Cardiac and cerebral coupling feature extraction method and system based on variational mode decomposition

Through variational modal decomposition and phase-amplitude cross-frequency coupling methods, the heart-brain coupling characteristics are extracted, and the problems of large scale differences and insufficient adversarial ability in nonlinear and non-stationary signal analysis by traditional methods are solved, and the sleep state is improved The accuracy of heart-brain interaction dynamic analysis and sleep disorder diagnosis.

CN118924302APending Publication Date: 2024-11-12BEIJING INST OF TECH

Patent Information

Application Number
CN202410992402.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Traditional heart-brain coupling methods have problems such as large scale differences and poor ability to combat nonlinear non-stationary signals when analyzing nonlinear and non-stationary signals, resulting in dynamic misjudgment of heart-brain interactions in sleep states, reducing the accuracy of diagnosis of sleep disorder-related diseases. Rate.

Method used

Using a method based on variational modal decomposition, the time series of R-R intervals of heartbeat intervals is extracted by pre-processing of electrocardiogram signals and EEG signals, and the eigenmodal modal decomposition algorithm is used to obtain the eigenmodal modality of different frequency components. Function, perform Hilbert transformation, calculate the phase-amplitude modulation index, construct a heart-to-brain coupling map, and extract the maximum modulation index in the region of interest as a heart-to-brain coupling feature.

Benefits of technology

The ability to combat nonlinear and non-stationary has been improved, a more accurate heart-to-brain coupling map has been built, and the accuracy of sleep structure evaluation and sleep disorder diagnosis has been improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118924302A_ABST
    Figure CN118924302A_ABST
Patent Text Reader

Abstract

The invention discloses a heart and brain coupling feature extraction method and system based on variational mode decomposition. The method comprises the following steps that collected original single-lead electrocardiosignals and electroencephalogram signals are preprocessed; extracting a heartbeat interval R-R interval time sequence from the preprocessed electrocardiosignals, and interpolating to enable the number of sample points of the R-R interval time sequence to be the same as that of the electroencephalogram signals and keep time synchronization with the electroencephalogram signals; processing the electroencephalogram signal time sequence and the R-R interval time sequence by using a variational mode decomposition algorithm to obtain intrinsic mode functions with different frequency components of the two sequences; performing Hilbert transform on the intrinsic mode function, extracting amplitude information of the electroencephalogram intrinsic mode function and phase information of the R-R interval intrinsic mode function, calculating phase-amplitude modulation indexes in pairs, and constructing a heart-brain coupling map during sleep; and extracting the maximum modulation index in the region of interest as a heart-brain coupling feature based on the sleep heart-brain coupling map.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram signal processing, and in particular to a method and system for extracting heart-brain coupling features based on variational mode decomposition. Background Art

[0002] At present, scholars have proposed a variety of heart-brain joint analysis methods based on ECG and EEG signals to study the interaction between the central nervous system and the autonomic nervous system during sleep. However, these methods simply analyze the correlation between ECG features and EEG features, and cannot accurately express the causal relationship between the two interactions.

[0003] Therefore, in recent years, the research on the heart-brain coupling network based on ECG and EEG signals has received widespread attention. Recent studies have used Granger causality, transfer entropy and other measurement methods to accurately measure the strength of heart-brain coupling by fusing different frequency components of EEG and ECG signals, thereby further revealing the sleep structure and evaluating and detecting sleep disorders.

[0004] However, these methods have defects when analyzing nonlinear and non-stationary signals. The nonlinear and non-stationary characteristics of the signals will cause false components in the heart-brain coupling measurement results, leading to misjudgment of the dynamics of heart-brain interaction during sleep, thereby reducing the diagnostic accuracy of sleep disorder-related diseases.

[0005] It can be seen that traditional heart-brain coupling has the problems of large scale differences and poor ability to combat nonlinear and non-stationary problems, which are problems that have not been solved by existing technologies. Summary of the invention

[0006] In view of this, the present invention provides a heart-brain coupling feature extraction method and system based on variational mode decomposition, which can perform high-dynamic heart-brain coupling feature extraction based on variational mode decomposition and phase-amplitude cross-frequency coupling, solving the problems of large electrical scale differences and poor ability to combat nonlinear and non-stationarity in traditional heart-brain coupling.

[0007] To achieve the above object, the present invention provides a method for extracting heart-brain coupling features based on variational mode decomposition, which specifically includes the following steps:

[0008] Step 1) pre-process the collected original single-lead ECG signal and EEG signal to avoid the influence of power frequency noise and irrelevant frequency components on the results.

[0009] Step 2) extracting the heart beat interval RR interval time series from the preprocessed ECG signal, and using cubic spline interpolation to make the number of sample points of the RR interval time series the same as the EEG signal, and to keep it in time synchronization with the EEG signal.

[0010] Step 3) The variational mode decomposition algorithm is used to process the EEG signal time series and the RR interval time series respectively to obtain the intrinsic mode functions of the above two time series with different frequency components.

[0011] Step 4) Perform Hilbert transform on the IMFs of different frequencies of the two time series, extract the amplitude information of the EEG IMF and the phase information of the RR interval IMF, calculate the phase-amplitude modulation index pairwise, and construct a heart-brain coupling map during sleep.

[0012] Step 5) Based on the sleep heart-brain coupling map, the maximum modulation index in the region of interest is extracted as the heart-brain coupling feature.

