A classification method for mental illness based on local coupling patterns of structural-functional brain networks

Abstract

The present invention provides a method for classifying mental illness based on the local coupling pattern of the structural-functional brain network, which belongs to the field of medical information intelligent diagnosis technology; its technical solution is: frequent subgraph mining is performed on the connection matrix of the structural brain network to obtain all frequently connected subgraphs; the coupling coefficient between the structural brain network and the functional brain network is calculated through all the frequently connected subgraphs mined; then the coupling coefficient obtained is feature selected using the lasso model through five-fold cross validation; the coupling coefficient obtained after attribute simplification is classified using random forests to evaluate the accuracy. The beneficial effects of the present invention are: the present invention can better capture the key features of the structural-brain network, significantly improve the accuracy and efficiency of mental illness classification, and help doctors formulate more effective treatment plans for patients.

Description

Technical Field

[0001] The present invention relates to the field of medical information intelligent diagnosis technology, and in particular to a method for classifying mental illnesses based on a local coupling pattern of a structure-function brain network. Background Art

[0002] Mental illness is mainly a group of neurological diseases that manifest as disorders in behavior and psychological activities. These diseases may cause emotional problems, cognitive difficulties, abnormal behavior or social disorders. The symptoms and severity vary depending on the type of disease and the individual. Therefore, early diagnosis and intervention of mental illness are very important.

[0003] Currently, traditional methods of classifying mental illness usually rely on diagnostic criteria and clinical assessments to classify mental illnesses based on the patient's symptoms and behaviors. This method is subjective in some cases and ignores the biological and neurological complexity of mental illness, so there is uncertainty in the diagnosis.

[0004] The paper Disrupted structural and functional brain networks in Alzheimer's disease used graph theory analysis to study the topological organization of structural brain networks and functional brain networks, and concluded that there were overlaps and different network disruptions between structural brain networks and functional brain networks, but did not deeply explore the frequent subgraphs of structural brain networks and functional brain networks, resulting in low classification accuracy.

[0005] The literature Coupling analysis between functional and structural brain networks in Alzheimer's disease uses the Pearson correlation coefficient method to analyze the correlation between structural brain networks and functional brain networks. However, the Pearson correlation coefficient method can only measure linear correlation and cannot effectively capture nonlinear relationships.

[0006] Therefore, analyzing the local coupling patterns of structural and functional brain networks can more accurately diagnose and classify mental illnesses. Summary of the invention

[0007] The purpose of the present invention is to provide a method for classifying mental illness based on the local coupling pattern of structure-function brain network, so as to improve the accuracy of mental illness identification.

[0008] The inventive concept of the present invention is: the local coupling pattern of structural-functional brain network refers to the interconnected pattern between brain structure and function, with special emphasis on the connection and collaborative work between local brain regions. The structural brain network focuses on the physical structure of the brain, reveals the anatomical connection and communication path between various brain regions, and is usually constructed using diffusion tensor data; the functional brain network focuses on the activity and function of the brain, describes the activity pattern and information transfer between different brain regions, and is usually constructed using functional magnetic resonance imaging data. Combining the structural and functional information of the brain, abnormal connections and collaborative work between brain regions can be found, which can better reflect the biological basis of mental illness and improve the accuracy, personalization and biological understanding of mental illness classification. Therefore, classifying mental illness by analyzing the local coupling pattern of structural brain network and functional brain network is an emerging method that combines the structural and functional information of the brain to more accurately diagnose and classify mental illness.

[0009] In order to better achieve the above-mentioned invention object, the present invention is implemented by the following technical solution: a classification method for mental illness based on the local coupling pattern of structure-function brain network, the specific steps are as follows:

[0010] A. In a mental illness dataset, there are two important brain map networks, namely the structural brain network and the connection matrix A. S ∈R N×N Representation and function of brain network, represented by the connection matrix A F ∈R N×N Indicates that they mainly represent different brain regions, where N is the number of brain network nodes. Graph G1 represents the structural brain network, and graph G2 represents the functional brain network, where each node corresponds to a brain region, and edges represent the connections between different brain regions. The gSpan frequent subgraph mining algorithm is used to mine all the frequently connected subgraphs g from the structural brain network graph data G1. These subgraphs can represent the common features of multiple graphs in the structural brain network graph data.

