A dynamic brain functional network learning and display method based on manifold optimization algorithm

Abstract

This specification discloses a method and device for learning and displaying a dynamic brain functional network based on a manifold optimization algorithm. The method and device can obtain blood oxygen signals within a period of time and preprocess the blood oxygen signals through a sliding window to obtain preprocessed signal data. The preprocessed signal data includes the BOLD signal of each brain node in each time window, and then construct an objective function. The objective function is used to characterize the product between the signal distance and the functional connection weight between brain nodes, and to represent the distance between the dynamic graph embeddings on the Grassmann manifold between adjacent time windows. Finally, based on the preprocessed signal data, with minimizing the objective function as the optimization goal, the dynamic brain functional connection network and the dynamic graph embedding corresponding to the dynamic brain functional connection network are determined and displayed, thereby effectively determining the dynamic brain functional connection network and dynamic graph embedding that reflect the actual functional connection status of the brain.

Description

Technical Field

[0001] This specification relates to the field of graph machine learning, and in particular to a method and device for learning and displaying dynamic brain functional networks based on a manifold optimization algorithm. Background Art

[0002] The brain can be viewed as a hierarchical network system composed of multiple interconnected brain regions. Analyzing the topological structure and functional connectivity of brain networks is crucial for a deeper understanding of information transmission and processing within the brain, as well as the interactions between different brain regions.

[0003] Most previous brain network analysis studies were based on static brain networks. The construction of static brain networks is based on the assumption that the brain's connectivity state is static over time. However, in reality, the brain is essentially a dynamic system, and different brain regions change over time or with task demands, cognitive state, and environment. Studies have also shown that the dynamic performance of the brain is closely related to the occurrence and progression of disease. Therefore, how to construct a more effective dynamic brain functional network that better conforms to the characteristics of brain networks can provide a more complete and comprehensive understanding of the evolution of the brain's dynamic connectivity. This is of great significance for understanding the brain's information processing mechanisms in different cognitive tasks, abnormal changes in disease states, and the development and aging of brain networks.

[0004] Constructing dynamic brain network connectivity is fundamental and the first step in dynamic brain network analysis. Traditional methods for constructing dynamic brain network functional connectivity typically use a sliding window to acquire fMRI signal segments and then calculate the Pearson coefficient of the fMRI signal at each brain region. This approach is significantly affected by noise and ignores the temporal consistency of dynamic changes in fMRI signals. Therefore, we propose a dynamic brain functional network learning method based on a manifold optimization algorithm to learn dynamic brain functional networks, which can more accurately reflect the dynamic functional connectivity changes in the brain. Summary of the Invention

[0005] This specification provides a dynamic brain function network learning method and device based on a manifold optimization algorithm to partially solve the above-mentioned problems existing in the prior art.

[0006] This manual adopts the following technical solutions:

[0007] This manual provides a method for learning and displaying a dynamic brain functional network based on a manifold optimization algorithm, including:

[0008] Acquire a blood oxygen signal over a period of time, and preprocess the blood oxygen signal using a sliding window to obtain preprocessed signal data, wherein the preprocessed signal data includes the BOLD signal of each brain node in each time window;

[0009] constructing an objective function for representing the product between the signal distance and the functional connection weight between each brain node and for representing the distance between adjacent time windows of the dynamic graph embedding on the Grassmann manifold;

[0010] Taking minimizing the objective function as the optimization goal, the adjacency matrix and graph embedding matrix corresponding to the preprocessed signal data are optimized based on the manifold optimization algorithm to determine the dynamic brain functional connection network and the dynamic graph embedding corresponding to the dynamic brain functional connection network and display them.

[0011] Optionally, preprocessing the blood oxygen signal by a sliding window method to obtain preprocessed signal data specifically includes:

[0012] Determining, based on the blood oxygen signal and a preset brain region of interest template, an average value of the BOLD-fMRI signal of each voxel in each brain node during the period of time to obtain two-dimensional BOLD signal data;

[0013] The signal data of the preset time lengths at the beginning and end of the two-dimensional BOLD signal data are removed, and the BOLD signal data in each time window is obtained by sliding the window as the preprocessed signal data.

[0014] Optionally, minimizing the objective function is used as an optimization goal, and the adjacency matrix and graph embedding matrix corresponding to the preprocessed signal data are optimized based on a manifold optimization algorithm to determine and display a dynamic brain functional connectivity network and a dynamic graph embedding corresponding to the dynamic brain functional connectivity network, specifically including:

[0015] determining an initial adjacency matrix and an initial graph embedding matrix based on the preprocessed signal data;

[0016] In a first round of iteration, a first objective function is obtained according to the initial adjacency matrix and the objective function, and the initial graph embedding matrix is ​​optimized with minimizing the first objective function as the optimization objective to obtain an optimized graph embedding matrix, the gradient of the first objective function relative to the graph embedding matrix is ​​calculated to obtain a second objective function, and the optimized graph embedding matrix is ​​brought into the second objective function, and the initial adjacency matrix is ​​optimized with minimizing the second objective function as the optimization objective to obtain an optimized adjacency matrix, and in subsequent iterations, the optimized graph embedding matrix is ​​continuously optimized by the first objective function, and the optimized adjacency matrix is ​​continuously optimized by the second objective function;

[0017] When the preset iteration termination condition is met, the iteration is terminated;

[0018] According to the optimized graph embedding matrix and the optimized adjacency matrix, a dynamic brain functional connectivity network and a dynamic graph embedding corresponding to the dynamic brain functional connectivity network are determined.

[0019] Optionally, the first objective function is:

[0020] Among them, F t It is used to represent the initial graph embedding matrix or the optimized graph embedding matrix in the t-th time window; the second objective function is Among them, W t Used to represent the initial adjacency matrix or optimized adjacency matrix under the t-th time window, is the graph embedding similarity matrix, is the signal similarity matrix, and β and γ are constraint weight parameters.

