Artificial intelligence-based brain neural network modeling analysis method and system

Abstract

The invention relates to the technical field of artificial intelligence, and provides a brain neural network modeling analysis method and system based on artificial intelligence, and the method comprises the steps: obtaining electroencephalogram signal data of a target multi-channel, carrying out the synchronization processing, obtaining a synchronous electroencephalogram signal sequence, carrying out the extraction of frequency domain features and time domain features, and obtaining a clustering result of a neural function region; modeling a cranial neural network connection mode according to the clustering result of the neural function area, obtaining a function connection graph of the cranial neural network, inputting the function connection graph into a preset classification analysis model, evaluating the function state of the cranial neural network, and obtaining an evaluation result of the brain function state. And generating a neural regulation strategy of the target brain region with the functional connection diagram of the brain neural network. The brain neural network function connection diagram is generated by acquiring the multi-channel electroencephalogram signal data and processing the multi-channel electroencephalogram signal data, so that the accuracy and reliability of electroencephalogram signal analysis are improved, and the problem that the accuracy and robustness are insufficient when complex electroencephalogram signal data are processed is solved.

Description

Technical Field

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a brain neural network modeling and analysis method and system based on artificial intelligence. Background Art

[0002] With the rapid development of artificial intelligence technology, brain neural network modeling and analysis have been widely used in brain science research, neurological disease diagnosis and treatment, etc. Through deep mining of EEG signal data, the spatiotemporal characteristics and functional state of brain neuron activity can be revealed, providing an important basis for the study of brain function.

[0003] Among the related technical means, the brain neural network analysis methods mainly rely on linear analysis and statistical models. These methods extract the time domain characteristics and frequency domain characteristics of EEG signals, establish the spatiotemporal distribution model of neuronal activity, and perform cluster analysis. They achieve the preliminary functional division and correlation analysis of brain neural activity areas, laying the foundation for the study of brain neural networks.

[0004] Regarding the above technical solution, although the preliminary analysis of brain neural network can be achieved through traditional linear analysis and statistical models, there are problems of insufficient accuracy and robustness when processing complex EEG signal data, such as extraction of nonlinear features and processing of multivariate data. Summary of the invention

[0005] In order to improve the problems of insufficient accuracy and robustness when processing complex EEG signal data, such as extraction of nonlinear features and processing of multivariate data, the present application provides a brain neural network modeling and analysis method and system based on artificial intelligence.

[0006] The present invention provides a brain neural network modeling and analysis method based on artificial intelligence, comprising: acquiring target multi-channel electroencephalogram signal data, and performing channel synchronization processing to obtain a synchronized electroencephalogram signal sequence, extracting frequency domain features and time domain features of the synchronized electroencephalogram signal sequence to obtain a frequency domain feature parameter set and a time domain feature parameter set, inputting the frequency domain feature parameter set and the time domain feature parameter set into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity; using a clustering algorithm to classify the neuron activity in the spatiotemporal distribution matrix to obtain active area clusters and static area clusters, performing functional division on the active area clusters to obtain an active area functional distribution map, and performing classification on the active area clusters according to the active area functional distribution map. The static area clusters are subjected to adjacent functional correlation analysis to obtain a static area functional association map; based on the active area functional distribution map and the static area functional association map, the interaction relationship between brain nerve activity areas is analyzed to obtain a clustering result of neural function areas; according to the clustering result of the neural function areas, the brain neural network connection pattern is modeled to obtain a functional connection map of the brain neural network, the functional connection map of the brain neural network is input into a preset classification analysis model, and the functional state of the brain neural network is evaluated based on the classification analysis model to obtain an evaluation result of the brain function state; based on the evaluation result of the brain function state and the functional connection map of the brain neural network, a neural regulation strategy for the target brain area is generated.

[0007] As a preferred solution, the steps of acquiring target multi-channel EEG signal data, performing channel synchronization processing to obtain a synchronized EEG signal sequence, extracting frequency domain features and time domain features of the synchronized EEG signal sequence to obtain a frequency domain feature parameter set and a time domain feature parameter set, inputting the frequency domain feature parameter set and the time domain feature parameter set into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity include: acquiring target multi-channel EEG signal data through an EEG device, performing channel synchronization processing using a synchronous sampling technique to obtain a synchronized EEG signal sequence, performing wavelet decomposition processing on the synchronized EEG signal sequence to obtain multi-resolution frequency domain components and time domain residual components; applying Fourier transform Perform frequency feature extraction on the frequency domain component to obtain a frequency domain feature parameter set, apply a sliding window method to extract time feature on the time domain residual component to obtain a time domain feature parameter set, normalize the frequency domain feature parameter set and the time domain feature parameter set to obtain a normalized feature parameter set; input the normalized feature parameter set into a preset deep neural network model, extract local features through a convolution layer in the deep neural network model, reduce the dimension of the local features by a pooling layer to obtain a local activity distribution of neurons, calculate the global activity distribution of neurons through a fully connected layer, fuse the local activity distribution with the global activity distribution to obtain a spatiotemporal distribution matrix of neuron activity.

[0008] As a preferred solution, the steps of using a clustering algorithm to classify the neuron activity in the spatiotemporal distribution matrix to obtain active area clusters and static area clusters, performing functional division on the active area clusters to obtain an active area functional distribution map, and performing adjacent functional correlation analysis on the static area clusters according to the active area functional distribution map to obtain a static area functional association map include: performing block processing on the spatiotemporal distribution matrix by a k-means clustering algorithm to obtain neuron activity distributions in multiple small areas, classifying all the neuron activity distributions by a density clustering algorithm to obtain active area clusters and static area clusters; performing functional division on the active area clusters by using a k-means++ algorithm to obtain an active area functional distribution map, analyzing the boundary relationship between the active area clusters and the static area clusters by a regional growing algorithm to obtain a static area functional boundary map; performing cross-analysis on the active area functional distribution map and the static area functional boundary map by a cross-analysis method based on Euclidean distance and cosine similarity to extract potential correlation features within the static area clusters to obtain a static area functional association map.

