Detection method, device and equipment for coronary heart disease severity classification and medium

Abstract

The invention provides a coronary heart disease severity classification detection method and device, equipment and a medium, and the method comprises the steps: processing an electrocardiosignal and a heart sound signal which are collected synchronously, and determining a plurality of ECG interval sequences and a plurality of PCG interval sequences; determining a multi-dimensional coupling feature based on a coupling relationship between the plurality of ECG interval sequences and the plurality of PCG interval sequences; and carrying out feature selection on the multi-dimensional coupling features, screening out target features having a significant distinguishing capability for coronary heart disease severity classification, inputting the target features into a coronary heart disease severity classification model for coronary heart disease severity classification detection, and outputting coronary heart disease severity levels. The deep coupling relationship between the ECG and PCG signal intervals is mined, and the severity of the coronary heart disease is finely classified by using the multi-dimensional coupling features, so that the severity identification accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to detection methods, devices, equipment and media for classifying the severity of coronary heart disease. Background Technology

[0002] Electrocardiogram (ECG) and heart sound signals reflect the electrical and mechanical activity of the heart, respectively. Current technologies primarily employ deep learning methods to detect coronary artery disease (CAD) using ECG and PCG signals. Examples include convolutional neural network models for feature extraction and classification of single ECG or PCG signals, automatically learning deep signal features to achieve CAD detection; multimodal feature-level fusion methods that extract manual features from ECG and PCG signals in the time, frequency, and time-frequency domains and then concatenate or weighted fuse them; and deep learning multimodal fusion methods. These existing technologies typically achieve high accuracy in CAD detection, but they cannot provide precise classification of the severity of CAD. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a detection method, device, equipment and medium for classifying the severity of coronary heart disease, which uses time series coupling analysis to explore the deep coupling relationship between ECG and PCG signal intervals, and uses multi-dimensional coupling features to perform fine classification of the severity of coronary heart disease, thereby improving the accuracy of severity identification.

[0004] This application provides a detection method for classifying the severity of coronary heart disease, the detection method comprising: The synchronously acquired electrocardiogram (ECG) signals and heart sound signals are processed to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals. A multi-dimensional coupling feature is determined based on the coupling relationship between multiple ECG interval sequences and multiple PCG interval sequences; wherein, the multi-dimensional coupling feature includes at least one of cross-entropy feature, multivariate entropy feature and non-entropy feature; Feature selection is performed on the multi-dimensional coupled features to screen out target features that have significant distinguishing ability for the severity grading of coronary heart disease. The target features are then input into the coronary heart disease severity classification model for coronary heart disease severity classification detection, and the severity level of coronary heart disease is output.

[0005] In one possible implementation, the processing of the synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals includes: The electrocardiogram (ECG) signal and the heart sound signal are filtered to determine the filtered ECG signal and the filtered heart sound signal; The filtered electrocardiogram signal is subjected to QRS complex detection processing and R wave peak localization processing to determine multiple ECG interval sequences; The filtered heart sound signal was periodically segmented using a hidden semi-Markov model combined with a dynamic programming algorithm to determine multiple PCG interval sequences.

[0006] In one possible implementation, regarding the cross-entropy feature, the determination of multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences includes: By combining any of the ECG interval sequences with any of the PCG interval sequences, multiple combined interval sequences can be determined. Based on the cross-sample entropy calculation formula, cross-fuzzy entropy calculation formula, joint distribution entropy calculation formula, high-resolution joint symbolic dynamic entropy calculation formula, and symmetric KL distance formula, feature extraction is performed on multiple combined interval sequences to determine multiple cross-entropy features; wherein, the cross-entropy features include at least one of cross-sample entropy features, cross-fuzzy entropy features, joint distribution entropy features, high-resolution joint symbolic dynamic entropy features, and DKL features.

[0007] In one possible implementation, for the multivariate entropy feature, determining the multidimensional coupling feature based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences includes: Based on the set reconstruction dimension, multiple ECG interval sequences and multiple PCG interval sequences are embedded and reconstructed to determine multiple reconstructed sequences; Determine the distance matrix density distribution between the reconstructed sequences; Based on the distance matrix density distribution between reconstructed sequences under the current reconstruction dimension and the distance matrix density distribution between reconstructed sequences under the next reconstruction dimension, the multivariate sample entropy feature and / or multivariate fuzzy entropy feature in the multivariate entropy feature are determined.

[0008] In one possible implementation, for the non-entropy feature, determining the multi-dimensional coupling feature based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences includes: By combining any of the ECG interval sequences with any of the PCG interval sequences, multiple combined interval sequences can be determined. Feature extraction is performed on multiple combined interval sequences based on mutual information function, amplitude squared coherence function, cross power spectral density, and coherent cross power product to determine non-entropy features; wherein, the non-entropy features include mutual information features, frequency domain features, and time domain features.

