A functional brain network construction method, device, medium and equipment

Abstract

The application discloses a functional brain network construction method and device, medium and equipment, and relates to the technical field of functional brain network construction. The method comprises the following steps: firstly, based on the resting-state brain functional magnetic resonance imaging of a patient with a potential brain disease, determining the time sequence characteristics of each brain region according to the brain region division of an automatic anatomical marker map, and constructing an initial functional brain network; then, performing spatial convolution on the time sequence characteristics of each brain region, fusing the information of adjacent brain regions of the initial functional brain network into the time sequence characteristics of each brain region after spatial convolution; then, performing clustering on the time sequence characteristics of each brain region after spatial convolution to obtain a clustering mapping matrix, and performing clustering mapping based on the clustering mapping matrix to obtain the time sequence characteristics of each brain region after clustering of the brain regions; and finally, constructing a functional brain network after clustering of the brain regions according to the time sequence characteristics of each brain region after clustering. The functional brain network constructed by the application can more directly represent the characteristics of the potential brain disease.

Description

Technical Field

[0001] This invention relates to the field of functional brain network construction technology, and in particular to a method, apparatus, medium and device for constructing a functional brain network. Background Technology

[0002] Currently, autism spectrum disorder is a common neurodevelopmental disorder in children, characterized by impaired social skills, language and communication difficulties, a narrow range of interests, and repetitive and stereotyped behaviors. At present, the diagnosis of autism largely relies on observation of clinical symptoms and the subjective experience of physicians, which may lead to a high rate of misdiagnosis and thus delay treatment. Therefore, how to use neuroscience to assist physicians in diagnosing autism, especially in the early stages of the disease, has become a key research focus in the field of neuroscience.

[0003] Resting-state functional magnetic resonance imaging (rs-fMRI), as a non-invasive technique, has been widely used in the early diagnosis of brain diseases. Taking autism as an example, since autism spectrum disorders often affect the neural connections in different areas of the brain, constructing high-quality brain networks has become an important means to help identify abnormalities in brain regions of autistic patients.

[0004] However, when traditional methods construct brain networks based on resting-state functional magnetic resonance imaging of patients with potential brain disorders, they are usually based on traditional brain regions. However, traditional brain region classification is too rigid, and the relationships between a large number of fine-grained brain regions are intricate and complex, which is not conducive to doctors' direct observation and judgment. Therefore, how to construct a more relevant brain network that can more directly characterize potential brain disorders is an urgent problem to be solved. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, medium, and device for constructing a functional brain network to address the aforementioned technical problems.

[0006] The present invention adopts the following technical solution:

[0007] This invention provides a method for constructing a functional brain network, comprising:

[0008] Based on resting-state functional magnetic resonance imaging of patients with potential brain diseases, the temporal characteristics of each brain region were determined according to the brain region division of the automated anatomical labeling atlas, and an initial functional brain network was constructed.

[0009] Based on the spatial relationships between nodes in the initial functional brain network, the time series features of each brain region are spatially convolved, so that the time series features of each brain region after spatial convolution are integrated with the information of its neighboring nodes.

[0010] Clustering is performed on the temporal series features of each brain region after spatial convolution to remove brain regions that are not related to potential brain diseases and to aggregate brain regions with similar temporal series features after spatial convolution to determine the clustering mapping matrix.

[0011] Clustering mapping is performed on the time series features of each brain region after spatial convolution based on the clustering mapping matrix, and the time series features of each brain region after clustering mapping are determined to construct the functional brain network after brain region clustering.

[0012] Optionally, the step of spatially convolving the time-series features of each brain region based on the spatial relationships between nodes in the initial functional brain network specifically includes:

[0013] The time-series features of each brain region and the initial functional brain network are input into a map-isomorphic network. The time-series features of each brain region are then spatially convolved based on the spatial relationships between nodes in the initial functional brain network using the following formula:

[0014]

[0015] H 0 =AX,

[0016]

[0017] Where Z represents the temporal series features of each brain region after spatial convolution, A represents the initial functional brain network, X represents the temporal series features of each brain region, and GIN_C() represents the graph isomorphic network expression. For the feature information of the input node of the k-th layer of the graph isomorphic network, N i Let be the node adjacent to the i-th node in the initial functional brain network, conv2d() be the convolution function, l be the preset maximum number of iterations for the graph isomorphic network, and ∈ k These are learnable parameters.

[0018] Optionally, the clustering of the time-series features of each brain region after spatial convolution specifically includes:

[0019] S = softmax(GIN_C p (A,Z)),

[0020]

[0021] Where S is the clustering mapping matrix, GIN_C p ( ) represents a graph isomorphic network with pooling layers as readout functions, and pool() represents the pooling layer function.

