Brain connection coupling analysis method based on grey matter morphological similarity-white matter fiber bundles

Abstract

The invention discloses a brain connection coupling analysis method based on grey matter morphological similarity-white matter fiber bundles, relates to the technical field of neuropsychiatric imaging, and has the technical key points that the method can be used for analyzing grey matter-white matter brain connection group coupling modes of the whole brain level, the level in brain network communities and the level between the brain network communities; the integration characteristics of structural tissues of the brain on different scales can be understood; the reliability of statistical analysis is improved by adopting a spatial autocorrelation correction method; based on a specific machine learning fusion model, the clinical application value of the model is verified.

Description

Technical Field

[0001] The present invention relates to the field of neuropsychiatric imaging technology, and in particular to a brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles. Background Art

[0002] At present, there are two main technical routes for constructing brain structural connectome based on magnetic resonance imaging: white matter fiber tract tracing network (Tractography Network) based on diffusion tensor imaging and gray matter morphological similarity network (Morphometric Similarity Network) based on high-resolution structural imaging. Among them, the white matter fiber tract network constructed by fiber tract tracing mainly indicates the transport of biological molecules across brain regions, while the gray matter morphological similarity network indicates the developmental coordination and cell architecture similarity between brain regions. It has been found that about 35% to 40% of morphological covariation occurs between cortical regions directly connected by fiber tracts.

[0003] However, the existing technologies have the following major problems: (1) They are limited to exploring the relationship between brain structure and function, lack systematic analysis of the structural characteristics of different levels of the brain, lack a gray matter-white matter integrated analysis model, and are difficult to measure the relationship between gray matter morphology and white matter fiber bundles; (2) There is a lack of quantitative analysis methods for the dissociation phenomenon between the delayed morphological maturation of the cortical and subcortical gray matter regions and the accelerated development of fiber bundles in individuals with mental disorders; (3) The brain structure-function coupling analysis method has limited ability to explain the pathological mechanisms of mental disorders and is difficult to effectively support the early diagnosis of mental disorders.

[0004] To this end, the present invention aims to provide a brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to solve the above problems and provide a brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles. The method can analyze the coupling patterns at the whole brain level, the level within the brain network community and the level between brain network communities, which is helpful to understand the integration characteristics of structural organization at different scales of the brain; the spatial autocorrelation correction method is used to improve the reliability of statistical analysis; based on a specific machine learning fusion model, the clinical application value of the model is verified.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows:

[0007] The present invention provides a brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles, the method comprising the following steps:

[0008] S1. Construction of gray matter morphological similarity network: preprocessing of T1-weighted images; extraction of brain gray matter voxel-level volume maps;

[0009] S2, calculating the morphological similarity between brain regions based on the probability density distribution function and obtaining the corresponding matrix;

[0010] S3. Construction of white matter fiber bundle connection network: preprocessing of diffusion images; extraction of brain white matter anisotropy fraction map;

[0011] S4, infer the fiber direction of the diffusion image and calculate the connection probability between brain regions. At the same time, the fiber bundle direction is restricted by registration and segmentation to obtain the fiber bundle probability matrix;

[0012] S5. Gray matter-white matter morphological structural connection coupling analysis: correlation coupling analysis of gray matter-white matter brain network matrix at the whole brain, within-community and between-community levels

[0013] S6. Evaluate the statistical test of spatial autocorrelation correction: Generate a replacement map through spatial autocorrelation correction and perform a statistical significance test;

[0014] S7. Establish a diagnostic classification model based on brain coupling features: establish a nested cross-validation framework; perform feature screening and preprocessing based on the brain network community level; perform classifier training; evaluate and verify the model to identify high-contribution features.

[0015] Gray matter morphological similarity network (MSN): This part extracts the morphological features of gray matter regions by preprocessing, segmenting, registering, and spatially smoothing T1-weighted images. These processes ensure the spatial consistency of gray matter tissues between different individuals and obtain individual gray matter volume maps. The resulting morphological similarity network lays the foundation for coupling analysis.

[0016] White matter fiber bundle connection network (FCN): Through diffusion image preprocessing, fiber direction estimation and fiber bundle tracking, the individual white matter anisotropy fraction map was extracted and the white matter fiber connection network between brain regions was constructed. FCN reflects the structural connection strength of different brain regions through white matter fibers, which provides a computational basis for subsequent coupling analysis.

