Method and device for positioning myocardial infarction through electrocardiogram based on geometric iterative network

Abstract

The invention provides a method and a device for positioning myocardial infarction by an electrocardiogram based on a geometric iterative network. The method comprises the following steps: establishing a corresponding relation between electrocardiogram data and myocardial infarction key nodes through an artificial intelligence model; wherein the corresponding relationship comprises a first corresponding sub-relationship, a second corresponding sub-relationship and a third corresponding sub-relationship; specifically, the first artificial intelligence sub-model establishes a first corresponding sub-relationship between electrocardiogram data and ventricular surface potential distribution; establishing a second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area through a second artificial intelligence sub-model; establishing a third corresponding sub-relationship between the myocardial infarction key area and the myocardial infarction key node through a third artificial intelligence sub-model; and obtaining target electrocardiogram data of a target patient, and determining a target myocardial infarction key node corresponding to the target electrocardiogram data through the corresponding relationship. The most critical nodes in the myocardial infarction area can be accurately identified through the geometric iterative network.

Description

Technical Field

[0001] The present application relates to the field of medical detection, and in particular to a method and device for locating myocardial infarction using an electrocardiogram based on a geometric iterative network. Background Art

[0002] Myocardial infarction, also known as acute myocardial infarction or coronary artery occlusion, refers to an acute, potentially life-threatening condition in which the heart muscle dies due to lack of blood supply and heart function is impaired due to acute blockage of the coronary artery.

[0003] Localizing myocardial infarction is necessary to inform clinical management and therapeutic intervention planning. It is crucial in guiding patient selection for revascularization procedures, helping to identify coronary artery obstruction and providing an assessment of severity. In addition, localizing myocardial infarction helps estimate the risk of associated complications, such as arrhythmias. Infarctions precipitate arrhythmias by altering the electrophysiological status of the myocardium. Therefore, localization of myocardial infarctions aids in the development of arrhythmia ablation strategies.

[0004] Determining the location of myocardial infarction is crucial for clinical management and treatment strategies. However, existing diagnostic tools for myocardial infarction either sacrifice practicality or are limited by spatial resolution. Summary of the invention

[0005] In view of the above problems, the present application is proposed to provide a method and device for locating myocardial infarction by electrocardiogram based on a geometric iteration network, which overcomes the above problems or at least partially solves the above problems, including:

[0006] A method for locating myocardial infarction by electrocardiogram based on a geometric iterative network, the method involving an artificial intelligence model, the artificial intelligence model comprising a first artificial intelligence sub-model, a second artificial intelligence sub-model and a third artificial intelligence sub-model; the method comprising:

[0007] A correspondence relationship between electrocardiogram data and a key node of myocardial infarction is established through an artificial intelligence model; wherein the correspondence relationship includes a first correspondence sub-relationship, a second correspondence sub-relationship and a third correspondence sub-relationship; specifically, a first correspondence sub-relationship between electrocardiogram data and ventricular surface potential distribution is established through the first artificial intelligence sub-model; a second correspondence sub-relationship between ventricular surface potential distribution and a key area of ​​myocardial infarction is established through the second artificial intelligence sub-model; a third correspondence sub-relationship between a key area of ​​myocardial infarction and a key node of myocardial infarction is established through the third artificial intelligence sub-model;

[0008] The target electrocardiogram data of the target patient is obtained, and the target myocardial infarction key node corresponding to the target electrocardiogram data is determined through the corresponding relationship pre-established by the artificial intelligence model.

[0009] Furthermore, the step of establishing the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution through the first artificial intelligence sub-model includes:

[0010] Determining the potential features of the electrocardiogram data in the time domain and the frequency domain using discrete multiwavelet operations;

[0011] Performing feature compensation on the potential features based on a preset sandwich structure operator to determine graphic information with a data dimension consistent with the ventricular surface potential distribution;

[0012] Geometric constraints are embedded based on the graphic information of nodes and edges, and the self-learning ability of the first artificial intelligence sub-model is utilized to establish the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution.

[0013] Furthermore, the step of using discrete multiwavelet operation to determine the potential features of the electrocardiogram data in the time domain and the frequency domain includes:

[0014] Decomposing the electrocardiogram data by using orthogonal multi-wavelet to obtain decomposed electrocardiogram data;

[0015] Determine simplified electrocardiogram information corresponding to the decomposed electrocardiogram data by downsampling and zero-filling operations;

[0016] Discrete multi-wavelet operation is used to determine the potential features of the simplified electrocardiogram information in the time domain and the frequency domain.

[0017] Furthermore, the step of embedding the graphic information based on nodes and edges with geometric constraints and establishing the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution by using the self-learning ability of the first artificial intelligence sub-model includes:

[0018]

[0019] u k =P l (y k -γf(y k )) (2)

[0020] y k+1 =u k +τ k+1 (u k -u k-1 ) (3)

[0021]

[0022] y k+1 =u k +τk (u k -u k-1 ) (5)

[0023] Where, T k+1 is the iteration step size, which is used to control the amplitude of variable update during the iteration process; u k is the intermediate variable in the iteration process; y k is the target variable in the iterative process; γ is the regularization parameter, which is used to control the size of the gradient descent step; is the transpose of the adaptive gradient operator during iteration, used to adjust u k Update direction of Hy k Initialize items H to y for the application k The result of ; where in step k, H is used as the initialization term; P represents the proximal operator supported by the network embedded with constraints; G is the adaptive gradient operator during iteration; τ k > 0 can be learned through training defined in equation (1); ρ is the shrinkage operator ρ(u k )=(|u k |-ρ)+sgn(u k ), ()+ is |u k |-ρ is the maximum value between 0 and 0, sgn represents the sign function, limited to ±1.

[0024] Furthermore, the step of establishing the second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area through the second artificial intelligence sub-model includes:

[0025] Acquire a medical image dataset; wherein the medical image dataset includes a normal myocardial image dataset, a myocardial infarction region image dataset, and a pathological change image dataset related to myocardial infarction;

[0026] The second artificial intelligence sub-model is trained based on the medical image data set to determine the second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area.

[0027] Furthermore, the second artificial intelligence sub-model includes a physical sub-model and a neural network, and the step of training the second artificial intelligence sub-model based on the medical image data set to determine the second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area includes:

[0028] The expression of the electric potential is approximated based on the neural network; the formula is as follows:

[0029] u(x,y,z,t)≈N(x,y,z,t;θ);

[0030] In the formula, θ is the parameter of the neural network; x, y, z are spatial coordinates; t represents time;

[0031] Substitute the expression of the electric potential into the physical sub-model for describing the cardiac electrophysiological process, train the second artificial intelligence sub-model, and make the neural network satisfy the laws of physics based on the loss function; the formula is as follows:

[0032]

[0033] Where α is the weight parameter that balances the two loss terms; p i is at position (x i ,y i ,t i ) Observed probability of myocardial infarction; N data is the total number of observed data points; is the true potential value at position j; is the potential value predicted by the neural network;

[0034] The back propagation algorithm is used to update the weight W and bias b; the formula is as follows:

[0035]

[0036] The update rules are as follows:

[0037]

[0038] Where L is the loss function used to evaluate the difference between the model's prediction and the actual observed value; y is the output of the neural network, representing the predicted potential value; η is the learning rate;

[0039] After each training, if the difference between the predicted value and the true value is still greater than the preset minimum tolerance difference, the neural network returns to continue the cyclic training until the difference between the predicted value and the true value is no greater than the preset minimum tolerance difference.