[0021] Optionally, the preset iteration termination condition is: using the graph embedding matrix and adjacency matrix obtained by optimization in the current round and the previous round respectively as inputs of the loss function, and when the value of the loss function is less than a preset threshold, the optimization ends.

[0022] Optionally, determining an initial adjacency matrix and an initial graph embedding matrix includes:

[0023] Calculating an initial adjacency matrix based on the similarity of BOLD signals between adjacent brain nodes in the preprocessed signal data;

[0024] An initial graph embedding matrix is ​​determined according to the Laplace matrix corresponding to the initial adjacency matrix.

[0025] This specification provides a dynamic brain function network learning and display device based on a manifold optimization algorithm, including:

[0026] A preprocessing module is used to obtain blood oxygen signals over a period of time and preprocess the blood oxygen signals through a sliding window to obtain preprocessed signal data, wherein the preprocessed signal data includes the BOLD signal of each brain node in each time window;

[0027] A construction module for constructing an objective function, wherein the objective function is used to represent the product between the signal distance and the functional connection weight between brain nodes and the distance between adjacent time windows embedded in the dynamic graph on the Grassmann manifold;

[0028] An optimization module is used to optimize the adjacency matrix and graph embedding matrix corresponding to the preprocessed signal data based on a manifold optimization algorithm with minimization of the objective function as the optimization goal, determine the dynamic brain functional connection network and the dynamic graph embedding corresponding to the dynamic brain functional connection network, and display it.

[0029] Optionally, the preprocessing module is specifically used to determine the average value of the BOLD-fMRI signal of each voxel in each brain node within the period of time based on the blood oxygen signal and a preset brain region of interest template to obtain two-dimensional BOLD signal data; remove the signal data of the preset time length at the beginning and end of the two-dimensional BOLD signal data, and obtain the BOLD signal data in each time window through a sliding window as the preprocessed signal data.

[0030] This specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned dynamic brain function network learning and display method based on the manifold optimization algorithm.

[0031] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, the above-mentioned dynamic brain functional network learning and display method based on the manifold optimization algorithm is implemented.

[0032] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0033] It can be seen from the above-mentioned dynamic brain functional network learning and display method based on the manifold optimization algorithm that the blood oxygen signal within a period of time is obtained, and the blood oxygen signal is preprocessed by means of a sliding window to obtain preprocessed signal data. The preprocessed signal data includes the BOLD signal of each brain node in each time window, and then an objective function is constructed. The objective function is used to characterize the product between the signal distance and the functional connection weight between brain nodes, and to represent the distance between the dynamic graph embeddings of adjacent time windows on the Grassmann manifold. Finally, based on the preprocessed signal data, with minimizing the objective function as the optimization goal, the dynamic brain functional connection network and the dynamic graph embedding corresponding to the dynamic brain functional connection network are determined and displayed.

[0034] This method uses fMRI signals to determine the brain's dynamic functional connectivity network and its low-dimensional graph embedding on a Grassmann manifold. The brain functional connectivity network learned through manifold optimization algorithms more accurately reflects the brain's true functional connectivity. Dynamic brain network analysis based on this network can more effectively understand the brain's information processing mechanisms during different cognitive tasks and abnormal changes in disease states.

[0035] The reason why the above-mentioned manifold optimization algorithm is used in this specification to optimize the adjacency matrix and the graph embedding matrix is ​​to combine the principle that the brain is a manifold (the brain is round). Therefore, when calculating the distance between the graph embedding matrix in adjacent time windows, the manifold distance is combined. And when optimizing the adjacency matrix and the graph embedding matrix using the manifold optimization algorithm, the graph embedding matrix is ​​first optimized in a manifold manner, and then the adjacency matrix is ​​optimized using the optimized graph embedding matrix. In this way, the dynamic brain functional connection network obtained by the final optimized adjacency matrix can better conform to the actual characteristics of the brain. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The exemplary embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation of this specification. In the drawings:

[0037] Figure 1 This is a flowchart of a dynamic brain functional network learning and display method based on a manifold optimization algorithm provided in this manual;

[0038] Figure 2 A schematic diagram of pre-processed signal data provided in this specification;

[0039] Figure 3 A schematic diagram of the optimization steps of an optimization algorithm provided in this specification;

[0040] Figure 4 This is a schematic diagram of a dynamic brain function network learning and display device based on a manifold optimization algorithm provided in this manual;

[0041] Figure 5 The corresponding Figure 1 Schematic diagram of electronic equipment. DETAILED DESCRIPTION

[0042] To make the objectives, technical solutions, and advantages of this specification more clear, the following will clearly and completely describe the technical solutions of this specification in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this specification.

[0043] The technical solutions provided by the embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0044] In this specification, a method is provided to analyze fMRI signals to obtain a dynamic brain functional connection network of the patient's or user's brain over a period of time and a graph embedding (dynamic graph embedding) corresponding to the dynamic brain functional connection network. The obtained dynamic brain functional connection network and the corresponding dynamic graph embedding can be used for disease analysis, prediction, medical research, etc. in medical scenarios.

[0045] It should be noted that when determining the dynamic brain functional connection network and the corresponding dynamic graph embedding using the above method, it is necessary to partition the brain to determine each brain area, that is, each brain node.

[0046] Specifically, in step (1), the brain can be divided into different brain regions by using techniques such as structural magnetic resonance imaging (sMRI), and brain nodes are used to represent different brain regions.

[0047] The brain is a complex system, with each region possessing its own unique structure and function. These regions interact and influence each other, working together to accomplish complex cognitive and behavioral tasks. The connections and interactions between these regions may also change during different cognitive and emotional processes. Structural Magnetic Resonance Imaging (sMRI) can be used to identify different brain regions. sMRI provides high-resolution images that reveal the structure and morphology of biological tissues, making it widely used in brain anatomical research.

