Nonlinear analysis method and system for EEG signals

Abstract

The present invention relates to a nonlinear analysis method for EEG signals, comprising: obtaining resting-state EEG signals within a preset total time period, preprocessing the resting-state EEG signals to obtain target EEG signals; dividing the target EEG signals to obtain a target EEG signal set corresponding to multiple frequency bands; constructing a time-frequency matrix based on the target EEG signal set corresponding to each frequency band; using the time-frequency matrix to construct multiple time-frequency domains containing partial sub-target EEG signals corresponding to each frequency band at a specific time point; respectively calculating the difference distance between each two time-frequency domains, and constructing a distance matrix between the two time-frequency domains based on the difference distance; reconstructing features for each frequency band in the distance matrix to obtain a feature reconstruction sequence; and calculating multi-band entropy based on the feature reconstruction sequence. The present invention also relates to a nonlinear analysis system for EEG signals. The present invention combines time and frequency band information for research to improve the accuracy, specificity and sensitivity of nonlinear analysis of EEG signals.

Description

Technical Field

[0001] The invention relates to a nonlinear analysis method and system for electroencephalogram signals. Background Art

[0002] EEG is a technology that records the activity of brain neurons. It can detect the discharge of neurons in the cerebral cortex, thereby generating a series of waveforms. EEG signals can be used to study depression, epilepsy, Alzheimer's disease, cognitive impairment and many other diseases. Entropy-based nonlinear analysis methods can effectively characterize the random change pattern of EEG signals.

[0003] Most of the current entropy methods study the information of the entire frequency band or a single frequency band. This obviously ignores the relationship between signals in different frequency bands. Since EEG signals are composed of signals in many frequency bands, the signals in each frequency band represent different brain activity patterns, and there are relationships between different frequency bands. For example, there is an inverse relationship between Alpha waves and Theta waves, that is, when the Alpha wave is very strong, the Theta wave is always weak, and vice versa. Therefore, by measuring signals in different frequency bands in the brain area at the same time, we can have a more comprehensive understanding of the overall situation of brain activity. However, due to the limitations of its own principles, traditional complexity analysis methods generally have methodological limitations. For example: ignoring the relationship between signals in different frequency bands, it is impossible to fully understand the overall situation of brain activity.

[0004] It can be seen that the existing technology can only study the EEG information of the entire frequency band or a single frequency band, and the problem of being unable to fully understand the overall situation of brain activity needs to be solved urgently. Summary of the invention

[0005] In view of this, it is necessary to provide a nonlinear analysis method and system for EEG signals.

[0006] The present invention provides a nonlinear analysis method for electroencephalogram (EEG) signals, which comprises the following steps: S1, acquiring resting-state electroencephalogram (EEG) signals within a preset total time period, preprocessing the resting-state electroencephalogram (EEG) signals, and obtaining target electroencephalogram (EEG) signals; S2, dividing the target electroencephalogram (EEG) signals to obtain target electroencephalogram (EEG) signal sets corresponding to multiple frequency bands; S3, constructing a time-frequency matrix based on the target electroencephalogram (EEG) signal sets corresponding to each frequency band; S4, using the time-frequency matrix to construct multiple time-frequency domains containing partial sub-target electroencephalogram (EEG) signals corresponding to each frequency band at a specific time point; S5, respectively calculating the difference distance between each two time-frequency domains, and constructing a distance matrix between the two time-frequency domains based on the difference distance; S6, performing feature reconstruction on each frequency band in the distance matrix to obtain a feature reconstruction sequence; S7, calculating multi-band entropy based on the feature reconstruction sequence.

[0007] Preferably, the preprocessing includes: channel positioning, removal of useless electrodes, filtering, segmentation, removal of bad segments and replacement of bad conductors, removal of ICA artifact components, extreme value artifact removal and re-referencing.

[0008] Preferably, the step S2 comprises:

[0009] The multi-band decomposition method is used to divide the target EEG signal into sub-target EEG signals corresponding to the Alpha band, Beta band, and Theta band:

[0010] ;

[0011] ;

[0012] ;

[0013] in, , , are the target EEG signals corresponding to each unit time point in the preset total time period N for the Alpha band, the Beta band, and the Theta band respectively; , , They are the target EEG signals corresponding to the Alpha band, Beta band and Theta band in the preset total time period N.

[0014] Preferably, the step S3 comprises:

[0015] The time-frequency matrix is ​​constructed based on the target EEG signals corresponding to the preset total time period N in the Alpha band, Beta band, and Theta band:

[0016]

[0017] in, Represents the time-frequency matrix composed of the target EEG signal within the preset total time period N; , , They are the target EEG signals corresponding to each unit time point in the preset total time period N for the Alpha band, the Beta band and the Theta band respectively.

