A multi-parameter fusion method for pulse wave signal quality assessment

Abstract

The present invention discloses a multi-parameter fusion pulse wave signal quality assessment method, which relates to the field of medical diagnosis technology, and aims to solve the technical problem that it is not convenient to refine the classification and give the specific signal abnormality type when evaluating the quality of the output signal of the existing pulse oximeter, and the actual use is limited. The method includes the following steps: S1, PPG signal and three-axis acceleration data acquisition; synchronously collect the red light, infrared light and three-axis acceleration raw data of the pulse wave, and represent each data with a specific symbol; S2, PPG signal endpoint detection; for the problem of signal saturation caused by individual and acquisition site differences, endpoint value detection is set, and labels are set according to the relationship between signal amplitude and endpoint value; S3, PPG signal non-pulsating component feature extraction. The present invention has the advantages of assisting clinicians in evaluating the credibility of the output results of the pulse oximeter, and can help clinicians directly eliminate corresponding problems and improve monitoring efficiency by giving the corresponding signal abnormality type.

Description

Technical Field

[0001] The present invention relates to the field of medical diagnosis technology, and more specifically, to a multi-parameter fusion pulse wave signal quality assessment method. Background Art

[0002] Blood oxygen saturation (SpO2) is the percentage of oxygenated hemoglobin (HbO2) in the blood to the total hemoglobin (Hb) capacity that can be combined, that is, the concentration of blood oxygen in the blood. It is an important physiological parameter of respiratory circulation. Functional oxygen saturation is the ratio of HbO2 concentration to HbO2+Hb concentration, which is different from the percentage of oxygenated hemoglobin. Therefore, monitoring arterial oxygen saturation can estimate the oxygenation of the lungs and the oxygen carrying capacity of hemoglobin. The oxygen saturation of normal human arterial blood is 98%, and that of venous blood is 75%.

[0003] The metabolic process of the human body is a biological oxidation process. The oxygen needed in the metabolic process enters the human blood through the respiratory system, combines with hemoglobin in red blood cells to form oxygenated hemoglobin, and is then transported to tissue cells in various parts of the human body. The ability of blood to carry and transport oxygen is measured by blood oxygen saturation.

[0004] Pulse oximeter is a non-invasive monitor for measuring pulse rate and blood oxygen saturation. However, pulse oximeter is susceptible to noise interference, which can have a negative impact on the accuracy of the output pulse rate and blood oxygen saturation. When the output pulse rate is not correlated with the ECG, clinicians are likely to question it. Therefore, it is necessary to evaluate the credibility of the output results of the pulse oximeter.

[0005] At present, the existing schemes for signal quality assessment mainly include: signal quality assessment based on acceleration and signal quality assessment based on feature extraction plus machine learning or deep learning. The signal vector is extracted based on the three-axis acceleration to judge whether there is body movement. This method relies on the extraction of the three-axis acceleration amplitude envelope and the detection threshold, which is easy to miss detection, and this method only detects the interference caused by body movement. In fact, there are interferences from other factors such as ambient light and weak perfusion. There are also methods based on feature extraction plus machine learning or deep learning. Machine learning or deep learning methods rely on the accuracy of the actual data set labels and the generalization of the model. It takes a long time to label the data set, and most of the evaluation results only divide the signal into available or unavailable, without detailed classification and giving specific signal abnormality types, which limits actual use. In view of this, we propose a multi-parameter fusion pulse wave signal quality assessment method. Summary of the invention

[0006] The purpose of the present invention is to provide a multi-parameter fusion pulse wave signal quality assessment method to solve the technical problem that when evaluating the quality of the output signal of the existing pulse oximeter, it is not easy to refine the classification and give the specific signal abnormality type, and the actual use is limited.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a multi-parameter fusion pulse wave signal quality assessment method, comprising the following steps:

[0008] S1, PPG signal and three-axis acceleration data acquisition;

[0009] Synchronously collect the raw data of red light, infrared light and three-axis acceleration of the pulse wave;

[0010] S2, PPG signal endpoint detection;

[0011] To solve the problem of signal saturation caused by individual and acquisition site differences, endpoint value detection is set, and labels are set according to the relationship between signal amplitude and endpoint value. ;

[0012] S3, feature extraction of non-pulsating components of PPG signals;

[0013] Use filters to extract non-pulsating components with large differences between wearing and not wearing, and set labels based on thresholds ;

[0014] S4, PPG signal pulsation component feature extraction;

[0015] The basic amplitude signal is extracted by time window processing and amplitude characteristics, set labels according to threshold range

[0016] S5, PPG signal abnormal pulsation component detection;

[0017] In the fundamental amplitude signal Based on this, the dynamic threshold is extracted to is the time window, is the window moving length, extract the mean and 3 times the standard deviation of the data between the upper and lower quartiles in each time window as the threshold in each time window, traverse the entire signal, output the pulsation component amplitude threshold, and set the label according to the relationship between the amplitude and the threshold

