Health Monitoring via Heart, Breath, and Korotkoff Sounds by Wearable Piezoelectret Patches

Abstract Real‐time monitoring of vital sounds from cardiovascular and respiratory systems via wearable devices together with modern data analysis schemes have the potential to reveal a variety of health conditions. Here, a flexible piezoelectret sensing system is developed to examine audio physiological signals in an unobtrusive manner, including heart, Korotkoff, and breath sounds. A customized electromagnetic shielding structure is designed for precision and high‐fidelity measurements and several unique physiological sound patterns related to clinical applications are collected and analyzed. At the left chest location for the heart sounds, the S1 and S2 segments related to cardiac systole and diastole conditions, respectively, are successfully extracted and analyzed with good consistency from those of a commercial medical device. At the upper arm location, recorded Korotkoff sounds are used to characterize the systolic and diastolic blood pressure without a doctor or prior calibration. An Omron blood pressure monitor is used to validate these results. The breath sound detections from the lung/ trachea region are achieved a signal‐to‐noise ration comparable to those of a medical recorder, BIOPAC, with pattern classification capabilities for the diagnosis of viable respiratory diseases. Finally, a 6×6 sensor array is used to record heart sounds at different locations of the chest area simultaneously, including the Aortic, Pulmonic, Erb's point, Tricuspid, and Mitral regions in the form of mixed data resulting from the physiological activities of four heart valves. These signals are then separated by the independent component analysis algorithm and individual heart sound components from specific heart valves can reveal their instantaneous behaviors for the accurate diagnosis of heart diseases. The combination of these demonstrations illustrate a new class of wearable healthcare detection system for potentially advanced diagnostic schemes.


Figure S12
. Dynamic sensitivity characterization using the customized circuit.A dynamic sensitivity property with two distinct linear regions are obtained (3.33 V kPa -1 for 0-8 kPa, 0.24 V kPa -1 for 8-18 kPa).A ±15 V supply is used to power the customized circuit, limiting the peak-to-peak value of the output voltage to within 30 V. The output voltage is close to the saturation threshold in the high-pressure region, resulting in a lower sensitivity.

Supplementary Note 1. Frequency response characterization of the piezoelectret sensor
Frequency response is a key parameter of dynamic sensing to acquire the resonant frequency and determine the operating range.Here, the frequency response of the prototype sensor is characterized using the experimental setup in Figure S4, with a varying frequency between 2-2000 Hz and a constant pressure amplitude of 1 kPa.Three different sensors with the same size (2 × 2 cm 2 ) and crisscross structure are involved in this characterization (Figure S14): FEP (piezoelectret) sensor with Corona charging (s1), FEP sensor without Corona charging (s2), and PET (non-piezoelectret) sensor without Corona charging (s3).Sensor s3 is the control group and the outputs mainly come from electronic noises and the triboelectric effect between the modal shaker and sensor electrodes (Figure S15).It is noted that the sensor s2 has a certain output intensity (rather than zero-output) without the Corona charging process, which could be attributed to the residual surface charges on the FEP film because of the friction between FEP and the outside during the fabrication process (Figure S16).Results of the sensor s1 show a first-order resonant frequency of about 740 Hz and a wide working range of 600 Hz as shown by the red line in Figure S14.The short-circuit current measurements from the three sensors with respect to frequency are presented in Figure S17.Furthermore, the frequency response is also characterized using the customized circuit, and similar results are obtained to show the first-order resonant frequency of about 820 Hz and the working range of 600 Hz (Figure S18).frequency plot (Figure S11).However, the peak current depends on both the amplitude and frequency of the pressure, and the linear relationship is obtained in the peak current vs. frequency plot (0-600 Hz in the inset).order Shannon energy envelope of the heart sounds waveform is firstly calculated, and the periodic local maximas of the envelope are considered as the S1 peaks (max_S1).Then, the intersections between the threshold (0.1* max_S1) and the Shannon envelope are sought within 0.15 s before and after the S1 peaks, as the start and end points of S1 during this cardiac cycle.If there is no intersection found within 0.15 s, the point at 0.1 s before or after the S1 peak will be indicated as the start or end point of S1.
Heart sounds are divided into several cardiac cycle segments based on the S1 peaks, and the local maxima point is found for each segment as the S2 peak (max_S2).Subsequently, the process similar to S1 is conducted.That is, the intersections of the threshold (0.1* max_S2) and the Shannon envelope are found within 0.1 s before and after the S2 peak, as the start and end points of S2.If there is no intersection found within 0.1 s, the point at 0.05 s before or after the S2 peak will be forced as the start or end point of S2.          partially overlap, and their waveforms tend to interfere with each other.breath sound will be classified into the category with the highest MFCCs similarity.The detailed algorithm procedure is described as follows: 1. Prepare templates for each breath sound category.In the prototype demonstration, three kinds of breath sounds are used in the classification process: normal breathing, panting after strenuous exercise, and snoring.The volunteer is asked to breathe in these three patterns to obtain the standard breath sound templates (Figure S41, Supporting Information).

