A method, system, and electronic equipment for classifying different premature ventricular contractions (PVCs) ECG signals

Abstract

This invention relates to a method, system, and electronic device for classifying different premature ventricular contractions (PVCs) on electrocardiogram (ECG) signals. The method includes: inputting an acquired ECG signal dataset X into a trained U-Net model to obtain an embedding representation dataset Z; classifying the samples in the embedding representation dataset Z by cluster to obtain cluster assignment Y; determining whether the number of clusters in the cluster assignment Y is higher than a preset number of clusters; if it is higher than the preset number of clusters, then using the cluster assignment Y as a pseudo-label; and simultaneously inputting the ECG signal dataset X into the trained U-Net model to retrain the trained U-Net model and obtain a new embedding representation dataset Z. new The clustering process is repeated until the number of clusters in cluster assignment Y is lower than a preset number. If it is lower than the preset number, the cluster assignment Y with a lower than preset number of clusters is output. The classification of different premature ventricular contractions (PVCs) ECG signals is obtained based on the cluster assignment Y with a lower than preset number of clusters. This invention is capable of classifying different PVC ECG signals.

Description

Technical Field

[0001] This invention relates to the field of electrocardiogram (ECG) signal classification technology, and in particular to a method, system, and electronic device for classifying different premature ventricular contractions (PVCs) ECG signals. Background Technology

[0002] An electrocardiogram (ECG) records the electrical currents generated by the activity of detected myocardial cells, representing the changes in electrical signals produced by cardiac activity using a one-dimensional signal. In a normal cardiac cycle, a typical ECG mainly consists of the P wave, Q wave, R wave, S wave, and T wave, as well as the U wave, which may appear in some leads. The P wave and QRS complex correspond to atrial and ventricular depolarization, respectively, while the T wave corresponds to ventricular repolarization. A specific waveform of an ECG cycle is shown below. Figure 1 As shown.

[0003] Premature beats (PVCs) are one of the most common arrhythmias, with main symptoms including palpitations, chest tightness, and dizziness. Ventricular premature beats (VPCs) are particularly dangerous and common, exhibiting diverse morphologies. These morphologies can aid in the diagnosis of organic heart disease. If a patient shows multiple morphologies of VPCs in the same ECG lead, it is considered multifocal or polymorphic VPCs, a more serious condition. Currently, long-term ECGs are commonly used clinically for PVC diagnosis. However, due to the long recording time of long-term ECGs, doctors are prone to fatigue during review and may miss some VPCs. Therefore, a technique is needed to classify VPCs with similar morphologies, enabling batch analysis of similar VPCs. Labeling the morphology of VPC ECG signals is time-consuming and difficult, and commonly used classification algorithms require large amounts of labeled data for training, making them unsuitable for VPC ECG signal morphology classification. Clustering is an unsupervised method that classifies data into clusters using a similarity metric. It can divide data without the need for labeled data for training. Clustering makes objects within a data cluster highly similar, while being very dissimilar to objects in other clusters.

[0004] In recent years, clustering technology has been continuously improved and innovated, and deep learning-based clustering methods (deep clustering) have achieved great development in the fields of image and video. Scholars have also gradually applied deep clustering methods to electrocardiogram (ECG) signals, but most studies are limited to clustering analysis based on the different ECG waveforms of different diseases. Currently, there is a lack of research on morphological clustering of premature ventricular contractions (PVCs). Therefore, researching a morphological clustering method for PVC ECG signals is of significant value and importance.

[0005] In existing technologies, some literature uses continuous wavelet transform (CWT) to convert ECG data into two-dimensional time-frequency domain representations, and then applies deep embedding clustering to map the two-dimensional data to a low-dimensional latent space for clustering. The drawbacks of this method are: firstly, converting one-dimensional ECG data into two-dimensional data requires additional computation, and secondly, the two-dimensional data consumes more computational resources; thirdly, this method is designed for the classification of arrhythmias, focusing on the differences in ECG waveforms between different diseases rather than the differences in premature ventricular contractions (PVCs); and finally, this method is only applicable to uniformly distributed clusters and requires prior knowledge of the number of clusters.