[0009] In one possible implementation, the feature selection of the multi-dimensional coupled features to screen out target features with significant discriminative ability for the severity grading of coronary heart disease includes: The distribution normality test is performed on the multi-dimensional coupling features to determine the normality test results; Based on the normality test results, a corresponding significance test method is determined. Based on the significance test method, the feature values ​​between different coronary heart disease severity groups are compared, and multiple candidate features with discriminative ability are screened out. The importance score of multiple candidate features is calculated based on the recursive feature elimination algorithm. In each iteration of the score calculation, the candidate feature with the smallest contribution rate is eliminated. The process is repeated recursively until the end of the loop. The remaining candidate features are then used as target features that have significant distinguishing ability for the severity grading of coronary heart disease.

[0010] In one possible implementation, the coronary artery disease severity classification model is determined through the following steps: The sample features of the electrocardiogram signal and the heart sound signal are input into the vector machine classification model to classify and detect the severity of coronary heart disease, and the severity level of coronary heart disease of the sample is output. The vector machine classification model is iteratively trained based on the loss value between the severity level of coronary heart disease in the sample and the actual severity level of coronary heart disease in the sample features, thereby determining the classification model.

[0011] This application embodiment also provides a detection device for classifying the severity of coronary heart disease, the detection device comprising: The signal processing module is used to process the synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals. The coupling feature determination module is used to determine multi-dimensional coupling features based on the coupling relationship between multiple ECG interval sequences and multiple PCG interval sequences; wherein, the multi-dimensional coupling features include at least one of cross-entropy features, multivariate entropy features, and non-entropy features; The classification module is used to select features from the multi-dimensional coupled features, filter out target features that have significant distinguishing ability for the severity grading of coronary heart disease, input the target features into the coronary heart disease severity classification model for coronary heart disease severity classification detection, and output the severity level of coronary heart disease.

[0012] This application also provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the detection method for classifying the severity of coronary heart disease as described above.

[0013] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the detection method for classifying the severity of coronary heart disease as described above.

[0014] This application provides a method, apparatus, device, and medium for classifying the severity of coronary heart disease (CHD). The detection method includes: processing synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals; determining multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences; wherein the multi-dimensional coupling features include at least one of cross-entropy features, multivariate entropy features, and non-entropy features; performing feature selection on the multi-dimensional coupling features to screen out target features with significant discriminative ability for CHD severity classification; inputting the target features into a CHD severity classification model for CHD severity classification detection; and outputting the CHD severity level. By employing time-series coupling analysis to mine the deep coupling relationship between ECG and PCG signal intervals and utilizing multi-dimensional coupling features to perform fine classification of CHD severity, the accuracy of severity identification is improved.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a method for classifying the severity of coronary heart disease, provided in an embodiment of this application; Figure 2 One of the structural schematic diagrams of a detection device for classifying the severity of coronary heart disease provided in an embodiment of this application; Figure 3 A second schematic diagram of a detection device for classifying the severity of coronary heart disease, provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0019] First, the applicable scenarios for this application will be introduced. This application can be applied to the field of data processing technology.

[0020] Research has found that electrocardiogram (ECG) and heart sound signals reflect the electrical and mechanical activity of the heart, respectively. Existing technologies primarily employ deep learning methods to detect coronary artery disease (CAD) using ECG and PCG signals. For example, convolutional neural network models are used to extract and classify features from single ECG or PCG signals, automatically learning deep signal features to achieve CAD detection; multimodal feature-level fusion methods extract manual features from ECG and PCG signals in the time, frequency, and time-frequency domains, then concatenate or weighted fuse them; and deep learning multimodal fusion methods are also used. These existing technologies typically achieve high accuracy in CAD detection, but they cannot achieve precise classification of the severity of CAD.

[0021] Based on this, this application provides a method for classifying the severity of coronary heart disease. It uses time series coupling analysis to mine the deep coupling relationship between ECG and PCG signal intervals and uses multi-dimensional coupling features to classify the severity of coronary heart disease in detail, thereby improving the accuracy of severity identification.

[0022] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for classifying the severity of coronary heart disease, as provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the detection method includes: S101: Process the synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals. In this step, the synchronously acquired electrocardiogram (ECG) signals and heart sound signals are processed to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals.

[0023] It should be noted that ECG interval sequences include RR interval sequences, QT interval sequences, corrected QT interval sequences, T peak-end interval sequences, and TpeI / QTI interval sequences.

[0024] It should be noted that PCG interval sequences include systolic interval sequences (STI) and diastolic interval sequences (DTI).

[0025] Here, ECG and PCG signals are acquired simultaneously at a sampling rate of 8000Hz. A standard lead I single-lead ECG signal and a synchronous PCG signal are acquired. In one possible implementation, the processing of the synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals includes: A: The electrocardiogram (ECG) signal and the heart sound signal are filtered to determine the filtered ECG signal and the filtered heart sound signal.

[0026] Here, a second-order bandpass filter is used to filter the ECG signal from 0.05 to 75 Hz, and a third-order polynomial filter is used to remove baseline drift. A 20 Hz high-pass filter is used to filter the PCG signal, and an IIR notch filter is used to eliminate 50 Hz power frequency interference.

[0027] B: The filtered ECG signal is processed by QRS complex detection and R wave peak localization to determine multiple ECG interval sequences.

[0028] Here, the Pan-Tompkins algorithm is used to perform QRS complex detection and R wave peak localization on the filtered ECG signal to determine multiple ECG interval sequences.

