Multivariate pattern analysis and generalized representation analysis method and system for brain signal

Abstract

The present invention relates to a multivariate pattern analysis and generalized representation analysis method for a brain signal, comprising the following steps: a) preprocessing brain representation data; b) performing multivariate pattern analysis on the preprocessed data; and c) performing generalized representation analysis on the data that has been subjected to multivariate pattern analysis. The present invention can more comprehensively mine information in a neural image and other forms of brain representation data, break through the limitation of single-modal analysis, and better perform data analysis.

Description

Methods and systems for multivariate pattern analysis and generalized representation analysis of brain signals Technical Field

[0001] This invention relates to a method and system for multivariate pattern analysis and generalized representation analysis of brain signals. Background Technology

[0002] Traditional brain signal data analysis mainly includes the following methods:

[0003] Univariate analysis methods compare mean activity levels under different conditions based on a single measurement (e.g., per pixel or per time point) for each data point. However, univariate analysis considers only the individual measurement of each data point, ignoring the relationships and combinations between different data points, leading to insufficient understanding of the overall patterns of neural activity. It also typically only detects differences in overall mean activity levels, lacking sensitivity to specific neural activity patterns or individual differences. Furthermore, it cannot handle high-dimensional data; as the dimensionality of the data increases, the efficiency and reliability of univariate analysis methods become limited.

[0004] Time series analysis methods primarily focus on patterns of temporal variation in data, such as using waveform analysis to study time series changes under different conditions. However, time series analysis methods typically only capture changes in the time dimension and cannot fully consider the spatial relationships between different data points.

[0005] Functional connectivity analysis (FHI) primarily reveals patterns of functional connectivity between different brain regions. It typically employs methods such as correlation analysis or graph theory to study inter-brain region connectivity. However, FHI focuses more on the connectivity between brain regions rather than patterns at individual data points.

[0006] With the rapid advancement of modern medical research tools, neuroscience techniques, including electroencephalography (EEG), magnetoencephalography (MEG), and functional near-infrared spectroscopy (fMRI), have been widely applied. These tools enable non-invasive observation and recording of brain activity, providing crucial means for studying brain function and structure. However, with the increasing volume and complexity of data, traditional analytical methods are no longer adequate for fully unlocking the potential of this data.

[0007] Multivariate Pattern Analysis (MVPA) addresses the limitations of univariate analysis by treating brain signal representation data as multidimensional patterns. Furthermore, MVPA considers both temporal and spatial conditions, and when combined with Generalizable Representations Analysis (GRA), it can more precisely identify brain regions or specific signals representing stimuli, thus comprehensively mining information from brain signal representation data. However, current research is primarily based on single-modal methods, while GRA can study the brain's response to the same type of multimodal stimuli, providing a deeper understanding of how the brain specifically represents such stimuli. Therefore, providing a method or system for multivariate pattern analysis and generalizable representation analysis of brain signals is essential.

[0008] Summary of the Invention

[0009] In view of this, it is necessary to provide a method and system for multivariate pattern analysis and generalized representation analysis of brain signals.

[0010] This invention provides a method for multivariate pattern analysis and generalized representation analysis of brain signals. The method includes the following steps: a. preprocessing brain representation data; b. performing multivariate pattern analysis on the preprocessed data; c. performing generalized representation analysis on the data after multivariate pattern analysis.

[0011] Preferably, the method further includes the following step before step a:

[0012] Experimental paradigm design and data acquisition; including:

[0013] The data acquisition refers to the collection of brain representation data, which includes fMRI image data and EEG electroencephalogram (EEG) signal data.

[0014] Preferably, step a includes:

[0015] The preprocessing of the fMRI image data was performed using MATLAB and SPM12;

[0016] The preprocessing of the EEG brain signal data was performed using EEGLAB.

