WO2023165005A1

WO2023165005A1 - Multi-lead elctrocardiogram signal processing method, device, apparatus, and storage medium

Info

Publication number: WO2023165005A1
Application number: PCT/CN2022/089174
Authority: WO
Inventors: 张楠; 王健宗; 瞿晓阳
Original assignee: 平安科技（深圳）有限公司
Priority date: 2022-03-04
Filing date: 2022-04-26
Publication date: 2023-09-07
Also published as: CN114711780A

Abstract

A multi-lead electrocardiogram signal processing method, device, apparatus, and storage medium used for improving the accuracy and richness of electrocardiosignal extraction. The multi-lead electrocardiogram signal processing method comprises: acquiring a to-be-processed multi-lead electrocardiogram signal, the to-be-processed multi-lead electrocardiogram signal being used for indicating heart detection information for a target object; performing data preprocessing on the to-be-processed multi-lead electrocardiogram signal to obtain processed electrocardiogram data; performing data framing processing on the processed electrocardiogram data to obtain multi-dimensional lead electrocardiogram data; and performing feature extraction and feature aggregation processing on the multi-dimensional lead electrocardiogram data by means of a deep neural network model incorporating a dual attention mechanism to obtain target electrocardiogram feature data.

Description

Multi-lead electrocardiogram signal processing method, device, equipment and storage medium

This application claims the priority of the Chinese patent application submitted to the China Patent Office on March 4, 2022, with the application number 202210218648.0, and the title of the invention is "multi-lead electrocardiogram signal processing method, device, equipment and storage medium", the entire content of which Incorporated in the application by reference.

technical field

The present application relates to the technical field of neural networks, and in particular to a multi-lead electrocardiogram signal processing method, device, equipment and storage medium.

Background technique

Heart disease is one of the main culprits threatening our health, and electrocardiogram is an important method for heart disease detection. Electrocardiogram is a technique in which electrocardiogram machines record the electrical activity changes generated by each cardiac cycle of the heart from the body surface. The electrocardiogram shows the health status of the heart rate and detects the abnormal heart rate of ordinary users through the electrocardiogram.

In the field of medical artificial intelligence, automatic ECG analysis methods mainly include manual feature extraction of typical waveforms and bands such as p waves and qrs waves, as well as feature extraction and ECG data classification of some deep learning classification networks. At present, most of the convolutional network (CNN) is used to train multi-lead ECG data to realize automatic analysis of ECG. The inventor realized that the convolutional layer in CNN is limited by the receptive field, which leads to the limitation of long signal context information, and ignores the The channel correlation among multiple channels of ECG signal leads to low accuracy of ECG signal extraction.

Contents of the invention

The present application provides a multi-lead electrocardiogram signal processing method, device, equipment and storage medium for improving the accuracy and richness of electrocardiogram signal extraction.

The first aspect of the present application provides a multi-lead electrocardiogram signal processing method, including: acquiring the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object; Perform data preprocessing on the multi-lead ECG signal to be processed to obtain processed ECG data; perform data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data; The deep neural network model of the mechanism performs feature extraction and feature aggregation processing on the ECG data of the multi-dimensional lead channel to obtain target ECG feature data.

The second aspect of the present application provides a multi-lead electrocardiogram signal processing device, including a memory, a processor, and computer-readable instructions stored in the memory and operable on the processor, and the processor executes the When the computer-readable instructions are described, the following steps are implemented: obtaining the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object; the multi-lead electrocardiogram signal to be processed Perform data preprocessing on the signal to obtain processed ECG data; perform data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data; integrate the deep neural network model of the dual attention mechanism to the described The multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing to obtain target ECG feature data.

The third aspect of the present application provides a computer-readable storage medium, where computer instructions are stored in the computer-readable storage medium, and when the computer instructions are run on the computer, the computer is made to perform the following steps: obtaining the multi- Lead electrocardiogram signals, the multi-lead electrocardiogram signals to be processed are used to indicate the heart detection information of the target object; data preprocessing is performed on the multi-lead electrocardiogram signals to be processed to obtain processed electrocardiogram data; The processed electrocardiogram data is subjected to data frame processing to obtain multi-dimensional lead channel ECG data; the multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing through a deep neural network model fused with a dual attention mechanism, Obtain target ECG characteristic data.

The fourth aspect of the present application provides a multi-lead electrocardiogram signal processing device, wherein the multi-lead electrocardiogram signal processing device includes: an acquisition module, configured to acquire multi-lead electrocardiogram signals to be processed, and the to-be-processed The multi-lead electrocardiogram signal is used to indicate the heart detection information of the target object; the preprocessing module is used to perform data preprocessing on the multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data; the framing module is used to Carry out data frame processing on the processed electrocardiogram data to obtain multi-dimensional lead channel ECG data; the aggregation module is used to process the multi-dimensional lead channel ECG data by integrating a deep neural network model of a dual attention mechanism Feature extraction and feature aggregation processing to obtain target ECG feature data.

