A method for constructing brain connectivity networks based on homogeneity-constrained multi-set canonical correlation analysis

Abstract

The present invention discloses a method for constructing a brain connectivity network based on homogeneous constrained multi-set canonical correlation analysis, comprising: 1. pre-processing collected functional magnetic resonance imaging (fMRI) images to extract time-series signals from all brain voxels, and organizing them by brain region to construct a data set; 2. calculating weight vectors corresponding to voxels within the brain region using a homogeneous constrained multi-set canonical correlation analysis algorithm; 3. mapping the weight vectors corresponding to each brain region to the individual level, calculating the representational signal of each brain region, and using the representational signal to calculate the pairwise correlation between brain regions as the connection strength; 4. constructing a brain covariant and inverse connectivity network based on the sign of the connection strength. The present invention utilizes fMRI data to effectively extract representational signals from each brain region, solving the problem of constructing brain covariant and inverse connectivity networks, contributing to the study of the functional partitioning and connectivity patterns of the brain and deepening people's understanding of the differentiation and integration of brain functions.

Description

Technical Field

[0001] The present invention relates to the field of medical signal processing, and in particular to a method for constructing a brain connection network. Background Art

[0002] Brain connectivity networks reveal the interactions between various brain regions, deepening our understanding of the brain and facilitating research and understanding of brain function. Previous studies have shown that the characteristics of brain connectivity networks can aid in the study of neurological diseases such as depression, Alzheimer's disease, and Parkinson's disease. Brain connectivity networks can be constructed using modern neuroimaging techniques such as magnetoencephalography (MEG), electroencephalography (EEG), and functional magnetic resonance imaging (fMRI). fMRI is widely used due to its high spatial resolution and non-invasive nature. Brain connectivity networks are typically constructed at the voxel level or at the region of interest (ROI) level. The former uses each voxel as a node in the network, resulting in high network dimensionality, computational complexity, and susceptibility to noise, which can affect subsequent analysis. The latter uses each ROI as a node, enabling more efficient brain connectivity network construction and garnering increasing attention.

[0003] In recent years, various methods for constructing brain connectivity networks at the ROI level based on fMRI have been proposed. The most common methods include correlation methods, clustering algorithms, and blind source separation algorithms such as independent component analysis (ICA), non-negative matrix factorization (NMF), and canonical correlation analysis (CCA).

[0004] However, most current methods still face the following problems. First, the brain connectivity network constructed at the ROI level needs to reasonably represent the activity of the brain region. The most commonly used method currently is to select a time series of a representative voxel or use the average time series of all voxels to represent the activity of the entire brain region. These practices ignore the differences between different voxels and can cause large deviations. Using the weighted average time series of all voxels, such as principal component analysis (PCA), to represent brain region activity has poor reproducibility. Moreover, when representing brain region activity, it is necessary to ensure that the selected voxels have strong homogeneity. In addition, when estimating brain connectivity networks, many methods only consider the covariation relationship between brain regions and ignore the inverse variation relationship. This leads to the study of positively correlated connectivity networks, without paying attention to negatively correlated connectivity networks, and cannot fully explore brain function and the interactions between brain regions. Summary of the Invention

[0005] The present invention aims to address the deficiencies of the above-mentioned prior art and proposes a method for constructing a brain connection network based on homogeneous constrained multi-set canonical correlation analysis, in order to more accurately characterize brain area activities for constructing connection networks, thereby enabling in-depth exploration of various connection patterns of the brain and providing a new method for studying brain functions.

[0006] In order to achieve the above-mentioned object, the present invention adopts the following technical solutions:

[0007] The method for constructing a brain connectivity network based on homogeneity-constrained multi-set canonical correlation analysis of the present invention is characterized by the following steps:

[0008] Step 1: Use the signal acquisition device to obtain the functional magnetic resonance imaging signals of S individuals and align them to the standard template, and then divide them according to the given M brain regions to obtain the total sample data set {X k |k=1,2,...,M}, represents the kth brain region sample dataset after S individuals are registered, T is the sample length, P k is the number of voxels in the kth brain region, and X k =[X k ' (1) ,X k ' (2) ,...,X k ' (i) ,...,X k ' (S) ]′, [·]′ represents the transpose of the matrix, represents the signal in the kth brain region of the i-th individual, X k ' (i) express The transpose of

