End-stage renal disease white matter function network clustering identification network construction method and system

Abstract

This invention belongs to the field of medical imaging and discloses a method and system for constructing a clustering identification network for white matter functional networks in end-stage renal disease. This invention obtains usable continuous functional magnetic resonance imaging signals of the brain by preprocessing resting-state functional magnetic resonance imaging data, and calculates the white matter voxel correlation matrix under the constraints of structural image segmentation and white matter masking. A stable white matter functional network is obtained through group-level clustering. This invention realizes the systematic construction and quantitative characterization of the white matter functional network from the voxel layer to the network layer, thereby overcoming the previous problem of difficulty in quantitatively assessing abnormalities in white matter functional networks.

Description

Technical Field

[0001] This invention belongs to the field of medical imaging, specifically relating to a method and system for constructing a clustering and identification network for white matter functional networks in end-stage renal disease. Background Technology

[0002] Patients with end-stage renal disease (ESRD) often experience multiple metabolic and internal environment disturbances, and have a high incidence of cognitive impairment, although the neuroimaging mechanisms are not fully understood. Previous studies have focused on changes in gray matter structure and function, suggesting that ESRD can cause brain network abnormalities and is associated with cognitive decline; however, white matter (WM) has long been considered the main structural pathway for information transmission, and whether it has measurable functional activity and network organization in the resting state, and whether ESRD leads to abnormalities in white matter functional networks, remains unclear.

[0003] From a pathophysiological perspective, ESRD patients are in a long-term uremia state. ESRD-specific clinical risk factors such as uremic toxin accumulation, calcium and phosphorus homeostasis imbalance, and anemia can affect the central nervous system through chronic inflammation, endothelial dysfunction, abnormal microvascular perfusion, and glial reactions. White matter, due to its relatively fragile blood supply and sensitivity to ischemia, hypoxia, and metabolic disorders, is considered one of the brain regions most susceptible to involvement. Although structural imaging studies have reported that ESRD-related white matter damage (such as abnormal diffusion parameters and increased white matter lesion burden) is associated with cognitive impairment, this evidence mainly reflects "structural damage" and is insufficient to explain the functional deficits in information integration, executive control, attention, and memory in ESRD patients.

[0004] Therefore, there is an urgent need to systematically assess, from the perspective of white matter functional networks, whether ESRD patients have white matter functional abnormalities, whether the interactions between these networks are disordered, and the association between these changes and cognitive impairment and ESRD-related clinical risk factors. This would provide a basis for identifying potential clinical phenotypes and discovering quantifiable neuroimaging biomarkers. However, current systems lack a systematic and quantifiable assessment framework for white matter functional networks and their inter-network interactions. Summary of the Invention

[0005] The purpose of this invention is to overcome the problem of being unable to quantify the abnormalities in white matter functional networks in ESRD patients, and to provide a method and system for clustering and identifying white matter functional networks in end-stage renal disease.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease, comprising the following steps: Acquire T1-weighted structural images and resting-state functional magnetic resonance imaging data of patients with end-stage renal disease; Structural image segmentation was performed on the T1-weighted structural image to obtain the segmentation results, resulting in functional images of white matter and gray matter. Preprocessing of resting-state functional magnetic resonance imaging data yields the required continuous functional magnetic resonance brain signals. Construct a white matter mask based on functional images of the brain white matter; Based on resting-state functional magnetic resonance imaging (fMRI) signals, continuous brain fMRI signal time series were obtained. The correlation between voxel time series was calculated within the brain white matter mask to obtain individual brain white matter voxel correlation matrices. After processing the individual brain white matter voxel correlation matrices, group-level matrices were obtained. K-means clustering was performed on the group-level matrices, and brain white matter functional networks were selected. A functional connectivity matrix of brain white matter was constructed using functional connectivity analysis. Several gray matter regions were preset and time series of white matter functional networks were extracted. The correlation matrix between the white matter network and gray matter regions was obtained to obtain the covariance relationship between white matter networks. Bivariate coefficient Granger causality analysis was used to quantify the sign-direction relationship of the brain white matter functional network and to correct for the region-specific hemodynamic delay effect of the brain white matter functional network. Based on the white matter functional connectivity matrix, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks, a corresponding classifier is constructed to obtain a clustering and recognition network for white matter functional networks in end-stage renal disease.

[0007] A further improvement of the present invention is that when performing structural image segmentation on the T1-weighted structural image to obtain the segmentation results, and obtaining the brain white matter functional image and the brain gray matter functional image, the statistical parameter mapping software 12 is used for segmentation.

