CN116831589A - An optimal lead signal selection method for long-term wearable ECG signals - Google Patents

Abstract

The invention provides a method for selecting an optimal lead signal of a long-term wearable electrocardiosignal, which comprises the following steps: respectively establishing three quality-level dynamic electrocardiosignal data sets, respectively calculating quality parameters KSQI, PSQI and QQI of the three data sets, fusing the results of the SQI, and calculating the comprehensive quality score of a certain lead signal; calculating a composite quality score for each lead signal over the same time period; the lead signal with the highest score is taken as the electrocardiosignal with the best quality to be selected. The cardiac beat information with the best quality obtained by the optimal lead signal selection method can be subjected to cardiac beat classification by using a deep learning algorithm, and can also be subjected to scatter diagram depiction and cardiac rhythm type detection according to the distribution condition of the scatter diagram and the slope of the attractor contour line. According to the invention, by analyzing the wearable electrocardiosignals in a long time, signals with better quality and more stability can be selected from signals with different leads in the same time period.

Description

An optimal lead signal selection method for long-term wearable ECG signals

Technical field

The invention belongs to the field of ECG data processing. Specifically, it relates to an optimal lead signal selection method for long-term wearable ECG signals.

Background technique

Electrocardiogram (ECG) is a technology that records the electrophysiological activity of the heart through the chest through time. The electrical potential transmission of the heart is detected using electrodes attached to the surface of human skin. ECG results are usually displayed in waveforms. A normal and complete heart beat basically includes P wave, QRS wave, and T wave. The measurement and evaluation of heart rate is represented by the interval between R wave and R wave. When measuring ECG signals, you need to choose different lead systems and connect sensor electrodes to multiple parts of the body. Common lead systems include single lead, 8-lead, 12-lead, 18-lead, etc. Signals without leads at the same time will have different intensities, but they can all reflect the heart's activity information to a certain extent.

Conventional electrocardiogram can record cardiac electrical activity, is simple to operate and low in cost. However, the onset of cardiovascular disease is random and non-sustained. Usually, patients cannot capture the disease well in the short time when ECG is measured in the hospital. With the rapid maturity of Internet of Things technology, a wearable dynamic ECG detector that collects 24-hour dynamic ECG signals in real time has been realized through the combination of smartphone APP and professional chart reading and analysis software. Holter electrocardiography is a process of recording the patient's cardiac electrical activity continuously for 24 hours or more in daily life through a dynamic electrocardiograph, and analyzes and processes it with the help of a computer to discover cardiac activity that is not easily found during routine surface electrocardiography examinations. traces of. Although the collection of dynamic electrocardiogram is not restricted by time and place and can be collected anytime and anywhere, the problem that comes with it is that the signal source of the collection equipment not only includes the physiological activity of the heart, but also includes the electromyographic signal generated by the patient's activity and external noise. , prone to unstable signal quality.

The ECG scatter plot (Lorenz plot) is a large number of continuous ECG R-R interval diagrams traced using an iterative method, which can reflect the special evolution rules of nonlinear systems. The use of scatter plots drawn by long-term ECG signals can help doctors understand the trajectory of cardiac activity faster and more accurately. The signal of the ECG scattergram generally only requires the R-R interval information of the dynamic ECG. In this way, simple information can be used to retain the trajectory information of the patient's cardiac activity, greatly reducing the storage space of the dynamic ECG signal. In order to ensure the accuracy of R-R interval information, it is necessary to conduct quality inspection on multi-lead ECG signals and select the lead signal with the best quality. Based on the best quality lead signal, the R-R interval is calculated to ensure the effectiveness of the scatter plot to a greater extent.

Solution 1: Assessment of ECG signal quality in intensive care patients

Comprehensive quality index (ECGSQI) of ECG signal:

bSQI(k) represents the matching degree of the two results obtained by using two R-wave detection algorithms to identify the same segment of ECG signals. iSQI represents beat-to-lead beat-to-beat matching signal quality. kSQI represents the kurtosis of the ECG signal, which is a measure of the Gaussianity of the signal. When the kurtosis value of the signal is greater than 5, kSQI is taken as 1. sSQI represents the power ratio of ECG signals in different frequency bands;

Disadvantage: When calculating bSQI, if a QRS recognition algorithm misses detection because the signal amplitude is too low or a tall P wave or T wave is mistakenly detected as a QRS wave, it is difficult for bSQI to truly reflect the signal quality. And the parameter selection is based on the MIT multi-parameter intelligent critical care database II, which is one-sided and not suitable for dynamic electrocardiography.

Option 2: Single-lead ECG signal SQI quality evaluation mechanism based on simple heuristic fusion and fuzzy comprehensive evaluation

According to each SQI of the ECG signal: R-wave matching degree qSQI; QRS power spectrum distribution pSQI; kurtosis kSQI and baseline relative power basSQI. Combining Cauchy distribution, rectangular distribution and trapezoidal distribution, the membership function of SQI is quantified and the fuzzy vector is established. Bounded operators are selected for fuzzy synthesis, and weighted membership functions are used for evaluation and classification.

Disadvantage: There is no golden indicator to determine the quality of ECG. This algorithm can only distinguish between high-quality and low-quality ECGs, but it is not conducive to selecting the optimal leads.

Contents of the invention

The purpose of the present invention is to address the shortcomings of the existing technology. The present invention provides an optimal lead signal selection method for long-term wearable ECG signals.

