Digital brain data assimilation system and method based on multi-modal data fusion

Abstract

The invention discloses a digital brain data assimilation system and method based on multi-modal data fusion. The system comprises an EEG forward model construction module, an MRI-EEG space mapping construction module, a network structure generation module, a multi-modal data assimilation module, a neural network simulation module and a digital brain similarity evaluation module. The multi-modal data assimilation module is improved on the basis of an existing hierarchical mesoscale data assimilation algorithm, and the model state is filtered on the basis of an augmented observation vector (an experiment / system simulation BOLD signal and an experiment EEG signal) in each updating step, so that estimated parameters can fit two observation signals at the same time. According to the method, fitting of multi-modal data with different temporal-spatial resolutions can be realized in the same digital twin brain system.

Description

Technical Field

[0001] The present invention belongs to the technical field of data assimilation, and in particular, relates to a digital brain data assimilation system and method based on multimodal data fusion. Background Art

[0002] Existing data assimilation methods estimate the parameters of digital brain models by fitting BOLD (blood oxygen level dependence) signals. However, due to the low temporal resolution of fMRI (functional magnetic resonance imaging) acquisition, it is difficult to capture neuronal dynamics during continuous tasks. EEG (electroencephalogram) signals have higher temporal resolution and can be used to decode continuous tasks, but their spatial resolution is poor. The goal of multimodal data fusion and assimilation is to simultaneously fit high-temporal-resolution EEG signals and high-spatial-resolution BOLD signals using the same set of parameters, thereby constructing a digital twin that more closely resembles the biological brain.

[0003] Based on the theory of mesoscopic data assimilation, and using multimodal neuroimaging data of the human brain obtained by high-precision magnetic resonance imaging equipment, on a supercomputer cluster connected by a high-speed network, based on a computational neural network model in voxels or functional columns, we have successfully inferred the parameters in the network model, and simulated the multimodal signals produced by the real human brain through the digital twin brain. Summary of the Invention

[0004] In order to fuse multi-spatial resolution and multi-modal signals, incorporate more biological constraints into the digital brain, and build a digital twin brain that is more consistent with the real biological brain, the present invention improves the existing data assimilation algorithm and designs experiments to verify simulated observations on a small network.

[0005] The technical solution of the present invention is specifically described as follows.

[0006] The present invention provides a digital brain data assimilation system based on multimodal data fusion, which includes an EEG forward model construction module, an MRI-EEG spatial mapping construction module, a network structure generation module, a multimodal data assimilation module, a neural network simulation module and a digital brain similarity evaluation module; wherein: EEG forward model building module, used to calculate the lead field matrix that represents the mapping relationship between source current and scalp electrode potential; The MRI-EEG spatial mapping construction module calculates the lead field sensitivity distribution of each electrode based on the EEG forward model and uses an optimization algorithm to find the voxel-electrode mapping relationship with the maximum total lead field sensitivity; The network structure generation module generates a computational brain model of the experimental subject based on the MRI (magnetic resonance imaging) signals measured by the experimental subject. The parameters that can be set in the computational brain model include the total number of neurons, the minimum number of neurons in the neural network, the network link degree, and the synaptic transmitter conductivity coefficient; The multimodal data assimilation module uses a multimodal integrated Kalman filtering method based on multi-layer Bayesian estimation theory to assimilate the synaptic conductance coefficients of the neurons in the minimum neural network and update the parameters in the assimilated and calculated brain model so that the output of the brain model is similar to the subject's EEG and BOLD signals. In the implementation of the multimodal integrated Kalman filtering method, the update step uses the mapping relationship established by the MRI-EEG spatial mapping construction module to correct the parameters of each voxel through the augmented observation vector composed of the EEG signal of the corresponding electrode and the measured BOLD signal. In time steps with only EEG observations, the BOLD signal predicted by the model is used to interpolate the empirical data to reduce the noise fluctuations introduced by the EEG high-frequency components in the BOLD signal simulation. Neural network simulation module, which performs network simulation on given assimilated parameters; The digital brain similarity assessment module is verified through multimodal neural signal comparison.

[0007] In the present invention, the EEG forward model construction module realizes the spatial registration of EEG electrode coordinates with individual MRI data based on the anatomical landmarks marked during the data acquisition process, calculates the lead field matrix based on the co-registered voxel positions, electrode positions and directions, and the conductivity of different brain tissues; and generates a brain mask for DTI tracking and BOLD signal extraction based on the voxel coordinates in the individual space.

