Construction method of brain structural network weight based on quantitative characteristics of microstructure

Abstract

The application discloses a brain structure network weight construction method based on microstructure quantitative characteristics. The method comprises the following steps: firstly, a brain structure fiber bundle sampling point template is obtained by processing a MNI space brain fiber bundle template; then, a population data set is acquired by processing the brain structure fiber bundle sampling point template according to an improved multi-modal magnetic resonance imaging method; an attention variational autoencoder model is constructed; the population data set is input into the attention variational autoencoder model for training; finally, the vector value of the brain structure feature map of a to-be-tested individual is input into the trained attention variational autoencoder model for processing, and the processing result is directly used as the weight value of an edge in the brain structure network. The application overcomes the information imbalance caused by the fiber bundle length difference, provides multiple different microstructure information of a multi-modal brain, and realizes providing a new direction for the brain structure network weight research field.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a method for constructing brain structural network weights based on quantitative microstructure features. Background Technology

[0002] The brain is the control center of the human body; its complex structure and functions interact to determine an individual's cognition, behavior, and emotions. To study how the brain processes information and its functional performance in different states, the construction and analysis of brain structural networks have become particularly important. A brain structural network refers to a complex network formed by neural connections (edges) between different regions of the brain. Nodes typically represent anatomical or functional areas of the brain, while edges represent the connections between different regions. The weights of network edges are usually used to measure the strength of connections or the information transmission capacity between different regions.

[0003] In the construction of brain structural networks, the calculation of weights is crucial. Currently, most weight calculations for brain structural networks rely on macroscopic features, such as the number of streamlines and fiber density in white matter fiber tracts. These methods describe the connection strength between different regions by analyzing the white matter fiber tracts of the brain. However, these macroscopic feature-based calculation methods often neglect information about microstructures and cannot fully reflect the true connections and network structure between different brain regions.

[0004] In recent years, with the advancement of neuroimaging technology, an increasing number of studies have begun to use microstructural features (such as microscopic tissue characteristics obtained by diffusion magnetic resonance imaging) to calculate the weights of brain structural networks. Microstructural features can provide more refined information about brain connectivity, revealing the functional and structural complexity of the brain. However, most existing methods rely on averaging of microstructural features, ignoring the rich details of quantitative analysis along fiber bundles. This simplification may lead to insufficient accuracy in capturing individual differences and structural variations, thus affecting the accuracy and reliability of brain network analysis. Although current quantitative analysis methods along fiber bundles have not been applied to the construction of brain structural network weights, they also have certain shortcomings. They use the same quantitative values ​​for fiber bundles of different lengths, ignoring the information imbalance caused by differences in fiber bundle length. In addition, traditional brain structural network analysis methods usually rely on static models, failing to fully consider the comprehensive utilization of multimodal data and ignoring the complex relationships between different data sources. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for constructing brain structural network weights based on quantitative microstructural features, or a related neuroimaging image detection method. This invention enables quantitative analysis of multimodal brain microstructural features along fiber bundles, selecting specific sampling numbers for fiber bundles of different lengths during the quantification process. An attentional variational autoencoder model is trained using the multimodal quantitative features of a baseline population, and this model is used to calculate the intensity of structural variability in individuals. This intensity of structural variability is ultimately defined as the weight of the brain network. This method can be widely applied in various fields such as brain structural network analysis and has high application value.

[0006] The technical method used in this invention is as follows:

[0007] S1. Establish a brain data model in the computer, and obtain the central template streamline by sequentially acquiring and processing the brain fiber bundle template in the MNI space and processing the line center. Then, process the microstructure quantitative sampling method according to the central template streamline to obtain the brain structure fiber bundle sampling point template.

[0008] The brain data model is a standard brain model obtained based on MRI scans of a large number of normal subjects. The MNI space brain fiber tract template can be an anatomical template of white matter fiber tracts in the Standard Neuroimaging Space. The MNI space brain fiber tract template provides the basis for tracking and segmenting the main fiber tracts in the brain structure, and operations performed on the MNI space brain fiber tract template provide a foundation for subsequent implementation. The brain structure fiber tract sampling point template consists of sampling points of all template streamlines obtained from the MNI space brain fiber tract template, facilitating the direct application of all template streamline sampling points to the brain structure feature map after multimodal magnetic resonance imaging in subsequent implementations.

[0009] S2. Using an improved multimodal magnetic resonance imaging method in a computer, several brain data models in a known database are scanned to obtain several brain structure images. Each brain structure image is processed by feature extraction to obtain a corresponding brain structure feature map. Based on the brain structure fiber bundle sampling point template obtained in step S1, all brain structure feature maps are processed to obtain the vector value of each brain structure feature map. The vector values ​​of all brain structure feature maps are combined to construct a population dataset.

[0010] The brain data model in the database is a standardized brain image constructed from a large amount of known neuroimaging data.

[0011] S3. Construct an attention variational autoencoder model in the computer and input the group dataset into the attention variational autoencoder model for training to obtain a trained attention variational autoencoder model.

[0012] S4. In the computer, the same method as in step S2 is used to obtain the vector value of the brain structure feature map of the individual to be tested. The vector value of the brain structure feature map of the individual to be tested is input into the trained attention variational autoencoder model for model processing. Then, the result of the model processing is directly used as the weight value of the edge in the brain structure network, and finally the weight map of all edges in the brain structure network is obtained.

[0013] The weighted graph of all edges in a brain structural network is a novel research subject. This invention provides a method for obtaining the weighted graph of all edges in a brain structural network, which can be used for further research in the field of brain structural network weights.

