ECG signal classification training method, classification method, device and storage medium

Abstract

The present invention discloses an electrocardiogram signal classification training method, a classification method, a device and a storage medium. The classification training method comprises: obtaining an electrocardiogram signal in a training set; training the electrocardiogram signal based on a deep neural network to obtain a predicted probability value of each of at least two types of abnormal heart rhythm; extracting the electrocardiogram signal based on a signal feature value to obtain a target feature value of each of the at least two types of abnormal heart rhythm; based on the predicted probability value and the target feature value of each of the at least two types of abnormal heart rhythm, training a classifier for identifying each of the at least two types of abnormal heart rhythm. For different types of abnormal heart rhythm, a classifier can be established using different target feature values ​​and predicted probability values, which is conducive to fully imitating the diagnosis process of medical experts, effectively utilizing the feature signals corresponding to different types of abnormal heart rhythm, and effectively improving the accuracy of electrocardiogram signal classification.

Description

Technical Field

[0001] The present invention relates to the field of signal analysis, and in particular to an electrocardiogram signal classification training method, a classification method, a device and a storage medium. Background Art

[0002] In related technologies, deep learning of a large number of ECG signals can often be performed based on machine learning, so as to classify normal ECG and ECG with regular heart rate. Among them, the structure of the deep neural network often affects the quality of the classification results.

[0003] In the related art, when learning features based on deep neural networks, there is no targeted use of expert feature signals corresponding to different arrhythmias. Specifically, no matter what kind of arrhythmia is to be judged, the same set of expert feature signals is used. However, many judgments on expert feature signals are based on medical knowledge, and doctors will make judgments based on the actual situation of the patient when making actual diagnoses. Therefore, the boundary of whether the expert feature signal is abnormal is not strictly consistent with the boundary of medical knowledge, resulting in a high false positive rate of existing classifiers. Summary of the invention

[0004] In view of this, the embodiments of the present invention provide an electrocardiogram signal classification training method, a classification method, an apparatus and a storage medium, aiming to improve the accuracy of electrocardiogram signal classification.

[0005] The technical solution of the embodiment of the present invention is achieved as follows:

[0006] The embodiment of the present invention provides an electrocardiogram signal classification training method, comprising:

[0007] Get the ECG signal in the training set;

[0008] Training the electrocardiogram signal based on a deep neural network to obtain a predicted probability value for each of at least two types of abnormal heart rhythm;

[0009] Extracting signal feature values ​​of the electrocardiogram signal to obtain target feature values ​​of each of the at least two types of abnormal heart rhythm;

[0010] Based on the predicted probability value and the target feature value of each of the at least two types of abnormal heart rhythm, a classifier for identifying each of the at least two types of abnormal heart rhythm is trained.

[0011] The embodiment of the present invention also provides an electrocardiogram signal classification method, comprising:

[0012] Acquire the electrocardiogram signal to be identified;

[0013] The ECG signal to be identified is input into an ECG signal classification model trained by the ECG signal classification training method as described in the embodiment of the present invention. The ECG signal classification model determines the predicted probability value and target feature value of each type of at least two types of abnormal heart rhythm for the ECG signal to be identified, and obtains classification results of each type based on the predicted probability value, the target feature value and classifier of the same type.

[0014] The embodiment of the present invention also provides an electrocardiogram signal classification training device, comprising:

[0015] A first acquisition module is used to acquire the ECG signal in the training set;

[0016] A first training module is used to train the electrocardiogram signal based on a deep neural network to obtain a predicted probability value of each of at least two types of abnormal heart rhythm;

[0017] A feature extraction module, configured to extract a signal feature value from the electrocardiogram signal to obtain a target feature value for each of the at least two types of abnormal heart rhythm;

[0018] The second training module is used to train a classifier for identifying each of the at least two types of abnormal heart rhythm based on the predicted probability value and the target feature value of each of the at least two types of abnormal heart rhythm.

[0019] The embodiment of the present invention further provides an electrocardiogram signal classification device, comprising:

[0020] A second acquisition module is used to acquire the electrocardiogram signal to be identified;

[0021] A classification module is used to input the ECG signal to be identified into the ECG signal classification model trained by the ECG signal classification training device of the embodiment of the present invention, wherein the ECG signal classification model determines the predicted probability value and target feature value of each type of at least two types of abnormal heart rhythm for the ECG signal to be identified, and obtains classification results of each type based on the predicted probability value, target feature value and classifier of the same type.

