A general multi-lead QRS wave detection method for electrocardiosignal

Abstract

This application provides a general multi-lead QRS wave detection method for electrocardiogram (ECG) signals, comprising: acquiring the multi-lead ECG signal to be tested, normalizing it with Z-score, and inputting it into a general multi-lead QRS wave detection model for ECG signals. The model includes a lead distillation module and a QRS wave detection module. The lead distillation module converts the channel dimension axis of the lead input into the width dimension, converting the multi-lead signal into a single-channel image. Then, multiple convolutional units are used to extract and aggregate features from the input multi-lead ECG signal to be tested to obtain the final distilled signal. The distilled signal is then input into the QRS wave detection module to obtain the position information of the QRS complex.

Description

Technical Field

[0001] This invention relates to the field of QRS wave detection technology, and in particular to a universal multi-lead QRS wave detection method for electrocardiogram signals. Background Technology

[0002] A standard electrocardiogram (ECG) waveform includes the P wave, QRS complex, and T wave. Among them, the QRS complex represents the depolarization process of the ventricle. Compared with other characteristic waveforms, it is tall and sharp, making it easy to detect. Therefore, it is often used to locate heartbeats, calculate parameters such as heart rate and RR interval, and analyze heart rate variability to assist in the diagnosis of underlying cardiac diseases such as myocardial, pericardial, and cardiac function, as well as diseases of other extracardiac systems. Thus, QRS complex detection is very important in the auxiliary diagnosis and treatment monitoring of cardiac diseases.

[0003] In recent years, various QRS waveform detection algorithms have been proposed, mainly divided into two categories: traditional methods and deep learning methods. Traditional methods typically include threshold detection, empirical mode decomposition, wavelet transform, and mathematical morphology. These methods are primarily based on fixed models and rules, extracting waveform-related features from empirical parameters obtained from cardiac physiology for detection. Their accuracy largely depends on the selection of hyperparameters and the accuracy of feature analysis. Machine learning and deep learning-based QRS detection algorithms typically construct models to autonomously learn the sampling point features of ECG signals, predicting the current ECG signal sampling features based on the features of past ECG signal sampling points. However, in implementing the technical solutions of this application, the inventors discovered at least the following technical problems with the above-mentioned technologies: most of these algorithms are designed based on single-lead ECG signals. While they can detect QRS waves relatively accurately on high-quality single-lead signals, their detection performance may significantly decrease once the lead on which these algorithms rely is subject to substantial noise interference. Therefore, for low-quality signals, multi-lead ECG signals can provide more usable information compared to single-lead ECG signals. The more leads a signal has, the higher the accuracy of QRS detection from low-quality ECG signals. Some researchers extract high-quality leads from multi-lead signals as input to QRS detectors; however, due to the uneven distribution of noise, this method may discard low-quality signals containing clear portions or select high-quality signals containing noisy portions. Other researchers predict the QRS wave location for each lead in a 12-lead ECG signal separately and then fuse these results to obtain the final QRS wave location. These methods can fuse multi-lead signals to obtain optimized QRS wave detection, but they typically require manually designing lead processing strategies. This inevitably involves complex procedures and manual threshold selection, resulting in poor compatibility with multi-lead signals. Summary of the Invention

[0004] To address the problems of low efficiency and poor compatibility of multi-lead ECG signal processing due to the complexity of existing multi-lead ECG signal processing procedures and the need for manual threshold selection, this invention proposes a universal multi-lead QRS wave detection method for ECG signals.

[0005] To address the aforementioned technical problems, this invention provides a universal multi-lead QRS wave detection method for electrocardiogram (ECG) signals, comprising: acquiring the multi-lead ECG signal to be tested, normalizing it using Z-score, and inputting it into a universal multi-lead QRS wave detection model for ECG signals. This model includes a lead distillation module and a QRS wave detection module. The lead distillation module converts the channel dimension axis of the lead input into the width dimension, transforming the multi-lead signal into a single-channel image. Then, multiple convolutional units are used to extract features from the input multi-lead ECG signal to be tested. The extracted feature signals are input into a classifier to select noise-free feature signals. The noise-free feature signals are summed along the width dimension to obtain the final distilled signal. The distilled signal is input into the QRS wave detection module for image segmentation to obtain the location information of the QRS complex.

