Visual task brain function network construction method under synchronization theory

Abstract

The invention discloses a visual task brain function network construction method under a synchronization theory, and belongs to the technical field of neuroscience and brain function imaging. According to the technical scheme adopted by the invention, the problems of signal acquisition and analysis in brain function network research in the prior art are effectively solved. By processing the high-resolution EEG time sequence, the community structure of the brain function network in the visual task and the relation between the community structure and the anatomical structure are successfully revealed, a brand new view angle and a theoretical basis are provided for deep understanding of a brain visual cognition mechanism, and development of basic research of neuroscience is promoted. The SAX algorithm is adopted for similarity measurement, a clustering method based on the synchronization theory is combined, large-scale EEG data can be efficiently processed, key information is reserved while the data dimension is reduced, and the efficiency and accuracy of data analysis are improved. The innovative data processing mode provides a reliable technical means for subsequent research, and also has important reference significance in the research of processing other similar time sequence data.

Description

Technical Field

[0001] The present invention belongs to the technical field of neuroscience and brain function imaging, and specifically relates to a method and a related system for revealing the hierarchical structure of brain functional networks in visual tasks by in-depth analysis of electroencephalogram (EEG) data with the help of synchronization theory. Background Art

[0002] In neuroscience research, the study of the human brain connectome has always been a very challenging topic. The human brain is composed of tens of billions of neurons and a large number of synaptic connections, and is an extremely complex system. With the continuous development of brain anatomy, people have gradually realized that the neuronal elements of the brain form a highly complex structural network that supports a variety of cognitive functions and neural activities. The study of brain functional connectivity, that is, measuring the temporal correlation between different brain regions and the statistical dependence of functional activities, helps to reveal the collaborative working mode between brain regions, and has become an important research direction in the field of neuroscience.

[0003] At present, researchers often use advanced technologies such as electroencephalography (EEG), magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI) to record brain neuron activity signals, abstract brain functional networks into complex networks, and use graph-based analysis methods to understand the internal working mechanisms of the brain. A large number of studies have shown that brain functional networks have small-world characteristics and scale-free characteristics, but the small-world characteristics only reflect the overall structure of the network and do not reveal enough specific information within the community in the network. In addition, although previous studies have constructed functional networks in the resting state of the human brain and revealed their modular structure, the hierarchical structure of brain functional networks in visual tasks is still not deeply studied.

[0004] Synchronization plays a key role in information processing in the cerebral cortex. However, existing research methods fail to make full use of synchronization theory to deeply explore the community structure of brain functional networks when analyzing brain functional networks, which in turn affects the comprehensive understanding of brain functional network connectivity. Therefore, a new technical solution is needed to more deeply reveal the hierarchical structure of brain functional networks in visual tasks based on synchronization theory.

[0005] When processing these signals, traditional analysis methods are unable to deeply explore the community structure and hierarchical relationships of the brain's functional network. The commonly used methods based on graph theory to analyze brain functional networks can reveal some overall characteristics of the network, such as small-world characteristics and scale-free characteristics, but lack in-depth exploration of the detailed structure of the community within the network and the hierarchical relationships between different communities. When studying the relationship between brain function and structure, it is impossible to accurately reveal how the community structure of the brain's functional network corresponds to the anatomical structure, and it is difficult to clarify the collaborative working mode and functional hierarchy of different brain regions in visual tasks, which limits the in-depth understanding of the brain's visual cognitive mechanism, and thus affects the development of related fields such as the diagnosis and treatment of neurological diseases, and the development of artificial intelligence visual systems. On the other hand, most of the current research work is about brain functional connectivity in the "resting state", while there is very little work using data-driven nonlinear dynamics methods based on synchronization theory to study the functional connectivity of brain networks in the "task state". Because EEG signals have nonlinear characteristics, the use of conventional linear analysis methods has obvious defects. Contents of the invention

[0006] The present invention solves the problem that the prior art cannot accurately reveal how the community structure of the brain functional network corresponds to the anatomical structure, and it is difficult to clarify the collaborative working mode and functional hierarchy of different brain regions in visual tasks; from the perspective of data mining, a data-driven analysis method based on synchronization theory is used to obtain the functional connectivity of the subject's brain network under visual tasks.