[0018] S6, PPG signal segment autocorrelation feature extraction;

[0019] According to the pulse wave cycle characteristics, the time window is selected to segment the signal autocorrelation to calculate the characteristic value, and the label is determined according to the threshold. ;

[0020] S7, PPG signal template Pearson correlation coefficient extraction;

[0021] Detect key points and output templates. Calculate correlation coefficients based on key point templates and set labels based on thresholds after processing. ;

[0022] S8, three-axis acceleration body motion detection;

[0023] Extract the first-order partial derivative of each axis of the three-axis data , calculate the first-order partial derivative vector , after filtering, we get the resultant vector envelope , set labels according to thresholds ;

[0024] S9, three-axis acceleration body position state change detection;

[0025] Determine the device status based on the amplitude and direction of the three-axis acceleration;

[0026] S10, three-axis acceleration body position change frequency detection;

[0027] Calculate the frequency based on the time of body position change and set the label based on the threshold ;

[0028] S11, abnormal signal display;

[0029] when When the signal abnormality type prompt is amplitude saturation, the signal abnormality prompt priority is 1, when When the signal abnormality type prompt is not worn, the signal abnormality prompt priority is 2. When the signal abnormality type prompt is low amplitude, the signal abnormality prompt priority is 3. When the signal abnormality type prompt is a large amplitude value, the signal abnormality prompt priority is 4, when When the signal abnormality type prompt is poor signal periodicity, the signal abnormality prompt priority is 5. When the signal abnormality type prompt is signal shape abnormality, the signal abnormality prompt priority is 6. When the signal abnormality type prompt is abnormal body motion amplitude, the signal abnormality prompt priority is 7. When the signal abnormality type prompt is abnormal device status change, the signal abnormality prompt priority is 8, and the signal abnormality prompt will give priority to the signal abnormality types with lower priority.

[0030] The present invention uses 8 evaluation methods to deeply integrate pulse waves and three-axis acceleration features, covering a wide range of conditions such as signal saturation, abnormal device wearing, abnormal signal amplitude, abnormal periodicity, abnormal morphological characteristics, abnormal body movement amplitude, and abnormal device status change frequency, providing a comprehensive and accurate evaluation system for the quality of pulse oximeter signals. For example, in signal saturation detection, the endpoint value is set according to the AD sampling maximum value to accurately identify the validity of the signal; in non-pulsating component analysis, features are extracted by using a specific filter and the wearing status is determined according to the individual scene threshold. The signal quality is accurately determined in a multi-dimensional collaborative manner, which effectively assists clinicians in accurately evaluating the output credibility of the pulse oximeter, effectively eliminating potential problems, and improving monitoring accuracy and efficiency.

[0031] Preferably, in S1, red light, infrared light and three-axis acceleration axis, axis, The original data of the axis are , , , , Indicates that the PPG signal is express, , the three-axis acceleration is express, .

[0032] Preferably, in S2, endpoint detection is performed on the pulse wave signal collected in S1, and the endpoint value based on the AD sampling maximum value is set , , when the signal amplitude value is equal to the endpoint value, the output label is 0, otherwise is 1, as shown in the following formula:

[0033] .

[0034] Preferably, in S3, a 4th order, zero phase IIR digital low pass filter is used to filter out interference components below 0.5 Hz and above 5 Hz, and to extract non-pulsating components of red light and infrared light, wherein the cutoff frequency of the zero phase IIR digital low pass filter is 0.1 Hz;

[0035] The low-frequency components between 0.5Hz and 5Hz are used as two features in S3. express, , set thresholds based on individual scenarios , ,when ,or , output label is 0, otherwise is 1, as shown in the following formula:

[0036] .

[0037] Preferably, in S4, 1 second is used as a fixed time window, the difference between the maximum and minimum values ​​of the signal in each time window is extracted, and the basic amplitude signal is output. , ,in, and Represent the basic amplitude signals of red light and infrared light respectively, and then convolve them with a rectangular window of length 5 to output the amplitude characteristics of the pulsating component , ,in, and Respectively represent the amplitude characteristics of the red light and infrared light pulsation components. At the experimental threshold within the range, or At the experimental threshold Within the range, output label Set to 0, otherwise is 1, as shown in the following formula:

[0038] .

[0039] Preferably, in S5, for , ,in, and Respectively represent the pulsation component amplitude thresholds of red light and infrared light. If the basic amplitudes of red light and infrared light are within the pulsation component amplitude threshold If the current second signal is abnormal, Set to 0, otherwise is 1, as shown in the following formula:

[0040] .