[0006] In existing technologies, some literature assumes that ECG data follows a vMF distribution and uses a variational autoencoder (VAE) based on a long short-term memory model to reduce the dimensionality of the ECG data. Then, it uses a nonparametric hidden Markov model based on vMF to cluster the dimensionality-reduced data. This method uses a variational autoencoder for dimensionality reduction, which is a generative approach. Its drawback is that it requires pre-assuming the distribution of the data to be clustered; if the assumed distribution is incorrect, clustering will fail. Additionally, this method requires prior knowledge of the number of clusters.

[0007] In existing technologies, there is also a cohesive clustering technique applied to image data. This technique first initializes cluster labels using cohesive clustering, then uses the cluster labels to calculate triplet loss to train a convolutional neural network (CNN) to reduce the dimensionality of the data. By alternately running cohesive clustering and training the CNN, the final clustering result is obtained. However, this method is primarily designed for image data and is not suitable for electrocardiogram (ECG) signals. Furthermore, this method is unstable during CNN training, and the local structure of the original data is easily disrupted.

[0008] It is not difficult to find that the shortcomings of the existing technology include: (1) Most of the existing methods applied to ECG waveform clustering focus on waveform clustering between diseases, and lack morphological clustering specifically for diseases such as premature ventricular contractions; (2) Existing ECG waveform clustering methods require prior knowledge of the number of data clusters or assumptions about the data distribution; (3) Existing deep clustering methods based on agglomerative clustering are limited to image data and are unstable. Summary of the Invention

[0009] Therefore, the technical problem to be solved by the present invention is to overcome the complexity of the existing methods for classifying different premature ventricular contractions (which require prior knowledge of the number or distribution of data clusters) and to overcome the instability of existing neural network methods.

[0010] To address the aforementioned technical problems, this invention provides a method for classifying different premature ventricular contractions (PVCs) electrocardiogram (ECG) signals, comprising:

[0011] Step S1: Input the acquired ECG signal dataset X into the trained U-Net model to obtain the embedding expression dataset Z;

[0012] Step S2: Classify the samples in the embedded expression dataset Z by cluster to obtain the cluster assignment Y;

[0013] Step S3: Determine whether the number of clusters in the clustering assignment Y is higher than the preset number of clusters. If it is higher than the preset number of clusters, then use the clustering assignment Y as a pseudo-label and the ECG signal dataset X as input to the trained U-Net model to retrain the trained U-Net model and obtain a new embedding expression dataset Z. new Repeat step S2 until the number of clusters in cluster assignment Y is lower than the preset number of clusters; if it is lower than the preset number of clusters, output the cluster assignment Y with the lower than the preset number of clusters.

[0014] Step S4: Based on the clustering assignment Y below the preset cluster number, the different premature ventricular contractions (PVCs) are classified.

[0015] In one embodiment of the present invention, the U-Net model includes an encoding subnetwork and a decoding subnetwork;

[0016] The coding sub-network includes a first convolutional module, a first SE module, a first downsampling module, a second convolutional module, a second SE module, a second downsampling module, a third convolutional module, a third SE module, a dimensionality reduction module, and an L2 normalization layer, which are connected in sequence.

[0017] The decoding sub-network includes a dimension-upgrading module, a fourth convolutional module, a first upsampling module, a fifth convolutional module, a sixth convolutional module, a second upsampling module, a seventh convolutional module, an eighth convolutional module, and a first pointwise convolutional layer connected in sequence.

[0018] The output of the dimensionality reduction module is also connected to the input of the dimensionality increase module;

[0019] Specifically, the output of the first SE module is concatenated with the output of the seventh convolution module, and the concatenated result is then input into the eighth convolution module; the output of the second SE module is concatenated with the output of the fifth convolution module, and the concatenated result is then input into the sixth convolution module.

[0020] In one embodiment of the present invention, the first to the eighth convolutional modules each include a one-dimensional convolutional layer, a BN layer and a LeakyReLU layer connected in sequence.

[0021] In one embodiment of the present invention, the dimensionality reduction module includes a second pointwise convolutional layer and a first fully connected layer connected in sequence.

[0022] In one embodiment of the present invention, the dimensionality-upgrading module includes a second fully connected layer and a third pointwise convolutional layer connected in sequence.

[0023] In one embodiment of the present invention, the first SE module, the second SE module, and the third SE module each include an average pooling layer, a third fully connected layer, a ReLU layer, a fourth fully connected layer, a Sigmoid layer, and a scaling layer connected in sequence, wherein the input of the average pooling layer is multiplied by the output of the scaling layer.