[0006] Preferably, the training process of the general multi-lead QRS wave detection model for electrocardiogram signals is as follows:

[0007] Prepare training data, which consists of electrocardiogram signals and corresponding label data selected from different category datasets, forming a training set;

[0008] The ECG signals in the training set are Z-score normalized, and then the normalized data is input into the lead distillation module to obtain the distillation signal. The distillation signal is then input into the QRS wave detection module, and the Adam algorithm is used for optimization training. After training, the model with the minimum loss is saved. This model is the general multi-lead QRS wave detection model for ECG signals.

[0009] The ECG signal label for the training set is a square wave. In this square wave, the area of ​​the QRS complex is marked with a value of 1, the width is 37 sampling points, there are 18 sampling points before and after the R peak, and the other areas are kept at a value of 0.

[0010] The loss function of the Adam algorithm is Diceloss, and its formula is shown below:

[0011]

[0012] Where N represents the signal length, g(x) i ) represents x i The true value of the label, p(x) i ) represents x i The network prediction value.

[0013] Preferably, the convolutional unit of the lead distillation module includes a conventional convolutional unit and a depthwise separable convolutional unit. The conventional convolutional unit includes a convolutional layer, a batch normalization layer, and a Leaky ReLU activation layer, and the depthwise separable convolutional unit includes a depthwise separable convolutional layer, a batch normalization layer, and a Leaky ReLU activation layer.

[0014] Preferably, the QRS wave detection module is any one of FCNet, U-Net, and DenseNet networks.

[0015] Preferably, the QRS detection module is a multi-layer network framework based on the U-net network, with the convolution kernels of corresponding layers having the same size. The multi-layer network framework adopts a skip-layer connection structure, which concatenates the intermediate results of the encoder with the corresponding decoder feature maps in the channel dimension. Finally, the feature signals are processed through the Sigmoid activation function to obtain the location information of the QRS wavegroup.

[0016] The present invention also provides a universal multi-lead QRS wave detection device for electrocardiogram signals, comprising:

[0017] Acquisition module: used to acquire the multi-lead electrocardiogram signal to be tested, and input the normalized multi-lead electrocardiogram signal to be tested into the pre-constructed general multi-lead QRS wave detection model of electrocardiogram signal; the general multi-lead QRS wave detection model of electrocardiogram signal includes a lead distillation module and a QRS wave detection module;

[0018] Lead distillation module: The multi-lead ECG signal to be tested is input into the lead distillation module. The lead distillation module converts the channel dimension axis of the lead input into the width dimension, converts the multi-lead signal into a single-channel image, uses multiple convolutional units to extract features from the converted multi-lead ECG signal to be tested, inputs the extracted feature signals into a classifier to select noise-free feature signals, and sums the noise-free feature signals according to the width dimension to obtain the final distilled signal;

[0019] QRS wave detection module: The distillation signal is input into the QRS wave detection module to achieve image segmentation and obtain the location information of the QRS wave group.

[0020] This invention also provides a universal multi-lead QRS wave detection device for electrocardiogram signals, comprising:

[0021] A multi-lead electrocardiogram (ECG) signal acquisition device is used to acquire ECG signals to be tested.

[0022] The host computer is communicatively connected to the multi-lead ECG signal acquisition device and is used to execute computer programs to implement the steps of the general multi-lead QRS wave detection method for ECG signals as described above.

[0023] The display device is connected to the host computer and is used to display the QRS complex wave label diagram of the ECG signal to be tested.

[0024] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a general multi-lead QRS wave detection method for electrocardiogram signals as described in any of the preceding claims.

[0025] The technical solution of the present invention has the following advantages compared with the prior art:

[0026] This invention discloses a universal multi-lead QRS wave detection method for electrocardiogram (ECG) signals. It constructs an end-to-end network model based on a convolutional neural network and a lead distillation module. The multi-lead ECG signal to be tested is Z-score normalized and then input into the network model. The lead distillation module converts the channel dimension axis of the lead input into the width dimension, transforming the multi-lead signal into a single-channel image. This conversion exhibits good compatibility across various lead structures, including six-lead, four-lead, three-lead, two-lead, and even single-lead systems. Within the lead distillation module, multiple convolutional units extract features from the converted multi-lead ECG signal. The extracted features are input into a classifier to select noise-free feature signals. These noise-free feature signals are then summed along the width dimension to obtain the final distilled signal, effectively filtering out noise present in the multi-lead ECG signal under test. The present invention proposes a universal multi-lead QRS wave detection method for electrocardiogram (ECG) signals, which can accurately detect the QRS wave position in low-quality multi-lead ECG signals. After training with a neural convolutional network and using the Adam optimization algorithm, the method avoids manual selection of training parameters, ensuring the accuracy of training results and improving the compatibility of multi-lead signals. Attached Figure Description

[0027] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein...

