Hopv model-based electroencephalogram transformation simulation system and method for different brain states

Abstract

The invention discloses an electroencephalogram transformation simulation system and method for different brain states based on a Hopv model, and the electroencephalogram transformation simulation system constructs a virtual brain model by using an objective nonlinear dynamic model, carries out the quantitative description of the brain states, and achieves the accurate simulation of the transformation between different brain states and the prediction of a target spot. According to the method, the weighted phase lag index and the similarity of the network attributes of the weighted phase lag index are used as a parameter optimization target when the virtual brain signal is constructed, the electroencephalogram simulation level of the virtual signal is improved, and the target spot in the brain state transition process can be recognized and predicted by extracting the co-activation brain area of the co-activation mode and combining virtual stimulation. According to the method, a quantitative basis is provided for virtual brain model construction and brain state regulation and control, and an individualized nerve regulation and control strategy can be formulated.

Description

Technical Field

[0001] The present invention relates to the technical field of computational neuroscience, and particularly to an electroencephalogram transition simulation system and method for different brain states based on the Hopf model. Background Art

[0002] Brain stimulation such as transcranial magnetic stimulation, direct current stimulation, etc. has been widely used in the treatment of clinical mental diseases. In this process, it is particularly important to find the stimulation targets for brain state transitions. In recent years, the development of neuroscience and computational models has promoted the construction of virtual brain models, which provides new ideas for exploring the complex dynamic processes of the brain. Existing studies have shown that when the brain is performing cognitive tasks or in a pathological state, it will experience transitions between multiple states, and these state transitions reflect the reorganization of the internal neural network of the brain and its functional regulation mechanism. However, most current electroencephalogram (EEG) analysis and brain state modeling methods mainly rely on data-driven algorithms, often ignoring the physiological interpretability of the models, and it is difficult to reveal the subtle changes and internal mechanisms in the early stage of brain state transitions.

[0003] Traditional methods have certain limitations in capturing brain state transitions: some models can only predict long-term stable states and cannot reflect the dynamic adjustments that occur in the early stage of the brain in a timely manner; in addition, these methods lack a systematic description of brain oscillations and nonlinear dynamic characteristics, which limits the accurate positioning of key regulatory targets during the transition process. At the same time, as a non-invasive measurement tool, electroencephalogram signals provide strong support for studying the instantaneous changes of brain states with their high temporal resolution, but how to make full use of this advantage is still an urgent problem to be solved currently.

[0004] Due to its excellent performance in describing the transition of a system from a stable state to a periodic oscillation state, the Hopf model based on nonlinear dynamics theory has gradually become an important tool for simulating brain dynamic behavior. The Hopf model can effectively capture the oscillation characteristics of neural population activities through simple mathematical formulas, providing a theoretical basis for revealing the internal mechanism of brain state transitions. However, applying the Hopf model to electroencephalogram data to achieve the simulation of different brain state transitions and the prediction of regulatory targets still faces technical challenges such as parameter optimization, refinement of state division, and dynamic capture of the transition process. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide an electroencephalogram transition simulation system and method for different brain states based on the Hopf model, which realizes the accurate simulation of transitions between different brain states and the prediction of key targets by quantitatively describing brain states using a nonlinear dynamics model, and solves the deficiencies of existing electroencephalogram state analysis methods in capturing early dynamic changes and target recognition.

[0006] Technical solution: An electroencephalogram transition simulation system for different brain states based on the Hopf model, comprising:

[0007] A signal preparation module, configured to collect, preprocess, and perform source localization on electroencephalogram data of the initial state and the target state, remove noise and artifacts, and extract clean source signals; it includes an electroencephalogram signal acquisition module, a preprocessing and source reconstruction module. The electroencephalogram signal acquisition module: acquires multi-channel EEG signals, ensuring sufficient sampling rate and spatial resolution; the preprocessing and source reconstruction module: filters and removes artifacts from the acquired EEG signals, and performs source localization in combination with a head model and an inversion algorithm to obtain neural activity signals at each brain region level;

[0008] A state feature extraction module, which performs weighted phase lag index network analysis and co-activation pattern analysis on the brain region signals after source reconstruction, extracts brain network feature indicators in different brain states, and provides them for the virtual signal construction module and the state transition prediction module;

[0009] A virtual signal construction module, which constructs a virtual brain model capable of simulating real brain states based on the extracted brain network connection features and neural dynamics models; the virtual brain signals output by the virtual brain model provide a controllable simulation environment for the state transition prediction module;

[0010] A state transition prediction module, which, based on the virtual brain model, evaluates and predicts the evolution process of brain states under different intervention methods, and searches for a combination of target points to achieve brain state transitions.

[0011] Furthermore, the state feature extraction module includes a weighted phase lag index and its network attribute extraction module and a co-activation pattern construction module. The weighted phase lag index and its network attribute extraction module: calculates the functional connection strength in the brain network based on the wPLI method, and further extracts network attributes to quantitatively describe the overall and local connection features of the brain in different states, and is used for parameter optimization in the process of the virtual signal construction module; the co-activation pattern construction module: identifies synchronous activation segments of multi-channel brain signals in the time dimension, extracts co-activation patterns with specific functional significance, and depicts the co-activation features of the brain under different states or task conditions, assisting the state transition prediction module to search for a combination of target points according to the perturbation strategy.