[0040] Furthermore, the step of establishing the third corresponding sub-relationship between the myocardial infarction key area and the myocardial infarction key node through the third artificial intelligence sub-model includes:

[0041] The node features in the myocardial infarction key area are input into a preset cross entropy loss function, and the output parameter of the cross entropy loss function is determined to be the corresponding myocardial infarction key node; wherein the cross entropy loss function is as follows:

[0042]

[0043] Where N is the number of nodes; y i is the true label of node i; is the model-predicted probability of infarction.

[0044] An electrocardiogram positioning myocardial infarction device based on a geometric iterative network, the device relates to an artificial intelligence model, the artificial intelligence model includes a first artificial intelligence sub-model, a second artificial intelligence sub-model and a third artificial intelligence sub-model; including:

[0045] A correspondence establishing module, used to establish a correspondence between electrocardiogram data and a key node of myocardial infarction through an artificial intelligence model; wherein the correspondence includes a first corresponding sub-relationship, a second corresponding sub-relationship and a third corresponding sub-relationship; specifically, a first corresponding sub-relationship between electrocardiogram data and ventricular surface potential distribution is established through the first artificial intelligence sub-model; a second corresponding sub-relationship between ventricular surface potential distribution and a key area of ​​myocardial infarction is established through the second artificial intelligence sub-model; a third corresponding sub-relationship between a key area of ​​myocardial infarction and a key node of myocardial infarction is established through the third artificial intelligence sub-model;

[0046] The data processing module is used to obtain the target electrocardiogram data of the target patient, and determine the target myocardial infarction key node corresponding to the target electrocardiogram data through the corresponding relationship pre-established by the artificial intelligence model.

[0047] A device comprises a processor, a memory and a computer program stored in the memory and capable of running on the processor, wherein when the computer program is executed by the processor, the steps of the method for locating myocardial infarction by electrocardiogram based on geometric iterative network are implemented as described above.

[0048] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for locating myocardial infarction by electrocardiogram based on geometric iterative network are implemented as described above.

[0049] This application has the following advantages:

[0050] In the embodiments of the present application, with respect to the technical problem in the prior art that "existing myocardial infarction diagnostic tools either sacrifice practicality or are limited by spatial resolution", the present application provides a solution for accurately locating myocardial infarction by reconstructing the ventricular surface potential of electrocardiogram data through an artificial intelligence model, specifically: establishing a correspondence between electrocardiogram data and key nodes of myocardial infarction through an artificial intelligence model; wherein the correspondence includes a first corresponding sub-relationship, a second corresponding sub-relationship and a third corresponding sub-relationship; specifically, establishing a first corresponding sub-relationship between electrocardiogram data and ventricular surface potential distribution through the first artificial intelligence sub-model; establishing a second corresponding sub-relationship between ventricular surface potential distribution and key areas of myocardial infarction through the second artificial intelligence sub-model; establishing a third corresponding sub-relationship between key areas of myocardial infarction and key nodes of myocardial infarction through the third artificial intelligence sub-model; acquiring target electrocardiogram data of a target patient, and determining the target key node of myocardial infarction corresponding to the target electrocardiogram data through the corresponding relationship. The first artificial intelligence sub-model is used to reconstruct the ventricular surface potential distribution of electrocardiogram data, showing excellent accuracy on both synthetic and clinical datasets; the second artificial intelligence sub-model is used to enable the network to not only learn patterns in the data, but also follow the laws of physics to determine the key areas of myocardial infarction; the third artificial intelligence sub-model is used to process graph-structured data and learn the relationships between nodes in the graph, so as to accurately identify the most critical nodes in the myocardial infarction area. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the description of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0052] Figure 1 It is a flowchart of the steps of a method for locating myocardial infarction using an electrocardiogram based on a geometric iteration network provided in one embodiment of the present application;

[0053] Figure 2 is a schematic diagram of the structure of a geometric iteration network provided in an embodiment of the present application;

[0054] Figure 3 It is a structural block diagram of an electrocardiogram locating myocardial infarction device based on a geometric iteration network provided in an embodiment of the present application;

[0055] Figure 4 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the objects, features and advantages of the present application more obvious and understandable, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0057] The inventors analyzed the prior art and found that in clinical practice, one standard for locating myocardial infarction is late gadolinium-enhanced cardiac magnetic resonance. However, this technology has some limitations. Late gadolinium-enhanced cardiac magnetic resonance requires patients to have sufficient breath-holding ability. It also requires gadolinium-based contrast agents, which have risks such as nephrogenic systemic sclerosis and long-term cerebral retention. In addition, large amounts of gadolinium are quite expensive for patients and the healthcare system. This is due to the high operating expenses associated with the equipment.

[0058] Body surface potential mapping has become an advanced tool for localizing myocardial infarction. As an extension of the 12-lead electrocardiogram, body surface potential mapping uses additional electrodes to capture comprehensive electrocardiogram profiles. It can be combined with a cardiac anatomical model to visualize and analyze cardiac electrical activity through an inverse solution. With this inverse solution, it is possible to accurately visualize and analyze cardiac electrical activity. This allows the combination of heart and data. Recently, various deep learning and machine learning techniques have been used to solve inverse problems in electrocardiograms, aiming to obtain robust inverse solutions. However, due to its additional operational complexity and expense, body surface potential mapping has not been widely adopted in clinical practice. This has hindered the rapid clinical promotion and adoption of inverse technology. Therefore, it is important to find a method to localize myocardial infarction using 12-lead electrocardiograms.

[0059] By reconstructing the transmembrane potential distribution through 12-lead electrocardiogram and identifying the abnormal potential distribution area, the purpose of locating myocardial infarction can be achieved. This reconstruction of the ventricular surface potential will greatly improve its clinical value. This method is expected to be more widely adopted and disseminated in clinical settings.

[0060] However, the 12-lead ECG has limitations in localization resolution and can only distinguish wider areas. These areas include inferior, lateral and posterior, or a combination of them, such as inferior lateral and inferior posterior infarction. This limitation is related to the indirect representation of electrical activity. The 12-lead ECG only provides a narrowed view of the electrical activity unfolding in space and time. Therefore, this hides key spatial information that is crucial in localization. Reconstructing the ventricular surface potential distribution from the 12-lead ECG presents substantial ill-posedness, which means that the same ECG data may correspond to different potential distributions. This may lead to physiologically unreasonable or even erroneous distribution results. This problem stems from the complex propagation of cardiac potentials in tissues such as lungs and skin. The different electrical conductivities in these tissues lead to attenuation and distortion. When these potentials are captured by body surface electrodes, the ECG signals contain mixed noise. These noises are amplified during the reconstruction process. This affects feature mapping and complicates the capture of accurate electrical activity. In addition, the 12-lead ECG only provides 12 surface electrical records, limiting its data dimensionality.

