Multi-label classification method, device and storage medium for electrocardiogram data

Abstract

The present invention relates to the field of artificial intelligence and digital medical technology, and provides a multi-label classification method, device and storage medium for ECG data. The method includes: obtaining an ECG data knowledge graph, wherein the nodes of the ECG data knowledge graph are classification labels, and the edges of the ECG data knowledge graph are attribute relationships between classification labels; extracting the feature vector of each node of the ECG data knowledge graph through a label feature extraction model, and constructing a classification label association matrix based on the distance information between the feature vectors of every two nodes; using the classification label association matrix, updating the parameters of the ECG data classification processing model to obtain an updated ECG data classification processing model; obtaining the ECG data to be processed, and inputting the ECG data into the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data. The above method can effectively improve the accuracy of multi-label classification of ECG data.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and digital medical technology, and in particular to a multi-label classification method, device and storage medium for electrocardiogram data. Background Art

[0002] Electrocardiogram (ECG) is a technology that uses an electrocardiograph to record the changes in electrical activity generated by each cardiac cycle from the human body surface. Through the electrocardiogram, a variety of human heart diseases can be characterized, and doctors can also judge the patient's heart condition based on the electrocardiogram.

[0003] In recent years, with the continuous development of artificial intelligence technology, especially the development of deep learning technology, various ECG data processing models have emerged. At present, most ECG data processing models are trained using existing ECG samples and their classification labels. However, existing ECG samples often have problems of unbalanced classification labels (i.e., a large number of labels in some categories, and a small number of labels in other categories) and inconsistent classification label standards (i.e., for the same type of ECG waveform, different samples will classify it into different types of labels). It is difficult to train a high-performance ECG data classification model based only on such ECG samples and their classification labels, especially to train a multi-label classification model for ECG data. Accordingly, the accuracy of the ECG data classification results obtained using such a model is also difficult to guarantee. Summary of the invention

[0004] In view of this, the present application provides a multi-label classification method, device and storage medium for ECG data, the main purpose of which is to solve the technical problem of low accuracy of multi-label classification results of ECG data.

[0005] According to a first aspect of the present invention, a multi-label classification method for electrocardiogram data is provided, the method comprising:

[0006] Obtain an ECG data knowledge graph, wherein the nodes of the ECG data knowledge graph are classification labels of the ECG data, and the edges of the ECG data knowledge graph are attribute relationships between classification labels;

[0007] The pre-trained label feature extraction model is used to extract the feature vector of each node in the ECG data knowledge graph, and the classification label association matrix is ​​constructed based on the distance information between the feature vectors of every two nodes.

[0008] Using the classification label association matrix, the parameters of the pre-trained ECG data classification processing model are updated to obtain an updated ECG data classification processing model;

[0009] The ECG data to be processed is obtained, and the ECG data is input into the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data.

[0010] According to a second aspect of the present invention, a multi-label classification device for electrocardiogram data is provided, the device comprising:

[0011] A knowledge graph acquisition module is used to acquire an ECG data knowledge graph, wherein the nodes of the ECG data knowledge graph are classification labels of the ECG data, and the edges of the ECG data knowledge graph are attribute relationships between classification labels;

[0012] The association matrix construction module is used to extract the feature vector of each node of the ECG data knowledge graph through the pre-trained label feature extraction model, and construct the classification label association matrix based on the distance information between the feature vectors of every two nodes;

[0013] A classification model updating module is used to update the parameters of the pre-trained ECG data classification processing model using the classification label association matrix to obtain an updated ECG data classification processing model;

[0014] The ECG data processing module is used to obtain the ECG data to be processed and input the ECG data into the updated ECG data classification processing model to obtain the multi-label classification result of the ECG data.

[0015] According to a third aspect of the present invention, there is provided a storage medium having a computer program stored thereon, which implements the above-mentioned multi-label classification method for ECG data when the program is executed by a processor.

[0016] According to a fourth aspect of the present invention, there is provided a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the multi-label classification method for electrocardiogram data when executing the program.

