Coronary Microcirculation Disorder Detection Model, Construction Method and Application

Abstract

The present application proposes a detection model, construction method and application for coronary microcirculation disorder, and uses the sample entropy, approximate entropy and heterogeneity index of the ECG ST-T segment and the VCG ST-T segment respectively to construct a detection model for coronary microcirculation disorder. The constructed detection model for coronary microcirculation disorder can achieve non-invasive and efficient early screening of coronary microcirculation disorder. The detection model for coronary microcirculation disorder is easy to be loaded on traditional electrocardiogram acquisition devices and has wide clinical application value.

Description

Technical Field

[0001] This application relates to the field of electrocardiogram signal processing, and in particular to a coronary microcirculation disorder detection model, a construction method and an application thereof. Background Art

[0002] Myocardial ischemia causes 16% of the global mortality rate and is listed as the number one killer by the World Health Organization. Obstructive coronary artery disease and coronary microcirculation disorder are the main causes of myocardial ischemia. However, for a long time, obstructive coronary artery disease has been considered the primary or even the only cause of myocardial ischemia. Therefore, the pathogenesis, treatment and diagnosis of obstructive coronary artery disease have been fully developed. However, with the continuous in-depth study of the mechanism of myocardial ischemia, coronary microcirculation disorder has received more and more attention in recent years because a considerable number of patients do not have obstructive coronary artery disease or still have ischemic symptoms after full treatment of the disease.

[0003] Existing clinical detection methods for coronary microcirculation disorder, such as FFR, index of microcirculatory resistance (IMR), and positron emission tomography (PET), are not suitable for early screening because they are invasive, radioactive, and technically difficult. The prior art CN111407458A provides a method for establishing a goat coronary microcirculation disorder model. This scheme is to inject microembolism balls into goats and then perform coronary angiography to intervene in the establishment of a goat coronary microcirculation disorder model. This method is not applicable to human detection and is also an invasive behavior. In view of the importance of coronary microcirculation disorder in clinical practice, there is an urgent need to establish a non-invasive and widely applicable detection method. Summary of the Invention

[0004] The embodiments of this application provide a coronary microcirculation disorder detection model, a construction method and an application thereof. This scheme constructs a coronary microcirculation disorder detection model that can detect coronary microcirculation disorder. This model uses the entropy of the ECG ST-segment as an effective feature, can non-invasively and efficiently detect coronary microcirculation disorder, and provides technical support for the early screening of this disease.

[0005] In a first aspect, the embodiments of this application provide a construction method for a coronary microcirculation disorder detection model, including the following steps:

[0006] Selection of training samples:

[0007] Collect at least one 12-lead electrocardiogram signal, convert each of the 12-lead ECG signals into a 3-lead vectorcardiogram signal, respectively intercept at least one lead's ST-T segment of the ECG signal and at least one lead's ST-T segment of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the sample entropy of the corresponding ECG ST-T segments and VCG ST-T segments to form a sample entropy feature set;

[0008] Training of the coronary microcirculation disorder detection model:

[0009] Use the sample entropy feature set to train a machine learning model to obtain a coronary microcirculation disorder detection model.

[0010] In a second aspect, an embodiment of the present application provides a method for constructing a coronary microcirculation disorder detection model, including the following steps:

[0011] Selection of training samples:

[0012] Collect at least one 12-lead electrocardiogram signal, convert each of the 12-lead ECG signals into a 3-lead vectorcardiogram signal, respectively intercept at least one lead's ST-T segment of the ECG signal and at least one lead's ST-T segment of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the approximate entropy of the corresponding ECG ST-T segments and VCG ST-T segments to form an approximate entropy feature set;

[0013] Training of the coronary microcirculation disorder detection model:

[0014] Use the approximate entropy feature set to train a machine learning model to obtain a coronary microcirculation disorder detection model.

[0015] In a third aspect, an embodiment of the present application provides a method for constructing a coronary microcirculation disorder detection model, including the following steps:

[0016] Selection of training samples:

[0017] Collect at least one 12-lead electrocardiogram signal, convert each of the 12-lead ECG signals into a 3-lead vectorcardiogram signal, respectively intercept at least one lead's ST-T segment of the ECG signal and at least one lead's ST-T segment of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the heterogeneity index of the corresponding ECG ST-T segments and VCG ST-T segments to form a heterogeneity index feature set;

[0018] Training of the coronary microcirculation disorder detection model:

[0019] A machine learning model is trained using a heterogeneity index feature set to obtain a coronary microcirculation disorder detection model.

[0020] In a fourth aspect, an embodiment of the present application provides a coronary microcirculation disorder detection model, which is trained according to the above training method.

[0021] The main contributions and innovations of the present invention are as follows:

[0022] The embodiments of the present application have for the first time confirmed that sample entropy, approximate entropy, and multi-scale entropy are effective features for the detection of coronary microcirculation disorders, and have established an artificial intelligence detection algorithm for non-invasive, economical, and user-friendly detection of coronary microcirculation disorders, realizing early screening of coronary microcirculation disorders. The algorithm established by the present invention can be easily loaded onto traditional electrocardiogram acquisition devices because this algorithm uses the 10-second ECG signals commonly used clinically, and VCG signals can be synthesized from ECG signals, so no additional leads or operations are required. The algorithm established by the present invention uses the clinically conventional 10-second ECG signals, and realizes automatic feature extraction and modeling without increasing the burden on operators or doctors, so it is more easily accepted clinically.