In order to achieve the above objects, the technical solutions adopted by the present invention are:

A first aspect of the present invention provides an optimal lead signal selection method for long-term wearable ECG signals, including:

Step 1: Read the dynamic ECG signal and establish three quality levels of dynamic ECG signal data sets, namely, a data set based on signal kurtosis, a data set based on signal power, and a matching data set based on different R peak detection algorithms;

Step 2: Calculate the quality parameters of dynamic ECG signal data sets of three quality levels respectively;

(1) Calculate the quality parameter kSQI based on the signal kurtosis data set:

in, is the kurtosis of the ECG signal;

kSQI is used to reflect the sharpness of the peak. The smaller the kurtosis of the ECG signal, the more likely it is noise. When kSQI<0, the signal quality is considered poor;

(2) Calculate the quality parameter pSQI based on the signal power data set:

According to the power spectrum distribution of the QRS wave, calculate the ratio of the energy of the QRS wave to the energy of the ECG signal;

in, It is the power of QRS wave in the 6Hz~30Hz band;/> It is the power of the ECG signal in the 0Hz~125Hz band; when pSQI is between 0.5~0.72, the signal quality is better;

(3) Calculate the quality parameter qSQI of matching data sets based on different R peak detection algorithms:

Compare the matching degree of R waves detected by two QRS wave detection algorithms; among them, the two QSR wave detection algorithms selected are the algorithm DF based on digital filtering and the algorithm LT based on length transformation;

Represents the number of QRS waves detected in the same segment of the ECG signal by the DF algorithm and the LT algorithm respectively. Represents the number of QRS waves detected by both algorithms at the same time;

Step 3: Fusion of the above multiple SQI results to calculate the comprehensive quality score of a certain lead signal:

Step 4: Calculate the comprehensive quality score of each lead signal in the same time period;

The lead signal with the highest score is selected as the best quality ECG signal.

A second aspect of the present invention provides an optimal lead signal selection system for long-term wearable ECG signals, including:

The data reading module is used to read dynamic ECG signals and establish three quality levels of dynamic ECG signal data sets, namely, based on signal kurtosis data sets, based on signal power data sets, and based on different R peak detection algorithm matching data sets. ;

The quality parameter calculation module of dynamic ECG signal data sets is used to calculate the quality parameters of dynamic ECG signal data sets of three quality levels, including kSQI calculation module, pSQI calculation module and qSQI calculation module;

The kSQI calculation module is used to calculate the quality parameter kSQI based on the signal kurtosis data set:

in, is the kurtosis of the ECG signal;

kSQI is used to reflect the sharpness of the peak. The smaller the kurtosis of the ECG signal, the more likely it is noise. When kSQI<0, the signal quality is considered poor;

The pSQI calculation module is used to calculate the quality parameter pSQI based on the signal power data set:

According to the power spectrum distribution of the QRS wave, calculate the ratio of the energy of the QRS wave to the energy of the ECG signal;

in, It is the power of QRS wave in the 6Hz~30Hz band;/> It is the power of the ECG signal in the 0Hz~125Hz band; when pSQI is between 0.5~0.72, the signal quality is better;

The qSQI calculation module is used to calculate the quality parameter qSQI of matching data sets based on different R peak detection algorithms:

Compare the matching degree of R waves detected by two QRS wave detection algorithms; among them, the two QSR wave detection algorithms selected are the algorithm DF based on digital filtering and the algorithm LT based on length transformation;

Represents the number of QRS waves detected in the same segment of the ECG signal by the DF algorithm and the LT algorithm respectively. Represents the number of QRS waves detected by both algorithms at the same time;

The optimal quality lead signal selection module is used to first fuse the results of the above multiple SQIs and calculate the comprehensive quality score of a certain lead signal. The comprehensive quality score is calculated as follows:

Then calculate the comprehensive quality score of each lead signal in the same time period;

Finally, the lead signal with the highest score is selected as the best quality ECG signal.

A third aspect of the present invention provides a heartbeat classification method, including:

Intercept heart beat information from the best quality ECG signal obtained through the optimal lead signal selection method of long-term wearable ECG signals;

Calculate the QRS complex and R-R interval of the intercepted heart beat information;

The calculated QRS complex and R-R interval are used as empirical features to fuse with the preprocessed ECG waveform and then input into the deep learning network model to classify the heart beats.

A fourth aspect of the present invention provides a scatter plot drawing method, which performs R peak identification on the long-term ECG signal with the best quality obtained through the optimal lead signal selection method of the long-term wearable ECG signal. Then draw a scatter plot; that is:

First, a bandpass filter is used to filter the ECG signal;

Then, perform nonlinear transformation on the denoised ECG signal;

Then, use the slope, amplitude and width judgment rules to judge the feature points and detect the QRS wave group;

Finally, calculate the R-R interval (R1, R2,...,Rn) based on the R peak position, and draw the scatter point set of {(R1, R2), (R2, R3),..., (Rn-1, Rn)} Figure, where R1 represents the time interval between the 1st and 2nd R waves.

A fifth aspect of the present invention provides a heart rhythm type detection method, including the following steps:

(1) Through the search window algorithm, extract the attractor contour line of the scatter plot B drawn by the scatter plot drawing method:

(1.1) The B picture search window is a rectangular window with a fixed width on the abscissa of the plane rectangular coordinate system. The width of the B picture search window is set to:

Among them, d is the width of the B picture search window, It is the maximum value of the abscissa from point (0,0) to (N,N),/> It is the minimum value of the abscissa from point (0,0) to (N,N), and 10% is the set threshold standard;

(1.2) The B picture search window goes through each square in sequence from (0,0) to (N,N) points;

(1.3) Within the range covered by a B-image search window, record the points M belonging to the scatter atlas, and define a threshold window as a rectangular window with a fixed height in the point set M. The height is set as:

Among them, h is the height of the threshold window in the point set M, is the maximum value of the ordinate of all points in the point set M,/> is the minimum value of the ordinate of all points in the point set M, and 10% is the set threshold standard;

(1.4) Store the point set P within the threshold window, continue to move the B picture search window to the right by the width of one pane, repeat steps (1.2) and (1.3) until the end, and finally obtain the B picture boundary point set ;