[0008] In the present invention, in the MRI-EEG spatial mapping construction module, the lead field sensitivity distribution of each scalp electrode is calculated based on the EEG forward model, the contribution of each voxel to the potential of each electrode is quantified, and a voxel-electrode bipartite graph model is constructed. With maximizing the overall lead field sensitivity as the optimization goal, the Hungarian algorithm is applied to solve the optimal matching problem and determine the voxel-electrode correspondence with the maximum total sensitivity.

[0009] In the present invention, the fMRI BOLD (blood oxygen level-dependent cerebral isotope) signal is processed as follows to obtain the measured BOLD signal for each voxel. Specifically, the following steps are performed: using the SPM software package to perform detrending and bandpass filtering on the signal to achieve standardization of the fMRI signal; and aligning the fMRI data to the individual T1 space and extracting the BOLD signal based on a brain mask, so that the BOLD signal and the EEG signal are in the same space.

[0010] In the present invention, the MRI signal includes brain diffusion tensor imaging (DTI) and cerebral cortical gray matter density as well as a brain mask generated in an EEG forward model building system.

[0011] In the present invention, a drawing program is encapsulated in the neural network simulation module to draw the average excitation rate and average neuron membrane potential of the minimum neural network; it also observes the neuron with a specific number by inputting the number, and outputs specific information including the membrane potential, synaptic current, and excitation time of the neuron.

[0012] In the present invention, the digital brain similarity assessment module quantitatively evaluates the similarity between the digital brain model and the biological brain in functional activities by calculating the correlation between the EEG / BOLD signals collected in the experiment and the analog signals.

[0013] The present invention also provides a digital brain data assimilation method based on multimodal data fusion, comprising the following steps: (1) Calculate the lead field matrix that represents the mapping relationship between source current and scalp electrode potential; (2) Based on the lead field sensitivity distribution of each electrode, an optimization algorithm is used to find the voxel-electrode mapping relationship with the maximum total lead field sensitivity; (3) Generate a computational brain model of the subject based on the MRI signals measured by the experimental subject; the parameters that can be set in the computational brain model include the total number of neurons, the minimum number of neurons in the neural network, the network link degree, and the synaptic transmitter conductivity coefficient; (4) Based on the multimodal integrated Kalman filtering method of multi-layer Bayesian estimation theory, the synaptic conductance coefficients of the neurons in the minimum neural network are assimilated, and the parameters in the assimilated and calculated brain model are updated so that the output of the brain model is similar to the EEG and BOLD signals of the subjects. In the implementation of the multimodal integrated Kalman filtering method, the update step uses the mapping relationship established by the MRI-EEG spatial mapping construction module to correct the parameters of each voxel through the augmented observation vector composed of the corresponding electrode EEG signal and the measured BOLD signal. In the time step with only EEG observation, the BOLD signal predicted by the model is used to interpolate the empirical data to reduce the noise fluctuation introduced by the EEG high-frequency component to the BOLD signal simulation. (5) Perform network simulation for given assimilated parameters; (6) Verification is achieved through multimodal neural signal comparison.

[0014] In the present invention, the BOLD (blood oxygen level-dependent brain signal) signal processing of fMRI includes: using the SPM software package to perform operations such as detrending and bandpass filtering on the signal to achieve standardized processing of the fRMI signal; and aligning the fMRI data to the individual T1 space and extracting the BOLD signal based on the brain mask, so that the BOLD signal and the EEG signal are in the same space.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention realizes the fitting of multimodal data with different spatiotemporal resolutions in the same digital twin brain system.

[0016] The present invention breaks through the limitations of existing single-modal signal fitting. The bottleneck of the existing technology is that although single-modal fMRI signals have millimeter-level spatial resolution, they are limited by low temporal resolution and cannot capture the rapid neural dynamics and high-frequency oscillation characteristics during continuous task execution; and although a single EEG signal can provide millisecond-level temporal resolution, its spatial resolution is at the centimeter level, making it difficult to accurately locate the source of neural activity at a finer granularity (such as the voxel level).