[0041] Preferably, in S6, the signal is segmented with 5 seconds as the time window and 1 second as the window moving length, and the signal in each time window is autocorrelated. Suppose the signal in a certain time window is , , is the signal index value, For a long time window, the autocorrelation calculation formula is as follows:

[0042]

[0043] , is the autocorrelation signal The independent variable;

[0044] Normalize the above formula to get the normalized autocorrelation value , as shown below:

[0045]

[0046] Extracting autocorrelation eigenvalues , , Quantitatively represents the periodicity of the pulse signal. The smaller the value, the better the periodicity. , is the normalized autocorrelation peak point, It means taking the maximum value among B, C, and D;

[0047] The periodicity of the pulse wave signal is evaluated by the following formula: Greater than threshold , Set to 0, otherwise is 1, as shown in the following formula:

[0048] ;

[0049] In the calculation, the weighted strategy dynamically weights the periodic component according to the signal characteristics to improve the evaluation accuracy. The weighted autocorrelation function is: , weight Based on the local energy of the signal With frequency characteristics Calculation, such as , optimize eigenvalue extraction and periodic evaluation, where, , To adjust the parameters, is the lag parameter in the autocorrelation function, Represents the pulse wave signal at discrete moments The sampling value of is the discrete time index, To calculate the local energy of the signal and frequency characteristics The window length parameter, Pulse wave signal The autocorrelation function of .

[0050] Preferably, in S7, four key points A, B, C, and D of the pulse wave are detected, and the data centered on the four key points are averaged to output four key point templates, and then the key point templates are respectively used to calculate the Pearson correlation coefficient with the data extracted centered on the key points, and the key point positions are used as independent variables and the Pearson correlation coefficient is used as the dependent variable to perform nonlinear interpolation so that the sampling rate of the interpolated Pearson correlation coefficient is the same as the sampling rate of the original signal, and then downsampling is performed to make the final output Pearson correlation coefficient The sampling rate is 1Hz, that is, one correlation coefficient value is output every 1 second. , Set to 0, otherwise is 1, as shown in the following formula:

[0051]

[0052] in Indicates the decision threshold for signal quality assessment using the Pearson correlation coefficient;

[0053] In S8, the first-order partial derivative extraction formula is as follows:

[0054]

[0055] in, is the index value of the three-axis discrete data independent variable, Represents the data of each axis of the three-axis data The first-order partial derivative of Indicates the original data of the three-axis acceleration The data on the axis, , , Respectively axis, Axis and axis;

[0056] Calculate the resultant vector of the first-order partial derivative , the calculation formula is as follows:

[0057]

[0058] right With 1 second as the filter window length and 0.5 second as the window moving length, a sliding average filter is performed to output the envelope of the resultant vector. ,when hour, Set to 0, otherwise is 1, as shown in the following formula:

[0059] ;

[0060] in, represents the body motion determination threshold, , and They are , , The first-order partial derivative of the axis data.

[0061] Preferably, in S9, the device state is determined according to the amplitude and direction of the three-axis acceleration. For example, if the device deflects to the right, it is determined as:

[0062] ;

[0063] in, Indicates the judgment that the device deflects to the right. , and They represent the three-axis accelerations in , , Amplitude of the axis;

[0064] In S10, the time point of the first position change As the starting point, if the time interval between the current body position change time point and the previous body position change time point is greater than 20s, the previous body position change time point is taken as the end point of the continuous body position change, recorded as , the number of changes between the starting and ending points of the position change is used The frequency of body position change is Calculated by the following formula:

[0065] ;

[0066] The frequency of body position changes is greater than a certain threshold ,but Set to 0, otherwise is 1, as shown in the following formula:

[0067] ;

[0068] Fusion of multi-sensor data to build a dynamic attitude model, using the Kalman filter algorithm to fuse gyroscope and accelerometer data into the attitude transfer matrix Update to improve the accuracy of body motion interference identification, including: , is the discrete time index, is the fusion algorithm update function, For at the moment The collected gyroscope data, For at the moment The collected accelerometer data.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] 1. The present invention uses 8 signal quality assessment methods, which respectively involve signal saturation, device wearing abnormality, signal amplitude abnormality, signal periodicity abnormality, signal morphology feature abnormality, body motion signal amplitude abnormality, and device state change frequency abnormality. According to the above abnormality types, the various characteristics of pulse wave and three-axis acceleration are integrated to effectively assess the quality of pulse wave signals, assist clinicians in assessing the credibility of pulse oximeter output results, and provide corresponding signal abnormality types to help clinicians directly eliminate corresponding problems and improve monitoring efficiency. This solves the problem that when evaluating the output signal quality of existing pulse oximeters, it is not easy to refine the classification and provide specific signal abnormality types, which limits actual use.

