阅读说明：本技术 少导联心电数据处理方法、装置、存储介质及计算机设备 (Few-lead electrocardiogram data processing method and device, storage medium and computer equipment ) 是由 徐啸 于 2021-08-31 设计创作，主要内容包括：本发明涉及人工智能及数字医疗领域,提供了一种少导联心电数据处理方法、装置、存储介质及计算机设备。该方法包括：获取少导联心电数据,并根据少导联心电数据的N个导联,确定少导联心电数据的M个待映射导联,N与M之和为12；通过心电特征提取器对少导联心电数据的N个导联进行特征提取,得到N导联特征向量组合；将N导联特征向量组合输入到特征向量映射器中,得到少导联心电数据的M个待映射导联的特征向量；对少导联心电数据的N导联特征向量组合和M个待映射导联的特征向量进行拼接,得到12导联特征向量组合；将12导联特征向量组合输入到心电数据处理模型中,得到少导联心电数据的分类结果。上述方法可以提高少导联心电数据的分类准确性。(The invention relates to the field of artificial intelligence and digital medical treatment, and provides a method and a device for processing few-lead electrocardiogram data, a storage medium and computer equipment. The method comprises the following steps: acquiring low-lead electrocardiogram data, and determining M leads to be mapped of the low-lead electrocardiogram data according to the N leads of the low-lead electrocardiogram data, wherein the sum of the N and the M is 12; performing characteristic extraction on N leads of the few-lead electrocardiogram data through an electrocardiogram characteristic extractor to obtain N-lead characteristic vector combinations; inputting the N lead eigenvector combination into an eigenvector mapper to obtain M eigenvectors of leads to be mapped of the few-lead electrocardiogram data; splicing the N lead eigenvector combination of the few-lead electrocardiogram data and the M eigenvectors of leads to be mapped to obtain a 12-lead eigenvector combination; and (4) inputting the 12-lead feature vector combination into the electrocardiogram data processing model to obtain a classification result of the few-lead electrocardiogram data. The method can improve the classification accuracy of the few-lead electrocardiogram data.)

1. A method for processing low-lead electrocardiographic data, the method comprising:

acquiring low-lead electrocardiogram data, and determining M leads to be mapped of the low-lead electrocardiogram data according to the N leads of the low-lead electrocardiogram data, wherein the N and the M are positive integers, and the sum of the N and the M is 12;

performing feature extraction on the N leads of the low-lead electrocardiographic data through a pre-trained electrocardiographic feature extractor to obtain an N-lead feature vector combination of the low-lead electrocardiographic data;

inputting the N lead eigenvector combination of the few-lead electrocardiographic data into a pre-trained eigenvector mapper to obtain M eigenvectors of leads to be mapped of the few-lead electrocardiographic data;

splicing the N-lead eigenvector combination of the few-lead electrocardiogram data and the M eigenvectors of the leads to be mapped to obtain a 12-lead eigenvector combination of the few-lead electrocardiogram data;

and combining and inputting the 12-lead feature vectors of the low-lead electrocardiogram data into a pre-trained electrocardiogram data processing model to obtain a classification result of the low-lead electrocardiogram data.

2. The method according to claim 1, wherein the ecg feature extractor comprises at least N trained convolutional neural network models, wherein the N convolutional neural network models correspond to N leads of the low-lead ecg data, respectively;

then, the extracting the features of the N leads of the low-lead electrocardiographic data by the pre-trained electrocardiographic feature extractor to obtain an N-lead feature vector combination of the low-lead electrocardiographic data, including:

respectively inputting the N leads of the low-lead electrocardiogram data into N convolutional neural network models of the electrocardiogram feature extractor to obtain feature vectors of the N leads of the low-lead electrocardiogram data;

and splicing the N lead eigenvectors of the low-lead electrocardiogram data to obtain the N lead eigenvector combination of the low-lead electrocardiogram data.

3. The method according to claim 1, wherein the feature vector mapper comprises at least M trained multi-layered perceptron prediction models, wherein the M multi-layered perceptron prediction models respectively correspond to M leads of the low-lead electrocardiographic data to be mapped;

then, the combining and inputting the N lead feature vectors of the low-lead electrocardiographic data into a pre-trained feature vector mapper to obtain M feature vectors of leads to be mapped of the low-lead electrocardiographic data, including:

and respectively inputting the N lead eigenvector combination of the low-lead electrocardiogram data into M multilayer perceptron prediction models of the eigenvector mapper to obtain M eigenvectors of leads to be mapped of the low-lead electrocardiogram data.

4. The method according to any one of claims 1-3, wherein the method for training the ECG feature extractor and the ECG data processing model comprises:

acquiring a plurality of groups of 12-lead electrocardiogram data samples, and constructing a group of convolutional neural network models and a multi-layer perceptron classification model according to the plurality of groups of 12-lead electrocardiogram data samples;

obtaining feature vector combinations of the multiple groups of 12-lead electrocardiogram data samples through the convolutional neural network model according to the multiple groups of 12-lead electrocardiogram data samples;

and synchronously training the convolutional neural network model and the multilayer perceptron classification model by taking the feature vector combination of the multiple groups of 12-lead electrocardiogram data samples as input and the classification labels of the multiple groups of 12-lead electrocardiogram data samples as output to obtain an electrocardiogram feature extractor and an electrocardiogram data processing model.