[0013] Furthermore, the acquisition frequencies of the original single-lead ECG and EEG signals are set to f s1 , f s2 , stored in the computer.

[0014] Furthermore, the collected original single-lead ECG and EEG signals are preprocessed to avoid the influence of power frequency noise and irrelevant frequency components on the results. The specific preprocessing method is as follows:

[0015] For EEG signals: use a bandpass filter to retain the set frequency band and remove other frequency bands, thereby eliminating the influence of power frequency noise and other artifacts on the results, and obtain the filtered EEG data; use independent component analysis ICA to decompose the filtered EEG data, remove the interference caused by eye movement, and obtain the preprocessed EEG signal.

[0016] For ECG signal: Use a bandpass filter on the entire ECG signal, retain the set frequency band, remove other frequency bands, avoid the influence of power frequency noise and irrelevant frequency components on the result, and obtain the preprocessed ECG signal.

[0017] Further, step 2) extracts the heart beat interval RR interval time series from the preprocessed ECG signal, and uses cubic spline interpolation to make the number of sample points of the RR interval time series the same as the EEG signal, and keeps it in time synchronization with the EEG signal, specifically:

[0018] The RR time series of heartbeat intervals is extracted from the preprocessed ECG signal, that is, the heartbeats are marked in the single-lead ECG signal, and the time interval between two adjacent heartbeats is calculated to obtain the RR time series.

[0019] After obtaining the RR interval time series, a sliding average filter with a window of 21 data points was used to remove outliers due to erroneous heartbeat detection. When the center point in the window was outside 30% of the mean, it was removed.

[0020] Afterwards, cubic spline interpolation is used to make the number of sample points of the processed RR interval time series consistent with the EEG signal and to keep it synchronized with the EEG signal in time.

[0021] Finally, the preprocessed EEG signals and RR interval time series are divided into multiple segments with a set length.

[0022] Furthermore, in step 3), the variational mode decomposition algorithm is used to process the EEG signal time series and the RR interval time series respectively to obtain the intrinsic mode functions of the above two time series with different frequency components, specifically:

[0023] For the EEG signals and RR interval time series that are synchronized in time, the variational mode decomposition is used to decompose the EEG signals and RR interval time series respectively to obtain the intrinsic mode functions of the two time series, as follows:

[0024]

[0025] where {u k}:={u 1 ,...,u K} represents the K finite bandwidth intrinsic mode functions obtained by decomposition, {ω k}:={ω 1 ,...,ω K} is the center frequency of K eigenmode functions, is the partial derivative with respect to time t, δ(t) is the impulse function, t is the time, * is the convolution symbol, j is the imaginary part symbol, and f(t) is the function to be decomposed, that is, the EEG signal segment or the RR interval time series segment.

[0026] Introducing the Lagrangian multiplication operator λ(t) and the quadratic penalty parameter α, we solve the constrained variation problem of formula (1); the augmented Lagrangian function is obtained as

[0027]

[0028] The alternating multiplier method is used to obtain the saddle point of formula (2), and u is iteratively updated in the frequency domain. k ,ω k and λ.

[0029] The EEG signal fragment and the RR interval time series fragment are respectively used as the function f(t) to be decomposed. The intrinsic mode functions with different frequency components of the above two time series are obtained by the above decomposition method. For specific EEG signal and RR interval time series fragments, the K 1 and K 2 The intrinsic mode function u i ,u j .

[0030] Furthermore, in step 4), the eigenmode functions of different frequencies of the two time series are subjected to Hilbert transform to extract the amplitude information of the EEG eigenmode function and the phase information of the RR interval eigenmode function, specifically using the following steps:

[0031] In step 3), for a specific EEG signal and RR interval time series fragment, each K 1 and K 2 The intrinsic mode function u i ,u j , and arranged in descending order according to the center frequency of the eigenmode function;

[0032] The Hilbert transform is used to process the intrinsic mode functions of the two time series and extract the phase information of the RR interval intrinsic mode functions. and the amplitude information of the intrinsic mode function of the EEG signal

[0033] Calculate the phase-amplitude modulation index two by two, specifically: Combine the phase information of the RR interval And the amplitude information A of the EEG signal i (t), calculate the phase-amplitude modulation index MI; divide the phase domain (0,2π) of the phase component into N intervals, calculate the average high-frequency amplitude on each low-frequency phase unit of each sample, and normalize the average amplitude to generate a distribution P, which represents the proportion of the high-frequency amplitude on each phase unit; calculate the KL distance between the phase amplitude distribution P and the uniform distribution U; finally, MI is defined as the ratio of the KL distance to logN.

[0034] Furthermore, after step 5), the method further includes the following steps: constructing a feature classifier, wherein the feature classifier takes the heart-brain coupling feature value as input and takes the wakefulness period, non-rapid eye movement sleep period and rapid eye movement sleep period as output;

[0035] The original single-lead ECG and EEG signals corresponding to the wakefulness period, non-rapid eye movement sleep period and rapid eye movement sleep period are respectively used as the acquisition signals, and steps 1) to 5) are executed to obtain the corresponding heart-brain coupling feature values, thereby constructing training samples and training the feature classifier.

[0036] The trained feature classifier is used to predict and identify wakefulness, non-REM sleep, and REM sleep.

