Surface electrophysiological signal processing method and system for removing signal phase shifts - Patents.com

Abstract

The present invention relates to the medical device field, specifically to a surface electrophysiological signal processing method and system for removing signal phase shift. The method of the present invention includes acquiring surface electrophysiological signals from the skin surface of the human body, processing the surface electrophysiological signal data using a zero-phase filter based on a stationary wavelet transform to obtain wavelet detail coefficients and approximation coefficients, and detecting QRS complexes for the detail coefficients of each layer, obtaining heartbeat detection results using a weighted voting scheme, and realizing information source fusion. The present invention also provides a system and apparatus that apply the above method. The present invention can accurately extract electrophysiological signal components with fixed time-domain waveform features, such as QRS complexes, from electrocardiogram signals in high-noise environments without generating waveform distortion of the extracted signal due to phase shift. The method is highly applicable to application scenarios such as acquiring long-term dynamic electrocardiograms from non-traditional sites such as the upper arm, behind the ear, and inside the ear, and has good prospects for application.

Description

[Technical Field]

[0001] The present invention relates to the field of medical devices, and more particularly to a surface electrophysiological signal processing method and system for removing phase shifts in the signals. [Background technology]

[0002] Cardiovascular disease is the leading cause of death worldwide. Approximately 50% of deaths are usually attributed to arrhythmias. Accurately recording and detecting various arrhythmias is crucial to preventing these arrhythmia-related deaths. Heartbeat detection is crucial for determining heart rate and associated arrhythmias. The electrocardiogram (ECG) is the most widely used clinical tool, graphically displaying cardiac electrical activity from the human body surface. Most importantly, the ECG waveform contains the QRS complex (representing ventricular depolarization), which plays a fundamental role in heartbeat detection. Therefore, accurately detecting the QRS complex is the most crucial step before any ECG analysis. One of the many methods for detecting QRS complexes in an ECG is the discrete wavelet transform (DWT). The DWT satisfies the law of energy conservation, achieving perfect signal reconstruction and offering efficient computation. Therefore, it is widely used for heartbeat detection. Researchers use different decomposition levels in the DWT and assume that the QRS wave is characterized by several detailed coefficients based on its spectrum. Considering the sampling frequency of the analyzed ECG signal, existing studies usually select detailed coefficients 1–4 (corresponding to frequencies between 11.25 and 180 Hz), 3–5 (corresponding to frequencies between 5.75 and 45 Hz), and 4 (corresponding to frequencies between 15.6 and 31.1 Hz) for further analysis.

[0003] However, the DWT also has drawbacks, including high sensitivity to signal offsets and reduced temporal resolution at coarse scales. Therefore, several subsequent studies have focused on reducing or eliminating the offset variance of the DWT. Here, the stationary wavelet transform (SWT) method is commonly used in ECG denoising. Instead of downsampling the filtered signal, the SWT upsamples the filter coefficients at each step. Therefore, the resulting coefficients have the same length as the original signal. Due to its redundant representation, the SWT is translation-invariant. Therefore, it is suitable for detecting the precise location of target events in the signal, such as edges. However, the SWT has a particular drawback: a phase shift with respect to the original signal. This is because most wavelet decomposition methods use FIR filter groups and many wavelet decomposition methods do not even have linear phase. Due to this phase shift, the QRS complexes are horizontally projected to different temporal positions with different SWT detail coefficients, all of which are different from their positions in the original signal. Clearly, this phase shift is undesirable for heartbeat detection. Although the offset amount is limited to the order of milliseconds, synchrony is a critical issue in QRS detection applications, especially in real-time applications. Furthermore, because the phase shift follows the SWT scale, it is expected that the phase shift increases as the scale increases. Therefore, the final detail coefficients and the final approximation coefficients have the largest phase shift compared to the original input signal. As a result, more information may be lost at higher ranges. Removing these phase shifts becomes a more sensitive issue in noisy environments where it is more difficult to locate the QRS complex. In the field of ECG data processing, existing techniques currently lack a method for removing such phase shifts. Therefore, existing QRS complex detection is performed using only one layer of wavelet detail coefficients, while the large amount of useful information contained in the wavelet detail coefficients of other layers is not analyzed.This can adversely affect both the accuracy and robustness of the QRS complex detection.

