Method for detecting brain structure-function relation based on multi-scale features

Abstract

The invention provides a method for detecting a brain structure-function relationship based on multi-scale features. The method comprises the steps of 1, calculating a brain structure connection matrix SC, a function connection matrix FC and a geodesic distance matrix DM; 2, calculating a brain gene expression similarity matrix GM; 3, obtaining a brain cell construction similarity matrix CM; and 4, fitting a brain structure-function relation coefficient R2 according to the structure connection matrix SC, the function connection matrix FC, the geodesic distance matrix DM, the gene expression similarity matrix GM and the cell construction similarity matrix CM obtained in the steps 1, 2 and 3. According to the method, a method of fusing multi-scale brain features is adopted to replace a common method of direct brain structure connection and function connection correlation coefficients, and micro-scale brain information including two features of gene expression and cell construction is added, so that the method has more biological significance when detecting the brain structure-function relationship.

Description

Technical Field

[0001] The present application relates to the technical field of brain function research, and in particular, to a method for detecting brain structure-function relationships based on multi-scale features. Background Art

[0002] The human brain is a complex multi-scale connection system, from genes, cells, and synapses at the microscopic scale, to cell architecture at the mesoscopic scale, to connections and networks at the macroscopic scale. Science emphasized several points in the special issue "Brain Connections": First, no neuron is an island, which shows the criticality of brain connections. Second, "scale" is very important in brain connection research. However, no research has explored the relationship between brain structure and brain function under the constraints of brain characteristics at different scales.

[0003] Magnetic resonance imaging (MRI) technology provides us with a non-invasive method to observe brain structure and function. Generally, the relationship between brain structure and brain function can be defined by directly measuring the correlation between the two. Taking Figure (1) as an example, diffusion tensor imaging can calculate the structural connectivity matrix (N×N, where N represents the number of brain regions), and functional magnetic resonance imaging can calculate the functional connectivity (N×N). The Pearson correlation coefficient is calculated for the corresponding rows or columns in the two matrices, and finally the brain structure and brain function coupling relationship matrix (N×1) can be obtained, which reflects the spatial coupling relationship between brain structure and function. The current method mainly considers the direct relationship between structural connectivity and functional connectivity, that is, directly calculating the linear correlation coefficient of the two connection matrices, but does not consider the influence of microscopic or other macroscopic brain structure characteristics on brain function, such as gene expression, cell architecture, geodesic distance, etc.

[0004] In addition to the direct calculation of the correlation coefficient between structural connectivity and functional connectivity as described above to define the relationship between brain structure and function, it is also possible to fit the spatial relationship between brain structure and function by constructing a simple linear relationship between structural connectivity and functional connectivity. Figure 2 As shown in the figure, some network properties can be calculated through structural connectivity, including path length and transmissibility, which describe the anatomical distance between brain regions, and Euclidean distance describes the physical distance between brain regions. This method mainly considers the simple linear relationship between multiple network characteristics of structural connectivity and functional connectivity, and also considers the influence of the physical space (Euclidean distance) of brain regions on brain functional relationships. However, there are still two shortcomings: First, the influence of microscopic brain structural characteristics on brain function is not considered, such as gene expression and cell architecture characteristics. Second, there are many grooves and gyri in the brain, such as Figure 3 As shown in the figure, the Euclidean distance is used to define the physical distance between brain regions. This method ignores the sulcus and gyrus information and is not accurate enough. Summary of the invention

[0005] In view of this, the present application provides a method for detecting brain structure-function relationships based on multi-scale features. It uses brain connections at different scales, such as brain gene expression, molecular metabolism, structural connectivity, and geodesic distance, to develop a hierarchical constraint method based on multi-scale connectivity to more accurately detect the brain's structure-function relationship.

