Target crowd identification method and device based on multilevel fusion of nerve blood flow features

Abstract

The invention relates to the technical field of brain function imaging, in particular to a target crowd recognition method and device based on neural blood flow feature multi-level fusion, and the method comprises the steps: collecting brain nerve blood flow signals of normal and target crowds by using synchronous EEG-fNIRS equipment, preprocessing and extracting single-mode features, and obtaining a single-mode feature set; a multi-frequency local nerve and blood vessel coupling analysis method is used for calculating nerve and blood vessel coupling strength characteristics, a space co-positioning method is used for unifying nerve and blood flow modal space information, the position precision is improved, and data layer fusion is achieved; and integrating and analyzing different modal feature space relationships by using a Yeo7 brain template to realize feature layer fusion, and finally training various machine learning classifiers by using significant difference features to realize identification of a target population. Therefore, the problems that in the prior art, macroscopic signal correlation is paid more attention to, so that analysis of the synergistic effect of cross-modal space information in a specific brain region is deficient, and deep understanding of a neurovascular function mechanism and brain region role revelation are restricted are solved.

Description

Technical Field

[0001] The present application relates to the technical field of brain function imaging, and in particular to a method and device for identifying a target population based on multi-level fusion of neural blood flow characteristics. Background Art

[0002] Among the related technologies, EEG (Electroencephalogram) and fNIRS (Functional Near-Infrared Spectroscopy) are often used in combination to study the brain function of people with substance use disorders and observe the neural blood flow information of the subjects.

[0003] However, in the related technologies, the synchronous EEG-fNIRS signal acquisition device has problems such as large and complex structure, mutual interference between electrodes and optical fiber probes that increases the difficulty of acquisition, and difficulty in signal integration. Moreover, when using the signal for research, although the two technologies can provide high-resolution brain activity information, there is a lack of analysis of the relationship between multimodal features and the neurovascular coupling conditions at specific locations, and only single-modal features are analyzed separately without discussion of the interaction mechanism, resulting in a lack of in-depth understanding of brain function and insufficient exploration of the potential for information fusion.

[0004] In summary, since relevant technologies focus more on signal correlation at the macro level, analysis of how spatial information across different modalities works synergistically in specific brain regions is still relatively scarce, which limits the understanding of higher-level mechanisms of neurovascular function and fails to fully reveal the unique roles of different brain regions in information processing and functional regulation, which urgently needs to be improved. Summary of the invention

[0005] The present application provides a method and device for identifying a target population based on multi-level fusion of neural blood flow characteristics, in order to solve the problem that related technologies focus more on macro signal correlation, resulting in a lack of analysis of the synergistic effect of cross-modal spatial information in specific brain regions, which restricts the in-depth understanding of the neurovascular function mechanism and the revelation of the role of brain regions.

[0006] The first aspect of the present application provides a method for identifying a target population based on multi-level fusion of neural blood flow features, comprising the following steps: based on a synchronous EEG-fNIRS acquisition device, collecting brain nerve blood flow signal data containing EEG signals and near-infrared signals of a normal population and a target population that meet preset conditions; aligning and preprocessing the brain nerve blood flow signal data, and dividing the EEG signal into multiple EEG components; based on the preprocessed brain nerve blood flow signal data, analyzing and extracting at least one single-modal similar feature parameter, and mapping the single-modal similar feature parameter to a specific channel position to obtain a specific channel feature; based on the preprocessed brain nerve blood flow signal data, analyzing and extracting at least one mixed-modal feature parameter, mapping the mixed-modal feature parameter to a specific channel position to obtain a specific channel feature, and based on the at least one mixed-modal feature parameter, realizing feature fusion of the data layer; using Yeo 7 The network integrates the single modal features and the mixed modal features in terms of position to achieve fusion at the feature level, and then analyzes the differences and connections between the single modal features and the mixed modal features on the same network, takes the similar feature parameters of the target population and the normal population as independent samples, and uses the mean hypothesis test to discover the sensitive brain features and brain networks of the target population; uses the sensitive brain features of the target population to train a variety of machine learning classifiers to achieve identification of the target population.

[0007] Through the above technical solutions, the embodiment of the present application can obtain brain nerve blood flow data by synchronously collecting EEG and fNIRS signals, providing rich information for comprehensive analysis of brain function. Data alignment and preprocessing as well as EEG component division ensure the data quality and effectiveness of analysis. The extraction and corresponding fusion of single-modality and mixed-modality feature parameters not only deeply explore the information of each modality itself, but also realize cross-modal information integration, which helps to more accurately characterize brain activity. Using the Yeo 7 network integration feature, feature differences and connections can be analyzed from the network level to discover sensitive features and networks of the target population. Finally, with the help of machine learning classifiers, target population identification is carried out, which improves the accuracy and efficiency of identification and provides strong support for neuroscience research, disease diagnosis, etc.

[0008] Optionally, in one embodiment of the present application, the analysis and extraction of at least one single-modality similar characteristic parameter based on the preprocessed brain nerve blood flow signal data includes: segmenting the EGG time domain data according to a preset time, calculating the correlation between each EGG channel signal on each segment by linear time-delay coherence, obtaining a functional connection matrix of the neural modality, averaging the functional connection matrix of the neural modality according to the EGG channel, and obtaining the functional connection strength on the specific EGG channel; calculating the correlation matrix between the time domain data of each fNIRS channel by Pearson correlation, and then performing nonlinear correction on the correlation matrix to obtain a functional connection matrix of the blood flow modality, averaging the functional connection matrix of the blood flow modality according to the fNIRS channel, and obtaining the functional connection strength on the specific fNIRS channel; based on the functional connection matrix of the neural modality and the functional connection matrix of the blood flow modality, using a preset wide threshold to set the sparsity, constructing a binary network according to the sparsity, calculating a variety of graph theory parameters by a graph theory method, and obtaining AUC values ​​for the various graph theory parameters to obtain a stable feature estimation value;

[0009] Through the above technical scheme, the embodiment of the present application can obtain the neural modal functional connection matrix by dividing the EGG time domain data into preset time segments and using linear time-delay coherence to calculate the correlation, and then average the specific channel functional connection strength by channel, which can accurately reflect the neural activity at the specific location; use Pearson correlation and nonlinear correction to process fNIRS channel data to obtain the blood flow modal functional connection matrix and average it, which helps to accurately grasp the blood flow activity at the specific location; construct a binary network based on the two modal functional connection matrices, and use certain thresholds and graph theory methods to calculate a variety of graph theory parameters, which can comprehensively and deeply analyze brain function characteristics and obtain stable feature estimates.

[0010] Optionally, in one embodiment of the present application, the analysis and extraction of at least one mixed modal characteristic parameter based on the preprocessed brain nerve blood flow signal data includes: performing multimodal EEG-fNIRS source estimation based on spatial positioning on the multiple EEG components to obtain cortical current source density; reconstructing the cortical current source density at the near-infrared channel position to convolve the current source density at the near-infrared channel position with a preset standard hemodynamic function to obtain expected blood flow changes; obtaining a real blood flow response based on the expected blood flow changes and the near-infrared signal, and using a general linear model to calculate a standardized slope to determine the multi-band local neurovascular coupling strength.