[0022] Optionally, the method further includes:

[0023] The temporal convolutional network is used to perform temporal convolution on the temporal series features of each brain region after brain region clustering using the following formula:

[0024] Tconv(Z')=relu(bn(conv2d'(Z')));

[0025] Spatiotemporal convolution is performed on the temporal series features of each brain region after temporal convolution using the following formula:

[0026] STGconv(Z')=Tconv(GIN_C(A',Z'))+bn(Tconv(Z'));

[0027] The Transformer model is used to integrate the temporal features of each brain region based on both temporal and spatiotemporal convolutions to obtain coarse-grained temporal series features of each brain region after spatiotemporal aggregation.

[0028] Q = W q (Tconv(Z' T )),

[0029] K = W k (STGconv(Z' T )),

[0030] V = W v (Tconv(Z' T )),

[0031]

[0032] Where Tconv() represents temporal convolution, conv2d'() is the convolution function with a kernel size of (1, kernel_size), kernel_size is the size of the temporal convolution kernel, Z' is the temporal series feature of each brain region after brain region clustering, bn() is the batch normalization function, relu() is the activation function, STGconv() represents spatiotemporal convolution, A' is the functional brain network after brain region clustering, Q is the query constructed by the Transformer model, and W... q ( ) represents the query constructor of the Transformer model, K is the key used to construct the Transformer model, and W k ( ) represents the key constructor of the Transformer model, V represents the value constructed by the Transformer model, and W represents the key constructor. v ( ) is the value constructor for the Transformer model, V out represents the coarse-grained time series features of each brain region after spatiotemporal aggregation, and n' represents the number of brain regions after brain region clustering.

[0033] Optionally, the resting-state functional magnetic resonance imaging of patients with potential brain disorders, determining the time-series characteristics of each brain region based on the brain region division of an automated anatomical labeling atlas, and constructing an initial functional brain network specifically includes:

[0034] Slice time correction and head movement correction were performed on the user's resting-state functional magnetic resonance imaging (fMRI) images, and the effects of central ventricular and white matter signals and higher-order head movement effects were removed from the corrected resting-state fMRI images to obtain pre-processed resting-state fMRI images.

[0035] The pre-processed resting-state functional magnetic resonance imaging (fMRI) images were registered to a standard space, and the registered resting-state fMRI images were filtered by a time bandpass filter of 0.01–0.1 Hz to reduce the influence of heartbeat and respiration, resulting in the final pre-processed resting-state fMRI images.

[0036] Based on the automated anatomical atlas, the brain is divided into multiple brain regions, and the average time-series features of each brain region are extracted from the final preprocessed resting-state functional magnetic resonance imaging (fMRI) images as the time-series features of each brain region.

[0037] This invention provides a functional brain network construction device, comprising:

[0038] The acquisition module is used to determine the time-series characteristics of each brain region based on resting-state functional magnetic resonance imaging of patients with potential brain diseases, according to the brain region division of the automatic anatomical labeling atlas, and to construct an initial functional brain network.

[0039] The spatial convolution module is used to perform spatial convolution on the time series features of each brain region based on the spatial relationship between nodes in the initial functional brain network, so that the time series features of each brain region after spatial convolution are integrated with the information of its neighboring nodes.

[0040] The clustering module is used to cluster the time-series features of each brain region after spatial convolution, in order to remove brain regions that are not related to potential brain diseases and aggregate brain regions with similar time-series features after spatial convolution, and determine the clustering mapping matrix.

[0041] The construction module is used to perform clustering mapping on the time series features of each brain region after spatial convolution based on the clustering mapping matrix, and to determine the time series features of each brain region after clustering mapping in order to construct the functional brain network after brain region clustering.

[0042] The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described brain network construction method.

[0043] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned brain network construction method.

[0044] The above-mentioned at least one technical solution adopted in this invention can achieve the following beneficial effects:

[0045] This invention first acquires a user's resting-state functional magnetic resonance imaging (fMRI) of the brain, determines the time-series features of brain regions based on traditional brain region division, and constructs an initial functional brain network based on this. Then, based on the spatial relationships between nodes in the initial functional brain network, spatial convolution is performed on the time-series features of each brain region. This allows the time-series features of each brain region after spatial convolution to incorporate information from its neighboring brain regions in the initial functional brain network. Next, the time-series features of each brain region after spatial convolution are clustered to remove brain regions unrelated to potential brain diseases and aggregate brain regions with similar time-series features after spatial convolution, thereby determining a clustering mapping matrix. Based on the clustering mapping matrix, the time-series features of each brain region after spatial convolution are clustered and mapped to determine the time-series features of each brain region after clustering mapping, thus constructing a functional brain network after brain region clustering.