[0018] Preferably, the step S4 comprises:

[0019] Based on the time-frequency matrix, a time-frequency domain containing the Alpha band, Beta band, and Theta band in the preset sub-time period n is constructed. ; where i is the time-frequency domain corresponding to a specific time point, 1≤i≤n-m+1, that is for :

[0020] (1)

[0021] (2)

[0022] Based on the above formulas (1) and (2), we can obtain:

[0023]

[0024] Among them, m is the preset embedding dimension; Represents the target EEG signal set corresponding to the specific time point i in the full frequency band with embedding dimension m; 1≤n≤N; 1≤i≤N-m+1; It refers to the target EEG signal set corresponding to each frequency band at a specific time point i; H represents the Alpha band, Beta band or Theta band.

[0025] Preferably, the step S5 comprises:

[0026] The maximum value of the difference between the corresponding frequency bands of the two time-frequency domains is calculated as the difference distance; and the distance matrix between any two time-frequency domains is calculated using the following formula:

[0027]

[0028] in, , are the target EEG signal sets corresponding to the embedding dimension m and the full frequency band at specific time points i and j respectively; , are the target EEG signal sets corresponding to each frequency band at specific time points i and j respectively; 1<i<n-m+1; 1<j<n-m+1; i≠j; H represents Alpha band, Beta band or Theta band.

[0029] Preferably, the step S6 comprises:

[0030] The distance matrix is ​​expressed using the following formula:

[0031]

[0032] in, Represents the maximum value of the difference between each frequency band at a specific time point i and j; H represents the Alpha band, Beta band or Theta band;

[0033] Arrange the distance matrix in ascending order according to the value The distance feature of each frequency band in the , and the feature sorting result is obtained; where the distance feature is the maximum value of the difference between each frequency band at a specific time point i and j , , ;

[0034] The feature sorting results are: ≤ ;

[0035] in, , , is the subscript index value of each distance feature in the reconstructed component;

[0036] Generate a feature reconstruction sequence based on the feature sorting results:

[0037]

[0038] in, is the distance matrix The corresponding feature reconstruction sequence; l =1,2... K ; K ≤3!.

[0039] Preferably, the step S7 comprises:

[0040] Based on the definition of Shannon entropy, the multi-band entropy is defined as:

[0041]

[0042] Calculate the probability of occurrence of different feature reconstruction sequences. The probability of each 3D feature reconstruction sequence is P1, P2, P3...P k , 1≤j≤k, The multi-band entropy corresponding to the preset total time period;

[0043] Normalize the calculated multi-band entropy:

[0044] 0≤ ≤ / ln(3!)≤1

[0045] in, The multi-band entropy corresponding to the preset total time period; is the target multi-band entropy corresponding to the preset total time period; 3! is the total number of occurrences of different feature reconstruction sequences.

[0046] The present invention also provides an EEG signal nonlinear analysis system, which includes a data preprocessing module, a signal frequency division module, a time-frequency matrix construction module, a time-frequency domain construction module, a distance calculation module, a feature reconstruction module and a multi-band entropy calculation module, wherein:

[0047] The data preprocessing module is used to obtain resting-state EEG signals within a preset total time period, preprocess the resting-state EEG signals, and obtain target EEG signals;

[0048] The signal frequency division module is used to divide the target EEG signal to obtain a target EEG signal set corresponding to multiple frequency bands;

[0049] The time-frequency matrix construction module is used to construct a time-frequency matrix based on the target EEG signal set corresponding to each frequency band;

[0050] The time-frequency domain construction module is used to construct multiple time-frequency domains containing part of the sub-target EEG signals corresponding to each frequency band at a specific time point using the time-frequency matrix;

[0051] The distance calculation module is used to respectively calculate the difference distance between each two time-frequency domains, and construct a distance matrix between the two time-frequency domains based on the difference distance;

[0052] The feature reconstruction module is used to perform feature reconstruction on each frequency band in the distance matrix to obtain a feature reconstruction sequence;

[0053] The multi-band entropy calculation module is used to calculate the multi-band entropy based on the feature reconstruction sequence.

[0054] The present invention can solve the problem that the existing technology can only study EEG information of the entire frequency band or a single frequency band, and cannot fully understand the overall situation of brain activity. The present invention deeply explores the pattern information between frequency bands, integrates time and frequency band information for research, and introduces the idea of ​​sorting symbolization to improve the noise resistance. It can be applied to the complexity analysis of EEG signal data, and improves the accuracy, specificity and sensitivity of nonlinear analysis of EEG signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of the nonlinear analysis method of EEG signal of the present invention;

[0056] Figure 2 This is a hardware architecture diagram of the EEG signal nonlinear analysis system of the present invention. DETAILED DESCRIPTION

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] See also Figure 1 , which is a flowchart of a preferred embodiment of the nonlinear analysis method of EEG signals of the present invention.