[0023] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more concise and understandable. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments and descriptions thereof are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0025] Figure 1 is a schematic flow chart of constructing a coronary microcirculation disorder detection model according to this solution;

[0026] Figure 2 is a schematic flow chart of the construction method of the coronary microcirculation disorder detection model according to this solution;

[0027] Figure 3 is a schematic flow chart of the detection method of the coronary microcirculation disorder detection model according to this solution;

[0028] Figure 4 is a schematic diagram of a low-quality ECG signal;

[0029] Figure 5 is a flow chart of optimal subset selection using the SBS algorithm;

[0030] Figure 6 is to compare the classification capabilities of different models in the intra-patient mode;

[0031] Figure 7 It is to compare the classification capabilities of different models in the inter-patient mode;

[0032] Figure 8 It is a schematic diagram of the hardware structure of the electronic device according to the embodiment of the present application. Detailed implementation manners

[0033] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with one or more embodiments of this specification. On the contrary, they are merely examples of devices and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0034] It should be noted that: In other embodiments, the steps of the corresponding methods are not necessarily executed in the order shown and described in this specification. In some other embodiments, the steps included in the method may be more or less than those described in this specification. In addition, a single step described in this specification may be decomposed into multiple steps for description in other embodiments; and multiple steps described in this specification may also be combined into a single step for description in other embodiments.

[0035] Electrocardiogram (ECG) has been widely used in the primary screening of myocardial ischemia due to its advantages such as economy, convenience, and non-invasiveness. The artificial intelligence detection algorithm based on ECG further improves the detection sensitivity. The existing artificial intelligence detection algorithms based on ECG mainly detect ischemia caused by obstructive coronary artery disease. Coronary microcirculation disorder can lead to ischemic changes in ECG, specifically manifested as ST segment elevation or depression, T wave changes, such as T wave inversion, bidirectional peak, etc. These changes increase the heterogeneity of the ECG signal. Entropy has been proven to be used to detect the heterogeneity of biological signals. Sample entropy (SampEn), approximate entropy (ApEn), and multiscale entropy (MSE) have been widely used in the detection of obstructive coronary artery disease.

[0036] Embodiment 1

[0037] The embodiment of the present application provides a method for constructing a detection model for coronary microcirculation disorder. Specifically, the method includes:

[0038] Selection of training samples:

[0039] Collect at least one 12-lead ECG signal, convert each of the 12-lead ECG signals into a 3-lead VCG signal, respectively intercept the ST-T segment of at least one lead of the ECG signal and the ST-T segment of at least one lead of the VCG signal, and splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the sample entropy of the corresponding ECG ST-T segments and VCG ST-T segments to form a sample entropy feature set;

[0040] Training of the coronary microcirculation disorder detection model:

[0041] Use the sample entropy feature set to train a machine learning model to obtain a coronary microcirculation disorder detection model.

[0042] Preferably, intercept the ST-T segments of leads I, III, aVR, V1, and V2 of the ECG signal and the ST-T segment of lead Vx of the VCG signal respectively.

[0043] The coronary microcirculation disorder detection model in the first embodiment is constructed based on the sample entropy feature set. The SampEn-based model can be used to name the coronary microcirculation disorder detection model to distinguish it from other coronary microcirculation disorder detection models. In some embodiments, the SampEn-based model is preferably constructed using an SVM model. Of course, the machine learning model can be selected as Support Vector Machine (SVM), Multilayer Perceptron (MLP), Gradient Boosting, Decision Tree, Random Forest, K-Nearest Neighbor (KNN), and ensemble algorithms Bagging Ensemble (Bagging) and Ada Boost.

[0044] The coronary microcirculation disorder detection method corresponding to the SampEn-based model includes the following steps:

[0045] Collect the 12-lead ECG signal of the subject, convert the 12-lead ECG signal into a 3-lead VCG signal, respectively intercept the ST-T segment of at least one lead of the ECG signal and the ST-T segment of at least one lead of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the sample entropy of the corresponding ECG ST-T segments and VCG ST-T segments to form a sample entropy feature set;

[0046] Input the sample entropy feature set into the above-mentioned coronary microcirculation disorder detection model obtained by training to obtain an output result, and judge whether there is coronary microcirculation disorder based on the output result.

[0047] Preferably, intercept the ST-T segments of leads I, III, aVR, V1, and V2 of the ECG signal and the ST-T segment of lead Vx of the VCG signal respectively.

[0048] In the step of selecting training samples in this embodiment, the 12-lead ECG signals of negative training samples and positive training samples are selected simultaneously, where the negative training samples are healthy people, and the positive training samples are patients with coronary microcirculation disorders excluding coronary heart disease. In this solution, patients are diagnosed with coronary microcirculation disorders when the fractional flow reserve (FFR)>0.8 and the index of microcirculatory resistance (IMR)>25. Specifically, this solution can collect 10-second 12-lead ECG signals.

