A coronary heart disease detection system based on electrocardio-dynamic static fusion and combined recursive analysis

Abstract

The present application belongs to the field of intelligent analysis of electrocardiosignal, and proposes a coronary heart disease detection system based on electrocardio dynamic and static fusion and joint recursive analysis. On the basis of variational mode decomposition of electrocardio vector signal and the change rule of the electrocardio vector signal, the joint recursive method is used to fuse the information of the static domain and the dynamic domain of the electrocardio signal, and the recursive fusion features that can complement each other are extracted. The limitation of insufficient features extracted from a single data domain is avoided, and the reliability of the diagnosis result is improved. The present application can capture the complementary information between the static domain and the dynamic domain of the electrocardio signal, fuse multi-level signals, reveal the internal rule of the electrocardio signal, better understand the pathogenesis, and provide a reliable basis for coronary heart disease detection.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent analysis of electrocardiosignal, and particularly relates to a coronary heart disease detection system based on electrocardio dynamic and static fusion and joint recursive analysis. BACKGROUND

[0002] Coronary heart disease seriously threatens human health, and if it is detected early and effective treatment is taken, it can effectively reduce malignant cardiovascular events and save patients' lives.

[0003] The inventor found that electrocardiogram, as a primary diagnosis method for heart disease, has the characteristics of non-invasiveness and convenience, and has been widely used in the diagnosis of coronary heart disease. However, the electrocardiosignal is weak and complex and diverse, so the diagnosis using electrocardiogram alone brings great challenges to people. The traditional method only extracts time (static) domain features based on morphology, statistics and signal transformation, but it is difficult to fully represent the electrocardiosignal by only obtaining features from the static domain, which affects the detection result of coronary heart disease. SUMMARY

[0004] In order to solve the above problems, the application provides a coronary heart disease detection system based on electrocardio dynamic and static fusion and joint recursive analysis, which uses a joint recursive method to fuse the information of the static domain and the dynamic domain (electrocardiosignal change rule) of the electrocardiosignal on the basis of variational mode decomposition, and extracts recursive fusion features that can complement each other between the two, thereby improving the detection performance.

[0005] In order to achieve the above purpose, the application provides a coronary heart disease detection system based on electrocardio dynamic and static fusion and joint recursive analysis, which uses the following technical scheme:

[0006] A coronary heart disease detection system based on electrocardio dynamic and static fusion and joint recursive analysis comprises:

[0007] A dynamic modeling module configured to perform local dynamic modeling on an acquired electrocardio vector signal to obtain a change rule of the electrocardio vector signal;

[0008] A variational mode decomposition module configured to perform variational mode decomposition on the electrocardio vector signal and the change rule of the electrocardio vector signal;

[0009] A joint recursive analysis module configured to use a joint recursive analysis method to perform fusion analysis on the static domain information of the decomposed electrocardio vector signal and the dynamic domain information of the change rule; extract complementary information between the static domain information and the dynamic domain information after the fusion analysis to obtain recursive fusion features;

[0010] A detection module configured to perform feature screening on the recursive fusion features to form a feature subset; and use the feature subset to detect coronary heart disease.

[0011] Further, the twelve-lead electrocardio signal is preprocessed, that is, the baseline drift noise of the twelve-lead electrocardio signal is removed, then converted into a three-lead electrocardio vector signal, and the ST-T segment reflecting the electrocardio repolarization process is intercepted.

[0012] Further, the ST-T segment intercepted from the three-lead electrocardio vector signal is defined as a nonlinear dynamics system including an ST-T vector ring.

[0013] Further, a radial basis function neural network is used to locally model the ST-T vector ring by using a determination learning algorithm, so as to obtain the change rule of the electrocardio vector signal.

[0014] Further, before the variational mode decomposition, the electrocardio vector signal and the change rule of the electrocardio vector signal are normalized.

[0015] Further, the electrocardio vector signal is decomposed into a plurality of discrete modes, the center frequency and bandwidth of each mode are determined, and iteration search is performed to minimize the sum of the estimated bandwidth of each mode.

[0016] Further, the static domain information of the decomposed electrocardio vector signal and the dynamic domain information of the change rule are fused and analyzed by using a joint recursive analysis method, so as to obtain a recursive fusion feature.

[0017] Further, the recursive fusion feature includes a recursive rate, certainty, average diagonal length, entropy, laminar flow and capture time.

[0018] Further, the importance of the recursive fusion feature is calculated by using a variance homogeneity test, the features with significant differences are input into a pre-trained coronary heart disease detection model according to the feature importance ranking, and a detection result is obtained.

