Classification system and method for mental diseases based on brain mapping data analysis

Abstract

This application provides a classification system and method for mental illnesses based on brain mapping data analysis. The system acquires brain network data and abnormal brain region feature data of the target subject through a data acquisition unit. A feature extraction unit extracts connectivity, topological, and network features of the target subject based on the brain network data, thereby determining the target subject's input features. A disease classification unit inputs the target subject's input features into a trained mental illness classification model to obtain and output the mental illness classification result for the target subject. The mental illness is schizophrenia, which includes multiple subtypes: paranoid, disorganized, undifferentiated, and residual. The classification result includes the probability of the target subject belonging to each subtype of schizophrenia. This solution can achieve accurate and reliable subtype classification of schizophrenia, providing support for precision treatment of schizophrenia patients.

Description

Technical Field

[0001] This application relates to the fields of medical image data processing and artificial intelligence technology, and more specifically, to a classification system and method for mental illnesses based on brain mapping data analysis. Background Technology

[0002] Schizophrenia is a complex and highly heterogeneous severe mental disorder with diverse clinical manifestations and a prolonged course, placing a heavy burden on patients, their families, and society. To facilitate clinical diagnosis, treatment, and research, the medical community has classified schizophrenia into subtypes based on the significant characteristics of symptom clusters. The main subtypes include paranoid, disorganized (also known as hebephrenic), undifferentiated, and residual. The paranoid type is characterized by relatively stable delusions and hallucinations, especially persecutory or grandiose delusions; the patient's thought patterns and emotional responses may remain relatively intact in the early stages of the illness. The disorganized type is characterized by significant disorganized thinking, disordered behavior, and emotional incoordination; impairment of social functioning is often more severe and appears earlier. The undifferentiated type is suitable for patients who meet the diagnostic criteria for schizophrenia but whose symptom patterns do not conform to the above specific subtypes. The residual type refers to the later stages of the illness where positive symptoms (such as hallucinations and delusions) lessen, but negative symptoms (such as emotional flatness and diminished volition) or mild positive symptoms remain. This subtype classification has long existed in authoritative standards such as the International Classification of Diseases (ICD-10) and the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV), with the original intention of providing a framework for clinical communication, prognosis, and treatment selection.

[0003] However, the current subtyping of schizophrenia faces challenges in clinical practice and research. Although the American Psychiatric Association has abandoned the traditional subtyping classification since DSM-V due to the difficulty of subtyping and some overlapping symptoms, and has instead shifted to symptom-based subtyping labels to assess the severity of symptoms from a symptom perspective, the subtyping concept is still widely used in many clinical settings and studies.

[0004] The accurate identification of schizophrenia subtypes has long faced multiple challenges, primarily due to the complexity and heterogeneity of the disease itself, as well as the limitations of current diagnostic methods. The lack of objective, quantitative biological markers is a fundamental challenge in subtype identification. Currently, the diagnosis of schizophrenia and its subtypes relies almost entirely on clinical interviews and subjective symptom assessments, lacking objective indicators.

[0005] With the development of neuroimaging techniques (such as fMRI and DTI), it has been discovered that schizophrenia patients exhibit widespread abnormalities in brain connectivity, changes in brain volume, and other brain abnormalities, providing new research directions for the subtype classification of schizophrenia. Currently, the mainstream approach is based on symptom classification (positive symptoms: such as delusions, hallucinations, and thought disorder; negative symptoms: such as emotional flattening, diminished volition, poverty of speech, anhedonia, and social withdrawal; and cognitive symptoms: such as impairments in attention, memory, and executive function). Supplementing this with objective subtype classification (paranoid, disorganized, undifferentiated, and residual types) holds promise for achieving precision medicine for different types of schizophrenia patients.

[0006] Therefore, developing a classification system for schizophrenia subtypes to achieve objective classification of schizophrenia subtypes and provide assistance for precise clinical treatment is a technical problem that needs to be solved in this field. Summary of the Invention

[0007] The purpose of this application is to provide a classification system and method for mental illnesses based on brain atlas data analysis. By utilizing brain network data and abnormal brain region feature data, targeted and diversified feature extraction is performed, and a classification model for mental illnesses is constructed based on support vector machines. This facilitates accurate and reliable classification of schizophrenia subtypes in scenarios with small samples and high-dimensional features, thus providing support for the precision treatment of schizophrenia patients.

