sensitive interface is designed near the neutral plane of the device to further attenuate the influence of bending stress on such interface48.
B4. Crosstalk prevention of the FITS array
In order to further improve the accuracy of the FITS array and to avoid misleading and inaccurate stimulus information due to crosstal47,48, the ionic coating of the ionic active layer is patterned into eight separated regions, which would significantly reduce the crosstalk between the units. To verify the effectiveness of this processing step, we have conducted experiments by comparing two types of sensors: one with an intact ionic coating and another with a separated one. To assess the crosstalk effect generated by neighboring channels, the fourth units of both sensors are subjected to a pressure of 25 kPa individually. As shown in Fig. 2k, the FITS array using intact ionic coating exhibits simultaneous outputs across all seven channels when pressure is applied to the fourth channel, resulting in a maximum crosstalk of 57% (defined as the signal amplitude of the neighboring sensor in reference to the one under loading)49. Whereas, as shown in Fig. 2l, the sensor used in the FAST system, with separated ionic coating, displays minimal output from all channels except the one under pressure, with a maximum crosstalk of only approximately 0.2%. This observation emphasizes the significant capability of the separated ionic coating to eliminate crosstalk in the FITS array, underscoring its paramount importance in practical applications.
C. Radial arterial pulse simulator
A radial arterial pulse simulator has been developed for the purpose of efficient characterization of the FAST system. By mimicking the radial arterial pulse waveforms, it allows fine adjustments on the dynamic characteristics of the waveforms and provides a standard testing platform to evaluate the performance of the FAST device. Notedly, the commercial noninvasive blood pressure simulator can only offer three features, i.e., pulse wave begin (PWB), pulse wave systolic peak (PWSP), and pulse wave end (PWE) of the radial arterial pulse waveforms, while our radial arterial pulse simulator is intended to provide the five major distinctive features, adding pulse diastolic notch (PDN) and pulse wave diastolic peaks (PWDP). Both PDN and PWDP are of critical importance in determining hemodynamic dynamics50. To this end, we have modified the commercial blood pressure simulator (Fluke BP Pump) by connecting a pneumatic pump to set the static pressure and adding another programmable branch to provide pneumatic oscillation synchronized with the primary simulated waveform.
In particular, the Fluke device can offer the simulated radial pulse waveforms with the primary features, including the controllable HR, SBP and DBP values. Moreover, by introducing PDN and PWDP features in the modified waveforms, the waveform area under the systolic curve can be directly analyzed, from which the CO, along with other key hemodynamic parameters, can be simultaneously assessed13. As a result, the radial arterial pulse simulator can offer the simulated pulse waveforms with the five key pulse waveform features, from which the corresponding hemodynamic parameters can be extracted.
Figure 3a shows a device illustration of the pulse simulator along with its control block diagram in Fig. 3b. As can be seen, the pressure sensor monitors the air pressure in the circuit in real time, while the Pump No.1 plays a role in setting up the static pressure for the pulse simulator, so that the commercial blood pressure simulator produces the waveform with the three primary features. Once the falling edge of the primary waveform is detected by the pressure sensor, the control unit in the programmable branch output will subsequently turn on the solenoid valve with a precise time delay, from which the Pump No.2 can release the two supplemental pulse features of PDN and PWDP. Thus, the tubes of the programmable branch outlet and the commercial simulator are connected to form the output of the radial arterial pulse simulator, allowing it to output artificial pulse waves with five programmable features. In order to mimic the practical scenario for the FAST device verification, the outlet of the pulse simulator is finally connected to an artificial model of the human hand. Specifically, the tube of the simulator’s outlet is linked with an artificial blood vessel, which is supported by the artificial wrist bone and surrounded by the silicone artificial skin. Then, the pulse waveform can result in the deformation of the blood vessel and subsequently oscillation of the artificial skin to simulate the physiological characteristics of the arterial pressure pulse.