[0011] Through the above technical solution, the embodiment of the present application can perform multimodal source estimation based on spatial positioning for multiple EEG components, and can accurately obtain the cortical current source density, providing an accurate basis for subsequent analysis. The current source density is reconstructed at the near-infrared channel position and convolved with the standard hemodynamic function to obtain the expected blood flow changes, and the real blood flow response is obtained by comparing with the near-infrared signal and the normalized slope is calculated using a general linear model to determine the neurovascular coupling strength, which realizes the effective correlation and analysis of EEG signals and blood flow signals at local positions, which helps to deeply understand the relationship between brain neural activity and blood flow changes.

[0012] Optionally, in one embodiment of the present application, the calculation formula of the multi-band local neurovascular coupling strength is:

[0013]

[0014] Among them, S i(i＝1,2,3,4,5) represents the source activity of multiple (five) brain waves, V EEG is the potential recorded by the EEG electrode, W is the weight matrix constructed according to the fNIRS activation area, L is the EEG lead matrix (representing the relationship between the electrode and the current source), H is the current source density selected at the near-infrared channel position, is the convolution of the current source density and the standard hemodynamic response function, Y is the actual hemodynamic change, β0 is the intercept of the regression model, and β1 is the regression coefficient (indicating the multi-band local neurovascular coupling strength).

[0015] Through the above technical solution, the embodiment of the present application can comprehensively and systematically construct a quantitative relationship model between neural activity and hemodynamic changes by clarifying multiple key variables. This precise mathematical expression can accurately calculate the local neurovascular coupling strength in multiple frequency bands, providing a solid theoretical basis and effective calculation tools for in-depth research on the brain's neurovascular coupling mechanism, and helping to more carefully understand the impact of neural activity on blood flow changes in the brain under different EEG states.

[0016] Optionally, in one embodiment of the present application, the positional integration of different modal features using the Yeo 7 network includes: using a preset standard system to determine the specific spatial positions of the neural modality and the blood flow modality channel, and mapping the specific spatial positions to a unified MNI standard space using a spatial registration method to obtain the MNI coordinates of each channel of each modality; using the Yeo 7 network to divide the MNI standard space into 7 brain networks, and determining the brain network to which each EEG electrode and the near-infrared channel belong according to their MNI coordinates, thereby achieving spatial positional integration of the single modality feature and the mixed modality feature; using the mean hypothesis to test the normal population and the target population, obtaining brain features with significant differences as brain features that are sensitive to the target population, and using the brain network where differences in multiple modal brain features of the neural modality, the blood flow modality and the mixed modality appear as the sensitive brain network of the target population.

[0017] Through the above technical solutions, the embodiment of the present application can determine and unify the spatial position of the modal channel through the standard system and spatial registration method, obtain MNI coordinates, and lay the foundation for accurate analysis. The Yeo 7 network is used to divide the brain network and determine the network to which the channel belongs, so as to realize the spatial integration of multimodal features, and facilitate the comprehensive study of the relationship and synergy between different modalities. The mean hypothesis test is used to obtain the sensitive brain features and brain networks of the target population, which can effectively distinguish between normal and target populations, and help to deeply understand the changes in brain function of the target population.

[0018] Optionally, in one embodiment of the present application, the use of sensitive brain features of the target population to train multiple machine learning classifiers to achieve identification of the target population includes: adopting multiple machine learning classification models and k-fold cross-validation methods, using the sensitive brain features of the target population to train multiple stable binary classifiers for distinguishing between the normal population and the target population, combining the multiple stable binary classifiers to classify and identify the target population to obtain multiple identification results, and combining the multiple identification results to achieve identification of the target population.

[0019] Through the above technical scheme, the embodiment of the present application can use a variety of machine learning classification models to give full play to the advantages of different models and improve the accuracy and generalization ability of classification. The k-fold cross-validation method effectively utilizes limited data, divides the training set and the validation set multiple times, verifies the parameter effect, ensures that the trained model is stable and reliable, and avoids overfitting or underfitting. By training multiple stable binary classifiers using the sensitive brain features of the target population, the normal population and the target population can be distinguished from multiple angles, thereby enhancing the accuracy of recognition. Finally, the recognition results of multiple classifiers are combined and comprehensively judged to further improve the accuracy and credibility of the target population identification, providing strong support for the accurate identification of the target population.

[0020] The second aspect of the present application provides a target population identification device based on multi-level fusion of neural blood flow features, including: an acquisition module, which is used to acquire brain nerve blood flow signal data containing EEG signals and near-infrared signals of normal people and target people that meet preset conditions based on a synchronous EEG-fNIRS acquisition device; a preprocessing module, which is used to align and preprocess the brain nerve blood flow signal data, and divide the EEG signal into multiple EEG components; a single-modal feature extraction module, which is used to analyze and extract at least one single-modal similar feature parameter based on the preprocessed brain nerve blood flow signal data, and map the single-modal similar feature parameter to a specific channel position to obtain specific channel features; a mixed-modal feature extraction module, which is used to analyze and extract at least one mixed-modal feature parameter based on the preprocessed brain nerve blood flow signal data, map the mixed-modal feature parameter to a specific channel position to obtain specific channel features, and realize feature fusion of the data layer based on the at least one mixed-modal feature parameter; a multi-modal feature integration module, which is used to use Yeo based on the at least one single-modal feature parameter and the at least one mixed-modal feature parameter. 7. The network integrates the single modal features and the mixed modal features in position to achieve feature layer fusion, and then analyzes the differences and connections between the single modal features and the mixed modal features on the same network, takes the similar feature parameters of the target population and the normal population as independent samples, and uses the mean hypothesis test to find the sensitive brain features and brain networks of the target population; an identification module is used to train multiple machine learning classifiers using the sensitive brain features of the target population to achieve identification of the target population.

[0021] Through the above technical solutions, the embodiment of the present application can obtain brain nerve blood flow data by synchronously collecting EEG and fNIRS signals, providing rich information for comprehensive analysis of brain function. Data alignment and preprocessing as well as EEG component division ensure the data quality and effectiveness of analysis. The extraction and corresponding fusion of single-modality and mixed-modality feature parameters not only deeply explore the information of each modality itself, but also realize cross-modal information integration, which helps to more accurately characterize brain activity. Using the Yeo 7 network integration feature, feature differences and connections can be analyzed from the network level to discover sensitive features and networks of the target population. Finally, with the help of machine learning classifiers, target population identification is carried out, which improves the accuracy and efficiency of identification and provides strong support for neuroscience research, disease diagnosis, etc.

[0022] Optionally, in one embodiment of the present application, the single-modal feature extraction module includes: a first calculation unit, which is used to divide the EGG time domain data into preset time segments, calculate the correlation between each EGG channel signal on each segment through linear time-delay coherence, obtain the functional connection matrix of the neural modality, and average the functional connection matrix of the neural modality according to the EGG channel to obtain the functional connection strength on the specific EGG channel; a second calculation unit, which is used to calculate the correlation matrix between the time domain data of each fNIRS channel through Pearson correlation, and then perform nonlinear correction on the correlation matrix to obtain the functional connection matrix of the blood flow modality, average the functional connection matrix of the blood flow modality according to the fNIRS channel to obtain the functional connection strength on the specific fNIRS channel; a third calculation unit, which is used to set the sparsity based on the functional connection matrix of the neural modality and the functional connection matrix of the blood flow modality using a preset wide threshold, construct a binary network according to the sparsity, calculate a variety of graph theory parameters through a graph theory method, and calculate the AUC value of the multiple graph theory parameters to obtain a stable feature estimation value.