[0059] Step S1, obtaining a resting EEG signal within a preset total time period, preprocessing the resting EEG signal, and obtaining a target EEG signal. Specifically:

[0060] The data set of the embodiment of the present invention is resting-state EEG data (electroencephalogram data) from the University of Iowa, and the collection objects include 14 Parkinson's disease patients and 14 controls, and all procedures comply with relevant regulations.

[0061] After determining the subjects, researchers at the University of Iowa collected resting-state EEG signals through experiments. In the specific experimental process, the resting-state EEG data of 28 subjects were recorded with their eyes open; Ag / AgCl electrodes were used with a frequency range of 0.1-100 Hz and a sampling rate of 500 Hz, and a 64-channel Brain Vision system was used for recording; the online reference electrodes set CPz (UNM) and Pz (Iowa) as baselines. Therefore, CPz and Pz data are missing in the data set, respectively. The EEG data set used in the embodiment of the present invention has a total of 62 channels, and eye movement noise is identified and removed by independent component analysis. Thus, the resting-state EEG signal acquisition of multiple channels of the subject's brain is completed.

[0062] Furthermore, the resting-state EEG signals are preprocessed to obtain the target EEG signals. Specifically, EEGLAB is used to preprocess the resting-state EEG data, including channel positioning, removal of useless electrodes, filtering, segmentation, bad segment removal and bad conductor replacement, removal of ICA (Independent Component Analysis) artifact components, extreme value artifact removal, and re-reference.

[0063] In the embodiment of the present invention, the preprocessing of the resting-state EEG signal is performed based on the Matlab toolbox EEGLAB. Among them, the channel positioning refers to matching the channel label of the electrode with the actual position; the removal of useless electrodes refers to removing useless electrode signals by using the function pop_select; the filtering refers to removing low-frequency or high-frequency noise, and the signal of 0.1 to 40Hz is retained in this embodiment; the bad segment removal and bad lead replacement refers to replacing the bad lead data segment caused by environmental noise or poor lead based on the pop_rejcont function; the removal of ICA artifact components refers to removing artifact components by applying the ICA independent component analysis algorithm based on the pop_runica function; the extreme value artifact removal refers to detecting and removing extreme values ​​using the function pop_rejchan; the re-reference refers to average reference of EEG data based on the pop_reref function to change the reference mode.

[0064] In the experiment of the embodiment of the present invention, target EEG data of 14 normal subjects and 13 subjects with Parkinson's disease were obtained after the above preprocessing operation, and each subject had 58 channels of target EEG data.

[0065] Step S2, frequency-divide the target EEG signal to obtain a target EEG signal set corresponding to multiple frequency bands. Specifically:

[0066] The embodiment of the present invention first adopts a multi-band decomposition method to divide the target EEG signal into sub-target EEG signals corresponding to the Alpha band, the Beta band, and the Theta band:

[0067] ;

[0068] ;

[0069] ;

[0070] in, , , are the target EEG signals corresponding to each unit time point in the preset total time period N for the Alpha band, the Beta band, and the Theta band respectively; , , They are the target EEG signals corresponding to the Alpha band, Beta band and Theta band in the preset total time period N.

[0071] That is, the embodiment of the present invention first divides the target EEG signal of each channel to obtain the sub-target EEG signals corresponding to the Alpha band (8~13Hz), Beta band (13~30Hz) and Theta band (4~8Hz). Taking the preset total time period N containing 200 sampling points as an example, each frequency band contains the target EEG signals corresponding to 200 sampling points. For example, It is the target EEG signal data corresponding to the Alpha frequency band at the first time point; is the target EEG signal data corresponding to the Beta frequency band at the second time point; is the target EEG signal data corresponding to the Theta frequency band at the Nth time point.

[0072] Step S3, constructing a time-frequency matrix based on the target EEG signal set corresponding to each frequency band. The time-frequency matrix is ​​a data set of the target EEG signal set corresponding to each frequency band. Specifically:

[0073] In the embodiment of the present invention, the target EEG signals corresponding to the three frequency bands in the preset total time period N can be summarized in the form of a time-frequency matrix, that is, the time-frequency matrix is ​​a data set of the target EEG signal set corresponding to each frequency band. Specifically, the time-frequency matrix is ​​constructed based on the target EEG signals corresponding to the Alpha band, the Beta band, and the Theta band in the preset total time period N:

[0074]

[0075] in, Represents the time-frequency matrix composed of the target EEG signal within the preset total time period N; , , They are the target EEG signals corresponding to each unit time point in the preset total time period N for the Alpha band, the Beta band and the Theta band respectively.

