A method for implementing a brain-computer interface based on microstates of brain electrical functional connectivity

Abstract

The present invention belongs to the technical field of biomedical signal processing and analysis, and proposes a brain-computer interface implementation method based on EEG functional connection microstates, aiming to solve the problem of sacrificing time accuracy in the traditional time-frequency analysis process, introduce microstate sequences, construct functional brain networks, represent the characteristics of brain-computer interaction in different dimensions, and use a different angle from traditional time-frequency analysis to parse EEG signals, which can not only retain the global measurement results of instantaneous brain activity with high time resolution, but also obtain information on the macroscopic spatiotemporal organization of the brain, thereby improving the information decoding and classification recognition effects of the brain-computer interface system, in order to improve the utilization rate of EEG information and the accuracy of classification.

Description

Technical Field

[0001] The present invention belongs to the technical field of biomedical signal processing and analysis, and mainly proposes a brain-computer interface based on EEG functional connection microstates based on EEG signals. Background Art

[0002] As a key method of brain-like research, the brain-computer interface (BCI) has potential application value in the fields of sports rehabilitation, neural regulation, and human-computer collaboration. It refers to a direct connection between the human or animal brain and external devices to achieve information exchange between the brain and the device. Specifically, it records the electrical signals of brain activity, analyzes and processes them, obtains characteristic values ​​that can be recognized by the machine and outputs them, thereby controlling external devices.

[0003] Brain-computer interface technology is indeed narrowing the distance between the human brain and computers, and has practical significance in providing medical support and other assistance to humans: for example, in 2019, a patient used his mind to control a robotic arm to drink beer with the help of the California Institute of Technology.

[0004] The processing of electroencephalogram (EEG) signals is a key method to obtain real-time information on the function of large-scale neural networks with high temporal resolution. Traditionally, EEG analysis of each channel is mainly transformed into the time-frequency domain through time-frequency analysis, etc., in order to extract features, which relies on the power changes of EEG signals in different frequency bands. However, most EEG signals are difficult to decode, especially EEG features related to high-level cognitive activities. Due to the principle of time-frequency uncertainty, frequency analysis without time accuracy will lose time domain information. In order to explain the short-term and continuous fluctuations in neuronal activity, it is necessary to take into account the temporal evolution of the scalp electric field topography in the time domain analysis method, because it represents the sum of all temporarily active sources in the brain without considering the frequency domain characteristics.

[0005] Therefore, we urgently need a new method of EEG information processing to make up for the sacrifice of time accuracy. Summary of the invention

[0006] In response to the problems and challenges of the existing technology, the present invention proposes a brain-computer interface based on EEG functional connection microstates. This study aims to solve the problem of sacrificing time accuracy in the traditional time-frequency analysis process, introduce microstate sequences, construct a functional brain network, represent the characteristics of brain-computer interaction in different dimensions, and use a completely different angle from traditional time-frequency analysis to parse EEG signals, in order to improve the utilization rate of EEG information and the accuracy of classification.

[0007] The technical solution of the present invention:

[0008] A method for implementing a brain-computer interface based on microstates of EEG functional connections, comprising the following steps:

[0009] Step 1: According to the international standard 10-20 EEG electrode positioning system, the user's scalp EEG signal is recorded, and the user is subjected to a 2-category left-right hand movement imagery experiment; then the scalp EEG signal is preprocessed, including removing 50 Hz power frequency interference, bandpass filtering, removing artifacts based on independent component analysis, and extracting the EEG segments for left-hand movement imagery and right-hand movement imagery respectively;

[0010] Step 2: Classify the microstates of the EEG segments obtained for leftward motor imagery and rightward motor imagery respectively;

[0011] (2.1) Global Field Power (GFP) is a measurement of scalp potential strength. It is based on the potential difference between all electrodes at each sampling point to obtain a scalar value of the field strength at each sampling point. It represents the instantaneous electric field strength of the brain and is used to describe rapid changes in brain activity.

[0012]

[0013] Where i represents each electrode, n represents the total number of electrodes, and u represents the voltage value measured by the electrode;

[0014] (2.2) Using the clustering algorithm of the modified K-means algorithm to calculate the microstates of EEG; extracting the GFP peak and using the modified K-means algorithm to derive four types of brain topography that can maximally explain the variance of the brain topography at the GFP peak, namely microstates A, B, C, and D; where the brain topography is a plane figure developed from a spherical scalp with different colors representing the power of each segment of the EEG;

[0015] (2.3) Finally, the generated microstates A, B, C, and D were used as templates to analyze the EEG spectra of all users at all target moments. The GFP peaks were assigned to one of the four microstates based on the maximum absolute value of the Pearson correlation. Continuous GFP peak maps assigned to the same class were considered to belong to one microstate class.