[0012] Furthermore, the virtual signal construction module includes a Hopf model construction module and a parameter optimization module. The Hopf model construction module: regards brain regions as coupled oscillators, and builds a Hopf model as a virtual brain model by setting the parameters of each oscillator; the parameter optimization module: adjusts and optimizes the parameters of the Hopf model based on the electroencephalogram data of the initial state, so that the dynamic behavior of the virtual brain model matches the real EEG signals.

[0013] Furthermore, the state transition prediction module includes a single-site perturbation module and a two-site perturbation module. The single-site perturbation module: Perturb a single brain region and observe the response and state migration of the global network; Two-site perturbation: Perturb two brain regions simultaneously to study the effects of multiple stimuli or co-regulation on brain state transitions and find target combinations.

[0014] An EEG transition simulation method for different brain states based on the Hopf model, which finds the targets for the transition between the initial state and the target state through the EEG transition simulation system as described in any of the above items, includes the following steps:

[0015] S1, Collect the EEG signals of all subjects in the initial state and the target state and perform preprocessing;

[0016] S2, Reconstruct and trace the EEG signal source to the cortical brain region level, calculate the wPLI matrix between brain regions of each subject, and calculate the network properties of the wPLI matrix and the structural connection matrix of each subject;

[0017] S3, Select the region of interest as the seed point, construct a co-activation pattern topological map, and extract the brain regions that are co-activated with the region of interest as sites;

[0018] S4, Construct a Hopf model and construct a virtual signal by solving the Hopf oscillator equation;

[0019] S5, Taking the maximum value of the Pearson correlation coefficient between the wPLI matrix of the virtual signal and the wPLI matrix of the real EEG signal as the optimization goal, use the grid search method to obtain the optimal virtual brain models for the initial state and the test state respectively;

[0020] S6, Based on the optimal virtual brain model, use the perturbation strategy to find the targets that promote brain state transitions and realize the transition from the initial state virtual signal to the target state virtual signal;

[0021] S7, According to the obtained targets, formulate personalized neuroregulation strategies for the subjects.

[0022] Furthermore, the network properties of the weighted phase lag index matrix include global clustering coefficient Clu, characteristic path length L, global efficiency Ge, and local efficiency Le; The process of calculating the wPLI matrix between brain regions is as follows:

[0023] Use the automated anatomical labeling template to divide the brain of the subject into N brain regions, each brain region as a node of the brain network, and then calculate the wPLI matrix between the EEG signals of the N brain regions; The calculation formula of wPLI is:

[0024]

[0025] Among them, wPLI represents the weighted phase lag index, Im(Δφ(t)) represents the imaginary part of the phase difference between two signals at time t; T represents the signal acquisition period; sign(sin(Δφ(t))) is used to take the sign of the sine of the phase difference at time t, either +1 or -1;

[0026] Construct the wPLI matrix M to represent the phase lag connection network. The expression of the wPLI matrix M is as follows:

[0027]

[0028] Among them, wPLI bc represents the weighted phase lag index matrix between the b-th brain region and the c-th brain region, where b = 1, 2,..., N; c = 1, 2,..., N;

[0029] For the clustering coefficient C of a certain node i in the matrix M i , it is defined as the ratio between the actual number of connections e i existing among the neighbor nodes of node i and all possible numbers of connections; then the global clustering coefficient Clu takes the average value of the clustering coefficients of all nodes, and the expression is as follows:

[0030]

[0031]

[0032] Among them, k i represents the degree of node i;

[0033] The characteristic path length L is defined as the average value of the shortest path lengths between all node pairs in the topological graph, and its formula is:

[0034]

[0035] Among them, d ij represents the shortest path length from node i to node j;

[0036] The global efficiency Ge measures the overall efficiency of information transmission in the brain network and is the average value of the reciprocals of the shortest paths of all node pairs:

[0037]

[0038] The local efficiency Le is the average value of the local efficiencies of all nodes:

[0039]

[0040] Among them, Le(i) represents the local efficiency of node i; N(i) represents the neighbor set of node i, represents in the subgraph G iThe shortest path length between the middle node j and the node h;

[0041] Taking the maximum Pearson correlation coefficient of the wPLI matrices of the initial state and the target state as the goal of fitting the EEG signals by the Hopf model; taking the minimum Euclidean distance of the network properties of the wPLI matrices of the perturbed state and the target state of the initial state as the index to evaluate the degree of state transition;

[0042] Calculate the structural connection matrix M of each subject SC , taking the normalized number of fiber bundles between brain regions as the connection strength, and the calculation formula is as follows:

[0043]

[0044] where C np represents the structural connection strength between the nth brain region and the pth brain region, n = 1, 2,..., N; p = 1, 2,..., N; C np The calculation formula of is as follows:

[0045]

[0046] where g ij represents the number of fiber bundles between brain region i and brain region j.

