Early detection model and method for coronary circulatory disorders based on VCG and CDG

Abstract

This invention provides an early detection model and method for coronary artery circulatory disorders based on VCG and CDG. A model is constructed that uses the STT segment of VCG and CDG to jointly characterize and predict coronary artery circulatory disorders. VCG temporal heterogeneity, VCG spatial heterogeneity, VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index are integrated as the VCG feature set. Similarly, CDG temporal heterogeneity, CDG spatial heterogeneity, CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index are integrated as the CDG feature set. The VCG and CDG feature sets are input into the trained early detection model for coronary artery circulatory disorders to obtain output values. Based on these output values, the existence of coronary artery circulatory disorders in the subject is predicted, enabling accurate and objective early detection of coronary artery circulatory disorders.

Description

Technical Field

[0001] This application relates to the field of electrocardiogram signal processing, and in particular to an early detection model and method for coronary circulatory disorders based on VCG and CDG. Background Technology

[0002] The coronary arteries mainly consist of two parts: the subepicardial vessels and the coronary microcirculation system, which account for 10% and 90% of myocardial blood volume, respectively. Narrowing of the epicardial vessels or obstruction of the coronary microcirculation system leads to myocardial ischemia. Therefore, coronary artery disease (CAD, defined as epicardial vessel stenosis greater than or equal to 50% as shown on coronary angiography) and coronary microcirculation disorder (CMD) are two major causes of myocardial ischemia. Clinical methods for detecting myocardial ischemia mainly include cardiac magnetic resonance imaging (MRI), PET, and the coronary microcirculation resistance index. However, most of these methods are expensive, involve radiation, or are invasive, making them unsuitable for early detection.

[0003] Myocardial ischemia leads to ischemic changes on electrocardiogram (ECG). In the early stages of myocardial ischemia, ECG signals are primarily affected, along with ST segment or T wave changes. Since ECG is a routine method for detecting heart disease, offering advantages such as cost-effectiveness, non-invasiveness, and convenience, its use in detecting myocardial ischemia is a growing trend. Existing ECG-based methods for detecting myocardial ischemia mainly include visual detection methods and AI-based methods based on ECG features. Compared to ECG, using a vector electrocardiogram (VCG) provides better detection sensitivity; therefore, indicators such as sample entropy, temporal heterogeneity, and spatial heterogeneity extracted from VCG are often used for myocardial ischemia detection.

[0004] However, existing electrocardiogram vector diagrams, such as sample entropy, temporal heterogeneity, and spatial heterogeneity, are mainly used for detecting CAD-induced myocardial ischemia. Similarly, myocardial ischemia detection algorithms based on coronary cardiomyography (CDG) are mainly aimed at myocardial ischemia induced by epicardial vascular stenosis, i.e., coronary heart disease (CMD). There are no related studies or reports on coronary artery circulatory disorders (CMD). Therefore, how to detect coronary artery circulatory disorders (CMD) is of great significance for the detection and prevention of myocardial ischemia. Summary of the Invention

[0005] This application provides an early detection model and method for coronary artery circulatory disorders based on VCG and CDG. The early detection model for coronary artery circulatory disorders is constructed based on VCG and CDG features, achieving the effect of non-invasive early detection of coronary artery circulatory disorders.

[0006] In a first aspect, embodiments of this application provide a method for early detection of coronary circulatory disorders based on VCG and CDG, including:

[0007] Collect 12-lead ECG signals from the subject;

[0008] The 12-lead ECG signal was converted into a 3-lead VCG signal, and the ST-T segment of each lead of the VCG signal was extracted.

[0009] The ST-T segments of the three leads of VCG are combined to form a three-dimensional ST-T loop. VCG temporal heterogeneity and VCG spatial heterogeneity are calculated based on the three-dimensional ST-T loop. The ST-T segments of each lead of the VGG signal are stitched together frame by frame to form the corresponding time series. Each time series is standardized to obtain the corresponding lead time series. The VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index of each lead time series are calculated. The VCG temporal heterogeneity, VCG spatial heterogeneity, VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index are integrated as the VCG feature set.

[0010] A deterministic learning algorithm is used to process the ST-T segments of three leads to obtain a three-dimensional CDG loop. Based on the three-dimensional CDG loop, CDG temporal heterogeneity and CDG spatial heterogeneity are calculated. After standardizing the signal of each channel of the three-dimensional CDG loop, the channel time series is obtained. The CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index of the channel time series are calculated. The CDG temporal heterogeneity, CDG spatial heterogeneity, CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index are integrated as a CDG feature set.

[0011] The effective feature subsets corresponding to the VCG feature set and the CDG feature set are input into the trained early detection model of coronary artery circulation disorder to obtain the output value, and the presence of coronary artery circulation disorder in the subject is predicted based on the output value.

[0012] Secondly, embodiments of this application provide a method for constructing an early detection model of coronary artery circulatory disorders based on VCG and CDG, comprising the following steps:

[0013] Establish training set: Select VCG feature set and CDG feature set of CMD patients and healthy individuals as training set;

[0014] Training the model: Input the VCG feature set and CDG feature set into the multilayer perceptron model for training to obtain an early detection model of coronary artery circulation disorders.

[0015] Thirdly, embodiments of this application provide an early detection model for coronary artery circulation disorders based on VCG and CDG, which is constructed according to the construction method of an early detection model for coronary artery circulation disorders based on VCG and CDG.

