12-lead electrocardiogram reconstruction method based on attention mechanism generative adversarial network

By introducing an attention mechanism into the GAN model, the problem of insufficient single-lead ECG information is solved, and the accurate reconstruction of 12-lead signals is achieved, which improves the accuracy of ECG diagnosis and the application potential of wearable devices.

CN119006882AActive Publication Date: 2024-11-22ZHEJIANG UNIV

Patent Information

Application Number
CN202410943241.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-22
Estimated Expiration
2044-07-15

AI Technical Summary

Technical Problem

The existing 12-lead ECG measurement method is difficult to achieve real-time monitoring in daily life. The single-lead ECG information is insufficient, so it is impossible to effectively diagnose ECG diseases. The existing GAN-based reconstruction method has not been implemented in wearable devices. Practical application.

Method used

Using a Generative Adversarial Network (GAN) model based on attention mechanism, 12-lead signals are reconstructed from single-lead ECG signals, and through the game process of generator and discriminator, multi-lead ECG similar to real signals is learned and generated. Data, using attention mechanisms to improve the sensitivity and accuracy of the model.

Benefits of technology

The accurate reconstruction of the ECG signal from single lead to 12 lead is achieved, overcoming the inadequate information and operation difficulty of traditional methods, and improving the accuracy of ECG diagnosis and the application potential of wearable devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006882A_ABST
    Figure CN119006882A_ABST
Patent Text Reader

Abstract

The invention discloses a 12-lead electrocardiogram reconstruction method based on an attention mechanism generative adversarial network, and the method comprises the steps: carrying out the preprocessing operation of filtering, time segmentation, data enhancement and the like on 12-lead electrocardiograms of various disease types to obtain electrocardiogram segments, extracting I-lead signals in the electrocardiogram segments, copying the I-lead signals, and inputting the I-lead signals into a network in a zero-filling manner; the corresponding 12-lead electrocardiogram is used as a true value label, the network is trained to learn the corresponding relation between the 12-lead electrocardiogram and the 12-lead electrocardiogram, game balance of the generator and the discriminator is achieved, and recovery from the single-lead electrocardiogram to the 12-lead electrocardiogram is achieved. The attention mechanism is added into the GAN model, and the attention mechanism is applied to the generator, so that attention to the salient region can be well realized, the salient features of the specific local region are highlighted, and the sensitivity of the model and the prediction precision are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of electrocardiogram signal reconstruction, and in particular relates to a 12-lead electrocardiogram reconstruction method based on an attention mechanism-generated adversarial network. Background Art

[0002] Electrocardiogram (ECG) is a common and non-invasive method for diagnosing cardiovascular diseases in clinical practice. It is a voltage graph that changes over time and reflects the electrical activity of myocardial depolarization and repolarization during each heartbeat. The ECG under normal rhythm consists of a series of waveforms, including P wave, QRS complex and T wave, and the signal also includes PR, ST and QT intervals. The ECG record reflects the electrical activity of the heart and can be used to detect and diagnose common heart diseases such as atrial fibrillation (AF) or myocardial infarct (MI). At present, the common ECG examination method in clinical practice is 12-lead ECG, which obtains ECG data by placing lead electrodes on the limbs and chest to analyze cardiovascular disease conditions; 12-lead ECG includes 3 standard limb leads (I, II, III), 3 enhanced limb leads (aVR, aVL, aVF) and 6 precordial leads (V1, V2, V3, V4, V5, V6).

[0003] However, in the 12-lead ECG measurement method, multiple electrodes need to be placed on the patient's limbs and chest, making it difficult to achieve real-time ECG monitoring in daily life. With the development of science and technology, the emergence of wearable devices has provided a portable and real-time detection method for the measurement of ECG signals. For example, smart watches detect the potential changes between the left and right arms of the human body caused by heart activity to obtain ECG waveforms; but this only records the potential changes of the heart in one direction, and can only obtain single-lead (I lead) data. The amount of information is much lower than the standard 12-lead ECG used clinically, and is limited for the diagnosis of ECG diseases. Therefore, the reconstruction of ECG signals from single lead to multi-lead has great clinical significance.