In the technical solution provided by the present application, the multi-lead electrocardiogram signal to be processed is obtained, and the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object; the multi-lead electrocardiogram signal to be processed is Data preprocessing to obtain processed electrocardiogram data; data frame processing is performed on the processed electrocardiogram data to obtain multidimensional lead channel ECG data; the multidimensional guide channel is processed by a deep neural network model that integrates a dual attention mechanism. Feature extraction and feature aggregation processing are performed on the joint-channel ECG data to obtain target ECG feature data. In the embodiment of the present application, the feature extraction and feature aggregation processing are performed on the multi-dimensional lead channel ECG data through the deep neural network model fused with the dual attention mechanism to obtain the target ECG feature data, that is, through two different attention mechanisms Integrate feature information of different dimensions to achieve extended context information. Among them, the dependencies between all positions of the spatial feature map are calculated through the global deep attention mechanism, which expands the receptive field of the architecture; while the cross-channel attention mechanism captures feature information between different channels. The features of these two attention mechanisms are finally aggregated to further improve feature representations that help enrich contextual information.

Description of drawings

Fig. 1 is a schematic diagram of an embodiment of a multi-lead electrocardiogram signal processing method in the embodiment of the present application;

2 is a schematic diagram of another embodiment of a multi-lead electrocardiogram signal processing method in the embodiment of the present application;

Fig. 3 is a schematic diagram of an embodiment of a multi-lead electrocardiogram signal processing device in the embodiment of the present application;

4 is a schematic diagram of another embodiment of a multi-lead electrocardiogram signal processing device in the embodiment of the present application;

Fig. 5 is a schematic diagram of an embodiment of a multi-lead electrocardiogram signal processing device in the embodiment of the present application.

Detailed ways

The embodiment of the present application provides a multi-lead electrocardiogram signal processing method, device, device and storage medium, which are used for feature extraction and processing of multi-dimensional lead channel electrocardiogram data through a deep neural network model fused with a dual attention mechanism. Feature aggregation processing to obtain target ECG feature data, that is, to integrate feature information of different dimensions through two different attention mechanisms to achieve extended context information and further improve feature representation that helps enrich context information.

The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and not necessarily Used to describe a specific sequence or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments described herein can be practiced in sequences other than those illustrated or described herein. Furthermore, the term "comprising" or "having" and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or device comprising a sequence of steps or elements is not necessarily limited to those explicitly listed instead, may include other steps or elements not explicitly listed or inherent to the process, method, product or apparatus.

For ease of understanding, the following describes the specific process of the embodiment of the present application. Please refer to FIG. 1. An embodiment of the multi-lead ECG signal processing method in the embodiment of the present application includes:

101. Acquire a multi-lead electrocardiogram signal to be processed, where the multi-lead electrocardiogram signal to be processed is used to indicate heart detection information of a target object.

Wherein, the heart detection information of the target object may include the case diagnosis information of the target object, and may also be normal detection information, and the multi-lead ECG signal to be processed includes one or more set bands of the heartbeat cycle. Specifically, the server receives the electrocardiogram data processing request sent by the target terminal, and extracts the multi-lead electrocardiogram signal to be processed from the electrocardiogram data processing request; the server stores the data of the multi-lead electrocardiogram signal to be processed, for example, the server can The processed multi-lead ECG signals are stored in a preset type of file, or the multi-lead ECG signals to be processed are stored in a preset memory database (redis, a remote dictionary service).

Further, the server queries the preset queue table to obtain the query result. When the query result is not empty, the server extracts the multi-lead ECG signal to be processed from the query result, and the multi-lead ECG signal to be processed is used to indicate Heart detection information of the target object.

It can be understood that the subject of execution of the present application may be a multi-lead electrocardiogram signal processing device, and may also be a terminal or a server, which is not specifically limited here. The embodiment of the present application is described by taking the server as an execution subject as an example.

102. Perform data preprocessing on the multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data.

It should be noted that more than 90% of the energy of the multi-lead ECG signal is concentrated between 0.5 and 35 Hz, and this part of energy contains the heart detection information of the target object. Baseline drift is generally caused by breathing and human body movement during signal acquisition. It is manifested as low-frequency slowly changing noise, and its frequency is generally less than 0.5Hz. Moreover, multi-lead ECG signals also include interference signals above 30Hz (that is, noise) . Therefore, the server can remove noise and baseline drift from the multi-lead ECG signal through a band-pass filter with a pass-band cut-off frequency of (0.5, 35) Hz to obtain processed ECG data.

103. Perform data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data.

It should be noted that the clinically collected data often have different lengths, so this application uses the frame division method to intercept the processed ECG data into 10 sections with a length of 1000 points, one section is one frame, and the data of each channel All can be regularized into a combination of 10 frames of data [10, 1000]. There are 12 lead channels in ECG data, so the dimension of multi-dimensional lead channel ECG data is [10, 1000, 12].