[0046] Traditional brain region segmentation is highly granular, resulting in numerous fine-grained brain regions with complex relationships, hindering direct observation and judgment. Generally, not all brain regions are related to potential brain disorders. Therefore, this invention addresses this by aggregating traditional brain regions and removing those irrelevant to potential brain disorders during the aggregation and mapping process. This improves the direct correlation between the constructed functional brain network and potential brain disorders, avoiding interference from irrelevant brain regions in the representation of potential brain disorders. Furthermore, it characterizes the function of brain regions through time-series features and aggregates functionally similar brain regions. By reducing the number of brain regions, the relationships between them are clarified and centralized, thereby reducing the complexity of the constructed functional brain network. The combination of these two aspects enhances the representational accuracy of the constructed functional brain network for potential brain disorder characteristics. Attached Figure Description

[0047] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0048] Figure 1 This is a schematic diagram of a method for constructing a functional brain network provided by the present invention;

[0049] Figure 2 This is a schematic diagram illustrating the specific process of a functional brain network construction method provided by the present invention;

[0050] Figure 3 This is a schematic diagram of a functional brain network construction device provided by the present invention;

[0051] Figure 4 A schematic diagram of a computer device for implementing a method for constructing a functional brain network, provided by the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0053] Currently, the Region of Interest (ROI) obtained from resting-state functional magnetic resonance imaging (fMRI) preprocessing is represented by a Blood Oxygen Level-Dependent (BOLD) signal, which consists of a time series. When using classic Graph Neural Networks (GNNs), fully connected layers as readout functions often disrupt the sequential nature of the time series while capturing the relationships between adjacent time series, making them suitable for processing BOLD signals.

[0054] When constructing brain networks, the time series data of brain regions themselves can be used as features, denoted as nodes, and the relationships between brain regions can be used as edges. This is very similar to a graph structure, so GNNs are often used to handle spatial relationships between brain regions. The advantage of graphs is that they can mathematically model unconventional interactions between data based on nodes and edges. In classical graph theory methods, the intrinsic properties of graphs, such as local efficiency, global efficiency, average path length, and small-world property, are relied upon to analyze brain networks. These features can be used to compare brain networks from different subjects, revealing the properties of brain networks and thus providing insights into biomarkers of brain diseases. The classification results of Graph Isomorphism Networks (GINs) in the Weisfeiler-Lehman (WL) graph isomorphism test neural network demonstrate that the GIN method can provide stronger classification performance. GINs are suitable for handling data with weak contextual relationships; however, BOLD signals are time series with strong correlations and are not suitable for GINs. Therefore, this invention improves GINs by using two-dimensional convolutions as the output layer. Two-dimensional convolutions have extremely strong local information aggregation capabilities and do not destroy the original temporal characteristics of the sequence.

[0055] The Transformer model is primarily used for processing sequential data, mainly capturing the relationships between different positions in the sequence through a self-attention system. While the classic Transformer model is suitable for processing text data, it performs poorly when processing brain region data. To better suit the characteristics of time series data, this invention uses spatiotemporal convolution and temporal convolution to construct the query Q, key K, and value V of the Transformer model. The spatiotemporal convolution here is a novel method, primarily focusing on the temporal dimension, with the spatial dimension playing a supporting role.

[0056] Furthermore, the main idea of ​​this invention is hierarchical spatiotemporal multi-granularity. It uses spatial convolution to aggregate the temporal dimension, constructing coarse-grained temporal information, while simultaneously aggregating fine-grained brain regions to form coarse-grained brain regions. The relationship coefficients between these coarse-grained brain regions exhibit more significant characteristics, facilitating the identification by physicians of potential brain disorders in the brain networks of test subjects.

[0057] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0058] Figure 1 This is a schematic diagram of a method for constructing a functional brain network according to the present invention, which specifically includes the following steps:

[0059] S101: Based on resting-state functional magnetic resonance imaging of patients with potential brain diseases, the time-series characteristics of each brain region were determined according to the brain region division of the automated anatomical labeling atlas, and an initial functional brain network was constructed.

[0060] S102: Spatial convolution is performed on the time series features of each brain region based on the spatial relationship between each brain region in the initial functional brain network, so that the time series features of each brain region after spatial convolution are integrated with the information of its neighboring brain regions.

[0061] S103: Cluster the time series features of each brain region after spatial convolution to remove brain regions that are not related to potential brain diseases and aggregate brain regions with similar time series features after spatial convolution to determine the clustering mapping matrix.