[0004] At present, the literature [Atoui, Hussein, Jocelyne Fayn, and Paul Rubel. "A novel neural-network model for deriving standard 12-lead ECGs from serial three-lead ECGs: application to self-care." IEEE transactions on information technology inbiomedicine 14.3 (2010): 883-890] recovers 12-lead ECGs from orthogonal three-leads (I, II, V2) through general and specific nonlinear reconstruction methods based on artificial neural networks (ANNs). However, the acquisition of three-lead ECG signals requires professional medical equipment to set electrode patches. Compared with the acquisition of single-lead signals, it is difficult to achieve daily monitoring.

[0005] Generative models are a type of model in the field of deep learning. They generate new data that are similar to but not exactly the same as real data by learning the distribution pattern of training data. The generative adversarial network (GAN) consists of two networks, the generator and the discriminator. The generator is responsible for generating fake data, while the discriminator is responsible for distinguishing the generated data from the real data. During the training iteration process, the two networks continue to compete against each other until a state of equilibrium is reached. The introduction of the generative adversarial network (GAN) meets the research needs of many fields and has been widely used in image vision, language and speech processing and other fields. For example, the literature [He, Yan, et al. "Efficient learning of healthcare data from IoT devices by edge convolution neural networks." Applied Sciences 10.24 (2020): 8934] uses a deep convolutional generative adversarial network (DCGAN) to generate additional ECG data from the arrhythmia database MIT-BIH, thereby improving the accuracy of remote health monitoring and edge computing. However, the generated data is only used for data enhancement to train the prediction model, and is not combined with common wearable devices for measuring single-lead ECGs, making it impossible to achieve actual scenario applications. Summary of the invention

[0006] In view of the above, the present invention provides a 12-lead ECG reconstruction method based on an attention mechanism-generated adversarial network, which can realize the reconstruction of ECG signals from a single lead to 12 leads, overcoming the limitations of modern ECG measurement methods.

[0007] A 12-lead electrocardiogram reconstruction method based on an attention mechanism-generated adversarial network comprises the following steps:

[0008] (1) Obtain 12-lead ECG signals from patients with normal heart rate, myocardial infarction, and various heart diseases, and divide these ECG signals into training sets and test sets after preprocessing;

[0009] (2) Construct a 12-lead ECG reconstruction model based on the generative adversarial network GAN, which includes a discriminator D and two generators G I and G L The single-lead ECG signals are input to G I and G L In, G I Used to reconstruct and generate the corresponding 12-lead ECG signal, G L It is equivalent to an autoencoder, and the latent space vector z l As G I The encoder outputs the latent space vector z i The label of G I The decoder of G can better learn the important characteristics of single-lead ECG signals; the discriminator D is used to I The generated 12-lead ECG signal and the original 12-lead ECG signal are distinguished as true or false;

[0010] (3) using the training set to train the above model, that is, using the 12-lead ECG signals in the training set as labels and extracting single-lead ECG signals from them as model input;

[0011] (4) Extract a single-lead ECG signal from the 12-lead ECG signal of the test set and input it into the generator G of the trained model I The corresponding 12-lead ECG signal can be reconstructed and output.

[0012] Furthermore, in the step (1), the 12-lead ECG signal obtained is first filtered to eliminate low-frequency drift and interference in the ECG signal, and then the ECG signal is divided into multiple signal segments of fixed time length, and several additional segments are generated for each group of ECG signals for data enhancement. The size of each signal segment is m×y, where m is the time length of the segment and y=12 is the number of leads.

[0013] Furthermore, the generator G IThe encoder of G is composed of five cascaded encoding modules, each of which is composed of a convolutional layer, a batch normalization layer, and a Leaky ReLU activation function layer connected in sequence; I The decoder is the inverse structure of the encoder, which is composed of five cascaded decoding modules. Except for the last decoding module, the remaining decoding modules are composed of an attention mechanism layer, a deconvolution layer, a batch normalization layer, and a ReLU activation function layer connected in sequence. Starting from the second decoding module, the output of the previous decoding module is concatenated with the output of the corresponding encoding module and then used as the input of the current decoding module. The last decoding module only contains a deconvolution layer.

[0014] Furthermore, the attention mechanism layer combines the input with the linear transformation matrix W Q , W K and W V Multiply them to get the query vector matrix Q, key vector matrix K, and value vector matrix V, and then calculate the output result by the following formula;

[0015]

[0016] Where: Attention(Q,K,V) is the output result of the attention mechanism layer, d is the number of columns of the query vector matrix Q, T represents transpose, and softmax() represents the normalized exponential function.

