A method and device for constructing a brain network by synchronously acquiring brain structure and metabolism images

Abstract

The application relates to the technical field of brain network construction, and particularly discloses a brain structure and metabolic image synchronously acquired brain network construction method and device, which comprises the following steps: acquiring a structure image and a metabolic image to be processed; registering the structure image to be processed to a standard brain template; registering the metabolic image to the standard brain template by taking the structure image as an intermediate; fusing brain structure information and metabolic information of the nuclear medicine image to obtain a metabolic image; performing brain partition processing on the metabolic image to form a multi-partition brain model; obtaining a co-correlation matrix by analyzing the correlation between multiple brain regions; and obtaining a metabolic brain network by optimizing the co-correlation matrix of the brain model. The brain structure and metabolic image synchronously acquired brain network construction method and device can more effectively find metabolic differences between brain disease patients and normal people in network attribute indexes, and provide valuable information for brain development, maturity and aging.

Description

Technical Field

[0001] This invention relates to the field of brain network construction and control technology, and in particular to a method and device for constructing brain networks by simultaneously acquiring brain structure and metabolic images. Background Technology

[0002] The brain primarily achieves its various functions through the operation of multiple systems of interacting units. These units range in size from clusters of neurons to different cortical and subcortical regions. One of the major challenges in neuroscience is understanding how the brain's rich functional diversity arises from such a relatively fixed anatomical structure, spanning multiple scales. Furthermore, the sheer number and complexity of the brain systems involved within its vast functional repertoire present unique challenges to the study of higher cognitive processes. Growing evidence suggests that the answer may lie in the brain's interconnected organization, which facilitates the global integration of local operations and functions. This type of organization builds intrinsic resilience and ensures adaptability, robustness, and functional diversity. Both structural and functional connectivity in the brain reflect the overall condition of the brain and play a crucial role in exploring and researching the pathological mechanisms of various brain diseases.

[0003] PET / MR, or positron emission tomography (PET) and magnetic resonance imaging (MR), is a large-scale functional metabolic and molecular imaging diagnostic device that combines the strengths of both. It integrates the diagnostic capabilities of PET and MR, achieving maximum synergy. As a representative of cutting-edge technology in the field of high-end medical imaging diagnostic equipment, PET / MR can perform multimodal and multiparameter imaging, which is of great significance for the accurate diagnosis and research of neurodegenerative diseases such as Parkinson's disease and Alzheimer's disease, as well as complex diseases such as liver cancer, pancreatic cancer, epilepsy, and multiple myeloma.

[0004] In existing technologies, most diagnostic and treatment methods for brain metabolism rely on physicians' experience to observe metabolic images. However, metabolic images cannot show the distribution of brain metabolism in the brain structure, making it difficult to reliably analyze and understand the physiological mechanisms of brain metabolism, and making it difficult to detect metabolic differences between patients with brain diseases and normal people in terms of standard indicators.

[0005] Therefore, those skilled in the art urgently need to find a new technical solution to address the aforementioned problems. Summary of the Invention

[0006] To address the technical problems in existing technologies, this invention provides a method for constructing a brain network by simultaneously acquiring brain structure and metabolic images. Through brain network analysis, the distribution of brain metabolism in the brain structure can be displayed intuitively and quantitatively. This method can more effectively discover metabolic differences in network attribute indicators between brain disease patients and normal individuals, providing valuable information for brain development, maturation, and aging.

[0007] A method for constructing brain networks using simultaneously acquired brain structural and metabolic images includes:

[0008] S1: Acquire the structural and metabolic images to be processed, and preprocess the structural and metabolic images to be processed;

[0009] S2: Register the structural image to be processed with the standard brain template, and then register the metabolic image with the structural image after registration processing, thereby fusing the brain structural information with the metabolic information of the metabolic image to obtain the registered metabolic image.

[0010] S3: Perform brain partitioning on metabolic images to form a multi-part brain model, which reflects the metabolic information of multiple brain regions.