[0076] Step S4, using the time-frequency matrix to construct multiple time-frequency domains containing the partial sub-target EEG signals corresponding to each frequency band at a specific time point. Specifically:

[0077] The embodiment of the present invention presets the embedding dimension, and constructs a time-frequency domain including the Alpha band, the Beta band, and the Theta band in the preset sub-time period n based on the time-frequency matrix. ; where i is the time-frequency domain corresponding to a specific time point, 1≤i≤n-m+1, that is for :

[0078] (1)

[0079] (2)

[0080] Based on the above formulas (1) and (2), we can obtain:

[0081]

[0082] Among them, m is the preset embedding dimension; Represents the target EEG signal set corresponding to the specific time point i in the full frequency band with embedding dimension m; 1≤n≤N; 1≤i≤N-m+1; It refers to the target EEG signal set corresponding to each frequency band at a specific time point i; H represents the Alpha band, Beta band or Theta band.

[0083] It should be noted that entropy calculation requires the selection of a suitable embedding dimension m. Studies have shown that the larger the m value, the more effective it is for reconstructing the dynamic development process of the system, but it cannot be too large. Preferably, the embedding dimension m of the phase space in the embodiment of the present invention is 4. It can be understood that the time-frequency domain in the embodiment of the present invention is a target EEG signal set corresponding to the full frequency band at a specific time point i. The specific time point i is any time point in the preset total time period N. The length of the time-frequency domain is n, that is, part of the time points in the preset total time period N.

[0084] Step S5, respectively calculate the difference distance between each two time-frequency domains, and construct a distance matrix between the two time-frequency domains based on the difference distance. Specifically:

[0085] In the embodiment of the present invention, the maximum value of the difference between the corresponding frequency bands in the time-frequency domain is calculated as the difference distance; and the distance matrix between any two time-frequency domains is calculated using the following formula:

[0086]

[0087] Among them, and are the embedding dimension m, the target EEG signal sets corresponding to all frequency bands at specific time points i and j, respectively; and are the target EEG signal sets corresponding to each frequency band at specific time points i and j, respectively; 1 < i < n - m + 1; 1 < j < n - m + 1; i ≠ j; H represents the Alpha band, Beta band or Theta band.

[0088] Taking i = 1 and j = 2 as an example, = d ={max{| - |}, max{| |}, max{| |}}.

[0089] According to Equation (2), max{| - |} = max{| - |, | - |,..., | - |}.

[0090] Thus, the embodiment of the present invention generates a total of n - m + 1 distance matrices.

[0091] Step S6, perform feature reconstruction on each frequency band in the distance matrix to obtain a feature reconstruction sequence. Specifically:

[0092] The embodiment of the present invention represents the distance matrix using the following formula:

[0093]

[0094] Among them, represents the maximum value of the difference between each frequency band at specific time points i and j; H represents the Alpha band, Beta band or Theta band.

[0095] Furthermore, arrange the distance matrix in ascending order of numerical values The distance feature of each frequency band in the , and the feature sorting result is obtained; where the distance feature is the maximum value of the difference between each frequency band at a specific time point i and j , , .

[0096] The feature sorting results are: ≤ ;

[0097] in, , , is the subscript index value of each distance feature in the reconstructed component;

[0098] Generate a feature reconstruction sequence based on the feature sorting results:

[0099]

[0100] in, is the distance matrix The corresponding feature reconstruction sequence; there are 3! ways to arrange the three frequency bands, so l =1,2... K ; K ≤3!. It can be understood that each distance matrix in the embodiment of the present invention is a 3-dimensional space, mapped to a 3-dimensional feature reconstruction sequence, and has 3! kinds of arrangement modes.

[0101] Step S7, calculating multi-band entropy based on the feature reconstruction sequence. The multi-band entropy is used to perform nonlinear analysis on the EEG signal. Specifically:

[0102] Based on the definition of Shannon entropy, the multi-band entropy in the embodiment of the present invention is defined as:

[0103]

[0104] Therefore, we first calculate the probability of occurrence of different feature reconstruction sequences. The probability of each 3D feature reconstruction sequence is P1, P2, P3...P k The embodiment of the present invention is based on P j Represents the occurrence probability of different feature reconstruction sequences in the target area; 1≤j≤k. It is the multi-band entropy corresponding to the preset total time period.

[0105] It can be understood that when the probability of reconstructing the sequence of each 3D feature is the same, that is, P j =1 / 3!, multi-band entropy The maximum value ln(3!) is obtained. Therefore, in the actual processing, the calculated multi-band entropy is normalized:

[0106] 0≤ ≤ / ln(3!)≤1

[0107] in, The multi-band entropy corresponding to the preset total time period; is the target multi-band entropy corresponding to the preset total time period; 3! is the total number of occurrences of different feature reconstruction sequences.