[0023] Through the above technical scheme, the embodiment of the present application can obtain the neural modal functional connection matrix by dividing the EGG time domain data into preset time segments and using linear time-delay coherence to calculate the correlation, and then average the specific channel functional connection strength by channel, which can accurately reflect the neural activity at the specific location; use Pearson correlation and nonlinear correction to process fNIRS channel data to obtain the blood flow modal functional connection matrix and average it, which helps to accurately grasp the blood flow activity at the specific location; construct a binary network based on the two modal functional connection matrices, and use certain thresholds and graph theory methods to calculate a variety of graph theory parameters, which can comprehensively and deeply analyze brain function characteristics and obtain stable feature estimates.

[0024] Optionally, in one embodiment of the present application, the hybrid modal feature extraction module includes: a source estimation unit, used to perform multimodal EEG-fNIRS source estimation based on spatial positioning on the multiple EEG components to obtain cortical current source density; a reconstruction unit, used to reconstruct the cortical current source density at the near-infrared channel position to convolve the current source density at the near-infrared channel position with a preset standard hemodynamic function to obtain expected blood flow changes; a fourth calculation unit, used to obtain a real blood flow response based on the expected blood flow changes and the near-infrared signal, and use a general linear model to calculate a standardized slope to determine the multi-band local neurovascular coupling strength.

[0025] Through the above technical solution, the embodiment of the present application can perform multimodal source estimation based on spatial positioning for multiple EEG components, and can accurately obtain the cortical current source density, providing an accurate basis for subsequent analysis. The current source density is reconstructed at the near-infrared channel position and convolved with the standard hemodynamic function to obtain the expected blood flow changes, and the real blood flow response is obtained by comparing with the near-infrared signal and the normalized slope is calculated using a general linear model to determine the neurovascular coupling strength, which realizes the effective correlation and analysis of EEG signals and blood flow signals at local positions, which helps to deeply understand the relationship between brain neural activity and blood flow changes.

[0026] Optionally, in one embodiment of the present application, the calculation formula of the multi-band local neurovascular coupling strength is:

[0027]

[0028] Among them, S i(i＝1,2,3,4,5) represents the source activity of multiple (five) brain waves, V EEG is the potential recorded by the EEG electrode, W is the weight matrix constructed according to the fNIRS activation area, L is the EEG lead matrix (representing the relationship between the electrode and the current source), H is the current source density selected at the near-infrared channel position, is the convolution of the current source density and the standard hemodynamic response function, Y is the actual hemodynamic change, β0 is the intercept of the regression model, and β1 is the regression coefficient (indicating the multi-band local neurovascular coupling strength).

[0029] Through the above technical solution, the embodiment of the present application can comprehensively and systematically construct a quantitative relationship model between neural activity and hemodynamic changes by clarifying multiple key variables. This precise mathematical expression can accurately calculate the local neurovascular coupling strength in multiple frequency bands, providing a solid theoretical basis and effective calculation tools for in-depth research on the brain's neurovascular coupling mechanism, and helping to more carefully understand the impact of neural activity on blood flow changes in the brain under different EEG states.

[0030] Optionally, in one embodiment of the present application, the multimodal feature integration module includes: a mapping unit, used to determine the specific spatial positions of the neural modality and the blood flow modality channel using a preset standard system, and map the specific spatial positions to a unified MNI standard space using a spatial registration method to obtain the MNI coordinates of each channel of each modality; an integration unit, used to divide the MNI standard space into 7 brain networks using the Yeo 7 network, and determine the brain network to which each EEG electrode and the near-infrared channel belong according to their MNI coordinates, so as to achieve the integration of the single modality feature and the mixed modality feature in spatial position; a testing unit, used to test the normal population and the target population using the mean hypothesis, obtain brain features with significant differences as brain features sensitive to the target population, and use the brain network where differences in the multiple modal brain features of the neural modality, the blood flow modality and the mixed modality appear as the sensitive brain network of the target population.

[0031] Through the above technical solutions, the embodiment of the present application can determine and unify the spatial position of the modal channel through the standard system and spatial registration method, obtain MNI coordinates, and lay the foundation for accurate analysis. The Yeo 7 network is used to divide the brain network and determine the network to which the channel belongs, so as to realize the spatial integration of multimodal features, and facilitate the comprehensive study of the relationship and synergy between different modalities. The mean hypothesis test is used to obtain the sensitive brain features and brain networks of the target population, which can effectively distinguish between normal and target populations, and help to deeply understand the changes in brain function of the target population.

[0032] Optionally, in one embodiment of the present application, the identification module includes: using multiple machine learning classification models and k-fold cross-validation methods, using the sensitive brain features of the target population to train multiple stable binary classifiers for distinguishing between the normal population and the target population, combining the multiple stable binary classifiers to classify and identify the target population to obtain multiple identification results, and combining the multiple identification results to achieve identification of the target population.

[0033] Through the above technical scheme, the embodiment of the present application can use a variety of machine learning classification models to give full play to the advantages of different models and improve the accuracy and generalization ability of classification. The k-fold cross-validation method effectively utilizes limited data, divides the training set and the validation set multiple times, verifies the parameter effect, ensures that the trained model is stable and reliable, and avoids overfitting or underfitting. By training multiple stable binary classifiers using the sensitive brain features of the target population, the normal population and the target population can be distinguished from multiple angles, thereby enhancing the accuracy of recognition. Finally, the recognition results of multiple classifiers are combined and comprehensively judged to further improve the accuracy and credibility of the target population identification, providing strong support for the accurate identification of the target population.

[0034] The third aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the target population identification method based on multi-level fusion of neural blood flow features as described in the above embodiment.

[0035] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above-mentioned target population identification method based on multi-level fusion of neural blood flow features.

[0036] The fifth aspect of the present application provides a computer program product, including a computer program, which, when executed, is used to implement the above-mentioned target population identification method based on multi-level fusion of neural blood flow features.

[0037] The embodiments of the present application can obtain rich brain nerve blood flow data by synchronously collecting EEG and fNIRS signals, and ensure data quality and analysis effectiveness through data alignment, preprocessing and EEG component division; use linear time-delay coherence, Pearson correlation and nonlinear correction to process two modal data respectively, obtain functional connection matrix and calculate related parameters, construct binary network to calculate graph theory parameters to obtain AUC value, and comprehensively and deeply analyze brain function characteristics; perform multimodal source estimation based on spatial positioning on EEG components, reconstruct current source density and convolve with standard function to obtain expected blood flow changes, compare the actual blood flow response to calculate neurovascular coupling strength, and deeply understand the relationship between nerves and blood flow; accurately calculate coupling strength by constructing quantitative relationship models through clear variables; adopt multiple machine learning classification models combined with k-fold cross validation, use sensitive brain features to train multiple stable binary classifiers, and combine their results for comprehensive judgment, thereby improving the accuracy and credibility of target population identification and providing strong support for neuroscience research, disease diagnosis, etc.