[0017] The present invention provides a multi-label classification method, device and storage medium for ECG data. First, an ECG data knowledge graph composed of classification labels of ECG data is obtained. Then, a classification label association matrix is ​​constructed through the ECG data knowledge graph so that each classification label of the ECG data has a certain internal correlation. Then, the parameters of the ECG data classification processing model are updated through the classification label association matrix to improve the classification performance of the ECG data classification processing model. Finally, the ECG data is processed through the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data. The above method combines the association between each classification label of the ECG data with the training of the ECG data classification processing model, so that the updated ECG data classification processing model can combine the association between the classification labels in the process of multi-label classification processing of the ECG data, thereby overcoming the problems of overlapping classification labels of multiple data sources and uneven distribution of classification labels, and effectively improving the accuracy of multi-label classification processing of ECG data.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0020] Figure 1 A schematic diagram of a process flow of a multi-label classification method for electrocardiogram data provided by an embodiment of the present invention is shown;

[0021] Figure 2 A schematic structural diagram of a multi-label classification device for electrocardiogram data provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0022] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0023] At present, most ECG data classification models are trained based on rich historical label data. However, most current ECG data classification models are trained separately for each classification label, without considering the correlation between multiple classification labels. This will cause the following two problems:

[0024] First, the problem is caused by the imbalance of classification labels of ECG data samples. For example, it is known that classification label A and classification label B often occur at the same time. Then, when the model determines that a certain data belongs to classification label A, the data will also belong to classification label B with a high probability. However, if classification label B appears less in the sample data, then the probability of the model giving classification label B is also relatively low. At this time, the model's prediction of classification label B will be inaccurate.

[0025] Second, the problem is caused by the inconsistency of the classification standards of ECG data samples. Large-scale ECG data samples often come from different data sources. Therefore, the quality of their classification labels is usually uneven. A typical case is that their respective label systems are inconsistent. When these data are combined, there will be label overlap problems. For example, classification label A includes classification label B and classification label C. Some data sources will attribute all ECG data samples to A, while some data sources will attribute ECG data samples to B or C. Similar waveforms are divided into different categories, which makes it difficult for the model to distinguish the correct results.

[0026] In response to the above problems, some existing technologies propose to use the co-occurrence of different labels of ECG data samples in the ECG data classification model to inject the correlation between multiple classification labels into the model. However, due to the limitation of sample data volume, label quality and other issues, this method is difficult to comprehensively and objectively reflect the correlation between multiple classification labels, which ultimately leads to low accuracy of multi-label classification of ECG data.

[0027] Based on this, in one embodiment, Figure 1 As shown, a multi-label classification method for ECG data is provided, and the method is applied to a computer device such as a server as an example for explanation, wherein the server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. The above method includes the following steps:

[0028] 101. Obtain an ECG data knowledge graph, wherein the nodes of the ECG data knowledge graph are classification labels of the ECG data, and the edges of the ECG data knowledge graph are attribute relationships between classification labels.

[0029] Specifically, the computer device can obtain the ECG data knowledge graph through a data interface or a network. In this embodiment, the nodes of the ECG data knowledge graph are the classification labels of the ECG data, and the edges of the ECG data knowledge graph are the attribute relationships between the classification labels. The ECG data knowledge graph can contain multiple nodes and multiple edges, wherein there is an edge between two associated nodes, and this edge represents the attribute relationship between the two nodes. The attribute relationship can be that classification label A belongs to classification label B, or classification label A leads to classification label B, etc. It can be understood that the relationship between classification labels is different, and the attributes are different.

[0030] 102. Through the pre-trained label feature extraction model, the feature vector of each node of the ECG data knowledge graph is extracted, and the classification label association matrix is ​​constructed based on the distance information between the feature vectors of every two nodes.

[0031] Specifically, the computer device can extract the feature vector of each node of the ECG data knowledge graph through a pre-trained label feature extraction model, wherein the label feature extraction model can be trained by models such as Graph Attention Network (GAT) and Multilayer Perceptron (MLP). Furthermore, the computer device can construct a classification label association matrix with the distance information between the feature vectors of each two nodes as an element, wherein the distance information can be used to characterize the similarity between the two feature vectors. For example, the distance information can be Euclidean distance, etc. In addition, the distance information can also be cosine similarity, Pearson correlation coefficient, Jaccard similarity coefficient, etc. The classification label association matrix is ​​a square matrix. In the classification label association matrix, R ij Represents the distance between the feature vectors of the i-th classification label and the j-th classification label, R ii The distance is 0.