[0062] S104: Based on the clustering mapping matrix, cluster the time series features of each brain region after spatial convolution, determine the time series features of each brain region after clustering mapping, and construct the functional brain network after brain region clustering.

[0063] For ease of explanation, the following description focuses solely on the server as the executing entity. The server mentioned in this invention can be a server set up on a business platform, or a device such as a desktop computer or laptop computer capable of executing the solution of this invention.

[0064] In practical applications, the server of the business platform typically first obtains the user's resting-state functional magnetic resonance imaging (fMRI) of the brain and uses it to construct the user's functional brain network. Of course, in one or more embodiments of the present invention, the server can first preprocess the user's resting-state fMRI of the brain using the DPARSF toolbox in MATLAB.

[0065] Specifically, the server can first perform slice time correction (removing the first p time points of the resting-state functional magnetic resonance imaging) and head motion correction on the user's resting-state functional magnetic resonance imaging, and remove the influence of central ventricular and white matter signals and higher-order head motion effects in the corrected resting-state functional magnetic resonance imaging, to obtain a pre-processed resting-state functional magnetic resonance imaging.

[0066] Then, the pre-processed resting-state functional magnetic resonance imaging (fMRI) images were registered to a standard space. The registered resting-state fMRI images were then subjected to a 0.01–0.1 Hz time bandpass filter to reduce the influence of heartbeat and respiration, resulting in the final pre-processed resting-state fMRI images. Registration to a standard space means that the fMRI image is typically a three-dimensional image with a time axis (length, width, height, and time). Since different brain regions acquired by different machines have different positions in three-dimensional space, specialized tools (such as FSL, SPM, AFNI, etc.) are needed to detect and correct head movements to ensure that all resting-state fMRI images are in the same coordinate system.

[0067] Finally, based on the automated anatomical atlas, the brain was divided into multiple brain regions, and the average time-series features of each brain region were extracted from the preprocessed resting-state functional magnetic resonance imaging (fMRI) images, serving as the time-series features for each brain region. The time-series features of each brain region can be represented as X = [x1, x2, ..., x...]. i ,…,x n ]∈R n×t Among them, x i ∈R n×t Let t represent the time series of the i-th brain region, t represent the length of the time series, and n represent the number of brain regions.

[0068] After obtaining the time-series features of each brain region, the server can construct an initial functional brain network using Pearson correlation. Here, the server can determine the correlation between brain regions based on the time-series features of each brain region through the Pearson correlation coefficient. Thus, each brain region is used as a node, and the edges between nodes are determined based on the correlation between brain regions. For example, the Pearson correlation coefficient between brain regions can be used as the edge between nodes. Alternatively, a preset correlation threshold can be used. Only when the Pearson correlation coefficient between brain regions is greater than the preset correlation threshold is it considered that there is an edge connection between nodes of corresponding two brain regions.

[0069] After obtaining the time-series features and initial functional brain networks of each brain region, the server can further capture spatial information between brain regions using a GIN network. Classical graph convolution operations use GNNs to capture spatial information and construct classification matrices. However, since GNNs perform poorly in processing BOLD signals, GINs, which have proven to have superior potential, are used instead. The expression for the GIN network is shown below: H 0 =AX, H 0 ∈R n×t ,

[0070] In the formula, A∈R n*n It is a brain network constructed using Pearson correlation, where X represents the time-series features of each brain region. It is the feature information of the input node of the k-th layer of the graph isomorphic network, ∈ k These are learnable parameters, MLP k It is the k-th fully connected layer. N represents the information surrounding the aggregation node. i is the node adjacent to the i-th node in the initial functional brain network.

[0071] In order to preserve the temporal characteristics of the BOLD signal and make the GIN model focus more on information at local time points, in one or more embodiments of the present invention, one-dimensional convolution is used to replace the fully connected layer operation, and the GIN_C formula constructed in this way is shown below: conv2d() is the convolution function.

[0072] Therefore, the server can use a homogeneous network of the time-series features of each brain region and the initial functional brain network input map to perform spatial convolution on the time-series features of each brain region according to the spatial relationship between nodes in the initial functional brain network, using the following formula:

[0073]

[0074] H 0 =AX,

[0075] In the formula, Z represents the time-series features of each brain region after spatial convolution, GIN_C() is the graph isomorphic network expression, and l is the preset maximum number of iterations for the graph isomorphic network.