[0008] A further improvement of this invention lies in the following method for preprocessing resting-state functional magnetic resonance imaging (fMRI) data to obtain the desired continuous brain fMRI signals: Acquire resting-state functional magnetic resonance imaging data and remove imbalanced data; Time-level correction is performed on the remaining time point data, and head motion correction is performed based on the average functional image. The functional image is then registered to the corresponding structural image. Remove functional images whose head motion data exceeds a preset threshold; Detrending and bandpass filtering were performed on the functional images to obtain the desired continuous functional magnetic resonance imaging (fMRI) signal of the brain.

[0009] A further improvement of this invention lies in the following specific method for constructing a white matter mask based on brain white matter functional images: Acquire functional images of brain white matter, register the brain white matter functional images to the functional image space, and construct a brain white matter mask; The voxels that are collectively covered by research subjects exceeding a preset threshold are defined as the group-level brain white matter mask range; The deep brain structures within the group-level white matter mask range are removed to obtain the final white matter mask.

[0010] A further improvement of this invention lies in the following: based on resting-state functional magnetic resonance imaging (fMRI) signals, the correlation between pairwise time series of voxels is calculated within a white matter mask to obtain an individual white matter voxel correlation matrix. After processing the individual white matter voxel correlation matrix, a group-level matrix is ​​obtained. K-means clustering is then performed on the group-level matrix, and the specific method for selecting the white matter functional network is as follows: A time series of continuous functional magnetic resonance imaging (fMRI) signals of the brain obtained based on resting-state fMRI signals; Pearson correlation coefficients of pairwise time series of voxels are calculated within the white matter mask to obtain the voxel correlation matrix of individual white matter. An interleaved grid strategy was used to resample the brain white matter mask into 4245 nodes. The group-level average correlation matrix was obtained by averaging the voxel correlation matrix of individual brain white matter, and K-means clustering algorithm was used to perform cluster analysis on the group-level matrix. Clustering stability was evaluated by four-fold cross-validation. The connection matrix was randomly divided into four subsets and clustered separately. The similarity of the clustering results of different subsets was quantified by the adjacency matrix dice coefficients, and the optimal number of clusters was determined, thereby selecting the required brain white matter functional network. The selected brain white matter functional networks were spatially superimposed to verify their anatomical correspondence. After passing the verification, the brain white matter functional networks were divided into three layers: surface, middle and deep.

[0011] A further improvement of this invention lies in the following method for constructing the brain white matter functional connectivity matrix using functional connectivity analysis: Voxel functional magnetic resonance imaging (fMRI) signals from the white matter functional network were processed to obtain the average time series of the white matter functional network. Calculate the Pearson correlation coefficients of the average time series between each pair of brain white matter functional networks, and construct the brain white matter functional connectivity matrix.

[0012] A further improvement of this invention lies in the following method for obtaining the covariance relationship between white matter networks by pre-setting several gray matter regions and extracting the time series of white matter functional networks: Several gray matter regions in the brain white matter functional network are predefined, and the time series of the brain white matter functional network is extracted to calculate the correlation matrix between the brain white matter network and the gray matter regions. The covariance relationship between white matter networks was constructed based on the correlation matrix between white matter networks and gray matter regions.

[0013] A further improvement of this invention lies in the following method for quantifying the sign-direction relationship of the brain white matter functional network using bivariate coefficient Granger causality analysis, and for correcting the region-specific hemodynamic delay effect of the brain white matter functional network: We acquired continuous functional magnetic resonance imaging (fMRI) signals of the brain without bandpass filtering, and used rsHRF to perform HRF estimation and deconvolution on the continuous fMRI signals of the brain. The mean voxel time series of each brain white matter functional network was extracted as the network-level time series, and Granger causality strength between networks was calculated using RESTplus. The sign and direction of the influence between networks are characterized by regression coefficients, where positive and negative regression coefficients represent excitatory and inhibitory pathways, respectively, thus constructing a directed asymmetric regression coefficient matrix; The graph theory indices were used to calculate the ingress and egress indices of each white matter functional network to quantify the degree to which they were affected as target white matter functional networks.

[0014] A further improvement of this invention lies in constructing a corresponding classifier based on the white matter functional connectivity matrix, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks. The specific method for obtaining the white matter functional network clustering recognition network for end-stage renal disease is as follows: The correlation between changes in the functional connectivity matrix of white matter, covariance between white matter networks, and sign-direction relationships of white matter functional networks and clinical and neuropsychological indicators was obtained, and multiple comparisons were corrected for by false discovery rate. The required classifier model was built using Python to verify whether the functional connectivity matrix of white matter, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks can effectively distinguish between patients with end-stage renal disease and healthy controls. Five-fold cross-validation was used to evaluate the generalization ability of the classifier, and the statistical significance of the model performance was verified by several permutation tests. The relative contribution of a single feature to the classification result is quantified based on the discriminative weight of the feature, thereby forming a clustering identification network for white matter functional networks in end-stage renal disease.