[0037] Another embodiment of the present invention further provides a heart-brain coupling feature extraction system based on variational mode decomposition, comprising a preprocessing module, an RR interval time series extraction module, a variational mode decomposition module, an amplitude-phase information extraction module, and a heart-brain coupling feature extraction module;

[0038] The preprocessing module is used to preprocess the collected original single-lead ECG and EEG signals to avoid the influence of power frequency noise and irrelevant frequency components on the results;

[0039] The RR interval time series extraction module is used to extract the RR interval time series of heart beat intervals from the preprocessed ECG signal, and use cubic spline interpolation to make the number of sample points of the RR interval time series the same as that of the EEG signal, and keep it in time synchronization with the EEG signal;

[0040] The variational mode decomposition module uses the variational mode decomposition algorithm to process the EEG signal time series and the RR interval time series respectively, and obtains the intrinsic mode functions of the above two time series with different frequency components;

[0041] The amplitude-phase information extraction module is used to perform Hilbert transform on the intrinsic mode functions of different frequencies of the two time series, extract the amplitude information of the EEG intrinsic mode function and the phase information of the RR interval intrinsic mode function, and calculate the phase-amplitude modulation index two by two to construct the heart-brain coupling map during sleep;

[0042] The heart-brain coupling feature extraction module is used to extract the maximum modulation index in the region of interest as the heart-brain coupling feature based on the sleep heart-brain coupling map.

[0043] Furthermore, the preprocessing module specifically adopts the following preprocessing method:

[0044] For EEG signals: use a bandpass filter to retain the set frequency band and remove other frequency bands, thereby eliminating the influence of power frequency noise and other artifacts on the results, and obtain the filtered EEG data; use independent component analysis ICA to decompose the filtered EEG data, remove the interference caused by eye movement, and obtain the preprocessed EEG signal;

[0045] For ECG signal: use a bandpass filter on the entire ECG signal, retain the set frequency band, remove other frequency bands, avoid the influence of power frequency noise and irrelevant frequency components on the result, and obtain the preprocessed ECG signal;

[0046] The RR interval time series extraction module specifically performs the following steps:

[0047] Extract the heartbeat interval RR time series from the preprocessed ECG signal, that is, mark the heartbeats in the single-lead ECG signal, calculate the time interval between two adjacent heartbeats, and obtain the RR interval time series;

[0048] After obtaining the RR interval time series, a sliding average filter with a window of 21 data points was used to remove outliers due to erroneous heartbeat detection. When the center point in the window was outside 30% of the mean, it was removed;

[0049] Afterwards, cubic spline interpolation was used to make the number of sample points of the processed RR interval time series consistent with the EEG signal and to synchronize it with the EEG signal in time;

[0050] Finally, the preprocessed EEG signals and RR interval time series are divided into multiple segments with a set length.

[0051] Furthermore, the variational mode decomposition module specifically performs the following steps:

[0052] For the EEG signals and RR interval time series that are synchronized in time, the variational mode decomposition is used to decompose the EEG signals and RR interval time series respectively to obtain the intrinsic mode functions of the two time series, as follows:

[0053]

[0054] where {u k}:={u 1 ,...,u K} represents the K finite bandwidth intrinsic mode functions obtained by decomposition, {ω k}:={ω 1 ,...,ω K} is the center frequency of K eigenmode functions, is the partial derivative with respect to time t, δ(t) is the impulse function, t is the time, * is the convolution symbol, j is the imaginary part symbol, and f(t) is the function to be decomposed, that is, the EEG signal segment or the RR interval time series segment.

[0055] Introducing the Lagrangian multiplication operator λ(t) and the quadratic penalty parameter α, we solve the constrained variation problem of formula (1); the augmented Lagrangian function is obtained as

[0056]

[0057] The alternating multiplier method is used to obtain the saddle point of formula (2), and u is iteratively updated in the frequency domain. k ,ω k and λ.

[0058] The EEG signal fragment and the RR interval time series fragment are respectively used as the function f(t) to be decomposed. The intrinsic mode functions with different frequency components of the above two time series are obtained by the above decomposition method. For specific EEG signal and RR interval time series fragments, the K 1 and K 2 The intrinsic mode function u i ,u j ;

[0059] The amplitude-phase information extraction module specifically adopts the following steps:

[0060] For a specific EEG signal and RR interval time series fragment, each K 1 and K 2 The intrinsic mode function u i ,u j , and arranged in descending order according to the center frequency of the eigenmode function;

[0061] The Hilbert transform is used to process the intrinsic mode functions of the two time series and extract the phase information of the RR interval intrinsic mode functions. and the amplitude information of the intrinsic mode function of the EEG signal

[0062] Calculate the phase-amplitude modulation index two by two, specifically: Combine the phase information of the RR interval And the amplitude information A of the EEG signal i (t), calculate the phase-amplitude modulation index MI; divide the phase domain (0,2π) of the phase component into N intervals, calculate the average high-frequency amplitude on each low-frequency phase unit of each sample, and normalize the average amplitude to generate a distribution P, which represents the proportion of the high-frequency amplitude on each phase unit; calculate the KL distance between the phase amplitude distribution P and the uniform distribution U; finally, MI is defined as the ratio of the KL distance to logN.