[0108] See also Figure 2 , which is a hardware architecture diagram of the EEG signal nonlinear analysis system 10 of the present invention. The system includes: a data preprocessing module 101, a signal frequency division module 102, a time-frequency matrix construction module 103, a time-frequency domain construction module 104, a distance calculation module 105, a feature reconstruction module 106 and a multi-band entropy calculation module 107. Among them:

[0109] The data preprocessing module 101 is used to obtain the resting EEG signals within a preset total time period, preprocess the resting EEG signals, and obtain the target EEG signals. Specifically:

[0110] The data set of the embodiment of the present invention is resting-state EEG data (electroencephalogram data) from the University of Iowa, and the collection objects include 14 Parkinson's disease patients and 14 controls, and all procedures comply with relevant regulations.

[0111] After determining the subjects, researchers at the University of Iowa collected resting-state EEG signals through experiments. In the specific experimental process, the resting-state EEG data of 28 subjects were recorded with their eyes open; Ag / AgCl electrodes were used with a frequency range of 0.1-100 Hz and a sampling rate of 500 Hz, and a 64-channel Brain Vision system was used for recording; the online reference electrodes set CPz (UNM) and Pz (Iowa) as baselines. Therefore, CPz and Pz data are missing in the data set, respectively. The EEG data set used in the embodiment of the present invention has a total of 62 channels, and eye movement noise is identified and removed by independent component analysis. Thus, the resting-state EEG signal acquisition of multiple channels of the subject's brain is completed.

[0112] Furthermore, the data preprocessing module 101 preprocesses the resting-state EEG signal to obtain the target EEG signal. Specifically, EEGLAB is used to preprocess the resting-state EEG data, including channel positioning, removing useless electrodes, filtering, segmentation, removing bad segments and replacing bad conductors, removing ICA (Independent Component Analysis) artifact components, removing extreme value artifacts, and re-referencing.

[0113] In the embodiment of the present invention, the preprocessing of the resting-state EEG signal is performed based on the Matlab toolbox EEGLAB. Among them, the channel positioning refers to matching the channel label of the electrode with the actual position; the removal of useless electrodes refers to removing useless electrode signals by using the function pop_select; the filtering refers to removing low-frequency or high-frequency noise, and the signal of 0.1 to 40Hz is retained in this embodiment; the bad segment removal and bad lead replacement refers to replacing the bad lead data segment caused by environmental noise or poor lead based on the pop_rejcont function; the removal of ICA artifact components refers to removing artifact components by applying the ICA independent component analysis algorithm based on the pop_runica function; the extreme value artifact removal refers to detecting and removing extreme values ​​using the function pop_rejchan; the re-reference refers to average reference of EEG data based on the pop_reref function to change the reference mode.

[0114] In the experimental process of the embodiment of the present invention, after the above preprocessing operation, target EEG data of 14 normal subjects and 13 subjects with Parkinson's disease were obtained, and each subject had 58 channels of target EEG data.

[0115] The signal frequency division module 102 is used to divide the target EEG signal to obtain a target EEG signal set corresponding to multiple frequency bands. Specifically:

[0116] The embodiment of the present invention first adopts a multi-band decomposition method to divide the target EEG signal into sub-target EEG signals corresponding to the Alpha band, the Beta band, and the Theta band:

[0117] ;

[0118] ;

[0119] ;

[0120] in, , , are the target EEG signals corresponding to each unit time point in the preset total time period N for the Alpha band, the Beta band, and the Theta band respectively; , , They are the target EEG signals corresponding to the Alpha band, Beta band and Theta band in the preset total time period N.

[0121] That is, the embodiment of the present invention first divides the target EEG signal of each channel to obtain the sub-target EEG signals corresponding to the Alpha band (8~13Hz), Beta band (13~30Hz) and Theta band (4~8Hz). Taking the preset total time period N containing 200 sampling points as an example, each frequency band contains the target EEG signals corresponding to 200 sampling points. For example, It is the target EEG signal data corresponding to the Alpha frequency band at the first time point; is the target EEG signal data corresponding to the Beta frequency band at the second time point; is the target EEG signal data corresponding to the Theta frequency band at the Nth time point.

[0122] The time-frequency matrix construction module 103 is used to construct a time-frequency matrix based on the target EEG signal set corresponding to each frequency band. The time-frequency matrix is ​​a data set of the target EEG signal set corresponding to each frequency band. Specifically:

[0123] In the embodiment of the present invention, the target EEG signals corresponding to the three frequency bands in the preset total time period N can be summarized in the form of a time-frequency matrix, that is, the time-frequency matrix is ​​a data set of the target EEG signal set corresponding to each frequency band. Specifically, the time-frequency matrix is ​​constructed based on the target EEG signals corresponding to the Alpha band, the Beta band, and the Theta band in the preset total time period N:

[0124]

[0125] in, Represents the time-frequency matrix composed of the target EEG signal within the preset total time period N; , , They are the target EEG signals corresponding to each unit time point in the preset total time period N for the Alpha band, the Beta band and the Theta band respectively.