5. The method of claim 4, wherein the set of convolutional neural network models comprises 12 convolutional neural network models, wherein the 12 convolutional neural network models correspond to 12 leads of the 12 lead electrocardiographic data samples, respectively;

obtaining a feature vector combination of the multiple groups of 12-lead electrocardiographic data samples through the convolutional neural network model according to the multiple groups of 12-lead electrocardiographic data samples, including:

respectively inputting 12-lead electrocardiogram data of the multiple groups of 12-lead electrocardiogram data samples into the 12 convolutional neural network models to obtain 12 eigenvectors of the multiple groups of 12-lead electrocardiogram data samples;

and respectively splicing the 12 eigenvectors of each group of 12-lead electrocardiogram data samples to obtain the eigenvector combination of the multiple groups of 12-lead electrocardiogram data samples.

6. The method according to any one of claims 1-3, wherein the training method of the feature vector mapper comprises:

acquiring a plurality of groups of 12-lead electrocardiogram data samples, and constructing a group of multilayer perceptron prediction models according to the plurality of groups of 12-lead electrocardiogram data samples;

obtaining feature vectors of 12 leads of the multiple groups of 12-lead electrocardiogram data samples through the pre-trained electrocardiogram feature extractor according to the multiple groups of 12-lead electrocardiogram data samples;

according to a preset training target, dividing feature vectors of N target leads and feature vectors of M leads to be mapped from feature vectors of 12 leads in the multiple groups of 12-lead electrocardiogram data samples;

and training the multilayer perceptron prediction model by taking the characteristic vectors of the N target leads of the multiple groups of 12-lead electrocardiogram data samples as input and taking the characteristic vectors of the M leads to be mapped of the multiple groups of 12-lead electrocardiogram data samples as output to obtain the characteristic vector mapper.

7. The method according to claim 6, wherein the set of multi-layered perceptron prediction models comprises M multi-layered perceptron prediction models, wherein the M multi-layered perceptron prediction models correspond to M leads of the 12-lead electrocardiographic data samples, respectively, to be mapped;

then, training the multilayer perceptron prediction model by taking the feature vectors of the N target leads of the multiple groups of 12-lead electrocardiograph data samples as input and taking the feature vectors of the M leads to be mapped of the multiple groups of 12-lead electrocardiograph data samples as output to obtain the feature vector mapper, including:

splicing the feature vectors of N target leads of the multiple groups of 12-lead electrocardiogram data samples to obtain N-lead feature vector combinations of the multiple groups of 12-lead electrocardiogram data samples;

and synchronously training the M multilayer perceptron prediction models by taking the N-lead characteristic vector combination of the multiple groups of 12-lead electrocardiogram data samples as input and taking the characteristic vector of each lead to be mapped of the multiple groups of 12-lead electrocardiogram data samples as output to obtain the characteristic vector mapper.

8. A low-lead electrocardiographic data processing apparatus, comprising:

the data acquisition module is used for acquiring the low-lead electrocardiogram data and determining M leads to be mapped of the low-lead electrocardiogram data according to the N leads of the low-lead electrocardiogram data;

the characteristic extraction module is used for carrying out characteristic extraction on N leads of the low-lead electrocardio data through a pre-trained electrocardio characteristic extractor to obtain an N-lead characteristic vector combination of the low-lead electrocardio data, wherein N and M are positive integers, and the sum of N and M is 12;

the characteristic mapping module is used for inputting the N lead characteristic vector combination of the low-lead electrocardio data into a pre-trained characteristic vector mapper to obtain M lead characteristic vectors to be mapped of the low-lead electrocardio data;

the characteristic splicing module is used for splicing the N lead characteristic vector combination of the low-lead electrocardiogram data and the M characteristic vectors of the leads to be mapped to obtain a 12 lead characteristic vector combination of the low-lead electrocardiogram data;

and the data processing module is used for inputting the 12-lead feature vector combination of the low-lead electrocardiogram data into a pre-trained electrocardiogram data processing model to obtain the classification result of the low-lead electrocardiogram data.

9. A storage medium having a computer program stored thereon, the computer program, when being executed by a processor, realizing the steps of the method of any one of claims 1 to 7.

10. A computer arrangement comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the computer program realizes the steps of the method of any one of claims 1 to 7 when executed by the processor.

Technical Field

The invention relates to the technical field of artificial intelligence technology and digital medical treatment, in particular to a method and a device for processing few-lead electrocardiogram data, a storage medium and computer equipment.

Background

Electrocardiography (ECG) is a technique for recording a pattern of changes in electrical activity generated in each cardiac cycle of the heart from the surface of the body of a human body by using an Electrocardiograph. By means of the electrocardiogram, various heart diseases of human can be represented, and doctors can judge the heart condition of patients according to the electrocardiogram. At present, a commonly used electrocardiogram generally consists of 12 leads, and these 12 leads of electrocardiographic data can comprehensively reflect the heart condition of a tested person, so the 12 leads of electrocardiographic data are often used as main training samples of a plurality of electrocardiographic data processing models.

At present, along with the popularization of wearable equipment, like wrist-watch, bracelet and implantation chip etc for electrocardio data acquisition's convenience has obtained very big promotion, and the user need not go the occasion of regulation again, just can anytime and anywhere the collection electrocardio data, and the heart situation of monitoring self that can be convenient. However, the portable electrocardiographic data acquisition device can only acquire single-lead electrocardiographic data or two-lead electrocardiographic data, and compared with complete 12-lead electrocardiographic data, the few-lead electrocardiographic data is difficult to reflect the actual condition of the heart comprehensively, so that the accuracy of an electrocardiographic data processing model constructed around the few-lead electrocardiographic data is insufficient, the classification performance is poor, and the classification accuracy of the few-lead electrocardiographic data is low.

