A method for extracting high-order features of brain cortex surface morphology

Abstract

The application discloses a brain cortex surface morphology high-order feature extraction method, which comprises the following steps: firstly, acquiring brain magnetic resonance T1 structural image data; for individual brain magnetic resonance T1 weighted structural image data, using an open source magnetic resonance data processing toolkit to reconstruct a brain cortex surface; secondly, acquiring brain cortex surface vertex original morphology features, constructing brain cortex surface morphology vertex node strength, and obtaining high-order features; then, constructing a brain region level complex network based on the cortex vertex morphology high-order features, and obtaining complex network features; finally, based on the high-order features and the complex network features, using feature selection combined with machine learning to realize a cognitive ability evaluation model, and completing effectiveness evaluation. The application helps to expand traditional brain cortex surface morphology research to the field of brain connection and brain network research, improves the validity of brain network research, reveals the brain cortex morphology connection mechanism of brain cognition and brain development, and promotes the progress of brain science research.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of neuroimaging, and relates to a brain cortex surface morphology high-order feature extraction method. BACKGROUND

[0002] Surface-based morphology (SBM) is a commonly used and mature method and technology in the field of neuroimaging. Surface-based morphology features are obtained based on T1 magnetic resonance imaging data and a neuroimaging toolkit. The surface-based morphology features include cortical thickness, volume, surface area, curvature, sulcal depth, complexity and the like. Based on the above features, researchers have shown that the surface-based morphology features have good sensitivity, specificity and individual differences. However, so far, few studies have focused on the morphological similarity (morphological connectivity) between brain regions and even between vertices.

[0003] In the aspect of cortical brain region level morphological connectivity, some researchers first proposed a brain region level morphological network construction method based on the similarity of the mean and standard deviation of cortical thickness. Some researchers analyzed a brain morphological connectivity construction method based on kernel density estimation of brain region surface morphology features, which has excellent repeatability. Some researchers proposed a complex network construction method based on the correlation coefficient of brain region surface morphology features, and a high-order brain connectivity construction method based on the distance of brain region surface morphology features, which has good reliability and validity. The construction method of the cortical surface morphology brain region level network can be summarized as: brain atlas selection, feature vector construction and similarity calculation. The purpose of brain atlas selection is to construct nodes, feature vector construction is to represent node attributes, and similarity calculation is to represent the connection of the network. The limit of the node resolution of the brain cortex atlas is the cortical vertex. So far, there has been no research on the cortical vertex level brain morphological connectivity.

[0004] In the aspect of cortical vertex level morphological connectivity, the brain cortex surface vertex is the network node. Since the brain cortex surface vertex has a series of morphological features, the connection of the vertex level brain network can be calculated by referring to the idea of brain region level morphological network, and further extracting the cortical vertex level high-order features. On this basis, the brain region level high-order morphological features are extracted, and the whole brain high-order connectivity and complex network are constructed. SUMMARY

[0005] The existing brain cortex surface morphological features reflect the morphological properties of each vertex. Due to the limitations of feature extraction methods and techniques, few studies focus on the morphological similarity (morphological connection) between brain cortex vertices or brain regions. In order to overcome the shortcomings of the existing research, the present application provides a brain cortex surface morphological high-order feature extraction method, which constructs an individualized brain network feature extraction method based on the existing brain cortex surface morphological data, to make up for the lack of high-order features in traditional brain cortex surface morphology, and expands the traditional brain surface morphology to the field of brain network research.

[0006] The present application constructs a vertex-level feature vector based on the existing brain cortex surface morphology toolkit, further constructs high-order features of vertex-level and brain region-level brain morphology, and proposes a series of brain morphology connection representation methods based on feature distance, and explores the cognitive evaluation ability of brain morphology high-order features through machine learning, to provide a new measurement method for brain disease research and neuroscience research.

