Electrocardiogram inverse problem solving method based on neural network

Abstract

The invention discloses an electrocardio inverse problem solving method based on a neural network, and the method comprises the following steps: obtaining the three-dimensional geometric information of a heart source model, and constructing an electrocardio data set containing a body surface potential and a transmembrane potential; dividing a training set and a test set, and preprocessing data; constructing a neural network model by using a bidirectional long short-term memory network (BiLSTM), and setting and optimizing network parameters; training the constructed neural network model by using a training set; performing simulation prediction on the trained model through a test set to realize reverse reconstruction of the heart potential; and evaluating and analyzing the result, and visualizing the result. According to the method, through the deep learning ability of the neural network model, the problem of discomfort in the electrocardio inverse problem is effectively solved, higher reconstruction precision can be provided, and important technical support is provided for accurate diagnosis and treatment of heart diseases.

Description

Technical Field

[0001] The present disclosure relates to the field of cardiac electrophysiological imaging, and in particular to a method for solving an electrocardiographic inverse problem based on a neural network. Background Art

[0002] Cardiovascular disease has become the leading cause of death worldwide, especially in my country, where the incidence and mortality of cardiovascular disease continue to rise. The electrical activity of the heart is closely related to the health of the heart, and the cardiac potential inverse reconstruction technology analyzes the surface ECG signals to infer the spatial distribution of the electrical activity inside the heart, providing an important basis for the diagnosis and treatment of heart disease.

[0003] In clinical practice, the traditional 12-lead electrocardiogram is widely used in the diagnosis of cardiovascular diseases. By analyzing the electrical signal characteristics in the electrocardiogram, doctors can judge the patient's heart condition based on experience. However, the electrocardiogram can only provide local electrical activity information and is easily affected by body noise and sparse distribution of electrodes, which limits its diagnostic accuracy and range.

[0004] To address this limitation, cardiac electrophysiological imaging (ECGI) has received widespread attention. ECGI technology collects a full range of surface potential signals by arranging multiple electrodes (usually more than 64) on the body surface, and combines the geometric model of the heart and trunk to invert the distribution of electrical activity inside the heart. Compared with traditional electrocardiograms, ECGI can provide more detailed and accurate images of electrical activity, thereby better locating and diagnosing complex heart diseases such as arrhythmias.

[0005] However, due to the complex structure of the heart, the ECG inverse problem has significant ill-posedness. Traditional solution methods, such as direct regularization and iterative regularization, often require the introduction of additional constraints to improve the stability of the solution. However, these methods are extremely sensitive to the choice of parameters, have high computational complexity, and have limited generalization capabilities, making it difficult to meet the requirements of clinical applications.

[0006] With the development of artificial intelligence technology, machine learning and deep learning have shown strong capabilities in solving complex nonlinear problems. In recent years, researchers have tried to apply deep learning to the solution of ECG inverse problems by treating the ECG inverse problem as a nonlinear regression problem and using neural networks such as BiLSTM to build prediction models. Compared with traditional methods, deep learning methods can automatically extract complex features in data and show stronger generalization ability and stability under large-scale data training, thereby providing more accurate inverse reconstruction results. Summary of the invention

[0007] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method for solving the ECG inverse problem based on a neural network. This method effectively solves the ill-posedness in the ECG inverse problem by introducing a bidirectional long short-term memory network (BiLSTM), achieves high-precision reconstruction of the electrical activity inside the heart, and provides technical support for the diagnosis and treatment of heart diseases.

[0008] To achieve the above objectives, the present disclosure provides the following technical solutions:

[0009] A method for solving an electrocardiogram inverse problem based on a neural network comprises the following steps:

[0010] S100: Acquire three-dimensional geometric information of a heart source model and electrocardiogram data of a normal heart in one cycle;

[0011] S200: constructing an ECG data set based on the ECG data, dividing the data into a training set and a test set, and processing the data;

[0012] S300: construct a neural network model and set network parameters;

[0013] S400: training the neural network model using the training set, and optimizing model parameters during the training process to improve prediction accuracy;

[0014] S500: Use the test set to simulate and predict the trained neural network model to achieve reverse reconstruction of cardiac potential;

[0015] S600: Evaluate the prediction results, generate a comparison diagram of the reconstructed waveform and the real waveform of the heart node, and visualize the electrocardiogram signal.