[0071] 2. When determining the periodicity of a signal, the present invention determines the periodicity of the signal by normalizing the autocorrelation eigenvalue. Compared with determining the periodicity of the signal by the duration of the peak interval, this can effectively avoid the problem of abnormal periodicity determination caused by abnormal peak detection.

[0072] 3. When determining abnormalities in signal morphological features, the present invention extracts multiple feature point templates and Pearson correlation coefficients at multiple feature points of the pulse wave, thereby increasing the resolution of the extracted Pearson correlation coefficients and effectively reducing the probability of false detection caused by abnormal template extraction.

[0073] 4. The present invention can reduce the influence of the baseline drift of the three-axis sensor on the body motion amplitude and improve the accuracy of body motion detection by calculating the resultant vector of the data after the first-order difference of the three-axis original data. As for whether there is body motion interference, the device position state change frequency parameter is introduced to judge whether there is body motion interference by changing the frequency, which can effectively avoid the interference to the pulse signal caused by the small body motion amplitude but the device position state has changed. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 The multi-parameter fusion pulse oximeter signal quality assessment flow chart of the present invention;

[0075] Figure 2 The waveform diagram of the present invention when wearing and not wearing red light and infrared light;

[0076] Figure 3 This is a noise detection waveform diagram of the infrared light amplitude exceeding the normal threshold range of the present invention;

[0077] Figure 4 The waveform diagram of the pulse wave autocorrelation and normalized autocorrelation of the present invention;

[0078] Figure 5A waveform diagram of the key points of the pulse wave of the present invention;

[0079] Figure 6 It is a schematic diagram of three-axis acceleration body motion detection of the present invention;

[0080] Figure 7 It is a schematic diagram of the three-axis acceleration body position state of the present invention. DETAILED DESCRIPTION

[0081] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0082] Example 1

[0083] The present invention provides a multi-parameter fusion pulse wave signal quality assessment method, comprising the following steps:

[0084] S1, PPG signal and three-axis acceleration data acquisition;

[0085] Synchronously collect the raw data of red light, infrared light and three-axis acceleration of the pulse wave;

[0086] S2, PPG signal endpoint detection;

[0087] To solve the problem of signal saturation caused by individual and acquisition site differences, endpoint value detection is set, and labels are set according to the relationship between signal amplitude and endpoint value. ;

[0088] S3, feature extraction of non-pulsating components of PPG signals;

[0089] Use filters to extract non-pulsating components with large differences between wearing and not wearing, and set labels based on thresholds ;

[0090] S4, PPG signal pulsation component feature extraction;

[0091] The basic amplitude signal is extracted by time window processing and amplitude characteristics, set labels according to threshold range

[0092] S5, PPG signal abnormal pulsation component detection;

[0093] In the fundamental amplitude signal Based on this, the dynamic threshold is extracted to is the time window, is the window moving length, extract the mean and 3 times the standard deviation of the data between the upper and lower quartiles in each time window as the threshold in each time window, traverse the entire signal, output the pulsation component amplitude threshold, and set the label according to the relationship between the amplitude and the threshold

[0094] S6, PPG signal segment autocorrelation feature extraction;

[0095] According to the pulse wave cycle characteristics, the time window is selected to segment the signal autocorrelation to calculate the characteristic value, and the label is determined according to the threshold. ;

[0096] S7, PPG signal template Pearson correlation coefficient extraction;

[0097] Detect key points and output templates. Calculate correlation coefficients based on key point templates and set labels based on thresholds after processing. ;

[0098] S8, three-axis acceleration body motion detection;

[0099] Extract the first-order partial derivative of each axis of the three-axis data , calculate the first-order partial derivative vector , after filtering, we get the resultant vector envelope , set labels according to thresholds ;

[0100] S9, three-axis acceleration body position state change detection;

[0101] Determine the device status based on the amplitude and direction of the three-axis acceleration;

[0102] S10, three-axis acceleration body position change frequency detection;

[0103] Calculate the frequency based on the time of body position change and set the label based on the threshold ;

[0104] S11, abnormal signal display;

[0105] when When the signal abnormality type prompt is amplitude saturation, the signal abnormality prompt priority is 1, when When the signal abnormality type prompt is not worn, the signal abnormality prompt priority is 2. When the signal abnormality type prompt is low amplitude, the signal abnormality prompt priority is 3. When the signal abnormality type prompt is a large amplitude value, the signal abnormality prompt priority is 4, when When the signal abnormality type prompt is poor signal periodicity, the signal abnormality prompt priority is 5. When the signal abnormality type prompt is signal shape abnormality, the signal abnormality prompt priority is 6. When the signal abnormality type prompt is abnormal body motion amplitude, the signal abnormality prompt priority is 7. When the signal abnormality type prompt is abnormal device status change, the signal abnormality prompt priority is 8, and the signal abnormality prompt will give priority to the signal abnormality types with lower priority.