[0007] A brain cortex surface morphological high-order feature extraction method, comprising the following steps:

[0008] Step one: obtaining brain magnetic resonance T1 structural image data;

[0009] The brain magnetic resonance T1 weighted structural image dataset comes from a neural image open source database and a clinical dataset. The field strength of the magnetic resonance scanner should be 1.5T or 3.0T, and the scanning protocol is a standard T1 weighted magnetic resonance sequence. The subject population includes a cognitive impairment group and a healthy control group. At the same time, the cognitive ability score of each subject is obtained through a scale. The cognitive ability includes attention, self-control, memory and communication ability, etc.

[0010] Step two: establishing a brain cortex surface model;

[0011] For individual brain magnetic resonance T1 weighted structural image data, an open source magnetic resonance data processing toolkit is used for brain cortex surface reconstruction. The open source magnetic resonance data processing toolkit includes: FreeSurfer, FastSurfer, Computational Anatomy Toolbox12 (CAT12) and the like. The processing flow includes: skull removal, brain tissue segmentation, spherical space registration, surface reconstruction, brain region segmentation, parameter extraction and statistics and the like.

[0012] Step three: obtaining brain cortex surface vertex original morphological features;

[0013] The original morphological features of the brain cortex surface vertex include: cortical thickness, area, volume, curvature, folding, sulcal depth, complexity, etc. The morphological feature data file of the cortex surface is in gii or mgh format. The original morphological features need to be resampled by FreeSurfer, FastSurfer or CAT12 to obtain standard data in the fsaverage coordinates of the average brain.

[0014] Step four: constructing the morphological vertex node strength of the brain cortex surface to obtain high-order features;

[0015] Based on the original morphological features of the brain cortex surface vertex, a feature vector is constructed, and the vertex feature vector is any combination of the original morphological features obtained in step three. High-order morphological connections are calculated by feature distance correlation coefficients between vertices, which can be divided into local morphological connections and global morphological connections, and the node morphological connection strength represents the high-order features of the brain network.

[0016] In terms of local morphological connection construction, for each vertex, a vertex neighborhood graph is constructed based on the fsaverage coordinate system to obtain the adjacent vertices of each vertex. The current vertex and the adjacent vertices form the nodes of the local connection graph, and the feature distance correlation coefficient of the vertex is taken as the morphological connection edge of the local connection graph. The node strength of the current vertex is calculated, and the local morphological connection strength of the current vertex, i.e. the high-order feature, is obtained.

[0017] In terms of global morphological connection construction, all vertices form the nodes of the global connection graph, and the feature distance correlation coefficient of the vertex is taken as the morphological connection edge of the global connection graph. The node strength of the current vertex is calculated, and the global morphological connection strength of the current vertex, i.e. the high-order feature, is obtained.

[0018] Here, the feature distance is selected from Euclidian distance, chebychev distance, city block distance, minkowski distance, etc.

[0019] The feature distance correlation coefficient is:

[0020] mc mn = corr(D m , D n )

[0021] m and n represent any two vertices of the brain cortex, mc mn represents the similarity between the vertices, D m and D n are the feature distances between the brain cortex vertex and other local nodes or other nodes in the whole brain, and corr is a linear correlation function.

[0022] Step five: Constructing brain region-level complex network based on cortical vertex morphological high-order features, and obtaining complex network features;

[0023] The construction of brain region-level complex network includes two parts: brain region morphological connection construction and complex network construction.

[0024] Firstly, based on the above local morphological connection, the mean and standard deviation of the brain region are extracted as the feature vector using the division method of the standard brain atlas, and the similarity of the feature distance is calculated using the exponential function normalization, that is:

[0025] s1 ij =exp(-dist(F i ,F j ))

[0026] Here, i and j represent any two nodes of the brain network, s ij represents the similarity between nodes, that is, the edge of the network, F i =[MEAN i ,STD i ], F j =[MEAN j ,STD j ], MEAN represents the mean, STD represents the standard deviation, and dist is the feature distance function: such as Euclidean distance, city block distance, Chebyshev distance, Minkowski distance, etc.

[0027] Secondly, based on the above global morphological connection, the kernel density distribution of the brain region is extracted as the feature vector using the division method of the standard brain atlas, and the similarity of the feature distance is calculated using the exponential function normalization, that is:

[0028] s2 ij =exp(-dist(kde i ,kde j ))

[0029] Here, kde i and kde j are the kernel density distribution vectors of the vertex global connection strength in the brain region.