[0016] Preferably, in step S100, the ECG data is clinical data collected by the Thomas Berger research team of the Medical University of Innsbruck, Austria, and obtained through the official website of ECGSIM software.

[0017] Furthermore, the ECG data includes body surface potential (BSP) and transmembrane potential (TMP).

[0018] A normal heart cycle is 505ms, and the simulation step is 1ms. BSP is the electrical signal measured on the human body surface through a 64-electrode patch, and TMP reflects the difference in the internal and external potentials of the myocardial cell membranes at 257 nodes in the heart.

[0019] Preferably, the ECG data set is a data set including BSP (64×3030) and TMP (257×3030) constructed by changing the waveform of the cardiac source signal and using different surface lead signals generated by ECGSIM, and is expanded to 6 groups.

[0020] Preferably, in step S200, the data set is divided to sample the electrical activity information at a time interval of 1 ms. Among them, 1 ms of sample data is taken every 4 ms, a total of 606 ms of sample data constitutes a test set, and the remaining 4 ms of sample data, a total of 2424 ms, constitute a training set. Further, the divided data is normalized and scaled to the range of [0,1] for better training and prediction.

[0021] Preferably, in step S300, the neural network model is constructed based on BiLSTM, and the specific construction process and parameter setting process are as follows:

[0022] S301: Establish an input layer to receive the preprocessed BSP data; establish a bidirectional LSTM layer to process the bidirectional dependencies of the time series; add a ReLU activation layer to introduce nonlinearity to help the network capture complex patterns in the data while avoiding the gradient vanishing problem; establish a fully connected layer to map features to the output space; establish a regression layer to process regression tasks and calculate the network loss;

[0023] S302: Set training options and use the Adam optimizer for gradient descent. To prevent overfitting, use the Dropout function to randomly discard a certain proportion of neurons during training, and set the discard rate to 0.5. To prevent gradient explosion, set the gradient threshold to 1, specify the initial learning rate to 0.001, and use a piecewise learning rate adjustment strategy to reduce the learning rate by multiplying it by a factor of 0.5 after 300 rounds of training. Use a validation set to help monitor the performance of the model during training, and perform validation every 50 iterations. Add data shuffling to reduce the model's tendency to learn patterns between batches and improve the model's generalization ability.

[0024] Preferably, in step S600, the evaluating the prediction result includes using the coefficient of determination (R²) and the mean absolute error (MAE).

[0025] Preferably, the evaluation index determination coefficient (R²) is defined as follows:

[0026] ,

[0027] In the formula, is the true value of cardiac transmembrane potential; is the average value of cardiac transmembrane potential; It is the predicted value of cardiac transmembrane potential.

[0028] Preferably, the mean absolute error (MAE) is defined as follows:

[0029]

[0030] In the formula, is the true value of cardiac transmembrane potential; It is the predicted value of cardiac transmembrane potential.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The method disclosed in the present invention can accurately solve the ECG inverse problem based on a neural network, and has significant advantages in the accuracy of solving the inverse problem, the generalization ability of the model, and the processing of complex electrical signals, providing a strong guarantee for the accurate diagnosis and treatment of heart diseases.

[0033] The overall characteristics and advantages of the disclosed invention can be specifically reflected in the following three aspects:

[0034] (1) The present invention is based on neural network technology, especially bidirectional long short-term memory network (BiLSTM), and effectively extracts spatial and temporal features from cardiac electrical signals by training a large amount of ECG data. Compared with traditional regularization methods (such as Tikhonov regularization and L-curve regularization), BiLSTM has stronger nonlinear fitting ability and time-dependent characteristics, and can more accurately infer the distribution of electrical activity inside the heart when dealing with ECG inverse problems. Traditional regularization methods often rely on the selection of regularization parameters by adding regularization terms to the objective function. Their performance varies on different data sets, the computational complexity is high, and there may be insufficient generalization ability when dealing with noise. However, through deep learning, especially BiLSTM network, it is possible to automatically extract key features from the complex mapping relationship between body surface potential and epicardial potential, reduce the dependence on manually set parameters, significantly improve the accuracy of solving ECG inverse problems, and perform particularly well on large-scale data sets.