[0022] An embodiment of the present invention also provides an ECG signal classification training device, comprising: a processor and a memory for storing a computer program that can be run on the processor, wherein the processor is used to execute the steps of the ECG signal classification training method described in the embodiment of the present invention when running the computer program.

[0023] An embodiment of the present invention further provides an ECG signal classification device, comprising: a processor and a memory for storing a computer program that can be run on the processor, wherein the processor is used to execute the steps of the ECG signal classification method described in the embodiment of the present invention when running the computer program.

[0024] An embodiment of the present invention further provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in any embodiment of the present invention are implemented.

[0025] The technical solution provided by the embodiment of the present invention trains the ECG signals in the training set based on a deep neural network to obtain predicted probability values ​​for each of at least two types of abnormal heart rhythms; and obtains target feature values ​​for each of the at least two types of abnormal heart rhythms based on the extraction of signal feature values; then, based on the predicted probability values ​​and target feature values ​​for each of the at least two types of abnormal heart rhythms, a classifier for identifying each of the at least two types of abnormal heart rhythms is trained, so that for different types of abnormal heart rhythms, different target feature values ​​and predicted probability values ​​can be used to establish a classifier, which is conducive to fully imitating the diagnosis process of medical experts, effectively utilizing the feature signals corresponding to different types of abnormal heart rhythms, and effectively improving the accuracy of ECG signal classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the process of the electrocardiogram signal classification training method according to an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of the principle of a central electrical signal classification training method according to an application example of the present invention;

[0028] Figure 3 A schematic diagram of the architecture of a multi-scale deep neural network in an application example of the present invention;

[0029] Figure 4 Schematic diagram of the process of the electrocardiogram signal classification method according to an embodiment of the present invention;

[0030] Figure 5 This is a schematic diagram of the structure of an electrocardiogram signal classification training device according to an embodiment of the present invention;

[0031] Figure 6 It is a structural schematic diagram of an electrocardiogram signal classification device according to an embodiment of the present invention;

[0032] Figure 7 This is a schematic diagram of the structure of an electrocardiogram signal classification training device according to an embodiment of the present invention;

[0033] Figure 8 It is a structural schematic diagram of an electrocardiogram signal classification device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0036] The embodiment of the present invention provides an electrocardiogram signal classification training method, such as Figure 1 The electrocardiogram signal training classification method comprises:

[0037] Step 101, obtaining the ECG signal in the training set;

[0038] Step 102, training the ECG signal based on a deep neural network to obtain a predicted probability value of each of at least two types of abnormal heart rhythm;

[0039] Step 103, extracting the signal feature value of the electrocardiogram signal to obtain target feature values ​​of each of the at least two types of abnormal heart rhythm;

[0040] Step 104: Based on the predicted probability value and the target feature value of each of the at least two types of abnormal heart rhythm, a classifier for identifying each of the at least two types of abnormal heart rhythm is trained.

[0041] In an embodiment of the present invention, the ECG signals in the training set are trained based on a deep neural network to obtain predicted probability values ​​for each of at least two types of abnormal heart rhythms; and based on the extraction of signal feature values, target feature values ​​for each of the at least two types of abnormal heart rhythms are obtained; then, based on the predicted probability values ​​and target feature values ​​for each of the at least two types of abnormal heart rhythms, a classifier for identifying each of the at least two types of abnormal heart rhythms is trained, so that for different types of abnormal heart rhythms, different target feature values ​​and predicted probability values ​​can be used to establish a classifier, which is conducive to fully imitating the diagnosis process of medical experts, effectively utilizing the feature signals corresponding to different types of abnormal heart rhythms, and can effectively improve the accuracy of ECG signal classification.

[0042] In some embodiments, before acquiring the ECG signals in the training set, the original ECG signals may be preprocessed.

[0043] Exemplarily, the multi-lead ECG signals may be preprocessed, wherein the preprocessing includes at least one of the following: normalization, baseline drift removal, high-frequency noise removal, and power frequency noise removal.