[0029] The specific processing steps are as follows: 1) Bandpass filtering: The input ECG signal is bandpass filtered from 5 to 15 Hz to suppress baseline drift, power line interference, and electromyography interference. 2) Differential operation: The differential operator is applied to the filtered signal to calculate the slope characteristics of the QRS complex wave. 3) Squaring operation: The differential result is squared to amplify the amplitude of the QRS complex wave and ensure that the signal value is non-negative. 4) Sliding window integration: A sliding window integrator is used to smooth the squared signal, and the width of the integration window is set according to the average width of the QRS complex wave. 5) Adaptive threshold detection: An adaptive threshold algorithm is applied to the integrated signal to detect peaks. The threshold is adjusted in real time based on the dynamic evaluation of the signal peak and the noise peak, thereby determining the position of the R-wave peak. 6) To locate the Q-wave initiation point, this invention uses a line fitting (FL) algorithm. This algorithm applies the least squares method to the locally normalized ECG signal within a specific interval before and after the detected R-wave peak, calculating two linear polynomial fitting lines respectively; the Q-wave initiation point is determined based on the gradient of the two fitting lines and the small angle formed by them. 7) For T waves, set a time window (typically 200-400 ms) after the detected R wave peak and search for the T wave peak within the window; then, locate the position where the T wave peak falls back to the ECG signal isoelectric line as the end point of the T wave.

[0030] Here, after completing the waveform segmentation above, the various interval sequences of the ECG are calculated as follows: RR Interval Sequence (RRI): Calculate the time interval between two adjacent R-wave peaks, i.e., RRI(i) = R(i+1) - R(i), where R(i) represents the time position of the i-th R-wave peak. QT Interval Sequence (QTI): Calculate the time interval between the start of the Q-wave and the end of the T-wave, i.e., QTI(i) = T_end(i) - Q_onset(i), where Q_onset(i) is the start of the Q-wave in the i-th cardiac cycle, and T_end(i) is the corresponding end of the T-wave. Corrected QT Interval Sequence (QTcI): Use the Bazett formula to correct the heart rate of the QT interval sequence, calculated as QTcI(i) = QTI(i) / √RRI(i), where RRI(i) is in seconds. T-peak-end interval sequence (TpeI): Calculates the time interval between the peak and end of the T-wave, i.e., TpeI(i) = T_end(i) - T_peak(i), where T_peak(i) is the position of the T-wave peak in the i-th cardiac cycle. TpeI / QTI ratio sequence: Calculates the ratio of TpeI to QTI in each cardiac cycle, i.e., Ratio(i) = TpeI(i) / QTI(i), forming a ratio time series. This enables the extraction of five key interval sequences from the ECG signal: RR interval sequence (RRI), QT interval sequence (QTI), corrected QT interval sequence (QTcI), T-peak-end interval sequence (TpeI), and the TpeI / QTI ratio.

[0031] C: The filtered heart sound signal is periodically segmented using a hidden semi-Markov model combined with a dynamic programming algorithm to determine multiple PCG interval sequences.

[0032] Here, based on the principles of cardiac sound auscultation, the cardiac sound signal is divided into four states: S1, systole, S2, and diastole. A hidden semi-Markov model combined with a dynamic programming algorithm is used to periodically segment the cardiac sound signal. The algorithm first models the cardiac sound signal as a Markov process with four physiological states (S1, systole, S2, and diastole), and uses a Gaussian distribution to accurately model the duration characteristics of each state. The algorithm extracts multidimensional features from the signal (mainly including homomorphic envelopes and Hilbert envelopes) as the observation sequence, and then combines a logistic regression model with Gaussian correction to calculate the observation probability. Finally, by integrating state transition constraints, observation probabilities, and duration probabilities, the improved Viterbi algorithm can calculate the most probable state sequence, achieving accurate localization and segmentation of key cardiac sound events such as S1 and S2.

[0033] The process for determining multiple PCG interval sequences is as follows: 1) For each state (S1, systole, S2, and diastole), calculate its mean μ and variance. For example, the duration of state S1 has a mean of mean_s1 and a variance of std_s1. The probability P of the state duration is described using the following Gaussian distribution formula:

[0034] Based on actual heart sound signal data and physiological patterns, a reasonable duration range (minimum and maximum value) is set for each state. For example, the maximum / minimum duration of S1 is the mean ± 3 times the standard deviation. This ensures that the duration of each state conforms to physiological patterns and is reasonably constrained within the model. 2) The hidden semi-Markov model combines the duration probability of each state with signal features for computation. Signal features (Hilbert envelope, power spectral density) are used as observations input to the hidden semi-Markov model. A logistic regression model is used to calculate the observation probability of each state, and these probabilities are corrected using a Gaussian distribution to ensure that the features conform to a Gaussian distribution.

[0035] in, Let be the observation value at time t. This refers to a specific heart sound state. The observation probability output by the logistic regression model represents the probability that the current observation belongs to a certain state. This is the Gaussian correction term. Let be the initial state probability. In order to be in The probability of observation.