[0016] Fourthly, embodiments of this application provide an application method for an early detection method of coronary artery circulatory disorders based on VCG and CDG, comprising: a 12-lead electrocardiogram device, an electronic data processing device, and a display component; the 12-lead electrocardiogram device is used to collect 12-lead ECG signals from the subject; the electronic data processing device executes the early detection method of coronary artery circulatory disorders; the display component is configured to display a myocardial ischemia alarm or warning when it is determined that coronary artery circulatory disorders exist.

[0017] The main contributions and innovations of this invention are as follows:

[0018] This protocol integrates VCG and CDG features to achieve early detection of coronary artery circulatory disorders, thereby enabling early detection of myocardial ischemia caused by these disorders. This protocol not only validates the effectiveness of VCG and CDG features in detecting coronary artery circulatory disorders (CMD), but also establishes a non-invasive early detection method for CMD based on VCG and CDG. This protocol can be integrated into electrocardiographs for clinical applications or embedded in wearable devices for daily monitoring in nursing homes, healthcare facilities, and homes to detect early myocardial ischemia caused by coronary artery circulatory disorders, exhibiting high sensitivity and accuracy. The early detection model for coronary artery circulatory disorders and the corresponding myocardial ischemia prediction device developed in this protocol can be used for early detection of myocardial ischemia caused by coronary artery circulatory disorders, greatly facilitating medical diagnosis.

[0019] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0021] Figure 1 This is a schematic diagram illustrating the construction logic of an early detection method for coronary artery circulatory disorders based on VCG and CDG.

[0022] Figure 2 This is a flowchart illustrating an early detection method for coronary artery circulatory disorders based on VCG and CDG.

[0023] Figure 3 This is a diagram showing the VCG and CDG of CMD patients and healthy individuals;

[0024] Figure 4 This is a schematic diagram comparing the VCG feature sets of CMD patients and healthy individuals;

[0025] Figure 5 This is a schematic diagram comparing the CDG feature sets of CMD patients and healthy individuals;

[0026] Figure 6 This is a performance comparison diagram of the three models provided in this solution;

[0027] Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0028] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0029] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0030] Example 1

[0031] This application provides a method for early detection of coronary artery circulatory disorders. Specifically, refer to... Figure 2 The methods include:

[0032] Collect 12-lead ECG signals from the subject;

[0033] The 12-lead ECG signal was converted into a 3-lead VCG signal, and the ST-T segment of each lead of the VCG signal was extracted.

[0034] The ST-T segments of the three leads of VCG are combined to form a three-dimensional ST-T loop. VCG temporal heterogeneity and VCG spatial heterogeneity are calculated based on the three-dimensional ST-T loop. The ST-T segments of each lead of the VGG signal are stitched together frame by frame to form the corresponding time series. Each time series is standardized to obtain the corresponding lead time series. The VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index of each lead time series are calculated. The VCG temporal heterogeneity, VCG spatial heterogeneity, VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index are integrated as the VCG feature set.

[0035] A deterministic learning algorithm is used to process the ST-T segments of three leads to obtain a three-dimensional CDG loop. Based on the three-dimensional CDG loop, CDG temporal heterogeneity and CDG spatial heterogeneity are calculated. After standardizing the signal of each channel of the three-dimensional CDG loop, the channel time series is obtained. The CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index of the channel time series are calculated. The CDG temporal heterogeneity, CDG spatial heterogeneity, CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index are integrated as a CDG feature set.

[0036] The effective feature subsets corresponding to the VCG feature set and the CDG feature set are input into the trained early detection model of coronary artery circulation disorder to obtain the output value, and the presence of coronary artery circulation disorder in the subject is predicted based on the output value.

[0037] This protocol integrates VCG and CDG features to achieve early detection of coronary circulatory disorders, thereby enabling early detection of myocardial ischemia caused by coronary circulatory disorders. This protocol not only validates the effectiveness of VCG and CDG features in detecting coronary circulatory disorders (CMD), but also establishes an early non-invasive detection method for CMD based on these features. This protocol can be integrated into electrocardiographs for clinical applications, or embedded in wearable devices for daily monitoring in nursing homes, healthcare facilities, and homes to detect early myocardial ischemia caused by coronary circulatory disorders, exhibiting high sensitivity and accuracy. In this protocol, VCG refers to the vector electrocardiogram (VCG) and CDG refers to the cardiac dynamics graph (CDG).

[0038] It's worth noting that using CDG features alone to characterize coronary artery disease (CMD) results in very low classification accuracy. This proposed method, however, combines CDG and VCG features to form a larger feature set, leading to more accurate detection of CMD. Furthermore, this method not only uses temporal and spatial heterogeneity as evaluation metrics but also incorporates sample entropy, approximate entropy, and heterogeneity index, further enhancing its accuracy in detecting CMD.

[0039] In this scheme, conventional ECG acquisition equipment can be used to acquire 12-lead ECG signals from the subject. To facilitate subsequent ECG signal processing, in some embodiments, this scheme requires first performing noise reduction filtering on all ECG signals to remove power line interference, electromyographic interference, and baseline drift. Specifically, this scheme uses median filtering to remove baseline drift and wavelet filtering to remove electromyographic interference. In one specific embodiment, Coif4 from the Coifiet wavelet system is used as the wavelet basis in the wavelet filtering, and the ECG signal is decomposed into four levels of Coif4. An adaptive threshold obtained using Stein's unbiased likelihood estimation principle is then applied to the wavelet coefficients, followed by soft threshold filtering. Finally, the ECG signal is reconstructed using inverse wavelet transform, thus filtering out power line interference and electromyographic interference. All ECG signals are standardized according to the 25mm / s and 10mm / mV standards.

