A deep learning-based electrocardiogram classification method and device, and a storage medium

Abstract

The present application relates to the field of signal processing and medical treatment, and discloses an electrocardiogram classification method based on deep learning, comprising: preprocessing a to-be-identified electrocardiogram to obtain frontal plane data of 6*n lead signals and horizontal plane data of 6*n lead signals; using a fusion model to process the frontal plane data and the horizontal plane data respectively to obtain first electrocardiogram plane data and second electrocardiogram plane data, and splicing the first electrocardiogram plane data and the second electrocardiogram plane data to obtain spliced data; performing feature extraction on the spliced data through a feature extraction model to obtain feature data; and using a classification function to classify the extracted feature data to obtain a classification result of the to-be-identified electrocardiogram. By using machine learning, the present application can classify normal and abnormal electrocardiogram signals, and in the medical field, can be used for intelligently classifying electrocardiogram signals of patients, so that doctors have more time to analyze the condition of abnormal electrocardiogram signals.

Description

Technical Field

[0001] This invention relates to the field of signal processing, and more particularly to an electrocardiogram (ECG) classification method, apparatus, and storage medium based on deep learning. Background Technology

[0002] As people's living standards continue to improve, they are paying more and more attention to their health. Electrocardiogram (ECG) is an important tool for the clinical diagnosis of heart diseases, assisting medical staff in making diagnoses.

[0003] Currently, various types of electrocardiographs are widely used in clinical practice in the medical field.

[0004] However, current electrocardiogram (ECG) collection requires manual interpretation of the ECG equipment to determine whether the patient's ECG is normal or abnormal. Manually checking for abnormalities in ECGs is extremely time-consuming and inefficient, causing doctors to spend a lot of time judging whether a patient's ECG is abnormal, thus diverting their energy from analyzing the causes of the abnormalities.

[0005] Therefore, current electrocardiogram (ECG) instruments have the problem of not being able to intelligently determine whether an ECG is abnormal. Summary of the Invention

[0006] In view of the above, it is necessary to provide a deep learning-based electrocardiogram (ECG) classification method, which aims to solve the technical problem in the existing technology of not being able to intelligently determine whether an ECG is abnormal.

[0007] This invention provides a deep learning-based electrocardiogram (ECG) classification method, comprising:

[0008] The electrocardiogram to be identified is preprocessed to obtain electrocardiogram signal data. The electrocardiogram signal data is divided into a 12*n two-dimensional array, which includes frontal plane data of 6*n leads and transverse plane data of 6*n leads.

[0009] The two-dimensional array is input into a preset fusion model. The frontal plane data is processed to obtain the first ECG plane data, and the transverse plane data is processed to obtain the second ECG plane data. The first ECG plane data and the second ECG plane data are then spliced ​​together to obtain spliced ​​data.

[0010] The spliced ​​data is subjected to feature extraction using a preset feature extraction model to obtain the extracted feature data.

[0011] The extracted feature data is classified using a preset classification function to obtain the classification result of the electrocardiogram to be identified.

[0012] Optionally, the electrocardiogram to be identified is preprocessed to obtain electrocardiogram signal data, including:

[0013] The electrocardiogram to be identified is denoised using a preset first filter;

[0014] The denoised electrocardiogram data is downsampled using a preset second filter to obtain electrocardiogram signal data with a preset sampling rate.

[0015] Optionally, processing the forehead data to obtain the first electrocardiogram plane data includes:

[0016] The fusion model's convolution kernel is used to fuse the 6*n lead signals of the frontal plane data to obtain the first ECG plane data with a data dimension of 1*n1*a, where a represents the number of channels of the convolution kernel and n1 is the data size obtained after convolution by the convolution kernel.

[0017] Optionally, processing the transverse data to obtain the second electrocardiogram plane data includes:

[0018] The fusion model's convolution kernel is used to fuse the 6*n lead signals of the frontal plane data to obtain second ECG plane data with a data dimension of 1*n1*a, where a represents the number of channels of the convolution kernel and n1 is the data size obtained after convolution by the convolution kernel.