[0076] After obtaining the temporal series features of each brain region after spatial convolution, the server can use differentiable graph pooling (diffpooling) to cluster the brain regions. In one or more embodiments of the present invention, the server can use the following formula to cluster the brain regions based on the temporal series features of each brain region after spatial convolution, thereby removing brain regions unrelated to potential brain diseases and aggregating brain regions with similar temporal series features, and determining the clustering mapping matrix: S = softmax(GIN_C p (A,Z)),

[0077] In the formula, S is the clustering mapping matrix, and GIN_C p ( ) represents a graph isomorphic network with pooling layers as readout functions, and pool() represents the pooling layer function.

[0078] Furthermore, the server can perform clustering mapping on the time-series features of each brain region after spatial convolution using the following formula, based on the clustering mapping matrix, to determine the time-series features of each brain region after clustering, and construct the functional brain network after brain region clustering based on the time-series features of each brain region after clustering: Z' = S T Z, A' = S T AS.

[0079] In the formula, Z' represents the time-series features of each brain region after brain region clustering, and A' represents the functional brain network after brain region clustering.

[0080] The process of obtaining the mapping matrix through differentiable graph pooling involves using F-regularization and information entropy loss to make the matrix as sparse as possible, aiming for each row to be a one-hot encoded form, thus ensuring the independence of aggregated brain regions—a hard aggregation method. Training constraints can be applied here using the following loss function:

[0081] Loss = αLoss en +(1-α)Loss LP ,

[0082]

[0083] In the formula, Loss LP To measure the value of S through clustering representation T The difference between the reconstructed brain network matrix S and the initial functional brain network matrix A, S T S represents the clustering similarity metric matrix. Assuming each node in the initial functional brain network corresponds to 116 brain regions, and each node in the clustered functional brain network corresponds to 64 brain regions, S is the clustering mapping matrix from 116 brain regions to 64 brain regions. TMultiplying the two results in a 116×116 matrix. This matrix must not deviate from the initial functional brain network; this is equivalent to incorporating prior conditions from the initial functional brain network to ensure the clustered functional brain network is as close as possible to the initial functional brain network. d is the normalization factor, which can be the square of the number of brain regions after clustering. Loss... en The loss avoids overly certain data point assignments by increasing uncertainty in cluster allocation, thus making the distribution more even rather than overly reliant on any single cluster. ij ε is the entropy measure of the non-negative part of matrix S. It is a very small positive number and is usually used to avoid the log(0) problem in the logarithmic function and to ensure numerical stability.

[0084] based on Figure 1 The method for constructing a functional brain network, as shown, first acquires the user's resting-state functional magnetic resonance imaging (fMRI) of the brain. Based on traditional brain region division, the time-series features of brain regions are determined, and an initial functional brain network is constructed using these features. Then, based on the spatial relationships between nodes in the initial functional brain network, the time-series features of each brain region are spatially convolved. This allows the time-series features of each brain region after spatial convolution to incorporate information from its neighboring brain regions in the initial functional brain network. Next, the time-series features of each brain region after spatial convolution are clustered to remove brain regions unrelated to potential brain diseases and aggregate brain regions with similar time-series features after spatial convolution. This determines the clustering mapping matrix. Based on the clustering mapping matrix, the time-series features of each brain region after spatial convolution are clustered and mapped to determine the time-series features of each brain region after clustering mapping, thus constructing a functional brain network after brain region clustering.

[0085] Traditional brain region segmentation is highly granular, resulting in numerous fine-grained brain regions with complex relationships, hindering direct observation and judgment. Generally, not all brain regions are related to potential brain disorders. Therefore, this invention addresses this by aggregating traditional brain regions and removing those irrelevant to potential brain disorders during the aggregation and mapping process. This improves the direct correlation between the constructed functional brain network and potential brain disorders, avoiding interference from irrelevant brain regions in the representation of potential brain disorders. Furthermore, it characterizes the function of brain regions through time-series features and aggregates functionally similar brain regions. By reducing the number of brain regions, the relationships between them are clarified and centralized, thereby reducing the complexity of the constructed functional brain network. The combination of these two aspects enhances the representational accuracy of the constructed functional brain network for potential brain disorder characteristics.

[0086] This invention uses spatial convolution to aggregate different brain regions at a fine granular level to form coarse-grained brain regions. The relationship coefficients between coarse-grained brain regions are more characteristic, making it easier for doctors to identify differences in functional brain networks among different subjects.

[0087] This invention aims to construct a good functional brain network based on complementary information from multi-perspective data, thereby finding biomarkers corresponding to neurological diseases. These biomarkers can assist doctors in identifying neurological diseases.

[0088] When applying the functional brain network construction method provided by this invention, it is not necessary to follow the... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this invention does not impose any restrictions on it.