[0014] The brain structure network refers to brain model data built using computers in cognitive neuroscience, which utilizes graph theory and network analysis methods to construct a network connecting the human brain structure. The edges of the brain structure network are the edge structures in the constructed network connecting the human brain structure, and the weight graph is a graph obtained through the connection relationships of all edges in the brain structure network and their weight values.

[0015] Step S1 specifically involves:

[0016] S11. Based on the MNI spatial brain fiber bundle template, obtain several template fiber bundles, several template streamlines in each template fiber bundle, and the length of each template streamline.

[0017] The template fiber bundle is a three-dimensional model of the fiber bundle obtained from the MNI spatial brain fiber bundle template. The template streamline is a three-dimensional model of the streamline obtained from the three-dimensional model of the fiber bundle. The length of the template streamline is the length of the streamline in the three-dimensional model. The fiber bundle is a group of white matter fibers in the brain that form intercortical or cortical-subcortical connections in the MNI spatial brain fiber bundle template.

[0018] S12. Perform line centering on all template streamlines in each template fiber bundle to obtain the center template streamline and the length of the center template streamline for each template fiber bundle.

[0019] The line center processing involves uniformly selecting m points for all template streamlines in each template fiber bundle, processing the i-th point of all template streamlines to obtain the center point of all t-th points, processing the center point of the t-th point in the same way to obtain m center points, and connecting the m center points to obtain the central template streamline.

[0020] S13. Based on the length l of the central template streamline of each template fiber bundlei Select the longest central template streamline l0 from the central template streamlines of all template fiber bundles, and set the number of sampling points N0 of the template fiber bundle corresponding to the longest central template streamline.

[0021] S14. The number of sampling points N for template fiber bundles other than the longest central template streamline is obtained by using a microstructure quantitative sampling method. i .

[0022] The microstructure quantitative sampling method is set according to the following formula:

[0023]

[0024] Where l0 is the length of the longest central template streamline, l i N is the length of the i-th central template streamline (excluding the longest central template streamline), and N0 is the number of samples in the template fiber bundle containing the longest central template streamline. i The number of sampling points for the i-th template fiber bundle, excluding the template fiber bundle containing the longest central template streamline.

[0025] S15. In each template fiber bundle, the number of sampling points N of the template fiber bundle is... i N is the number of sampling points for all template streamlines. i .

[0026] S16. Based on the number of sampling points N of all template streamlines in each template fiber bundle i The brain structure fiber bundle sampling point template was obtained by combining the samples.

[0027] The brain structure fiber tract sampling point template contains the specific number N of sampling points for all template streamlines. i The subsequent brain structure feature map is directly based on the specific number N of the template streamlines in the brain structure fiber bundle sampling point template. i Sampling is performed; here it is assumed that the brain structure fiber bundle sampling point template and the brain structure feature map have the same structure (fiber bundles and streamlines, etc.), and the error in the length of fiber bundles and streamlines in the two is not considered, so it is applied directly here.

[0028] Step S2 specifically involves:

[0029] S201. An improved multimodal magnetic resonance imaging method is used to scan several brain data models in a known database to obtain several brain structure imaging images. Each brain structure imaging image is processed by feature extraction to obtain a corresponding brain structure feature map.

[0030] The improved multimodal magnetic resonance imaging method is the first in this invention to use multimodal magnetic resonance imaging to scan several brain data models in a known database. In the prior art, only a single modal magnetic resonance imaging method is used for scanning, and then a single feature extraction method of the single modal magnetic resonance imaging method is used for processing.

[0031] S202. Obtain several fiber bundles and several streamlines in each fiber bundle from each brain structure feature map.

[0032] S203. Based on the brain structure fiber tract sampling point template obtained in step S1, perform tracking sampling on all streamlines of each fiber tract in each brain structure feature map to obtain the sampling points and the number of sampling points N for each streamline. i And the sampling number x for each sampling point.

[0033] The tracking sampling process involves sampling all streamlines of each fiber bundle in each brain structure feature map according to the brain structure fiber bundle sampling point template.

[0034] S204. Divide all sampling points in each fiber bundle according to the same sampling sequence number x to obtain N in each fiber bundle. i Group number sampling point.

[0035] S205. Perform point center processing on each group of sequential sampling points to obtain the center node k of each group of sequential sampling points.

[0036] The point center processing involves processing N samples of the same group's sampling sequence number x. i Processing each sampling point yields N samples with the same sampling sequence number x. i The center point of each sampling point is the center node k.

[0037] S206. In each group of numbered sampling points, for N i Each sampling point is processed with the center node k using Mahalanobis distance, resulting in the Mahalanobis distance D(t) from each sampling point to the corresponding center node k.

[0038] The Mahalanobis distance is the covariance distance between two points.

[0039] S207. The Mahalanobis distance D(t) from each sampling point to the corresponding center node k is inversely proportional to obtain the influence weight W(t) from each sampling point to the corresponding center node k.

[0040] S208. In each group of numbered sampling points, for N i The weights W(t) of each sampling point to the central node k are normalized and summed to obtain the quantitative value P(k) of each central node k.

[0041] S209. In each fiber bundle of each brain structural feature map, construct a vector P = [P(1), P(2), ..., P(k), ..., P(N)] for the quantitative values ​​P(k) of all central nodes k. i )] T As the vector value of each fiber bundle, the vector values ​​of all fiber bundles are combined to obtain the vector value of a brain structure feature map, which is the individual data.

[0042] S210. Combine all individual data to construct a population dataset.