[0017] Furthermore, the generator G L The structure of the generator G I The structure is similar, the difference is that G L The decoding module in does not contain an attention mechanism layer and is not concatenated with the output of the encoding module as input.

[0018] Furthermore, the discriminator adopts a 1D U-Net network, whose structure is similar to that of the generator G L The structure is the same.

[0019] Furthermore, in the step (3), the ECG signal of any lead is extracted from the 12-lead ECG signals of the training set, 12 copies are made, and two additional rows are added at the top and bottom for zero padding (a total of 16 rows), and then the batches are combined and input into the model for training. The data dimension of the input model is b×m×n×c, where b is the number of signal segments in a batch, c is the number of channels, and n=16.

[0020] Furthermore, the optimization objective function expression used in the model training process in step (3) is as follows:

[0021]

[0022] in: Represents the final trained generator G I , To generate the loss function of the adversarial network GAN, For the generator G I The loss function is For the generator G L The loss function (which is only used to train the generator G L , not included in the overall optimization objective function), is the loss function for the latent space vector, λ and α are hyperparameters, represents the expected function, ||||1 represents the L1 norm, e i is the single-lead ECG signal input to the model, e o is the real 12-lead ECG signal, D(e o ) indicates that the discriminator is o The judgment result of G I (e i ) represents the generator G I For e i The reconstructed 12-lead ECG signal, D(G I (e i )) indicates that the discriminator is I (e i ) is the result of the judgment, G L (e i ) represents the generator G L For e i Reconstructed single-lead ECG signal.

[0023] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the above-mentioned 12-lead electrocardiogram reconstruction method.

[0024] A computer-readable storage medium stores a computer program, wherein the computer program is executed by a processor to implement the 12-lead electrocardiogram reconstruction method.

[0025] The present invention performs preprocessing operations such as filtering, time segmentation, and data enhancement on the 12-lead electrocardiograms of various diseases to obtain electrocardiogram segments, extracts the I lead signal therein, replicates and zero-fills it into the network, takes the corresponding 12-lead electrocardiogram as the true value label, trains the network to learn the correspondence between the two, achieves the game balance between the generator and the discriminator, and realizes the recovery from a single-lead electrocardiogram to a 12-lead electrocardiogram.

[0026] The attention mechanism is added to the generative adversarial network (GAN) model used in the present invention. The attention mechanism is applied to the generator, which can better achieve attention to the significant area, highlight the significant features of the specific local area, and improve the sensitivity of the model and the accuracy of prediction. Electrocardiograms of different diseases were used in the experiment, and the superiority of the present invention was proved according to various evaluation indicators. Compared with other deep learning methods, the present invention is closer to the true value in the experiment, and the reconstruction effect is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a schematic diagram of the structure of a 12-lead electrocardiogram reconstruction model based on a generative adversarial network in the present invention.

[0028] Figure 2 This is a schematic diagram of the structure of the attention mechanism.

[0029] Figure 3 Schematic diagram of a data set preprocessing process according to an embodiment of the present invention.

[0030] Figure 4 It is a pie-shaped statistical chart of each disease type in the training data set of the present invention.

[0031] Figure 5 ECG reconstruction results of various networks for patients with myocardial infarction.

[0032] Figure 6 ECG reconstruction results of various networks for patients with ventricular hypertrophy.

[0033] Figure 7 ECG reconstruction results for various networks for people with normal heart rate.

[0034] Figure 8 ECG reconstruction results of various networks for patients with myocardial infarction. DETAILED DESCRIPTION

[0035] In order to describe the present invention more specifically, the technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0036] The present invention provides a 12-lead electrocardiogram reconstruction method based on an attention mechanism-generated adversarial network, comprising the following steps:

[0037] (1) Prepare data.

[0038] The ECG public data set PTB-XL was selected, which includes clinical 12-lead ECGs of nearly 20,000 patients, including normal heart rate, myocardial infarction, and ST segment T wave changes. These ECG data with a time length of 10s were first filtered through a filter to eliminate the low-frequency drift and network interference of the ECG signal, and then divided into training sets and test sets in proportion; then the signals in the training subset and the test subset were divided into signal segments with a time length of 1.28s, and finally 10 additional segments were generated for each signal in the training subset for data enhancement, and finally a training subset of size s×m×y was obtained, where s=214624 represents the signal segment, m=128 represents the time length of the segment, y=12 represents the number of leads, and the test subset contains 22095 segments.