[0044] In some embodiments, the training of the ECG signal based on a deep neural network to obtain a predicted probability value of each of at least two types of abnormal heart rhythms includes:

[0045] Classifying the ECG signal based on a convolution module, a bidirectional GRU (Gated Recurrent Unit) module, a fully connected layer, and a classification module to obtain a predicted probability value of each of at least two types of abnormal heart rhythm;

[0046] Among them, the convolution module includes: at least two residual network branches and a channel attention mechanism (channel attention) unit corresponding to each residual network branch, the size of the convolution kernel in each residual network branch is different, and the channel attention mechanism unit is used to adjust the weight value of the residual network branch.

[0047] Here, due to the different sizes of the convolution kernels in the residual network branches, the ECG signal features of different scales can be sensed. In this way, it is possible to observe the waveform differences between different heartbeats within a period of time, and to observe whether the time intervals and relative amplitudes before each wave in a single heartbeat are within the normal medical range, so as to better capture the abnormal features of the ECG signal; in addition, the channel attention mechanism unit can assign different weights to the observation channels of the ECG signal at different scales, among which the branches that can better represent the ECG features have higher weights, so that the ECG signal features can be better learned; again, the combination of CNN (convolutional neural network) and GRU structure not only applies the powerful feature extraction function of CNN, but also utilizes the ability of RNN (Recurrent Neural Network) to extract the temporal relationship between the front and back of the ECG signal.

[0048] In some embodiments, the method further comprises:

[0049] The hidden layer nodes of the bidirectional GRU module are randomly screened based on a preset node ratio to determine the reserved nodes for training.

[0050] In practical applications, in order to prevent the model from overfitting, dropout can be performed during each training of the ECG signal, that is, the hidden layer nodes of the trained bidirectional GRU module are randomly screened according to the preset node ratio to obtain the reserved nodes for training the input signal. The preset node ratio can be reasonably adjusted according to the needs.

[0051] In some embodiments, the extracting of the signal feature value of the electrocardiogram signal to obtain the target feature value of each of the at least two types of abnormal heart rhythms includes:

[0052] Identify P wave, Q wave, R wave, S wave and T wave in the electrocardiogram signal;

[0053] Determining a signal characteristic value of the electrocardiogram signal based on the P wave, Q wave, R wave, S wave and T wave;

[0054] Based on the signal characteristic value, a target characteristic value of the electrocardiogram signal corresponding to each of the at least two types of abnormal heart rhythm is determined.

[0055] Here, the P wave is the wave that initially produces the deviation, which reflects the potential change of the atrial depolarization process and represents the depolarization of the two atria. Exemplarily, X_P can be used to represent the waveform sequence of the P wave (i.e., the time represented by the abscissa of the electrocardiogram), and ecg(X_P) can be used to represent the electrocardiogram signal amplitude sequence of the P wave (i.e., the voltage amplitude represented by the ordinate of the electrocardiogram).

[0056] Here, the Q wave is a wave that is downward (negative) before the R wave. Exemplarily, X_Q can be used to represent the waveform sequence of the Q wave, and ecg(X_Q) can be used to represent the electrocardiogram signal amplitude sequence of the Q wave.

[0057] Here, the R wave is the first upward (positive) wave. Exemplarily, X_R can be used to represent the waveform sequence of the R wave, and ecg(X_R) can be used to represent the electrocardiogram signal amplitude sequence of the R wave.

[0058] Here, the S wave is the first downward wave after the R wave. Exemplarily, X_S may be used to represent the waveform sequence of the S wave, and ecg(X_S) may be used to represent the electrocardiographic signal amplitude sequence of the S wave.

[0059] Here, the T wave is a wave formed by the current generated by the rapid repolarization process of the ventricle, representing the potential change caused by the repolarization process of the ventricular muscle; it appears after the QRS complex. Exemplarily, X_T can be used to represent the waveform sequence of the T wave, and ecg(X_T) can be used to represent the electrocardiographic signal amplitude sequence of the T wave.