[0038] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0040] Figure 1 A flowchart of a target population identification method based on multi-level fusion of neural blood flow features provided according to an embodiment of the present application;

[0041] Figure 2 is a schematic diagram of a multi-band local neurovascular coupling analysis method according to an embodiment of the present application;

[0042] Figure 3 A schematic diagram of a method for classifying target population and normal persons using machine learning according to an embodiment of the present application;

[0043] Figure 4 It is a structural schematic diagram of a target population identification device based on multi-level fusion of neural blood flow features provided according to an embodiment of the present application;

[0044] Figure 5 The figure is a structural example diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0046] The following describes the target population identification method and device based on multi-level fusion of neural blood flow features of the embodiment of the present application with reference to the accompanying drawings. In view of the fact that the related technologies mentioned in the above background technology focus more on the correlation of macroscopic signals, resulting in a lack of analysis of the synergistic effect of cross-modal spatial information in specific brain regions, which restricts the in-depth understanding of the neurovascular functional mechanism and the problem of revealing the role of brain regions, the present application provides a target population identification method based on multi-level fusion of neural blood flow features, in which brain neural blood flow data can be obtained by synchronously collecting EEG and fNIRS signals, providing rich information for comprehensive analysis of brain function. Data alignment and preprocessing as well as EEG component division ensure the data quality and effectiveness of analysis. The extraction and corresponding fusion of single-modal and mixed-modal feature parameters not only deeply explores the information of each modality itself, but also realizes cross-modal information integration, which helps to more accurately portray brain activity. Using the Yeo 7 network integration feature, the feature differences and connections can be analyzed from the network level to find the sensitive features and networks of the target population. Finally, the target population is identified with the help of a machine learning classifier, which improves the accuracy and efficiency of identification and provides strong support for neuroscience research, disease diagnosis, etc. This solves the problem that related technologies focus more on the correlation of macroscopic signals, resulting in a lack of analysis of the synergistic effects of cross-modal spatial information in specific brain regions, which restricts the in-depth understanding of the neurovascular function mechanism and the revelation of the role of brain regions.

[0047] Specifically, Figure 1 A flowchart of a target population identification method based on multi-level fusion of neural blood flow features provided in an embodiment of the present application.

[0048] like Figure 1As shown, the target population identification method based on multi-level fusion of neural blood flow features includes the following steps:

[0049] In step S101, based on the synchronous EEG-fNIRS acquisition device, brain nerve blood flow signal data including EEG signals and near infrared signals of normal people and target people who meet preset conditions are collected.

[0050] It can be understood that EEG and fNIRS are non-invasive, flexible, compact, user-friendly, and easy to access. EEG can provide information on neural electrical activity with high temporal resolution, which is particularly suitable for studying instantaneous brain responses and neural dynamic processes; while fNIRS can provide hemodynamic information with high spatial resolution, and compared with methods such as functional magnetic resonance imaging, it has greater operational flexibility and lower equipment cost.

[0051] A synchronous EEG-fNIRS signal acquisition platform was built by assembling EEG electrodes, fNIRS light sources and receivers in hardware and designing multimodal synchronous signal acquisition software. The acquisition software supports synchronous marking of two signals and controls the dual-modal signal acquisition error within 100ms. Based on the synchronous EEG-fNIRS signal acquisition platform, resting whole-brain neural blood flow signal data are collected. The brain neural blood flow signal data includes EEG signals and near-infrared signals.

[0052] The embodiment of the present application can not only realize the synchronous acquisition of two signals through carefully designed hardware assembly and multi-modal synchronous signal acquisition software, but also control the acquisition error within a very small range, effectively ensuring the accuracy and synchronization of the data, and laying a solid and reliable foundation for subsequent analysis and research based on resting whole-brain neural blood flow signal data.

[0053] In step S102, the brain nerve blood flow signal data is aligned and preprocessed, and the EEG signal is divided into multiple EEG components.

[0054] Specifically, the brain nerve blood flow signal data collected in step S101 is aligned and preprocessed, and the signal is intercepted according to the synchronization tag obtained by the synchronization marking of step S101, and the time alignment is performed. Further, the EEG signal is preprocessed, and the preprocessing operations include but are not limited to: downsampling, re-reference, bandpass filtering and notch filtering, ICA (Independent Component Analysis), and the EEG signal is divided into frequency bands for analysis, including the characteristics of five commonly used EEG signal components: δ (1-3Hz), θ (3-8Hz), α (8-13Hz), β (13-30Hz), and γ (30-40Hz); the near-infrared signal is preprocessed, and the preprocessing operations include: filtering, improved Beer-Lambert change, TDDR (emporal Derivative Distribution Repair) head movement correction, and the oxygenated hemoglobin HbO with relatively obvious changes in the near-infrared signal is used as the basis for the analysis of hemodynamic changes.

[0055] The embodiment of the present application can align and preprocess the brain nerve blood flow signal data, and realize time alignment based on the synchronization tag interception signal, which can ensure the accuracy and consistency of the data in the time dimension and facilitate subsequent analysis. A series of preprocessing operations are carried out on EEG signals, such as downsampling, re-referencing, etc., and the EEG signals are divided into five components according to frequency to optimize the signal quality, which helps to more carefully mine the information contained in the EEG signals. The preprocessing methods such as filtering, head movement correction and the use of oxygenated hemoglobin HbO as the basis for analysis of near-infrared signals can effectively improve the signal quality and accurately capture the changes in hemodynamics. On the whole, it provides a high-quality and reliable data foundation for the subsequent in-depth study of brain function and analysis of brain nerve activity.

[0056] In step S103, based on the preprocessed brain nerve blood flow signal data, at least one single-modality similar feature parameter is analyzed and extracted, and the single-modality similar feature parameter is mapped to a specific channel position to obtain a specific channel feature.

[0057] It is understandable that single-modality homogeneous characteristic parameters include but are not limited to analyzing and extracting functional connections of nerve and blood flow signals, and graph theory parameters.

[0058] Optionally, in one embodiment of the present application, based on the preprocessed brain nerve blood flow signal data, at least one single modality similar feature parameter is analyzed and extracted, including: dividing the EGG time domain data into preset time segments, calculating the correlation between each EGG channel signal in each segment through linear time-delay coherence, obtaining the functional connection matrix of the neural modality, averaging the functional connection matrix of the neural modality according to the EGG channel, and obtaining the functional connection strength on the specific EGG channel; calculating the correlation matrix between the time domain data of each fNIRS channel through Pearson correlation, and then performing nonlinear correction on the correlation matrix to obtain the functional connection matrix of the blood flow modality, averaging the functional connection matrix of the blood flow modality according to the fNIRS channel, and obtaining the functional connection strength on the specific fNIRS channel; based on the functional connection matrix of the neural modality and the functional connection matrix of the blood flow modality, using a preset wide threshold to set the sparsity, constructing a binary network according to the sparsity, calculating multiple graph theory parameters through graph theory methods, and obtaining AUC values ​​for multiple graph theory parameters to obtain stable feature estimation values.

[0059] Specifically, the EEG time domain data was segmented into 10-second segments, and the correlation between each EEG channel signal in each segment was calculated by linear time-delay coherence to reduce the impact of EEG signal volume conduction; the Pearson correlation between the time domain data of each fNIRS channel was calculated, and then nonlinear correction was performed through Fisher's Z change to obtain the functional connection matrix of the two modal data. Then, the calculated functional connection matrix of the neural blood flow signal was averaged according to the channel to obtain the functional connection strength on the specific channel, so as to achieve the purpose of analyzing each specific brain area.