Disclosure of Invention

In view of this, the present application provides a method, an apparatus, a storage medium, and a computer device for processing cardiac electrical data with fewer leads, and mainly aims to solve the technical problem of low classification accuracy of cardiac electrical data with fewer leads.

According to a first aspect of the present invention, there is provided a method for processing low-lead electrocardiographic data, the method comprising:

acquiring low-lead electrocardiogram data, and determining M leads to be mapped of the low-lead electrocardiogram data according to the N leads of the low-lead electrocardiogram data, wherein N and M are positive integers, and the sum of N and M is 12;

performing feature extraction on N leads of the low-lead electrocardiographic data through a pre-trained electrocardiographic feature extractor to obtain an N-lead feature vector combination of the low-lead electrocardiographic data;

inputting N lead eigenvector combination of the low-lead electrocardiogram data into a pre-trained eigenvector mapper to obtain M eigenvectors of leads to be mapped of the low-lead electrocardiogram data;

splicing the N-lead eigenvector combination of the low-lead electrocardiogram data and the M eigenvectors of leads to be mapped to obtain a 12-lead eigenvector combination of the low-lead electrocardiogram data;

and (3) combining and inputting the 12-lead feature vectors of the low-lead electrocardiogram data into a pre-trained electrocardiogram data processing model to obtain a classification result of the low-lead electrocardiogram data.

According to a second aspect of the present invention, there is provided a low-lead electrocardiographic data processing apparatus, comprising:

the data acquisition module is used for acquiring the low-lead electrocardiogram data and determining M leads to be mapped of the low-lead electrocardiogram data according to the N leads of the low-lead electrocardiogram data, wherein N and M are positive integers, and the sum of N and M is 12;

the characteristic extraction module is used for carrying out characteristic extraction on N leads of the low-lead electrocardio data through a pre-trained electrocardio characteristic extractor to obtain an N-lead characteristic vector combination of the low-lead electrocardio data;

the characteristic mapping module is used for inputting N lead characteristic vector combinations of the low-lead electrocardiogram data into a pre-trained characteristic vector mapper to obtain M characteristic vectors of leads to be mapped of the low-lead electrocardiogram data;

the characteristic splicing module is used for splicing the N-lead characteristic vector combination of the low-lead electrocardiogram data and the M characteristic vectors of leads to be mapped to obtain a 12-lead characteristic vector combination of the low-lead electrocardiogram data;

and the data processing module is used for inputting the 12-lead feature vector combination of the low-lead electrocardiogram data into the pre-trained electrocardiogram data processing model to obtain the classification result of the low-lead electrocardiogram data.

According to a third aspect of the present invention, there is provided a storage medium having stored thereon a computer program which, when executed by a processor, implements the method of low-lead electrocardiographic data processing described above.

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 on the memory and executable on the processor, the processor implementing the method for processing cardiac electrical data with fewer leads as described above when executing the program.

The invention provides a few-lead electrocardiogram data processing method, a device, a storage medium and a computer device, which are characterized in that firstly, few-lead electrocardiogram data to be processed are obtained, then, an electrocardiogram feature extractor is utilized to extract N lead feature vectors of the few-lead electrocardiogram data, then, M lead feature vectors to be mapped of the few-lead electrocardiogram data are mapped through a feature vector mapper according to the N lead feature vectors, finally, the N lead feature vectors and the M lead feature vectors to be mapped are spliced into a 12 lead feature vector combination, and a classification result of the few-lead electrocardiogram data is obtained according to the 12 lead feature vector combination. The method fully utilizes the incidence relation among all leads in the electrocardiogram data, deduces 12-lead electrocardiogram data through N-lead electrocardiogram data mapping of the few-lead electrocardiogram data, and obtains the classification result of the electrocardiogram data through the 12-lead electrocardiogram data. The method effectively enriches the lead information of the low-lead electrocardiogram data and improves the classification accuracy of the low-lead electrocardiogram data.

The foregoing description is only an overview of the technical solutions of the present application, and the present application can be implemented according to the content of the description in order to make the technical means of the present application more clearly understood, and the following detailed description of the present application is given in order to make the above and other objects, features, and advantages of the present application more clearly understandable.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:

FIG. 1 is a flow chart of a method for processing low-lead electrocardiographic data according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of a few-lead electrocardiograph data processing apparatus according to an embodiment of the present invention.

Detailed Description

The invention will be described in detail hereinafter with reference to the accompanying drawings in conjunction with embodiments. It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.

At present, most of the electrocardiogram data processing models are obtained by 12-lead electrocardiogram data training, and a small part of the electrocardiogram data models obtained by the few-lead electrocardiogram data training obviously have the problems of inaccurate classification and less label coverage. However, the 12-lead electrocardiographic data is easily limited by the instruments and the application places in the collection mode, so that it is difficult to monitor at any time, and the portable device capable of monitoring the electrocardiographic data at any time can only collect the single-lead or dual-lead electrocardiographic data with few leads, which results in that the processing result of the few-lead data actually measured by the portable device hardly meets the expected requirement.

In an embodiment, as shown in fig. 1, a method for processing electrocardiographic data with few leads is provided, which is described by taking the method as an example when the method is applied to a server and a computer device such as an electrocardiographic acquisition device, where the server may be an independent server, or a cloud server that provides basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a middleware service, a domain name service, a security service, a Content Delivery Network (CDN), a big data and artificial intelligence platform, and the like. The method comprises the following steps:

101. acquiring the low-lead electrocardiogram data, and determining M leads to be mapped of the low-lead electrocardiogram data according to the N leads of the low-lead electrocardiogram data, wherein N and M are positive integers, and the sum of N and M is 12.