[0048] It should be noted that in this embodiment, the AAL (Automated Anatomical Labeling) method can be used to partition the brain. The construction process of the AAL template mainly includes the following steps:

[0049] 1. Obtain reference brain imaging data: Use technologies such as magnetic resonance imaging to obtain brain imaging data from multiple healthy individuals.

[0050] 2. Standardization: These brain image data are standardized so that they have the same spatial coordinate system and anatomical structure.

[0051] 3. Segmentation of different brain regions: Use automatic segmentation algorithms, manual operations, or a combination of both to segment the brain into different brain regions. In the AAL template, the brain is segmented into 45 left regions and 45 right regions, each with a representative rectangular area.

[0052] 4. Naming different brain regions: Based on existing anatomical knowledge, different brain regions are named. These names are usually based on the location of the region (such as the dorsolateral gyrus), its connection with other regions (such as the anterior cingulate gyrus), or its function (such as the temporal pole).

[0053] 5. Evaluation and correction: Evaluate the divided brain areas and make corrections if errors or inaccuracies are found.

[0054] By using the AAL template, we can quickly and accurately divide the brain into different brain regions and use these brain regions as regions of interest in research. Each brain region is represented by a brain node, and the node set is represented as: V = {v1, v2, v3, ..., v 90}, with the adjacency matrix The brain functional connection network is encoded in the form of ij Represents node v i and v j The functional connection weight between the two brain regions represents the strength of the functional connection between them.

[0055] Correlation analysis is usually used to calculate w ij This method determines the strength of the connection between two or more brain regions by calculating the correlation coefficient of the blood oxygen signals. This method can use the Pearson correlation coefficient, Spearman rank correlation coefficient, or covariance matrix, etc.

[0056] Specifically, the method involves dividing the brain into several regions. Then, using techniques such as fMRI, the signal strength data for each region is acquired over different time periods. The Pearson coefficient is then calculated between each pair of regions to determine the functional connectivity weight between them.

[0057] The detailed processing steps are as follows: For two brain regions A and B, first express their signal intensity data at different time points as x A (t) and x B (t), then the Pearson coefficient between them can be calculated by the following formula:

[0058]

[0059] Where T represents the total number of time periods, and Respectively represent the average signal intensity data of brain areas A and B at all time points.

[0060] For the adjacency matrix (mentioned below) W, the Laplace matrix L can be calculated by the formula L = DW, where D = diag (d1, d2, ..., d90 ), each element on the main diagonal The Laplace matrix is ​​an important tool for brain network analysis. It not only reveals the topological structure of the brain network, but also can obtain a set of eigenvectors and eigenvalues ​​by performing eigenvalue decomposition on the Laplace matrix. These eigenvectors and eigenvalues ​​represent the spectral characteristics of the brain network and describe the spectral characteristics of the network.

[0061] That is, for the adjacency matrix W, its symmetric graph Laplace matrix L can be calculated by the formula: L = DW, where the diagonal matrix D = diag (d1, d2, ..., d N ), each element on the main diagonal The Laplacian matrix contains the topological information of the brain network.

[0062] The signal at each node of the brain network is: X∈R N×m =[x1,...,x N ] T , where each row x of X i ∈R n Represents the signal on node i, such as fMRI signals.

[0063] After determining each brain region based on the above method, we can start to solve the dynamic brain functional connectivity network and the corresponding dynamic graph embedding from the real blood oxygen signal. It should be noted that after determining the final adjacency matrix and graph embedding matrix, we can construct a visual dynamic brain functional connectivity network and the corresponding dynamic graph embedding for display.

[0064] The functional connection weights between each two brain nodes contained in the adjacency matrix are the edge weights between each two brain nodes in the dynamic brain functional connection network. Therefore, the display mentioned here may refer to displaying the functional connection weights in the adjacency matrix in the dynamic brain functional connection network in a visual form, and the determined dynamic graph embedding may also be displayed in an encoded form.

[0065] It should also be noted that for blood oxygen signals over a period of time, this method can not only determine the dynamic brain functional connectivity network (i.e., dynamic graph) and the corresponding graph embedding. In practical applications, this method can also be used to determine the dynamic brain functional connectivity network of Alzheimer's patients at different stages (such as the early, middle, and late stages of the disease), and through the functional connection weights between each node in the dynamic brain functional connectivity network, determine the target brain region and display it to relevant personnel (such as patients, doctors, etc.) to assist doctors in decision-making. The target brain region is the brain region determined to be associated with the onset of Alzheimer's disease.

[0066] The target brain region mentioned above may refer to the brain region corresponding to the brain nodes with higher functional connection weights in the dynamic brain functional connection network.

[0067] Figure 1 The following is a flowchart of a method for learning and displaying a dynamic brain functional network based on a manifold optimization algorithm provided in this specification, which specifically includes the following steps:

[0068] S100: Acquire a blood oxygen signal within a period of time, and preprocess the blood oxygen signal through a sliding window to obtain preprocessed signal data, wherein the preprocessed signal data includes the BOLD signal of each brain node in each time window.

[0069] First, the server can obtain the blood oxygen signal within a period of time and preprocess the blood oxygen signal through a sliding window to obtain the preprocessed signal data. The preprocessed signal data includes the BOLD signal of each brain node in each time window.

[0070] The blood oxygen signal within the above period of time (the functional magnetic resonance signal of the original blood oxygen dependence level) is in the form of a four-dimensional matrix, that is, three-dimensional space×one-dimensional time.

[0071] Then, based on the blood oxygen signal and the preset brain region of interest template, the average value of the BOLD-fMRI signal of each voxel in each brain node over a period of time can be determined to obtain two-dimensional BOLD signal data, and the signal data of the preset time length at the beginning and end of the two-dimensional BOLD signal data can be removed (denoising), and the BOLD signal data in each time window can be obtained by sliding the window as the preprocessed signal data.