[0016] Time points between GFP peaks were assigned to the microstate class of the temporally closest GFP peak;

[0017] Step 3: Extract the time series signals of users with the same microstate class, calculate the phase lag index PLI between the EEG data of each channel, and obtain the adjacency matrix; extract the network statistical feature quantities: clustering coefficient, global efficiency, characteristic path length and comprehensive index;

[0018] (3.1) The PLI (Phase Lag Index, PLI) value reflects the consistency of the lead or lag of two EEG recording electrodes, i.e., the nodes of the network, from the perspective of phase. The PLI value between the EEG signals of any two nodes i and j is calculated by the following formula:

[0019]

[0020] in, is the instantaneous phase difference between nodes i and j at time k, indicating the sign function; 〈·〉 indicates taking the average value in the time domain; when hour, This means that the phases of the two nodes are synchronized; if hour, This means that the phases of the two nodes are not synchronized; therefore, the value range of PLI is between 0 and 1. The larger the PLI value, the stronger the phase coupling between the two nodes.

[0021] (3.2) The calculated PLI values ​​are used as elements to construct the connection matrix A, thereby obtaining the connection matrix under each GFP peak; then the following four characteristic parameters are calculated to quantify the brain network;

[0022] Clustering coefficient C i : It is the ratio of the number of connections between the nearest neighbor nodes of the selected node to the number of possible connections between the selected node and the nearest neighbor nodes; the value of the clustering coefficient is between 0 and 1, and the formula is as follows:

[0023]

[0024] Among them, E i represents the number of edges connecting the adjacent nodes of node i, k i Represents the number of all connections on node i;

[0025] Path length L: the average value of the shortest paths between all node pairs; the shortest path length d(i,j) is defined as the minimum number of connections that must be traversed from node i to node j; when d(i,j)≥1, if node i to node j are adjacent nodes, d(i,j)=1; if node i to node j are not connected, d(i,j)→∞; the path length reflects the functional dispersion of node pairs in the network. The smaller L is, the better the integration is, and the higher the information transmission efficiency is. The calculation formula is as follows:

[0026]

[0027] Among them, d(i,j) is the shortest path length, N is the number of all nodes in the network, and V is the set of all nodes;

[0028] Assuming that the information flow is parallel, the communication efficiency between two nodes is inversely proportional to the shortest path length; the efficiency of a group of nodes is the sum of the efficiencies of all node pairs, normalized by the maximum number of links N(N-1) / 2; when the node set V contains all nodes, the third characteristic parameter will be calculated by the following formula, which is the global efficiency E global :

[0029]

[0030] Since the global efficiency is inversely proportional to the characteristic path length, the ratio of the product of the global efficiency and the clustering coefficient to the characteristic path length is taken as the network feature-comprehensive index η(w), E global (w), C(w) and L(w) represent the global efficiency, clustering coefficient and characteristic path length of the sliding window w respectively, and the calculation formula is as follows:

[0031]

[0032] Step 4: Perform validity check on the statistical feature quantity to obtain a feature value that effectively distinguishes leftward and rightward, and classify based on this feature value to control the external device.

[0033] Beneficial effects of the invention: The invention proposes a brain-computer interface based on the microstate of EEG functional connection, which makes use of the advantages of scalp brain topography without considering frequency domain characteristics and can also represent the sum of all temporarily active sources in the brain, making up for the sacrifice of time accuracy. At the same time, the introduction of brain functional connections can not only retain the global measurement results of instantaneous brain activity with high time resolution, but also obtain information on the macroscopic spatiotemporal organization of the brain, thereby improving the information decoding and classification recognition effect of the brain-computer interface system. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of a brain-computer interface based on EEG functional connection microstates of the present invention.

[0035] Figure 2 Computational flow chart for microstate analysis.

[0036] Figure 3 Computational flowchart of functional brain networks. DETAILED DESCRIPTION

[0037] In order to deepen the understanding of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The embodiments do not limit the protection scope of the present invention.

[0038] Reference Figure 1One embodiment of the present invention provides a brain-computer interface based on EEG functional connection microstates. In motor imagery, obtaining the category labels of left and right in real time includes the following steps:

[0039] Step 1: According to the international standard 10-20 EEG electrode positioning system, the user's scalp EEG signal is recorded, and the user is subjected to a 2-category left-right hand motor imagery task experiment. Then the EEG signal is preprocessed, including removing 50Hz power frequency interference, bandpass filtering (0.5-30Hz), removing artifacts based on independent component analysis, and extracting the EEG segments for left and right motor imagery respectively;

[0040] Step 2: Perform microstate analysis on the EEG segments of leftward and rightward motor imagery to obtain GFP peaks, and perform dynamic functional brain network analysis on the data of GFP peak points.