[0017] Preferably, the preprocessing of the fMRI image data is performed using MATLAB and SPM12, including:

[0018] Perform slice time correction: Correct the influence caused by the acquisition of signals from different brain regions at different times, set the slice order and repetition time so that the signals of all slices are corrected to the same time point;

[0019] Perform head movement correction: correct the effects of head movements caused by the subject during the scanning process and filter out images with excessive head movements;

[0020] Spatial registration: The subjects' fMRI image data were mapped to a standard space for population-level statistical analysis;

[0021] Spatial smoothing: Gaussian kernel smoothing is applied to fMRI image data to reduce noise and increase the signal-to-noise ratio.

[0022] Preferably, the preprocessing of the EEG brainwave signal data is performed using EEGLAB, including:

[0023] First, import the EEG brainwave signal data, select the EEG data file, and set the corresponding parameters according to the format and sampling rate of the EEG data file;

[0024] The second step is to confirm the location of the electrode points by checking the electrode location diagram;

[0025] The third step is to remove useless electrodes;

[0026] The fourth step is to rereference the EEG brainwave signal data;

[0027] The fifth step is to filter the EEG brainwave signal data to remove noise or highlight signals at specific frequencies;

[0028] Step 6: If it is necessary to reduce the data sampling rate, then perform downsampling.

[0029] The seventh step is to segment the data according to experimental requirements or data characteristics, using markers to divide the data into different segments.

[0030] The eighth step is to remove bad segments and interpolation bad leads, eliminating any damaged electrodes or data segments to ensure data integrity.

[0031] The ninth step is to run independent component analysis to separate and remove the independent components from the EEG signal.

[0032] Preferably, step b includes:

[0033] The first step is feature extraction: For fMRI image data, all voxels of the region of interest are selected as features; for EEG brain signal data, if it is necessary to study the brain region, relocalization is performed first, and then the signals of appropriate electrodes are selected as features; if it is necessary to study the relationship between stimulus and time, the signals of electrodes are selected as features according to different time points.

[0034] The second step is to build a model: select machine learning algorithms and statistical methods based on the data type and problem type, and build the corresponding model.

[0035] The third step is to evaluate the established model to compare the performance of different models and select the optimal model.

[0036] The fourth step is to determine which parts of the brain process which stimuli based on the classification results, analyze the regions of interest for subsequent generalization representation analysis, or obtain the brain's processing of stimuli over time.

[0037] Preferably, step c includes:

[0038] First, establish the representation RDM: For fMRI image data, calculate the estimated β value of each voxel in the region of interest using a general linear model, list the estimated β values ​​of each voxel in the region of interest as a vector, and obtain the correlation distance using Euclidean distance or 1-person's correlation coefficient based on pairwise comparisons of domain, subdomain, condition, and subject vectors to establish the RDM; For EEG electroencephalogram data, establish a time window, list the current magnitude of each electrode as a vector, and establish the RDM using Euclidean distance or 1-person's correlation coefficient based on pairwise comparisons of domain, subdomain, condition, and subject vectors.

[0039] Then, domain RDM, subdomain RDM, research RDM, subject RDM, and a constant RDM were constructed; each RDM model was built as an independent predictor to explain generalized brain activity.

[0040] Finally, the observed RDM activity was fitted using a general linear model.

[0041] Preferably, the domain RDM is used not only to explain cross-conditional generalization, but also to explain the activity mode of cross-subdomain generalization;

[0042] The subdomain RDM is used to interpret activity patterns that generalize across conditions;

[0043] The research RDM is used to explain the characteristics of specific conditions;

[0044] Each row of the subject RDM represents a participant, and each element represents the dissimilarity of brain activity patterns between two participants. Data from all participants across all conditions are combined in a matrix to construct a model-based RDM to explain the different components of the psychological hierarchy.

[0045] The constant RDM is an RDM composed of constants that is used for subsequent general linear model fitting.