Figure 3c compares the simulated pulse waveforms generated with the three characteristic features identified by the Fluke device (from 0-4sec), along with those produced by the modified arterial pulse simulator with the five distinctive features marked (from 4-8sec). In order to verify that the pulse simulator can simulate different physiological characteristics of the human body and provide a functionally rich test environment, three sets of experiments have been conducted, during which the simulated pulse waveforms have been recorded by the pressure sensor. We have verified the simulator's capability to generate the pulse waveforms with different heart rates, blood pressures and interval times between the systolic peak and the dicrotic notch of the waveforms. Specifically, each of the above parameters was varied three times while maintaining all other conditions undisturbed.
As shown in Fig. 3d, the Fluke device is able to directly set the heart rates with 1BPM increment. The artificial pulse waveforms at typical heart rates of, e.g., 60BPM, 70BPM, 80BPM, can be obtained. Figure 3e shows highly reproducible results from the above three heart rate cases (Cases1-3), in which the mean absolute errors (MAEs) between the expected and actual beat-to-beat values of the heart rate are 0.10BPM, 0.37BPM and 0.85BPM, respectively, demonstrating the high accuracy of the simulator. Furthermore, with the combined output of the Fluke device and Pump No.1, the simulator can also set both the systolic and diastolic blood pressures. Figure 3f exhibits the artificial pulse waveforms at different systolic and diastolic blood pressures, e.g., 100/60mmHg, 120/80mmHg, and 150/100mmHg. Similarly, as shown in Fig. 3g, the experimental results suggest highly reproducible and accurate outcomes in the three blood pressure measurements (Cases4-6), in which the MAEs between the expected and actual beat-to-beat values of the systolic and diastolic blood pressure are 0.20/0.78mmHg, 0.34/0.37mmHg and 0.78/0.34mmHg, respectively. In addition, by adjusting the pressure output of the Pump No.2 and the timing of the valve open, the simulator can also alter interval times between the systolic peak and the dicrotic notch, as illustrated in Fig. 3h. In particular, Fig. 3i illustrates the repeatability of the interval time in each of the three cases (Cases7-9), of which the MAEs are 1.08ms, 0.42ms and 3.92ms, respectively, thus demonstrating the stability of this pulse simulator. According to the above results, the simulator can steadily produce pulse waveforms that resemble the physiological characteristics of human hemodynamics, making it a convenient and reliable characterization system for the FAST device and other similar pulse wave tracking wearables.
D. Determination of radial arterial location
To evaluate the locating accuracy of the FITS array, the FAST device is randomly mounted on the radial arterial pulse simulator with a known vessel location of 3mm (Fig. 3b). And then, the pressure data from the skin are collected from the multi-channel sensor array for analysis, from which the location of the desired arterial position could be determined. The results are further used to compare with the physical position of the vessel in the artificial model (Supplementary Fig. 9). The experimental results are summarized in Figs. 3j-k. The radial pulse wave amplitudes acquired by different sensing units, i.e., channels, are depicted in the histogram when the FAST device is placed at the different positions over the pulse simulator. Notably, the pulse wave amplitudes vary across the serial FAST channels and are highly dependent on the relative position to the arterial location. In the three test cases, we have intended to place the FITS array to the left, middle, and right of the simulated radial artery (Fig. 3j), respectively, and compared the computed location of the artery from the pulse wave signals with the actual location of the artificial artery. As expected, the channels closer to the arterial location exhibit higher signal amplitudes, as compared to those further away from the radial artery. As shown in Fig. 3k, in the three test cases, Channels 6, 4 and 2 are determined to be the closest units to the arterial location, while the relative errors from the closest units to the center of the radial artery are 0.3mm, -0.4mm and 0.1mm, respectively. Thus, the experiment proves that the expected locating precision is within half of the spatial resolution of the sensor array, i.e., 0.5mm. Taking into account the diameter of the radial artery and the feature size of the sensing unit, the computed closest unit can be guaranteed to be positioned directly above the radial artery. In brief, the FAST device is capable of automated search for the desired arterial position for optimal pulse waveform detection.