[0106] In the embodiment of the present invention, in S1, red light, infrared light and three-axis acceleration axis, axis, The original data of the axis are , , , , Indicates that the PPG signal is express, , the three-axis acceleration is express, .

[0107] In the embodiment of the present invention, in S2, in practice, there are differences in the AC and DC components of pulse waves of different individuals, and there are also differences in the strength of signals collected at different parts. When the device performs AD sampling, the numerical range of the actual output signal is fixed. Some device firmware does not involve automatic dimming, which can easily cause signal saturation. When the signal is saturated, the collected signals are all invalid values. In order to avoid signal anomalies caused by individuals with large AC and DC components or differences between the collection parts and the collection parts during hardware design, endpoint detection is performed on the pulse wave signal collected in S1, and the endpoint value based on the maximum value of AD sampling is set. , , when the signal amplitude value is equal to the endpoint value, the output label is 0, otherwise is 1, as shown in the following formula:

[0108] .

[0109] In an embodiment of the present invention, Figure 2 As shown in Figure 1, in S3, there is a large difference in the numerical distribution range of the non-pulsating components between wearing and not wearing. The frequency of the non-pulsating components is relatively low, and the actual pulse wave frequency range is 30-240 times / min. Therefore, a 4th-order, zero-phase IIR digital low-pass filter is used to filter out interference components below 0.5Hz and above 5Hz, effectively filter out the interference of the lowest respiratory frequency on the non-pulsating components, and extract the red and infrared non-pulsating components. The cutoff frequency of the zero-phase IIR digital low-pass filter is 0.1Hz.

[0110] The low-frequency components between 0.5Hz and 5Hz are used as two features in S3. express, , set thresholds based on individual scenarios , ,when ,or , output label is 0, otherwise is 1, as shown in the following formula:

[0111] .

[0112] The threshold selection is determined by the difference between different scenes when different individuals are wearing and not wearing. Table 1 shows the minimum value of non-pulsating components when different samples are wearing, and Table 2 shows the maximum value of non-pulsating components in different scenes when not wearing. The simulated scenes mainly include sensor backlight, shading, direct exposure to backlit mobile phone light, and direct exposure to shading mobile phone light. There is no overlap between the minimum value under actual wearing and the maximum value under different scenes when not wearing. In this way, the judgment threshold is determined to eliminate the interference when not wearing. The attached table is as follows:

[0113]

[0114]

[0115] In the embodiment of the present invention, in S4, 1 second is used as a fixed time window, the difference between the maximum and minimum values ​​of the signal in each time window is extracted, and the basic amplitude signal is output. , ,in, and Represent the basic amplitude signals of red light and infrared light respectively, and then convolve them with a rectangular window of length 5 to output the amplitude characteristics of the pulsating component , ,in, and They represent the amplitude characteristics of the red light and infrared light pulsation components respectively. There is a significant difference in the amplitude of the normal pulse wave signal and the noise signal when not wearing the device. In order to distinguish them, when the amplitude characteristics At the experimental threshold within the range, or At the experimental threshold Within the range, output label Set to 0, otherwise is 1, as shown in the following formula:

[0116] .

[0117] In an embodiment of the present invention, in S5, for , ,in, and Respectively represent the pulsation component amplitude thresholds of red light and infrared light. If the basic amplitudes of red light and infrared light are within the pulsation component amplitude threshold If the current second signal is abnormal, Set to 0, otherwise is 1, as shown in the following formula:

[0118] .

[0119] In the embodiment of the present invention, in S6, according to the actual pulse wave signal frequency range, the longest signal period is 2s. To ensure that there are at least two pulse periods in each time window, 5 seconds is used as the time window and 1 second is used as the window moving length. The signal is segmented, and the signal in each time window is autocorrelated. Suppose the signal in a certain time window is , , is the signal index value, For a long time window, the autocorrelation calculation formula is as follows:

[0120]

[0121] , is the autocorrelation signal The independent variable;

[0122] The autocorrelation within a time window is Figure 4 As shown, the signal has four effective peak points, namely ;

[0123] Normalize the above formula to get the normalized autocorrelation value , as shown below:

[0124]

[0125] This normalization method can effectively avoid the difference in autocorrelation amplitude caused by the difference in the number of zero padding. The normalized autocorrelation waveform within a certain time window is as follows: Figure 4 As shown in the figure is the normalized autocorrelation peak point;

[0126] Extracting autocorrelation eigenvalues , Quantitatively represents the periodicity of the pulse signal. The smaller the value, the better the periodicity. , is the normalized autocorrelation peak point, It means taking the maximum value among B, C, and D;

[0127] The periodicity of the pulse wave signal is evaluated by the following formula: Greater than threshold , Set to 0, otherwise is 1, as shown in the following formula:

[0128] ;

[0129] In the calculation, the weighted strategy dynamically weights the periodic component according to the signal characteristics to improve the evaluation accuracy. The weighted autocorrelation function is: , weight Based on the local energy of the signal With frequency characteristics Calculation, such as , optimize eigenvalue extraction and periodic evaluation, where, , To adjust the parameters, is the lag parameter in the autocorrelation function, Represents the pulse wave signal at discrete moments The sampling value of is the discrete time index, To calculate the local energy of the signal and frequency characteristics The window length parameter, Pulse wave signal The autocorrelation function of .

