Methods and systems for calculating long- and short-range connectivity of brain structures

By combining random forest fiber tracing with sMRI-DTI differential fusion, the problems of loss of directional information in fiber crossing regions and differences in the anatomical microenvironment of long and short-range fibers were solved, achieving high-quality fiber trajectory reconstruction and accurate structural connectivity calculation.

CN123118698APending Publication Date: 2026-11-17BEIJING TECH & BUSINESS UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611378506.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-07
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, fiber tracking methods lose directional information in fiber crossing areas, lack comprehensive utilization of neighborhood information, and cannot handle fan-shaped and curved fiber bundles. Furthermore, sMRI-DTI multimodal fusion schemes ignore the differences in the anatomical microenvironment of long and short-range fibers, resulting in a mismatch between fusion weights and the physical meaning of fiber types, and lacking a false positive suppression mechanism.

Method used

A method based on random forest fiber tracking and sMRI-DTI differential fusion was adopted. Fiber tracking was performed by combining a random forest classifier with neighborhood multi-point sampling voting. The global tracking confidence was calculated, and the structural connection edges were classified into short-range and long-range connections based on the bypass ratio. The connectivity strength was calculated by fusion of sMRI and DTI data with emphasis respectively.

Benefits of technology

It solves the problems of loss of directional information and high false positive rate in fiber tracking, realizes differentiated fusion of long and short-range fibers, improves the quality and accuracy of fiber trajectory reconstruction and fusion, and makes reasonable use of multimodal information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118698A_ABST
    Figure CN123118698A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of medical image processing and brain network computing technology, and relates to a method and system for calculating the short- and long-range connectivity of brain structures. It takes diffusion tensor imaging (DTI) data as input, and uses a random forest classifier combined with neighborhood multi-point sampling voting to perform streamlined fiber tracing, outputting a set of whole-brain fiber tracts and the tracing confidence score of each fiber tract. It then takes structural magnetic resonance imaging (sMRI) data as input, performs brain atlas registration and brain region node segmentation, and extracts the anatomical morphological feature vectors of each brain region node. The results from the first two parts are combined, and connectivity strength is calculated using a fusion method emphasizing sMRI (short-range) and a fusion method emphasizing DTI plus confidence score (long-range), respectively, ultimately outputting two structural connectivity matrices: short-range and long-range. This invention can solve the quality problems of upstream fiber trajectory reconstruction and the problems of spatial alignment and quantitative extraction of anatomical features from multimodal data in existing technologies.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of medical image processing and brain network computing technology, and relates to a method and system for calculating long and short-range connectivity of brain structures, particularly a method and system for calculating long and short-range connectivity of brain structures based on random forest fiber tracing and sMRI-DTI differential fusion. Background Technology

[0002] Diffusion tensor imaging (DTI) is currently the only magnetic resonance imaging technique capable of non-invasively visualizing white matter fiber tracts in living organisms. Fiber tracing based on DTI is a core technique for constructing brain connectomes. Existing mainstream fiber tracing methods are mainly divided into two categories: mathematical model-based tracing schemes and machine learning-based tracing schemes.

[0003] In mathematical model-based tracking schemes, deterministic tracking using diffusion tensor models assumes that water molecule diffusion within each voxel follows a Gaussian distribution, progressing point-by-point along the principal eigenvector of the tensor until the anisotropy fraction falls below a preset threshold or the turning angle exceeds a preset curvature threshold. This method cannot address the fiber crossing problem. To solve this, researchers have developed higher-order diffusion models such as constrained spherical deconvolution and Q-ball imaging, which identify multiple orientations in crossing regions by reconstructing multiple fiber orientation peaks within each voxel. Additionally, there are global tracking schemes that model the whole-brain fiber tract as a global optimization problem, but these are computationally extremely expensive and still rely on the selection of underlying models.

[0004] Among machine learning-based tracking schemes, researchers have proposed a fiber tracking method based on random forests. This method uses the original diffusion-weighted signal value within the current voxel's 3×3×3 neighborhood and the previous propagation direction as input. A pre-trained random forest classifier outputs the direction probability distribution and the non-fiber termination probability. During tracking, 30 to 50 sampling points are generated around the current position for voting decisions. This method performed excellently in the International Fiber Tracking Challenge. However, the research objective of this method stops at fiber trajectory reconstruction; its output is only a sequence of fiber trajectory coordinates in three-dimensional space. The probability information generated during tracking is never passed to subsequent brain region topological mapping and structural connectivity calculations.

[0005] In calculating brain structural connectivity, traditional methods map fiber bundle endpoints to brain region nodes defined by standard brain atlases after fiber tracing, and construct a structural connectivity matrix by statistically analyzing indicators such as fiber count and average FA (anisotropy score). In recent years, researchers have begun exploring multimodal fusion schemes of sMRI and DTI, incorporating sMRI anatomical features and DTI structural connectivity into a unified framework for differentiated weighting, used for cognitive prediction and disease classification. Other studies have utilized fMRI, DTI, and sMRI to construct functional, structural, and morphological networks, respectively, and used combinations of these three networks to classify autism spectrum disorders with high accuracy, demonstrating that sMRI and DTI can provide complementary information. Furthermore, studies on long- and short-range structural connectivity have shown significant differences between short-range associated fibers (located in the superficial white matter thin layer) and long-range deep white matter fiber bundles in microstructural indicators such as FA, axial diffusion coefficient, and orientation dispersion index, suggesting fundamental differences in the anatomical microenvironment between long- and short-range fibers. Ultrastructural studies have further revealed that the axon diameter and myelin sheath thickness of short-range fibers are smaller than those of long-range fibers.

[0006] However, traditional tracking schemes based on mathematical modeling have the following problems: (1) The mathematical model assumptions do not match the actual biological tissues. DTI loses directional information in the fiber crossover area. Multi-directional models such as CSD are sensitive to noise and are prone to false peaks. (2) Traditional methods rely on manually set termination thresholds. Hard thresholds such as FA below 0.15 cause a large number of shallow white matter short-range U-shaped fibers to terminate prematurely before reaching the cortex. (3) Traditional methods only make directional decisions in the current voxel and lack comprehensive utilization of neighborhood information. It is difficult to handle asymmetric configurations such as fan-shaped and curved fiber bundles. (4) Traditional methods lack quantifiable tracking confidence. Downstream, it is impossible to distinguish between high-reliability and low-reliability fibers.