[0019] Compared with the prior art, the present application has the following beneficial effects:

[0020] 1、On the basis of variational mode decomposition of the electrocardio vector signal and the change rule of the electrocardio vector signal, the joint recursive method is used to fuse the information of the static domain and the dynamic domain of the electrocardio signal, and the recursive fusion features that can complement each other are extracted; the limitation of insufficient features extracted from a single data domain is avoided, the reliability of the diagnosis result is improved, the complementary information between the static domain and the dynamic domain of the electrocardio signal can be captured, multiple levels of signals can be fused, the internal rule of the electrocardio signal can be revealed, the physiological mechanism of the disease can be better understood, and a reliable basis for coronary heart disease detection can be provided.

[0021] 2、The application can improve the diagnosis performance of coronary heart disease in clinical environment; pericarditis, electrolyte disturbance and other factors can cause repolarization abnormal change in addition to coronary heart disease, and the repolarization change caused by different causes can interfere with the judgment of clinicians, and the application can accurately identify coronary heart disease from the interference of other diseases. BRIEF DESCRIPTION OF DRAWINGS

[0022] The drawings constituting a part of this embodiment are used to provide further understanding of the embodiment, and the illustrative embodiment and its description are used to explain the embodiment, and do not constitute improper limitation on the embodiment.

[0023] Figure 1 The flow chart is realized for the system of the embodiment 1 of the application;

[0024] Figure 2 The process of obtaining the recurrence graph by joint recursive analysis for a healthy person in the embodiment 1 of the application;

[0025] Figure 3 The x-lead decomposition result of the variable mode decomposition result of a certain case in the embodiment 1 of the application;

[0026] Figure 4 The change rule decomposition result of the x-lead of the variable mode decomposition result of a certain case in the embodiment 1 of the application;

[0027] Figure 5 The y-lead decomposition result of the variable mode decomposition result of a certain case in the embodiment 1 of the application;

[0028] Figure 6 The change rule decomposition result of the y-lead of the variable mode decomposition result of a certain case in the embodiment 1 of the application;

[0029] Figure 7 The z-lead decomposition result of the variable mode decomposition result of a certain case in the embodiment 1 of the application;

[0030] Figure 8 The change rule decomposition result of the z-lead of the variable mode decomposition result of a certain case in the embodiment 1 of the application;

[0031] Figure 9 The scatter plot of the healthy and coronary heart disease patients in the embodiment 1 of the application. DETAILED DESCRIPTION

[0032] The application will be further described below in combination with the drawings and embodiments.

[0033] It should be pointed out that the following detailed description is exemplary and is intended to provide further description of the application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the application belongs.

[0034] Embodiment 1

[0035] Traditional methods only extract time (static) domain features based on morphology, statistics and signal transformation, but it is difficult to fully represent electrocardiogram signals by only obtaining features from the static domain, which affects the detection result of coronary heart disease; To solve this problem, the embodiment provides a coronary heart disease detection system based on electrocardiogram dynamic and static fusion and joint recursive analysis, comprising:

[0036] A dynamic modeling module is configured to locally model the acquired electrocardiogram vector signal to obtain the change rule of the electrocardiogram vector signal.

[0037] A variational mode decomposition module is configured to perform variational mode decomposition on the electrocardiogram vector signal and the change rule of the electrocardiogram vector signal.

[0038] A joint recursive analysis module is configured to use a joint recursive analysis method to fuse and analyze the static domain information of the decomposed electrocardiogram vector signal and the dynamic domain information of the change rule; extract complementary information between the static domain information and the dynamic domain information after fusion analysis to obtain recursive fusion features.

[0039] A detection module is configured to perform feature screening on the recursive fusion features to form a feature subset; and use the feature subset to detect coronary heart disease.

[0040] By determining the learning to model the dynamic of the complex process of the electrocardiogram signal, the weak electrocardiogram changes can be captured; from the perspective of diagnosis, single type of diagnostic information is not comprehensive and accurate, and obtaining multi-dimensional information from multiple aspects and using it can be more accurate in diagnosis. Therefore, based on the joint recursive fusion of electrocardiogram static domain and dynamic domain information, complementary diagnostic information is extracted to obtain a more comprehensive electrocardiogram signal representation, which has important value for improving the performance of coronary heart disease diagnosis. In the embodiment, the electrocardiogram signal is first preprocessed, the change rule of the electrocardiogram signal is obtained by using the deterministic learning, and the electrocardiogram signal and its change rule are decomposed by variational mode decomposition, and then the joint recursive analysis method is used to fuse the multi-data domain information to realize data layer fusion, eliminate redundant information, extract complementary recursive fusion features with diagnostic significance between the two, and fully represent the pathogenesis of coronary heart disease.