[0011] B. Convert all frequently connected subgraphs g found in the form of adjacency lists into corresponding connection matrices according to the mapping relationship between labels and nodes, and let B∈R N×N Represents, through the subgraph connection matrix B∈R N×N The node and edge information in the brain network are used to extract relevant network subsets from the connection matrices of the structural brain network and the functional brain network, respectively. Then, based on the obtained structural subnetwork and functional subnetwork, the Spearman correlation coefficient method is used to calculate the structure-function coupling coefficient. These coupling coefficients reflect the degree of association between the structural brain network and the functional brain network.

[0012] C. Through five-fold cross validation, the lasso model is used to perform feature selection on the calculated coupling coefficients. In this process, the obtained coupling coefficients are divided into five subsets, four of which are used to train the model, and the fifth subset is used to test the performance of the model. This process will be repeated multiple times, using different subsets as test sets each time to ensure a robust model evaluation. In each cross validation, the lasso model will evaluate the importance of each feature and select the feature with the greatest predictive power for the target attribute. Through this feature selection method, the coupling coefficients after attribute simplification are obtained. These simplified coupling coefficients can reflect the degree of association between structural and functional brain networks.

[0013] D. Use the coupling coefficient after attribute simplification as a feature and apply the random forest classification model for further analysis. Random forest can integrate the results of multiple decision trees to improve model performance. Through random forest, the coupling coefficient can be classified and its accuracy in diagnosing mental illness can be evaluated.

[0014] As the mental illness classification method based on the local coupling pattern of structure-function brain network provided by the present invention, the specific steps of step A are as follows:

[0015] a. Use graph data G1 to represent the structural brain network connection matrix A S ∈R N×N , where graph G1 exists in the form of an adjacency list. Assume that the node label set is {A, B, C, ...}, the edge label set is {a, b, c, ...}, and the support threshold is set to minsup;

[0016] b. Sort the labels in G1 by frequency, remove the nodes and edges with lower frequencies according to minsup, and obtain a new graph G1'. Relabel the remaining nodes and edges in graph G1', and keep the mapping relationship between the original labels and the new labels. Combine the edges with higher frequencies into a set S and sort them by their frequencies. The edges in set S can be considered as the simplest frequent subgraph S'.

[0017] c. For each edge in subgraph S', first determine whether it satisfies the minimum DFS encoding. If so, perform the rightmost expansion on subgraph S' to obtain S", and then determine whether S" satisfies the support minsup. If so, repeat this step and recursively mine S" until no new frequent subgraphs are generated. All subgraphs found are saved in the form of adjacency lists.

[0018] As the mental illness classification method based on the local coupling pattern of structure-function brain network provided by the present invention, the specific steps of step B are as follows:

[0019] a. Assume that we get n subgraphs, represented as g1, g2, …, g n , according to the mapping relationship between labels and nodes, it is converted into the corresponding connection matrix, represented by B∈R N×N It means that for each subgraph g i , i = 1, 2, ..., n, there is an adjacency matrix B i , i=1,2,...,n;

[0020] b. Through the subgraph adjacency matrix B i , i=1,2,...,n, the node and edge information in the functional brain network A F ∈R N ×N Extract the adjacency matrix B of the resulting subgraph i , i=1,2,...,n, the relevant functional network subset;

[0021] c. Through the subgraph adjacency matrix B i , i = 1, 2, ..., n, the node and edge information in the structured brain network A S ∈R N ×N Extract the adjacency matrix B of the resulting subgraph i , i=1,2,...,n, the relevant structural network subset;