[0063] Beneficial effects:

[0064] 1: The present invention provides a method for extracting heart-brain coupling features based on variational modal decomposition. Based on the excellent frequency band decomposition capability of variational modal decomposition and the advantages of phase-amplitude cross-frequency coupling in constructing high-dynamic coupling maps, a new algorithm for extracting heart-brain coupling features integrating variational modal decomposition and phase-amplitude cross-frequency coupling is proposed for the first time. Compared with the traditional heart-brain coupling algorithm, the heart-brain coupling map obtained based on variational modal decomposition has a better ability to combat nonlinearity and non-stationarity.

[0065] 2: The present invention provides a heart-brain coupling feature extraction system based on variational mode decomposition. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flow chart of a method for extracting heart-brain coupling features based on variational mode decomposition provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0068] The present invention provides a method to overcome the technical difficulties of the traditional heart-brain coupling, which has large electrical scale differences and poor ability to resist nonlinear non-stationarity. The present invention proposes a high-dynamic heart-brain coupling algorithm based on variational mode decomposition and phase-amplitude cross-frequency coupling, and realizes the evaluation and analysis of sleep structure. The technical flow chart of the present invention is as follows: Figure 1 The detailed process is as follows:

[0069] Step 1) Collect the ECG and EEG signals of the subject during sleep, and set the sampling frequencies to f s1 , f s2 , stored in the computer.

[0070] Step 2) Preprocess the original EEG signals and ECG signals. The detailed steps are as follows.

[0071] For EEG signals: use a bandpass filter to retain the set frequency band and remove other frequency bands. Here, the frequency band is set to 0.5-40Hz in the scenario of sleep analysis. For other analysis scenarios, the passband can be adjusted as needed. For example, in emotion analysis, the passband can be set to 0.5-45Hz to eliminate the influence of power frequency noise and other artifacts on the results; use independent component analysis ICA to decompose the filtered EEG data to remove interference caused by eye movement.

[0072] For ECG signal: use a bandpass filter on the entire ECG signal, retain the 0.5-30Hz frequency band, remove other frequency bands, and avoid the influence of power frequency noise and irrelevant frequency components on the results; extract the RR time series of heartbeat intervals from the preprocessed ECG signal, that is, mark the heartbeats in the single-lead ECG signal, calculate the time interval between two adjacent heartbeats, and obtain the RR time series; after obtaining the RR time series, use a sliding average filter with a 21-data point window to remove outliers caused by erroneous heartbeat detection. When the center point in the window is outside 30% of the average value, it will be removed. After that, use cubic spline interpolation to make the number of sample points of the processed RR interval time series consistent with the EEG signal, and keep it synchronized with the EEG signal in time. Finally, the preprocessed EEG signal and RR interval time series are divided into multiple segments of set lengths. For the sleep analysis scenario, the segment length of the EEG and RR interval data is 30s, while for other analysis scenarios, the segment length of the EEG and RR interval data can be adjusted. For example, the segment length can be set to 60s in the emotion analysis scenario.

[0073] Step 3) Decompose the EEG signal and RR interval time series of the 30-second EEG signal and RR interval time series synchronized in time using variational mode decomposition to obtain the intrinsic mode functions of the two time series. The details are as follows:

[0074]

[0075] where {u k}:={u 1 ,...,u K} represents the K finite bandwidth intrinsic mode functions obtained by decomposition, which can be expressed as u k (t); {ω k}:={ω 1 ,...,ω K} is the center frequency of the eigenmode function, is the partial derivative with respect to time t, δ(t) is the impulse function, t is the time, * is the convolution symbol, j is the imaginary part symbol, f(t) is the function to be decomposed, which can be an EEG signal segment or a RR interval time series segment.

[0076] In order to find the optimal solution to the constrained variational problem, the Lagrangian multiplication operator λ(t) and the quadratic penalty parameter α are introduced to solve the constrained variational problem of formula (1). The augmented Lagrangian function is obtained as follows:

[0077]

[0078] The alternating multiplier method (ADMM) is used to obtain the saddle point of equation (2), and u is iteratively updated in the frequency domain. k ,ω k And λ. That is

[0079]

[0080] The superscript n represents the nth iteration, and ρ is the time step in the dual ascent method, that is, the learning rate of parameter update.

[0081] by As an example, the solution of

[0082]

[0083] Due to the equivalence of Fourier transform, Parseval's theorem is used to solve in the frequency domain, where ω is the frequency variable:

[0084]

[0085] Taking the derivative of the above formula, we can know:

[0086]

[0087] Similarly, yes have:

[0088]

[0089] The stopping condition is:

[0090]

[0091] The EEG signal fragment and the RR interval time series fragment are respectively used as the function f(t) to be decomposed. The intrinsic mode functions with different frequency components of the above two time series are obtained by the above decomposition method. For specific EEG signal and RR interval time series fragments, the K 1 and K 2 The intrinsic mode function u i ,u j .