(1.5) Combine the obtained boundary point set of B picture Perform a fitting to get the slope of graph B/> ;

(2) Through the search window algorithm, extract the attractor contour line of the scatter plot C drawn by the scatter plot drawing method:

(2.1) The C-picture search window is a rectangular window with a fixed width on the ordinate of the plane rectangular coordinate system. The width of the C-picture search window is set to:

Among them, d is the width of the C picture search window, It is the maximum value of the abscissa from point (0,0) to (N,N),/> It is the minimum value of the abscissa from point (0,0) to (N,N), and 10% is the set threshold standard;

(2.2) The C graph search window goes through each square in sequence from (0,0) to (N,N) points;

(2.3) Within the range covered by a C-graph search window, record the points M belonging to the scatter atlas, and define a threshold window as a rectangular window with a fixed height in the point set M. This height is set as:

Among them, h is the height of the threshold window in the point set M, is the maximum value of the abscissa of all points in the point set M,/> is the minimum value of the abscissa of all points in the point set M, and 10% is the set threshold standard;

(2.4) Store the point set P within the threshold window, continue to move the C-graph search window upward by the width of one pane, repeat steps (2.2) and (2.3) until the end, and finally obtain the C-graph boundary point set ;

(2.5) Combine the already obtained set of boundary points of the C graph Perform a fitting and obtain the slope of the C graph/> ;

(3) Detect the heart rhythm type based on the distribution of the scatter plot and the slope of the attractor contour

(3.1) Determine whether the graph is a distribution graph based on the conditions of graphs B and C. If the slope of graphs B and C is not detected, it is a distribution graph;

For example, pictures B and C are symmetrical about the 45° axis, which is a three-distribution figure or a sector figure;

If both picture B and picture C are detected, it is a three-distribution pattern or a four-distribution pattern;

(3.2) Determine the heart rhythm type based on the graph distribution and the slope of line B:

(3.2.1) The scatter plot is a distribution graph, which is detected as sinus beat;

(3.2.2) The scatter plot is a three-distribution graph, with the slope of line B Between 0.18~0.80;

Or the scatter plot is a four-distribution graph, with the slope of line B Between 0.132 and 050, it is detected as premature supraventricular contraction;

(3.2.3) The scatter plot is a four-distribution graph, with the slope of line B <0.132;

Or the scatter plot is a three-distribution graph, with the slope of line B Between 0~0.08;

Or the scatter plot is a two-distribution graph, with the slope of line B If it is 0, it is detected as premature ventricular contraction;

(3.2.4) The scatter plot is a fan-shaped graph, and the slope of line B is >0.11, atrial fibrillation is detected;

(3.2.5) The scatter plot is a grid-like ordered multi-distribution graph, which detects atrial flutter with changes in conduction proportion;

(3.2.6) The scatter plot is a linear graph parallel to the X-axis below the bottom edge of the fan-shaped graph, or a linear graph with a slope of approximately zero overlaps in the fan-shaped area, which is detected as atrial fibrillation with premature ventricular contractions;

(3.2.7) The scatter plot is a linear graph that completely overlaps with the bottom edge of the fan shape, or is completely parallel to it above, and is detected as atrial fibrillation with intraventricular differential conduction.

A sixth aspect of the present invention provides an optimal lead signal selection device, including:

memory; and

A processor coupled to the memory, the processor being configured to execute the optimal lead signal selection method for long-term wearable ECG signals based on instructions stored in the memory.

A seventh aspect of the present invention provides a non-transitory computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the optimal lead signal selection of the long-term wearable ECG signal is realized. method.

An eighth aspect of the present invention provides a wearable dynamic ECG monitor, which uses the optimal lead signal selection method for long-term wearable ECG signals to select the best quality lead signal; For the lead signals, the heart beat classification method is used for heart beat classification; or for the selected lead signals, the heart rhythm type detection method is used for heart rhythm type detection;

or,

Configure the optimal lead signal selection system for long-term wearable ECG signals to select the lead signals with the best quality; classify the selected lead signals using the heart beat classification method ; Or for the selected lead signal, use the described heart rhythm type detection method to detect the heart rhythm type;

or,

Set up the optimal lead signal selection device and use the device to select the lead signal with the best quality; use the heart beat classification method to classify the selected lead signals; or classify the selected leads Connect the signal, and use the described heart rhythm type detection method to detect the heart rhythm type;

or,

Preset the non-transitory computer-readable storage medium, execute the program in the non-transitory computer-readable storage medium to obtain the best quality lead signal; use the described heart beat classification method for the obtained lead signal Classify heart beats; or use the heart rhythm type detection method to detect heart rhythm types on the obtained lead signals.

Compared with the existing technology, the present invention has outstanding substantive features and significant progress. Specifically:

The present invention analyzes long-term wearable ECG signals through a signal quality assessment algorithm, and can select better quality and more stable signals from signals from different leads within the same time period, thereby improving wearable dynamic ECG signals. Availability of ECG signals from electrical detectors.

Description of the drawings

Figure 1 is a flow chart of the method of Embodiment 1.

Figure 2 is an example of evaluation of multi-lead ECG signal quality in Embodiment 1.

Figure 3 shows the ECG signals of four typical diseases of MIT-BIH.

Figure 4 shows the difference results of various cardiac QRS wave complexes.

Figure 5 shows the difference results of R-R intervals of various heart beats.

Figure 6 is the model structure diagram of ResNet18.

Figure 7 is the model structure diagram of ResNet34.

Figure 8 shows three characteristic inputs of ECG signals.

Figure 9 is a schematic diagram of the implementation of the search window algorithm in Embodiment 5.

Figure 10 is a common arrhythmia scatter plot diagnostic model.