[0061] To overcome the above problems, the inventors introduced a frequency-enhanced geometrically constrained iterative network. The network first mines potential features in the time and frequency domains from ECG data. Subsequently, the method increases the dimensionality of ECG data and uses convolutional layers to capture complex features. Finally, the geometrically constrained iterative network uses the ventricular geometry as a constraint on the surface potential distribution. It assigns variable weights to different edges.

[0062] Reference Figure 1 , showing a method for locating myocardial infarction using an electrocardiogram based on a geometric iteration network provided in an embodiment of the present application;

[0063] The method comprises:

[0064] S1. Establishing a correspondence between electrocardiogram data and key nodes of myocardial infarction through an artificial intelligence model; wherein the artificial intelligence model includes a first artificial intelligence sub-model, a second artificial intelligence sub-model and a third artificial intelligence sub-model; specifically, establishing a first correspondence sub-relationship between electrocardiogram data and ventricular surface potential distribution through the first artificial intelligence sub-model; establishing a second correspondence sub-relationship between ventricular surface potential distribution and key areas of myocardial infarction through the second artificial intelligence sub-model; establishing a third correspondence sub-relationship between key areas of myocardial infarction and key nodes of myocardial infarction through the third artificial intelligence sub-model;

[0065] S2. Obtain target electrocardiogram data of a target patient, and determine the key node of myocardial infarction corresponding to the target electrocardiogram data through the corresponding relationship pre-established by the artificial intelligence model.

[0066] In the embodiments of the present application, with respect to the technical problem in the prior art that "existing myocardial infarction diagnostic tools either sacrifice practicality or are limited by spatial resolution", the present application provides a solution for accurately locating myocardial infarction by reconstructing the ventricular surface potential of electrocardiogram data through an artificial intelligence model, specifically: establishing a correspondence between electrocardiogram data and key nodes of myocardial infarction through an artificial intelligence model; wherein the correspondence includes a first corresponding sub-relationship, a second corresponding sub-relationship and a third corresponding sub-relationship; specifically, establishing a first corresponding sub-relationship between electrocardiogram data and ventricular surface potential distribution through the first artificial intelligence sub-model; establishing a second corresponding sub-relationship between ventricular surface potential distribution and key areas of myocardial infarction through the second artificial intelligence sub-model; establishing a third corresponding sub-relationship between key areas of myocardial infarction and key nodes of myocardial infarction through the third artificial intelligence sub-model; acquiring target electrocardiogram data of a target patient, and determining the target key node of myocardial infarction corresponding to the target electrocardiogram data through the corresponding relationship. The first artificial intelligence sub-model is used to reconstruct the ventricular surface potential distribution of electrocardiogram data, showing excellent accuracy on both synthetic and clinical datasets; the second artificial intelligence sub-model is used to enable the network to not only learn patterns in the data, but also follow the laws of physics to determine the key areas of myocardial infarction; the third artificial intelligence sub-model is used to process graph-structured data and learn the relationships between nodes in the graph, so as to accurately identify the most critical nodes in the myocardial infarction area.

[0067] Next, a method for locating myocardial infarction using an electrocardiogram based on a geometric iterative network in this exemplary embodiment will be further described.

[0068] As described in step S1, a correspondence between electrocardiogram data and key nodes of myocardial infarction is established through an artificial intelligence model; specifically, a first corresponding sub-relationship between electrocardiogram data and ventricular surface potential distribution is established through the first artificial intelligence sub-model; a second corresponding sub-relationship between ventricular surface potential distribution and key areas of myocardial infarction is established through the second artificial intelligence sub-model; and a third corresponding sub-relationship between key areas of myocardial infarction and key nodes of myocardial infarction is established through the third artificial intelligence sub-model.

[0069] It should be noted that, refer to Figure 2 , the artificial intelligence model is a geometric constraint iterative network, and the artificial intelligence model includes a first artificial intelligence sub-model, a second artificial intelligence sub-model and a third artificial intelligence sub-model.

[0070] The first artificial intelligence sub-model is a frequency-enhanced geometrically constrained iterative network, which includes a basic framework, a frequency-enhanced feature module, and a geometrically constrained embedding module. By introducing a frequency-enhanced geometrically constrained iterative network to reconstruct the ventricular surface potential distribution of a standard 12-lead electrocardiogram, it shows excellent accuracy on both synthetic and clinical datasets.

[0071] The second AI sub-model is a physical information neural network, which combines deep learning with physical laws, using the powerful nonlinear mapping capabilities of neural networks to learn complex functional relationships through large amounts of data. The physical information neural network is a deep learning model that combines physical laws with the powerful function approximation capabilities of neural networks, embedding physical equations in the neural network so that the network can not only learn patterns in the data, but also follow the laws of physics to determine the key areas of myocardial infarction.

[0072] The third artificial intelligence sub-model is a graph attention mechanism, which is used to enhance the interaction and information transfer between nodes, and iteratively aggregate the information of neighboring nodes to learn nodes. The graph attention mechanism is used to process graph structure data, learn the relationship between nodes in the graph, and assign different attention weights to each node based on these relationships to identify the most critical nodes in the myocardial infarction area.

[0073] In one embodiment of the present invention, the specific process of "establishing the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution through the first artificial intelligence sub-model" in step S1 can be further explained in combination with the following description.

[0074] S111, using discrete multi-wavelet operation to determine the potential features of the electrocardiogram data in the time domain and the frequency domain;

[0075] It should be noted that the electrocardiogram data refers to the electrocardiogram sequence obtained from the patient's 12-lead electrocardiogram, and these data contain the electrical activity information of the heart at different time points.

[0076] It should be noted that the time domain describes the change of a signal over time, captures the instantaneous characteristics, and focuses on the instantaneous behavior and timing relationship of the signal; the frequency domain reveals the frequency distribution and spectral characteristics of the signal, and understands the contributions of different frequencies. Through time-domain analysis, the dynamic behavior of the electrocardiogram (ECG) signal on the time axis can be intuitively observed, including key information such as the instantaneous characteristics, waveform shape, amplitude, and phase of the signal. In the analysis of electrocardiogram (ECG) signals, time-domain analysis can help doctors identify abnormal rhythms and waveform changes of the heart by analyzing the frequency components of the electrocardiogram signal and understanding the activity characteristics of the heart in different states. Through frequency-domain analysis, the frequency components of the signal are revealed. By representing the signal as a superposition of frequency components, frequency-domain analysis can describe the frequency characteristics of the signal, including the different frequency components contained in the signal and their relative contributions.

