A method for estimating heterogeneous functional brain networks

Abstract

The application discloses a heterogeneous functional brain network estimation method, comprising the following steps: pre-processing resting-state functional magnetic resonance imaging data output by a functional magnetic resonance device; dividing a brain function image after pre-processing into a plurality of brain regions based on a standard template; calculating connection weights between the brain regions as candidates based on a plurality of methods; and improving an orthogonal matching pursuit algorithm, configuring different types of weights for each pair of brain regions, and generating a heterogeneous functional brain network. The method is based on a new assumption that connection types between different brain regions are different, and at most only one type of interaction mode exists between any two brain regions, so that the estimated heterogeneous functional brain network can better depict complex interaction modes of the brain.

Description

Technical Field

[0001] This invention relates to the field of biomedical information processing technology, specifically a method for estimating heterogeneous functional brain networks. Background Technology

[0002] Autism spectrum disorder is characterized by reduced social interaction and impaired language communication. This brain dysfunction has a serious impact on the mental and physical health and daily life of patients. It has been reported that the functional brain networks constructed under different interaction patterns differ across brain regions in individuals with autism spectrum disorder. Therefore, constructing an appropriate functional brain network is crucial for exploring biomarkers associated with autism spectrum disorder.

[0003] In recent years, estimating functional brain networks using resting-state functional magnetic resonance imaging (fMRI) data to quantify interactions between different brain regions has been widely used for the early diagnosis of autism spectrum disorder. In current research, the presence of connectivity is typically determined based on correlation measurements of blood oxygenation levels in two brain regions, resulting in an adjacency matrix of interaction information between the regions, where weights represent the connection strength between the two regions. Therefore, research on functional brain networks offers an opportunity to identify abnormal brain functional connectivity in patients with brain disorders, which is crucial for exploring potential biomarkers associated with autism spectrum disorder classification.

[0004] Currently, all proposed functional brain network analysis models, whether using traditional machine learning methods or deep learning methods, first estimate a homogeneous functional brain network. Then, based on this functional brain network, the nodes (regions of interest in the brain) or edges (connections between brain regions) are analyzed; for example, Zhang H, Chen X, Shi F, et al. Topographical information-based high-order functional connectivity and its application in abnormality detection for mild cognitive impairment[J]. Journal of Alzheimer's Disease, 2016, 54(3): 1095-1112; Wen G, Cao P, Bao H, et al. MVS-GCN: A prior brain structure learning-guided multi-view graphconvolution network for autism spectrum disorder diagnosis[J]. Computers in Biology and Medicine, 2022, 142: 105239; Jiang X, Zhang L, Qiao L, et al. Estimating functional connectivity networks via low-rank tensor approximationwith applications to MCI identification[J]. IEEE Transactions on BiomedicalEngineering, 2019, 67(7): 1912-1920.

[0005] The functional brain networks estimated in the aforementioned literature are all based on the assumption that the interaction patterns between different brain regions are identical. Taking brain regions A, B, and C as an example: using the Pearson correlation coefficient to estimate the functional brain network, the interaction patterns between brain regions A and B, A and C, and B and C are all linearly correlated. We call this type of brain network estimated using the same correlation a homogeneous functional brain network. However, the brain is a very complex system, and using homogeneous functional brain networks to characterize the interaction patterns between different brain regions has certain limitations. Summary of the Invention

[0006] This invention proposes a heterogeneous functional brain network: different brain regions can have different types of interactions, and any two brain regions can have at most one type of interaction. The technical problem this invention aims to solve is to provide a method for estimating heterogeneous functional brain networks: an improved orthogonal matching pursuit algorithm is used to assign different weights to each pair of brain regions, which are then estimated using different methods to generate the heterogeneous functional brain network. The heterogeneous functional brain network better characterizes the complex interactions between different brain regions, and the effectiveness of the proposed model and algorithm is verified on an autism spectrum disorder recognition task.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A method for estimating heterogeneous functional brain networks includes the following steps:

[0009] (1) For resting-state functional magnetic resonance imaging data acquired by the functional magnetic resonance device, the following preprocessing was performed using the DPARSF toolbox: 1) Remove the anterior end of each subject's data. 1) Unstable time points; 2) Head movement and time-level correction; 3) Regression of the influence of ventricles, white matter, and global signals; 4) Registration to standard space using the Montreal Neurological Institute template; 5) Using A bandpass filter is used to filter the airflow to reduce the effects of heartbeat and respiration.