[0030] Finally, based on the brain region-level morphological connection s1 ij and s2 ij , the whole brain brain region-level complex network can be constructed. The extractable complex network features include: brain region node features (node degree, shortest path, node efficiency, etc.) and global features (small world property, global efficiency, etc.). Here, the complex network feature extraction is realized through the brain connectivity toolbox (Brain connectivity toolbox, BCT).

[0031] Step six; verify the effectiveness of the high-order features of the cerebral cortex surface morphology.

[0032] Based on the high-order features of the fourth step and the complex network features of the fifth step, the cognitive ability evaluation model can be realized by using feature selection combined with machine learning, and the effectiveness evaluation is completed. Age can also be used as a prediction index to verify the effectiveness of the model. In addition to single cognitive ability prediction model, multi-task learning can also be used to evaluate multiple cognitive abilities.

[0033] As preferred, in the step one, the brain magnetic resonance data is obtained by a 3.0T magnetic resonance scanner, and the scanning resolution is 1mm*1mm*1mm. The experimental subjects are school-age children, including a disease group and a healthy control group. The image data of the same experiment should come from the same magnetic resonance scanner. The cognitive ability score is obtained by cognitive scale evaluation.

[0034] Compared with the prior art, the present application has the beneficial effects that:

[0035] By adopting the above technical solution, based on the brain T1 structural magnetic resonance analysis toolkit, the surface morphology features of the individual cerebral cortex vertex are obtained, the vertex-level and brain region-level morphology connection features are extracted using feature distance, so as to reflect the high-order brain morphology connection information, which helps to expand the traditional cerebral cortex surface morphology research to the brain connection and brain network research field. The present application can be applied to different T1 magnetic resonance data sets, and is not affected by the magnetic resonance scanner and the scanning parameters, which helps to explore the high-order morphology connection mechanism of the cerebral cortex and the imaging markers.

[0036] Based on the brain T1 magnetic resonance image and the open source data processing and analysis toolkit, the high-order features of the cerebral cortex morphology network can be constructed, which is not restricted by the magnetic resonance machine and the subject group, and the subjects do not need to participate in the task, so the application scope is wide, and it is especially suitable for infants, teenagers, the elderly and people with cognitive impairment.

[0037] Based on the high-order features of the cerebral cortex morphology, the individual differences and cognitive prediction ability of the high-order morphology features of the cerebral cortex are verified by using the method of machine learning, and the validity of the brain network research is improved.

[0038] Compared with the existing cerebral cortex surface morphology research, this scheme can construct a brain complex network based on the high-order features of the cortex vertex, provide a new method and technology for brain network research, reveal the cerebral cortex morphology connection mechanism of brain cognition and brain development, and promote the progress of brain science research. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0040] Figure 1 A flowchart of a brain cortex surface morphology high-order feature extraction method of the present application;

[0041] Figure 2 A flowchart of a brain cortex vertex high-order morphology feature extraction of the present application;

[0042] Figure 3 A flowchart of a brain region level complex network construction of the present application;

[0043] Figure 4 A brain age prediction model performance result graph based on brain region morphology connection;

[0044] Figure 5 Brain morphology connection related to brain age. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0046] A brain cortex surface morphology high-order feature extraction method comprises the following steps:

[0047] Step 1: Obtain brain magnetic resonance T1 structural image data;

[0048] The brain magnetic resonance T1 weighted structural image dataset comes from a neural image open source database and a clinical dataset. The field strength of the magnetic resonance scanner should be 1.5T or 3.0T, and the scanning protocol is a standard T1 weighted magnetic resonance sequence. The subject population includes a cognitive impairment group and a healthy control group. At the same time, the cognitive ability score of each subject is obtained through a scale. The cognitive ability includes attention, self-control, memory and communication ability, etc.