Figure 11 is the analysis flow of the central rhythm type detection method in Embodiment 5.

Detailed ways

The technical solution of the present invention will be further described in detail below through specific embodiments.

Example 1

As shown in Figure 1, this embodiment provides an optimal lead signal selection method for long-term wearable ECG signals, including:

Step 1: Read the dynamic ECG signal and establish three quality levels of dynamic ECG signal data sets, namely, a data set based on signal kurtosis, a data set based on signal power, and a matching data set based on different R peak detection algorithms;

Step 2: Calculate the quality parameters of dynamic ECG signal data sets of three quality levels respectively;

(1) Calculate the quality parameter kSQI based on the signal kurtosis data set:

in, is the kurtosis of the ECG signal;

kSQI is used to reflect the sharpness of the peak. The smaller the kurtosis of the ECG signal, the more likely it is noise. When kSQI<0, the signal quality is considered poor;

(2) Calculate the quality parameter pSQI based on the signal power data set:

According to the power spectrum distribution of the QRS wave, calculate the ratio of the energy of the QRS wave to the energy of the ECG signal;

in, It is the power of QRS wave in the 6Hz~30Hz band;/> It is the power of the ECG signal in the 0Hz~125Hz band; when pSQI is between 0.5~0.72, the signal quality is better;

(3) Calculate the quality parameter qSQI of matching data sets based on different R peak detection algorithms:

Compare the matching degree of R waves detected by two QRS wave detection algorithms; among them, the two QSR wave detection algorithms selected are the algorithm DF based on digital filtering and the algorithm LT based on length transformation;

Represents the number of QRS waves detected in the same segment of the ECG signal by the DF algorithm and the LT algorithm respectively. Represents the number of QRS waves detected by both algorithms at the same time;

Step 3: Fusion of the above multiple SQI results to calculate the comprehensive quality score of a certain lead signal:

Step 4: Calculate the comprehensive quality score of each lead signal in the same time period;

The lead signal with the highest score is selected as the best quality ECG signal.

Result verification

Figure 2 shows the results obtained based on SQI using the method of this embodiment. Among them, the frequency of the ECG signal used is 250Hz, and the 8-lead system is selected as the lead, which are I, II, V1, V2, V3, V4, V5, and V6. The eight sub-figures respectively represent the ECG waveforms of eight lead signals within the same 10 seconds. The lead of V3 is the lead signal with the best quality selected by the algorithm of this embodiment.

Example 2

This embodiment provides an optimal lead signal selection system for long-term wearable ECG signals, including:

The data reading module is used to read dynamic ECG signals and establish dynamic ECG signal data sets of three quality levels;

The quality parameter calculation module of dynamic ECG signal data sets is used to calculate the quality parameters of dynamic ECG signal data sets of three quality levels, including kSQI calculation module, pSQI calculation module and qSQI calculation module;

The best quality lead signal selection module is used to select the best quality leads.

For the specific implementation method of the system in this embodiment, please refer to the method described in Embodiment 1, which will not be described again here.

Example 3

This embodiment provides a heartbeat classification method, which includes: intercepting heartbeat information from the best-quality ECG signal obtained through Embodiment 1 or 2, and calculating the QRS wave complex and R-R interval as empirical features; and then using the experience The features are fused with the preprocessed ECG signal and input into ResNet to classify heartbeats.

(1) QRS waveform detection based on digital filtering

Methods based on digital filtering generally consist of three parts: linear filtering, nonlinear transformation and decision rules. First, a band-pass filter is used to filter and denoise the selected ECG signal; then, the filtered and denoised ECG signal is subjected to nonlinear transformation; finally, the slope, amplitude and width judgment rules are used to judge the feature points. , detect the QRS wave complex and intercept it as heartbeat information. This method has simple operation and fast speed, and is more suitable for processing large batches of long-term dynamic ECG data to provide early warning of diseases as soon as possible.

(2) Acquisition of prior knowledge

The four different heartbeat types of the MIT-BIH arrhythmia database are shown in Figure 3. Figure 3(a) is a normal heartbeat, represented by N; Figure 3(b) is a supraventricular premature beat, represented by S; Figure 3( c) is a premature ventricular beat, represented by V; Figure 3(d) is other cardiac beats, represented by O. After experiments, it was found that there are obvious differences in the QRS complexes and R-R intervals of various types of heart beats. The distribution of QRS complexes and R-R intervals of each type of heart beats is shown in Figures 4 and 5. In the knowledge of clinical artificial electrocardiogram diagnosis, the QRS wave complex characteristics and the R-R interval characteristics of the ECG signal are important criteria for classifying the ECG signal and heart beats. Therefore, these two time-limited empirical features are combined with the morphological characteristics of the heartbeat waveform to jointly serve as residual The input of the difference network increases the interpretability of the model and improves the residual network's ability to identify abnormal heart beats.

(3) Heart beat classification algorithm based on deep learning

Learn the morphological features of ECG signals through ResNet18 and ResNet34. Figures 6 and 7 show the network structures of ResNet18 and ResNet34 respectively.

In order to enable feature fusion, the QRS complex features and R-R interval features are converted into waveforms that are as long as the heart beat. Since the QRS complex is smaller than the length of the heartbeat waveform, in order to highlight the QRS complex, a rectangular pulse with a duration of 128 sampling points is used to represent the timing characteristics of the QRS complex. The R-R interval generally exceeds the duration of a single heart beat, so the R-R interval is converted into a feature vector with a duration of 128 sampling points and an amplitude value of the R-R interval duration. The morphological characteristics, QRS time limit characteristics and R-R interval characteristics of the cardiac beat waveform are shown in Figure 8. The equal-length QRS complex features, R-R interval features, and preprocessed ECG waveforms are divided into three feature dimensions and simultaneously input into the residual network model for feature fusion.