[0043] The specific steps of S201 are as follows:

[0044] For each individual's brain data model, first obtain magnetic resonance imaging (MRI) images using at least two MRI methods, and then process them using the feature extraction methods corresponding to each MRI image to obtain several brain structure feature maps; or, for each individual's brain data model, first obtain MRI images using one MRI method, and then process them using at least two feature extraction methods corresponding to the MRI image to obtain several brain structure feature maps.

[0045] Step S201 only excludes the case where an image is obtained by a magnetic resonance imaging method and then processed by only a feature extraction method. All other cases conform to the situation described in step S201 of the present invention.

[0046] The magnetic resonance imaging method includes diffusion magnetic resonance imaging, functional magnetic resonance imaging, or quantitative magnetic resonance imaging; the feature extraction method corresponding to the magnetic resonance image obtained by the diffusion magnetic resonance imaging method includes diffusion tensor imaging or neural synapse direction dispersion and density imaging; the feature extraction method corresponding to the magnetic resonance image obtained by the functional magnetic resonance imaging method includes low-frequency amplitude analysis; the feature extraction method corresponding to the magnetic resonance image obtained by the quantitative magnetic resonance imaging method includes magnetic susceptibility quantitative imaging or magnetic resonance fingerprinting.

[0047] The inverse ratio processing in step S207 is set according to the following formula:

[0048]

[0049] Where W(t) is the influence weight from the sampling point to the corresponding center node, D(t) is the Mahalanobis distance from the sampling point to the corresponding center node, and k is a preset proportional weight value.

[0050] The attention variational autoencoder model in step S3 includes an encoder, a multi-head attention mechanism, and a decoder connected in series; the encoder and decoder are both encoders and decoders in the variational autoencoder model, and the multi-head attention mechanism is the multi-head attention mechanism in the Transformer model.

[0051] Step S4 specifically involves:

[0052] S401. Obtain the vector values ​​of the brain structural feature map of the individual to be tested using the same method as in step S2.

[0053] S402. Input the vector values ​​of the brain structural feature map of the individual to be tested into the trained attention variational autoencoder model to obtain the structural variation intensity value of each fiber bundle in the brain structural feature map of the individual to be tested.

[0054] S403. The structural variation intensity value of each fiber bundle in the brain structural feature map of the individual to be tested is used as the weight of the edge between the two endpoints of each fiber bundle in the brain network, thereby constructing a weight map of all edges in the brain structural network.

[0055] By combining the weighted graph of all edges in the obtained brain structural network with functional connectivity data in the field of brain structural networks, the interaction between the function and structure of different brain regions can be further explored. Based on the relevant results, functional analysis of various brain regions in the human brain can be performed, which can be used for drug efficacy evaluation and for detection, analysis and judgment of CT images.

[0056] The innovation of this invention lies in adopting reasonable sampling points for fiber bundles of different lengths, using multimodal magnetic resonance imaging to extract features from the brain, and the first comprehensive method of using structural variation intensity values ​​as the weight values ​​of edges in brain structural networks. This provides a new direction for the research of brain structural network weights and brings the beneficial effect of extracting more brain network feature information.

[0057] The beneficial effects of this invention are:

[0058] 1. This invention fully considers the information imbalance caused by the difference in fiber bundle length and provides an adaptive sampling and quantification scheme for the physical length of fiber bundles, so that more information can be characterized for longer fiber bundles.

[0059] 2. This invention uses an improved multimodal magnetic resonance imaging technique to overcome the shortcomings of traditional single-modal methods, which result in a lack of microstructural information and a single type of information in brain structural feature maps, and provides multimodal brain microstructural information.

[0060] 3. This invention is the first to use the structural variation intensity value as a comprehensive method for the weight value of edges in brain structural networks, which provides a new direction for the research of brain structural network weights and brings the beneficial effect of extracting more brain network feature information. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of FA and NDI in Embodiment 2 of the present invention.

[0062] Figure 2 This is a schematic diagram of the center streamlines and sampling number of the left IFOF template and the left UF template in Embodiment 2 of the present invention.

[0063] Figure 3 This is a graph showing the quantitative results of FA and NDI in the left ventricular fossa of an individual in Embodiment 2 of the present invention.

[0064] Figure 4 This is a model diagram of the attention variational autoencoder in an embodiment of the present invention.

[0065] Figure 5 This is a schematic diagram illustrating how variational autoencoders predict structural variation intensity and define it as weights in a brain structural network in Embodiment 2 of the present invention. Detailed Implementation

[0066] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be noted that the following embodiments are intended to facilitate the understanding of the present invention and do not constitute any limitation thereof.

[0067] Embodiment 1 of the present invention is carried out according to the following steps:

[0068] S1. Establish a brain data model in the computer and store it in the memory. Based on the MNI spatial brain fiber bundle template, sequentially perform acquisition processing and line center processing to obtain the central template streamline. Based on the central template streamline, use the computer processor to process the microstructure quantitative sampling method to obtain the brain structure fiber bundle sampling point template.

[0069] The brain data model is a standard brain model obtained from MRI scans of a large number of normal subjects. The MNI space brain fiber tract template can be used as an anatomical template for white matter fiber tracts in the Standard Neuroimaging Space. The MNI space brain fiber tract template provides the basis for tracking and segmenting the main fiber tracts in the brain structure in this invention. Operations on the MNI space brain fiber tract template provide a foundation for subsequent implementation. The brain structure fiber tract sampling point template consists of sampling points of all template streamlines obtained from the MNI space brain fiber tract template. This facilitates the direct application of all template streamline sampling point numbers to the brain structure feature map after multimodal magnetic resonance imaging scans in subsequent implementations.

[0070] Step S1 is as follows:

[0071] S11. Based on the MNI spatial brain fiber bundle template, obtain several template fiber bundles, several template streamlines in each template fiber bundle, and the length of each template streamline.