[0004] Meanwhile, several studies have focused on designing ECG devices with non-standard ECG leads, such as those on the upper arm, to improve device wearability, convenience, reusability, and the potential for home use. However, the surface electrophysiological signals collected by these devices in non-clinical standard positions contain not only ECG signals but also electromyogram (EMG) signals. Therefore, one of the main challenges in research on these devices is how to improve the acquisition quality of ECG signals under the interference of EMG artifacts, which have more energy than ECG signals. Due to the presence of EMG artifacts, feature extraction and analysis without phase shift compared to ECG signals is crucial for heartbeat detection. Considering the widespread use of SWT in these scenarios, removing the resulting phase shift becomes even more important.

[0005] Considering the importance of minimizing phase shift, Daubechies formulated an approximate zero-phase filter widely used in the DWT (Ingrid Daubechies, “Ten Lectures on Wavelets,” 1992). However, these filters cannot achieve exact zero phase. To solve this problem, Percival introduced a zero-phase wavelet called the Zefret transform (D. Percival, “Discrete Wavelet Transforms Based on Zero-Phase Daubechies Filters.”). Subsequently, Lenis et al. proposed an improved version of the SWT to solve the lack of phase shift properties (Biomed. Eng. / Biomed. Tech., vol. 61, no. 1, pp. 37–56, February). While applying the standard SWT to a signal, they invert the signal and apply the SWT to the resulting sequence. Then, they invert the transform again and sum the transformed signal at each scale. This ensures that the phase shift of the resulting coefficients is zero. Therefore, we claim that this method can detect the onset of the P wave with very high accuracy. Removing the phase shift at higher scales provides an opportunity to combine information from different scales to achieve more accurate heartbeat detection. However, these existing methods for processing high-noise data were not designed for ECG. When applied to ECG data, they still suffer from poor processing effectiveness and low robustness. This makes them less applicable to data collected by ECG equipment configured with non-standard ECG leads, such as those on the upper arm. Summary of the Invention [Problem to be solved by the invention]

[0006] In response to the shortcomings of existing techniques, the present invention provides a surface electrophysiological signal processing method and system for removing signal phase shifts. The objective of the present invention is to process electrocardiogram signals by combining an SWT-based zero-phase filter group and a voting policy for different SWT scales, thereby improving noise handling ability and method robustness, so that the method can be more applicable to data collected by ECG equipment configured with non-standard ECG leads, such as those on the upper arm. [Means for solving the problem]

[0007] Surface electrophysiological signal processing methods for removing signal phase shifts include: Step 1: acquiring a surface electrophysiological signal from a skin surface of a human body; Step 2: Processing the surface electrophysiological signals using a zero-phase filter based on stationary wavelet transform to obtain wavelet detail coefficients and approximation coefficients of several layers; and step 3 of detecting QRS complexes from the wavelet detail coefficients of each layer and obtaining heartbeat detection results from the wavelet detail coefficients of each layer using a weighted voting method.

[0008] Preferably, the surface electrophysiological signal comprises: 1) Clinical electrocardiogram signals are collected from standard electrode positions defined by the clinical electrocardiogram. or 2) acquired based on collection from a non-clinical standard location.

[0009] Preferably, the non-clinical standard locations include the upper arm, wrist, behind the ear, in the ear, lower back, both legs or both feet.