[0046] This invention also provides a system for multivariate pattern analysis and generalized representation analysis of brain signals. The system includes a preprocessing module, a multivariate pattern analysis module, and a generalized representation analysis module, wherein: the preprocessing module is used to preprocess brain representation data; the multivariate pattern analysis module is used to perform multivariate pattern analysis on the preprocessed data; and the generalized representation analysis module is used to perform generalized representation analysis on the data after multivariate pattern analysis.

[0047] Preferably, the system further includes a data acquisition module, which is specifically used for: experimental paradigm design and data acquisition; wherein:

[0048] The data acquisition refers to the collection of brain representation data, which includes fMRI image data and EEG electroencephalogram (EEG) signal data.

[0049] The beneficial effects of this invention include:

[0050] First, this invention addresses the limitations of univariate analysis methods by treating fMRI image data as a multidimensional model using the MVPA method. Furthermore, MVPA offers advantages such as considering both temporal and spatial conditions, and not merely focusing on brain regions, allowing for a more comprehensive extraction of information from neuroimaging and other forms of brain representation data.

[0051] Secondly, this invention utilizes MVPA to apply brain information data collected by other tools, thereby providing a better data analysis method.

[0052] Finally, this invention utilizes generalized characterization analysis, which can overcome the limitations of single-modal analysis and perform combined analysis of similar stimuli from multiple modalities. Attached Figure Description

[0053] Figure 1 is a flowchart of the multivariate pattern analysis and generalized representation analysis method for brain signals of the present invention;

[0054] Figure 2 is a hardware architecture diagram of the brain signal multivariate pattern analysis and generalized representation analysis system of the present invention;

[0055] Figure 3 is a schematic diagram of cross-modal characterization of domains, subdomains, and conditions using GRA in an embodiment of the present invention. Detailed Implementation

[0056] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0057] Referring to Figure 1, it is a flowchart of a preferred embodiment of the multivariate pattern analysis and generalized representation analysis method for brain signals of the present invention.

[0058] Step S1: Experimental paradigm design and data acquisition. Specifically:

[0059] Experimental paradigm design: requires multimodal stimulation of the same type. For example, to study the brain's perceptual memory of intensity changes, use intensity-changing stimuli from three modalities: visual, auditory, and tactile, and divide the stimuli into domains, subdomains, and conditions from largest to smallest.

[0060] Data acquisition, also known as brain representation data acquisition, involves using fMRI to record brain neural responses and perform subsequent analysis, or using EEG electroencephalography to study event-related potentials and other aspects.

[0061] Step S2 involves preprocessing the collected data. Specifically:

[0062] fMRI image data preprocessing was performed using MATLAB and SPM12, specifically including:

[0063] First, slice timing correction is performed to correct for the effects of signals from different brain regions being acquired at different times. The slice order and repetition time (TR) are set to ensure all slice signals are aligned to the same time point. Typically, the slice order and TR values ​​used during the scan need to be known and correctly set in the software. Next, head movement correction is performed to correct for head movements during the scan, filtering out images with excessive head movement. Excessive head movement is defined as 2mm and 2°; images exceeding these standards are considered to have excessive head movement and are filtered out. Then, spatial registration (coregister) is performed to map the subject's fMRI data to a standard space (such as MNI space) for population-level statistical analysis. A suitable standard space template is selected, and parameters such as the interpolation method are set. The accuracy of this step significantly impacts subsequent analysis results. Finally, spatial smoothing (smoothing) is performed on the fMRI data using Gaussian kernel smoothing to reduce noise and increase the signal-to-noise ratio.