[0007] The existing sMRI-DTI multimodal fusion scheme has the following problems: (1) The existing method ignores the differences in the anatomical microenvironment of long and short fibers and uniformly assigns fusion weights of sMRI and DTI to all connecting edges, resulting in a mismatch between the fusion weights and the physical meaning of the fiber type. Short fibers are close to the cortex and their orientation is highly dependent on the local cortical morphology (sMRI sensitive), while long fibers cross deep white matter and their integrity is determined by the myelin sheath and axon density (DTI sensitive). The uniform assignment strategy results in the complementary information of the two modalities not being used properly; (2) The existing scheme does not introduce tracking confidence to calibrate the fusion weights. Due to the cumulative tracking error, the false positive rate of long fibers is much higher than that of short fibers, but there is no suppression mechanism.

[0008] In summary, although random forest tracking provides high-quality upstream fiber data, long- and short-range microstructural differences provide biological evidence, and sMRI-DTI fusion studies validate the effectiveness of multimodal fusion, there are significant technical gaps among these studies: the confidence parameter of random forest tracking has not been used for downstream fusion, long- and short-range biological differences have not been translated into differentiated fusion weights, and existing fusion schemes lack false positive suppression mechanisms based on tracking confidence.

[0009] Therefore, in view of the shortcomings of existing technologies, there is an urgent need for a new method and system for calculating the long- and short-range connectivity of brain structures. Summary of the Invention

[0010] To address the problems existing in the prior art, this invention proposes a method and system for calculating long and short-range connectivity of brain structures. This method and system is based on the differential fusion of random forest fiber tracing and sMRI-DTI, and can overcome the technical defects existing in the prior art.

[0011] To achieve the above objectives, the present invention provides the following technical solution: A method for calculating the long-range and short-range connectivity of brain structures, characterized by comprising the following steps: S1: Based on diffusion tensor imaging data, streamlined fiber tracing is performed by combining a random forest classifier with neighborhood multi-point sampling voting to obtain the whole brain fiber bundle set and the global tracking confidence of each fiber bundle. S2: Based on structural magnetic resonance imaging data, perform brain atlas registration and brain region node division, and extract the anatomical morphological feature vectors of each brain region node; S3: Map fiber bundles to structural connection edges between brain region nodes, then calculate the bypass ratio based on the structural connection edges, and classify the structural connection edges into short-range connections and long-range connections based on the bypass ratio. Then, calculate the connectivity strength using a short-range fusion method that emphasizes structural magnetic resonance imaging data and a long-range fusion method that emphasizes diffusion tensor imaging data, respectively, to obtain the connectivity matrices of short-range connections and long-range connections.

[0012] Preferably, step S1 specifically includes: S11: Uniformly sample seed points from whole-brain white matter voxels in diffusion tensor imaging data; S12: Starting from the location of the seed point, generate multiple neighborhood sampling points; S13: Input the seed point and the neighborhood sampling point into the random forest classifier respectively to obtain the probability value of each point in each direction and the termination probability; S14: Perform a probability-weighted summation based on the probability values ​​of each point in each direction and the unit vector of the effective candidate directions to obtain the direction voting vector of each point; S15: Adaptively weighted aggregate the direction voting vectors of each point to obtain the weighted aggregated total direction vector, and determine the unit vector of the next forward direction based on the weighted aggregated total direction vector, and determine the single-step tracking confidence based on the weighted aggregated total direction vector and the unit vector of the previous forward direction. S16: Advance and update the position of the seed point in the next direction with a fixed step size to obtain the current position of the seed point; S17: Starting from the current position of the seed point, repeat steps S12-S16 until the termination condition is met, thereby obtaining the fiber bundle; S18: Determine the global tracking confidence of the fiber bundle based on the single-step tracking confidence.

[0013] Preferably, the termination condition in step S17 is any one of the following: the total direction vector after weighted aggregation is a zero vector; the number of termination votes exceeds 50% of the total number of sampling points; the current position moves out of the white matter region; the total length of the fiber bundle exceeds the maximum fiber length.

[0014] 4. Preferably, step S2 specifically includes: S21: Perform bias field correction on structural magnetic resonance imaging data to eliminate grayscale drift caused by field inhomogeneity; S22: Perform brain tissue segmentation on the bias-field rectified structural magnetic resonance imaging data, dividing it into three categories: gray matter, white matter, and cerebrospinal fluid, and generating a brain tissue probability map. S23: Align the probability map of brain tissue to the b=0 image space of diffusion tensor imaging data by linear affine transformation and nonlinear deformation registration; S24: The standard brain atlas is mapped to the structural magnetic resonance imaging data space through nonlinear registration, thereby obtaining the set of voxels contained in each brain region node in the structural magnetic resonance imaging data space. S25: Calculate the average cortical thickness, gray matter volume, and gyral index within each brain region node based on the set of voxels contained in each brain region node. S26: Determine the anatomical morphological feature vectors of each brain region node based on the average cortical thickness, gray matter volume, and gyral index.

[0015] Preferably, step S25 specifically includes: S251: For each voxel of a brain region node, search outward along the gray matter-white matter interface within the brain region node for the gray matter-cerebrospinal fluid interface. Measure the distance between the two interfaces in multiple directions and take the average of all measurements as the cortical thickness of the voxel. Then take the average of the cortical thicknesses of all voxels within the brain region node to obtain the average cortical thickness of the brain region node. S252: Count the number of gray matter voxels in a brain region node and multiply it by the volume of a single gray matter voxel to obtain the gray matter volume of the brain region node. S253: In the cortical region corresponding to the brain region node, extract the inner surface contour of the cortex along the gray-white matter interface and calculate its total length. Extract the outer surface contour along the gray-cerebrospinal fluid interface and calculate its total length. Use the ratio of the total length of the inner surface contour to the total length of the outer surface contour as the gyral index of the brain region node.

[0016] Preferably, step S3 specifically includes: S31: Map the two endpoints of each fiber bundle to the nearest brain region node respectively, count the set of fiber bundles between any two brain region nodes, and define a structural connection edge between the two brain region nodes if there is at least one fiber bundle between them. S32: Calculate the actual travel length of the fiber bundle and the Euclidean distance between brain regions for the structural connection edge, and obtain the detour ratio based on the actual travel length of the fiber bundle and the Euclidean distance between brain regions; S33: Based on the detour ratio and the preset threshold, determine whether the structural connection edge is a short-range connection or a long-range connection; S34: The connectivity strength of short-range and long-range connections is calculated by using a short-range fusion method that emphasizes structural magnetic resonance imaging data and a long-range fusion method that emphasizes diffusion tensor imaging data, respectively, to obtain the connectivity matrices of short-range and long-range connections.