[0015] Secondly, this invention provides a white matter functional network clustering and identification network model for end-stage renal disease, comprising: The data acquisition module is used to acquire T1-weighted structural images and resting-state functional magnetic resonance imaging data of patients with end-stage renal disease. The image segmentation module is used to perform structural image segmentation on the T1-weighted structural image to obtain the segmentation results, which yield brain white matter functional images and brain gray matter functional images. The data preprocessing module preprocesses the resting-state functional magnetic resonance imaging data to obtain the required continuous functional magnetic resonance brain signals. A mask construction module is used to construct a white matter mask based on functional images of the white matter. The network clustering module is used to calculate the correlation between pairwise time series of continuous brain functional magnetic resonance signals obtained based on resting-state functional magnetic resonance signals, obtain individual brain white matter voxel correlation matrices, process the individual brain white matter voxel correlation matrices to obtain group-level matrices, perform K-means clustering on the group-level matrices, and select brain white matter functional networks. The FC analysis module is used to construct the brain white matter functional connectivity matrix using functional connectivity analysis. The FCC analysis module is used to preset several gray matter regions and extract the time series of white matter functional networks to obtain the correlation matrix between the brain white matter network and gray matter regions, and to obtain the covariance relationship between white matter networks. The cGCA analysis module is used to quantify the sign-direction relationship of the brain white matter functional network using bivariate coefficient Granger causality analysis and to correct for the region-specific hemodynamic delay effect of the brain white matter functional network. The model building module is used to construct corresponding classifiers based on the white matter functional connectivity matrix, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks, thereby obtaining a clustering and recognition network for white matter functional networks in end-stage renal disease.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention obtains usable continuous functional magnetic resonance imaging (fMRI) signals of the brain through preprocessing of resting-state fMRI data, calculates the white matter voxel correlation matrix under structural image segmentation and white matter masking constraints, and obtains a stable white matter functional network through group-level clustering. This achieves the systematic construction and quantitative characterization of the white matter functional network from the voxel layer to the network layer, thereby overcoming the previous problem of difficulty in quantitatively assessing abnormalities in the white matter functional network. This invention characterizes the synchronicity changes between various white matter functional networks based on functional connectivity matrices, directly reflecting the degree of disorder in the overall coordinated activity of white matter networks. By pre-setting gray matter regions and calculating the correlation matrix between white matter networks and gray matter regions, this invention obtains the covariant relationships between white matter networks. This allows white matter network abnormalities to not only manifest in internal white matter connectivity but also to obtain a more stable high-dimensional phenotypic characterization through changes in coupling patterns with multiple gray matter regions, improving the sensitivity and robustness to abnormal structural patterns in white matter networks. Furthermore, this invention employs bivariate coefficient Granger causality analysis to obtain signed directional relationships after correcting for region-specific hemodynamic delays. This reveals the dominant direction of information transmission between white matter functional networks and potential excitation / inhibition pattern changes, overcoming the inadequacy of relying solely on correlation indicators to distinguish direction and influence patterns. Based on the aforementioned multidimensional network features, this invention constructs a classifier that can transform white matter functional network abnormalities into a discriminative model that can be used for individual identification. This improves the detection rate and discrimination of white matter functional network abnormalities in patients with end-stage renal disease, and supports clustering identification and phenotypic stratification of different abnormal patterns. It provides more interpretable and generalizable quantitative evidence for cognitive impairment risk assessment, clinical risk factor association analysis, and screening of potential neuroimaging biomarkers. Attached Figure Description

[0017] Figure 1 This is a flowchart of the present invention; Figure 2 This is a system diagram of the present invention; Figure 3 This diagram illustrates the cluster stability under different numbers of clusters. Figure 4 This diagram illustrates the differences in functional connectivity and functional covariance connectivity between brain white matter functional networks in patients with end-stage renal disease and healthy controls. (a) shows the differences in functional connectivity and functional covariance connectivity between 11 brain white matter networks when the white matter masking threshold is >70%, and (b) shows the differences in functional covariance connectivity between 11 brain white matter networks when the white matter masking threshold is >70%. Figure 5 This is a box plot showing the average frame shift values ​​between patients with end-stage renal disease and healthy controls. Detailed Implementation

[0018] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0019] Example 1: See Figure 1 A method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease includes the following steps: S1: Acquire T1-weighted structural images and resting-state functional magnetic resonance imaging data of patients with end-stage renal disease.

[0020] S2, perform structural image segmentation on the T1-weighted structural image to obtain the segmentation results, resulting in functional images of white matter and gray matter.

[0021] S3, preprocessing the resting-state functional magnetic resonance imaging data to obtain the required continuous functional magnetic resonance brain signals.

[0022] S4, construct a white matter mask based on the brain white matter functional image.

