Adaptive heart shock signal measurement and feature extraction method and device based on smartphone imu

Abstract

This invention belongs to the field of biosignal processing and mobile health monitoring, and relates to an adaptive cardiac oscillation signal measurement and feature extraction method and device based on a smartphone IMU. The smartphone is placed against the user's chest, and the built-in IMU simultaneously acquires three-axis acceleration and three-axis angular velocity signals. After preprocessing, the signal-to-noise ratio, periodicity index, and kurtosis characteristics of the six-axis signals are calculated for quality assessment. Once the quality meets the standards, a weighted scoring mechanism adaptively selects the optimal signal axis or performs multi-axis weighted fusion. Complementary filtering is used to fuse accelerometer and gyroscope signals to reconstruct the cardiac oscillation SCG signal. The amplitude envelope is extracted, and AO feature point detection is achieved by combining adaptive thresholding and cardiac cycle constraints. Key cardiac oscillation feature parameters are extracted based on the AO points. This invention solves the problems of difficult signal axis selection due to attitude changes and unstable feature point detection under low signal-to-noise ratio in smartphone cardiac oscillation measurement, realizing convenient attitude-independent cardiac oscillation measurement.

Description

Technical Field

[0001] This invention belongs to the field of biosignal processing and mobile health monitoring, and relates to an adaptive method and device for measuring and extracting features of oscillatory signals based on a smartphone IMU. Background Technology

[0002] Routine monitoring of cardiac function is crucial for cardiac health management. Traditional methods of assessing cardiac function, such as electrocardiograms (ECG) and echocardiography, typically require professionals using specialized equipment, which presents significant challenges, including high equipment costs, complex procedures, and limitations in testing scenarios. This hinders the widespread adoption of routine cardiac function monitoring and early screening in homes and at the grassroots level.

[0003] Seismography, a non-invasive technique for assessing cardiac function by measuring the mechanical vibrations of the heart, has received increasing attention in recent years. The seismogram signal consists of micro-vibrations in the thoracic cavity generated during the heart's contraction and relaxation, reflecting key information about cardiac mechanical function such as myocardial contractility, valve opening and closing status, and hemodynamic changes. The aortic valve opening point (AO point, marking the beginning of ventricular ejection of blood into the aorta) is one of the most important feature points in the seismogram signal; its accurate detection is crucial for accurate heart rate calculation, ejection time measurement, and assessment of cardiac systolic function.

[0004] Currently, seismogram acquisition mainly relies on dedicated high-sensitivity accelerometers, which require precise fixation of the sensor at a specific location on the sternum and a specific supine posture for the subject. This acquisition method has the following drawbacks: First, it requires dedicated hardware, increasing the cost and operational barrier, hindering widespread application; second, the sensor's placement and orientation have a significant impact on signal quality, with even slight deviations potentially causing significant changes in the signal waveform, affecting measurement repeatability; third, it imposes strict requirements on the user's posture, making it inflexible for everyday use.

[0005] In recent years, inertial measurement units (IMUs) built into smartphones have made significant progress in accuracy and sampling rate, and are now capable of sensing micro-vibrations in the chest cavity. Existing research has shown that the accelerometers built into smartphones can detect surface vibration signals caused by heartbeats. However, as a general-purpose consumer electronic device, smartphones face a core technological challenge during use: users hold their phones in vastly different ways and at varying angles, such as placing the phone flat against the chest, holding it against the chest, or placing it on the abdomen. This causes the cardiac vibration signal components captured by the IMU along each axis to vary drastically with the phone's placement angle and position. In a specific posture, the main vibration component of the cardiac vibration signal may be concentrated on the X-axis, while in another posture it may shift to the Y-axis or Z-axis. A fixed single-axis signal extraction strategy will inevitably lead to signal loss or a decrease in the signal-to-noise ratio.

[0006] Furthermore, in practical use, users may experience poor signal quality due to improper placement without being aware of it in time, resulting in unreliable measurement results. Current technology lacks a mechanism for real-time evaluation and feedback of the acquired signal quality.

[0007] In addition, existing methods for detecting AO feature points in cardiac oscillation signals are mostly based on signal design obtained by dedicated sensors under standardized acquisition conditions, which are not adaptable to low signal-to-noise ratio and morphologically deformable signals obtained by smartphone IMUs under non-ideal acquisition conditions.

