CN115886833A - Electrocardiosignal classification method and device, computer readable medium and electronic equipment - Google Patents

Abstract

The present disclosure specifically relates to the field of computer technology, in particular to a method and device for classifying electrocardiographic signals, a computer-readable medium, and electronic equipment. The method includes: preprocessing the ECG signal to be processed, and obtaining two-dimensional images corresponding to each cardiac beat; inputting the two-dimensional images into an ECG signal classification model based on a VGG network, and using several continuous feature extraction units to analyze the The two-dimensional image is subjected to feature extraction, and a feature map is output; wherein, the feature extraction unit includes: a convolutional layer based on Ghost Module, a maximum pooling layer; a classification unit is used to classify according to the feature map, and obtain the electrocardiogram to be processed The classification result corresponding to the signal. This solution can realize a lightweight ECG signal classification model structure, and can be easily deployed on edge portable devices with limited processing capabilities.

Description

Electrocardiogram signal classification method and device, computer readable medium, and electronic device

Technical Field

The present disclosure relates to the field of computer technology, and in particular to an electrocardiogram signal classification method, an electrocardiogram signal classification device, a computer-readable medium, and an electronic device.

Background Art

Electrocardiogram (ECG) is the most widely used clinical method for diagnosing cardiovascular diseases. In related technologies, in order to ensure the performance of ECG signal classification, most ECG signal classification methods based on deep learning are used. In order to improve the accuracy of the model, related algorithms usually use computationally intensive network models, which require high computing power and memory capacity. However, such a solution is not suitable for wearable devices with limited capacity and processing power in edge computing scenarios.

It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field.

Summary of the invention

The present disclosure provides an electrocardiogram signal classification method, an electrocardiogram signal classification device, a computer-readable medium, and an electronic device, which can effectively overcome the defects of the prior art such as complex models, high computational complexity, and inability to be applied to wearable devices with limited capacity and processing power in edge computing scenarios.

Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

According to a first aspect of the present disclosure, a method for classifying an electrocardiogram signal is provided, the method comprising:

Preprocess the ECG signal to be processed to obtain a two-dimensional image corresponding to each heartbeat;

Input the two-dimensional image into an electrocardiogram signal classification model based on a VGG network, extract features from the two-dimensional image using a plurality of continuous feature extraction units, and output a feature map; wherein the feature extraction unit includes: a convolution layer and a maximum pooling layer based on a Ghost Module;

The classification unit is used to perform classification according to the feature map to obtain the classification result corresponding to the electrocardiogram signal to be processed.

In some exemplary embodiments, preprocessing the electrocardiogram signal to be processed to obtain a two-dimensional image corresponding to each heartbeat includes:

Segment the ECG signal to be processed and obtain the corresponding heart beat signal;

The heartbeat signal is converted to obtain a two-dimensional image in grayscale format corresponding to the heartbeat signal.

In some exemplary embodiments, the method further comprises: training the ECG signal classification model, comprising:

Collecting marked ECG sample data, and slicing the ECG sample data according to the R wave peak value to obtain heart beat sample data;

Convert the heartbeat sample data to obtain corresponding two-dimensional image sample data as training sample data;

Dividing the training sample data to obtain corresponding training sets, validation sets and test sets;

Input the training set into the ECG signal classification model based on the VGG network to perform model training until the model converges; and input the verification set into the ECG signal classification model to verify the model;

The test set data is input into the trained ECG signal classification model, and the corresponding ECG signal classification result is output.

In some exemplary embodiments, the method further comprises:

Cropping the two-dimensional image sample data according to a preset cropping rule to obtain a cropped image;

The cropped image is resized to obtain enhanced image sample data, which is used as the training sample data.

In some exemplary embodiments, the VGG network-based ECG signal classification model includes a plurality of feature extraction units arranged in series, and a classification unit; wherein the feature extraction unit includes at least two Ghost Module-based convolutional layers and a maximum pooling layer arranged in sequence; the classification unit includes a first fully connected layer and a second fully connected layer arranged in sequence.

In some exemplary embodiments, a dropout layer is provided between the first fully connected layer and the second fully connected layer.