[0009] Step 2: Use the homogeneity constraint multi-set canonical correlation analysis algorithm to analyze the total sample data set {X k |k=1,2,...,M} is calculated to obtain the weight vector {w k |k=1,2,...,M},where w k Represents the weight vector of the k-th brain region sample data set;

[0010] Step 3: Calculate the representation signal of the kth brain region in the i-th individual And the representation signal in the kth brain region of the i-th individual and the representation signal of the jth brain area The correlation coefficient between them is used as the connection strength between the kth brain and the jth brain region in the i-th individual, thereby obtaining the connection strength between the kth brain and the j-th brain region in S individuals; represents the signal in the jth brain region of the i-th individual, w j Represents the j-th brain region sample dataset X j The weight vector of

[0011] Step 4: Perform a sign test on the S connection strengths between the kth brain region and the jth brain region. If the test result is positive, it means that the connection relationship between the kth brain region and the jth brain region is a covariant connection; if the test result is negative, it means that the connection relationship between the kth brain region and the jth brain region is an inverse connection; if the test result is not significant, it means that the connection relationship between the kth brain region and the jth brain region is not significant;

[0012] Step 5: A brain covariant connection network is formed by the brain regions whose connection relationship among each brain region is covariant connection; a brain inverse connection network is formed by the brain regions whose connection relationship among each brain region is inverse connection.

[0013] The method for constructing a brain covariant-inverse connectivity network based on homogeneity-constrained multi-set canonical correlation analysis of the present invention is also characterized in that the homogeneity-constrained multi-set canonical correlation analysis algorithm is performed according to the following steps:

[0014] Step 2.1: Use formula (1) to construct an optimization model with the goal of maximizing the sum of squares of correlation coefficients between the M brain region representation signals under the constraint of sign consistency of the weight vector:

[0015]

[0016] In formula (1), w k ≥0 or w k≤0 represents the weight vector w of the k-th brain region sample dataset k P k The elements are all non-negative or non-positive, w′ k represents w k The transpose of

[0017] Step 2.2: Use the interior point method and Newton method to iteratively solve the optimization model:

[0018] Step 2.2.1: Use equation (2) to construct the loss function L(w,s,λ):

[0019]

[0020] In formula (2), w represents the weight matrix, and w = [w′1,...,w′ k ,...,w′ M ]′,s=[s′1,...,s′ m ,...,s′ M+1 ]′ is the relaxation matrix, s m represents the mth slack variable, s′ m Indicates s m The transpose of λ=[λ1,...,λ m ,...,λ M+1 ]′ is the Lagrange multiplier matrix, λ m represents the mth Lagrange multiplier, μ is a positive barrier parameter, f(w) = -trace(ZZ′) is the inverse of the sum of squares of the correlation coefficients between the two brain regions of M brain regions, trace represents the trace of the matrix, Z = [w′ k X′ k X j w j ] M×M Represents the correlation matrix of brain area activity, g=[g1,...,g m ,...,g M+1 ]' is the constraint function matrix, g m represents the mth constraint function, and g m (w)=w'A m w-1+s m , X m represents the sample dataset of the mth brain region, X′ m Represents X m The transpose of diag represents the diagonal matrix, and the M+1 constraint function g M+1 (w)=-w+s M+1 , s M+1 represents the M+1th slack variable;

[0021] Step 2.2.2: Define the number of outer iterations as n, define the number of inner iterations as t, give the parameter σ∈(0,1), and initialize n=1, and set the M weight vectors under the nth outer iteration Initialized to M unit vectors, where Represents the kth weight vector under the nth outer iteration and is initialized to the kth unit vector. The relaxation matrix s under the nth outer iteration is (n) Initialize it to a matrix of all 1s and convert the Lagrange multiplier matrix λ under the nth outer iteration (n) Initialize it to a matrix of all 1s and set the barrier parameter μ under the nth outer iteration (n) Initialized to 1;

[0022] Step 2.2.3: Initialize t=1;