[0077] It should be noted that the geometric constraint iterative network with frequency enhancement (the first artificial intelligence sub-model) consists of a geometric constraint iterative network framework, a frequency enhancement feature attention mechanism, and corresponding geometric constraint embeddings. Among them, the frequency enhancement feature attention mechanism includes a frequency operation module, discrete multi-wavelet operation, a long-term memory self-attention mechanism, and feature compensation. In this step, the potential features of the electrocardiogram data in the time domain and the frequency domain are calculated based on the frequency operation module, discrete multi-wavelet operation, and long-term memory self-attention mechanism.

[0078] In an embodiment of the present invention, the specific process of "determining the potential features of the electrocardiogram data in the time domain and the frequency domain by using discrete multi-wavelet operation" can be further described in combination with the following description.

[0079] As described in the following steps, the electrocardiogram data is decomposed by using orthogonal multi-wavelets to obtain decomposed electrocardiogram data; downsampling and zero-padding operations are used to determine the simplified electrocardiogram information corresponding to the decomposed electrocardiogram data; discrete multi-wavelet operation is used to determine the potential features of the simplified electrocardiogram information in the time domain and the frequency domain.

[0080] It should be noted that the frequency operation module: The frequency operation module works in the frequency domain and processes the decomposed sequence Y∈C L×T To capture the prominent features related to electrical activity and minimize the computational load, a downsampling operation S is introduced. This results in a simplified expression where M << L. Then, the feature extraction kernel EK∈C T ×T×M is applied to whose learnable parameters are as follows:

[0081]

[0082] In the formula, represents the original frequency domain sequence; C represents the complex domain; L represents the number of original frequency components; T represents time; M represents the number of frequency components after downsampling; t i ∈[1,T] represents the input channel and lead number; t o ∈[1,T] represents the output channel; m represents the index of the frequency component; in order to restore the data dimension reduced by the S operation, a zero padding operation is then applied to obtain

[0083]

[0084] In the formula, Represents the frequency domain sequence after the downsampling operation; Padding represents the zero padding operation.

[0085] It should be noted that the discrete multiwavelet operation: This block includes two key operations: decomposition and representation. Given an input sequence, orthogonal multiwavelets are used for decomposition due to their robust performance in noise reduction and signal recovery applications. The multiwavelet transform coefficients at scale j and position l are defined as

[0086]

[0087] In the formula, represents the approximate coefficient at scale j, which is used to represent the low-frequency part of the signal; f represents the original input signal; k represents the number of approximate coefficients at scale j; i represents the index of the approximate coefficient; and the measure μ is used. j and Characterization, represents the inner product; Wavelet orthogonal basis representing piecewise polynomials; represents the detail function at scale j, which is used to capture the detail information of the signal. The cross-scale decomposition is defined as:

[0088]

[0089] Wherein, the linear coefficients (H(0), H(1), G(0), G(1)) are used as multiwavelet decomposition filters using Legendre polynomials, determined by Gaussian orthogonality and Gram-Schmidt orthogonality. Based on empirical evidence and the length of the ECG data, the decomposition level is set to 4. The filter length is set to 480. Zero padding is selected as the boundary processing method. Each ECG signal is cropped to 480 time steps (some redundant parts after removing the T wave). The representation in the multiwavelet frequency domain is defined as in Represents multi-scale, multi-wavelet coefficients, with H and G as low-pass filters and high-pass filters, respectively. (A j , B j , Cj ) is an independent frequency operation module for processing different sequences during decomposition, r is the coarsest scale during recursive decomposition, It is the fully connected layer for processing the decomposed sequence. The multi-wavelet basis, reconstruction level, filter coefficients and other implementation settings are consistent with those used in the decomposition process.

[0090] By analyzing the distribution of different frequency components in the ECG signal, it is possible to determine whether the heart has an abnormal rhythm, such as arrhythmia. Decomposing the ECG signal into frequency components of different scales helps to filter out noise and interference, enhance frequency domain features, thereby improving signal quality and providing more accurate data support for subsequent ECG signal analysis and processing.

[0091] It should be noted that the long memory self-attention mechanism: when analyzing the electrocardiogram data y∈R L×T When we use an independent fully connected layer, the adaptive weight matrix w Q ,w K ,w V ∈R L×L Modify y to get y Q =w Q ·y,y K =w K ·y,y V =w V ·y. Then, for y Q ,y K ,y V Discrete multi-wavelet operation is performed to enhance the frequency domain characteristics, and finally the Finally, the self-attention mechanism is applied to the frequency domain, and the activation function is where d K is the dimension L. This self-attention mechanism eventually produces Keep the basic features ( It is also the y during iteration step k in equation (2) k ). The benefits of applying the self-attention mechanism are as follows: First, self-attention enhances the analysis of the ECG by adaptively focusing on diagnostically critical regions associated with the infarct site. Therefore, it can improve the efficiency of providing useful features for the following blocks. Second, it can also capture long-term dependencies across the ECG data. This is very beneficial because the information about the location of central myocardial infarction in the ECG data usually spans multiple time steps.

[0092] S112, performing feature compensation on the potential features based on a preset sandwich structure operator to determine graphic information whose data dimension is consistent with the ventricular surface potential distribution;

[0093] It should be noted that feature compensation in the frequency-enhanced geometrically constrained iterative network is used to output data dimensions consistent with the ventricular surface potential distribution. Feature compensation involves creating a sandwich structure operator Γ, which consists of a two-layer convolution (layer C and layer D) containing an active ReLU layer: Γ(·) = D(ReLU(C(·))). To enlarge the network depth to encapsulate the complex forward relationship, an additional convolution layer E is added before the double Γ operation: In order to maintain a closed process, the left inverse of Φ is introduced, denoted as Φ -1 , satisfying Φ -1 Φ = I (where I represents the unit operator), resulting in

[0094] This block implements the proximal mapping of features extracted from the 12-lead ECG (input from the long-memory self-attention block) to an output whose data dimension is consistent with the ventricular surface potential distribution (input geometry-constrained embedding mechanism).

[0095] S113. Perform geometric constraint embedding based on the graphic information of nodes and edges, and utilize the self-learning ability of the first artificial intelligence sub-model to establish the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution.

[0096] It should be noted that the first artificial intelligence model is established, and geometric constraints are set for the first artificial intelligence sub-model. The geometric constraint embedding module uses the ventricular geometry as a constraint for the surface potential distribution, and simulates the propagation mode of cardiac electrical activity by assigning variable weights to different edges, thereby outputting a reconstructed ventricular surface potential distribution with high accuracy and robustness, overcoming the instability of reconstructing the ventricular surface potential distribution from the 12-lead electrocardiogram.