In some exemplary embodiments, the type of the ECG signal includes any one of a NOR type, a PVC type, an APC type, a RBBB type, and a LBBB type.

According to a second aspect of the present disclosure, there is provided an electrocardiogram signal classification device, comprising:

A preprocessing module, used to preprocess the ECG signal to be processed and obtain a two-dimensional image corresponding to each heartbeat;

A feature extraction module, used to input the two-dimensional image into an ECG signal classification model based on a VGG network, perform feature extraction on the two-dimensional image using a plurality of consecutive feature extraction units, and output a feature map; wherein the feature extraction unit includes: a convolution layer and a maximum pooling layer based on a Ghost Module;

The classification result output module is used to use the classification unit to perform classification according to the feature map to obtain the classification result corresponding to the electrocardiogram signal to be processed.

According to a third aspect of the present disclosure, a computer-readable medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned electrocardiogram signal classification method is implemented.

According to a fourth aspect of the present disclosure, there is provided an electronic device, including:

Processor; and

A memory, configured to store executable instructions of the processor;

Wherein, the processor is configured to implement the above-mentioned electrocardiogram signal classification method by executing the executable instructions.

An ECG signal classification method provided by an embodiment of the present disclosure obtains a two-dimensional image corresponding to each heartbeat by preprocessing the ECG signal to be processed; after inputting the two-dimensional image into an ECG signal classification model based on a VGG network, a plurality of continuous feature extraction units are used to extract features from the two-dimensional image and output a feature map; and then a classification unit is used to classify according to the feature map to obtain a classification result corresponding to the ECG signal to be processed. By using the two-dimensional image of the ECG signal as the input of the model and using the ECG signal classification model based on a VGG network, it is not necessary to perform noise filtering on the ECG signal to be processed, and a lightweight ECG signal classification model structure can be realized, which can be easily deployed on edge portable devices with limited processing capabilities.

It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

FIG1 schematically shows a schematic diagram of an electrocardiogram signal classification method in an exemplary embodiment of the present disclosure;

FIG2 schematically shows a schematic diagram of a model architecture of an electrocardiogram classification model according to an exemplary embodiment of the present disclosure;

FIG3 schematically shows a schematic diagram of a change state of a loss function for training an electrocardiogram signal classification model according to an exemplary embodiment of the present disclosure;

FIG4 schematically shows a schematic diagram of a model accuracy change state of an electrocardiogram signal classification model training in an exemplary embodiment of the present disclosure;

FIG5 schematically shows a schematic diagram of a confusion matrix of classification results of an electrocardiogram signal classification model in an exemplary embodiment of the present disclosure;

FIG6 schematically shows a composition diagram of an electrocardiogram signal classification device in an exemplary embodiment of the present disclosure;

FIG. 7 schematically shows a composition diagram of an electronic device in an exemplary embodiment of the present disclosure.

DETAILED DESCRIPTION

Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and/or processor devices and/or microcontroller devices.

In view of the shortcomings and deficiencies of the prior art, this example embodiment provides an ECG signal classification method that can be applied to low-configuration, low-power wearable smart electronic devices such as mobile phones, sports watches, and smart bracelets. Referring to FIG1 , the ECG signal analysis method may include:

Step S11, preprocessing the ECG signal to be processed to obtain a two-dimensional image corresponding to each heartbeat;

Step S12, inputting the two-dimensional image into an ECG signal classification model based on a VGG network, extracting features from the two-dimensional image using a plurality of consecutive feature extraction units, and outputting a feature map; wherein the feature extraction unit includes: a convolution layer and a maximum pooling layer based on a Ghost Module;

Step S13, using a classification unit to perform classification according to the feature graph, to obtain a classification result corresponding to the electrocardiogram signal to be processed.

The ECG signal classification method provided in this example implementation can simplify the preprocessing process of the ECG signal. After the ECG signal is segmented into heartbeats, it is converted into a corresponding two-dimensional image and used as the input data of the ECG signal classification model; the trained ECG signal classification model based on the VGG network is used for classification to obtain the classification result of the ECG signal to be processed. By using the two-dimensional image as the input data of the model, additional noise filtering and feature extraction steps are avoided. In addition, by using the ECG signal classification model based on the VGG network for classification, the Ghost-VGG-16 network is more suitable for edge computing devices with limited performance, and can achieve an average accuracy of 99.74% at a lower computing and memory cost.