The electrocardiographic data refers to a graph which is acquired from the body surface of a human body through electrocardiographic equipment and records the change of electrical activity generated by each cardiac cycle of the heart, and generally, the complete electrocardiographic data usually comprises 12 leads of electrocardiographic data. On the basis, the low-lead electrocardiogram data only comprises part of 12 leads. The few-lead electrocardiogram data can be collected through intelligent collecting equipment such as an intelligent bracelet, a watch, clothes, an implanted chip and the like, the collected few-lead electrocardiogram data comprises at least one lead electrocardiogram data, and each lead electrocardiogram data consists of a series of electrocardiogram waveforms. In these electrocardiographic waveforms, a great deal of feature information is contained, and the feature information can be extracted by some deep learning models.

Specifically, the computer device may acquire the to-be-processed low-lead electrocardiographic data through the data interface, and then may compare N leads included in the acquired low-lead electrocardiographic data with the complete 12-lead electrocardiographic data to determine the missing leads in the low-lead electrocardiographic data, where the missing leads are to-be-mapped leads of the low-lead electrocardiographic data, and the number of the to-be-mapped leads of the low-lead electrocardiographic data is M. In this embodiment, N and M are both positive integers, and the sum of N and M is 12. For example, the N leads of the low-lead electrocardiographic data are lead 1 and lead 2, and then the leads to be mapped of the low-lead electrocardiographic data are leads 3 to 12.

102. And performing feature extraction on N leads of the low-lead electrocardiogram data through a pre-trained electrocardiogram feature extractor to obtain an N-lead feature vector combination of the low-lead electrocardiogram data.

Specifically, the computer device can input the N-lead electrocardiographic data of the low-lead electrocardiographic data into each deep learning model of the trained electrocardiographic feature extractor, so as to obtain N-lead feature vectors of the low-lead electrocardiographic data, and then the computer device can splice the N-lead feature vectors of the low-lead electrocardiographic data, so as to obtain an N-lead feature vector combination of the low-lead electrocardiographic data.

In this embodiment, the feature vectors of the leads of the low-lead electrocardiographic data can be extracted through some deep learning models. For example, the computer device may construct a set of convolutional neural network models through some electrocardiographic data samples, where the number of the set of convolutional neural network models is the same as the number of leads of the electrocardiographic data samples, that is, each convolutional neural network may correspond to one lead of electrocardiographic data, and may convert the corresponding lead of electrocardiographic data into a feature vector, and the trained convolutional neural network is the electrocardiographic feature extractor. It can be understood that the electrocardiographic feature extractor at least comprises N deep learning models, and the N deep learning models are respectively in one-to-one correspondence with the N leads of the low-lead electrocardiographic data.

103. And inputting the N lead feature vector combination of the low-lead electrocardiogram data into a pre-trained feature vector mapper to obtain M lead feature vectors of the low-lead electrocardiogram data to be mapped.

Specifically, after obtaining the N-lead feature vector combination of the low-lead electrocardiographic data, the computer device may input the N-lead feature vector combination of the low-lead electrocardiographic data to each feature prediction model of the trained feature vector mapper, so as to obtain M feature vectors of the low-lead electrocardiographic data to be mapped.

In the present embodiment, the feature vectors of the leads (i.e. the leads to be mapped) missing from the low-lead electrocardiographic data can be predicted by some neural network models with complex computing power. For example, the computer device may construct a group of multilayer perceptron models by using 12-lead electrocardiographic data samples, where the number of the group of multilayer perceptron models may be the same as the number of leads to be mapped of the few-lead electrocardiographic data, that is, each multilayer perceptron model may correspond to electrocardiographic data of one lead to be mapped, and may predict a feature vector of the lead to be mapped by using a feature vector of a known lead, and the group of trained multilayer perceptron models is a feature vector mapper. It can be understood that the feature vector mapper at least includes M neural network models, and the M neural network models respectively correspond to M leads of the low-lead electrocardiographic data to be mapped one by one.

104. And splicing the N-lead eigenvector combination of the low-lead electrocardiogram data and the M eigenvectors of the leads to be mapped to obtain a 12-lead eigenvector combination of the low-lead electrocardiogram data.

Specifically, after obtaining the N-lead eigenvector combination of the low-lead electrocardiographic data and the M eigenvectors of the leads to be mapped, the computer device may splice the N-lead eigenvector combination of the low-lead electrocardiographic data and the M eigenvectors of the leads to be mapped, thereby obtaining the 12-lead eigenvector combination of the low-lead electrocardiographic data. In this embodiment, the 12-lead eigenvector combination includes both the original lead information and the missing lead information in the low-lead electrocardiographic data, so that the information coverage of the 12-lead eigenvector combination is much richer than that of the low-lead electrocardiographic data before processing, and further, the classification result obtained by using the 12-lead eigenvector combination is more accurate than that obtained by directly using the low-lead electrocardiographic data, and the coverage range of the classification label is wider.

105. And (3) combining and inputting the 12-lead feature vectors of the low-lead electrocardiogram data into a pre-trained electrocardiogram data processing model to obtain a classification result of the low-lead electrocardiogram data.