[0017] This invention achieves the simultaneous fusion of fMRI and EEG bimodal neural signals, constructing a brain activity analysis system with both spatial and temporal resolution. Technically, this overcomes the limitations of existing single-modality fitting by integrating the millimeter-level spatial resolution of fMRI with the millisecond-level temporal resolution of EEG, improving the parameter estimation accuracy of digital brain models and providing a technical foundation for building a highly biologically reliable digital brain platform. In terms of application, this invention is applied to the execution of continuous tasks. For example, taking a driving task as an example, by collecting fMRI and EEG signals during the task, a virtual current input corresponding to the real stimulus is inferred based on a multimodal assimilation framework. This current input is then applied to the input brain region of the digital brain (such as the primary visual cortex), achieving simultaneous fitting of the fMRI and EEG signals, thereby reproducing the brain's dynamic behavior during task execution. By analyzing the spatiotemporal activation patterns of different brain regions after stimulation, the inherent laws of brain information encoding and the mechanism of cross-regional information transmission are revealed, providing a new analytical dimension for cognitive neuroscience research. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of EEG and BOLD signals in digital twin brain simulation experiments.

[0019] Figure 2 Verification of multimodal data assimilation methods on toy models. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0021] This invention is used to fit multimodal data of varying spatiotemporal resolution within a single digital twin brain system. Using an improved data assimilation algorithm, the parameters of the digital brain model are inferred from multimodal signals (BOLD and EEG). Simulations are then performed using the estimated parameters, and the BOLD and EEG signals generated by the digital brain are observed and compared with experimental data. Based on existing brain imaging technology, this invention constructs the digital brain using structural data (DTI and T1w) from a real brain and neuroimaging data (BOLD signals acquired from fMRI and EEG signals acquired using an MRI-compatible 64-channel cap). This improves upon an existing hierarchical mesoscopic data assimilation algorithm, filtering the model state based on the augmented observation vector (experimental / system-simulated BOLD signals and experimental EEG signals) at each update step. This allows the estimated parameters to fit both observation signals simultaneously.

[0022] The digital brain data assimilation system based on multimodal data fusion provided by the present invention includes: an EEG forward model construction module, an MRI-EEG space mapping construction module, a network structure generation module, a multimodal data assimilation module, a neural network simulation module and a digital brain similarity evaluation module.

[0023] The EEG forward model building module primarily calculates the lead field matrix, which represents the mapping relationship between source current and scalp electrode potentials. First, the Brainstorm toolbox is used to spatially align the EEG electrode coordinates with the individual MRI data based on the anatomical landmarks (left and right auricular points, and nasion) marked during data acquisition. This aligns the voxel coordinates and electrode spatial positions in the individual T1w space. The OpenMEEG boundary element method (BEM) is then used to calculate the lead field matrix. The resulting voxel coordinates are used to generate a brain mask in the individual T1w space, which is then used for subsequent DTI fiber tracking and BOLD signal extraction.

[0024] The MRI-EEG spatial mapping module primarily addresses the spatial resolution discrepancy between MRI and EEG data. Based on the EEG forward model, the system calculates the lead field sensitivity distribution for each electrode and uses an optimization algorithm to find the voxel-electrode mapping relationship with the maximum total lead field sensitivity. This mapping is then used in the subsequent multimodal data assimilation module.

[0025] The network structure generation module generates a computational brain model (mathematical brain model) of the subject based on the MRI signals measured by the experimental subject, including brain diffusion tensor imaging DTI and cerebral cortical gray matter density, as well as the brain mask generated in the EEG forward model construction system. The parameters that can be set include the total number of neurons, the number of neurons included in the minimum neural network (voxel or functional column), the network link degree, the synaptic transmitter conductivity coefficient, etc.

[0026] The multimodal data assimilation module is the core of the algorithm. It mainly uses the multimodal integrated Kalman filtering method based on the multi-layer Bayesian estimation theory to assimilate the synaptic conductance coefficients of neurons in the minimum neural network (voxel or functional column). Here, we assume that the synaptic conductance coefficients of neurons in the same minimum neural network (voxel or functional column) follow the same probability distribution, so only the hyperparameters of the distribution need to be assimilated. In the implementation of the multimodal integrated Kalman filtering method: (1) The update step uses the mapping relationship established by the MRI-EEG spatial mapping system to modify the parameters of each voxel through the augmented observation vector (composed of the corresponding electrode EEG signal and the measured BOLD signal spliced ​​together); (2) In the time step with only EEG observations, the BOLD signal predicted by the model is used to interpolate the empirical data to reduce the noise fluctuations introduced by the EEG high-frequency components in the BOLD signal simulation.