[0023] S5. Based on the resting-state functional magnetic resonance imaging (fMRI) signal, the time series of continuous brain fMRI signals are obtained. The correlation between voxel time series is calculated within the brain white matter mask to obtain the individual brain white matter voxel correlation matrix. After processing the individual brain white matter voxel correlation matrix, the group level matrix is ​​obtained. K-means clustering is performed on the group level matrix, and the brain white matter functional network is selected.

[0024] S6 uses functional connectivity analysis to construct a functional connectivity matrix of brain white matter.

[0025] S7. Preset several gray matter regions and extract the time series of white matter functional networks to obtain the correlation matrix between the brain white matter network and gray matter regions, and obtain the covariance relationship between white matter networks.

[0026] S8 uses bivariate coefficient Granger causality analysis to quantify the sign-direction relationship of the brain white matter functional network and corrects for the region-specific hemodynamic delay effect of the brain white matter functional network.

[0027] S9. Based on the white matter functional connectivity matrix, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks, a corresponding classifier is constructed to obtain the white matter functional network clustering recognition network for end-stage renal disease.

[0028] Example 2: See Figure 2 A clustering and identification network model for white matter functional networks in end-stage renal disease, including: The data acquisition module is used to acquire T1-weighted structural images and resting-state functional magnetic resonance imaging data of patients with end-stage renal disease.

[0029] The image segmentation module is used to perform structural image segmentation on the T1-weighted structural image to obtain the segmentation results, which yield brain white matter functional images and brain gray matter functional images.

[0030] The data preprocessing module preprocesses the resting-state functional magnetic resonance imaging (fMRI) data to obtain the required continuous functional magnetic resonance brain signals.

[0031] The mask construction module is used to construct a white matter mask based on functional images of the white matter.

[0032] The network clustering module is used to calculate the correlation between pairwise time series of continuous brain functional magnetic resonance signals obtained based on resting-state functional magnetic resonance signals within the brain white matter mask, thereby obtaining the individual brain white matter voxel correlation matrix. After processing the individual brain white matter voxel correlation matrix, the group-level matrix is ​​obtained. K-means clustering is performed on the group-level matrix, and the brain white matter functional network is selected.

[0033] The FC analysis module is used to construct the functional connectivity matrix of brain white matter using functional connectivity analysis.

[0034] The FCC analysis module is used to preset several gray matter regions and extract the time series of white matter functional networks to obtain the correlation matrix between the brain white matter network and gray matter regions, and to obtain the covariance relationship between white matter networks.

[0035] The cGCA analysis module is used to quantify the sign-direction relationship of the brain white matter functional network using bivariate coefficient Granger causality analysis and to correct for the region-specific hemodynamic delay effect of the brain white matter functional network.

[0036] The model building module is used to construct corresponding classifiers based on the white matter functional connectivity matrix, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks, thereby obtaining a clustering and recognition network for white matter functional networks in end-stage renal disease.

[0037] Example 3: This embodiment preprocesses T1-weighted structural images and resting-state functional magnetic resonance imaging data.

[0038] Using the New Segment algorithm in Statistical Parametric Mapping Software 12 (SPM12, version 7487; website: www.fil.ion.ucl.ac.uk / spm), T1-weighted structural images were segmented into gray matter, white matter, and cerebrospinal fluid components, and then normalized to the Montreal Institute for Neurology (MNI) standard template. The white matter functional image preprocessing steps were as follows: Individual T1-weighted segmented images were registered to the functional image space, and a white matter mask was constructed (threshold set to 0.5, which has been validated by previous studies). To avoid confusion between white matter and gray matter signals, spatial smoothing was performed within the white matter or gray matter mask range using a 4 mm full width at half maximum (FWHM) Gaussian kernel. The smoothed white matter functional images were normalized to the MNI standard template and resampled to 3×3×3 mm. 3 The voxels were used for subsequent analysis.

[0039] Resting-state functional magnetic resonance imaging (fMRI) data were preprocessed using the DPARSF Advanced Edition (version 5.4; website: www.restfmri.net), SPM12 software, and a self-written MATLAB script (https: / / mind.huji.ac.il / white-matter.aspx). Data from the first 10 time points were removed to eliminate the influence of initial signal imbalance and to allow subjects to adapt to scan noise. Temporal correction was performed on the remaining time-point data, followed by head motion correction based on the average functional images. The functional images were then registered to the corresponding structural images. Subjects with head motion displacement >3 mm or angle >3° were excluded; ultimately, 5 patients with end-stage renal disease were excluded due to excessive head motion, while no healthy controls were excluded. Detrending processing was performed on the functional images to eliminate linear drift; regression was used to remove interfering variables, including 24 head motion parameters and the average cerebrospinal fluid signal. White matter and whole-brain signals were preserved to avoid removing target signals. A time-point cleaning method based on head movement "peaks" (frame displacement value [FD] > 0.5 mm) was employed to further reduce the impact of head movements. Referring to previous studies, a data loss threshold of 10% was set, resulting in approximately 95% of the original data being retained (i.e., an average of about 5% of the data was discarded). Temporal bandpass filtering of 0.01–0.15 Hz was applied to the images to reduce interference from non-neuronal activity on blood oxygenation level-dependent signal fluctuations.