[0064] EEG signal data preprocessing was performed using EEGLAB, specifically including:

[0065] First, import the EEG data. Select the EEG data file and set the appropriate parameters according to the file format and sampling rate. Second, locate the electrode points by checking the Electrode Locations map. Third, remove useless electrodes using the Channel Editing function in EEGLAB, such as those for electrooculography (EOG), electrocardiography (ECG), and electromyography (EMG). Fourth, rereference the data using the Re-Reference function. For example, select bilateral mastoid processes, whole-brain average, or zero reference as new reference points. Fifth, filter the data using an appropriate filtering method to remove noise or highlight specific frequencies. Sixth, downsample the data if necessary, performing downsampling in EEGLAB. Seventh, segment the data using markers based on experimental requirements or data characteristics. Eighth, remove damaged segments and interpolation leads using EEGLAB's removal and interpolation functions to eliminate damaged electrodes or data segments, ensuring data integrity. The ninth step is Independent Component Analysis (ICA), which involves running ICA to separate and remove independent components from the EEG signal, such as electrooculography (EOG) and electrocardiogram (ECG).

[0066] Brain representation data collected by other tools also undergo similar preprocessing, which will not be elaborated here.

[0067] Step S3: Perform multivariate pattern analysis (MVPA) on the preprocessed data. Specifically:

[0068] The first step is feature extraction. For fMRI, all voxels in the region of interest are selected as features. For EEG, if the brain region needs to be studied, relocalization is performed first, and then signals from appropriate electrodes are selected as features. If the relationship between stimulus and time needs to be studied, such as how the representation of a stimulus changes over time, then electrode signals are selected as features based on different time points. Multivariate pattern analysis is then used to analyze the selected data.

[0069] The second step is to build the model. Based on the data type and problem type, select appropriate machine learning algorithms and statistical methods, and build the corresponding model.

[0070] The third step is model evaluation. Methods such as cross-validation are used to evaluate the established model, comparing the performance of different models and selecting the optimal model.

[0071] The fourth step is to analyze the classification results to determine which brain data represent which types of stimuli, thereby identifying which parts of the brain process which stimuli, and analyzing the Region of Interest (ROI) for subsequent GRA, or to obtain the brain's processing of stimuli over time.

[0072] Step S4 involves performing generalization representation analysis (GRA) on the data after multivariate pattern analysis. Please refer to Figure 3 for details. Specifically:

[0073] First, a Representational Dissimilarity Matrix (RDM) is constructed: For fMRI image data, the estimated β value of each voxel in the Region of Interest (ROI) is calculated using a general linear model (GLM). The estimated β values ​​of each voxel within the ROI are then vectorized. The correlation distance is calculated using Euclidean distance or 1-person correlation coefficient based on pairwise comparisons of the domain, sub-domain, condition, and subject vectors to construct the RDM. For EEG data, a time window is established, the current magnitude of each electrode is vectorized, and the RDM is constructed using pairwise comparisons of the domain, sub-domain, condition, and subject vectors using Euclidean distance or 1-person correlation coefficient.

[0074] Then, domain RDM, subdomain RDM, research RDM, subject RDM, and a constant RDM were constructed; each RDM model was built as an independent predictor to explain generalized brain activity. Among them:

[0075] The domain RDM is used not only to explain cross-conditional generalization but also to explain the activity patterns of cross-subdomain generalization, thus being generalizable within each domain; the subdomain RDM is used to explain the activity patterns of cross-conditional generalization; the research RDM is used to explain the characteristics of specific conditions; each row of the participant RDM represents a participant, and each element represents the dissimilarity of brain activity patterns between two participants; data from all participants across all conditions are combined in a matrix to construct a model-based RDM to explain the different components of the psychological hierarchy; the constant RDM is an RDM composed of constants, which is used for subsequent GLM fitting.

[0076] Finally, the observed RDM activity was fitted using GLM, and the parameter estimate (β) represents the generalizable form of activity that captures the correlation of activity patterns. It is regarded as a generalization index, and a significant β value indicates that the domain-related coherent activity patterns in the ROI are shared among subdomains, conditions, or participants.