[0009] As a preferred solution, the steps of applying the k-means++ algorithm to perform functional division on the active region clusters to obtain an active region functional distribution map, analyzing the boundary relationship between the active region cluster and the static region cluster by a region growing algorithm to obtain a functional boundary map of the static region include: applying the k-means++ algorithm to randomly select a cluster center, and selecting the point farthest from the cluster center as a new cluster center, and after all cluster centers are selected by analogy, calculating the mean of the neural activity features in each active region cluster, and dividing the neural activity features in each active region cluster to the nearest cluster center according to the mean to obtain the divided active region clusters; visualizing the divided active region clusters according to functional features, using color coding to represent the functional strength of different clusters, and obtaining an active region functional distribution map; using the boundary between the active region cluster and the static region cluster as a growth seed point, setting a preset growth threshold around the growth seed point by a region growing algorithm, judging whether adjacent points meet the functional similarity requirements, and if so, incorporating them into the current boundary area to obtain a functional boundary map of the static area; wherein the boundary area is a transition area between the active region and the static region.

[0010] As a preferred scheme, the step of analyzing the interaction relationship between brain nerve activity areas based on the active area function distribution map and the static area function association map to obtain the clustering results of neural function areas includes: using the interaction influence analysis method to perform spatiotemporal dynamic coupling analysis on the active area function distribution map and the static area function association map to obtain the interaction influence matrix of neural function areas, extracting high contribution factors of the interaction influence matrix through principal component analysis to obtain a preliminary functional area clustering map; applying an iterative optimization algorithm to refine the preliminary functional area clustering map to obtain a quantitative structure of the interaction relationship between regions, calculating the interaction weight matrix between regions based on the quantitative structure, and establishing a hierarchical relationship network of neural function areas based on the interaction weight matrix; classifying the hierarchical relationship network into functional areas through a hierarchical clustering algorithm to generate clustering results of neural function areas.

[0011] As a preferred scheme, the steps of modeling the brain neural network connection pattern according to the clustering results of the neural function areas to obtain a functional connection map of the brain neural network, inputting the functional connection map of the brain neural network into a preset classification analysis model, and evaluating the functional state of the brain neural network based on the classification analysis model to obtain the evaluation result of the brain function state include: constructing a functional connection matrix between regional nodes based on the clustering results of the neural function areas, calculating the optimal connection path between each node using the shortest path algorithm, and generating a brain neural network functional connection map; applying a graph convolutional network algorithm to perform high-level feature extraction on the brain neural network functional connection map to generate an optimized brain neural network functional connection map, inputting the optimized brain neural network functional connection map into a support vector machine classifier, performing functional state classification analysis of the brain neural network, and obtaining a preliminary evaluation result; and applying a Bayesian network to perform state credibility analysis on the preliminary evaluation result to generate an evaluation result of the brain function state.

[0012] As a preferred embodiment, the step of generating a neural regulation strategy for a target brain area based on the evaluation results of the brain function state and the functional connection map of the brain neural network comprises: jointly analyzing the evaluation results of the brain function state and the functional connection map of the brain neural network to obtain abnormal connection characteristics and a functional area distribution map of the target brain area, optimizing and adjusting the functional area distribution of the target brain area using a back propagation algorithm, analyzing the mutual relationship of abnormal connections based on the adjusted functional area distribution, and confirming the abnormal connection characteristics of the target brain area; designing a brain stimulation parameter set according to the abnormal connection characteristics, and generating a neural regulation strategy for the target brain area based on the brain stimulation parameter set.

[0013] The present application also provides a brain neural network modeling and analysis system based on artificial intelligence, including: an acquisition unit, used to acquire target multi-channel EEG signal data, and perform channel synchronization processing to obtain a synchronized EEG signal sequence, extract frequency domain features and time domain features of the synchronized EEG signal sequence to obtain a frequency domain feature parameter set and a time domain feature parameter set, input the frequency domain feature parameter set and the time domain feature parameter set into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity; a classification unit, used to classify the neuron activity in the spatiotemporal distribution matrix using a clustering algorithm to obtain active area clusters and static area clusters, perform functional division on the active area clusters to obtain an active area functional distribution map, and classify the static area clusters according to the active area functional distribution map. The regional clusters are used to perform adjacent functional correlation analysis to obtain a static regional functional association map; an analyzing unit is used to analyze the interaction relationship between brain nerve activity areas based on the active regional functional distribution map and the static regional functional association map to obtain a clustering result of the neural function areas; a modeling unit is used to model the brain neural network connection pattern according to the clustering result of the neural function areas to obtain a functional connection map of the brain neural network, input the functional connection map of the brain neural network into a preset classification analysis model, and evaluate the functional state of the brain neural network based on the classification analysis model to obtain an evaluation result of the brain function state; a generating unit is used to generate a neural regulation strategy for the target brain area based on the evaluation result of the brain function state and the functional connection map of the brain neural network.

[0014] Compared with the prior art, the present application has the following beneficial effects: high accuracy and sufficient robustness. By acquiring multi-channel EEG signal data and performing channel synchronization processing, extracting frequency domain and time domain feature parameters, and inputting the deep neural network model, the spatiotemporal distribution matrix of neuron activity is obtained; the neuron activity is classified by clustering algorithm to obtain active area clusters and static area clusters, and functional division and correlation analysis are performed to obtain active area functional distribution maps and static area functional association maps; based on the functional distribution map and the association map, the interaction relationship between brain nerve activity areas is analyzed to obtain the clustering results of neural function areas; according to the clustering results, the brain neural network connection mode is modeled to obtain the functional connection map, and the functional state of the brain neural network is evaluated to generate the neural regulation strategy of the target brain area, which improves the accuracy and reliability of EEG signal analysis, generates a more accurate brain neural network functional connection map, and improves the problem of insufficient accuracy and robustness when processing complex EEG signal data, such as the extraction of nonlinear features and the processing of multivariate data. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0016] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0017] Figure 1 It is a flowchart of a brain neural network modeling and analysis method based on artificial intelligence provided by an embodiment of the present invention; Figure 2 It is a schematic block diagram of the structure of the brain neural network modeling and analysis system based on artificial intelligence provided by an embodiment of the present invention.