[0017] Preferably, step S34 specifically includes: S341: Calculate the anatomical feature similarity between two brain region nodes:

[0018] In the formula, Representing brain region nodes With brain region nodes Anatomical feature similarity; and These represent brain region nodes. and brain region nodes Anatomical morphological feature vectors; express and The dot product; and They represent and The modulus length; S342: Calculate the average anisotropy fraction of the structural connection edges between two brain region nodes:

[0019]

[0020] In the formula, Representing brain region nodes With brain region nodes The average anisotropy fraction of the structural connecting edges between them; Representing brain region nodes With brain region nodes The total number of fiber bundles within the structural connection edges between them; Indicates the first The total number of voxels traversed by the root fiber bundle; Indicates the first The first fiber bundle passed through Anisotropy fraction of individual elements; Indicates the first The average of the anisotropy fractions of all voxels traversed by the root fiber bundle; S343: Based on the above and Determine the short-range fusion method for the structurally focused magnetic resonance imaging data:

[0021] In the formula, Representing brain region nodes With brain region nodes The structural connection edges between them represent the connectivity strength value for short-range connections. and To integrate the weighting coefficients, satisfy the following conditions: and ; S344: Based on the above and Determine the long-range fusion formula for the diffusion tensor imaging data:

[0022] in, Representing brain region nodes With brain region nodes The structural connection edges between them represent the connectivity strength value for long-range connections. Indicates the first Global tracking confidence of root fiber bundles; and To integrate the weighting coefficients, satisfy the following conditions: and .

[0023] Furthermore, the present invention also provides a computational system for tracking long- and short-range connectivity of brain structures, characterized in that it comprises: The fiber tracing module is used to perform streamlined fiber tracing based on diffusion tensor imaging data by combining a random forest classifier with neighborhood multi-point sampling voting, to obtain the whole brain fiber bundle set and the global tracing confidence of each fiber bundle. The brain region segmentation module is used to perform brain atlas registration and brain region node division based on structural magnetic resonance imaging data, and to extract the anatomical morphological feature vectors of each brain region node. The long-range and short-range fusion module maps fiber bundles to structural connection edges between brain region nodes, calculates the bypass ratio based on the structural connection edges, and classifies the structural connection edges into short-range and long-range connections based on the bypass ratio. Then, it calculates the connectivity strength using a short-range fusion method that emphasizes structural magnetic resonance imaging data and a long-range fusion method that emphasizes diffusion tensor imaging data, respectively, to obtain the connectivity matrices of short-range and long-range connections.

[0024] Furthermore, the present invention also provides a brain structure long- and short-range connectivity computing device, characterized in that it comprises: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the brain structure long- and short-range connectivity calculation method as described above.

[0025] Finally, the present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the steps of the brain structure long- and short-range connectivity calculation method as described above.

[0026] Compared with the prior art, the brain structure long-range and short-range connectivity calculation method and system of the present invention have one or more of the following beneficial technical effects: (1) In view of the problems in traditional mathematical model tracking schemes, such as the mathematical model assumptions not matching the actual biological tissues, loss of directional information in the fiber crossing area, reliance on manually set termination thresholds leading to premature termination of short-range fibers in superficial white matter, and lack of quantifiable tracking confidence, this invention adopts a fiber tracking method based on random forest and neighborhood multi-point sampling voting. It does not rely on mathematical model assumptions, does not require manually set termination thresholds, and can synchronously output the directional decision confidence of each step during the tracking process, thus solving the quality problem of upstream fiber trajectory reconstruction.

[0027] (2) In view of the problems that existing sMRI-DTI multimodal fusion schemes ignore the differences in the anatomical microenvironment of long and short fibers, uniformly assign fusion weights to all connection edges, and thus fail to make reasonable use of the complementary information of the two modes, this invention adopts an automatic classification method of long and short range based on the detour ratio. The structural connection edges are divided into short-range connections and long-range connections according to the ratio of the actual fiber running length to the Euclidean distance between brain regions. This lays the foundation for differentiated fusion and can solve the problems of spatial alignment of multimodal data and quantitative extraction of anatomical features.

[0028] (3) In view of the problems in the existing fusion schemes, such as the mismatch between fusion weight and the physical meaning of fiber type, the failure to utilize the sMRI sensitivity of short-range fibers and the DTI sensitivity of long-range fibers respectively, and the lack of suppression mechanism for false positives of long-range fibers, this invention adopts a long-short-range differentiated fusion strategy, so that short-range connections focus on sMRI anatomical features, long-range connections focus on the product of DTI microstructure indicators and tracking confidence, and introduces tracking confidence as a penalty factor in long-range fusion to suppress false positives, thereby improving the accuracy of fusion. Attached Figure Description

[0029] Figure 1 This is a flowchart of the brain structure long- and short-range connectivity calculation method of the present invention.

[0030] Figure 2 This is a schematic diagram of the brain structure long-range and short-range connectivity calculation system of the present invention.

[0031] Figure 3 This is a structural block diagram of the brain structure long-range and short-range connectivity computing device of the present invention. Detailed Implementation

[0032] Before detailing any embodiment of the invention, it should be understood that the invention, in its application, is not limited to the details of the construction and arrangement of the components set forth in the following description or illustrated in the following figures. The invention can have other embodiments and can be practiced or carried out in various ways. Furthermore, it should be understood that the wording and terminology used herein are for descriptive purposes and should not be considered limiting. The use of “comprising” or “having” and variations thereof in this invention is intended to cover the items set forth below and their equivalents, as well as any additional items. Unless otherwise specified or limited, the terms “installation,” “connection,” “support,” and “linkage,” and variations thereof are used broadly and cover both direct and indirect installation, connection, support, and linking. Moreover, “connection” and “linkage” are not limited to physical or mechanical connections or links.

[0033] Furthermore, firstly, in the disclosure of this invention, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting the invention. Secondly, the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple. The term "a" should not be construed as a limitation on the quantity.

[0034] Figure 1 A flowchart of the brain structure long- and short-range connectivity calculation method of the present invention is shown. Figure 1 As shown, the brain structure long-range and short-range connectivity calculation method of the present invention includes the following steps: S1: Fiber tracking.

[0035] Based on diffusion tensor imaging (DTI) data, streamlined fiber tracing was performed using a random forest classifier combined with neighborhood multi-point sampling voting to obtain the whole brain fiber bundle set and the global tracking confidence of each fiber bundle.

[0036] Specifically, the fiber tracking includes: 1. Sampling seed points.