[0040] In the step of “converting a 12-lead ECG signal to a 3-lead VCG signal”, the following formula (1) is used to convert the 12-lead ECG signal to a 3-lead VCG signal:

[0041]

[0042] Where I, II, V1, V2, V3, V4, V5, and V6 represent the leads of the ECG signal, respectively; and Vx, Vy, and Vz represent the leads of the VCG signal, respectively.

[0043] In the step of “extracting the ST-T segment of each lead of the VCG signal”, the ST-T segment of each lead of each VCG signal is extracted based on the wavelet ECG baseline detection method.

[0044] In the step of “calculating VCG temporal heterogeneity and VCG spatial heterogeneity based on three-dimensional ST-T ring”, the Lyapunov exponent of the three-dimensional ST-T ring is calculated as the VCG spatial heterogeneity. The specific calculation formula is as follows (2):

[0045]

[0046] Where SHI is the spatial heterogeneity of VCG, d n1 d represents the distance from the nth point in the three-dimensional ST-T ring to the nearest point. n2 Represents relative to d n1 The distance from the nth point after ten steps to the nearest point, where L is the number of all points on the three-dimensional ST-T ring.

[0047] The formula for calculating the VCG time heterogeneity of the three-dimensional ST-T ring is as follows (3):

[0048]

[0049] Where THI is the VCG temporal heterogeneity, F represents the Fourier transform of the time series of the three-dimensional ST-T ring, and λ i It represents 1: length(F), i = Vx, Vy, Vz.

[0050] In some embodiments, the VCG temporal heterogeneity is named VCGTHI, and the VCG spatial heterogeneity is named VCGSHI.

[0051] In the step of “sponging the ST-T segments of each lead of the VGG signal into a corresponding time series and standardizing each time series to obtain the corresponding lead time series”, the standard deviation of the time series composed of the ST-T segments of each lead is adjusted to 1 to obtain the lead time series. Specifically, the standard deviation of the time series is adjusted by subtracting the mean from each point and then dividing by the standard deviation.

[0052] In this embodiment of the scheme, the specific steps for calculating the VCG sample entropy of each lead time series are as follows:

[0053] Take the average similarity probability of the m subvectors with embedding dimension m in the time series of each lead, and the average similarity probability of the m+1 subvectors with embedding dimension m+1. Calculate the negative logarithm of the quotient of the average similarity probability of the m subvectors and the average similarity probability of the m+1 subvectors as the VGG sample entropy of the time series of that lead.

[0054] The specific formula for calculating sample entropy is shown in equation (4) below:

[0055]

[0056] Where SampEn is the sample entropy, ψ m (ε) and ψ m+1 (ε) is the average similarity probability of subvectors with embedding dimensions m and m+1, where ε is the error tolerance range of the similar region, and m represents the embedding dimension.

[0057] In this embodiment of the scheme, the specific steps for calculating the approximate VCG entropy of each lead time series are as follows:

[0058] Take the average probability of the m-th sub-vector with embedding dimension m in the time series of each lead, and the average probability of the m+1 sub-vector with embedding dimension m+1. Take the difference between the average probability of the m-th sub-vector and the average probability of the m+1 sub-vector as the VCG approximate entropy of the time series of that lead.

[0059] The specific formula for calculating the approximate entropy is shown in equation (5) below:

[0060] ApEn(ε,m)=θ m (ε)-θm+1 (ε) (5)

[0061] Where ApEn is the approximate entropy, θ m (ε) and θ m+1 (ε) represents the average probability of m sub-vectors with embedding dimension m, ε is the error tolerance range of similar regions, and m represents the embedding dimension.

[0062] In an embodiment of this scheme, the specific steps for calculating the VCG heterogeneity index of each lead time series are as follows: calculate the multiscale entropy of each lead time series, and calculate the area of ​​the multiscale entropy on the time scale as the VCG heterogeneity index.

[0063] After coarsening the time series of each lead, a coarse-grained time series is obtained. The sample entropy is calculated using the coarse-grained time series according to formula (4) to obtain the multi-scale entropy. The area of ​​the multi-scale entropy on the time scale is taken as the VCG heterogeneity index.

[0064] The multi-scale entropy calculation algorithm integrates time series coarsening and sample entropy calculation algorithms. At each scale τ, the coarse-grained time series is represented as: in

[0065] M represents the maximum value of the time scale, t(l) represents a point in the time series, τ represents the time scale, and L represents the sequence length of the coarse-grained time series.

[0066] The formula for calculating multiscale entropy (MSE) is shown in equation (6):

[0067] MSE(τ, m, ε) = SampEn(y (τ) ,m,ε),1≤τ≤M (6)

[0068] The formula for calculating the heterogeneity index CI is shown in equation (7) below:

[0069]

[0070] In this scheme, the corresponding VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index are calculated for the lead time series of the ST-T segment of the VCG in three leads, resulting in VxSampEn, VySampEn, VzSampEn, VxApen, VyApen, VzApen, VxCI, VyCI, and VzCI. Here, Vx, Vy, and Vz correspond to the ST-T segment of the VCG in the three leads, SampEn corresponds to the sample entropy, Apen corresponds to the approximate entropy, and CI corresponds to the heterogeneity index.