[0006] To achieve the above objectives, the technical solutions adopted in this application are as follows:

[0007] The present application provides a method for detecting brain structure-function relationships based on multi-scale features, including:

[0008] Step 1: Calculate the brain structure connection matrix SC, functional connection matrix FC and geodesic distance matrix DM;

[0009] Step 2: Calculate the brain gene expression similarity matrix GM;

[0010] Step 3: Obtain the brain cell architecture similarity matrix CM;

[0011] Step 4: According to the structural connectivity matrix SC, functional connectivity matrix FC, geodesic distance matrix DM, gene expression similarity matrix GM and cell architecture similarity matrix CM obtained in steps 1, 2 and 3, fit the brain structure-function relationship coefficient R 2 .

[0012] Furthermore, the brain structure connection matrix SC is obtained by the following method:

[0013] 101: Acquire original T1-weighted MRI data and field-map-corrected diffusion-weighted images of the brain;

[0014] 102: Use unweighted diffusion b 0 Image and field map correction Diffusion-weighted images calculate distortion caused by eddy currents to reduce the effects of head motion;

[0015] 103: Use the MRtrix3 toolkit to perform probabilistic fiber tract tracking on the corrected diffusion-weighted images;

[0016] 104: Segment the T1-weighted structural image into different tissue types, including cortex, subcortical gray matter, white matter, and cerebrospinal fluid. Then transform the T1 segmented different tissue images into individual diffusion-weighted image space using FSL's FLIRT command. Then estimate the response function of multiple tissues, and perform constrained spherical deconvolution and intensity normalization.

[0017] 105: All white matter voxels are used as seed points and streamline tracing is performed using the toolkit’s default algorithm iFOD2;

[0018] 106: Use SIFT2 algorithm to optimize the cross-sectional magnification of each streamline to match the whole brain fiber bundle map;

[0019] 107: Map the streamlines to a specific atlas containing m brain regions, and then use log transformation to obtain the m×m structural connection matrix SC.

[0020] Furthermore, the functional connectivity matrix FC is obtained by the following method:

[0021] 201: Acquiring fMRI data;

[0022] 202: Preprocessing of functional MRI data using the DPABI toolbox;

[0023] 203: extracting time series signals of corresponding brain regions using a specific atlas containing m brain regions;

[0024] 204: Perform Pearson correlation analysis on the signal to obtain an m×m functional connection matrix FC.

[0025] Furthermore, the geodesic distance matrix DM is obtained by the following method:

[0026] 301: Preprocess the T1-weighted structural images using the recon-all command of FreeSurfer software;

[0027] 302: Use the command for calculating geodesic distance in the ConnectomeWorkBench toolkit to finally obtain an m×m geodesic distance matrix DM.

[0028] Furthermore, the specific method for calculating the brain gene expression similarity matrix GM is:

[0029] 401: Obtaining microarray data of a gene expression similarity matrix and performing preprocessing, wherein the preprocessing is specifically as follows:

[0030] First, the gene probes were repositioned, and the probes with an intensity threshold below 0.5 were filtered, and the single probe ΔS(p) with the highest differential stability was selected to represent each gene, where the differential stability was calculated as:

[0031]

[0032] The p is a single probe p in two donor brains B i and B j Spearman rank correlation of expression across regions, N is the total number of donor brains; the probes are used by the Allen Atlas Working Group when collecting microarray data, and their function is to detect nucleic acid sequences.

[0033] 402: mirroring the tissue sample to the left and right hemispheres, then finding the region closest to the specific atlas, and assigning the brain tissue sample containing the gene expression data to the corresponding brain region using the corrected MNI coordinates;

[0034] 403: Samples assigned to the same brain region were averaged for each donor separately;

[0035] 404: Use the sigmoid function to standardize the gene expression values ​​of each donor, and finally obtain an average m×n gene expression matrix GM in each donor, where m corresponds to the brain region and n corresponds to the retained genes;

[0036] 405: Find k genes that are specifically expressed in the brain among the n retained genes, calculate the co-expression similarity of these k genes between brain regions using Pearson correlation, and finally obtain the m×m gene expression similarity matrix GM.