[0130] In the embodiment of the present invention, there are four key points of the pulse wave, such as Figure 5 As shown, in S7, four key points A, B, C, and D of the pulse wave are detected, point A is the starting point of the pulse wave, point B is the peak point of the main wave, point C is the descending isthmus point, and point D is the dicrotic wave point. The data centered on the above four key points are averaged to output four key point templates, and then the key point templates are respectively compared with the data extracted centered on the key points to calculate the Pearson correlation coefficient. The key point position is used as the independent variable and the Pearson correlation coefficient is used as the dependent variable. Nonlinear interpolation is performed to make the sampling rate of the interpolated Pearson correlation coefficient the same as the sampling rate of the original signal, and then downsampling is performed to make the final output Pearson correlation coefficient The sampling rate is 1Hz, that is, one correlation coefficient value is output every 1 second. , Set to 0, otherwise is 1, as shown in the following formula:

[0131]

[0132] in Indicates the decision threshold for signal quality assessment using the Pearson correlation coefficient;

[0133] Normally, the signal quality assessment through the template is mainly based on a single pulse wave key position point, extracting the template. This method is prone to template extraction abnormality, and the sampling rate of the extracted correlation coefficient is low, and the error is large. By increasing the key point position and extracting the correlation coefficient, the probability of false detection caused by template extraction abnormality can be effectively reduced. Table 3;

[0134] In S8, the first-order partial derivative extraction formula is as follows:

[0135]

[0136] in, is the index value of the three-axis discrete data independent variable, Represents the data of each axis of the three-axis data The first-order partial derivative of Indicates the original data of the three-axis acceleration The data on the axis, , , Respectively axis, Axis and axis;

[0137] Calculate the resultant vector of the first-order partial derivative , the calculation formula is as follows:

[0138]

[0139] right With 1 second as the filter window length and 0.5 second as the window moving length, a sliding average filter is performed to output the envelope of the resultant vector. ,when hour, Set to 0, otherwise is 1, as shown in the following formula:

[0140] ;

[0141] represents the body motion determination threshold, , and They are , , The first-order partial derivative of the three-axis data is used to calculate the resultant vector after the first-order difference of the three-axis original data, which can reduce the influence of the baseline drift of the three-axis sensor on the body motion amplitude and improve the accuracy of body motion detection. The body motion detection output is as follows: Figure 6 shown.

[0142] In the embodiment of the present invention, in S9, in sleep monitoring, there are five main sleeping postures: supine, prone, left side, right side, and upright. When the sleeping posture changes, there will definitely be interference to the pulse wave signal. Therefore, the state of the device is judged according to the amplitude and direction of each axis of the three-axis acceleration. For example, if the device deflects to the right, it is judged as:

[0143] ;

[0144] in, Indicates the judgment that the device deflects to the right. , and They represent the three-axis accelerations in , , Amplitude of the axis;

[0145] Figure 7 It is the three-axis acceleration body position state. The S, P, L, R, Up / Down in the figure correspond to the supine position, prone position, left lateral position, right lateral position and upright position respectively;

[0146] In S10, the time point of the first position change As the starting point, if the time interval between the current body position change time point and the previous body position change time point is greater than 20s, the previous body position change time point is taken as the end point of the continuous body position change, recorded as , the number of changes between the starting and ending points of the position change is used The frequency of body position change is Calculated by the following formula:

[0147] ;

[0148] The frequency of body position changes is greater than a certain threshold ,but Set to 0, otherwise is 1, as shown in the following formula:

[0149] ;

[0150] like Figure 7 , between 2617.53 seconds and 2630 seconds, the change frequency is 7.38 times / second. Therefore, if the threshold is determined If it is set to 5, it can be judged that there is body motion interference in this time period;

[0151] Fusion of multi-sensor data to build a dynamic attitude model, using the Kalman filter algorithm to fuse gyroscope and accelerometer data into the attitude transfer matrix Update to improve the accuracy of body motion interference identification, including: , is the discrete time index, is the fusion algorithm update function, For at the moment The collected gyroscope data, For at the moment The collected accelerometer data.