[0040] Example 4: This embodiment performs cluster analysis on resting-state functional magnetic resonance imaging data: A unified white matter mask at the group level was constructed based on T1-weighted image segmentation results. The range of the group-level white matter mask was defined as the voxels commonly covered by more than 60% of the subjects. Deep brain structures (including the thalamus, caudate nucleus, putamen, globus pallidus, and nucleus accumbens, with anatomical partitioning based on the Harvard-Oxford atlas) were removed to optimize the mask range. The final group-level white matter mask contained 17,064 voxels, which were registered to the functional image space and resampled to the same dimensions as the voxels in resting-state functional magnetic resonance imaging. The Pearson correlation coefficient between any two voxels within the white matter mask for each subject was calculated. To reduce computational load, an interleaved grid strategy was used to resample the mask to 4,245 nodes. The individual correlation matrices were averaged to obtain the group-level average correlation matrix. K-means clustering was used to perform cluster analysis on the group-level matrix, with the number of clusters ranging from 2 to 22. Cluster stability was evaluated using four-fold cross-validation: the connectivity matrix was randomly divided into four subsets (each containing 1061 voxels), and clustering was performed on each subset. The similarity of the clustering results across different subsets was quantified using adjacency matrix dice coefficients. The optimal number of clusters was determined based on the highest stability and fine-grained network discriminative power. Based on the dice coefficient results corresponding to each cluster size, the most stable and highest-resolution number of white matter functional networks was determined to be 11. (See [link to relevant documentation]). Figure 3 Referring to previous studies, each white matter functional network was registered to the Johns Hopkins University White Matter fasciculation atlas (containing 20 fasciculations) and the International Magnetoencephalography Consortium Diffusion Tensor Imaging Working Group atlas (containing 48 fasciculations), and their anatomical correspondence was verified by spatial overlay. Subsequently, the white matter functional network was divided into three layers: surface, middle and deep.

[0041] Example 5: This embodiment uses functional connectivity (FC) analysis to quantify the associations between functional networks in the brain's white matter.

[0042] The average time series of each predefined white matter functional network is extracted. Specifically, for each subject, the blood oxygen level-dependent signal values ​​of all voxels within a specific white matter functional network are summarized and their mean values ​​are calculated to obtain a time series reflecting the overall functional activity of that network. Subsequently, the Pearson correlation coefficients of the average time series between all white matter functional networks for each subject are calculated, resulting in an individual-level symmetric functional connectivity matrix. Each element in the matrix represents the strength of the linear temporal association between two target white matter functional networks. All correlation coefficients are subjected to a Fisher z-transform to satisfy the normal distribution assumption.

[0043] Example 6: This embodiment uses functional covariance connectivity (FCC) analysis to quantify the covariance relationships among brain white matter functional networks.

[0044] Ninety-six gray matter regions were defined based on the Harvard-Oxford gray matter map. A gray matter mask was applied to limit the analysis to gray matter voxels, and time series data of voxels within the intersection of the mask and the map were extracted. The Pearson correlation coefficient between each white matter functional network and each gray matter region was calculated, resulting in a K×96-dimensional correlation matrix (where K is the number of white matter functional networks). Finally, the functional covariance connectivity values ​​between all pairs of white matter functional networks were calculated, constructing a K×K-dimensional functional covariance connectivity matrix; the functional covariance connectivity values ​​were then subjected to a Fisher z-transform.

[0045] Example 7: In this embodiment, bivariate coefficient Granger causality analysis (cGCA) was used to quantify the directional influence of band signs among 11 white matter functional networks and to correct for the region-specific hemodynamic delay effect of white matter functional networks.