[0008] To achieve the above objectives, the embodiments of this application are implemented in the following manner:

[0009] This application provides a mental illness classification system based on brain atlas data analysis, comprising: a data acquisition unit for acquiring brain network data and abnormal brain region feature data of a target object, wherein the brain network data includes m brain region nodes and the connection strength between brain region nodes, the connection strength between the m brain region nodes is represented by an m×m functional connectivity matrix, and the abnormal brain region feature data is the abnormality index of m brain regions; a feature extraction unit for extracting connection features, topological features, and network features of the target object based on the brain network data, and determining the input features of the target object based on the connection features, topological features, network features, and abnormal brain region feature data of the target object; and a disease classification unit for inputting the input features of the target object into a trained mental illness classification model to obtain and output the mental illness classification result of the target object, wherein the mental illness is schizophrenia, which includes multiple subtypes: paranoid, disorganized, undifferentiated, and residual, and the mental illness classification result includes the probability of the target object in each subtype of schizophrenia.

[0010] Furthermore, the feature extraction unit is specifically used for: extracting features from the functional connectivity matrix to determine the connectivity features corresponding to each brain region node, wherein the connectivity features corresponding to each brain region node are 1×m row vectors; for each brain region node: based on the connectivity features corresponding to the brain region node, calculating the weighted node degree and node centrality of the brain region node to generate the topological features corresponding to the brain region node; calculating the node curvature index between brain region nodes based on the weighted node degree as the network features corresponding to the brain region node; fusing the connectivity features, topological features, network features, and brain region abnormality features corresponding to the brain region node to obtain the fused features of each brain region node; and determining the input features of the target object based on the fused features of each brain region node.

[0011] Furthermore, the feature extraction unit is specifically used to: for each brain region node, determine the length of the target path with the highest propagation efficiency from that brain region node to other brain region nodes; and based on the target path length from each brain region node to other brain region nodes, determine the weight between every two brain region nodes. The weighted node degree of brain region nodes is calculated using the following method:

[0012] ,

[0013] in, brain region nodes The weighted node degree, brain region nodes With brain region nodes The strength of the connection between them brain region nodes With brain region nodes The weights between them.

[0014] Furthermore, the feature extraction unit is specifically used to calculate the propagation efficiency of brain region nodes in the following manner:

[0015] ,

[0016] in, brain region nodes to brain region nodes The overall dissemination efficiency brain region nodes to brain region nodes The set of propagation paths, the propagation paths in the set of propagation paths transmission path , brain region nodes , brain region nodes , transmission path The number of nodes is The number of sides is , Represents the set of propagation paths The maximum calculated value in Representing brain region nodes to brain region nodes The strength of the connection between brain region nodes to brain region nodes There is no effective transmission path, making =0; Determine brain region nodes to brain region nodes propagation efficiency Corresponding number of edges brain region nodes to brain region nodes Target path length , among which, if =0, confirm If the propagation efficiency If there are multiple propagation paths with the maximum computed value, take the number of edges of the shortest path. brain region nodes to brain region nodes Target path length .

[0017] Furthermore, the feature extraction unit is specifically used to calculate brain region nodes using the following methods. With brain region nodes Weights between :

[0018] ,

[0019] in, brain region nodes With brain region nodes The weights between them.

[0020] Furthermore, the feature extraction unit is specifically used to calculate the node centrality of brain region nodes in the following manner:

[0021] ,

[0022] in, brain region nodes The node centrality, brain region nodes The set of neighboring nodes, For the set of neighbor nodes The number of neighboring nodes in the data. brain region nodes With brain region nodes The strength of the connection between them.

[0023] Furthermore, the feature extraction unit is specifically used to calculate the node curvature index of brain region nodes in the following manner:

[0024] ,

[0025] in, brain region nodes With brain region nodes The curvature index between nodes, brain region nodes With brain region nodes The strength of the connection between them brain region nodes The weighted node degree, brain region nodes The weighted degree of nodes.