[0023] Step 2.2.4: Substitute the M weight vectors of the t-th inner iteration under the n-th outer iteration Initialized as a weight vector in, represents the kth weight vector of the tth inner iteration under the nth outer iteration and is initialized to The relaxation vector s of the t-th inner iteration under the n-th outer iteration (n,t) Initialized to s (n) , the Lagrange multiplier matrix λ of the t-th inner iteration under the n-th outer iteration (n,t) Initialized to λ (n) ;

[0024] Step 2.2.5: Calculate the triplet of changes (Δw (n ,t) ,Δs (n,t) ,Δλ (n,t) ):

[0025]

[0026] In formula (3), Δw (n,t) represents the tth inner iteration w under the nth outer iteration (n,t) The change in Δs (n,t) Indicates the tth inner iteration s under the nth outer iteration (n,t) The change in Δλ (n,t) represents the λ of the tth inner iteration under the nth outer iteration (n,t) Change, H (n,t) represents the Hessian matrix of the loss function of the t-th inner iteration under the n-th outer iteration, and f(·) represents the inverse of the sum of squares of the correlation coefficients, represents the Lagrange multiplier matrix λ of the tth inner iteration under the nth outer iteration(n,t) The mth Lagrange multiplier in , represents the gradient operator, express The constraint function matrix g(w (n,t) )’s Jacobian matrix, J′ g (n,t) express The transpose of S (n,t) It is made by (n,t) The diagonal matrix composed of the diagonal elements in , Λ (n,t) is given by λ (n,t) The diagonal matrix consists of the diagonal elements in , I represents the identity matrix, and e represents the all-1 vector;

[0027] Step 2.2.6: Use formula (4) to obtain the triplet (w (n,t+1) ,s (n ,t+1) ,λ (n,t+1) ):

[0028] (w (n,t+1) ,s (n,t+1) ,λ (n,t+1) )=(w (n,t) +Δw (n,t) ,s (n,t) +Δs (n,t) ,λ (n,t) +Δλ (n,t) ) (4)

[0029] In formula (4), w (n,t+1) represents the weight vector of the t+1th inner iteration under the nth outer iteration, s (n,t+1) represents the relaxation vector of the t-th inner iteration under the n-th outer iteration, λ (n,t+1) represents the Lagrange multiplier matrix of the tth inner iteration under the nth outer iteration;

[0030] Step 2.2.7: After assigning t+1 to t, return to step 2.2.5 and execute sequentially until |Δw (n,t) |+|Δs (n,t) |+|Δλ (n,t) |<δ, the inner iteration ends; where δ represents the threshold;

[0031] Step 2.2.8: Use formula (5) to obtain the triplet (w (n+1) ,s (n+1) ,λ (n+1) ):

[0032] (w (n+1) ,s (n+1) ,λ (n+1) )=(w(n,t) ,s (n,t) ,λ (n,t) ) (5)

[0033] Step 2.2.9: Calculate the barrier parameter μ for the n+1th outer iteration (n+1) =σμ (n) ;

[0034] Step 2.2.10: After assigning n+1 to n, return to step 2.2.3 and execute sequentially until μ (n) <δ, the outer iteration ends;

[0035] Step 2.2.11: Set the kth weight vector under the n+1th outer iteration As the weight vector w of the k-th brain region sample data set k , thus obtaining the weight vector {w k |k=1,2,...,M}.

[0036] The electronic device of the present invention includes a memory and a processor, and is characterized in that the memory is used to store a program that supports the processor to execute the brain connection network construction method, and the processor is configured to execute the program stored in the memory.

[0037] The present invention provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, which is characterized in that when the computer program is run by a processor, the steps of the brain connection network construction method are executed.

[0038] Compared with existing brain connection network construction methods, the advantages of this invention are:

[0039] 1. In step 2 of this invention, we extend the traditional multi-set CCA algorithm by integrating domain knowledge and propose a homogeneity-constrained multi-set canonical correlation analysis algorithm. This algorithm ensures the functional homogeneity of each voxel within the same brain region by constraining the sign of the weight vector. Furthermore, based on the homogeneity-constrained weight vector, we perform a weighted average of the temporal activity of the voxels, thereby more accurately characterizing the functional activity of the brain region.