E. Establishment of the targeted applanation level
Following the determination of the radial arterial position, the applanation level needs to be established to obtain high-quality pulse wave signals, based on the arterial tonometry principle24. In order to assess the quality of the pulse waveform at the desired applanation level, we have performed a validation experiment on 3 healthy volunteers and each subject test has been repeated five times, in terms of measurement accuracy and repeatability. Supplementary Fig. 10 illustrates that the arterial pulse waves are spatially measured in real-time with the FAST device and compared with that from the standard tonometry approach (TL-400) in both time and frequency domains from one subject. In the time domain analysis, the applanation is applied by the step-motor and the pulse waveforms at different pressure states are selected to compare with that from the TL-400 with the correlation coefficients calculated. Then, the measurement accuracy can be evaluated by comparing the statistical differences of both the pulse waveforms collected at the desired applanation level by the FAST device and the TL-400 as well as the key characteristic features extracted from the waveforms18. Moreover, the first six harmonics from the Fourier transform, covering the vast majority of the waveform energy, can be used to assess the medical-grade accuracy of the FAST device in the frequency domain51.
Figure 4a plots the pressure measurement results during the search of the applanation level, in which the solid blue line is generated from the sensor readings and the dashed line is the calculated average values, representing the applanation level. As expected, when the step-motor moves gradually downwards, the applanation pressure rises accordingly. Subsequently, the pulse waveforms at different pressure levels are extracted (i.e., the thin green line) and compared with the applanated arterial pulse waveform measured by the TL-400. As can be seen, at either a low pressure of 42mmHg (Fig. 4b) or a high pressure of 126mmHg (Fig. 4d), the pulse waveforms become distorted with reduced correlations, i.e., 0.914 and 0.938, to the controlled assessments, respectively. While at the applanation level of 87mmHg (Fig. 4c), the pulse measurements show high consistence of the correlation coefficient of 0.988 with the standard waveforms, reflecting the capacity of the FAST device to capture the applanated arterial pulse waveform accurately. Figure 4e exhibits the simultaneous signal collection of the arterial pulse waveforms, one from the FAST device at the desired applanation level while the other from the TL-400. As exhibited, the characteristic waveforms measured by the FAST device have shown no appreciable differences as compared to the values measured by the TL-400 in the time-domain sequence. And then, the normalized pulse waveforms in each single period of the 5 repeated experiments have been superimposed in Fig. 4f with the average values drawn in the solid lines and the standard deviations highlighted in the colored shades. The relative errors have been calculated from 0.017 to 0.045 from the two normalized pulse waveforms, demonstrating the beat-to-beat accuracy of the FAST measurements in a qualitative manner. Moreover, the standard deviations among the 5 repeated experiments are also calculated between 0.030 and 0.120 for the FAST device, which is in a good agreement with that obtained by the TL-400, i.e., ranging from 0.020 to 0.082. It reflects the high reliability and repeatability of the FAST measurements. Furthermore, the key characteristic features extracted from the FAST device and the TL-400 measurements have also been compared. Figure 4g presents the comparison results of the five characteristic features extracted from the pulse waveform of one subject with all their average values and standard deviations. As can be seen, the comparison sets of the averaged five features are measured as 80.16mmHg/77.83mmHg (of PWB), 122.49mmHg/119.05mmHg (of PWSP), 93.63 mmHg/92.75mmHg (of PDN), 99.89mmHg/97.86mmHg (of PWDP) and 80.16mmHg/77.83mmHg (of PWE). Among the 5 comparison sets, the largest difference calculated as 3.44mmHg is considered within the widely accepted AAMI SP10 Protocol 51 (i.e., the standard error of BP monitoring device’s error tolerance is as high as 5mmHg). Finally, the frequency-domain analysis has been performed by splitting pulse waveform contours into harmonics. Figure 4h illustrates the analytical results from the Fourier transform with the first six harmonics. As can be seen, the results of the first six harmonics are measured as 40.68mmHg/40.55mmHg (of 1st harmonic), 36.21mmHg/36.22mmHg (of 2nd harmonic), 19.26mmHg/17.66mmHg (of 3rd harmonic), 11.29mmHg/12.261mmHg (of 4th harmonic), 4.22mmHg/3.91mmHg (of 5th harmonic) and 3.41mmHg/3.13mmHg (of 6th harmonic). As a conclusion, the comparison of these harmonic values reveals no substantial difference statistically (< 10%) between the two techniques, which again confirms the medical-grade precision of the FAST measurements. Similar findings can be also found from the measurements conducted on the other subjects which are included in the Supplementary Materials (Supplementary Figs. 11–12). Overall, the FAST measurement results have been all within the relevant medical standards in both time and frequency domains, confirming its high accuracy and high repeatability, in comparison with that of the clinical TL-400 tonometric device.