[0072] The raw signal data (blood oxygen signal acquired over a period of time) can be converted into a two-dimensional matrix V × T, where V represents the number of voxels in the magnetic resonance image and T is the number of sampling time points. Based on the ROI template of the brain region of interest (the brain region of interest template described above), the average BOLD-fMRI signal of all voxels in each ROI is calculated. This average value represents the BOLD-fMRI signal value of the ROI region. The two-dimensional matrix can be converted into M × T, where M is the number of ROI brain regions and T is the number of sampling time points.

[0073] Figure 2 This is a schematic diagram of pre-processed signal data provided in this specification.

[0074] like Figure 2As shown in the figure, the idea of ​​sliding window is used to process the two-dimensional BOLD signal. Set the time window size to M×W, where M is the number of ROI brain regions and W is the number of sampling points. Set the sliding step size to S sampling points. Use the sliding window to segment the BOLD signal data, and you can get a three-dimensional matrix of M×W×N. (preprocessed signal data), where N represents the number of time windows, which is also the number of graphs in the constructed dynamic brain functional connection network.

[0075] In the present invention, the experimental data are processed in the following manner:

[0076] The brain was divided into 90 regions using the AAL90 brain region template, and signals were collected from each region 170 times, with a 2-second interval between each signal. This yielded a two-dimensional BOLD signal matrix of 90 × 170 pixels.

[0077] During MRI, magnetic resonance noise is generated by the rapid change in the magnetic field and the current generated by the gradient coils. This noise is particularly noticeable when the MRI machine is started and shut down. To eliminate the effects of this noise, the first 10 seconds and the last 10 seconds of the BOLD signal are removed. After removing the noise, the BOLD signal matrix size is 90×160. Setting the time window size to 90×25 and the sliding step size to 15 results in a 90×25×10 three-dimensional BOLD signal matrix, namely:

[0078] S102: Constructing an objective function, wherein the objective function is used to characterize the product between the signal distance and the functional connection weight between brain nodes and to represent the distance between adjacent time windows embedded in the dynamic graph on the Grassmann manifold.

[0079] After signal preprocessing, the objective function for solving the dynamic brain functional connection network and dynamic graph embedding can be determined. The objective function can be used to represent the product between the signal distance between brain nodes and the functional connection weight, and to represent the distance between adjacent time windows in the dynamic graph embedding on the Grassmann manifold. The objective function can be specifically formula (11) in the following derivation process.

[0080] The following is the derivation process of the objective function:

[0081] For node pairs with high functional connectivity strength (v i ,v j ), it is clear that there are more neuronal connections between nodes, and this connection usually means that certain functions or tasks require coordination and cooperation between these areas. Therefore, when there are high-strength connections between these nodes, the BOLD signals on the nodes will show similar signal characteristics.

[0082] Specifically, as shown in formula (1), the calculation node pair (v i ,v j ) on the BOLD signal (x i ,x j ) between the square of the Euclidean distance, and use the square of the Euclidean distance to measure the signal (x i ,x j ) similarity.

[0083]

[0084] d(v i ,v j ) is the pairwise distance. Obviously, if d(v i ,v j ) value is larger, the node pair (v i ,v j ) the lower the functional connectivity strength. On the contrary, if d(v i ,v j ) value is smaller, the node pair (v i ,v j ) has a higher functional connectivity strength.

[0085] Furthermore, as shown in formula (2), the pairwise distances of all node pairs in the brain network are calculated and multiplied by the functional connection weight values ​​of the node pairs.

[0086]

[0087] Where W ij represents the connection weight between node i and node j, and tr() represents the trace of the matrix.

[0088] Furthermore, solving the Laplace matrix of the brain functional connection network is transformed into solving the following optimization problem:

[0089]

[0090] Since the Laplace matrix obtained by optimization of formula (3) is difficult to solve the brain functional connection matrix from the Laplace matrix, we rewrite formula (3) to establish the objective function for solving the brain functional connection matrix. The specific operations are as follows:

[0091] First, define Formula (2) can be equivalent to formula (4):

[0092]

[0093] in Represents the Hadamard product.

[0094] Then, solving the brain functional connection network is transformed into solving the following optimization problem:

[0095]

[0096] Finally, according to the desired brain functional connection network, the Laplace matrix can be obtained using L=DW.

[0097] The aforementioned method for solving the Laplacian matrix simply analyzes and models the BOLD signal and obtains the result through optimization. Numerous studies have shown that the neural networks formed by the brain's neuronal connections have a manifold structure. This means that the brain's neuronal connections can be viewed as a complex spatial form and can be modeled and analyzed using manifold learning methods.

[0098] The BOLD signal obtained from step (2) Obviously, the BOLD signals of adjacent time windows are more similar than those of non-adjacent time windows because the acquisition time interval is short and contains repeated sampling segments with a length of 10 sampling points. Then, the dynamic functional connectivity network obtained from the BOLD signal is In the , adjacent functional connection networks should also show higher similarity. In order to measure this similarity, the manifold learning method is used for analysis and modeling. The specific operations are as follows:

[0099] First, we solve the dynamic graph embedding on the Grassmann manifold.