[0039] (2) Build a 12-lead ECG reconstruction model based on the generative adversarial model GAN, such as Figure 1 shown.

[0040] Single lead signal i Before inputting the network model, it is necessary to first extract, copy and zero-fill the training set. The data dimension of the input network is b×m×n×c, where b is the batch number, m=128 is the signal length, n=16 is the result after copying and zero-filling, and c=1 is the number of channels. The vector matrix is ​​input into the encoding and decoding structure of the generator G. The output matrix of the encoding and decoding structure of the generator G has the same dimension as the input matrix. The 12 leads generated by the generator G and the original 12 leads are then input into the discriminator D for discrimination.

[0041] The network model consists of a generator G and a discriminator D, where the generator G includes a 12-lead generator G I and label generator G L , single lead signal e i Input into two generators respectively, generator G I Generates the corresponding 12-lead signal, and the label generator G L It is equivalent to an autoencoder, whose encoder obtains the latent space vector z l As G I The latent space vector z obtained by the encoder i The label of G I The decoder of G can better learn the important characteristics of the input single lead. I The generated 12-lead ECG and the original 12-lead ECG are discriminated, and this process is repeated to train the generator G and the discriminator D.

[0042] 12-lead generator G in the network model IThe encoder consists of five encoding blocks, each of which includes convolution, batch normalization and Leaky ReLU layers. The decoder is the inverse structure of the encoder, including five decoding blocks, each of which includes Attention, splicing layer, deconvolution, batch normalization and ReLU. The first decoding block does not contain a splicing layer, and the last decoding block only contains a splicing layer and deconvolution. Starting from the second decoding block, the splicing layer splices the output of the previous decoding block with the output of the corresponding encoding block and uses it as the input of the next decoding block.

[0043] 12-lead generator G in the network model I The key to the decoder's attention mechanism is to learn a weight distribution and then act on the features; Figure 2 As shown, it first calculates the similarity of the query for each key. From the perspective of signal processing, the inner product represents the similarity of two signals. The distance between the query and the keys is calculated through the inner product to obtain the weight coefficient:

[0044]

[0045] Where d is the number of columns of matrices Q and K.

[0046] The weights are then normalized to obtain directly usable weights; the softmax operation can organize the original calculated scores into a probability distribution where the sum of the weights of all elements is 1, and the calculation is performed on each row of the matrix:

[0047]

[0048] Finally, the value is weighted and summed according to the weight coefficient to obtain the attention value for the query:

[0049]

[0050] Label GeneratorG L The structure is mostly the same as the 12-lead generator G I The discriminator D is a 1D U-Net, which is different from the label generator G. L With the same layer structure, the 12-lead generator G I The output and the true value 12-lead ECG are used as input for discrimination.

[0051] (3) The single-lead data e in the training set i Input network, e oThe real 12-lead data, i.e. the label of the network, is used to train the network model to generate a 12-lead ECG from a single-lead ECG. The training process is as follows:

[0052] 3.1 Set the learning rate, batch size and maximum number of iterations, and initialize the network parameters;

[0053] 3.2 The single-lead data in the training set e i Input network, calculate generator G I The loss of the generated 12 leads and the true value of the label, the generator G I Latent space variable z i and the true value of the label, i.e., the generator G L Latent space variable z l The loss of the generator G I The adversarial loss function with the discriminator D is:

[0054]

[0055] Where: e i is the input single-lead data, e o For real 12-lead data, is the loss of the GAN network itself. In addition, a 12-lead generator G is introduced. I Loss and the loss of the latent space vector in Only used to train the label generator G L , which is not included in the final objective function; the hyperparameters λ and α are set to 50 and 1, respectively.

[0056] 3.3 Use gradient descent to update the network parameters. When the optimization target converges or the number of training iterations reaches the maximum number, the training is completed.

[0057] (4) The single-lead data e of the test set i The trained generator is input to obtain a generated 12-lead electrocardiogram. In this embodiment, the electrocardiogram signal of the standard limb lead I is selected for reconstruction.