[0092] Step 4) After the calculation in the previous step, for a specific EEG signal and RR interval time series fragment, each K 1 and K 2 The intrinsic mode function u i ,u j , and arranged in descending order according to the center frequency of the eigenmode function. The eigenmode functions of the two time series are processed using the Hilbert transform to extract the phase information of the RR interval eigenmode function respectively. and the amplitude information of the intrinsic mode function of the EEG signal

[0093] Combined with the phase information of RR interval And the amplitude information A of the EEG signal i (t), calculate the modulation index (MI) between phase and amplitude. The modulation index aims to quantify the coupling strength between low-frequency phase and high-frequency amplitude (here low-frequency and high-frequency are relative concepts, that is, the phase of a signal component with a relatively low center frequency modulates the amplitude of another signal component with a relatively high center frequency). It is widely used to quantify the strength of phase-amplitude coupling, mainly by measuring the degree to which the distribution of high-frequency amplitude locked on the low-frequency phase deviates from the uniform distribution to quantify the strength of coupling. The phase domain (0,2π) of the phase component is divided into N intervals (for example, N=20), the average high-frequency amplitude on each low-frequency phase unit of each sample is calculated, and the average amplitude is normalized to generate a distribution P, which represents the proportion of high-frequency amplitude on each phase unit. The Kullback-Leibler (KL) distance, D, of the phase amplitude distribution P and the uniform distribution U is calculated. KL (P,U) = log(N) - H(P), where logN is the Shannon entropy of uniform distribution and H(P) is the Shannon entropy of distribution P, defined as Finally, MI is defined as the ratio of KL distance to logN, that is, When the average amplitude values ​​over all possible phase units are similar, that is, the distribution is approximately uniform, the MI value is close to 0.

[0094] To verify that the modulation index in the above calculation has a unique temporal modulation, both the phase and amplitude time series are “block mixed” at a periodic frequency to generate random permutations of these IMF blocks. To determine the empirical distribution of the i-th given amplitude component and the j-th given phase component, A i (t) and Perform random block mixing, the length of the block can be different. For each A i (t) and All blocks are assigned an index, and then the block index is randomly and fully shuffled to produce an almost random block arrangement. Finally, the blocks are concatenated according to the randomly permuted block indexes. In the resulting shuffled phase and amplitude time series and The temporal relationship between the high-frequency amplitude and the low-frequency phase is disordered, but the phase and amplitude curves within each block remain intact. This block mixing process is repeated 100 times for each pair of EEG signals and RR interval sequence intrinsic mode functions, and each time a shuffled phase and amplitude time series is obtained. and The MI value was calculated and the z score of the MI obtained in the 100-time block mixing process was set as the threshold, and the values ​​with a significance level above p = 0.05 were ignored.

[0095] Step 5) Based on the sleep heart-brain coupling map, the maximum modulation index in the region of interest is extracted as the heart-brain coupling feature.

[0096] By calculating the MI value between each EEG signal and the intrinsic mode function of the RR interval, we can obtain the phase-amplitude cross-frequency coupling spectrum of the heart and brain, and observe the distribution of MI values ​​generated by the phase-frequency-amplitude frequency pair. Repeat steps (1-6) for multi-channel EEG signals to obtain the phase-amplitude coupling spectrum between the EEG of different channels and each frequency of the RR interval. Finally, obtain the maximum value of MI in the region of interest in the phase-amplitude coupling spectrum, and use the obtained MI value as the classification feature of the machine learning classifier. By training the classifier and using the MI value as input, the wakefulness period, non-rapid eye movement sleep period and rapid eye movement sleep period can be predicted and identified.

[0097] The present invention has been comprehensively compared and evaluated with traditional methods in simulation experiments, and has been effectively verified on MASS-SS3 and ISRUC-S3 real sleep data sets. By simulating coupling signals under different degrees of nonlinearity and non-stationarity to construct coupling maps, it is found that compared with the traditional heart-brain coupling algorithm, the heart-brain coupling map based on variational mode decomposition has better ability to combat nonlinearity and non-stationarity. In addition, in the analysis of real sleep data sets, the new heart-brain coupling features show statistically significant differences between different sleep stages, providing support for subsequent sleep staging and sleep structure revelation.

[0098] Based on the above heart-brain phase frequency coupling feature extraction method, another embodiment of the present invention further provides a heart-brain coupling feature extraction system based on variational mode decomposition, characterized in that it includes a preprocessing module, an RR interval time series extraction module, a variational mode decomposition module, an amplitude-phase information extraction module and a heart-brain coupling feature extraction module;

[0099] The preprocessing module is used to preprocess the collected original single-lead ECG and EEG signals to avoid the influence of power frequency noise and irrelevant frequency components on the results;

[0100] The RR interval time series extraction module is used to extract the RR interval time series of heart beat intervals from the preprocessed ECG signal, and use cubic spline interpolation to make the number of sample points of the RR interval time series the same as that of the EEG signal, and keep it in time synchronization with the EEG signal;

[0101] The variational mode decomposition module uses the variational mode decomposition algorithm to process the EEG signal time series and the RR interval time series respectively, and obtains the intrinsic mode functions of the above two time series with different frequency components;

[0102] The amplitude-phase information extraction module is used to perform Hilbert transform on the intrinsic mode functions of different frequencies of the two time series, extract the amplitude information of the EEG intrinsic mode function and the phase information of the RR interval intrinsic mode function, and calculate the phase-amplitude modulation index two by two to construct the heart-brain coupling map during sleep;

[0103] The heart-brain coupling feature extraction module is used to extract the maximum modulation index in the region of interest as the heart-brain coupling feature based on the sleep heart-brain coupling map.

[0104] The preprocessing module specifically adopts the following preprocessing methods:

[0105] For EEG signals: use a bandpass filter to retain the set frequency band and remove other frequency bands, thereby eliminating the influence of power frequency noise and other artifacts on the results, and obtain the filtered EEG data; use independent component analysis ICA to decompose the filtered EEG data, remove the interference caused by eye movement, and obtain the preprocessed EEG signal.