[0037] Furthermore, the method for obtaining the brain cell architecture similarity matrix CM is:

[0038] 501: Obtain brain cell density data through BigBrain atlas;

[0039] 502: The MPC is calculated by the following formula:

[0040]

[0041] where i and j represent the corresponding brain regions, and r ij is the Pearson correlation coefficient of the cell density of brain regions i and j in the BigBrain atlas, r ic is the correlation coefficient between the cell density of brain region i and the average cell density of the entire cerebral cortex, and MPC represents the cytoarchitecture similarity coefficient between brain regions;

[0042] 503: Using a specific atlas containing m brain regions, the calculation steps of formula (2) above are performed on each pair of brain regions, and finally an m×m cell architecture similarity matrix CM can be calculated.

[0043] Furthermore, the brain structure-function relationship coefficient R is fitted based on the structural connection matrix SC, functional connection matrix FC, geodesic distance matrix DM, gene expression similarity matrix GM and cell architecture similarity matrix CM obtained in steps one, two and three. 2 The specific method is as follows:

[0044] FC=β 0 +β 1 SC+β 2 DM+β 3 GM+β 4CM; (3)

[0045] Among them, FC, SC, DM, GM and CM represent five multi-scale brain features, namely functional connectivity, structural connectivity, geodesic distance, gene expression and cell architecture, and β represents the influence of each brain feature. Then, the ordinary least squares method of MATLAB, that is, the function fitlm.m, is used to solve the regression coefficient β of the linear model. 0 , β 1 , β 2 , β 3 and β 4 , use mdl.Rsquared.Ordinary to get the coefficient of determination R of the linear model 2 , the coefficient of determination is calculated by the following formulas (4) and (5):

[0046]

[0047] Among them, fc represents the model prediction value obtained using the fitted regression coefficient, and FC represents the actual observed value. Representing FC i The average value of R 2 The larger the coefficient, the closer the brain structure-function relationship is.

[0048] The beneficial effects of this application are:

[0049] Compared with the prior art, the beneficial effects of this application are:

[0050] 1. This application adopts a method of integrating multi-scale brain features, replacing the commonly used method of direct brain structure connection and functional connection correlation coefficient, and adds microscopic brain information, including gene expression and cell architecture, making the detection of brain structure-function relationship more biologically meaningful;

[0051] 2. This application uses geodesic distance to describe the physical distance between different brain regions, replacing the commonly used Euclidean distance method, so that the brain's sulcus information can be considered when measuring the physical distance of brain regions, so that the detection of brain structure is more accurate and more realistic;

[0052] 3. Integrating the characteristics of the brain at the microscale and macroscale, multi-modal and multi-source information will help to better understand the relationship between the structure and function of the cerebral cortex, thereby helping to explore the physiological mechanisms of brain diseases;

[0053] 4. The linear model used can analyze the structure-function relationship of each specific brain region and more intuitively reflect the anatomical position of the brain. Therefore, specific brain regions can be specifically located for analysis in clinical practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 Schematic diagram of direct correlation measurements of brain structure and brain function provided for this application;

[0056] Figure 2 A schematic diagram for predicting brain functional connectivity through path length, propagability, and Euclidean distance of brain structural connections provided for this application;

[0057] Figure 3 A schematic diagram of the Euclidean distance and geodesic distance between brain area A and brain area B provided in this application;

[0058] Figure 4 A schematic diagram of the method for detecting brain structure-function relationships based on multi-scale features in this application;

[0059] Figure 5 This is a flow chart of the method for detecting brain structure-function relationships based on multi-scale features in this application. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0061] like Figure 4 As shown, a method for detecting brain structure-function relationship based on multi-scale features is provided, including:

[0062] S10: Calculate the brain structural connectivity matrix SC, functional connectivity matrix FC and geodesic distance matrix DM;

[0063] Furthermore, the brain structure connection matrix SC is obtained by the following method:

[0064] 101: Acquire original T1-weighted MRI data and field-map-corrected diffusion-weighted images of the brain;

[0065] In this embodiment, the diffusion-weighted image can be acquired through magnetic resonance diffusion-weighted imaging technology, specifically through a magnetic resonance machine. The magnetic resonance diffusion-weighted imaging technology is different from conventional magnetic resonance imaging (MRI) in that it is based on the movement of water molecules and provides information based on the physiological state of the brain.