[0072] The template fiber bundle is a 3D model of the fiber bundle obtained from the MNI spatial brain fiber bundle template. The template streamline is a 3D model of the streamline obtained from the 3D model of the fiber bundle. The length of the template streamline is the length of the streamline in the 3D model. The fiber bundle is a group of white matter fibers in the brain that form intercortical or cortical-subcortical connections in the MNI spatial brain fiber bundle template.

[0073] S12. Perform line centering on all template streamlines in each template fiber bundle to obtain the center template streamline and the length of the center template streamline for each template fiber bundle.

[0074] The center processing involves uniformly selecting m points for all template streamlines in each template fiber bundle, processing the i-th point of all template streamlines to obtain the center point of all t-th points, processing the center point of the t-th point in the same way to obtain m center points, and connecting the m center points to obtain the center template streamline.

[0075] S13. Based on the length l of the central template streamline of each template fiber bundle i Select the longest central template streamline l0 from the central template streamlines of all template fiber bundles, and set the number of sampling points N0 of the template fiber bundle corresponding to the longest central template streamline.

[0076] In practice, the number of sampling points N0 corresponding to the longest central template streamline is set to 100.

[0077] S14. The number of sampling points N for template fiber bundles other than the longest central template streamline is obtained by using a microstructure quantitative sampling method. i .

[0078] The quantitative sampling method for microstructures is set according to the following formula:

[0079]

[0080] Where l0 is the length of the longest central template streamline, l i N is the length of the i-th central template streamline (excluding the longest central template streamline), and N0 is the number of samples in the template fiber bundle containing the longest central template streamline. i The number of sampling points for the i-th template fiber bundle, excluding the template fiber bundle containing the longest central template streamline.

[0081] S15. In each template fiber bundle, the number of sampling points N of the template fiber bundle is... i N represents the number of sampling points for all template streamlines in the template fiber bundle. i ;

[0082] S16. Based on the number of sampling points N of all template streamlines in each template fiber bundle i The brain structure fiber bundle sampling point template was obtained by combining the samples.

[0083] The brain structure fiber tract sampling point template contains the specific number N of sampling points for all template streamlines. i The subsequent brain structure feature map is directly based on the specific number N of the template streamlines in the brain structure fiber bundle sampling point template. i Sampling is performed; here it is assumed that the brain structure fiber bundle sampling point template and the brain structure feature map have the same structure (fiber bundles and streamlines, etc.), and the error in the length of fiber bundles and streamlines in the two is not considered, so it is applied directly here.

[0084] S2. Using an improved multimodal magnetic resonance imaging method in a computer, the brain data models of several individuals stored in a known database in the computer memory are scanned to obtain several brain structure imaging maps. Each brain structure imaging map is processed by feature extraction to obtain a corresponding brain structure feature map. Based on the brain structure fiber bundle sampling point template obtained in step S1, all brain structure feature maps are processed to obtain the vector value of each brain structure feature map. The vector values ​​of all brain structure feature maps are combined to construct a population dataset.

[0085] The brain data model in the database is a standardized brain image constructed from a large amount of known neuroimaging data.

[0086] Step S2 is as follows:

[0087] S201. An improved multimodal magnetic resonance imaging method is used to scan the brain data models of several individuals stored in a known database in a computer memory to obtain several brain structure imaging maps. Each brain structure imaging map is processed by feature extraction to obtain the corresponding brain structure feature map.

[0088] Step S201 is as follows:

[0089] For each individual's brain data model, first obtain magnetic resonance imaging (MRI) images using at least two MRI methods, and then process them using at least one feature extraction method corresponding to each MRI image to obtain several brain structure feature maps; or, for each individual's brain data model, first obtain MRI images using one MRI method, and then process them using at least two feature extraction methods corresponding to the MRI image to obtain several brain structure feature maps.

[0090] The magnetic resonance imaging method includes diffusion magnetic resonance imaging, functional magnetic resonance imaging, or quantitative magnetic resonance imaging; the feature extraction method corresponding to the magnetic resonance image obtained by the diffusion magnetic resonance imaging method includes diffusion tensor imaging or neural synapse direction dispersion and density imaging; the feature extraction method corresponding to the magnetic resonance image obtained by the functional magnetic resonance imaging method includes low-frequency amplitude analysis; the feature extraction method corresponding to the magnetic resonance image obtained by the quantitative magnetic resonance imaging method includes magnetic susceptibility quantitative imaging or magnetic resonance fingerprinting.

[0091] S202. Obtain several fiber bundles and several streamlines in each fiber bundle from each brain structure feature map.

[0092] S203. Based on the brain structure fiber tract sampling point template obtained in step S1, perform tracking sampling on all streamlines of each fiber tract in each brain structure feature map to obtain the sampling points and the number of sampling points N for each streamline of each fiber tract in each brain structure feature map. i And the sampling number x for each sampling point.

[0093] In practice, the brain structures in the MNI space brain fiber bundle template are assumed to be consistent with the brain structures in the database, ignoring structural differences.

[0094] The tracking sampling process involves sampling all streamlines of each fiber bundle in each brain structure feature map according to the brain structure fiber bundle sampling point template. In specific implementation, during the tracking sampling process, each streamline is sampled simultaneously from the same side. All streamlines will simultaneously generate the first sampling point, the second sampling point, and so on until the last sampling point. The x-th point is the sequence number of that sampling point.

[0095] S204. Divide all sampling points in each fiber bundle of each brain structure feature map according to the same sampling sequence number x to obtain N in each fiber bundle. i Group number sampling point.