[0060] For example, the positions of typical waves such as P, Q, R, S, and T in each heartbeat can be identified, and then the width, amplitude, interval, etc. of each wave group can be calculated according to the parameter target to be calculated, and then the expert characteristic signal value (i.e., signal characteristic value) directly related to the diagnosis of arrhythmia that doctors pay more attention to during diagnosis can be calculated, including but not limited to QT interval, QT interval corrected by heart rate, QRS interval, PP interval, RR interval, P wave width, electric axis degree, SV1, RV5, ventricular rate, atrial rate, etc. Among them, QT interval is the total time from the beginning of depolarization to the completion of repolarization of the entire ventricular muscle; QRS interval is the depolarization time of the ventricular muscle; PP interval is the sinus cycle, i.e., the time between two P waves; RR interval is the ventricular cycle, i.e., the time between two R waves; SV1 represents the S wave of the V1 lead; RV5 represents the R wave of the V5 lead.

[0061] In some embodiments, the at least two types of abnormal heart rhythms include: at least two of: premature beats, conduction block, atrial fibrillation, atrial flutter, tachycardia, bradycardia and arrhythmia; the target feature value and the classifier both correspond one-to-one to each type of the at least two types of abnormal heart rhythms.

[0062] The electrocardiogram signal classification training method according to the embodiment of the present invention is further described in detail below with reference to application examples.

[0063] like Figure 2 As shown, the ECG signal classification training method includes:

[0064] 1) Pretreatment;

[0065] Here, the multi-lead ECG signal (ie, the original ECG signal) may be preprocessed, including normalization, removal of baseline drift, removal of high-frequency noise, removal of power frequency noise, and the like.

[0066] 2) Deep neural network training part;

[0067] The preprocessed ECG signal dataset is trained through a multi-classification and multi-scale deep neural network to obtain a multi-classification model that can distinguish between normal and multiple abnormal heart rhythms. The multiple classifications include but are not limited to normal, premature beats, conduction block, atrial fibrillation, atrial flutter, tachycardia, bradycardia, arrhythmia, and others.

[0068] For example, the specific architecture of the multi-scale deep neural network (ResAttentionGRUNet) is as follows: Figure 3As shown. Among them, the multi-lead ECG signal can be 12-lead or 8-lead, and each lead is a one-dimensional signal. In this application example, the input signal (INPUT) is also processed as a one-dimensional signal, different leads are input as different channels, and subsequent convolution models are all one-dimensional. Exemplarily, the input signal first enters 4 (this number is not limited, 4 is taken as an example) residual network branches (i.e., Res1 to Res4), each residual network branch is a convolution module composed of 6 one-dimensional residual blocks, and a channel attention module is subsequently added, in which the convolution kernel of each branch has a different size (in this example, 2, 4, 8, and 16, respectively) to sense the ECG signal characteristics of different scales. Because the diagnosis of ECG signals requires not only observing the waveform differences between different heartbeats within a period of time, but also observing whether the time intervals and relative amplitudes between waves in a single heartbeat are within the normal medical range, the deep neural network in this application example, the channelattention module, can give different weights to channels of ECG signals at different scales, so that the weight of channels that can better represent ECG features is higher. In this way, multiple residual network branches constitute a multi-scale module, and then the features obtained after multi-scale processing by multiple branches are fused, that is, through Figure 3 The concat layer, Dense layer, and Average pooling layer shown in the figure are used to obtain the fused features, which are then processed through a multi-layer bidirectional GRU (BiGRU) module ( Figure 3 It is shown as two layers, and more layers can be set according to needs in actual applications), Dense layer, Averagepooling layer, and finally multi-classification is performed through softmax to obtain the prediction probability value of each type (Multi-classes probability). For example, in order to prevent the model from overfitting, dropout can be performed in each training process of the ECG signal, that is, the hidden layer nodes of the trained GRU neural network are randomly screened according to the preset node ratio to obtain the retained nodes for training the input signal.

[0069] The multi-scale deep neural network in this application example has the following advantages: 1. Multi-scale fusion can combine the long-term and short-term features of the ECG to better capture the abnormal features of the ECG signal; 2. The application of the channelattention mechanism indirectly interprets the importance of ECG signals of different leads by assigning different weights to different channels; 3. The combination of CNN and GRU structures not only applies the powerful feature extraction function of CNN, but also utilizes the ability of RNN to extract the temporal relationship between ECG signals.