[0007] The technical solution of the present invention is a method for constructing a brain functional network for visual tasks under synchronization theory, the method comprising:

[0008] Step 1: Collect data and pre-process the collected data;

[0009] Step 2: Calculate the similarity between the signals collected by different electrodes;

[0010] Step 3: Clustering based on similarity;

[0011] Given a set of discretized data Each data point The ε-similarity neighborhood Nb ε is defined as:

[0012]

[0013] in is the similarity measurement function; after obtaining the ε similarity neighborhood, the data point Each dimension of is dynamically evolved according to formula (5);

[0014]

[0015] in and Respectively The ε similarity neighborhood and its size, ω i，k represents the original frequency of the k-th dimension of the ith data point, x i，k represents the k-th dimension data of the ith data point; let dt = Δt, and the discrete form of equation (5) is obtained as:

[0016]

[0017] Simplifying the constant term Δt*K to 1, we get equation (6):

[0018]

[0019] The synchronization level during the evolution process is characterized by the local order parameter r, which is given by:

[0020]

[0021] Where N represents the number of all data points, the value of r gradually tends to 1, and the dynamic clustering process stops immediately;

[0022] Step 4: Divide the clustering results of step 3 into visual area, visual temporal lobe area, cognitive area, action area, and perception area. According to the positions of the corresponding electrodes, the visual area, visual temporal lobe area, cognitive area, action area, and perception area corresponding to the collector's head are obtained.

[0023] Furthermore, the specific method of step 1 is:

[0024] High-resolution EEG time series were recorded through 238 scalp electrodes at a sampling rate of 256 Hz; the subject's screen presented 5 horizontally arranged blank squares with a fixation point in the upper center, highlighted squares of different colors indicating the focus position, and white disks appeared randomly in the squares. The subject was required to press a response button when the disk appeared in the focus position.

[0025] Furthermore, the method of step 2 is:

[0026] Step 2.1: Regularize the signal collected by each electrode;

[0027] Step 2.2: The regularized signal is subjected to dimension reduction by PAA, where PAA stands for piecewise aggregate approximation.

[0028] A time series of length n In w-dimensional space, it is represented as The i-th element in Calculated by the following formula:

[0029]

[0030] Step 2.3: Discretization;

[0031] Map the time series after PAA dimensionality reduction to the corresponding letters and finally discretize them into strings

[0032]

[0033]

[0034] represents the jth letter of the English alphabet; β j It means that the regularized time series generates multiple intervals with equal areas under the Gaussian curve, where the endpoints of the jth interval are.

[0035] Step 2.4: Calculate similarity;

[0036]

[0037] Among them, the value of the dist() function is obtained by looking up the table. Each corresponds to a discretization of the signal collected by an electrode as a character string; Represents the i-th character in the corresponding string.

[0038] Furthermore, in the dist() function of step 2.4, n is set to 7 and w is set to 8.

[0039] The technical solution adopted by the present invention effectively solves the problem of signal acquisition and analysis in the existing technology in the study of brain functional networks. By processing high-resolution EEG time series, the community structure of the brain functional network in visual tasks and its relationship with the anatomical structure are successfully revealed, which provides a new perspective and theoretical basis for in-depth understanding of the brain's visual cognitive mechanism and promotes the development of basic research in neuroscience. The SAX algorithm is used for similarity measurement, combined with a clustering method based on synchronization theory, which can efficiently process large-scale EEG data, reduce the data dimension while retaining key information, and improve the efficiency and accuracy of data analysis. This innovative data processing method provides a reliable technical means for subsequent research, and is also of great reference significance in the research of processing other similar time series data. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The flowchart of the algorithm.