[0032] 103. Using the classification label association matrix, update the parameters of the pre-trained ECG data classification processing model to obtain an updated ECG data classification processing model.

[0033] Specifically, the computer device can use the classification label association matrix to update the parameters of the pre-trained ECG data classification processing model. For example, the classification label association matrix can be used to transform the output of multiple batches of ECG data samples, thereby converting the multi-label classification prediction probability of the ECG data samples into a multi-label classification prediction probability with correlation, and then the original ECG data samples and the converted multi-label classification prediction probability are used to iteratively update the parameters of the ECG data classification processing model to obtain an updated ECG data classification processing model.

[0034] 104. Obtain the ECG data to be processed, and input the ECG data into the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data.

[0035] Specifically, the computer device can obtain the ECG data to be processed through a data interface or an ECG data acquisition device, and then input the ECG data to be processed into an updated ECG data classification processing model to obtain a multi-label classification result of the ECG data, that is, obtain the predicted probabilities of multiple classification labels corresponding to the ECG data.

[0036] The present embodiment provides a multi-label classification method for ECG data, which first obtains an ECG data knowledge graph composed of classification labels of ECG data, then constructs a classification label association matrix through the ECG data knowledge graph, so that each classification label of the ECG data has a certain internal correlation, and then updates the parameters of the ECG data classification processing model through the classification label association matrix to improve the classification performance of the ECG data classification processing model, and finally processes the ECG data through the updated ECG data classification processing model to obtain the multi-label classification result of the ECG data. The above method combines the association between each classification label of the ECG data with the training of the ECG data classification processing model, so that the updated ECG data classification processing model can combine the association between the classification labels in the process of multi-label classification processing of the ECG data, thereby overcoming the problems of overlapping classification labels of multiple data sources and uneven distribution of classification labels, and effectively improving the accuracy of multi-label classification processing of ECG data.

[0037] In one embodiment, the training method of the label feature extraction model in the above step 102 can be implemented by the following method: first, an ECG data knowledge graph and multiple groups of ECG data samples are obtained, wherein each group of ECG data samples contains multiple classification labels, and the multiple classification labels are a subset of the ECG data knowledge graph; then, based on the above ECG data knowledge graph and the multiple groups of ECG data samples, a graph attention network and a multi-layer perceptron prediction model are constructed; and then, through the graph attention network, the initial feature vector of each node of the ECG data knowledge graph is extracted, and based on the initial feature vector of each node of the ECG data knowledge graph, a feature vector combination of multiple classification labels of each group of ECG data samples is obtained; finally, with the feature vector combination of multiple classification labels of each group of ECG data samples as input and the multiple classification labels of each group of ECG data samples as output, the above-mentioned graph attention network and the multi-layer perceptron prediction model are synchronously iteratively trained to obtain a label feature extraction model.

[0038] In the above embodiment, the ECG data knowledge graph used for model training can be the same knowledge graph as the ECG data knowledge graph in step 101 and step 102, or it can be a different knowledge graph. The similarities are that the nodes of the two ECG data knowledge graphs are classification labels of ECG data, the edges of the ECG data knowledge graphs are attribute relationships between classification labels, and the multiple classification labels of the ECG data samples used for model training are a subset of the ECG data knowledge graph. Furthermore, the method of extracting the initial feature vector of each node of the ECG data knowledge graph through the graph attention network can refer to the existing method, that is, the feature vectors of the surrounding nodes of the ECG data knowledge graph can be read out through the graph attention network, and then the feature vectors of the surrounding nodes of the knowledge graph are weighted summed to obtain the feature vector of the previous node, and the feature vectors of all nodes are obtained by continuous recursion. Furthermore, after obtaining the initial feature vector of each node of the ECG data knowledge graph, the feature vector combination of each group of ECG data samples can be obtained by splicing and other methods. Finally, the feature vector combination of the ECG data samples is used as input and multiple classification labels of the ECG data samples are used as output. The graph attention network and the multi-layer perceptron prediction model are synchronously iteratively trained to obtain a label feature extraction model.