[0071] This scheme employs a deterministic learning algorithm to extract information related to myocardial ischemia from the ST-T segments of the VCG in three leads. Specifically, the ST-T segments of the VCG in the three leads are used as input information for the deterministic learning algorithm, which then extracts the three-dimensional CDG loop.

[0072] Specifically, the formula for calculating CDG is shown in equation (8) below:

[0073]

[0074] Where V stt =[v xstt v ystt v zstt ] represents the ST-T segment of the three leads of VCG, F(V stt η) is a nonlinear function vector representing the electrodynamic characteristics of the ST-T segment, where η is a constant vector of system parameters, F(V stt ;η) represents the three-dimensional CDG ring.

[0075] Specifically, in determining the learning algorithm, three locally dynamic RBF neural networks are used to approximate the electrocardiographic features F(V). stt ;η).

[0076] The calculation formula for the RBF neural network is shown in equation (9) below:

[0077]

[0078] in As the state vector of the dynamic RBF neural network, A is a constant matrix in which each element is greater than 0. S represents an approximation of the optimal weights. i (V stt )=[s i1 (V stt -ζ1), ..., s iN (V stt -ζ N )] is a radial basis function, where s ij (·) is a Gaussian function, ζ i It is a point in the state space.

[0079] Update according to the following formula (10):

[0080]

[0081] in ω * δ represents the optimal weight, and δ is a very small constant.

[0082] By using three RBF neural networks with Equation (11) and setting the initial weights to zero, The output of the RBF neural network is F(V) stt ;η),

[0083]

[0084] in represent The arithmetic mean of the terms, where err represents the approximation error.

[0085] In other words, the ST-T segments of the three VCG leads are input into the corresponding RBF neural networks for learning to obtain the electrocardiographic features of the leads. The electrocardiographic features of the three leads form a three-dimensional CDG loop.

[0086] After obtaining the three-dimensional CDG ring, the temporal heterogeneity and spatial heterogeneity of the CDG are calculated based on the three-dimensional CDG ring.

[0087] Specifically, in the step of “calculating CDG temporal heterogeneity and CDG spatial heterogeneity based on the three-dimensional CDG ring”, the Lyapunov exponent of the three-dimensional CDG ring is calculated as the CDG spatial heterogeneity. The specific calculation formula is as follows (2):

[0088]

[0089] Where SHI is the spatial heterogeneity of CDG, d n1 d represents the distance from the nth point in the 3D CDG ring to the nearest point. n2 Represents relative to d n1 The distance from the nth point after ten steps to the nearest point, where L is the number of all points on the 3D CDG ring.

[0090] The formula for calculating the CDG temporal heterogeneity of the three-dimensional CDG loop is as follows (3):

[0091]

[0092] Where THI is the CDG temporal heterogeneity, F represents the Fourier transform of the time series of the three-dimensional CDG ring, and λ i It represents 1: length(F), i = W1S1, W2S2, W3S3.

[0093] In some embodiments, the CDG temporal heterogeneity is named CDGTHI, and the CDG spatial heterogeneity is named CDGSHI.

[0094] The channel time series is obtained by standardizing the signal of each channel of the 3D CDG loop. Similarly, this scheme standardizes the signal of each channel of the 3D CDG loop so that the standard deviation of the channel time series is 1.

[0095] In this embodiment of the scheme, the specific steps for calculating the CDG sample entropy of each channel time series are as follows:

[0096] Take the average similarity probability of the m sub-vectors with embedding dimension m in each channel time series, and the average similarity probability of the m+1 sub-vectors with embedding dimension m+1. Calculate the negative logarithm of the quotient of the average similarity probability of the m sub-vectors and the average similarity probability of the m+1 sub-vectors as the CDG sample entropy of that channel time series.

[0097] The specific formula for calculating sample entropy is shown in equation (4) below:

[0098]

[0099] Where SampEn is the sample entropy, ψ m (ε) and ψ m+1 (ε) is the average similarity probability of subvectors with embedding dimensions m and m+1, where ε is the error tolerance range of the similar region, and m represents the embedding dimension.

[0100] In this embodiment of the scheme, the specific steps for calculating the CDG approximate entropy of each channel time series are as follows:

[0101] Take the average probability of the m-th sub-vector with embedding dimension m in each channel time series, and the average probability of the m+1 sub-vector with embedding dimension m+1. Take the difference between the average probability of the m-th sub-vector and the average probability of the m+1 sub-vector as the CDG approximate entropy of the channel time series.

[0102] The specific formula for calculating the approximate entropy is shown in equation (5) below:

[0103] ApEn(ε,m)=θ m (ε)-θ m+1 (ε) (5)

[0104] Where ApEn is the approximate entropy, θ m (ε) and θ m+1 (ε) represents the average probability of m sub-vectors with embedding dimension m, ε is the error tolerance range of similar regions, and m represents the embedding dimension.

[0105] In an embodiment of this scheme, the specific steps for calculating the CDG heterogeneity index of each channel time series are as follows: calculate the multi-scale entropy of each channel time series, and calculate the area of ​​the multi-scale entropy on the time scale as the CDG heterogeneity index.

[0106] After coarsening the time series of each channel, a coarse-grained time series is obtained. The sample entropy is calculated using the coarse-grained time series according to formula (4) to obtain the multi-scale entropy. The area of ​​the multi-scale entropy on the time scale is taken as the CDG heterogeneity index.