[0047] Furthermore, the steps to construct the co-activation pattern of the region of interest are as follows:

[0048] First, extract the EEG signals of the prior frequency band of the prior brain region, arrange the numerical values of the EEG signals from large to small, then take the time points corresponding to the top-ranked numerical values according to a certain proportion, and the threshold line is set at the top 20% of the time points; select the time points in the EEG signals that exceed the threshold line, integrate the topological maps of all subjects' selected time points together, and use the KMeans algorithm to cluster the topological maps;

[0049] Then, average the topological maps belonging to the same class to obtain the co-activation pattern, and calculate the time ratio TR of each co-activation pattern:

[0050]

[0051] where T i represents the number of time points belonging to the ith co-activation pattern;

[0052] Select the co-activation pattern with the highest time ratio as the dominant brain activation pattern in this state, and take the activated brain regions of this co-activation pattern as the sites.

[0053] Furthermore, the parameter inputs of the Hopf model are the connection weights of the structural connections, the natural frequencies of each site, the noise intensity, the global coupling strength, and the bifurcation coefficient, and the output is the discrete virtual signal that changes with time.

[0054] Furthermore, the perturbation strategy includes: synchronous stimulation and noise stimulation, as well as single-site perturbation and two-site perturbation; synchronous stimulation corresponds to an increase in the bifurcation parameter of the Hopf model, and noise stimulation corresponds to a decrease in the bifurcation parameter of the Hopf model; single-site perturbation refers to perturbing the adjustment parameter of only one brain region, and two-site perturbation refers to perturbing the adjustment parameters of two brain regions simultaneously;

[0055] Optimization strategy: For the new signal generated after applying stimulation to the virtual signal in the initial state, calculate the four network attributes of the new signal's wPLI matrix, and then calculate the Euclidean distance between the four network attributes of the wPLI of the new signal and the virtual EEG signal in the target state. The smaller the distance, the more similar it is. Find the site corresponding to the minimum distance as the target point, and the change amount of the corresponding bifurcation parameter value as the optimal stimulation intensity.

[0056] Compared with the prior art, the remarkable effects of the present invention are as follows:

[0057] 1. The system of the present invention can effectively avoid the deviation caused by subjective judgment in traditional methods by using an objective non-linear dynamics model to quantitatively describe the brain state, realize the accurate simulation of the transition between different brain states and the prediction of target points, thereby providing a scientific basis for neuromodulation, early diagnosis of brain diseases and individualized treatment, and solving the deficiencies of existing EEG state analysis methods in capturing early dynamic changes and target point recognition;

[0058] 2. The method of the present invention improves the EEG simulation level of the virtual signal by using the weighted phase lag index and the similarity of its network attributes as the parameter optimization target when constructing the virtual brain signal. By extracting the co-activated brain regions of the co-activation pattern and combining with virtual stimulation, the target points during the process of brain state transition can be identified and predicted, improving the physiological interpretability of the model, providing a quantitative basis for the construction of the virtual brain model and brain state regulation, and helping to formulate individualized neuromodulation strategies;

[0059] 3. The method of the present invention focuses on the microscopic dynamic changes in the early stage of brain state transition, can be used as a tool for EEG signal analysis in neuroscience research and clinical applications, and reduces the workload of manual intervention and the risk of misjudgment. Description of the Drawings

[0060] Figure 1 It is a schematic structural diagram of the EEG transition simulation system of the present invention;

[0061] Figure 2 It is a flowchart of the EEG transition simulation method of the present invention;

[0062] Figure 3 It is an exploration diagram of virtual EEG hyperparameter optimization for the resting state and smoking state;

[0063] Figure 4 Heat map of bifurcation parameter change for promoting state transition by single-site stimulation;

[0064] Figure 5 Optimization result graph for promoting state transition by two-site stimulation. Specific implementation manners

[0065] The present invention will be further described in detail below in conjunction with the accompanying drawings of the specification and specific implementation manners.

[0066] As Figure 1 shown, an electroencephalogram (EEG) transition simulation system for different brain states based on the Hopf model includes:

[0067] A signal preparation module, configured to collect, preprocess, and perform source localization on real EEG data, remove noise and artifacts, and extract clean source signals, and the "clean source signals" output by this module serve as the input of the state feature extraction module. Among them, the real EEG data includes initial state EEG data and target state EEG data.

[0068] The signal preparation module includes an EEG signal acquisition module, a preprocessing and source reconstruction module. The EEG signal acquisition module: acquires multi-channel EEG signals to ensure sufficient sampling rate and spatial resolution. The preprocessing and source reconstruction module: filters the acquired EEG signals, removes artifacts (such as eye movement, electromyogram interference, etc.), and performs source localization in combination with a head model and an algorithm (such as an inversion algorithm) to obtain neural activity signals at each brain region level.

[0069] A state feature extraction module, configured to perform weighted phase lag index network analysis and co-activation pattern analysis on the brain region signals after source reconstruction, and extract brain network feature indicators in different brain states. The output brain network connection features, spatio-temporal patterns, etc. are used by the virtual signal construction module, the single-site perturbation module, and the two-site perturbation module.