[0040] 2. The present invention maximizes the sum of squares of correlation coefficients of brain area activities, extracts positively correlated and negatively correlated connection components of each brain area, and divides the connections into covariant and inverse categories based on the sign test, thereby constructing brain covariant and inverse networks, which helps to comprehensively explore various connection patterns of the brain and provides a new method for modeling brain connection networks.

[0041] 3. In steps 3 and 4 of the present invention, based on the obtained group-level brain network, the weight vector is mapped to the individual data to obtain the brain connection network at the individual level, which is helpful for group analysis and statistical testing and is of great significance for the study of brain function and brain-related diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Flow chart of the method of the present invention;

[0043] Figure 2 A brain connection network diagram obtained by the present invention using the first set of data;

[0044] Figure 3 This is a brain connection network diagram obtained by the present invention using the second set of data;

[0045] Figure 4 A comparison of the reproducibility of brain network construction using different methods. DETAILED DESCRIPTION

[0046] In this embodiment, a method for constructing a brain connection network based on homogeneous constrained multiset canonical correlation analysis (homogeneous multiset CCA, HMCCA) is to preprocess the collected functional magnetic resonance images, extract the time series signals of all voxels in the brain, and organize them according to brain regions to construct a data set; then, the weight vector corresponding to each brain region is calculated through the homogeneous constrained multiset canonical correlation analysis algorithm proposed by the present invention; then, the weight vector corresponding to each brain region is mapped to the individual level, the representation signal of each brain region is calculated, and the correlation between the two brain regions is calculated as the connection strength using the representation signal; finally, the brain covariance and inversion connection network is constructed according to the sign of the connection strength. Specifically, if Figure 1 As shown, proceed as follows:

[0047] Step 1: The experiment used a signal acquisition device to obtain two sets of repeated measurement functional magnetic resonance imaging signals of S = 100 individuals, and aligned them to the standard template respectively. According to the given M = 7 brain regions: nucleus accumbens (ACC), amygdala (AMY), caudate nucleus (CAU), hippocampus (HIP), globus pallidus (PAL), putamen (PUT) and thalamus (THA), the total sample data set {X k |k=1,2,...,M}, represents the kth brain region sample dataset after S individuals are registered, T = 1200 is the sample length, P k is the number of voxels in the kth brain region; in this embodiment, P1 = 275, P2 = 647, P3 = 1483, P4 = 1559, P5 = 557, P6 = 2070, P7 = 2536, and [·]′ represents the transpose of the matrix, represents the signal in the kth brain region of the i-th individual, X k ' (i) express The transpose of

[0048] Step 2: Use the homogeneity constraint multi-set canonical correlation analysis algorithm to analyze the first set of total sample data sets {X k |k=1,2,...,M} is calculated to obtain the weight vector {w k |k=1,2,...,M},where w k Represents the weight vector of the k-th brain region sample data set:

[0049] Step 2.1: Use formula (1) to construct an optimization model with the goal of maximizing the sum of squares of correlation coefficients between the M brain region representation signals under the constraint of sign consistency of the weight vector:

[0050]

[0051] In formula (1), w k ≥0 or w k ≤0 represents the weight vector w of the k-th brain region sample dataset k P k The elements are all non-negative or non-positive, w′ k represents w k The transpose of

[0052] Step 2.2: Use the interior point method and Newton method to iteratively solve the optimization model:

[0053] Step 2.2.1: Use equation (2) to construct the loss function L(w,s,λ):

[0054]

[0055] In formula (2), w represents the weight matrix, and w = [w1′, ..., w′ k ,...,w′ M ]′,s=[s1′,...,s′ m ,...,s′ M+1 ]′ is the relaxation matrix, s m represents the mth slack variable, s′ m Indicates s m The transpose of λ=[λ1,...,λ m ,...,λ M+1 ]′ is the Lagrange multiplier matrix, λ mrepresents the mth Lagrange multiplier, μ is a positive barrier parameter, f(w) = -trace(ZZ′) is the inverse of the sum of squares of the correlation coefficients between the two brain regions of M brain regions, trace represents the trace of the matrix, Z = [w′ k X k 'X j w j ] M×M Represents the correlation matrix of brain area activity, g=[g1,...,g m ,...,g M+1 ]' is the constraint function matrix, g m represents the mth constraint function, and g m (w)=w'A m w-1+s m , X m represents the sample dataset of the mth brain region, X′ m Represents X m The transpose of diag represents the diagonal matrix, and the M+1 constraint function g M+1 (w)=-w+s M+1 , s M+1 represents the M+1th slack variable;