[0008] In summary, the following problems exist in the existing technology: (1) There is a lack of an adaptive signal acquisition mechanism to cope with multi-pose measurement of smartphones; (2) There is a lack of an instant evaluation and feedback mechanism for the quality of acquired signals, and users cannot make timely adjustments during the measurement process; (3) There is a lack of robust AO feature point detection algorithms for the low signal-to-noise ratio signal characteristics of smartphone IMUs; (4) A complete and automated processing flow from the raw IMU signal to the output of cardiac vibration feature parameters has not yet been formed. Summary of the Invention

[0009] This invention aims to overcome the problems of existing seismogram measurement technologies, such as reliance on dedicated equipment, strict sensor positioning requirements, user posture limitations, and difficulties in signal axis selection and lack of signal quality feedback in smartphone IMU measurements. To address these limitations, this invention proposes an adaptive seismogram signal measurement and feature extraction method and device based on a smartphone IMU. The method involves placing the smartphone against the chest and simultaneously acquiring three-axis motion signals using a built-in accelerometer and gyroscope. The acquired signals are evaluated in real-time; if the quality is substandard, the user is prompted to adjust and re-acquire the signal. Once the quality meets the standard, an adaptive axis selection mechanism driven by signal quality automatically adapts to different phone placement angles and user postures. Complementary filtering is used to fuse the built-in accelerometer and gyroscope signals to reconstruct the seismogram signal. Robust detection of AO feature points is achieved using a combination of Hilbert envelope analysis and cardiac cycle physiological constraints. Finally, key seismogram feature parameters such as heart rate, cardiac cycle duration, and ejection time are output. This invention effectively lowers the equipment threshold for seismogram measurement, enabling convenient cardiac function monitoring anytime, anywhere.

[0010] The specific technical solution of this invention is as follows: An adaptive method for measuring and extracting features of oscillatory signals based on a smartphone IMU includes the following steps: Step 1: Place the smartphone against the user's chest and use the smartphone's built-in inertial measurement unit (IMU) to simultaneously acquire triaxial acceleration and triaxial angular velocity signals.

[0011] Step 2: Signal preprocessing; The acquired triaxial acceleration and triaxial angular velocity signals are processed to remove DC components, and drift interference and noise interference are filtered out by a bandpass filter to obtain preprocessed triaxial acceleration and triaxial angular velocity signals. The bandpass filter is a Butterworth bandpass filter with a passband frequency range of 1~40Hz.

[0012] Step 3: Signal quality assessment; The preprocessed triaxial acceleration and triaxial angular velocity signals are evaluated for quality. Three quality indicators, namely signal-to-noise ratio, periodicity index, and kurtosis, are calculated for the triaxial acceleration and triaxial angular velocity signals. The results are used to determine whether the currently acquired signal meets the preset quality threshold requirements. If the quality threshold is not met, the user is prompted through the smartphone interface to adjust the phone's placement or measurement posture and reacquire the signal until the signal quality meets the standard. If the quality threshold is met, the subsequent processing steps are performed.

[0013] The signal-to-noise ratio (SNR) is calculated using the ratio of signal power to noise power in decibels, and the formula is as follows: in, Signal-to-noise ratio (unit: ), For signal segment power, This represents the power of the noise segment.

[0014] The periodicity index is quantified by the peak characteristics of the autocorrelation function, and the calculation formula is as follows: in, As a periodic indicator, To calculate the autocorrelation function of the signal, The lag time corresponding to the first non-zero lag peak. For is the autocorrelation function in lag time The value of time, The closer the value is to 1, the stronger the periodicity.

[0015] The kurtosis characteristic reflects the sharpness of the signal distribution, and the calculation formula is as follows: in, Kurtosis is a characteristic feature. The mathematical expectation of the variable, The sampling sequence of the signal, and These represent the signal's mean and standard deviation, respectively. A higher kurtosis value indicates more pronounced impulse characteristics and a more prominent cardiac oscillation signal component. Kurtosis does not directly participate in quality assessment but is used for adaptive axis selection in subsequent steps.

[0016] Step 4: Adaptive axis selection and cardiac signal reconstruction; The aforementioned adaptive axis selection and cardiac signal reconstruction include adaptive axis selection and cardiac signal reconstruction.