[0049] Step 2: Establish a brain cortex surface model;

[0050] For individual brain magnetic resonance T1-weighted structural image data, brain cortical surface reconstruction is performed using an open source magnetic resonance data processing toolkit. The open source magnetic resonance data processing toolkit includes: FreeSurfer, FastSurfer, Computational Anatomy Toolbox12 (CAT12), etc. The processing flow includes: skull removal, brain tissue segmentation, spherical space registration, surface reconstruction, brain region segmentation, parameter extraction and statistics, etc.

[0051] Step three: obtain the original morphological features of the brain cortical surface vertex;

[0052] The original morphological features of the brain cortical surface vertex include: cortical thickness, area, volume, curvature, folding, sulcal depth, complexity, etc. The cortical surface morphological feature data file is in gii or mgh format. The original morphological features need to be resampled by FreeSurfer, FastSurfer or CAT12 to obtain standard data in fsaverage (fs represents FreeSurfer, and average represents average brain) coordinates.

[0053] Step four: construct the morphological vertex node strength of the brain cortical surface to obtain high-order features;

[0054] Based on the original morphological features of the brain cortical surface vertex, a feature vector is constructed, and the vertex feature vector is any combination of the original morphological features obtained in step three. High-order morphological connections are calculated by feature distance correlation coefficients between vertices, which can be divided into local morphological connections and global morphological connections. It should be noted that the number of brain cortical surface vertices is as high as 160,000, and due to the limitation of storage space, the node morphological connection strength is used to represent the high-order features of the brain network.

[0055] In terms of local morphological connection construction, for each vertex, a vertex neighborhood graph is constructed based on the fsaverage coordinate system to obtain the adjacent vertices of each vertex. The current vertex and the adjacent vertices form the nodes of the local connection graph, and the feature distance correlation coefficient of the vertex is used as the morphological connection edge of the local connection graph. The node strength of the current vertex is calculated, and the local morphological connection strength (high-order feature) of the current vertex is obtained.

[0056] In terms of global morphological connection construction, all vertices form the nodes of the global connection graph, and the feature distance correlation coefficient of the vertex is used as the morphological connection edge of the global connection graph. The node strength of the current vertex is calculated, and the global morphological connection strength (high-order feature) of the current vertex is obtained.

[0057] Here, the feature distance is selected from Euclidian distance, chebychev distance, city block distance, minkowski distance, etc.

[0058] The feature distance correlation coefficient is:

[0059] mc mn = corr(D m , D n )

[0060] m and n represent any two vertices of the cerebral cortex, mc mn represents the similarity between the vertices, D m , D n is the feature distance between the cerebral cortex vertices and the local other nodes or the whole brain other nodes, and corr is a linear correlation function.

[0061] Step five: constructing a brain region level complex network based on the morphological high-order features of the cortical vertices, and obtaining the complex network features;

[0062] The construction of the brain region level complex network includes two parts: the construction of the brain region morphological connection and the construction of the complex network.

[0063] Firstly, based on the above local morphological connection, the mean and standard deviation of the brain region are extracted as the feature vector using the division method of the standard brain atlas, and the similarity of the feature distance is calculated using the exponential function normalization, i.e.:

[0064] s1 ij = exp(-dist(F i , F j ))

[0065] Here, i and j represent any two nodes of the brain network, s ij represents the similarity between the nodes, i.e. the edge of the network, F i = [MEAN i , STD i ], F j = [MEAN j , STD j ], MEAN represents the mean, STD represents the standard deviation, and is the mean and standard deviation of the local connection strength of the vertices in the brain region, and dist is a feature distance function: such as Euclidean distance, city block distance, chebychev distance, minkowski distance, etc.

[0066] Secondly, based on the above global morphological connection, the kernel density distribution of the brain region is extracted as the feature vector using the division method of the standard brain atlas, and the similarity of the feature distance is calculated using the exponential function normalization, i.e.:

[0067] s2 ij = exp(-dist(kde i ,kde j ))

[0068] Here, i and j represent any two nodes of the brain network, s ij represents the similarity between nodes, i.e., the edge of the network, kde i and kde j are the kernel density distribution vectors of the global connection strength of the vertices within the brain region, and dist is a feature distance function: such as Euclidean distance, city block distance, Chebyshev distance, Minkowski distance, etc.