[0026] Furthermore, the feature extraction unit is specifically used for: for each brain region node: to extract the brain region node... Corresponding connectivity features, weighted node degree, node centrality, node curvature index, and brain region nodes Brain region abnormality feature data were spliced ​​together to obtain The row vectors, as brain region nodes The corresponding fusion features; the fusion features corresponding to each brain region node are arranged according to the brain region number to form a fusion feature matrix of m brain regions, which is used as the input features of the target object. The matrix dimension is... .

[0027] Furthermore, the classification model for mental illnesses uses a support vector machine model, which is obtained after training with a sample set.

[0028] This application embodiment also provides a method for classifying mental illnesses based on brain atlas data analysis, applicable to any of the aforementioned mental illness classification systems based on brain atlas data analysis, including: using a data acquisition unit to perform step S10: acquiring brain network data and brain region abnormality feature data of the target object, wherein the brain network data includes m brain region nodes and the connection strength between brain region nodes, the connection strength between the m brain region nodes is represented by an m×m functional connectivity matrix, and the brain region abnormality feature data is the abnormality index of m brain regions; using a feature extraction unit to perform step S20: extracting the target object based on the brain network data. The target object's connectivity features, topological features, and network features are used to determine the target object's input features based on the target object's connectivity features, topological features, network features, and brain region abnormality features. Step S30 is executed using the disease classification unit: the target object's input features are input into the trained mental illness classification model to obtain and output the target object's mental illness classification result. The mental illness is schizophrenia, which includes multiple subtypes: paranoid, disorganized, undifferentiated, and residual. The mental illness classification result includes the probability of the target object in each subtype of schizophrenia.

[0029] Beneficial Effects: This approach utilizes brain network data and abnormal brain region feature data of the target subjects to perform multi-level and multi-dimensional feature extraction. Compared to traditional brain imaging-based classification studies (which typically focus only on simple attributes of the functional connectivity matrix or single brain region activity indicators), this approach constructs a rich feature space through a four-level extraction and fusion of "connectivity features, topological features, network features, and abnormal features." This allows for a comprehensive and three-dimensional characterization of the complex brain network disorders that may exist behind different subtypes of schizophrenia. Starting from the most basic connectivity features, the original connection strength information (1×m row vectors) of each brain region node with all other brain regions is retained. Further mining of the topological attributes contained in the connectivity data is conducted. An innovative weighted node degree and node centrality calculation scheme is used, which is not a simple binary connection calculation. Instead, the length of the target path with the highest propagation efficiency between nodes is used as the basis for weighting. This considers not only the strength of direct connections but also the global efficiency of information exchange between nodes through multi-hop paths. Therefore, even if a node has a weak direct connection with a distant node, as long as an efficient indirect path exists, its contribution will be reasonably included in the importance assessment of that node. Compared to traditional node degree, it more accurately reflects the true state of information integration in brain networks. Node centrality, on the other hand, focuses on assessing the influence of a node within its local neighborhood environment, reflecting the modularity of brain networks. Furthermore, the introduction of network features, specifically the node curvature index, reflects the degree of local curvature in the brain network, sensitively capturing subtle changes in network structure. For example, a weak connection between two nodes with high weighted node degrees indicates insufficient network integration, while a strong connection between two nodes with low weighted node degrees suggests abnormal functional compensation or pathological overconnection. This feature can reveal subtle network structural variations that traditional topological indicators cannot capture, providing potential feature markers for distinguishing clinically diverse subtypes (such as the relative preservation in paranoid schizophrenia and the widespread dysfunction in disorganized schizophrenia). Moreover, this scheme fuses the calculated weighted node degree, node centrality, node curvature index, and other features with brain region abnormality feature data (reflecting the probability of abnormality in each brain region). This multimodal data fusion effectively enhances the robustness and interpretability of the features. By concatenating each brain region node into a 1×(2m+3) row vector and finally forming an m×(2m+3) fusion feature matrix, the scheme constructs a comprehensive input that includes both global network attributes and retains the unique information of each brain region, providing a solid data foundation for the subsequent high-precision classification of various subtypes of schizophrenia.