[0077] Referring to Figure 2, this is a hardware architecture diagram of the brain signal multivariate pattern analysis and generalized representation analysis system 10 of the present invention. The system includes: an acquisition module 101, a preprocessing module 102, a multivariate pattern analysis module 103, and a generalized representation analysis module 104. Wherein:

[0078] The acquisition module 101 is used for experimental paradigm design and data acquisition. Specifically:

[0079] Experimental paradigm design: requires multimodal stimulation of the same type. For example, to study the brain's perceptual memory of intensity changes, use intensity-changing stimuli from three modalities: visual, auditory, and tactile, and divide the stimuli into domains, subdomains, and conditions from largest to smallest.

[0080] Data acquisition, also known as brain representation data acquisition, involves using fMRI to record brain neural responses and perform subsequent analysis, or using EEG electroencephalography to study event-related potentials and other aspects.

[0081] The preprocessing module 102 is used to preprocess the collected data. Wherein:

[0082] fMRI image data preprocessing was performed using MATLAB and SPM12, specifically including:

[0083] First, slice timing correction is performed to correct for the effects of signals from different brain regions being acquired at different times. The slice order and repetition time (TR) are set to align all slice signals to the same time point. Typically, the slice order and TR values ​​used during the scan need to be known and correctly set in the software. Next, head movement correction is performed to correct for the effects of head movements during the scan, filtering out images with excessive head movement. Excessive head movement is defined as 2mm and 2°; images exceeding these standards are considered to have excessive head movement and are filtered out. Then, spatial registration (coregister) is performed to map the subject's fMRI data to a standard space (such as MNI space) for population-level statistical analysis. A suitable standard space template is selected, and parameters such as the interpolation method are set. Finally, spatial smoothing (smoothing) is performed on the fMRI data using Gaussian kernel smoothing to reduce noise and increase the signal-to-noise ratio.

[0084] EEG signal preprocessing was performed using EEGLAB, specifically including:

[0085] First, import the EEG data. Select the EEG data file and set the appropriate parameters according to the file format and sampling rate. Second, locate the electrode points by checking the Electrode Locations map. Third, remove useless electrodes using the Channel Editing function in EEGLAB, such as those for electrooculography (EOG), electrocardiography (ECG), and electromyography (EMG). Fourth, rereference the data using the Re-Reference function. For example, select bilateral mastoid processes, whole-brain average, or zero reference as new reference points. Fifth, filter the data using an appropriate filtering method to remove noise or highlight specific frequencies. Sixth, downsample the data if necessary, performing downsampling in EEGLAB. Seventh, segment the data using markers based on experimental requirements or data characteristics. Eighth, remove damaged segments and interpolation leads using EEGLAB's removal and interpolation functions to eliminate damaged electrodes or data segments, ensuring data integrity. The ninth step is Independent Component Analysis (ICA), which involves running ICA to separate and remove independent components from the EEG signal, such as electrooculography (EOG) and electrocardiogram (ECG).

[0086] Brain representation data collected by other tools also undergo similar preprocessing, which will not be elaborated here.

[0087] The multivariate pattern analysis module 103 is used to perform multivariate pattern analysis (MVPA) on the preprocessed data. Specifically:

[0088] The first step is feature extraction. For fMRI, all voxels in the region of interest are selected as features. For EEG, if the brain region needs to be studied, relocalization is performed first, and then signals from appropriate electrodes are selected as features. If the relationship between stimulus and time needs to be studied, such as how the representation of a stimulus changes over time, then electrode signals are selected as features based on different time points. Multivariate pattern analysis is then used to analyze the selected data.

[0089] The second step is to build the model. Based on the data type and problem type, select appropriate machine learning algorithms and statistical methods, and build the corresponding model.

[0090] The third step is model evaluation. Methods such as cross-validation are used to evaluate the established model, comparing the performance of different models and selecting the optimal model.

[0091] The fourth step is to analyze the classification results to determine which brain data represent which types of stimuli, thereby identifying which parts of the brain process which stimuli, and analyzing the Region of Interest (ROI) for subsequent GRA, or to obtain the brain's processing of stimuli over time.