[0100] for Find each L t The smallest k eigenvalues, denoted as σ k (L t ), the eigenvector combination corresponding to the k eigenvalues ​​is used as the graph embedding F t In this specific implementation case, k is defined as 12, then F t ∈R 90×12 It can be expressed as L t The low-dimensional embedding of the Laplace matrix satisfies the orthogonal constraint, that is: F t T F t =I. Perform the above operation on each time window to get the dynamic graph embedding Furthermore, the graph embedding is mapped onto the Grassmann manifold, as shown in Equation (6), and the adjacent graph embedding F is calculated. t and F t+1 Distance on a Grassmann manifold:

[0101]

[0102] Where p is equal to the number of columns in the graph embedding, that is, p = k = 12. The similarity of adjacent functional connection networks is measured by the distance between the graph embeddings on the Grassmann manifold. Furthermore, as shown in Equation (7), solving the dynamic graph embedding is transformed into solving the following optimization problem:

[0103]

[0104] Combining equations (6) and (7), we get the objective function as follows:

[0105]

[0106] Then, the brain functional connectivity network is solved from the graph embedding and BOLD signal based on the Grassmann manifold. The optimized graph embedding is regarded as the signal on the node. Therefore, the signal on the node contains two parts: BOLD signal and graph embedding signal, as shown in Equation (9): Solving the dynamic Laplace matrix is ​​transformed into solving the following optimization problem:

[0107]

[0108] Furthermore, by combining Equation (8) and Equation (9), we construct the energy function for obtaining the dynamic Laplacian matrix and dynamic graph embedding:

[0109]

[0110] It should be noted that it is difficult to obtain the dynamic functional connectivity network by solving the dynamic Laplace matrix obtained from equation (10). Therefore, we first solve it by rewriting equation (11) The detailed steps are as follows:

[0111] Introduction where x i ,x j are the blood oxygen signals on nodes i and j respectively, f i ,f j are the vectors in the graph embedding matrix corresponding to nodes i and j, respectively, so we can get:

[0112]

[0113] Formula (10) can be transformed into Formula (11) to construct the energy function for obtaining the dynamic brain functional connectivity matrix and dynamic graph embedding:

[0114]

[0115] Solved Then, through L=DW, we can get

[0116] That is, formula (11) is the final objective function, which combines formula (5) and formula (6). Therefore, it can be used to represent the product between the signal distance and the functional connection weight between brain nodes, and can also be used to represent the distance between the dynamic graphs of adjacent time windows embedded on the Grassmann manifold.

[0117] S104: Based on the preprocessed signal data, with minimizing the objective function as the optimization goal, a dynamic brain functional connectivity network and a dynamic graph corresponding to the dynamic brain functional connectivity network are determined to be embedded and displayed.

[0118] It should be noted that when determining the dynamic brain functional connectivity network and the dynamic graph embedding corresponding to the dynamic brain functional connectivity network, the initial adjacency matrix and the initial graph embedding matrix of the dynamic brain functional connectivity network can be determined first. The method for determining the initial adjacency matrix and the initial graph embedding matrix can be the method mentioned in step 1:

[0119] That is, we can first calculate the initial adjacency matrix based on the similarity of BOLD signals between adjacent brain nodes (such as the Pearson correlation coefficient), and then determine the Laplace matrix corresponding to the initial adjacency matrix and perform eigenvalue decomposition on the Laplace matrix to obtain the eigenvectors and eigenvalues ​​corresponding to the Laplace matrix to obtain the initial graph embedding matrix. Specifically, we can take the Laplace matrix Each L t The smallest 12 eigenvalues ​​σ 12 (L t ), combine the eigenvectors corresponding to the eigenvalues ​​to obtain Among them F t ∈R 90×12 .

[0120] Then, the initial adjacency matrix and the initial graph embedding matrix are optimized by the objective function in the above step S102 to obtain the optimized adjacency matrix and the optimized graph embedding matrix. During the optimization, a preset manifold optimization algorithm needs to be used for optimization. It is necessary to first fix the adjacency matrix, optimize the graph embedding matrix, and fix the graph embedding matrix before optimizing the adjacency matrix.

[0121] The reason why the above-mentioned manifold optimization algorithm is used to optimize the adjacency matrix and the graph embedding matrix in this specification is to combine the principle that the brain is a manifold (the brain is round). Therefore, when calculating the distance of the graph embedding matrix under adjacent time windows, combining the manifold distance, and optimizing the adjacency matrix and the graph embedding matrix using the manifold optimization algorithm, the graph embedding matrix is ​​first optimized in a manifold manner, and then the adjacency matrix is ​​optimized using the optimized graph embedding matrix. In this way, the dynamic brain functional connection network obtained by the final optimized adjacency matrix can better conform to the actual characteristics of the brain.

[0122] Specifically, in the first round of iteration, a first objective function can be obtained based on the initial adjacency matrix and the objective function, and the initial graph embedding matrix is ​​optimized with minimizing the first objective function as the optimization goal to obtain the optimized graph embedding matrix, and the gradient of the first objective function relative to the graph embedding matrix is ​​calculated to obtain the second objective function, and the optimized graph embedding matrix is ​​brought into the second objective function, and the initial adjacency matrix is ​​optimized with minimizing the second objective function as the optimization goal to obtain the optimized adjacency matrix. In subsequent iterations, the optimized graph embedding matrix continues to be optimized by the first objective function, and the optimized adjacency matrix is ​​optimized by the second objective function.

[0123] The first objective function can be: Among them, F t It is used to represent the initial graph embedding matrix or the optimized graph embedding matrix in the t-th time window; the second objective function is Among them, W t Used to represent the initial adjacency matrix or optimized adjacency matrix under the t-th time window, is the graph embedding similarity matrix, is the similarity matrix of the signal, β and γ are constrained weight parameters, the role of parameter β is to keep the edges in more connected, and the role of parameter γ is to control the sparsity. The larger the value of parameter γ, the sparser the solution.

[0124] Among them, the above optimization process needs to be implemented through optimization algorithms, as follows:

[0125] For the objective function (11) established above, a step-by-step optimization method can be used to solve it. The specific steps are as follows:

[0126] Step 1: Fix (i.e., fixing the adjacency matrix, is the Laplace matrix corresponding to the adjacency matrix), optimize (Graph Embedding Matrix).

[0127] optimization The objective function is shown in formula (12):

[0128]

[0129] Optimize F on the Grassmann manifold t It involves two steps:

[0130] 1. In F t Calculate the Grassmann gradient ΔF in the tangent space t .