[0030] Considering the limited sample size (difficulty in obtaining tens of thousands of samples) and high feature dimensionality of each subtype of schizophrenia, this approach uses a support vector machine (SVM) as the classifier. This avoids the overfitting problem of complex deep learning models and is more suitable for classifying schizophrenia subtypes. The final output, determining the probability of a target individual belonging to each subtype of schizophrenia, is more valuable for clinical decision support, assisting doctors in making comprehensive judgments based on clinical interviews and facilitating precision medicine.

[0031] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 A framework diagram of a mental illness classification system based on brain atlas data analysis provided in this application embodiment.

[0034] Figure 2 This is a flowchart of a classification method for mental illnesses based on brain mapping data analysis.

[0035] Figure 3 A schematic diagram illustrating the visualization of brain region correlations.

[0036] Figure 4 This is a schematic diagram visualizing the strength of connectivity between brain regions.

[0037] Figure 5 This is a schematic diagram illustrating the visualization of brain region abnormality analysis.

[0038] Figure 6 A schematic diagram illustrating the visualization of abnormalities in brain regions. Detailed Implementation

[0039] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0040] In this embodiment, mental illness specifically refers to schizophrenia, which includes multiple subtypes: paranoid, disorganized, undifferentiated, and residual. There are other subtypes (such as catatonic), but these have less research significance (the symptoms are relatively clear and easy to distinguish; after a patient is diagnosed with schizophrenia, it can be determined whether they belong to the catatonic type. If they do, they are marked as catatonic, and there is no need to use this scheme for classification. If they do not belong to the catatonic type, this scheme is used for specific subtype classification). Furthermore, task-based data is required for assistance, so they are not within the scope of classification in this embodiment. The classification results of mental illnesses include the probability of the target object belonging to each subtype of schizophrenia.

[0041] Please see Figure 1 and Figure 2 , Figure 1 This is a framework diagram of a classification system for mental disorders based on brain mapping data analysis. Figure 2 This is a flowchart of a mental illness classification method based on brain atlas data analysis. In this embodiment, the mental illness classification system based on brain atlas data analysis may include a data acquisition unit, a feature extraction unit, and a disease classification unit. The mental illness classification method based on brain atlas data analysis includes steps S10, S20, and S30. The data acquisition unit performs step S10, the feature extraction unit performs step S20, and the disease classification unit performs step S30.

[0042] First, the data acquisition unit can run step S10.

[0043] Step S10: Obtain brain network data and brain region abnormality feature data of the target object. The brain network data includes m brain region nodes and the connection strength between brain region nodes. The connection strength between m brain region nodes is represented by an m×m functional connectivity matrix. The brain region abnormality feature data is the abnormality index of m brain regions.

[0044] In this embodiment, the data acquisition unit can acquire brain network data and brain region abnormal feature data of the target object. The brain network data includes m brain region nodes and the connection strength between brain region nodes. The connection strength between m brain region nodes is represented by an m×m functional connectivity matrix. The brain region abnormal feature data is the abnormal index of m brain regions.

[0045] The process of acquiring brain network data can be based on brain imaging data. A network graph generation method based on brain imaging data is used to generate a brain network graph of the target object, and then the brain network data of the target object (mainly the m brain region nodes and the connection strength between brain region nodes) is exported. In the process of generating the brain network graph of the target object, the patent application 1 published by our unit (application number: 202510585743.8, title: Network Graph Generation Method Based on Brain Imaging Data) is used. When calculating the time series data of each brain region, strategy 2 in the network graph generation method based on brain imaging data is used. Strategy 2 means: for each brain region, the time series of each voxel in this brain region is calculated based on the spatial distance weighted average, which is used as the time series data of this brain region. It should be noted that in the description of patent application 1, the calculation scheme of strategy 2 is specifically in its formula (3), the character explanation is not provided. Indicates brain regions voxels in With voxels The relative distance between them lacks information on voxels. Explanation of voxels brain region To avoid insufficient explanation of the cited prior art, this supplementary explanation is provided.