[0070] The state feature extraction module includes a weighted phase lag index and its network attribute extraction module and a co-activation pattern construction module. The weighted phase lag index and its network attribute extraction module: calculates the functional connection strength in the brain network based on the wPLI method, and further extracts brain network attributes (clustering coefficient, characteristic path length, global efficiency, local efficiency) to quantitatively describe the overall and local connection features of the brain in different states, and is used for parameter optimization in the process of the virtual signal construction module. The co-activation pattern (CAP) construction module: identifies synchronous activation segments of multi-channel brain signals in the time dimension, extracts CAPs with specific functional significance, thereby characterizing the co-activation features of the brain in different states or task conditions, and assisting the state transition prediction module to find target combinations according to the perturbation strategy.

[0071] The virtual signal construction module constructs a virtual brain model that can simulate the real brain state based on the extracted brain network connection features and the neural dynamics model. The virtual brain signals output by the virtual brain model provide a controllable simulation environment for the state transition prediction module; by comparing with the real EEG signals, the physiological rationality and prediction accuracy of the state transition prediction module can be verified.

[0072] The virtual signal construction module includes a Hopf model construction module and a parameter optimization module. The Hopf model construction module: regards brain regions as coupled oscillators, and initially constructs a virtual brain model (i.e., the Hopf model) by setting the parameters of each oscillator. The parameter optimization module: adjusts and optimizes the parameters of the Hopf model based on real EEG data, so that the dynamic behavior of the virtual brain signals matches that of the real EEG signals, thereby improving the physiological rationality and accuracy of the virtual brain model.

[0073] The state transition prediction module evaluates and predicts the evolution process of the brain state under different intervention methods based on the virtual brain model, and finds the target combination for realizing the brain state transition.

[0074] The state transition prediction module includes a single-site perturbation module and a two-site perturbation module. The single-site perturbation module: applies a perturbation to a single brain region and observes the response and state migration of the global network. The two-site perturbation: applies perturbations to two brain regions simultaneously to study the impact of multiple stimuli or cooperative regulation on the brain state transition and find the target combination.

[0075] As Figure 2 shown, an EEG transition simulation method for different brain states based on the Hopf model includes the following steps:

[0076] Step 1, collect the EEG signals of all subjects in the initial state and the target state, and perform preprocessing;

[0077] The EEG signal acquisition module collects the EEG signals of all subjects in the initial state and the target state, and then the preprocessing and source reconstruction module preprocesses the collected EEG signals.

[0078] Step 2, construct a weighted phase lag index matrix (wPLI (weighted Phase Lag Index) matrix) between brain regions;

[0079] The preprocessing and source reconstruction module reconstructs and traces the collected EEG signals back to the cortical brain region level, and calculates the weighted phase lag index matrix between brain regions in the initial state and the target state for each subject. And calculate the network properties of the weighted phase lag index matrix, including the global clustering coefficient Clu (Clustering Coefficient), the characteristic path length L (Characteristic Path Length), the global efficiency Ge (Global Efficiency), and the global-local efficiency Le (Local Efficiency).

[0080] Furthermore, the construction process of the weighted phase lag index matrix at the brain source level is as follows: Reconstruct the collected EEG signals to the cortical brain region level, and use the automated anatomical labeling template (AAL structural template) to divide the subject's brain into N brain regions, and each brain region serves as a node of the brain network. Then calculate the wPLI value between the EEG signals of the N brain regions. The wPLI calculation formula is:

[0081]

[0082] where wPLI represents the weighted phase lag index, Im(Δφ(t)) represents the imaginary part of the phase difference between two signals at time t; T represents the period of the collected signal; sign(sin(Δφ(t))) is used to take the sign (+1 or -1) of the sine of the phase difference at time t, indicating whether the phase difference is positive or negative at this time, and this step plays a "weighting" role in the subsequent summation.

[0083] Construct the wPLI matrix M from the wPLI values between brain regions to represent the phase lag connection network. The calculation formula of the wPLI matrix M is as follows:

[0084]

[0085] where wPLI bc represents the weighted phase lag index between the b-th brain region and the c-th brain region, b = 1, 2,..., N; c = 1, 2,..., N.

[0086] In this embodiment, N = 90, and then calculate the four network properties Clu, L, Ge, and Le of the wPLI matrix M.

[0087] Global clustering coefficient Clu: For a certain node i in the matrix, the clustering coefficient C of node i i is defined as the ratio between the actual number of connections e i existing between its neighbor nodes and all possible numbers of connections. The formula is:

[0088]

[0089] The global clustering coefficient Clu is the average of the clustering coefficients of all nodes:

[0090]

[0091] where k i represents the degree of node i;

[0092] The characteristic path length L: The characteristic path length is defined as the average of the shortest path lengths between all node pairs in the graph, and its formula is:

[0093]

[0094] where d ij represents the shortest path length from node i to node j.