[0058] The following simulation experiments are used to verify the effectiveness of the present invention. The experiment uses the public data set PTB-XL. The data preprocessing process is as follows: Figure 3 As shown, the training set and test set are divided into 8:2 ratios. The training data set is composed as follows Figure 4 As shown in the figure, the time length of each set of data is 1.28s. i Input network, e oThe real 12-lead data is used as labels for training. During the training process, the learning rate is set to 0.0001, the batch number is set to 32, and the maximum number of iterations is set to 10. The network of the present invention is compared with CycleGAN.

[0059] Table 1 shows the quantitative analysis results of the 12-lead images. The root mean square error (RMSE) is a commonly used measure of the difference between measured values. The specific formula is:

[0060]

[0061] The mean absolute error (MAE) describes the error between the predicted value and the true value, and measures the absolute deviation between the predicted result and the actual result. The specific formula is:

[0062]

[0063] Where: s represents the ECG segment, and M is the sample length.

[0064] The table shows the other 11 generated leads from top to bottom, and from left to right are the result analysis of CycleGAN, EKGAN, and EKGAN with attention mechanism. It can be seen from the table that the mean values ​​of the evaluation indicators of the present invention are higher than those of the other two methods, which proves the superiority of the present invention.

[0065] Table 1

[0066]

[0067]

[0068] Figure 5 The following are the ECG reconstruction results of various networks for patients with myocardial infarction. The black waveform, red waveform and blue waveform correspond to the true image, EKGAN (removing the attention mechanism) and CycleGAN ECG reconstruction images. The figures show the reconstruction results of limb leads (II, III, aVR, aVL, aVF) and precordial leads (V1, V2, V3, V4, V5, V6) respectively. It can be seen that the reconstruction result of CycleGAN has a large error, and the reconstruction result of EKGAN is closer to the true image.

[0069] Figure 6The following are the ECG reconstruction results of various networks for patients with ventricular hypertrophy. The black waveform, red waveform, and blue waveform correspond to the true image, EKGAN (removing the attention mechanism) and CycleGAN ECG reconstruction images. The figures show the reconstruction results of limb leads (II, III, aVR, aVL, aVF) and precordial leads (V1, V2, V3, V4, V5, V6) respectively. It can be seen that the reconstruction result of CycleGAN has a large error, and the reconstruction result of EKGAN is closer to the true image.

[0070] Figure 7 The figures are the ECG reconstruction results of various networks for people with normal heart rate. The black waveform, red waveform, and blue waveform correspond to the true value image, the ECG reconstruction image of the network of the present invention, and EKGAN (without attention mechanism). The figures show the reconstruction results of limb leads (II, III, aVR, aVL, aVF) and precordial leads (V1, V2, V3, V4, V5, V6) in sequence. It can be seen that the reconstruction result of the network of the present invention is closer to the true value, which shows the superiority of the present invention.

[0071] Figure 8 The figures show the ECG reconstruction results of various networks for patients with myocardial infarction. The black waveform, red waveform and blue waveform correspond to the true value image, the ECG reconstruction diagram of the network of the present invention and EKGAN (without attention mechanism). The figures show the reconstruction results of limb leads (II, III, aVR, aVL, aVF) and precordial leads (V1, V2, V3, V4, V5, V6) in sequence. It can be seen that the reconstruction result of the present invention is closer to the true value, which shows the superiority of the present invention.

[0072] Experimental results show that the 12-lead ECG reconstruction method based on the attention generative adversarial network (GAN) of the present invention is effective. Applying the attention mechanism to the generator of the GAN network can better focus on the salient areas, highlight the significant features of specific local areas, and improve the sensitivity of the model and the accuracy of prediction.

[0073] The above description of the embodiments is to facilitate the understanding and application of the present invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made to the present invention by those skilled in the art based on the disclosure of the present invention should be within the scope of protection of the present invention.

Claims

1. A 12-lead electrocardiogram reconstruction method based on an attention mechanism-generated adversarial network, comprising the following steps: (1) Obtain 12-lead ECG signals from patients with normal heart rate, myocardial infarction, and various heart diseases, and divide these ECG signals into training sets and test sets after preprocessing; (2) Construct a 12-lead ECG reconstruction model based on the generative adversarial network GAN, which includes a discriminator D and two generators G I and G L The single-lead ECG signals are input to G I and G L In, G I Used to reconstruct and generate the corresponding 12-lead ECG signal, G L It is equivalent to an autoencoder, and the latent space vector z l As G I The encoder outputs the latent space vector z i The label of G I The decoder of G can better learn the important characteristics of single-lead ECG signals; the discriminator D is used to I The generated 12-lead ECG signal and the original 12-lead ECG signal are distinguished as true or false; (3) using the training set to train the above model, that is, using the 12-lead ECG signals in the training set as labels and extracting single-lead ECG signals from them as model input; (4) Extract a single-lead ECG signal from the 12-lead ECG signal of the test set and input it into the generator G of the trained model I The corresponding 12-lead ECG signal can be reconstructed and output.