[0037] Seed points are uniformly sampled from whole-brain white matter voxels in diffusion tensor imaging data. If a seed point is located near the boundary between white and gray matter, it is used as the starting point to advance bidirectionally along the principal direction of the tensor. A set of candidate directions is predefined. Since fibers follow forward, they theoretically only deflect forward and not directly turn backward; therefore, the search range is limited to the forward hemisphere, eliminating the need for a full 360° spatial search to reduce computation. Thus, the forward hemisphere space is uniformly discretized into 100 directions, each represented by a three-dimensional unit vector. In each decision step, directions with an angle less than a preset curvature threshold of 45 degrees to the previous direction are selected as valid candidate directions from these 100, while other directions are excluded.

[0038] 2. Generate neighborhood sampling points.

[0039] Starting from the location of the seed point, multiple neighborhood sampling points are generated.

[0040] Specifically, at the current position of the seed point Surrounding generation There are 1 sampling point, and the neighborhood sampling points are distributed in the following order: Center and radius of the sphere On a forward hemisphere of 0.25 to 0.5 times the size of a voxel, the orientation of the forward hemisphere is determined by the direction of movement in the previous step. The locations of the neighboring sampling points are then:

[0041] in, For the first The three-dimensional spatial coordinates of each sampling point =1,2,..., N ; The three-dimensional spatial coordinates of the current position of the seed point; The sampling radius is 0.25 to 0.5 times the voxel size, typically 0.3 mm to 0.6 mm. For the first Unit vectors in each sampling direction are uniformly and randomly distributed on the forward hemisphere. N The value range is 30 to 50, current position It also serves as a sampling point, corresponding to =0. The spread signal values ​​at all sampling points were obtained from the DTI data using trilinear interpolation.

[0042] 3. Single sampling point classification and directional voting.

[0043] The seed point and neighborhood sampling points are input into a random forest classifier to obtain the probability values ​​and termination probabilities of each point in each direction. Furthermore, a probability-weighted sum is performed based on the probability values ​​of each point in each direction and the unit vectors of valid candidate directions to obtain the direction voting vector for each point.

[0044] Specifically, for each sampling point Extract to The 3×3×3 neighborhood diffusion signal centered on the sample point and the previous moving direction are input into a pre-trained random forest classifier to obtain the directional probability value of that sample point. and termination probability For each sampling point, its opinion on the next step is represented as a directional voting vector:

[0045] in, For the first The directional voting vector of each sampling point is a three-dimensional vector; For the first The sampling point at the th sampling point The probability values ​​in each direction are output by the random forest classifier, ranging from 0 to 1, and the sum of the probability values ​​in all directions is 1. For the first The unit vectors of the _ ... If the termination threshold is exceeded, then... Let it be the zero vector. This indicates that voting at that sampling point has ended.

[0046] 4. Adaptive weighted aggregation.

[0047] Adaptive weighted aggregation of the directional voting vectors at each point is performed to obtain the weighted aggregated total directional vector, and the unit vector of the next forward direction is determined based on the weighted aggregated total directional vector.

[0048] Specifically, different sampling points should contribute equally to the final decision. Sampling points located ahead of the current position and closer to the fiber center should have higher weights. Therefore, the sampling point weights are:

[0049] in, For the first Adaptive weights for each sampling point j=1,2,...,N ; For the first The position offset vector of each sampling point ; This is the unit vector representing the direction of movement in the previous step; Indicates the first The projection of the offset vector of each sampling point onto the forward direction of the previous step, with values ​​ranging from negative to positive. r To Zheng r ; This is a bandwidth parameter that controls the rate at which the weights decay with angle. This is a natural exponential function. The formula means that the closer the sampling point is to the front of the streamline, the more... and The weight is higher when the direction is consistent, and lower when it is closer to the side or back. Current position weight .

[0050] The total direction vector after weighted aggregation is:

[0051] in, The total direction vector after weighted aggregation is a three-dimensional vector; For the first Adaptive weights for each sampling point; For the first The directional voting vector of each sampling point.

[0052] If all If all are zero vectors, then If the vector is zero, terminate the iteration. Otherwise, proceed in the following direction:

[0053] in, This is the unit vector indicating the direction of movement for the next step after normalization. for The modulus, i.e., the Euclidean norm. Dividing by the modulus is a normalization operation to ensure... It is a unit vector. If the number of termination votes is... The number of sampling points that are zero vectors exceeds the total number of sampling points. 50%, regardless The iteration terminates regardless of whether the vector is zero.

[0054] Furthermore, to prevent abrupt changes in direction between adjacent steps, this invention introduces a direction smoothing mechanism to perform multi-candidate smoothing of the step direction. Through this direction smoothing mechanism, the actual forward direction adopted for the next step is:

[0055] in, This is the unit vector representing the actual direction of movement for the next step after smoothing. The smoothing factor represents the weight percentage of the aggregation direction in the current step; This indicates the weight percentage of the direction of movement in the previous step; This represents the vector normalization operation. This smoothing mechanism allows for slow changes in direction as the fiber bends, avoiding drastic jitter caused by single-step noise.

[0056] 5. Advance.

[0057] The position of the seed point is updated by advancing along the next direction with a fixed step size to obtain the current position of the seed point.

[0058] Specifically, with a fixed step size along Direction advances and updates current position:

[0059] in, The updated current position coordinates; These are the position coordinates before the update; To maintain a fixed step size, The size of ploidytes is typically 0.5 mm to 0.75 mm. This is the unit vector of the smoothed forward direction.

[0060] 6. Termination of judgment.

[0061] Repeat steps 2 to 5 above, that is, repeat the neighborhood sampling, single-sampling point classification, directional voting, weighted aggregation, directional smoothing and advancement process until any of the following termination conditions are met: The following conditions must be met: the vector is zero; the number of termination votes exceeds 50% of the total number of sampling points; the current position moves out of the white matter region, i.e., the probability of being classified as a non-fiber by the random forest classifier is continuously greater than 0.8 for more than 5 consecutive steps; the total fiber length exceeds the maximum fiber length of 300mm. If the total fiber length is less than the minimum fiber length of 20mm, then the fiber is discarded.

[0062] Therefore, for a single seed point, a fiber bundle can be obtained. By repeating this process for all seed points, a set of fiber bundles from the entire brain can be obtained.