[0018] Description of reference numerals: 10. Brain neural network modeling and analysis system based on artificial intelligence; 11. Acquisition unit; 12. Classification unit; 13. Analysis unit; 14. Modeling unit; 15. Generation unit. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0021] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0022] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0023] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0024] Embodiment 1: like Figure 1 As shown, the present application provides a brain neural network modeling and analysis method based on artificial intelligence, including steps S100 to S500.

[0025] Step S100, obtain the target multi-channel EEG signal data, and perform channel synchronization processing to obtain a synchronized EEG signal sequence, extract the frequency domain features and time domain features of the synchronized EEG signal sequence to obtain a frequency domain feature parameter set and a time domain feature parameter set, input the frequency domain feature parameter set and the time domain feature parameter set into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity.

[0026] In this step, the acquired target multi-channel EEG signal data is first preprocessed, including noise filtering and baseline drift correction. Then, the EEG signals of multiple channels are synchronously processed by a synchronization algorithm to obtain a synchronized EEG signal sequence. Specifically, the frequency domain features of the synchronized EEG signal sequence are extracted by fast Fourier transform (FFT) to obtain a frequency domain feature parameter set; at the same time, the time domain features of the synchronized EEG signal sequence are extracted by wavelet transform (WT) to obtain a time domain feature parameter set. The extracted frequency domain feature parameter set and time domain feature parameter set are input into the preset convolutional neural network (CNN) model to obtain the spatiotemporal distribution matrix of neuronal activity.

[0027] For example, when processing EEG signal data containing 32 channels, the data length of each channel is 10,000 sample points. After the above processing, 32 frequency domain feature parameter sets and 32 time domain feature parameter sets can be obtained. After input into the CNN model, a 32×10,000 spatiotemporal distribution matrix of neuron activity is obtained.

[0028] Step S200, using a clustering algorithm to classify the neuronal activity in the spatiotemporal distribution matrix to obtain active area clusters and static area clusters, performing functional division on the active area clusters to obtain an active area functional distribution map, performing adjacent functional correlation analysis on the static area clusters based on the active area functional distribution map to obtain a static area functional association map.

[0029] In this step, the K-means clustering algorithm is used to classify the neuron activity in the spatiotemporal distribution matrix, and neurons with similar activity characteristics are grouped into the same cluster. Specifically, areas with higher neuron activity are divided into active area clusters, and areas with lower activity are divided into static area clusters. Then, by analyzing the neuron functions in the active area clusters, an active area function distribution map is drawn. Next, based on the active area function distribution map, the static area clusters are subjected to adjacent function correlation analysis, and a static area function association map is drawn.

[0030] For example, in a 32×10000 spatiotemporal distribution matrix, neurons are divided into 5 active region clusters and 5 static region clusters by the K-means clustering algorithm. After analyzing the neuronal functions in the active region clusters, an active region function distribution map containing 5 functional regions can be drawn. Based on this function distribution map, after performing adjacent function correlation analysis on the static region clusters, a static region function association map is obtained.

[0031] Step S300: Analyze the interaction relationship between brain nerve activity areas based on the active area function distribution map and the static area function association map to obtain a clustering result of the nerve function areas.

[0032] In this step, the interaction between different brain nerve activity areas is identified by analyzing the connection patterns of each functional area in the active area functional distribution map. Specifically, the connection strength and correlation between each functional area are calculated using graph theory analysis methods to obtain the clustering results of brain nerve functional areas.

[0033] For example, by calculating the Pearson correlation coefficient between every two functional areas in the active area functional distribution map, a connection matrix of the functional areas is constructed, and the spectral clustering algorithm is used to perform cluster analysis on the connection matrix to obtain the clustering results of multiple cranial nerve functional areas.

[0034] Step S400: Model the brain neural network connection pattern according to the clustering results of the neural function areas to obtain a functional connection diagram of the brain neural network, input the functional connection diagram of the brain neural network into a preset classification analysis model, evaluate the functional state of the brain neural network based on the classification analysis model, and obtain an evaluation result of the brain function state.

[0035] In this step, the functional connection map of the brain neural network is constructed by modeling the clustering results of the brain nerve functional areas. Specifically, the structural equation modeling (SEM) method is used to establish the connection model between the functional areas and generate the functional connection map of the brain neural network. Then, the functional connection map is input into the preset random forest classification model to evaluate the functional state of the brain neural network and obtain the evaluation result of the brain functional state.

[0036] For example, by modeling five brain nerve functional areas, a brain neural network functional connection diagram containing five nodes is generated, and the random forest classification model is used to evaluate the brain function status of different samples to obtain the functional status classification results of each sample.

[0037] Step S500: Generate a neural regulation strategy for the target brain region based on the evaluation results of the brain function state and the functional connection diagram of the brain neural network.

[0038] In this step, the target brain regions that need to be neuromodulated are identified by analyzing the evaluation results of brain functional status. Specifically, a neuromodulation strategy is formulated based on the status and connection strength of each functional region in the functional connectivity map.

[0039] For example, if the activity of a functional area is abnormally high or low, neuromodulatory strategies can be used to regulate the activity of that area and restore it to the normal range.