[0041] Step three, calculate the phase lag index between the EEG data of each channel, obtain the adjacency matrix, and extract the network statistical characteristics: clustering coefficient, global efficiency, characteristic path length and comprehensive index.

[0042] Step 4: Perform validity check on the statistical feature quantity to obtain a feature value that can effectively distinguish between left and right, and classify based on this feature value to control the external device.

[0043] Among them, refer to Figure 2 , step 2 of the present invention is described in detail:

[0044] (1) In microstate analysis, a very important concept is global field power (GFP). GFP is a measurement of scalp potential intensity. It is based on the potential difference between all electrodes at each sampling point, thereby obtaining a scalar value of the field strength at each sampling point. It represents the instantaneous electric field strength of the brain and is therefore often used to describe rapid changes in brain activity.

[0045]

[0046] GFP is calculated by taking the average of the squares of the potential differences between all electrodes and their average instantaneous potential and then taking the root, where i represents each electrode, n represents the total number of electrodes, and u represents the voltage value measured by the electrode.

[0047] Since the signal-to-noise ratio of EEG signals is the highest at the GFP peak, the spatial configuration of EEG is considered to be stable, and the instantaneous voltage amplitude of each electrode at the GFP peak time point was selected for cluster analysis.

[0048] (2) Use the modified K-means algorithm to calculate EEG microstates. The K-means algorithm is to first randomly select 4 topographic maps, correlate the 4 topographic maps with the remaining topographic maps, and select the topographic maps that are most similar to the 4 topographic maps. Calculate these similar topographic maps, update them to obtain 4 new topographic maps, and calculate the global explained variance (Global Explained Variance, GEV). With the iteration of the topographic maps, GEV will gradually tend to be flat, and the final 4 topographic maps will be obtained.

[0049] (3) The topographic map sequences of the GFP of the two groups ("leftward motor imagery" and "rightward motor imagery") are clustered into four microstate classes A, B, C, and D. Finally, the generated A, B, C, and D are used as templates to assign the four microstate classes to the original continuous EEG signal sequence of each user to obtain the microstate sequence.

[0050] Among them, refer to Figure 3 , step three of the present invention is described in detail as follows:

[0051] (1) PLI reflects the consistency of the lead or lag of two leads from the perspective of phase. The PLI value between the EEG signals of node i and node j can be calculated by the following formula:

[0052]

[0053] in is the instantaneous phase difference between nodes i and j at time k. The symbol at this time represents a sign function. <·> means taking the average value in the time domain. hour, It means that the phases of the two nodes are synchronized. hour, It means that the phases of the two nodes are not synchronized. Therefore, the value of PLI ranges from 0 to 1. The larger the PLI value, the stronger the phase coupling of the two nodes.

[0054] (2) The calculated PLI values ​​are used as elements to construct the connection matrix A, thereby obtaining the connection matrix under each GFP peak. The following four characteristic parameters are then calculated to quantify the brain network:

[0055] First is the clustering coefficient: the clustering coefficient C of a node i is the ratio of the number of connections between the nearest neighbors of the selected node to the number of possible connections between them, E i represents the number of edges connecting the adjacent nodes of node i, k iRepresents the number of connections between node i and all its adjacent nodes. The clustering coefficient C of the network is the average of the clustering coefficients of all nodes. By definition, the value of the clustering coefficient is between 0 and 1, and the formula is as follows:

[0056]

[0057] The second is the path length (PathLength): the average value of the shortest path between all node pairs. The shortest path length d(i,j) is defined as the minimum number of connections that must be traversed from node i to node j. Therefore, d(i,j)≥1. If i and j are adjacent nodes, d(i,j)=1; if the nodes are not connected, d(i,j)→∞. It reflects the functional dispersion of node pairs in the network. d(i,j) is the shortest path length, N is the number of all nodes in the network, and V is the set of all nodes. The smaller the characteristic path length L, the better the integration and the higher the information transmission efficiency. The calculation formula is as follows:

[0058]

[0059] Assuming that the information flow is parallel, the communication efficiency between two nodes is inversely proportional to the shortest path length. The efficiency of a set of nodes is the sum of the efficiencies of all node pairs, normalized by the maximum number of links N(N-1) / 2. When the node set (V) contains all nodes, the third characteristic parameter will be calculated by the following formula, which is the global efficiency (GlobalEfficiency):

[0060]

[0061] Since the global efficiency is inversely proportional to the characteristic path length, the ratio of the product of the global efficiency and the clustering coefficient to the characteristic path length is taken as the fourth characteristic parameter (Comprehensive Index) of the network feature, E global (w), C(w), and L(w) represent the global efficiency, clustering coefficient, and characteristic path length of the sliding window w (1 second), respectively, and the calculation formula is as follows:

[0062]

[0063] The above describes the basic principles, main features and advantages of the present invention, but the present invention is not limited to the above methods. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that some improvements and substitutions made by technicians familiar with the field without violating the spirit of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for implementing a brain-computer interface based on EEG functional connection microstates, characterized in that: Here are the steps: Step 1: According to the international standard 10-20 EEG electrode positioning system, the user's scalp EEG signal is recorded, and the user is subjected to a 2-category left-right hand movement imagery experiment; then the scalp EEG signal is preprocessed, including removing 50Hz power frequency interference, bandpass filtering, removing artifacts based on independent component analysis, and extracting the EEG segments for left-hand movement imagery and right-hand movement imagery respectively; Step 2: Classify the microstates of the EEG segments obtained for leftward motor imagery and rightward motor imagery respectively; (2.1) Global field strength GFP is a measurement of scalp potential intensity. It is based on the potential difference between all electrodes at each sampling point to obtain a scalar value of the field strength at each sampling point. It represents the instantaneous electric field strength of the brain and is used to describe rapid changes in brain activity. Where i represents each electrode, n represents the total number of electrodes, and u represents the voltage value measured by the electrode; (2.2) Using the clustering algorithm of the modified K-means algorithm to calculate the microstates of EEG; extracting the GFP peak and using the modified K-means algorithm to derive four types of brain topography that can maximally explain the variance of the brain topography at the GFP peak, namely microstates A, B, C, and D; where the brain topography is a plane figure developed from a spherical scalp with different colors representing the power of each segment of the EEG; (2.3) Finally, the generated microstates A, B, C, and D were used as templates to analyze the EEG spectra of all users at all target moments. The GFP peaks were assigned to one of the four microstates based on the maximum absolute value of the Pearson correlation. Continuous GFP peak maps assigned to the same class were considered to belong to one microstate class. Time points between GFP peaks were assigned to the microstate class of the temporally closest GFP peak; Step 3: Extract the time series signals of users with the same microstate class, calculate the phase lag index PLI between the EEG data of each channel, and obtain the adjacency matrix; extract the network statistical feature quantities: clustering coefficient, global efficiency, characteristic path length and comprehensive index; (3.1) The PLI value reflects the consistency of the lead or lag of two EEG recording electrodes, i.e., the nodes of the network, from the perspective of phase. The PLI value between the EEG signals of any two nodes i and j is calculated by the following formula: in, is the instantaneous phase difference between nodes i and j at time k, which represents the sign function; <·> represents the average value in the time domain; when hour, This means that the phases of the two nodes are synchronized; if hour, This means that the phases of the two nodes are not synchronized; therefore, the value range of PLI is between 0 and 1. The larger the PLI value, the stronger the phase coupling between the two nodes. (3.2) The calculated PLI values ​​are used as elements to construct the connection matrix A, thereby obtaining the connection matrix under each GFP peak; then the following four characteristic parameters are calculated to quantify the brain network; Clustering coefficient C i : It is the ratio of the number of connections between the nearest neighbor nodes of the selected node to the number of possible connections between the selected node and the nearest neighbor nodes; the value of the clustering coefficient is between 0 and 1, and the formula is as follows: Among them, E i represents the number of edges connecting the adjacent nodes of node i, k i Represents the number of all connections on node i; Path length L: is the average value of the shortest paths between all node pairs; the shortest path length d(i, j) is defined as the minimum number of connections that must be traversed from node i to node j; when d(i, j) ≥ 1, if node i to node j are adjacent nodes, d(i, j) = 1; if node i to node j are not connected, d(i, j) → ∞; the path length reflects the functional dispersion of node pairs in the network. The smaller L is, the better the integration is, and the higher the information transmission efficiency is. The calculation formula is as follows: Among them, d(i, j) is the shortest path length, N is the number of all nodes in the network, and V is the set of all nodes; Assuming that the information flow is parallel, the communication efficiency between two nodes is inversely proportional to the shortest path length; the efficiency of a group of nodes is the sum of the efficiencies of all node pairs, normalized by the maximum number of links N(N-1) / 2; when the node set V contains all nodes, the third characteristic parameter will be calculated by the following formula, which is the global efficiency E global : Since the global efficiency is inversely proportional to the characteristic path length, the ratio of the product of the global efficiency and the clustering coefficient to the characteristic path length is taken as the network feature-comprehensive index η(w), E global (w), C(w) and L(w) represent the global efficiency, clustering coefficient and characteristic path length of the sliding window w respectively, and the calculation formula is as follows: Step 4: Perform validity check on the statistical feature quantity to obtain a feature value that effectively distinguishes leftward and rightward, and classify based on this feature value to control the external device.

Images