A shape feature-based subcortical individual morphological brain network construction method

Abstract

The application relates to the field of data processing and discloses a shape-feature-based subcortical individual morphological brain network construction method, which comprises the following steps: acquiring a structural magnetic resonance image of a subject, and extracting a binary mask therefrom; respectively performing shape modeling on each binary mask to obtain an individualized shape grid of each subcortical region; calculating the deformation variables of all vertices of each shape grid relative to corresponding vertices in an average template to obtain a deformation variable set of each subcortical region; taking a plurality of subcortical regions as network nodes, calculating the probability distribution similarity of any two deformation variable sets according to the deformation variable set of each subcortical region, taking the similarity as the edge strength of the two subcortical regions, constructing a subcortical individual morphological brain network, extracting network features of the brain network, and constructing a classification model according to the network features for the classification of brain images. The application improves the construction precision of the subcortical morphological network and the classification precision of brain images.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, terminal, and computer-readable storage medium for constructing subcortical individual morphological brain networks based on shape features. Background Technology

[0002] Morphological brain networks, constructed based on statistical correlations of morphological features between brain regions, are an important branch of network neuroscience and have been widely applied in research on brain structural development and the pathological mechanisms of neurological diseases. Their development has shifted from population-level covariance networks to individualized modeling. Individualized morphological networks offer advantages such as high robustness, reproducibility, and low cost, making them more suitable for multi-center, large-sample studies and clinical translation.

[0003] Currently, numerous studies have confirmed the early abnormalities of subcortical structures in neurological diseases. As a core hub for brain function regulation, the correlation between the morphological characteristics of subcortical structures and cognitive function, as well as the pathological mechanisms of neurological diseases, has become an important direction in the field of neuroimaging. However, current methods for constructing individualized morphological brain networks mainly focus on the cortical level, while systematic research on subcortical morphological networks remains scarce, and existing methods are limited to population-level construction, resulting in low accuracy and reliability of the constructed subcortical morphological networks.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, terminal, and computer-readable storage medium for constructing subcortical morphological brain networks based on shape features, aiming to solve the problems of insufficient research, inadequate construction accuracy, and limited clinical translational value in the existing technology of subcortical morphological networks.

[0006] To achieve the above objectives, this invention provides a method for constructing a shape-feature-based subcortical individual morphological brain network, which includes the following steps:

[0007] Structural magnetic resonance images of the subject are acquired, and the structural magnetic resonance images are preprocessed and segmented. Binary masks of multiple subcortical regions are extracted from the segmentation results.

[0008] Based on a deep learning model, the shape of the binary mask for each subcortical region is modeled to obtain the individualized shape mesh corresponding to each subcortical region.

[0009] Calculate the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template to obtain the deformation set of each subcortical region;

[0010] Using multiple subcortical regions as network nodes, the similarity of probability distributions corresponding to the deformation sets of any two subcortical regions is calculated based on the deformation set of each subcortical region, and the similarity is used as the connection strength between the two subcortical regions to construct a subcortical individual morphological brain network.

[0011] The network features of the subcortical individual morphological brain network are extracted, a brain image classification model is constructed based on the network features, and a target brain image is obtained. The target brain image is classified based on the brain image classification model to obtain the classification result of the target brain image.

[0012] Optionally, the method for constructing a shape-feature-based subcortical individual morphological brain network includes the following subcortical regions: the caudate nucleus, putamen, thalamus, globus pallidus, hippocampus, amygdala, and nucleus accumbens of both hemispheres.

[0013] Optionally, the method for constructing individual subcortical morphological brain networks based on shape features, wherein the step of performing shape modeling on the binary mask of each subcortical region based on a deep learning model to obtain an individualized shape mesh corresponding to each subcortical region specifically includes:

[0014] Based on the binary masks of all subcortical regions, an average mask corresponding to each subcortical region is constructed, and the average mask is converted into a three-dimensional template shape model by the moving cube algorithm. The three-dimensional template shape model is used as the starting reference mesh for shape modeling.

[0015] Align the surface point cloud of the binary mask for each subcortical region of the subject with the corresponding 3D template shape model to obtain the initial shape point cloud for each subcortical region;

[0016] Using the PointNet deep learning model based on a multilayer perceptron architecture, the initial shape point cloud of each subcortical region is iteratively optimized. By predicting the displacement vector of each vertex, the corresponding 3D template shape model is made to fit the individual target point cloud, thus obtaining the individualized shape mesh corresponding to each subcortical region.

[0017] Optionally, the method for constructing individual subcortical morphological brain networks based on shape features, wherein calculating the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template to obtain the deformation set of each subcortical region specifically includes:

[0018] Based on all individualized shape grids, an average template grid corresponding to each subcortical region is constructed, and the average template grid is used as a statistical reference grid for deformation calculation;

[0019] For each subcortical region, a correspondence is established between the vertices of the individualized shape mesh and the corresponding vertices of the average template mesh;

[0020] Calculate the signed Euclidean distance of each vertex in the individualized shape mesh relative to the corresponding vertex in the average template mesh, and use the signed Euclidean distance as the deformation of the vertex, wherein the sign is determined by the dot product of the vertex displacement vector and the surface normal vector of the average template mesh;

[0021] By aggregating the deformations of all vertices within each subcortical region, we obtain the deformation set for each subcortical region.