[0060] Furthermore, the functional connectivity matrices calculated using the time domain information of EEG and fNIRS were used to construct binary networks. The nodes of the networks were each EEG or near-infrared channel. Since there was no consensus on the threshold for the edges of the networks, a wide threshold (0.05≤T≤0.5, with an interval of 0.05) was used. The sparsity was set according to the threshold, and 11 binary networks were obtained for each functional connectivity matrix. Various graph theory parameters were calculated using graph theory methods, including but not limited to small-world properties, local efficiency, global efficiency, node efficiency, etc., and the AUC (Area Under the Curve) value was calculated for each graph theory feature of each matrix to obtain a stable feature estimate.

[0061] The embodiment of the present application can extract single-modal similar feature parameters based on the pre-processed brain nerve blood flow signal data, covering the functional connection and graph theory parameters of the analysis of nerve and blood flow signals, and the content is comprehensive and in-depth. Specifically, the EEG time domain data is reasonably segmented and the correlation is calculated using linear time-delay coherence to reduce the volume conduction effect of the EEG signal; the functional connection matrix is ​​obtained by Pearson correlation and nonlinear correction of the fNIRS channel data, and the operation is scientific and rigorous. The specific channel strength is obtained according to the channel average functional connection matrix, which is conducive to the analysis of each brain area. Construct a binary network and use a wide range of thresholds to set the sparsity, combine graph theory methods to calculate a variety of graph theory parameters, and then calculate the AUC value to obtain a stable feature estimate. It can accurately portray brain function characteristics from multiple dimensions, and provide reliable and valuable data support for subsequent in-depth research on brain activity, disease diagnosis, etc.

[0062] In step S104, based on the preprocessed brain nerve blood flow signal data, at least one mixed modal feature parameter is analyzed and extracted, the mixed modal feature parameter is mapped to a specific channel position to obtain a specific channel feature, and based on at least one mixed modal feature parameter, feature fusion of the data layer is achieved.

[0063] Optionally, in one embodiment of the present application, based on the preprocessed cranial nerve blood flow signal data, at least one mixed modal characteristic parameter is analyzed and extracted, including: performing multimodal EEG-fNIRS source estimation based on spatial positioning on multiple EEG components to obtain cortical current source density; reconstructing the cortical current source density at the near-infrared channel position to convolve the current source density at the near-infrared channel position with a preset standard hemodynamic function to obtain expected blood flow changes; obtaining the real blood flow response based on the expected blood flow changes and the near-infrared signal, and using a general linear model to calculate the standardized slope to determine the multi-band local neurovascular coupling strength.

[0064] It should be noted that, using the preprocessed brain nerve blood flow signal data, the time series needs to be further processed to calculate the multi-frequency local neurovascular coupling degree characteristics, such as Figure 2 shown.

[0065] In the actual implementation process, since it is difficult for the equipment to collect neural blood flow signals at the same position at the same time, the current source density information on the cortex is obtained by performing a multimodal EEG-fNIRS source estimation method based on spatial co-localization on multiple brain waves. Specifically, the activation area provided by the fNIRS signal is used as the prior spatial information H of the EEG to reduce the solution space of the EEG source estimation. Then, the EEG electrical signal source positioning is combined with the fNIRS blood flow signal, and the two modal information is jointly estimated using a Bayesian model to enhance the spatial accuracy of the source estimation. Among them, the spatial distribution of the fNIRS signal can be expressed as:

[0066] H=H1,H2,...,H n (1)

[0067] Among them, H i represents the hemodynamic response corresponding to the ith channel, and n is the number of fNIRS channels.

[0068] The EEG source estimation problem is usually modeled as a linear inverse problem as follows:

[0069] V EEG =LS (2)

[0070] Among them, V EEG are five types of brain waves (observation data), L is the lead matrix, and S is the source activity.

[0071] Define a weighted source estimation constraint matrix W, where active regions identified by fNIRS will be given higher weights, while the weights of other regions are reduced:

[0072] W=diag(w1,w2,...,w p ) (3)

[0073] Among them, if H i is greater than the threshold, w1=1; if H i Less than the threshold, w1=0.

[0074] In order to combine the fNIRS blood flow signal and the EEG electrical signal, Bayesian inference is used for multimodal fusion. Under the Bayesian framework, EEG source estimation is jointly estimated by prior information (fNIRS signal) and observed data (EEG signal). The posterior probability of source activity S can be expressed as:

[0075] P(S|V EEG ,H)=P(V EEG |S)P(S|H)(4)

[0076] Among them, P(V EEG |S) is the likelihood function of the EEG observation data, assuming a Gaussian distribution:

[0077]

[0078] in, is the reconstruction error of the source estimation, which represents the difference between the EEG observation data and the source estimation model. P(S|H) is the prior distribution provided by fNIRS, which represents the constraint of the fNIRS activated area on the EEG source activity. It can be modeled as:

[0079]

[0080] in, is the constraint provided by fNIRS, limiting EEG source localization to the active area determined by fNIRS. Z1 and Z2 are normalization constants, σ 2 and τ 2 are the EEG observation error and the fNIRS prior uncertainty, respectively. By maximizing the posterior probability P(S|V EEG ,H), and obtain the optimal estimate S of the source activity * , expressed as:

[0081]

[0082] Furthermore, the EEG component signals of the scalp are mapped to multiple voxels of the cerebral cortex to obtain the current source density results on each voxel. Then, the current source density of a single voxel closest to the near-infrared channel is selected as the reconstructed current source density to minimize the impact of volume conduction of the EEG signal. Secondly, the expected blood flow change is obtained by convolution with the standard hemodynamic function, and then the expected blood flow change signal and the actual hemodynamic change obtained by the near-infrared signal analysis are aligned in length and amplitude through downsampling and z-score normalization operations.

[0083] After completing the above data processing, the blood flow signals predicted by the five EEG components of each subject and the real blood flow signals were put into the general linear model (Formula 1) as independent variables and dependent variables respectively. The standardized slope is the degree to which the change of the real blood flow of the dependent variable is affected by the neural activity of the independent variable, that is, the degree of multi-frequency local neurovascular coupling. The calculation formula for the strength of multi-frequency local neurovascular coupling is:

[0084]

[0085] Among them, S i(i＝1,2,3,4,5) represents the source activity of multiple (five) brain waves, V EEG is the potential recorded by the EEG electrode, W is the weight matrix constructed according to the fNIRS activation area, L is the EEG lead matrix (representing the relationship between the electrode and the current source), H is the current source density selected at the near-infrared channel position, is the convolution of the current source density and the standard hemodynamic response function, Y is the actual hemodynamic change, β0 is the intercept of the regression model, and β1 is the regression coefficient (indicating the multi-band local neurovascular coupling strength).

[0086] The embodiment of the present application can extract mixed modal feature parameters based on the pre-processed brain nerve blood flow signal data and realize data layer feature fusion. By performing a multimodal EEG-fNIRS source estimation method based on spatial co-localization on multiple brain waves, the fNIRS signal is cleverly used to provide prior spatial information for EEG, reducing the EEG source estimation solution space, and then the two modal information is jointly estimated with the help of the Bayesian model, thereby enhancing the spatial accuracy of source estimation. From calculating the cortical current source density, to reconstructing and convolving at the near-infrared channel position to obtain the expected blood flow changes, and then comparing with the actual blood flow response, the standardized slope is calculated using the general linear model to determine the multi-band local neurovascular coupling strength, and the coupling strength is further quantified, which can effectively deeply fuse the EEG and blood flow signals and accurately depict the relationship between the two.