[0066] 102: Use unweighted diffusion b 0 Image and field map correction Diffusion-weighted images calculate distortion caused by eddy currents to reduce the effects of head motion;

[0067] The field image correction refers to static and dynamic distortion correction of imaging.

[0068] 103: Use the MRtrix3 toolkit to perform probabilistic fiber tract tracking on the corrected diffusion-weighted images;

[0069] 104: segmenting the T1-weighted structural image into different tissue types, including cortical and subcortical gray matter, white matter and cerebrospinal fluid, and then transforming the segmented different tissue type images into individual diffusion-weighted image space through the FLIRT command of FSL, and then estimating the response function of multiple tissues, and performing constrained spherical deconvolution and intensity normalization processing;

[0070] The FLIRT is a linear correction tool of FSL, and the official provides a rich calling method for sharing.

[0071] 105: Using all white matter voxels as seed points, the toolkit default algorithm iFOD2 is used for streamline tracing

[0072] 106: Use SIFT2 algorithm to optimize the cross-sectional magnification of each streamline to match the whole brain fiber bundle map;

[0073] 107: Map the streamlines to a specific atlas containing m brain regions, and then use log transformation to obtain the m×m structural connection matrix SC.

[0074] The log transformation is a logarithmic transformation.

[0075] Furthermore, the functional connectivity matrix FC is obtained by the following method:

[0076] 201: Acquisition of functional magnetic resonance imaging data;

[0077] The functional magnetic resonance data may be acquired by a magnetic resonance imaging machine.

[0078] 202: Preprocessing of functional MRI data using the DPABI toolbox;

[0079] The DPABI toolbox is run in SPM12 software. In one embodiment, the preprocessing includes but is not limited to: removing the first 10 time point images, performing time layer correction and head motion correction on the remaining images, spatial normalization to the Montreal standard space, and resampling the voxel point size to 3×3×3mm 3, regress white matter signals, ventricular signals, and 24 head motion parameters, take 0.5mm as the threshold, interpolate the time point when the head motion parameter exceeds the threshold and the two time points before and after this point, and use a Gaussian smoothing kernel with a FWHM (full width at half maxima) of 8mm for spatial smoothing. Finally, remove linear correction and perform temporal bandpass filtering with a bandpass frequency of 0.01-0.08Hz.

[0080] 203: extracting time series signals of corresponding brain regions using a specific atlas containing m brain regions;

[0081] 204: Perform Pearson correlation analysis on the signal to obtain an m×m functional connection matrix FC.

[0082] For example, the time signal of brain area A is [t1, t2, t3, ..., tn], and the time signal of brain area B is a vector of [t'1, t'2, t'3, ..., t'n], where n is the number of time points in the time series. The Pearson correlation of the two vectors is calculated to obtain the correlation coefficient r1. There are a total of m brain areas, and the correlation is calculated for each of them to obtain an m×m correlation coefficient matrix.

[0083] Furthermore, the geodesic distance matrix DM is obtained by the following method:

[0084] 301: Preprocess the T1-weighted structural images using the recon-all command of FreeSurfer software;

[0085] The FreeSurfer software is an open source software for analyzing and visualizing structural, functional and diffusion neuroimaging data for cross-sectional and longitudinal studies. It was developed by the Computational Neuroimaging Laboratory of the Athinoula A.Martinos Center for Biomedical Imaging and is mainly used to analyze brain neural data, providing a series of algorithms to quantify the function, connectivity and structural properties of the human brain. FreeSurfer can perform three-dimensional reconstruction of high-resolution magnetic resonance images, generate flattened or expanded images, and obtain anatomical parameters such as cortical thickness, area, and gray matter volume.