Specifically, the server pre-emphasizes the processed ECG data to obtain the aggravated ECG data; the server performs frame processing on the aggravated ECG data to obtain multiple frames of ECG data; The electrocardiographic data is smoothed to obtain each frame of ECG data after windowing; the server sequentially performs signal transformation processing (such as Fourier transform or wavelet transform) and splicing processing on each frame of ECG data after windowing to obtain a multidimensional guide Link channel ECG data.

104. Perform feature extraction and feature aggregation processing on the ECG data of multi-dimensional lead channels through the deep neural network model integrating the double attention mechanism, and obtain the target ECG feature data.

It should be noted that the deep neural network model (that is, DACnet) that incorporates the dual attention mechanism includes the residual network layer resnet, the dual attention network layer Dual-Attention and the fully connected network layer, the residual network layer, and the dual attention network. There is a preset connection mode and placement relationship between the placement position i and the fully connected network layer. Resnet is used to extract the feature map of multi-dimensional lead channel ECG data, while Dual-Attention is used to obtain the global information of the local features generated by resnet. The number of resnets is multiple, and each resnet contains a preset number (N1, N2, ..., N6) of stacked sub-blocks with the same number of channels, each sub-block consists of a two-dimensional convolutional layer, a batch normalization layer and activation function ReLU. Dual Attention includes a cross-channel attention mechanism and a global depth attention mechanism. A cross-channel attention mechanism is used to model the contextual information of any two locations on the feature map. The global deep attention mechanism is used to capture feature information between different channels, and the deep neural network model fused with the dual attention mechanism aggregates the features extracted by the two attention mechanisms respectively. Specifically, the server performs feature extraction and feature aggregation processing on the ECG data of the multi-dimensional lead channel through the residual network layer, the double attention network layer and the fully connected network layer to obtain target ECG feature data. Further, the server stores the target ECG characteristic data in the block chain database, which is not limited here.

In the embodiment of the present application, the feature extraction and feature aggregation processing are performed on the multi-dimensional lead channel ECG data through the deep neural network model fused with the dual attention mechanism to obtain the target ECG feature data, that is, through two different attention mechanisms Integrate feature information of different dimensions to achieve extended context information. Among them, the dependencies between all positions of the spatial feature map are calculated through the global deep attention mechanism, which expands the receptive field of the architecture; while the cross-channel attention mechanism captures feature information between different channels. The features of these two attention mechanisms are finally aggregated to further improve feature representations that help enrich contextual information. This solution can be applied in the field of smart medical care, thereby promoting the construction of smart cities.

Please refer to Fig. 2, another embodiment of the multi-lead electrocardiogram signal processing method in the embodiment of the present application includes:

201. Acquire a multi-lead electrocardiogram signal to be processed, where the multi-lead electrocardiogram signal to be processed is used to indicate heart detection information of a target object.

The specific execution process of step 201 is similar to the execution process of step 101, and details are not repeated here.

202. Perform data preprocessing on the multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data.

Specifically, the server can also use a low-pass filter to clean up the noise in the multi-lead ECG signal to obtain denoised ECG data; and use a high-pass filter to eliminate baseline drift on the denoised ECG data to obtain processed ECG data.

Optionally, the server removes noise from the multi-lead ECG signal to be processed through a preset band-pass filter to obtain denoised ECG data; the server eliminates baseline drift on the denoised ECG data to obtain processed ECG data.

203. Perform data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data.

It should be noted that the processed ECG data is a long time-domain signal, and usually the server divides the processed ECG data into multiple frame signals to obtain multi-dimensional lead channel ECG data. Optionally, the server performs length statistics on the processed ECG data to obtain the target data length; the server obtains the frame length and frame number, and performs difference calculation on the target data length and frame length to obtain the target difference; the server bases the target difference on and the frame number to determine the frame shift, and determine the ECG data of the multi-dimensional lead channel based on the frame shift, frame length and frame number.

204. Perform feature extraction and feature aggregation processing on the ECG data of the multi-dimensional lead channel through the deep neural network model integrated with the double attention mechanism, and obtain the target ECG feature data.

It should be noted that the server has pre-trained the deep neural network model that integrates the dual attention mechanism. Specifically, the server obtains the initial multi-lead ECG sample data, and performs data preprocessing on the initial multi-lead ECG sample data to obtain the target multiple Lead ECG sample data, for example, the server performs processing such as eliminating abnormal data, filling missing data, and data format conversion on the initial multi-lead ECG sample data; the server divides the target multi-lead ECG sample data according to the preset ratio, and obtains Multi-lead ECG training set, multi-lead ECG verification set and multi-lead ECG test set, for example, the preset ratio can be 8:1:1, or 6:2:2, which is not limited here; The server performs model training on the initial hybrid model based on the multi-lead ECG training set, the multi-lead ECG verification set and the multi-lead ECG test set, and obtains a deep neural network model incorporating a dual attention mechanism.