[0070] 3) Characteristic signal calculation part;

[0071] The preprocessed signal identifies the positions of typical waves such as P, Q, R, S, T in each heartbeat, and then calculates the width, amplitude, interval, etc. of each wave group according to the parameter target to be calculated, and then calculates the expert characteristic signal values ​​that doctors are more concerned about during diagnosis and are directly related to the diagnosis of arrhythmia, including but not limited to QT interval, QT interval corrected by heart rate, PT interval, QRS interval, P wave width, electrical axis degree, SV1, RV5, ventricular rate, atrial rate, etc.

[0072] 4) Training classifiers by combining traditional feature values ​​with deep neural network specificity;

[0073] Each ECG signal can obtain the probability value of each type of arrhythmia through the multi-classification and multi-scale deep neural network model trained in 2). At the same time, various characteristic signal values ​​(such as QT interval, QT interval corrected by heart rate, PT interval, QRS interval, P wave width, electrical axis degree, SV1, RV5, ventricular rate, atrial rate, etc.) can be obtained through calculation in 3).

[0074] Then, according to the characteristic signal value corresponding to each type of arrhythmia and the predicted probability value of the class, a machine learning classification model (such as XGBoost, random forest, etc.) of the class is established. If there are N types of abnormal heart rhythm, a total of N classifiers are established. For example, when constructing an atrioventricular block classifier, the probability of atrioventricular block of each ECG, plus the PR interval, atrial rate, ventricular rate, P wave number / R wave number ratio, and the maximum and minimum mean of the RR interval in the characteristic signal value, are taken as the input of the machine learning model to train the classifier of whether there is atrioventricular block; for another example, when constructing an arrhythmia classifier, the probability of arrhythmia of each ECG, plus the atrial rate, ventricular rate, the maximum and minimum mean of the PP interval, and the maximum and minimum mean of the RR interval in the characteristic signal value, are taken as the input of the machine learning model to train the classifier of whether there is an arrhythmia; for another example, when constructing an atrial fibrillation classifier, the probability of atrial fibrillation of each ECG, plus the number of P waves, P wave amplitude, atrial rate, ventricular rate, the maximum and minimum mean of the RR interval, Q wave amplitude, R wave amplitude, and S wave amplitude in the characteristic signal value, are taken as the input of the machine learning model to train the classifier of whether there is atrial fibrillation.

[0075] In this way, the training process of the ECG signal classification model according to the embodiment of the present invention is completed.

[0076] The embodiment of the present invention also provides a method for classifying electrocardiogram signals. Figure 4 Said, including:

[0077] Step 401, obtaining an electrocardiogram signal to be identified;

[0078] Step 402, input the ECG signal to be identified into an ECG signal classification model trained by an ECG signal classification training method, the ECG signal classification model determines the predicted probability value and target feature value of each type of at least two types of abnormal heart rhythm for the ECG signal to be identified, and obtains classification results of each type based on the predicted probability value, the target feature value and the classifier of the same type.

[0079] In the embodiment of the present invention, the ECG signal classification training method described in the above embodiment is used to train the ECG signal classification model. Since the method of establishing the classifier by jointly using the target feature value and the predicted probability of the deep neural network enables the ECG signal classification model to automatically learn the contribution of the expert feature signal to the presence or absence of a certain type of abnormal heart rhythm through big data, and automatically learn the weights of different parts, which can give full play to the advantages of machine learning. Compared with the method of judging feature signals by threshold rule method in the previous technology, it is more suitable for practical application scenarios based on ECG signals and improves the accuracy of ECG signal classification.

[0080] In some embodiments, the judgment result of the ECG signal to be identified may be the union of the classification results of each abnormal heart rhythm type judged. If the judgment of each abnormal heart rhythm type is negative, the ECG signal belongs to the normal class.

[0081] In order to implement the method of the embodiment of the present invention, the embodiment of the present invention also provides an ECG signal classification training device, which corresponds to the above-mentioned ECG signal classification training method, and each step in the above-mentioned ECG signal classification training method embodiment is also fully applicable to the present ECG signal classification training device embodiment.

[0082] like Figure 5 As shown, the ECG signal classification training device includes: a first acquisition module 501, a first training module 502, a feature extraction module 503 and a second training module 504; wherein the first acquisition module 501 is used to acquire ECG signals in a training set; the first training module 502 is used to train the ECG signals based on a deep neural network to obtain predicted probability values ​​for each of at least two types of abnormal heart rhythms; the feature extraction module 503 is used to extract signal feature values ​​from the ECG signals to obtain target feature values ​​for each of the at least two types of abnormal heart rhythms; the second training module 504 is used to train a classifier for identifying each of the at least two types of abnormal heart rhythms based on the predicted probability values ​​and the target feature values ​​for each of the at least two types of abnormal heart rhythms.