[0086] The preprocessing mainly includes cortical segmentation and surface reconstruction. The recon-all command in FreeSurfer is used to construct the anatomical cortical layer for the T1-weighted structural image of each subject's individual space, including skull stripping, cortical extraction, and separation of left and right cerebral hemispheres and subcortical tissue. The surface reconstruction includes the identification of gray / white matter surface and meningeal surface.

[0087] 302: Use the command for calculating geodesic distance in the ConnectomeWorkBench toolkit to finally obtain an m×m geodesic distance matrix DM.

[0088] S20: Calculate the brain gene expression similarity matrix GM;

[0089] Microscopic-scale brain metabolism image data can be obtained through PET-fMRI data analysis.

[0090] The gene expression similarity data is obtained by analyzing the co-expression similarity of these genes specifically expressed in brain tissue in different brain regions, and the cell architecture similarity data is obtained by analyzing the co-variation similarity of cell density with cortical depth in different brain regions. Both of them define similarity by calculating the Pearson correlation coefficient.

[0091] Furthermore, the specific method for calculating the brain gene expression similarity matrix GM is:

[0092] 401: Obtaining microarray data of a gene expression similarity matrix and performing preprocessing, wherein the preprocessing is specifically as follows:

[0093] First, the probes were repositioned and the probes with an intensity threshold below 0.5 were filtered, and the single probe with the highest differential stability ΔS(p) was selected to represent each gene, where the differential stability was calculated as:

[0094]

[0095] The p is a single probe p in two donor brains B i and B j Spearman rank correlation of expression across regions, where N is the total number of donor brains; this process retains N probes, each representing a unique gene.

[0096] The microarray data of the gene expression similarity matrix can be downloaded from the Allen Human Brain Atlas (http: / / human.brain-map.org / ), and the downloaded data are preprocessed using the abagen toolkit.

[0097] 402: mirroring the tissue sample to the left and right hemispheres, then finding the region closest to the specific atlas, and assigning the brain tissue sample containing the gene expression data to the corresponding brain region using the corrected MNI coordinates;

[0098] The region closest to the specific atlas is the region with a range of no more than 2 mm. If a brain region is not assigned any sample according to the above procedure, the sample closest to the centroid of the region is selected to ensure that all brain regions are assigned a value.

[0099] 403: Samples assigned to the same brain region were averaged for each donor separately;

[0100] 404: Use the sigmoid function to standardize the gene expression values ​​of each donor, and finally obtain an average m×n gene expression matrix GM in each donor, where m corresponds to the brain region and n corresponds to the retained genes;

[0101] 405: Find k genes that are specifically expressed in the brain among the n retained genes, calculate the co-expression similarity of these k genes between brain regions using Pearson correlation, and finally obtain the m×m gene expression similarity matrix GM.

[0102] If there is a value of zero in the matrix GM, it means that the gene is not expressed in the brain region. Referring to the genes specifically expressed in brain tissue used in related studies, k of them have expression values ​​in all brain regions, and the remaining genes have no gene expression values ​​in certain brain regions. Therefore, the k genes in m brain regions are used to perform co-expression correlation analysis to obtain an m×m gene co-expression matrix GM.

[0103] S30: Obtain brain cell architecture similarity matrix CM;

[0104] Furthermore, the method for obtaining the brain cell architecture similarity matrix CM is:

[0105] 501: Obtain brain cell density data through BigBrain atlas;

[0106] 502: Calculated by the following formula:

[0107]

[0108] where i and j represent the corresponding brain regions, and r ij is the Pearson correlation coefficient of the cell density of brain regions i and j in the BigBrain atlas, r ic is the correlation coefficient between the cell density of brain region i and the average cell density of the entire cerebral cortex, and MPC represents the cytoarchitecture similarity coefficient between brain regions;

[0109] 503: Using a specific atlas containing m brain regions, the calculation steps of formula (2) above are performed on each pair of brain regions, and finally an m×m cell architecture similarity matrix CM can be calculated.