[0011] S4: By analyzing the correlation between multiple brain regions, the cocorrelation matrix of the brain network is obtained. The cocorrelation matrix of the brain network is then optimized to obtain a metabolic brain network in matrix form.

[0012] Furthermore, methods for preprocessing structural and metabolic images include:

[0013] S11: Perform preliminary screening of structural and metabolic images. By roughly browsing the images, remove data with severe artifacts, occupancy, or other image partitions.

[0014] S12: Clean and desensitize structural and metabolic imaging data to remove sensitive personal information of users;

[0015] S13: If there are differences in resolution between structural and metabolic images due to reasons such as manufacturer or scanning method, then the resolution of structural images shall be unified.

[0016] Furthermore, the method for registering the structural images to be processed with standard brain templates includes:

[0017] S21: Import the structural image M and the standard brain template, and find a geometrically optimal alignment. This is achieved by finding an image matrix T(M) that maximizes the similarity between the two volumetric brain images of the standard brain template and the structural image M, where the image matrix T(M) satisfies:

[0018] T(M)=argminC(N,T(M))

[0019] Where C(N,T(M)) is the cost function, and N is the registration matrix of the standard brain template;

[0020] S22: Perform affine transformation of the structural image on each coordinate point of the image matrix T(M). The affine transformation of the structural image covers four dimensions of transformation: translation, rotation, scaling and shearing, to complete the registration of the main structure of the brain.

[0021] S23: Perform a nonlinear transformation on each coordinate point of the image matrix T(M) after affine transformation of the structural image, register local regions such as sulci and ventricles, and thus obtain the registered structural image M. * .

[0022] Furthermore, methods for registering metabolic images with structural images after registration processing include:

[0023] S24: Import metabolic image O and registered structural image M * The goal is to find a geometrically optimal alignment by finding an image matrix T(O) such that the structural image M... * Two volumetric brain images of metabolic image O have the greatest similarity, where the image matrix T(O) satisfies:

[0024] T(O) = argminC(T) * (M),T(O))

[0025] Among them, C(T) * (M),T(O)) are cost functions;

[0026] S25: Perform affine transformation of metabolic image on each coordinate point of image matrix T(O). The affine transformation of metabolic image covers transformations in four dimensions: translation, rotation, scaling and shearing. Register the main brain structure in metabolic image O with the registered structural image.

[0027] S26: Perform a nonlinear transformation on each coordinate point of the image matrix T(O) after affine transformation of the metabolic image to register local regions such as sulci and ventricles in the metabolic image O, thereby obtaining the registered metabolic image O. * .

[0028] Furthermore, methods for processing metabolic images into brain regions to form multi-region brain models include:

[0029] S31: Import metabolic images and construct the vertices and edges of the brain model through the modeler. The vertices are abstract indicators of the brain voxels with coordinates and gray values ​​in space, and the edges are line segments connecting two vertices. The brain sulcus surface is approximated by as many end-to-end line segments as possible.

[0030] S32: Brain regions are constructed using Bayesian parameter estimation, that is, maximizing the following probabilities:

[0031]

[0032] Where G represents the brain surface shape of the metabolic image, f is a nonlinear spherical transformation function, r represents the vertex, and P represents the partition. Given G and f, calculate and maximize the probability that partition P appears in a certain spatial location of the brain.

[0033] S33: Metabolic information of multiple brain regions is obtained based on different standard brain templates. The metabolic information includes the maximum, minimum, average and standard deviation of the SUV in each brain region, as well as the SUV range.

[0034] Furthermore, methods for obtaining the cocorrelation matrix of brain networks by analyzing the correlations between multiple brain regions include:

[0035] S41: Brain networks are constructed based on the user's age group, using individual brain regions as vertices and Pearson correlation coefficients between different brain regions as edges, to obtain the co-correlation matrix of the metabolic brain network. The Pearson correlation coefficient is calculated according to the following formula:

[0036]

[0037] Where r represents the Pearson correlation coefficient, which ranges from -1 to 1, where -1 indicates a perfect negative correlation between the two variables and 1 indicates a perfect positive correlation between the two variables. i and Y i These represent values ​​for different brain regions. and The set of Pearson correlation coefficients is calculated, where m = n*(n-1) / 2.