[0039] In one embodiment, a method for obtaining a feature vector combination of ECG data samples during the training process of a label feature extraction model can be implemented by the following method: first, extract the classification label whose classification result is the first classification result in each group of ECG data samples, wherein the first classification result is the classification result of the ECG data sample belonging to the classification label; then, based on the initial feature vector of each node of the ECG data knowledge graph, determine the initial feature vector of the classification label whose classification result is the first classification result in each group of ECG data samples; finally, concatenate the initial feature vectors of the classification label whose classification result is the first classification result in each group of ECG data samples to obtain a feature vector combination of multiple classification labels for each group of ECG data samples.

[0040] In the above embodiment, it is assumed that N groups of ECG data samples are given, namely D = {D1, D2, ..., D N}, the classification label corresponding to the ECG data sample is Y = {Y1, Y2, ..., Y N}, where Y i is a multi-valued vector {yi1, yi2, ..., yi n}, a bit of 1 indicates that a sample belongs to a certain classification label, and a bit of 0 indicates that a sample does not belong to a certain classification label. Each ECG data sample has n classification labels. For this group of ECG data samples, a multi-layer perceptron can be used for auto-encoder to train the parameters of the entire network (including the parameters of the graph attention network and the multi-layer perceptron prediction model). Among them, the input of the multi-layer perceptron is Y i The feature vector combination is obtained by summing the feature vectors of the corresponding nodes on the ECG data knowledge graph for all classification labels corresponding to the bits that are 1 in the data. The output of the multilayer perceptron is the original Y i After the model training is completed, the feature vectors of the nodes in the ECG data knowledge graph corresponding to all classification labels in the ECG data samples can be extracted through the graph attention network, and the classification label association matrix can be constructed based on the distance between the feature vectors of every two nodes.

[0041] In one embodiment, the training method of the ECG data classification processing model in the above step 103 can be implemented in the following manner: first, multiple groups of ECG data samples are obtained, and a group of convolutional neural networks and a multi-layer perceptron classification model are constructed according to the multiple groups of ECG data samples, wherein each group of ECG data samples contains multiple classification labels, then, according to the multiple groups of ECG data samples, through the convolutional neural network, the characteristic vector combination of the multiple groups of ECG data samples can be obtained, and finally, the characteristic vector combination of the multiple groups of ECG data samples is used as input, and the multiple classification labels of the multiple groups of ECG data samples are used as output, and the convolutional neural network model and the multi-layer perceptron classification model are synchronously iteratively trained through a preset parameter adjustment strategy, so as to obtain the ECG data classification processing model. In this embodiment, the parameters of the multi-layer perceptron model mainly include the connection weights and biases between the layers in the model, and the parameter adjustment strategy can be selected according to the actual situation. A relatively simple parameter adjustment strategy is the gradient descent method (SGD), that is, firstly randomly initialize the parameters in the model, and then iteratively train, continuously calculate the gradient and update the parameters until a certain condition is met (such as when the error is small enough and the number of iterations is large enough), and the model training is completed.

[0042] In one embodiment, a group of convolutional neural networks includes multiple convolutional neural networks corresponding to multiple lead data of the ECG data samples. Then, the method for obtaining the feature vector combination of multiple groups of ECG data samples can be implemented in the following manner: first, the multiple lead data of the multiple groups of ECG data samples are respectively input into the corresponding multiple convolutional neural networks to obtain multiple feature vectors of each group of ECG data samples, and then the multiple feature vectors of each group of ECG data samples are spliced ​​to obtain the feature vector combination of each group of ECG data samples. In this embodiment, when splicing the feature vector combination, the N feature vectors of the ECG data samples can be first separated, and then the separated multiple feature vectors can be spliced. For example, the feature vectors of each lead can be separated by [SEP] as a delimiter, and then the [CLS] delimiter is spliced ​​at the beginning of each sample, and finally the feature vectors of each lead after separation are spliced ​​together to obtain the feature vector combination of each sample. Among them, the [SEP] delimiter can separate the feature vectors of each lead in the sample, and the [CLS] delimiter can facilitate the extraction of subsequent feature vector combinations. This embodiment obtains the feature vector combination of the unlabeled ECG sample by splicing, which can effectively improve the generation efficiency and extraction efficiency of the feature vector combination.