2. The 12-lead electrocardiogram reconstruction method according to claim 1, characterized in that: In the step (1), the 12-lead ECG signal obtained is first filtered to eliminate low-frequency drift and interference in the ECG signal, and then the ECG signal is divided into multiple signal segments of fixed time length, and several additional segments are generated for each group of ECG signals for data enhancement. The size of each signal segment is m×y, where m is the time length of the segment and y=12 is the number of leads.

3. The 12-lead electrocardiogram reconstruction method according to claim 1, characterized in that: The generator G I The encoder of G is composed of five cascaded encoding modules, each of which is composed of a convolutional layer, a batch normalization layer, and a Leaky ReLU activation function layer connected in sequence; I The decoder is the inverse structure of the encoder, which is composed of five cascaded decoding modules. Except for the last decoding module, the remaining decoding modules are composed of an attention mechanism layer, a deconvolution layer, a batch normalization layer, and a ReLU activation function layer connected in sequence. Starting from the second decoding module, the output of the previous decoding module is concatenated with the output of the corresponding encoding module and then used as the input of the current decoding module. The last decoding module only contains a deconvolution layer.

4. The 12-lead electrocardiogram reconstruction method according to claim 3, characterized in that: The attention mechanism layer combines the input with the linear transformation matrix W Q , W K and W V Multiply them to get the query vector matrix Q, key vector matrix K, and value vector matrix V, and then calculate the output result by the following formula; Where: Attention(Q,K,V) is the output result of the attention mechanism layer, d is the number of columns of the query vector matrix Q, T represents transpose, and softmax() represents the normalized exponential function.

5. The 12-lead electrocardiogram reconstruction method according to claim 3, characterized in that: The generator G L The structure of the generator G I The structure is similar, the difference is that G L The decoding module in does not contain an attention mechanism layer and is not concatenated with the output of the encoding module as input.

6. The 12-lead electrocardiogram reconstruction method according to claim 5, characterized in that: The discriminator uses a 1D U-Net network, which has the same structure as the generator G L The structure is the same.

7. The 12-lead electrocardiogram reconstruction method according to claim 1, characterized in that: In the step (3), an ECG signal of any lead is extracted from the 12-lead ECG signals of the training set, 12 copies are made, two additional rows are added at the top and bottom for zero padding, and then the copies are combined in batches and input into the model for training. The data dimension of the input model is b×m×n×c, where b is the number of signal segments in a batch, m is the time length of the segment, c is the number of channels, and n=16.

8. The 12-lead electrocardiogram reconstruction method according to claim 1, characterized in that: The optimization objective function expression used in the model training process in step (3) is as follows: in: Represents the final trained generator G I , To generate the loss function of the adversarial network GAN, For the generator G I The loss function is For the generator G L The loss function is is the loss function for the latent space vector, λ and α are hyperparameters, represents the expected function, || ||1 represents the L1 norm, e i is the single-lead ECG signal input to the model, e o is the real 12-lead ECG signal, D(e o ) indicates that the discriminator is o The judgment result of G I (e i ) represents the generator G I For e i The reconstructed 12-lead ECG signal, D(G I (e i )) indicates that the discriminator is I (e i ) is the result of the judgment, G L (e i ) represents the generator G L For e i Reconstructed single-lead ECG signal.

9. A computer device comprising a memory and a processor, characterized in that: The memory stores a computer program, and the processor is used to execute the computer program to implement the 12-lead electrocardiogram reconstruction method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the 12-lead electrocardiogram reconstruction method according to any one of claims 1 to 8 is implemented.

Citation Information

Patent Citations

  • Cardiac transmembrane potential imaging method based on electrocardiogram combined physiological model

    CN117557667A

Cited By

  • Accurate rhythm generation method based on generative adversarial network

    CN121456467A