[0063] Meanwhile, in this invention, the single-step tracking confidence is calculated in each loop, and the calculation method is as follows:

[0064] in, This represents the confidence level for single-step tracking, with a value ranging from 0 to 1. This is the unit vector representing the weighted aggregation direction; This is the unit vector representing the direction of movement in the previous step; Let be the inverse cosine function, used to calculate the angle between two vectors, with values ​​ranging from 0 to π. The normalization coefficient maps the angle values ​​to the range of 0 to 1. This formula means that the smaller the angle between the total direction after weighted aggregation and the direction of movement in the previous step, the more consistent the voting direction is with the historical trend, and the higher the confidence level. Finally, the global fiber confidence level of the fiber bundle is:

[0065] in, The global confidence level for the entire fiber bundle ranges from 0 to 1. For step index, t=1, 2,...,T , T This represents the total number of steps in the fiber, i.e., the number of cycles. For the first The step size is a constant value. ; For the first The single-step tracking confidence score for each step. Since the step size is constant, this formula is equivalent to the arithmetic mean of the single-step tracking confidence scores for each step.

[0066] Therefore, after tracking all seed points, the whole brain fiber bundle set is output. F = {f 1 , f 2 , …, f M} Each fiber bundle consists of a sequence of three-dimensional spatial coordinate points, and the global tracking confidence level of each fiber bundle is also considered. .

[0067] In addition, in this invention, the random forest classifier can be pre-trained before use.

[0068] Before pre-training, DTI data needs to be acquired and preprocessed. Specifically, multi-directional diffusion-weighted DTI data of the subjects is acquired, with scanning parameters including at least 30 diffusion gradient directions and b values ​​ranging from 1000 to 3000 s / mm². Simultaneously, at least one undiffusion-weighted reference image with b=0 is acquired. Furthermore, eddy current correction is performed on the raw DTI data to eliminate image distortion caused by gradient switching; head motion correction is performed, and images in different diffusion directions are aligned to the b=0 image space through rigid registration; diffusion tensor fitting is completed, and the tensor matrix and anisotropy score of each voxel are calculated. The preprocessed data retains the original signal intensity values ​​of each voxel in each diffusion direction, providing input features for the subsequent random forest classifier.

[0069] In the pre-training of the random forest classifier, the training samples are constructed by taking a 3×3×3 neighborhood of 27 voxels centered on the current voxel and extracting the original diffusion-weighted signal intensity value of each voxel in all diffusion gradient directions. To eliminate the difference in the number of gradient directions between different scanning devices, the signal values ​​are resampled to 100 standard directions uniformly distributed across the entire hemisphere using spherical harmonic basis functions. The above signal features are concatenated with the unit vector of the previous step's forward direction to form the complete input feature vector of the random forest classifier. For the seed point position, since there is no previous forward direction, the direction vector is set to zero and used as a separate training sample category. The training label generation method uses the constrained spherical deconvolution deterministic tracking method to generate reference fiber bundles on the same data. The tangent direction of the reference fiber bundle at each voxel is used as the standard direction, and this direction is classified into the direction label with the smallest angle among the 100 discrete directions. The classification formula is:

[0070] in, The discrete orientation label number assigned to the current voxel, with a value ranging from 1 to 100; The unit vector of the tangent direction of the reference fiber at its current position; For the first A unit vector in a standard direction; The inverse cosine function is used to calculate the angle between two vectors. This indicates taking the absolute value, ensuring the non-directionality of the direction, i.e., the positive and negative directions of the fiber are equivalent; This means taking the value that makes the value inside the parentheses the smallest. .

[0071] A random forest classifier is trained using training sample features and corresponding orientation labels. The random forest classifier consists of 50 decision trees, each with a maximum depth of 30, and uses the Gini coefficient as the splitting criterion. After training, the random forest classifier outputs a 100-dimensional orientation probability distribution and a non-fiber termination probability for any input feature vector, where the orientation probability distribution satisfies the condition that the sum of all orientation probabilities is 1.

[0072] S2: Brain region segmentation.

[0073] Based on structural magnetic resonance imaging data, brain atlas registration and brain region node division were performed, and the anatomical morphological feature vectors of each brain region node were extracted.

[0074] Specifically, the brain region segmentation includes: 1. Structural magnetic resonance imaging (sMRI) data acquisition and preprocessing.

[0075] High-resolution T1-weighted sMRI data of the subjects were acquired. Offset field correction was performed on the raw sMRI data to eliminate grayscale drift caused by MRI field inhomogeneities, ensuring consistent signal intensity for the same tissue at different spatial locations. The bias-field-corrected structural MRI data was then segmented into three categories: gray matter, white matter, and cerebrospinal fluid, generating a brain tissue probability map to provide tissue labeling for subsequent brain atlas registration and anatomical feature extraction.

[0076] 2. Cross-modal space registration.

[0077] Preprocessed sMRI data, i.e., brain tissue probability maps, are precisely aligned to the b=0 image space of diffusion tensor imaging (DTI) data through linear affine transformation and nonlinear deformation registration. The linear affine transformation employs a 12-parameter affine model, including translation, rotation, scaling, and shearing in three directions, to correct for overall positional discrepancies between sMRI and DTI. The nonlinear deformation registration uses a B-spline-based free deformation model, optimizing deformation field parameters to maximize mutual information between sMRI and DTI, achieving precise correspondence between the two modalities in local anatomical structures. After registration, all voxels from sMRI and DTI are in the same three-dimensional coordinate system, ensuring spatial consistency between subsequent fiber tract endpoint mapping and brain region node segmentation.

[0078] 3. Brain map registration and brain region division.

[0079] Standard brain atlases were mapped to individual sMRI space using nonlinear registration. The standard brain atlas was selected from the AAL atlas (which divides the brain into 90 regions). The boundaries of each brain region in the standard brain atlas were deformed from the standard space to the individual sMRI space using nonlinear deformation, thereby obtaining the set of voxels contained in each brain region node in the individual sMRI space, and dividing the brain into N brain region nodes. .

[0080] 4. Extraction of anatomical morphological feature vectors.

[0081] Three metrics are calculated for each brain region node: mean cortical thickness, gray matter volume, and gyral index. These three values ​​are then used to construct the anatomical morphological feature vector of that brain region node.

[0082] The average cortical thickness is calculated by searching for the gray matter-cerebrospinal fluid interface outward from the gray matter-white matter interface within the brain region node, measuring the distance between the two interfaces in multiple directions, taking the average of all measurements as the cortical thickness of that voxel, and then taking the average of the cortical thicknesses of all voxels within the brain region node, with the unit being millimeters.

[0083] The gray matter volume is calculated by multiplying the number of gray matter voxels within a brain region node by the volume of a single voxel, with the unit being cubic millimeters. The gray matter volume reflects the total number of neuronal bodies in that brain region node and is a commonly used morphological indicator for measuring the size of brain region nodes. Let the number of gray matter voxels within a brain region node be... The volume of a single voxel is (Equal to the product of the resolutions of voxels in the x, y, and z directions), then the gray matter volume is:

[0084] in, The gray matter volume of a brain region node is expressed in mm³. The number of voxels that are segmented into gray matter within a brain region node; The volume of a single gray matter voxel. , These represent the resolution of a voxel in three directions, in mm. A larger volume usually indicates a higher number or density of neurons in that brain region.