[0038] S42: Perform Benjamini-Yekutieli type multiple test correction on the set of Pearson correlation coefficients. If the Pearson correlation coefficients of two randomly selected brain regions are independent, then m p-values ​​will be obtained in the Benjamini-Yekutieli type multiple test correction. i Given the values ​​(i = 1, 2, 3, ..., m), sort these p values ​​in ascending order to obtain:

[0039] p1≤p2≤…≤p m

[0040] S43: Set a significance threshold of q, and find the largest k that satisfies:

[0041]

[0042] In the correlation test of brain regions corresponding to the p-value sequence, it is considered that the top k brain regions are correlated.

[0043] Furthermore, methods for obtaining matrix-form metabolic brain networks by optimizing the cocorrelation matrix of brain networks include:

[0044] S44: Obtain the cocorrelation matrix R through S41:

[0045]

[0046] The absolute values ​​of the negative correlation coefficients are processed, and a correlation threshold r is set. δ Based on the magnitude of the absolute value of the correlation, for those with an absolute value lower than r δ An edge is considered to connect two unrelated brain regions, and the corresponding value in the binarization matrix Δ is set to 0 otherwise. The elements in the binarization matrix Δ are obtained by the following formula:

[0047]

[0048] S45: The Hadamard product of the binarized matrix Δ and R is used to obtain the metabolic brain network in matrix form.

[0049] A device, characterized in that it comprises:

[0050] Memory, used to store computer programs;

[0051] A processor for executing a computer program to implement a method for constructing a brain network from synchronously acquired brain structure and metabolic images as claimed in any one of claims 1 to 7.

[0052] This invention discloses a method for constructing a brain network by simultaneously acquiring brain structure and metabolic images. The method involves registering a structural image to be processed with a standard brain template, and then registering the metabolic image with the structural image after registration. This fuses the brain structure information with the metabolic information of the metabolic image to obtain a registered metabolic image. The metabolic image is then processed by brain partitioning to form a multi-partition brain model, reflecting the metabolic information of multiple brain regions and providing a more intuitive representation of the distribution of metabolism within the brain structure. By analyzing the correlations between multiple brain regions, a co-correlation matrix of the brain network is obtained. This co-correlation matrix is ​​then optimized to obtain a matrix-form metabolic brain network. Through brain network analysis, the distribution of brain metabolism within the brain structure can be intuitively and quantitatively displayed. By comparing the normal metabolic brain network, the differences in metabolic indicators of network attributes between patients with brain diseases and normal individuals can be more effectively identified, providing valuable information for brain development, maturation, and aging. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart of a method for constructing a brain network from synchronously acquired brain structure and metabolic images, according to an embodiment of the present invention.

[0055] Figure 2 This is a flowchart illustrating the preprocessing of structural and metabolic images to be processed according to an embodiment of the present invention.

[0056] Figure 3 This is a flowchart illustrating how the registered metabolic image is obtained by fusing brain structural information with metabolic information from a metabolic image, according to an embodiment of the present invention.

[0057] Figure 4 In this embodiment of the invention, metabolic images are processed into brain regions to form a multi-region brain model, which reflects the metabolic information of multiple brain regions.