[0040] In this embodiment, a synchronized EEG signal sequence is obtained by acquiring EEG signal data of the target multi-channel and performing channel synchronization processing. Then, the frequency domain features and time domain features of the synchronized EEG signal sequence are extracted to obtain a frequency domain feature parameter set and a time domain feature parameter set. The frequency domain feature parameter set and the time domain feature parameter set are input into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuronal activity. Next, a clustering algorithm is used to classify the neuronal activity in the spatiotemporal distribution matrix to obtain active area clusters and static area clusters. The active area clusters are functionally divided to obtain an active area function distribution map, and the static area clusters are subjected to adjacent function correlation analysis based on the active area function distribution map to obtain a static area function association map. Based on the active area function distribution map and the static area function association map, the interaction relationship between the brain nerve activity areas is analyzed to obtain the clustering results of the neural function areas. Finally, the brain neural network connection pattern is modeled according to the clustering results of the neural functional areas to obtain the functional connection map of the brain neural network, and the functional connection map of the brain neural network is input into the preset classification analysis model. The functional state of the brain neural network is evaluated based on the classification analysis model to obtain the evaluation results of the brain functional state, and the neural regulation strategy of the target brain area is generated based on the evaluation results and the functional connection map. Improve the accuracy and reliability of EEG signal analysis and generate a more accurate brain neural network functional connection map. It not only helps to understand the activity and functional state of brain neurons more comprehensively, but also improves the problems of insufficient accuracy and robustness when processing complex EEG signal data, such as the extraction of nonlinear features and the processing of multivariate data.

[0041] Embodiment 2: In step S100, target multi-channel EEG signal data is acquired through an EEG device, and channel synchronization processing is performed using synchronous sampling technology to obtain a synchronized EEG signal sequence, which is then subjected to wavelet decomposition processing to obtain multi-resolution frequency domain components and time domain residual components.

[0042] The data quality is ensured by preprocessing the target multi-channel EEG signal data, including denoising and baseline correction. Specifically, the wavelet decomposition method is used to decompose the synchronized EEG signal sequence into multiple frequency domain components and time domain residual components, and the EEG signal is analyzed in detail in a multi-resolution manner.

[0043] For example, using Daubechies wavelet to decompose a 32-channel EEG signal with a signal length of 10,000 sample points in each channel, 32 multi-resolution frequency domain components and 32 time domain residual components can be obtained.

[0044] Fourier transform is applied to extract frequency features of frequency domain components to obtain a frequency domain feature parameter set. A sliding window method is applied to extract time features of time domain residual components to obtain a time domain feature parameter set. The frequency domain feature parameter set and the time domain feature parameter set are normalized to obtain a normalized feature parameter set.

[0045] The frequency domain components are converted into frequency features through Fourier transform. Specifically, the power spectrum density of each frequency component is calculated. The statistical features of the time domain residual components, such as mean value, variance, etc., are extracted through the sliding window method. Finally, the extracted frequency domain feature parameter set and time domain feature parameter set are normalized to reduce the difference between different feature magnitudes.

[0046] For example, the power spectral density of the frequency domain component and the mean, variance and other characteristic parameters of the time domain residual component are normalized to the range of [0,1] to facilitate subsequent model input.

[0047] The normalized feature parameter set is input into the preset deep neural network model. In the deep neural network model, local features are extracted through the convolution layer, and the local features are reduced in dimension using the pooling layer to obtain the local activity distribution of neurons. The global activity distribution of neurons is calculated through the fully connected layer, and the local activity distribution and the global activity distribution are fused to obtain the spatiotemporal distribution matrix of neuron activity.

[0048] The normalized feature parameter set is processed through a deep neural network model. Specifically, the local features of the EEG signal are first extracted through multiple convolutional layers, and then the features are reduced in dimension through the pooling layer to reduce the amount of calculation. Then, the global activity distribution of neurons is calculated through the fully connected layer. Finally, the local activity distribution and the global activity distribution are fused to obtain a complete spatiotemporal distribution matrix of neuron activity.

[0049] For example, after inputting the normalized feature parameter set, the size of the feature map extracted by the convolution layer is 32×50, the size of the feature map after dimensionality reduction by the pooling layer is 32×25, and the size of the global activity distribution output by the fully connected layer is 1×25. Finally, a 32×25 neuron activity spatiotemporal distribution matrix is ​​obtained by fusion.

[0050] In step S200, the spatiotemporal distribution matrix is ​​divided into blocks using the k-means clustering algorithm to obtain the neuron activity distribution of multiple small areas, and all the neuron activity distributions are classified using the density clustering algorithm to obtain active area clusters and static area clusters.

[0051] By applying the k-means clustering algorithm to the spatiotemporal distribution matrix, the matrix is ​​divided into multiple small regions, specifically, each region represents neurons with similar activity characteristics. Subsequently, the distribution of neuronal activity in small regions is further classified using a density clustering algorithm (such as DBSCAN), with highly active regions being classified as active region clusters and low-activity regions being classified as static region clusters.

[0052] For example, k-means clustering is performed on the 32×25 spatiotemporal distribution matrix to obtain 5 small regions. Then, the 5 small regions are further classified into 3 active region clusters and 2 static region clusters through the DBSCAN algorithm.

[0053] The k-means++ algorithm is used to divide the active area clusters into functional groups and obtain the functional distribution map of the active area. The region growing algorithm is used to analyze the boundary relationship between the active area clusters and the static area clusters and obtain the functional boundary map of the static area.

[0054] The k-means++ algorithm is used to perform detailed functional division of the active region clusters. Specifically, a cluster center is randomly selected, and then the point farthest from the current cluster center is selected as the new cluster center, and so on, until all cluster centers are selected. According to the feature mean of each cluster center, the neurons are divided into the clusters with the closest distance to obtain the active region function distribution map. Then, the region growing algorithm is used to analyze the junction of the active region cluster and the static region cluster. According to the preset growth threshold, the functional similarity of adjacent points is judged to obtain the functional boundary map of the static region.

[0055] For example, in the active region function distribution map, color coding is used to represent the functional strength of different clusters, and the 32×25 distribution map is visualized. Through the region growing algorithm, the boundary area is used as the seed point, the growth threshold is set to 0.5, and the functional similarity of adjacent points is judged to obtain the functional boundary map of the static area.

[0056] The cross-analysis method based on Euclidean distance and cosine similarity was used to cross-analyze the functional distribution map of the active area and the functional boundary map of the static area, extract the potential correlation features within the static area cluster, and obtain the static area functional association map.