Specifically, the computer device may combine and input the spliced 12-lead feature vectors of the low-lead electrocardiographic data into the pre-trained electrocardiographic data processing model to obtain the classification result of the low-lead electrocardiographic data, i.e., obtain the name of the disease corresponding to the low-lead electrocardiographic data or the prediction probability of a certain disease corresponding to the low-lead electrocardiographic data. In this embodiment, the electrocardiographic data processing model may be trained in advance by using 12-lead electrocardiographic data samples with classification labels, where the classification labels of the electrocardiographic data samples may be names of diseases corresponding to the electrocardiographic data samples, or prediction probabilities of certain diseases corresponding to the electrocardiographic data samples, and the implementation is not specifically described here.

The method for processing the few-lead electrocardiographic data includes the steps of firstly obtaining the few-lead electrocardiographic data to be processed, then extracting feature vectors of N leads of the few-lead electrocardiographic data by using an electrocardiographic feature extractor, then mapping out M feature vectors of the few-lead electrocardiographic data to be mapped according to the feature vectors of the N leads by using a feature vector mapper, finally splicing the feature vectors of the N leads and the feature vectors of the M leads to be mapped into a 12-lead feature vector combination, and obtaining a classification result of the few-lead electrocardiographic data according to the result. The method fully utilizes the incidence relation among all leads in the electrocardiogram data, deduces 12-lead electrocardiogram data through N-lead electrocardiogram data mapping of the few-lead electrocardiogram data, and obtains the classification result of the electrocardiogram data through the 12-lead electrocardiogram data. The method effectively enriches the lead information of the low-lead electrocardiogram data and improves the classification accuracy of the low-lead electrocardiogram data.

In one embodiment, the ecg feature extractor of step 102 includes at least N trained convolutional neural network models, where the N convolutional neural network models correspond to N leads of the low-lead ecg data, respectively. Based on this, step 102 can be specifically realized by the following method: firstly, respectively inputting N leads of the low-lead electrocardiogram data into N convolutional neural network models of an electrocardiogram feature extractor to obtain characteristic vectors of the N leads of the low-lead electrocardiogram data, and then splicing the characteristic vectors of the N leads of the low-lead electrocardiogram data to obtain an N-lead characteristic vector combination of the low-lead electrocardiogram data. In this embodiment, when splicing the N-lead eigenvector combination of the low-lead electrocardiographic data, the N eigenvectors of the low-lead electrocardiographic data may be first separated, and then the separated plural eigenvectors may be spliced. For example, the [ SEP ] can be used as a separator to separate the eigenvectors of the N leads, then the [ CLS ] separator is spliced at the beginning, and finally the eigenvectors of the N leads after separation are spliced together to obtain the N-lead eigenvector combination of the electrocardiographic data with few leads. Wherein, the [ SEP ] separator can separate the feature vectors of N leads of the electrocardio data with few leads, and the [ CLS ] separator can facilitate the extraction of feature vector combination. In the embodiment, the feature vector combination of the few-lead electrocardiogram data is obtained in a splicing mode, and the generation efficiency and the extraction efficiency of the N-lead feature vector combination can be improved. It can be understood that the electrocardiogram feature extractor may also include 12 trained convolutional neural network models, and then, when in application, N of the convolutional neural network models are selected for feature extraction.

In one embodiment, the feature vector mapper in step 103 includes at least M trained multi-layer perceptron prediction models, where the M multi-layer perceptron prediction models respectively correspond to M leads to be mapped of the low-lead electrocardiographic data, that is, each multi-layer perceptron prediction model can predict a feature vector of a specific lead based on the feature vectors of existing leads. Based on this, step 103 can be specifically realized by the following method: and respectively inputting the N lead eigenvector combination of the low-lead electrocardiogram data into M multilayer perceptron prediction models of an eigenvector mapper to obtain M eigenvectors of leads to be mapped of the low-lead electrocardiogram data. For example, if N leads of the few-lead electrocardiographic data are lead 1 and lead 2, then the feature vector combinations of lead 1 and lead 2 of the few-lead electrocardiographic data are input into the multi-layer perceptron prediction models corresponding to lead 3 to lead 12, so that the feature vectors of lead 3 to lead 12 of the few-lead electrocardiographic data can be obtained. It is understood that the feature vector mapper may also include 12 trained multi-layer perceptron prediction models, and then select M multi-layer perceptron prediction models for feature mapping when applying the feature vector mapper.

In an embodiment, the electrocardiographic feature extractor in step 102 and the electrocardiographic data processing model in step 105 may be specifically trained by the following method: firstly, obtaining a plurality of groups of 12-lead electrocardiogram data samples with classification labels, constructing a group of convolutional neural network models and a multi-layer perceptron classification model according to the plurality of groups of 12-lead electrocardiogram data samples, then obtaining feature vector combinations of the plurality of groups of 12-lead electrocardiogram data samples through the convolutional neural network models according to the plurality of groups of 12-lead electrocardiogram data samples, finally taking the feature vector combinations of the plurality of groups of 12-lead electrocardiogram data samples as input, taking the classification labels of the plurality of groups of 12-lead electrocardiogram data samples as output, and carrying out synchronous training on the convolutional neural network models and the multi-layer perceptron classification model to obtain an electrocardiogram feature extractor and an electrocardiogram data processing model. In this embodiment, the electrocardiographic data processing model is trained based on 12-lead electrocardiographic data samples, so that the performance of the model is more accurate than that of a model obtained by training a few-lead electrocardiographic data sample, and the coverage range of the classification note is wider. Furthermore, a more accurate classification result can be obtained by inputting the 12-lead electrocardiogram data combination obtained by mapping the few-lead electrocardiogram data into the electrocardiogram data processing model.