[0058] Figure 5 This invention provides a flowchart of how a cocorrelation matrix of a brain network is obtained by analyzing the correlation between multiple brain regions, and how a matrix-form metabolic brain network is obtained by optimizing the cocorrelation matrix of the brain network. Detailed Implementation

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0060] This invention includes a method for constructing a brain network based on simultaneously acquired brain structure and metabolic images, the method comprising:

[0061] S1: Acquire structural and metabolic images to be processed, and preprocess the structural and metabolic images to be processed; the acquisition equipment for structural and metabolic images can be an integrated PET / MR device;

[0062] S2: The structural images to be processed are registered with a standard brain template. The structural images are then registered with the metabolic images to fuse the brain structural information with the metabolic information of the metabolic images, thus obtaining the registered metabolic images. Due to individual differences in the human brain, the coordinates of the scanned structural images and metabolic images in space are also different. During the study, it is necessary to first eliminate individual differences and unify the coordinates. That is, it is necessary to correct all the subjects' brains to the standard template to facilitate subsequent statistical analysis and improve the accuracy of network construction.

[0063] S3: The metabolic images are processed into brain regions to form a multi-region brain model, which reflects the metabolic information of multiple brain regions. The brain regions show the structural and functional connections in the brain, and the metabolic situation is displayed through the brain model.

[0064] S4: By analyzing the correlation between multiple brain regions, a cocorrelation matrix of the brain network is obtained. By optimizing the cocorrelation matrix of the brain network, a matrix-form metabolic brain network is obtained. Through brain network analysis, the distribution of brain metabolism in the brain structure can be displayed intuitively and quantitatively. By comparing the normal metabolic brain network, the metabolic differences in network attribute indicators between patients with brain diseases and normal people can be more effectively discovered, providing valuable information for brain development, maturation, and aging.

[0065] Specifically, such as Figure 2 As shown, the methods for preprocessing structural and metabolic images include:

[0066] S11: Perform preliminary screening of structural and metabolic images. By roughly browsing the images, remove data with severe artifacts, occupancy, or other image partitions.

[0067] S12: Clean and desensitize structural and metabolic imaging data to remove sensitive personal information of users;

[0068] S13: If there are differences in resolution between structural and metabolic images due to reasons such as manufacturer or scanning method, then the resolution of structural images shall be unified.

[0069] Specifically, such as Figure 3 As shown, the method for registering the structural image to be processed with a standard brain template includes:

[0070] S21: Import the structural image M and the standard brain template, and find a geometrically optimal alignment. This is achieved by finding an image matrix T(M) that maximizes the similarity between the two volumetric brain images of the standard brain template and the structural image M, where the image matrix T(M) satisfies:

[0071] T(M)=argminC(N,T(M))

[0072] Where C(N,T(M)) is the cost function, and N is the registration matrix of the standard brain template; the standard brain template can be the MNI-152 template, the Chinese2020 template, etc.

[0073] S22: Perform an affine transformation of the structural image on each coordinate point of the image matrix T(M). The affine transformation of the structural image encompasses four dimensions: translation, rotation, scaling, and shearing, to complete the registration of the main structure of the brain. The formula for the affine transformation of the structural image is:

[0074]

[0075] Where (x, y, z) are the original coordinates of the points in T(M), and (x', y', z') are the new coordinates of the points in T(M). M It is a 4x4 matrix that covers the transformation parameters of translation, rotation, scaling and shearing, that is, to realize the transformation operations of coordinate points in four dimensions: translation, rotation, scaling and shearing.

[0076] S23: Perform a nonlinear transformation on each coordinate point of the image matrix T(M) after affine transformation of the structural image, register local regions such as sulci and ventricles, refine local details, and thus obtain the registered structural image M. * Then, a nonlinear transformation is performed on the new coordinates (x', y', z'). The formula for the nonlinear transformation is:

[0077]

[0078] (x”, y”, z”) are the new coordinates in T(M) after nonlinear transformation, d x (x,y,z),d y (x,y,z) and d z (x,y,z) represents the warp field in three directions. The registered structural image is obtained through the above nonlinear transformation.