[0131] Because the first time window is only adjacent to the second time window, and the last window is only adjacent to the previous time window, the energy function J(F t ), calculate J(F t ) t The gradient of is shown in formula (13):

[0132]

[0133] Then the Grassmann gradient ΔF t , the Euclidean gradient can be transformed into Projected into the tangent space, as shown in formula (14):

[0134]

[0135] 2. Map the Grassmann gradient from the tangent space to the Grassmann manifold and then update F t As shown in formula (14):

[0136]

[0137] Where U, ∑ and V are The result of singular value decomposition (SVD), for example: Where β is a positive parameter that controls the step size during the optimization process.

[0138] 3. Define Optimization The loss function

[0139] Define the loss function: in and F t They are respectively the graph embedding optimization results of the new round and the old round during the optimization process. Define the threshold ε=1e-3, when the loss is less than ε, the optimization end condition is met and the optimization result is output

[0140] Step 2: Optimized by step 1 optimization

[0141] Constructing a dynamic brain functional connectivity network Energy function:

[0142]

[0143] Further: In order to prevent the optimization of W t A trivial solution appears, namely: W t =0, and W t Should be appropriately sparse, that is: W tThe number of values ​​​​of 0 should be as large as possible. Add constraints to formula (16), as shown in formula (17):

[0144]

[0145] where 1=[1,...,1] T , logarithmic constraint log(W t 1) Make the degree of the nodes in W positive, but do not make the elements in W become 0, so as to avoid the trivial solution of W, which improves the overall connectivity of the graph without affecting the sparsity. However, adding only the logarithmic constraint term will make W very sparse, so the norm constraint term is added To control the sparsity of the graph, the larger the value of the parameter γ, the sparser the solved graph.

[0146] The W solved by formula (17) is in W m space, When using specific optimization techniques to solve the optimization problem of Equation (17), in order to make the optimization easier, W can be transformed into m Space is transformed into W v space, Therefore, it is not necessary to deal with the symmetry problem of W, thus reducing the difficulty of optimization. v In space, matrices W and Z are represented in the form of vectors, specifically described as: w = vec(W t ),w∈R 4005 ×1 ,z=vec(Z Ft +Z Xt ),z∈R 4005×1 . Some in W v Space and W v The equivalent terms in space are shown in the following table:

[0147]

[0148] Therefore, as shown in Equation (18), Equation (17) can be transformed into the sum of the following three functions so that it can be solved using the primal-dual algorithm:

[0149]

[0150] Where K is a linear operator satisfying W1=Kw. Further, define:

[0151] f1(w)=I{w≥0}+2w T z

[0152] f2(d)=-α1 T log(d)

[0153] f3(w)=β||w|| 2

[0154] Where I{} is the indicator function, when the conditions in the brackets are met, the value is zero, otherwise, the value is infinite. Where d = Kw∈R m Then, the primal-dual algorithm is used to solve the objective function. The specific optimization process is as follows:

[0155] Input:z,α,β,γ,w 0 ∈W v , γ,toleranceε

[0156] for i=1,...,i max do

[0157] y i =w i -γ(2βw i +S T d i )

[0158]

[0159] p i =max(0,y i -2γz)

[0160]

[0161] q i =p i -γ(2βp i +S T p i )

[0162]

[0163] w i =w i -y i +p i ;

[0164]

[0165] if||w i -w i-1 || / ||w i-1 ||<εand

[0166] ||d i -d i-1 || / ||d i-1 ||<εthen

[0167] break

[0168] end if

[0169] end for

[0170] Finally, for each W t , calculated by the formula L=DW

[0171] It should be noted that when the preset iteration termination condition is met, the iteration can be terminated, and the dynamic brain functional connection network and the dynamic graph embedding corresponding to the dynamic brain functional connection network are determined based on the optimized graph embedding matrix and the optimized adjacency matrix, that is, the optimized graph embedding matrix and the optimized adjacency matrix obtained in the last iteration, the optimized adjacency matrix is ​​used as the adjacency matrix of the final dynamic brain functional connection network, and the optimized graph embedding matrix is ​​the dynamic graph embedding corresponding to the dynamic brain functional connection network, which is used to characterize the topological structure of the dynamic brain functional connection network.

[0172] Figure 3 This is a schematic diagram of the optimization steps of an optimization algorithm provided in this specification.

[0173] like Figure 3 As shown, the optimization algorithm needs to go through the following steps:

[0174] 1. Define the number of iterations iter, and the BOLD signal obtained after preprocessing As input, the initial And calculate the initial

[0175] 2. Update graph embedding based on Grassmann manifold Then, from and In the combined signal, the optimization algorithm is used to solve the problem. and

[0176] 3. Calculate the loss value, as shown in formula (19), and define the energy function:

[0177]

[0178] As shown in formula (20), the loss function is defined as

[0179] and the F solved in the previous iteration t , L t Substitute into the energy function of formula (19) and calculate the loss value of each iteration:

[0180]

[0181] Define the threshold ε = 1e-3. When the loss value is less than ε, the optimization end condition is met and the dynamic functional connection network and dynamic graph embedding are output. If the threshold condition is not met, repeat step 2 and continue optimization. and and The optimization ends when the loss value is less than ε.

[0182] In addition, in this specification, the results of determining the target brain region are only provided to an evaluator (e.g., a doctor) who assesses the target user's mental health status. This provides the evaluator with reference biological indicators when assessing the target user's mental health status, allowing the evaluator to obtain a more accurate assessment result. Furthermore, the embodiments of this specification do not limit the manner in which the evaluator uses the predicted results as a reference to assess the target user's mental health status.

[0183] It should be noted that, for the sake of ease of description, the execution subject of this method is described as a server. The execution subject of this method can be a computer, a large service platform, etc., which is not limited here.

[0184] At least one of the above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: the present invention first processes the fMRI data to obtain the BOLD signal, and uses a sliding window to divide the BOLD signal into multiple time periods; then, a manifold optimization algorithm based on the Grassmann manifold is designed to optimize and obtain a graph embedding based on the Grassmann manifold; finally, the BOLD signal and the graph embedding data are combined, and a dynamic brain functional connection matrix is ​​optimized from the combined signal, namely, dynamic graph learning.