[0097] It should be noted that the geometric constraint iterative network framework: integrates the neural network into equations (2) and (3), The estimation of is transformed into the following process of equations (4) and (5):

[0098]

[0099] u k =P l (y k -γf(y k )) (2)

[0100] y k+1 =u k +τ k+1 (u k -u k-1 ) (3)

[0101]

[0102] y k+1 =u k +τ k (u k -u k-1 ) (5)

[0103] Where, T k+1 is the iteration step size, which is used to control the amplitude of variable update during the iteration process; u k is the intermediate variable in the iteration process; y k is the target variable in the iterative process; γ is the regularization parameter, which is used to control the size of the gradient descent step; is the transpose of the adaptive gradient operator during iteration, used to adjust u k Update direction of Hy k Initialize items H to y for the application k The result of ; where in step k, H is used as the initialization term; P represents the proximal operator supported by the network embedded with constraints; G is the adaptive gradient operator during iteration; τ k > 0 can be learned through training defined in equation (1); ρ is the shrinkage operator ρ(u k )=(|u k |-ρ)+sgn(u k ), ()+ is |u k |-ρ is the maximum value between 0 and 0, sgn represents the sign function, limited to ±1.

[0104] Implementation details in P: The network implementation in P is twofold: the proximal part and the constrained part. The proximal part mainly involves the frequency-enhanced feature attention mechanism. The constrained part involves geometrically constrained embedding.

[0105] It should be noted that geometric constraints are embedded: In order to correctly implement the regularizer in non-Euclidean space, it is crucial to analyze the spatial relationships between nodes. These relationships are intrinsically related to the ventricular anatomy. Given that the focus is only on the ventricular surface, we restrict our considerations to a geometric representation using triangular surface elements. The representation consists of node indices υ = [N, 3] and an adjacency matrix A = [N, N], which are defined by the ventricular triangle mesh. Therefore, the ventricular undirected graph is described as G = (υ, A). The edge indices ε = [2, num_edges] are formed by merging the node indices. The edge weights are obtained from the adjacency matrix using the corresponding indices.

[0106] In summary, the problems of complex ECG signal propagation and noise interference are solved by mining the potential features in the time and frequency domains, increasing the data dimension, using convolutional layers to capture complex features, and embedding with geometric constraints. The frequency domain features of ECG signals are enhanced through operations such as downsampling, feature extraction kernels, and zero padding, providing high-quality data for subsequent processing. Through decomposition and representation operations, ECG signals are decomposed into frequency components of different scales, noise and interference are filtered out, and frequency domain features are enhanced. Through the self-attention mechanism, the analysis efficiency of ECG is enhanced and the accuracy of myocardial infarction location is improved. Through operations such as convolutional layers and active ReLU layers, complex forward relations are encapsulated to maintain a closed processing process. Through steps such as initialization terms, proximal operators, and adaptive gradient operators, the real ventricular surface potential distribution is gradually approached. Through geometric constraint embedding, the propagation pattern of cardiac electrical activity is simulated to ensure that the potential around the infarction boundary is consistent with the physiological local structural characteristics.

[0107] In one embodiment of the present invention, the specific process of "training the second artificial intelligence sub-model based on the medical imaging data set to determine the second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area" described in step S1 can be further explained in combination with the following description.

[0108] It should be noted that the physical information neural network (the second artificial intelligence sub-model) is a machine learning method that combines a physical model and a deep learning network. In locating myocardial infarction, this embodiment constructs a neural network that not only learns patterns in the data, but also must satisfy the physical laws of the cardiac electrophysiological model. The physical information neural network includes five blocks: data collection, physical model, neural network, loss function, and cyclic training. The five blocks of the physical information neural network are introduced in detail in conjunction with the following content.

[0109] S121, acquiring a medical image dataset; wherein the medical image dataset includes a normal myocardial image dataset, a myocardial infarction region image dataset, and a pathological change image dataset related to myocardial infarction;

[0110] It should be noted that data collection: First, collect medical imaging data sets containing myocardial infarction, including normal myocardium, myocardial infarction area, and pathological changes related to myocardial infarction. These data can borrow some of the data used in the frequency-enhanced geometric constraint iterative network, as long as they meet certain standards. In order to ensure the training effect, a certain number of CT, MRI or echocardiography images of normal myocardium need to be obtained from the hospital's medical imaging database. Make sure that the dataset covers individuals of different ages, genders, body shapes, and health conditions to increase the generalization ability of the model. Next, accurately label the normal myocardial area for the supervised learning process during model training.

[0111] S122. Train a second artificial intelligence sub-model based on the medical image data set to determine the second corresponding sub-relationship between the ventricular surface potential distribution and the key area of ​​myocardial infarction.

[0112] It should be noted that the physical model is used to describe the cardiac electrophysiological process. A common cardiac electrophysiological model is the dual-domain model, in which myocardial cells are divided into two regions: intracellular and extracellular. The physical model includes:

[0113] ① Intracellular potential equation

[0114]

[0115] In the formula, is the electric potential u i The rate of change over time; represents the divergence operator; σ i is the conductivity within the cell; ρ i is the resistivity inside the cell; I i is the ionic current within the cell.

[0116] ②Extracellular potential equation

[0117]

[0118] In the formula, is the electric potential u e The rate of change over time; represents the divergence operator; σ e is the extracellular conductivity; ρ e is the resistivity outside the cell; I e is the extracellular ionic current.

[0119] ③Ion current

[0120] Ion current I i and I e is generated by ion channel activity and can usually be expressed as:

[0121]

[0122] In the formula, g oj and g ej is the current density of the ion channel; and is the equilibrium potential of the ion channel.

[0123] It should be noted that the neural network: In the physical model, the key function we use is the electric potential u. Using the previously collected and annotated data, we can use a neural network to approximate the expression of u. Define a neural network N, whose input is (x, y, z, t) and whose output is the electric potential u:

[0124] u(x,y,z,t)≈N(x,y,z,t;θ)

[0125] Where θ is the parameter of the neural network; x, y, z are spatial coordinates; t represents time.

[0126] In this way, the known approximate expression of u can be substituted back into the formula of the physical model to train the neural network.

[0127] It should be noted that the loss function: The loss function ensures that the neural network satisfies the laws of physics:

[0128]

[0129] Where α is the weight parameter that balances the two loss terms; p i is at position (x i ,y i ,t i ) Observed probability of myocardial infarction; N data is the total number of observed data points; is the true potential value at position j; is the potential value predicted by the neural network.

[0130] It should be noted that during the training process, we use the back propagation algorithm to update the weight W and bias b:

[0131]

[0132] The update rules are as follows:

[0133]

[0134] Where L is the loss function, which is used to evaluate the difference between the model's prediction and the actual observed value; y is the output of the neural network, representing the predicted potential value; and η is the learning rate.

[0135] A minimum tolerance difference between the predicted value and the true value is set in advance. After each training, if the difference is still greater than the set minimum tolerance difference, the neural network is returned to continue the cyclic training. After a certain number of cycles, if it is no greater than the set minimum tolerance difference, the training of the fitting function is completed.

[0136] In summary, the powerful function approximation ability of neural networks is used to learn patterns in the data and follow the laws of physics to identify the critical areas of myocardial infarction.

[0137] In one embodiment of the present invention, the specific process of "establishing the third corresponding sub-relationship between the myocardial infarction key area and the myocardial infarction key node through the third artificial intelligence sub-model" in step S1 can be further explained in combination with the following description.