[0046] Given that combining filtering with Granger causality analysis may lead to erroneous causal inferences or omission of true causal relationships. 67 Subsequent analyses used preprocessed oxygenation level-dependent signals without bandpass filtering. To eliminate interference from the hemodynamic response function (HRF), the open-source software package rsHRF (version 2.4.0, download link: https: / / www.nitrc.org / projects / rshrf) was used. This software can estimate and deconvolve the hemodynamic response function based on the oxygenation level-dependent signal. The time series mean of all voxels within each functional network was extracted as the individual preprocessed oxygenation level-dependent time series. The Granger causality strength between networks was calculated using the Resting-State Functional Magnetic Resonance Imaging Data Processing Extension Toolkit (RESTplus, version 1.27; website: www.restfmri.net). Referring to previous studies, regression coefficients were used to characterize the sign and direction of pairwise influences between networks: positive / negative regression coefficients represent excitatory / inhibitory pathways, respectively (i.e., source network activity can predict subsequent increases / decreases in target network activity). An 11×11 dimensional directed asymmetric regression coefficient matrix was constructed for each study subject. The graph theory index analysis coefficients and Granger causality analysis results are used to quantify the strength of the incoming and outgoing influence of each network: Input strength: Represents the sum of the absolute values ​​of the regression coefficients of the incoming connections, reflecting the total strength of the influence of other networks (i.e. the degree to which it is significantly predicted by other networks) when the network is used as the "target network".

[0047] Outgoing strength: Represents the sum of the absolute values ​​of the regression coefficients of outgoing connections, reflecting the total strength of the influence of this network on other networks when it acts as a "source network" (i.e. the degree to which it significantly predicts other networks).

[0048] Example 8: To clarify the influence pattern of the brain white matter functional network within each group (i.e., the existence of statistically significant effective connections within each cohort), one-sample t-tests were performed on the pairwise mean Granger causality strength of the networks in the end-stage renal disease group and the healthy control group. The null hypothesis was that the mean influence strength was equal to 0 (i.e., no effective connections existed). After error discovery rate correction, p < 0.05 was considered statistically significant, and significant effective connections were preserved. Independent samples t-tests were used to compare the differences in directed connections between groups, with covariates including age, sex, mean frame shift value, and Euclidean distance between each pair of networks (corrected p < 0.05). Permutation tests (5000 iterations) were used to verify the significance of the difference distribution, with covariates consistent with the above (corrected p < 0.05).

[0049] Based on classical graph theory indices, the causal input and output strengths of each white matter functional network were analyzed to further quantify their overall impact on the network interaction model. The 11 white matter functional networks were divided into surface, intermediate, and deep networks. An independent samples t-test was used to compare the differences in input and output strengths of the three networks between the end-stage renal disease patient group and the healthy control group (corrected p < 0.05).

[0050] Pearson or Spearman correlation analysis was used to explore the correlation between changes in brain white matter functional network indices (functional connectivity, functional covariance connectivity, and Granger causality model) and clinical and neuropsychological indices. Multiple comparisons were adjusted for false discovery rate, and a p-value < 0.05 was defined as statistically significant.

[0051] Using the sklearn toolkit in Python 3.11, six commonly used classifier models were constructed, including k-nearest neighbors, logistic regression, support vector machine, adaptive boosting, categorical Naive Bayes, and random forest. The study verified whether inter-group differences in white matter functional network indices could effectively distinguish between patients with end-stage renal disease and healthy controls. A 5-fold cross-validation strategy was employed to ensure the reliability of the classifier's generalization ability assessment. To verify the statistical significance of the effectiveness of each classification model, a permutation test was performed (the label vectors of all subjects were randomly shuffled and iterated 5000 times); the same training and testing process was used after each permutation. Based on the results of 5000 permutations, the null distributions of each effectiveness index (accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve) were generated, and 95% confidence intervals were calculated. The p-value for each index was calculated as the proportion of iterations in which the permuted index value exceeded the original model index value out of the total number of iterations; p < 0.05 was defined as statistically significant. Finally, based on the discriminant weights of each feature, the relative contribution of a single feature to the classification results was quantified.

[0052] Example 9: This embodiment performs a verification analysis of the present invention, and the results confirm the robustness of the invention. This embodiment uses group level masks with thresholds of 70% and 80% for analysis, and the main results obtained are consistent with the analysis results using a mask with a threshold of 60% (see Tables 1-4). Figure 3 This indicates that even with stricter masking standards, the core findings of the study remain stable. Second, there were no significant differences in the functional network connectivity patterns of the brain white matter between male and female patients with end-stage renal disease (Supplementary Tables 10-11). Third, there were also no significant differences in the functional network connectivity patterns of the brain white matter among patients with different primary disease types of end-stage renal disease (Supplementary Tables 12-13). Fourth, there was no significant difference in the mean frame shift values ​​between the two groups (t=-1.12, p=0.904, see...). Figure 5 Finally, the results of 5-fold cross-validation are basically consistent with those of leave-one-out cross-validation, as shown in Tables 5 and 6.

[0053] Table 1. Intergroup differences in functional connectivity (FC) of white matter (WM) functional networks between patients with end-stage renal disease (ESRD) and healthy controls (HCs) (FDR corrected, white matter mask threshold >70%).

[0054] * indicates that the difference is statistically significant after FDR correction.