[0027] The neural network simulation module primarily simulates the network given assimilated parameters. The system includes a plotting program that can plot the average firing rate and average neuron membrane potential of the minimum neural network. A specific neuron number can also be entered to observe that number, outputting detailed information including the neuron's membrane potential, synaptic current, firing time, and more.

[0028] The digital brain similarity assessment module is primarily validated through multimodal neural signal comparison. By calculating the correlation (Pearson correlation coefficient) between the experimentally collected EEG / BOLD signals and the simulated signals, it quantitatively assesses the similarity between the digital brain model and the biological brain in terms of functional activity.

[0029] The workflow of the digital brain data assimilation system based on multimodal data fusion provided by the present invention is as follows: Step 1: Lead field matrix calculation. Individual T1w images were segmented using Freesurfer and imported into the Brainstorm toolbox, where standardized MRI registration and MNI spatial normalization procedures were performed. The experimental EEG data were then imported and spatially co-registered with the MRI structural image and EEG electrode coordinates. At this point, both the electrode spatial positions and voxel coordinates were in individual space. The "compute head model" function in Brainstorm was invoked, selecting the OpenMEEG boundary element model (BEM) and calculating the lead field matrix using the default conductivity parameters for each tissue (gray matter / skull / scalp). Finally, a brain mask was generated based on the voxel coordinates in individual T1w space.

[0030] Step 2: MRI-EEG spatial mapping calculation. Based on the EEG forward model, the lead field sensitivity distribution of each scalp electrode is calculated, and the contribution of each voxel to the potential of each electrode is quantified. A voxel-electrode bipartite graph model is constructed and the Hungarian algorithm is used to solve the optimal matching problem (optimization goal: maximize overall lead field sensitivity) to determine the voxel-electrode correspondence with the greatest overall sensitivity.

[0031] Step 3: The network structure is generated by inputting the storage address of the MRI obtained by scanning the subject, including the DTI and gray matter density data, as well as the storage address of the brain mask. The network scale parameters are set to generate the computational brain model of the subject, and the brain model storage address is output to the data assimilation module.

[0032] Step 4: After receiving the brain model storage address and the subject's EEG and BOLD signals, the data assimilation module runs the data assimilation algorithm, updating the parameters of the assimilated brain model so that the brain model output is similar to the subject's EEG and BOLD signals. The assimilated parameters are then stored and input into the neural network simulation system. The data assimilation algorithm is as follows: Based on the EEG forward model, the lead field sensitivity of the electrode to the voxel is calculated. The correspondence between voxels and electrodes is calculated with the goal of maximizing the overall lead field sensitivity. During the update step, the parameters of each voxel are modified by an augmented observation vector generated primarily by concatenating the corresponding electrode EEG signal and its BOLD signal. During time steps when only the experimental EEG signal is observed, the experimental BOLD signal is interpolated using the BOLD signal predicted by the model in the previous time step as the observation to reduce the noise fluctuations in the simulated BOLD signal caused by the high-frequency information of the EEG.

[0033] Step 5: After receiving the assimilated parameters, the neural network simulation system starts to simulate the network. Note that the simulation system does not The system uses the subject's EEG / normalized BOLD signal as input, yet can still simulate output signals similar to the subject's EEG / BOLD signal. Leveraging multi-GPU parallel computing, it achieves assimilation and simulation of large-scale neural networks, reducing the reduction ratio to less than 100.

[0034] Step 6: Drawing and storage. After the simulation system is finished running, the drawing module will draw and display the assimilated parameter sequence, simulated EEG / BOLD signals and original EEG / BOLD signals in the whole process. In addition, you can choose to display the average membrane potential, average firing rate, firing time of sampled neurons and other information of the smallest neural network (brain area or voxel), and store all the results in the memory.

[0035] Step 7: The Pearson correlation coefficient between the EEG / BOLD signals obtained from the digital twin brain and the original EEG / BOLD signals is calculated to measure the similarity between the functional activities of the digital brain model and the biological brain.