[0138] S131, inputting the node features in the myocardial infarction key area into a preset cross entropy loss function, and determining the output parameter of the cross entropy loss function as the corresponding myocardial infarction key node; wherein the cross entropy loss function is as follows:

[0139]

[0140] Where N is the number of nodes; y i is the true label of node i; is the model-predicted probability of infarction.

[0141] It should be noted that the graph attention mechanism (the third AI sub-model): applying the graph attention mechanism to myocardial infarction localization can use the vascular network structure of the heart to represent the various parts of the heart through a graph, and then use the graph attention mechanism to identify and locate the myocardial infarction area. The graph attention mechanism includes:

[0142] ⑴ Input layer

[0143] The input layer receives the node feature matrix X∈R N×D , where N is the number of nodes and D is the feature dimension.

[0144] ⑵Attention weight calculation

[0145] For any edge e between nodes i and j ij , calculate the attention weight a between them ij (h i ,h j ). This can be achieved using the following formula:

[0146]

[0147] In the formula, h i is the feature vector of node i; h j is the feature vector of node j; W a is the weight matrix, used for linear transformation; b a is the bias term; Represents the dot product of the feature vector; LeakyReLU is a nonlinear activation function used to introduce nonlinear relationships.

[0148] ⑶Model training

[0149] A cross entropy loss function is introduced to optimize the model parameters:

[0150]

[0151] In the formula, y i is the true label of node i (e.g., whether it is an infarct area); is the model-predicted probability of infarction.

[0152] In summary, through the graph attention mechanism, the most critical nodes in the myocardial infarction area can be identified, and the accuracy of myocardial infarction positioning can be improved.

[0153] As described in step S2, the target electrocardiogram data of the target patient is obtained, and the key node of myocardial infarction corresponding to the target electrocardiogram data is determined through the corresponding relationship pre-established by the artificial intelligence model.

[0154] The target 12-lead ECG data of the target patient's six limb leads and six precordial leads are collected by an ECG acquisition device, and the target 12-lead ECG data are input into the trained artificial intelligence model, which is used to analyze the target patient's ECG data and output key nodes of myocardial infarction.

[0155] The beneficial effects achieved by the present invention include:

[0156] (1) Compared with the existing surface potential mapping, the present invention solves the pathological nature of surface potential reconstruction by solving two causes of surface potential reconstruction: complex propagation and noise. For complex propagation, the graphical information of nodes and edges is used for geometric constraint embedding to learn the relationship between the 12-lead electrocardiogram and the ventricular surface potential distribution. This embedding controls the direction of feature changes at the potential level by introducing weight distribution at the edges between geometric nodes on the ventricular surface, thereby simulating the propagation pattern of cardiac electrical activity. This effectively constrains the reconstructed output to conform to the ventricular transmembrane potential distribution pattern, especially ensuring that the potential around the infarct boundary is consistent with the physiological local structural characteristics.

[0157] (2) For noise, the frequency-enhanced geometrically constrained iterative network imposes a frequency-enhanced feature attention mechanism to reduce the noise component while highlighting the target features during reconstruction. This mechanism focuses not only on time domain features but also on frequency domain features through discrete multi-wavelet decomposition. This decomposition uses a set of wavelet bases to decompose the ECG signal into approximate components (low frequency, long time scale) and detail components (high frequency, short time scale) of different scales. Each wavelet basis captures specific frequency information at a specific time point without interfering with each other. This allows the high-frequency part of the signal to be analyzed at a finer time scale while maintaining a global view of the low-frequency part. It provides frequency information at different scales while preserving the temporal distribution of these frequency components.

[0158] (3) With the help of the self-attention mechanism, the frequency-enhanced geometrically constrained iterative network is able to focus on the target features related to infarct localization from these decomposed components while reducing the noise component. After signal denoising and feature extraction, the corresponding inverse transform can be used to represent the signal. Therefore, the noise in the 12-lead ECG that may affect the reconstruction process can be greatly reduced.

[0159] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0160] Reference Figure 3 , showing an electrocardiogram positioning myocardial infarction device based on a geometric iterative network provided by an embodiment of the present application, the device involves an artificial intelligence model, and the artificial intelligence model includes a first artificial intelligence sub-model, a second artificial intelligence sub-model and a third artificial intelligence sub-model;

[0161] Specifically include:

[0162] The corresponding relationship establishing module 310 is used to establish the corresponding relationship between the electrocardiogram data and the key nodes of myocardial infarction through the artificial intelligence model; wherein the corresponding relationship includes a first corresponding sub-relationship, a second corresponding sub-relationship and a third corresponding sub-relationship; specifically, the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution is established through the first artificial intelligence sub-model; the second corresponding sub-relationship between the ventricular surface potential distribution and the key area of ​​myocardial infarction is established through the second artificial intelligence sub-model; the third corresponding sub-relationship between the key area of ​​myocardial infarction and the key node of myocardial infarction is established through the third artificial intelligence sub-model;

[0163] The data processing module 320 is used to obtain the target electrocardiogram data of the target patient, and determine the target myocardial infarction key node corresponding to the target electrocardiogram data through the corresponding relationship pre-established by the artificial intelligence model.

[0164] In one embodiment of the present invention, the corresponding relationship establishing module 310 includes:

[0165] A discrete multi-wavelet operation submodule, used for determining the potential features of the electrocardiogram data in the time domain and the frequency domain by using discrete multi-wavelet operation;

[0166] A feature compensation submodule, used for performing feature compensation on the potential features based on a preset sandwich structure operator, and determining graphic information whose data dimension is consistent with the ventricular surface potential distribution;

[0167] The geometric constraint embedding submodule is used to perform geometric constraint embedding based on the graphic information of nodes and edges, and utilize the self-learning ability of the first artificial intelligence submodel to establish the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution.

[0168] In one embodiment of the present invention, the discrete multi-wavelet operator module includes:

[0169] A data decomposition unit, used for decomposing the electrocardiogram data by using orthogonal multi-wavelet to obtain decomposed electrocardiogram data;

[0170] An information simplification unit, configured to determine simplified ECG information corresponding to the decomposed ECG data by downsampling and zero-filling operations;

[0171] The feature capture unit is used to determine the potential features of the simplified ECG information in the time domain and the frequency domain by using discrete multi-wavelet operation.

[0172] In one embodiment of the present invention, the geometric constraint embedding submodule includes:

[0173] The first sub-model building unit is used to establish the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution, and the formula for setting geometric constraints for the first artificial intelligence sub-model is as follows:

[0174]

[0175] u k =P l (y k -γf(y k )) (2)

[0176] y k+1 =u k +τ k+1 (u k -u k-1 ) (3)

[0177]

[0178] y k+1 =u k +τ k (u k -u k-1 ) (5)

[0179] Where, T k+1 is the iteration step size, which is used to control the amplitude of variable update during the iteration process; u k is the intermediate variable in the iteration process; y kis the target variable in the iterative process; γ is the regularization parameter, which is used to control the size of the gradient descent step; is the transpose of the adaptive gradient operator during iteration, used to adjust u k Update direction of Hy k Initialize items H to y for the application k The result of ; where in step k, H is used as the initialization term; P represents the proximal operator supported by the network embedded with constraints; G is the adaptive gradient operator during iteration; τ k > 0 can be learned through training defined in equation (1); ρ is the shrinkage operator ρ(u k )=(|u k |-ρ)+sgn(u k ), ()+ is |u k |-ρ is the maximum value between 0 and 0, sgn represents the sign function, limited to ±1.