F. Self-calibration algorithm for continuous BP monitoring
Once the optimal arterial pulse waveforms are extracted from the tonometric device, the calibration step becomes necessary to align the signal magnitude to that of the intra-arterial pressure52. It is an important step to achieve continuous BP monitoring with the medical-grade precision. Remarkably, the FAST system can carry out its calibration protocol during the applanation process without the need of any additional equipment or procedure. Figure 4a plots the oscillometric envelope generated from the Gaussian-fitting curve of the above data sets (i.e., the thin green line), from which the MBP, SBP and DBP can be calculated. Specifically, the MBP represents the peak point of the oscillometric envelope (i.e., the thick green line). Accordingly, the SBP and DBP can be quantitatively predicted by the principle of oscillometric blood pressure measurement. This data analysis completes the entire self-calibration process.
To validate the measurement accuracy of the self-calibration feature of the FAST device, we have conducted a comparison study with 10 different subjects between the FAST system and a medical-grade pressure cuff device (HEM-7136, Omron, Kyoto, Japan). The traditional cuff sphygmomanometer is based on the oscillometric method33. Figure 4i displays the data sets (represented by blue dots) measured by the FAST and Omron devices in a regression model. The data corresponds to MBP values ranging from 70mmHg to 130mmHg and exhibits a high correlation coefficient of 0.9884. This highly linear relationship implies that the FAST device possesses high accuracy comparable to that of the medical-grade devices in terms of blood pressure calibration. In brief, following the calibration procedure, the FAST system can be utilized to conduct continuous hemodynamic monitoring with direct access to the calibrated values and resulted high accuracy.
G. Validation of continuous monitoring of BP, CO and HR on human subjects
The high-precision arterial pulse waveforms detected by the FAST device can subsequently be utilized to compute key hemodynamic parameters, including BP, HR, and CO, in a continuous manner as described in the principle section36. As shown in Fig. 5a, to validate continuous beat-to-beat blood pressure monitoring, we have applied the FAST device and the medical-grade TL-400 device simultaneously to 3 healthy subjects, worn on the right and left hands, respectively. In addition, the standard 3-lead ECG electrodes have been applied onto the RA (right arm), LA (left arm) and LL (left leg) of the subjects, which is considered the clinical gold standard for heart rate monitoring22,53. All the tests have been performed under an IRB protocol approved by the University of Science and Technology of China (YXLLSH-SQ-2023-06-01). In order to verify the measurement accuracy of the FAST system, several physiological maneuvers, including Valsalva maneuver (Vals), passive leg raising (PLR) and air cycling (AC), have been performed by the subjects within the 10 min testing period, which can induce intended changes in BP and other hemodynamic parameters.