[0107] The multi-scale entropy calculation algorithm integrates time series coarsening and sample entropy calculation algorithms. At each scale τ, the coarse-grained time series is represented as: in

[0108] M represents the maximum value of the time scale, t(l) represents a point in the time series, τ represents the time scale, and L represents the sequence length of the coarse-grained time series.

[0109] The formula for calculating the heterogeneity index CI is shown in equation (7) below:

[0110]

[0111] In this scheme, the corresponding CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index are calculated for each of the three channel time series, resulting in W1S1SampEn, W2S2SampEn, W3S3SampEn, W1S1Apen, W2S2Apen, W3S3Apen, W1S1CI, W2S2CI, and W3S3CI. Here, W1S1 represents the first channel time series, W2S2 represents the second channel time series, W3S3 represents the third channel time series, SampEn corresponds to the sample entropy, Apen corresponds to the approximate entropy, and CI corresponds to the heterogeneity index.

[0112] like Figure 1As shown, it is worth mentioning that this scheme obtains an effective subset of the VCG feature set, CDG feature set, and combined feature set through the SBS (Sequential Backward Selection) algorithm, and then inputs it into the trained early detection model of coronary artery circulation disorder to obtain the output value. The output values ​​of all models are then evaluated. The effective feature subsets obtained by applying the SBS (Sequential Backward Selection) algorithm to the VCG feature set are VxApEn, VxCI, and VyApen. The effective feature subsets obtained by applying the SBS (Sequential Backward Selection) algorithm to the CDG feature set are CDGSHI, W1S1ApEn, W1S1CI, W2S2ApEn, W2S2CI, and W3S3CI. The effective feature subsets obtained by applying the SBS (Sequential Backward Selection) algorithm to the combined feature set are VCGSHI, VGGTHI, VxApEn, VxCI, VyApen, VzSampEn, VzApen, VzCI, CDGSHI, CDGTHI, W1S1ApEn, W2S2SampEn, W2S2Apen, W3S3Apen, and W3S3CI.

[0113] Specifically, the effective feature subsets in this scheme are obtained based on the SBS (Sequential Backward Selection) algorithm. These effective feature subsets are selected from the VCG feature set, CDG feature set, and combined feature set using the backward sequence selection algorithm. First, the VCG feature set, CDG feature set, and combined feature set are selected as candidate subsets. The iteration variable n is set to the initial number of features, where each subset includes n-1 features. Five-fold cross-validation is used to train and evaluate the classification performance of these subsets on the training set, and the subset with the highest accuracy is selected. In the next iteration, n-2 subsets are created from the selected n-1 subsets, and these subsets are evaluated and trained. The best subset is selected, and this process is repeated until n=1.

[0114] After obtaining the effective feature subsets, this scheme trains an early detection model for coronary artery circulation disorders using the effective feature subsets corresponding to each model. The early detection model for coronary artery circulation disorders in this scheme employs a multilayer perceptron (MLP) model. In this scheme, an STT-based MLP model is constructed with the effective feature subset of the VCG feature set as input, an MLP model with the effective feature subset of the CDG feature set as input is named the CDG-based MLP model, and an STT-CDG-based MLP model uses the effective feature subset of the combined feature set as common input.

[0115] The training methods for the three early detection models of coronary artery circulatory disorders are as follows:

[0116] Establish training set: Select VCG and CDG feature sets from healthy individuals and CMD patients as training sets;

[0117] The corresponding training set is input into the multilayer perceptron model for training to obtain the early detection model of coronary artery circulation disorders.

[0118] Furthermore, since this scheme selects a multilayer perceptron as the early detection model for coronary artery circulation disorders, and the size of the hidden layers is a key parameter in the multilayer perceptron model, this invention uses the Sparrow Swarm Algorithm to optimize it. However, the Sparrow Algorithm suffers from a decrease in population diversity with increasing iterations, which can easily lead to local optima and prevent the algorithm from obtaining the global optimum. Therefore, this scheme adds mixed variables and Gaussian heterogeneity to the basic Sparrow Algorithm in the multilayer perceptron.

[0119] Specifically, in the sparrow algorithm, the position of the i-th sparrow is set as follows:

[0120] P i =[p i1 ,L,p ij ,L,p iD ], i = 1, 2, K, N, where p ij Let N represent the position of the i-th sparrow in the j-th dimension, and let N and D represent the number of sparrows and the dimension with optimization parameters, respectively.

[0121] First, the mixed variables use a Bernoulli distribution to increase the diversity of the sparrows' initial positions, thereby improving the algorithm's global search capability. Compared with the random variables used in the basic algorithm, the Bernoulli distribution has ergodic performance, making the initial positions of the sparrows more uniform.

[0122] The Bernoulli distribution formula is shown in equation (12) below.

[0123]

[0124] Where λ = 0.4, b ij Let represent the value of the i-th hybrid variable, Sparrow, in the j-th dimension.

[0125] Next, the hybrid variables are transformed into the solution space using the following formula (13):

[0126] p ij =p min +(p max -p min )×b ij (13)

[0127] Where pmin and p max The minimum and maximum values ​​of the parameters to be optimized are respectively achieved.

[0128] Secondly, the fitting function value corresponding to the i-th sparrow is defined as shown in equation (14):

[0129] fun(P i ) = 1 - argmin(F1score)(14)

[0130] Here, F1 score represents the F1 score of five-fold cross-validation on the training set.

[0131] In each iteration, the position of the producer in the sparrow is updated as shown in equation (15):

[0132]

[0133] Where t represents the number of iterations, iter max The maximum number of iterations is represented by `rand(0,1)∈[0,1]`, where `rand(0,1)` is a random variable, `Q` represents a random number that follows a standard normal distribution, `R2∈[0,1]` represents the warning value, and `ST∈[0.5,1]` represents the safety threshold.