[0046] like Figure 3 and Figure 4 As shown, Figure 3 This is a schematic diagram visualizing the correlation between brain regions. Figure 4 This is a schematic diagram visualizing the connectivity strength between brain regions. The connectivity strength between brain region nodes is derived as an m×m functional connectivity matrix.

[0047] The abnormal brain region feature data is obtained by using our published patent application 2 (application number: 202510559840.X, title: A method for identifying abnormal regions in brain map data), which yields the abnormal probability of each brain region, denoted as a 1×m row vector. Each element in the row vector reveals the abnormal probability of the corresponding brain region.

[0048] like Figure 5 and Figure 6 As shown, Figure 5 This is a schematic diagram visualizing the analysis of abnormalities in brain regions. Figure 6 This is a schematic diagram for visualizing brain region abnormalities (because single-row or single-column charts are too long, multiple segments of a single-column chart are cropped and displayed in the same chart). Export the abnormality feature data of m brain region nodes as a 1×m row vector.

[0049] After obtaining the brain network data and abnormal brain region feature data of the target object, the feature extraction unit can run step S20.

[0050] Step S20: Extract the connectivity features, topological features, and network features of the target object based on brain network data, and determine the input features of the target object based on the connectivity features, topological features, network features, and abnormal brain region features of the target object.

[0051] In this embodiment, the feature extraction unit can extract features from the functional connectivity matrix to determine the connectivity features corresponding to each brain region node. The connectivity features corresponding to each brain region node are 1×m row vectors. Here, the connectivity features corresponding to the brain region node are the connectivity strengths between the brain region node and all brain region nodes (including its connectivity strength with itself, where its connectivity strength with itself is 1).

[0052] For example, in the brain network data of the target object, the m×m functional connectivity matrix is:

[0053] , (1)

[0054] in, For the first The correlation coefficient between the first brain region node and the first brain region node (also the first brain region node) (The numerical value of the connection strength between each brain region node and the first brain region node). For the first brain region and the second The correlation coefficient between the first and second brain regions (also the correlation coefficient between the first and second brain regions) (The numerical value of the correlation coefficient of the connection strength of individual brain regions).

[0055] So, brain region nodes The connection characteristics are:

[0056] , (2)

[0057] in, brain region nodes The connection characteristics, .

[0058] After extracting the connectivity features of each brain region node, for each brain region node: the feature extraction unit can calculate the weighted node degree and node centrality of the brain region node based on the connectivity features corresponding to the brain region node, and generate the topological features corresponding to the brain region node.

[0059] For example, for each brain region node: the feature extraction unit can determine the length of the target path with the highest propagation efficiency from the current brain region node to other brain region nodes.

[0060] First, for each brain region node: the feature extraction unit can identify the brain region node. to brain region nodes The set of propagation paths For the set of propagation paths propagation path , This allows us to determine the set of propagation paths between each brain region node and other brain region nodes.

[0061] Because excessively long propagation paths not only have limited reference value (the most efficient propagation paths are generally within 3 hops), but also lead to a significant increase in computation when calculating propagation paths from brain region nodes to other brain region nodes, the number of propagation hops is limited to no more than 3, i.e., the number of brain region nodes. to brain region nodes If each propagation path has no more than 3 edges, then the number of nodes containing brain regions is... and brain region nodes The number of brain region nodes in each propagation path will not exceed 4. This not only reduces the computational cost but also sufficiently reflects the propagation efficiency of brain region nodes, ensuring high reliability.

[0062] After determining the set of propagation paths between each brain region node and other brain region nodes, the feature extraction unit can calculate the propagation efficiency between brain region nodes in the following way:

[0063] , (3)

[0064] in, brain region nodes to brain region nodes The overall dissemination efficiency brain region nodes to brain region nodes The set of propagation paths, the propagation paths in the set of propagation paths transmission path , brain region nodes , brain region nodes , transmission path The number of nodes is The number of sides is , Represents the set of propagation paths The maximum calculated value in Representing brain region nodes to brain region nodes The strength of the connection between brain region nodes to brain region nodes There is no effective transmission path, making =0.