[0035] (2) When processing complex cardiac electrical signals, traditional methods are prone to model performance degradation due to ill-posed problems. The present invention uses a deep learning model, especially the bidirectional information processing mechanism of BiLSTM, to maintain a high generalization ability in the face of different pathological conditions, ensuring stability and accuracy in clinical applications.

[0036] (3) After solving the inverse ECG problem, the present invention combines the three-dimensional geometric model and uses visualization technology to intuitively present the spatial distribution and change process of cardiac electrical activity. This visualization function not only helps researchers better understand the mechanism of cardiac electrical activity, but also provides clinicians with an intuitive auxiliary diagnostic tool to improve the accuracy and efficiency of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flowchart of a method for solving an electrocardiogram inverse problem based on a neural network provided by an embodiment of the present disclosure;

[0038] Figure 2 It is a panel schematic diagram of obtaining a cardiac source geometric model and an electrocardiogram data set through ECGSIM software provided by another embodiment of the present disclosure;

[0039] Figure 3 It is a waveform diagram of the potential of lead No. 30 and the transmembrane potential of the heart node No. 30 provided by another embodiment of the present disclosure;

[0040] Figure 4 is a schematic diagram of a bidirectional long short-term memory network BiLSTM network structure provided by another embodiment of the present disclosure;

[0041] Figure 5 is a comparison diagram of the reconstructed waveform and the real waveform of the heart node No. 150 provided by another embodiment of the present disclosure;

[0042] Figure 6 This is an effect diagram of three-dimensional visualization of reconstructed cardiac transmembrane potentials through Map3D at different test time points provided by another embodiment of the present disclosure. DETAILED DESCRIPTION

[0043] The following will refer to the attached Figures 1 to 6 Specific embodiments of the present disclosure are described in detail. Although specific embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0044] It should be noted that certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different nouns to refer to the same component. This specification and claims do not use the difference in nouns as a way to distinguish components, but use the difference in the functions of the components as the criterion for distinction. As mentioned throughout the specification and claims, "including" or "comprising" is an open term, so it should be interpreted as "including but not limited to". The subsequent description of the specification is a preferred embodiment of the present disclosure, but the description is based on the general principles of the specification and is not used to limit the scope of the present disclosure. The scope of protection of the present disclosure shall be determined by the attached claims.

[0045] To facilitate understanding of the embodiments of the present disclosure, further explanation will be given below using specific embodiments as examples in conjunction with the accompanying drawings, and each of the accompanying drawings does not constitute a limitation on the embodiments of the present disclosure.

[0046] The present invention provides a method for solving the ECG inverse problem based on a neural network. A clinical data set is obtained through ECGSIM as data input, and then the ECG inverse problem is solved based on a BILSTM neural network.

[0047] In one embodiment, Figure 1 As shown, a method for solving the inverse ECG problem based on a neural network includes the following steps:

[0048] S100: The three-dimensional geometric information of the heart source model and the body surface potential (BSP) and transmembrane potential (TMP) data of a normal heart in one cycle are obtained through ECGSIM software;

[0049] S200: construct an ECG data set based on the acquired BSP and TMP data, divide the data into a training set and a test set, and normalize the data and scale it to [0,1] for better training and prediction;

[0050] S300: Build a neural network model based on the bidirectional long short-term memory network (BiLSTM) and set network parameters;