[0022] d. The function-structure coupling coefficient is calculated for the extracted functional subnetwork and structural subnetwork using the Spearman correlation coefficient method. The formula is as follows:

[0023]

[0024] Among them, ρ is the structure-function coupling coefficient, which is used to measure the degree of association between the structural brain network and the functional brain network, d is the difference between the ranks of the structural brain network and the functional brain network, and n is the number of sample data points of the structural brain network and the functional brain network.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. The mental illness classification method proposed in the present invention adopts the local coupling model of structural-functional brain networks, eliminating the subjectivity of traditional mental illness classification methods, thereby providing a more objective and accurate mental illness classification.

[0027] 2. By analyzing the coupling coefficient between structural brain networks and functional brain networks, personalized mental illness diagnosis and treatment recommendations can be provided to each patient, helping doctors better meet the specific needs of each patient and develop more effective treatment plans.

[0028] 3. The present invention adopts frequent subgraph mining and feature selection methods, which can better capture the key features of structural-functional brain networks, thereby significantly improving the accuracy and efficiency of classification, helping to identify and intervene in mental illness earlier, and improving the treatment prospects of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention but do not constitute a limitation of the present invention.

[0030] Figure 1 This is the overall framework diagram of the mental illness classification method based on the local coupling pattern of the structural-functional brain network of the present invention.

[0031] Figure 2 This is a model diagram of the mental illness classification method based on the local coupling pattern of the structural-functional brain network of the present invention. DETAILED DESCRIPTION

[0032] In order to deepen the understanding of the present invention, the present invention will be further described in detail below in conjunction with embodiments and drawings. The embodiments are only used to explain the present invention and do not constitute a limitation on the protection scope of the present invention.

[0033] See also Figure 1 , Figure 2 This embodiment is specifically implemented by the following technical solution: a method for classifying mental illness based on the local coupling pattern of structure-function brain network, the specific steps are as follows:

[0034] A. In a mental illness dataset, there are two important brain map networks, namely the structural brain network and the connection matrix A. S ∈R N×N Representation and function of brain network, represented by the connection matrix A F ∈R N×N Indicates that they mainly represent different brain regions, where N is the number of brain network nodes. Graph G1 represents the structural brain network, and graph G2 represents the functional brain network, where each node corresponds to a brain region, and the edge represents the connection between different brain regions. The gSpan frequent subgraph mining algorithm is used to mine all the frequently connected subgraphs g from the structural brain network graph data G1. These subgraphs can represent the common features of multiple graphs in the structural brain network graph data. The specific steps are as follows:

[0035] a. Use graph data G1 to represent the structural brain network connection matrix A S ∈R N×N , where graph G1 exists in the form of an adjacency list. Assume that the node label set is {A, B, C, ...}, the edge label set is {a, b, c, ...}, and the support threshold is set to minsup;

[0036] b. Sort the labels in G1 by frequency, remove the nodes and edges with lower frequencies according to minsup, and obtain a new graph G1'. Relabel the remaining nodes and edges in graph G1', and keep the mapping relationship between the original labels and the new labels. Combine the edges with higher frequencies into a set S and sort them by their frequencies. The edges in set S can be considered as the simplest frequent subgraph S'.

[0037] c. For each edge in subgraph S', first determine whether it satisfies the minimum DFS encoding. If so, perform the rightmost expansion on subgraph S' to obtain S", and then determine whether S" satisfies the support minsup. If so, repeat this step and recursively mine S" until no new frequent subgraphs are generated. All subgraphs found are saved in the form of adjacency lists.

[0038] The above model was tested using a mental illness brain map dataset, with 217 samples and decision attributes divided into two categories: patients with confirmed mental illness and normal people. The connection matrices of the functional brain network and the structural brain network are A F ∈R 90×90 and A S ∈R 90×90 , setting the support minsup to 200, the node upper limit to 5, and the node lower limit to 3. According to the gSpan frequent subgraph mining algorithm, 3574 frequently connected subgraphs can be obtained in the end.