[0041] Figure 2 It is the labeling result of each activation area detected by the algorithm and the coupling degree of the corresponding functional partition.

[0042] Figure 3 The figure shows the 3D head model fitted according to the activation areas detected by the algorithm. DETAILED DESCRIPTION

[0043] When analyzing the preprocessed EEG data, each EEG channel is regarded as an oscillator, and the clustering method based on synchronization theory is used to obtain the results. The clustering results are affected by the value of the neighborhood interaction range. By changing the number of nearest neighbor electrodes interacting at each focus position, the influence of the parameters on the results is studied, and finally the parameter value when the change tends to be stable is selected.

[0044] The coupling degree between each electrode cluster obtained by the algorithm and the functional substructures of different anatomical divisions was calculated, and the functional substructure with the highest coupling degree was selected and annotated on the corresponding electrode cluster, and the coupling degree value was recorded. The results showed that there was significant consistency between the functional activation areas detected by the algorithm and the corresponding anatomical substructures, especially with the functional substructures of anatomical divisions such as cognition, vision, visual temporal lobe, and action. This shows that when completing the visual selective attention experiment, these anatomical substructures work closely together, further verifying the effectiveness of the algorithm in revealing the relationship between brain function and structure, and deepening the understanding of the relationship between brain function and structure. The clusters detected by the algorithm were displayed on a 3D head model based on the fitting of the electrode spatial position to intuitively present the clustering results. By observing the 3D head model at different attention positions and analyzing the distribution of each functional substructure, the effectiveness of the algorithm and the characteristics of the brain functional network were further verified.

[0045] The technical solution adopted by the present invention to solve the technical problem is divided into the following steps:

[0046] 1. Data collection and preprocessing: This paper uses the data collected by A. Delorme et al. in the visual spatial selective attention task at the SWARTZ Center of the University of California, San Diego, and uses the Bio-semi Active Two system to record high-resolution EEG time series at a sampling rate of 256Hz through 238 scalp electrodes. In the experiment, the subject's screen was presented with 5 horizontally arranged blank squares, with a fixation point in the center, and squares highlighted in different colors to indicate the focus position. White discs appeared randomly in the squares, and the subject had to press the response button when the disc appeared in the focus position; the experimental records were divided into 5 data sets;

[0047] 2. Similarity measurement: Appropriate similarity measurement plays a vital role in clustering tasks; it needs to accurately quantify the similarities between all electrode pairs in a systematic way to capture subtle correlations; the symbolic aggregate approximation algorithm (SAX) proposed on the basis of piecewise aggregate approximation (PAA) is a technology for time series data compression and representation; it converts continuous time series data into a series of symbols, thereby reducing the dimension and computational cost of the data and avoiding the occurrence of "dimensionality disaster". At the same time, it also allows the distance metric defined on the symbolic representation to be smaller than the corresponding distance metric defined on the original sequence, thereby retaining the key features of the original data. The main steps are as follows:

[0048] (1) Regularization: The EEG time series is regularized to have a mean of 0 and a standard deviation of 1;

[0049] (2) Dimensionality reduction through PAA: A time series of length n can be represented in w-dimensional space as The i-th element in It can be calculated by the following formula:

[0050]

[0051] (3) Discretization: The regularized time series has a Gaussian distribution, so the "breakpoints" can be determined, resulting in multiple regions of equal area below the Gaussian curve. These breakpoints can be found online, so that the time series processed by PAA dimensionality reduction can be mapped to corresponding letters and finally discretized into a string.

[0052]

[0053] (4) Similarity measurement: Under the guarantee that SAX allows the definition of lower bound distance measurement in the symbol space, the similarity measurement formula between any two time series is obtained:

[0054]

[0055] The value of the dist() function can be obtained by looking up the table. If a small parameter a is selected, it will lead to weak constraints, but when the parameter a takes a larger value, the effect will not be so good. In addition, the study also shows that the parameters are not too critical, and the alphabet size selection in the range of 5 to 8 seems to be a good choice. Therefore, in our study, a is set to 7 and w is set to 8.