[0041] Specifically, on the basis of variational mode decomposition of the electrocardio vector signal and the change rule of the electrocardio vector signal, a joint recursive method is used to fuse the information of the static domain and the dynamic domain of the electrocardio signal, and the recursive fusion features complementary between the two are extracted; the limitation of insufficient features extracted from a single data domain is avoided, and the reliability of the diagnosis result is improved; the complementary information between the static domain and the dynamic domain of the electrocardio signal can be captured, multi-level signals can be fused, the internal rule of the electrocardio signal can be revealed, and the disease physiological mechanism can be better understood, thereby providing a reliable basis for the detection of coronary heart disease.

[0042] The main steps of the coronary heart disease detection system based on electrocardio dynamic and static fusion and joint recursive analysis in the embodiment can include:

[0043] S1, data preprocessing is performed on twelve-lead electrocardio signals, the twelve-lead electrocardio signals are converted into three-lead electrocardio vector signals, and the ST-T segment reflecting the electrocardio repolarization process is intercepted;

[0044] S2, then a local accurate dynamic modeling of the three-lead electrocardio vector signals is performed by using a deterministic learning algorithm, so as to obtain the change rule of the electrocardio vector signals;

[0045] S3, the three-lead electrocardio vector signals and the change rule thereof are normalized and subjected to variational mode decomposition;

[0046] S4, the static domain information of the decomposed electrocardio vector signals and the dynamic domain information of the change rule thereof are fused and analyzed by using a joint recursive analysis method, and recursive rate, certainty, average diagonal length, entropy, laminarity and capture time and other recursive features of dynamic and static fusion are extracted;

[0047] S5, the obtained recursive fusion features are subjected to feature selection by using a filtering type feature selection method such as F-test to form a feature subset, and then the features are input into a classifier to evaluate the performance of coronary heart disease detection.

[0048] Optionally, the implementation manner of step S1 is:

[0049] The twelve-lead electrocardio signals are preprocessed to remove noise such as baseline drift, and are converted into three-lead electrocardio vector signals by using a Kors matrix, and the ST-T segment reflecting the electrocardio repolarization process is intercepted.

[0050] Optionally, the implementation manner of step S2 is:

[0051] Firstly, the three-lead electrocardio vector signals are regarded as a quasi-periodic trajectory generated by the following nonlinear dynamic system:

[0052]

[0053] Wherein x = [x1, x2, x3] T∈R 3 represents a three-dimensional electrocardiogram ST-T vector loop, wherein x(t) is a one-dimensional time series, and t is a sampling time; F(x;p)=[f1(x;p), f2(x;p), f3(x;p)] T is a nonlinear system dynamics, wherein F(x;p) is a nonlinear vector function.

[0054] Secondly, a radial basis function neural network identifier is used to locally and accurately model the ST-T vector loop:

[0055]

[0056] wherein, is a state estimation error; T S is a sampling time; is a learning gain; S is a regression vector of the RBF neural network; is a weight of the RBF neural network.

[0057]

[0058] wherein, that is, the variation law of the electrocardiogram signal is obtained by determining learning.

[0059] Optionally, the implementation manner of step S3 is:

[0060] The three-lead electrocardiogram vector signal and its variation law are normalized and subjected to variational mode decomposition, so that the electrocardiogram vector signal is decomposed into K discrete modes u k and the center frequency ω k and bandwidth of each mode are determined, and iterative search is performed to minimize the sum of the bandwidths of each mode:

[0061]

[0062] wherein, is a partial derivative of a function; ω k is a center frequency of each mode component; δ is a Dirac function, t is time, and j is a complex square root of -1.

[0063] Optionally, the implementation manner of step S4 is:

[0064] The decomposed signal is subjected to joint recursive analysis through formula (5), and recursive rate, certainty, average diagonal length, entropy, laminarity and capture time and other dynamic and static data layer fusion recursive features are extracted:

[0065]

[0066] wherein, i, j=1,..., N; and are respectively the radius threshold of the electrocardio vector signal and its variation law; and are respectively the trajectory path of the electrocardio vector signal and its variation law.