[0051] S400: The neural network model is trained using the training set. During the training process, the model parameters are continuously adjusted through various optimization strategies to improve the prediction accuracy. In the early stage of training, the model structure is simple and the default parameters are used (SGD optimizer, fixed learning rate 0.01, batch size 32, no Dropout layer added, no gradient threshold set). At this time, the mean absolute error (MAE) of the model is 2.76±0.69 mV and the coefficient of determination (R²) is 93.6±1.37%. The model can perform basic reconstruction of cardiac potentials, but has weak ability to capture complex temporal features. Through the optimization strategy, the optimizer is switched to the Adam optimizer, the initial learning rate is set to 0.001, and a piecewise learning rate adjustment strategy is adopted (50% decrease every 300 rounds). In addition, to prevent gradient explosion, the gradient threshold is set to 1. The batch size is adjusted to 20, and the Dropout function is introduced with a drop rate of 0.5. After the optimization, the mean absolute error (MAE) of the model was reduced to 0.79±0.139mV, and the coefficient of determination (R²) increased to 99.91±0.05%. During the optimization process, the performance of the validation set was monitored every 50 iterations to ensure that the model could maintain high prediction accuracy on different data sets.

[0052] S500: Use the test set to simulate and predict the trained neural network model to achieve reverse reconstruction of cardiac potential;

[0053] S600: The prediction results were evaluated by the coefficient of determination (R²) and mean absolute error (MAE), a comparison chart of the reconstructed waveform and the real waveform of the cardiac node was generated, and the ECG signal was visualized using Map3D software.

[0054] The above embodiments constitute a complete technical solution of the present disclosure. The method described in this embodiment can quickly and effectively solve the ECG inverse problem through a bidirectional long short-term memory network and reconstruct the internal electrical activity of the heart. It has significant advantages in accuracy and stability, and provides a more reliable technical means for the diagnosis and treatment of cardiovascular diseases.

[0055] In another embodiment, clinical data collected by Thomas Berger's research team at the Medical University of Innsbruck, Austria, was obtained through the official website of ECGSIM software. ECGSIM is based on the equivalent surface source model and believes that the potential generated by the heart on the body surface is completely determined by the transmembrane potential on the myocardial surface. Figure 2 As shown in the figure, the three-dimensional geometric information of the heart source model and the BSP (64*505) and TMP (257*505) data of a normal heart in one cycle (505ms, simulation step length 1ms) were successfully obtained through ECGSIM software. Among them, BSP is the electrical signal measured on the human body surface through 64 electrode patches, and TMP reflects the difference in the potential inside and outside the myocardial cell membrane at 257 nodes of the heart. In order to further enrich the data set, ECGSIM is used to change the waveform of the heart source signal to generate different surface lead signals, merge and extend to 3030ms, and construct an ECG data set containing BSP (64×3030) and TMP (257×3030). The summarized ECG data collectively represents the waveform of the potential of lead No. 30 and the transmembrane potential of the heart node No. 30. Figure 3 shown.

[0056] In another embodiment, the data set is divided and the electrical activity information is sampled at 1 ms intervals, wherein 1 ms of sample data is taken every 4 ms, a total of 606 ms, to form a test set, and the remaining 4 ms of sample data, a total of 2424 ms, to form a training set. The divided data is normalized and scaled to the range of [0, 1] for better training and prediction.

[0057] It should be noted that the normalization operation is implemented through the following program, the program example is as follows:

[0058] %% Data normalization

[0059] [p_train, ps_input] = mapminmax(P_train, 0, 1); % Normalize the input training set and scale it to the [0, 1] interval

[0060] p_test = mapminmax('apply', P_test, ps_input); % Normalize the input test set using the same normalization parameters as the training set

[0061] [t_train, ps_output] = mapminmax(T_train, 0, 1); % Normalize the output training set and scale the output data to the [0, 1] interval

[0062] t_test = mapminmax('apply', T_test, ps_output); % Normalize the output test set using the same normalization parameters as the training set

[0063] In another embodiment, a BiLSTM neural network model is constructed, and the network structure is as follows: Figure 4 As shown, the construction process and parameter setting process are as follows:

[0064] Establish an input layer to receive the preprocessed BSP data; establish a bidirectional LSTM layer to process the bidirectional dependencies of the time series; add a ReLU activation layer to introduce nonlinearity to help the network capture complex patterns in the data while avoiding the gradient vanishing problem; establish a fully connected layer to map features to the output space; establish a regression layer to process regression tasks and calculate the network loss.