[0056] Step 2.2.2: Define the number of outer iterations as n, define the number of inner iterations as t, give the parameter σ∈(0,1), in the specific implementation, σ=0.1, and initialize n=1, and set the M weight vectors under the nth outer iteration Initialized to M unit vectors, where Represents the kth unit vector under the nth outer iteration and is initialized to a unit vector. The relaxation matrix s under the nth outer iteration is (n) Initialize it to a matrix of all 1s and convert the Lagrange multiplier matrix λ under the nth outer iteration (n) Initialize it to a matrix of all 1s and set the barrier parameter μ under the nth outer iteration (n) Initialized to 1;

[0057] Step 2.2.3: Initialize t=1;

[0058] Step 2.2.4: Substitute the M weight vectors of the t-th inner iteration under the n-th outer iteration Initialized as a weight vector in, represents the kth weight vector of the tth inner iteration under the nth outer iteration and is initialized to The relaxation vector s of the t-th inner iteration under the n-th outer iteration (n,t) Initialized to the relaxation vector s (n) , the Lagrange multiplier matrix λ of the t-th inner iteration under the n-th outer iteration(n,t) Initialized to λ (n) ;

[0059] Step 2.2.5: Calculate (Δw (n,t) ,Δs (n,t) ,Δλ (n,t) ):

[0060]

[0061] In formula (3), Δw (n,t) represents the tth inner iteration w under the nth outer iteration (n,t) The change in Δs (n,t) Indicates the tth inner iteration s under the nth outer iteration (n,t) The change in Δλ (n,t) represents the λ of the tth inner iteration under the nth outer iteration (n,t) Change, H (n,t) Represents the Hessian matrix of the loss function, that is f(·) represents the inverse of the sum of squares of the correlation coefficients, represents the Lagrange multiplier matrix λ of the tth inner iteration under the nth outer iteration (n,t) The mth Lagrange multiplier in , represents the gradient operator, express The constraint function matrix g(w (n,t) )’s Jacobian matrix, J′ g (n,t) express The transpose of S (n,t) It is made by (n,t) The diagonal matrix composed of the diagonal elements in , Λ (n,t) is given by λ (n,t) The diagonal matrix consists of the diagonal elements in , I represents the identity matrix, and e represents the all-1 vector;

[0062] Step 2.2.6: Use formula (4) to obtain the triplet (w (n,t+1) ,s (n ,t+1) ,λ (n,t+1) ):

[0063] (w (n,t+1) ,s (n,t+1) ,λ (n,t+1) )=(w (n,t) +Δw (n,t) ,s (n,t) +Δs (n,t) ,λ (n,t) +Δλ (n,t) ) (4)

[0064] In formula (4), w (n,t+1) represents the weight vector of the t+1th inner iteration under the nth outer iteration, s (n,t+1) represents the relaxation vector of the t-th inner iteration under the n-th outer iteration, λ (n,t+1) represents the Lagrange multiplier matrix of the t-th inner iteration under the n-th outer iteration,

[0065] Step 2.2.7: After assigning t+1 to t, return to step 2.2.5 and execute sequentially until |Δw (n,t) |+|Δs (n,t) |+|Δλ (n,t) |<δ, the inner iteration ends; where δ represents the threshold; in this embodiment, δ=10 -16 ;

[0066] Step 2.2.8: Use formula (5) to obtain the triplet (w (n+1) ,s (n+1) ,λ (n+1) ):

[0067] (w (n+1) ,s (n+1) ,λ (n+1) )=(w (n,t) ,s (n,t) ,λ (n,t) ) (5)

[0068] Step 2.2.9: Calculate the barrier parameter μ for the n+1th outer iteration (n+1) =σμ (n) ;

[0069] Step 2.2.10: After assigning n+1 to n, return to step 2.2.3 and execute sequentially until μ (n) <δ, the outer iteration ends;

[0070] Step 2.2.11: Set the kth weight vector under the n+1th outer iteration As the weight vector w of the k-th brain region sample data set k , thus obtaining the weight vector {w k |k=1,2,...,M}.