[0106] For ECG signal: Use a bandpass filter on the entire ECG signal, retain the set frequency band, remove other frequency bands, avoid the influence of power frequency noise and irrelevant frequency components on the result, and obtain the preprocessed ECG signal.

[0107] The RR interval time series extraction module specifically performs the following steps:

[0108] The RR time series of heartbeat intervals is extracted from the preprocessed ECG signal, that is, the heartbeats are marked in the single-lead ECG signal, and the time interval between two adjacent heartbeats is calculated to obtain the RR time series.

[0109] After obtaining the RR interval time series, a sliding average filter with a window of 21 data points was used to remove outliers due to erroneous heartbeat detection. When the center point in the window was outside 30% of the mean, it was removed.

[0110] Afterwards, cubic spline interpolation is used to make the number of sample points of the processed RR interval time series consistent with the EEG signal and to keep it synchronized with the EEG signal in time.

[0111] Finally, the preprocessed EEG signals and RR interval time series are divided into multiple segments with a set length.

[0112] The variational mode decomposition module specifically performs the following steps:

[0113] For the EEG signals and RR interval time series that are synchronized in time, the variational mode decomposition is used to decompose the EEG signals and RR interval time series respectively to obtain the intrinsic mode functions of the two time series, as follows:

[0114]

[0115] where {u k}:={u 1 ,...,u K} represents the K finite bandwidth intrinsic mode functions obtained by decomposition, {ω k}:={ω 1 ,...,ω K} is the center frequency of K eigenmode functions, is the partial derivative with respect to time t, δ(t) is the impulse function, t is the time, * is the convolution symbol, j is the imaginary part symbol, and f(t) is the function to be decomposed, that is, the EEG signal segment or the RR interval time series segment.

[0116] Introducing the Lagrangian multiplication operator λ(t) and the quadratic penalty parameter α, we solve the constrained variation problem of formula (1); the augmented Lagrangian function is obtained as

[0117]

[0118] The alternating multiplier method is used to obtain the saddle point of formula (2), and u is iteratively updated in the frequency domain. k ,ω k and λ.

[0119] The EEG signal fragment and the RR interval time series fragment are respectively used as the function f(t) to be decomposed. The intrinsic mode functions with different frequency components of the above two time series are obtained by the above decomposition method. For specific EEG signal and RR interval time series fragments, the K 1 and K 2 The intrinsic mode function u i ,u j .

[0120] The amplitude-phase information extraction module specifically adopts the following steps: for a specific EEG signal and RR interval time series fragment, each K 1 and K 2 The intrinsic mode function u i ,u j , and are arranged in descending order according to the center frequency of the eigenmode function.

[0121] The Hilbert transform is used to process the intrinsic mode functions of the two time series and extract the phase information of the RR interval intrinsic mode functions. and the amplitude information of the intrinsic mode function of the EEG signal

[0122] Calculate the phase-amplitude modulation index two by two, specifically: Combine the phase information of the RR interval And the amplitude information A of the EEG signal i (t), calculate the phase-amplitude modulation index MI; divide the phase domain (0,2π) of the phase component into N intervals, calculate the average high-frequency amplitude on each low-frequency phase unit of each sample, and normalize the average amplitude to generate a distribution P, which represents the proportion of the high-frequency amplitude on each phase unit; calculate the KL distance between the phase amplitude distribution P and the uniform distribution U; finally, MI is defined as the ratio of the KL distance to logN.

[0123] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting heart-brain coupling features based on variational mode decomposition, characterized in that: The steps include: Step 1) preprocessing the collected original single-lead ECG signal and EEG signal to avoid the influence of power frequency noise and irrelevant frequency components on the results; Step 2) extracting the heart beat interval RR interval time series from the preprocessed ECG signal, and using cubic spline interpolation to make the number of sample points of the RR interval time series the same as the EEG signal, and to keep it in time synchronization with the EEG signal; Step 3) using the variational mode decomposition algorithm to process the EEG signal time series and the RR interval time series respectively, and obtaining the intrinsic mode functions of the above two time series with different frequency components; Step 4) performing Hilbert transform on the IMFs of different frequencies of the two time series, extracting the amplitude information of the EEG IMFs and the phase information of the RR interval IMFs, and calculating the phase-amplitude modulation indexes pairwise to construct a brain-heart coupling map during sleep; Step 5) Based on the sleep heart-brain coupling map, the maximum modulation index in the region of interest is extracted as the heart-brain coupling feature.

2. The method for extracting heart-brain coupling features based on variational mode decomposition according to claim 1, characterized in that: The acquisition frequencies of the original single-lead ECG and EEG signals are set to f s1 , f s2 , stored in the computer.

3. The method for extracting heart-brain coupling features based on variational mode decomposition according to claim 1, characterized in that: The collected original single-lead ECG signal and EEG signal are preprocessed to avoid the influence of power frequency noise and irrelevant frequency components on the results. The specific preprocessing method is as follows: For EEG signals: use a bandpass filter to retain the set frequency band and remove other frequency bands, thereby eliminating the influence of power frequency noise and other artifacts on the results, and obtain the filtered EEG data; use independent component analysis ICA to decompose the filtered EEG data, remove the interference caused by eye movement, and obtain the preprocessed EEG signal; For ECG signal: Use a bandpass filter on the entire ECG signal, retain the set frequency band, remove other frequency bands, avoid the influence of power frequency noise and irrelevant frequency components on the result, and obtain the preprocessed ECG signal.