[0022] Optionally, the method for constructing a shape-feature-based subcortical individual morphological brain network, wherein the step of using multiple subcortical regions as network nodes and calculating the probability distribution similarity between any two subcortical regions' shape variable sets based on the shape variable set of each subcortical region specifically includes:

[0023] By evaluating the impact of different sampling point numbers on the stability of the probability density function, and using the Fraser distance to quantify the differences in the probability density function under different sampling point numbers, the optimal number of sampling points for kernel density estimation is determined.

[0024] Based on the optimal number of sampling points, kernel density estimation is performed on the set of deformable variables corresponding to each subcortical region to obtain the probability density function of the set of deformable variables at the sampling points;

[0025] For any two subcortical regions, calculate the Jensen-Shannon divergence between the probability density functions, convert the Jensen-Shannon divergence into Jensen-Shannon similarity, and use the Jensen-Shannon similarity as the network edge strength between the two subcortical regions.

[0026] Optionally, the method for constructing a subcortical individual morphological brain network based on shape features, wherein the construction of the subcortical individual morphological brain network by using the similarity as the connection strength between two subcortical regions, further includes:

[0027] Graph theory analysis was performed on the subcortical individual morphological brain network to obtain the network's graph theory indices. Based on the graph theory indices, the network's global and node attributes were calculated, and the test-retest reliability of the global and node attributes was quantified using the intragroup correlation coefficient.

[0028] The global attributes and node attributes are filtered to obtain the global attributes and node attributes whose test-retest reliability meets the preset conditions, and the global attributes and node attributes whose test-retest reliability meets the preset conditions are used as high reliability indicators.

[0029] The global attributes include clustering coefficient, shortest path length, local efficiency, global efficiency, and modularity, while the node attributes include node degree, node efficiency, and betweenness centrality.

[0030] Optionally, the method for constructing subcortical individual morphological brain networks based on shape features, wherein the global attribute and node attribute whose test-retest reliability meets preset conditions are used as high reliability indicators, further includes:

[0031] Based on the aforementioned subcortical individual morphological brain network and the aforementioned high reliability index, a comparative analysis of clinical application was conducted between the case group and the healthy control group. Intergroup statistical tests were performed on the network edge strength, global attributes, and node attributes, and multiple tests were performed using the false discovery rate to screen out network features with significant differences.

[0032] Furthermore, to achieve the above objectives, the present invention also provides a shape-feature-based system for constructing subcortical individual morphological brain networks, wherein the shape-feature-based system for constructing subcortical individual morphological brain networks includes:

[0033] The binary mask extraction module is used to acquire the structural magnetic resonance image of the subject, preprocess and segment the structural magnetic resonance image, and extract the binary mask of multiple subcortical regions from the segmentation result.

[0034] The shape modeling module is used to perform shape modeling on the binary mask of each subcortical region based on the deep learning model, so as to obtain the individualized shape mesh corresponding to each subcortical region.

[0035] The deformation calculation module is used to calculate the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template, so as to obtain the deformation set of each subcortical region;

[0036] The brain network construction module is used to construct an individual subcortical morphological brain network by using multiple subcortical regions as network nodes, calculating the probability distribution similarity between any two subcortical regions based on the set of deformation variables of each subcortical region, and using the similarity as the connection strength between the two subcortical regions.

[0037] The classification application module is used to extract network features of the subcortical individual morphological brain network, construct a brain image classification model based on the network features, acquire a target brain image, classify the target brain image based on the brain image classification model, and obtain the classification result of the target brain image.

[0038] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a shape feature-based subcortical individual morphological brain network construction program stored in the memory and executable on the processor, wherein when the shape feature-based subcortical individual morphological brain network construction program is executed by the processor, it implements the steps of the shape feature-based subcortical individual morphological brain network construction method as described above.

[0039] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a shape-feature-based subcortical individual morphological brain network construction program, which, when executed by a processor, implements the steps of the shape-feature-based subcortical individual morphological brain network construction method described above.

[0040] In this invention, structural magnetic resonance imaging (SMRI) images of subjects are acquired, preprocessed, and segmented. Multiple binary masks of subcortical regions are extracted from the segmentation results. Based on a deep learning model, shape modeling is performed on the binary mask of each subcortical region to obtain an individualized shape mesh corresponding to each region. The deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template is calculated to obtain the deformation set of each subcortical region. Using multiple subcortical regions as network nodes, the probability distribution similarity between any two deformation sets of subcortical regions is calculated based on the deformation set of each region. This similarity is used as the connection strength between the two subcortical regions to construct a subcortical individual morphological brain network. Network features of the brain network are extracted, and a classification model is constructed based on these features for brain image classification. The subcortical individual shape network constructed in this invention exhibits good test-retest stability, effectively reproduces the small-world properties and modular features of healthy individuals' brain networks, and improves the classification accuracy of brain images. Attached Figure Description

[0041] Figure 1 This is a flowchart of a preferred embodiment of the method for constructing subcortical individual morphological brain networks based on shape features according to the present invention;

[0042] Figure 2 This is a structural diagram of a preferred embodiment of the shape-feature-based subcortical individual morphological brain network construction system of the present invention;

[0043] Figure 3 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0044] This application provides a method, system, and terminal for constructing subcortical individual morphological brain networks based on shape features. To make the objectives, technical solutions, and effects of this application clearer and more explicit, the following detailed description is provided with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application.