[0134] The participant's position is updated using the following formula (16):

[0135]

[0136] in, This represents the worst position at present. It represents the optimal position occupied by the participant in the iteration. A is a 1×D matrix whose elements are randomly divided into -1 and 1, where A + =A T (AA T ) -1 R is a unit row vector of D columns.

[0137] The position update of the monitor is shown in equation (17):

[0138]

[0139] in, This is the current optimal position. N(0,1) is a random number that follows a standard normal distribution. K∈[-1,1] is a random number. best and fun worst These represent the best and worst fitted function values, respectively, with γ being a very small value to avoid zero in the denominator.

[0140] After the first iteration, Gaussian alienation was introduced to update the sparrow positions to increase population diversity, as shown in equation (18):

[0141]

[0142] fun avg N represents the mean of the fitted function. T It represents p ij Quantity, p ijnew The sparrow's new position is constantly being updated.

[0143] When fun(P) i ) <fun avg Gaussian alienation is used to avoid population aggregation, and Gaussian mutation can prevent sparrows from clustering. Gaussian alienation has enhanced local search capabilities. When the problem to be optimized has many local optima, it can help the algorithm effectively search for the global optimum. When fun(P) i )>fun avg This invention uses mixed perturbation to avoid algorithm dispersion. Subsequently, the sparrow's position is updated using a position value with a lower fitted function value.

[0144] After constructing the CDG-STT-based model, the CDG-based model, and the STT-based model, this approach compared the performance metrics of the CDG-STT-based model and the CDG-based model, as well as the CDG-STT-based model and the STT-based model, on the validation set. The validation results are shown in Table 1. The CDG-STT-based model significantly outperforms the CDG-based and STT-based models in all performance metrics. Therefore, the CDG-STT-based model is determined to be the optimal early detection model for coronary artery circulation disorders.

[0145] The three models showed similar trends on the test dataset, such as Figure 6 As shown in Table 1, the CDG-STT-based model outperforms the other two models on the test set. All metrics for this model on the test set are greater than 0.8, with an accuracy greater than 0.9. Therefore, the CDG-STT-based model is the optimal model for early detection of coronary artery circulatory disorders.

[0146] Table 1

[0147]

[0148] In addition, this scheme selects ECG signals from healthy individuals and CMD patients, processes them through the above steps to obtain VCG and CDG feature sets for healthy individuals, and VCG and CDG feature sets for CMD patients. The results of comparing the features of healthy individuals and CMD patients are as follows: Figure 4 and Figure 5 As shown.

[0149] like Figure 4 As shown, Figure 4 This diagram illustrates the extraction of VCG features from healthy individuals and CMD patients. It is clearly visible that, compared to healthy individuals, CMD patients have significantly lower VCG THI, VzSampEn, VzApen, and VzCI, but significantly higher VCG THI, VxSampEn, and VxCI. This demonstrates a significant difference in this feature set between healthy individuals and CMD patients.

[0150] like Figure 5 As shown, Figure 5 This is a schematic diagram illustrating the extraction of CDG features from healthy individuals and CMD patients. It is very clear that, compared to healthy individuals, CMD patients have significantly higher levels of W1S1SampEn, W2S2SampEn, W1S1Apen, W1S1CI, W2S2CI, W3S3CI, and CDGTHI. This feature set is significantly different between healthy individuals and CMD patients.

[0151] Example 2

[0152] This solution provides a method for constructing an early detection model of coronary artery circulatory disorders based on VCG and CDG, including the following steps:

[0153] Establish training set: Select VCG feature set and CDG feature set of CMD patients and healthy individuals as training set;

[0154] Training the model: Input the effective feature subset into the multilayer perceptual layer model for training to obtain an early detection model for coronary artery circulation disorders.

[0155] In some embodiments, the effective feature subset includes VCGSHI, VGGTHI, VxApEn, VxCI, VyApen, VzSampEn, VzApen, VzCI, CDGSHI, CDGTHI, W1S1ApEn, W2S2SampEn, W2S2Apen, W3S3Apen, and W3S3CI.

[0156] In some embodiments, the multilayer perception layer model is optimized using the Sparrow Intelligence Group algorithm, and mixed variables and Gaussian alienation are added to the Sparrow algorithm for optimization training.

[0157] This solution provides an early detection model for coronary artery circulatory disorders trained using the above method.

[0158] Example 3

[0159] This solution provides an application of an early detection method for coronary artery circulatory disorders based on VCG and CDG. In some embodiments, this method can be applied to an early detection device for coronary artery circulatory disorders. In this case, the myocardial ischemia prediction device includes a 12-lead electrocardiogram (ECG) device, an electronic data processing device, and a display component. The electronic data processing device performs the following steps:

[0160] Collect 12-lead ECG signals from the subject;

[0161] The 12-lead ECG signal was converted into a 3-lead VCG signal, and the ST-T segment of each lead of the VCG signal was extracted.

[0162] The ST-T segments of the three leads of VCG are combined to form a three-dimensional ST-T loop. VCG temporal heterogeneity and VCG spatial heterogeneity are calculated based on the three-dimensional ST-T loop. The ST-T segments of each lead of the VGG signal are stitched together frame by frame to form the corresponding time series. Each time series is standardized to obtain the corresponding lead time series. The VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index of each lead time series are calculated. The VCG temporal heterogeneity, VCG spatial heterogeneity, VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index are integrated as the VCG feature set.