In an embodiment, the set of convolutional neural network models required for the training of the electrocardiograph feature extractor may specifically include 12 convolutional neural network models, where the 12 convolutional neural network models correspond to 12 leads of the 12-lead electrocardiograph data samples, respectively. Based on the method, the splicing method of the feature vector combination of the 12-lead electrocardiogram data sample comprises the following steps: firstly, 12-lead electrocardiogram data of a plurality of groups of 12-lead electrocardiogram data samples are respectively input into 12 convolutional neural network models to obtain 12 characteristic vectors of the plurality of groups of 12-lead electrocardiogram data samples, and then the 12 characteristic vectors of each group of 12-lead electrocardiogram data samples are respectively spliced to obtain characteristic vector combinations of the plurality of groups of 12-lead electrocardiogram data samples. In this embodiment, the splicing manner of the feature vector combination of the 12-lead electrocardiographic data sample may refer to the splicing manner of the feature vector combination of the few-lead electrocardiographic data, which is not described herein again.

In one embodiment, the feature vector mapper in step 103 may be specifically trained by: firstly, obtaining a plurality of groups of 12-lead electrocardiogram data samples, constructing a group of multilayer perceptron prediction models according to the plurality of groups of 12-lead electrocardiogram data samples, then obtaining characteristic vectors of 12 leads of the plurality of groups of 12-lead electrocardiogram data samples through a pre-trained electrocardiogram characteristic extractor according to the plurality of groups of 12-lead electrocardiogram data samples, further dividing characteristic vectors of N target leads and characteristic vectors of M leads to be mapped from the characteristic vectors of 12 leads of the plurality of groups of 12-lead electrocardiogram data samples according to a preset training target, finally taking the characteristic vectors of N target leads of the plurality of groups of 12-lead electrocardiogram data samples as input, taking the characteristic vectors of M leads to be mapped of the plurality of groups of 12-lead electrocardiogram data samples as output, training the multilayer perceptron prediction models, and obtaining a characteristic vector mapper. In this embodiment, N target leads of the 12-lead electrocardiographic data sample correspond to N original leads of the low-lead electrocardiographic data, and M leads to be mapped of the 12-lead electrocardiographic data sample correspond to M missing leads of the low-lead electrocardiographic data. It can be understood that the division mode of the 12-lead electrocardiographic data sample can be flexibly adjusted according to a set training target, and the input data and the output data of model training can also be flexibly adjusted according to the set training target, so that the feature vector mapper provided by the embodiment can be applied to any feature mapping scene of the few-lead electrocardiographic data, and the application range is very wide.

In one embodiment, the set of multi-layered perceptron prediction models required for the training of the feature vector mapper includes M multi-layered perceptron prediction models, where the M multi-layered perceptron prediction models correspond to M leads of the 12-lead electrocardiographic data samples to be mapped, respectively. Based on the above, the method for training the multi-layer perceptron prediction model is as follows: firstly, splicing the characteristic vectors of N target leads of a plurality of groups of 12-lead electrocardiogram data samples to obtain N-lead characteristic vector combinations of the plurality of groups of 12-lead electrocardiogram data samples, then taking the N-lead characteristic vector combinations of the plurality of groups of 12-lead electrocardiogram data samples as input, respectively taking the characteristic vector of each lead to be mapped of the plurality of groups of 12-lead electrocardiogram data samples as output, and synchronously training M multilayer perceptron prediction models to obtain a characteristic vector mapper. In this embodiment, the splicing manner of the feature vector combination of the 12-lead electrocardiographic data sample may refer to the splicing manner of the feature vector combination of the few-lead electrocardiographic data, which is not described herein again.

Further, as a refinement and an extension of the specific implementation of the above embodiment, in order to fully describe the implementation process of the embodiment, the embodiment provides a method for processing electrocardiographic data with fewer leads, which first trains an electrocardiographic data processing model and a feature vector mapping model between leads according to a large amount of 12-lead electrocardiographic data, then uses the feature vector mapping model of each lead to estimate electrocardiographic data of all leads through the electrocardiographic data with fewer leads, and finally performs classification processing on the estimated electrocardiographic data with 12 leads to obtain a classification result of the electrocardiographic data with fewer leads. Specifically, the method comprises the following steps:

201. acquiring multiple groups of 12-lead electrocardiogram data D_aAnd each group of 12-lead electrocardiographic data D_aCorresponding class label Y_a。

202. And aiming at each lead in each group of the electrocardiographic data, obtaining a characteristic vector (embedding) of each lead in each group of the electrocardiographic data by utilizing a one-dimensional convolutional neural network model (CNN model).

203. Splicing the imbedding of each lead of each group of electrocardiogram data to obtain a feature vector combination of each group of electrocardiogram data, inputting the feature vector combination of each group of electrocardiogram data into an MLP (multi-layer perceptron classification) model, and utilizing a classification label Y of each group of electrocardiogram data_aTraining the MLP model to obtain an electrocardiogram data processing model (denoted as M)_d) (ii) a And meanwhile, adjusting parameters of the CNN model to obtain the trained CNN model.

204. Constructing 12-n MLP models (multilayer perceptron prediction models), obtaining a feature vector combination of each group of electrocardiogram data by using the trained CNN models, then respectively inputting the feature vector combination of each group of electrocardiogram data into the 12-n MLP models, and synchronously training the 12-n MLP models by using the feature vector of each lead of each group of electrocardiogram data to obtain a feature vector mapping model (denoted as M)_m)。

205. Inputting n-lead electrocardiogram data (n is less than 12) of the low-lead electrocardiogram data to be processed into the trained CNN model to obtain n-lead eigenvectors of the low-lead electrocardiogram data.