[0180] In one embodiment of the present invention, the corresponding relationship establishing module 310 further includes:

[0181] A data acquisition submodule is used to acquire a medical image data set; wherein the medical image data set includes a normal myocardial image data set, a myocardial infarction area image data set, and a pathological change image data set related to myocardial infarction;

[0182] The second sub-model establishment sub-module is used to train the second artificial intelligence sub-model based on the medical image data set to determine the second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area.

[0183] In one embodiment of the present invention, the second artificial intelligence sub-model includes a physical sub-model and a neural network, and the second sub-model establishes a sub-module, including:

[0184] The potential expression unit is used to approximate the potential expression based on the neural network; the formula is as follows:

[0185] u(x,y,z,t)≈N(x,y,z,t;θ);

[0186] In the formula, θ is the parameter of the neural network; x, y, z are spatial coordinates; t represents time;

[0187] A training unit is used to substitute the expression of the electric potential into the physical sub-model for describing the cardiac electrophysiological process, train the second artificial intelligence sub-model, and make the neural network satisfy the laws of physics based on the loss function; the formula is as follows:

[0188]

[0189] Where α is the weight parameter that balances the two loss terms; p i is at position (x i ,y i ,t i ) Observed probability of myocardial infarction; N data is the total number of observed data points; is the true potential value at position j; is the potential value predicted by the neural network;

[0190] The update unit is used to update the weight W and bias b using the back propagation algorithm; the formula is as follows:

[0191]

[0192] The update rules are as follows:

[0193]

[0194]

[0195] Where L is the loss function used to evaluate the difference between the model's prediction and the actual observed value; y is the output of the neural network, representing the predicted potential value; η is the learning rate;

[0196] The loop unit is used to return to the neural network to continue the loop training after each training if the difference between the predicted value and the true value is still greater than the preset minimum tolerance difference, until the difference between the predicted value and the true value is no greater than the preset minimum tolerance difference.

[0197] In one embodiment of the present invention, the corresponding relationship establishing module 310 further includes:

[0198] The third sub-model establishment sub-module is used to input the node features in the myocardial infarction key area into a preset cross entropy loss function, and determine the output parameter of the cross entropy loss function as the corresponding myocardial infarction key node; wherein the cross entropy loss function is as follows:

[0199]

[0200] Where N is the number of nodes; y i is the true label of node i; is the model-predicted probability of infarction.

[0201] Reference Figure 4 , showing a computer device of a method for locating myocardial infarction by electrocardiogram based on a geometric iteration network of the present invention, which may specifically include the following:

[0202] The computer device 12 is in the form of a general-purpose computing device, and the components of the computer device 12 may include but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 connecting different system components (including the system memory 28 and the processing unit 16).

[0203] The bus 18 represents one or more of several types of bus 18 structures, including a memory bus 18 or memory controller, a peripheral bus 18, an accelerated graphics port, a processor or a local bus 18 using any of a variety of bus 18 architectures. These architectures include, by way of example, but are not limited to, an Industry Standard Architecture (ISA) bus 18, a Micro Channel Architecture (MAC) bus 18, an Enhanced ISA bus 18, an Audio Video Electronics Standards Association (VESA) local bus 18, and a Peripheral Component Interconnect (PCI) bus 18.

[0204] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0205] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write to non-removable, non-volatile magnetic media (commonly referred to as a "hard drive"). Although Figure 4 Not shown, a disk drive for reading and writing to a removable non-volatile disk (such as a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (such as a CD-ROM, DVD-ROM or other optical medium) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory may include at least one program product having a set (e.g., at least one) of program modules 42, which are configured to perform the functions of various embodiments of the present invention.

[0206] A program / utility 40 having a set (at least one) of program modules 42 may be stored in, for example, a memory, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules 42, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0207] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, cameras, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the computer device 12 may also communicate with one or more networks (e.g., local area networks (LANs)), wide area networks (WANs), and / or public networks (e.g., the Internet) via a network adapter 20. Figure 4 As shown, the network adapter 20 communicates with other modules of the computer device 12 via the bus 18. It should be understood that although Figure 4 Not shown, other hardware and / or software modules may be used in conjunction with the computer device 12, including but not limited to: microcode, device drivers, redundant processing units 16, external disk drive arrays, RAID systems, tape drives, and data backup storage systems 34, etc.

[0208] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing a method for locating myocardial infarction using an electrocardiogram based on a geometric iterative network provided in an embodiment of the present invention.

[0209] That is, when the processing unit 16 executes the program, it realizes: establishing a correspondence between the electrocardiogram data and the key nodes of myocardial infarction through the artificial intelligence model; wherein the correspondence includes a first corresponding sub-relationship, a second corresponding sub-relationship and a third corresponding sub-relationship; specifically, establishing a first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution through the first artificial intelligence sub-model; establishing a second corresponding sub-relationship between the ventricular surface potential distribution and the key area of ​​myocardial infarction through the second artificial intelligence sub-model; establishing a third corresponding sub-relationship between the key area of ​​myocardial infarction and the key nodes of myocardial infarction through the third artificial intelligence sub-model; acquiring the target electrocardiogram data of the target patient, and determining the target myocardial infarction key node corresponding to the target electrocardiogram data through the corresponding relationship.

[0210] In an embodiment of the present invention, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for locating myocardial infarction by electrocardiogram based on a geometric iterative network as provided in all embodiments of the present application:

[0211] That is, when the program is executed by the processor, it is achieved that: a correspondence between electrocardiogram data and key nodes of myocardial infarction is established through an artificial intelligence model; wherein the correspondence includes a first corresponding sub-relationship, a second corresponding sub-relationship and a third corresponding sub-relationship; specifically, a first corresponding sub-relationship between electrocardiogram data and ventricular surface potential distribution is established through the first artificial intelligence sub-model; a second corresponding sub-relationship between ventricular surface potential distribution and key areas of myocardial infarction is established through the second artificial intelligence sub-model; a third corresponding sub-relationship between key areas of myocardial infarction and key nodes of myocardial infarction is established through the third artificial intelligence sub-model; the target electrocardiogram data of the target patient is acquired, and the target myocardial infarction key node corresponding to the target electrocardiogram data is determined through the corresponding relationship.

[0212] Any combination of one or more computer-readable media may be used. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A 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 thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: 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 or 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 thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0213] Computer-readable signal media may include a data signal propagated in 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 a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0214] The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet). The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other.