[0049] Before the step of "converting each of the 12-lead ECG signals into 3-lead VCG signals", signal processing can also be performed on the collected 12-lead ECG signals. Specific signal processing includes steps: cleaning low-quality ECG signals, filtering, noise reduction, removing baseline drift, power frequency interference, and electromyographic noise, etc.

[0050] Specifically, first perform ECG signal cleaning, and delete low-quality ECG signals caused by poor electrode contact, electrode detachment, etc., such as Figure 4 Two low-quality ECG signals are shown. The left figure shows the poor contact between leads V1 and V2 and the skin of the measured object, and the right figure shows the detachment of lead avF. Subsequently, a sliding median filter is used to remove baseline drift and low-frequency interference, and a wavelet filter is used to remove high-frequency noise and interference. In wavelet filtering, Coif4 in the Coifiet wavelet family is used as the wavelet basis, and the ECG signal is decomposed at level 4 of Coif4. The adaptive threshold obtained by the principle of Stein's unbiased likelihood estimation is used, and then the wavelet coefficients are subjected to soft threshold filtering. After that, the electrocardiogram signal is reconstructed using wavelet inverse transform to remove power frequency and electromyographic interference.

[0051] Before the step of "converting each of the 12-lead ECG signals into 3-lead VCG signals", it not only includes the action of signal processing on the 12-lead ECG signals, but also includes the step of normalizing the 12-lead ECG signals. In a specific embodiment of this solution, all ECG signals are normalized according to the standards of 25 mm / s and 10 mm / mV.

[0052] In the step of "converting each of the 12-lead ECG signals into 3-lead VCG signals", the following formula is used to convert the 12-lead ECG signals into 3-lead VCG signals:

[0053]

[0054] where I, II, V1, V2, V3, V4, V4, V5, and V6 respectively represent the leads of the ECG signals; Vx, Vy, and Vz respectively represent the leads of the VCG.

[0055] In the step of "sequentially stitching the intercepted ECG ST-T segments and VCG ST-T segments beat by beat to form a time series", for the ST-T segments of each lead of the ECG and VCG, they are sequentially stitched beat by beat to obtain a time series {x(n)} = x(1), x(2),..., x(N), where N represents the total number of points of the ST-T segment of each lead, and the following formula is used to normalize the time series where σ and μ respectively represent the standard deviation and mean of the time series.

[0056] In the step of "calculating the sample entropy of the corresponding ECG ST-T segment and VCG ST-T segment to form a sample entropy feature set" or "calculating the sample entropy of the ST-T segments of each lead of the corresponding ECG and each lead of the VCG to form a sample entropy feature set", the calculation methods of the sample entropy of the ECG ST-T segment and the VCG ST-T segment are the same. The following focuses on the calculation method of the sample entropy of the ECG ST-T segment and the VCG ST-T segment:

[0057] Extract subvectors with an embedding dimension of m from the time series to form a first subvector sequence, calculate the similarity probability between any subvector in the first subvector sequence and all other subvectors, and count the average value of the similarity probabilities of all the subvectors to obtain its average probability; extract vectors with an embedding dimension of m + 1 from the standard time series to form a second subvector sequence, calculate the similarity probability between any subvector in the second subvector sequence and all other vectors, and count the average value of the similarity probabilities of all the subvectors to obtain its average probability, and obtain the sample entropy of the ECG ST-T segment or the VCG ST-T segment based on the average probability of the first subvector sequence and the average probability of the second subvector sequence.

[0058] Specifically, calculate the negative logarithm of the quotient of the average probability of the first subvector sequence and the average probability of the second subvector sequence as the sample entropy.

[0059] The specific calculation formula is as follows:

[0060] First, extract subvector sequences with embedding dimensions of m and m + 1 from the time series:

[0061] X m (1),...,X m (N - m + 1) and X m+1 (1),...,X m+1 (N - m)

[0062] Any first sub - vector sequence with an embedding dimension of m is X m (j) = {x(j), x(j + 1),..., x(j + m - 1)}, 1 ≤ j ≤ N - m + 1, and any second sub - vector sequence with an embedding dimension of m + 1 is X m+1 (j) = {x(j), x(j + 1),..., x(j + m)}, 1 ≤ j ≤ N - m;

[0063] Subsequently, the distance between any two sub - vectors with an embedding dimension of m is defined as:

[0064]

[0065] Similarly, the distance between any two sub - vectors with an embedding dimension of m + 1 is defined as:

[0066]

[0067] Calculate the similarity probability between any sub - vector Xm(j) and any sub - vector Xm(z):

[0068]

[0069] Calculate the similarity probability between any sub - vector Xm + 1(j) and any sub - vector Xm + 1(z):

[0070]

[0071] Where represents the number of distances between any two sub - vectors with an embedding dimension of m that are less than r, d{X m (j), X m (z))} < r; represents the number of distances between any two sub - vectors with an embedding dimension of m + 1 that are less than r, d{X m+1 (j), X m+1 (z)} < r. r is the error tolerance range of the similarity region.