[0010] (2) After preprocessing the resting-state functional magnetic resonance imaging (fMRI) data, the brain was divided into several brain regions based on an automated anatomical marker template, and the average time series of each brain region was extracted and represented as follows: ,in Indicates the first Time series of brain regions Indicates the length of the time series. Indicates the number of brain regions;

[0011] (3) Based on the extracted average time series of brain regions, multiple candidate functional brain networks are estimated for each subject's brain region, denoted as follows: , Indicates the first A candidate functional brain network; due to It is a symmetric matrix. To avoid feature redundancy, we choose... The upper triangular elements are then concatenated sequentially to form a row vector.

[0012] (4) Divide the various types of connections between brain regions corresponding to multiple candidate functional brain networks spliced ​​into row vectors into a group, that is, the candidate connections between any two brain regions estimated by multiple methods are represented as follows: ,in This indicates the group containing any two brain regions. This represents the number of candidate functional brain networks; multiple connections in each subject's brain region are represented as... ,in This indicates the number of groups, i.e., the number of brain region pairs. Indicates the first Multiple types of connections to brain regions, namely the first Between brain regions Types of connections, represented as ];

[0013] The estimation model for heterogeneous functional brain networks is as follows:

[0014]

[0015]

[0016] in Indicates the first A projection column vector of a brain region, express The label vector of each subject; through constraints There can be at most one way for each pair of brain regions to interact;

[0017] (5) Improved Orthogonal Matching Pursuit Algorithm: Define a set Store the index of the group containing the selected heterogeneous join, where Indicates the first The index selected in the next iteration; This is used to represent the selected connections between each pair of brain regions, where Indicates the first The selected in the nth iteration Group 1 One connection, ;

[0018] (6) Initialize residuals ,gather Empty; in the first In the next iteration, Select with the The residual obtained in the next iteration The index with the maximum related join, i.e.:

[0019]

[0020] (7) Index Add to collection In, that is Simultaneous indexing All other connections within the corresponding group become 0, meaning they are not selected. , Indicates index All connections for the corresponding group; then update the residuals using the least squares method:

[0021]

[0022] Update residuals:

[0023]

[0024] (8) Repeat the above process, selecting connections in sequence, until the number of iterations reaches the set value. Or, when the sum of squares of the residuals: When the time comes, stop updating; the final generated This refers to heterogeneous functional brain networks.

[0025] This invention proposes a novel hypothesis to explore the interaction patterns between different brain regions: different brain regions can have different types of interaction patterns, and any two brain regions can have at most one type of interaction. Taking brain regions A, B, and C as an example: multiple functional brain networks are estimated using various methods, and a modified orthogonal matching pursuit algorithm is used to estimate a heterogeneous functional brain network. This determines the specific type of interaction between any two brain regions within regions A and B, A and C, and B and C, i.e., whether each pair of brain regions is linearly correlated, biased, or has a higher-order correlation. This results in brain networks with different interaction patterns between different brain regions. The aforementioned brain network is the heterogeneous functional brain network proposed in this invention, and it can better characterize the complex interaction patterns between different brain regions.

[0026] As a further improvement to this technical solution:

[0027] Step (3) uses three methods to estimate the functional brain network based on the extracted average time series of brain regions. These are, respectively, fully correlated, partially correlated, and correlated correlations;

[0028] Taking three brain regions A, B, and C as an example, the method of this invention uses three pairs of brain regions, namely brain regions A and B, brain regions A and C, and brain regions B and C, to estimate the correlation between each pair of brain regions, so that there are brain networks with different types of interaction between different brain regions, namely the heterogeneous functional brain network proposed in this invention. This heterogeneous functional brain network can better characterize the complex interaction between different brain regions.