[0010] Preferably, step 2 specifically comprises: Step 2.1: Perform a second-order Daubechies wavelet transform as the mother wavelet, decompose it into 4 to 7 layers, and obtain 4 to 7 wavelet detail coefficients and one approximation coefficient. Step 2.2: Remove the phase shift when performing the wavelet decomposition at each layer using the following steps: 1) First, we perform standard wavelet filtering, and the equation becomes X(z)H(z)=D(z) (where z is the z-transform operator, X(z) is the z-transform of the signal, H(z) is the wavelet detail function of this layer, and D(z) is the coefficient of this layer.) 2) By performing a time-reverse operation, the equivalent equation in the z-transform domain is JPEG2025527016000002.jpg161703) Repeat the filtering operation in step 1 on the output of step 2, JPEG2025527016000003.jpg171704) Perform a time reverse operation on the output of step 3, JPEG2025527016000004.jpg15170D`(z) is the final zero-phase shift wavelet detail coefficient of this layer, Then, step 2.2 performs the same operation on the approximation coefficients by the wavelet approximation function G(z) of this layer.

[0011] Preferably, the wavelet detail coefficients have five layers, and the frequency components of the five layers of wavelet detail coefficients are 64 to 128 Hz, 32 to 64 Hz, 16 to 32 Hz, 8 to 16 Hz, and 4 to 8 Hz, respectively.

[0012] Preferably, in step 3, the method for performing QRS complex detection is the Pan-Tompkins algorithm.

[0013] Preferably, in step 3, the specific content of the voting method is: After detecting the QRS complex for each layer's detailed coefficients, a voting sparse with corresponding weights is obtained. The final result of weighting the voting coefficients of all layers determines the final QRS detection result. The voting process is as follows: A moving window of length 200 ms is used over the detected heartbeat sequence, and if the result of weighting the coefficients within the window exceeds a pre-set threshold, a heartbeat is detected and the position of the detected heartbeat is set to the average time position of all positive votes. If at least two heartbeats were detected with a time interval of less than 200 ms, they were combined into one heartbeat using the time averaging method.

[0014] The present invention further provides a system for implementing the above-described surface electrophysiological signal processing method, the system comprising: an input module for inputting surface electrophysiological signal data; a feature extraction module for processing the surface electrophysiological signal data using a zero-phase filter based on stationary wavelet transform to obtain 4 to 7 layers of wavelet detail coefficients and one approximation coefficient; and a voting module for detecting QRS complexes for all wavelet detail coefficients and obtaining heartbeat detection results from all wavelet detail coefficients using a weighted voting method.

[0015] The present invention further provides a heart rate detection device comprising a surface electrophysiological signal detection device and the above system.

[0016] The present invention further provides a computer-readable storage medium having stored thereon a computer program for implementing the above-described surface electrophysiological signal processing method.

[0017] "Surface electrophysiological signals" of the present invention include signals indicative of cardiac electrical activity at these locations (ECG signals), as well as signals from other sources such as skeletal muscle electrical signals, smooth muscle electrical signals, and neuronal electrical signals.

[0018] The present invention provides a zero-phase filter group based on SWT and voting policies for different SWT scales that can accurately extract electrophysiological signal components with fixed time-domain waveform features, such as the QRS complex in an electrocardiogram signal, from the surface electrophysiological signal without generating waveform distortion due to phase shift. The method of the present invention enhances robustness against noise. Experiments have demonstrated that the method of the present invention outperforms existing techniques in two indicators: sensitivity and positive predictive value. Therefore, the present invention can accurately detect QRS complexes in high-noise environments (e.g., high skeletal muscle electrical signals, smooth muscle electrical signals, and neuronal electrical signals), and is highly applicable to application scenarios such as electrocardiogram detection on the upper arm, demonstrating promising prospects for application.

[0019] It is apparent that, based on the above content of the present invention, various other modifications, substitutions or changes can be made in accordance with ordinary technical knowledge and conventional means in the art without departing from the above basic technical idea of the present invention.