[0163] A deterministic learning algorithm is used to process the ST-T segments of three leads to obtain a three-dimensional CDG loop. Based on the three-dimensional CDG loop, CDG temporal heterogeneity and CDG spatial heterogeneity are calculated. After standardizing the signal of each channel of the three-dimensional CDG loop, the channel time series is obtained. The CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index of the channel time series are calculated. The CDG temporal heterogeneity, CDG spatial heterogeneity, CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index are integrated as a CDG feature set.

[0164] The effective feature subsets corresponding to the VCG feature set and the CDG feature set are input into the trained early detection model of coronary artery circulation disorder to obtain the output value, and the presence of coronary artery circulation disorder in the subject is predicted based on the output value.

[0165] The display unit is configured to show an alarm or warning if it is determined that there is a coronary artery circulatory obstruction.

[0166] The technical content of the method executed by the electronic data processing device is the same as that of Embodiment 1, and the repeated content will not be described again here.

[0167] Example 4

[0168] This embodiment also provides an electronic device, see reference. Figure 7 It includes a memory 404 and a processor 402, the memory 404 storing a computer program and the processor 402 being configured to run the computer program to perform the steps in any of the embodiments of the above-described methods for early detection of coronary circulatory disorders based on VCG and CDG.

[0169] Specifically, the processor 402 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0170] The memory 404 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 404 may include removable or non-removable (or fixed) media. Where appropriate, the memory 404 may be internal or external to a data processing device. In a particular embodiment, the memory 404 is non-volatile memory. In a particular embodiment, the memory 404 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0171] The memory 404 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 402.

[0172] The processor 402 reads and executes computer program instructions stored in the memory 404 to implement any of the myocardial ischemia prediction methods based on 12-lead ECG in the above embodiments.

[0173] Optionally, the electronic device may further include a transmission device 406 and an input / output device 408, wherein the transmission device 406 is connected to the processor 402, and the input / output device 408 is connected to the processor 402.

[0174] The transmission device 406 can be used to receive or send data via a network. Specific examples of the network described above may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 406 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0175] The input / output device 408 is used to input or output information. In this embodiment, the input information may be the acquired ECG and VCG signals, etc., and the output information may be the coronary artery circulation disorder detection results, etc.

[0176] Optionally, in this embodiment, the processor 402 can be configured to perform the following steps via a computer program:

[0177] Collect 12-lead ECG signals from the subject;

[0178] The 12-lead ECG signal was converted into a 3-lead VCG signal, and the ST-T segment of each lead of the VCG signal was extracted.

[0179] The ST-T segments of the three leads of VCG are combined to form a three-dimensional ST-T loop. VCG temporal heterogeneity and VCG spatial heterogeneity are calculated based on the three-dimensional ST-T loop. The ST-T segments of each lead of the VGG signal are stitched together frame by frame to form the corresponding time series. Each time series is standardized to obtain the corresponding lead time series. The VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index of each lead time series are calculated. The VCG temporal heterogeneity, VCG spatial heterogeneity, VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index are integrated as the VCG feature set.

[0180] A deterministic learning algorithm is used to process the ST-T segments of three leads to obtain a three-dimensional CDG loop. Based on the three-dimensional CDG loop, CDG temporal heterogeneity and CDG spatial heterogeneity are calculated. After standardizing the signal of each channel of the three-dimensional CDG loop, the channel time series is obtained. The CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index of the channel time series are calculated. The CDG temporal heterogeneity, CDG spatial heterogeneity, CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index are integrated as a CDG feature set.

[0181] The effective feature subsets corresponding to the VCG feature set and the CDG feature set are input into the trained early detection model of coronary artery circulation disorder to obtain the output value, and the presence of coronary artery circulation disorder in the subject is predicted based on the output value.