[0065] Based on this, the feature extraction unit can identify brain region nodes. to brain region nodes propagation efficiency Corresponding number of edges brain region nodes to brain region nodes Target path length , among which, if =0, confirm If the propagation efficiency If there are multiple propagation paths with the maximum computed value, take the number of edges of the shortest path. brain region nodes to brain region nodes Target path length .

[0066] Then, the feature extraction unit can determine the weights between every two brain region nodes based on the target path length from each brain region node to other brain region nodes. Specifically, the feature extraction unit calculates brain region nodes using the following method. With brain region nodes Weights between :

[0067] , (4)

[0068] in, brain region nodes With brain region nodes The weights between them.

[0069] Accordingly, the feature extraction unit can calculate the weighted node degree of brain region nodes in the following way:

[0070] , (5)

[0071] in, brain region nodes The weighted node degree, brain region nodes With brain region nodes The strength of the connection between them brain region nodes With brain region nodes The weights between them.

[0072] Starting from the most basic connectivity features, this study preserves the original connection strength information (1×m row vectors) between each brain region node and all other brain regions. It further mines the topological attributes inherent in the connectivity data, innovatively utilizing a weighted node degree and node centrality calculation scheme. Instead of simply binarizing connections, it calculates the length of the target path with the highest propagation efficiency between nodes as the basis for weighting. This considers not only the strength of direct connections but also the global efficiency of information exchange between nodes through multi-hop paths. Therefore, even if a node has a weak direct connection to a distant node, as long as an efficient indirect path exists, its contribution will be reasonably included in the importance assessment of that node. Compared to traditional node degree, this approach more accurately reflects the true state of information integration in the brain network.

[0073] Furthermore, the feature extraction unit can calculate the node centrality of brain region nodes in the following manner:

[0074] , (6)

[0075] in, brain region nodes The node centrality, brain region nodes The set of neighboring nodes, For the set of neighbor nodes The number of neighboring nodes in the data. brain region nodes With brain region nodes The strength of the connection between them.

[0076] Node centrality focuses on assessing the influence of a node in its local neighborhood environment and can reflect the modularity of brain networks.

[0077] Subsequently, the feature extraction unit can calculate the node curvature index between brain region nodes based on the weighted node degree, which serves as the network feature corresponding to the brain region node.

[0078] Specifically, the feature extraction unit can calculate the node curvature index of brain region nodes in the following way:

[0079] , (7)

[0080] in, brain region nodes With brain region nodes The curvature index between nodes, brain region nodes With brain region nodes The strength of the connection between them brain region nodes The weighted node degree, brain region nodes The weighted node degree. In this embodiment, brain region nodes... With brain region nodes When they are nodes in the same brain region, .

[0081] Introducing network features, specifically the node curvature index, can reflect the degree of curvature in local brain networks and sensitively capture subtle changes in network structure. For example, a weak connection between two highly weighted nodes indicates insufficient network integration, while a strong connection between two low-weighted nodes suggests abnormal functional compensation or pathological overconnection. This feature can reveal subtle network structural variations that traditional topological indicators cannot capture, providing potential markers for differentiating clinical subtypes (such as the relative preservation in paranoid schizophrenia versus the widespread dysfunction in disorganized schizophrenia).

[0082] After completing the calculation and processing of the connectivity features, topological features, and network features corresponding to brain region nodes, the feature extraction unit can fuse the connectivity features, topological features, network features, and abnormal brain region feature data corresponding to brain region nodes to obtain the fused features of each brain region node.

[0083] For example, brain region nodes The corresponding fusion features are:

[0084] , (8)

[0085] in, brain region nodes Corresponding fusion features, brain region nodes Corresponding abnormal feature values ​​of brain regions reveal brain region nodes The probability of belonging to an abnormal brain region. Therefore, the dimension of the fused feature is... .

[0086] After determining the fusion features corresponding to each brain region node, the feature extraction unit can determine the input features of the target object based on the fusion features of each brain region node. That is, the feature extraction unit can use the fusion features corresponding to each brain region node, arranged by brain region number, to form a fusion feature matrix of m brain regions as the input features of the target object, with a matrix dimension of... .