[0024] In one embodiment of the present invention, the first downsampling module and the second downsampling module employ max pooling operation; the first upsampling module and the second upsampling module employ linear interpolation operation.

[0025] To solve the above-mentioned technical problems, the present invention provides a classification system for different premature ventricular contractions (PVCs) electrocardiogram signals, comprising:

[0026] Data construction module: used to input the acquired ECG signal dataset X into the trained U-Net model to obtain the embedding expression dataset Z;

[0027] Cluster classification module: used to classify samples in the embedded expression dataset Z into clusters to obtain cluster assignment Y;

[0028] The judgment module is used to determine whether the number of clusters in the clustering assignment Y is higher than a preset number of clusters. If it is higher than the preset number of clusters, the clustering assignment Y is used as a pseudo-label, and the ECG signal dataset X is simultaneously input into the trained U-Net model to retrain the trained U-Net model and obtain a new embedding expression dataset Z. new Then repeat the process of the cluster classification module until the number of clusters in the cluster assignment Y is lower than the preset number of clusters; if it is lower than the preset number of clusters, then output the cluster assignment Y with a lower than the preset number of clusters.

[0029] Classification module: used to classify different premature ventricular contractions (PVCs) ECG signals based on the clustering assignment Y that is below the preset cluster number.

[0030] To solve the above-mentioned technical problems, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: when the processor executes the computer program, it implements the steps of the above-mentioned different premature ventricular contraction ECG signal classification method.

[0031] To solve the above-mentioned technical problems, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that: when the computer program is executed by a processor, it implements the steps of the above-mentioned different premature ventricular contraction ECG signal classification method.

[0032] The technical solution of the present invention has the following advantages compared with the prior art:

[0033] This invention uses an agglomerative clustering algorithm and a U-Net model to alternately optimize the embedded expression dataset Z, which enables the clustering of the same premature ventricular contraction (PVC) ECG signals, and thus the classification of different PVC ECG signals.

[0034] This invention improves the U-Net model by: First, adding an SE module (channel attention module) to the U-Net model, which emphasizes channels that are more beneficial to the results; Second, improving the convolutional layers in the convolution module (converting two dimensions to one dimension and using large convolutional kernels), so that the improved convolution module can effectively accelerate the convergence speed of the network; Third, adding a dimensionality reduction module, which provides an embedding layer space for clustering assignment of samples; Fourth, adding an L2 normalization layer, which enhances the stability of the U-Net model.

[0035] Experiments have shown that this invention has achieved remarkable results in morphological clustering of premature ventricular contractions (PVCs), solving the problem of the lack of automatic analysis of PVC ECG signal morphology in clinical practice. This invention can classify PVCs according to their morphology in a short time and does not require prior assumptions about the data, thus achieving high efficiency.

[0036] This invention not only enables clinicians to quickly analyze premature ventricular contractions (PVCs) ECG signals in batches, but also provides assistance for the data annotation of PVC ECG signals. Attached Figure Description

[0037] 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.

[0038] Figure 1 This is a waveform diagram of one electrocardiogram cycle in the existing technology;

[0039] Figure 2 This is a flowchart of the method of the present invention;

[0040] Figure 3 This is a schematic diagram of the improved U-Net model structure in an embodiment of the present invention. Detailed Implementation

[0041] 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.

[0042] Example 1

[0043] Reference Figure 2 As shown, the present invention provides a method for classifying different premature ventricular contractions (PVCs) electrocardiogram signals, including:

[0044] Step S1: Input the acquired ECG signal dataset X into the trained U-Net model to obtain the embedding expression dataset Z;

[0045] Step S2: Classify the samples in the embedded expression dataset Z by cluster to obtain the cluster assignment Y;

[0046] Step S3: Determine whether the number of clusters in the clustering assignment Y is higher than the preset number of clusters. If it is higher than the preset number of clusters, then use the clustering assignment Y as a pseudo-label and the ECG signal dataset X as inputs to the trained U-Net model to retrain the trained U-Net model (using the triplet loss function and the mean squared error loss function) and obtain a new embedding representation dataset Z. new Then repeat step S2 until the number of clusters in cluster assignment Y is lower than the preset number of clusters. It should be noted that in this embodiment, the new embedding expression dataset Z is used. new As the embedding representation dataset Z (i.e., Z←Z) new Return to step S2; if the number of clusters is lower than the preset number, output the clustering assignment Y that is lower than the preset number of clusters;

[0047] Step S4: Based on the clustering assignment Y below the preset cluster number, the different premature ventricular contractions (PVCs) are classified.