[0095] The global efficiency Ge: The global efficiency measures the overall efficiency of information transmission in the network and is the average of the reciprocals of the shortest paths for all node pairs:

[0096]

[0097] The local efficiency Le: For each node i, the local efficiency is defined as the global efficiency of its neighbor subgraph:

[0098]

[0099] where N(i) represents the neighbor set of node i, represents the shortest path length between node j and node h in the subgraph G i ; Le(i) represents the local efficiency of node i;

[0100] Then the local efficiency Le is the average of the local efficiencies of all nodes:

[0101]

[0102] Take the maximum Pearson correlation coefficient of the wPLI matrices of the initial state and the target state as the goal of fitting the EEG signals by the Hopf model. Take the minimum Euclidean distance of the network properties of the wPLI matrices of the perturbed state of the initial state and the target state as the index to evaluate the degree of state transition.

[0103] Calculate the structural connection matrix M SC for each subject, and use the normalized number of fiber bundles between brain regions as the connection strength. The calculation formula is as follows:

[0104]

[0105] where C nprepresents the structural connection strength (i.e., the connection weight of the structural connection) between the nth brain region and the pth brain region, where n = 1, 2, …, N; p = 1, 2, …, N. C np The calculation formula is as follows:

[0106]

[0107] where g ij represents the number of fiber bundles between brain region i and brain region j.

[0108] Step 3: For the initial state and the target state of each subject, construct the co-activation pattern CAP (Co-Activation Pattern) of the region of interest;

[0109] According to relevant priors, select the region of interest ROI (Region of Interest) as the seed point, construct the co-activation pattern topological graph, and extract the brain regions that are co-activated with the ROI as the sites (the site refers to the specific position or node where the perturbation (or stimulus) is applied in the Hopf model).

[0110] Furthermore, the method for constructing the co-activation pattern CAP of the region of interest is as follows: First, extract the electroencephalogram signals of the prior frequency band of the prior brain region, select the time points in the electroencephalogram signals that exceed the threshold line (first arrange the values of the electroencephalogram signals from large to small, and then take the time points corresponding to the top-ranked values according to a certain proportion. The threshold line is set at the top 20% of the time points), integrate the topological graphs of all the selected time points of all the subjects together, and use the KMeans algorithm to cluster the topological graphs. Then, average the topological graphs belonging to the same class to obtain the co-activation pattern (CAP). Calculate the temporal ratio TR (Temporal Ratios) of each CAP:

[0111]

[0112] The TR of each CAP refers to the ratio between the number of time points it contains in each subject and the total number of time points of each subject, where T i represents the number of time points belonging to the ith CAP.

[0113] Select the CAP with the highest temporal ratio as the dominant brain activation pattern in this state (this state includes the initial state and the target state), and use the co-activated brain regions of this CAP as the target points (the target points refer to the brain regions or nodes that show significant dynamic changes in the whole brain network after perturbation and play a key role in state transition).

[0114] Step 4: Construct the Hopf model and construct the virtual signal by solving the Hopf oscillator equation;

[0115] Input the connection weights of the structural connections, the natural frequency of each site, the noise intensity, the global coupling strength, and the bifurcation coefficient as the parameters of the Hopf model, solve the Hopf oscillator equation, and obtain a set of discrete virtual oscillation signals that vary with time (i.e., obtain a set of virtual signals);

[0116] Furthermore, the expressions of the Hopf model are as shown in formulas (12) and (13):

[0117]

[0118] Among them, and are the amplitude modulation terms of the Hopf oscillator; a is a key parameter that determines the oscillation characteristics of the Hopf model: if a > 0, the Hopf model will generate a limit cycle oscillation near the radius ; if a < 0, the equilibrium state of the Hopf model is often a stationary point and there is no spontaneous oscillation; is the square of the amplitude of the current oscillator state (r in polar coordinates 2 ); -w n y n and +w n x n are the parts that determine the phase rotation of the oscillator, and w n represents the natural angular frequency of the oscillator. and are the coupling terms, where G is the global coupling coefficient, which controls the "weight" of the interaction between nodes in the network. C np is the connection relationship between the nth oscillator and the pth oscillator in the network, which is obtained from the subject's structural connection matrix M SC (SC, Structural Connectivity). (x p -x n ) and (y p -y n ) are the differential coupling terms. βη n (t) is the external input term, β represents the amplitude or weight of the influence of the external input signal on the Hopf model, and η n (t) represents the external drive for the nth node (set as random noise).

[0119] The essence of solving the Hopf oscillator equation to construct virtual signals is to solve a set of Hopf differential equations. For the Hopf model, a set of virtual oscillation discrete signals can be obtained by inputting each set of different parameters.

[0120] Step 5, optimize the hyperparameters of the Hopf model according to the fitting target to obtain the optimal virtual brain models for the initial state and the target state respectively;

[0121] The fitting objective is to find the maximum Pearson correlation coefficient between the wPLI matrix of the virtual signal and the wPLI matrix of the real EEG signals (including the EEG signals in the initial state and the target state). It is found using the grid search method, and the hyperparameters of the Hopf model when the fitting is optimal for the initial state and the target state are obtained respectively, and the optimal Hopf models for the initial state and the target state are saved.