Further, the server forms an initial mixed model based on the initial deep neural network model and the initial double attention mechanism model, and initializes each network parameter in the initial mixed model. The initial mixed model includes a residual network layer, a double attention network layer and a fully connected Network layer, wherein each network parameter includes parameters such as learning rate, learning step size, number of iterations, and gradient descent rate; the server performs model training on the initial mixed model according to the multi-lead ECG training set to obtain the trained mixed model; The lead ECG verification set performs model verification and fine-tuning of each network parameter on the trained hybrid model to obtain the target hybrid model; the server performs model testing on the target hybrid model according to the multi-lead ECG test set to obtain the test result. When the test result is greater than or When it is equal to the preset target value, set the target hybrid model to be a deep neural network model with dual attention mechanism.

Optionally, first, the server extracts the features of the ECG data of the multi-dimensional lead channel by fusing the residual network layer in the deep neural network model of the dual attention mechanism to obtain the initial ECG local feature data; specifically, the server fuses the In the deep neural network model of the dual attention mechanism, the residual network extracts the convolutional features of the multi-dimensional lead channel ECG data to obtain the initial ECG local feature data, wherein the residual network includes multiple superimposed residual modules.

Secondly, based on the dual attention network layer in the deep neural network model that integrates the dual attention mechanism, the server performs deep feature processing on the initial ECG local feature data to obtain the initial ECG global feature data. The dual attention network layer includes cross-channel attention. force mechanism and global depth attention mechanism; specifically, the server losslessly transmits the initial ECG local feature data to the double attention network in the deep neural network model, and the double attention network includes a cross-channel attention mechanism and a global depth attention mechanism; The server extracts the details that are easily lost in the initial ECG local feature data through the cross-channel attention mechanism (embedded compression and excitation network) and the global deep attention mechanism, and obtains the initial ECG global feature data.

Finally, the server performs feature aggregation processing on the initial ECG global feature data by integrating the fully connected network layer in the deep neural network model of the dual attention mechanism to obtain the target ECG feature data. It should be noted that after the compression and excitation network converts the input signal X into a feature map through the last resnet, the squeeze operation aggregates the features across spatial dimensions into a channel descriptor of size 1×1×C as the channel global expression of information. The target ECG feature data include ECG abnormal feature data, sinus rhythm feature data, sinus tachycardia feature data, sinus arrhythmia feature data, sinus bradycardia feature data, atrial premature beat feature data, atrial fibrillation feature data Data, characteristic data of left ventricular high voltage, characteristic data of abnormal leads or poor data quality, characteristic data of ST-T changes, characteristic data of ventricular premature beats, characteristic data of T wave changes, characteristic data of limited right bundle branch block and abnormal Q Wave characteristic data, etc.

205. Update the target ECG feature data to a preset knowledge graph database, and generate an electrocardiogram analysis report based on the preset knowledge graph database.

Specifically, the server sequentially performs semantic analysis on the user questions of the target ECG feature data to obtain the analyzed ECG feature data; the server writes the analyzed ECG feature data into the preset knowledge graph database; the server passes The preset map analysis task sequentially performs data extraction, data fusion, data storage, and data calculation on the map database in the preset knowledge map library to obtain the ECG data; the server obtains the map template, and the server generates an ECG based on the map template and the ECG data Spectrum analysis report.

206. Send the electrocardiogram analysis report to the preset cloud storage terminal and the target terminal respectively, so that the target terminal displays the electrocardiogram analysis report.

Specifically, the server calls the preset application interface to send the electrocardiogram analysis report to the preset cloud storage terminal, so that the cloud storage terminal safely stores the electrocardiogram analysis report and responds to the file download request requested by the target terminal; The electrocardiogram analysis report is sent to the target terminal, so that the target terminal draws and displays the electrocardiogram analysis report.

In the embodiment of the present application, the feature extraction and feature aggregation processing are performed on the multi-dimensional lead channel ECG data through the deep neural network model fused with the dual attention mechanism to obtain the target ECG feature data, that is, through two different attention mechanisms Integrate feature information of different dimensions to achieve extended context information. Among them, the dependencies between all positions of the spatial feature map are calculated through the global deep attention mechanism, which expands the receptive field of the architecture; while the cross-channel attention mechanism captures feature information between different channels. The features of these two attention mechanisms are finally aggregated to further improve feature representations that help enrich contextual information. This program belongs to the field of smart medical care, through which the construction of smart cities can be promoted.