[0069] Finally, based on the regional morphological connection s1 ij and s2 ij , the whole brain regional complex network can be constructed. The extractable complex network features include: regional node features (node degree, shortest path, node efficiency, etc.) and global features (small-world property, global efficiency, etc.). Here, the complex network feature extraction is realized through the brain connectivity toolbox (BCT).

[0070] Step six: verify the effectiveness of the morphological high-order features of the brain cortex surface.

[0071] Based on the fourth step high-order features and the fifth step complex network features, cognitive ability evaluation models can be realized using feature selection combined with machine learning, and effectiveness evaluation can be completed. Feature selection algorithms include correlation coefficient significance method, maximum correlation minimum redundancy method, genetic algorithm, ant colony algorithm, particle swarm algorithm, etc. Machine learning algorithms include support vector machine regression, correlation vector machine regression, lasso regression, neural network, etc. Cognitive abilities include: attention, self-control, working memory, social skills, etc. Age can also be used as a predictor to verify the effectiveness of the model. In addition to single cognitive ability prediction models, multi-task learning can also be used to evaluate multiple cognitive abilities at the same time.

[0072] As a preferred, in the step one, the brain magnetic resonance data is obtained by a 3.0T magnetic resonance scanner, and the scanning resolution is 1mm x 1mm x 1mm. The experimental subjects are school-age children, including a disease group and a healthy control group. The image data of the same experiment should come from the same magnetic resonance scanner. The cognitive ability score is obtained by cognitive scale evaluation.

[0073] As preferred, in the step two, the brain structure image magnetic resonance data is preprocessed by the FreeSurfer toolkit. The FreeSurfer can be accelerated by a computing cluster multi-core CPU in runtime, so as to improve the efficiency of preprocessing. The software version of the FreeSurfer needs to be unified, and is based on the same version of the operating system. The processing flow includes the following steps: skull removal, brain tissue segmentation, spherical space registration, surface reconstruction, brain region segmentation, parameter extraction and statistics.

[0074] As preferred, in the step three, the original morphological features of the brain cortex surface vertex obtained by the FreeSurfer toolkit include four types of features: cortex thickness, area, volume and curvature. The original morphological features need to be resampled to obtain the standard mgh format medical image data under the fsaverage coordinate. Spatial smoothing is not required here.

[0075] As preferred, in the step four, a feature vector V i =[thickness, area, volume, curvature] is constructed for each vertex. Here, thickness represents the cortex thickness, area represents the cortex surface area, volume represents the volume, and curvature represents the cortex curvature. The local morphological connection strength and the global morphological connection strength of the vertex are calculated respectively. The feature distance is the city block distance. The high-order morphological connection is characterized by the feature distance correlation coefficient.

[0076] As preferred, in the step five, the whole brain is divided into 68 spatially independent brain regions as network nodes by using the division method of the standard Desikan-Killiany (DK) brain atlas, the mean value and the standard deviation of the local connection of the vertex in the brain region are extracted as the feature vector, and the similarity (morphological connection) of the feature distance is calculated by using the exponential function. The feature distance is the city block distance. The global attribute and the node attribute of the whole brain complex network can be further extracted.

[0077] As preferred, in the step six, the genetic algorithm is combined with the neural network algorithm to realize the cognitive ability evaluation model, so as to verify the effectiveness and individual difference of the high-order morphological features of the brain cortex surface, and reveal the brain network coding basis of the cognitive ability. Here, the cognitive ability can be selected as the widely used indicators such as attention and memory.

[0078] The present application provides a brain cortex surface morphological high-order feature extraction method, and it needs to be explained that only representative embodiments are disclosed in the present application, and the present application is not limited to the specific methods described herein, but can also have other embodiments or combinations of other embodiments.

[0079] The brain cortex surface morphological high-order feature extraction method flow is as follows:Figure 1 As shown in the figure, the high-order brain network feature extraction based on the cortical surface morphological characteristics in this paper is taken as an example, and the specific implementation is described as follows.