[0089] Furthermore, in one or more embodiments of the present invention, the time-series features of each brain region after aggregation can be further processed in the time dimension, that is, further coarsened in the time dimension to concentrate key information more directly, so as to obtain a brain network that can more directly characterize potential brain diseases.

[0090] For learning about time, a Transformer assisted by temporal convolution and spatial convolution can be used.

[0091] Temporal convolution is typically used to learn information from local time points and aggregate fine-grained time points into coarse-grained time points. For example, aggregating 60 time points in seconds into one time point in minutes inevitably contains information from the 60 seconds. The problem to be solved here is how to integrate the information from those 60 seconds. Whether using averaging or max pooling, a significant amount of important information is inevitably lost, and the temporal characteristics are compromised. This paper uses an improved temporal convolution for aggregation, allowing the model to autonomously determine which information is useful and needs to be retained, and which can be discarded.

[0092] The server can use the following formula to perform temporal convolution on the time series features of each brain region after brain region clustering: Tconv(Z')=relu(bn(conv2d'(Z'))).

[0093] In the formula, Tconv() represents temporal convolution, conv2d'() is the convolution function with a kernel size of (1, kernel_size), where kernel_size is the size of the temporal convolution kernel, representing the number of aggregated time points, Z' is the temporal series feature of each brain region after brain region clustering, bn() is the batch normalization function, and ReLU() is the activation function. By limiting the kernel size of the two-dimensional convolution to (1, kernel_size), it can be ensured that the convolution operation is performed only in the temporal dimension, thereby capturing information in the temporal dimension, reconstructing coarse-grained temporal information, and without losing the inherent logic of the temporal information.

[0094] Spatiotemporal convolution is performed on the temporal series features of each brain region after temporal convolution using the following formula:

[0095] STGconv(Z')=Tconv(GIN_C(A',Z'))+bn(Tconv(Z')).

[0096] In the formula, STGconv() represents spatiotemporal convolution. Spatiotemporal convolution performs a gin operation before temporal convolution, using spatial information as a priori to assist the convolution in capturing temporal information.

[0097] The Transformer model is used to integrate the temporal features of individual brain regions based on both temporal and spatiotemporal convolutions to obtain coarse-grained temporal series features of each brain region after spatiotemporal aggregation, using the following formula: Q = W q (Tconv(Z' T K = W k (STGconv(Z' T V = W v (Tconv(Z' T )),

[0098] In the formula, Q is the query constructed by the Transformer model, and W... q ( ) represents the query constructor of the Transformer model, K is the key used to construct the Transformer model, and W k ( ) represents the key constructor of the Transformer model, V represents the value constructed by the Transformer model, and W represents the key constructor. v ( ) is the value constructor for the Transformer model, V out represents the coarse-grained time series features of each brain region after spatiotemporal aggregation, and n' represents the number of brain regions after clustering.

[0099] Here, a linear transformation can be used to construct the query Q, key K, and value V. Z' is (brain region, time), and the transpose of Z' is... T It is (time, brain region). By constructing the attention level at each time point through dot product attention, and then optimizing the global information of the data by constructing the relationship matrix of time points × time points.

[0100] Then V out The input can be fed into the readout function, which can be an average pooling function, a fully connected layer mapping function, or an orthogonal readout function. The readout function in this example is shown below:

[0101] In the formula, A r This represents a coarse-grained brain network after spatiotemporal aggregation, with Pearson() being the Pearson correlation function.

[0102] That is, Pearson correlation is reused here to construct a coarse-grained brain network. To reduce data redundancy, only W is retained here. r The upper half of the matrix is ​​stretched into a vector and used as the output of the readout function.

[0103] Furthermore, this invention also provides an embodiment of a method for constructing executive function brain networks, the specific process of which is as follows: Figure 2 As shown:

[0104] (1) In this example, the Autism Brain Imaging Data Exchange (ABIDE) dataset from the New York site was used. This dataset contains 184 subjects, including 105 normal individuals and 79 autistic patients. Functional brain images acquired by the magnetic resonance imaging (MRI) device were format-converted and read. To ensure the stability of the time series, the first 10 time points of the MRI images were removed, and head movement and time-level corrections were applied to the images. The corrected images were registered to standard space, and a 0.01–0.1 Hz time bandpass filter was applied to the subjects' images to reduce the influence of heart rate and respiration.

[0105] (2) A standardized brain region template was selected. In this example, the human brain was divided into 116 regions (90 regions of the cerebrum and 16 regions of the cerebellum) according to the AAL template. These 116 regions represent 116 nodes of the brain's functional connectivity network. In reality, not all brain regions are functional. In this example, through model learning, brain regions unrelated to autism were eliminated, and functionally similar brain regions were aggregated, ultimately resulting in 64 brain regions.