[0056] 3. Clustering method based on synchronization theory: Each data point (representing the EEG channel in this new study) is regarded as a phase oscillator.

[0057] Given a set of data points Each data point The ε similarity neighborhood is defined as:

[0058]

[0059] in is the similarity metric function mentioned above. After obtaining the ε similarity neighborhood, the data point Each dimension of can be dynamically evolved according to formula (5).

[0060]

[0061] in and Respectively The ε-similarity neighborhood and its size. Let dt = Δt, and the discrete form of equation (5) can be obtained as:

[0062]

[0063] Given that we lack prior knowledge about the dataset, all oscillators have the same natural frequency at the beginning. It should be emphasized that the second term on the right side of equation (6) is a constant and can be safely omitted because it has no effect on the final clustering result. Therefore, we simplify the constant term Δt*K to 1 to simplify the analysis process. Finally, equation (6) can be simplified to:

[0064]

[0065] The level of synchronization during the evolution of coupled oscillators is characterized by the local order parameter r, which is given by:

[0066]

[0067] Note that r is defined as the degree of local synchronization of the oscillators. As more of the nearest neighbors of each oscillator synchronize together in the time evolution, the value of r gradually tends to 1. This indicates that the oscillators within the cluster have their own phases.

[0068] In general, the algorithm consists of the following steps:

[0069] (1) In the initial state (t = 0), all data points, i.e., phase oscillators, have an initial state {x i,1 (0),x i,2 (0),...,x i,d(0)}, i = 1, 2, ..., N;

[0070] (2) Each phase oscillator is coupled to the electrodes in their ε-similarity neighborhood and evolves according to equation (8). These phase oscillators corresponding to data points with similar properties will synchronize together and form a cluster. It is worth noting that the ε-similarity neighborhood of the phase oscillators is updated simultaneously with their states;

[0071] (3) Finally, the local order parameter is calculated according to equation (8) to characterize the local synchronization process. When r converges, the dynamic clustering process stops.

[0072] It should be noted that the time complexity of each dynamic clustering analysis using this algorithm is O(L×N 2 ), N is the number of EEG channels, and L is the number of steps for r to converge, which is usually between 5 and 20. In order to produce a stable interaction between each oscillator, the value of ε can be dynamically set in each time step according to the average of the k nearest neighbor similarities of each EEG channel.

[0073] 4. Anatomical clustering based on functional zoning of the cerebral cortex: EEG channels were clustered according to the anatomical segmentation scheme of the cerebral cortex of the Brodmann area. The Brodmann template image (resolution 181×217×181, voxel size 1mm×1mm×1mm) limited to the standard MNI space provided by MRIcro was used to map the electrode positions to the MNI space through the SPM8 toolbox. According to the Brodmann area segmentation scheme, each cerebral hemisphere is divided into 41 regions, and the same label indicates that the functional regions of the two hemispheres are the same. Check the Brodmann area to which the electrode belongs, cluster the electrodes in the same area, and mark them with corresponding numbers. 235 electrodes are clustered into 25 functional groups. In addition, according to physiological functions, the Brodmann area is roughly divided into 9 main substructures, which further cluster the EEG channels. In the "visual attention" task, the substructures involved in vision, visual temporal lobe, cognition, action, perception, etc. are fitted to the 3D head model annotation. Although there are errors in model fitting, it can assist analysis.

