Electroencephalogram signal feature extraction method based on artificial intelligence

Abstract

The invention relates to the technical field of electroencephalogram signal feature extraction, and discloses an electroencephalogram signal feature extraction method based on artificial intelligence. An electroencephalogram data frequency band component extraction step; calculating an adjacent matrix of each frequency band; carrying out feature selection on values in the plurality of frequency band adjacent matrixes; calculating an adjacent matrix between every two frequency bands; feature selection is carried out on values in the adjacent matrixes of every two frequency bands; and in the electroencephalogram data preprocessing step, a 50Hz wave trap and independent component analysis are utilized. By constructing a cross-channel electroencephalogram network, accurate extraction of cross-channel electroencephalogram features is realized, a cross-frequency electroencephalogram network is constructed, on this basis, deep mining of cross-frequency electroencephalogram features is further realized, electroencephalogram information of different scales is effectively fused, and by cooperating with a deep neural network, the accuracy of the cross-channel electroencephalogram features is improved. The electroencephalogram activity in the chronic pain state can be analyzed more comprehensively, and the accuracy and efficiency of feature extraction are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram signal feature extraction, and in particular to an electroencephalogram signal feature extraction method based on artificial intelligence. Background Art

[0002] Chronic pain refers to pain that lasts for more than 3 to 6 months. Unlike acute pain, chronic pain is an independent disease that lasts longer than the expected healing time. Chronic pain not only affects patients' daily eating and conversation, but also causes emotional, affective and cognitive distress. Chronic pain can easily induce sleep disorders, insomnia, anxiety, depression and other psychological diseases because it causes pain to patients over a long period of time.

[0003] When treating patients with chronic pain, it is difficult for patients to observe their own mental state in real time when implementing the psychological habit correction program, which makes it impossible to achieve self-monitoring. Therefore, it is necessary to use the calculation of EEG signal features to reflect the patient's pain state. However, the existing EEG signal feature extraction methods only focus on a single EEG channel or brain area, and do not consider the EEG connection characteristics across channels or brain areas. It is impossible to comprehensively measure the entire brain as a network as a whole, and it is difficult to effectively discover the spatial features related to chronic pain. Or it only focuses on a single EEG frequency range, and does not consider the correlation between brain functions reflected by different EEG frequency ranges. It is difficult to effectively discover the frequency domain features related to chronic pain, and thus it is difficult to comprehensively consider different spatiotemporal features to effectively discover the multi-scale EEG features related to chronic pain. Summary of the invention

[0004] The purpose of the present invention is to provide an artificial intelligence-based EEG signal feature extraction method that can effectively solve the problems in the background technology.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] An artificial intelligence-based EEG signal feature extraction method includes an EEG data preprocessing step; an EEG data frequency band component extraction step;

[0007] Calculate the adjacency matrix of each frequency band; perform feature selection on the values ​​in the adjacency matrices of multiple frequency bands;

[0008] The adjacency matrix is ​​calculated between each pair of frequency bands; and feature selection is performed on the values ​​in the adjacency matrix of each pair of frequency bands.

[0009] As a further preferred embodiment of the present invention, the EEG data preprocessing step utilizes a 50HZ notch filter and independent component analysis to preprocess each channel of EEG data collected from chronic pain patients. The 50HZ notch filter eliminates the industrial frequency interference when extracting EEG data of each channel. The independent component analysis separates independent signal sources from the mixed signal and removes artifacts, electrooculograms, and electromyography, while retaining useful components in the EEG signal.

[0010] As a further preferred embodiment of the present invention, in the step of extracting the EEG data frequency band components, a zero-phase digital bandpass filter is used to extract the EEG data components of the ALPHA, BETA, THETA and GAMMA frequency bands from the EEG data of each channel;

[0011] Among them, the ALPHA band represents 8~12HZ, the BETA band represents 13~30HZ, the THETA band represents 4~7HZ, and the GAMMA band represents 30~90HZ.

[0012] As a further preferred embodiment of the present invention, the adjacency matrix of each frequency band is calculated by using the correlation of the signals between channels to represent the distance of the brain network between channels, and the size of the adjacency matrix calculated is C×C, where C is the number of EEG channels;

[0013] In each adjacency matrix, the value of the i-th row and j-th column is calculated by the Pearson product-moment correlation coefficient between the i-th channel EEG and the j-th channel EEG, and the calculation formula is:

[0014]

[0015] Among them, m k，ij represents the value of the i-th row and j-th column in the adjacency matrix of the k-th frequency band, X k，j and X k，j They represent the discrete EEG signals of the ith channel in the kth frequency band and the jth channel in the kth frequency band, respectively. The matrix sizes of these discrete EEG signals are both 1×N s , N s is the number of sampling points.