[0020] The above content of the present invention will be described in more detail below through specific embodiments in the form of examples. However, this should not be understood as meaning that the scope of the above subject matter of the present invention is limited to the following examples. All technologies realized based on the above content of the present invention belong to the scope of the present invention. [Brief explanation of the drawings]

[0021] [Figure 1] FIG. 1 is a block diagram of a zero-phase filter group based on an SWT. [Figure 2] This is a diagram showing the voting method, and the output of Pan-Tompkins applied to different detail coefficients di is shown in the diagram. An example of a moving window is shown in the pink rectangle. If we want to consider only d3 or d2 in this window, the current heartbeat data will be lost, but the system can prevent this information from being lost. [Figure 3]Figure 1 shows the electrode positions in a data detection experimental paradigm designed to validate this zero-phase surface electrophysiological signal system. Three high-density electrode arrays, each containing 64 electrodes (yellow), were placed on the left upper arm, with a reference electrode placed at the elbow. Simultaneously, the lead II channel of a standard electrocardiogram was collected; the signal electrode was placed at the LA (upper right shoulder) and the reference electrode was placed at the RL (right hip). [Figure 4] Figure 1 shows the detailed and approximate coefficients of applying SWT and zero-phase filter group to standard ECG lead II (left) and typical brachial (right) signals, respectively. SWT has obvious phase differences in various layers, but there is no corresponding phase difference in the zero-phase filter group. [Figure 5] This figure shows the performance of the SWT-based zero-phase filter group and the conventional SWT method when detecting heartbeats from the upper arm signal before voting (i.e., using only the detailed and approximate coefficients). The sensitivity (left) and positive predictive value (right) are shown. The D value on the horizontal axis is the maximum allowable difference (in milliseconds) between the detected Q wave peak position and the Q wave peak position in lead II of the standard electrocardiogram. The error bars represent the mean value + / - standard deviation of the SE (left) and PPV (right) across all subjects and trials. It is clear that when the requirements for the Q wave peak position are high (the D value is small), the performance of the zero-phase filter group is significantly higher than that of the conventional SWT method. [Figure 6]This figure shows the performance of the SWT-based zero-phase filter group and conventional SWT in detecting heartbeats from brachial signals after voting, i.e., sensitivity (left) and positive predictive value (right). The D value on the horizontal axis is the maximum allowable difference (in milliseconds) between the detected Q wave peak position and the Q wave peak position in lead II of a standard electrocardiogram. The error bars represent the mean value + / - standard deviation of the SE (left) and PPV (right) across all subjects and trials. Clearly, when the requirements for the Q wave peak position are high (the D value is small), the performance of the zero-phase filter group is significantly higher than that of the conventional SWT method. At the same time, the performance is significantly improved compared to the results in Figure 5 (before voting). DETAILED DESCRIPTION OF THE INVENTION

[0022] It should be particularly noted that algorithms for steps such as data collection, transmission, storage, and processing, which are not specifically described in the embodiments, as well as hardware structures, circuit connections, etc., which are not specifically described, can all be realized by content disclosed in existing technology.

[0023] Example 1: Surface electrophysiological signal processing method and system for removing signal phase shift The system of the present invention comprises: an input module used to input surface electrophysiological signal data, which in some preferred solutions is collected from the upper arm, wrist, behind the ear or inside the ear; a feature extraction module for processing the surface electrophysiological signal data using a zero-phase filter based on stationary wavelet transform to obtain 4 to 7 layers of wavelet detail coefficients and one approximation coefficient; and a voting module for detecting QRS complexes for all wavelet detail coefficients and obtaining heartbeat detection results from all wavelet detail coefficients using a weighted voting method.