[0043] In one embodiment, the above step 103 can be specifically implemented by the following method: obtaining multiple groups of ECG data samples, and according to the multiple groups of ECG data samples, obtaining the multi-label prediction probability of the multiple groups of ECG data samples through the ECG data classification processing model, and then using the classification label association matrix to convert the multi-label prediction probability of the multiple groups of ECG data samples into the multi-label correlation prediction probability, and according to the multi-label correlation prediction probability of the multiple groups of ECG data samples, calculating the cross entropy loss function of the multiple groups of ECG data samples, and updating the parameters of the ECG data classification processing model through the cross entropy loss function to obtain the updated ECG data classification processing model. In this embodiment, the classification label association matrix R can be used to convert the prediction probability Y′ of each classification label of the ECG data sample into the prediction probability of the multi-label correlation prediction probability. i Convert to Y″ i , the reference formula is as follows:

[0044]

[0045] Then, we can use Y″ ij The cross entropy loss function is calculated to update the parameters of the ECG data classification processing model, including the parameters of the convolutional neural network and the multi-layer perceptron classification model. The formula of the cross entropy loss function is as follows:

[0046]

[0047] After training, given an ECG data, we can first use a convolutional neural network to obtain the feature vectors of each lead of the ECG data, and then splice them and input them into the multi-layer perceptron classification model to obtain the multi-label classification results of the ECG data, that is, to obtain the predicted probabilities of multiple classification labels corresponding to the ECG data.

[0048] In one embodiment, step 104 can be specifically implemented by the following method: first, multiple lead data of the ECG data are respectively input into multiple convolutional neural networks of the updated ECG data classification processing model to obtain multiple feature vectors of the ECG data, and then the multiple feature vectors of the ECG data are spliced ​​to obtain a feature vector combination of the ECG data, and finally the feature vector combination of the ECG data is input into the multi-layer perceptron classification model of the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data. In this embodiment, the splicing method of the feature vector combination of the ECG data can refer to the splicing method of the feature vector combination of the ECG data sample, which will not be repeated in this embodiment. Further, the multi-label classification result of the ECG data is the predicted probability that the ECG data belongs to each classification label.

[0049] Further, as Figure 1 The specific implementation of the method shown in the embodiment provides a multi-label classification device for electrocardiogram data, such as Figure 2 As shown, the device includes: a knowledge graph acquisition module 21, an association matrix construction module 22, a classification model update module 23 and an ECG data processing module 24.

[0050] The knowledge graph acquisition module 21 can be used to acquire the ECG data knowledge graph, wherein the nodes of the ECG data knowledge graph are the classification labels of the ECG data, and the edges of the ECG data knowledge graph are the attribute relationships between the classification labels;

[0051] The association matrix construction module 22 can be used to extract the feature vector of each node of the ECG data knowledge graph through a pre-trained label feature extraction model, and construct a classification label association matrix based on the distance information between the feature vectors of every two nodes;

[0052] The classification model updating module 23 can be used to update the parameters of the pre-trained ECG data classification processing model using the classification label association matrix to obtain an updated ECG data classification processing model;

[0053] The ECG data processing module 24 can be used to obtain the ECG data to be processed, and input the ECG data into the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data.

[0054] In a specific application scenario, the training method of the label feature extraction model in the association matrix construction module 22 includes: obtaining an ECG data knowledge graph and multiple groups of ECG data samples, wherein each group of ECG data samples contains multiple classification labels, and the multiple classification labels are a subset of the ECG data knowledge graph; constructing a graph attention network and a multi-layer perceptron prediction model based on the ECG data knowledge graph and the multiple groups of ECG data samples; extracting the initial feature vector of each node of the ECG data knowledge graph through the graph attention network, and obtaining a feature vector combination of multiple classification labels of each group of ECG data samples based on the initial feature vector of each node of the ECG data knowledge graph; taking the feature vector combination of multiple classification labels of each group of ECG data samples as input and taking the multiple classification labels of each group of ECG data samples as output, synchronously iteratively training the graph attention network and the multi-layer perceptron prediction model to obtain a label feature extraction model.