[0083] In some embodiments, the first training module 502 is specifically used to:

[0084] Classifying the ECG signal based on a convolution module, a bidirectional GRU module, a fully connected layer, and a classification module to obtain a predicted probability value of each of at least two types of abnormal heart rhythm;

[0085] Among them, the convolution module includes: at least two residual network branches and a channel attention mechanism unit corresponding to each residual network branch, the size of the convolution kernel in each residual network branch is different, and the channel attention mechanism unit is used to adjust the weight value of the residual network branch.

[0086] In some embodiments, the first training module 502 is further configured to:

[0087] The hidden layer nodes of the bidirectional GRU module are randomly screened based on a preset node ratio to determine the reserved nodes for training.

[0088] In some embodiments, the feature extraction module 503 is specifically used to:

[0089] Identify P wave, Q wave, R wave, S wave and T wave in the electrocardiogram signal;

[0090] Determining a signal characteristic value of the electrocardiogram signal based on the P wave, Q wave, R wave, S wave and T wave;

[0091] Based on the signal characteristic value, a target characteristic value of the electrocardiogram signal corresponding to each of the at least two types of abnormal heart rhythm is determined.

[0092] In some embodiments, the at least two types of abnormal heart rhythms include: at least two of: premature beats, conduction block, atrial fibrillation, atrial flutter, tachycardia, bradycardia and arrhythmia; the target feature value and the classifier both correspond one-to-one to each type of the at least two types of abnormal heart rhythms.

[0093] In actual application, the first acquisition module 501, the first training module 502, the feature extraction module 503 and the second training module 504 can be implemented by a processor in the electrocardiogram signal classification training device. Of course, the processor needs to run the computer program in the memory to implement its function.

[0094] It should be noted that: the ECG signal classification training device provided in the above embodiment only uses the division of the above program modules as an example when performing ECG signal classification training. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the processing described above. In addition, the ECG signal classification training device provided in the above embodiment and the ECG signal classification training method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0095] In order to implement the method of the embodiment of the present invention, the embodiment of the present invention also provides an ECG signal classification device, which corresponds to the above-mentioned ECG signal classification method, and each step in the above-mentioned ECG signal classification method embodiment is also completely applicable to the embodiment of the ECG signal classification device.

[0096] like Figure 6 As shown, the ECG signal classification device includes: a second acquisition module 601 and a classification module 602, wherein the second acquisition module 601 is used to acquire the ECG signal to be identified; the classification module 602 inputs the ECG signal to be identified into an ECG signal classification model trained by the ECG signal classification training device as described in the aforementioned embodiment, and the ECG signal classification model determines the predicted probability value and target feature value of each type of the ECG signal to be identified in at least two types of abnormal heart rhythm, and obtains classification results of each type based on the predicted probability value, the target feature value and the classifier of the same type.

[0097] In actual application, the second acquisition module 601 and the classification module 602 can be implemented by a processor in the electrocardiogram signal classification device. Of course, the processor needs to run the computer program in the memory to realize its function.

[0098] It should be noted that: the ECG signal classification device provided in the above embodiment only uses the division of the above program modules as an example when performing ECG signal classification. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the processing described above. In addition, the ECG signal classification device provided in the above embodiment and the ECG signal classification method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0099] Based on the hardware implementation of the above program modules and in order to implement the method of the embodiment of the present invention, the embodiment of the present invention further provides an electrocardiogram signal classification training device. Figure 7 Only an exemplary structure of the electrocardiogram signal classification training device is shown, not all structures, and it can be implemented as needed. Figure 7 Partial or complete structure shown.

[0100] like Figure 7As shown, the ECG signal classification training device 700 provided in the embodiment of the present invention includes: at least one processor 701, a memory 702, a user interface 703 and at least one network interface 704. The various components in the ECG signal classification training device 700 are coupled together through a bus system 705. It can be understood that the bus system 705 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 705 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 705 is not used in the embodiment of the present invention. Figure 7 Various buses are labeled as bus system 705.