[0092] The generalization characterization analysis module 104 is used to perform generalization characterization analysis (GRA) on the data after multivariate pattern analysis, as shown in Figure 3. Specifically:

[0093] First, a Representational Dissimilarity Matrix (RDM) is constructed: For fMRI image data, the estimated β value of each voxel in the Region of Interest (ROI) is calculated using a general linear model (GLM). The estimated β values ​​of each voxel within the ROI are then vectorized. The correlation distance is calculated using Euclidean distance or 1-person correlation coefficient based on pairwise comparisons of the domain, sub-domain, condition, and subject vectors to construct the RDM. For EEG data, a time window is established, the current magnitude of each electrode is vectorized, and the RDM is constructed using pairwise comparisons of the domain, sub-domain, condition, and subject vectors using Euclidean distance or 1-person correlation coefficient.

[0094] Then, domain RDM, subdomain RDM, research RDM, subject RDM, and a constant RDM were constructed; each RDM model was built as an independent predictor to explain generalized brain activity. Among them:

[0095] The domain RDM is used not only to explain cross-conditional generalization but also to explain the activity patterns of cross-subdomain generalization, thus being generalizable within each domain; the subdomain RDM is used to explain the activity patterns of cross-conditional generalization; the research RDM is used to explain the characteristics of specific conditions; each row of the participant RDM represents a participant, and each element represents the dissimilarity of brain activity patterns between two participants; data from all participants across all conditions are combined in a matrix to construct a model-based RDM to explain the different components of the psychological hierarchy; the constant RDM is an RDM composed of constants, which is used for subsequent GLM fitting.

[0096] Finally, the observed RDM activity was fitted using GLM, and the parameter estimate (β) represents the generalizable form of activity that captures the correlation of activity patterns. It is regarded as a generalization index, and a significant β value indicates that the domain-related coherent activity patterns in the ROI are shared among subdomains, conditions, or participants.

[0097] Using this invention, MVAP can more comprehensively reveal the characteristics of brain activity based on data such as EEG and MEG, providing a completely new perspective for neuroscience research. It can more effectively analyze and understand brain activity data; it can process high-dimensional, complex neuroscience data and discover potential patterns and regularities within the data, allowing researchers to gain a deeper understanding of brain activity mechanisms. GRA can explore and understand the brain's processing of stimuli more deeply based on multimodal stimuli.

[0098] Although the present invention has been described with reference to the present preferred embodiments, those skilled in the art should understand that the above preferred embodiments are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of multivariate pattern analysis and generalized representational analysis of brain signals, characterized in that, The method includes the following steps: a. Preprocessing brain representation data; b. Perform multivariate pattern analysis on the preprocessed data; c. Perform generalization characterization analysis on the data after multivariate pattern analysis.

2. The method as described in claim 1, characterized in that, The steps preceding step a also include: Experimental paradigm design and data acquisition; including: The data acquisition refers to the collection of brain representation data, which includes fMRI image data and EEG electroencephalogram (EEG) signal data.

3. The method as described in claim 2, characterized in that, Step a includes: The preprocessing of the fMRI image data was performed using MATLAB and SPM12; The preprocessing of the EEG brain signal data was performed using EEGLAB.

4. The method as described in claim 3, characterized in that, The preprocessing of the fMRI image data was performed using MATLAB and SPM12, including: Perform slice time correction: Correct the influence caused by the acquisition of signals from different brain regions at different times, set the slice order and repetition time so that the signals of all slices are corrected to the same time point; Perform head movement correction: correct the effects of head movements caused by the subject during the scanning process and filter out images with excessive head movements; Spatial registration: The subjects' fMRI image data were mapped to a standard space for population-level statistical analysis; Spatial smoothing: Gaussian kernel smoothing is applied to fMRI image data to reduce noise and increase the signal-to-noise ratio.