Figure S1 .
Figure S1.Fabrication process of the folded double-layer FEP piezoelectret sensor.

Figure S2 .
Figure S2.Characterization of the FEP piezoelectret sensor topography.(a) Photograph of the piezoelectret sensors, highlighting the crisscross grooves structure.Scale bar, 1 cm.(b) Topography characterization of a single-periodic groove in the dotted box in a using the surface profilometer.The FEP film (thickness of 25 μm) is grooved by a laser cutter, and the minimum thickness of the residual film is less than 2 μm.

Figure S3 .
Figure S3.Photographic image of the piezoelectret sensor.(a) Double-layer FEP film placed on the shielding layer (copper tape).(b) The piezoelectret sensor after folding.Unit: cm.

Figure S4 .
Figure S4.The thickness of the piezoelectret sensor.(a) Schematic diagram of the thickness of each layer in the sensor.The thickness dimension of the sensor has been greatly exaggerated.The shielding layer on the right side is opened to facilitate marking the thickness size of each layer.(b) The total thickness of the entire sensor.

Figure S5 .
Figure S5.Working principle of the piezoelectret sensor.The thickness dimension of the sensor has been greatly exaggerated.The external electrode wrapping the entire sensor forms a grounded shield.The shielding layer on the left side is opened to highlight the charge distribution and electrical connection of each layer.The amount of inducted charges on the internal electrodes fluctuate periodically in response to the applied pressure cycles, generating alternating currents in the external circuit.

Figure S6 .
Figure S6.Experimental setup for characterizing the mechanical properties of the piezoelectret sensor.The modal shaker and force sensor mounted on an optical table are used apply pressure with controllable amplitude and frequency to the DUT.The output of the DUT is amplified by the current preamplifier and acquired by the NI USB 6009.

Figure S7 .
Figure S7.Variation of the dynamic sensitivity versus time of the piezoelectret sensor.(a) Results of transferred charges versus applied pressures at the specific date after the initial corona charging.(b) Results of dynamic sensitivity versus time.(i) Low-pressure region; (ii)High-pressure region.The sensitivity attenuation after the initial corona charging process is observed due to the charge neutralization process by the internal defects of FEP.After an initial attenuation of about one week, the dynamic sensitivity of the piezoelectret sensor eventually becomes stable.

Figure S8 .
Figure S8.Transferred charges under different pressures.(a) Low-pressure region of 0-8 kPa.(b) High-pressure region of 8-18 kPa.(c) Transferred charges and (d) dynamic sensitivity of the same piezoelectret sensor in higher pressure regions (18-25 kPa).The dynamic sensitivity of this piezoelectret sensor in the range of 18-25 kPa is comparable to that of 8-18 kPa (275 pC kPa -1 v.s.290 pC kPa -1 ), indicating the potential to operate at higher pressures.

Figure S9 .
Figure S9.Pressure resolution test of the piezoelectret sensor.(a) Applied pressures.(b) Corresponding transferred charges.When the applied pressure is slightly changed around 4255 Pa (P0, which is the reference pressure chosen arbitrarily in the low linearity region), the corresponding transferred charges are measured.Although limited by the experimental setup, the applied pressure fluctuates within tens of Pascal, it can still be considered that a minute change in the applied pressure of 5 Pa causes a significant change in the transferred charge for the statistical results over a period of time.

Figure S10 .
Figure S10.Scheme of the customized circuit.(a) Function modules of the circuit.(b) Detailed circuit diagram.

Figure S11 .
Figure S11.Amplitude-frequency response characterization of the customized circuit.This circuit contains a 50 Hz notch filter and a 2000 Hz low-pass filter with an amplitude gain of about 6 V nC -1 .

Figure S13 .
Figure S13.Pressure resolution characterization.Mixed pressures with a frequency difference of 0.1 Hz have been applied on the DUT and the time/frequency domain results of corresponding transferred charges are obtained.

Figure S14 .
Figure S14.Frequency response characterization.The applied pressure has a variable frequency of 2-2000 Hz and a constant intensity of 1 kPa.Five identical measurements are performed for each frequency point with error bars showing the mean, maximum and minimum.

Figure S15 .
Figure S15.Enlarged view of the control group's amplitude-frequency response curve.

Figure S16 .
Figure S16.Surface potentials of the (a) FEP film with corona charging, (b) FEP film without corona charging and (c) PET film without corona charging.

Figure S17 .
Figure S17.Output peak currents of the three sensors as the function of frequency.The amount of transferred charges is only related to the pressure amplitude, so the flat response results are observed in the working frequency band (0-600 Hz) of the transferred charge vs.