[0071] Step 3: Calculate the representation signal of the kth brain region in the i-th individual And the representation signal in the kth brain region of the i-th individual and the representation signal of the jth brain area The correlation coefficient between them is used as the connection strength between the kth brain and the jth brain region in the i-th individual, thereby obtaining the connection strength between the kth brain and the j-th brain region in S individuals; represents the signal in the jth brain region of the i-th individual, w j Represents the j-th brain region sample dataset X j The weight vector of

[0072] Step 4: Perform a sign test on the S connection strengths between the kth brain region and the jth brain region. If the test result is positive, it means that the connection relationship between the kth brain region and the jth brain region is a covariant connection; if the test result is negative, it means that the connection relationship between the kth brain region and the jth brain region is an inverse connection; if the test result is not significant, it means that the connection relationship between the kth brain region and the jth brain region is not significant;

[0073] Step 5: A brain covariant connection network is formed by the brain regions whose connection relationship among each brain region is covariant connection; a brain inverse connection network is formed by the brain regions whose connection relationship among each brain region is inverse connection.

[0074] Similarly, repeat steps 2, 3, 4, and 5 for the second set of sample data sets to construct the brain connection network. The connection networks obtained from the two sets of samples are as follows: Figure 2 and Figure 3 As shown, the solid line represents the covariant network, the dotted line represents the inverse network, and the group-level connection strength matrix is ​​displayed in the upper right corner of the picture.

[0075] In this embodiment, an electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the above-mentioned brain connection network construction method, and the processor is configured to execute the program stored in the memory.

[0076] In this embodiment, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned brain connection network construction method are executed.

[0077] Furthermore, in order to verify the reliability of the brain connection network constructed by the present invention, cosine similarity is used to measure the repeatability of the connection strength obtained from the two sets of data sets. The higher the cosine similarity, the stronger the repeatability and the more reliable the brain connection network obtained. The brain area activity characterization method using average signal (Average) and principal component analysis (PCA) is used to compare with the algorithm of the present invention (HMCCA), and the results are as follows: Figure 4 The repeatability of HMCCA is significantly higher than that of the other two traditional methods, which shows that the present invention can extract reliable brain activity features and then construct brain covariant and inverse connection networks, which has high clinical practical value.

[0078] In summary, this method can simultaneously extract covariant and inverse connectivity networks across the brain at both the group and individual levels, ensuring voxel homogeneity within brain regions while effectively characterizing regional activity. This provides an effective means for comprehensively studying functional connectivity patterns in the brain, and is of great significance for advancing research into brain function and related brain diseases.