[0057] Through the cross-analysis method of Euclidean distance and cosine similarity, the functional distribution map of the active area and the functional boundary map of the static area are analyzed in detail. Specifically, the Euclidean distance and cosine similarity between each point in the static area cluster are calculated to identify potential correlation features and construct a static area functional association map.

[0058] For example, for each neuron in the static region cluster, its Euclidean distance and cosine similarity with other neurons are calculated, and potential association features are extracted based on the similarity threshold (such as 0.7) to obtain a static region functional association map containing multiple functional association nodes.

[0059] The steps of applying the k-means++ algorithm to perform functional division on the active region clusters to obtain the active region functional distribution map, analyzing the boundary relationship between the active region clusters and the static region clusters by the region growing algorithm to obtain the functional boundary map of the static region include: applying the k-means++ algorithm to randomly select a cluster center, and selecting the point farthest from the cluster center as the new cluster center, and after all cluster centers are selected in this way, calculating the mean of the neural activity features in each active region cluster, and dividing the neural activity features in each active region cluster to the nearest cluster center according to the mean to obtain the divided active region clusters.

[0060] The k-means++ algorithm is used to perform functional division on each active region cluster. Specifically, a cluster center is first randomly selected, and then the point farthest from the current cluster center is selected as the new cluster center, and so on, until all cluster centers are selected. Next, the mean of the neural activity features in each active region cluster is calculated, and the neural activity features in each active region cluster are divided into the nearest cluster center according to the mean, thus obtaining the divided active region cluster.

[0061] For example, in a spatiotemporal distribution matrix containing 32 active area clusters, 5 cluster centers are selected through the k-means++ algorithm, and the mean of the neural activity features in each cluster is calculated. The neural activity features are divided into the cluster centers closest to them, and finally 5 divided active area clusters are obtained.

[0062] The divided active area clusters are visualized according to their functional characteristics, and color coding is used to represent the functional strength of different clusters to obtain the active area functional distribution map.

[0063] The divided active region clusters are visualized according to their functional characteristics through visualization tools. Specifically, color coding is used to represent the functional strength of different clusters to help intuitively understand the functional distribution of each active region cluster.

[0064] For example, through color coding, the functional strength of the five active area clusters is represented by red, orange, yellow, green, and blue, respectively, generating a 32×25 active area functional distribution map.

[0065] The junction of the active area cluster and the static area cluster is used as the growth seed point. The preset growth threshold is set around the growth seed point through the region growing algorithm to determine whether the adjacent points meet the functional similarity requirements. If the conditions are met, they will be included in the current boundary area to obtain the functional boundary map of the static area; among which, the boundary area is the transition area between the active area and the static area.

[0066] Through the region growing algorithm, the junction of the active region cluster and the static region cluster is used as the growth seed point. Specifically, a preset growth threshold is set around the growth seed point to determine whether the adjacent points meet the functional similarity requirements. If the conditions are met, the adjacent points are included in the current boundary region, and finally a functional boundary map of the static region is formed. The boundary region is defined as the transition area between the active region and the static region.

[0067] For example, in a 32×25 spatiotemporal distribution matrix, the intersection points of 5 active region clusters and static region clusters are selected as growth seed points, the growth threshold is set to 0.5, and the functional similarity of adjacent points is judged by the region growing algorithm to obtain the functional boundary map of the static region.

[0068] In step S300, the interaction influence analysis method is used to perform spatiotemporal dynamic coupling analysis on the active area function distribution map and the static area function association map to obtain the interaction influence matrix of the neural function area. The high contribution factors of the interaction influence matrix are extracted through principal component analysis to obtain a preliminary functional area clustering map.

[0069] The interactive influence analysis method is used to couple the active area function distribution map and the static area function association map. Specifically, the interactive influence between the regions at different time points is calculated to form an interactive influence matrix. Then, principal component analysis (PCA) is used to reduce the dimension of the interactive influence matrix, extract high contribution factors, and obtain a preliminary functional area clustering map.

[0070] For example, the interaction matrix (size 32×32) at 32 time points was reduced to three principal components through PCA to generate a preliminary functional region clustering map, in which each region represented the clustering result of a high contribution factor.

[0071] An iterative optimization algorithm was applied to refine the preliminary functional region clustering map to obtain the quantitative structure of the interaction relationship between regions. The interaction weight matrix between regions was calculated based on the quantitative structure, and the hierarchical relationship network of neural functional regions was established based on the interaction weight matrix.

[0072] Through the iterative optimization algorithm, the preliminary functional region clustering map is refined. Specifically, the region boundaries are gradually adjusted, the interaction relationship between regions is optimized, and a quantitative interaction structure is formed. Then, the interaction weights between regions are calculated to generate a weight matrix. Based on this matrix, a hierarchical relationship network of neural functional regions is constructed to show the hierarchical structure and interaction between regions.

[0073] For example, through the iterative optimization algorithm, the 32 preliminary functional areas were refined to obtain the optimized regional interaction structure, the interaction weights between the 32 areas were calculated, a 32 × 32 weight matrix was generated, and a hierarchical relationship network containing 32 nodes was constructed.

[0074] The hierarchical relationship network is classified into functional areas through a hierarchical clustering algorithm to generate clustering results of neural functional areas.

[0075] The constructed hierarchical relationship network is classified into functional areas through a hierarchical clustering algorithm. Specifically, similar functional areas are gradually merged to form a hierarchical clustering tree, and finally the clustering results of neural functional areas are obtained.

[0076] For example, through the hierarchical clustering algorithm, the hierarchical relationship network of 32 nodes is gradually merged into 6 functional areas, each of which represents a highly correlated neural function area, generating a clustering result map of the neural function areas.

[0077] In step S400, a functional connection matrix between regional nodes is constructed based on the clustering results of neural function areas, and the optimal connection path between nodes is calculated using the shortest path algorithm to generate a brain neural network functional connection diagram.