[0087] For example, the fusion feature matrix of m brain regions is represented as:

[0088] (9)

[0089] in, This is the fusion feature matrix of m brain regions, which represents the input features of the target object.

[0090] In this embodiment, the disease classification unit can perform step S30.

[0091] Step S30: Input the input features of the target object into the trained mental illness classification model to obtain and output the mental illness classification result of the target object. The mental illness is schizophrenia, which includes multiple subtypes: paranoid, disordered, undifferentiated, and residual. The mental illness classification result includes the probability of the target object in each subtype of schizophrenia.

[0092] In this embodiment, the disease classification unit can input the input features of the target object into a trained mental illness classification model to obtain and output the mental illness classification result of the target object. The mental illness is schizophrenia, which includes multiple subtypes: paranoid, disordered, undifferentiated, and residual. The mental illness classification result includes the probability of the target object in each subtype of schizophrenia.

[0093] In this embodiment, considering the limited sample size (difficulty in obtaining tens of thousands of samples) and high feature dimensionality of each subtype of schizophrenia, this solution adopts support vector machine as the classifier to avoid the overfitting problem of complex deep learning models, which is more suitable for the classification scenario of each subtype of schizophrenia.

[0094] First, a sample set needs to be constructed, collecting sample data carrying sample labels (specific subtypes of schizophrenia), and adding sample data of some normal subjects (the sample data of normal subjects corresponds to the sample label of health). Then, each sample data is processed by this system to obtain the corresponding input features, associate the corresponding sample labels, form a sample set, and then divide the sample set into training set, validation set and test set according to 7:2:1.

[0095] Since a support vector machine model is chosen, it is necessary to process the input features ( The matrix is ​​flattened into a one-dimensional vector and used as input to the support vector machine (SVM) model (standardization can also be performed, such as Z-score standardization, to make the mean of each feature 0 and the standard deviation 1), and the sample labels are converted into a form that the SVM model can process. A radial basis function (RBF) is set as the kernel function, and a grid search strategy is used to optimize the key hyperparameters of the SVM model (such as the penalty coefficient C and the parameter γ of the RBF kernel function) on the training set through cross-validation. The number of folds in the cross-validation can be set to the commonly used 5-fold or 10-fold to balance computational cost and performance. After the hyperparameters are optimized, the final SVM multi-classification model is trained on the entire training sample set using the optimal hyperparameter configuration to obtain a mental illness classification model. The model is evaluated by accuracy, precision, recall, and F1 score. Finally, the trained mental illness classification model is deployed on a server for the system's disease classification unit to run.

[0096] Accordingly, the disease classification unit can flatten the input features of the target object after obtaining them, and then input them into a trained mental illness classification model to obtain and output the classification result of the target object in mental illness. The output result is in the form of the probability of the target object in each subtype of schizophrenia (paranoid, disorganized, undifferentiated, and residual) and healthy. The final output of the probability of the target object in each subtype of schizophrenia is more valuable for clinical decision support, assisting doctors in making comprehensive judgments in conjunction with clinical interviews, and facilitating the realization of precision medicine.