[0055] Table 2. Intergroup differences in functional connectivity (FC) of white matter (WM) functional networks between patients with end-stage renal disease (ESRD) and healthy controls (HCs) (FDR corrected, white matter mask threshold >80%)

[0056] * indicates that the difference is statistically significant after FDR correction.

[0057] Table 3. Between-group differences in functional covariance connectivity (FCC) of white matter (WM) network between patients with end-stage renal disease (ESRD) and healthy controls (HCs) (FDR corrected, white matter mask threshold >70%).

[0058] Table 4. Between-group differences in functional covariance connectivity (FCC) of white matter (WM) networks between ESRD patients and healthy controls (HCs) (FDR corrected, WM mask threshold >80%)

[0059] Table 5. Classification results based on white matter (WM) functional network indicators (5-fold cross-validation)

[0060] Table 6. Classification results based on white matter (WM) functional network indices (Leave-one-out cross-validation, LOOCV)

[0061] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a clustering and recognition network for white matter functional networks in end-stage renal disease, characterized in that, Includes the following steps: Acquire T1-weighted structural images and resting-state functional magnetic resonance imaging data of patients with end-stage renal disease; Structural image segmentation was performed on the T1-weighted structural image to obtain the segmentation results, resulting in functional images of white matter and gray matter. Preprocessing of resting-state functional magnetic resonance imaging data yields the required continuous functional magnetic resonance brain signals. Construct a white matter mask based on functional images of the brain white matter; Based on resting-state functional magnetic resonance imaging (fMRI) signals, continuous brain fMRI signal time series were obtained. The correlation between voxel time series was calculated within the brain white matter mask to obtain individual brain white matter voxel correlation matrices. After processing the individual brain white matter voxel correlation matrices, group-level matrices were obtained. K-means clustering was performed on the group-level matrices, and brain white matter functional networks were selected. A functional connectivity matrix of brain white matter was constructed using functional connectivity analysis. Several gray matter regions were preset and time series of white matter functional networks were extracted. The correlation matrix between the white matter network and gray matter regions was obtained to obtain the covariance relationship between white matter networks. Bivariate coefficient Granger causality analysis was used to quantify the sign-direction relationship of the brain white matter functional network and to correct for the region-specific hemodynamic delay effect of the brain white matter functional network. Based on the white matter functional connectivity matrix, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks, a corresponding classifier is constructed to obtain a clustering and recognition network for white matter functional networks in end-stage renal disease.

2. The method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease according to claim 1, characterized in that, Structural image segmentation was performed on the T1-weighted structural image to obtain the segmentation results. When obtaining the brain white matter functional image and brain gray matter functional image, the statistical parameter mapping software 12 was used for segmentation.

3. The method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease according to claim 1, characterized in that, The specific method for preprocessing resting-state functional magnetic resonance imaging (fMRI) data to obtain the required continuous functional magnetic resonance brain signals is as follows: Acquire resting-state functional magnetic resonance imaging data and remove imbalanced data; Time-level correction is performed on the remaining time point data, and head motion correction is performed based on the average functional image. The functional image is then registered to the corresponding structural image. Remove functional images whose head motion data exceeds a preset threshold; Detrending and bandpass filtering were performed on the functional images to obtain the desired continuous functional magnetic resonance imaging (fMRI) signal of the brain.

4. The method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease according to claim 1, characterized in that, The specific method for constructing a white matter mask based on functional images of the brain white matter is as follows: Acquire functional images of brain white matter, register the brain white matter functional images to the functional image space, and construct a brain white matter mask; The voxels that are collectively covered by research subjects exceeding a preset threshold are defined as the group-level brain white matter mask range; The deep brain structures within the group-level white matter mask range are removed to obtain the final white matter mask.

5. The method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease according to claim 1, characterized in that, Based on resting-state functional magnetic resonance imaging (fMRI) signals, continuous brain fMRI signal time series were obtained. The correlation between pairwise time series of voxels was calculated within a white matter mask to obtain individual white matter voxel correlation matrices. After processing these individual white matter voxel correlation matrices, group-level matrices were obtained. K-means clustering was then performed on these group-level matrices, and the specific method for selecting brain white matter functional networks is as follows: A time series of continuous functional magnetic resonance imaging (fMRI) signals of the brain obtained based on resting-state fMRI signals; Pearson correlation coefficients of pairwise time series of voxels are calculated within the white matter mask to obtain the voxel correlation matrix of individual white matter. An interleaved grid strategy was used to resample the brain white matter mask into 4245 nodes. The group-level average correlation matrix was obtained by averaging the voxel correlation matrix of individual brain white matter, and K-means clustering algorithm was used to perform cluster analysis on the group-level matrix. Clustering stability was evaluated by four-fold cross-validation. The connection matrix was randomly divided into four subsets and clustered separately. The similarity of the clustering results of different subsets was quantified by the adjacency matrix dice coefficients, and the optimal number of clusters was determined, thereby selecting the required brain white matter functional network. The selected brain white matter functional networks were spatially superimposed to verify their anatomical correspondence. After passing the verification, the brain white matter functional networks were divided into three layers: surface, middle and deep.