[0039] B. Convert all frequently connected subgraphs g found in the form of adjacency lists into corresponding connection matrices according to the mapping relationship between labels and nodes, and let B∈R N×N Represents, through the subgraph connection matrix B∈R N×N The node and edge information in the brain network are used to extract relevant network subsets from the connection matrices of the structural brain network and the functional brain network, respectively. Then, based on the obtained structural subnetwork and functional subnetwork, the Spearman correlation coefficient method is used to calculate the structure-function coupling coefficient. These coupling coefficients reflect the degree of association between the structural brain network and the functional brain network. The specific steps are as follows:

[0040] a. Assume that we get n subgraphs, represented as g1, g2, …, g n , according to the mapping relationship between labels and nodes, it is converted into the corresponding connection matrix, represented by B∈R N×N It means that for each subgraph g i , i = 1, 2, ..., n, there is an adjacency matrix B i , i=1,2,...,n;

[0041] b. Through the subgraph adjacency matrix Bi , i=1,2,...,n, the node and edge information from the functional brain network A F ∈R N×N Extract the adjacency matrix B of the resulting subgraph i , i=1,2,...,n, the relevant functional network subset;

[0042] c. Through the subgraph adjacency matrix B i , i = 1, 2, ..., n, the node and edge information in the structured brain network A S ∈R N ×N Extract the adjacency matrix B of the resulting subgraph i , i=1,2,...,n, the relevant structural network subset;

[0043] d. The function-structure coupling coefficient is calculated for the extracted functional subnetwork and structural subnetwork using the Spearman correlation coefficient method. The formula is as follows:

[0044]

[0045] Among them, ρ is the structure-function coupling coefficient, which is used to measure the degree of association between the structural brain network and the functional brain network, d is the difference between the ranks of the structural brain network and the functional brain network, and n is the number of sample data points of the structural brain network and the functional brain network.

[0046] The 3574 frequently connected subgraphs found (by B i ∈R 90×90 Represented by, where i = 1, 2, ..., 3574, extracting the functional brain network connection matrix A F ∈R 90×90 and the structural brain network connection matrix A S ∈R 90×90 For the subset of related networks, using the Spearman correlation coefficient method, 3574×90×90 coupling coefficients can be obtained, and their value range remains between -1 and 1.

[0047] C. Through five-fold cross validation, the lasso model is used to perform feature selection on the calculated coupling coefficients. In this process, the obtained coupling coefficients are divided into five subsets, four of which are used to train the model, and the fifth subset is used to test the performance of the model. This process will be repeated multiple times, using different subsets as test sets each time to ensure a robust model evaluation. In each cross validation, the lasso model will evaluate the importance of each feature and select the feature with the greatest predictive power for the target attribute. Through this feature selection method, the coupling coefficients after attribute simplification are obtained. These simplified coupling coefficients can reflect the degree of association between structural and functional brain networks.

[0048] D. Use the coupling coefficient after attribute simplification as a feature and apply the random forest classification model for further analysis. Random forest can integrate the results of multiple decision trees to improve model performance. Through random forest, the coupling coefficient can be classified and its accuracy in diagnosing mental illness can be evaluated.