[0048] This embodiment improves the structure of the U-Net model, which enhances the model's robustness to noise and artifacts in the input ECG signal data, improves the network's convergence speed, and makes the U-Net model more stable, making subsequent feature extraction more effective.

[0049] The improved U-Net model in this embodiment will be described in detail below:

[0050] Please see Figure 3 The improved U-Net model in this embodiment includes an encoding subnetwork and a decoding subnetwork;

[0051] The coding subnetwork includes a first convolutional module, a first SE module, a first downsampling module, a second convolutional module, a second SE module, a second downsampling module, a third convolutional module, a third SE module, a dimensionality reduction module, and L2 normalization, which are connected in sequence.

[0052] The decoding sub-network includes a dimension-upgrading module, a fourth convolutional module, a first upsampling module, a fifth convolutional module, a sixth convolutional module, a second upsampling module, a seventh convolutional module, an eighth convolutional module, and a first pointwise convolutional layer connected in sequence.

[0053] The output of the dimensionality reduction module is also connected to the input of the dimensionality increase module;

[0054] The output of the first SE module is concatenated with the output of the seventh convolution module, and the concatenated result is then input into the eighth convolution module; the output of the second SE module is concatenated with the output of the fifth convolution module, and the concatenated result is then input into the sixth convolution module.

[0055] It is important to note that in this embodiment, the embedded representation dataset Z is output by the L2 normalization layer in the encoding subnetwork, while the output of the first pointwise convolutional layer in the decoding subnetwork is only used during the training of the U-Net model. The output of the first pointwise convolutional layer is used to reconstruct the loss function (mean squared error loss function). The loss function continuously encourages the reduction of the gap between the U-Net model input and output, that is, to ensure the similarity between the input and output, so as to make the U-Net model training more effective.

[0056] Furthermore, each of the first to eighth convolutional modules includes a one-dimensional convolutional layer, a BN layer, and a LeakyReLU layer connected in sequence.

[0057] Specifically, since the main purpose of the existing U-Net model is to segment target regions in biomedical images, it uses 3x3 two-dimensional convolutional layers (2DConv). However, the ECG signal (electrocardiogram signal) in this embodiment is a one-dimensional signal, so a one-dimensional convolutional layer (1DConv) is used instead of the original two-dimensional convolutional layer. In the model of this embodiment, a one-dimensional convolutional layer (1DConv) with a kernel size of 5 is selected, which has a larger receptive field and enhances the model's robustness to noise and artifacts.

[0058] Batch Normalization (BN) layers are an adaptive parameter optimization method that slowly maps each hidden layer neuron to a nonlinear function, allowing the input value after nonlinear transformation to enter a region that is more sensitive to input features. This alleviates the gradient vanishing problem that occurs during network training. At the same time, since the gradient can remain relatively large, it can improve the convergence speed of the network.

[0059] Compared to the common ReLU activation function, the LeakyReLU layer retains a small portion of the gradient in the parts with values ​​less than 0, allowing neurons to continue learning and training, thereby further increasing the stability of the training process.

[0060] It is easy to see that this embodiment can effectively accelerate the convergence speed of the network by combining a one-dimensional convolutional layer with a large convolutional kernel (kernel=5), a BN layer, and a LeakyReLU layer into a convolutional module.

[0061] Furthermore, this embodiment adds a dimensionality reduction module to the bottom layer of the encoding subnetwork of the U-Net model. The dimensionality reduction module consists of two parts: a 1x1 second pointwise convolutional layer and a first fully connected layer. The 1x1 second pointwise convolutional layer performs channel transformation on the feature map encoded by the encoding subnetwork. The first fully connected layer requires a large number of parameters, but due to its flexibility, only the input and output parameters need to be changed for different embedding layer dimensions. Therefore, the dimensionality reduction module designed in this embodiment uses the first fully connected layer to establish a mapping relationship between the channel-transformed feature map and the embedding layer space of a specified dimension. This embedding layer space can be simply understood as the space required for subsequent agglomerative clustering algorithms, i.e., the feature space suitable for agglomerative clustering. After careful parameter selection, this embodiment sets the dimension of the embedding layer space to 25. In addition, this embodiment also performs L2 normalization on the output of the dimensionality reduction module to further enhance the stability of the encoding subnetwork.