[0055] In a specific application scenario, according to the initial feature vector of each node of the ECG data knowledge graph, a feature vector combination of multiple classification labels of each group of ECG data samples is obtained, including: extracting the classification label whose classification result is the first classification result in each group of ECG data samples, wherein the first classification result is the classification result of the ECG data sample belonging to the classification label; according to the initial feature vector of each node of the ECG data knowledge graph, determining the initial feature vector of the classification label whose classification result is the first classification result in each group of ECG data samples; splicing the initial feature vectors of the classification label whose classification result is the first classification result in each group of ECG data samples, to obtain a feature vector combination of multiple classification labels for each group of ECG data samples.

[0056] In a specific application scenario, the training method of the ECG data classification processing model in the classification model update module 23 includes: obtaining multiple groups of ECG data samples, and constructing a group of convolutional neural networks and a multi-layer perceptron classification model based on the multiple groups of ECG data samples, wherein each group of ECG data samples contains multiple classification labels; obtaining a combination of feature vectors of the multiple groups of ECG data samples through a convolutional neural network based on the multiple groups of ECG data samples; taking the combination of feature vectors of the multiple groups of ECG data samples as input and taking the multiple classification labels of the multiple groups of ECG data samples as output, synchronously iteratively training the convolutional neural network model and the multi-layer perceptron classification model to obtain the ECG data classification processing model.

[0057] In a specific application scenario, a group of convolutional neural networks includes multiple convolutional neural networks corresponding one-to-one to multiple lead data of ECG data samples; then, based on the multiple groups of ECG data samples, a combination of feature vectors of the multiple groups of ECG data samples are obtained through the convolutional neural network, including: inputting the multiple lead data of the multiple groups of ECG data samples into the corresponding multiple convolutional neural networks respectively, to obtain multiple feature vectors of each group of ECG data samples; and splicing the multiple feature vectors of each group of ECG data samples respectively, to obtain a combination of feature vectors of each group of ECG data samples.

[0058] In a specific application scenario, the classification model update module 23 can be specifically used to obtain multiple groups of ECG data samples, and based on the multiple groups of ECG data samples, obtain the multi-label prediction probabilities of the multiple groups of ECG data samples through the ECG data classification processing model; use the classification label association matrix to convert the multi-label prediction probabilities of the multiple groups of ECG data samples into multi-label correlation prediction probabilities; calculate the cross-entropy loss function of the multiple groups of ECG data samples based on the multi-label correlation prediction probabilities of the multiple groups of ECG data samples, and update the parameters of the ECG data classification processing model through the cross-entropy loss function to obtain an updated ECG data classification processing model.

[0059] In a specific application scenario, the ECG data processing module 24 can be specifically used to input multiple lead data of the ECG data into multiple convolutional neural networks of an updated ECG data classification processing model to obtain multiple feature vectors of the ECG data; concatenate the multiple feature vectors of the ECG data to obtain a feature vector combination of the ECG data; input the feature vector combination of the ECG data into a multi-layer perceptron classification model of the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data.

[0060] It should be noted that for other corresponding descriptions of the functional units involved in the multi-label classification device for electrocardiogram data provided in this embodiment, please refer to Figure 1 The corresponding description in will not be repeated here.

[0061] Based on the above Figure 1 The method shown in the embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned Figure 1 Multi-label classification method for ECG data shown.

[0062] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the multi-label classification method of ECG data in various implementation scenarios of the present application.

[0063] Based on the above Figure 1 The method shown, and Figure 2 In order to achieve the above-mentioned purpose, the present embodiment also provides a physical device for multi-label classification of ECG data, which can be a personal computer, a server, a smart phone, a tablet computer, a smart watch, or other network devices, etc. The physical device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1 The method shown.

[0064] Optionally, the physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0065] Those skilled in the art will appreciate that the physical device structure for multi-label classification of ECG data provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different arrangements of components.