[0017] Adaptive axis selection: Based on the three quality indicators of the three-axis acceleration signal and three-axis angular velocity signal calculated in step 3, the axes are comprehensively sorted through a weighted scoring mechanism, and the single principal axis with the best signal quality is automatically selected, or the multi-axis signals are weighted and fused to adapt to different smartphone placement angles and user measurement postures, and to generate optimized acceleration and angular velocity signals. The mass vector is composed of three mass indices from the triaxial acceleration signal and the triaxial angular velocity signal. , ,in, The first Signal-to-noise ratio, periodicity index, and kurtosis characteristics of the individual axis signals. , , It is a triaxial acceleration signal. , , The signals are triaxial angular velocity signals; each indicator is normalized and linearly mapped to the [0,1] interval; the overall quality score of each signal is calculated through weighted scoring. in, The overall quality score for each signal is calculated. , , These are the normalized th Signal-to-noise ratio, periodicity index, and kurtosis characteristics of the individual axis signals. , , These are the weighting coefficients for each item.

[0018] Adaptive axis selection supports two operating modes: Mode 1 (Single-axis optimization mode): Selects the axis signal with the highest overall quality score as the main signal. This mode has low computational cost and is suitable for real-time processing scenarios. Mode 2 (Weighted Fusion Mode): Select the top two or three axis signals with the highest overall quality scores, and use their normalized scores as weights to perform weighted average fusion of the signals. This mode has higher signal fidelity and is suitable for precise analysis scenarios.

[0019] Heartbeat signal reconstruction: The acceleration and angular velocity signals after adaptive axis selection are fused, and the low-frequency response advantage of the accelerometer is combined with the high-frequency response advantage of the gyroscope through complementary filtering to reconstruct the heartbeat signal. Signal.

[0020] Heart Tremor The signal expression is as follows: in, For heart shock Signal, For time, These are complementary filter coefficients. The acceleration signal obtained after adaptive axis selection. The angular velocity signal obtained after adaptive axis selection.

[0021] Step 5: Detection of AO feature points (aortic valve opening points); Reconstructed cardiac vibration The signal amplitude envelope is extracted using Hilbert transform. Candidate point search is performed on the amplitude envelope. Specifically, an adaptive thresholding method is used to detect local peaks as AO candidate points. The adaptive threshold is dynamically adjusted based on the moving average and moving standard deviation of the current signal segment, and the threshold coefficient is adaptively adjusted based on the signal noise level. Physiological constraints are used for screening based on the cardiac cycle to eliminate spurious peaks that do not conform to physiological laws and determine the final AO feature point location.

[0022] In the process of amplitude envelope extraction, the reconstructed cardiac vibration is first processed. The signal is subjected to a fourth-order Butterworth bandpass filter with a passband range of 5–30 Hz to further suppress low-frequency motion artifacts and high-frequency noise. Then, the Hilbert transform is used to calculate the analytical signal and extract the amplitude envelope. The calculation formula is as follows: in, To analyze the signal, For amplitude envelope, Filtered cardiac vibration Signal, For Hilbert transform, It is the imaginary unit.

[0023] During the candidate point search process, peak detection is performed on the amplitude envelope, the moving mean and moving standard deviation of the amplitude envelope are calculated, and an adaptive threshold is set. The calculation formula is as follows: in, The moving average of the amplitude envelope. The sliding standard deviation of the amplitude envelope; The threshold coefficient is adaptively adjusted based on the noise level of the current signal: [Settings are missing from the original text.] The base value is The fluctuation range is limited to [2.0, 5.0]. The signal-to-noise ratio is calculated in real time. When fluctuations occur, Based on adaptive updates using a linear piecewise function: The amplitude envelope is greater than Points that are local maxima are marked as AO candidate points, and the minimum peak spacing constraint must be satisfied between adjacent candidate points.

[0024] During the physiological constraint screening process, periodic constraints are applied based on the physiological range of heart rate: the cardiac cycle is 0.3~2.0 seconds; for adjacent AO candidate points, if the time interval... If the time interval falls outside the range of [0.3, 2.0] seconds, candidate points with smaller amplitudes are removed. Further verification is performed using the consistency between cardiac cycles: the median and interquartile range of the time intervals of all retained candidate points are calculated, and abnormal AO candidate points whose time intervals deviate from the median by more than 1.5 times the interquartile range are removed, thus obtaining the final AO feature point sequence.