[0028] Figure 1 A flowchart of the universal multi-lead QRS wave detection method for electrocardiogram signals provided by the present invention;

[0029] Figure 2 The electrocardiogram signal within a normal cardiac cycle;

[0030] Figure 3 This is a structural diagram of the multi-lead electrocardiogram QRS wave detector model provided by the present invention;

[0031] Figure 4 This is a structural diagram of the leaded distillation module provided by the present invention;

[0032] Figure 5 This is a structural diagram of the QRS wave detection module provided by the present invention;

[0033] Figure 6 This is a flowchart illustrating the training process of the end-to-end network model based on a convolutional neural network and a lead distillation module provided in this invention.

[0034] Figure 7 This is a schematic diagram of the input multi-lead electrocardiogram signal and QRS wave label provided by the present invention. Detailed Implementation

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0036] This invention provides a universal multi-lead QRS wave detection method for electrocardiogram signals. It designs a multi-lead QRS wave detection model for low-quality signals based on a lead distillation module and a convolutional neural network. The lead distillation module can automatically extract effective sampling points from each lead, without relying on certain leads. It has higher accuracy and better robustness for low-quality signals affected by noise interference, and has good multi-lead signal compatibility.

[0037] Reference Figure 1 As shown in the flowchart of the general multi-lead QRS wave detection method for electrocardiogram signals provided in this embodiment of the invention, the method includes the following steps:

[0038] Step S101: Acquire the multi-lead ECG signal to be tested;

[0039] The waveform of the electrocardiogram signal within a normal cardiac cycle is as follows: Figure 2 As shown in the figure, a standard electrocardiogram waveform includes the P wave, QRS complex, and T wave. Each wave represents a different process in the heartbeat cycle. Among them, the QRS complex represents the depolarization process of the ventricle. Compared with other characteristic waveforms, it is tall and sharp, and easy to detect.

[0040] The multi-lead ECG signal to be tested can be any combination of leads, including six-lead, four-lead, three-lead, two-lead, or even single-lead.

[0041] Step S102: After normalizing the Z-score of the multi-lead ECG signal to be tested, input it into the general multi-lead QRS wave detection model for ECG signals, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the multi-lead electrocardiogram signal QRS wave detector model provided by the present invention. The general multi-lead QRS wave detection model for electrocardiogram signals includes a lead distillation module and a QRS wave detection module.

[0042] Step S103: Input the normalized multi-lead ECG signal to be tested into the lead distillation module to obtain the distillation signal that fuses the multi-lead feature signals;

[0043] Reference Figure 4 As shown, Figure 4 The diagram shows the structure of the lead distillation module provided by this invention. The lead distillation module provided in this embodiment converts the channel dimension axis of the lead input into the width dimension, converts the multi-lead signal into a single-channel image, and uses multiple convolutional units to extract and aggregate features from the converted multi-lead ECG signal to be tested, thereby obtaining the final distilled signal.

[0044] The feature extraction utilizes multiple convolutional units, including conventional convolutional units and depthwise separable convolutional units. Conventional convolutional units include convolutional layers, batch normalization layers, and Leaky ReLU activation layers, while depthwise separable convolutional units include depthwise separable convolutional layers, batch normalization layers, and Leaky ReLU activation layers. Depthwise separable convolutional units have fewer parameters, which can avoid extracting redundant features from multi-lead ECG signals.

[0045] The feature aggregation process involves selecting noise-free feature signals from the feature signals extracted by the convolutional units using the softmax function, and then summing the noise-free feature signals along the width dimension to obtain the final distillation signal.