The above describes the multi-lead electrocardiogram signal processing method in the embodiment of the present application, and the following describes the multi-lead electrocardiogram signal processing device in the embodiment of the present application. Please refer to FIG. 3, the multi-lead electrocardiogram signal processing device in the embodiment of the present application An example of includes:

An acquisition module 301, configured to acquire a multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object;

A preprocessing module 302, configured to perform data preprocessing on the multi-lead electrocardiogram signal to be processed, to obtain processed electrocardiogram data;

Framing module 303, for performing data framing processing on the processed electrocardiogram data to obtain multidimensional lead channel electrocardiographic data;

The aggregation module 304 is configured to perform feature extraction and feature aggregation processing on the ECG data of multi-dimensional lead channels by integrating the deep neural network model of the dual attention mechanism to obtain target ECG feature data.

Further, the target ECG feature data is stored in the block chain database, which is not limited here.

In the embodiment of the present application, the feature extraction and feature aggregation processing are performed on the multi-dimensional lead channel ECG data through the deep neural network model fused with the dual attention mechanism to obtain the target ECG feature data, that is, through two different attention mechanisms Integrate feature information of different dimensions to achieve extended context information. Among them, the dependencies between all positions of the spatial feature map are calculated through the global deep attention mechanism, which expands the receptive field of the architecture; while the cross-channel attention mechanism captures feature information between different channels. The features of these two attention mechanisms are finally aggregated to further improve feature representations that help enrich contextual information.

Please refer to Fig. 4, another embodiment of the multi-lead electrocardiogram signal processing device in the embodiment of the present application includes:

Optionally, the preprocessing module 302 can also be specifically used for:

Denoise the multi-lead ECG signal to be processed through a preset band-pass filter to obtain denoised ECG data;

The baseline drift is eliminated for the denoised electrocardiogram data, and the processed electrocardiogram data is obtained.

Optionally, the framing module 303 can also be specifically used for:

Perform length statistics on the processed ECG data to obtain the target data length;

Obtain the frame length and frame number, and perform a difference operation on the target data length and frame length to obtain the target difference;

The frame shift is determined based on the target difference and the frame number, and the multi-dimensional lead channel ECG data is determined based on the frame shift, frame length and frame number.

Optionally, the aggregation module 304 may also be specifically used for:

By integrating the residual network layer in the deep neural network model of the dual attention mechanism, the feature extraction of the ECG data of the multi-dimensional lead channel is performed to obtain the initial ECG local feature data;

The dual attention network layer in the deep neural network model based on the fusion of dual attention mechanism performs deep feature processing on the initial ECG local feature data to obtain the initial ECG global feature data. The dual attention network layer includes cross-channel attention mechanism and Global deep attention mechanism;

By integrating the fully connected network layer in the deep neural network model of the dual attention mechanism, the initial ECG global feature data is aggregated to obtain the target ECG feature data.

Optionally, the multi-lead ECG signal processing device may also include:

A processing module 305, configured to acquire initial multi-lead ECG sample data, and perform data preprocessing on the initial multi-lead ECG sample data to obtain target multi-lead ECG sample data;

A division module 306, configured to divide the target multi-lead ECG sample data proportionally according to a preset ratio to obtain a multi-lead ECG training set, a multi-lead ECG verification set and a multi-lead ECG test set;

The training module 307 is used to perform model training on the initial hybrid model based on the multi-lead ECG training set, the multi-lead ECG verification set and the multi-lead ECG test set, and obtain a deep neural network model incorporating a dual attention mechanism.

Optionally, the training module 307 can also be specifically used for:

Based on the initial deep neural network model and the initial double attention mechanism model, the initial mixed model is formed, and each network parameter in the initial mixed model is initialized. The initial mixed model includes a residual network layer, a double attention network layer and a fully connected network layer;

Perform model training on the initial mixed model according to the multi-lead ECG training set to obtain the trained mixed model;

Through the multi-lead electrocardiogram verification set, the model verification and fine-tuning of each network parameter are carried out on the trained hybrid model to obtain the target hybrid model;

The target hybrid model is tested according to the multi-lead ECG test set, and the test result is obtained. When the test result is greater than or equal to the preset target value, the target hybrid model is set as a deep neural network model with a dual attention mechanism.

Optionally, the multi-lead ECG signal processing device may also include:

An update module 308, configured to update the target ECG characteristic data into a preset knowledge graph database, and generate an electrocardiogram analysis report based on the preset knowledge graph database;

The sending module 309 is configured to send the electrocardiogram analysis report to the preset cloud storage terminal and the target terminal respectively, so that the target terminal displays the electrocardiogram analysis report.

The above Fig. 3 and Fig. 4 describe the multi-lead ECG signal processing device in the embodiment of the present application in detail from the perspective of modularization, and the following describes the multi-lead ECG signal processing device in the embodiment of the present application in detail from the perspective of hardware processing.

Fig. 5 is a schematic structural diagram of a multi-lead ECG signal processing device provided by an embodiment of the present application. The multi-lead ECG signal processing device 500 may have relatively large differences due to different configurations or performances, and may include one or more than one Processor (central processing units, CPU) 510 (for example, one or more processors) and memory 520, one or more storage media 530 for storing application programs 533 or data 532 (for example, one or more mass storage devices). Wherein, the memory 520 and the storage medium 530 may be temporary storage or persistent storage. The program stored in the storage medium 530 may include one or more modules (not shown in the figure), and each module may include a series of computer program operations on the multi-lead ECG signal processing device 500 . Furthermore, the processor 510 may be configured to communicate with the storage medium 530 , and execute a series of computer program operations in the storage medium 530 on the multi-lead ECG signal processing device 500 .