[0096] In practice, the first sampling point of all streamlines in each fiber bundle is taken as the first group of sequential sampling points, the second sampling point of all streamlines is taken as the second group of sequential sampling points, and so on until the sampling ends.

[0097] S205. Perform point centering on each group of numbered sampling points in each fiber bundle of each brain structure feature map to obtain the center node k of each group of numbered sampling points.

[0098] Point center processing involves N sampling numbers x within the same group. i Processing each sampling point yields N samples with the same sampling sequence number x. i The center point of each sampling point is the center node k.

[0099] S206. In each group of numbered sampling points, for N i Each sampling point is processed with the center node k using Mahalanobis distance, resulting in the Mahalanobis distance D(t) from each sampling point to the corresponding center node k.

[0100] Mahalanobis distance is the covariance distance between two points.

[0101] S207. The Mahalanobis distance D(t) from each sampling point to the corresponding center node k is inversely proportional to obtain the influence weight W(t) from each sampling point to the corresponding center node k.

[0102] The inverse ratio processing is set according to the following formula:

[0103]

[0104] Where W(t) is the influence weight from the sampling point to the corresponding center node, D(t) is the Mahalanobis distance from the sampling point to the corresponding center node, and k is a preset proportional weight value.

[0105] S208. In each group of numbered sampling points, for N i The weights W(t) of each sampling point to the central node k are normalized and summed to obtain the quantitative value P(k) of each central node k.

[0106] S209. In each fiber bundle of each brain structural feature map, construct a vector P = [P(1), P(2), ..., P(k), ..., P(N)] for the quantitative values ​​P(k) of all central nodes k. i )] TAs the vector value of each fiber bundle, the vector values ​​of all fiber bundles are combined to obtain the vector value of a brain structure feature map, which is the individual data.

[0107] S210. Combine all individual data to construct a population dataset.

[0108] S3. Construct an attention variational autoencoder model in the computer and input the group dataset into the attention variational autoencoder model for training to obtain a trained attention variational autoencoder model.

[0109] The attention variational autoencoder model in step S3 includes an encoder, a multi-head attention mechanism, and a decoder connected in series. The encoder and decoder are both encoders and decoders in the variational autoencoder model, and the multi-head attention mechanism is the multi-head attention mechanism in the Transformer model.

[0110] S4. In the computer, the same method as in step S2 is used to obtain the vector value of the brain structure feature map of the individual to be tested. The vector value of the brain structure feature map of the individual to be tested is input into the trained attention variational autoencoder model for model processing. Then, the result of the model processing is directly used as the weight value of the edge in the brain structure network, and finally the weight map of all edges in the brain structure network is obtained.

[0111] The brain structural network is a computer-generated brain model in cognitive neuroscience, which uses graph theory and network analysis methods to construct a network connecting the structures of the human brain. The edges of the brain structural network represent the edge structure within the constructed network, and the weight graph is a graph obtained by considering the connections and weights of all edges in the brain structural network. By default, the fiber bundles of the brain structural feature map of the test individual are considered as edges in the brain structural network. Similarly, all fiber bundles have corresponding edges in the brain structural network, thus constructing a network graph. The structural variation intensity value of the fiber bundle is used as the weight value of the corresponding edge, ultimately resulting in a weight graph of all edges in the brain structural network.

[0112] Step S4 is as follows:

[0113] S401. Obtain the vector values ​​of the brain structural feature map of the individual to be tested using the same method as in step S2.

[0114] S402. Input the vector values ​​of the brain structural feature map of the individual to be tested into the trained attention variational autoencoder model to obtain the structural variation intensity value of each fiber bundle in the brain structural feature map of the individual to be tested.

[0115] S403. The structural variation intensity value of each fiber bundle in the brain structural feature map of the individual to be tested is used as the weight of the edge between the two endpoints of each fiber bundle in the brain network, thereby constructing a weight map of all edges in the brain structural network.

[0116] By default, the fiber bundles in the brain structural feature map of the individual being tested are regarded as an edge of the brain structural network. Similarly, all fiber bundles have corresponding edges in the brain structural network, thus constructing a network graph. The structural variation intensity value of the fiber bundle is used as the weight value of the corresponding edge, and finally the weight graph of all edges in the brain structural network is obtained.

[0117] Example 2

[0118] In this embodiment, 108 participants were selected as the baseline group, including 71 males and 52 females, with a mean age of 18.33 years. One additional participant was selected as the test individual.

[0119] Using the MNI spatial brain fiber tract template proposed by Garyfallidis et al., for fiber tract i in the template, the geometric center of all streamlines within the fiber tract is calculated and defined as the central streamline. The length l of the central streamline of each fiber tract in the template is then obtained. i .

[0120] The left inferior frontal-occipital fasciculus (IFOF) was used as the reference for the quantitative microstructure sampling method, and the left uncinate fasciculus (UF) was used as the research object. The quantitative microstructure sampling method for the left UF sample size N1 was set according to the following formula:

[0121]

[0122] Where l0 is the length of the central template streamline of the left IFOF, N0 is the number of samples of the left IFOF, l1 is the length of the central template streamline of the left UF, and N1 is the number of samples of the left UF.

[0123] Substituting the length of the central template streamline of the left IFOF (l0 = 151.37 mm), the sampling number of the left IFOF (N0 = 100), and the length of the central template streamline of the left UF (l1 = 77.44 mm) into the formula for the quantitative sampling method of microstructures, we obtain the sampling number of the left UF (N1 = 51). Figure 2 As shown, 1 represents the left IFOF, 2 represents the left UF, 3 represents the center streamline of the left IFOF with a length of 151.37 mm and a quantitative sampling number N0 of 100, and 4 represents the center streamline of the left UF with a length of 77.44 mm and a sampling number N1 of 51. The sampling numbers obtained by processing using the above method correspond to the actual length of the fiber bundle, avoiding information imbalance caused by differences in fiber bundle length during subsequent sampling along the fiber bundle.