[0087] In step S105, the Yeo 7 network is used to integrate the unimodal features and the mixed modal features in terms of position to achieve feature layer fusion, and then the differences and connections between the unimodal features and the mixed modal features on the same network are analyzed. The similar feature parameters of the target population and the normal population are taken as independent samples, and the mean hypothesis test is used to discover the sensitive brain features and brain networks of the target population.

[0088] After completing all the above steps, that is, extracting all the features, it is necessary to find the brain functions that have changed significantly in the target population. According to the MNI coordinates of each EEG and near-infrared channel, find the corresponding Yeo 7 network, use the Yeo 7 network to integrate the features of different modalities of the same network, and determine the brain features and brain networks that have changed significantly in the target population compared to normal people, that is, the sensitive brain features of the target population.

[0089] Optionally, in one embodiment of the present application, the Yeo 7 network is used to integrate the different modal features in position, including: using a preset standard system to determine the specific spatial positions of the neural modality and blood flow modality channels, and mapping the specific spatial positions to a unified MNI standard space using a spatial registration method to obtain the MNI coordinates of each channel of each modality; using the Yeo 7 network to divide the MNI standard space into 7 brain networks, and determining the brain network to which each EEG electrode and near-infrared channel belongs according to its MNI coordinates, thereby realizing the spatial integration of single-modality features and mixed-modality features; using the mean hypothesis to test the normal population and the target population, obtaining brain features with significant differences as the sensitive brain features of the target population, and using the brain network where differences in multiple modal brain features of the neural modality, blood flow modality and mixed modality appear as the sensitive brain network of the target population.

[0090] Specifically, the standard system set therein is the 10-20 international standard lead system.

[0091] The following describes step S105 in detail by taking etomidate use disorder patients as the identification target as an example.

[0092] Specifically, there are significant differences in age and education between people with etomidate use disorder and normal people. Linear regression was used to eliminate the differences in age and education. After ensuring that there is no influence from other factors other than etomidate use, an independent sample T test was performed on each feature of the two groups of people. For multiple similar features, multiple verification correction was used to avoid false positive results. After completing the hypothesis test, since the features of different modalities cannot correspond in position (EEG unimodal features: features at the position of EEG channels; near-infrared unimodal features, neurovascular coupling features: features at the position of near-infrared channels), the two device channels were mapped to the same group of networks according to their positions with the help of the Yeo 7 network template, and channels with significant functional changes and clustered in the network were found to find the Yeo 7 brain network with significant brain function changes. When a brain network where significant change features appear in a concentrated manner is found, the network can be considered to be a sensitive network for the target population. For example, in this embodiment, the features of significant changes in low-frequency EEG functional connectivity, graph theory, and neurovascular coupling are concentrated in the somatomotor network and dorsal attention network, that is, these two Yeo 7 networks are etomidate-sensitive networks, and etomidate has a concentrated impact on the functions of these two brain network areas.

[0093] The embodiment of the present application can integrate different modal features in position and fuse feature layers using the Yeo 7 network based on single-modal and mixed-modal feature parameters, which can effectively break the analysis limitations caused by the position correspondence problem of different modal features. By taking the same feature parameters of the target population and the normal population as independent samples, applying the mean hypothesis test, and combining rigorous methods such as linear regression to eliminate the influence of other factors, and multiple verification corrections to avoid false positive results, the sensitive brain features and brain networks of the target population can be accurately discovered.

[0094] In step S106, multiple machine learning classifiers are trained using the sensitive brain features of the target population to achieve identification of the target population.

[0095] It is understandable that after finding the brain features and brain networks that are sensitive to the target population, all significantly different features can be selected as machine learning parameters for training, or the difference features on the sensitive network can be selected as parameters for training. This can reduce the cost of recognition. For example, in the embodiment of step S105 where people with etomidate use disorder are used as the recognition target, only the significantly different features on the somatomotor network and the dorsal attention network can be used as parameters to train the machine learning model. That is to say, in the subsequent targeted signal acquisition process, the equipment can be simplified, and only the information of a few leads can be used to achieve the same recognition effect as analyzing all the lead information.

[0096] Optionally, in one embodiment of the present application, multiple machine learning classifiers are trained using sensitive brain features of the target population to achieve identification of the target population, including: using multiple machine learning classification models and k-fold cross-validation methods, training multiple stable binary classifiers using sensitive brain features of the target population to distinguish between a normal population and a target population, combining multiple stable binary classifiers to classify and identify the target population to obtain multiple identification results, and combining multiple identification results to achieve identification of the target population.

[0097] In the training process of the machine learning model, three machine learning classification models were used, including SVM (Support Vector Machine), RF (Bootstrap Aggregating, Random Forest), and XGBoost (Extreme Gradient Boosting). The sensitive brain features of the target population were used to adjust the parameters respectively, and three stable binary classifiers of normal people and the target population were trained to find the optimal parameters under the specified feature quantity to achieve the best classification effect.

[0098] Alternatively, if the data set is limited and there is less data for training and validation, the k-fold cross-validation method can be used to divide the training set of the data into k subsets, and repeat the training k times in the three classifiers respectively, each time using one of the k subsets as the validation set and the remaining k-1 subsets as the training set, and taking the average of the classification results of 10 trainings as the classification effect of the current classifier and parameters. By dividing the training set and validation set multiple times and testing the parameter effect, a stable model parameter is obtained.

[0099] After obtaining a stable classifier, the sample features to be identified are placed in three classifiers, and the classification results of multiple classifiers are combined to identify the category of the analyzed sample. Figure 3 shown.

[0100] The embodiments of the present application can use the sensitive brain features of the target population to train multiple machine learning classifiers to identify the target population. All significant difference features can be selected, and the difference features on the sensitive network can also be selected as parameters, which can effectively reduce the recognition cost. For example, in the relevant embodiments, the device collection information can be simplified. The use of multiple machine learning classification models such as SVM, RF, XGBoost, etc. combined with the k-fold cross-validation method can give full play to the advantages of different models. In the case of limited data, the parameter effect can be tested by dividing the training set and the validation set multiple times, and a stable binary classifier can be trained. The classification results of multiple classifiers are integrated for identification, which further improves the accuracy, stability and reliability of classification, and provides a strong guarantee for the accurate identification of the target population.

[0101] According to the target population identification method based on multi-level fusion of neural blood flow features proposed in the embodiment of the present application, brain neural blood flow data can be obtained by synchronously collecting EEG and fNIRS signals, providing rich information for comprehensive analysis of brain function. Data alignment and preprocessing as well as EEG component division ensure the data quality and effectiveness of analysis. The extraction and corresponding fusion of single-modal and mixed-modal feature parameters not only deeply explore the information of each modality itself, but also realize cross-modal information integration, which helps to more accurately characterize brain activity. Using the Yeo 7 network integration feature, feature differences and connections can be analyzed from the network level to discover sensitive features and networks of the target population. Finally, the target population is identified with the help of a machine learning classifier, which improves the accuracy and efficiency of identification and provides strong support for neuroscience research, disease diagnosis, etc.