The evaluation of BP measurements has been conducted first. As can be seen, Fig. 5b plots DBP (bottom), SBP (top) and MBP (middle) data continuously measured by the FAST device and the TL-400 from one health subject. The remaining data from other subjects can be found in Supplementary Figs. 13–14 in the Supplementary Materials. Specifically, at the beginning (0-120s), the subject keeps still, while the blood pressure remains almost constant. Subsequently, the subject engages in 3 physiological maneuvers in sequence, i.e., the Valsava breathing between 120-180s, the passive leg raising between 240-360s, and the air cycling between 420-540s. All these maneuvers can induce blood pressure to climb35,54,55. Thus, the hemodynamics of the subject experience considerable variations during the procedures, of which the MBP ranges from 88mmHg to 144mmHg. It also demonstrates the real-time tracking capability of the FAST system comparable to that of the medical-grade device. To further assess the accuracy and precision of the beat-to-beat hemodynamics, Bland–Altman analysis is subsequently conducted to assess the measurement differences of MBP, DBP and SBP between the FAST device and the TL-400. Figures 5c-e exhibit the extracted mean biases from the comparison tests of MBP, DBP and SBP between the FAST device and the TL-400, the dashed line in which represents the 95% confidence interval (CI) of the limits of agreement. In particular, the means of the differences and the standard deviations of DBP, SBP and MBP are 2.93 ± 6.20mmHg, 4.98 ± 6.10mmHg and 3.05 ± 5.99mmHg, respectively, which are well in accordance with the AAMI criteria (5 ± 8mmHg)56. Therefore, the comparison results suggest that the FAST system is capable of measuring BP within the medical-grade precision.
In the following data analysis, the heart rate monitoring by the FAST device is assessed, by comparing that of the simultaneous ECG measurements. Figure 5f displays the HR values measured continuously by the FAST and the ECG devices from the same person above, while the remaining data from other subjects can be found in Supplementary Figs. 13–14 in the Supplementary Materials. As the subjects conduct the three physiological maneuvers in consequence, the heart rate values fluctuate significantly, accompanied by a rising trend, which gives the subject a wide range of heart rate variations from 70BPM to 104BPM. This shows that the ability of the FAST device to track heart rates is comparable to that of the medical-grade device. To verify the accuracy and precision of the beat-to-beat HR, the 10-minute continuous data from both the FAST and the ECG devices are plotted in Fig. 5g. The linear correlation analysis shows a high correlation with a correlation coefficient of 0.98. Moreover, the mean of the differences and its standard deviation are − 0.50 ± 1.71 BPM between the FAST system and conventional ECG measurements. These results indicate that the HR values calculated from the FAST device have a high statistical precision compared to that of the clinical gold standard and are compliant with the published standard57 (ANSI/AAMI EC13:2002), of which the acceptable accuracy is within ± 5BPM.
Finally, to assess the measurement accuracy of the CO, the same datasets acquired by the FAST device and the TL-400 have been again applied. As shown in Fig. 5h, several characteristic features have been extracted from the pulse waveform, such as SBP, DBP, MBP and pulse pressure (PP), along with the relevant predetermined biometric information (i.e., gender, age, height, and weight) can be utilized to predict the CO values through the non-linear regression model13. Figure 5i shows the continuous CO values calculated by the FAST device and the TL-400 from the same test subject above (and the remaining data from other subjects can be found in Supplementary Figs. 13–14 in the Supplementary Materials). Similar to the BP and HR analyses, the CO value of the subject has experienced appreciable fluctuations during the 3 physiological maneuvers. Specifically, the Valsava breathing leads to a decreasing trend of the CO, while the air cycling along with passive leg raising spurs an increase in the CO values, which has been all observed by the FAST device and the TL-400. Furthermore, the Bland–Altman analysis of CO measurement has been again conducted to evaluate the accuracy of the FAST system compared with the medical-grade device, as demonstrated in Fig. 5j. The mean of the differences and the standard deviations between the FAST device and the TL-400 are − 0.01 ± 0.94L/min. Besides, to meet the acceptable clinical agreement of the CO measurement devices, which is defined by Critchley58, percentage error (PE) is introduced to assess the accuracy of CO calculation of the FAST system. The PE of the CO values between the FAST device and the TL-400 is 27.8%, which stays below the clinical threshold of 30%, demonstrating the CO measurement of the FAST device can achieve the relevant medical-grade standard.
Besides the key hemodynamic parameters, i.e., BP, HR and CO, the results of other hemodynamic parameters, i.e., SV, SVI, CI, etc. can also be found in Supplementary Figs. 15–17 in the Supplementary Materials. In summary, the proposed FAST system, as a truly wearable tonometric device, is capable of continuously tracking key hemodynamic parameters, including BP, HR and CO, within the medical-grade accuracy.