[0072]

[0073]

[0074] Where and They represent the average probabilities of sub-vectors with embedding dimensions of m and m + 1 respectively.

[0075] Finally, SampEn is calculated using the following formula:

[0076]

[0077] Among them, the parameter settings are as follows: r = 0.1, m = 2.

[0078] It is worth mentioning that in this solution, the sample entropy of the ECG ST-T segment and the VCG ST-T segment of each lead is selected as the training sample of the SampEn-based model. In the feature selection stage, the method of sequential backward selection is used for selection. As Figure 5 shown, this solution first selects the sample entropy of the ECG ST-T segment and the VCG ST-T segment of 15 leads as the alternative subset set. The iteration variable n is set to 15. Each subset includes n - 1 sample entropies. Five-fold cross-validation is used to train and evaluate the classification performance of these subsets in the training set entropy, and the subset with the highest accuracy is selected. In the next iteration, a series of subsets with n - 2 are established from the selected subset with n - 1 entropies and evaluated and trained, and the best subset is selected. This process is repeated until n = 1. Explain to obtain a series of subsets with n from 15 to 1, and then select S I , S III , S aVR , S V1 , S V2 , S Vx as the optimal subset.

[0079] Embodiment 2

[0080] The embodiment of the present application provides a method for constructing a coronary microcirculation disorder detection model. Specifically, the method includes:

[0081] Selection of training samples:

[0082] Collect at least one 12-lead ECG signal, convert each 12-lead ECG signal into a 3-lead VCG signal, respectively intercept the ST-T segment of at least one lead of the ECG signal and the ST-T segment of at least one lead of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the approximate entropy of the corresponding ECG ST-T segments and VCG ST-T segments to form an approximate entropy feature set;

[0083] Training of the coronary microcirculation disorder detection model:

[0084] Use the approximate entropy feature set to train a machine learning model to obtain a coronary microcirculation disorder detection model.

[0085] Preferably, the ST-T segments of each lead of the ECG signals I, II, V1, V2, V3, V5, and V6 and the ST-T segment of the Vy lead of the VCG signal are intercepted respectively.

[0086] The coronary microcirculation disorder detection model in the second embodiment is constructed based on the approximate entropy feature set. The ApEn-based model can be used to name the coronary microcirculation disorder detection model to distinguish it from other coronary microcirculation disorder detection models. In some embodiments, the ApEn-based model is preferably constructed using the SVM model. Of course, the machine learning model can be selected as Support Vector Machine (SVM), Multilayer Perceptron (MLP), Gradient Boosting, Decision Tree, Random Forest, K-Nearest Neighbor (KNN), and ensemble algorithms Bagging ensemble (Bagging) and Ada boost.

[0087] The coronary microcirculation disorder detection method corresponding to the ApEn-based model includes the following steps:

[0088] Collect the 12-lead ECG signals of the subject, convert the 12-lead ECG signals into 3-lead VCG signals, intercept the ST-T segments of at least one lead of the ECG signals and the ST-T segments of at least one lead of the VCG signals respectively, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the approximate entropy of the corresponding ECG ST-T segments and VCG ST-T segments to form an approximate entropy feature set;

[0089] Input the approximate entropy feature set into the above-trained coronary microcirculation disorder detection model to obtain an output result, and judge whether there is coronary microcirculation disorder based on the output result.

[0090] Preferably, the ST-T segments of each lead of the ECG signals I, II, V1, V2, V3, V5, and V6 and the ST-T segment of the Vy lead of the VCG signal are intercepted respectively.

[0091] For the other ECG processing technical means in the second embodiment, they are the same as those in the first embodiment. The only difference is that the calculation process of the approximate entropy of the corresponding ECG ST-T segment and VCG ST-T segment is different from the calculation process of the sample entropy. Therefore, the following focuses on the calculation process of the approximate entropy of the ECG ST-T segment and VCG ST-T segment:

[0092] Extract sub-vectors with an embedding dimension of m from the time series and form a third sub-vector sequence. Calculate the similarity probability between any sub-vector in the third sub-vector sequence and all sub-vectors, and count the average value of the logarithms of all the similarity probabilities of the sub-vectors to obtain its average similarity probability; extract vectors with an embedding dimension of m + 1 from the standard time series and form a fourth sub-vector sequence. Calculate the similarity probability between any sub-vector in the fourth sub-vector sequence and all the vectors, and count the average value of the logarithms of all the similarity probabilities of the sub-vectors to obtain its average similarity probability. Based on the difference between the average probability of the third sub-vector sequence and the average probability of the fourth sub-vector sequence, obtain the approximate entropy of the ECG ST-T segment or VCG ST-T segment.