[0062] Furthermore, the dimensionality-upgrading module includes a second fully connected layer and a third pointwise convolutional layer connected in sequence.

[0063] Furthermore, this embodiment adds several channel attention modules (SE modules) to the U-Net model. Specifically, the first, second, and third SE modules each include an average pooling layer, a third fully connected layer, a ReLU layer, a fourth fully connected layer, a Sigmoid layer, and a scaling layer (used to maintain the output dimension size consistent with the input dimension size) connected in sequence. The input of the average pooling layer is also multiplied by the output of the scaling layer. This embodiment uses SE modules to add an attention mechanism in the channel dimension, thereby effectively improving the accuracy of the U-Net model. The SE modules can enhance the channel features of the input feature map, and the final output of the SE modules does not change the size of the input feature map.

[0064] Furthermore, the first downsampling module and the second downsampling module employ max pooling.

[0065] Furthermore, the first upsampling module and the second upsampling module employ linear interpolation.

[0066] In step S1, an electrocardiogram (ECG) signal dataset X is obtained. In this embodiment, the ECG signal data sampling rate is 360 Hz, and the input length of the ECG signal is fixed at a uniform length (176 points, including 72 points before the R peak and 104 points after the R peak). All ECGs are used to form the ECG signal dataset X. The ECG signal dataset X is then input into the trained U-Net model, and the embedded expression dataset Z is output.

[0067] In step S2, a directed graph is generated for the embedded expression dataset Z, where each sample is called a vertex, and the weight of the edge between two vertices is determined by whether they are K-nearest neighbors. The isolated vertices and vertices on weakly connected branches are initialized into a single cluster by the agglomerative clustering algorithm to obtain the initial clustering assignment Y. The clustering assignment Y is the classification of different premature ventricular contraction ECG signals.

[0068] In step S3, it is determined whether the number of clusters in the initial clustering assignment Y is higher than the preset number of clusters. If it is higher than the preset number of clusters, then the clustering assignment Y is used as a pseudo-label. At the same time, the ECG signal dataset X is input into the trained U-Net model to retrain the trained U-Net model and obtain a new embedding expression dataset Z. Finally, the new embedding expression dataset Z is... new The embedded representation dataset Z is returned to step S2 until the number of clusters in the cluster assignment Y is lower than the preset number of clusters; if it is lower than the preset number of clusters, the final trained U-Net model and the cluster assignment Y with a lower than the preset number of clusters are output.

[0069] In step S4, the classification of different premature ventricular contractions (PVCs) is obtained based on the clustering assignment Y, which is lower than the preset cluster number.

[0070] In short, this embodiment measures the similarity between two clusters by the sum of the weights of the edges between all vertices in the two clusters, and repeatedly finds the two most similar clusters and merges them, that is, merges the same premature ventricular contraction (PVC) ECG signals, and finally obtains the classification of different PVC ECG signals.

[0071] Example 2

[0072] This embodiment provides a classification system for different premature ventricular contractions (PVCs) ECG signals, including:

[0073] Data construction module: used to input the acquired ECG signal dataset X into the trained U-Net model to obtain the embedding expression dataset Z;

[0074] Cluster classification module: used to classify samples in the embedded expression dataset Z into clusters to obtain cluster assignment Y;

[0075] The judgment module is used to determine whether the number of clusters in the clustering assignment Y is higher than a preset number of clusters. If it is higher than the preset number of clusters, the clustering assignment Y is used as a pseudo-label, and the ECG signal dataset X is simultaneously input into the trained U-Net model to retrain the trained U-Net model and obtain a new embedding expression dataset Z. new Then repeat the process of the cluster classification module until the number of clusters in the cluster assignment Y is lower than the preset number of clusters; if it is lower than the preset number of clusters, then output the cluster assignment Y with a lower than the preset number of clusters.