[0016] As a further preferred embodiment of the present invention, when performing feature selection on the values ​​in the adjacency matrix of multiple frequency bands, a correlation that is too small is regarded as a distance between two channels that is too far and leads to no connection. Therefore, the correlation distance between the discrete EEG signals of the ith channel and the jth channel of the kth frequency band used for subsequent feature fusion is:

[0017]

[0018] Among them, m k， thRepresents the threshold value of the kth frequency band for filtering channels that are too far away. In addition, because the adjacency matrix constructed by this correlation calculation is a diagonal matrix and all the elements on the diagonal are 1, only the upper half-corner matrix value of the adjacency matrix of each frequency band is taken as the feature.

[0019] As a further preferred embodiment of the present invention, when calculating the adjacency matrix between two frequency bands, the correlation between the two channel signals between the frequency bands is used to represent the distance of the brain network between the frequency bands. The size of the calculated adjacency matrix is ​​also C×C. For the adjacency matrix between the qth frequency band and the pth frequency band, the value of the i-th row and the j-th column is calculated by the Pearson product-moment correlation coefficient between the i-th channel EEG of the qth frequency band and the j-th channel EEG of the pth frequency band. The calculation formula is:

[0020]

[0021] Among them, m pq，ij represents the value of the i-th row and j-th column in the adjacency matrix between the q-th frequency band and the p-th frequency band, X p，i and X q，j They represent the discrete EEG signals of the i-th channel in the p-th frequency band and the discrete EEG signals of the j-th channel in the q-th frequency band respectively.

[0022] As a further preferred embodiment of the present invention, when performing feature selection on the values ​​in the pairwise frequency band adjacency matrix, a correlation that is too small is regarded as a distance between two channels that is too far and leads to no connection. Therefore, the correlation distance between the discrete EEG signal of the ith channel of the pth frequency band and the discrete EEG signal of the jth channel of the qth frequency band used for subsequent feature fusion is:

[0023]

[0024] Among them, m pq，th Represents the threshold used to filter channels that are too far away in the adjacency matrix between the qth frequency band and the pth frequency band.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] In the present invention, by constructing a cross-channel EEG network, accurate extraction of cross-channel EEG features is achieved, and a cross-frequency EEG network is constructed. On this basis, deep mining of cross-frequency EEG features is further achieved, and EEG information of different scales is effectively integrated. By cooperating with a deep neural network, EEG activity in chronic pain states can be more comprehensively analyzed, significantly improving the accuracy and efficiency of feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The figure is a schematic diagram of the EEG signal feature extraction of the present invention. DETAILED DESCRIPTION

[0028] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0029] like Figure 1 The present invention provides an artificial intelligence-based EEG signal feature extraction method, which includes an EEG data preprocessing step; an EEG data frequency band component extraction step; calculating an adjacency matrix for each frequency band; performing feature selection on values ​​in the adjacency matrices of multiple frequency bands; calculating an adjacency matrix between two frequency bands; and performing feature selection on values ​​in the adjacency matrices of two frequency bands.

[0030] The EEG data preprocessing step uses a 50 Hz notch filter and independent component analysis to preprocess each channel of EEG data collected from chronic pain patients. The 50 Hz notch filter eliminates the power frequency interference when extracting EEG data from each channel. Independent component analysis separates independent signal sources from the mixed signal and removes artifacts, electrooculograms, and electromyography, while retaining useful components in the EEG signal.

[0031] When calculating the adjacency matrix for each frequency band:

[0032] The adjacency matrix of each frequency band is calculated by using the correlation of inter-channel signals to represent the distance of the brain network between channels. The size of the calculated adjacency matrix is ​​C×C, where C is the number of EEG channels. In each adjacency matrix, the value of the i-th row and j-th column is calculated by the Pearson product-moment correlation coefficient between the i-th channel EEG and the j-th channel EEG. The calculation formula is: Among them, m k，ij represents the value of the i-th row and j-th column in the adjacency matrix of the k-th frequency band, X k，i and X k，j They represent the discrete EEG signals of the ith channel in the kth frequency band and the jth channel in the kth frequency band, respectively. The matrix sizes of these discrete EEG signals are both 1×N s , N s is the number of sampling points; when performing feature selection on the values ​​in the adjacency matrix of multiple frequency bands, a correlation that is too small is regarded as the distance between the two channels being too far and resulting in no connection. Therefore, the correlation distance between the discrete EEG signals of the ith channel and the jth channel of the kth frequency band used for subsequent feature fusion is: Among them, m k，th Represents the threshold value of the kth frequency band for filtering channels that are too far away. In addition, because the adjacency matrix constructed by this correlation calculation is a diagonal matrix and all the elements on the diagonal are 1, only the upper half-corner matrix value of the adjacency matrix of each frequency band is taken as the feature.