[0126] The time-frequency domain construction module 104 is used to construct multiple time-frequency domains containing the sub-target EEG signals corresponding to each frequency band at a specific time point using the time-frequency matrix. Specifically:

[0127] The embodiment of the present invention presets the embedding dimension, and constructs a time-frequency domain including the Alpha band, the Beta band, and the Theta band in the preset sub-time period n based on the time-frequency matrix. ; where i is the time-frequency domain corresponding to a specific time point, 1≤i≤n-m+1, that is for :

[0128] (1)

[0129] (2)

[0130] Based on the above formulas (1) and (2), we can obtain:

[0131]

[0132] where m is the preset embedding dimension; represents the target EEG signal set corresponding to the embedding dimension m and the full frequency band at a specific time point i; 1 ≤ n ≤ N; 1 ≤ i ≤ N - m + 1; refers to the target EEG signal set corresponding to each frequency band at a specific time point i; H represents the Alpha frequency band, Beta frequency band or Theta frequency band.

[0133] It should be noted that for entropy calculation, an appropriate embedding dimension m needs to be selected. Research indicates that the larger the value of m, the more effective it is for reconstructing the dynamic development process of the system, but it cannot be too large. Preferably, the phase space embedding dimension m in the embodiments of the present invention takes the value of 4. It can be understood that the time-frequency domain in the embodiments of the present invention is the target EEG signal set corresponding to the full frequency band at a specific time point i, and the specific time point i is any time point in the preset total time period N. The length of the time-frequency domain is n, that is, a part of the time points in the preset total time period N.

[0134] The distance calculation module 105 is used to calculate the difference distance between every two time-frequency domains respectively, and construct a distance matrix between the two time-frequency domains based on the difference distance. Specifically:

[0135] In the embodiments of the present invention, the maximum value of the differences between the corresponding frequency bands of two time-frequency domains is calculated as the difference distance; and the distance matrix between any two time-frequency domains is calculated using the following formula:

[0136]

[0137] where and are respectively the target EEG signal sets corresponding to the embedding dimension m, the full frequency band at specific time points i and j; and are respectively the target EEG signal sets corresponding to each frequency band at specific time points i and j; 1 < i < n - m + 1; 1 < j < n - m + 1; i ≠ j; H represents the Alpha frequency band, Beta frequency band or Theta frequency band.

[0138] Taking i = 1 and j = 2 as an example, = d ={max{| - |}, max{| |}, max{| 丨}}。

[0139] According to Equation (2), max{| - 丨}=max{| - 丨,丨 - 丨,...,丨 - 丨}。

[0140] Thus, the embodiments of the present invention generate a total of n - m + 1 distance matrices.

[0141] The feature reconstruction module 106 is used to perform feature reconstruction on each frequency band in the distance matrix to obtain a feature reconstruction sequence. Specifically:

[0142] The embodiments of the present invention represent the distance matrix using the following formula:

[0143]

[0144] where represents the maximum value of the difference between each frequency band at specific time points i and j; H represents the Alpha band, Beta band, or Theta band.

[0145] Further, the distance features of each frequency band in the distance matrix are sorted in ascending order of numerical value to obtain a feature sorting result; where the distance feature is the maximum value of the difference between each frequency band at specific time points i and j 、 、 。

[0146] The feature sorting result is: ≤ ;

[0147] where 、 、 are the subscript index values of each distance feature in the reconstructed component;

[0148] Generate a feature reconstruction sequence according to the feature sorting result:

[0149]

[0150] where is the feature reconstruction sequence corresponding to the distance matrix ; there are a total of 3! permutation ways among 3 frequency bands, so l = 1, 2... K ; K≤3!. It can be understood that each distance matrix in the embodiment of the present invention is a 3-dimensional space, mapped to a 3-dimensional feature reconstruction sequence, and has 3! kinds of arrangement modes.

[0151] The multi-band entropy calculation module 107 is used to calculate the multi-band entropy based on the feature reconstruction sequence. The multi-band entropy is used to perform nonlinear analysis on the EEG signal. Specifically:

[0152] Based on the definition of Shannon entropy, the multi-band entropy in the embodiment of the present invention is defined as:

[0153]

[0154] Therefore, we first calculate the probability of occurrence of different feature reconstruction sequences. The probability of each 3D feature reconstruction sequence is P1, P2, P3...P k The embodiment of the present invention is based on P j Represents the occurrence probability of different feature reconstruction sequences in the target area; 1≤j≤k. It is the multi-band entropy corresponding to the preset total time period.