[0036] 3) Finally, the improved Viterbi algorithm is used to decode the optimal state sequence of the heart sound signal, and dynamic programming is performed by combining the state duration and observation probability. First, the initial probability of each state at time 0 is calculated. Combining the initial state probability π and the observation probability, the optimal probability of each state is calculated for each subsequent time point, and the backtracking matrix is ​​updated. Finally, the optimal state sequence is gradually recovered from the last time point using the backtracking matrix. This yields the start and end times of S1, systolic interval, S2, and diastolic interval in each cardiac cycle. After heart sound segmentation, the heart sound time intervals are calculated as follows: Systolic Interval Sequence (STI): The time interval between the end of S1 and the start of S2 is calculated, i.e., STI(i) = S2_onset(i) - S1_end(i), where S1_end(i) is the end time of S1 in the i-th cardiac cycle, and S2_onset(i) is the corresponding start time of S2. Diastolic Interval Sequence (DTI): Calculate the time interval between the end of S2 and the start of the next cardiac cycle S1, i.e., DTI(i) = S1_onset(i+1) - S2_end(i), where S2_end(i) is the end time of S2 in the i-th cardiac cycle, and S1_onset(i+1) is the start time of S1 in the next cardiac cycle.

[0037] S102: Determine multidimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences; wherein, the multidimensional coupling features include at least one of cross-entropy features, multivariate entropy features, and non-entropy features.

[0038] In this step, multidimensional coupling characteristics are determined based on the coupling relationship between multiple ECG interval sequences and multiple PCG interval sequences.

[0039] In one possible implementation, regarding the cross-entropy feature, the determination of multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences includes: (1): Combine any ECG interval sequence with any PCG interval sequence to determine multiple combined interval sequences.

[0040] (2): Based on the cross-sample entropy calculation formula, cross-fuzzy entropy calculation formula, joint distribution entropy calculation formula, high-resolution joint symbolic dynamic entropy calculation formula and symmetric KL distance formula, feature extraction is performed on multiple combined interval sequences to determine multiple cross-entropy features; wherein, the cross-entropy features include at least one of cross-sample entropy features, cross-fuzzy entropy features, joint distribution entropy features, high-resolution joint symbolic dynamic entropy features and DKL features. Here, the formula for calculating cross-sample entropy is:

[0041] Wherein, the embedding dimension m=2, the threshold parameter r=0.2×SD (SD is the standard deviation of the time series), the lag coefficient τ=3, N is the sequence length, A() is the Heaviside function (i.e., if x ≥ 0, then A()= 1, otherwise A()= 0), and d is the maximum norm of the distance between sequences. By combining any interval sequence of ECG (RRI, QTI, QTcI, TpeI, TpeI / QTI) with any interval sequence of PCG (STI, DTI), multiple cross-sample entropy features are extracted, where i is the i-th ECG interval sequence and j is the j-th PCG interval sequence.

[0042] It should be noted that the cross-fuzzy calculation formula is an improved formula for cross-sample entropy. It is implemented by replacing the Heaviside function with a Gaussian function and using the same sequence combination method as cross-sample entropy to extract multiple cross-fuzzy entropy features.

[0043] Here, the formula for calculating the joint distribution entropy is:

[0044] Specifically, the distance matrix is ​​measured globally using a combination of distance matrix and distribution entropy algorithms, avoiding parameter dependencies caused by local analysis. An embedding dimension of m=2 is chosen, and the number of predefined histograms M is obtained using the Doane formula. The probability of an element falling into the t-th histogram is denoted as... (1≤t≤M).

[0045] Here, the formula for calculating the high-resolution joint symbol dynamic entropy is: ,

[0046] This method is used to quantify the interaction between two physiological signals. The basic principle is to analyze the coupling between sequences based on symbolic analysis of binary dynamic processes. The binary interphase sequence is converted into a binary symbol vector. Where n is the sequence length and r = 0.25 × SD. Then, the symbol sequence is converted into a short word symbol sequence of length 3, and based on the word distribution density matrix... = 0.25 Estimate the normalized joint probability of each word This allows for the use of the high-resolution joint symbolic dynamic entropy calculation formula to obtain multiple high-resolution joint conforming dynamic entropy characteristics.

[0047] Here, the symmetric KL distance formula is used to quantify the difference between the two probability distributions of the interval sequence after quantification.

[0048] In this application, by calculating the above five cross-entropy features, the mutual similarity and synchronicity between ECG and PCG interval sequences can be quantified from different perspectives, providing rich coupling information for subsequent coronary heart disease severity grading.

[0049] In one possible implementation, for the multivariate entropy feature, determining the multidimensional coupling feature based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences includes: a: Based on the set reconstruction dimension, multiple ECG interval sequences and multiple PCG interval sequences are embedded and reconstructed to determine multiple reconstructed sequences.

[0050] Here, we first analyze the Δ channel time series. (1≤p≤Δ, 1≤i≤N) Perform multidimensional reconstruction and select the reconstruction dimension. (1≤p≤Δ, m=2), construct the reconstructed sequence .

[0051] b: Determine the distance matrix density distribution between the reconstructed sequences.