Figure S18 .
Figure S18.Amplitude-frequency response curves measured by the customized circuit.The similar linear working range (0-600 Hz) is obtained, with a gain of about 3.44 V kPa -1 .

Figure S19 .
Figure S19.Mechanical stability test of the piezoelectret sensor.More than 1.1 million cycles of the (a) applied force and (b) corresponding transferred charges for 5000 s.(c) Fluctuation in the amount of transferred charges throughout the test.

Figure S20 .
Figure S20.Improvement in SNR using the shielding layer and electrical ground.A comparative experiment involving 4 sensors has been conducted: (a, i) using the external shielding electrode and the electrical ground, (b, i) using the external shielding electrode and without the electrical ground, (c, i) using a device without the external shielding electrode, and (d, i) a basic circuit with background noises without electrodes (control group).Physiological signals at the heart apex are measured simultaneously.Comparing the (ii) original signals, and (iii) pulse signals (0-10 Hz), (iv) heart sound signals (20-200 Hz) separated by corresponding filters, it can be found that shielding and grounding significantly improve the SNR of physiological signals (especially the weak physiological signals, such as heart sounds).

Figure S21 .
Figure S21.Flexibility presentation and mechanical parameters of the piezoelectret sensor.(a) Bending stability.The sensor is taped on the 3-DOF displacement stage and experiences 2000 'stretching-bending-stretching' cycles.There is no significant difference between the output of before and after the bending test, verifying the good bending stability.Inset: the sensor experiences severe bending and the cross-section becomes semi-circular, indicating the good flexibility of the sensor.(b) Statistical results of the peak charges before and after the bending test under the same applied pressure (220 Hz, 1 kPa).The output charges of more than 10 5 pressure cycles are summarized as the two box-whisker plots.The five horizontal lines from top to bottom of each box are the maximum value, the 25th percentile, the median, the 75th percentile, and the minimum value.(c) The piezoelectret sensor adhered to skin surfaces with different curvatures, and the corresponding pulse waveforms, indicating the good flexibility.(d) Mechanical parameters for each layer of the piezoelectret sensor.TheYoung's modulus of the FEP electret with crisscross cavities is measured by a thermal analysis system (DMA1, Mettler-Toledo, LLC.).The Young's modulus of the copper tape and PI spacer layer is obtained from the datasheet provided by the seller.

Figure S22 .
Figure S22.Continuous results of multiple physiological activities processed by filters of different frequency ranges.The 'Audio waves' in Figure 1c is actually segmental spliced after processing the original signal by different filters to highlight corresponding physiological activities.Here, the continuous filtered results with specific amplitudes are presented.

Figure S23 .
Figure S23.Effects of the exercise on respiration and cardiac activity.(a) Breathing and heart beating patterns before and after the exercise in amplitude (top) and frequency (bottom) plots.(b) The respiratory wave, pulse wave, and heart sound at specific moments (before exercise, after exercise and after rest).

Figure S24 .
Figure S24.Comparison of the amplitude fluctuations of heart sound S1 and S2 before and after the exercise.

Figure S25 .
Figure S25.Time-frequency domain results under increasing respiration intensity at the heart apex.(a) Original signal and (b) filtered signals acquired from the continuous monitoring at the heart apex.(c) STFT spectrum of the original signal at the frequency range of 0-4 Hz, showing the fluctuations of heart beating and breathing.(d) Schematic diagram describing the measurement location.

Figure S26 .
Figure S26.Time-frequency domain results at the neck.(a) Original signal and (b) filtered signals obtained at the neck.(c) STFT spectrum of the original signal at the frequency range of 0-4 Hz.(d) Schematic diagram describing the measurement location.

Figure S27 .
Figure S27.Time-frequency domain results at the brachial artery.(a) Original signal and (b) filtered signals obtained at the brachial artery.The sounds resulting from the heart beating at the brachial artery are also called "Korotkoff sounds".(c) STFT spectrum of the original

Figure S29 .
Figure S29.Further analysis of the cardiac cycle.Length of the (a) cardiac cycle, (b) systolic period and (c) diastolic period acquired from the ECG reference, piezoelectret sensor (HS Sensor) and medical physiological recorder (HS BIOPAC).Correlation between the cardiac cycle and (d) systolic period and (e) diastolic period.The length of the cardiac cycle is mainly affected by the diastolic period when the volunteer kept the resting state.PCC: Pearson correlation coefficient.

Figure S30 .
Figure S30.Simultaneously recorded (a) heart sounds and (b) Korotkoff sounds, and the corresponding STFT spectrums.For each cardiac cycle, the heart sounds contain two components, S1, S2, while the Korotkoff sounds appear to have only one identifiable component.