In the following, each step of the electrocardiogram signal classification method in this exemplary implementation will be described in more detail with reference to the accompanying drawings and embodiments.

In step S11, the electrocardiographic signal to be processed is preprocessed to obtain a two-dimensional image corresponding to each heartbeat.

In this example implementation, the above step S11 may include:

Step S111, segmenting the ECG signal to be processed to obtain the corresponding heart beat signal;

Step S112: convert the heartbeat signal to obtain a two-dimensional image in grayscale format corresponding to the heartbeat signal.

Specifically, the ECG signal to be processed may be an ECG signal collected in real time by a smart wearable device worn by a user. For example, it may be an ECG signal within a period of time. For example, the ECG signal collection and classification cycle may be configured in advance on the smart wearable device; and the ECG signal for a period of time may be collected according to the specified data collection cycle, and the corresponding ECG signal classification task may be generated and added to the task list. For example, the ECG signal collection cycle may be 10 seconds, 30 seconds, or 60 seconds, etc., and the specific duration may be set according to the user-defined duration.

For the collected ECG signal to be processed, the heartbeat segmentation can be performed to obtain the corresponding heartbeat signal, and then the heartbeat signal can be converted into the corresponding two-dimensional grayscale image. Specifically, the duration of a complete ECG signal heartbeat cycle is between 0.6s and 0.8s, and the number of sampling points of a complete heartbeat is between 216 and 288. Each ECG signal beat is sliced according to the R wave peak time. Taking the R wave as the reference, 150 and 149 points are taken on the left and right sides, and a total of 300 points are taken as a heartbeat. For the heartbeat signal, the ECG signal can be converted into an ECG image using 2D-CNN (two-dimensional convolutional neural network). 2D-CNN and pooling layers are more suitable for extracting the spatial locality of ECG images, which can also solve the problem that some ECG beats are easily ignored in noise filtering and feature extraction. The intercepted one-dimensional ECG data is converted into a single two-dimensional 128*128 grayscale image. After obtaining the ECG data, there is no need to perform noise filtering and feature extraction steps.

In step S12, the two-dimensional image is input into an ECG signal classification model based on a VGG network, and features of the two-dimensional image are extracted using a plurality of consecutive feature extraction units to output a feature map; wherein the feature extraction unit includes: a convolution layer and a maximum pooling layer based on a Ghost Module.

In this example implementation, specifically, with reference to FIG2, the ECG classification model 20 may include five feature extraction units (a first feature extraction unit 201, a second feature extraction unit 202, a third feature extraction unit 203, a fourth feature extraction unit 204, and a fifth feature extraction unit 205) and a classification unit. Each feature extraction unit may include at least two convolutional layers based on Ghost Module and a maximum pooling layer. For example, with reference to Table 1, the first feature extraction unit and the second feature extraction unit may include two convolutional layers based on Ghost Module and a Maxpooling maximum pooling layer; the third feature extraction unit, the fourth feature extraction unit, and the fifth feature extraction unit may include three convolutional layers based on Ghost Module and a Maxpooling maximum pooling layer. The classification unit may include a first fully connected layer FC and a second fully connected layer FC. Among them, the first fully connected layer contains only 512 neurons, and the second fully connected layer is also the output layer, which uses the Softmax function to classify the data and output the classification results.

Table 1

In step S13, a classification unit is used to perform classification according to the feature map to obtain a classification result corresponding to the electrocardiogram signal to be processed.

In this example implementation, referring to the model structure shown in Table 1, the classification unit may include two first fully connected layers and a second fully connected layer that are consecutively arranged. The fully connected layer may use a Softmax function to classify the data and output the classification result of the ECG signal.

In addition, a dropout layer may be provided between the first fully connected layer and the second fully connected layer to perform a dropout operation and discard the number of neurons in the fully connected layer with a probability of 0.5 to avoid overfitting.