[0110] For example, the cell density data of brain area A is [d1, d2, d3, ..., dn], and the cell density data of brain area B is a vector of [d'1, d'2, d'3, ..., d'n], where n is the number of samples at different depths of the cortex. The Pearson correlation of the two vectors is calculated to obtain the correlation coefficient r. There are a total of m brain areas, and the correlation is calculated for each of them to obtain an m×m correlation coefficient matrix, where each element in the matrix is ​​r.

[0111] In addition, the cytoarchitectonic similarity matrix can also be obtained from https: / / github.com / MICA-MNI / micaopen / tree / master / MPC.

[0112] S40: According to the structural connection matrix SC, functional connection matrix FC, geodesic distance matrix DM, gene expression similarity matrix GM and cell architecture similarity matrix CM obtained in steps 1, 2 and 3, the relationship coefficient R between brain structure and function is fitted. 2 The specific method is as follows:

[0113] FC=β 0 +β 1 SC+β 2 DM+β 3 GM+β 4 CM; (3)

[0114] Among them, FC, SC, DM, GM and CM represent five multi-scale brain features, namely functional connectivity, structural connectivity, geodesic distance, gene expression and cell architecture, and β represents the influence of each brain feature. Then, the ordinary least squares method of MATLAB, that is, the function fitlm.m, is used to solve the regression coefficient β of the linear model. 0 , β 1 , β 2 , β 3 and β 4 , use mdl.Rsquared.Ordinary to get the coefficient of determination R of the linear model 2 , the coefficient of determination is calculated by the following formulas (4) and (5):

[0115]

[0116] Among them, fc represents the model prediction value obtained using the fitted regression coefficient, and FC represents the actual observed value. Representing FC i The larger the R2 coefficient, the closer the relationship between brain structure and function.

[0117] This application adopts a method of fusing multi-scale brain features, replacing the commonly used method of direct brain structure connection and functional connection correlation coefficient, and adds microscopic scale brain information, including two features of gene expression and cell architecture, so that the detection of brain structure-function relationship is more biologically meaningful; geodesic distance is used to describe the physical distance between different brain regions, replacing the commonly used Euclidean distance method, so that the brain's sulcus information can be considered when measuring the physical distance between brain regions, so that the detection of brain structure is more accurate and more practical.

[0118] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for detecting brain structure-function relationships based on multi-scale features, characterized in that: include: Step 1: Calculate the brain structure connection matrix SC, functional connection matrix FC and geodesic distance matrix DM; Step 2: Calculate the brain gene expression similarity matrix GM; Step 3: Obtain the brain cell architecture similarity matrix CM; Step 4: According to the structural connectivity matrix SC, functional connectivity matrix FC, geodesic distance matrix DM, gene expression similarity matrix GM and cell architecture similarity matrix CM obtained in steps 1, 2 and 3, fit the brain structure-function relationship coefficient R 2 .

2. The method for detecting brain structure-function relationship based on multi-scale features according to claim 1, characterized in that: The brain structure connection matrix SC is obtained by the following method: 101: Obtain original T1-weighted MRI data and diffusion-weighted images of the brain; 102: Use the non-weighted diffusion b0 image and field map to correct the diffusion weighted image to calculate the distortion caused by eddy currents to reduce the influence of head movement; 103: Use the MRtrix3 toolkit to perform probabilistic fiber tract tracking on the corrected diffusion-weighted images; 104: Segment the T1-weighted structural image into different tissue types, including cortex, subcortical gray matter, white matter, and cerebrospinal fluid. Then transform the T1 segmented different tissue images into individual diffusion-weighted image space using FSL's FLIRT command. Then estimate the response function of multiple tissues, and perform constrained spherical deconvolution and intensity normalization. 105: All white matter voxels are used as seed points and streamline tracing is performed using the toolkit’s default algorithm iFOD2; 106: Use SIFT2 algorithm to optimize the cross-sectional magnification of each streamline to match the whole brain fiber bundle map; 107: Map the streamlines to a specific atlas containing m brain regions, and then use log transformation to obtain the m×m structural connection matrix SC.