[0066] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware of the physical device and the software resources to be identified, and supports the operation of the information processing program and other software and / or programs to be identified. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing physical device.

[0067] Through the description of the above implementation methods, the technicians in this field can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware. By applying the technical solution of the present application, the ECG data knowledge graph composed of the classification labels of the ECG data is first obtained, and then the classification label association matrix is ​​constructed through the ECG data knowledge graph, so that each classification label of the ECG data has a certain internal correlation, and then the parameters of the ECG data classification processing model are updated through the classification label association matrix to improve the classification performance of the ECG data classification processing model, and finally the ECG data is processed by the updated ECG data classification processing model to obtain the multi-label classification result of the ECG data. The above method combines the association between the classification labels of the ECG data with the training of the ECG data classification processing model, so that the updated ECG data classification processing model can combine the association between the classification labels in the process of multi-label classification processing of the ECG data, thereby overcoming the problems of overlapping classification labels of multiple data sources and uneven distribution of classification labels, and effectively improving the accuracy of multi-label classification processing of ECG data.

[0068] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0069] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A multi-label classification method for electrocardiogram data, characterized in that: The method comprises: Acquire an ECG data knowledge graph, wherein the nodes of the ECG data knowledge graph are classification labels of the ECG data, and the edges of the ECG data knowledge graph are attribute relationships between classification labels; The feature vector of each node of the ECG data knowledge graph is extracted through a pre-trained label feature extraction model, and a classification label association matrix is ​​constructed according to the distance information between the feature vectors of every two nodes; Using the classification label association matrix, updating the parameters of the pre-trained ECG data classification processing model to obtain an updated ECG data classification processing model, specifically comprising: acquiring multiple groups of ECG data samples, and according to the multiple groups of ECG data samples, obtaining the multi-label prediction probabilities of the multiple groups of ECG data samples through the ECG data classification processing model; using the classification label association matrix, converting the multi-label prediction probabilities of the multiple groups of ECG data samples into multi-label correlation prediction probabilities; calculating the cross entropy loss function of the multiple groups of ECG data samples according to the multi-label correlation prediction probabilities of the multiple groups of ECG data samples, and updating the parameters of the ECG data classification processing model through the cross entropy loss function to obtain an updated ECG data classification processing model; Acquire the ECG data to be processed, and input the ECG data into the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data; The formula for converting the multi-label prediction probability of the ECG data sample into the multi-label correlation prediction probability using the classification label association matrix is ​​as follows: Among them, y″ ij is the multi-label correlation prediction probability of the j-th classification label of the i-th ECG data sample, R is the classification label correlation matrix, R jk is the distance between the feature vectors of the jth classification label and the kth classification label, y′ ij The multi-label prediction probability of the j-th classification label for the i-th ECG data sample.

2. The method according to claim 1, characterized in that: The training method of the label feature extraction model includes: Acquire an ECG data knowledge graph and multiple groups of ECG data samples, wherein each group of the ECG data samples includes multiple classification labels, and the multiple classification labels are a subset of the ECG data knowledge graph; According to the ECG data knowledge graph and multiple groups of ECG data samples, a graph attention network and a multi-layer perceptron prediction model are constructed; Extracting the initial feature vector of each node of the ECG data knowledge graph through the graph attention network, and obtaining a feature vector combination of multiple classification labels of each group of ECG data samples according to the initial feature vector of each node of the ECG data knowledge graph; Taking the feature vector combination of multiple classification labels of each group of ECG data samples as input and the multiple classification labels of each group of ECG data samples as output, the graph attention network and the multi-layer perceptron prediction model are synchronously iteratively trained to obtain the label feature extraction model.

3. The method according to claim 2, characterized in that The method of obtaining a combination of feature vectors of multiple classification labels of each group of ECG data samples according to the initial feature vector of each node of the ECG data knowledge graph includes: Extracting a classification label whose classification result is a first classification result in each group of the electrocardiogram data samples, wherein the first classification result is a classification result of the electrocardiogram data sample belonging to the classification label; According to the initial feature vector of each node of the electrocardiogram data knowledge graph, determining the initial feature vector of the classification label whose classification result is the first classification result in each group of the electrocardiogram data samples; The initial feature vectors of the classification labels whose classification results in each group of the electrocardiogram data samples are the first classification results are concatenated to obtain a feature vector combination of multiple classification labels of each group of electrocardiogram data samples.