In this example implementation, the type of the ECG signal includes any one of a normal fluctuation (NOR) type, a premature ventricular contraction (PVC) type, a premature atrial contraction (APC) type, a right bundle branch block beat (RBBB) type, and a left bundle branch block beat (LBBB) type.

In this example implementation, the above method may further include: training the electrocardiogram signal classification model. Specifically, the training model may include:

Step S31, collecting the marked ECG sample data, and slicing the ECG sample data according to the R wave peak value to obtain heart beat sample data;

Step S32, converting the heartbeat sample data to obtain corresponding two-dimensional image sample data as training sample data;

Step S33, dividing the training sample data to obtain corresponding training sets, validation sets and test sets;

Step S34, inputting the training set into the ECG signal classification model based on the VGG network to perform model training until the model converges; and inputting the verification set into the ECG signal classification model to verify the model;

Step S35, inputting the test set data into the trained ECG signal classification model, and outputting the corresponding ECG signal classification result.

In this example implementation, specifically, in the above-mentioned step S31, the labeled ECG sample data can be collected from the MIT-BIH arrhythmia database of the Massachusetts Institute of Technology. The database contains different beat types obtained from 48 30-minute dual-channel dynamic electrocardiogram records of 47 volunteers. Two or more independent ECG experts diagnose and mark all recorded heartbeats to ensure the correctness of the results. Because the sampling frequency of the MIT-BIH database is 360Hz, the duration of a complete ECG signal heartbeat cycle is between 0.6s and 0.8s, and the number of sampling points for a complete heartbeat is between 216 and 288. Each ECG beat is sliced according to the R wave peak time, and 150 and 149 points are taken on the left and right sides respectively, with a total of 300 points as a heartbeat. Based on the time information, the range of a single ECG beat can be defined as follows:

T(R _peak (n)-150)≤T(n)≤T(R _peak (n)+149)

In this example implementation, in the above-mentioned step S32, a two-dimensional convolutional neural network can be used to convert the ECG signal into an ECG image form because the 2D-CNN and pooling layer are more suitable for extracting the spatial locality of the ECG image, which can also solve the problem that some ECG beats are easily ignored in noise filtering and feature extraction. The intercepted one-dimensional ECG data is converted into a single two-dimensional 128*128 grayscale image. After obtaining the ECG data, no noise filtering and feature extraction steps are required. The MIT-BIH database contains approximately 110,000 ECG records, from which 45,987 images are selected as experimental samples. The experimental data set includes five ECG beat types, namely NOR, PVC, APC, RBBB and LBBB.

In this example implementation, in the above step S33, all data can be divided into three parts: training set, validation set and test set, where 60% of the data is used to train the neural network, 20% as the validation set, and 20% for the test phase. The experimental data set division is shown in Table 2.

Table 2

In addition, in this example implementation, the method may further include: cropping the two-dimensional image sample data according to a preset cropping rule to obtain a cropped image; and resizing the cropped image to obtain enhanced image sample data as the training sample data.

Specifically, for the converted two-dimensional images, data augmentation can also be performed on them. Specifically, for the converted two-dimensional heartbeat images, data augmentation techniques can be further used to expand the dataset to avoid overfitting problems. Therefore, nine different cropping methods (upper left, upper middle, upper right, upper middle, middle right, lower left, lower middle, and lower right) are used to enhance the ECG abnormality category. As a result of cropping, we obtain multiple ECG images of reduced size (96×96), which are then resized to 128×128 images.

In this example implementation, the activation function of the ECG signal classification model uses an exponential linear unit (ELU) instead of ReLU, providing a smaller negative value. The ELU function may specifically include:

Among them, the value of the hyperparameter γ can be set to 1.0.

In this example implementation, the VGG network-based ECG signal classification model includes a plurality of feature extraction units arranged in series, and a classification unit; wherein the feature extraction unit includes at least two convolutional layers based on Ghost Module arranged in sequence, and a maximum pooling layer; the classification unit includes a first fully connected layer and a second fully connected layer arranged in sequence.

In this example implementation, a dropout layer is provided between the first fully connected layer and the second fully connected layer.