[0045] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0046] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0047] The preferred embodiment of the present invention describes a method for constructing subcortical individual morphological brain networks based on shape features, such as... Figure 1 As shown, the method for constructing a shape-feature-based subcortical individual morphological brain network includes the following steps:

[0048] Step S10: Obtain the structural magnetic resonance image of the subject, and preprocess and segment the structural magnetic resonance image to extract a binary mask of multiple subcortical regions from the segmentation result.

[0049] In this embodiment, the online brain MRI (Magnetic Resonance Imaging) volume analysis platform volBrain was used to automatically preprocess and segment the T1-weighted images from the structural MRI, extracting binary masks for 14 subcortical regions in the two hemispheres, including the caudate nucleus, putamen, thalamus, globus pallidus, hippocampus, amygdala, and nucleus accumbens. The segmentation results were then subjected to quality control and verification to ensure the accuracy of subcortical structure identification.

[0050] Specifically, three-dimensional T1-weighted structural magnetic resonance images of the subjects were acquired. To ensure data quality, the acquired images underwent preprocessing, including: spatial normalization to register the images to standard anatomical space; field correction to eliminate grayscale differences caused by magnetic field inhomogeneity; and denoising to improve the image signal-to-noise ratio. The preprocessed images were then input into the automated segmentation platform volBrain for fine segmentation of brain structures. This platform, based on multi-map registration and patch fusion technology, performs voxel-by-voxel labeling on the images, accurately segmenting deep subcortical structures in both hemispheres, such as the caudate nucleus, putamen, thalamus, globus pallidus, hippocampus, amygdala, and nucleus accumbens. After segmentation, a corresponding binary mask was generated for each subcortical structure, i.e., a three-dimensional binary image where the voxel containing the region is labeled 1 and the remaining voxels are labeled 0. All segmentation results underwent rigorous quality control and were visually verified by experienced radiologists to ensure the accuracy and integrity of the boundaries of each subcortical structure. Ultimately, each subject obtained binary masks for 14 subcortical regions, which served as input data for subsequent shape modeling.

[0051] Step S20: Based on the deep learning model, shape modeling is performed on the binary mask of each subcortical region to obtain the individualized shape mesh corresponding to each subcortical region.

[0052] Specifically, based on the binary masks of all subcortical regions, an average mask corresponding to each subcortical region is constructed, and the average mask is converted into a three-dimensional template shape model through the moving cube algorithm. The three-dimensional template shape model is used as the starting reference mesh for shape modeling.

[0053] In this embodiment, an average mask corresponding to the same subcortical region is constructed based on the binary mask of all subjects. Specifically, the binary masks of all subjects are superimposed in a standard space, and the probability of each voxel being labeled as the region is calculated. By setting a threshold (e.g., probability greater than 0.5), the population-level average mask of the region is obtained. This average mask reflects the typical spatial distribution and morphological characteristics of the subcortical structure in the population. Subsequently, the moving cube algorithm is used to convert the average mask into a three-dimensional mesh model. That is, by extracting isosurfaces, the voxelized binary image is transformed into a three-dimensional geometric mesh composed of multiple vertices and faces connecting the vertices, thus obtaining the three-dimensional template shape model corresponding to the subcortical region.

[0054] The 3D template shape model serves as the initial reference mesh for subsequent individual shape modeling. Its function is to provide a standardized geometric benchmark for each subcortical region, ensuring that shape modeling for different subjects and different regions has a unified topological structure and vertex correspondence. This template model not only preserves the typical morphological features of the subcortical structure but also optimizes the uniformity of vertex distribution through adaptive remeshing, providing a high-quality initial mesh for subsequent deep learning-based individualized deformation optimization.

[0055] Furthermore, the surface point cloud of the binary mask for each subcortical region of the subject is aligned with the corresponding 3D template shape model to obtain the initial shape point cloud corresponding to each subcortical region.

[0056] Understandably, after obtaining the 3D template shape model corresponding to each subcortical region, for each subcortical region of the same subject, the generated binary mask is used to extract surface point clouds, obtaining a discrete point cloud representation of the individual morphology of that region. Subsequently, the iterative nearest point algorithm is used to rigidly register the surface point cloud of that region with the corresponding 3D template shape model. By continuously iterating and optimizing the rotation matrix and translation vector, the Euclidean distance between the point cloud and the corresponding points of the template model is minimized, so that the individual point cloud is initially aligned with the template model in spatial position. After the above registration, the individual point cloud is transformed into the coordinate space of the template model, while retaining the original morphological information, forming the initial shape point cloud corresponding to that subcortical region.

[0057] Furthermore, using the PointNet deep learning model based on a multilayer perceptron architecture, the initial shape point cloud of each subcortical region is iteratively optimized, and the corresponding 3D template shape model is made to fit the individual target point cloud by predicting the displacement vector of each vertex, thus obtaining the individualized shape mesh corresponding to each subcortical region.