After feature fusion, QRS complex features, R-R interval features, and preprocessed ECG waveforms are input to the residual network model composed of ResNet18 and ResNet34 for heart beat classification training. By learning various types of ECG data with existing labels, Features, improved model; among them, the size of the input data of the residual network model is 2×128×3, where 2 refers to the dual-channel data, 128 refers to the heartbeat duration, and 3 refers to the dimension of the heartbeat feature.

The residual network model analyzed 22 pieces of test ECG signal data in the MIT-BIH arrhythmia database, including 44,218 normal heart beats, 1,836 supraventricular premature beats, 3,213 premature ventricular beats, and 1,849 other types of heart beats. Classification. The detection model based on the auxiliary data set (QT data set) provides the residual network with two additional prior features of QRS complex and R-R interval, which reduces the adverse impact of the limited sample size of the MIT-BIH arrhythmia database and improves improves the generalization ability of the network.

Example 4

This embodiment provides a method for drawing a scatter plot. The best quality long-term ECG signal obtained through Embodiment 1 or 2 is identified by R peak and then a scatter plot is drawn.

First, a bandpass filter is used to filter the ECG signal;

Then, perform nonlinear transformation on the denoised ECG signal;

Then, use the slope, amplitude and width judgment rules to judge the feature points and detect the QRS wave group;

Finally, calculate the R-R interval (R1, R2,...,Rn) based on the R peak position, and draw the scatter point set of {(R1, R2), (R2, R3),..., (Rn-1, Rn)} Figure, where R1 represents the time interval between the 1st and 2nd R waves.

Example 5

This embodiment provides a heart rhythm type detection method, which includes the following steps:

(1) As shown in Figure 9, through the search window algorithm, extract the attractor contour line of the scatter plot B drawn using the scatter plot drawing method described in Embodiment 4:

(1.1) The B picture search window is a rectangular window with a fixed width on the abscissa of the plane rectangular coordinate system. The width of the B picture search window is set to:

Among them, d is the width of the B picture search window, It is the maximum value of the abscissa from point (0,0) to (N,N),/> It is the minimum value of the abscissa from point (0,0) to (N,N), and 10% is the set threshold standard;

(1.2) The B picture search window goes through each square in sequence from (0,0) to (N,N) points;

(1.3) Within the range covered by a B-image search window, record the points M belonging to the scatter atlas, and define a threshold window as a rectangular window with a fixed height in the point set M. The height is set as:

Among them, h is the height of the threshold window in the point set M, is the maximum value of the ordinate of all points in the point set M,/> is the minimum value of the ordinate of all points in the point set M, and 10% is the set threshold standard;

(1.4) Store the point set P within the threshold window, continue to move the B picture search window to the right by the width of one pane, repeat steps (1.2) and (1.3) until the end, and finally obtain the B picture boundary point set ;

(1.5) Combine the obtained boundary point set of B picture Perform a fitting to get the slope of graph B/> ;

(2) Through the search window algorithm, extract the attractor contour line of the scatter plot C drawn using the scatter plot drawing method described in Example 4:

(2.1) The C-picture search window is a rectangular window with a fixed width on the ordinate of the plane rectangular coordinate system. The width of the C-picture search window is set to:

Among them, d is the width of the C picture search window, It is the maximum value of the abscissa from point (0,0) to (N,N),/> It is the minimum value of the abscissa from point (0,0) to (N,N), and 10% is the set threshold standard;

(2.2) The C graph search window goes through each square in sequence from (0,0) to (N,N) points;

(2.3) Within the range covered by a C-graph search window, record the points M belonging to the scatter atlas, and define a threshold window as a rectangular window with a fixed height in the point set M. This height is set as:

Among them, h is the height of the threshold window in the point set M, is the maximum value of the abscissa of all points in the point set M,/> is the minimum value of the abscissa of all points in the point set M, and 10% is the set threshold standard;

(2.4) Store the point set P within the threshold window, continue to move the C-graph search window upward by the width of one pane, repeat steps (2.2) and (2.3) until the end, and finally obtain the C-graph boundary point set ;

(2.5) Combine the already obtained set of boundary points of the C graph Perform a fitting and obtain the slope of the C graph/> ;

(3) Detect the heart rhythm type based on the distribution of the scatter plot and the slope of the attractor contour

As shown in Figure 10, the ECG scatter plot attractor presents different shapes corresponding to different arrhythmia categories. As shown in Figure 10-1, the one-distribution rod pattern is sinus rhythm, as shown in Figure 10-2, the three-distribution rod pattern is mostly judged as supraventricular premature beats, and the one shown in Figure 10-3 is four-distribution supraventricular premature beats. Premature beats, shown in Figure 10-4 are four-distribution premature ventricular beats. The three-distribution shape in Figure 10-5 is determined to be three-distribution ventricular premature beats bigemy. The two-distribution pattern shown in Figure 10-6 is Sustained ventricular premature bigeminy, and most of the two distributions are ventricular premature. The fan-shaped distribution shown in Figure 10-7 is atrial fibrillation, and the grid-shaped distribution shown in Figure 10-8 is atrial flutter, as shown in Figure 10- Figure 9 shows atrial fibrillation with premature ventricular contractions, Figure 10-11 shows atrial fibrillation combined with atrial flutter, and Figure 10-12 shows a special three distribution. Different diseases in multi-distribution may have the same type of scatter cluster set. In the Lorenz scatter diagram model, further judgment needs to be made with the slope of the B line, that is, a more accurate judgment of the arrhythmia type is made through chaos theory.