[0185] Dynamic brain functional connectivity networks are a method for measuring the coordination between different brain regions. They study the information transmission and interaction patterns between brain regions and can be applied to many fields of neuroscience. Their research significance includes but is not limited to the following:

[0186] 1. Understand how the brain works: Dynamic brain functional connectivity networks can help us gain a deeper understanding of how the brain works, including interactions between different regions, information processing and transmission, etc.

[0187] 2. Exploring neurological diseases: Studies have found that the pathogenesis of mental illness may involve changes in the functional connectivity matrix of brain networks. For example, patients with schizophrenia may have abnormal connectivity in the prefrontal cortex-amygdala pathway, and patients with depression may have weakened connectivity in the cingulate gyrus-amygdala pathway. Therefore, the dynamic graph learning method proposed in this invention can timely and accurately represent the evolution of the brain's functional connectivity network, as well as the Laplace matrix that can represent the brain's topological structure, which directly plays an important role in the pathogenesis, prevention, and treatment of many diseases.

[0188] In summary, the study of dynamic brain functional connectivity networks is of great significance for understanding the complexity of the brain as well as the mechanisms and treatments of neurological diseases.

[0189] The above is a dynamic brain function network learning and display method based on the manifold optimization algorithm provided by one or more embodiments of this specification. Based on the same idea, this specification also provides a dynamic brain function network learning and display device based on the manifold optimization algorithm, such as Figure 4 shown.

[0190] Figure 4 This specification provides a schematic diagram of a dynamic brain function network learning and display device based on a manifold optimization algorithm, including:

[0191] A preprocessing module 401 is configured to acquire a blood oxygen signal over a period of time and preprocess the blood oxygen signal using a sliding window to obtain preprocessed signal data, wherein the preprocessed signal data includes the BOLD signal of each brain node in each time window;

[0192] A construction module 402 is configured to construct an objective function, wherein the objective function is used to represent the product between the signal distance and the functional connection weight between brain nodes and the distance between adjacent time windows embedded in the dynamic graph on the Grassmann manifold;

[0193] The optimization module 403 is used to optimize the adjacency matrix and graph embedding matrix corresponding to the preprocessed signal data based on the manifold optimization algorithm with minimization of the objective function as the optimization goal, determine the dynamic brain functional connection network and the dynamic graph embedding corresponding to the dynamic brain functional connection network, and display it.

[0194] Optionally, the preprocessing module 401 is specifically used to obtain the original signal data corresponding to the blood oxygen signal within the period of time; determine the average value of the BOLD-fMRI signal of each voxel in each brain node within the period of time based on the original signal data and a preset brain region of interest template to obtain two-dimensional BOLD signal data; remove the signal data of the preset time length at the beginning and end of the two-dimensional BOLD signal data, and obtain the BOLD signal data in each time window through a sliding window as the preprocessed signal data.

[0195] Optionally, the optimization module 403 is specifically used to determine an initial adjacency matrix and an initial graph embedding matrix based on the preprocessed signal data; in a first round of iteration, a first objective function is obtained based on the initial adjacency matrix and the objective function, and the initial graph embedding matrix is ​​optimized with minimizing the first objective function as the optimization goal to obtain an optimized graph embedding matrix, the gradient of the first objective function relative to the graph embedding matrix is ​​calculated to obtain a second objective function, and the optimized graph embedding matrix is ​​brought into the second objective function, and the initial adjacency matrix is ​​optimized with minimizing the second objective function as the optimization goal to obtain an optimized adjacency matrix, and in subsequent iterations, the optimized graph embedding matrix continues to be optimized by the first objective function, and the optimized adjacency matrix is ​​optimized by the second objective function; when the preset iteration termination condition is met, the iteration is terminated; according to the optimized graph embedding matrix and the optimized adjacency matrix, the dynamic brain functional connection network and the dynamic graph embedding corresponding to the dynamic brain functional connection network are determined.

[0196] Optionally, the first objective function is:

[0197] Among them, F t It is used to represent the initial graph embedding matrix or the optimized graph embedding matrix in the t-th time window; the second objective function is Among them, W t Used to represent the initial adjacency matrix or optimized adjacency matrix under the t-th time window, is the graph embedding similarity matrix, is the signal similarity matrix, and β and γ are constraint weight parameters.

[0198] Optionally, the preset iteration termination condition is: using the graph embedding matrix and adjacency matrix obtained by optimization in the current round and the previous round respectively as inputs of the loss function, and when the value of the loss function is less than a preset threshold, the optimization ends.

[0199] Optionally, the optimization module 403 is specifically configured to calculate an initial adjacency matrix based on the similarity of BOLD signals between adjacent brain nodes in the preprocessed signal data; and determine an initial graph embedding matrix based on a Laplacian matrix corresponding to the initial adjacency matrix.

[0200] This specification also provides a computer-readable storage medium, which stores a computer program. The computer program can be used to execute the above-mentioned dynamic brain function network learning and display method based on the manifold optimization algorithm.

[0201] This manual also provides Figure 5 The schematic structure diagram of the electronic device shown in FIG. Figure 5 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, memory, and non-volatile storage, and may also include other hardware required for its operation. The processor reads the corresponding computer program from the non-volatile storage into the memory and then runs it to implement the dynamic brain functional network learning and display method based on the manifold optimization algorithm.

[0202] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0203] In the 1990s, technological improvements could be clearly distinguished as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with the advancement of technology, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD through their own programming, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0204] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.