[0079] Specifically, such as Figure 3 As shown, the methods for registering metabolic images with structural images after registration processing include:

[0080] S24: Import metabolic image O and registered structural image M * The goal is to find a geometrically optimal alignment by finding an image matrix T(O) such that the structural image M... * Two volumetric brain images of metabolic image O have the greatest similarity, where the image matrix T(O) satisfies:

[0081] T(O) = argminC(T)* (M),T(O))

[0082] Among them, C(T) * (M),T(O)) are cost functions;

[0083] S25: Perform affine transformation of metabolic image on each coordinate point of image matrix T(O). The affine transformation of metabolic image covers transformations in four dimensions: translation, rotation, scaling and shearing. Register the main brain structure in metabolic image O with the registered structural image.

[0084] S26: Perform a nonlinear transformation on each coordinate point of the image matrix T(O) after affine transformation of the metabolic image to register local regions such as sulci and ventricles in the metabolic image O, thereby obtaining the registered metabolic image O. * The affine and nonlinear transformations of metabolic images are consistent with the transformation methods in S22 and S23, and will not be described again.

[0085] Specifically, such as Figure 4 As shown, methods for processing metabolic images into brain regions to form multi-region brain models include:

[0086] S31: Import metabolic images and construct the vertices and edges of the brain model through the modeler. The vertices are abstract indicators of the brain voxels with coordinates and gray values ​​in space, and the edges are line segments connecting two vertices. The brain sulcus surface is approximated by as many end-to-end line segments as possible.

[0087] S32: Brain regions are constructed using Bayesian parameter estimation, that is, maximizing the following probabilities:

[0088]

[0089] Where G represents the brain surface shape of the metabolic image, f is a nonlinear spherical transform function, r represents the vertex, and P represents the partition. Given G and f, calculate and maximize the probability that partition P appears in a certain spatial location of the brain.

[0090] Considering that the noise at each vertex is independent of that at other vertices, the above probabilities can also be written in the form of a product, i.e.:

[0091]

[0092] S33: Obtain metabolic information for 90 or 109 brain regions based on different standard brain templates. The metabolic information includes the maximum, minimum, average, and standard deviation of the SUV for each brain region, as well as the SUV range.

[0093] Specifically, such as Figure 5 As shown, methods for constructing brain networks from brain models and obtaining the cocorrelation matrix of the brain network by analyzing the correlations between multiple brain regions include:

[0094] S41: Brain networks are constructed based on the user's age group, using individual brain regions as vertices and Pearson correlation coefficients between different brain regions as edges, to obtain the co-correlation matrix of the metabolic brain network. The Pearson correlation coefficient is calculated according to the following formula:

[0095]

[0096] Where r represents the Pearson correlation coefficient, which ranges from -1 to 1, where -1 indicates a perfect negative correlation between the two variables and 1 indicates a perfect positive correlation between the two variables. i and Y i These represent values ​​for different brain regions. and The set of Pearson correlation coefficients is calculated, where m = n*(n-1) / 2.

[0097] S42: Perform Benjamini-Yekutieli multiple comparisons correction on the set of Pearson correlation coefficients. If the Pearson correlation coefficients of two randomly selected brain regions are independent, then m p-values ​​will be obtained in the Benjamini-Yekutieli multiple comparisons correction. i Given the values ​​(i = 1, 2, 3, ..., m), sort these p values ​​in ascending order to obtain:

[0098] p1≤p2≤…≤p m

[0099] S43: Set a significance threshold of q, and find the largest k that satisfies:

[0100]

[0101] In the correlation test of brain regions corresponding to the p-value sequence, it is considered that the top k brain regions are correlated.

[0102] Specifically, such as Figure 5 As shown, methods for obtaining a matrix-form metabolic brain network by optimizing the cocorrelation matrix of the brain network include:

[0103] S44: Obtain the cocorrelation matrix R through S41:

[0104]

[0105] The absolute values ​​of the negative correlation coefficients are processed, and a correlation threshold r is set. δ Based on the magnitude of the absolute value of the correlation, for those with an absolute value lower than r δ An edge is considered to connect two unrelated brain regions, and the corresponding value in the binarization matrix Δ is set to 0 otherwise. The elements in the binarization matrix Δ are obtained by the following formula:

[0106]

[0107] S45: The Hadamard product is performed between the binarized matrix Δ and R to obtain the metabolic brain network in matrix form; the processing of S44 and S45 is used to remove irrelevant matrix points, thereby enhancing the contrast of the metabolic brain network.