As shown in Figure 11, the analysis algorithm flow based on attractor distribution and slope:

(3.1) Determine whether the graph is a distribution graph based on the conditions of graphs B and C. If the slope of graphs B and C is not detected, it is a distribution graph;

For example, pictures B and C are symmetrical about the 45° axis, which is a three-distribution figure or a sector figure;

If both picture B and picture C are detected, it is a three-distribution pattern or a four-distribution pattern.

(3.2) Determine the heart rhythm type based on the graph distribution and the slope of line B:

(3.2.1) The scatter plot is a distribution graph, which is detected as sinus beat;

(3.2.2) The scatter plot is a three-distribution graph, with the slope of line B Between 0.18~0.80;

Or the scatter plot is a four-distribution graph, with the slope of line B Between 0.132 and 050, it is detected as premature supraventricular contraction;

(3.2.3) The scatter plot is a four-distribution graph, with the slope of line B <0.132;

Or the scatter plot is a three-distribution graph, with the slope of line B Between 0~0.08;

Or the scatter plot is a two-distribution graph, with the slope of line B If it is 0, it is detected as premature ventricular contraction;

(3.2.4) The scatter plot is a fan-shaped graph, and the slope of line B is >0.11, atrial fibrillation is detected;

(3.2.5) The scatter plot is a grid-like ordered multi-distribution graph, which detects atrial flutter with changes in conduction proportion;

(3.2.6) The scatter plot is a linear graph parallel to the X-axis below the bottom edge of the fan-shaped graph, or a linear graph with a slope of approximately zero overlaps in the fan-shaped area, which is detected as atrial fibrillation with premature ventricular contractions;

(3.2.7) The scatter plot is a linear graph that completely overlaps with the bottom edge of the fan shape, or is completely parallel to it above, and is detected as atrial fibrillation with intraventricular differential conduction.

Example 6

This embodiment provides an optimal lead signal selection device, including:

memory; and

A processor coupled to the memory, the processor being configured to execute the optimal lead signal selection method for long-term wearable ECG signals described in Embodiment 1 based on instructions stored in the memory. .

The memory may include, for example, system memory, fixed non-volatile storage media, etc. The system memory stores, for example, an operating system, application programs, a boot loader, and other programs.

The optimal lead signal selection device may also include input and output interfaces, network interfaces, storage interfaces, etc. These interfaces as well as the memory and the processor can be connected via a bus, for example. Among them, the input and output interface provides connection interfaces for input and output devices such as monitors, mice, keyboards, and touch screens. Network interfaces provide connection interfaces for various networked devices. The storage interface provides connection interfaces for external storage devices such as SD cards and USB disks.

Example 7

This embodiment provides a non-transitory computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the optimal lead signal of the long-term wearable ECG signal described in Embodiment 1 is achieved. Select method.

Example 8

This embodiment provides a wearable dynamic ECG monitor:

The optimal lead signal selection method for long-term wearable ECG signals described in Example 1 is used to select the lead signal with the best quality; for the selected lead signals, the heart beat classification described in Example 3 is used The method performs cardiac beat classification; or, for the selected lead signals, the heart rhythm type detection method described in Embodiment 5 is used to detect the heart rhythm type.

or,

Configure the optimal lead signal selection system for long-term wearable ECG signals described in Embodiment 2, and select the lead signal with the best quality; for the selected lead signal, use the method described in Embodiment 3 The heart beat classification method is used to classify the heart beats; or the selected lead signals are used to detect the heart rhythm type using the heart rhythm type detection method described in Embodiment 5.

or,

Set up the optimal lead signal selection device described in Embodiment 6, and use this device to select the lead signal with the best quality; classify the selected lead signals using the heart beat classification method described in Embodiment 3; Or, for the selected lead signals, the heart rhythm type detection method described in Embodiment 5 is used to detect the heart rhythm type.

or,

Preset the non-transitory computer-readable storage medium described in Embodiment 7, execute the program in the non-transitory computer-readable storage medium to obtain the lead signal with the best quality; for the obtained lead signal, use the method in Embodiment 3 Classify heart beats using the heart beat classification method described above; or use the heart rhythm type detection method described in Embodiment 5 to detect heart rhythm types on the obtained lead signals.

It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the invention may take the form of a computer program product embodied on one or more computer non-transitory readable storage media (including, but not limited to, magnetic disk storage, CD-ROM, optical storage, etc.) embodying computer program code therein. .

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each process and/or block in the flowchart illustrations and/or block diagrams, and combinations of processes and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a use A device for realizing the functions specified in a process or processes in a flowchart and/or a block or blocks in a block diagram.

These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions The device implements the functions specified in a process or processes in the flowchart and/or in a block or blocks in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operating steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby executing on the computer or other programmable device. Instructions provide steps for implementing the functions specified in a process or processes of a flowchart diagram and/or a block or blocks of a block diagram.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention but not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the present invention can still be modified. Modifications to the specific embodiments of the invention or equivalent substitutions of some of the technical features without departing from the spirit of the technical solution of the present invention shall be covered by the scope of the technical solution claimed by the present invention.