[0101] The user interface 703 may include a display, a keyboard, a mouse, a trackball, a click wheel, keys, buttons, a touch pad or a touch screen.

[0102] The memory 702 in the embodiment of the present invention is used to store various types of data to support the operation of the electrocardiogram signal classification training device. Examples of such data include: any computer program used to operate on the electrocardiogram signal classification training device.

[0103] The ECG signal classification training method disclosed in the embodiment of the present invention can be applied to the processor 701, or implemented by the processor 701. The processor 701 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the ECG signal classification training method can be completed by the hardware integrated logic circuit in the processor 701 or the instructions in the form of software. The above-mentioned processor 701 can be a general-purpose processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 701 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor are combined to execute. The software module can be located in a storage medium, which is located in the memory 702. The processor 701 reads the information in the memory 702 and completes the steps of the ECG signal classification training method provided in the embodiment of the present invention in combination with its hardware.

[0104] In an exemplary embodiment, the ECG signal classification training device 700 can be implemented by one or more application-specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD), FPGA, general-purpose processor, controller, microcontroller (MCU), microprocessor, or other electronic components to execute the aforementioned method.

[0105] Based on the hardware implementation of the above program modules and in order to implement the method of the embodiment of the present invention, the embodiment of the present invention further provides an electrocardiogram signal classification device. Figure 8 Only an exemplary structure of the electrocardiogram signal classification device is shown, not all structures, and it can be implemented as needed. Figure 8 Partial or complete structure shown.

[0106] like Figure 8 As shown, the ECG signal classification device 800 provided in the embodiment of the present invention includes: at least one processor 801, a memory 802, a user interface 803 and at least one network interface 804. The various components in the ECG signal classification device 800 are coupled together through a bus system 805. It can be understood that the bus system 805 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 805 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, Figure 8 Various buses are labeled as bus system 805 .

[0107] The user interface 803 may include a display, a keyboard, a mouse, a trackball, a click wheel, keys, buttons, a touch pad or a touch screen.

[0108] The memory 802 in the embodiment of the present invention is used to store various types of data to support the operation of the electrocardiogram signal classification device. Examples of such data include: any computer program used to operate on the electrocardiogram signal classification device.

[0109] The ECG signal classification training method disclosed in the embodiment of the present invention can be applied to the processor 801, or implemented by the processor 801. The processor 801 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the ECG signal classification training method can be completed by the hardware integrated logic circuit or software instructions in the processor 801. The above-mentioned processor 801 can be a general-purpose processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 801 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general-purpose processor can be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in the memory 802. The processor 801 reads the information in the memory 802 and completes the steps of the ECG signal classification method provided in the embodiment of the present invention in combination with its hardware.

[0110] In an exemplary embodiment, the ECG signal classification device 800 may be implemented by one or more ASICs, DSPs, PLDs, CPLDs, FPGAs, general purpose processors, controllers, MCUs, Microprocessors, or other electronic components to perform the aforementioned method.

[0111] It can be understood that the memory 702, 802 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM, SyncLink Dynamic Random Access Memory), and direct RAM bus random access memory (DRRAM, Direct Rambus Random Access Memory).The memories described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memories.

[0112] In an exemplary embodiment, the present invention also provides a storage medium, namely a computer storage medium, which can be a computer-readable storage medium, for example, a memory 702 storing a computer program, which can be executed by a processor 701 of an ECG signal classification training device 700 to complete the steps described in the ECG signal classification training method of the present invention; for another example, a memory 802 storing a computer program, which can be executed by a processor 801 of an ECG signal classification device 800 to complete the steps described in the ECG signal classification method of the present invention. The computer-readable storage medium can be a memory such as a ROM, a PROM, an EPROM, an EEPROM, a Flash Memory, a magnetic surface memory, an optical disk, or a CD-ROM.

[0113] It should be noted that: "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0114] In addition, the technical solutions described in the embodiments of the present invention can be arbitrarily combined without conflict.