[0108] A device, characterized in that it comprises:

[0109] Memory, used to store computer programs;

[0110] A processor for executing a computer program to implement a method for constructing a brain network from synchronously acquired brain structure and metabolic images as claimed in any one of claims 1 to 7.

[0111] The present invention has been further described above with reference to specific embodiments. However, it should be understood that the specific description herein should not be construed as limiting the nature and scope of the present invention. Various modifications made to the above embodiments by those skilled in the art after reading this specification are all within the scope of protection of the present invention.

Claims

1. A method for constructing a brain network that simultaneously acquires brain structure and metabolic images, characterized in that, The method includes: S1: Acquire the structural and metabolic images to be processed, and preprocess the structural and metabolic images to be processed; S2: The structural image to be processed is registered with a standard brain template. The metabolic image is then registered with the structural image after registration processing, thereby fusing the brain structural information with the metabolic information of the metabolic image to obtain the registered metabolic image. The process of registering the structural image to be processed using a standard brain template includes: S21: Import the structural image M and the standard brain template, find a geometrically optimal alignment, and achieve this by finding an image matrix. This ensures that the standard brain template and the structural image M have the greatest similarity. S22: For the image matrix The values ​​corresponding to each coordinate point are subjected to affine transformation of the structural image to complete the registration of the main structure of the brain; S23: For the image matrix After affine transformation of the structural image, the values ​​corresponding to each coordinate point of the resulting image matrix undergo a nonlinear transformation, and local regions are registered to obtain the registered structural image. The local region includes cerebral sulci and ventricles; The registration of metabolic images with structural images after registration processing includes: S24: Import the metabolic image O and the registered structural image. The goal is to find a geometrically optimal alignment by finding an image matrix. Make structural images It has the greatest similarity to metabolic imaging O; S25: For the image matrix The values ​​corresponding to each coordinate point are subjected to affine transformation of the metabolic image, and the main brain structure in the metabolic image O is registered with the main brain structure in the registered structural image. S26: For the image matrix After affine transformation of the metabolic image, the values ​​corresponding to each coordinate point of the image matrix are subjected to nonlinear transformation, and the local regions in the metabolic image O are registered to obtain the registered metabolic image. ; S3: Perform brain partitioning processing on the registered metabolic images to form a multi-partition brain model. The brain model reflects metabolic information from multiple brain regions. The process of performing brain partitioning processing on the registered metabolic images to form a multi-partition brain model includes: S31: Import the registered metabolic image, and construct the vertices and edges of the brain model through the modeler. The vertex is an abstract representation of the brain in space with coordinates and gray values, and the edge is a line segment connecting two vertices. The brain sulcus surface is approximated by multiple line segments connected end to end. S32: The brain regions are constructed using Bayesian parameter estimation, i.e.: in, Let G be the general probability function, G represent the brain surface shape of the registered metabolic image, f be the nonlinear spherical transformation function, and r represent the vertex. With the sign of proportionality, P represents a brain region. Given G and f, calculate and maximize the probability that brain region P appears in a certain spatial location of the brain. S33: Obtain metabolic information for multiple brain regions based on different standard brain templates. The metabolic information includes the maximum, minimum, average, and standard deviation of the SUV for each brain region, as well as the SUV range information. S4: By analyzing the correlation between multiple brain regions, the cocorrelation matrix of the brain network constructed by the brain model is obtained. The metabolic brain network in matrix form is obtained by optimizing the cocorrelation matrix of the brain network.

2. A device for simultaneously acquiring brain structure and metabolic images of brain networks, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method for constructing a brain network that simultaneously acquires brain structure and metabolic images as described in claim 1.

Images