[0019] Optionally, the step of stitching together the first ECG plane data and the second ECG plane data to obtain stitched data includes:

[0020] The first ECG plane data with a data dimension of 1*n1*a after being fused by the fusion model, and the second ECG plane data with a data dimension of 1*n1*a are concatenated to obtain spliced ​​data with a data dimension of 1*n1*2a, where n1 is the data size obtained after convolution by the convolution kernel of the fusion model, and a represents the number of channels of the convolution kernel.

[0021] Optionally, classifying the extracted feature data using a preset classification function to obtain the classification result of the electrocardiogram to be identified includes:

[0022] The ECG results to be identified are classified using the softmax function. The extracted feature data is mapped to a (0,1) value by the softmax function, and the output is 0 or 1. 0 represents an abnormal ECG to be identified, and 1 represents a normal ECG to be identified.

[0023] Optionally, the step of using a preset feature extraction model to extract features from the concatenated data to obtain extracted feature data includes:

[0024] The concatenated data is subjected to a predetermined number of convolutional processes using the feature extraction model.

[0025] The feature data is obtained by extracting features from the concatenated data after convolution processing using the fully connected layer of the feature extraction model.

[0026] To address the aforementioned problems, the present invention also provides a device for classifying electrocardiogram (ECG) signal images, the device comprising:

[0027] The data processing module is used to preprocess the electrocardiogram to be identified to obtain electrocardiogram signal data, and divide the electrocardiogram signal data into a 12*n two-dimensional array, wherein the two-dimensional array includes frontal plane data of 6*n leads and transverse plane data of 6*n leads.

[0028] The data layer fusion module is used to input the two-dimensional array into a preset fusion model, process the frontal plane data to obtain the first ECG plane data, process the transverse plane data to obtain the second ECG plane data, and splice the first ECG plane data and the second ECG plane data to obtain spliced ​​data.

[0029] The feature extraction module is used to extract features from the spliced ​​data using a preset feature extraction model to obtain the extracted feature data.

[0030] The classification module is used to classify the extracted feature data using a preset classification function to obtain the classification result of the electrocardiogram to be identified.

[0031] To address the above problems, the present invention also provides an electronic device, the electronic device comprising:

[0032] At least one processor; and,

[0033] A memory communicatively connected to the at least one processor; wherein,

[0034] The memory stores a deep learning-based electrocardiogram (ECG) classification program that can be executed by the at least one processor. The deep learning-based ECG classification program is executed by the at least one processor so that the at least one processor can execute the aforementioned deep learning-based ECG classification method.

[0035] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing a deep learning-based electrocardiogram (ECG) classification program, which can be executed by one or more processors to implement the aforementioned deep learning-based ECG classification method.

[0036] Compared to existing technologies, this method captures the characteristics of synchronized ECG signals from different leads, then fuses them, and finally classifies them.

[0037] This invention preprocesses the electrocardiogram (ECG) to be identified to obtain ECG signal data. This data is then divided into a 12*n two-dimensional array, comprising frontal plane data and transverse plane data of 6*n leads. The array is then input into a preset fusion model. The frontal plane data is processed to obtain first ECG plane data, and the transverse plane data is processed to obtain second ECG plane data. The first and second ECG plane data are then concatenated to obtain concatenated data. This fusion of features from the ECG facilitates subsequent feature extraction. A preset feature extraction model is then used to extract features from the concatenated data. Finally, a preset classification function is used to classify the extracted feature data, resulting in a classification result for the ECG to be identified. This allows for the classification of the extracted feature data.

[0038] Therefore, the above-mentioned technical solution of the present invention, by fusing electrocardiogram (ECG) activity information collected from different planes, reflects the original information of cardiac activity, which is more conducive to subsequent feature extraction and classification. Furthermore, by fusing multi-lead ECG signals through a convolutional neural network to obtain fused signals from different planes for feature extraction, and then classifying abnormal ECG signals, this method can intelligently determine whether an ECG is abnormal in the medical field, thereby improving the efficiency of ECG signal classification and allowing doctors more time to analyze abnormal ECGs. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating a deep learning-based electrocardiogram classification method according to an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of a deep learning-based electrocardiogram classification device according to an embodiment of the present invention;