[0124] The magnetic resonance imaging (MRI) scan was performed on a Siemens 3T Prisma system, acquiring T1-weighted imaging (T1W) and diffusion-weighted imaging (DWI) sequences. T1W used MPRAGE sequences with a voxel size of 1×1×1 mm. 3 The DWI used an SMS-EPI sequence, with a voxel size of 1.5 × 1.5 × 1.5 mm. 3 GRAPPA factor = 2, SMS factor = 3, b value is 1000, 2000 and 3000 s / mm 3 Each direction has 30 directions, providing phase coding direction PA and inverse coding direction AP to reduce artifacts and improve signal-to-noise ratio.

[0125] PCA denoising, Gibbs artifact removal, eddy current correction, and distortion correction were performed on DWI images using MRtrix3 and FSL software. Two different feature extraction methods based on DWI were selected to obtain brain structural feature maps. A diffusion tensor model was constructed using MRtrix3 software to calculate fractional anisotropy (FA) brain structural feature maps. The neurite orientation dispersion and density imaging (NODDI) toolbox in Matlab was used to construct neurite orientation dispersion and density imaging to calculate axon density fraction (NDI) brain structural feature maps. Figure 1 The diagram shows the FA brain structure feature map and the NDI brain structure feature map of the test individuals.

[0126] like Figure 1 This is a schematic diagram of FA and NDI. The left side represents FA, and the right side represents NDI. The brighter the voxel color, the higher the value at that location.

[0127] The T1W image is registered with the DWI image, and the fiber orientation distribution function is obtained to obtain the distribution of all fiber bundles in the brain of the individual in the database. This embodiment takes the left UF in the brain of an individual as an example:

[0128] Based on the previous sampling number N1=51 of the left UF obtained from the MNI space brain fiber tract template, the left UF in the individual's brain was sampled.

[0129] like Figure 3 As shown, Figure 3In (a), the left UF is the relative position in the human brain. The starting sampling point is position 1, which is close to the lower side, and the ending sampling point is position 51, which is close to the upper side.

[0130] All streamlines j in the left UF are divided into N1 sampling points. The geometric center of all streamlines in each group of sampling points is calculated to obtain N1 central nodes, which form the central streamlines. For each central node k, the Mahalanobis distance D(t) between each sampling point and the corresponding central node is calculated. The weight W(t) of the sampling point on the central node is inversely proportional to D(t).

[0131] For the FA brain structure feature map and the NDI brain structure feature map, the sampled values ​​of the feature map are weighted and mapped to the corresponding central node k with weight W(t). After weighting, the sampled values ​​are normalized and summed to obtain the quantitative value P(k) of the central node.

[0132] The vector value of the left-hand UF in the FA brain structure feature map can be represented as vector P. FA =[P FA (1), P FA (2), ..., P FA (k), ..., P FA (N1)] T The vector value of the left-side UF in the NDI brain structure feature map can be represented as vector P. NDI =[P NDI (1), P NDI (2), ..., P NDI (k), ..., P NDI (N1)] T .

[0133] Quantitative results of UF on the left side of individual samples as follows Figure 3 As shown, Figure 3 (a) shows the spatial location of the left ulnar fossa in the brain. Figure 3 Figure (b) shows the quantitative results of FA and NDI. The horizontal axis represents the sampling point location, points 1-51 represent the quantitative results of FA, and points 52-102 represent the quantitative results of NDI. The vertical axis represents the values ​​of FA and NDI. As can be seen from Figure a, quantitative values ​​were obtained at 51 locations along the left UF from bottom to top. Figure b shows that 51 features were obtained for each of the two microstructural features, FA and NDI. The two microstructural features have different characteristics, representing the multimodal microstructural information of the brain.

[0134] The same method described above is used to process and obtain the vector values ​​of all fiber bundles. The vector values ​​of all fiber bundles are combined to obtain the vector values ​​of a brain structure feature map, which is the individual data. All individual data are combined to construct the population dataset.

[0135] An attention-based variational autoencoder model was constructed to represent brain structure through quantitative features of multimodal microstructures, such as... Figure 4 As shown. The attention variational autoencoder model consists of an encoder, a multi-head attention mechanism, and a decoder connected in series. Both the encoder and decoder are derived from the variational autoencoder model, and the multi-head attention mechanism is based on the Transformer model. The specific steps are as follows:

[0136] First, the model's input data comes from a population dataset containing n = 108 samples. Each sample yields R = 2 brain structural feature maps through the steps described above, with R*N values ​​for fiber tract i. i There are several quantitative values, of which N i This represents the number of sampling points on fiber bundle i. The input data shape of the model is (n, R*N). i That is, each input sample has R*N. i Each input feature.

[0137] The input data is standardized by setting the mean of each feature to 0 and the variance to 1, thereby eliminating the impact of differences in feature dimensions and scales on model training. The encoder contains one fully connected layer and a self-attention layer. To balance the correlation between features, the input data is processed through the multi-head self-attention mechanism of Transform, generating query, key, and value matrices. The dot product of the query and key is calculated to obtain attention weights. These weights are then used to perform a weighted average of the input features, generating a weighted input representation that allows the model to automatically focus on the most relevant input features.

[0138] The weighted input representation is mapped to the latent space, and the mean μ and variance σ of the latent variables are output. 2 This generates the latent variable z; by introducing a reparameterization technique, the sampling process of the latent space is represented as:

[0139] z=μ+σ·ε

[0140] Where ε is the noise sampled from the standard normal distribution N(0,1).