Specifically, in the existing VGG-16 network model architecture, as the number of network layers increases, even with the use of small convolution kernel stacking, the number of VGGNet-16 model parameters is still very large. In view of the problems of large number of parameters, large amount of calculation, and slow reasoning speed of the VGG-16 network model, this solution follows the basic network structure of VGG-16, as shown in Table 1, and uses Ghost Module to replace the traditional convolution layer in the VGG-16 network, reducing the number of parameters and calculation, thereby achieving lightweight network models, which is conducive to the application of portable devices.

In this example implementation, the input of the ECG signal classification model is a relatively simple 128×128 grayscale image. The current classic convolutional neural network model generally does not have high requirements for the fully connected layer. Therefore, the solution also modified the fully connected layer of VGG-16. The first fully connected layer contains only 512 neurons, and the second fully connected layer is also the output layer. The Softmax function is used to classify the data and output the arrhythmia type, namely NOR, LBBB, RBBB, PVC and APC. A dropout operation is added between the fully connected layers to discard the number of neurons in the fully connected layer with a probability of 0.5 to avoid overfitting. The specific steps of Ghost-VGG-16 are shown in Table 1. Ghost Module and Maxpooling represent the feature extraction steps, and the fully connected layer represents the classification step.

The main idea of Ghost Module is to replace part of the traditional convolution with a series of simple linear operations in a more cost-effective way. Specifically, the traditional convolution is divided into two parts, where the first part uses traditional convolution to generate half of the feature map, and then uses cheap linear transformation to obtain the other part of the feature map, which is combined with the previously generated feature map to obtain more feature maps. The linear transformation in Ghost Module can be replaced by channel convolution, which can reduce the number of parameters and calculations in the Ghost module without changing the size of the output feature map.

Assume that the input feature map is H×W×M, the size of the traditional convolution kernel is k×k, the size of the first part of the convolution of the Ghost Module is k×k, the size of the second part of the convolution kernel is d×d, and the size of the output feature map is H′×W′×N. The number of parameters of the traditional convolution is:

The parameters of Ghost Module are:

和

分别为Ghost Module操作使用传统卷积和线性变换的参数量；其中，in,

and

The parameters of traditional convolution and linear transformation are used for Ghost Module operation respectively;

The computational complexity of traditional convolution is: Calc _T = H×W×M×N×k×k

The calculation amount of Ghost Module operation is: Calc _Ghost = Calc _Conv + Calc _Linear

Among them, Calc _Conv and Calc _Linear are the parameters of the Ghost Module operation using traditional convolution and linear transformation respectively.

The ratio of the number of parameters using the Ghost Module operation to the number of parameters using the traditional convolution operation r _Param and the ratio of the amount of calculation r _Calc are:

From the above formula, it can be concluded that under the premise of obtaining the same number of feature maps, using the Ghost Module operation can reduce the number of parameters and calculations by about half, and the inference speed is also faster.

The loss function of the ECG signal classification model can adopt the cross entropy loss function, and the formula can include:

Where n is the number of training data (or batch size), y is the expected value, and a is the actual value of the output layer.

In this example implementation, the training set and the validation set are first used to train the ECG signal classification model based on the Ghost-VGG-16 network. After 10 iterations, the model loss function and training accuracy can be obtained, as shown in Figures 3 and 4, respectively. It can be observed that after 10 iterations, the model gradually becomes stable, so we set the epoch to 10. As shown in Figure 3, as the number of iterations increases, the loss function also decreases. After 5 iterations, the loss function gradually becomes stable, which indicates that the proposed Ghost-VGG16 network model has converged. As shown in Figure 4, with the continuous optimization of the loss function, the training accuracy and the validation set accuracy will also increase, and gradually stabilize between 0.99 and 1. After the model parameters are trained, the model is saved, and the model classification results are verified with the test set. The classification result confusion matrix of the ECG signal classification model is shown in Figure 5. In the confusion matrix, each row represents the classification of the real heartbeat recognized by the model in the test set, and each column shows the number of heartbeat types recognized by the model in the five heartbeat types.