[0017] MSN and FCN provide individualized data from the perspectives of gray matter morphology and white matter connectivity, respectively. Through Spearman rank correlation analysis, the coupling relationship between these two brain network connection groups can be explored and the mutual influence between gray matter morphology and white matter connectivity can be evaluated. Coupling analysis reveals the coordination between gray matter morphology and white matter connectivity by calculating the correlation between MSN and FCN. For example, whether regions with similar gray matter morphology also show similar strength in white matter connectivity, and vice versa.

[0018] In the S1-S2 step, the T1 weighted image preprocessing is as follows: the T1 weighted image is processed by SPM8 software, and the image is segmented into gray matter, white matter, and cerebrospinal fluid by a unified segmentation model; then the DARTEL method is used for high-precision registration, and the image is mapped to the standard space (MNI space). Nonlinear modulation is applied to maintain gray matter registration and obtain gray matter volume maps, and then spatial smoothing is used to improve the signal-to-noise ratio of the data, thereby reducing the difference in anatomical structure;

[0019] Calculation of morphological similarity between regions: Kullback-Leibler divergence was used to calculate the morphological similarity of different brain regions; the gray matter volume distribution of each brain region was calculated by the kernel density estimation method, and then the divergence formula was used to calculate the similarity between two brain regions to generate a similarity matrix;

[0020] The kernel density formula is: p(x) = (1 / nh) × ∑K((x-xi) / h), where n is the number of samples, h is the bandwidth parameter, K is the Gaussian sum function, and xi is the observed value;

[0021] The divergence formula is: KLS(P||Q)=∑P(x)log(P(x) / Q(x)), where P and Q are the probability density functions of the two brain regions and x is the sampling point.

[0022] In the steps S3-S4, diffusion image preprocessing: spatial registration of diffusion images is performed by FLIRT to improve image consistency; eddy current and motion correction is performed by FSL tools to reduce image distortion; EPI image distortion is corrected based on field maps by FUGUE tools; and a brain white matter anisotropy fraction map is obtained;

[0023] Probabilistic fiber tract tracing: Use FSLbedpostx to estimate the fiber direction of the diffusion image and infer the main fiber direction of each voxel. Use FSLprobtrackx to perform fiber tract tracing and calculate the connection probability between brain regions. Use registration and segmentation to limit the direction of fiber tracts and avoid crossing non-white matter areas.

[0024] In the step S5, the gray matter-white matter brain connection matrix (MSN-FCN matrix) is subjected to correlation coupling analysis: Spearman rank correlation is used to perform correlation analysis on the gray matter matrix and the white matter matrix, and by calculating the coupling strength (ρ value) between brain regions, the relationship between gray matter morphology and white matter connection can be evaluated. If the coupling strength is a positive value, it means that the two are tending to be consistent, otherwise it means that there is an inverse relationship between the two, and the absolute value of the coupling represents the coupling strength; according to the functional network of the brain, the coupling strength of a certain brain region within the same network community and between different network communities is calculated to evaluate the coordination of the structural and functional connections of different brain regions at the network community level; in the step S6, the spatial autocorrelation correction statistical test value is evaluated: multiple alternative distribution maps are generated through spatial autocorrelation correction, and compared with the original values ​​(p value) to evaluate whether the coupling relationship is statistically significant.

[0025] The i-th row vector of the gray matter matrix is: MSN_i=[m_i1,m_i2,…,m_in], and the i-th row vector of the white matter matrix is: FCN_i=[f_i1,f_i2,…,f_in], where n is the total number of brain region segmentations of the preset brain template;

[0026] The brain region coupling strength formula is: ρ = 1-(6×∑d 2 ) / (n×(n 2 -1)), where d is the order difference and n is the number of paired samples.

[0027] In step S7, the nested cross-validation framework: the model is trained and evaluated using repeated nested cross-validation, the inner cross-validation is used to optimize the hyperparameters, and the outer cross-validation is used to evaluate the model performance, and the influence of accidental factors is reduced through multiple repetitions;

[0028] Feature selection and preprocessing: The extracted features are standardized, scaled and sorted based on the group of the network community model. The F-score is used for feature selection to select the most predictive features. The significance of the features is evaluated by permutation test (n=1000) to ensure that the features input into the model are reliable. The feature selection step ensures the efficiency of the classification model. Selecting the most important coupling features helps improve the performance of the model, and removing irrelevant features reduces the risk of overfitting.

[0029] Classifier training: L2 regularized support vector machine (SVM) classifier is used; the contribution of each feature is evaluated through the weight vector during the training process. By training known coupled features, new sample data can be predicted or classified, making this analysis process have practical diagnostic application value.