[0093] The specific calculation formula is as follows:

[0094] First, extract sub-vector sequences with embedding dimensions of m and m + 1 from the time series:

[0095] X m (1),...,X m (N - m + 1) and X m+1 (1),...,X m+1 (N - m);

[0096] Where any sub-vector with an embedding dimension of m is X m (j) = {x(j), x(j + 1),..., x(j + m - 1)}, 1 ≤ j ≤ N - m + 1, and any sub-vector with an embedding dimension of m + 1 is X m+1 (j) = {x(j), x(j + 1),..., x(j + m)}, 1 ≤ j ≤ N - m;

[0097] Subsequently, the distance between any two sub-vectors with an embedding dimension of m is defined as:

[0098]

[0099] Similarly, the distance between any two sub-vectors with an embedding dimension of m + 1 is defined as:

[0100]

[0101] and They are calculated respectively using the following formulas:

[0102]

[0103]

[0104] where represents the number of pairs of sub-vectors with an embedding dimension of m whose distance is less than r, d{X m (j), X m (z))} < r; represents the number of pairs of sub-vectors with an embedding dimension of m + 1 whose distance is less than r, d{X m+1 (j), X m+1 (z)} < r. r is the error tolerance range of the similar region.

[0105]

[0106]

[0107] where φ m (r) and φ m+1 (r) represent the average similarity probabilities of sub-vectors with embedding dimensions of m and m + 1, respectively. Finally, ApEn is calculated by the following formula:

[0108] ApEn(N, m, r) = φ m (r) - φ m+1 (r);

[0109] where the parameters are set as r = 0.1 and m = 2.

[0110] It is worth mentioning that in this scheme, the approximate entropy of the ECG ST-T segment and the VCG ST-T segment of each lead is selected as the training sample of the ApEn-based model. In the feature selection stage, this scheme uses the method of sequential backward selection for selection. As Figure 5 shown, this scheme first selects the approximate entropy of the ECG ST-T segment and the VCG ST-T segment of 15 leads as the alternative subset set, the iteration variable n is set to 15, where each subset includes n - 1 approximate entropies, and the classification performance of these subsets is trained and evaluated using five-fold cross-validation on the training set entropy. The subset with the highest accuracy is selected. In the next iteration, a series of subsets with n - 2 are established from the selected subset with n - 1 entropies and evaluated and trained, and the best subset is selected. This process is repeated until n = 1. Explain to obtain a series of subsets from n = 15 to 1, and then select A I , A II , A V1 , A V2 , A V3 , AV5 , A V6 , A Vy as the optimal subset.

[0111] Example 3

[0112] The embodiment of the present application provides a method for constructing a coronary microcirculation disorder detection model. Specifically, the method includes:

[0113] Selection of training samples:

[0114] Collect at least one 12-lead ECG signal, convert each 12-lead ECG signal into a 3-lead VCG signal, respectively intercept the ST-T segment of at least one lead of the ECG signal and the ST-T segment of at least one lead of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the heterogeneity index of the corresponding ECG ST-T segments and VCG ST-T segments to form a heterogeneity index feature set;

[0115] Training of the coronary microcirculation disorder detection model:

[0116] Use the heterogeneity index feature set to train a machine learning model to obtain a coronary microcirculation disorder detection model.

[0117] Preferably, intercept the ST-T segments of leads I, V1, V3, V4, V5, and V6 of the ECG signal and the ST-T segments of leads Vx and Vy of the VCG signal respectively.

[0118] Regarding the definition and explanation of the heterogeneity index, the heterogeneity index refers to the area of multi-scale entropy at each scale.

[0119] The coronary microcirculation disorder detection model in this Example 3 is constructed based on the heterogeneity index feature set. The CI-based model can be used to name this coronary microcirculation disorder detection model to distinguish it from other coronary microcirculation disorder detection models. In some embodiments, the CI-based model is preferably constructed using an MLP model. Of course, the machine learning model can be selected as Support Vector Machine (SVM), Multilayer Perceptron (MLP), Gradient Boosting, Decision Tree, Random Forest, K-Nearest Neighbor (KNN), and ensemble algorithms Bagging ensemble (Bagging) and Ada boost.

[0120] The method for detecting coronary microcirculation disorder corresponding to the CI-based model includes the following steps:

[0121] Collect the 12-lead ECG signals of the subject, convert the 12-lead ECG signals into 3-lead VCG signals, respectively intercept the ST-T segments of at least one lead of the ECG signals and the ST-T segments of the Vx and Vy leads of the VCG signals, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the heterogeneity index of the corresponding ECG ST-T segments and VCG ST-T segments to form a heterogeneity index feature set;

[0122] Input the heterogeneity index feature set into the above-mentioned trained coronary microcirculation disorder detection model to obtain an output result, and judge whether there is coronary microcirculation disorder based on the output result.

[0123] Preferably, intercept the ST-T segments of the I, V1, V3, V4, V5, and V6 leads of the ECG signals and the ST-T segments of the Vx and Vy leads of the VCG signals respectively.

[0124] Regarding other technical means for ECG processing in the third embodiment, they are the same as those in the first embodiment. The only difference is that: the calculation process of calculating the heterogeneity index of the corresponding ECG ST-T segments and VCG ST-T segments is different from the calculation process of calculating sample entropy. Therefore, the following focuses on the calculation process of the heterogeneity index of the ECG ST-T segments and VCG ST-T segments:

[0125] Use the scale factor to coarsely grain the time series of each scale factor to obtain a coarsened time series, calculate the sample entropy at each scale factor, and obtain the area sum of the sample entropy at each scale to obtain the heterogeneity index.