[0036] like Figure 1 As shown, Figure 1 Schematic diagram of using a digital twin brain to simulate experimental EEG and BOLD signals. The basic architecture of the digital twin brain is generated based on individual structural terms (T1w, DTI) and functional column connectivity tables. The synaptic conductance of neurons is estimated by assimilating and fitting the experimental EEG and BOLD signals through multimodal data. A simulation of the neural network is run with the estimated parameters, with an iteration step of 1 millisecond. During the simulation, neuronal firing time series and the postsynaptic current of each voxel are recorded. For the BOLD signal, the average firing rate of each voxel is calculated based on the firing of neurons and used as input to the balloon model to generate the BOLD signal. For the EEG signal, each voxel is modeled as an equivalent dipole. The magnitude of the source current is given by the postsynaptic current of that voxel, and the direction is determined by the direction vector between that voxel and its connected voxels (the weight of the direction vector is the ratio of the number of connected edges to the total number of connected edges). The source currents of all voxels are input to the forward model, which outputs the EEG signal.

[0037] Figure 2Schematic diagram of multimodal data assimilation (left) and validation results on a toy model (right). Left: At each assimilation update step, the system state is corrected based on the augmented observation vector, which is obtained by concatenating the BOLD signal of each voxel and the EEG signal of its corresponding electrode. The correspondence between voxels and electrodes determines the electrode corresponding to each voxel. When only the experimental EEG signal is observed, the voxel's BOLD signal uses the BOLD signal predicted by the dynamic model at the previous time step as the estimated BOLD observation at the current time step to reduce the interference of high-frequency EEG noise on the BOLD signal estimate. The state correction is calculated using a diffuse ensemble Kalman filter. Right: The toy model was constructed as follows: the network consists of 10 voxels, with 2000 neurons per voxel (excitatory: inhibitory ratio = 4:1). The average in-degree of each neuron is 100, and the ratio of inter-voxel excitatory edges to intra-voxel excitatory edges to inhibitory edges is 2:4:1. Interneuron connection weights are sampled from a uniform distribution between [0, 1]. Inhibitory connections exist only within voxels, and inter-voxel connection probabilities are given by an artificially generated 10x10 matrix (row sums 1, diagonal zeros). The lead field matrix (single electrode) and voxel positions for the EEG forward model were artificially generated. Network simulations were run with a pre-given set of nontrivial parameters to generate observed EEG and BOLD signals. A single-neuron model was fitted to the observed signals using a leaky integrate-and-fire model. Multimodal data assimilation was then performed to fit the observed signals. Network simulations were run again with the estimated parameters, and the Pearson correlations between the simulated and observed signals reached 80% (BOLD signal) and 82% (EEG signal).

Claims

1. A digital brain data assimilation system based on multimodal data fusion, characterized in that: It includes an EEG forward model building module, an MRI-EEG spatial mapping building module, a network structure generation module, a multimodal data assimilation module, a neural network simulation module, and a digital brain similarity assessment module; among which: EEG forward model building module, used to calculate the lead field matrix that represents the mapping relationship between source current and scalp electrode potential; The MRI-EEG spatial mapping construction module calculates the lead field sensitivity distribution of each electrode based on the EEG forward model and uses an optimization algorithm to find the voxel-electrode mapping relationship with the maximum total lead field sensitivity; The network structure generation module generates a computational brain model of the experimental subject based on the MRI signals measured by the experimental subject. The parameters that can be set in the computational brain model include the total number of neurons, the minimum number of neurons included in the neural network, the network link degree, and the synaptic transmitter conductivity coefficient; The multimodal data assimilation module uses a multimodal integrated Kalman filtering method based on multi-layer Bayesian estimation theory to assimilate the synaptic conductance coefficients of the neurons in the minimum neural network and update the parameters in the assimilated and calculated brain model so that the output of the brain model is similar to the subject's EEG and BOLD signals. In the implementation of the multimodal integrated Kalman filtering method, the update step uses the mapping relationship established by the MRI-EEG spatial mapping construction module to correct the parameters of each voxel through the augmented observation vector composed of the EEG signal of the corresponding electrode and the measured BOLD signal. In time steps with only EEG observations, the BOLD signal predicted by the model is used to interpolate the empirical data to reduce the noise fluctuations introduced by the EEG high-frequency components in the BOLD signal simulation. Neural network simulation module, which performs network simulation on given assimilated parameters; The digital brain similarity assessment module is verified through multimodal neural signal comparison.