[0155] It can be understood that when the probability of reconstructing the sequence of each 3D feature is the same, that is, P j =1 / 3!, multi-band entropy The maximum value ln(3!) is obtained. Therefore, in the actual processing, the calculated multi-band entropy is normalized:

[0156] 0≤ ≤ / ln(3!)≤1

[0157] in, The multi-band entropy corresponding to the preset total time period; is the target multi-band entropy corresponding to the preset total time period; 3! is the total number of occurrences of different feature reconstruction sequences.

[0158] It is understandable that multi-band entropy is used to perform nonlinear analysis on EEG signals. By performing nonlinear analysis on EEG signals, the physiological activities of brain nerve cells in the cerebral cortex or on the surface of the scalp can be reflected, which is an effective and recognized tool for recording brain activity. Studies have shown that diseases such as depression, epilepsy, Alzheimer's disease and cognitive impairment are all highly correlated with changes in EEG signals. The nonlinear analysis of EEG signals based on multi-band entropy proposed in the embodiment of the present invention can be applied but not limited to the study of diseases such as depression, epilepsy, Alzheimer's disease and mild cognitive impairment.

[0159] The embodiment of the present invention systematically compares the ability of two classic multivariate entropy methods: Multivariate Permutation Entropy (mvPE) and Multivariate Sample Entropy (mvSE), as well as Multi-Frequency Entropy (mFE) proposed in the embodiment of the present invention to distinguish the EEG signals of normal people and patients.

[0160] Specifically, the resting EEG data is first divided to obtain EEG data of the Alpha band, Beta band and Theta band. Then the divided EEG data is segmented, each segment is 200 time points. Then the time-frequency matrix corresponding to 200 unit time points in 3 frequency bands is calculated by the mvPE method, the mvSE method and the mFE method proposed in the embodiment of the present invention. Finally, an entropy matrix of size 58×x can be obtained, where x is the number of segments. The entropy matrices of normal subjects and subjects with Parkinson's disease are spliced ​​into matrices of size 58×X1 and 58×X2, respectively, where X1 and X2 represent the sum of the number of segments of all subjects in normal people and patients, respectively. Then the entropy matrices corresponding to normal subjects and subjects with Parkinson's disease are trained and adjusted with SVM classifiers, and the parameter adjustment process is the process of selecting the optimal parameters. Finally, the accuracy, specificity and sensitivity of the three methods are calculated based on the test data.

[0161] In the embodiment of the present invention, when the entropy value is calculated by the mvPE, mvSE and mFE methods respectively, the dimension m of the phase space is taken as 4; the sample entropy parameter r is taken as 0.15SD to obtain more reasonable statistical characteristics. The parameter selection range based on the SVM classifier is: the penalty parameter C value range C=2^C range, the RBF kernel function parameter gamma=2^gamma_range, where C_range=[-5:2:15], and the gamma range value range=[-15:2:3]. Based on the above calculation method and parameter settings, the accuracy, specificity and sensitivity of the entropy value calculated based on the mvPE, mvSE and mFE methods are obtained, as shown in Table 1 below:

[0162] Table 1

[0163]

[0164] By comparison, the mFE method proposed in the embodiment of the present invention has higher accuracy, specificity and sensitivity than the two classic multivariate entropy methods (mvPE and mvSE).