[0122] Furthermore, the method for optimizing hyperparameters according to the fitting objective is as follows: The global coupling coefficient G and the bifurcation parameter a in the Hopf equation are used as hyperparameters, and w n is set to 2πf, where f is the number of Hertz of the signal frequency band to be simulated, and β is selected as 0.2. The similarity between the wPLI matrix of the virtual signal generated by different combinations of G and a values and the wPLI matrix of the real signal is explored, and the G and a values when the similarity is the highest are found as the optimal hyperparameters of the Hopf model, and the optimal Hopf model is saved as the optimal virtual brain model. The index for measuring the similarity is the Pearson correlation coefficient after arranging the wPLI matrices of the two groups of signals as vectors.

[0123] Step 6, based on the optimal virtual brain models for the initial state and the target state, a perturbation strategy is adopted to find the target points that promote the transformation of the brain state, and the virtual signal in the initial state is transformed into the virtual signal in the target state;

[0124] Finding the perturbation strategy is to realize the characteristic transformation of the virtual signal in the initial state into the virtual signal in the target state. The measurement index of the transformation degree is the Euclidean distance between the network attributes of the wPLI matrix in the initial state and the network attributes of the wPLI matrix in the target state, and the change amount of the bifurcation parameter corresponding to the minimum distance is found.

[0125] The perturbation strategy includes: synchronous stimulation and noise stimulation, as well as single-site perturbation and two-site perturbation. Synchronous stimulation corresponds to an increase in the bifurcation parameter of the Hopf model, and noise stimulation corresponds to a decrease in the bifurcation parameter of the Hopf model. Single-site perturbation means only perturbing the adjustment parameter of one brain region, and two-site perturbation means perturbing the adjustment parameters of two brain regions simultaneously.

[0126] Experiments have shown that the two-site perturbation of the co-activated brain region combination found in Step 3 is superior to the single-site perturbation of the whole brain, and the double target points that promote the transformation of the brain state are finally found through the perturbation strategy.

[0127] Further, the perturbation bifurcation parameter realizes the state transition process as follows: First, perform single-site stimulation on each brain region of the virtual EEG in the initial state. Increase the value of a to achieve synchronous stimulation, and decrease the value of a to achieve noise stimulation. Find the optimal a-value change for each site as the relative stimulation intensity. Optimization strategy: For the new signal generated after applying stimulation to the virtual signal in the initial state, calculate the four network properties of the wPLI matrix of the new signal, and then calculate the Euclidean distance between the four network properties of the wPLI of the new signal and the virtual EEG signal in the target state. The smaller the distance, the more similar it is. Find the site corresponding to the minimum distance as the target, and the corresponding a-value change as the optimal stimulation intensity. Then, use the co-activated brain regions found in step 3 as potential targets for two-site stimulation for optimization. The two-site stimulation combination is the combination of the seed point brain region of CAP and other co-activated brain regions. The optimization strategy is the same as that of single-site stimulation. Find the site combination corresponding to the minimum distance as the target combination, and the corresponding a-value change as the optimal stimulation intensity.

[0128] Step 7, according to the found targets, formulate personalized neuroregulation strategies for the subjects.

[0129] In this example, 23 subjects of smokers with severe nicotine dependence were selected, and the scores on the Fagerström Test for Nicotine Dependence (FTND scale) were 6 - 10 points. Electroencephalogram (EEG) signals of all subjects in the resting state (i.e., the initial state) and smoking state (i.e., the target state) were collected. After preprocessing and source reconstruction, the targets for the transition between the resting state and the smoking state were found through the EEG transition simulation system of the present invention, including the following steps:

[0130] Step F1, calculate the weighted phase lag index matrix wPLI between brain regions using source-level EEG signals, and calculate the network properties of the wPLI matrix, including clustering coefficient Clu, characteristic path length L, global efficiency Ge, and local efficiency Le. Then, use diffusion tensor imaging to construct the brain structural connection matrix;

[0131] Step F2, according to the prior selection, the medial prefrontal cortex (MPFC) was used as the seed point to construct the co-activation pattern topological map of smoking EEG. After clustering, 2 CAPs were obtained. Among them, the time ratio TR of CAP1 was significantly higher than that of CAP2. CAP1 was selected as the main activated brain state during smoking. The brain regions activated in CAP1, including the middle frontal gyrus, medial prefrontal cortex, superior parietal lobule, angular gyrus, insula, anterior cingulate cortex, middle temporal gyrus, and superior temporal gyrus, were used as potential targets for subsequent verification of two-site stimulation;

[0132] Step F3, solve the Hopf oscillator equation to construct virtual brain models for the resting state and the smoking state respectively. The connection weights of the structural connection, the natural frequency of each site, the noise intensity, the global coupling intensity, and the bifurcation coefficient were used as the inputs of the virtual brain models, and the discrete virtual signals that change with time were used as the outputs of the virtual brain models;

[0133] Step F4: Optimize the parameters of the virtual signals in the resting state and smoking state. Use the grid search method to find the hyperparameters G and a of the Hopf model in the resting state and smoking state when the fitting is optimal. The optimization results are as Figure 3 shown. The horizontal axis of the heat map represents the G value, and the vertical axis represents the a value;