[0058] In this embodiment, the initial point cloud is then optimized using the PointNet architecture (a landmark neural network that directly processes unordered 3D point clouds) based on a multilayer perceptron (MLP). Specifically, the template mesh data (containing vertex coordinates and face indices) and individualized target point clouds of the subjects to be processed are first loaded. All data are converted into tensors, and the PointNetOpt (an optimized version of PointNet) deep learning model, AdamW optimizer (an improved version of Adam optimizer with weight decay), and MSELoss (Mean Squared Error) are initialized. Loss functions such as mean squared error loss are used. After setting the maximum number of iterations (default 5001), the optimization loop is started. In each iteration, the vertex features of the current template mesh are extracted and input into the model. The model predicts the 3D displacement vector of each vertex. The displacement vector is superimposed on the original vertex coordinates to obtain the updated vertex set. Then, the distance loss and regularization loss between the updated vertices and the original vertices are calculated. The model parameters are optimized through backpropagation to reduce the loss. At the same time, the learning rate is gradually reduced with the help of the learning rate scheduler. During the iteration process, the loss log is printed every 100 rounds and the updated vertex coordinates are saved every 500 rounds until all iterations are completed. Finally, after continuous displacement prediction and accumulation, the vertices of the template mesh accurately fit the individualized shape of the target point cloud. Moreover, the MSELoss loss constraint avoids the vertices from deviating too much from their original positions, ensuring the physiological rationality of the subcortical structure morphology.

[0059] Step S30: Calculate the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template to obtain the deformation set of each subcortical region.

[0060] Specifically, based on all individualized shape grids, an average template grid corresponding to each subcortical region is constructed, and the average template grid is used as a statistical reference grid for deformation calculation.

[0061] Understandably, after completing individualized shape modeling for all subjects, for each subcortical region, the individualized shape meshes of all subjects for that region are spatially aligned, and the arithmetic mean of the coordinates for each vertex is calculated to construct the average template mesh corresponding to that region. This average template mesh consists of multiple vertices and faces connecting the vertices, representing the standard morphology of the subcortical structure at the population level, providing a foundation for subsequent deformation analysis and network construction.

[0062] Furthermore, for each subcortical region, a correspondence is established between the vertices of the individualized shape mesh and the corresponding vertices of the average template mesh.

[0063] It is understandable that for each subcortical region, since the individualized shape mesh and the average template mesh have the same topological structure, there is a one-to-one correspondence between their vertices. Therefore, each vertex of the individualized mesh can be directly mapped to the indirectly indexed vertices in the average template mesh, thereby determining the displacement vector of each vertex between the individual and the template.

[0064] Furthermore, the signed Euclidean distance of each vertex in the individualized shape mesh relative to the corresponding vertex in the average template mesh is calculated, and the signed Euclidean distance is used as the deformation of the vertex.

[0065] By aggregating the deformations of all vertices within each subcortical region, we obtain the deformation set for each subcortical region.

[0066] The symbol is determined by the dot product of the vertex displacement vector and the surface normal vector of the average template mesh.

[0067] In this embodiment, the signed Euclidean distance from each vertex in the individual mesh to the corresponding vertex of the average template is used as a local deformation. The sign of the signed Euclidean distance is determined by the dot product of the displacement vector and the template surface normal vector: positive when the dot product is greater than zero, and negative when the dot product is less than or equal to zero. Finally, the deformations of all vertices in each subcortical region are aggregated to obtain the deformation set of each subcortical region.

[0068] Step S40: Using multiple subcortical regions as network nodes, calculate the probability distribution similarity between any two subcortical regions based on the set of deformation variables for each subcortical region, and use the similarity as the connection strength between the two subcortical regions to construct a subcortical individual morphological brain network.

[0069] The method of using multiple subcortical regions as network nodes and calculating the similarity of probability distributions corresponding to the deformation sets of any two subcortical regions based on the deformation set of each subcortical region specifically includes:

[0070] By evaluating the impact of different sampling point numbers on the stability of the probability density function, and using the Fraser distance to quantify the differences in the probability density function under different sampling point numbers, the optimal number of sampling points for kernel density estimation is determined.

[0071] Based on the optimal number of sampling points, kernel density estimation is performed on the set of deformable variables corresponding to each subcortical region to obtain the probability density function of the set of deformable variables at the sampling points;

[0072] For any two subcortical regions, calculate the Jensen-Shannon divergence between the probability density functions, convert the Jensen-Shannon divergence into Jensen-Shannon similarity, and use the Jensen-Shannon similarity as the network edge strength between the two subcortical regions.

[0073] Understandably, in kernel density estimation, the choice of the number of sampling points directly affects the smoothness and stability of the probability density function. To determine the optimal number of sampling points, multiple candidate sampling point values ​​are first selected within a preset range, and kernel density estimation is performed on the set of deformation variables of the same subcortical region to obtain probability density curves with different numbers of sampling points. Then, taking the probability density curves corresponding to adjacent numbers of sampling points as the object, the Fraser distance is used to quantify the difference between them. The Fraser distance can measure the degree of similarity between two curves in terms of geometric shape. When increasing the number of sampling points causes the Fraser distance to tend to stabilize, the corresponding number of sampling points is the optimal value.