[0025] Step 6: Extraction and output of cardiac vibration features; Based on the detected AO feature points, cardiac vibration is extracted. The key characteristic parameters of the signal are identified, and the cardiac measurement results are output.

[0026] The key characteristic parameters include instantaneous heart rate, average heart rate, heart rate variability, cardiac cycle duration, aortic valve opening amplitude, ventricular ejection time, and morphological characteristics of cardiac vibration signals.

[0027] The formula for calculating instantaneous heart rate is as follows: in, For the first Instantaneous heart rate (in beats / minute) at AO feature points. For the first The temporal location of each AO feature point For the first The temporal location of each AO feature point.

[0028] The average heart rate is the median of all instantaneous heart rates used as the average heart rate for the measurement segment.

[0029] The heart rate variability index is calculated based on the time interval between adjacent AO feature points (denoted as the AO-AO interval, equivalent to the RR interval of the electrocardiogram signal), specifically including: (1)SDNN: Calculate the standard deviation of all adjacent AO-AO intervals to assess the degree of overall heart rate variability.

[0030] (2) RMSSD: Calculates the root mean square of the difference between adjacent AO-AO intervals, i.e. ,in For the first There are AO-AO intervals, among which, This represents the total number of AO feature points; used to assess the regulatory activity of the vagus nerve (parasympathetic nervous system) on heart rate.

[0031] The aortic valve opening amplitude is determined by taking each AO feature point in the cardiac resonance. The amplitude in the signal.

[0032] The ventricular ejection time (VET) is defined as the time from the AO feature point to the cardiac vibration. The next zero-crossing point or local minimum point is searched backward in the signal as the aortic valve closure point (AC point). Since the mitral valve closure (MC) point is difficult to capture in the cardiac resonant signal, this invention uses the time interval from the AO point to the AC point to approximate the ejection time.

[0033] The morphological features of the cardiac oscillation signal include parameters describing the dynamic characteristics of ventricular contraction, such as the slope, symmetry, and area of ​​the waveform near the AO feature point.

[0034] An adaptive cardiac signal measurement and feature extraction device based on a smartphone IMU includes: at least one processor and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the processor, which are then executed by the at least one processor to cause the at least one processor to perform the aforementioned adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU.

[0035] The present invention has the following beneficial effects: (1) This invention utilizes only the built-in IMU of a smartphone to complete the acquisition and feature extraction of cardiac vibration signals without any external dedicated sensors, which greatly reduces the equipment threshold for cardiac vibration measurement and realizes self-monitoring of cardiac function in remote and home scenarios.

[0036] (2) The present invention adds a real-time signal quality assessment step after signal preprocessing. When the signal quality is not up to standard, it actively prompts the user to adjust the position and re-acquire the signal, ensuring that the input signal in subsequent processing always meets the minimum quality requirements and avoiding the problem of unreliable measurement due to poor signal quality.

[0037] (3) The adaptive axis selection mechanism proposed in this invention automatically adapts to different mobile phone placement angles and user postures through multi-dimensional signal quality assessment and weighted scoring, solving the core problem of signal axis selection difficulty in smartphone cardiac measurement and ensuring the consistency and reliability of cardiac signals under different usage scenarios.

[0038] (4) The AO feature point detection algorithm based on Hilbert envelope and cardiac cycle physiological constraints proposed in this invention can achieve robust and accurate AO point localization under the low signal-to-noise ratio condition of smartphone IMU, and effectively suppress the influence of motion artifacts and noise interference on the detection results.

[0039] (5) This invention constructs a complete automated processing flow from smartphone chest-touch acquisition, signal preprocessing, quality assessment feedback, adaptive axis selection, signal reconstruction, AO point detection to cardiac vibration feature output, providing an end-to-end technical solution for daily monitoring of cardiac function based on smartphones. Attached Figure Description

[0040] Figure 1 This is the overall flowchart of the adaptive cardiac signal measurement and feature extraction method based on the built-in IMU of a smartphone according to the present invention; Figure 2 This is a flowchart of the adaptive axis selection and cardiac signal reconstruction of the present invention; Figure 3 This is a flowchart of the AO feature point detection method of the present invention. Detailed Implementation