[0152] The device outputs a signal quality label based on the algorithm and displays the corresponding signal abnormality type, as shown in Table 3. Clinicians can determine whether the current output value is credible based on the displayed signal abnormality type, and can also adjust the monitoring status based on the output signal abnormality type to obtain a signal with better quality. The signal prompt priority in the table means that when the signal meets both the conditions of signal quality assessment 1 and the conditions of signal quality assessment 3, the signal abnormality prompt displays the signal abnormality type in quality assessment 1, that is, the signal abnormality type with a smaller signal quality assessment value is displayed.

[0153]

[0154] The embodiments of the present invention disclose preferred embodiments, but are not limited thereto. A person skilled in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. However, as long as they do not deviate from the spirit of the present invention, they are all within the protection scope of the present invention.

Claims

1. A multi-parameter fusion pulse wave signal quality assessment method, characterized in that: The following steps are involved: S1, PPG signal and three-axis acceleration data acquisition; Synchronously collect the raw data of red light, infrared light and three-axis acceleration of the pulse wave; S2, PPG signal endpoint detection; To solve the problem of signal saturation caused by individual and acquisition site differences, endpoint value detection is set, and labels are set according to the relationship between signal amplitude and endpoint value. ; S3, feature extraction of non-pulsating components of PPG signals; Use filters to extract non-pulsating components with large differences between wearing and not wearing, and set labels based on thresholds ; S4, PPG signal pulsation component feature extraction; The basic amplitude signal is extracted by time window processing and amplitude characteristics, set labels according to threshold range S5, PPG signal abnormal pulsation component detection; In the fundamental amplitude signal Based on this, the dynamic threshold is extracted to is the time window, is the window moving length, extract the mean and 3 times the standard deviation of the data between the upper and lower quartiles in each time window as the threshold in each time window, traverse the entire signal, output the pulsation component amplitude threshold, and set the label according to the relationship between the amplitude and the threshold S6, PPG signal segment autocorrelation feature extraction; According to the pulse wave cycle characteristics, the time window is selected to segment the signal autocorrelation to calculate the characteristic value, and the label is determined according to the threshold. ; S7, PPG signal template Pearson correlation coefficient extraction; Detect key points and output templates. Calculate correlation coefficients based on key point templates and set labels based on thresholds after processing. ; S8, three-axis acceleration body motion detection; Extract the first-order partial derivative of each axis of the three-axis data , calculate the first-order partial derivative vector , after filtering, we get the resultant vector envelope , set labels according to thresholds ; S9, three-axis acceleration body position state change detection; Determine the device status based on the amplitude and direction of the three-axis acceleration; S10, three-axis acceleration body position change frequency detection; Calculate the frequency based on the time of body position change and set the label based on the threshold ; S11, abnormal signal display; when When the signal abnormality type prompt is amplitude saturation, the signal abnormality prompt priority is 1, when When the signal abnormality type prompt is not worn, the signal abnormality prompt priority is 2. When the signal abnormality type prompt is low amplitude, the signal abnormality prompt priority is 3. When the signal abnormality type prompt is a large amplitude value, the signal abnormality prompt priority is 4, when When the signal abnormality type prompt is poor signal periodicity, the signal abnormality prompt priority is 5. When the signal abnormality type prompt is signal shape abnormality, the signal abnormality prompt priority is 6. When the signal abnormality type prompt is abnormal body motion amplitude, the signal abnormality prompt priority is 7. When the signal abnormality type prompt is abnormal device status change, the signal abnormality prompt priority is 8, and the signal abnormality prompt will give priority to the signal abnormality types with lower priority.

2. A multi-parameter fusion pulse wave signal quality assessment method according to claim 1, characterized in that: In S1, red light, infrared light and three-axis acceleration axis, axis, The original data of the axis are , , , , Indicates that the PPG signal is express, , the three-axis acceleration is express, .