Claims (10)

1. An optimal lead signal selection method for long-term wearable ECG signals, which is characterized by including: Step 1: Read the dynamic ECG signal and establish three quality levels of dynamic ECG signal data sets, namely, a data set based on signal kurtosis, a data set based on signal power, and a matching data set based on different R peak detection algorithms; Step 2: Calculate the quality parameters of dynamic ECG signal data sets of three quality levels respectively; (1) Calculate the quality parameter kSQI based on the signal kurtosis data set: in, is the kurtosis of the ECG signal; kSQI is used to reflect the sharpness of the peak. The smaller the kurtosis of the ECG signal, the more likely it is noise. When kSQI<0, the signal quality is considered poor; (2) Calculate the quality parameter pSQI based on the signal power data set: According to the power spectrum distribution of the QRS wave, calculate the ratio of the energy of the QRS wave to the energy of the ECG signal; in, It is the power of QRS wave in the 6Hz~30Hz band;/> It is the power of the ECG signal in the 0Hz~125Hz band; when pSQI is between 0.5~0.72, the signal quality is better; (3) Calculate the quality parameter qSQI of matching data sets based on different R peak detection algorithms: Compare the matching degree of R waves detected by two QRS wave detection algorithms; among them, the two QSR wave detection algorithms selected are the algorithm DF based on digital filtering and the algorithm LT based on length transformation; Represents the number of QRS waves detected in the same segment of the ECG signal by the DF algorithm and the LT algorithm respectively,/> Represents the number of QRS waves detected by both algorithms at the same time; Step 3: Fusion of the above multiple SQI results to calculate the comprehensive quality score of a certain lead signal: Step 4: Calculate the comprehensive quality score of each lead signal in the same time period; The lead signal with the highest score is selected as the best quality ECG signal. 2. An optimal lead signal selection system for long-term wearable ECG signals, which is characterized by including: The data reading module is used to read dynamic ECG signals and establish three quality levels of dynamic ECG signal data sets, namely, based on signal kurtosis data sets, based on signal power data sets, and based on different R peak detection algorithm matching data sets. ; The quality parameter calculation module of dynamic ECG signal data sets is used to calculate the quality parameters of dynamic ECG signal data sets of three quality levels, including kSQI calculation module, pSQI calculation module and qSQI calculation module; The kSQI calculation module is used to calculate the quality parameter kSQI based on the signal kurtosis data set: in, is the kurtosis of the ECG signal; kSQI is used to reflect the sharpness of the peak. The smaller the kurtosis of the ECG signal, the more likely it is noise. When kSQI<0, the signal quality is considered poor; The pSQI calculation module is used to calculate the quality parameter pSQI based on the signal power data set: According to the power spectrum distribution of the QRS wave, calculate the ratio of the energy of the QRS wave to the energy of the ECG signal; in, It is the power of QRS wave in the 6Hz~30Hz band;/> It is the power of the ECG signal in the 0Hz~125Hz band; when pSQI is between 0.5~0.72, the signal quality is better; The qSQI calculation module is used to calculate the quality parameter qSQI of matching data sets based on different R peak detection algorithms: Compare the matching degree of R waves detected by two QRS wave detection algorithms; among them, the two QSR wave detection algorithms selected are the algorithm DF based on digital filtering and the algorithm LT based on length transformation; Represents the number of QRS waves detected in the same segment of the ECG signal by the DF algorithm and the LT algorithm respectively,/> Represents the number of QRS waves detected by both algorithms at the same time; The optimal quality lead signal selection module is used to first fuse the results of the above multiple SQIs and calculate the comprehensive quality score of a certain lead signal. The comprehensive quality score is calculated as follows: Then calculate the comprehensive quality score of each lead signal in the same time period; Finally, the lead signal with the highest score is selected as the best quality ECG signal. 3. A heart beat classification method, characterized by including: Intercept the heart beat information from the best quality ECG signal obtained through the optimal lead signal selection method of long-term wearable ECG signal according to claim 1; Calculate the QRS complex and R-R interval of the intercepted heart beat information; The calculated QRS complex and R-R interval are used as empirical features to fuse with the preprocessed ECG waveform and then input into the deep learning network model to classify the heart beats. 4. The heartbeat classification method according to claim 3, wherein the method for intercepting heartbeat information is: First, a band-pass filter is used to filter and denoise the selected ECG signal; Then, perform nonlinear transformation on the filtered and denoised ECG signal; Finally, the slope, amplitude and width judgment rules are used to judge the feature points, detect the QRS wave complex and intercept it as heart beat information. 5. The heartbeat classification method according to claim 3, characterized in that the method for classifying heartbeats is: First, various types of features are fused; in order to perform feature fusion, the QRS wave complex features and R-R interval features are converted into waveforms with the same length as the heart beat; among them, a rectangular pulse with a duration of 128 sampling points is used to represent the QRS wave complex features. , convert the R-R interval features into a feature vector with a duration of 128 sampling points and an amplitude value equal to the R-R interval duration; divide the equal-length QRS complex features, R-R interval features, and preprocessed ECG waveforms into three Feature dimensions are simultaneously input into the residual network model for feature fusion; After feature fusion, QRS complex features, R-R interval features, and preprocessed ECG waveforms are input to the residual network model composed of ResNet18 and ResNet34 for heart beat classification training. By learning various types of ECG data with existing labels, Features, improved models; Among them, the size of the input data of the residual network model is 2×128×3, where 2 refers to the dual-channel data, 128 refers to the heartbeat duration, and 3 refers to the dimension of the heartbeat feature. 6. A scatter plot drawing method, characterized in that: the long-term ECG signal with the best quality obtained through the optimal lead signal selection method of the long-term wearable ECG signal according to claim 1 is carried out. After the R peak is identified, a scatter plot is drawn; that is: First, a bandpass filter is used to filter the ECG signal; Then, perform nonlinear transformation on the denoised ECG signal; Then, use the slope, amplitude and width judgment rules to judge the feature points and detect the QRS wave group; Finally, calculate the R-R interval (R1, R2,...,Rn) based on the R peak position, and draw the scatter point set of {(R1, R2), (R2, R3),..., (Rn-1, Rn)} Figure, where R1 represents the time interval between the 1st and 2nd R waves. 