[0106] (3) Calculate the time series average of all voxels in each brain region in the magnetic resonance image, and extract the time series features corresponding to different standard brain regions. The extracted data can be represented as X = [x1, x2, ..., x...]. i ,…,x n ]∈R 116×175 , where x i ∈R t Let represent the time series of the i-th brain region, t represent the number of times in the time series, and n represent the number of brain regions. An initial brain network was constructed based on 116 brain regions using Pearson correlation.

[0107] (4) Based on the above ideas, a model is constructed, consisting of a hierarchical spatial learning module, a hierarchical temporal learning module, and a readout function.

[0108] (5) A hierarchical spatial learning module, as shown below:

[0109]

[0110] H 0 =AX,

[0111] S = softmax(GIN_C p (A,Z)),

[0112] Z' = S T Z, A' = S T AS.

[0113] Among them, H 0 ∈R 116*175 The input bold signal is captured by GIN_C(), which captures the spatial relationships between brain regions to facilitate subsequent aggregation of functionally similar brain regions. S is the clustering mapping matrix, whose main purpose is to aggregate fine-grained brain regions into coarse-grained brain regions. GIN_C p ( ) The last layer is a pooling layer, which pools R 116*175 Pooling to R 116*64 Normalization is performed using softmax to obtain S∈R 116*64 .

[0114] (6) Next, we will construct a tiered time learning module:

[0115] Q = W q (relu(bn(conv2d'(Z' T ))))∈R 88*64 ,

[0116] K = W k (Tconv(GIN_C(Z' T ))+bn(Tconv(Z' T )))∈R 88*64 ,

[0117] V = W v (relu(bn(conv2d'(Z' T ))))∈R 88*64 ,

[0118]

[0119] For time information, two-dimensional convolution is mainly used for operation, and the convolution kernel can be controlled to be (9,1) to ensure that the length of the time dimension output is 88.

[0120] (7) In step (6), coarse-grained brain region data is obtained. In this embodiment, Pearson correlation reconstructed brain network A is used. r ∈R 64*64 .

[0121] The above describes a method for constructing a functional brain network according to one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding device for constructing a functional brain network, such as... Figure 3 As shown.

[0122] Figure 3 A schematic diagram of a functional brain network construction device provided by the present invention includes:

[0123] The acquisition module 201 is used to determine the time-series characteristics of each brain region based on resting-state functional magnetic resonance imaging of patients with potential brain diseases, according to the brain region division of the automatic anatomical labeling atlas, and to construct an initial functional brain network.

[0124] Spatial convolution module 202 is used to perform spatial convolution on the time series features of each brain region according to the spatial relationship between each brain region in the initial functional brain network, so that the time series features of each brain region after spatial convolution are integrated with the information of its neighboring brain regions.

[0125] Clustering module 203 is used to cluster the time series features of each brain region after spatial convolution, so as to remove brain regions that are not related to potential brain diseases and aggregate brain regions with similar time series features after spatial convolution, and determine the clustering mapping matrix.

[0126] The construction module 204 is used to perform clustering mapping on the time series features of each brain region after spatial convolution based on the clustering mapping matrix, and to determine the time series features of each brain region after clustering mapping in order to construct a functional brain network after brain region clustering.

[0127] Specific limitations regarding the functional brain network construction device can be found in the limitations of the functional brain network construction method described above, and will not be repeated here. Each module in the aforementioned functional brain network construction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0128] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The provided method for constructing functional brain networks.

[0129] The present invention also provides Figure 4 The schematic diagram of the computer device shown is as follows: Figure 4 As shown, at the hardware level, this computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above. Figure 1 The provided method for constructing functional brain networks.

[0130] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0131] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the invention.