[0041] Figure 3 A schematic diagram of the structure of an electronic device that implements a deep learning-based electrocardiogram classification method according to an embodiment of the present invention;

[0042] Figure 4 a shows the lead positions of the 12 leads in a standard ECG;

[0043] Figure 4 b shows a schematic diagram of the positions of the six frontal plane leads I, II, III, aVR, aVL, and aVF in the 12-lead configuration;

[0044] Figure 4 c shows a schematic diagram of the positions of the six transverse leads V1, V2, V3, V4, V5, and V6 in a 12-lead configuration;

[0045] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0047] It should be noted that the descriptions involving "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0048] Reference Figure 1 The diagram shown is a flowchart illustrating a deep learning-based electrocardiogram (ECG) classification method according to an embodiment of the present invention. This method is executed by an electronic device.

[0049] In this embodiment, a deep learning-based electrocardiogram (ECG) classification method includes:

[0050] S1. Preprocess the electrocardiogram to be identified to obtain electrocardiogram signal data, and divide the electrocardiogram signal data into a 12*n two-dimensional array. The two-dimensional array includes frontal plane data of 6*n leads and transverse plane data of 6*n leads.

[0051] In this embodiment, electrocardiography (ECG) is a primary method for recording and monitoring cardiac electrical activity signals. It diagnoses cardiac abnormalities by observing the regularity of the heartbeat, including arrhythmias, ischemic heart disease, and myocardial infarction. The ECG signal data consists of 12 leads, including 6 limb leads (frontal plane data, I, II, III, aVR, aVL, aVF) and 6 chest leads (transverse plane data, V1–V6). The limb leads include standard bipolar leads (I, II, and III) and augmentation leads (aVR, aVL, and aVF). Figure 4 As shown:

[0052] These 12 lead data are time-series signals, reflecting the electrocardiographic activity of the heart at different angles. A two-dimensional array is essentially an array of arrays, i.e., an array of arrays. The preprocessing of the ECG to be identified involves denoising and dimensionality reduction of the ECG data. The first filter is a noise reduction filter; in this embodiment, an EMI filter can be used to denoise the ECG data. The second filter is a downsampling filter; in this embodiment, a downsampling digital filter can be used. A second downsampling filter is used to downsample the ECG data, resulting in a sampling rate of 250Hz and 2000 samples per lead. These sampling values ​​and sampling rate are not subject to further limitation. The final data is processed into a 12*2000 two-dimensional array, with each row representing the ECG signal data of one lead, reducing the number of samples and thus reducing the computer's processing time.

[0053] In one embodiment, a binning method can also be used to denoise the ECG data to be identified, which will not be elaborated further. Step S1 allows the ECG data to be identified to be divided into a 12*n two-dimensional array after denoising and downsampling. This facilitates direct input of the two-dimensional array into a preset fusion model for fusion. In the medical field, ECGs of patients seeking consultation at a hospital are obtained using an ECG instrument, and then the above operations are performed to obtain a 12*n two-dimensional array of the patient's ECG data.

[0054] S2. Input the two-dimensional array into a preset fusion model, process the frontal plane data to obtain the first ECG plane data, process the transverse plane data to obtain the second ECG plane data, and splice the first ECG plane data and the second ECG plane data to obtain spliced ​​data.

[0055] In this embodiment, the ECG sampling signals of the 12 leads in the ECG data to be identified are processed into a two-dimensional array. For example, the 12 leads are processed into a 12*2000 two-dimensional array. According to the leads of the frontal plane data and the transverse plane data, the 12 leads are divided into groups of 6 leads each, resulting in two 6*2000 two-dimensional arrays. One 6*2000 two-dimensional array is the frontal plane data, and the other 6*2000 two-dimensional array is the transverse plane data.

[0056] Finally, these two 6*2000 two-dimensional arrays are simultaneously processed through a preset fusion model.

[0057] The preset fusion model is a convolutional neural network with a kernel size of 6*N (N can be customized according to requirements, without further limitation here). Then, the signal fusion of the 6*2000 two-dimensional array of frontal data is performed. The kernel of the fusion model is used to convolve the 6*2000 first lead data of the frontal data to obtain the first ECG plane data. At the same time, the preset fusion model is a convolutional neural network with a kernel size of 6*N to convolve the 6*2000 two-dimensional array of transverse data to obtain the second ECG plane data.