The four evaluation index values of the calculated model are shown in Table 3.

Type Acc(%) Sp(%) Se(%) PPV(%) NOR 99.95 99.93 99.98 99.94 LBBB 99.43 99.99 96.28 99.93 RBBB 99.36 99.34 99.55 94.11 PVC 99.98 99.97 100 99.82 APC 99.99 100 99.76 100 Total 99.74 (average) 99.85 (average) 99.11 (average) 98.76 (average)

Table 3

Among the indicators in Table 3, four parameters, namely, accuracy (Acc), specificity (Sp), sensitivity (Se) and positive predictive value (PPV), are used as indicators to measure classification performance. Acc is the ratio between the number of correctly classified samples and the total number of test samples. Sp is the fraction of negative test results correctly identified as normal. Se represents the ability to identify the disease. The higher the Se, the more likely the method is to correctly diagnose the disease and the less likely the patient is to be missed. PPV reflects the ability to correctly detect the disease in the diagnosed case. The definition of each indicator may include:

Among them, True Positive (TP) indicates the correct classification of arrhythmia, True Negative (TN) indicates the correct classification of normal samples, False Positive (FP) indicates the incorrect classification of arrhythmia, and False Negative (FN) indicates the incorrect classification of normal samples.

The performance of this method is compared with other related algorithms based on VGGNet, as shown in Table 4. In Table 4, weights represents the number of parameters of the model size, floating-point operations (Flops) represents the amount of calculation of the model, and the unit M represents one million. As can be seen from Table 5, compared with other previous works, the proposed Ghost-VGG-16 not only achieves excellent performance (average sensitivity 99.11%, average specificity 99.85%, average accuracy 99.74%, average positive predictive value 98.76%), but also greatly reduces the number of parameters and the amount of calculation of the model.

Table 4

As shown in Table 5, the ECG signal classification model of the lightweight neural network provided by the present invention is compared with other advanced ECG arrhythmia automatic classification algorithms. The results show that the proposed lightweight convolutional neural network Ghost-VGG-16 based on Ghost Module achieves the best performance in terms of Acc, Sp, Se and PPV. At the same time, compared with the model proposed by us, several other methods [11-17] directly use one-dimensional CNN to classify arrhythmias, and the classification accuracy of the model is lower ([15]-96.40% and [17]-93.60%). This is due to the flexibility of two-dimensional data augmentation. Compared with other algorithms based on 2D-CNN, the algorithm proposed by us achieves higher accuracy under the premise of increasing the number of network layers.

Model Acc(%) Sp(%) Se(%) PPV (%) RNN[11] 98.06 97.78 98.15 - LS-SVM[12] 95.82 99.17 86.16 97.01 KNN[14] 97.00 95.80 96.60 1-D CNN[17] 96.40 99.50 68.80 79.20 2-D CNN[19] 99.11 99.61 97.91 98.58 Ghost-VGG-16 99.74 99.85 99.11 98.76

Table 5

Through the above comparison, the invented Ghost-VGG-16-based ECG signal classification model is more suitable for application on portable devices in edge computing scenarios, thereby realizing automatic classification of ECG arrhythmias; the signal classification results can be used to prevent cardiovascular diseases.

In this example implementation, during the test process, the above ECG signal classification method can use Raspberry Pi 3B as a program running carrier to test the program in a low computing power scenario. Raspberry Pi is an ARM-based microcomputer motherboard with an SD/MicroSD card as a memory hard disk. There are 1/2/4 USB ports and a 10/100 Ethernet port (Type A has no network port) around the card motherboard, which can be connected to a keyboard, mouse and network cable. It also has a TV output interface for video analog signals and an HDMI high-definition video output interface. All of the above components are integrated on a motherboard that is only slightly larger than a credit card. It has all the basic functions of a PC. Just connect the TV and keyboard to perform functions such as spreadsheets, word processing, playing games, and playing high-definition videos. Raspberry Pi B only provides a computer board without memory, power supply, keyboard, or chassis. Ordinary computer motherboards rely on hard disks to store data, but Raspberry Pi uses SD cards as "hard disks" and can also connect external USB hard disks. Raspberry Pi can be used to edit Office documents, browse the web, play games, etc. The low price of the Raspberry Pi means that it has a wide range of uses, and it is also a good choice to turn it into an excellent multimedia center.