[0024] Specifically, the method for performing surface electrophysiological signal processing using the above system includes: Step 1: acquiring a surface electrophysiological signal from a skin surface of a human body; Step 2: processing the surface electrophysiological signals using a zero-phase filter based on stationary wavelet transform to obtain wavelet detail coefficients and approximation coefficients of several layers; Specifically, step 2 includes: Step 2.1: Perform a second-order Daubechies wavelet transform as the mother wavelet, decompose it into 4 to 7 layers, and obtain 4 to 7 wavelet detail coefficients and one approximation coefficient. Step 2.2: Remove the phase shift when performing the wavelet decomposition at each layer using the following steps: 1) First, we perform standard wavelet filtering, and the equation becomes X(z)H(z)=D(z) (where z is the z-transform operator, X(z) is the z-transform of the signal, H(z) is the wavelet detail function of this layer, and D(z) is the coefficient of this layer.) 2) By performing a time-reverse operation, the equivalent equation in the z-transform domain is JPEG2025527016000005.jpg181703) Repeat the filtering operation in step 1 on the output of step 2, JPEG2025527016000006.jpg171704) Perform a time reverse operation on the output of step 3, JPEG2025527016000007.jpg17170D`(z) is the final zero-phase shift wavelet detail coefficient of this layer, Next, step 2.2 involves performing the same operation on the approximation coefficients by the wavelet approximation function G(z) of this layer, and step 3 of detecting QRS complexes for the wavelet detail coefficients of each layer using the Pan-Tompkins algorithm, and obtaining heartbeat detection results from the wavelet detail coefficients of each layer using a weighted voting method.

[0025] The specific details of the voting method are as follows: After performing QRS complex detection for all wavelet detail coefficients, votes with equal weights are obtained, and the votes determine whether a heartbeat is detected or not; After detecting the QRS complex for each layer's detailed coefficients, a voting sparse with corresponding weights is obtained. The final result of weighting the voting coefficients of all layers determines the final QRS detection result. The voting process is as follows: A moving window of 200 ms in length is used in the detected heartbeat sequence. If the result of weighting the coefficients within the window exceeds a preset threshold, a heartbeat is detected, and the position of the detected heartbeat is set to the average value of the time positions of all positive votes. The weighting coefficients may be set according to the equal weight method or based on the information contained in the wavelet detail coefficients of each layer. If at least two heartbeats are detected with a time interval of less than 200 ms, they are combined into one heartbeat by time averaging.

[0026] The technical solution of the present invention will be further explained below through experiments, and the steps not specifically described are all the same as those in Example 1.

[0027] Experimental Example 1: SWT-based zero-phase filter group combined with voting policy to detect heartbeat from human upper arm 1. Experimental Method 1. Description of the dataset The dataset in this study consisted of a total of 193 (64 *Data was collected from nine participants (all male, aged 24–34 years) using a 3+1 channel recording system. All participants were free of cardiac disease. As shown in Figure 3, the electrodes included one channel on the right shoulder and three high-density 64-channel grids positioned around the upper arm. The multi-channel grids were equidistant from each other, with a reference point located at the elbow. The reference point for the right shoulder electrode was placed on the left hip, corresponding to standard ECG lead II. Each participant signed a written informed consent form before the experiment began. This experiment was approved by the University of Waterloo's Research Ethics Office.

[0028] Participants were asked to sit in a comfortable chair and rest their left arm on the chair. Data was then simultaneously recorded from 193 channels using an EMG-USB2+ biosignal amplifier and a hardware bandpass filter ranging from 0.1 Hz to 500 Hz. The sampling frequency for these records was 2048 Hz. The data was then low-pass filtered using a Butterworth filter with a cutoff frequency of 100 Hz. The data was then downsampled to a sampling frequency of 256 Hz.

[0029] The data in this example experiment was collected from the upper arm and falls within the scenario of extracting cardiac electrical activity in a high-noise environment.

[0030] 2. Data processing Experimental group: The method and system used in this part are the same as in Example 1. Comparison group: Using the standard stationary wavelet transform (SWT) in existing technology.

[0031] 3. Evaluate and statistically analyze the sensitivity (SE) and positive predictive value (PPV). The formula is, JPEG2025527016000008.jpg37170 (Here, TP is a true positive, a heart rate (from lead II ECG) detected within a predefined time interval of the actual heart rate. This time interval (D) ranges from 50ms to 20ms and has a duration of 10ms. FN is a false negative, a heart rate that was not detected (missed), and FP is a heart rate that was incorrectly detected.)

[0032] The ability of the algorithm to detect heartbeats was demonstrated by SE, while PPV demonstrated the accuracy of detection.

[0033] Each participant's data (240 seconds) is divided into 12 20-second trials. Therefore, a t-test is used to compare the differences in the performance vectors (SE and PPV) of the two algorithms of length 96 (8 × 12 = 96) under different conditions.