[0085] The gyral index is calculated by taking the ratio of the inner surface length to the outer surface length of the cortex within a brain region node. The gyral index is a classic morphological indicator for measuring the degree of folding in the cerebral cortex, reflecting the complexity of the gyri and sulci. Within the cortical region corresponding to a brain region node, the inner surface contour is extracted along the gray-white matter interface, and its total length is calculated. The outer surface contour was extracted along the cortical gray matter-cerebrospinal fluid interface and its total length was calculated. The gyri index is:

[0086] in, The gyri index; The total length of the contour of the inner surface of the cortex (gray matter-white matter interface), in mm; This is the total length of the outline of the outer surface of the cortex (gray matter-cerebrospinal fluid interface), in mm. This ratio reflects the degree of gyri folding; a larger ratio indicates more complex gyri folding.

[0087] The above three indicators are used to construct the sMRI anatomical morphological feature vector for each brain region node:

[0088] in, For the first Anatomical morphological feature vectors (three-dimensional vectors) of brain region nodes. For the first The average cortical thickness of each brain region node, in mm; For the first The gray matter volume of each brain region node, in mm³; For the first The gyral index of each brain region node.

[0089] The final output is the whole brain. Set of anatomical morphological feature vectors of brain region nodes .

[0090] S3: Long-range and short-range fusion.

[0091] Fiber bundles are mapped to structural connection edges between brain region nodes. Then, the bypass ratio is calculated based on the structural connection edges, and the structural connection edges are classified into short-range connections and long-range connections based on the bypass ratio. Then, the connectivity strength is calculated by using a short-range fusion method that focuses on structural magnetic resonance imaging data and a long-range fusion method that focuses on diffusion tensor imaging data, respectively, to obtain the connectivity matrices of short-range connections and long-range connections.

[0092] In this invention, the long-range and short-range fusion specifically includes: 1. Construction of structural connection edges and mapping of fiber bundle endpoints.

[0093] Specifically, the two endpoints of each fiber bundle obtained in step S1 are mapped to the nearest brain region nodes identified in step S2. The distance between the fiber bundle endpoints and the brain region nodes is measured using Euclidean distance in three-dimensional space. This involves calculating the nearest distance between the fiber bundle endpoint and all voxels within each brain region node, and selecting the brain region node with the smallest distance as the node to which that endpoint belongs—that is, the nearest brain region node. The distance between any two brain region nodes is then statistically analyzed. and fiber bundles between ,like Non-empty means two brain region nodes and If there is at least one fiber bundle between two brain regions, then two brain region nodes are defined. and There is a structural connection edge between them. .

[0094] 2. Calculate the detour ratio.

[0095] For each structural connection edge, calculate the actual travel length of the fiber bundle and the Euclidean distance between brain regions. The actual travel length of the fiber bundle is taken as the average of the actual bending lengths of all fiber bundles within that structural connection edge. The actual bending length of a single fiber bundle is the sum of the straight-line distances between all adjacent points along its trajectory.

[0096] in, For the first f The actual travel length of the root fiber bundle, in mm; For the first fiber bundle Three-dimensional spatial coordinate points ( m=1,2,...,M ); This represents the total number of points on the fiber bundle; This represents the Euclidean distance between two adjacent points.

[0097] Actual travel length of fiber bundles at structural connection edges This is the average of the actual travel lengths of all fiber bundles within the connecting edge of the structure:

[0098] in, For structural connection edges The actual fiber travel length, in mm; This represents the total number of fiber bundles within the connecting edge of the structure. The first one in the connecting plate of this structure f The actual travel length of the root fiber bundle.

[0099] The Euclidean distance between brain regions is the three-dimensional straight-line distance between the centroids of two brain region nodes. The centroid of a brain region is the average of the coordinates of all voxels within that brain region node.

[0100] in, For the first The centroid coordinates of each brain region node are a three-dimensional coordinate vector. The unit is mm; For the first The total number of voxels within each brain region node; For the first The first brain region node Three-dimensional spatial coordinates of individual elements .

[0101] The Euclidean distance between the centroids of the two brain regions is:

[0102] in, brain region nodes With brain region nodes The Euclidean distance between the centroids, in mm; and These are the centroid coordinates of two brain region nodes.

[0103] Detour ratio It equals the actual length of the fiber divided by the Euclidean distance between brain regions:

[0104] in, For connecting edges The detour ratio; This represents the actual fiber travel length, in mm. This represents the Euclidean distance between brain regions, in mm. The closer to 1, the straighter the fiber path. The larger the value, the more the fiber loops around.

[0105] 3. Long-range and short-range classification.

[0106] Structural connecting edges are classified into two categories based on their geometric detour characteristics: short-range connections and long-range connections. That is, based on the detour ratio... Preset threshold for distribution throughout the brain Classify, if If the connection edge of the structure is determined to be a short-range connection, corresponding to a shallow white U-shaped fiber; if If it is determined to be a long-range connection, it corresponds to deep white matter projection fibers, joint fibers, and conjoint fibers.

[0107] 4. Calculation of connectivity strength.

[0108] After completing the long-range and short-range classification, this invention uses different fusion weight ratios for the connection edges of the two types of structures to achieve differentiated connectivity strength calculation.

[0109] Short-range connectivity is located in the superficial white matter layer, closely adhering to the cerebral cortex. Its course and distribution are closely related to the morphological characteristics of the local cortex. Therefore, the calculation of short-range connectivity strength should focus on sMRI anatomical features. Specifically, short-range connectivity strength is equal to the weighted sum of the cosine similarity of the sMRI anatomical morphological feature vectors of the two brain regions and the average FA value of the structural connectivity edges.

[0110] The formula for calculating the cosine similarity of sMRI anatomical morphological feature vectors of two brain regions is as follows:

[0111] in, Representing brain region nodes With brain region nodes sMRI anatomical feature similarity; and These represent brain region nodes. and brain region nodes The anatomical morphological feature vectors are all three-dimensional vectors. ; express and The dot product; and They represent and The modulus length. This cosine similarity reflects the overall similarity between two brain region nodes in three anatomical dimensions: cortical thickness, gray matter volume, and gyral index. The closer the value is to 1, the more similar the anatomical morphology of the two brain region nodes is.