Claims

1. A method for constructing a brain connectivity network based on homogeneity-constrained multi-set canonical correlation analysis, characterized in that: The steps are as follows: Step 1: Use signal acquisition equipment to obtain After the functional magnetic resonance imaging signals of each individual are registered to the standard template, the The brain regions are divided to obtain the total sample data set , express After individual registration Brain region sample datasets, is the sample length, For the The number of voxels in a brain region, and , represents the transpose of the matrix, Indicates the Among individuals Signals within brain regions, express The transpose of Step 2: Use the homogeneity constrained multi-set canonical correlation analysis algorithm to analyze the total sample data set Calculate and get the weight vector of each brain region sample data set ,in, Indicates the The weight vector of the brain region sample data set; Step 2.1: Use formula (1) to build the weight vector under the sign consistency constraint to maximize The optimization model takes the sum of squares of correlation coefficients between the representation signals of brain regions as the target: (1) In formula (1), Representative The weight vector of the brain region sample dataset of The elements are all non-negative or non-positive, express The transpose of Step 2.2: Use the interior point method and Newton method to iteratively solve the optimization model: Step 2.2.1: Use formula (2) to construct the loss function : (2) In formula (2), represents the weight matrix, and , is the relaxation matrix, Indicates the slack variables, express The transpose of is the Lagrange multiplier matrix, Indicates the Lagrange multipliers, is a positive barrier parameter, for The inverse of the sum of squares of the correlation coefficients between the two brain regions, trace represents the trace of the matrix, The correlation matrix representing brain region activity, is the constraint function matrix, Indicates the constraint functions, and , , Indicates the Brain region sample datasets, express The transpose of represents a diagonal matrix, Constraint Function , Indicates the slack variables; Step 2.2.2: Define the number of outer iterations as n, the number of inner iterations as t, and the given parameters , and initialize n=1, and the nth outer iteration weight vector Initialized to unit vectors, where Represents the kth weight vector under the nth outer iteration and is initialized to the kth unit vector. The relaxation matrix under the nth outer iteration is Initialize it to a matrix of all 1s and convert the Lagrange multiplier matrix under the nth outer iteration into Initialize it to a matrix of all 1s and set the barrier parameter under the nth outer iteration Initialized to 1; Step 2.2.3: Initialize t=1; Step 2.2.4: Substitute the results of the tth inner iteration under the nth outer iteration weight vector Initialized as a weight vector ,in, represents the kth weight vector of the tth inner iteration under the nth outer iteration and is initialized to , the relaxation vector of the t-th inner iteration under the n-th outer iteration is Initialized to , the Lagrange multiplier matrix of the tth inner iteration under the nth outer iteration Initialized to ; Step 2.2.5: Calculate the triplet of changes in the tth inner iteration under the nth outer iteration according to formula (3): : (3) In formula (3), Indicates the tth inner iteration under the nth outer iteration The amount of change, Indicates the tth inner iteration under the nth outer iteration The amount of change, Indicates the tth inner iteration under the nth outer iteration Change amount, represents the Hessian matrix of the loss function of the t-th inner iteration under the n-th outer iteration, and , represents the inverse of the sum of squares of the correlation coefficients, represents the Lagrange multiplier matrix of the tth inner iteration under the nth outer iteration Middle Lagrange multipliers, represents the gradient operator, express The constraint function matrix The Jacobian matrix, express The transpose of is The diagonal matrix consisting of the diagonal elements in , is The diagonal matrix consisting of the diagonal elements in , represents the identity matrix, represents a vector of all 1s; Step 2.2.6: Use formula (4) to obtain the triplet of the t+1th inner iteration under the nth outer iteration : (4) In formula (4), represents the weight vector of the t+1th inner iteration under the nth outer iteration, represents the relaxation vector of the t-th inner iteration under the n-th outer iteration, represents the Lagrange multiplier matrix of the tth inner iteration under the nth outer iteration; Step 2.2.7: Assign to Then return to step 2.2.5 and execute it sequentially until Until the inner iteration ends; among them, Indicates the threshold value; Step 2.2.8: Use formula (5) to obtain the triples under the n+1th outer iteration : (5) Step 2.2.9: Calculate the barrier parameters for the n+1th outer iteration ; Step 2.2.10: Assign to Then return to step 2.2.3 and execute sequentially until So far, the outer iteration ends; Step 2.2.11: Set the kth weight vector under the n+1th outer iteration As the first The weight vector of the brain region sample dataset , thus obtaining the weight vector ; Step 3: Calculate the Among individuals Representational signals in brain regions , and the Among individuals Representational signals within brain regions Hedi Representational signals in brain regions The correlation coefficient between Among individuals The brain and The connection strength of each brain region is obtained Among individuals The brain and The connection strength of brain regions; Indicates the Among individuals Signals within brain regions, Indicates the Brain region sample dataset The weight vector of Step 4: brain regions and between brain regions The sign of the connection strength is tested. If the test result is positive, it means that the brain regions and The connection between brain regions is covariant connection; if the test result is negative, it means that brain regions and The connection between the brain regions is an inverse connection; if the test result is not significant, it means that the brain regions and The connectivity between brain regions was not significant; Step 5: By The brain regions with covariant connections among the brain regions constitute a brain covariant connection network; The brain regions in which the connection relationship between each brain region is inverse connection constitute a brain inverse connection network.

2. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the brain connection network construction method according to claim 1, and the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the brain connection network construction method according to claim 1 are executed.

Images