4. The method for extracting heart-brain coupling features based on variational mode decomposition according to claim 1, characterized in that: The step 2) extracts the heart beat interval RR interval time series from the preprocessed ECG signal, and uses cubic spline interpolation to make the number of sample points of the RR interval time series the same as the EEG signal, and keeps it in time synchronization with the EEG signal, specifically: Extract the heartbeat interval RR time series from the preprocessed ECG signal, that is, mark the heartbeats in the single-lead ECG signal, calculate the time interval between two adjacent heartbeats, and obtain the RR interval time series; After obtaining the RR interval time series, a sliding average filter with a window of 21 data points was used to remove outliers due to erroneous heartbeat detection. When the center point in the window was outside 30% of the mean, it was removed; Afterwards, cubic spline interpolation was used to make the number of sample points of the processed RR interval time series consistent with the EEG signal and to synchronize it with the EEG signal in time; Finally, the preprocessed EEG signals and RR interval time series are divided into multiple segments with a set length.

5. The method for extracting heart-brain coupling features based on variational mode decomposition as claimed in claim 4, characterized in that: The step 3) uses the variational mode decomposition algorithm to process the EEG signal time series and the RR interval time series respectively to obtain the intrinsic mode functions of the two time series with different frequency components, specifically: For the EEG signals and RR interval time series that are synchronized in time, the variational mode decomposition is used to decompose the EEG signals and RR interval time series respectively to obtain the intrinsic mode functions of the two time series, as follows: where {u k }:={u1,...,u K } represents the K finite bandwidth intrinsic mode functions obtained by decomposition, {ω k }:={ω1,...,ω K } is the center frequency of K eigenmode functions, is the partial derivative with respect to time t, δ(t) is the impulse function, t is the time, * is the convolution symbol, j is the imaginary part symbol, and f(t) is the function to be decomposed, i.e., the EEG signal segment or the RR interval time series segment; Introducing the Lagrangian multiplication operator λ(t) and the quadratic penalty parameter α, we solve the constrained variation problem of formula (1); the augmented Lagrangian function is obtained as The alternating multiplier method is used to obtain the saddle point of formula (2), and u is iteratively updated in the frequency domain. k ,ω k and λ; The EEG signal fragment and the RR interval time series fragment are respectively used as the function f(t) to be decomposed. The eigenmode functions with different frequency components of the above two time series are obtained by the above decomposition method. For a specific EEG signal and RR interval time series fragment, K1 and K2 eigenmode functions u are obtained respectively. i ,u j .

6. The method for extracting heart-brain coupling features based on variational mode decomposition as claimed in claim 5, characterized in that: In step 4), the eigenmode functions of different frequencies of the two time series are subjected to Hilbert transform to extract the amplitude information of the EEG eigenmode function and the phase information of the RR interval eigenmode function, specifically using the following steps: In step 3), for a specific EEG signal and RR interval time series fragment, K1 and K2 eigenmode functions u are obtained respectively. i ,u j , and arranged in descending order according to the center frequency of the eigenmode function; The Hilbert transform is used to process the intrinsic mode functions of the two time series and extract the phase information of the RR interval intrinsic mode functions. and the amplitude information of the intrinsic mode function of the EEG signal The two-by-two calculation of the phase-amplitude modulation index is specifically: combining the phase information of the RR interval And the amplitude information A of the EEG signal i (t), calculate the phase-amplitude modulation index MI; divide the phase domain (0,2π) of the phase component into N intervals, calculate the average high-frequency amplitude on each low-frequency phase unit of each sample, and normalize the average amplitude to generate a distribution P, which represents the proportion of the high-frequency amplitude on each phase unit; The KL distance between the phase amplitude distribution P and the uniform distribution U is calculated; finally, MI is defined as the ratio of the KL distance to logN.

7. The method for extracting heart-brain coupling features based on variational mode decomposition according to claim 6, characterized in that: After step 5), the method further includes the following steps: constructing a feature classifier, wherein the feature classifier takes the heart-brain coupling feature value as input and takes the wakefulness period, non-rapid eye movement sleep period and rapid eye movement sleep period as output; Using the original single-lead ECG signals and EEG signals corresponding to the wakefulness period, non-rapid eye movement sleep period and rapid eye movement sleep period as the acquisition signals, executing steps 1) to 5) to obtain the corresponding heart-brain coupling feature values, thereby constructing training samples and training the feature classifier; The trained feature classifier is used to predict and identify wakefulness, non-REM sleep, and REM sleep.

8. A heart-brain coupling feature extraction system based on variational mode decomposition, characterized in that: It includes a preprocessing module, an RR interval time series extraction module, a variational mode decomposition module, an amplitude-phase information extraction module, and a heart-brain coupling feature extraction module; The preprocessing module is used to preprocess the collected original single-lead ECG signal and EEG signal to avoid the influence of power frequency noise and irrelevant frequency components on the results; The RR interval time series extraction module is used to extract the RR interval time series of heart beat intervals from the preprocessed electrocardiogram signal, and use cubic spline interpolation to make the number of sample points of the RR interval time series the same as that of the electroencephalogram signal, and keep it in time synchronization with the electroencephalogram signal; The variational mode decomposition module uses a variational mode decomposition algorithm to process the EEG signal time series and the RR interval time series respectively to obtain the intrinsic mode functions of the two time series with different frequency components; The amplitude-phase information extraction module is used to perform Hilbert transform on the intrinsic mode functions of different frequencies of the two time series, extract the amplitude information of the EEG intrinsic mode function and the phase information of the RR interval intrinsic mode function, and calculate the phase-amplitude modulation index in pairs to construct a heart-brain coupling spectrum during sleep; The heart-brain coupling feature extraction module is used to extract the maximum modulation index in the region of interest as the heart-brain coupling feature based on the sleep heart-brain coupling map.