4. The method according to claim 1, characterized in that: The training method of the electrocardiogram data classification processing model comprises: Acquire multiple groups of ECG data samples, and construct a group of convolutional neural networks and a multi-layer perceptron classification model according to the multiple groups of ECG data samples, wherein each group of ECG data samples contains multiple classification labels; According to the multiple groups of ECG data samples, obtaining a combination of feature vectors of the multiple groups of ECG data samples through the convolutional neural network; Taking the feature vector combination of the multiple groups of ECG data samples as input and the multiple classification labels of the multiple groups of ECG data samples as output, the convolutional neural network model and the multi-layer perceptron classification model are synchronously iteratively trained to obtain the ECG data classification processing model.

5. The method according to claim 4, characterized in that The group of convolutional neural networks includes a plurality of convolutional neural networks corresponding one-to-one to a plurality of lead data of the electrocardiogram data samples; then, according to the plurality of groups of electrocardiogram data samples, the feature vector combinations of the plurality of groups of electrocardiogram data samples are obtained through the convolutional neural networks, including: Inputting the plurality of lead data of the plurality of groups of ECG data samples into the corresponding plurality of convolutional neural networks respectively, to obtain a plurality of feature vectors of each group of the ECG data samples; The multiple feature vectors of each group of the electrocardiogram data samples are concatenated to obtain a feature vector combination of each group of the electrocardiogram data samples.

6. The method according to claim 1, characterized in that The step of inputting the ECG data into the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data includes: Inputting the multiple lead data of the ECG data into the multiple convolutional neural networks of the updated ECG data classification processing model respectively to obtain multiple feature vectors of the ECG data; splicing a plurality of feature vectors of the electrocardiogram data to obtain a feature vector combination of the electrocardiogram data; The feature vector combination of the electrocardiogram data is input into the multi-layer perceptron classification model of the updated electrocardiogram data classification processing model to obtain a multi-label classification result of the electrocardiogram data.

7. A multi-label classification device for electrocardiogram data, characterized in that: The device comprises: A knowledge graph acquisition module, used to acquire an ECG data knowledge graph, wherein the nodes of the ECG data knowledge graph are classification labels of the ECG data, and the edges of the ECG data knowledge graph are attribute relationships between classification labels; An association matrix construction module is used to extract the feature vector of each node of the ECG data knowledge graph through a pre-trained label feature extraction model, and to construct a classification label association matrix based on the distance information between the feature vectors of every two nodes; The classification model updating module is used to update the parameters of the pre-trained ECG data classification processing model using the classification label association matrix to obtain an updated ECG data classification processing model, specifically comprising: acquiring multiple groups of ECG data samples, and obtaining the multi-label prediction probabilities of the multiple groups of ECG data samples through the ECG data classification processing model according to the multiple groups of ECG data samples; using the classification label association matrix, converting the multi-label prediction probabilities of the multiple groups of ECG data samples into multi-label correlation prediction probabilities; calculating the cross entropy loss functions of the multiple groups of ECG data samples according to the multi-label correlation prediction probabilities of the multiple groups of ECG data samples, and updating the parameters of the ECG data classification processing model through the cross entropy loss function to obtain an updated ECG data classification processing model; An ECG data processing module, used for acquiring the ECG data to be processed, and inputting the ECG data into the updated ECG data classification processing model to obtain a multi-label classification result of the ECG data; The formula for converting the multi-label prediction probability of the ECG data sample into the multi-label correlation prediction probability using the classification label association matrix is ​​as follows: Among them, y″ ij is the multi-label correlation prediction probability of the j-th classification label of the i-th ECG data sample, R is the classification label correlation matrix, R jk is the distance between the feature vectors of the jth classification label and the kth classification label, y′ ij The multi-label prediction probability of the j-th classification label for the i-th ECG data sample.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images