[0049] The model was trained by dividing it into different training sets, test sets and validation sets five times to obtain different accuracy rates, among which the maximum accuracy rate was 91.7% and the average accuracy rate was 86.2%. Therefore, it can be verified that the model can improve the accuracy of identifying mental illness.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for classifying mental illnesses based on local coupling patterns of structural-functional brain networks, characterized in that: The following steps are involved: S1. In a mental illness dataset, there are two important brain map networks, namely the structural brain network and the connection matrix A. S ∈R N×N Representation and function of brain network, represented by the connection matrix A F ∈R N×N Indicates that they represent different brain regions, where N is the number of brain network nodes. Graph G1 represents the structural brain network, and graph G2 represents the functional brain network, where each node corresponds to a brain region, and edges represent the connections between different brain regions. The gSpan frequent subgraph mining algorithm is used to mine all the frequently connected subgraphs g from the structural brain network graph data G1. These subgraphs represent the common features of multiple graphs in the structural brain network graph data. The steps of performing frequent subgraph mining on the structural brain network connection matrix in step S1 are as follows: S11. Use graph data G1 to represent the structural brain network connection matrix A S ∈R N×N , where graph G1 exists in the form of an adjacency list. Assume that the node label set is {A, B, C, ...}, the edge label set is {a, b, c, ...}, and the support threshold is set to minsup; S12, sort the labels in G1 by frequency, remove the nodes and edges with lower frequency according to minsup, and thus obtain a new graph G1', relabel the remaining nodes and edges in graph G1', and retain the mapping relationship between the original labels and the new labels, combine the edges with higher frequency into a set S, and sort them according to the frequency of the edges. The edges in set S are considered to be the simplest frequent subgraph S'; S13. For each edge in the subgraph S', first determine whether it satisfies the minimum DFS encoding. If so, perform the rightmost expansion on the subgraph S' to obtain S", and then determine whether S" satisfies the support minsup. If so, repeat this step and recursively mine S" until no new frequent subgraphs are generated. All subgraphs found are saved in the form of adjacency lists. S2. Convert all frequently connected subgraphs g found in the form of adjacency lists into corresponding connection matrices according to the mapping relationship between labels and nodes, and let B∈R N×N Represents, through the subgraph connection matrix B∈R N×N The node and edge information in the brain network are respectively extracted from the connection matrices of the structural brain network and the functional brain network, and then the structure-function coupling coefficient is calculated using the Spearman correlation coefficient method according to the obtained structural subnetwork and functional subnetwork. These coupling coefficients reflect the degree of association between the structural brain network and the functional brain network; the steps for calculating the function-structure coupling coefficient in step S2 are as follows: S21. Assume that n subgraphs are obtained, represented as g1, g2, ..., g n , according to the mapping relationship between labels and nodes, it is converted into the corresponding connection matrix, represented by B∈R N×N It means that for each subgraph g i , i = 1, 2, ..., n, there is an adjacency matrix B i , i=1,2,...,n; S22, through the subgraph adjacency matrix B i , i=1,2,...,n, the node and edge information in the functional brain network A F ∈R N×N Extract the adjacency matrix B of the resulting subgraph i , i=1,2,...,n, the relevant functional network subset; S23, through the subgraph adjacency matrix B i , i = 1, 2, ..., n, the node and edge information in the structured brain network A S ∈R N×N Extract the adjacency matrix B of the resulting subgraph i , i=1,2,...,n, the relevant structural network subset; S24. The function-structure coupling coefficient is calculated for the extracted functional subnetwork and structural subnetwork using the Spearman correlation coefficient method. The formula is as follows: Where ρ is the structure-function coupling coefficient, which is used to measure the degree of association between the structural brain network and the functional brain network, d is the difference between the ranks of the structural brain network and the functional brain network, and n is the number of sample data points of the structural brain network and the functional brain network; S3. Through five-fold cross validation, the lasso model is used to perform feature selection on the calculated coupling coefficients. In this process, the obtained coupling coefficients are divided into five subsets, four of which are used to train the model, and the fifth subset is used to test the performance of the model. This process will be repeated multiple times, each time using a different subset as the test set to ensure a robust model evaluation. In each cross validation, the lasso model will evaluate the importance of each feature and select the feature with the greatest predictive power for the target attribute. Through this feature selection method, the coupling coefficients after attribute simplification are obtained. These simplified coupling coefficients can reflect the degree of association between structural and functional brain networks. S4. Use the coupling coefficient after attribute simplification as a feature and apply the random forest classification model for further analysis. Random forest can integrate the results of multiple decision trees to improve model performance. Through random forest, the coupling coefficients are classified and their accuracy in the diagnosis of mental illness is evaluated.

Images