[0033] When calculating the adjacency matrix between two frequency bands, the correlation between the two channel signals between the frequency bands is used to represent the distance of the brain network between the frequency bands. The size of the calculated adjacency matrix is ​​also C×C. For the adjacency matrix between the qth frequency band and the pth frequency band, the value of the i-th row and the j-th column is calculated by the Pearson product-moment correlation coefficient between the i-th channel EEG of the qth frequency band and the j-th channel EEG of the pth frequency band. The calculation formula is: Among them, m pq，ij represents the value of the i-th row and j-th column in the adjacency matrix between the q-th frequency band and the p-th frequency band, X p，i and X q，j They represent the discrete EEG signals of the ith channel in the pth frequency band and the jth channel in the qth frequency band respectively;

[0034] When performing feature selection on the values ​​in the pairwise frequency band adjacency matrix, a correlation that is too small is considered to be that the distance between the two channels is too far and thus there is no connection. Therefore, the correlation distance between the discrete EEG signals of the ith channel of the pth frequency band and the jth channel of the qth frequency band used for subsequent feature fusion is:

[0035] Among them, m pq，th Represents the threshold used to filter channels that are too far away in the adjacency matrix between the qth frequency band and the pth frequency band.

[0036] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for extracting EEG signal features based on artificial intelligence, characterized in that: It includes the steps of preprocessing EEG data and extracting frequency band components of EEG data; Calculate the adjacency matrix of each frequency band; perform feature selection on the values ​​in the adjacency matrices of multiple frequency bands; The adjacency matrix is ​​calculated between each pair of frequency bands; and feature selection is performed on the values ​​in the adjacency matrix of each pair of frequency bands.

2. The method for extracting EEG signal features based on artificial intelligence according to claim 1, characterized in that: The EEG data preprocessing step uses a 50 Hz notch filter and independent component analysis to preprocess each channel of EEG data collected from chronic pain patients.

3. The method for extracting EEG signal features based on artificial intelligence according to claim 2, characterized in that: The 50 Hz notch filter eliminates power frequency interference when extracting EEG data from each channel. The independent component analysis separates independent signal sources from mixed signals and removes artifacts, electrooculograms, and electromyography, while retaining useful components in the EEG signal.

4. The method for extracting EEG signal features based on artificial intelligence according to claim 1, characterized in that: The step of extracting the frequency band components of the EEG data uses a zero-phase digital bandpass filter to extract the EEG data components of the alpha, beta, theta, and gamma frequency bands from the EEG data of each channel during the process of extracting the frequency band components of the EEG data; Among them, the alpha band represents 8~12HZ, the beta band represents 13~30HZ, the theta band represents 4~7HZ, and the gamma band represents 30~90HZ.

5. The method for extracting EEG signal features based on artificial intelligence according to claim 1, characterized in that: The adjacency matrix of each frequency band is calculated by using the correlation of the signals between channels to represent the distance of the brain network between channels. The size of the adjacency matrix calculated is C×C, where C is the number of EEG channels. In each adjacency matrix, the value of the i-th row and j-th column is calculated by the Pearson product-moment correlation coefficient between the i-th channel EEG and the j-th channel EEG, and the calculation formula is: Among them, m k，ij represents the value of the i-th row and j-th column in the adjacency matrix of the k-th frequency band, X k，i and X k，j They represent the discrete EEG signals of the ith channel in the kth frequency band and the jth channel in the kth frequency band, respectively. The matrix sizes of these discrete EEG signals are both 1×N s , N s is the number of sampling points.

6. The method for extracting EEG signal features based on artificial intelligence according to claim 5, characterized in that: When performing feature selection on the values ​​in the adjacency matrix of multiple frequency bands, a correlation that is too small is regarded as a lack of connection between two channels due to the distance between them being too far. Therefore, the correlation distance between the discrete EEG signals of the ith channel and the jth channel of the kth frequency band used for subsequent feature fusion is: Among them, m k，th Represents the threshold value of the kth frequency band for filtering channels that are too far away. In addition, because the adjacency matrix constructed by this correlation calculation is a diagonal matrix and all the elements on the diagonal are 1, only the upper half-corner matrix value of the adjacency matrix of each frequency band is taken as the feature.

7. The method for extracting EEG signal features based on artificial intelligence according to claim 1, characterized in that: When calculating the adjacency matrix between two frequency bands, the correlation between the two channel signals between the frequency bands is used to represent the distance of the brain network between the frequency bands. The size of the calculated adjacency matrix is ​​also C×C. For the adjacency matrix between the qth frequency band and the pth frequency band, the value of the i-th row and j-th column is calculated by the Pearson product-moment correlation coefficient between the i-th channel EEG of the qth frequency band and the j-th channel EEG of the pth frequency band. The calculation formula is: Among them, m pq，ij represents the value of the i-th row and j-th column in the adjacency matrix between the q-th frequency band and the p-th frequency band, X p，i and X q，j They represent the discrete EEG signals of the i-th channel in the p-th frequency band and the discrete EEG signals of the j-th channel in the q-th frequency band respectively.

8. The method for extracting EEG signal features based on artificial intelligence according to claim 7, characterized in that: When performing feature selection on the values ​​in the pairwise frequency band adjacency matrix, a correlation that is too small is considered to be that the distance between the two channels is too far and thus there is no connection. Therefore, the correlation distance between the discrete EEG signals of the ith channel of the pth frequency band and the jth channel of the qth frequency band used for subsequent feature fusion is: Among them, m pq，th Represents the threshold used to filter channels that are too far away in the adjacency matrix between the qth frequency band and the pth frequency band.

Images