3. The method for detecting brain structure-function relationship based on multi-scale features according to claim 1 or 2, characterized in that: The functional connectivity matrix FC is obtained by the following method: 201: Acquiring fMRI data; 202: Preprocessing of functional MRI data using the DPABI toolbox; 203: extracting time series signals of corresponding brain regions using a specific atlas containing m brain regions; 204: Perform Pearson correlation analysis on the signal to obtain an m×m functional connection matrix FC.

4. The method for detecting brain structure-function relationship based on multi-scale features as claimed in claim 3, characterized in that: The geodesic distance matrix DM is obtained by the following method: 301: Preprocess the T1-weighted structural images using the recon-all command of FreeSurfer software; 302: Use the command for calculating geodesic distance in the ConnectomeWorkBench toolkit to finally obtain an m×m geodesic distance matrix DM.

5. The method for detecting brain structure-function relationship based on multi-scale features according to claim 4, characterized in that: The specific method for calculating the brain gene expression similarity matrix GM is: 401: Obtain microarray data from the Allen human brain atlas and perform preprocessing, wherein the preprocessing is specifically as follows: First, the probes were repositioned and the probes with an intensity threshold below 0.5 were filtered, and the single probe with the highest differential stability ΔS(p) was selected to represent each gene, where the differential stability was calculated as: The p is a single probe p in two donor brains B i and B j Spearman rank correlation of expression across regions, N is the total number of donor brains; 402: mirroring the tissue sample to the left and right hemispheres, then finding the region closest to the specific atlas, and assigning the brain tissue sample containing gene expression data to the brain region using the corrected MNI coordinates; 403: Samples assigned to the same brain region were averaged for each donor separately; 404: Use the sigmoid function to standardize the gene expression values ​​of each donor, and finally obtain an average m×n gene expression matrix GM in each donor, where m corresponds to the brain region and n corresponds to the retained genes; 405: Find k genes that are specifically expressed in the brain among the n retained genes, calculate the co-expression similarity of these k genes between brain regions using Pearson correlation, and finally obtain the m×m gene expression similarity matrix GM.

6. The method for detecting brain structure-function relationship based on multi-scale features according to claim 5, characterized in that: The method for obtaining the brain cell architecture similarity matrix CM is: 501: Obtain brain cell density data through BigBrain atlas; 502: The MPC is calculated by the following formula: where i and j represent the corresponding brain regions, and r ij is the Pearson correlation coefficient of the cell density of brain regions i and j in the BigBrain atlas, r ic is the correlation coefficient between the cell density of brain region i and the average cell density of the entire cerebral cortex, and MPC represents the cytoarchitecture similarity coefficient between brain regions; 503: Using a specific atlas containing m brain regions, the calculation steps of formula (2) above are performed on each pair of brain regions, and finally an m×m cell architecture similarity matrix CM can be calculated.

7. The method for detecting brain structure-function relationship based on multi-scale features according to claim 6, characterized in that: According to the structural connection matrix SC, functional connection matrix FC, geodesic distance matrix DM, gene expression similarity matrix GM and cell architecture similarity matrix CM obtained in steps one, two and three, the relationship coefficient R between brain structure and function is fitted. 2 The specific method is as follows: FC=β0+β1SC+β2DM+β3GM+β4CM; (3) Among them, FC, SC, DM, GM and CM represent five multi-scale brain features, namely functional connectivity, structural connectivity, geodesic distance, gene expression and cell architecture, respectively, and β represents the influence of each brain feature. Then, the ordinary least squares method of MATLAB, namely the function fitlm.m, is used to solve the regression coefficients β0, β1, β2, β3 and β4 of the linear model, and mdl.Rsquared.Ordinary is used to obtain the coefficient of determination R of the linear model. 2 , the coefficient of determination is calculated by the following formulas (4) and (5): Among them, fc represents the model prediction value obtained using the fitted regression coefficient, and FC represents the actual observed value. Representing FC i The average value, m represents the brain region number, R 2 The larger the coefficient, the closer the brain structure-function relationship is.

Images