[0102] Next, a target population identification device based on multi-level fusion of neural blood flow features proposed in an embodiment of the present application is described with reference to the accompanying drawings.

[0103] Figure 4 It is a block diagram of a target population identification device based on multi-level fusion of neural blood flow features according to an embodiment of the present application.

[0104] like Figure 4 As shown, the target population identification device 10 based on multi-level fusion of neural blood flow features includes: an acquisition module 100, a preprocessing module 200, a single-modal feature extraction module 300, a mixed-modal feature extraction module 400, a multi-modal feature integration module 500 and an identification module 600.

[0105] Specifically, the acquisition module 100 is used to acquire brain nerve blood flow signal data including EEG signals and near-infrared signals of normal people and target people who meet preset conditions based on a synchronous EEG-fNIRS acquisition device.

[0106] The preprocessing module 200 is used to align and preprocess the brain nerve blood flow signal data and divide the EEG signal into multiple EEG components.

[0107] The single-modal feature extraction module 300 is used to analyze and extract at least one single-modal similar feature parameter based on the preprocessed brain nerve blood flow signal data, and map the single-modal similar feature parameter to a specific channel position to obtain a specific channel feature.

[0108] The mixed modal feature extraction module 400 is used to analyze and extract at least one mixed modal feature parameter based on the preprocessed brain nerve blood flow signal data, map the mixed modal feature parameter to a specific channel position to obtain specific channel features, and realize feature fusion of the data layer based on at least one mixed modal feature parameter.

[0109] The multimodal feature integration module 500 uses the Yeo 7 network to positionally integrate the unimodal features and the mixed modal features to achieve feature layer fusion, and then analyzes the differences and connections between the unimodal features and the mixed modal features on the same network, takes the similar feature parameters of the target population and the normal population as independent samples, and uses the mean hypothesis test to discover the sensitive brain features and brain networks of the target population.

[0110] The identification module 600 is used to train a variety of machine learning classifiers using the sensitive brain features of the target population to achieve identification of the target population.

[0111] Optionally, in one embodiment of the present application, the unimodal feature extraction module 300 includes: a first computing unit, a second computing unit and a third computing unit.

[0112] Among them, the first calculation unit is used to divide the EGG time domain data into segments according to preset time segments, calculate the correlation between each EGG channel signal in each segment through linear time-delay coherence, obtain the functional connection matrix of the neural modality, and average the functional connection matrix of the neural modality according to the EGG channel to obtain the functional connection strength on the specific EGG channel.

[0113] The second calculation unit is used to calculate the correlation matrix between the time domain data of each fNIRS channel through Pearson correlation, and then perform nonlinear correction on the correlation matrix to obtain the functional connection matrix of the blood flow modality, and average the functional connection matrix of the blood flow modality according to the fNIRS channels to obtain the functional connection strength on the specific fNIRS channel.

[0114] The third computing unit is used to set the sparsity based on the functional connectivity matrix of the neural modality and the functional connectivity matrix of the blood flow modality by using a preset wide threshold, construct a binary network according to the sparsity, calculate a variety of graph theory parameters through graph theory methods, and obtain AUC values ​​for the various graph theory parameters to obtain stable feature estimation values.

[0115] Optionally, in one embodiment of the present application, the mixed modality feature extraction module 400 includes: a source estimation unit, a reconstruction unit and a fourth calculation unit.

[0116] The source estimation unit is used to perform multimodal EEG-fNIRS source estimation based on spatial positioning on multiple EEG components to obtain cortical current source density.

[0117] The reconstruction unit is used to reconstruct the cortical current source density at the near-infrared channel position, so as to convolve the current source density at the near-infrared channel position with a preset standard hemodynamic function to obtain the expected blood flow change.

[0118] The fourth calculation unit is used to obtain the real blood flow response according to the expected blood flow change and the near-infrared signal, and to calculate the standardized slope using the general linear model to determine the multi-band local neurovascular coupling strength.

[0119] Optionally, in one embodiment of the present application, the calculation formula of the multi-band local neurovascular coupling strength is:

[0120]

[0121] Among them, S i(i＝1,2,3,4,5) represents the source activity of multiple (five) brain waves, V EEG is the potential recorded by the EEG electrode, W is the weight matrix constructed according to the fNIRS activation area, L is the EEG lead matrix (representing the relationship between the electrode and the current source), H is the current source density selected at the near-infrared channel position, is the convolution of the current source density and the standard hemodynamic response function, Y is the actual hemodynamic change, β0 is the intercept of the regression model, and β1 is the regression coefficient (indicating the multi-band local neurovascular coupling strength).

[0122] Optionally, in one embodiment of the present application, the multimodal feature integration module 500 includes: a mapping unit, an integration unit and a verification unit.

[0123] Among them, the mapping unit is used to determine the specific spatial positions of the neural modality and blood flow modality channels using a preset standard system, and map the specific spatial positions to a unified MNI standard space using a spatial registration method to obtain the MNI coordinates of each channel of each modality.

[0124] The integration unit is used to divide the MNI standard space into 7 brain networks using the Yeo 7 network, determine the brain network to which each EEG electrode and near-infrared channel belongs according to its MNI coordinates, and realize the integration of single-modal features and mixed-modal features in spatial position.

[0125] The testing unit is used to test the normal population and the target population using the mean hypothesis, obtain the brain features with significant differences as the sensitive brain features of the target population, and take the brain network with differences in the multiple modal brain features of neural modalities, blood flow modalities and mixed modalities as the sensitive brain network of the target population.

[0126] Optionally, in one embodiment of the present application, the identification module 600 includes: using multiple machine learning classification models and k-fold cross-validation methods, using sensitive brain features of the target population to train multiple stable binary classifiers for distinguishing between a normal population and a target population, combining multiple stable binary classifiers to classify and identify the target population to obtain multiple identification results, and combining multiple identification results to achieve identification of the target population.

[0127] It should be noted that the aforementioned explanation of the embodiment of the target population identification method based on multi-level fusion of neural blood flow features is also applicable to the target population identification device based on multi-level fusion of neural blood flow features in this embodiment, and will not be repeated here.

[0128] According to the target population identification device based on multi-level fusion of neural blood flow features proposed in the embodiment of the present application, brain neural blood flow data can be obtained by synchronously collecting EEG and fNIRS signals, providing rich information for comprehensive analysis of brain function. Data alignment and preprocessing as well as EEG component division ensure the data quality and effectiveness of analysis. The extraction and corresponding fusion of single-modal and mixed-modal feature parameters not only deeply explore the information of each modality itself, but also realize cross-modal information integration, which helps to more accurately characterize brain activity. Using the Yeo 7 network integration feature, feature differences and connections can be analyzed from the network level to discover sensitive features and networks of the target population. Finally, with the help of machine learning classifiers, target population identification is carried out to improve the accuracy and efficiency of identification, providing strong support for neuroscience research, disease diagnosis, etc.

[0129] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0130] A memory 501 , a processor 502 , and a computer program stored in the memory 501 and executable on the processor 502 .

[0131] When the processor 502 executes the program, the target population identification method based on multi-level fusion of neural blood flow features provided in the above embodiment is implemented.

[0132] Furthermore, the electronic device further comprises:

[0133] The communication interface 503 is used for communication between the memory 501 and the processor 502 .