Claims

1. A method for constructing a functional brain network, characterized in that, include: Based on resting-state functional magnetic resonance imaging of patients with potential brain diseases, the temporal characteristics of each brain region were determined according to the brain region division of the automated anatomical labeling atlas, and an initial functional brain network was constructed. The temporal series features of each brain region and the initial functional brain network are input into a graph isomorphic network. The temporal series features of each brain region are spatially convolved according to the spatial relationships between nodes in the initial functional brain network using the following formula, so that the temporal series features of each brain region after spatial convolution are integrated with the information of its neighboring brain regions: ， ； ； The following formula is used to cluster brain regions based on their temporal series features after spatial convolution, employing differentiable graph pooling to remove brain regions unrelated to potential brain disorders and aggregate brain regions with similar temporal series features after spatial convolution, thus determining the clustering mapping matrix: ， ； Based on the clustering mapping matrix, the time series features of each brain region after spatial convolution are clustered and mapped to determine the time series features of each brain region after clustering and mapping in order to construct a functional brain network after brain region clustering. in, These are the time-series features of each brain region after spatial convolution. For the initial functional brain network, The time-series features of each brain region, For graph isomorphic network expressions, For graph isomorphic networks k Feature information of layer input nodes, Let be the nodes adjacent to the i-th node in the initial functional brain network. It is a convolution function. l The preset maximum number of iterations for graph isomorphic networks. These are learnable parameters. This is the clustering mapping matrix. To read out graph isomorphic networks with pooling layers as the readout function, This is the pooling layer function.

2. The method for constructing a functional brain network as described in claim 1, characterized in that, The method further includes: The temporal convolutional network is used to perform temporal convolution on the temporal series features of each brain region after brain region clustering using the following formula: ； Spatiotemporal convolution is performed on the temporal series features of each brain region after temporal convolution using the following formula: ； The Transformer model is used to integrate the temporal features of each brain region based on both temporal and spatiotemporal convolutions to obtain coarse-grained temporal series features of each brain region after spatiotemporal aggregation. ， ， ， ； in, Represents temporal convolution. It is a convolution function with a kernel size of (1, kernal_ size ), kernel_size The size of the convolution kernel in the time dimension. The time-series features of each brain region after brain region clustering. For batch normalization function, For activation function, Represents spatiotemporal convolution. This represents the functional brain network after brain region clustering. Queries constructed for the Transformer model This is the query constructor for the Transformer model. Keys constructed for the Transformer model For the key constructor of the Transformer model, The values ​​constructed for the Transformer model. Construct the value constructor for the Transformer model. This represents the coarse-grained temporal series features of each brain region after spatiotemporal aggregation. This represents the number of brain regions after clustering.

3. The method for constructing a functional brain network as described in claim 1, characterized in that, The resting-state functional magnetic resonance imaging (fMRI) of patients with potential brain disorders determines the time-series characteristics of each brain region based on the brain region division of an automated anatomical marker atlas, specifically including: Slice time correction and head movement correction were performed on the user's resting-state functional magnetic resonance imaging (fMRI) images, and the effects of central ventricular and white matter signals and higher-order head movement effects were removed from the corrected resting-state fMRI images to obtain pre-processed resting-state fMRI images. The pre-processed resting-state functional magnetic resonance imaging (fMRI) images were registered to a standard space, and the registered resting-state fMRI images were subjected to a 0.01~0.1 Hz time bandpass filter to reduce the influence of heartbeat and respiration, resulting in the final pre-processed resting-state fMRI images. Based on the automated anatomical atlas, the brain is divided into multiple brain regions, and the average time-series features of each brain region are extracted from the final preprocessed resting-state functional magnetic resonance imaging (fMRI) images as the time-series features of each brain region.

4. A functional brain network construction device, characterized in that, include: The acquisition module is used to determine the time-series characteristics of each brain region based on resting-state functional magnetic resonance imaging of patients with potential brain diseases, according to the brain region division of the automatic anatomical labeling atlas, and to construct an initial functional brain network. The spatial convolution module is used to integrate the temporal series features of each brain region with the initial functional brain network input map into a homogeneous network. It performs spatial convolution on the temporal series features of each brain region based on the spatial relationships between nodes in the initial functional brain network, thereby fusing information from neighboring brain regions into the spatially convolved temporal series features of each brain region. , ; ; The clustering module is used to cluster brain regions based on their temporal series features after spatial convolution using differentiable graph pooling, thereby removing brain regions unrelated to potential brain disorders and aggregating brain regions with similar temporal series features after spatial convolution, and determining the clustering mapping matrix: , ; The construction module is used to perform clustering mapping on the time series features of each brain region after spatial convolution based on the clustering mapping matrix, and to determine the time series features of each brain region after clustering mapping in order to construct the functional brain network after brain region clustering. in, These are the time-series features of each brain region after spatial convolution. For the initial functional brain network, The time-series features of each brain region, For graph isomorphic network expressions, For graph isomorphic networks k Feature information of layer input nodes, Let be the nodes adjacent to the i-th node in the initial functional brain network. It is a convolution function. l The preset maximum number of iterations for graph isomorphic networks. These are learnable parameters. This is the clustering mapping matrix. To read out graph isomorphic networks with pooling layers as the readout function, This is the pooling layer function.

5. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method as described in any one of claims 1 to 3.

6. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1 to 3.

Images