206. Splicing the n lead eigenvectors of the few-lead electrocardiogram data and respectively inputting the spliced n lead eigenvectors into 12-n M leads_mAnd obtaining 12-n mapping feature vectors. For example, if the low-lead electrocardiographic data has lead 1 and lead 2, then the feature vector combinations of lead 1 and lead 2 of the low-lead electrocardiographic data are respectively input into 12-n M_mAnd obtaining embedding of mapping leads 3-12 of the low-lead electrocardiogram data.

207. Splicing n eigenvectors of the low-lead electrocardiographic data and 12-n mapping eigenvectors to obtain a 12-lead eigenvector combination of the low-lead electrocardiographic data, and finally inputting the 12-lead eigenvector combination of the low-lead electrocardiographic data into M_dAnd obtaining the classification result of the few-lead electrocardiogram data.

The method comprises the steps of firstly obtaining a large amount of 12-lead electrocardiogram data, training the data to obtain an electrocardiogram data processing model and a characteristic vector model, then deducing the electrocardiogram data of unknown leads by using the existing electrocardiogram data of the leads through the characteristic vector model to obtain all 12-lead electrocardiogram data, and finally obtaining a final classification result through an electrocardiogram diagnosis model according to the 12-lead electrocardiogram data. The method fully utilizes the incidence relation among all leads in the electrocardiogram data, solves the mapping problem between the few leads and the 12 leads, obtains an accurate electrocardiogram data classification result by utilizing the electrocardiogram data model constructed based on the electrocardiogram data with 12 leads, effectively improves the classification accuracy of the few-lead electrocardiogram data, and in addition, the method also improves the coverage range of classification labels through abundant lead information in the electrocardiogram data processing model.

Further, as a refinement and an extension of the specific implementation of the foregoing embodiment, in order to fully illustrate the implementation process of this embodiment, this embodiment provides a method for processing electrocardiographic data with fewer leads, which includes the following steps:

further, as a specific implementation of the method shown in fig. 1, the present embodiment provides a few-lead electrocardiographic data processing apparatus, as shown in fig. 2, the apparatus includes: the system comprises a data acquisition module 21, a feature extraction module 22, a feature mapping module 23, a feature splicing module 24 and a data processing module 25, wherein:

the data acquisition module 21 is configured to acquire the low-lead electrocardiographic data, and determine M leads to be mapped of the low-lead electrocardiographic data according to N leads of the low-lead electrocardiographic data, where N and M are positive integers, and a sum of N and M is 12;

the feature extraction module 22 is configured to perform feature extraction on the N leads of the low-lead electrocardiographic data by using a pre-trained electrocardiographic feature extractor to obtain an N-lead feature vector combination of the low-lead electrocardiographic data;

the feature mapping module 23 is configured to input N-lead feature vector combinations of the low-lead electrocardiographic data into a pre-trained feature vector mapper to obtain M feature vectors of leads to be mapped of the low-lead electrocardiographic data;

the feature splicing module 24 is configured to splice N-lead feature vector combinations of the low-lead electrocardiographic data and M feature vectors of leads to be mapped to obtain 12-lead feature vector combinations of the low-lead electrocardiographic data;

the data processing module 25 is configured to combine and input the 12-lead feature vectors of the low-lead electrocardiographic data into the pre-trained electrocardiographic data processing model to obtain a classification result of the low-lead electrocardiographic data.

In a specific application scenario, the electrocardio-feature extractor comprises at least N trained convolutional neural network models, wherein the N convolutional neural network models correspond to N leads of the low-lead electrocardio data respectively; the feature extraction module 22 is specifically configured to input the N leads of the low-lead electrocardiographic data into the N convolutional neural network models of the electrocardiographic feature extractor, respectively, to obtain feature vectors of the N leads of the low-lead electrocardiographic data; and splicing the N lead eigenvectors of the low-lead electrocardiogram data to obtain the N lead eigenvector combination of the low-lead electrocardiogram data.

In a specific application scenario, the feature vector mapper comprises at least M trained multilayer perceptron prediction models, wherein the M multilayer perceptron prediction models respectively correspond to M leads to be mapped of the low-lead electrocardiogram data; the feature mapping module 23 is specifically configured to input N-lead feature vector combinations of the low-lead electrocardiographic data into M multi-layer perceptron prediction models of the feature vector mapper, respectively, to obtain M feature vectors of leads to be mapped of the low-lead electrocardiographic data.

In a specific application scenario, the device further comprises a model training module 26, wherein the model training module 26 is specifically configured to obtain multiple groups of 12-lead electrocardiographic data samples, and construct a group of convolutional neural network models and a multi-layer perceptron classification model according to the multiple groups of 12-lead electrocardiographic data samples; obtaining feature vector combinations of the multiple groups of 12-lead electrocardiogram data samples through a convolutional neural network model according to the multiple groups of 12-lead electrocardiogram data samples; and synchronously training the convolutional neural network model and the multi-layer perceptron classification model by taking the characteristic vector combination of the multiple groups of 12-lead electrocardiogram data samples as input and the classification labels of the multiple groups of 12-lead electrocardiogram data samples as output to obtain the electrocardiogram characteristic extractor and the electrocardiogram data processing model.