[0030] Model evaluation and validation: Use balanced accuracy BAC = (sensitivity + specificity) / 2 to evaluate model performance, balance sensitivity and specificity, and ensure that the classification results have good overall accuracy; use cross-validation ratio CVR = median(w) / std(w) to evaluate the importance of each feature in the model, and select significant features for further analysis, where w is the weight vector obtained in all cross-validation processes; use independent external samples to validate the model.

[0031] Independent external sample validation: By testing the model on an independent validation set, we can ensure that the established model has good generalization ability. After validation, the model can be applied to new patients or experimental data. External sample validation makes the established model more robust and practical, thus providing a reliable diagnostic tool for clinical or scientific research.

[0032] Data preprocessing (processing of T1-weighted images and diffusion images) provides clear basic data for subsequent morphological similarity and white matter fiber bundle connection analysis.

[0033] Gray matter morphological similarity and white matter fiber connection network analyze brain regions from the perspective of morphology and connectivity, respectively, providing two perspectives on brain structure.

[0034] Through coupling analysis, MSN and FCN data are combined to explore the synergistic or antagonistic relationship between morphology and function, providing effective features for further classification analysis.

[0035] The construction of the classification model involves steps such as feature selection, model training and evaluation, and uses the features obtained from MSN and FCN to perform disease diagnosis or other prognosis monitoring tasks.

[0036] Compared with the prior art, this solution has the following beneficial effects:

[0037] The brain connection coupling analysis method of the present invention can analyze the gray matter-white matter brain connection group coupling patterns at the whole brain, within brain network communities and between communities, which helps to understand the integration characteristics of structural organizations at different scales of the brain; the spatial autocorrelation correction method is used to improve the reliability of statistical analysis; based on a specific machine learning fusion model, the potential clinical application value of the model is verified; the explanatory power of the pathological mechanism of neuropsychiatric disorders is enhanced, and the early diagnosis and evaluation of neuropsychiatric disorders are effectively supported. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flowchart of a brain connection coupling analysis method in an embodiment of the present invention;

[0039] Figure 2 4 is a flow chart of a method for analyzing brain connection coupling in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solution of the present invention will be further described in detail below in conjunction with the embodiments of the present invention and the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0041] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will be described in detail below in conjunction with the embodiments.

[0042] Example:

[0043] The solution provided by the embodiment of the present invention is as described in the above invention content, and provides a brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles, comprising the following steps:

[0044] 1. Construction of the Gray Matter Morphometric Similarity Network (MSN)

[0045] 1.1 T1-weighted image preprocessing

[0046] Use SPM8 software to process T1-weighted images. The specific steps are as follows:

[0047] 1) Unified Segmentation Model: The individual T1 image is segmented into three types of tissues: gray matter, white matter, and cerebrospinal fluid. The segmentation is performed based on the tissue probability map and spatial prior information, and the normalized tissue probability map is output.

[0048] 2) DARTEL method normalization processing: The Diffeomorphic Anatomical Registration Through Exponentiated LieAlgebra method is used to accurately align the gray matter map to the MNI standard space, maintain the relative differences in local voxel values, and generate flow field deformation parameters (Flow Fields) for spatial transformation.

[0049] 3) Nonlinear Modulation: Nonlinear modulation is performed on the normalized gray matter image to compensate for the volume change during the spatial normalization process and keep the total amount of regional gray matter unchanged. The modulated gray matter volume is equal to the volume in the original image multiplied by the Jacobian determinant.

[0050] 4) Spatial smoothing (Smoothing) uses an 8mm full width at half maximum (FWHM) Gaussian kernel for smoothing to increase the normality of the data, reduce the differences in anatomical structures between individuals, and improve the signal-to-noise ratio.

[0051] 1.2 Calculation of morphological similarity between regions

[0052] The Kullback-Leibler divergence similarity was used to calculate the morphological covariation relationship between brain regions:

[0053] 1) Kernel density estimation: Kernel density estimation is performed on the gray matter volume distribution of each brain region. The Gaussian kernel function is used for density estimation. The probability density function is expressed as: p(x) = (1 / nh) × ∑K((x-xi) / h).

[0054] Where n is the number of samples, h is the bandwidth parameter, K is the Gaussian sum function, and xi is the observed value.