[0052] Here, the distance matrix density distribution between the reconstructed sequences is determined using the following formula:

[0053] in, , i = 1,2,...,Nn d is the maximum norm of the distance between the reconstructed sequences, and A() is the Heaviside function. For the i-th reconstructed ECG interval sequence under the current reconstruction dimension, For the j-th reconstructed PCG interval sequence under the current reconstruction dimension, Let r be the distance matrix density distribution between reconstructed sequences under the current reconstructed dimension m, where r = 0.2 × SD, and SD is the standard deviation of the time series.

[0054] c: Based on the distance matrix density distribution between reconstructed sequences under the current reconstruction dimension and the distance matrix density distribution between reconstructed sequences under the next reconstruction dimension, determine the multivariate sample entropy feature or / and multivariate fuzzy entropy feature in the multivariate entropy feature.

[0055] Here, the multivariate entropy is determined using the following formula:

[0056] It should be noted that the multivariate sample entropy features of the combined interphase sequences are extracted using multi-sample entropy. Multivariate fuzzy entropy is to introduce the ability to process multi-channel data based on fuzzy entropy. Similar to multivariate sample entropy, multivariate fuzzy entropy can be obtained by replacing the Heaviside function in multivariate sample entropy with a Gaussian function. Using the same combination method, multiple variable fuzzy entropy features are extracted.

[0057] In a possible implementation manner, for the non-entropy features, the multi-dimensional coupling features determined based on the coupling relationship between the multiple ECG interphase sequences and the multiple PCG interphase sequences include: Combining any one of the ECG interphase sequences with any one of the PCG interphase sequences to determine multiple combined interphase sequences; extracting features from the multiple combined interphase sequences based on the mutual information function, amplitude squared coherence function, cross-power spectral density, and coherent cross-power product to determine non-entropy features; where the non-entropy features include mutual information features, frequency domain features, and time domain features.

[0058] Here, the mutual information function is mutual information(X,Y)=H(X)+H(Y)-H(X,Y). For two sequences x(i) and y(i) (1 < i < N), and respectively represent the marginal distributions and of the entropy, represents the joint distribution of the entropy. x(i) is the i-th feature in the ECG interphase sequence, y(i) is the i-th feature in the PCG interphase sequence, and N is the sequence length. After determining the mutual information, feature extraction is performed to obtain mutual information features.

[0059] Here, the amplitude squared coherence function can identify the frequency domain correlation between two time series. For two sequences and (1 < i < N), the amplitude squared coherence function is defined as:

[0060] where, f is the frequency, N is the sequence length, and are respectively and of the power spectral density estimates, is and of the cross-power spectral density estimate. Using the same combination method, the mean and standard deviation of the amplitude squared coherence function are extracted, and multiple frequency domain features are extracted from multiple combined interphase sequences in total.

[0061] Here, the similarity between combined power estimation sequences is used. For two sequences x(i) and y(i) (1 < i < N), it is defined as:

[0062] where denotes the conjugate complex number, X and Y are respectively and the Fourier transforms of, denotes X the phase difference between and Y, and respectively denote the amplitudes of X and Y. Using the same combination method, the real part and imaginary part of the cross-power spectral density j are extracted for their means and standard deviations, and multiple time-domain features are extracted for multiple combined interphase sequences.

[0063] It should be noted that the coherent cross-power product: when both signals oscillate with large amplitudes, coupling can be measured through cross-power; when the two oscillations are synchronized with each other (i.e., they maintain a constant phase relationship), this effect can be evaluated through coherence. Therefore, the coherent cross-power product is used to measure both effects simultaneously. The calculation formula is: coherent cross-power product = coherence between two signals × cross-power.

[0064] In this application, through the calculation of the above non-entropy features, the correlation between ECG and PCG interphase sequences is comprehensively quantified from both time-domain and frequency-domain perspectives, providing a more comprehensive feature description for the cardiac electro-mechanical coupling relationship.

[0065] S103: Perform feature selection on the multi-dimensional coupling features, screen out the target features with significant discrimination ability for classifying the severity of coronary heart disease, input the target features into the coronary heart disease severity classification model for coronary heart disease severity classification detection, and output the coronary heart disease severity level.

[0066] In this step, feature selection is performed on the multi-dimensional coupling features, the target features with significant discrimination ability for classifying the severity of coronary heart disease are screened out, the target features are input into the coronary heart disease severity classification model for coronary heart disease severity classification detection, and the coronary heart disease severity level is output. It should be noted that the coronary heart disease severity levels can include severe level one, severe level two, severe level three, and healthy, and the higher the level, the more severe the coronary heart disease.

[0067] In this application, a feature extraction strategy based on physiological mechanisms enhances the interpretability and clinical application value of the method. Unlike existing technologies that mostly employ black-box deep learning models, the time interval sequences extracted in this application have clear physiological significance, and their coupling relationships can directly reflect changes in the cardiac electromechanical coupling mechanism under pathological conditions. Analysis of the optimal sequence combination revealed that the QTI-DTI combination performed best among multiple groups. This finding is highly consistent with cardiovascular physiological theory: the QT interval reflects the total time of ventricular depolarization and repolarization, while the diastolic interval reflects the cardiac mechanical relaxation process; the coupling relationship between the two can sensitively reflect changes in the functional state of the cardiovascular system. This physiological mechanism-based analysis method not only improves diagnostic accuracy but also provides clinicians with understandable and reliable diagnostic evidence.