[0078] Through the clustering results of neural functional areas, a functional connection matrix between regional nodes is constructed. Specifically, the connection weights between each node are calculated, and the shortest path algorithm is used to determine the optimal connection path between nodes to generate a functional connection map of the brain neural network.

[0079] For example, the connection weights of the six functional areas are calculated to generate a 6×6 functional connection matrix, and the optimal connection path between the nodes is calculated through the Dijkstra algorithm to generate a functional connection diagram of the brain neural network.

[0080] The graph convolutional network algorithm is used to extract high-level features of the brain neural network functional connection map to generate an optimized brain neural network functional connection map. The optimized brain neural network functional connection map is input into the support vector machine classifier to perform classification analysis on the functional state of the brain neural network and obtain preliminary evaluation results.

[0081] The high-level features of the brain neural network functional connection map are extracted through the graph convolutional network algorithm. Specifically, the global features in the functional connection map are extracted through a multi-layer graph convolutional network. Then, the optimized functional connection map is input into the support vector machine classifier to classify and analyze the functional state of the brain neural network and obtain preliminary evaluation results.

[0082] For example, a three-layer graph convolutional network is used to extract features from the functional connection graph of six nodes to generate a high-level feature vector, which is then input into a support vector machine for classification to obtain a preliminary evaluation result of the functional state of the brain neural network.

[0083] The Bayesian network was used to conduct state credibility analysis on the preliminary assessment results and generate assessment results of brain function status.

[0084] The state credibility analysis of the preliminary evaluation results is performed through the Bayesian network. Specifically, the credibility of each functional state is calculated using the Bayesian reasoning method to generate the final brain function state evaluation results.

[0085] For example, the Bayesian network is used to conduct credibility analysis on the results of support vector machine classification, determine the probability distribution of each functional state, and generate the final evaluation results of the brain functional state.

[0086] In step S500, the evaluation results of the brain function state and the functional connection map of the brain neural network are jointly analyzed to obtain the abnormal connection characteristics and functional area distribution map of the target brain area, and the functional area distribution of the target brain area is optimized and adjusted using the back propagation algorithm. The relationship between the abnormal connections is analyzed based on the adjusted functional area distribution to confirm the abnormal connection characteristics of the target brain area.

[0087] By jointly analyzing the evaluation results of brain function status and the functional connection map of the brain neural network, the abnormal connection characteristics of the target brain area are identified, and the functional area distribution map of the target brain area is drawn. The back propagation algorithm is used to optimize the functional area distribution of the target brain area, analyze the mutual relationship of abnormal connections, and confirm the abnormal connection characteristics of the target brain area.

[0088] For example, based on the evaluation results and functional connectivity maps, the target brain area was determined to be the left prefrontal cortex, and the functional area distribution was optimized through the back-propagation algorithm, and finally the abnormal connection characteristics were confirmed.

[0089] A brain stimulation parameter set is designed according to the abnormal connection characteristics, and a neural regulation strategy for the target brain area is generated based on the brain stimulation parameter set.

[0090] By designing a set of brain stimulation parameters suitable for the target brain area based on the abnormal connection characteristics, specifically, determining parameters such as stimulation frequency, intensity and duration, a neural regulation strategy is generated.

[0091] For example, based on the abnormal connectivity characteristics of the left prefrontal cortex, appropriate stimulation frequency (such as 10 Hz), stimulation intensity (such as 2 mA) and stimulation duration (such as 20 minutes) are designed to generate a neural regulation strategy for this target brain area.

[0092] In this embodiment, the EEG signal data of the target multi-channel is obtained by using an EEG device, and synchronous sampling technology is used to obtain a synchronous EEG signal sequence, and then the synchronous EEG signal sequence is subjected to wavelet decomposition to obtain multi-resolution frequency domain components and time domain residual components. The frequency domain and time domain feature parameters are extracted respectively by Fourier transform and sliding window methods, and normalized, and the obtained normalized feature parameter set is input into a preset deep neural network model, and the spatiotemporal distribution matrix of neuron activity is obtained by processing through convolution layer, pooling layer and fully connected layer. The spatiotemporal distribution matrix is ​​divided into blocks by k-means clustering algorithm, and classified by density clustering algorithm to obtain active area clusters and static area clusters. The k-means++ algorithm is further applied to functionally divide the active area clusters to obtain the active area functional distribution map, and the boundary relationship between the active area clusters and the static area clusters is analyzed by the regional growth algorithm to obtain the functional boundary map of the static area. Based on the cross-analysis method of Euclidean distance and cosine similarity, the functional distribution map of active areas and the functional boundary map of static areas were cross-analyzed to extract the potential correlation features within the static area cluster and obtain the static area functional correlation map. The interactive influence analysis method was used to analyze the spatiotemporal dynamic coupling of active and static areas to generate an interactive influence matrix. The functional areas were further refined through principal component analysis and iterative optimization algorithm to construct the interaction weight matrix between regions and form a hierarchical relationship network of neural functional areas. Finally, the hierarchical relationship network was classified into functional areas based on the hierarchical clustering algorithm to construct the brain neural network functional connection map. The graph convolution network and support vector machine were used to extract high-level features and classify the functional connection map. Finally, the Bayesian network was combined for state credibility analysis to generate the evaluation results of brain function state. The neural regulation strategy of the target brain area was formulated based on the evaluation results and the functional connection map, which effectively improved the analysis accuracy and reliability of EEG signal data and generated a more accurate brain neural network functional connection map.

[0093] Embodiment 3: like Figure 2As shown, the present application also provides an artificial intelligence-based brain neural network modeling and analysis system 10, including an acquisition unit 11, a classification unit 12, an analysis unit 13, a modeling unit 14 and a generation unit 15.

[0094] The acquisition unit 11 is mainly used to acquire the target multi-channel EEG signal data and perform channel synchronization processing to obtain a synchronized EEG signal sequence, extract frequency domain features and time domain features of the synchronized EEG signal sequence to obtain a frequency domain feature parameter set and a time domain feature parameter set, input the frequency domain feature parameter set and the time domain feature parameter set into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity.