2. The digital brain data assimilation system based on multimodal data fusion according to claim 1 is characterized in that: The EEG forward model construction module realizes the spatial registration of EEG electrode coordinates with individual MRI data based on the anatomical landmarks marked during the data acquisition process, calculates the lead field matrix based on the co-registered voxel position, electrode position and orientation, and the conductivity of different brain tissues; and generates a brain mask for DTI tracking and BOLD signal extraction based on the voxel coordinates in the individual space.

3. The digital brain data assimilation system based on multimodal data fusion according to claim 1 is characterized in that: In the MRI-EEG spatial mapping construction module, the lead field sensitivity distribution of each scalp electrode is calculated based on the EEG forward model, the contribution of each voxel to the potential of each electrode is quantified, and a voxel-electrode bipartite graph model is constructed. With maximizing the overall lead field sensitivity as the optimization goal, the Hungarian algorithm is applied to solve the optimal matching problem and determine the voxel-electrode correspondence with the maximum total sensitivity.

4. The digital brain data assimilation system based on multimodal data fusion according to claim 1, characterized in that: The MRI signals include brain diffusion tensor imaging (DTI) and cerebral cortical gray matter density, as well as brain masks generated in the EEG forward model building system.

5. The digital brain data assimilation system based on multimodal data fusion according to claim 1 is characterized in that: The neural network simulation module encapsulates a drawing program that is used to draw the average firing rate and average neuron membrane potential of the minimum neural network. It also observes the neuron with a specific number by inputting that number, and outputs specific information including the neuron's membrane potential, synaptic current, and firing time.

6. The digital brain data assimilation system based on multimodal data fusion according to claim 1, characterized in that: The digital brain similarity assessment module quantitatively evaluates the similarity between the digital brain model and the biological brain in functional activities by calculating the correlation between the EEG / BOLD signals collected in the experiment and the analog signals.

7. A digital brain data assimilation method based on multimodal data fusion, which is based on the digital brain data assimilation system according to claim 1, characterized in that: The following steps are involved: (1) Calculate the lead field matrix that represents the mapping relationship between source current and scalp electrode potential; (2) Based on the lead field sensitivity distribution of each electrode, an optimization algorithm is used to find the voxel-electrode mapping relationship with the maximum total lead field sensitivity; (3) Generate a computational brain model of the subject based on the MRI signals measured by the experimental subject; the parameters that can be set in the computational brain model include the total number of neurons, the minimum number of neurons in the neural network, the network link degree, and the synaptic transmitter conductivity coefficient; (4) Based on the multimodal integrated Kalman filtering method of multi-layer Bayesian estimation theory, the synaptic conductance coefficients of the neurons in the minimum neural network are assimilated, and the parameters in the assimilated and calculated brain model are updated so that the output of the brain model is similar to the EEG and BOLD signals of the subjects. In the implementation of the multimodal integrated Kalman filtering method, the update step uses the mapping relationship established by the MRI-EEG spatial mapping construction module to correct the parameters of each voxel through the augmented observation vector composed of the corresponding electrode EEG signal and the measured BOLD signal. In the time step with only EEG observation, the BOLD signal predicted by the model is used to interpolate the empirical data to reduce the noise fluctuation introduced by the EEG high-frequency component to the BOLD signal simulation. (5) Perform network simulation for given assimilated parameters; (6) Verification is achieved through multimodal neural signal comparison.

8. The digital brain data assimilation method according to claim 7, characterized in that: In step (1), the EEG forward model construction module realizes the spatial registration of EEG electrode coordinates with individual MRI data based on the anatomical landmarks marked during the data acquisition process, calculates the lead field matrix based on the co-registered voxel positions, electrode positions and directions, and the conductivity of different brain tissues; and generates a brain mask for DTI tracking and BOLD signal extraction based on the voxel coordinates in the individual space.

9. The digital brain data assimilation method according to claim 7, characterized in that: In step (2), the lead field sensitivity distribution of each scalp electrode is calculated based on the EEG forward model, the contribution of each voxel to the potential of each electrode is quantified, and a voxel-electrode bipartite graph model is constructed. With the maximization of the overall lead field sensitivity as the optimization goal, the Hungarian algorithm is applied to solve the optimal matching problem and determine the voxel-electrode correspondence with the maximum total sensitivity.

Images