6. The method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease according to claim 1, characterized in that, The specific method for constructing the brain white matter functional connectivity matrix using functional connectivity analysis is as follows: Voxel functional magnetic resonance imaging (fMRI) signals from the white matter functional network were processed to obtain the average time series of the white matter functional network. Calculate the Pearson correlation coefficients of the average time series between each pair of brain white matter functional networks, and construct the brain white matter functional connectivity matrix.

7. The method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease according to claim 1, characterized in that, The specific method for obtaining the covariance relationship between white matter networks by pre-setting several gray matter regions and extracting time series data of white matter functional networks is as follows: Several gray matter regions in the brain white matter functional network are predefined, and the time series of the brain white matter functional network is extracted to calculate the correlation matrix between the brain white matter network and the gray matter regions. The covariance relationship between white matter networks was constructed based on the correlation matrix between white matter networks and gray matter regions.

8. The method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease according to claim 1, characterized in that, The specific method for quantifying the sign-direction relationship of the brain white matter functional network and correcting for the region-specific hemodynamic delay effect of the brain white matter functional network using bivariate coefficient Granger causality analysis is as follows: We acquired continuous functional magnetic resonance imaging (fMRI) signals of the brain without bandpass filtering, and used rsHRF to perform HRF estimation and deconvolution on the continuous fMRI signals of the brain. The mean voxel time series of each brain white matter functional network was extracted as the network-level time series, and Granger causality strength between networks was calculated using RESTplus. The sign and direction of the influence between networks are characterized by regression coefficients, where positive and negative regression coefficients represent excitatory and inhibitory pathways, respectively, thus constructing a directed asymmetric regression coefficient matrix; The graph theory indices were used to calculate the ingress and egress indices of each white matter functional network to quantify the degree to which they were affected as target white matter functional networks.

9. The method for constructing a clustering and identification network for white matter functional networks in end-stage renal disease according to claim 1, characterized in that, Based on the white matter functional connectivity matrix, the covariance between white matter networks, and the sign and direction relationships of white matter functional networks, a corresponding classifier is constructed. The specific method for obtaining the white matter functional network clustering recognition network for end-stage renal disease is as follows: The correlation between changes in the functional connectivity matrix of white matter, covariance between white matter networks, and sign-direction relationships of white matter functional networks and clinical and neuropsychological indicators was obtained, and multiple comparisons were corrected for by false discovery rate. The required classifier model was built using Python to verify whether the functional connectivity matrix of white matter, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks can effectively distinguish between patients with end-stage renal disease and healthy controls. Five-fold cross-validation was used to evaluate the generalization ability of the classifier, and the statistical significance of the model performance was verified by several permutation tests. The relative contribution of a single feature to the classification result is quantified based on the discriminative weight of the feature, thereby forming a clustering identification network for white matter functional networks in end-stage renal disease.

10. A clustering and recognition network model for white matter functional networks in end-stage renal disease, characterized in that, include: The data acquisition module is used to acquire T1-weighted structural images and resting-state functional magnetic resonance imaging data of patients with end-stage renal disease. The image segmentation module is used to perform structural image segmentation on the T1-weighted structural image to obtain the segmentation results, which yield brain white matter functional images and brain gray matter functional images. The data preprocessing module preprocesses the resting-state functional magnetic resonance imaging data to obtain the required continuous functional magnetic resonance brain signals. A mask construction module is used to construct a white matter mask based on functional images of the white matter. The network clustering module is used to calculate the correlation between pairwise time series of continuous brain functional magnetic resonance signals obtained based on resting-state functional magnetic resonance signals, obtain individual brain white matter voxel correlation matrices, process the individual brain white matter voxel correlation matrices to obtain group-level matrices, perform K-means clustering on the group-level matrices, and select brain white matter functional networks. The FC analysis module is used to construct the brain white matter functional connectivity matrix using functional connectivity analysis. The FCC analysis module is used to preset several gray matter regions and extract the time series of white matter functional networks to obtain the correlation matrix between the brain white matter network and gray matter regions, and to obtain the covariance relationship between white matter networks. The cGCA analysis module is used to quantify the sign-direction relationship of the brain white matter functional network using bivariate coefficient Granger causality analysis and to correct for the region-specific hemodynamic delay effect of the brain white matter functional network. The model building module is used to construct corresponding classifiers based on the white matter functional connectivity matrix, the covariance between white matter networks, and the sign-direction relationship of white matter functional networks, thereby obtaining a clustering and recognition network for white matter functional networks in end-stage renal disease.

Images