[0034] 2. Experimental results As shown in Figure 4, the left panel shows the results of applying the original SWT and the zero-phase filter group resulting from step 3 of Example 1 to an ECG signal. The left panel shows the detail and approximation coefficients when applying the SWT to an ECG signal, and compares them with the coefficients of the zero-phase filter group based on the SWT. As can be seen, this filter group can prevent information loss due to zero delay, especially at higher levels where there is cumulative delay. Next, the same comparison was performed on selected channels of the upper arm (shown in the right panel of Figure 4). The comparison results showed that step 3 of Example 1 eliminated the delay, and all wavelet detail coefficient scales achieved synchronization with the ECG signal.

[0035] Statistical analysis of SE and PPV Before applying the voting method, we compare the performance of the zero-phase filter group used in Step 3 of Example 1 with the original SWT. The results for all scales and four time intervals (D = 20, 30, 40, 50 ms) are shown in Figure 5. We can see that the zero-phase filter group has significantly better performance than the SWT at the fourth and fifth scales in all time intervals. In more challenging situations, the difference becomes more pronounced as D gets lower.

[0036] For D = 50 ms, d4 achieves SEs of 0.82 and 0.97 for the SWT and zero-phase filter groups, respectively, and this performance decreases as D decreases. At the most challenging time interval (D = 20 ms), the zero-phase filter group outperforms the SWT with an SE of 0.14 vs. 0.56. Meanwhile, for D = 50 ms, d5 already has a significant advantage over the zero-phase filter group (0.11 vs. 0.74). For D = 20 ms, this difference becomes 0.02 vs. 0.48. This means that in the most challenging situations, almost no heartbeats are detected by the SWT final detail coefficients, while almost half of the heartbeats are detected by the zero-phase filter group. When D is set to 40 ms and 50 ms, there is no obvious difference between d1 through d3. However, when D is set to 20 ms and 30 ms, the zero-phase filter pulse achieves a significant performance advantage over d3. PPV also follows the same trend as SE in all time intervals.

[0037] Figure 6 shows the results of the voting scheme using SWT d1-d4 and the zero-phase filter group. The zero-phase filter group had significantly higher SE and PPV at all Ds (p<0.05 for D=50ms, p<0.01 for D=20, 30, 40ms). This indicates that the zero-phase filter group is more robust to shortening of the allowable interval between heartbeats in detection applications.

[0038] Comparing the results of the voting method in the zero-phase filter group with the best individual ad before voting, when D = 50 ms, a.d4 had a lower SE (0.98 vs. 0.97) but a higher PPV (0.95 vs. 0.96). When D was set to 40 ms, the voting method outperformed the a.d4 method, with an SE of 0.94 vs. 0.96. The voting method also had an advantage in PPV, with a parameter of 0.93 vs. 0.94. When D = 30 ms, the voting method increased SE (0.86 vs. 0.87) compared to a.d4, but decreased PPV (0.84 vs. 0.85). When D was set to 20 ms, the voting method had the greatest advantage. Compared to a.d3, SE (0.67 vs. 0.75) and PPV (0.67 vs. 0.73) increased significantly.

[0039] As can be seen from the above experimental results, when using the method of the present invention, the SE and PPV are 0.98±0.04 and 0.95±0.09, respectively, when detecting heartbeats at 50 ms intervals. When the intervals are 40 ms, 30 ms, and 20 ms, the SE is 0.96±0.07, 0.87±0.12, and 0.75±0.15, respectively. PPV also follows a similar trend, with PPV being 0.94±0.07, 0.84±0.14, and 0.73±0.16, respectively, at intervals of 40 ms, 30 ms, and 20 ms from the actual heartbeat. Furthermore, both the SE and PPV of the method of the present invention are superior to those of the original SWT. This indicates that the method of the present invention has excellent accuracy and robustness when detecting high-noise data.