[0134] Step F5: Perturb the bifurcation parameter to achieve state transition. First, perform single-site stimulation on each brain region of the virtual EEG in the resting state separately, and find the optimal change amount of a value for each site as the relative stimulation intensity. Optimization strategy: For the new signal generated after applying stimulation to the virtual signal in the resting state, calculate the four network attributes of the new signal wPLI, and then calculate the Euclidean distance between the four network attributes of the new signal and the wPLI of the virtual EEG signal in the smoking state, and find the site and its a value change amount corresponding to the minimum distance. The single-site stimulation results are as Figure 4 shown, where the horizontal axis represents the stimulation intensity, that is, the change amount of a value, the vertical axis represents the brain region site, and the darker the color, the closer the distance and the better the stimulation effect. Then, use the co-activated brain regions found in Step F2 as potential targets for two-site stimulation to optimize. The two-site stimulation combinations are the combinations of the seed point brain regions of CAP and other co-activated brain regions. The optimization strategy is the same as that of single-site stimulation, and find the site combination and its a value change amount corresponding to the minimum distance. The two-site stimulation results are as Figure 5 shown, where the horizontal axis represents the site combination, the vertical axis represents the Euclidean distance between the network attributes of the wPLI of the virtual signal after stimulation and the network attributes of the wPLI of the target state virtual signal, and different broken lines represent different change amounts of a value. Take the site combination corresponding to the minimum distance as the target combination, and the corresponding change amount of a value as the optimal stimulation intensity.

[0135] Step F6: According to the found target combination, formulate different smoking cessation regulation strategies for each subject.

Claims

1. A system for simulating EEG transitions of different brain states based on the Hopf model, characterized in that: include: The signal preparation module is used to collect, preprocess and locate the EEG data of the initial state and the target state, remove noise and artifacts and extract clean source signals; It includes EEG signal acquisition module, preprocessing and source reconstruction module. The EEG signal acquisition module acquires multi-channel EEG signals to ensure sufficient sampling rate and spatial resolution. The preprocessing and source reconstruction module filters and removes artifacts from the acquired EEG signals, and combines the head model and inversion algorithm to locate the source, so as to obtain neural activity signals at the level of each brain region. The state feature extraction module performs weighted phase lag index network analysis and co-activation pattern analysis on the brain area signals after source reconstruction, extracts brain network feature indicators under different brain states, and provides them for use by the virtual signal construction module and the state transition prediction module; The virtual signal construction module builds a virtual brain model that can simulate the real brain state based on the extracted brain network connection characteristics and neural dynamics model; the virtual brain signal output by the virtual brain model provides a controllable simulation environment for the state transition prediction module; The state transition prediction module, based on the virtual brain model, evaluates and predicts the evolution of brain states under different intervention methods, and searches for target combinations to achieve brain state transition.

2. The EEG transition simulation system of different brain states based on the Hopf model according to claim 1, characterized in that: The state feature extraction module includes a weighted phase lag index and its network attribute extraction module and a co-activation pattern construction module. The weighted phase lag index and its network attribute extraction module calculates the functional connection strength in the brain network based on the wPLI method, and further extracts network attributes, quantitatively describes the overall and local connection characteristics of the brain under different states, and is used for parameter optimization in the virtual signal construction module process; the co-activation pattern construction module identifies the synchronous activation fragments of multi-channel brain signals in the time dimension, extracts co-activation patterns with specific functional significance, and describes the coordinated activation characteristics of the brain under different states or task conditions, and assists the state transition prediction module in finding target combinations according to the perturbation strategy.

3. The EEG transition simulation system of different brain states based on the Hopf model according to claim 1, characterized in that: The virtual signal construction module includes a Hopf model construction module and a parameter optimization module. The Hopf model construction module regards the brain area as a coupled oscillator, and builds a Hopf model as a virtual brain model by setting the parameters of each oscillator; the parameter optimization module adjusts and optimizes the Hopf model parameters based on the initial state EEG data so that the dynamic behavior of the virtual brain model matches the real EEG signal.

4. The EEG transition simulation system of different brain states based on the Hopf model according to claim 1, characterized in that: The state transition prediction module includes a single-site perturbation module and a dual-site perturbation module. The single-site perturbation module applies perturbations to a single brain region to observe the response and state transition of the global network; the dual-site perturbation module applies perturbations to two brain regions at the same time to study the effects of multiple stimulations or coordinated regulation on brain state transitions and find target combinations.

5. A method for simulating EEG transitions of different brain states based on the Hopf model, characterized in that: The method of searching for the target point of the transition between the initial state and the target state by using the EEG transition simulation system as claimed in any one of claims 1 to 4 comprises the following steps: S1, collect the EEG signals of all subjects in the initial state and target state and perform preprocessing; S2, reconstruct the EEG signal source to the cortical brain area level, calculate the wPLI matrix between brain areas of each subject, and calculate the network properties of the wPLI matrix and the structural connection matrix of each subject; S3, select the region of interest as the seed point, construct the co-activation pattern topology map, and extract the brain areas that are in the same activated state as the region of interest as the sites; S4, constructing the Hopf model and constructing the virtual signal by solving the Hopf oscillator equation; S5, taking the maximum value of the Pearson correlation coefficient of the wPLI matrix of the virtual signal and the wPLI matrix of the real EEG signal as the optimization target, using the grid search method, the optimal virtual brain model of the initial state and the test state are obtained respectively; S6, based on the optimal virtual brain model, a perturbation strategy is used to find the target that promotes the brain state transition, and realize the transition from the initial state virtual signal to the target state virtual signal; S7, develop personalized neuromodulation strategies for the subjects based on the obtained targets.