[0029] The beneficial technical effects of this invention after adopting the above technical solution are as follows: a heterogeneous functional brain network is proposed, which can better characterize the complex interaction mode of the brain, thereby solving the limitation of homogeneous functional brain networks in the prior art, that is, different brain regions use the same method to calculate the correlation; an improved orthogonal matching pursuit algorithm is proposed, which solves the assumption that there is at most one interaction mode between each pair of brain regions; and the effectiveness of the proposed model is verified on real brain disease datasets. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the implementation of the heterogeneous functional brain network estimation model of this invention. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings and specific implementation examples.

[0032] like Figure 1 As shown, a specific implementation example of a heterogeneous functional brain network estimation model includes the following steps:

[0033] (1) In this example, a publicly available Autism Brain Imaging Data Exchange (ABIDE) dataset was used, selecting data from the two largest sites: New York University (NYU) and the University of Michigan (UM). The NYU dataset contained 184 participants, including 79 autistic patients and 105 healthy controls; the UM dataset contained 145 participants, including 68 autistic patients and 77 healthy controls. The resting-state functional magnetic resonance imaging (fMRI) data acquired by the MRI machine were read and converted in format. The first 10 unstable time points for each participant were excluded; head movement and time-layer correction were performed; the effects of ventricle drop, white matter, and global signals were regressed; the data were registered to standard space using the Montreal Neurological Institute template; and the data were used to... A bandpass filter is used to filter the airflow to reduce the effects of heartbeat and respiration.

[0034] (2) Select a standardized brain region template. In this example, the AAL template is selected to divide the brain into 116 brain regions, including 90 cerebral brain regions and 26 cerebellar brain regions. The 116 brain regions represent 116 nodes of the functional brain network.

[0035] (3) The average time series of each brain region is extracted and represented as follows: ,in Indicates the first Time series of brain regions Indicates the length of the time series. Indicates the number of brain regions;

[0036] (4) Based on the extracted average time series of brain regions, multiple candidate functional brain networks are estimated for each subject's brain region, denoted as follows: , Indicates the first A candidate functional brain network; due to It is a symmetric matrix. To avoid feature redundancy, we choose... The upper triangular elements are then concatenated sequentially to form a row vector.

[0037] (5) Divide the various types of connections between brain regions corresponding to multiple candidate functional brain networks spliced ​​into row vectors into a group, that is, the candidate connections between any two brain regions estimated by multiple methods are represented as follows: ,in This indicates the group containing any two brain regions. This represents the number of candidate functional brain networks; multiple connections in each subject's brain region are represented as... ,in This indicates the number of groups, i.e., the number of brain region pairs. Indicates the first Multiple types of connections to brain regions, namely the first Between brain regions Types of connections, represented as ];

[0038] The estimation model for heterogeneous functional brain networks is as follows:

[0039]

[0040]

[0041] in Indicates the first A projection column vector of a brain region, express The label vector of each subject; through constraints There can be at most one way for each pair of brain regions to interact;

[0042] (6) Improved Orthogonal Matching Pursuit Algorithm: Define a set Store the index of the group containing the selected heterogeneous join, where Indicates the first The index selected in the next iteration; This is used to represent the selected connections between each pair of brain regions, where Indicates the first The selected in the nth iteration Group 1 One connection, ;

[0043] (7) Initialize residuals ,gather Empty; in the first In the next iteration, Select with the The residual obtained in the next iteration The index with the maximum related join, i.e.:

[0044]

[0045] (8) Index Add to collection In, that is Simultaneous indexing All other connections within the corresponding group become 0, meaning they are not selected. , Indicates index All connections for the corresponding group; then update the residuals using the least squares method:

[0046]

[0047] Update residuals:

[0048]

[0049] (9) Repeat the above process in the experiment, selecting connections sequentially until the number of iterations reaches the set value. Or, when the sum of squares of the residuals: When the time comes, stop updating; the final generated This refers to heterogeneous functional brain networks;

[0050] (10) In step (4), based on the extracted brain region mean time series, this example uses three types of correlations to calculate the functional brain network in different brain regions. These are, respectively, full correlation (Pearson correlation), partial correlation (sparse representation), and correlational correlation (higher-order correlation).