[0055] 2) Kullback-Leibler divergence similarity calculation: Generate 2 7 Evenly distributed sampling points are used to calculate the divergence similarity value between two brain regions: KLS(P||Q)=∑P(x)log(P(x) / Q(x)). P and Q are the probability density functions of the two brain regions, and x is the sampling point. The closer the similarity is to 1, the higher the similarity.

[0056] 3) Similarity matrix construction: The existing gray matter segmentation template can be used to calculate the Kullback-Leibler divergence similarity between all brain region pairs and construct a matrix.

[0057] 2. Construction of the White Matter Fiber Connection Network (FCN)

[0058] 2.1 Diffusion image preprocessing

[0059] 1) Spatial registration: FLIRT (FMRIB's Linear Image Registration Tool) was used for linear registration. The binary mask of the standard anisotropy score map was registered to the diffusion reference image (b = 0) of each subject. The registration parameters were set as follows: the degrees of freedom was 12, the maximum number of iterations was 500, and the normalized mutual information was used as the similarity measure.

[0060] 2) Eddy current and motion correction: FSL eddy tool was used to correct the influence of eddy current and head motion on diffusion-weighted images using Gaussian process model, and the diffusion gradient vector was adjusted according to the head motion correction.

[0061] 3) Field map and distortion correction: FUGUE (FMRIB's Utility for Geometrically UnwarpingEPIs) tool was used to perform EPI distortion correction based on the field map to obtain individual-specific anisotropy fractional maps.

[0062] 2.2 Probabilistic fiber tract tracing

[0063] 1) Fiber direction estimation (FSLbedpostx): The ball-and-stick diffusion model was used to estimate the main fiber direction distribution and diffusion parameter distribution using Markov chain Monte Carlo sampling, with a maximum of three fiber directions allowed per voxel. The ARD (Automatic Relevance Determination) prior was used and 1000 iterations were set.

[0064] 2) Fiber tract tracking (FSL probtrackx): Tracking parameters were set as follows: 5000 fibers were emitted per voxel, the step size was 0.5 mm, the curvature threshold was 0.2, and the minimum fiber length was 20 mm; the connection probability was calculated for each pair of brain regions: the number of fibers from seed region A to target region B was divided by the total number of emitted fibers to obtain the connection probability; connection strength was normalized: the average volume of the two brain regions was used for normalization to eliminate the influence of brain region size.

[0065] 3) Atlas registration and segmentation: The pre-selected brain segmentation atlas is registered to the T1 image, and the registered atlas is mapped to the diffusion space. The white matter segmentation is used as waypoints constraint to ensure that the fiber bundles pass through the white matter area and avoid passing through the cerebrospinal fluid or gray matter.

[0066] 4) Connection matrix construction: The existing gray matter segmentation template can be used to calculate the probability tracking structural connection network matrix between all brain region pairs. The matrix elements represent the fiber bundle connection probability (volume-normalized connection strength).

[0067] 3. Gray matter-white matter morphological structural connectivity coupling analysis

[0068] 3.1 MSN-FCN matrix correlation coupling analysis (calculating Spearman rank correlation)

[0069] 1) Based on the construction of the gray matter morphological similarity network and the white matter fiber bundle connection network, for each brain region i, the following is obtained: MSN matrix i-th row vector: MSN_i = [m_i1, m_i2, ..., m_in], excluding self-connected m_ii; FCN matrix i-th row vector: FCN_i = [f_i1, f_i2, ..., f_in], excluding self-connected f_ii. Where n is the total number of brain region segmentations of the preset brain template.

[0070] 2) Brain region coupling strength ρ = 1-(6×Σd 2 ) / (n×(n 2 -1)). Where d is the rank difference and n is the number of paired samples (self-connection needs to be removed). The ρ value range is [-1,1]. A positive value indicates that the structural connection and morphological covariation patterns are consistent, and a negative value indicates that the two patterns are opposite. The absolute value indicates the coupling strength.

[0071] 3.2 Coupling Analysis of Network Community

[0072] 1) Based on the template’s own prior large-scale brain network division scheme, the brain network affiliation information of each brain region can be obtained, such as the default mode network, central executive network, sensorimotor network and other sub-networks.

[0073] 2) Calculation of coupling within the network community: For brain region i in network k: Within_Coupling_i = corr(MSN_i(k), FCN_i(k)). Where MSN_i(k) is the MSN value of region i and other regions in the same network, FCN_i(k) is the corresponding FCN value, and corr() is the Spearman correlation coefficient.