Figure S31 .
Figure S31.Fluctuation trends of PWV and HSWV under different static pressures.

Figure S32 .
Figure S32.Applied static pressure of 150-40 mmHg by a cuff attached to a mercury manometer.Inset shows the BP reference results from Omron.BP measurement is performed twice using the Omron monitor and the mean is taken as the BP reference result.

Figure S33 .
Figure S33.Calculation of the parameter SNR_K at specific pressures.(a, b) Near the systolic pressure.(c, d) Pressure between the systolic and diastolic pressure.(e, f) Near the diastolic pressure.These data are from i~vi respectively in Figure 3c.

Figure S34 .
Figure S34.Detailed BP measurement results by the Korotkoff sound method for seven volunteers, with all using the same parameter threshold of SNR_K = 2. Insets show the BP results from the commercial BP monitor (Omron BP7211).BP measurement is performed using the Omron monitor before and after each Korotkoff sound method detection, and the average of the twice results is taken as the BP reference value.

Figure S35 .
Figure S35.BP measurement results by the Korotkoff sounds at the radial artery.Insets exhibit the BP reference results from the Omron monitor.

Figure S36 .
Figure S36.Detailed presentation of the radial pulses and Korotkoff sounds from the ninth volunteer (# 9).

Figure S37 .
Figure S37.Comparison of the BP results (a) before and (b) after caffeine consumption for the sixth volunteer (# 6).Insets exhibit the BP reference results from the Omron monitor.

Figure S38 .
Figure S38.Enlarged view of the (a) breath holding and (b) deep breathing fragments in Figure 4a.The frequency bands of heart sounds (20-200 Hz) and breath sounds (100-600 Hz)

Figure S41 .
Figure S41.Prepared templates for three breathing categories of (a) normal breathing, (b) panting after strenuous exercise, and (c) snoring.

Figure S44 .
Figure S44.Algorithm procedure of the breath sound classification.(a) Unknown breath sound is collected and the MFCCs are calculated.(b) The similarity between the MFCCs of unknown breath sound and those of three templates is compared.(c) The parameter σ h 2 /σ v 2 is introduced to quantify the similarity.After the MFCCs results of two breath sounds are mapped into the scatter plot, the two-dimensional Gaussian distribution is used to fit these mapped points, and the variances along the directions parallel to and perpendicular to the diagonal axis are obtained.(d) The variance ratio between the unknown breath sound and the three templates is calculated and the maximum value is found.The unknown breath sound will be classified into the corresponding category with the highest σ h 2 /σ v 2 value.

Figure S45 .
Figure S45.Original results of the parameter σ h 2 /σ v 2 between the collected breath sounds and three templates.(a) 84 cases of normal breathing.(b) 32 cases of panting.(c) 25 cases of

Figure S46 .
Figure S46.Breath sounds classification by DTW algorithm.(a) DTW matrix of MFCC from different breath sounds (containing the three templates corresponding to each kind of breath sounds).A small DTW distance (blue color) means a high similarity, while a large DTW distance (orange color) indicates a high dissimilarity.Three blue squares along the diagonal line illustrate high similarity among the breath sounds from the same pattern.(b) Confusion matrix of the classification results using the DTW distance.(c) F1-scores for the three breathing patterns.

Figure S47 .
Figure S47.Breath sounds classification for human identity recognition.Breath sounds from three volunteers (84 cases for v#1, 42 cases for v#2, and 14 cases for v#3) participated in the classification process.(a) Presentation of distinguishing different volunteers through MFCC.(b) Confusion matrix of the classification results.(c) F1-scores for the three volunteers.

Figure S48 .
Figure S48.Heart sounds sensor array and data acquisition system.(a) Photographic image of the 6×6 sensor array adhered on a flexible printed circuit board (fPCB).(b) Data acquisition system for the sensor array.The output signal of each sensor unit is processed by the corresponding customized circuit (Figure S8) and collected by NI USB-6255 in the differential mode.Independent structure design and circuit composition ensure the low crosstalk among the 36 sensor units.

Figure S50 .
Figure S50.Measurement of the heart sounds distribution for the same volunteer using the medical physiological recorder (BIOPAC ® ).(a) Summarization of the heart sounds results

Figure S51 .
Figure S51.Mapping distribution of the heart sounds and pulses measured simultaneously from the 6×6 sensor array.

Figure S52 .
Figure S52.Heart sounds mapping and unmixing from the second volunteer.(a) Summarization of the heart sounds results acquired from the 36 sensor units.Mapping distributions of (b) S1 and (c) S2 intensities.The original 6×6 results were smoothed by cubic spline interpolation to highlight the position-specific amplitude differences.(d) Demonstration of heart sounds separation by ICA.(i) Original heart sounds from the four valve areas.(ii) Corresponding four valve components after the ICA unmixing.