[0134] The memory 501 is used to store computer programs that can be executed on the processor 502 .

[0135] The memory 501 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0136] If the memory 501, the processor 502 and the communication interface 503 are implemented independently, the communication interface 503, the memory 501 and the processor 502 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0137] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can communicate with each other through an internal interface.

[0138] The processor 502 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0139] The embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the target population identification method based on multi-level fusion of neural blood flow features as described above.

[0140] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-mentioned target population identification method based on multi-level fusion of neural blood flow features.

[0141] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0142] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0143] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0144] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.

[0145] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one or a combination of multiple of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0146] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0147] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0148] The storage medium mentioned above may be a read-only memory, a disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A target population identification method based on multi-level fusion of neural blood flow features, characterized in that: The following steps are involved: Based on the synchronous EEG-fNIRS acquisition equipment, the brain nerve blood flow signal data including EEG signals and near-infrared signals of normal people and target people who meet the preset conditions are collected; Aligning and preprocessing the brain nerve blood flow signal data, and dividing the EEG signal into multiple EEG components; Based on the preprocessed brain nerve blood flow signal data, analyzing and extracting at least one single-modal similar feature parameter, and mapping the single-modal similar feature parameter to a specific channel position to obtain a specific channel feature; Based on the preprocessed brain nerve blood flow signal data, at least one mixed modal feature parameter is analyzed and extracted, the mixed modal feature parameter is mapped to a specific channel position to obtain a specific channel feature, and based on the at least one mixed modal feature parameter, feature fusion of the data layer is achieved; The single modal features and the mixed modal features are integrated in position using the Yeo7 network to achieve feature layer fusion, and then the differences and connections between the single modal features and the mixed modal features on the same network are analyzed, and the similar feature parameters of the target population and the normal population are used as independent samples, and the mean hypothesis test is used to find the sensitive brain features and brain networks of the target population; The sensitive brain features of the target population are used to train a variety of machine learning classifiers to achieve identification of the target population.

2. The method according to claim 1, characterized in that The method of analyzing and extracting at least one single-modal similar characteristic parameter based on the preprocessed cranial nerve blood flow signal data includes: The EGG time domain data is segmented according to a preset time segment, and the correlation between each EGG channel signal in each segment is calculated by linear time-delay coherence to obtain a functional connection matrix of the neural modality, and the functional connection matrix of the neural modality is averaged according to the EGG channel to obtain the functional connection strength on the specific EGG channel; Calculating the correlation matrix between the time domain data of each fNIRS channel by Pearson correlation, then performing nonlinear correction on the correlation matrix to obtain the functional connection matrix of the blood flow modality, averaging the functional connection matrix of the blood flow modality according to the fNIRS channels, and obtaining the functional connection strength on the specific fNIRS channel; Based on the functional connection matrix of the neural modality and the functional connection matrix of the blood flow modality, a preset wide threshold is used to set the sparsity, a binary network is constructed according to the sparsity, a variety of graph theory parameters are calculated by graph theory methods, and AUC values ​​are calculated for the various graph theory parameters to obtain stable feature estimation values.

3. The method according to claim 1, characterized in that The step of analyzing and extracting at least one mixed modal characteristic parameter based on the preprocessed cranial nerve blood flow signal data includes: performing multimodal EEG-fNIRS source estimation based on spatial co-localization on the multiple EEG components to obtain cortical current source density; reconstructing the cortical current source density at the near-infrared channel position, so as to convolve the current source density at the near-infrared channel position with a preset standard hemodynamic function to obtain an expected blood flow change; The actual blood flow response is obtained according to the expected blood flow change and the near-infrared signal, and the normalized slope is calculated using a general linear model to determine the multi-band local neurovascular coupling strength.

4. The method according to claim 3, characterized in that The calculation formula of the multi-band local neurovascular coupling strength is: in, represents the source activity of multiple (five) brain waves, V EEG is the potential recorded by the EEG electrode, W is the weight matrix constructed according to the fNIRS activation area, L is the EEG lead matrix (representing the relationship between the electrode and the current source), H is the current source density selected at the near-infrared channel position, is the convolution of the current source density and the standard hemodynamic response function, Y is the actual hemodynamic change, β0 is the intercept of the regression model, and β1 is the regression coefficient (indicating the multi-band local neurovascular coupling strength).

5. The method according to claim 1, characterized in that: The Yeo 7 network is used to integrate different modal features in position, including: Determine the specific spatial positions of the neural modality and the blood flow modality channels using a preset standard system, and map the specific spatial positions to a unified MNI standard space using a spatial registration method to obtain the MNI coordinates of each channel of each modality; The MNI standard space is divided into 7 brain networks using the Yeo 7 network, and the brain network to which each EEG electrode and the near-infrared channel belongs is determined according to the MNI coordinates, so as to achieve the integration of the single-modality feature and the mixed-modality feature in spatial position; The normal population and the target population are tested by adopting the mean hypothesis to obtain brain features with significant differences as the sensitive brain features of the target population, and the brain network in which differences appear in the multiple modal brain features of the neural modality, the blood flow modality and the mixed modality is taken as the sensitive brain network of the target population.

6. The method according to claim 1, characterized in that The method of using the sensitive brain features of the target population to train multiple machine learning classifiers to achieve identification of the target population includes: A variety of machine learning classification models and k-fold cross-validation methods are used to train multiple stable binary classifiers using brain features that are sensitive to the target population or features on the brain network that are sensitive to the target population to distinguish between the normal population and the target population. The target population is classified and identified by combining the multiple stable binary classifiers to obtain multiple identification results, and the multiple identification results are combined to achieve identification of the target population.

7. A target population identification device based on multi-level fusion of neural blood flow features, characterized in that: include: An acquisition module is used to acquire brain nerve blood flow signal data including EEG signals and near-infrared signals of normal people and target people who meet preset conditions based on a synchronous EEG-fNIRS acquisition device; A preprocessing module, used for aligning and preprocessing the brain nerve blood flow signal data, and dividing the EEG signal into multiple EEG components; A single-modal feature extraction module, used to analyze and extract at least one single-modal similar feature parameter based on the preprocessed brain nerve blood flow signal data, and map the single-modal similar feature parameter to a specific channel position to obtain a specific channel feature; A mixed modality feature extraction module, used for analyzing and extracting at least one mixed modality feature parameter based on the preprocessed brain nerve blood flow signal data, mapping the mixed modality feature parameter to a specific channel position to obtain a specific channel feature, and realizing feature fusion of the data layer based on the at least one mixed modality feature parameter; A multimodal feature integration module is used to integrate the single modal features and the mixed modal features in position using the Yeo 7 network to achieve feature layer fusion, and then analyze the differences and connections between the single modal features and the mixed modal features on the same network, take the similar feature parameters of the target population and the normal population as independent samples, and use the mean hypothesis test to discover the sensitive brain features and brain networks of the target population; The recognition module is used to train multiple machine learning classifiers using the sensitive brain features of the target population to achieve recognition of the target population.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the target population identification method based on multi-level fusion of neural blood flow features as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement a target population identification method based on multi-level fusion of neural blood flow features as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that The computer program is executed to implement the target population identification method based on multi-level fusion of neural blood flow features as described in any one of claims 1 to 6.

Images