The multi-lead ECG signal processing device 500 may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input and output interfaces 560, and/or, one or more operating systems 531, Such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that the structure of the multi-lead ECG signal processing device shown in FIG. components, or a different arrangement of components.

The present application also provides a computer-readable storage medium. The computer-readable storage medium may be a non-volatile computer-readable storage medium. The computer-readable storage medium may also be a volatile computer-readable storage medium. A computer program is stored in the computer-readable storage medium, and when the computer program is run on the computer, the computer is made to perform the steps of the multi-lead electrocardiogram signal processing method:

Acquiring the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object;

Perform data preprocessing on the multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data;

Perform data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data;

The multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing by means of a deep neural network model fused with a dual attention mechanism to obtain target ECG feature data.

The present application also provides a multi-lead electrocardiogram signal processing device. The multi-lead electrocardiogram signal processing device includes a memory and a processor, and a computer program is stored in the memory. When the computer program is executed by the processor, the processor executes The steps of the multi-lead electrocardiogram signal processing method in the above-mentioned embodiments:

Further, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function, etc.; The data created using the node, etc.

The blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain (Blockchain), essentially a decentralized database, is a series of data blocks associated with each other using cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify its Validity of information (anti-counterfeiting) and generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

Those skilled in the art can clearly understand that for the convenience and brevity of the description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several computer programs to enable a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disc and other media that can store program codes. .

As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still understand the foregoing The technical solutions recorded in each embodiment are modified, or some of the technical features are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the application.

The present application also provides a computer-readable storage medium. The computer-readable storage medium may be a non-volatile computer-readable storage medium. The computer-readable storage medium may also be a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium, and when the instructions are run on the computer, the computer is made to execute the steps of the multi-lead electrocardiogram signal processing method.

The present application also provides a multi-lead electrocardiogram signal processing device. The multi-lead electrocardiogram signal processing device includes a memory and a processor. Instructions are stored in the memory. When the instructions are executed by the processor, the processor executes the above-mentioned The steps of the multi-lead electrocardiogram signal processing method in the embodiment.

Claims

A multi-lead electrocardiogram signal processing method, wherein, the multi-lead electrocardiogram signal processing method comprises:

Acquiring the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object;

Perform data preprocessing on the multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data;

Perform data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data;

The multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing by means of a deep neural network model fused with a dual attention mechanism to obtain target ECG feature data.
The multi-lead electrocardiogram signal processing method according to claim 1, wherein said performing data preprocessing on said multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data comprises:

Denoising the multi-lead electrocardiogram signal to be processed through a preset bandpass filter to obtain denoised electrocardiogram data;

Eliminate baseline drift on the denoised electrocardiogram data to obtain processed electrocardiogram data.
The multi-lead electrocardiogram signal processing method according to claim 1, wherein, performing data frame processing on the processed electrocardiogram data to obtain multi-dimensional lead channel electrocardiogram data, comprising:

Carry out length statistics on the processed electrocardiogram data to obtain the target data length;

Obtaining the frame length and the number of frames, and performing a difference operation on the target data length and the frame length to obtain a target difference;

A frame shift is determined based on the target difference and the frame number, and multidimensional lead channel ECG data is determined based on the frame shift, the frame length, and the frame number.
The multi-lead electrocardiogram signal processing method according to claim 1, wherein, the described multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing through the deep neural network model of the fusion dual attention mechanism to obtain the target ECG characteristic data, including:

By fusing the residual network layer in the deep neural network model of the dual attention mechanism, the ECG data of the multi-dimensional lead channel is extracted to obtain initial ECG local feature data;

Based on the double attention network layer in the deep neural network model of the fusion double attention mechanism, the initial ECG local feature data is subjected to feature deep processing to obtain the initial ECG global feature data, and the double attention network layer includes Cross-channel attention mechanism and global depth attention mechanism;

Through the fully connected network layer in the deep neural network model fused with the dual attention mechanism, the feature aggregation processing is performed on the initial ECG global feature data to obtain the target ECG feature data.
The multi-lead electrocardiogram signal processing method according to any one of claims 1-4, wherein, in the acquisition of the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the target Before detecting the heart information of the object, the multi-lead electrocardiogram signal processing method includes:

Obtain initial multi-lead ECG sample data, and perform data preprocessing on the initial multi-lead ECG sample data to obtain target multi-lead ECG sample data;

Proportionally dividing the target multi-lead ECG sample data according to a preset ratio to obtain a multi-lead ECG training set, a multi-lead ECG verification set and a multi-lead ECG test set;