[0205] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0206] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0207] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0208] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0209] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0210] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0211] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0212] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0213] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0214] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0215] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Thus, this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0216] This specification may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0217] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0218] The foregoing is merely an example of the present invention and is not intended to limit the present invention. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A dynamic brain functional network learning and display method based on manifold optimization algorithm, characterized in that: include: Acquire a blood oxygen signal over a period of time, and preprocess the blood oxygen signal using a sliding window to obtain preprocessed signal data, wherein the preprocessed signal data includes the BOLD signal of each brain node in each time window; constructing an objective function for representing the product between the signal distance and the functional connection weight between brain nodes and for representing the distance between adjacent time windows of the dynamic graph embedding on the Grassmann manifold; Taking minimization of the objective function as the optimization goal, optimizing the adjacency matrix and graph embedding matrix corresponding to the preprocessed signal data based on a manifold optimization algorithm, determining a dynamic brain functional connectivity network and a dynamic graph embedding corresponding to the dynamic brain functional connectivity network and displaying the same, including: determining an initial adjacency matrix and an initial graph embedding matrix based on the preprocessed signal data; In a first round of iteration, a first objective function is obtained according to the initial adjacency matrix and the objective function, and the initial graph embedding matrix is ​​optimized with minimizing the first objective function as the optimization objective to obtain an optimized graph embedding matrix, the gradient of the first objective function relative to the graph embedding matrix is ​​calculated to obtain a second objective function, and the optimized graph embedding matrix is ​​brought into the second objective function, and the initial adjacency matrix is ​​optimized with minimizing the second objective function as the optimization objective to obtain an optimized adjacency matrix, and in subsequent iterations, the optimized graph embedding matrix is ​​continuously optimized by the first objective function, and the optimized adjacency matrix is ​​continuously optimized by the second objective function; When the preset iteration termination condition is met, the iteration is terminated; Determining a dynamic brain functional connectivity network and a dynamic graph embedding corresponding to the dynamic brain functional connectivity network based on the optimized graph embedding matrix and the optimized adjacency matrix; Wherein, the first objective function is: ,in, It is used to represent the initial graph embedding matrix or the optimized graph embedding matrix in the t-th time window; the second objective function is ,in, Used to represent the initial adjacency matrix or optimized adjacency matrix under the t-th time window, is the graph embedding similarity matrix, is the similarity matrix of the signal, β and γ is the constraint weight parameter.

2. The method according to claim 1, wherein Preprocessing the blood oxygen signal in a sliding window manner to obtain preprocessed signal data specifically includes: Determining, based on the blood oxygen signal and a preset brain region of interest template, an average value of the BOLD-fMRI signal of each voxel in each brain node during the period of time to obtain two-dimensional BOLD signal data; The signal data of the preset time lengths at the beginning and end of the two-dimensional BOLD signal data are removed, and the BOLD signal data in each time window is obtained by sliding the window as the preprocessed signal data.

3. The method according to claim 1, wherein The preset iteration termination condition is: using the graph embedding matrix and adjacency matrix obtained by optimization in the current round and the previous round respectively as inputs of the loss function, and when the value of the loss function is less than a preset threshold, the optimization ends.

4. The method according to claim 1, wherein Determine the initial adjacency matrix and the initial graph embedding matrix, specifically including: Calculating an initial adjacency matrix based on the similarity of BOLD signals between adjacent brain nodes in the preprocessed signal data; An initial graph embedding matrix is ​​determined according to the Laplace matrix corresponding to the initial adjacency matrix.

5. A dynamic brain function network learning and display device based on manifold optimization algorithm, characterized in that: include: A preprocessing module is used to obtain blood oxygen signals over a period of time and preprocess the blood oxygen signals through a sliding window to obtain preprocessed signal data, wherein the preprocessed signal data includes the BOLD signal of each brain node in each time window; A construction module for constructing an objective function, wherein the objective function is used to represent the product between the signal distance and the functional connection weight between brain nodes and the distance between adjacent time windows embedded in the dynamic graph on the Grassmann manifold; an optimization module, configured to optimize the adjacency matrix and graph embedding matrix corresponding to the preprocessed signal data based on a manifold optimization algorithm with minimization of the objective function as the optimization goal, determine a dynamic brain functional connectivity network and a dynamic graph embedding corresponding to the dynamic brain functional connectivity network, and display the result; The optimization module is specifically configured to: determine an initial adjacency matrix and an initial graph embedding matrix based on the preprocessed signal data; In a first round of iteration, a first objective function is obtained according to the initial adjacency matrix and the objective function, and the initial graph embedding matrix is ​​optimized with minimizing the first objective function as the optimization objective to obtain an optimized graph embedding matrix, the gradient of the first objective function relative to the graph embedding matrix is ​​calculated to obtain a second objective function, and the optimized graph embedding matrix is ​​brought into the second objective function, and the initial adjacency matrix is ​​optimized with minimizing the second objective function as the optimization objective to obtain an optimized adjacency matrix, and in subsequent iterations, the optimized graph embedding matrix is ​​continuously optimized by the first objective function, and the optimized adjacency matrix is ​​continuously optimized by the second objective function; When the preset iteration termination condition is met, the iteration is terminated; Determining a dynamic brain functional connectivity network and a dynamic graph embedding corresponding to the dynamic brain functional connectivity network based on the optimized graph embedding matrix and the optimized adjacency matrix; Wherein, the first objective function is: ,in, It is used to represent the initial graph embedding matrix or the optimized graph embedding matrix in the t-th time window; the second objective function is ,in, Used to represent the initial adjacency matrix or optimized adjacency matrix under the t-th time window, is the graph embedding similarity matrix, is the similarity matrix of the signal, β and γ is the constraint weight parameter.

6. The device according to claim 5, characterized in that The preprocessing module is specifically used to determine the average value of the BOLD-fMRI signal of each voxel in each brain node within the period of time based on the blood oxygen signal and a preset brain region of interest template to obtain two-dimensional BOLD signal data; remove the signal data of the preset time lengths at the beginning and end of the two-dimensional BOLD signal data, and obtain the BOLD signal data in each time window through a sliding window as the preprocessed signal data.

7. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 4 is implemented.

Images