[0095] The classification unit 12 is mainly used to classify the neuronal activity in the spatiotemporal distribution matrix using a clustering algorithm to obtain active area clusters and static area clusters, perform functional division on the active area clusters to obtain an active area functional distribution map, perform adjacent functional correlation analysis on the static area clusters based on the active area functional distribution map, and obtain a static area functional association map.

[0096] The analysis unit 13 is mainly used to analyze the interaction relationship between brain nerve activity areas based on the active area function distribution map and the static area function association map to obtain the clustering result of the nerve function area.

[0097] The modeling unit 14 is mainly used to model the brain neural network connection pattern according to the clustering results of the neural function areas, obtain the functional connection map of the brain neural network, input the functional connection map of the brain neural network into a preset classification analysis model, evaluate the functional state of the brain neural network based on the classification analysis model, and obtain the evaluation result of the brain function state.

[0098] The generation unit 15 is mainly used to generate a neural regulation strategy for the target brain area based on the evaluation results of the brain function state and the functional connection diagram of the brain neural network.

[0099] In this embodiment, the EEG signal data of the target multi-channel is acquired by the acquisition unit 11, and the channel synchronization processing is performed to obtain a synchronized EEG signal sequence. Then, the frequency domain features and time domain features of the synchronized EEG signal sequence are extracted to obtain a frequency domain feature parameter set and a time domain feature parameter set. The frequency domain feature parameter set and the time domain feature parameter set are input into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity. The classification unit 12 uses a clustering algorithm to classify the neuron activity in the spatiotemporal distribution matrix to obtain active area clusters and static area clusters, and performs functional division on the active area clusters to generate an active area function distribution map. According to the active area function distribution map, the static area cluster is analyzed for adjacent functional correlation to generate a static area function association map. The analysis unit 13 analyzes the interaction relationship between the brain nerve activity areas based on the active area function distribution map and the static area function association map to obtain the clustering results of the neural function areas. The modeling unit 14 models the brain neural network connection pattern according to the clustering results of the neural function areas to generate a functional connection map of the brain neural network. The functional connection diagram of the brain neural network is input into a preset classification analysis model, and the functional state of the brain neural network is evaluated by the classification analysis model to obtain an evaluation result of the brain functional state. The generation unit 15 generates a neural regulation strategy for the target brain area based on the evaluation result of the brain functional state and the functional connection diagram of the brain neural network. The system provides more accurate brain neural network modeling and functional state evaluation by effectively processing and analyzing EEG signal data, and ultimately generates a highly targeted neural regulation strategy.

[0100] It should be noted that technical personnel in the relevant technical field can clearly understand that, for the convenience and conciseness of description, the specific working process of the system and each unit described above can refer to the corresponding process in the aforementioned artificial intelligence-based brain neural network modeling and analysis method and system implementation example, and will not be repeated here.

[0101] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A brain neural network modeling and analysis method based on artificial intelligence, characterized in that: include: Acquire target multi-channel EEG signal data, and perform channel synchronization processing to obtain a synchronized EEG signal sequence, extract frequency domain features and time domain features of the synchronized EEG signal sequence to obtain a frequency domain feature parameter set and a time domain feature parameter set, input the frequency domain feature parameter set and the time domain feature parameter set into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity; Using a clustering algorithm to classify the neuron activity in the spatiotemporal distribution matrix to obtain active region clusters and static region clusters, performing functional division on the active region clusters to obtain an active region functional distribution map, performing adjacent functional correlation analysis on the static region clusters according to the active region functional distribution map to obtain a static region functional association map; Analyzing the interaction relationship between brain nerve activity areas based on the active area function distribution map and the static area function association map to obtain a clustering result of nerve function areas; Modeling the brain neural network connection pattern according to the clustering result of the neural function area to obtain a functional connection map of the brain neural network, inputting the functional connection map of the brain neural network into a preset classification analysis model, and evaluating the functional state of the brain neural network based on the classification analysis model to obtain an evaluation result of the brain functional state; Based on the evaluation results of the brain functional state and the functional connection diagram of the brain neural network, a neural regulation strategy for the target brain area is generated.

2. The method for modeling and analyzing a brain neural network based on artificial intelligence according to claim 1, characterized in that: The steps of acquiring target multi-channel EEG signal data and performing channel synchronization processing to obtain a synchronized EEG signal sequence, extracting frequency domain features and time domain features of the synchronized EEG signal sequence to obtain a frequency domain feature parameter set and a time domain feature parameter set, inputting the frequency domain feature parameter set and the time domain feature parameter set into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity include: Acquire target multi-channel EEG signal data through an EEG device, perform channel synchronization processing using a synchronous sampling technique to obtain a synchronized EEG signal sequence, perform wavelet decomposition processing on the synchronized EEG signal sequence to obtain a multi-resolution frequency domain component and a time domain residual component; Applying Fourier transform to extract frequency features of the frequency domain component to obtain a frequency domain feature parameter set, applying a sliding window method to extract time features of the time domain residual component to obtain a time domain feature parameter set, and normalizing the frequency domain feature parameter set and the time domain feature parameter set to obtain a normalized feature parameter set; The normalized feature parameter set is input into a preset deep neural network model. In the deep neural network model, local features are extracted through a convolution layer, and the local features are reduced in dimension using a pooling layer to obtain the local activity distribution of neurons. The global activity distribution of neurons is calculated through a fully connected layer, and the local activity distribution and the global activity distribution are fused to obtain a spatiotemporal distribution matrix of neuron activity.