[0080] The brain magnetic resonance T1 structural image data is obtained through the Autism Brain Imaging Data Exchange (ABIDE), a total of 539 cases of autism patients and 573 cases of healthy controls. The ABIDE database is composed of data from 16 sites, so the T1 weighted brain structural image data comes from different scanners and scanning protocols, each site has different different image parameters and spatial resolution.

[0081] The T1 weighted brain structural image data is preprocessed using FreeSurfer, and all the preprocessing results can be obtained from the website (http: / / preprocessed-connectomes-project.org / abide / ). The website provides a data download script (https: / / github.com / preprocessed-connectomes-project / abide / ), which can be downloaded in batches through the script.

[0082] The original morphological features of the brain cortex surface vertex obtained by the FreeSurfer toolkit include: cortical thickness, area, volume, and curvature, a total of four types of features. The original morphological features are resampled to obtain standard mgh format medical image data under the fsaverage coordinate. Here, the spatial smoothing data does not need to be downloaded.

[0083] The extraction process of the high-order morphological features of the vertex-level brain cortex surface is as shown in the figure Figure 2 Based on the four types of features of cortical thickness, area, volume, and curvature, a feature vector V i =[thickness, area, volume, curvature] is constructed for each vertex. The local morphological connection strength and the global morphological connection strength of the vertex are calculated respectively. The feature distance is the city block distance. The high-order morphological connection is the correlation coefficient of the feature distance. This step needs to save the results in the same mgh format as the original features such as cortical thickness.

[0084] The extraction process of the high-order morphological features of the brain cortex surface at the brain region level is as shown in the figure Figure 3 Using the standard DK brain atlas division method, the left and right hemispheres of the brain cortex are divided into 34 brain regions, respectively. The mean and standard deviation of the brain region are extracted as the feature vector, and the similarity of the feature distance is calculated using the exponential function. Among them, the feature distance is the city block distance. Further, the global properties and node properties of the whole brain complex network can be extracted.

[0085] To validate the effectiveness and individual difference of the high-order features of brain cortical surface morphology, we extracted the morphological connection features of brain regions based on Euclidean distance, city block distance, Chebyshev distance, and Minkowski distance, respectively. We used the correlation coefficient feature selection (p < 0.01) combined with the support vector machine regression method to realize the brain age estimation model. We used 1000 ten-fold cross-validation method to test the performance of the model. Finally, we established the brain age estimation model and revealed the morphological connection mechanism of brain development.

[0086] The experimental results show that the morphological connection of brain regions can significantly predict brain age, as shown in Figure 4 Further, it can reveal the age-related key brain morphological connections, as shown in Figure 5 It provides a new image marker for brain mechanism research.

Claims

1. A method for extracting high-order morphological features of the cerebral cortex surface, characterized in that, Includes the following steps: Step 1: Obtain T1-weighted MRI images of the brain; Step 2: For individual brain MRI T1-weighted structural image data, use an open-source MRI data processing toolkit to reconstruct the surface of the cerebral cortex; Step 3: Obtain the original morphological features of the vertices on the surface of the cerebral cortex. The original morphological features of the vertices on the surface of the cerebral cortex include: cortical thickness, area, volume, curvature, folding degree, sulcus depth and complexity. The original morphological features were resampled using an open-source magnetic resonance data processing toolkit to obtain standard data in the average brain fsaverage coordinates. Step 4: Construct the morphological vertex node strength of the cerebral cortex surface to obtain higher-order features. The specific process is as follows: construct feature vectors based on the original morphological features of the cerebral cortex surface vertices. The vertex feature vectors are any combination of the original morphological features obtained in Step 3. Higher-order morphological connections are calculated using the correlation coefficient of feature distances between vertices and are divided into local morphological connections and global morphological connections. The strength of node morphological connections is used to represent the higher-order features of the brain network. The specific process of constructing the local morphological connection is as follows: For each vertex, construct a vertex nearest neighbor graph based on the fsaverage coordinate system to obtain the neighboring vertices of each vertex; use the current vertex and its nearest neighbor vertices to form the nodes of the local connection graph, use the feature distance correlation coefficient of the vertex as the morphological connection edge of the local connection graph, calculate the node strength of the current vertex, and obtain the local morphological connection strength of the current vertex, i.e., the higher-order feature. The specific process of constructing the global morphological connection is as follows: all vertices form the nodes of the global connection graph, the feature distance correlation coefficient of the vertex is used as the morphological connection edge of the global connection graph, the node strength of the current vertex is calculated, and the global morphological connection strength of the current vertex is obtained, i.e., the higher-order feature. The characteristic distance is selected from Euclidean distance, Chebyshev distance, city block distance or Minkowski distance; The feature distance correlation coefficient is: mc mn = corr(D m , D n ) m and n are any two vertices of the cerebral cortex, mc mn is the similarity between the vertices, D m , D n is the feature distance between the vertices of the cerebral cortex and the local other nodes or the whole brain other nodes, corr is a linear correlation function; Step 5: Construct a brain region-level complex network based on higher-order morphological features of cortical vertices to obtain complex network features; Step 6: Based on high-order features and complex network features, use feature selection combined with machine learning to implement a cognitive ability assessment model and verify the effectiveness of high-order features of cortical surface morphology.