[0041] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0042] An adaptive method for cardiac signal measurement and feature extraction based on a smartphone IMU, such as... Figure 1 As shown. The specific steps are as follows: Step 1: Place the smartphone against the user's chest and start data collection; The user opens the heart rate measurement application on their smartphone and follows the on-screen instructions to place the back of the smartphone against the sternum area on their chest, maintaining slight pressure to ensure stable contact. The user can choose to lie flat, sit, or semi-recline. After confirming readiness, the user clicks the "Start Measurement" button, and the smartphone's built-in IMU begins synchronously acquiring triaxial acceleration signals (denoted as...). , , ) and triaxial angular velocity signals (denoted as , , The IMU sampling rate should be set to 100~200Hz, the accelerometer range to ±2g or ±4g, and the gyroscope range to ±250dps or ±500dps to ensure that the fine signal characteristics of chest cavity micro-vibrations can be captured. The recommended duration of a single acquisition is 30 seconds to 2 minutes. During this period, the user should maintain natural breathing and avoid large body movements.

[0043] Step 2: Signal preprocessing; The DC components (mean values) of the acquired triaxial IMU raw angular velocity and acceleration signals are first removed, and then filtered through a fourth-order Butterworth bandpass filter with a passband frequency range of 1~40Hz to eliminate baseline drift (<1Hz) caused by low-frequency breathing and body swaying, as well as high-frequency muscle tremors and electronic noise (>40Hz), to obtain the preprocessed triaxial IMU signal.

[0044] Step 3: Signal quality assessment; The preprocessed triaxial signals were paired to obtain three sets of angular velocity and acceleration axis signals: , , For each set of axis signals, three quality indicators are calculated: (a) Signal-to-noise ratio (SNR) The signal is divided into segments with a window length of 2 seconds, and the signal power in each segment is... The mean square value of the highest 10% amplitude samples within the signal segment is used to calculate the noise power. The signal-to-noise ratio is calculated by taking the mean square value of the lowest 50% amplitude samples within the signal segment. The unit is ; (b) Periodic indicators ( ): Calculate the autocorrelation function of the signal, and take the ratio of the first non-zero lag peak to the zero lag peak as a periodicity measure, i.e. ,in This represents the lag time corresponding to the first peak; the closer this value is to 1, the stronger the periodicity. For is the autocorrelation function in lag time The value at time; (c) Kurtosis characteristics ): Calculate the kurtosis characteristics of the signal ,in The mathematical expectation of the variable, The sampling sequence of the signal, and These are the mean and standard deviation of the signal, respectively. A higher kurtosis value indicates that the pulse characteristics in the signal are more significant and the cardiac vibration signal component is more prominent. Comprehensive evaluation: Among all axis signals, if at least one axis signal... Exceeding the preset minimum threshold (e.g., 3) )and If the signal quality exceeds the preset minimum threshold (e.g., 0.3), the signal quality is deemed acceptable and the process proceeds to step 4. Otherwise, the signal quality is deemed unacceptable, and the user is prompted with text or graphics on the smartphone interface with the message "Signal quality is insufficient. Please adjust the phone position and measure again." The user can adjust the phone's contact position, pressure, or body posture to trigger the acquisition again and return to step 1. Step 4, Adaptive axis selection and cardiac signal reconstruction, such as Figure 2 As shown; Adaptive axis selection and cardiac signal reconstruction are the core components of this invention, which include two sub-steps: adaptive axis selection and cardiac signal reconstruction. In the adaptive axis selection sub-step, a quality vector is formed using the three quality indices for each axis calculated in step 3. ( The indicators are normalized and linearly mapped to the [0,1] interval; the overall quality score for each axis is calculated using weighted scoring. in , , These are the three indicators after normalization. , , These are the weighting coefficients for each item; the default value can be set to... , , It can also be determined through pre-calibration optimization.