5. The method as described in claim 4, characterized in that, The preprocessing of the EEG signal data was performed using EEGLAB, including: First, import the EEG brainwave signal data. Select the EEG data file and then import the data according to the EEG parameters. Set the appropriate parameters according to the file format and sampling rate; The second step is to confirm the location of the electrode points by checking the electrode location diagram; The third step is to remove useless electrodes; The fourth step is to rereference the EEG brainwave signal data; The fifth step is to filter the EEG brainwave signal data to remove noise or highlight signals at specific frequencies; Step 6: If it is necessary to reduce the data sampling rate, then perform downsampling. The seventh step is to segment the data according to experimental requirements or data characteristics, using markers to divide the data into different segments. The eighth step is to remove bad segments and interpolation bad leads, eliminating any damaged electrodes or data segments to ensure data integrity. The ninth step is to run independent component analysis to separate and remove the independent components from the EEG signal.

6. The method as described in claim 5, characterized in that, Step b includes: The first step is feature extraction: For fMRI image data, all voxels of the region of interest are selected as features; for EEG brain signal data, if it is necessary to study the brain region, relocalization is performed first, and then the signals of appropriate electrodes are selected as features; if it is necessary to study the relationship between stimulus and time, the signals of electrodes are selected as features according to different time points. The second step is to build a model: select machine learning algorithms and statistical methods based on the data type and problem type, and build the corresponding model. The third step is to evaluate the established model to compare the performance of different models and select the optimal model. The fourth step is to determine which parts of the brain process which stimuli based on the classification results, analyze the regions of interest for subsequent generalization representation analysis, or obtain the brain's processing of stimuli over time.

7. The method as described in claim 6, characterized in that, Step c includes: First, establish the Representation Model (RDM): For fMRI image data, estimate the β value of each voxel in the region of interest using a general linear model. Array the estimated β values ​​of each voxel within the region of interest into a vector, and calculate the correlation distance using Euclidean distance or 1-person's correlation coefficient based on pairwise comparisons of the domain, subdomain, conditions, and subject vectors, thus establishing the RDM. For EEG data, establish the time... The window lists the current magnitude of each electrode as a vector, and establishes the RDM by comparing the domain, subdomain, conditions, and subject vectors pairwise using Euclidean distance or 1-person's correlation coefficient. Then, domain RDM, subdomain RDM, research RDM, subject RDM, and a constant RDM were constructed; each RDM model was built as an independent predictor to explain generalized brain activity. Finally, the observed RDM activity was fitted using a general linear model.

8. The method as described in claim 7, characterized in that: The domain RDM is used not only to explain cross-conditional generalization, but also to explain the activity mode of cross-subdomain generalization; The subdomain RDM is used to interpret activity patterns that generalize across conditions; The research RDM is used to explain the characteristics of specific conditions; Each row of the subject RDM represents a participant, and each element represents the dissimilarity of brain activity patterns between two participants. Data from all participants across all conditions are combined in a matrix to construct a model-based RDM to explain the different components of the psychological hierarchy. The constant RDM is an RDM composed of constants that is used for subsequent general linear model fitting.

9. A multivariate pattern analysis and generalized representation analysis system for brain signals, characterized in that, The system includes a preprocessing module, a multivariate pattern analysis module, and a generalization characterization analysis module, wherein: The preprocessing module is used to preprocess brain representation data; The multivariate pattern analysis module is used to perform multivariate pattern analysis on the preprocessed data; The generalized characterization analysis module is used to perform generalized characterization analysis on the data after multivariate pattern analysis.

10. The system as described in claim 9, characterized in that, The system also includes a data acquisition module, which is specifically used for: Experimental paradigm design and data acquisition; including: The data acquisition refers to the collection of brain representation data, which includes fMRI image data and EEG electroencephalogram (EEG) signal data.

Images