[0058] The data dimensions of both the first and second ECG plane data are 1*n1*a, where n1 is the data size obtained after convolution and a represents the channels of the convolution kernel.

[0059] The first and second ECG plane data are obtained by fusing them using a fusion model with the same convolutional kernel. The data has a dimension of 1*n1*a, meaning that the two dimensions (height and width) are the same. When the first and second ECG plane data are directly input into the computer, the computer can connect data of the same size (height and width) to generate three-dimensional data, resulting in spliced ​​data with a dimension of 1*n1*2a, where n1 is the size of the data obtained after convolution, and 2a represents the number of channels of the convolutional kernel.

[0060] In step S2, the data from two different planes, the frontal plane data and the transverse plane data, can be fused together using a fusion model to obtain the original features and spliced ​​data. This allows the features of the electrocardiogram to be identified to be fused together, facilitating subsequent feature extraction. In the medical field, the feature information contained in the electrocardiogram of the patient seeking consultation can be obtained.

[0061] S3. Use a preset feature extraction model to extract features from the spliced ​​data to obtain the extracted feature data.

[0062] In this embodiment, the concatenated data with a data dimension of 1*n1*2a is input into the preset feature extraction model. The preset feature extraction model uses a convolutional residual network model involving 17 convolutional layers, including 4 residual blocks containing two layers of residual learning units, and the 18th layer is a fully connected layer.

[0063] First, the concatenated data is used as the input value X1 and fed into the first layer for a 1x50x4 convolution. The output feature is represented by F1(X). After the second layer of 1x50x4 convolution and the third layer of 1x50x8 convolution, the output feature of the third layer is represented by F3(X). F3(X) is fed into the residual block, which is divided into two branches. One branch goes through shortcut connections and directly reaches the input position of the sixth layer. The other branch, F3(X), continues to perform a fourth layer of 1x50x8 convolution to output F4(X). F4(X) then goes through a fifth layer of 1x50x8 convolution to output F5(X). At this time, the input value of the sixth layer is X6 = F5(X) + F3(X).

[0064] Then, X6 is convolved into a sixth layer of 1x50x8, and the output F6(X) is input into a seventh layer of 1x50x16. The output features are represented by F7(X).

[0065] The input to F7(X) is a residual block. One path goes through shortcut connections and directly reaches the input position of the tenth layer. The other path F7(X) continues to perform a 1x50x16 convolution in the eighth layer and outputs F8(X). F8(X) then goes through a 1x50x16 convolution in the ninth layer and outputs F9(X). At this time, the input value of the tenth layer is X10 = F9(X) + F7(X).

[0066] X10 is convolved into a 1x50x16 layer at the tenth layer, and the output F10(X) is input into the eleventh layer for a 1x50x32 convolution. The output features are represented by F11(X).

[0067] The input to F11(X) is a residual block. One path goes through shortcut connections and directly reaches the input position of the fourteenth layer. The other path F11(X) continues to perform a 1x50x32 convolution in the twelfth layer and outputs F12(X). F12(X) then goes through a 1x50x32 convolution in the thirteenth layer and outputs F13(X). At this time, the input value of the fourteenth layer is X14 = F13(X) + F11(X).

[0068] X14 is convolved in the fourteenth layer with a 1x50x16 convolution, and the output F14(X) is input into the fifteenth layer with a 1x50x32 convolution. The output features are represented by F15(X).

[0069] The input to F15(X) is a residual block. One path goes through shortcut connections and directly reaches the input position of the eighteenth layer. The other path F15(X) continues to perform a 1x50x32 convolution in the sixteenth layer and outputs F16(X). F16(X) then goes through a 1x50x32 convolution in the seventeenth layer and outputs F17(X). At this time, the input value of the eighteenth layer is X18 = F17(X) + F15(X).

[0070] Input X18 into the fully connected layer.

[0071] In step S3, features of the electrocardiogram to be identified can be extracted, resulting in feature data from two different planes.

[0072] S4. Classify the extracted feature data using a preset classification function to obtain the classification result of the electrocardiogram to be identified.