When testing the program in a low computing power scenario, first reset the Raspberry Pi, apply the algorithm file through burning, and use the Raspberry Pi as the main body of the project test run, and visualize the results. In this project, compared with the computing power of the PC, the Raspberry Pi with weaker application storage, communication and computing capabilities better simulates the application scenarios of low computing power terminals, verifies the applicability of the ECG signal automatic classification model based on lightweight neural network in low computing power scenarios, and the applicability of the model to wearable devices, thus fully proving that the model is both innovative and practical.

It should be noted that the above figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

Further, referring to FIG6 , an electrocardiogram signal classification device 60 is provided in the embodiment of this example, and the device includes: a preprocessing module 601, a feature extraction module 602, and a classification result output module 603.

The preprocessing module 601 can be used to preprocess the ECG signal to be processed and obtain a two-dimensional image corresponding to each heartbeat.

The feature extraction module 602 can be used to input the two-dimensional image into an ECG signal classification model based on a VGG network, use a number of consecutive feature extraction units to extract features from the two-dimensional image, and output a feature map; wherein the feature extraction unit includes: a convolution layer and a maximum pooling layer based on a Ghost Module.

The classification result output module 603 may be used to perform classification according to the feature graph using the classification unit to obtain the classification result corresponding to the electrocardiogram signal to be processed.

In some exemplary embodiments, the preprocessing module 601 may be used to segment the electrocardiogram signal to be processed to obtain a corresponding heartbeat signal; and to convert the heartbeat signal to obtain a two-dimensional image in grayscale format corresponding to the heartbeat signal.

In some exemplary embodiments, the apparatus may further include: a model training module.

The model training module can be used to collect labeled ECG sample data, slice the ECG sample data according to the R wave peak value to obtain heartbeat sample data; convert the heartbeat sample data to obtain corresponding two-dimensional image sample data as training sample data; divide the training sample data to obtain corresponding training set, verification set and test set; input the training set into the ECG signal classification model based on the VGG network to perform model training until the model converges; and input the verification set into the ECG signal classification model to verify the model; input the test set data into the trained ECG signal classification model, and output the corresponding ECG signal classification result.

In some exemplary embodiments, the apparatus further comprises: a data enhancement module.

The data enhancement module can be used to crop the two-dimensional image sample data according to a preset cropping rule to obtain a cropped image; and to perform a size transformation on the cropped image to obtain enhanced image sample data as the training sample data.

In some exemplary embodiments, the VGG network-based ECG signal classification model includes a plurality of feature extraction units arranged in series, and a classification unit; wherein the feature extraction unit includes at least two Ghost Module-based convolutional layers and a maximum pooling layer arranged in sequence; the classification unit includes a first fully connected layer and a second fully connected layer arranged in sequence.

In some exemplary embodiments, a dropout layer is provided between the first fully connected layer and the second fully connected layer.

In some exemplary embodiments, the type of the ECG signal includes any one of a NOR type, a PVC type, an APC type, a RBBB type, and a LBBB type.

The specific details of each module in the above-mentioned electrocardiogram signal classification device 60 have been described in detail in the corresponding electrocardiogram signal classification method, so they will not be repeated here.

It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

FIG. 7 shows a schematic diagram of an electronic device suitable for implementing an embodiment of the present invention.

It should be noted that the electronic device 1000 shown in FIG. 7 is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

As shown in FIG. 7 , the electronic device 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage part 1008 to a random access memory (RAM) 1003. Various programs and data required for system operation are also stored in the RAM 1003. The CPU 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input/output (I/O) interface 1005 is also connected to the bus 1004.

The following components are connected to the I/O interface 1005: an input section 1006 including a keyboard, a mouse, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I/O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed.

In particular, according to an embodiment of the present invention, the process described below with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part 1009, and/or installed from a removable medium 1011. When the computer program is executed by a central processing unit (CPU) 1001, various functions defined in the system of the present application are executed.