In a specific application scenario, a group of convolutional neural network models comprises 12 convolutional neural network models, wherein the 12 convolutional neural network models respectively correspond to 12 leads of 12-lead electrocardiogram data samples; the model training module 26 may be further configured to input the 12-lead electrocardiographic data of the multiple sets of 12-lead electrocardiographic data samples into the 12 convolutional neural network models, respectively, to obtain 12 feature vectors of the multiple sets of 12-lead electrocardiographic data samples; and respectively splicing the 12 eigenvectors of each group of 12-lead electrocardiogram data samples to obtain the eigenvector combination of the multiple groups of 12-lead electrocardiogram data samples.

In a specific application scenario, the model training module 26 may be further configured to obtain multiple groups of 12-lead electrocardiographic data samples, and construct a group of multilayer perceptron prediction models according to the multiple groups of 12-lead electrocardiographic data samples; obtaining feature vectors of 12 leads of the multiple groups of 12-lead electrocardiogram data samples through a pre-trained electrocardiogram feature extractor according to the multiple groups of 12-lead electrocardiogram data samples; according to a preset training target, dividing feature vectors of N target leads and feature vectors of M leads to be mapped from feature vectors of 12 leads of a plurality of groups of 12-lead electrocardiogram data samples; and training a multilayer perceptron prediction model by taking the characteristic vectors of N target leads of the multiple groups of 12-lead electrocardiogram data samples as input and taking the characteristic vectors of M leads to be mapped of the multiple groups of 12-lead electrocardiogram data samples as output to obtain a characteristic vector mapper.

In a specific application scenario, a group of multi-layer perceptron prediction models comprises M multi-layer perceptron prediction models, wherein the M multi-layer perceptron prediction models respectively correspond to M leads to be mapped of 12-lead electrocardiogram data samples; the model training module 26 may be further configured to splice the feature vectors of the N target leads of the multiple groups of 12-lead electrocardiographic data samples to obtain N-lead feature vector combinations of the multiple groups of 12-lead electrocardiographic data samples; n lead eigenvector combinations of a plurality of groups of 12 lead electrocardiographic data samples are used as input, eigenvectors of each lead to be mapped of the plurality of groups of 12 lead electrocardiographic data samples are respectively used as output, and M multilayer perceptron prediction models are synchronously trained to obtain an eigenvector mapper.

It should be noted that other corresponding descriptions of the functional units related to the few-lead electrocardiograph data processing apparatus provided in this embodiment may refer to the corresponding description in fig. 1, and are not described herein again.

Based on the method shown in fig. 1, correspondingly, the present embodiment further provides a storage medium, on which a computer program is stored, and the program, when executed by a processor, implements the method for processing cardiac electrical data with few leads shown in fig. 1.

Based on such understanding, the technical solution of the present application may be embodied in the form of a software product, and the software product to be identified may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, or the like), and include several instructions for enabling a computer device (which may be a personal computer, a server, or a network device) to execute the method according to the implementation scenarios of the present application.

Based on the method shown in fig. 1 and the embodiment of the small-lead electrocardiographic data processing apparatus shown in fig. 2, in order to achieve the above object, this embodiment further provides an entity device for small-lead electrocardiographic data processing, which may specifically be a personal computer, a server, a smart phone, a tablet computer, a smart watch, or other network devices, and the entity device includes a storage medium and a processor; a storage medium for storing a computer program; a processor for executing a computer program for implementing the above-described method as shown in fig. 1.

Optionally, the entity device may further include a user interface, a network interface, a camera, a Radio Frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, and the like. The user interface may include a Display screen (Display), an input unit such as a keypad (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 (e.g., WI-FI interface), etc.

Those skilled in the art will appreciate that the physical device structure for processing the cardiac electrical data with few leads provided by the present embodiment does not constitute a limitation to the physical device, and may include more or more N components, or combine some components, or arrange different components.

The storage medium may further include an operating system and a network communication module. The operating system is a program for managing the hardware of the above-mentioned entity 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 for realizing communication among components in the storage medium and communication with other hardware and software in the information processing entity device.

Through the above description of the embodiments, those skilled in the art will clearly understand that the present application can be implemented by software plus a necessary general hardware platform, and can also be implemented by hardware. By applying the technical scheme of the application, firstly, the low-lead electrocardiographic data to be processed is obtained, then the characteristic vectors of N leads of the low-lead electrocardiographic data are extracted by the electrocardiographic feature extractor, then, the characteristic vectors of M leads to be mapped of the low-lead electrocardiographic data are mapped through the characteristic vector mapper according to the characteristic vectors of the N leads, finally, the characteristic vectors of the N leads and the characteristic vectors of the M leads to be mapped are spliced into a 12-lead characteristic vector combination, and the classification result of the low-lead electrocardiographic data is obtained. Compared with the prior art, the method fully utilizes the incidence relation among all leads in the electrocardiogram data, deduces 12-lead electrocardiogram data through N-lead electrocardiogram data mapping of the few-lead electrocardiogram data, and obtains the classification result of the electrocardiogram data through the 12-lead electrocardiogram data. The method effectively enriches the lead information of the low-lead electrocardiogram data and improves the classification accuracy of the low-lead electrocardiogram data.

Those skilled in the art will appreciate that the figures are merely schematic representations of one preferred implementation scenario and that the blocks or flow diagrams in the figures are not necessarily required to practice the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario may be distributed in the devices in the implementation scenario according to the description of the implementation scenario, or may be located in one or more devices different from the present implementation scenario with corresponding changes. The modules of the implementation scenario may be combined into one module, or may be further split into a plurality of sub-modules.

The above application serial numbers are for description purposes only and do not represent the superiority or inferiority of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of the present application, but the present application is not limited thereto, and any variations that can be made by those skilled in the art are intended to fall within the scope of the present application.