[0115] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for training electrocardiogram signal classification, characterized in that: include: Get the ECG signal in the training set; Training the electrocardiogram signal based on a deep neural network to obtain a predicted probability value for each of at least two types of abnormal heart rhythm; Extracting signal feature values ​​of the electrocardiogram signal to obtain target feature values ​​of each of the at least two types of abnormal heart rhythm; Based on the predicted probability value and the target feature value of each of the at least two types of abnormal heart rhythm, a classifier for identifying each of the at least two types of abnormal heart rhythm is trained; the target feature value and the classifier both correspond one-to-one to each of the at least two types of abnormal heart rhythm; The extracting of the signal feature value of the electrocardiogram signal to obtain the target feature value of each type of the at least two types of abnormal heart rhythms includes: Identify P wave, Q wave, R wave, S wave and T wave in the electrocardiogram signal; Determining a signal characteristic value of the electrocardiogram signal based on the P wave, Q wave, R wave, S wave and T wave; Based on the signal characteristic value, a target characteristic value of the electrocardiogram signal corresponding to each of the at least two types of abnormal heart rhythm is determined.

2. The method according to claim 1, characterized in that The step of training the electrocardiogram signal based on a deep neural network to obtain a predicted probability value of each of at least two types of abnormal heart rhythm includes: Classifying the ECG signal based on a convolution module, a bidirectional gated recurrent unit (GRU) module, a fully connected layer, and a classification module to obtain a predicted probability value of each of at least two types of abnormal heart rhythm; Among them, the convolution module includes: at least two residual network branches and a channel attention mechanism unit corresponding to each residual network branch, the size of the convolution kernel in each residual network branch is different, and the channel attention mechanism unit is used to adjust the weight value of the residual network branch.

3. The method according to claim 2, characterized in that The method further comprises: The hidden layer nodes of the bidirectional gated recurrent unit (GRU) module are randomly screened based on a preset node ratio to determine the reserved nodes for training.

4. The method according to claim 1, characterized in that: The at least two types of abnormal heart rhythm include: at least two of: premature beats, conduction block, atrial fibrillation, atrial flutter, tachycardia, bradycardia and arrhythmia.

5. A method for classifying electrocardiogram signals, characterized in that: include: Acquire the electrocardiogram signal to be identified; The ECG signal to be identified is input into an ECG signal classification model trained by the ECG signal classification training method as described in any one of claims 1 to 4, wherein the ECG signal classification model determines the predicted probability value and target feature value of each type of at least two types of abnormal heart rhythm for the ECG signal to be identified, and obtains classification results of each type based on the predicted probability value, the target feature value and classifier of the same type.

6. An electrocardiogram signal classification training device, characterized in that: include: A first acquisition module is used to acquire the ECG signal in the training set; A first training module is used to train the electrocardiogram signal based on a deep neural network to obtain a predicted probability value of each of at least two types of abnormal heart rhythm; A feature extraction module, configured to extract a signal feature value from the electrocardiogram signal to obtain a target feature value for each of the at least two types of abnormal heart rhythm; A second training module is used for training a classifier for identifying each of the at least two types of abnormal heart rhythm based on the predicted probability value and the target feature value of each of the at least two types of abnormal heart rhythm; the target feature value and the classifier both correspond one-to-one to each of the at least two types of abnormal heart rhythm; Wherein, the feature extraction module is specifically used for: Identify P wave, Q wave, R wave, S wave and T wave in the electrocardiogram signal; Determining a signal characteristic value of the electrocardiogram signal based on the P wave, Q wave, R wave, S wave and T wave; Based on the signal characteristic value, a target characteristic value of the electrocardiogram signal corresponding to each of the at least two types of abnormal heart rhythm is determined.

7. An electrocardiogram signal classification device, characterized in that: include: A second acquisition module is used to acquire the electrocardiogram signal to be identified; A classification module inputs the ECG signal to be identified into an ECG signal classification model trained by the ECG signal classification training device as described in claim 6, wherein the ECG signal classification model determines the predicted probability value and target feature value of each type of at least two types of abnormal heart rhythm for the ECG signal to be identified, and obtains classification results of each type based on the predicted probability value, the target feature value and the classifier of the same type.

8. An electrocardiogram signal classification training device, characterized in that: include: A processor and a memory for storing a computer program that can be executed on the processor, wherein: The processor is used to execute the steps of the method according to any one of claims 1 to 4 when running a computer program.

9. An electrocardiogram signal classification device, characterized in that: include: A processor and a memory for storing a computer program that can be executed on the processor, wherein: The processor is used to execute the steps of the method according to claim 5 when running a computer program.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images