[0074] 3) Calculation of coupling between network communities: For brain region i in network k: Between_Coupling_i = corr(MSN_i(~k),FCN_i(~k)). Where MCN_i(~k) is the MCN value between region i and other network regions, and FCN_i(~k) is the corresponding FCN value.

[0075] 3.3 Statistical significance assessment based on spatial autocorrelation correction

[0076] 1) Substitution map generation: The spatial autocorrelation structure of the original data is maintained, and 5000 spatial substitution maps are generated using the BrainSMASH toolbox. The correlation coefficient r or t statistic is calculated for each substitution map.

[0077] 2) Significance test: Calculate the original statistic T_ori, and count the proportion of substitution maps that exceed T_ori. P value calculation: p = (Σ(|T_surr|>=|T_ori|)) / 5000. T_surr is the statistic of the substitution map.

[0078] 4 Diagnostic classification model based on brain coupling characteristics

[0079] 4.1 Establishing a Nested Cross-Validation Framework

[0080] A nested cross-validation scheme was used, with 10-fold cross-validation for the outer layer and 5-fold cross-validation for the inner layer. The inner cross-validation was used to optimize the model hyperparameters, and the outer cross-validation was used to evaluate the model performance. To reduce the sampling effect, the outer cross-validation was randomly repeated 5 times.

[0081] 4.2 Feature Selection and Preprocessing

[0082] Taking the group of the network community model as the unit, the gray matter-white matter morphological structure connection coupling characteristics are evaluated and screened:

[0083] 1) Scaling and standardization preprocessing of input features;

[0084] 2) Sort the features based on the F-score and select the top 15%, 45%, or 75% features;

[0085] 3) The features of the network community-specific model that reached a significant level (p<0.05) in the permutation test (n=1000) were used as input to the fusion machine learning model.

[0086] 4.3 Classifier Training

[0087] The L2 loss support vector classification algorithm (LIBLINEAR) with L2 regularization is used to establish the classifier, and its optimization objective function is: min(w)1 / 2wTw+CΣmax(0,1-yi<w,xi> ) 2 . Where w is the weight vector, C is the regularization parameter, yi is the category label, and xi is the feature vector.

[0088] 4.4 Model Evaluation

[0089] 1) Use balanced accuracy (BAC) as the model evaluation indicator: BAC = (sensitivity + specificity) / 2.

[0090] 2) Calculate the cross validation ratio (CVR) to evaluate the importance of brain image features: CVR = median (w) / std (w), where w is the weight vector obtained in all cross validation processes.

[0091] 3) Signed-based consistency mapping was used to judge feature significance, and features that met the following conditions were selected as significant predictors: |CVR|>3 and pFDR<0.05.

[0092] 4.5 Independent external sample verification:

[0093] 1) Correct the validation sample characteristics: x′val=xval-(μdis-μval). x′val is the corrected validation sample characteristic value, xval is the original validation sample characteristic value, μdis is the discovery sample mean, and μval is the validation sample mean.

[0094] 2) Use the trained classifier to predict the corrected validation samples and calculate the diagnostic accuracy of the pre-trained model for external samples.

[0095] The above specific embodiments are merely explanations of the present invention and are not limitations of the present invention. After reading this specification, those skilled in the art may make modifications to the embodiments without any creative contribution as needed. However, such modifications are protected by the patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles, characterized by: The method comprises the following steps: S1. Construction of gray matter morphological similarity network: preprocessing of T1-weighted images; extraction of brain gray matter voxel-level volume maps; S2, calculate the morphological similarity between brain regions based on the probability density distribution function and obtain the gray matter matrix; S3. Construction of white matter fiber bundle connection network: preprocessing of diffusion images; extraction of brain white matter anisotropy fraction map; S4, infer the fiber direction of the diffusion image and calculate the connection probability between brain regions. At the same time, the fiber bundle direction is restricted by registration and segmentation to obtain the fiber bundle probability matrix; S5. Gray matter-white matter morphological structural connection coupling analysis: correlation coupling analysis of gray matter-white matter brain network matrix at the whole brain, within-community and between-community levels S6. Evaluate the statistical test of spatial autocorrelation correction: Generate a replacement map through spatial autocorrelation correction and perform a statistical significance test; S7. Establish a diagnostic classification model based on brain coupling features: Establish a nested cross-validation framework; Perform feature screening and preprocessing based on the brain network community level; perform classifier training; evaluate and verify the model to identify high-contribution features.