[0126] This solution uses the MSE algorithm to calculate the heterogeneity index of the ECG ST-T segments and VCG ST-T segments. The MSE algorithm includes two parts: sample entropy and time series coarsening. The key in the MSE calculation process is to ensure that the length of the time series at each time scale is long enough to obtain a stable entropy value. Since there is literature proving that a time series of no less than 200 points is required to ensure consistent sample entropy values, the length of the time series in this solution is selected to be between 900 and 1200. And the maximum scale factor in this solution is set to 3.

[0127] The specific calculation formula is as follows:

[0128] At each scale, the coarsened time series of the time series coarsening is:

[0129]

[0130]

[0131] where The τ scale factor, and x(i) is the data in the time series {x(n)} = x(1), x(2), …, x(N), where N represents the length of the time series.

[0132] SampEn is used to calculate the entropy value at each scale factor:

[0133] MSE(x, τ, m, r, k) = SampEn(y( τ ), m, r), 1 # kτ

[0134] In the MSE calculation process, the parameters are set as r = 0.1, m = 2, τ = 3. After the parameter setting, CI is defined as the sum of the areas of MSE at each scale and is calculated by the following formula:

[0135] It is worth mentioning that this scheme selects the heterogeneity index of the ECG ST-T segment and the VCG ST-T segment of each lead as the training samples of the CI-based model. In the feature selection stage, this scheme uses the sequential backward selection method for selection. As Figure 5 shown, this scheme first selects the heterogeneity indexes of the ECG ST-T segment and the VCG ST-T segment of 15 leads as the alternative subset set, sets the iteration variable n to 15, where each subset includes n - 1 heterogeneity indexes, uses five-fold cross-validation to train and evaluate the classification performance of these subsets in the training set entropy, and selects the subset with the highest accuracy. In the next iteration, a series of subsets with n - 2 are established from the selected subsets with n - 1 entropies and are evaluated and trained, and the best subset is selected. This process is repeated until n = 1. Explain to obtain a series of subsets with n from 15 to 1, and then select C II , C IV1 , C IV3 , C IV4 , C IV5 , C IV6 , C IVx , C IVy as the optimal subset. The above scheme constructs three models: the SampEn-based model, the ApEn-based model, and the CI-based model. The following is an evaluation of these models:

[0136] In this invention, 156 10-second ECG signals of 44 patients with coronary microcirculation and 203 ECG signals of 107 healthy people were selected. The dataset was divided and the model was evaluated using two modes: intra-patient and inter-patient. In the intra-patient mode, the 10-second ECG signals were randomly divided into training sets and test sets for positive and negative samples at a ratio of 4:1. In the inter-patient mode, the positive and negative samples were divided into training sets and test sets at a ratio of 4:1 for each patient. Five-fold cross-validation was used on the training set, with one part as the validation set and the rest as the training set. The classification effect of the model was measured using accuracy, sensitivity, specificity, F1-score, and the area under the curve AUC as indicators.

[0137] In the intra-patient mode, based on the validation set, a model with the highest classification performance was selected as the alternative model for the SampEn-based model, ApEn-based model, and CI-based model respectively. The SampEn-based, ApEn-based, and CI-based alternative models were tested on the test set. Finally, the performances of the three models were compared, and the best classification model in this mode was selected for the detection of coronary microcirculation disorders in this mode.

[0138] In the inter-patient mode, based on the validation set, a model with the highest classification performance was selected as the alternative model for the SampEn-based model, ApEn-based model, and CI-based model respectively. The SampEn-based, ApEn-based, and CI-based alternative models were tested on the test set. Finally, the performances of the three models were compared, and the best classification model in this mode was selected for the detection of coronary microcirculation disorders in this mode.

[0139] The experimental results are as Figure 6As shown, the first, second, and third columns respectively show the classification performance of the SampEn-based, ApEn-based, and CI-based models on the validation set. It can be inferred from the first column that, except for a slightly lower sensitivity than the MLP model, the classification performance of the SVM model is significantly better than other models. Therefore, the SVM model was selected as the alternative model for the SampEn-based model. Similarly, SVM and MLP were respectively selected as the alternative models for the ApEn-based and CI-based models. By comparing the classification performance of these three alternative models (SampEn-based SVM, ApEn-based SVM, and CI-based MLP) on the validation set, it was found that the ApEn-based SVM model has the best accuracy, sensitivity, and F1-score. The ApEn-based SVM model has a specificity comparable to that of the SampEn-based SVM model and an AUC comparable to that of the CI-based MLP model, as Figure 6 shown. Further comparing the classification performance of these three alternative models on the test set, it was found that only all evaluation indicators of the ApEn-based SVM model are higher than 0.9, as shown in Table 1. Therefore, the ApEn-based SVM model was selected as the best detection model for coronary microcirculation disorders in the intra-patient mode, but other models can also detect coronary microcirculation disorders.