[0112] While obtaining sMRI anatomical similarity, it is also necessary to consider the microstructural information provided by DTI. The average FA value of structural connector edges reflects the average level of anisotropy fraction of all fiber bundles within that structural connector edge on each voxel, and is an important indicator for measuring the integrity of white matter fiber microstructure. Its calculation formula is as follows:

[0113]

[0114] In the formula, Representing brain region nodes With brain region nodes The structural connection edges between The average of the anisotropy fractions of all voxels through which all fiber bundles pass, with a value ranging from 0 to 1; Representing brain region nodes With brain region nodes The total number of fiber bundles within the structural connection edges between them; Indicates the first The total number of voxels traversed by the root fiber bundle; Indicates the first The first fiber bundle passed through Anisotropy fraction of individual elements; Indicates the first The average of the anisotropy fractions of all voxels through which the root fiber bundle passes.

[0115] The above sMRI anatomical similarity With DTI microstructure index By performing a weighted combination, the short-range connectivity strength is obtained, calculated using the following formula:

[0116] In the formula, Representing brain region nodes With brain region nodes The structural connection edges between This represents the connectivity strength value for short-range connections, ranging from 0 to 1. and To integrate the weighting coefficients, satisfy the following conditions: and This formula assigns higher weight to sMRI anatomical similarity, allowing the calculation of short-range connectivity strength to fully reflect its close association with cortical morphology.

[0117] For long-range connectivity, since it traverses deep white matter and is far from the cortex, its integrity is mainly determined by the degree of myelination and axon density. Therefore, the calculation of long-range connectivity strength should focus on DTI microstructural indices. Meanwhile, long-range fibers have a much higher false positive rate than short-range fibers due to their wide traversal area and large cumulative tracking error. Therefore, it is necessary to introduce the global tracking confidence obtained in step S1 to penalize the reliability of each fiber bundle. The long-range connectivity strength is equal to the average FA of all fiber bundles within the structural connection edge multiplied by the average global confidence of that fiber bundle and the weighted sum of the similarity of sMRI anatomical features between the two brain regions. The calculation formula is as follows:

[0118] in, Representing brain region nodes With brain region nodes The structural connection edges between This represents the connectivity strength value for long-distance connections, ranging from 0 to 1. Indicates the first The global tracking confidence level of the root fiber bundle, with a value ranging from 0 to 1; It serves as a comprehensive indicator of microstructure integrity and tracking reliability; and To integrate the weighting coefficients, satisfy the following conditions: and This formula, while retaining sMRI anatomical information as an auxiliary reference, effectively suppresses the weight of unreliable long-range fibers in the fusion calculation through the global tracking confidence product term, thereby reducing the false positive rate of long-range connectivity.

[0119] After obtaining the short-range and long-range connectivity strength values, the final output is a bistructural connectivity matrix. The short-range structural connectivity matrix is ​​an N x N matrix, where the rows and columns represent brain region node numbers. For all structural connections identified as short-range, the corresponding position is filled with the value of that structural connection. The value is filled in, and 0 is filled in other positions. The long-range structural connectivity matrix is ​​also an N x N matrix. For all structural connections that are determined to be long-range, fill in the corresponding position with the value of that structural connection. Enter the given value, and fill in 0 for the rest. The two matrices are the final output, which can be directly used for subsequent brain network analysis, disease classification, or cognitive prediction.

[0120] Figure 2 A schematic diagram of the brain structure long- and short-range connectivity calculation system of the present invention is shown. Figure 2 As shown, the brain structure long-range and short-range connectivity calculation system of the present invention includes: 1. Fiber tracking module.

[0121] The module is used to perform streamlined fiber tracing based on diffusion tensor imaging data, using a random forest classifier combined with neighborhood multi-point sampling voting, to obtain the whole brain fiber bundle set and the global tracking confidence of each fiber bundle.

[0122] 2. Brain region segmentation module.

[0123] The brain region segmentation module is used to perform brain map registration and brain region node division based on structural magnetic resonance imaging data, and to extract the anatomical morphological feature vectors of each brain region node.

[0124] 3. Long-range and short-range fusion module.

[0125] The long-range and short-range fusion module is used to map fiber bundles as structural connection edges between brain region nodes, then calculate the bypass ratio based on the structural connection edges, and classify the structural connection edges into short-range connections and long-range connections based on the bypass ratio. Then, the connectivity strength is calculated by using a short-range fusion method that focuses on structural magnetic resonance imaging data and a long-range fusion method that focuses on diffusion tensor imaging data, respectively, to obtain the connectivity matrices of short-range connections and long-range connections.

[0126] Furthermore, this invention also provides a device for calculating the long- and short-range connectivity of brain structures. For example... Figure 3As shown, the brain structure long- and short-range connectivity computing device of the present invention includes: a memory 11 for storing one or more programs; one or more processors 12; when the one or more programs are executed by the one or more processors 12, the one or more processors 12 implement the brain structure long- and short-range connectivity computing method of the present invention.

[0127] Finally, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the brain structure long- and short-range connectivity calculation method of the present invention.

[0128] The computer-readable storage medium includes both permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device. As defined in this invention, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0129] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0130] The steps of the methods or algorithms described in conjunction with the embodiments disclosed in this invention can be implemented in hardware, software modules executed by a processor, or a combination of both. The software modules can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Those skilled in the art can modify or make equivalent substitutions to the technical solutions of the present invention based on the concept of the present invention, without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for calculating the long- and short-range connectivity of brain structures, characterized in that, Includes the following steps: S1: Based on diffusion tensor imaging data, streamlined fiber tracing is performed by combining a random forest classifier with neighborhood multi-point sampling voting to obtain the whole brain fiber bundle set and the global tracking confidence of each fiber bundle. S2: Based on structural magnetic resonance imaging data, perform brain atlas registration and brain region node division, and extract the anatomical morphological feature vectors of each brain region node; S3: Map fiber bundles to structural connection edges between brain region nodes, then calculate the bypass ratio based on the structural connection edges, and classify the structural connection edges into short-range connections and long-range connections based on the bypass ratio. Then, calculate the connectivity strength using a short-range fusion method that emphasizes structural magnetic resonance imaging data and a long-range fusion method that emphasizes diffusion tensor imaging data, respectively, to obtain the connectivity matrices of short-range connections and long-range connections.