Based on the multi-lead electrocardiogram training set, the multi-lead electrocardiogram verification set and the multi-lead electrocardiogram test set, model training is performed on the initial mixed model to obtain a deep neural network model incorporating a dual attention mechanism.
The multi-lead electrocardiogram signal processing method according to claim 5, wherein, said initial mixed model based on said multi-lead electrocardiogram training set, said multi-lead electrocardiogram verification set and said multi-lead electrocardiogram test set Carry out model training to obtain a deep neural network model that incorporates a dual attention mechanism, including:

An initial mixed model is formed based on the initial deep neural network model and the initial double attention mechanism model, and each network parameter in the initial mixed model is initialized, and the initial mixed model includes a residual network layer, a double attention network layer and a full connection Network layer;

Carrying out model training on the initial mixed model according to the multi-lead ECG training set to obtain a trained mixed model;

Perform model verification and fine-tuning of each network parameter on the trained hybrid model through the multi-lead electrocardiogram verification set to obtain a target hybrid model;

Carry out a model test to the target mixed model according to the multi-lead electrocardiogram test set to obtain a test result, and when the test result is greater than or equal to a preset target value, set the target mixed model to be a combination of a dual attention mechanism Deep neural network models.
According to the multi-lead electrocardiogram signal processing method according to any one of claims 1-4, wherein, the multi-dimensional lead channel electrocardiogram data is subjected to feature extraction in the deep neural network model of the fusion dual attention mechanism and feature aggregation processing, after obtaining the target ECG feature data, the multi-lead electrocardiogram signal processing method also includes:

Updating the target ECG feature data into a preset knowledge graph library, and generating an electrocardiogram analysis report based on the preset knowledge graph library;

The electrocardiogram analysis report is respectively sent to a preset cloud storage terminal and a target terminal, so that the target terminal displays the electrocardiogram analysis report.
A multi-lead electrocardiogram signal processing device, comprising a memory, a processor, and computer-readable instructions stored on the memory and operable on the processor, the processor implements the computer-readable instructions when executing the computer-readable instructions Follow the steps below:

Acquiring the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object;

Perform data preprocessing on the multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data;

Perform data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data;

The multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing by means of a deep neural network model fused with a dual attention mechanism to obtain target ECG feature data.
The multi-lead electrocardiogram signal processing device according to claim 8, wherein said performing data preprocessing on said multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data comprises:

Denoising the multi-lead electrocardiogram signal to be processed through a preset bandpass filter to obtain denoised electrocardiogram data;

Eliminate baseline drift on the denoised electrocardiogram data to obtain processed electrocardiogram data.
The multi-lead electrocardiogram signal processing device according to claim 8, wherein, performing data frame processing on the processed electrocardiogram data to obtain multi-dimensional lead channel electrocardiogram data, comprising:

Carry out length statistics on the processed electrocardiogram data to obtain the target data length;

Obtaining the frame length and the number of frames, and performing a difference operation on the target data length and the frame length to obtain a target difference;

A frame shift is determined based on the target difference and the frame number, and multidimensional lead channel ECG data is determined based on the frame shift, the frame length, and the frame number.
The multi-lead electrocardiogram signal processing device according to claim 8, wherein, the multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing by the deep neural network model fused with a dual attention mechanism to obtain the target ECG characteristic data, including:

By fusing the residual network layer in the deep neural network model of the dual attention mechanism, the ECG data of the multi-dimensional lead channel is extracted to obtain initial ECG local feature data;

Based on the double attention network layer in the deep neural network model of the fusion double attention mechanism, the initial ECG local feature data is subjected to feature deep processing to obtain the initial ECG global feature data, and the double attention network layer includes Cross-channel attention mechanism and global depth attention mechanism;

Through the fully connected network layer in the deep neural network model fused with the dual attention mechanism, the feature aggregation processing is performed on the initial ECG global feature data to obtain the target ECG feature data.
The multi-lead electrocardiogram signal processing device according to any one of claims 8-11, wherein, in the acquisition of the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the target Before detecting the heart information of the object, the multi-lead electrocardiogram signal processing method includes:

Obtain initial multi-lead ECG sample data, and perform data preprocessing on the initial multi-lead ECG sample data to obtain target multi-lead ECG sample data;

Proportionally dividing the target multi-lead ECG sample data according to a preset ratio to obtain a multi-lead ECG training set, a multi-lead ECG verification set and a multi-lead ECG test set;

Based on the multi-lead electrocardiogram training set, the multi-lead electrocardiogram verification set and the multi-lead electrocardiogram test set, model training is performed on the initial mixed model to obtain a deep neural network model incorporating a dual attention mechanism.
The multi-lead electrocardiogram signal processing device according to claim 12, wherein the initial mixed model is based on the multi-lead electrocardiogram training set, the multi-lead electrocardiogram verification set and the multi-lead electrocardiogram test set Carry out model training to obtain a deep neural network model that incorporates a dual attention mechanism, including:

An initial mixed model is formed based on the initial deep neural network model and the initial double attention mechanism model, and each network parameter in the initial mixed model is initialized, and the initial mixed model includes a residual network layer, a double attention network layer and a full connection Network layer;

Carrying out model training on the initial mixed model according to the multi-lead ECG training set to obtain a trained mixed model;

Perform model verification and fine-tuning of each network parameter on the trained hybrid model through the multi-lead electrocardiogram verification set to obtain a target hybrid model;

Carry out a model test to the target mixed model according to the multi-lead electrocardiogram test set to obtain a test result, and when the test result is greater than or equal to a preset target value, set the target mixed model to be a combination of a dual attention mechanism Deep neural network models.
The multi-lead electrocardiogram signal processing device according to any one of claims 8-11, wherein the feature extraction is performed on the multi-dimensional lead channel ECG data through the deep neural network model of the fusion dual attention mechanism and feature aggregation processing, after obtaining the target ECG feature data, the multi-lead electrocardiogram signal processing method also includes:

Updating the target ECG feature data into a preset knowledge graph library, and generating an electrocardiogram analysis report based on the preset knowledge graph library;

The electrocardiogram analysis report is respectively sent to a preset cloud storage terminal and a target terminal, so that the target terminal displays the electrocardiogram analysis report.
A computer-readable storage medium, wherein computer instructions are stored in the computer-readable storage medium, and when the computer instructions are run on the computer, the computer is made to perform the following steps:

Acquiring the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object;

Perform data preprocessing on the multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data;

Perform data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data;

The multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing by means of a deep neural network model fused with a dual attention mechanism to obtain target ECG feature data.
The computer-readable storage medium according to claim 15, wherein said performing data preprocessing on said multi-lead electrocardiogram signal to be processed to obtain processed electrocardiogram data comprises:

Denoising the multi-lead electrocardiogram signal to be processed through a preset bandpass filter to obtain denoised electrocardiogram data;

Eliminate baseline drift on the denoised electrocardiogram data to obtain processed electrocardiogram data.
The computer-readable storage medium according to claim 15, wherein, performing data frame processing on the processed ECG data to obtain multi-dimensional lead channel ECG data includes:

Carry out length statistics on the processed electrocardiogram data to obtain the target data length;

Obtaining the frame length and the number of frames, and performing a difference operation on the target data length and the frame length to obtain a target difference;

A frame shift is determined based on the target difference and the frame number, and multidimensional lead channel ECG data is determined based on the frame shift, the frame length, and the frame number.
The computer-readable storage medium according to claim 15, wherein, the multi-dimensional lead channel ECG data is subjected to feature extraction and feature aggregation processing by the deep neural network model fused with a dual attention mechanism to obtain the target ECG Characteristic data, including:

By fusing the residual network layer in the deep neural network model of the dual attention mechanism, the ECG data of the multi-dimensional lead channel is extracted to obtain initial ECG local feature data;

Based on the double attention network layer in the deep neural network model of the fusion double attention mechanism, the initial ECG local feature data is subjected to feature deep processing to obtain the initial ECG global feature data, and the double attention network layer includes Cross-channel attention mechanism and global depth attention mechanism;

Through the fully connected network layer in the deep neural network model fused with the dual attention mechanism, the feature aggregation processing is performed on the initial ECG global feature data to obtain the target ECG feature data.
The computer-readable storage medium according to any one of claims 15-18, wherein, in the acquisition of the multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the Before the heart detection information, the multi-lead electrocardiogram signal processing method includes:

Obtain initial multi-lead ECG sample data, and perform data preprocessing on the initial multi-lead ECG sample data to obtain target multi-lead ECG sample data;

Proportionally dividing the target multi-lead ECG sample data according to a preset ratio to obtain a multi-lead ECG training set, a multi-lead ECG verification set and a multi-lead ECG test set;

Based on the multi-lead electrocardiogram training set, the multi-lead electrocardiogram verification set and the multi-lead electrocardiogram test set, model training is performed on the initial mixed model to obtain a deep neural network model incorporating a dual attention mechanism.
A multi-lead electrocardiogram signal processing device, wherein the multi-lead electrocardiogram signal processing device includes:

An acquisition module, configured to acquire a multi-lead electrocardiogram signal to be processed, the multi-lead electrocardiogram signal to be processed is used to indicate the heart detection information of the target object;

A preprocessing module, configured to perform data preprocessing on the multi-lead electrocardiogram signal to be processed, to obtain processed electrocardiogram data;

A framing module, configured to perform data framing processing on the processed ECG data to obtain multidimensional lead channel ECG data;

The aggregation module is used to perform feature extraction and feature aggregation processing on the ECG data of the multi-dimensional lead channel through a deep neural network model fused with a dual attention mechanism, so as to obtain target ECG feature data.