2. The method for extracting high-order morphological features of the cerebral cortex surface according to claim 1, characterized in that, In step one, the brain magnetic resonance T1-weighted structural image dataset comes from open-source neuroimaging databases and clinical datasets. The subjects included are school-aged children, with a cognitive impairment group and a healthy control group. At the same time, the cognitive ability score of each subject is obtained through a scale. The cognitive abilities mentioned include attention, self-control, memory, and communication skills.

3. The method for extracting high-order morphological features of the cerebral cortex surface according to claim 2, characterized in that, In step two: the open-source magnetic resonance data processing toolkit includes: the open-source cortical reconstruction toolkit, the fast cortical reconstruction toolkit, and the computational anatomy toolkit version 12; The open-source magnetic resonance imaging (MRI) data processing workflow includes: skull removal, brain tissue segmentation, spherical spatial registration, surface reconstruction, brain region segmentation, parameter extraction and statistics.

4. The method for extracting high-order morphological features of the cerebral cortex surface according to claim 3, characterized in that, Step 5 describes the construction of complex brain regions, which includes two parts: the construction of brain region morphological connectivity and the construction of complex networks. First, based on local morphological connectivity, using the standard brain atlas segmentation method, the mean and standard deviation of brain regions are extracted as feature vectors. The similarity of feature distances is then calculated using exponential function normalization, i.e.: s1 ij = exp(-dist(F i ,F j )) Here, i and j represent any two nodes of the brain network, s ij represents the similarity between nodes, i.e., the edge of the network, F i = [MEAN i , STD i ], F j = [MEAN j , STD j ], MEAN represents the mean, STD represents the standard deviation, dist is the feature distance function; Secondly, based on the aforementioned global morphological connectivity, using the standard brain atlas segmentation method, the nuclear density distribution of brain regions is extracted as a feature vector. The similarity of feature distances is then calculated using exponential function normalization, i.e.: s2 ij = exp(-dist(kde i ,kde j )) Among them, kde i and KDE j This represents the kernel density distribution vector of global connectivity strength at the vertices within the brain region; Finally, based on brain region-level morphological connectivity s1 ij and s2 ij We constructed a complex network at the whole-brain level and extracted its features.

5. The method for extracting high-order morphological features of the cerebral cortex surface according to claim 4, characterized in that, The complex network features include: brain region node features and global features; The feature extraction of the complex network is achieved through the Brain Connectivity Toolkit; The brain region node features include node degree, shortest path, and node efficiency; The global features include small-world properties and global efficiency.

6. The method for extracting high-order morphological features of the cerebral cortex surface according to claim 5, characterized in that, Step six also includes using age as a predictive indicator to verify effectiveness; In addition to single cognitive ability prediction models, it also includes using multi-task learning to simultaneously assess multiple cognitive abilities.

Images