[0073] Softmax is a known classification function in this field. This step can use the softmax function to classify the ECG result to be identified. The extracted feature data is mapped to a value of (0,1) by the softmax function, and outputs 0 or 1. 0 represents that the ECG to be identified is abnormal, and 1 represents that the ECG to be identified is normal. No further explanation is needed.

[0074] In step S4, the extracted feature data can be classified. In the medical field, the electrocardiograms of patients who come to the hospital for consultation can be intelligently classified without manual judgment. This allows doctors to know whether the electrocardiograms obtained from the electrocardiogram instrument are abnormal, thus giving them more time to analyze abnormal electrocardiograms.

[0075] like Figure 2 The diagram shown is a schematic diagram of a module of an electrocardiogram signal image classification device provided in an embodiment of the present invention.

[0076] The electrocardiogram (ECG) signal image classification device 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the ECG signal image classification device 100 may include a data processing module 110, a data layer fusion module 120, a feature extraction module 130, and a classification module 140. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0077] In this embodiment, the functions of each module / unit are as follows:

[0078] The data processing module 110 is used to preprocess the electrocardiogram to be identified to obtain electrocardiogram signal data, and divide the electrocardiogram signal data into a 12*n two-dimensional array, wherein the two-dimensional array includes frontal plane data of 6*n lead signals and transverse plane data of 6*n lead signals.

[0079] The data layer fusion module 120 is used to input the two-dimensional array into a preset fusion model, process the frontal plane data to obtain the first ECG plane data, process the transverse plane data to obtain the second ECG plane data, and splice the first ECG plane data and the second ECG plane data to obtain spliced ​​data.

[0080] Feature extraction module 130 is used to extract features from the spliced ​​data using a preset feature extraction model to obtain extracted feature data;

[0081] The classification module 140 is used to classify the extracted feature data using a preset classification function to obtain the classification result of the electrocardiogram to be identified.

[0082] In one embodiment, the preprocessing of the electrocardiogram to be identified to obtain electrocardiogram signal data includes:

[0083] The electrocardiogram to be identified is denoised using a preset first filter;

[0084] The denoised electrocardiogram data is downsampled using a preset second filter to obtain electrocardiogram signal data with a preset sampling rate.

[0085] In one embodiment, processing the frontal plane data to obtain first electrocardiogram plane data includes:

[0086] The fusion model's convolution kernel is used to fuse the 6*n lead signals of the frontal plane data to obtain the first ECG plane data with a data dimension of 1*n1*a, where a represents the number of channels of the convolution kernel and n1 is the data size obtained after convolution by the convolution kernel.

[0087] In one embodiment, processing the transverse plane data to obtain the second electrocardiogram plane data includes:

[0088] The fusion model's convolution kernel is used to fuse the 6*n lead signals of the frontal plane data to obtain second ECG plane data with a data dimension of 1*n1*a, where a represents the number of channels of the convolution kernel and n1 is the data size obtained after convolution by the convolution kernel.

[0089] In one embodiment, the step of stitching together the first ECG plane data and the second ECG plane data to obtain stitched data includes:

[0090] The first ECG plane data with a data dimension of 1*n1*a after being fused by the fusion model, and the second ECG plane data with a data dimension of 1*n1*a are concatenated to obtain spliced ​​data with a data dimension of 1*n1*2a, where n1 is the data size obtained after convolution by the convolution kernel of the fusion model, and a represents the number of channels of the convolution kernel.

[0091] In one embodiment, classifying the extracted feature data using a preset classification function to obtain the classification result of the electrocardiogram to be identified includes:

[0092] The ECG results to be identified are classified using the softmax function. The extracted feature data is mapped to a (0,1) value by the softmax function, and the output is 0 or 1. 0 represents an abnormal ECG to be identified, and 1 represents a normal ECG to be identified.

[0093] In one embodiment, the step of extracting features from the concatenated data using a preset feature extraction model to obtain the extracted feature data includes:

[0094] The concatenated data is subjected to a predetermined number of convolutional processes using the feature extraction model.

[0095] The feature data is obtained by extracting features from the concatenated data after convolution processing using the fully connected layer of the feature extraction model.