[0215] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0216] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0217] The above is a detailed introduction to the method and device for locating myocardial infarction by electrocardiogram based on geometric iterative network provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for locating myocardial infarction using electrocardiogram based on geometric iterative network, characterized in that: The method relates to an artificial intelligence model, the artificial intelligence model includes a first artificial intelligence sub-model, a second artificial intelligence sub-model and a third artificial intelligence sub-model; the method includes: A correspondence relationship between electrocardiogram data and a key node of myocardial infarction is established through an artificial intelligence model; wherein the correspondence relationship includes a first correspondence sub-relationship, a second correspondence sub-relationship and a third correspondence sub-relationship; specifically, a first correspondence sub-relationship between electrocardiogram data and ventricular surface potential distribution is established through the first artificial intelligence sub-model; a second correspondence sub-relationship between ventricular surface potential distribution and a key area of ​​myocardial infarction is established through the second artificial intelligence sub-model; a third correspondence sub-relationship between a key area of ​​myocardial infarction and a key node of myocardial infarction is established through the third artificial intelligence sub-model; Target electrocardiogram data of a target patient is acquired, and a target myocardial infarction key node corresponding to the target electrocardiogram data is determined through the corresponding relationship.

2. The method according to claim 1, characterized in that: The step of establishing the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution by using the first artificial intelligence sub-model comprises: Determining the potential features of the electrocardiogram data in the time domain and the frequency domain using discrete multiwavelet operations; Performing feature compensation on the potential features based on a preset sandwich structure operator to determine graphic information with a data dimension consistent with the ventricular surface potential distribution; Geometric constraints are embedded based on the graphic information of nodes and edges, and the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution is established by utilizing the self-learning ability of the first artificial intelligence sub-model.

3. The method according to claim 2, characterized in that The step of using discrete multiwavelet operation to determine the potential features of the electrocardiogram data in the time domain and the frequency domain comprises: Decomposing the electrocardiogram data by using orthogonal multi-wavelet to obtain decomposed electrocardiogram data; Determine simplified electrocardiogram information corresponding to the decomposed electrocardiogram data by downsampling and zero-filling operations; Discrete multi-wavelet operation is used to determine the potential features of the simplified electrocardiogram information in the time domain and the frequency domain.

4. The method according to claim 2, characterized in that: The step of embedding the graphic information based on nodes and edges with geometric constraints and establishing the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution by using the self-learning ability of the first artificial intelligence sub-model comprises: The formula for establishing the first corresponding sub-relationship between the electrocardiogram data and the ventricular surface potential distribution and setting geometric constraints on the first artificial intelligence sub-model is as follows: or k =P l (and k -γf(and k )) (2) y k+1 =u k +τ k+1 (u k -u k-1 ) (3) y k+1 =u k +τ k (u k -u k-1 ) (5) Where, T k+1 is the iteration step size, which is used to control the amplitude of variable update during the iteration process; u k is the intermediate variable in the iteration process; y k is the target variable in the iterative process; γ is the regularization parameter used to control the size of the gradient descent step; is the transpose of the adaptive gradient operator during iteration, used to adjust u k Update direction of Hy k Initialize items H to y for the application k The result of ; where in step k, H is used as the initialization term; P represents the proximal operator supported by the network embedded with constraints; G is the adaptive gradient operator during iteration; τ k > 0 can be learned through training defined in equation (1); ρ is the shrinkage operator ρ(u k )=(|u k |-ρ)+sgn(u k ), ()+ is |u k |-ρ is the maximum value between 0 and 0, sgn represents the sign function, limited to ±1.

5. The method according to claim 1, characterized in that The step of establishing the second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area through the second artificial intelligence sub-model comprises: Acquire a medical image dataset; wherein the medical image dataset includes a normal myocardial image dataset, a myocardial infarction region image dataset, and a pathological change image dataset related to myocardial infarction; The second artificial intelligence sub-model is trained based on the medical image data set to determine the second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area.

6. The method according to claim 5, characterized in that The second artificial intelligence sub-model includes a physical sub-model and a neural network, and the step of training the second artificial intelligence sub-model according to the medical image data set to determine the second corresponding sub-relationship between the ventricular surface potential distribution and the myocardial infarction key area includes: The expression of the electric potential is approximated based on the neural network; the formula is as follows: u(x,y,z,t)≈N(x,y,z,t;θ); In the formula, θ is the parameter of the neural network; x, y, z are spatial coordinates; t represents time; Substitute the expression of the electric potential into the physical sub-model for describing the cardiac electrophysiological process, train the second artificial intelligence sub-model, and make the neural network satisfy the laws of physics based on the loss function; the formula is as follows: Where α is the weight parameter that balances the two loss terms; p i is at position (x i ,y i ,t i ) Observed probability of myocardial infarction; N data is the total number of observed data points; is the true potential value at position j; is the potential value predicted by the neural network; The back propagation algorithm is used to update the weight W and bias b; the formula is as follows: The update rules are as follows: Where L is the loss function used to evaluate the difference between the model's prediction and the actual observed value; y is the output of the neural network, representing the predicted potential value; η is the learning rate; After each training, if the difference between the predicted value and the true value is still greater than the preset minimum tolerance difference, the neural network returns to continue the cyclic training until the difference between the predicted value and the true value is no greater than the preset minimum tolerance difference.

7. The method according to claim 1, characterized in that The step of establishing the third corresponding sub-relationship between the myocardial infarction key area and the myocardial infarction key node through the third artificial intelligence sub-model comprises: The node features in the myocardial infarction key area are input into a preset cross entropy loss function, and the output parameter of the cross entropy loss function is determined to be the corresponding myocardial infarction key node; wherein the cross entropy loss function is as follows: Where N is the number of nodes; y i is the true label of node i; is the model-predicted probability of infarction.

8. An electrocardiogram locating myocardial infarction device based on geometric iteration network, characterized in that: The device relates to an artificial intelligence model, which includes a first artificial intelligence sub-model, a second artificial intelligence sub-model and a third artificial intelligence sub-model; including: A correspondence establishing module, used to establish a correspondence between electrocardiogram data and a key node of myocardial infarction through an artificial intelligence model; wherein the correspondence includes a first corresponding sub-relationship, a second corresponding sub-relationship and a third corresponding sub-relationship; specifically, a first corresponding sub-relationship between electrocardiogram data and ventricular surface potential distribution is established through the first artificial intelligence sub-model; a second corresponding sub-relationship between ventricular surface potential distribution and a key area of ​​myocardial infarction is established through the second artificial intelligence sub-model; a third corresponding sub-relationship between a key area of ​​myocardial infarction and a key node of myocardial infarction is established through the third artificial intelligence sub-model; The data processing module is used to obtain the target electrocardiogram data of the target patient, and determine the target myocardial infarction key node corresponding to the target electrocardiogram data through the corresponding relationship pre-established by the artificial intelligence model.

9. A device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program implements the method according to any one of claims 1 to 7 when executed by the processor.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images