[0067] The recurrence rate is calculated by formula (6):

[0068]

[0069] Wherein, M is the number of points on the phase space trajectory.

[0070] The determinacy is calculated by formula (7):

[0071]

[0072] Wherein, M is the number of points on the diagonal line of the recurrence plot; s(d) is the number of line segments with length d.

[0073] The average diagonal line length is calculated by formula (8):

[0074]

[0075] The entropy is calculated by formula (9):

[0076]

[0077] Wherein,

[0078] The laminarity is calculated by formula (10):

[0079]

[0080] The capture time is calculated by formula (11):

[0081]

[0082] Wherein, M is the number of points of the vertical line of the recurrence plot; s(v) is the number of line segments with vertical line length v, v min Can be 2.

[0083] Optionally, the implementation manner of step S5 is:

[0084] The obtained recurrence features are used to calculate the feature importance by variance homogeneity test (F-test), and the features with significant differences are input and trained to obtain the coronary heart disease detection model according to the feature importance ranking.

[0085] In order to verify and further illustrate the method in the embodiment, in the embodiment, the PTB database is analyzed, the database includes 549 electrocardiogram records of 290 subjects from the cardiology department of an outpatient clinic of a university, the 290 subjects include 148 patients with severe coronary heart disease and 52 healthy subjects, each record contains 15 synchronous cardiac detection signals, namely, 12 traditional electrocardiogram leads and 3 Frank lead vectorcardiogram signals. The signals are digitized in the range of ± 16.384 mV, with a sampling rate of 1 kHz and a resolution of 16 bits. 200 electrocardiograms of 148 coronary heart disease patients and 52 healthy controls are used for coronary heart disease evaluation, and the classifier is used for testing, and good classification results are obtained.

[0086] The above only describes the preferred embodiments of the present embodiment and is not intended to limit the present embodiment. Those skilled in the art can make various modifications and changes to the present embodiment. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present embodiment shall be included in the protection scope of the present embodiment.

Claims

1. A coronary artery disease detection system based on electrocardiographic-static fusion and joint recursive analysis, characterized in that, include: The dynamics modeling module is configured to perform local dynamics modeling on the acquired electrocardiogram vector signal to obtain the variation law of the electrocardiogram vector signal. The variational mode decomposition module is configured to perform variational mode decomposition on the electrocardiogram vector signal and the variation law of the electrocardiogram vector signal; The joint recursive analysis module is configured to: use the joint recursive analysis method to perform fusion analysis on the static domain information and dynamic domain information of the decomposed ECG vector signal; extract the complementary information between the static domain information and the dynamic domain information after fusion analysis to obtain the recursive fusion features; The detection module is configured to: perform feature filtering on the recursive fusion features to form a feature subset; and use the feature subset to detect coronary heart disease. The twelve-lead ECG signal was preprocessed, that is, the baseline drift noise of the twelve-lead ECG signal was removed and it was converted into a three-lead ECG vector signal, and the ST-T segment reflecting the ECG repolarization process was extracted. The ECG vector signal is decomposed into multiple discrete modes, the center frequency and bandwidth of each mode are determined, and an iterative search is performed to minimize the sum of the estimated bandwidths of each mode. The static domain information and dynamic domain information of the decomposed electrocardiogram vector signal are fused and analyzed using the joint recursive analysis method to obtain the recursive fusion features. Recursive fusion features include recursion rate, determinism, average diagonal length, entropy, laminarity, and capture time.

2. The coronary artery disease detection system based on electrocardiographic-static fusion and joint recursive analysis as described in claim 1, characterized in that, The ST-T segment extracted from the three-lead ECG vector signal is defined as a nonlinear dynamic system including the ST-T vector loop.

3. The coronary artery disease detection system based on electrocardiographic-static fusion and joint recursive analysis as described in claim 2, characterized in that, By employing a deterministic learning algorithm and a radial basis function neural network to model the local dynamics of the ST-T vector loop, the variation law of the electrocardiogram vector signal was obtained.

4. The coronary artery disease detection system based on electrocardiographic-static fusion and joint recursive analysis as described in claim 1, characterized in that, Before performing variational mode decomposition, the ECG vector signal and its variation law are normalized.

5. The coronary artery disease detection system based on electrocardiographic-static fusion and joint recursive analysis as described in claim 1, characterized in that, The importance of the recursive fusion features is calculated using the homogeneity of variance test. Based on the importance of the features, they are ranked, and the features with significant differences are input into the pre-trained coronary heart disease detection model to obtain the detection results.

Images