[0074] Furthermore, based on the determined optimal number of sampling points, for each set of deformation variables in the subcortical region, the kernel density estimation method is used to calculate the probability density distribution of the deformation variables at the sampling points. Specifically, with each sampling point as the center, the contribution of the deformation variables in the neighborhood is weighted and summed using a kernel function to obtain the probability density value corresponding to that sampling point, and finally, the probability density function of that region is generated.

[0075] Furthermore, for any two subcortical regions, assuming their probability density functions for deformation are P and Q, respectively, the Jensen-Shannon divergence metric is used to quantify the difference between them and convert it into similarity as the network connection strength. The calculation formula is as follows:

[0076] ;

[0077] ;

[0078] in, Let P represent the Jensen-Shannon similarity between regions P and Q, where P and Q are probability density functions of the shape variables of different brain regions. JSD is a quantitative value representing the degree of difference in morphological variation patterns between regions P and Q. The value ranges from 0 to 1, with larger values ​​indicating higher similarity between regions. n is the number of sampling points. The ordinal number of the sampling point. Represents probability distribution In the The probability density value at each sampling point Describe the probability distribution Q at the th... The probability density value at each sampling point Indicates the average distribution No. The probability density value at each sampling point.

[0079] Furthermore, for any two subcortical regions, the calculated Jensen-Shannon similarity is used as the strength of the connection between these two regions. This strength value ranges from 0 to 1, with a larger value indicating a more similar morphological distribution between the two regions. After calculating the similarity pairwise for all regions, a symmetric weighted adjacency matrix is ​​obtained, which represents the subject's individual subcortical morphological brain network.

[0080] Step S50: Extract the network features of the subcortical individual morphological brain network, construct a brain image classification model based on the network features, obtain the target brain image, classify the target brain image based on the brain image classification model, and obtain the classification result of the target brain image.

[0081] In this embodiment, the extracted subcortical individual morphological brain network features are preprocessed and feature-selected. These network features are then used as model input to construct a brain image classification model for classifying magnetic resonance imaging (MRI) images of brain structures. The model is trained and optimized using a labeled sample dataset. In practical applications, a target brain image is acquired, and the corresponding subcortical individual morphological brain network and its features are extracted. This data is then input into the trained brain image classification model. Through model calculation and discrimination, the target brain image is automatically classified, ultimately outputting accurate and objective classification results, providing a reliable basis for subsequent clinical analysis, assisted diagnosis, and prognostic assessment.

[0082] Furthermore, the method of constructing a subcortical individual morphological brain network by using the similarity as the connection strength between two subcortical regions further includes:

[0083] Graph theory analysis was performed on the subcortical individual morphological brain network to obtain the network's graph theory indices. Based on the graph theory indices, the network's global and node attributes were calculated, and the test-retest reliability of the global and node attributes was quantified using the intragroup correlation coefficient.

[0084] The global attributes and node attributes are filtered to obtain the global attributes and node attributes whose test-retest reliability meets the preset conditions, and the global attributes and node attributes whose test-retest reliability meets the preset conditions are used as high reliability indicators.

[0085] The global attributes include clustering coefficient, shortest path length, local efficiency, global efficiency, and modularity, while the node attributes include node degree, node efficiency, and betweenness centrality.

[0086] In this embodiment, a toolkit is used to calculate the network's global attributes (clustering coefficient Cp, shortest path length Lp, local efficiency Eloc, global efficiency Eglob, modularity Q) and node attributes (node ​​degree k, node efficiency e, betweenness centrality b). All global attributes are normalized by dividing by the mean of 100 matched random networks to verify the network's small-world properties (normalized Cp>1, normalized Lp≈1, normalized Eloc>1, normalized Eglob≈1, normalized Q>1). The area under the curve (AUC) of each index across the entire sparsity range is calculated as a summary scalar measurement to avoid errors caused by a single threshold. The AUC of the three node attributes is calculated and averaged among subjects. Brain regions with attribute values ​​exceeding a preset threshold are defined as core nodes, clarifying the key functional hubs of the subcortical network.

[0087] The intraclass correlation coefficient (ICC) was used to quantify the test-retest reliability of network connectivity patterns, global and node attributes. The number and proportion of indicators with excellent reliability (ICC≥0.75) and good reliability (0.5≤ICC<0.75) were counted to verify the stability of the indicators. Only indicators with high reliability were retained for subsequent clinical application.

[0088] Furthermore, the step of using the global and node attributes whose test-retest reliability meets preset conditions as high reliability indicators further includes:

[0089] Based on the aforementioned subcortical individual morphological brain network and the aforementioned high reliability index, a comparative analysis of clinical application was conducted between the case group and the healthy control group. Intergroup statistical tests were performed on the network edge strength, global attributes, and node attributes, and multiple tests were performed using the false discovery rate to screen out network features with significant differences.