[0068] In one possible implementation, the feature selection of the multi-dimensional coupled features to screen out target features with significant discriminative ability for the severity grading of coronary heart disease includes: i: Perform a normality test on the distribution of the multi-dimensional coupling features to determine the normality test result.

[0069] Here, the results of the normality test include characteristics that conform to a normal distribution as well as characteristics that do not conform to a normal distribution.

[0070] ii: Based on the normality test results, a corresponding significance test method is determined. Based on the significance test method, the feature values ​​between different coronary heart disease severity groups are compared, and multiple candidate features with discriminative ability are screened out.

[0071] It should be noted that if the feature follows a normal distribution, a parametric test (Student's t-test) is used; if it does not follow a normal distribution, a non-parametric test (Mann-Whitney U test) is used.

[0072] It should be noted that the test criteria are: p<0.05: statistically significant; p<0.01: highly statistically significant.

[0073] Here, the differences in eigenvalues ​​between different coronary heart disease grade groups are compared, and features with significant differences between groups are selected based on p-values; features with p < 0.05 are retained as a preliminary set of candidate features with discriminative ability.

[0074] iii: Based on the recursive feature elimination algorithm, the importance score is calculated for multiple candidate features. In each iteration of the score calculation, the candidate feature with the smallest contribution rate is eliminated. The process is repeated recursively until the end of the loop. The remaining candidate features are then used as target features that have significant distinguishing ability for the severity grading of coronary heart disease.

[0075] Here, a Support Vector Machine-based Recursive Feature Elimination (SVM-RFE) algorithm is used for feature selection. This algorithm is based on the maximum margin principle of SVM, using a prediction model to train the original features and obtain an importance score ranking for each feature. In each iteration, the feature with the lowest contribution is eliminated, and this process is recursively repeated until a satisfactory feature set is selected. Each iteration re-scores the importance of each feature, finding and removing the least important feature, recording the remaining features in a feature list, and finally selecting the target feature from the feature list after the loop ends.

[0076] In one possible implementation, the coronary heart disease severity classification model is determined through the following steps: inputting sample features of sample electrocardiogram signals and sample heart sound signals into a vector machine classification model to perform coronary heart disease severity classification detection, and outputting the severity level of the sample coronary heart disease; iteratively training the vector machine classification model based on the loss value between the severity level of the sample coronary heart disease and the actual severity level of the sample features, and determining the classification model.

[0077] Here, Support Vector Machine (SVM) is used for four-class classification. SVM has the advantages of strong generalization ability and low computational burden in solving small-sample, nonlinear, and high-dimensional pattern recognition problems. Linear kernels and radial basis function (RBF) kernels are used respectively. Linear kernels are advantageous due to their ability to handle high-dimensional data, model simplicity, and fast computation speed; RBF kernels perform better in handling nonlinearly separable data, have better robustness, and are less sensitive to outliers. The penalty parameter C in the kernel function is used to control model overfitting, and γ is used to control the degree of nonlinearity of the model. Nested five-fold cross-validation is used for optimal hyperparameter selection and model performance evaluation. In the inner loop of the nested five-fold cross-validation, a grid search algorithm is used to find the optimal hyperparameters C and γ, and these selected optimal hyperparameters are used in the outer loop to evaluate the model's performance on the test set.

[0078] In this application, the physiological mechanisms of cardiac electromechanical coupling are deeply explored through the extraction of multi-dimensional coupling features, significantly improving the accuracy and reliability of detection. Unlike existing technologies that primarily employ simple feature-level or decision-level fusion, this invention performs deep coupling analysis at the signal level, comprehensively characterizing the complex coupling relationship between ECG and PCG signals through three dimensions: cross-entropy features, multivariate entropy features, and non-entropy features. In particular, multivariate entropy features can simultaneously consider the mutual similarity between sequences and the complexity of each sequence itself.

[0079] This application provides a detection method for classifying the severity of coronary heart disease (CHD). The method includes: processing synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals; determining multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences; wherein the multi-dimensional coupling features include at least one of cross-entropy features, multivariate entropy features, and non-entropy features; performing feature selection on the multi-dimensional coupling features to screen out target features with significant discriminative ability for CHD severity classification; inputting the target features into a CHD severity classification model for CHD severity classification detection; and outputting the CHD severity level. By employing time-series coupling analysis to mine the deep coupling relationship between ECG and PCG signal intervals and utilizing multi-dimensional coupling features for fine-grained classification of CHD severity, the accuracy of severity identification is improved.

[0080] Please see Figure 2 , Figure 3 , Figure 2 One of the structural schematic diagrams of a detection device for classifying the severity of coronary heart disease provided in an embodiment of this application; Figure 3 This is a second schematic diagram of a detection device for classifying the severity of coronary heart disease, provided in an embodiment of this application. Figure 2 As shown, the detection device 200 includes: The signal processing module 210 is used to process the synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals. The coupling feature determination module 220 is used to determine multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences; wherein, the multi-dimensional coupling features include at least one of cross-entropy features, multivariate entropy features, and non-entropy features; The classification module 230 is used to perform feature selection on the multi-dimensional coupled features, filter out target features that have significant distinguishing ability for the severity grading of coronary heart disease, input the target features into the coronary heart disease severity classification model for coronary heart disease severity classification detection, and output the severity level of coronary heart disease.