[0065] In this example, the training options are set and the Adam optimizer is used for gradient descent. To prevent overfitting, the Dropout function is used to randomly discard a certain proportion of neurons during the training process, and the discard rate is set to 0.5. To prevent gradient explosion, the gradient threshold is set to 1, the initial learning rate is specified to 0.001, and a piecewise learning rate adjustment strategy is used to reduce the learning rate by multiplying it by a factor of 0.5 after 300 rounds of training. During the training process, a validation set is used to help monitor the performance of the model, and validation is performed every 50 iterations. Data shuffling is added to reduce the model's tendency to learn patterns between batches and improve the model's generalization ability.

[0066] It should be noted that the BiLSTM neural network is built through the following program, the program example is as follows:

[0067] %% Create model

[0068] layers = [

[0069] sequenceInputLayer(inputSize) % input layer

[0070] bilstmLayer(500, 'OutputMode', 'sequence') % BiLSTM layer

[0071] dropoutLayer(0.5) % Dropout layer

[0072] reluLayer % ReLU activation layer

[0073] fullyConnectedLayer(outputSize) % fully connected layer

[0074] regressionLayer]; % regression layer

[0075] %% Parameter settings

[0076] options = trainingOptions('adam', ... % Adam optimizer

[0077] 'MaxEpochs', 900, ... % Maximum number of iterations

[0078] 'InitialLearnRate', 0.001, ... % The initial learning rate is adjusted to 0.001

[0079] 'LearnRateSchedule', 'piecewise', ... % piecewise learning rate adjustment

[0080] 'LearnRateDropFactor', 0.5, ... % Learning rate drop factor

[0081] 'LearnRateDropPeriod', 300, ... % Learning rate drop period

[0082] 'ValidationData', {p_test, t_test}, ...% validation data set

[0083] 'ValidationFrequency', 50, ... % Validation frequency

[0084] 'MiniBatchSize', 20, ... % Mini-batch size

[0085] 'GradientThreshold', 1, ... % Gradient threshold

[0086] 'Shuffle', 'every-epoch', ... % Shuffle data after each epoch

[0087] 'Plots', 'training-progress', ... % Plot the training progress curve

[0088] 'Verbose', false); % Disable detailed output

[0089] In another embodiment, the BiLSTM network architecture is trained according to the training set using the defined network structure and training options, and the operation is implemented by the trainNetwork function in MATLAB specifically used for training deep learning networks. After the training is completed, the trained neural network model net is simulated and predicted using the test set to achieve reverse reconstruction of the cardiac potential. After the simulation prediction is completed, the prediction results are denormalized to ensure that the output of the model has the same scale and range as the actual data for practical application or further analysis.

[0090] It should be noted that the BiLSTM neural network implements the training model, simulation prediction and data denormalization operations through the following program. The program example is as follows:

[0091] %% Training model

[0092] net = trainNetwork(p_train, t_train, layers, options);

[0093] %% Simulation prediction

[0094] t_sim1 = predict(net, p_train);

[0095] t_sim2 = predict(net, p_test );

[0096] %% Data denormalization

[0097] T_sim1 = mapminmax('reverse', t_sim1, ps_output);

[0098] T_sim2 = mapminmax('reverse', t_sim2, ps_output);

[0099] In another embodiment, the model reconstruction results are evaluated and analyzed, and the evaluation indicators determination coefficient (R²) and mean absolute error (MAE) are defined as follows:

[0100]

[0101]

[0102] In the formula, is the true value of cardiac transmembrane potential; is the average value of cardiac transmembrane potential; is the predicted value of cardiac transmembrane potential. For the whole heart, the determination coefficient of BiLSTM The absolute error is 99.91±0.05%, close to 1. It is 0.79±0.139mV, which is close to 0. It can be seen that the BiLSTM algorithm effectively realizes the reverse reconstruction of the cardiac potential.