2. The method for calculating long- and short-range connectivity of brain structures according to claim 1, characterized in that, Step S1 specifically includes: S11: Uniformly sample seed points from whole-brain white matter voxels in diffusion tensor imaging data; S12: Starting from the location of the seed point, generate multiple neighborhood sampling points; S13: Input the seed point and the neighborhood sampling point into the random forest classifier respectively to obtain the probability value of each point in each direction and the termination probability; S14: Perform a probability-weighted summation based on the probability values ​​of each point in each direction and the unit vector of the effective candidate directions to obtain the direction voting vector of each point; S15: Adaptively weighted aggregate the direction voting vectors of each point to obtain the weighted aggregated total direction vector, and determine the unit vector of the next forward direction based on the weighted aggregated total direction vector, and determine the single-step tracking confidence based on the weighted aggregated total direction vector and the unit vector of the previous forward direction. S16: Advance and update the position of the seed point in the next direction with a fixed step size to obtain the current position of the seed point; S17: Starting from the current position of the seed point, repeat steps S12-S16 until the termination condition is met, thereby obtaining the fiber bundle; S18: Determine the global tracking confidence of the fiber bundle based on the single-step tracking confidence.

3. The method for calculating long- and short-range connectivity of brain structures according to claim 2, characterized in that, The termination condition in step S17 is any one of the following: the total direction vector after weighted aggregation is zero; the number of termination votes exceeds 50% of the total number of sampling points; the current position moves out of the white matter region; the total length of the fiber bundle exceeds the maximum fiber length.

4. The method for calculating long- and short-range connectivity of brain structures according to claim 1, characterized in that, Step S2 specifically includes: S21: Perform bias field correction on structural magnetic resonance imaging data to eliminate grayscale drift caused by field inhomogeneity; S22: Perform brain tissue segmentation on the bias-field rectified structural magnetic resonance imaging data, dividing it into three categories: gray matter, white matter, and cerebrospinal fluid, and generating a brain tissue probability map. S23: Align the probability map of brain tissue to the b=0 image space of diffusion tensor imaging data by linear affine transformation and nonlinear deformation registration; S24: The standard brain atlas is mapped to the structural magnetic resonance imaging data space through nonlinear registration, thereby obtaining the set of voxels contained in each brain region node in the structural magnetic resonance imaging data space. S25: Calculate the average cortical thickness, gray matter volume, and gyral index within each brain region node based on the set of voxels contained in each brain region node. S26: Determine the anatomical morphological feature vectors of each brain region node based on the average cortical thickness, gray matter volume, and gyral index.

5. The method for calculating long- and short-range connectivity of brain structures according to claim 4, characterized in that, Step S25 specifically includes: S251: For each voxel of a brain region node, search outward along the gray matter-white matter interface within the brain region node for the gray matter-cerebrospinal fluid interface. Measure the distance between the two interfaces in multiple directions and take the average of all measurements as the cortical thickness of the voxel. Then take the average of the cortical thicknesses of all voxels within the brain region node to obtain the average cortical thickness of the brain region node. S252: Count the number of gray matter voxels in a brain region node and multiply it by the volume of a single gray matter voxel to obtain the gray matter volume of the brain region node. S253: In the cortical region corresponding to the brain region node, extract the inner surface contour of the cortex along the gray-white matter interface and calculate its total length. Extract the outer surface contour along the gray-cerebrospinal fluid interface and calculate its total length. Use the ratio of the total length of the inner surface contour to the total length of the outer surface contour as the gyral index of the brain region node.

6. The method for calculating long- and short-range connectivity of brain structures according to claim 1, characterized in that, Step S3 specifically includes: S31: Map the two endpoints of each fiber bundle to the nearest brain region node respectively, count the set of fiber bundles between any two brain region nodes, and define a structural connection edge between the two brain region nodes if there is at least one fiber bundle between them. S32: Calculate the actual travel length of the fiber bundle and the Euclidean distance between brain regions for the structural connection edge, and obtain the detour ratio based on the actual travel length of the fiber bundle and the Euclidean distance between brain regions; S33: Based on the detour ratio and the preset threshold, determine whether the structural connection edge is a short-range connection or a long-range connection; S34: The connectivity strength of short-range and long-range connections is calculated by using a short-range fusion method that emphasizes structural magnetic resonance imaging data and a long-range fusion method that emphasizes diffusion tensor imaging data, respectively, to obtain the connectivity matrices of short-range and long-range connections.

7. The method for calculating long- and short-range connectivity of brain structures according to claim 6, characterized in that, Step S34 specifically includes: S341: Calculate the anatomical feature similarity between two brain region nodes: In the formula, Representing brain region nodes With brain region nodes Anatomical feature similarity; and These represent brain region nodes. and brain region nodes Anatomical morphological feature vectors; express and The dot product; and They represent and The modulus length; S342: Calculate the average anisotropy fraction of the structural connection edges between two brain region nodes: In the formula, Representing brain region nodes With brain region nodes The average anisotropy fraction of the structural connecting edges between them; Representing brain region nodes With brain region nodes The total number of fiber bundles within the structural connection edges between them; Indicates the first The total number of voxels traversed by the root fiber bundle; Indicates the first The first fiber bundle passed through Anisotropy fraction of individual elements; Indicates the first The average of the anisotropy fractions of all voxels traversed by the root fiber bundle; S343: Based on the above and Determine the short-range fusion method for the structurally focused magnetic resonance imaging data: In the formula, Representing brain region nodes With brain region nodes The structural connection edges between them represent the connectivity strength value for short-range connections. and To integrate the weighting coefficients, satisfy the following conditions: and ; S344: Based on the above and Determine the long-range fusion formula for the diffusion tensor imaging data: in, Representing brain region nodes With brain region nodes The structural connection edges between them represent the connectivity strength value for long-range connections. Indicates the first Global tracking confidence of root fiber bundles; and To integrate the weighting coefficients, satisfy the following conditions: and .

8. A brain structure long- and short-range connectivity calculation system, characterized in that, include: The fiber tracing module is used to perform streamlined fiber tracing based on diffusion tensor imaging data by combining a random forest classifier with neighborhood multi-point sampling voting, to obtain the whole brain fiber bundle set and the global tracing confidence of each fiber bundle. The brain region segmentation module is used to perform brain atlas registration and brain region node division based on structural magnetic resonance imaging data, and to extract the anatomical morphological feature vectors of each brain region node. The long-range and short-range fusion module maps fiber bundles to structural connection edges between brain region nodes, calculates the bypass ratio based on the structural connection edges, and classifies the structural connection edges into short-range and long-range connections based on the bypass ratio. Then, it calculates the connectivity strength using a short-range fusion method that emphasizes structural magnetic resonance imaging data and a long-range fusion method that emphasizes diffusion tensor imaging data, respectively, to obtain the connectivity matrices of short-range and long-range connections.

9. A brain structure long- and short-range connectivity computing device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the brain structure long- and short-range connectivity calculation method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the brain structure long- and short-range connectivity calculation method as described in any one of claims 1-7.