[0081] Furthermore, the signal processing module 210 is used to process the synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals, including: The electrocardiogram (ECG) signal and the heart sound signal are filtered to determine the filtered ECG signal and the filtered heart sound signal; The filtered electrocardiogram signal is subjected to QRS complex detection processing and R wave peak localization processing to determine multiple ECG interval sequences; The filtered heart sound signal was periodically segmented using a hidden semi-Markov model combined with a dynamic programming algorithm to determine multiple PCG interval sequences.

[0082] Furthermore, the coupling feature determination module 220 is used to determine multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences, specifically for the cross-entropy features. By combining any of the ECG interval sequences with any of the PCG interval sequences, multiple combined interval sequences can be determined. Based on the cross-sample entropy calculation formula, cross-fuzzy entropy calculation formula, joint distribution entropy calculation formula, high-resolution joint symbolic dynamic entropy calculation formula, and symmetric KL distance formula, feature extraction is performed on multiple combined interval sequences to determine multiple cross-entropy features; wherein, the cross-entropy features include at least one of cross-sample entropy features, cross-fuzzy entropy features, joint distribution entropy features, high-resolution joint symbolic dynamic entropy features, and DKL features.

[0083] Furthermore, the coupling feature determination module 220 is used to determine multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences for the multivariate entropy features: Based on the set reconstruction dimension, multiple ECG interval sequences and multiple PCG interval sequences are embedded and reconstructed to determine multiple reconstructed sequences; Determine the distance matrix density distribution between the reconstructed sequences; Based on the distance matrix density distribution between reconstructed sequences under the current reconstruction dimension and the distance matrix density distribution between reconstructed sequences under the next reconstruction dimension, the multivariate sample entropy feature and / or multivariate fuzzy entropy feature in the multivariate entropy feature are determined.

[0084] Furthermore, the coupling feature determination module 220 is used to determine multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences for the non-entropy features: By combining any of the ECG interval sequences with any of the PCG interval sequences, multiple combined interval sequences can be determined. Feature extraction is performed on multiple combined interval sequences based on mutual information function, amplitude squared coherence function, cross power spectral density, and coherent cross power product to determine non-entropy features; wherein, the non-entropy features include mutual information features, frequency domain features, and time domain features.

[0085] Furthermore, the classification module 230 is used to perform feature selection on the multi-dimensional coupled features, filtering out target features that have significant discriminative ability in classifying the severity of coronary heart disease: The distribution normality test is performed on the multi-dimensional coupling features to determine the normality test results; Based on the normality test results, a corresponding significance test method is determined. Based on the significance test method, the feature values ​​between different coronary heart disease severity groups are compared, and multiple candidate features with discriminative ability are screened out. The importance score of multiple candidate features is calculated based on the recursive feature elimination algorithm. In each iteration of the score calculation, the candidate feature with the smallest contribution rate is eliminated. The process is repeated recursively until the end of the loop. The remaining candidate features are then used as target features that have significant distinguishing ability for the severity grading of coronary heart disease.

[0086] Furthermore, such as Figure 3 The detection device 200 shown also includes a model training module 240, which determines the coronary heart disease severity classification model through the following steps: The sample features of the electrocardiogram signal and the heart sound signal are input into the vector machine classification model to classify and detect the severity of coronary heart disease, and the severity level of coronary heart disease of the sample is output. The vector machine classification model is iteratively trained based on the loss value between the severity level of coronary heart disease in the sample and the actual severity level of coronary heart disease in the sample features, thereby determining the classification model.

[0087] This application provides a detection device for classifying the severity of coronary heart disease (CHD). The device includes: a signal processing module for processing synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals; a coupling feature determination module for determining multi-dimensional coupling features based on the coupling relationship between the multiple ECG interval sequences and the multiple PCG interval sequences; wherein the multi-dimensional coupling features include at least one of cross-entropy features, multivariate entropy features, and non-entropy features; and a classification module for feature selection of the multi-dimensional coupling features, screening out target features with significant discriminative ability for CHD severity classification, inputting the target features into a CHD severity classification model for CHD severity classification detection, and outputting the CHD severity level. By employing time-series coupling analysis to mine the deep coupling relationship between ECG and PCG signal intervals and utilizing multi-dimensional coupling features for fine classification of CHD severity, the accuracy of severity identification is improved.

[0088] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.

[0089] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, they can perform the operations described above. Figure 1 The steps of the detection method for classifying the severity of coronary heart disease in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0090] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The steps of the detection method for classifying the severity of coronary heart disease in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0091] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0092] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0093] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0095] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A detection method for classifying the severity of coronary heart disease, characterized in that, The detection method includes: The synchronously acquired electrocardiogram (ECG) signals and heart sound signals are processed to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals. A multi-dimensional coupling feature is determined based on the coupling relationship between multiple ECG interval sequences and multiple PCG interval sequences; wherein, the multi-dimensional coupling feature includes at least one of cross-entropy feature, multivariate entropy feature and non-entropy feature; Feature selection is performed on the multi-dimensional coupled features to screen out target features that have significant distinguishing ability for the severity grading of coronary heart disease. The target features are then input into the coronary heart disease severity classification model for coronary heart disease severity classification detection, and the severity level of coronary heart disease is output.