[0074] 5. Result analysis: The algorithm detection results were visualized in a 3D head model. The results at different attention positions all showed functional substructures involving vision, visual temporal lobe, cognition, and action, proving that these functional substructures work together when performing the "visual selective attention task", intuitively presenting the distribution of functional substructures in the cerebral cortex, and strengthening the understanding of the effectiveness of the algorithm and the characteristics of brain functional networks. The coupling degree between the electrode clusters obtained by the algorithm and the functional substructures of different anatomical partitions was calculated, and the functional substructures with the highest coupling degree were selected and marked on the electrode clusters and the values ​​were recorded. The results showed that the functional activation areas detected by the algorithm were significantly consistent with the corresponding anatomical substructures, especially with the functional substructures of anatomical partitions such as cognition, vision, visual temporal lobe, and action, which verified the effectiveness of the algorithm in revealing the relationship between brain function and structure.

[0075] The effectiveness of the algorithm was verified by comparing the functional activation areas with the results of anatomical segmentation of the cerebral cortex. The algorithm can accurately detect the functional areas of the brain related to visual tasks, and reveals the close connection between these areas in function and anatomical structure, providing potential biomarkers and therapeutic targets for the diagnosis and treatment of neurological diseases, and has important clinical application value. The visualization results intuitively present the structure and functional relationship of the brain functional network, which is convenient for researchers, medical personnel and people in related fields to understand and analyze. It is conducive to communication and cooperation among multiple disciplines, promotes the development of artificial intelligence visual systems, enables them to better simulate the visual cognitive process of the human brain, and improves the performance and application scope of intelligent visual technology.

Claims

1. A method for constructing a brain functional network for visual tasks under synchronization theory, the method comprising: Step 1: Collect data and pre-process the collected data; Step 2: Calculate the similarity between the signals collected by different electrodes; Step 3: Clustering based on similarity; Given a set of discretized data Each data point The ε-similarity neighborhood Nb ε is defined as: in is the similarity measurement function; after obtaining the ε similarity neighborhood, the data point Each dimension of is dynamically evolved according to formula (5); in and Respectively The ε similarity neighborhood and its size, ω i,k represents the original frequency of the k-th dimension of the ith data point, x i,k represents the k-th dimension data of the ith data point; let dt = Δt, and the discrete form of equation (5) is obtained as: Simplifying the constant term Δt*K to 1, we get equation (6): The synchronization level during the evolution process is characterized by the local order parameter r, which is given by: Where N represents the number of all data points, the value of r gradually tends to 1, and the dynamic clustering process stops immediately; Step 4: Divide the clustering results of step 3 into visual area, visual temporal lobe area, cognitive area, action area, and perception area. According to the positions of the corresponding electrodes, the visual area, visual temporal lobe area, cognitive area, action area, and perception area corresponding to the collector's head are obtained.

2. A method for constructing a brain functional network for visual tasks under synchronization theory as claimed in claim 1, characterized in that: The specific method of step 1 is: High-resolution EEG time series were recorded through 238 scalp electrodes at a sampling rate of 256 Hz; the subject's screen presented 5 horizontally arranged blank squares with a fixation point in the upper center, highlighted squares of different colors indicating the focus position, and white disks appeared randomly in the squares. The subject was required to press a response button when the disk appeared in the focus position.

3. The method for constructing a brain functional network for visual tasks under synchronization theory as claimed in claim 1, characterized in that: The method of step 2 is: Step 2.1: Regularize the signal collected by each electrode; Step 2.2: The regularized signal is subjected to dimension reduction by PAA, where PAA stands for piecewise aggregate approximation. A time series of length n In w-dimensional space, it is represented as The i-th element in Calculated by the following formula: Step 2.3: Discretization; Map the time series after PAA dimensionality reduction to the corresponding letters and finally discretize them into strings represents the jth letter of the English alphabet; β j It means that the regularized time series generates multiple intervals with equal areas under the Gaussian curve, where the endpoints of the jth interval are. Step 2.4: Calculate similarity; Among them, the value of the dist() function is obtained by looking up the table. Each corresponds to a discretization of the signal collected by an electrode as a character string; Represents the i-th character in the corresponding string.

4. A method for constructing a brain functional network for visual tasks under synchronization theory as claimed in claim 3, characterized in that: In the dist() function of step 2.4, n is set to 7 and w is set to 8.

Images