7. A heart rhythm type detection method, characterized by comprising the following steps: (1) Through the search window algorithm, extract the attractor contour line of the scatter plot B drawn by the scatter plot drawing method described in claim 6: (1.1) The B picture search window is a rectangular window with a fixed width on the abscissa of the plane rectangular coordinate system. The width of the B picture search window is set to: Among them, d is the width of the B picture search window, It is the maximum value of the abscissa from point (0,0) to (N,N),/> It is the minimum value of the abscissa from point (0,0) to (N,N), and 10% is the set threshold standard; (1.2) The B picture search window goes through each square in sequence from (0,0) to (N,N) points; (1.3) Within the range covered by a B-image search window, record the points M belonging to the scatter atlas, and define a threshold window as a rectangular window with a fixed height in the point set M. The height is set as: Among them, h is the height of the threshold window in the point set M, is the maximum value of the ordinate of all points in the point set M,/> is the minimum value of the ordinate of all points in the point set M, and 10% is the set threshold standard; (1.4) Store the point set P within the threshold window, continue to move the B picture search window to the right by the width of one pane, repeat steps (1.2) and (1.3) until the end, and finally obtain the B picture boundary point set ; (1.5) Combine the obtained boundary point set of B picture Perform a fitting to get the slope of graph B/> ; (2) Through the search window algorithm, extract the attractor contour line of the scatter plot C drawn by the scatter plot drawing method described in claim 6: (2.1) The C-picture search window is a rectangular window with a fixed width on the ordinate of the plane rectangular coordinate system. The width of the C-picture search window is set to: Among them, d is the width of the C picture search window, It is the maximum value of the abscissa from point (0,0) to (N,N),/> It is the minimum value of the abscissa from point (0,0) to (N,N), and 10% is the set threshold standard; (2.2) The C graph search window goes through each square in sequence from (0,0) to (N,N) points; (2.3) Within the range covered by a C-graph search window, record the points M belonging to the scatter atlas, and define a threshold window as a rectangular window with a fixed height in the point set M. This height is set as: Among them, h is the height of the threshold window in the point set M, is the maximum value of the abscissa of all points in the point set M,/> is the minimum value of the abscissa of all points in the point set M, and 10% is the set threshold standard; (2.4) Store the point set P within the threshold window, continue to move the C-graph search window upward by the width of one pane, repeat steps (2.2) and (2.3) until the end, and finally obtain the C-graph boundary point set ; (2.5) Combine the already obtained set of boundary points of the C graph Perform a fitting and obtain the slope of the C graph/> ; (3) Detect the heart rhythm type based on the distribution of the scatter plot and the slope of the attractor contour (3.1) Determine whether the graph is a distribution graph based on the conditions of graphs B and C. If the slope of graphs B and C is not detected, it is a distribution graph; For example, pictures B and C are symmetrical about the 45° axis, which is a three-distribution figure or a sector figure; If both picture B and picture C are detected, it is a three-distribution pattern or a four-distribution pattern; (3.2) Determine the heart rhythm type based on the graph distribution and the slope of line B: (3.2.1) The scatter plot is a distribution graph, which is detected as sinus beat; (3.2.2) The scatter plot is a three-distribution graph, with the slope of line B Between 0.18~0.80; Or the scatter plot is a four-distribution graph, with the slope of line B Between 0.132 and 050, it is detected as premature supraventricular contraction; (3.2.3) The scatter plot is a four-distribution graph, with the slope of line B <0.132; Or the scatter plot is a three-distribution graph, with the slope of line B Between 0~0.08; Or the scatter plot is a two-distribution graph, with the slope of line B If it is 0, it is detected as premature ventricular contraction; (3.2.4) The scatter plot is a fan-shaped graph, and the slope of line B is >0.11, atrial fibrillation is detected; (3.2.5) The scatter plot is a grid-like ordered multi-distribution graph, which detects atrial flutter with changes in conduction proportion; (3.2.6) The scatter plot is a linear graph parallel to the X-axis below the bottom edge of the fan-shaped graph, or a linear graph with a slope of approximately zero overlaps in the fan-shaped area, which is detected as atrial fibrillation with premature ventricular contractions; (3.2.7) The scatter plot is a linear graph that completely overlaps with the bottom edge of the fan shape, or is completely parallel to it above, and is detected as atrial fibrillation with intraventricular differential conduction. 8. An optimal lead signal selection device, including: memory; and A processor coupled to the memory, the processor being configured to execute the optimal lead signal selection method for long-term wearable ECG signals according to claim 1 based on instructions stored in the memory. . 9. A non-transitory computer-readable storage medium on which a computer program is stored, which when executed by a processor implements the optimal lead signal selection method for long-term wearable ECG signals according to claim 1 . 10. A wearable dynamic electrocardiogram monitor, characterized by: The optimal lead signal selection method for long-term wearable ECG signals described in claim 1 is used to select the lead signal with the best quality; for the selected lead signal, any one of claims 3-5 is used The described heart beat classification method performs heart beat classification; or the selected lead signal is used to perform heart rhythm type detection using the heart rhythm type detection method described in claim 7; or, Configure the optimal lead signal selection system for long-term wearable ECG signals according to claim 2, and select the lead signal with the best quality; for the selected lead signal, any of claims 3-5 is used The heart beat classification method described in one item is used to classify heart beats; or the selected lead signal is used to detect the heart rhythm type using the heart rhythm type detection method described in claim 7; or, An optimal lead signal selection device according to claim 8 is provided, and the device is used to select the lead signal with the best quality; for the selected lead signal, the heart beat classification described in any one of claims 3-5 is used The method performs heart beat classification; or uses the heart rhythm type detection method of claim 7 to detect the heart rhythm type of the selected lead signal; or, Preset the non-transitory computer-readable storage medium described in claim 9, execute the program in the non-transitory computer-readable storage medium to obtain the lead signal with the best quality; for the obtained lead signal, use claim 3- 5. Use the heart beat classification method described in any one of 5 to perform heart beat classification; or use the heart rhythm type detection method described in claim 7 to perform heart rhythm type detection on the obtained lead signals.