[0045] The adaptive axis selection strategy supports two operating modes: Mode 1 (Single-axis optimization mode): Selects the axis signal with the highest overall quality score as the main signal. This mode has low computational cost and is suitable for real-time processing scenarios. Mode 2 (Weighted Fusion Mode): Select the top two or three axis signals with the highest comprehensive quality scores, and use their normalized scores as weights to perform weighted average fusion to obtain the fused axis signal. This mode has higher signal fidelity and is suitable for precise analysis scenarios. In the cardiac signal reconstruction sub-step, the acceleration signal obtained after adaptive axis selection... and angular velocity signal A fusion and reconstruction process is performed. Since the accelerometer has better response characteristics in the low-frequency range (<10Hz), while the gyroscope has lower noise levels in the high-frequency range (10~40Hz), a complementary filter is used to fuse the two in the frequency domain. in, For heart shock Signal, For time, These are complementary filter coefficients, with a value range of 0.90 to 0.98, and a default value of 0.95. This fusion strategy preserves the reliability of the accelerometer in the low-frequency band while enhancing high-frequency details by utilizing the angle information from the gyroscope's angular velocity integral. Step 5, AO feature point detection, such as Figure 3 As shown; AO feature points are among the most critical feature points in cardiac resonance signals, marking the moment when the aortic valve opens, corresponding to the instant when the left ventricle begins to eject blood into the aorta; the AO detection process of this invention includes three stages: envelope extraction, candidate point search, and physiological constraint screening; In the envelope extraction stage, the reconstructed... The signal undergoes a fourth-order Butterworth bandpass filter with a passband range of 5–30 Hz to further suppress low-frequency motion artifacts and high-frequency noise. Then, the Hilbert transform is used to calculate the analytic signal and extract the amplitude envelope: Let the filtered signal… The signal is Its Hilbert transformation is The analytic signal is ,in The imaginary unit is , and the amplitude envelope is . ; During the candidate point search phase, the amplitude envelope is... Peak detection: Calculate the moving average of the envelope. and sliding standard deviation (Window length is 2 seconds), set adaptive threshold. ,in This is the threshold coefficient, typically ranging from 2.0 to 5.0, which is adaptively adjusted based on the noise level of the current signal. The amplitude envelope is greater than Points that are local maxima are marked as AO candidate points, and the minimum peak spacing constraint (200 milliseconds) must be satisfied between adjacent candidate points. During the physiological constraint screening phase, cycle constraints are applied based on the physiological range of heart rate: the normal resting heart rate for adults is approximately 60-100 beats / minute, with an extreme range of 30-200 beats / minute, corresponding to a cardiac cycle of 0.3-2.0 seconds; for adjacent AO candidate points, if their time interval... If the time interval falls outside the range of [0.3, 2.0] seconds, candidate points with smaller amplitudes are removed. Further verification is performed using the consistency between cardiac cycles: the median and interquartile range of the time intervals of all retained candidate points are calculated, and abnormal candidate points whose time intervals deviate from the median by more than 1.5 times the interquartile range are removed, thus obtaining the final AO feature point sequence.

[0046] Step 6: Extraction and output of cardiac vibration features; Based on the detected AO feature point sequence, the following key cardiac resonance feature parameters are extracted: (a) Instantaneous heart rate: (times / minute), of which For the first The temporal location of each AO feature point; (b) Average heart rate: The median of all instantaneous heart rates is taken as the average heart rate of the measurement segment; (c) Heart rate variability index: Calculate the standard deviation (SDNN) of the interval between adjacent AO feature points and the root mean square (RMSSD) of the difference between adjacent intervals. (d) Aortic valve opening amplitude: Take each AO feature point at... Amplitude in the signal; (e) Ventricular ejection time (VET): Starting from the AO characteristic point, at Searching backwards in the signal for the next significant zero-crossing point or local minimum point is taken as the aortic valve closure point (AC point). The time interval from the AO characteristic point to the AC point can be approximated as the ventricular ejection time, with a normal range of about 250~350 milliseconds. (f) Morphological characteristics of cardiac oscillation signals: including parameters describing the dynamic characteristics of ventricular contraction, such as the slope, symmetry, and area of ​​the waveform near the AO feature point; The aforementioned feature parameters can be displayed on a smartphone interface in a structured data format, and can also be stored locally or in the cloud for later viewing and analysis.