Specifically, the electronic device may be a smart mobile electronic device such as a mobile phone, a tablet computer or a laptop computer, or may be a smart electronic device such as a desktop computer.

It should be noted that the computer-readable medium shown in the embodiment of the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

The units involved in the embodiments of the present invention may be implemented by software or hardware, and the units described may also be arranged in a processor. The names of these units do not, in some cases, limit the units themselves.

It should be noted that, as another aspect, the present application also provides a computer-readable medium, which may be included in an electronic device; or may exist independently without being assembled into the electronic device. The above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed by an electronic device, the electronic device implements the method described in the following embodiments. For example, the electronic device may implement the steps shown in Figure 1.

In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims (10)

1. A method for classifying electrocardiogram signals, characterized in that the method comprises: Preprocess the ECG signal to be processed to obtain a two-dimensional image corresponding to each heartbeat; Input the two-dimensional image into an ECG signal classification model based on a VGG network, extract features from the two-dimensional image using a plurality of consecutive feature extraction units, and output a feature map; wherein the feature extraction unit includes: a convolution layer and a maximum pooling layer based on a GhostModule; The classification unit is used to perform classification according to the feature map to obtain the classification result corresponding to the electrocardiogram signal to be processed. 2. The ECG signal classification method according to claim 1, characterized in that the preprocessing of the ECG signal to be processed to obtain a two-dimensional image corresponding to each heartbeat comprises: Segment the ECG signal to be processed and obtain the corresponding heart beat signal; The heartbeat signal is converted to obtain a two-dimensional image in grayscale format corresponding to the heartbeat signal. 3. The ECG signal classification method according to claim 1, characterized in that the method further comprises: training the ECG signal classification model, comprising: Collecting marked ECG sample data, and slicing the ECG sample data according to the R wave peak value to obtain heart beat sample data; Convert the heartbeat sample data to obtain corresponding two-dimensional image sample data as training sample data; Dividing the training sample data to obtain corresponding training sets, validation sets and test sets; Input the training set into the ECG signal classification model based on the VGG network to perform model training until the model converges; and input the verification set into the ECG signal classification model to verify the model; The test set data is input into the trained ECG signal classification model, and the corresponding ECG signal classification result is output. 4. The electrocardiogram signal classification method according to claim 1, characterized in that the method further comprises: Cropping the two-dimensional image sample data according to a preset cropping rule to obtain a cropped image; The cropped image is resized to obtain enhanced image sample data, which is used as the training sample data. 5. The ECG signal classification method according to claim 1 or 3 is characterized in that the ECG signal classification model based on the VGG network includes a plurality of feature extraction units and a classification unit arranged in succession; wherein the feature extraction unit includes at least two convolutional layers based on Ghost Module and a maximum pooling layer arranged in sequence; the classification unit includes a first fully connected layer and a second fully connected layer arranged in sequence. 6. The electrocardiogram signal classification method according to claim 5, characterized in that a dropout layer is provided between the first fully connected layer and the second fully connected layer. 7 . The ECG signal classification method according to claim 1 , wherein the type of the ECG signal comprises any one of a NOR type, a PVC type, an APC type, a RBBB type and a LBBB type. 8. An electrocardiogram signal classification device, characterized in that the device comprises: A preprocessing module, used to preprocess the ECG signal to be processed and obtain a two-dimensional image corresponding to each heartbeat; A feature extraction module, used to input the two-dimensional image into an ECG signal classification model based on a VGG network, perform feature extraction on the two-dimensional image using a plurality of consecutive feature extraction units, and output a feature map; wherein the feature extraction unit includes: a convolution layer and a maximum pooling layer based on a Ghost Module; The classification result output module is used to use the classification unit to perform classification according to the feature map to obtain the classification result corresponding to the electrocardiogram signal to be processed. 9. A computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the electrocardiogram signal classification method according to any one of claims 1 to 7 is implemented. 10. An electronic device, comprising: Processor; and A memory, configured to store executable instructions of the processor; The processor is configured to perform the electrocardiogram signal classification method according to any one of claims 1 to 7 by executing the executable instructions.

Images