[0141] Gradient updates are performed on the latent space during backpropagation, enabling the VAE to generate higher-quality reconstructed data by optimizing the distribution of the latent space. The latent variable z is input to the decoder, passing through a fully connected layer to generate a reconstructed representation of the input data. The loss function consists of two parts: reconstruction error and KL divergence. The reconstruction error measures the difference between the input data and its reconstructed data, while the KL divergence measures the difference between the latent space distribution and the standard normal distribution. During training, He initialization and the Adam optimizer are used, with a learning rate α = 10. -3 Set the maximum number of training epochs to 200, and stop training when the loss function value is less than 0.01 or the maximum number of training epochs is reached.

[0142] For a single test individual, based on the obtained R=2 brain structural feature maps, the R*N for the left UF of the fiber tract is calculated. i Let x be the quantitative value of a microstructure. The individual's feature data is input into the encoder part of a pre-trained variational autoencoder. The encoder processes the input data through multiple fully connected layers and self-attention layers to generate the mean and variance of a latent space, which represents the low-dimensional features of the input data. Based on the mean and variance, a reparameterization technique is used to sample latent variables z from the Gaussian distribution of the latent space. The latent variables z are input into the decoder part of the variational autoencoder. The decoder maps the latent variables back to the original input data space through multiple fully connected layers, generating the reconstructed feature data of the individual to be predicted, which serves as the predicted output x'. Figure 5 As shown.

[0143] The structural variability intensity of an individual is the mean squared error (MSE) between the true and predicted values, calculated as follows:

[0144]

[0145] Where, x j Let x be the i-th feature value of the input data. j ′ represents the predicted value generated by the decoder, and m represents the number of input features. A higher MSE value indicates that the individual deviates significantly from the distribution of the training data, suggesting that the individual may have more significant structural variations, which in turn affect the weights of the edges in the brain structural network.

[0146] The brain regions connected by the left UF are the orbitofrontal cortex and the inferior temporal gyrus, such as... Figure 5 As shown, in the construction of the brain structure network, two brain regions are abstracted as nodes 5 and 7, and the structural variation intensity is defined as the weight 6 of the connection edge between the two nodes.

[0147] By combining the weighted graph of all edges in the obtained brain structural network with functional connectivity data in the field of brain structural networks, the interaction between the function and structure of different brain regions can be further explored. Based on the relevant results, functional analysis of various brain regions in the human brain can be performed, which can be used for drug efficacy evaluation and for detection, analysis and judgment of CT images.

[0148] The above results verify the feasibility of the present invention, which can innovatively provide a brain structure network method.

[0149] The weighted graph of all edges in a brain structural network is a new research subject. This invention ultimately provides a method for obtaining the weighted graph of all edges in a brain structural network, which can be used for further research in the field of brain structural network weights.

[0150] The innovation of this invention lies in adopting reasonable sampling points for fiber bundles of different lengths, using multimodal magnetic resonance imaging to extract features from the brain, and the first comprehensive method of using structural variation intensity values ​​as the weight values ​​of edges in brain structural networks. This provides a new direction for the research of brain structural network weights and brings the beneficial effect of extracting more brain network feature information.

[0151] The above embodiments are only two examples of the present invention and are not intended to limit the scope of the present invention. Therefore, all variations made according to the method of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for constructing brain structural network weights based on quantitative features of microstructures, characterized in that, Comprise the following steps: S1, a brain data model is established in the computer, and center template streamlines are obtained by sequentially performing acquisition processing and line center processing according to the MNI space brain fiber bundle template; the brain structure fiber bundle sampling point template is obtained by performing microstructure quantitative sampling method processing according to the center template streamlines; S2, a plurality of brain structure imaging graphs are obtained by scanning a plurality of brain data models in a known database by using an improved multi-modal magnetic resonance imaging method in the computer, each brain structure imaging graph is processed by feature extraction to obtain a corresponding brain structure feature graph, and the vector values of each brain structure feature graph are obtained by comprehensively processing all brain structure feature graphs according to the brain structure fiber bundle sampling point template obtained in step S1; and the vector values of all brain structure feature graphs are combined to construct a population data set; S3, an attention variational autoencoder model is constructed in the computer, and the population data set is input into the attention variational autoencoder model for training to obtain a trained attention variational autoencoder model; S4, the vector values of the brain structure feature graph of the to-be-measured individual are obtained by using the same method as in step S2, and the vector values of the brain structure feature graph of the to-be-measured individual are input into the trained attention variational autoencoder model for model processing, and then the model processing result is directly used as the weight value of the edge in the brain structure network, and finally the weight graph of all edges in the brain structure network is obtained; The step S1 is specifically: S11, a plurality of template fiber bundles, a plurality of template streamlines in each template fiber bundle and the length of each template streamline are obtained according to the MNI space brain fiber bundle template; S12, the line center processing is performed on all template streamlines in each template fiber bundle to obtain the center template streamline and the length of the center template streamline; The line center processing is that m points are uniformly taken from all template streamlines in each template fiber bundle, the center point of all tth points is obtained by processing the ith point of all template streamlines, m center points are obtained by processing in the same way as the center point of the tth point, and the m center points are connected to obtain the center template streamline; S13, the length of the center template streamline of each template fiber bundle , the center template streamline with the longest length is selected from the center template streamlines of all template fiber bundles, and the number of sampling points of the template fiber bundle corresponding to the center template streamline with the longest length is set ;​ S14, the microstructure quantitative sampling method is used to process to obtain the sampling point number of the template fiber bundle corresponding to the template fiber bundle except the longest center template streamline ; The microstructure quantitative sampling method is set according to the following formula: wherein, L is the length of the longest center template streamline, Li is the length of the i-th center template streamline other than the longest center template streamline, N is the number of samples of the template fiber bundle where the longest center template streamline is located, Ni is the number of samples of the i-th template fiber bundle other than the template fiber bundle where the longest center template streamline is located; S15, in each template fiber bundle, the number of sampling points of the template fiber bundle is determined according to the number of sampling points of the template streamline and the number of sampling points of the template streamline as the number of sampling points of all template streamlines ; S16, number of sampling points of all template streamlines in each template fiber bundle The brain structural fiber bundle sampling point template is obtained by combination.