[0090] In this embodiment, focusing on the strength of network edges, permutation tests or two-sample tests are used to compare the Jensen-Shannon similarity distributions of the two groups of subjects on each edge to identify any significant differences, thereby identifying abnormally enhanced or weakened morphological associations between regions in the case group. Secondly, for global attributes (such as clustering coefficient, shortest path length, global efficiency, etc.) and node attributes (such as node degree, node efficiency, betweenness centrality, etc.), inter-group statistical tests are performed to quantify the direction and magnitude of changes in indicators between the case group and the control group, recording the mean, standard deviation, and raw p-value. Since multiple comparisons (multiple edges, multiple attributes) are involved, multiple test corrections are applied to all p-values ​​obtained from the tests to control for false positives. Features with a corrected p-value less than 0.05 are considered statistically significant, ultimately identifying network features with stable differences between the case group and the healthy control group. These significantly different features can serve as potential imaging biomarkers for subsequent classification model construction or efficacy evaluation.

[0091] As can be seen, this invention provides a precise, stable, and individualized method for constructing subcortical morphological networks, effectively quantifying brain structural variation patterns, and providing reliable technical support for disease mechanism research and imaging biomarker mining.

[0092] Furthermore, such as Figure 2 As shown, based on the above-mentioned method for constructing subcortical individual morphological brain networks based on shape features, the present invention also provides a system for constructing subcortical individual morphological brain networks based on shape features, wherein the system for constructing subcortical individual morphological brain networks based on shape features includes:

[0093] Binary mask extraction module 51 is used to acquire the structural magnetic resonance image of the subject, preprocess and segment the structural magnetic resonance image, and extract binary masks of multiple subcortical regions from the segmentation results.

[0094] Shape modeling module 52 is used to perform shape modeling on the binary mask of each subcortical region based on a deep learning model, so as to obtain the individualized shape mesh corresponding to each subcortical region.

[0095] The deformation calculation module 53 is used to calculate the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template, so as to obtain the deformation set of each subcortical region;

[0096] The brain network construction module 54 is used to construct a subcortical individual morphological brain network by using multiple subcortical regions as network nodes, calculating the probability distribution similarity between any two subcortical regions based on the set of deformation variables of each subcortical region, and using the similarity as the connection strength between the two subcortical regions.

[0097] The classification application module 55 is used to extract the network features of the subcortical individual morphological brain network, construct a brain image classification model based on the network features, obtain the target brain image, classify the target brain image based on the brain image classification model, and obtain the classification result of the target brain image.

[0098] Furthermore, such as Figure 3 As shown, based on the above-mentioned method and system for constructing subcortical individual morphological brain networks based on shape features, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 3 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0099] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a shape-feature-based subcortical morphological brain network construction program 40, which can be executed by the processor 10 to implement the shape-feature-based subcortical morphological brain network construction method of this application.

[0100] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the shape feature-based method for constructing subcortical individual morphological brain networks.

[0101] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components of the terminal communicate with each other via a system bus.

[0102] In one embodiment, when processor 10 executes the shape-feature-based subcortical individual morphological brain network construction program 40 in memory 20, the following steps are performed:

[0103] Structural magnetic resonance images of the subject are acquired, and the structural magnetic resonance images are preprocessed and segmented. Binary masks of multiple subcortical regions are extracted from the segmentation results.

[0104] Based on a deep learning model, the shape of the binary mask for each subcortical region is modeled to obtain the individualized shape mesh corresponding to each subcortical region.

[0105] Calculate the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template to obtain the deformation set of each subcortical region;

[0106] Using multiple subcortical regions as network nodes, the similarity of probability distributions corresponding to the deformation sets of any two subcortical regions is calculated based on the deformation set of each subcortical region, and the similarity is used as the connection strength between the two subcortical regions to construct a subcortical individual morphological brain network.

[0107] The network features of the subcortical individual morphological brain network are extracted, a brain image classification model is constructed based on the network features, and a target brain image is obtained. The target brain image is classified based on the brain image classification model to obtain the classification result of the target brain image.

[0108] The multiple subcortical regions include: the caudate nucleus, putamen, thalamus, globus pallidus, hippocampus, amygdala, and nucleus accumbens of both hemispheres.

[0109] Specifically, the deep learning model-based method involves shaping the binary mask for each subcortical region to obtain an individualized shape mesh for each subcortical region.

[0110] Based on the binary masks of all subcortical regions, an average mask corresponding to each subcortical region is constructed, and the average mask is converted into a three-dimensional template shape model by the moving cube algorithm. The three-dimensional template shape model is used as the starting reference mesh for shape modeling.

[0111] Align the surface point cloud of the binary mask for each subcortical region of the subject with the corresponding 3D template shape model to obtain the initial shape point cloud for each subcortical region;

[0112] Using the PointNet deep learning model based on a multilayer perceptron architecture, the initial shape point cloud of each subcortical region is iteratively optimized. By predicting the displacement vector of each vertex, the corresponding 3D template shape model is made to fit the individual target point cloud, thus obtaining the individualized shape mesh corresponding to each subcortical region.

[0113] Specifically, calculating the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template to obtain the deformation set of each subcortical region includes:

[0114] Based on all individualized shape grids, an average template grid corresponding to each subcortical region is constructed, and the average template grid is used as a statistical reference grid for deformation calculation;

[0115] For each subcortical region, a correspondence is established between the vertices of the individualized shape mesh and the corresponding vertices of the average template mesh;

[0116] Calculate the signed Euclidean distance of each vertex in the individualized shape mesh relative to the corresponding vertex in the average template mesh, and use the signed Euclidean distance as the deformation of the vertex, wherein the sign is determined by the dot product of the vertex displacement vector and the surface normal vector of the average template mesh;

[0117] By aggregating the deformations of all vertices within each subcortical region, we obtain the deformation set for each subcortical region.