6. The method for simulating EEG transitions of different brain states based on the Hopf model according to claim 5, characterized in that: The network properties of the weighted phase lag index matrix include the global clustering coefficient Clu, the characteristic path length L, the global efficiency Ge and the local efficiency Le; the process of calculating the wPLI matrix between brain regions is as follows: The subject's brain was divided into N brain regions using an automatic anatomical labeling template, each brain region was used as a node of the brain network, and then the wPLI matrix between the EEG signals of the N brain regions was calculated; the calculation formula of wPLI was: Wherein, wPLI represents the weighted phase lag index, Im(Δφ(t)) represents the imaginary part of the phase difference between two signals at time t; T represents the signal acquisition period; sign(sin(Δφ(t))) is used to take the sign of the sine of the phase difference at time t, +1 or -1; Construct the wPLI matrix M to represent the phase-lag connection network. The expression of the wPLI matrix M is as follows: Among them, wPLI bc represents the weighted phase lag index matrix between the bth brain region and the cth brain region, b=1,2,…,N; c=1,2,…,N; For a node i in the matrix M, the clustering coefficient C i , defined as the number of actual connections between node i’s neighbor nodes e i The ratio of all possible connections; the global clustering coefficient Clu takes the average value of the clustering coefficients of all nodes, and the expression is as follows: Among them, k i represents the degree of node i; The characteristic path length L is defined as the average of the shortest path lengths between all node pairs in the topological graph, and its formula is: Among them, d ij represents the shortest path length from node i to node j; The global efficiency Ge measures the overall efficiency of information transmission in the brain network, which is the average of the reciprocal of the shortest path of all nodes: The local efficiency Le is the average of the local efficiencies of all nodes: Among them, Le(i) represents the node local efficiency of node i; N(i) represents the neighbor set of node i, In the subgraph G i The shortest path length between node j and node h; The maximum Pearson correlation coefficient of the wPLI matrix of the initial state and the target state is used as the target of the Hopf model to fit the EEG signal; the minimum Euclidean distance of the network attributes of the wPLI matrix of the initial state after the disturbance and the target state is used as an indicator to evaluate the degree of state transition; Calculate the structural connection matrix M for each subject SC , the normalized number of fiber bundles between brain regions is used as the connection strength, and the calculation formula is as follows: Among them, C np represents the strength of structural connection between n brain regions and the pth brain region, n = 1, 2, ..., N; p = 1, 2, ..., N; C np The calculation formula is as follows: Among them, g ij Represents the number of fiber bundles between brain area i and brain area j.

7. The method for simulating EEG transitions of different brain states based on the Hopf model according to claim 5, characterized in that: The steps to construct the co-activation pattern of the region of interest are as follows: First, extract the EEG signals of the prior brain region and the prior frequency band, arrange the values ​​of the EEG signals from large to small, and then take the time points corresponding to the top values ​​according to a certain proportion, and the threshold line is set at the top 20% time points; Select the time points that exceed the threshold line in the EEG signal, integrate the topological maps of all the selected time points of the subjects, and cluster the topological maps using the KMeans algorithm; Then, the topological maps belonging to the same category are averaged to obtain the co-activation pattern, and the time proportion TR of each co-activation pattern is calculated: Among them, T i represents the number of time points belonging to the i-th co-activation pattern; The co-activation pattern with the highest time proportion is selected as the dominant brain activation pattern in this state, and the activated brain area of ​​this co-activation pattern is taken as the site.

8. The method for simulating EEG transitions of different brain states based on the Hopf model according to claim 5, characterized in that: The parameter input of the Hopf model is the connection weight of the structural connection, the natural frequency of each site, the noise intensity, the global coupling strength and the bifurcation coefficient, and the output is a discrete virtual signal that changes with time.

9. The method for simulating EEG transitions of different brain states based on the Hopf model according to claim 5, characterized in that: The perturbation strategies include: synchronous stimulation and noise stimulation, as well as single-site perturbation and dual-site perturbation; synchronous stimulation corresponds to an increase in the bifurcation parameter of the Hopf model, and noise stimulation corresponds to a decrease in the bifurcation parameter of the Hopf model; single-site perturbation refers to perturbing the adjustment parameters of only one brain region, and dual-site perturbation refers to perturbing the adjustment parameters of two brain regions at the same time; Optimal judgment strategy: After applying stimulation to the virtual signal in the initial state, the new signal is generated, the four network attributes of the wPLI matrix of the new signal are calculated, and then the Euclidean distance between the new signal and the four network attributes of the wPLI of the virtual EEG signal in the target state is calculated. The smaller the distance, the more similar they are. The site corresponding to the minimum distance is found as the target point, and the corresponding change in the bifurcation parameter value is used as the optimal stimulation intensity.

Images