3. A multi-parameter fusion pulse wave signal quality assessment method according to claim 2, characterized in that: In S2, the pulse wave signal collected in S1 is subjected to endpoint detection, and the endpoint value based on the AD sampling maximum value is set , , when the signal amplitude value is equal to the endpoint value, the output label is 0, otherwise is 1, as shown in the following formula: 。 4. A multi-parameter fusion pulse wave signal quality assessment method according to claim 3, characterized in that: In S3, a 4th-order, zero-phase IIR digital low-pass filter is used to filter out interference components below 0.5 Hz and above 5 Hz, and to extract non-pulsating components of red light and infrared light, where the cutoff frequency of the zero-phase IIR digital low-pass filter is 0.1 Hz; The low-frequency components between 0.5Hz and 5Hz are used as two features in S3. express, , set thresholds based on individual scenarios , ,when ,or , output label is 0, otherwise is 1, as shown in the following formula: 。 5. A multi-parameter fusion pulse wave signal quality assessment method according to claim 4, characterized in that: In S4, 1 second is used as a fixed time window, the difference between the maximum and minimum values ​​of the signal in each time window is extracted, and the basic amplitude signal is output. , ,in, and Represent the basic amplitude signals of red light and infrared light respectively, and then convolve them with a rectangular window of length 5 to output the amplitude characteristics of the pulsating component , ,in, and Respectively represent the amplitude characteristics of the red light and infrared light pulsation components. At the experimental threshold within the range, or At the experimental threshold Within the range, output label Set to 0, otherwise is 1, as shown in the following formula: 。 6. A multi-parameter fusion pulse wave signal quality assessment method according to claim 5, characterized in that: In S5, for , ,in, and Respectively represent the pulsation component amplitude thresholds of red light and infrared light. If the basic amplitudes of red light and infrared light are within the pulsation component amplitude threshold If the current second signal is abnormal, Set to 0, otherwise is 1, as shown in the following formula: 。 7. A multi-parameter fusion pulse wave signal quality assessment method according to claim 6, characterized in that: In S6, the signal is segmented with 5 seconds as the time window and 1 second as the window moving length. The signal in each time window is autocorrelated. Suppose the signal in a certain time window is , , is the signal index value, For the time window length, the autocorrelation calculation formula is as follows: , is the autocorrelation signal The independent variable; Normalize the above formula to get the normalized autocorrelation value , as shown below: Extracting autocorrelation eigenvalues , , Quantitatively represents the periodicity of the pulse signal. The smaller the value, the better the periodicity. , is the normalized autocorrelation peak point, It means taking the maximum value among B, C, and D; The periodicity of the pulse wave signal is evaluated by the following formula: Greater than threshold , Set to 0, otherwise is 1, as shown in the following formula: ； In the calculation, the weighted strategy dynamically weights the periodic component according to the signal characteristics to improve the evaluation accuracy. The weighted autocorrelation function is: , weight Based on the local energy of the signal With frequency characteristics Calculation, such as , optimize eigenvalue extraction and periodic evaluation, where, , To adjust the parameters, is the lag parameter in the autocorrelation function, Represents the pulse wave signal at discrete moments The sampling value of is the discrete time index, To calculate the local energy of the signal and frequency characteristics The window length parameter, Pulse wave signal The autocorrelation function of .

8. A multi-parameter fusion pulse wave signal quality assessment method according to claim 7, characterized in that: In S7, four key points of the pulse wave A, B, C, and D are detected, and the data centered on the above four key point positions are averaged to output four key point templates. Then, the key point templates are respectively used to calculate the Pearson correlation coefficient with the data extracted centered on the key points. The key point positions are used as independent variables and the Pearson correlation coefficient is used as the dependent variable. Nonlinear interpolation is performed to make the sampling rate of the interpolated Pearson correlation coefficient the same as the sampling rate of the original signal. Then, downsampling is performed to make the final output Pearson correlation coefficient The sampling rate is 1Hz, that is, one correlation coefficient value is output every 1 second. , Set to 0, otherwise is 1, as shown in the following formula: in Indicates the decision threshold for signal quality assessment using the Pearson correlation coefficient; In S8, the first-order partial derivative extraction formula is as follows: in, is the index value of the three-axis discrete data independent variable, Represents the data of each axis of the three-axis data The first-order partial derivative of Indicates the original data of the three-axis acceleration The data on the axis, , , Respectively axis, Axis and axis; Calculate the resultant vector of the first-order partial derivative , the calculation formula is as follows: right With 1 second as the filter window length and 0.5 second as the window moving length, a sliding average filter is performed to output the envelope of the resultant vector. ,when hour, Set to 0, otherwise is 1, as shown in the following formula: ； in, represents the body motion determination threshold, , and They are , , The first-order partial derivative of the axis data.

9. A multi-parameter fusion pulse wave signal quality assessment method according to claim 8, characterized in that: In S9, the device state is determined according to the amplitude and direction of the three-axis acceleration. For example, if the device deflects to the right, it is determined as: ； in, Indicates the judgment that the device deflects to the right. , and They represent the three-axis accelerations in , , Amplitude of the axis; In S10, the time point of the first position change As the starting point, if the time interval between the current body position change time point and the previous body position change time point is greater than 20s, the previous body position change time point is taken as the end point of the continuous body position change, recorded as , the number of changes between the starting point and the end point of the position change is used The frequency of body position change is Calculated by the following formula: ； The frequency of body position changes is greater than a certain threshold ,but Set to 0, otherwise is 1, as shown in the following formula: ； Fusion of multi-sensor data to build a dynamic attitude model, using the Kalman filter algorithm to fuse gyroscope and accelerometer data into the attitude transfer matrix Update to improve the accuracy of body motion interference identification, including: , is the discrete time index, is the fusion algorithm update function, For at the moment The collected gyroscope data, For at the moment The collected accelerometer data.

Images