[0103] In addition, a comparison diagram of the reconstructed waveform and the real waveform of the cardiac node is generated, such as Figure 5 As shown in the figure, a comparison diagram of the reconstructed waveform and the true waveform of the 150th node of the heart is given. It can be seen from the figure that the resting potential of the reconstructed waveform is close to -80mV, and the action potential peak of the reconstructed waveform is slightly less than 20mV, which is in line with the theoretical level. In addition, the amplitude of the BiLSTM reconstructed potential is basically consistent with the true value, which shows that the reconstruction effect of the BiLSTM algorithm is better.

[0104] At the same time, Map3D software is used to visualize the ECG signals by combining the three-dimensional geometric information of the heart source model and the predicted and true values ​​of TMP. Figure 6 As shown in the figure, the cardiac transmembrane potential at four moments, 235ms, 260ms, 267ms, and 272ms, is visualized. It can be seen from the figure that the reverse reconstruction effect of the BiLSTM algorithm is very close to the actual situation.

[0105] The above general description of the invention involved in this application and the description of its specific implementation methods should not be understood as limiting the technical solutions of the invention. Based on the disclosure of this application, those skilled in the art can add, reduce or combine the disclosed technical features in the above general description and / or specific implementation methods (including examples) without violating the constituent elements of the invention involved, to form other technical solutions within the protection scope of this application.

Claims

1. A method for solving the inverse electrocardiogram problem based on a neural network, characterized in that: The method comprises the following steps: S100: Acquire three-dimensional geometric information of a heart source model and electrocardiogram data of a normal heart in one cycle; S200: constructing an ECG data set based on the ECG data, dividing the data into a training set and a test set, and processing the data; S300: construct a neural network model and set network parameters; S400: training the neural network model using the training set, and optimizing model parameters during the training process to improve prediction accuracy; S500: Use the test set to simulate and predict the trained neural network model to achieve reverse reconstruction of cardiac potential; S600: Evaluate the prediction results, generate a comparison diagram of the reconstructed waveform and the real waveform of the heart node, and visualize the ECG signal.

2. The method for solving the inverse ECG problem based on a neural network according to claim 1, characterized in that: Preferably, in step S100, the period is 505 ms.

3. The method for solving the inverse ECG problem based on a neural network according to claim 1, characterized in that: In step S100, the electrocardiogram data includes body surface potential and transmembrane potential.

4. The method for solving the inverse ECG problem based on a neural network according to claim 3, characterized in that: The body surface potential is an electrical signal measured on the human body surface through a 64-electrode patch.

5. The method for solving the inverse ECG problem based on a neural network according to claim 3, characterized in that: The transmembrane potential reflects the difference in potential between the inner and outer membranes of the cardiac myocardial cells at 257 nodes of the heart.

6. The method for solving the inverse ECG problem based on a neural network according to claim 3, characterized in that: The ECG data set is a data set including body surface potential and transmembrane potential, which is constructed by changing the waveform of the cardiac source signal and using different body surface lead signals generated by ECGSIM and extending to 6 groups.

7. The method for solving the inverse ECG problem based on a neural network according to claim 1, characterized in that: In step S200, the data set is divided by sampling the electrical activity information at a time interval of 1 ms.

8. The method for solving the inverse ECG problem based on a neural network according to claim 1, characterized in that: In step S600, the evaluation of the prediction result includes using the determination coefficient R² and the mean absolute error MAE.

9. The method for solving the inverse ECG problem based on a neural network according to claim 8, characterized in that: The determination coefficient R² of the evaluation index is defined as follows: ； In the formula, is the true value of cardiac transmembrane potential; is the average value of cardiac transmembrane potential; It is the predicted value of cardiac transmembrane potential.

10. The method for solving the inverse ECG problem based on a neural network according to claim 8, characterized in that: The mean absolute error MAE is defined as follows: ； In the formula, is the true value of cardiac transmembrane potential; It is the predicted value of cardiac transmembrane potential.

Images