[0076] Classification module: used to classify different premature ventricular contractions (PVCs) ECG signals based on the clustering assignment Y that is below the preset cluster number.

[0077] Example 3

[0078] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the different premature ventricular contraction (PVC) ECG signal classification method described in Embodiment 1.

[0079] Example 4

[0080] This embodiment provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the different premature ventricular contraction ECG signal classification method described in Embodiment 1.

[0081] 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 implemented 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. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0082] 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.

[0083] 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 1The function specified in one or more boxes.

[0084] 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.

[0085] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0086] 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 method for classifying different premature ventricular contractions (PVCs) on ECG signals, characterized in that: include: Step S1: Obtain the ECG signal dataset Input the trained U-Net model to obtain the embedding representation dataset. ; Step S2: The embedded representation dataset The samples are classified into clusters to obtain cluster assignments. ; Step S3: Determine the cluster assignment If the number of clusters in the cluster is higher than the preset number of clusters, then the clustering will be assigned. As pseudo-labels, and the ECG signal dataset Simultaneously, the pre-trained U-Net model is input to retrain it and obtain a new embedding representation dataset. Repeat step S2 until cluster assignment is performed. The number of clusters in the system is lower than the preset number of clusters; If the number of clusters is less than the preset number, then output the clustering assignment for the number of clusters less than the preset number. ; The U-Net model includes an encoding sub-network and a decoding sub-network; The coding sub-network includes a first convolutional module, a first SE module, a first downsampling module, a second convolutional module, a second SE module, a second downsampling module, a third convolutional module, a third SE module, a dimensionality reduction module, and an L2 normalization layer, which are connected in sequence. The decoding sub-network includes a dimension-upgrading module, a fourth convolutional module, a first upsampling module, a fifth convolutional module, a sixth convolutional module, a second upsampling module, a seventh convolutional module, an eighth convolutional module, and a first pointwise convolutional layer connected in sequence. The output of the dimensionality reduction module is also connected to the input of the dimensionality increase module; Specifically, the output of the first SE module is concatenated with the output of the seventh convolution module, and the concatenated result is then input into the eighth convolution module; the output of the second SE module is concatenated with the output of the fifth convolution module, and the concatenated result is then input into the sixth convolution module. The first to the eighth convolutional modules each include a one-dimensional convolutional layer, a BN layer, and a LeakyReLU layer connected in sequence; The dimensionality reduction module includes a second pointwise convolutional layer and a first fully connected layer connected in sequence; The dimension-upgrading module includes a second fully connected layer and a third pointwise convolutional layer connected in sequence; The first SE module, the second SE module, and the third SE module each include an average pooling layer, a third fully connected layer, a ReLU layer, a fourth fully connected layer, a Sigmoid layer, and a scaling layer connected in sequence. The input of the average pooling layer is also multiplied by the output of the scaling layer. The first downsampling module and the second downsampling module use max pooling; the first upsampling module and the second upsampling module use linear interpolation. Step S4: Assign clusters based on the number of clusters below the preset number. Different types of premature ventricular contractions (PVCs) were classified into different ECG signals.

2. A classification system for different premature ventricular contractions (PVCs) ECG signals, used to implement the classification method for different PVCs ECG signals as described in claim 1, characterized in that: include: Data building module: used to process the acquired electrocardiogram signal dataset. Input the trained U-Net model to obtain the embedding representation dataset. ; Cluster classification module: used to classify the embedded representation dataset The samples are classified into clusters to obtain cluster assignments. ; Judgment module: used to determine the clustering assignment If the number of clusters in the cluster is higher than the preset number of clusters, then the clustering will be assigned. As pseudo-labels, and the ECG signal dataset Simultaneously, the pre-trained U-Net model is input to retrain it and obtain a new embedding representation dataset. Then repeat the process of the cluster classification module until cluster assignment is performed. The number of clusters in the system is lower than the preset number of clusters; If the number of clusters is less than the preset number, then output the clustering assignment for the number of clusters less than the preset number. ; Classification module: used for clustering assignment based on the number of clusters below the preset number. Different types of premature ventricular contractions (PVCs) were classified into different ECG signals.

3. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the different premature ventricular contractions ECG signal classification method as described in claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the different premature ventricular contractions ECG signal classification method as described in claim 1.

Images