[0051] (11) To verify the effectiveness of the selected features (connections), a classification experiment was conducted on the autism brain imaging data exchange dataset, comparing autistic patients with healthy controls. Furthermore, to avoid the influence of complex feature selection and classifiers on the experimental results, this example used... Two samples Validation and default parameters of The -normSVM classifier performs feature selection and classification; model effectiveness evaluation metrics include accuracy, sensitivity, specificity, and the area under the cost-sensitivity curve (AUC); the definitions of the metrics are as follows: , , TP, TN, FP, and FN represent true positive, true negative, false positive, and false positive, respectively.

[0052] (12) As shown in Table 1, these are the performance comparison tables of the heterogeneous functional brain network classification indexes based on three baseline methods: PC, SR, and CC, and this embodiment.

[0053] Table 1: Comparison of classification index performance based on three baseline methods (PC, SR, and CC) and the heterogeneous functional brain network proposed in this invention.

[0054]

[0055] As shown in Table 1, this embodiment achieved the best performance across all metrics: Accuracy=74.46%, Sensitivity=74.68%, Specificity=74.29%, and AUC=76.83% on the NUY dataset; and Accuracy=68.97%, Sensitivity=66.18%, Specificity=71.43%, and AUC=69.12% on the UM dataset.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, and partial additions made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for estimating heterogeneous functional brain networks, characterized in that... Includes the following steps: (1) For resting-state functional magnetic resonance imaging data acquired by the functional magnetic resonance device, the following preprocessing was performed using the DPARSF toolbox: (1.1) Remove the anterior end of each subject's data. An unstable time point; (1.2) Perform head movement and time-level correction; (1.3) Regression of the effects of ventricles, white matter, and global signals; (1.4) Registration to standard space using templates from the Montreal Institute of Neuroscience; (1.5) Filtering with a 0.01-0.1 Hz bandpass filter to reduce the effects of heartbeat and respiration; (2) After preprocessing the resting-state functional magnetic resonance imaging (fMRI) data, the brain was divided into several brain regions based on an automated anatomical marker template, and the average time series of each brain region was extracted and represented as follows: ,in Indicates the first Time series of brain regions Indicates the length of the time series. Indicates the number of brain regions; (3) Based on the extracted average time series of brain regions, multiple candidate functional brain networks are estimated for each subject's brain region, denoted as follows: , Indicates the first A candidate functional brain network; due to It is a symmetric matrix. To avoid feature redundancy, we choose... The upper triangular elements are then concatenated sequentially to form a row vector. (4) Divide the various types of connections between brain regions corresponding to multiple candidate functional brain networks spliced ​​into row vectors into a group, that is, the candidate connections between any two brain regions estimated by multiple methods are represented as follows: ,in This indicates the group containing any two brain regions. This represents the number of candidate functional brain networks; multiple connections in each subject's brain region are represented as... ,in This indicates the number of groups, i.e., the number of brain region pairs. Indicates the first Multiple types of connections to brain regions, namely the first Between brain regions Types of connections, represented as ]; The estimation model for heterogeneous functional brain networks is as follows: in Indicates the first A projection column vector of a brain region, express The label vector of each subject; through constraints There can be at most one way for each pair of brain regions to interact; (5) Improved Orthogonal Matching Pursuit Algorithm: Define a set Store the index of the group containing the selected heterogeneous join, where Indicates the first The index selected in the next iteration; This is used to represent the selected connections between each pair of brain regions, where Indicates the first The selected in the nth iteration Group 1 One connection, ; (6) Initialize residuals ,gather Empty; in the first In the next iteration, Select with the first The residual obtained in the next iteration The index with the largest related join, i.e.: (7) Index Add to collection In, that is Simultaneous indexing All other connections within the corresponding group become 0, meaning they are not selected. , Indicates index All connections for the corresponding group; then update the residuals using the least squares method: Update residuals: (8) Repeat the above process, selecting connections in sequence, until the number of iterations reaches the set value. Or, when the sum of squares of the residuals: When the time comes, stop updating; the final generated This refers to heterogeneous functional brain networks.

2. The method for estimating heterogeneous functional brain networks according to claim 1, characterized in that, Step (3) uses three methods to estimate the functional brain network based on the extracted average time series of brain regions: full correlation, partial correlation, and higher-order correlation.

Images