2. The method of claim 1, wherein the microstructural quantitative features are selected from the group consisting of: The step S2 is specifically: ​ S201, a plurality of brain structure imaging graphs are obtained by scanning a plurality of brain data models in a known database by using an improved multi-modal magnetic resonance imaging method, and each brain structure imaging graph is processed by feature extraction to obtain a corresponding brain structure feature graph; S202, a plurality of fiber bundles and a plurality of streamlines in each fiber bundle are obtained from each brain structure feature graph; S203, according to the brain structure fiber bundle sampling point template obtained in step S1, tracking sampling processing is performed on all streamlines of each fiber bundle in each brain structure feature map, to obtain sampling points, sampling point numbers and sampling serial numbers x of each streamline of each sampling point S204, dividing all sampling points in each fiber bundle by the same sampling serial number x to obtain the sampling points in each fiber bundle group serial number sampling point; S205, the point center processing is performed on each group of serial number sampling points to obtain the center node k of each group of serial number sampling points; The point center processing is processing of the same group of sampling serial number x The center point of the same group of sampling serial number x The center node k is obtained by processing the same group of sampling serial number x S206、In each group of serial number sampling points, Mahalanobis distance processing is performed on each sampling point and the center node k respectively to obtain the Mahalanobis distance D(t) of each sampling point to the corresponding center node k. In each group of serial number sampling points, Mahalanobis distance processing is performed on each sampling point and the center node k respectively to obtain the Mahalanobis distance D(t) of each sampling point to the corresponding center node k. S207, the inverse processing is performed on the Mahalanobis distance D(t) from each sampling point to the corresponding center node k to obtain the action weight W(t) from each sampling point to the corresponding center node k; S208. In each group of numbered sampling points, for The effect weights W(t) from each sampling point to the central node k are normalized and summed to obtain the quantitative value P(k) of each central node k. S209. In each fiber bundle, the quantitative values P(k) of all center nodes k are constructed into a vector As the vector value of each fiber bundle, the vector values of all fiber bundles are combined to obtain a vector value of a brain structure feature map, i.e. individual data; S210, all individual data are combined to construct a population data set.

3. The method of claim 2, wherein the microstructural quantitative features are selected from the group consisting of: The step S201 is specifically: ​ The brain data model of each individual is first obtained by at least two magnetic resonance imaging methods to obtain magnetic resonance imaging images, and then processed by each corresponding feature extraction method of each magnetic resonance imaging image to obtain several brain structure feature images; or the brain data model of each individual is first obtained by one magnetic resonance imaging method to obtain a magnetic resonance imaging image, and then processed by at least two corresponding feature extraction methods of the magnetic resonance imaging image to obtain several brain structure feature images.

4. The brain structure network weight construction method based on microstructure quantitative features according to claim 3, characterized in that: The magnetic resonance imaging method includes a diffusion magnetic resonance imaging method, a functional magnetic resonance imaging method or a quantitative magnetic resonance imaging method; the feature extraction method corresponding to the magnetic resonance imaging image obtained by the diffusion magnetic resonance imaging method includes a diffusion tensor imaging method or a neurite direction dispersion and density imaging method; the feature extraction method corresponding to the magnetic resonance imaging image obtained by the functional magnetic resonance imaging method includes a low-frequency amplitude analysis method; and the feature extraction method corresponding to the magnetic resonance imaging image obtained by the quantitative magnetic resonance imaging method includes a magnetic susceptibility quantitative imaging method or a magnetic resonance fingerprint imaging method.

5. The brain structure network weight construction method based on microstructure quantitative features according to claim 2, characterized in that: The inverse processing in the step S207 is set according to the following formula: wherein, is a weight of the sampling point to the corresponding center node, is a Mahalanobis distance of the sampling point to the corresponding center node, is a preset proportional weight value.

6. The brain structure network weight construction method based on microstructure quantitative features according to claim 1, characterized in that: The attention variational autoencoder model in the step S3 includes an encoder, a multi-head attention mechanism and a decoder connected in sequence; the encoder and the decoder both adopt the encoder and the decoder in the variational autoencoder, and the multi-head attention mechanism adopts the multi-head attention mechanism in the Transformer model. 7.The method of constructing brain structural network weights based on microstructural quantitative features according to claim 1, wherein, The step S4 is specifically: S401, obtaining the vector value of the brain structure feature image of the to-be-tested individual by the same method as that in the step S2; S402, inputting the vector value of the brain structure feature image of the to-be-tested individual into the trained attention variational autoencoder model to obtain the structural variation intensity value of each fiber bundle in the brain structure feature image of the to-be-tested individual; S403, taking the structural variation intensity value of each fiber bundle in the brain structure feature image of the to-be-tested individual as the weight of the edge between the two end points of each fiber bundle in the brain network, so as to construct the weight map of all edges in the brain structure network.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to realize the steps of the method in any one of claims 1 to 7.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method in any one of claims 1 to 7.

Images