[0118] Specifically, the step of using multiple subcortical regions as network nodes and calculating the similarity of probability distributions corresponding to the deformation sets of any two subcortical regions based on the deformation set of each subcortical region includes:

[0119] By evaluating the impact of different sampling point numbers on the stability of the probability density function, and using the Fraser distance to quantify the differences in the probability density function under different sampling point numbers, the optimal number of sampling points for kernel density estimation is determined.

[0120] Based on the optimal number of sampling points, kernel density estimation is performed on the set of deformable variables corresponding to each subcortical region to obtain the probability density function of the set of deformable variables at the sampling points;

[0121] For any two subcortical regions, calculate the Jensen-Shannon divergence between the probability density functions, convert the Jensen-Shannon divergence into Jensen-Shannon similarity, and use the Jensen-Shannon similarity as the network edge strength between the two subcortical regions.

[0122] The method of constructing a subcortical individual morphological brain network by using the similarity as the connection strength between two subcortical regions further includes:

[0123] Graph theory analysis was performed on the subcortical individual morphological brain network to obtain the network's graph theory indices. Based on the graph theory indices, the network's global and node attributes were calculated, and the test-retest reliability of the global and node attributes was quantified using the intragroup correlation coefficient.

[0124] The global attributes and node attributes are filtered to obtain the global attributes and node attributes whose test-retest reliability meets the preset conditions, and the global attributes and node attributes whose test-retest reliability meets the preset conditions are used as high reliability indicators.

[0125] The global attributes include clustering coefficient, shortest path length, local efficiency, global efficiency, and modularity, while the node attributes include node degree, node efficiency, and betweenness centrality.

[0126] The method of using global and node attributes whose test-retest reliability meets preset conditions as high reliability indicators further includes:

[0127] Based on the aforementioned subcortical individual morphological brain network and the aforementioned high reliability index, a comparative analysis of clinical application was conducted between the case group and the healthy control group. Intergroup statistical tests were performed on the network edge strength, global attributes, and node attributes, and multiple tests were performed using the false discovery rate to screen out network features with significant differences.

[0128] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a shape feature-based subcortical individual morphological brain network construction program, which, when executed by a processor, implements the steps of the shape feature-based subcortical individual morphological brain network construction method described above.

[0129] In summary, this invention proposes a method for constructing a subcortical individual morphological brain network based on shape features. The method includes: acquiring structural magnetic resonance images of a subject, preprocessing and segmenting the images, and extracting binary masks of multiple subcortical regions from the segmentation results; performing shape modeling on the binary mask of each subcortical region based on a deep learning model to obtain an individualized shape mesh corresponding to each subcortical region; calculating the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template to obtain a deformation set for each subcortical region; using multiple subcortical regions as network nodes, calculating the probability distribution similarity between any two deformation sets of subcortical regions based on the deformation set of each subcortical region, and using this similarity as the connection strength between the two subcortical regions to construct a subcortical individual morphological brain network; extracting network features from the brain network; and constructing a classification model based on the network features for brain image classification. The subcortical individual shape network constructed in this invention exhibits good test-retest stability, effectively reproduces the small-world properties and modular features of healthy individuals' brain networks, and improves the classification accuracy of brain images.

[0130] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0131] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0132] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for constructing subcortical individual morphological brain networks based on shape features, characterized in that, The method for constructing subcortical individual morphological brain networks based on shape features includes: Structural magnetic resonance images of the subject are acquired, and the structural magnetic resonance images are preprocessed and segmented. Binary masks of multiple subcortical regions are extracted from the segmentation results. Based on a deep learning model, the shape of the binary mask for each subcortical region is modeled to obtain the individualized shape mesh corresponding to each subcortical region. Calculate the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template to obtain the deformation set of each subcortical region; Using multiple subcortical regions as network nodes, the similarity of probability distributions corresponding to the deformation sets of any two subcortical regions is calculated based on the deformation set of each subcortical region, and the similarity is used as the connection strength between the two subcortical regions to construct a subcortical individual morphological brain network. The network features of the subcortical individual morphological brain network are extracted, a brain image classification model is constructed based on the network features, and a target brain image is obtained. The target brain image is classified based on the brain image classification model to obtain the classification result of the target brain image. The deep learning model-based approach involves shape modeling of the binary mask for each subcortical region to obtain an individualized shape mesh corresponding to each subcortical region. Specifically, this includes: Based on the binary masks of all subcortical regions, an average mask corresponding to each subcortical region is constructed, and the average mask is converted into a three-dimensional template shape model by the moving cube algorithm. The three-dimensional template shape model is used as the starting reference mesh for shape modeling. Align the surface point cloud of the binary mask for each subcortical region of the subject with the corresponding 3D template shape model to obtain the initial shape point cloud for each subcortical region; Using the PointNet deep learning model based on a multilayer perceptron architecture, the initial shape point cloud of each subcortical region is iteratively optimized, and the corresponding 3D template shape model is made to fit the individual target point cloud by predicting the displacement vector of each vertex, so as to obtain the individualized shape mesh corresponding to each subcortical region. The calculation of the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template, to obtain the deformation set of each subcortical region, specifically includes: Based on all individualized shape grids, an average template grid corresponding to each subcortical region is constructed, and the average template grid is used as a statistical reference grid for deformation calculation; For each subcortical region, a correspondence is established between the vertices of the individualized shape mesh and the corresponding vertices of the average template mesh; Calculate the signed Euclidean distance of each vertex in the individualized shape mesh relative to the corresponding vertex in the average template mesh, and use the signed Euclidean distance as the deformation of the vertex, wherein the sign is determined by the dot product of the vertex displacement vector and the surface normal vector of the average template mesh; By aggregating the deformations of all vertices within each subcortical region, we obtain the deformation set for each subcortical region.