[0140] Figure 6 Indicates the classification ability of different models in the intra-patient mode: (A) Accuracy of the SampEn-based model. (B) Specificity of the SampEn-based model. (C) Sensitivity of the SampEn-based model. (D) F1 index of the SampEn-based model. (E) AUC of the SampEn-based model. (F) Accuracy of the ApEn-based model. (G) Specificity of the ApEn-based models model. (H) Sensitivity of the ApEn-based model. (I) F1 index of the ApEn-based model. (J) AUC of the ApEn-based model. (K) Accuracy of the CI-based model. (L) Specificity of the CI-based model. (M) Sensitivity of the CI-based model. (N) F1 index of the CI-based model. (O) AUC of the CI-based model.

[0141] Table 1. Classification performance of ApEn-based, SampEn-based, and CI-based models on the test set in the intra-patient mode

[0142]

[0143] Figure 7 Shows the classification performance of ApEn-based, SampEn-based, and CI-based models in the validation set in the inter-patient mode. It can be inferred from the first column that the SampEn-based SVM model has better classification performance than the other seven models and is thus selected as the alternative model. Therefore, except that its sensitivity is lower than that of the MLP model, other metrics are higher than those of the other seven models. Similarly, from the second and third columns, it can be inferred that MLP and SVM are selected as the alternative models for ApEn-based and CI-based, respectively. Further comparing the three alternative models (SampEn-based SVM, ApEn-based MLP, and CI-based SVM), the SampEn-based SVM model has better classification performance than the other two models in the validation set, except that its sensitivity is lower than that of the ApEn-based MLP model and its specificity is lower than that of the SampEn-based SVM model. In the test set, all performance metrics of only the SampEn-based SVM model are higher than 0.8, as shown in Table 2. Therefore, the SampEn-based SVM model is selected as the best model for detecting coronary microcirculation disorders in the inter-patient mode.

[0144] Figure 7 Represents the comparison of the classification capabilities of different models in the inter-patient mode. (A) Accuracy of the SampEn-based model. (B) Specificity of the SampEn-based model. (C) Sensitivity of the SampEn-based model. (D) F1 metric of the SampEn-based model. (E) AUC of the SampEn-based model. (F) Accuracy of the ApEn-based model. (G) Specificity of the ApEn-based models model. (H) Sensitivity of the ApEn-based model. (I) F1 metric of the ApEn-based model. (J) AUC of the ApEn-based model. (K) Accuracy of the CI-based model. (L) Specificity of the CI-based model. (M) Sensitivity of the CI-based model. (N) F1 metric of the CI-based model. (O) AUC of the CI-based model.

[0145] Table 2. Comparison of the classification performance of ApEn-based, SampEn-based, and CI-based models in the test set in the inter-patient mode

[0146]

[0147] Embodiment 4

[0148] This embodiment also provides an electronic device. Referring to Figure 8 , it includes a memory 404 and a processor 402. A computer program is stored in the memory 404, and the processor 402 is configured to run the computer program to execute the steps in any of the above embodiments of the method for constructing a coronary microcirculation disorder detection model.

[0149] Specifically, the above-mentioned processor 402 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured with one or more integrated circuits for implementing the embodiments of the present application.

[0150] Among them, the memory 404 may include a mass storage 404 for data or instructions. By way of example and not limitation, the memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a 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 the data processing device. In a particular embodiment, the memory 404 is non-volatile memory. In a particular embodiment, the memory 404 includes a read-only memory (ROM) and a random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM), or a flash memory, or a combination of two or more of these. Where appropriate, the RAM may be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM may be a fast page mode dynamic random access memory (FPMDRAM), an extended date out dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0151] The memory 404 can be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 402.

[0152] The processor 402 reads and executes the computer program instructions stored in the memory 404 to implement the construction method of any one of the coronary microcirculation disorder detection models in the above embodiments.

[0153] Optionally, the above electronic device may further include a transmission device 406 and an input / output device 408. Among them, the transmission device 406 is connected to the above processor 402, and the input / output device 408 is connected to the above processor 402.

[0154] The transmission device 406 can be used to receive or send data via a network. Specific examples of the above network may include wired or wireless networks provided by a communication provider of the electronic device. In one example, the transmission device includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one example, the transmission device 406 can be a radio frequency (Radio Frequency, abbreviated as RF) module, which is used to communicate with the Internet wirelessly.

[0155] The input / output device 408 is used for electrocardiogram, etc., and the output information can be the detection result of coronary microcirculation disorder, etc.

[0156] Optionally, in this embodiment, the above processor 402 can be set to execute the following steps through a computer program:

[0157] Selection of training samples:

[0158] Collect at least one 12-lead ECG signal, convert each of the 12-lead ECG signals into 3-lead VCG signals, intercept the ECG ST-T segment of each lead ECG signal and the VCG ST-T segment of each lead VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat to form a time series, and calculate the sample entropy or approximate entropy or heterogeneity index of the corresponding ECG ST-T segments and VCG ST-T segments to form a feature set;

[0159] Training of the coronary microcirculation disorder detection model:

[0160] Use the sample entropy feature set to train a machine learning model to obtain a coronary microcirculation disorder detection model.