[0046] Step S104: Input the distillation signal into the QRS wave detection module to achieve image segmentation and obtain the location information of the QRS wave group;

[0047] In this embodiment of the invention, the QRS wave detection module is any one of FCNet, U-Net, and DenseNet networks;

[0048] Reference Figure 5 As shown, Figure 5 The diagram shows the structure of the QRS wave detection module provided by this invention. The QRS wave detection module provided in this embodiment is a multi-layer network framework built on the U-net network, with convolutional kernels of the same size at each level. Correspondingly, the downsampling operation at each level reduces the mapping length of the input feature signal by half, while the upsampling operation at the corresponding level doubles the mapping length of the input feature signal. For more accurate localization, this invention uses a skip-layer connection structure, concatenating the intermediate results of the encoder with the corresponding decoder feature map in the channel dimension, combining shallow positional information and deep semantic information. Finally, the feature signal is processed through the Sigmoid activation function to obtain the QRS wave group's positional information.

[0049] The Sigmoid activation function is a commonly used non-linear activation function that maps all real numbers to the (0, 1) interval. It uses a non-linear method to normalize the data. By comparing the correspondence between categories and labels obtained from the training model, as shown in Table 1, the QRS wave detection results of the multi-lead ECG signal to be detected are obtained.

[0050] Table 1. Correspondence between categories and labels

[0051] category QRS wave Non-QRS wave Label 1 0

[0052] Reference Figure 6 As shown in the figure, the end-to-end network model training process based on a convolutional neural network and a lead distillation module provided in this embodiment of the invention is as follows:

[0053] Step S601: Prepare training data (N*L*2500), which is selected from ECG signals and corresponding label data (N*L*2500) of different category datasets to form a training set, where N is the number of training data samples and L is the number of leads;

[0054] Among them, the label corresponding to the electrocardiogram signal is a square wave, such as Figure 7 As shown, in this square wave, the area of ​​the QRS group is marked with a value of 1, the width is 37 sampling points, there are 18 sampling points before and after the R peak, and the other areas remain at a value of 0.

[0055] Step S602: Perform Z-score normalization on the input signal, and then input the normalized data into the network for training. The optimizer is the Adam algorithm.

[0056] The parameters of the Adam algorithm are set as follows: learning rate lr = 0.001, beta_1 = 0.9, beta_2 = 0.999, epsilon = 1e-08, and the loss function is Diceloss, the formula of which is shown below:

[0057]

[0058] Where N represents the signal length, g(x) i ) represents x i The true value of the label, p(x) i ) represents x i The network prediction value.

[0059] First, the normalized data is input into the lead distillation module. The parameters of the lead distillation module are set to automatically extract features of multi-lead ECG signals. The parameters of the lead distillation module are shown in Table 2.

[0060] Table 2 Parameters of the Lead Distillation Module

[0061]

[0062] The distilled signal is input into the QRS wave detection module for QRS group location detection. The network layer parameters of the QRS wave detection model include depth, filter size, number of levels, and kernel size of each level. The specific parameter settings are shown in Table 3.

[0063] Table 3. Network layer parameters of the QRS wave detection model

[0064]

[0065] Step 603: After training, save the model with the minimum loss to obtain the pre-constructed general multi-lead QRS wave detection model for ECG signals, which is used to detect the location of QPR wave groups;

[0066] In this embodiment, an end-to-end network model based on a convolutional neural network and a lead distillation module was trained, including a lead distillation module and a QRS wave detection module. The lead distillation module is used to extract useful feature signals from low-quality ECG signals in multiple leads, and the QRS wave detection module determines the location information of the QRS complex from the useful feature signals.

[0067] A specific embodiment of the present invention also provides a universal multi-lead QRS wave detection device for electrocardiogram signals, comprising:

[0068] A multi-lead electrocardiogram signal acquisition device is used to acquire multi-lead electrocardiogram signals to be tested;

[0069] The host computer is communicatively connected to the multi-lead ECG signal acquisition device and is used to execute computer programs to implement the steps of the general multi-lead QRS wave detection method for ECG signals as described above.

[0070] The display device is connected to the host computer and is used to display the QRS complex wave tag map of the multi-lead electrocardiogram signal to be tested.

[0071] A specific embodiment of the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described general multi-lead QRS wave detection method for electrocardiogram signals.