2. The brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles as claimed in claim 1, characterized in that: In the step S1, the T1 weighted image preprocessing is as follows: the T1 weighted image is processed by SPM8 software, and the image is segmented into gray matter, white matter, and cerebrospinal fluid by a unified segmentation model; the DARTEL method is used for high-precision registration, and the image is mapped to a standard space; nonlinear modulation is applied, and the signal-to-noise ratio of the data is improved by spatial smoothing; In the step S2, the morphological similarity between regions is calculated by using Kullback-Leibler divergence to calculate the morphological similarity of different brain regions; the gray matter volume distribution of each brain region is calculated by the kernel density estimation method, and then the divergence formula is used to calculate the similarity between two brain regions to generate a similarity matrix; The kernel density formula is: p(x) = (1 / nh) × ∑K((x-xi) / h), where n is the number of samples, h is the bandwidth parameter, K is the Gaussian sum function, and xi is the observed value; The divergence formula is: KLS(PQ)=∑P(x)log(P(x) / Q(x)), where P and Q are the probability density functions of the two brain regions and x is the sampling point.

3. The brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles according to claim 1, characterized in that: In the step S3, the diffusion image is preprocessed by: spatially registering the diffusion image using FLIRT; performing eddy current and motion correction using the FSL tool; and correcting the EPI image distortion based on the field map using the FUGUE tool to obtain a brain white matter anisotropy fraction map. In the S4 step, probabilistic fiber bundle tracing: fiber direction estimation is performed on the diffusion image by FSLbedpostx to infer the main fiber direction of each voxel; FSLprobtrackx was used to track fiber bundles and calculate the connection probability between brain regions; the direction of fiber bundles was restricted by registration and segmentation.

4. The brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles according to claim 1, characterized in that: In the step S5, the gray matter-white matter brain network matrix is ​​subjected to correlation coupling analysis: Spearman rank correlation is used to perform correlation analysis on the gray matter matrix and the white matter matrix, and the relationship between the gray matter morphology and the white matter connection is evaluated by calculating the coupling strength between the brain regions. If the coupling strength is a positive value, it means that the two tend to be consistent, otherwise it means that there is an inverse relationship between the two. The absolute value of the coupling represents the coupling strength; according to the functional network of the brain, the coupling strength of a certain brain region within the same network community and between different network communities is calculated to evaluate the coordination of the structural and functional connections of different brain regions at the network community level; In the step S6, the spatial autocorrelation correction statistical test quantity is evaluated: a plurality of alternative distribution maps are generated by spatial autocorrelation correction and compared with the original values ​​to evaluate whether the coupling relationship is statistically significant.

5. The brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles according to claim 1, characterized in that: The i-th row vector of the gray matter matrix is: MSN_i=[m_i1,m_i2,…,m_in], and the i-th row vector of the white matter matrix is: FCN_i=[f_i1,f_i2,…,f_in], where n is the total number of brain region segmentations of the preset brain template; The brain region coupling strength formula is: ρ = 1-(6×∑d 2 ) / (n×(n 2 -1)), where d is the order difference and n is the number of paired samples.

6. The brain connection coupling analysis method based on gray matter morphological similarity-white matter fiber bundles according to claim 1, characterized in that: In step S7, the nested cross-validation framework: the model is trained and evaluated using repeated nested cross-validation, the inner cross-validation is used to optimize the hyperparameters, and the outer cross-validation is used to evaluate the model performance, and the influence of accidental factors is reduced through multiple repetitions; Feature selection and preprocessing: Taking the group of the network community model as the unit, the extracted features are standardized, scaled and sorted, and the F-score is used for feature selection to screen out the features with the most predictive ability; Classifier training: L2 regularized support vector machine classifier is used, and the optimization objective function is: min(w)1 / 2wTw+C∑max(0,1-yi<w,xi> ) 2 , where w is the weight vector, C is the regularization parameter, yi is the category label, and xi is the feature vector; the contribution of each feature is evaluated by the weight vector during training; Model evaluation and validation: using balanced accuracy BAC = (sensitivity + specificity) / 2 to evaluate model performance, balance sensitivity and specificity, and ensure that the classification results have good overall accuracy; through the cross-validation ratio CVR = median (w) / std(w) evaluates the importance of each feature in the model and selects significant features for further analysis, where w is the weight vector obtained in all cross-validation processes; the model is validated using independent external samples, and the validation formula is: x′val=xval-(μdis-μval), where x′val is the corrected validation sample eigenvalue, xval is the original validation sample eigenvalue, μdis is the discovery sample mean, and μval is the validation sample mean.

Images