3. The method for modeling and analyzing a brain neural network based on artificial intelligence according to claim 1, characterized in that: The steps of using a clustering algorithm to classify the neuron activity in the spatiotemporal distribution matrix to obtain active region clusters and static region clusters, performing functional division on the active region clusters to obtain an active region functional distribution map, and performing adjacent functional correlation analysis on the static region clusters according to the active region functional distribution map to obtain a static region functional association map include: The spatiotemporal distribution matrix is ​​divided into blocks by a k-means clustering algorithm to obtain neuron activity distributions in multiple small areas, and all the neuron activity distributions are classified by a density clustering algorithm to obtain active area clusters and static area clusters; Using the k-means++ algorithm to perform functional division on the active region cluster to obtain an active region functional distribution map, and using the region growing algorithm to analyze the boundary relationship between the active region cluster and the static region cluster to obtain a static region functional boundary map; A cross analysis method based on Euclidean distance and cosine similarity is used to cross-analyze the active region function distribution map and the static region function boundary map, extract potential correlation features within the static region cluster, and obtain a static region function correlation map.

4. The method for modeling and analyzing a brain neural network based on artificial intelligence according to claim 3 is characterized in that: The step of applying the k-means++ algorithm to perform functional division on the active region cluster to obtain an active region functional distribution map, and analyzing the boundary relationship between the active region cluster and the static region cluster by a region growing algorithm to obtain a static region functional boundary map comprises: Apply the k-means++ algorithm to randomly select a cluster center, and select the point farthest from the cluster center as the new cluster center. After selecting all cluster centers in this way, calculate the mean of the neural activity features in each active area cluster, and divide the neural activity features in each active area cluster to the nearest cluster center according to the mean, so as to obtain the divided active area clusters; The divided active region clusters are visualized according to functional characteristics, and the functional strengths of different clusters are represented by color coding to obtain an active region functional distribution map; The junction of the active area cluster and the static area cluster is used as a growth seed point. A preset growth threshold is set around the growth seed point through a regional growing algorithm to determine whether adjacent points meet the functional similarity requirements. If the conditions are met, the points are included in the current boundary area to obtain a functional boundary map of the static area; wherein the boundary area is a transition area between the active area and the static area.

5. The method for modeling and analyzing brain neural network based on artificial intelligence according to claim 1, characterized in that: The step of analyzing the interaction relationship between brain nerve activity areas based on the active area function distribution map and the static area function association map to obtain the clustering result of nerve function areas includes: The interactive influence analysis method is used to perform spatiotemporal dynamic coupling analysis on the active area function distribution map and the static area function association map to obtain the interactive influence matrix of the neural function area, and the high contribution factors of the interactive influence matrix are extracted by principal component analysis to obtain a preliminary functional area clustering map; Applying an iterative optimization algorithm to refine the preliminary functional region clustering map, obtaining a quantitative structure of the interaction relationship between regions, calculating an interaction weight matrix between regions based on the quantitative structure, and establishing a hierarchical relationship network of neural functional regions based on the interaction weight matrix; The hierarchical relationship network is classified into functional areas through a hierarchical clustering algorithm to generate a clustering result of neural functional areas.

6. The method for modeling and analyzing brain neural network based on artificial intelligence according to claim 1, characterized in that: The step of modeling the brain neural network connection pattern according to the clustering result of the neural function area to obtain a functional connection map of the brain neural network, inputting the functional connection map of the brain neural network into a preset classification analysis model, and evaluating the functional state of the brain neural network based on the classification analysis model to obtain an evaluation result of the brain functional state includes: Based on the clustering results of the neural function areas, a functional connection matrix between regional nodes is constructed, and the optimal connection path between each node is calculated using the shortest path algorithm to generate a brain neural network functional connection diagram; Applying a graph convolutional network algorithm to extract high-level features of the brain neural network functional connection map to generate an optimized brain neural network functional connection map, inputting the optimized brain neural network functional connection map into a support vector machine classifier to perform classification analysis on the functional state of the brain neural network to obtain a preliminary evaluation result; A Bayesian network is used to perform state credibility analysis on the preliminary assessment results to generate assessment results of brain function status.

7. The method for modeling and analyzing brain neural network based on artificial intelligence according to claim 1, characterized in that: The step of generating a neural regulation strategy for a target brain region based on the evaluation result of the brain function state and the functional connection diagram of the brain neural network comprises: Performing a joint analysis on the evaluation result of the brain function state and the functional connection map of the brain neural network to obtain abnormal connection characteristics and functional area distribution map of the target brain area, optimizing and adjusting the functional area distribution of the target brain area using a back propagation algorithm, analyzing the mutual relationship of abnormal connections based on the adjusted functional area distribution, and confirming the abnormal connection characteristics of the target brain area; A brain stimulation parameter set is designed according to the abnormal connection characteristics, and a neural regulation strategy for the target brain area is generated based on the brain stimulation parameter set.

8. A brain neural network modeling and analysis system based on artificial intelligence, characterized in that: include: An acquisition unit is used to acquire EEG signal data of a target multi-channel and perform channel synchronization processing to obtain a synchronized EEG signal sequence, extract frequency domain features and time domain features of the synchronized EEG signal sequence to obtain a frequency domain feature parameter set and a time domain feature parameter set, input the frequency domain feature parameter set and the time domain feature parameter set into a preset deep neural network model to obtain a spatiotemporal distribution matrix of neuron activity; a classification unit, configured to classify the neuron activity in the spatiotemporal distribution matrix by using a clustering algorithm to obtain active region clusters and static region clusters, perform functional division on the active region clusters to obtain an active region functional distribution map, and perform adjacent functional correlation analysis on the static region clusters according to the active region functional distribution map to obtain a static region functional association map; An analysis unit, configured to analyze the interaction relationship between brain nerve activity areas based on the active area function distribution map and the static area function association map, and obtain a clustering result of nerve function areas; A modeling unit, used to model a brain neural network connection mode according to the clustering result of the neural function area, obtain a functional connection map of the brain neural network, input the functional connection map of the brain neural network into a preset classification analysis model, evaluate the functional state of the brain neural network based on the classification analysis model, and obtain an evaluation result of the brain functional state; A generating unit is used to generate a neural regulation strategy for a target brain region based on the evaluation result of the brain functional state and the functional connection diagram of the brain neural network.

Images