Claims

1. An adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU, characterized in that, Includes the following steps: Step 1: Place the smartphone against the user's chest and use the smartphone's built-in inertial measurement unit (IMU) to simultaneously acquire triaxial acceleration and triaxial angular velocity signals. Step 2: Signal preprocessing; The acquired triaxial acceleration and triaxial angular velocity signals are processed to remove DC components, and drift interference and noise interference are filtered out by a bandpass filter to obtain preprocessed triaxial acceleration and triaxial angular velocity signals. Step 3: Signal quality assessment; The preprocessed triaxial acceleration and triaxial angular velocity signals are evaluated for quality. Three quality indicators—signal-to-noise ratio, periodicity index, and kurtosis—are calculated for the triaxial acceleration and triaxial angular velocity signals. The results are then used to determine whether the currently acquired signal meets the preset quality threshold requirements. If the quality threshold is not met, the user is prompted via a smartphone interface to adjust the phone's placement or measurement posture and reacquire the signal until the signal quality meets the standards. If the quality threshold is met, the process proceeds to the next processing step. Step 4: Adaptive axis selection and cardiac signal reconstruction; The aforementioned adaptive axis selection and cardiac signal reconstruction include adaptive axis selection and cardiac signal reconstruction; Adaptive axis selection: Based on the three quality indicators of the three-axis acceleration signal and three-axis angular velocity signal calculated in step 3, the axes are comprehensively sorted through a weighted scoring mechanism, and the single principal axis with the best signal quality is automatically selected, or the multi-axis signals are weighted and fused to adapt to different smartphone placement angles and user measurement postures, and to generate optimized acceleration and angular velocity signals. Heartbeat signal reconstruction: The acceleration and angular velocity signals after adaptive axis selection are fused, and the low-frequency response advantage of the accelerometer is combined with the high-frequency response advantage of the gyroscope through complementary filtering to reconstruct the heartbeat signal. Signal; Step 5: AO feature point detection; Reconstructed cardiac vibration The signal amplitude envelope is extracted using Hilbert transform. Candidate point search is performed on the amplitude envelope, specifically: an adaptive thresholding method is used to detect local peaks as AO candidate points. The adaptive threshold is dynamically adjusted based on the moving average and moving standard deviation of the current signal segment, and the threshold coefficient is adaptively adjusted based on the signal noise level. Physiological constraints are used for screening based on the cardiac cycle to eliminate spurious peaks that do not conform to physiological laws, thus determining the final AO feature point locations. Step 6: Extraction and output of cardiac vibration features; Based on the detected AO feature points, cardiac vibration is extracted. The key characteristic parameters of the signal are identified, and the cardiac measurement results are output. The key characteristic parameters include instantaneous heart rate, average heart rate, heart rate variability, cardiac cycle duration, aortic valve opening amplitude, ventricular ejection time, and morphological characteristics of cardiac vibration signals.

2. The adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU according to claim 1, characterized in that, In step 2, the bandpass filter is a Butterworth bandpass filter with a passband frequency range of 1~40Hz.

3. The adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU according to claim 1, characterized in that, In step 3, The signal-to-noise ratio (SNR) is calculated using the ratio of signal power to noise power in decibels, and the formula is as follows: in, Signal-to-noise ratio (unit: ), For signal segment power, Power of the noise section; The periodicity index is quantified by the peak characteristics of the autocorrelation function, and the calculation formula is as follows: in, As a periodic indicator, To calculate the autocorrelation function of the signal, The lag time corresponding to the first non-zero lag peak. For is the autocorrelation function in lag time The value at time; The kurtosis characteristic reflects the sharpness of the signal distribution, and the calculation formula is as follows: in, Kurtosis is a characteristic feature. The mathematical expectation of the variable, The sampling sequence of the signal, and These are the mean and standard deviation of the signal, respectively.

4. The adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU according to claim 1, characterized in that, In step 3, the comprehensive judgment method is as follows: Of all axis signals, if at least one axis signal... Exceeding the preset minimum threshold and If the signal quality exceeds the preset minimum threshold, the signal quality is deemed acceptable, and the process proceeds to step 4. Otherwise, if the signal quality is deemed unqualified, the user will be prompted via text or graphics on the smartphone interface to adjust the phone's contact position, pressure, or body posture before triggering the acquisition again, and the process will return to step 1.

5. The adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU according to claim 1, characterized in that, In step 4, The mass vector is composed of three mass indices from the triaxial acceleration signal and the triaxial angular velocity signal. , ,in, The first Signal-to-noise ratio, periodicity index, and kurtosis characteristics of the individual axis signals. , , It is a triaxial acceleration signal. , , The signals are triaxial angular velocity signals; each indicator is normalized and linearly mapped to the [0,1] interval; the overall quality score of each signal is calculated through weighted scoring. in, The overall quality score for each signal is calculated. , , These are the normalized th Signal-to-noise ratio, periodicity index, and kurtosis characteristics of the individual axis signals. , , These are the weighting coefficients for each item; Adaptive axis selection supports two operating modes: Single-axis optimization mode: Selects the axis signal with the highest overall quality score as the main signal, suitable for real-time processing scenarios; Weighted fusion mode: Select the top two or three axis signals with the highest comprehensive quality scores, and use their normalized scores as weights to perform weighted average fusion of the signals. This mode is suitable for precise analysis scenarios.