[0161] Of course, it is also possible to select the sample entropy, approximate entropy, or heterogeneity index of the ECG ST-T segment and VCG ST-T segment of the subject's lead ECG signal to form a test feature set, and input the test feature set into the corresponding coronary microcirculation disorder detection model for detection.

[0162] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation manners, and will not be elaborated herein.

[0163] Generally, various embodiments can be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. Some aspects of the present invention can be implemented in hardware, while other aspects can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, but the present invention is not limited thereto. Although the various aspects of the present invention can 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, the blocks, devices, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuits or logic, general hardware or a controller or other computing device, or some combination thereof.

[0164] The embodiments of the present invention can be implemented by computer software, which can be executed by a data processor of a mobile device, such as in a processor entity, or implemented by hardware, or implemented by a combination of software and hardware. A computer software or program (also referred to as a program product), 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. The computer program product can include one or more computer-executable components that are configured to execute the embodiments when the program runs. The one or more computer-executable components can be at least one software code or a part thereof. Additionally, in this regard, it should be noted that any block in the logical flow in the figure can represent a program step, or an interconnected logical circuit, block, and function, or a combination of a program step and a logical circuit, block, and function. The software can be stored on physical media such as memory chips or storage blocks implemented within the processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and their data variants, CDs. The physical media is a non-transitory medium.

[0165] Those skilled in the art should understand that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0166] The above embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for constructing a detection model of coronary microcirculation disorder, characterized in that, The method includes: Selection of training samples: Collect at least one 12-lead electrocardiogram signal, convert each 12-lead ECG signal into a 3-lead vectorcardiogram signal, respectively intercept the ST-T segment of at least one lead of the ECG signal and the ST-T segment of at least one lead of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, calculate the approximate entropy of the corresponding ECG ST-T segments and VCG ST-T segments to form an approximate entropy feature set, wherein, extract sub-vectors with an embedding dimension of m from the time series and form a third sub-vector sequence, calculate the similarity probability between any sub-vector in the third sub-vector sequence and all sub-vectors, and statistically calculate the average value of the logarithms of all the similarity probabilities of the sub-vectors to obtain its average similarity probability; extract vectors with an embedding dimension of m + 1 from the standard time series and form a fourth sub-vector sequence, calculate the similarity probability between any sub-vector in the fourth sub-vector sequence and all sub-vectors, and statistically calculate the average value of the logarithms of all the similarity probabilities of the sub-vectors to obtain its average similarity probability, and obtain the approximate entropy of the ECG ST-T segment or VCG ST-T segment based on the difference between the average probability of the third sub-vector sequence and the average probability of the fourth sub-vector sequence; Training of the coronary microcirculation disorder detection model: Use the approximate entropy feature set to train a machine learning model to obtain a coronary microcirculation disorder detection model.

2. The method for constructing a coronary microcirculation disorder detection model according to claim 1, wherein Collect at least one 12-lead electrocardiogram signal, convert each 12-lead ECG signal into a 3-lead vectorcardiogram signal, respectively intercept the ST-T segment of at least one lead of the ECG signal and the ST-T segment of at least one lead of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, calculate the approximate entropy of the corresponding ECG ST-T segments and VCG ST-T segments to form an approximate entropy feature set; input the approximate entropy feature set into the above-trained coronary microcirculation disorder detection model to obtain an output result, and judge whether there is a coronary microcirculation disorder based on the output result.

3. A method for constructing a detection model of coronary microcirculation disorder, characterized in that, The method includes: Selection of training samples: Collect at least one 12-lead electrocardiogram signal, convert each 12-lead ECG signal into a 3-lead vectorcardiogram signal, respectively intercept the ST-T segment of at least one lead of the ECG signal and the ST-T segment of at least one lead of the VCG signal, splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, calculate the heterogeneity index of the corresponding ECG ST-T segments and VCG ST-T segments to form a heterogeneity index feature set, wherein, use a scale factor to coarsely granulate the time series of each scale factor to obtain a coarsened time series, calculate the sample entropy at each scale factor, and obtain the area sum of the sample entropy at each scale to obtain the heterogeneity index; Training of the coronary microcirculation disorder detection model: Use the heterogeneity index feature set to train a machine learning model to obtain a coronary microcirculation disorder detection model.

4. The method for constructing a coronary microcirculation disorder detection model according to claim 3, wherein Collect at least one 12-lead electrocardiogram signal, convert each 12-lead ECG signal into a 3-lead vectorcardiogram signal, respectively intercept the ST-T segment of at least one lead of the ECG signal and the ST-T segment of at least one lead of the VCG signal, and splice the intercepted ECG ST-T segments and VCG ST-T segments beat by beat according to the leads to form a time series, and calculate the heterogeneity index of the corresponding ECG ST-T segments and VCG ST-T segments to form a heterogeneity index feature set; Input the heterogeneity index feature set into the above-mentioned trained coronary microcirculation disorder detection model to obtain an output result, and judge whether there is a coronary microcirculation disorder based on the output result.

5. A detection model for coronary microcirculation disorder, characterized in that, It is constructed according to the construction method of the coronary microcirculation disorder detection model described in any one of claims 1-2, 3-4.

Images