2. The detection method according to claim 1, characterized in that, The process of processing the synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals includes: The electrocardiogram (ECG) signal and the heart sound signal are filtered to determine the filtered ECG signal and the filtered heart sound signal; The filtered electrocardiogram signal is subjected to QRS complex detection processing and R wave peak localization processing to determine multiple ECG interval sequences; The filtered heart sound signal was periodically segmented using a hidden semi-Markov model combined with a dynamic programming algorithm to determine multiple PCG interval sequences.

3. The detection method according to claim 1, characterized in that, Regarding the cross-entropy feature, the multi-dimensional coupling feature determined based on the coupling relationship between multiple ECG interval sequences and multiple PCG interval sequences includes: By combining any of the ECG interval sequences with any of the PCG interval sequences, multiple combined interval sequences can be determined. Based on the cross-sample entropy calculation formula, cross-fuzzy entropy calculation formula, joint distribution entropy calculation formula, high-resolution joint symbolic dynamic entropy calculation formula, and symmetric KL distance formula, feature extraction is performed on multiple combined interval sequences to determine multiple cross-entropy features; wherein, the cross-entropy features include at least one of cross-sample entropy features, cross-fuzzy entropy features, joint distribution entropy features, high-resolution joint symbolic dynamic entropy features, and DKL features.

4. The detection method according to claim 1, characterized in that, Regarding the multivariate entropy features, the multidimensional coupling features determined based on the coupling relationship between multiple ECG interval sequences and multiple PCG interval sequences include: Based on the set reconstruction dimension, multiple ECG interval sequences and multiple PCG interval sequences are embedded and reconstructed to determine multiple reconstructed sequences; Determine the distance matrix density distribution between the reconstructed sequences; Based on the distance matrix density distribution between reconstructed sequences under the current reconstruction dimension and the distance matrix density distribution between reconstructed sequences under the next reconstruction dimension, the multivariate sample entropy feature and / or multivariate fuzzy entropy feature in the multivariate entropy feature are determined.

5. The detection method according to claim 1, characterized in that, Regarding the non-entropy features, the multi-dimensional coupling features determined based on the coupling relationship between multiple ECG interval sequences and multiple PCG interval sequences include: By combining any of the ECG interval sequences with any of the PCG interval sequences, multiple combined interval sequences can be determined. Feature extraction is performed on multiple combined interval sequences based on mutual information function, amplitude squared coherence function, cross power spectral density, and coherent cross power product to determine non-entropy features; wherein, the non-entropy features include mutual information features, frequency domain features, and time domain features.

6. The detection method according to claim 1, characterized in that, The feature selection process for the multi-dimensional coupled features, filtering out target features with significant discriminative ability for classifying the severity of coronary heart disease, includes: The distribution normality test is performed on the multi-dimensional coupling features to determine the normality test results; Based on the normality test results, a corresponding significance test method is determined. Based on the significance test method, the feature values ​​between different coronary heart disease severity groups are compared, and multiple candidate features with discriminative ability are screened out. The importance score of multiple candidate features is calculated based on the recursive feature elimination algorithm. In each iteration of the score calculation, the candidate feature with the smallest contribution rate is eliminated. The process is repeated recursively until the end of the loop. The remaining candidate features are then used as target features that have significant distinguishing ability for the severity grading of coronary heart disease.

7. The detection method according to claim 1, characterized in that, The severity classification model for coronary artery disease is determined through the following steps: The sample features of the electrocardiogram signal and the heart sound signal are input into the vector machine classification model to classify and detect the severity of coronary heart disease, and the severity level of coronary heart disease of the sample is output. The vector machine classification model is iteratively trained based on the loss value between the severity level of coronary heart disease in the sample and the actual severity level of coronary heart disease in the sample features, thereby determining the classification model.

8. A detection device for classifying the severity of coronary heart disease, characterized in that, The detection device includes: The signal processing module is used to process the synchronously acquired electrocardiogram (ECG) signals and heart sound signals to determine multiple ECG interval sequences of the ECG signals and multiple PCG interval sequences of the heart sound signals. The coupling feature determination module is used to determine multi-dimensional coupling features based on the coupling relationship between multiple ECG interval sequences and multiple PCG interval sequences; wherein, the multi-dimensional coupling features include at least one of cross-entropy features, multivariate entropy features, and non-entropy features; The classification module is used to select features from the multi-dimensional coupled features, filter out target features that have significant distinguishing ability for the severity grading of coronary heart disease, input the target features into the coronary heart disease severity classification model for coronary heart disease severity classification detection, and output the severity level of coronary heart disease.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. The machine-readable instructions are executed by the processor to perform the steps of the detection method for classifying the severity of coronary heart disease as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the detection method for classifying the severity of coronary heart disease as described in any one of claims 1 to 7.

Images