6. The adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU according to claim 1, characterized in that, In step 4, Heart Tremor The signal expression is as follows: in, For heart shock Signal, For time, These are complementary filter coefficients. The acceleration signal obtained after adaptive axis selection. The angular velocity signal obtained after adaptive axis selection.

7. The adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU according to claim 1, characterized in that, Step 5 is as follows: In the process of amplitude envelope extraction, the reconstructed cardiac vibration is first processed. The signal is subjected to a fourth-order Butterworth bandpass filter with a passband range of 5–30 Hz to further suppress low-frequency motion artifacts and high-frequency noise. Then, the Hilbert transform is used to calculate the analytical signal and extract the amplitude envelope. The calculation formula is as follows: in, To analyze the signal, For amplitude envelope, Filtered cardiac vibration Signal, For Hilbert transform, The imaginary unit; During the candidate point search process, peak detection is performed on the amplitude envelope, the moving mean and moving standard deviation of the amplitude envelope are calculated, and an adaptive threshold is set. The calculation formula is as follows: in, The moving average of the amplitude envelope. The sliding standard deviation of the amplitude envelope; The threshold coefficient is adaptively adjusted based on the noise level of the current signal: [Settings are missing from the original text.] The base value is The fluctuation range is limited to [2.0, 5.0]; when the real-time calculated signal-to-noise ratio When fluctuations occur, Based on adaptive updates using a linear piecewise function: The amplitude envelope is greater than Points that are local maxima are marked as AO candidate points, and the minimum peak spacing constraint must be satisfied between adjacent candidate points; During the physiological constraint screening process, periodic constraints are applied based on the physiological range of heart rate: the cardiac cycle is 0.3~2.0 seconds; for adjacent AO candidate points, if the time interval... If the time interval falls outside the range of [0.3, 2.0] seconds, candidate points with smaller amplitudes are removed. Further verification is performed using the consistency between cardiac cycles: the median and interquartile range of the time intervals of all retained candidate points are calculated, and abnormal AO candidate points whose time intervals deviate from the median by more than 1.5 times the interquartile range are removed, thus obtaining the final AO feature point sequence.

8. The adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU according to claim 1, characterized in that, In step 6, The formula for calculating instantaneous heart rate is as follows: in, For the first Instantaneous heart rate at each AO feature point For the first The temporal location of each AO feature point For the first The temporal location of each AO feature point; The average heart rate is the median of all instantaneous heart rates used as the average heart rate for the measurement segment. The heart rate variability index is calculated based on the time interval between adjacent AO feature points, specifically including: SDNN: Calculates the standard deviation of all adjacent AO-AO intervals to assess the degree of overall heart rate variability; RMSSD: Calculates the root mean square of the difference between adjacent AO-AO intervals, i.e. ,in For the first There are AO-AO intervals, among which, This represents the total number of AO feature points; used to assess the vagus nerve's regulatory activity on heart rate. The aortic valve opening amplitude is determined by taking each AO feature point in the cardiac resonance. Amplitude in the signal; The ventricular ejection time is defined as the time starting from the AO feature point, within the cardiac vibration. The next zero-crossing point or local minimum point is searched backward in the signal as the aortic valve closure point, i.e., the AC point. The time interval between the AO point and the AC point is used to approximate the ejection time. The morphological features of the cardiac oscillation signal include parameters describing the dynamic characteristics of ventricular contraction, such as the slope, symmetry, and area of ​​the waveform near the AO feature point.

9. An adaptive cardiac signal measurement and feature extraction device based on a smartphone IMU, characterized in that, include: At least one processor, and a memory communicatively connected to said at least one processor; The memory stores instructions executable by the processor, which are executed by the at least one processor to cause the at least one processor to perform the adaptive cardiac signal measurement and feature extraction method based on a smartphone IMU as described in any one of claims 1-8.

Images