2. The method for constructing subcortical individual morphological brain networks based on shape features according to claim 1, characterized in that, The multiple subcortical regions include: the caudate nucleus of both hemispheres, putamen, thalamus, globus pallidus, hippocampus, amygdala, and nucleus accumbens.

3. The method for constructing subcortical individual morphological brain networks based on shape features according to claim 1, characterized in that, The method of using multiple subcortical regions as network nodes and calculating the similarity of probability distributions corresponding to the deformation sets of any two subcortical regions based on the deformation set of each subcortical region specifically includes: By evaluating the impact of different sampling point numbers on the stability of the probability density function, and using the Fraser distance to quantify the differences in the probability density function under different sampling point numbers, the optimal number of sampling points for kernel density estimation is determined. Based on the optimal number of sampling points, kernel density estimation is performed on the set of deformable variables corresponding to each subcortical region to obtain the probability density function of the set of deformable variables at the sampling points; For any two subcortical regions, calculate the Jensen-Shannon divergence between the probability density functions, convert the Jensen-Shannon divergence into Jensen-Shannon similarity, and use the Jensen-Shannon similarity as the network edge strength between the two subcortical regions.

4. The method for constructing subcortical individual morphological brain networks based on shape features according to claim 1, characterized in that, The method of constructing a subcortical individual morphological brain network by using the similarity as the connection strength between two subcortical regions further includes: Graph theory analysis was performed on the subcortical individual morphological brain network to obtain the network's graph theory indices. Based on the graph theory indices, the network's global and node attributes were calculated, and the test-retest reliability of the global and node attributes was quantified using the intragroup correlation coefficient. The global attributes and node attributes are filtered to obtain the global attributes and node attributes whose test-retest reliability meets the preset conditions, and the global attributes and node attributes whose test-retest reliability meets the preset conditions are used as high reliability indicators. The global attributes include clustering coefficient, shortest path length, local efficiency, global efficiency, and modularity, while the node attributes include node degree, node efficiency, and betweenness centrality.

5. The method for constructing subcortical individual morphological brain networks based on shape features according to claim 4, characterized in that, The step of using global and node attributes whose test-retest reliability meets preset conditions as high reliability indicators further includes: Based on the aforementioned subcortical individual morphological brain network and the aforementioned high reliability index, a comparative analysis of clinical application was conducted between the case group and the healthy control group. Intergroup statistical tests were performed on the network edge strength, global attributes, and node attributes, and multiple tests were performed using the false discovery rate to screen out network features with significant differences.

6. A shape-feature-based system for constructing subcortical individual morphological brain networks, characterized in that, The shape-feature-based subcortical individual morphological brain network construction system is used to implement the shape-feature-based subcortical individual morphological brain network construction method according to any one of claims 1-5, wherein the shape-feature-based subcortical individual morphological brain network construction system comprises: The binary mask extraction module is used to acquire the structural magnetic resonance image of the subject, preprocess and segment the structural magnetic resonance image, and extract the binary mask of multiple subcortical regions from the segmentation result. The shape modeling module is used to perform shape modeling on the binary mask of each subcortical region based on the deep learning model, so as to obtain the individualized shape mesh corresponding to each subcortical region. The deformation calculation module is used to calculate the deformation of all vertices of each individualized shape mesh relative to the corresponding vertices in the average template, so as to obtain the deformation set of each subcortical region; The brain network construction module is used to construct an individual subcortical morphological brain network by using multiple subcortical regions as network nodes, calculating the probability distribution similarity between any two subcortical regions based on the set of deformation variables of each subcortical region, and using the similarity as the connection strength between the two subcortical regions. The classification application module is used to extract network features of the subcortical individual morphological brain network, construct a brain image classification model based on the network features, acquire a target brain image, classify the target brain image based on the brain image classification model, and obtain the classification result of the target brain image.

7. A terminal, characterized in that, The terminal includes: a memory, a processor, and a shape feature-based subcortical individual morphological brain network construction program stored in the memory and executable on the processor. When the shape feature-based subcortical individual morphological brain network construction program is executed by the processor, it implements the steps of the shape feature-based subcortical individual morphological brain network construction method as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a shape-feature-based subcortical individual morphological brain network construction program, which, when executed by a processor, implements the steps of the shape-feature-based subcortical individual morphological brain network construction method as described in any one of claims 1-5.

Images