9. The heart-brain coupling feature extraction system based on variational mode decomposition as claimed in claim 8, characterized in that: The preprocessing module specifically adopts the following preprocessing method: For EEG signals: use a bandpass filter to retain the set frequency band and remove other frequency bands, thereby eliminating the influence of power frequency noise and other artifacts on the results, and obtain the filtered EEG data; use independent component analysis ICA to decompose the filtered EEG data, remove the interference caused by eye movement, and obtain the preprocessed EEG signal; For ECG signal: use a bandpass filter on the entire ECG signal, retain the set frequency band, remove other frequency bands, avoid the influence of power frequency noise and irrelevant frequency components on the result, and obtain the preprocessed ECG signal; The RR interval time series extraction module specifically performs the following steps: Extract the heartbeat interval RR time series from the preprocessed ECG signal, that is, mark the heartbeats in the single-lead ECG signal, calculate the time interval between two adjacent heartbeats, and obtain the RR interval time series; After obtaining the RR interval time series, a sliding average filter with a window of 21 data points was used to remove outliers due to erroneous heartbeat detection. When the center point in the window was outside 30% of the mean, it was removed; Afterwards, cubic spline interpolation was used to make the number of sample points of the processed RR interval time series consistent with the EEG signal and to synchronize it with the EEG signal in time; Finally, the preprocessed EEG signals and RR interval time series are divided into multiple segments with a set length.

10. The heart-brain coupling feature extraction system based on variational mode decomposition as claimed in claim 8, characterized in that: The variational mode decomposition module specifically performs the following steps: For the EEG signals and RR interval time series that are synchronized in time, the variational mode decomposition is used to decompose the EEG signals and RR interval time series respectively to obtain the intrinsic mode functions of the two time series, as follows: where {u k }:={u1,...,u K } represents the K finite bandwidth intrinsic mode functions obtained by decomposition, {ω k }:={ω1,...,ω K } is the center frequency of K eigenmode functions, is the partial derivative with respect to time t, δ(t) is the impulse function, t is the time, * is the convolution symbol, j is the imaginary part symbol, and f(t) is the function to be decomposed, i.e., the EEG signal segment or the RR interval time series segment; Introducing the Lagrangian multiplication operator λ(t) and the quadratic penalty parameter α, we solve the constrained variation problem of formula (1); the augmented Lagrangian function is obtained as The alternating multiplier method is used to obtain the saddle point of formula (2), and u is iteratively updated in the frequency domain. k ,ω k and λ; The EEG signal fragment and the RR interval time series fragment are respectively used as the function f(t) to be decomposed. The eigenmode functions with different frequency components of the above two time series are obtained by the above decomposition method. For a specific EEG signal and RR interval time series fragment, K1 and K2 eigenmode functions u are obtained respectively. i ,u j ; The amplitude-phase information extraction module specifically adopts the following steps: For a specific EEG signal and RR interval time series fragment, K1 and K2 intrinsic mode functions u will be obtained respectively. i ,u j , and arranged in descending order according to the center frequency of the eigenmode function; The Hilbert transform is used to process the intrinsic mode functions of the two time series and extract the phase information of the RR interval intrinsic mode functions. and the amplitude information of the intrinsic mode function of the EEG signal The two-by-two calculation of the phase-amplitude modulation index is specifically: combining the phase information of the RR interval And the amplitude information A of the EEG signal i (t), calculate the phase-amplitude modulation index MI; divide the phase domain (0,2π) of the phase component into N intervals, calculate the average high-frequency amplitude on each low-frequency phase unit of each sample, and normalize the average amplitude to generate a distribution P, which represents the proportion of the high-frequency amplitude on each phase unit; The KL distance between the phase amplitude distribution P and the uniform distribution U is calculated; finally, MI is defined as the ratio of the KL distance to logN.

Citation Information

Patent Citations

  • Single-channel heart-brain information based coupling analysis method and system

    CN111248900A

  • Variational mode decomposition (VMD) and non-linear twin support vector machine (NLTWSVM)-based multi-feature epilepsy signal classification method

    CN113558643A

  • Emotion recognition method and device based on variational phase amplitude coupling

    CN114343635A

  • Cardiopulmonary coupling feature extraction method and system based on synchronous extrusion transformation

    CN115630290A

  • Sleep dynamics analysis method based on variational cardiopulmonary coupling

    CN115886742A

Cited By

  • A high-precision anesthesia state monitoring method, system and device

    CN119745347A

  • Emotion prediction method and system based on cardiac-computerized electrical coupling

    CN120514386A

  • Brain function assessment method and system based on variational modal information coefficient

    CN121313194A

  • Cardio-cerebral magnetic coupling index acquisition method and device, equipment and storage medium

    CN121971098A

  • A method, apparatus, device, and storage medium for acquiring cardiac-brain magnetic coupling indicators.

    CN121971098B