[0165] Although the present invention has been described with reference to the current preferred embodiments, those skilled in the art should understand that the above-mentioned preferred embodiments are only used to illustrate the present invention and are not used to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A nonlinear analysis method for EEG signals, characterized in that: The method comprises the following steps: S1, obtaining a resting-state EEG signal within a preset total time period, preprocessing the resting-state EEG signal, and obtaining a target EEG signal; S2, dividing the target EEG signal into different frequency bands to obtain a target EEG signal set corresponding to multiple frequency bands; S3, constructing a time-frequency matrix based on the target EEG signal set corresponding to each frequency band; S4, using the time-frequency matrix to construct multiple time-frequency domains containing the EEG signals of some sub-targets corresponding to each frequency band at a specific time point; S5, respectively calculating the difference distance between each two time-frequency domains, and constructing a distance matrix between the two time-frequency domains based on the difference distance; S6, reconstructing the features of each frequency band in the distance matrix to obtain a feature reconstruction sequence; S7, calculating multi-band entropy based on feature reconstruction sequence; The step S4 comprises: Based on the time-frequency matrix, a time-frequency domain containing the Alpha band, Beta band, and Theta band in the preset sub-time period n is constructed. ; where i is a specific time point, 1≤i≤n-m+1, that is for : (1) (2) Based on the above formulas (1) and (2), we can obtain: Wherein, m is the preset embedding dimension; 1≤n≤N; N is the preset total time period; 1≤i≤n-m+1; It refers to the target EEG signal set corresponding to each frequency band at a specific time point i; H represents the Alpha band, Beta band or Theta band; when H is A, Refers to the target EEG signal set corresponding to the Alpha frequency band at a specific time point i; when H is B, Refers to the target EEG signal set corresponding to the Beta frequency band at a specific time point i; when H is T, Refers to the target EEG signal set corresponding to the Theta frequency band at a specific time point i; The step S5 comprises: The maximum value of the difference between the corresponding frequency bands of the two time-frequency domains is calculated as the difference distance; and the distance matrix between any two time-frequency domains is calculated using the following formula (3): （3） in, , are the target EEG signal sets corresponding to the embedding dimension m and the full frequency band at specific time points i and j respectively; , are the target EEG signal sets corresponding to the Alpha frequency band at specific time points i and j respectively; , are the target EEG signal sets corresponding to the Beta frequency band at specific time points i and j respectively; , are the target EEG signal sets corresponding to theta frequency band at specific time points i and j respectively; 1<i<n-m+1; 1<j<n-m+1; i≠j; The step S6 comprises: The distance matrix is ​​expressed using the following formula (4): in, Represents the maximum value of the difference between the Alpha band at a specific time point i and j; Represents the maximum value of the difference between the Beta frequency band at a specific time point i and j; Represents the maximum value of the difference between the Theta frequency band at a specific time point i and j; Arrange the distance matrix in ascending order according to the value The distance feature of each frequency band in the , and the feature sorting result is obtained; where the distance feature is the maximum value of the difference between each frequency band at a specific time point i and j , , ; The feature sorting results are: ≤ ; in, , , is the subscript index value of each distance feature in the reconstructed component; Generate a feature reconstruction sequence based on the feature sorting results: in, is the distance matrix The corresponding feature reconstruction sequence; there are 3! ways to arrange the three frequency bands, so l =1,2... K ; K ≤3!.

2. The method according to claim 1, characterized in that The preprocessing includes: channel positioning, removing useless electrodes, filtering, segmenting, removing bad segments and replacing bad conductors, removing ICA artifact components, removing extreme value artifacts and re-referencing.

3. The method according to claim 2, characterized in that The step S2 comprises: The multi-band decomposition method is used to divide the target EEG signal into sub-target EEG signals corresponding to the Alpha band, Beta band, and Theta band: ； ； ； in, , , are the target EEG signals corresponding to the Alpha band, Beta band, and Theta band at the Nth time point; , , They are the target EEG signal sets corresponding to the Alpha band, Beta band and Theta band in the preset total time period N.

4. The method according to claim 3, characterized in that The step S3 comprises: The time-frequency matrix is ​​constructed based on the target EEG signals corresponding to the preset total time period N in the Alpha band, Beta band, and Theta band: in, Represents the time-frequency matrix composed of the target EEG signals within the preset total time period N.

5. The method according to claim 4, characterized in that The step S7 comprises: Based on the definition of Shannon entropy, the multi-band entropy is defined as: Calculate the probability of occurrence of different feature reconstruction sequences. The probability of each 3D feature reconstruction sequence is P1, P2, P3...P k , 1≤t≤k, The multi-band entropy corresponding to the preset total time period; Normalize the calculated multi-band entropy: 0≤ ≤ / ln(3!)≤1 in, The multi-band entropy corresponding to the preset total time period; is the target multi-band entropy corresponding to the preset total time period; 3! is the total number of occurrences of sequences reconstructed with different features.

6. A nonlinear analysis system for EEG signals using the nonlinear analysis method for EEG signals according to claim 1, characterized in that: The system includes a data preprocessing module, a signal frequency division module, a time-frequency matrix construction module, a time-frequency domain construction module, a distance calculation module, a feature reconstruction module and a multi-band entropy calculation module, wherein: The data preprocessing module is used to obtain the resting-state EEG signals within a preset total time period, preprocess the resting-state EEG signals, and obtain the target EEG signals; The signal frequency division module is used to divide the target EEG signal to obtain a target EEG signal set corresponding to multiple frequency bands; The time-frequency matrix construction module is used to construct a time-frequency matrix based on the target EEG signal set corresponding to each frequency band; The time-frequency domain construction module is used to construct multiple time-frequency domains containing part of the sub-target EEG signals corresponding to each frequency band at a specific time point using the time-frequency matrix; The distance calculation module is used to respectively calculate the difference distance between each two time-frequency domains, and construct a distance matrix between the two time-frequency domains based on the difference distance; The feature reconstruction module is used to perform feature reconstruction on each frequency band in the distance matrix to obtain a feature reconstruction sequence; The multi-band entropy calculation module is used to calculate the multi-band entropy based on the feature reconstruction sequence.

Images