[0040] As can be seen from the above examples and experimental examples, the method and system of the present invention can accurately detect QRS complexes in high-noise environments, and is highly applicable to application scenarios such as detecting electrocardiograms on the upper arm, and has good prospects for application.

Claims

1. 1. A surface electrophysiological signal processing method for removing a phase shift in a signal, comprising: Step 1: acquiring a surface electrophysiological signal from a skin surface of a human body; Step 2: Processing the surface electrophysiological signals using a zero-phase filter based on stationary wavelet transform to obtain wavelet detail coefficients and approximation coefficients for several layers; and (3) detecting QRS complexes for the wavelet detail coefficients of each layer, and obtaining heartbeat detection results from the wavelet detail coefficients of each layer by a weighted voting method.

2. The surface electrophysiological signal comprises: 1) Acquired from standard electrode positions defined by clinical electrocardiogram signals; 2) The surface electrophysiological signal processing method according to claim 1, wherein the signal is acquired based on a form of collection from a non-clinical standard position.

3. 3. The surface electrophysiological signal processing method of claim 2, wherein the non-clinical standard locations include the upper arm, the wrist, behind the ear, in the ear, the lower back, both legs, or both feet.

4. Specifically, step 2 is Step 2.1: performing a second-order Daubechies wavelet transform as a mother wavelet, and decomposing it into 4-7 layers to obtain 4-7 wavelet detail coefficients and one approximation coefficient; Step 2.2 removing the phase shift when performing wavelet decomposition at each layer using the following steps: 1) First, we perform standard wavelet filtering, and the equation becomes X(z)H(z)=D(z) (where z is the z-transform operator, X(z) is the z-transform of the signal, H(z) is the wavelet detail function of this layer, and D(z) are the coefficients of this layer), 2) Perform a time-reverse operation, and find that the equivalent expression in the z-transform domain is 3) Repeat the filtering operation of step 1 on the output of step 2. 4) Perform a time-reverse operation on the output of step 3; D'(z) is the final zero-phase shift wavelet detail coefficient of this layer, 2.

2. A method for processing surface electrophysiological signals according to claim 1, further comprising the step of: performing the same operation on the approximation coefficients by the wavelet approximation function G(z) of this layer.

5. 5. The surface electrophysiological signal processing method according to claim 4, wherein the wavelet detail coefficients are in five layers, and the frequency components of the wavelet detail coefficients in the five layers are 64 to 128 Hz, 32 to 64 Hz, 16 to 32 Hz, 8 to 16 Hz, and 4 to 8 Hz, respectively.

6. 2. The method of claim 1, wherein in step 3, the method for detecting QRS complexes is the Pan-Tompkins algorithm.

7. In step 3, the specific contents of the voting method are as follows: After detecting the QRS complex for the detailed coefficients of each layer, a voting sparse with corresponding weights is obtained. The final result of weighting the voting coefficients of all layers determines the final QRS detection result. The voting process is as follows: A moving window of length 200 ms is used on the detected heartbeat sequence, and if the result of weighting the coefficients within the window exceeds a pre-set threshold, a heartbeat is detected and the position of the detected heartbeat is set to the average time position of all positive votes; 2. The method of claim 1, wherein if at least two heartbeats are detected with a time interval of less than 200 ms, they are combined into one heartbeat by a time averaging method.

8. A system for implementing the surface electrophysiological signal processing method according to any one of claims 1 to 7, comprising: an input module for inputting surface electrophysiological signal data; a feature extraction module for processing the surface electrophysiological signal data using a zero-phase filter based on stationary wavelet transform to obtain wavelet detail coefficients of 4 to 7 layers and one approximation coefficient; a voting module for performing QRS complex detection on all wavelet detail coefficients and obtaining heart rate detection results from all wavelet detail coefficients using a weighted voting scheme.

9. A heartbeat detection device, characterized in that it comprises a surface electrophysiological signal detection device and a system according to claim 8.

10. A computer-readable storage medium storing a computer program for implementing the surface electrophysiological signal processing method according to any one of claims 1 to 7.

Images