[0096] like Figure 3 The diagram shown is a schematic representation of an electronic device that implements a deep learning-based electrocardiogram classification method according to an embodiment of the present invention.

[0097] In this embodiment, the electronic device 1 includes, but is not limited to, a memory 11, a processor 12, and a network interface 13 that can be interconnected via a system bus. The memory 11 stores a ECG signal image classification program 10, which can be executed by the processor 12. Figure 3 Only an electronic device 1 with components 11-13 and an electrocardiogram signal image classification program 10 is shown. Those skilled in the art will understand that... Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0098] The memory 11 includes RAM and at least one type of readable storage medium. The RAM provides a cache for the operation of the electronic device 1; the readable storage medium can be a non-volatile storage medium such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the readable storage medium can be an internal storage unit of the electronic device 1; in other embodiments, the non-volatile storage medium can also be an external storage device of the electronic device 1, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 1. In this embodiment, the readable storage medium of the memory 11 is typically used to store the operating system and various application software installed on the electronic device 1, such as storing the code of an electrocardiogram signal image classification program 10 according to an embodiment of the present invention. Furthermore, the memory 11 can also be used to temporarily store various types of data that have been output or will be output.

[0099] In some embodiments, processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. This processor 12 is typically used to control the overall operation of the electronic device 1, such as performing control and processing related to data interaction or communication with other devices. In this embodiment, the processor 12 is used to run program code stored in the memory 11 or process data, for example, running a program 10 based on electrocardiogram signal image classification.

[0100] The network interface 13 may include a wireless network interface or a wired network interface, which is used to establish a communication connection between the electronic device 1 and the terminal (not shown in the figure).

[0101] Optionally, the electronic device 1 may further include a user interface, which may include a display, an input unit such as a keyboard, and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0102] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0103] The memory 11 in the electronic device 1 stores a deep learning-based electrocardiogram classification program 10, which is a combination of multiple instructions. When run in the processor 12, it can achieve the following:

[0104] The electrocardiogram to be identified is preprocessed to obtain electrocardiogram signal data. The electrocardiogram signal data is divided into a 12*n two-dimensional array, which includes frontal plane data of 6*n leads and transverse plane data of 6*n leads.

[0105] The two-dimensional array is input into a preset fusion model, the frontal plane data is processed to obtain the first ECG plane data, and the transverse plane data is processed to obtain the second ECG plane data; the first ECG plane data and the second ECG plane data are then connected to obtain spliced ​​data.

[0106] The spliced ​​data is processed by a pre-defined feature extraction model to extract features from the spliced ​​data.

[0107] The extracted feature data is classified using a preset classification function to obtain the classification result of the electrocardiogram to be identified.

[0108] Specifically, the processor 12 can refer to the specific implementation method of the aforementioned deep learning-based electrocardiogram classification program 10. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0109] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable medium can be non-volatile or non-combustible. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0110] The computer-readable storage medium stores a deep learning-based electrocardiogram (ECG) classification program 10, which can be executed by one or more processors. The specific implementation of the computer-readable storage medium of the present invention is basically the same as the above-described embodiments of the deep learning-based ECG classification method, and will not be described in detail here.

[0111] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0112] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0113] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0114] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0115] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0116] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The term "second class" is used to indicate names and does not indicate any specific order.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A deep learning-based electrocardiogram (ECG) classification method, characterized in that, The method includes: The electrocardiogram to be identified is preprocessed to obtain electrocardiogram signal data. The electrocardiogram signal data is divided into a 12*n two-dimensional array. The two-dimensional array includes frontal plane data of 6*n lead signals and transverse plane data of 6*n lead signals. The frontal plane data is the limb lead data in the electrocardiogram signal data, and the transverse plane data is the chest lead data in the electrocardiogram signal data. The two-dimensional array is input into a preset fusion model. The fusion model's convolution kernel is used to fuse the 6*n lead signals of the frontal plane data to obtain first ECG plane data with a data dimension of 1*n1*a. The fusion model's convolution kernel is used to fuse the 6*n lead signals of the transverse plane data to obtain second ECG plane data with a data dimension of 1*n1*a. The first ECG plane data and the second ECG plane data are concatenated to obtain spliced ​​data with a data dimension of 1*n1*2a, where n1 is the data size obtained after convolution by the convolution kernel, and a represents the number of channels of the convolution kernel. The spliced ​​data is subjected to feature extraction using a preset feature extraction model to obtain the extracted feature data. The extracted feature data is classified using a preset classification function to obtain the classification result of the electrocardiogram to be identified.