[0182] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0183] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0184] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products), including software routines, applets, and / or macros, can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. A computer program product may include one or more computer-executable components configured to perform embodiments when the program is run. One or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted that any block in the logical flow of the figures may represent a program step, or interconnected logical circuitry, blocks and functions, or a combination of program steps and logical circuitry, blocks and functions. The software may be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0185] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0186] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for early detection of coronary artery circulatory disorders based on VCG and CDG, characterized in that, Includes the following steps: Collect 12-lead ECG signals from the subject; The 12-lead ECG signal is converted into a 3-lead VCG signal, and the ST-T segment of each lead of the VCG signal is extracted. A three-dimensional ST-T loop is formed by combining the ST-T segments of the three leads of the VCG signal. Based on this three-dimensional ST-T loop, the VCG temporal heterogeneity and VCG spatial heterogeneity are calculated. The ST-T segments of each lead of the VCG signal are stitched together frame by frame to form a corresponding time series. Each time series is standardized to obtain the corresponding lead time series. The VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index of each lead time series are calculated. The VCG temporal heterogeneity, VCG spatial heterogeneity, VCG sample entropy, VCG approximate entropy, and VCG heterogeneity index are integrated as a VCG feature set. A deterministic learning algorithm is used to process the ST-T segments of the three leads to obtain a three-dimensional CDG loop. Based on the three-dimensional CDG loop, CDG temporal heterogeneity and CDG spatial heterogeneity are calculated. After standardizing the signal of each channel of the three-dimensional CDG loop, a channel time series is obtained. The CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index of the channel time series are calculated. The CDG temporal heterogeneity, CDG spatial heterogeneity, CDG sample entropy, CDG approximate entropy, and CDG heterogeneity index are integrated as a CDG feature set, where CDG is an electrocardiogram. Effective feature subsets are selected from the combined feature set of the VCG feature set and the CDG feature set. These effective feature subsets are input into the early detection model of coronary artery circulatory disorders to obtain output values. Based on these output values, the model predicts whether the subject has coronary artery circulatory disorders. The effective feature subsets are selected from the VCG feature set, CDG feature set, and combined feature set using a backward sequence selection algorithm. First, the VCG feature set, CDG feature set, and combined feature set are selected as candidate subsets. The iteration variable n is set to the initial number of features, where each subset includes n-1 features. Five-fold cross-validation is used to train and evaluate the classification performance of these subsets on the training set. The subset with the highest accuracy is selected. In the next iteration, n-2 subsets are built from the selected n-1 subsets and evaluated and trained. The best subset is selected. This process is repeated until n=1. The effective feature subset includes VCG spatial heterogeneity VCGSHI, VCG temporal heterogeneity VCGTHI, VCG x-axis lead approximate entropy VxApEn, VCG x-axis lead heterogeneity index VxCI, VCG y-axis lead approximate entropy VyApen, VCG z-axis lead sample entropy VzSampEn, VCG z-axis lead approximate entropy VzApen, VCG z-axis lead heterogeneity index VzCI, CDG spatial heterogeneity CDGSHI, CDG temporal heterogeneity CDGTHI, CDG first channel time series approximate entropy W1S1ApEn, CDG second channel time series sample entropy W2S2SampEn, CDG second channel time series approximate entropy W2S2Apen, CDG third channel time series approximate entropy W3S3Apen, and CDG third channel time series heterogeneity index W3S3CI.

2. The method for early detection of coronary circulatory disorders based on VCG and CDG according to claim 1, characterized in that, Calculate the multiscale entropy of the time series for each lead, and calculate the area of ​​the multiscale entropy on the time scale as the VCG heterogeneity index.

3. The method for early detection of coronary circulatory disorders based on VCG and CDG according to claim 1, characterized in that, The VCGST-T segments of the three leads are input into the RBF neural network with a defined learning model to obtain the electrocardiographic features of the lead. The electrocardiographic features of the three leads form a three-dimensional CDG loop.

4. The method for early detection of coronary artery circulatory disorders based on VCG and CDG according to claim 1, characterized in that, Multilayer perceptron was chosen as the early detection model for coronary artery circulatory disorders, and mixed variables and Gaussian heterogeneity were added to the sparrow algorithm used in the multilayer perceptron.

5. A method for constructing an early detection model of coronary artery circulatory disorders based on VCG and CDG, characterized in that, Includes the following steps: Establish training set: Select VCG feature set and CDG feature set of CMD patients and healthy people as training set, wherein the VCG feature set and CDG feature set are obtained by the method described in any one of claims 1-4, and CMD patients are patients with coronary microcirculation disorders; Training the model: An effective feature subset is input into a multilayer perceptron model for training to obtain an early detection model for coronary artery circulation disorders. The effective feature subset is selected from the VCG feature set, CDG feature set, and combined feature set using a backward sequence selection algorithm. First, the VCG feature set, CDG feature set, and combined feature set are selected as candidate subsets. The iteration variable n is set to the initial number of features, where each subset includes n-1 features. Five-fold cross-validation is used to train and evaluate the classification performance of these subsets on the training set. The subset with the highest accuracy is selected. In the next iteration, n-2 subsets are created from the selected n-1 subsets and evaluated and trained. The best subset is selected. This process is repeated until n=1. The effective feature subset includes VCG spatial heterogeneity (VCGSHI). VCG temporal heterogeneity VCGTHI, VCG x-axis lead approximate entropy VxApEn, VCG x-axis lead heterogeneity index VxCI, VCG y-axis lead approximate entropy VyApen, VCG z-axis lead sample entropy VzSampEn, VCG z-axis lead approximate entropy VzApen, VCG z-axis lead heterogeneity index VzCI, CDG spatial heterogeneity CDGSHI, CDG temporal heterogeneity CDGTHI, CDG first channel time series approximate entropy W1S1ApEn, CDG second channel time series sample entropy W2S2SampEn, CDG second channel time series approximate entropy W2S2Apen, CDG third channel time series approximate entropy W3S3Apen, and CDG third channel time series heterogeneity index W3S3CI.

6. The method for constructing an early detection model of coronary artery circulation disorders based on VCG and CDG according to claim 5, characterized in that, The multi-layer perception model is optimized using the Sparrow Intelligence Group algorithm, and mixed variables and Gaussian alienation methods are added to the Sparrow algorithm.

7. An early detection model for coronary artery circulatory disorders based on VCG and CDG, characterized in that, The model for early detection of coronary circulatory disorders based on VCG and CDG, as described in claim 5, was constructed.

8. An application method for an early detection method of coronary circulatory disorders based on VCG and CDG according to any one of claims 1-4, characterized in that, include: The device comprises a 12-lead electrocardiogram (ECG) device, an electronic data processing device, and a display component; the 12-lead ECG device is used to acquire 12-lead ECG signals from a subject; the electronic data processing device performs the method for early detection of coronary artery circulatory disorders as described in any one of claims 1-4; and the display component is configured to display a myocardial ischemia alarm or warning when it is determined that coronary artery circulatory disorders are present.

Images