[0072] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A general-purpose multi-lead QRS wave detection method for electrocardiosignal, characterized in that, include: The multi-lead electrocardiogram (ECG) signal to be tested is acquired, and the normalized multi-lead ECG signal is input into a general multi-lead QRS wave detection model for ECG signals. The general multi-lead QRS wave detection model for ECG signals includes a lead distillation module and a QRS wave detection module. The multi-lead ECG signal to be tested is input into the lead distillation module. The lead distillation module converts the channel dimension axis of the lead input into the width dimension, converts the multi-lead signal into a single-channel image, uses multiple convolutional units to extract features from the converted multi-lead ECG signal to be tested, inputs the extracted feature signals into a classifier to select noise-free feature signals, and sums the noise-free feature signals according to the width dimension to obtain the final distillation signal. The distillation signal is input into the QRS wave detection module to achieve image segmentation and obtain the location information of the QRS wave group.

2. The general multi-lead QRS wave detection method for electrocardiogram signals according to claim 1, characterized in that, The training process of the general multi-lead QRS wave detection model for electrocardiogram signals is as follows: Prepare the training data, which consists of electrocardiogram signals and corresponding label data selected from different category datasets, to form the training set; The ECG signals in the training set are Z-score normalized, and then the normalized data is input into the lead distillation module to obtain the distillation signal. The distillation signal is then input into the QRS wave detection module, and the Adam algorithm is used for optimization training. After training, the model with the minimum loss is saved. This model is the general multi-lead QRS wave detection model for ECG signals.

3. The general multi-lead QRS wave detection method for electrocardiogram signals according to claim 2, characterized in that, The label is a square wave in which the area of ​​the QRS group is marked with a value of 1, the width is 37 sampling points, there are 18 sampling points before and after the R peak, and the other areas are kept at a value of 0.

4. The general multi-lead QRS wave detection method for electrocardiogram signals according to claim 2, characterized in that, The loss function of the Adam algorithm is Diceloss, and its formula is shown below: ， Where N represents the signal length, The actual value of the label representing i. The network prediction value represents i.

5. The general multi-lead QRS wave detection method for electrocardiogram signals according to claim 1, characterized in that, The convolutional units of the lead distillation module include conventional convolutional units and depthwise separable convolutional units. The conventional convolutional unit includes a convolutional layer, a batch normalization layer, and a Leaky ReLU activation layer, while the depthwise separable convolutional unit includes a depthwise separable convolutional layer, a batch normalization layer, and a Leaky ReLU activation layer.

6. The general multi-lead QRS wave detection method for electrocardiogram signals according to claim 1, characterized in that, The QRS wave detection module can be any one of FCNet, U-Net, or DenseNet.

7. The general multi-lead QRS wave detection method for electrocardiogram signals according to claim 1, characterized in that, The QRS wave detection module is a multi-layer network framework based on the U-net network. The convolutional kernels at the corresponding levels are of the same size. The multi-layer network framework adopts a skip-layer connection structure, which concatenates the intermediate results of the encoder with the corresponding decoder feature maps in the channel dimension. Finally, the feature signals are activated by the Sigmoid activation function to obtain the location information of the QRS wave group.

8. A universal multi-lead QRS wave detection device for electrocardiogram signals, characterized in that, include: Acquisition module: used to acquire the multi-lead ECG signal to be tested, and input the normalized multi-lead ECG signal to be tested into a pre-constructed general multi-lead QRS wave detection model for ECG signals; the general multi-lead QRS wave detection model for ECG signals includes a lead distillation module and a QRS wave detection module; Lead distillation module: The multi-lead ECG signal to be tested is input into the lead distillation module. The lead distillation module converts the channel dimension axis of the lead input into the width dimension, converts the multi-lead signal into a single-channel image, uses multiple convolutional units to extract features from the converted multi-lead ECG signal to be tested, inputs the extracted feature signals into a classifier to select noise-free feature signals, and sums the noise-free feature signals according to the width dimension to obtain the final distilled signal; QRS wave detection module: The distillation signal is input into the QRS wave detection module to achieve image segmentation and obtain the location information of the QRS wave group.

9. A universal multi-lead QRS wave detection device for electrocardiogram signals, characterized in that, include: A multi-lead electrocardiogram signal acquisition device is used to acquire multi-lead electrocardiogram signals to be tested; The host computer is communicatively connected to the multi-lead ECG signal acquisition device and is used to execute a computer program to implement the general multi-lead QRS wave detection method for ECG signals as described in any one of claims 1 to 7. The display device is connected to the host computer and is used to display the QRS complex wave tag map of the multi-lead electrocardiogram signal to be tested.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the general multi-lead QRS wave detection method for electrocardiogram signals as described in any one of claims 1 to 7.

Images