2. The deep learning-based electrocardiogram classification method as described in claim 1, characterized in that, The preprocessing of the electrocardiogram to be identified to obtain electrocardiogram signal data includes: The electrocardiogram to be identified is denoised using a preset first filter; The ECG data after noise reduction is downsampled using a preset second filter to obtain ECG signal data with a preset sampling rate.

3. The deep learning-based electrocardiogram classification method as described in claim 1, characterized in that, The process of processing the forehead surface data to obtain the first electrocardiogram plane data includes: The fusion model's convolution kernel is used to fuse the 6*n lead signals of the frontal plane data to obtain the first ECG plane data with a data dimension of 1*n1*a, where a represents the number of channels of the convolution kernel and n1 is the data size obtained after convolution by the convolution kernel.

4. The deep learning-based electrocardiogram classification method as described in claim 1, characterized in that, The process of processing the transverse data to obtain the second electrocardiogram plane data includes: The fusion model's convolution kernel is used to fuse the 6*n lead signals of the horizontal data to obtain second ECG plane data with a data dimension of 1*n1*a, where a represents the number of channels of the convolution kernel and n1 is the data size obtained after convolution by the convolution kernel.

5. The deep learning-based electrocardiogram classification method as described in claim 1, characterized in that, The step of classifying the extracted feature data using a preset classification function to obtain the classification result of the electrocardiogram to be identified includes: The ECG results to be identified are classified using the softmax function. The extracted feature data is mapped to a (0,1) value by the softmax function, and the output is 0 or 1. 0 represents an abnormal ECG to be identified, and 1 represents a normal ECG to be identified.

6. The deep learning-based electrocardiogram classification method as described in claim 1, characterized in that, The step of extracting features from the concatenated data using a preset feature extraction model to obtain extracted feature data includes: The concatenated data is subjected to a predetermined number of convolutional processes using the feature extraction model. The feature data is obtained by extracting features from the concatenated data after convolution processing using the fully connected layer of the feature extraction model.

7. A device for classifying electrocardiogram (ECG) signal images, used to implement the deep learning-based ECG classification method as described in any one of claims 1 to 6, characterized in that, The device includes: The data processing module is used to preprocess the electrocardiogram to be identified to obtain electrocardiogram signal data, and divide the electrocardiogram signal data into a 12*n two-dimensional array. The two-dimensional array includes frontal plane data of 6*n lead signals and transverse plane data of 6*n lead signals. The frontal plane data is the limb lead data in the electrocardiogram signal data, and the transverse plane data is the chest lead data in the electrocardiogram signal data. The data layer fusion module is used to input the two-dimensional array into a preset fusion model, and use the convolution kernel of the fusion model to perform data signal fusion on the 6*n lead signals of the frontal plane data to obtain the first ECG plane data with a data dimension of 1*n1*a, and to perform data signal fusion on the 6*n lead signals of the transverse plane data to obtain the second ECG plane data with a data dimension of 1*n1*a. The first ECG plane data and the second ECG plane data are connected to obtain spliced ​​data with a data dimension of 1*n1*2a, where n1 is the data size obtained after convolution by the convolution kernel, and a represents the number of channels of the convolution kernel. The feature extraction module is used to extract features from the spliced ​​data using a preset feature extraction model; The classification module is used to classify the extracted feature data using a preset classification function to obtain the classification result of the electrocardiogram to be identified.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores an electrocardiogram (ECG) signal image classification program that can be executed by the at least one processor, the ECG signal image classification program being executed by the at least one processor to enable the at least one processor to perform a deep learning-based ECG classification method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for classifying electrocardiogram (ECG) signal images, which can be executed by one or more processors to implement a deep learning-based ECG classification method as described in any one of claims 1 to 6.

Images