[0097] In summary, this application provides a classification system and method for mental illnesses based on brain atlas data analysis. This approach utilizes brain network data and abnormal brain region feature data of the target subjects to perform multi-level and multi-dimensional feature extraction. Compared to traditional brain imaging-based classification studies (which typically focus only on simple attributes of the functional connectivity matrix or single brain region activity indicators), this approach constructs a rich feature space through a four-level extraction and fusion of "connectivity features, topological features, network features, and abnormal features." This allows for a comprehensive and three-dimensional characterization of the complex brain network disorders that may exist behind different subtypes of schizophrenia. Starting from the most basic connectivity features, the original connection strength information (1×m row vectors) of each brain region node with all other brain regions is retained. Further mining of the topological attributes contained in the connectivity data is conducted. An innovative weighted node degree and node centrality calculation scheme is used, which is not a simple binary connection calculation. Instead, the target path length with the highest propagation efficiency between nodes is calculated as the basis for weighting. This considers not only the strength of direct connections but also the global efficiency of information exchange between nodes through multi-hop paths. Therefore, even if a node has a weak direct connection to a distant node, its contribution will still be reasonably included in the importance assessment of that node as long as an efficient indirect path exists. Compared to traditional node degree, this better reflects the true state of information integration in the brain network. Node centrality, on the other hand, focuses on assessing the influence of a node in its local neighborhood environment, reflecting the modularity of the brain network. Furthermore, the introduction of network features, namely the node curvature index, can reflect the degree of curvature in the local brain network, sensitively capturing subtle changes in network structure. For example, a weak connection between two nodes with high weighted node degrees indicates insufficient network integration, while a strong connection between two nodes with low weighted node degrees suggests abnormal functional compensation or pathological overconnection. This feature can reveal subtle network structural variations that traditional topological indicators cannot capture, providing potential feature markers for distinguishing subtypes with different clinical manifestations (such as the relative preservation in paranoid schizophrenia and the widespread dysfunction in disorganized schizophrenia). Finally, this scheme fuses the calculated weighted node degree, node centrality, node curvature index, and other features with brain region abnormality feature data (reflecting the probability of abnormality in each brain region). This multimodal data fusion effectively enhances the robustness and interpretability of features. By concatenating each brain region node into a 1×(2m+3) row vector and ultimately forming an m×(2m+3) fusion feature matrix, the scheme constructs a comprehensive input that includes both global network attributes and retains the unique information of each brain region, providing a solid data foundation for the subsequent high-precision classification of various subtypes of schizophrenia.

[0098] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A mental illness classification system based on brain atlas data analysis, characterized by, include: The data acquisition unit is used to acquire brain network data and brain region abnormal feature data of the target object. The brain network data includes m brain region nodes and the connection strength between the brain region nodes. The connection strength between the m brain region nodes is represented by an m×m functional connectivity matrix. The brain region abnormal feature data is the abnormal index of m brain regions. The feature extraction unit is used to extract the connectivity features, topological features, and network features of the target object based on brain network data, and to determine the input features of the target object based on the connectivity features, topological features, network features, and abnormal brain region features of the target object. The disease classification unit is used to input the input features of the target object into a trained mental illness classification model, obtain the mental illness classification result of the target object and output it. The mental illness is schizophrenia, which includes multiple subtypes: paranoid, disorganized, undifferentiated and residual. The mental illness classification result includes the probability of the target object in each subtype of schizophrenia.

2. The mental illness classification system based on brain atlas data analysis according to claim 1, characterized in that, The feature extraction unit is specifically used for: Feature extraction is performed on the functional connectivity matrix to determine the connectivity features corresponding to each brain region node. The connectivity features corresponding to each brain region node are 1×m row vectors. For each brain region node: Based on the connection features corresponding to the brain region node, calculate the weighted node degree and node centrality of the brain region node, and generate the topological features corresponding to the brain region node; The node curvature index between brain region nodes is calculated based on weighted node degree and used as the network feature corresponding to the brain region node. The connectivity features, topological features, network features, and abnormal features of brain region nodes are fused to obtain the fused features of each brain region node; based on the fused features of each brain region node, the input features of the target object are determined. 3.The mental illness classification system based on brain atlas data analysis of claim 2, wherein, The feature extraction unit is specifically used for: For each brain region node, determine the length of the target path with the highest propagation efficiency from that brain region node to other brain region nodes; determine the weight between each two brain region nodes based on the target path length from each brain region node to other brain region nodes ; The weighted node degree of brain region nodes is calculated using the following method: ， wherein, is the weighted node degree of a brain region node , is the connection strength between a brain region node and a brain region node , is the weight between a brain region node and a brain region node .

4. The mental illness classification system based on brain atlas data analysis according to claim 3, characterized in that, The feature extraction unit is specifically used for: The propagation efficiency of brain region nodes is calculated using the following method: ， in, brain region nodes to brain region nodes The overall dissemination efficiency brain region nodes to brain region nodes The set of propagation paths, the propagation paths in the set of propagation paths transmission path , brain region nodes , brain region nodes , transmission path The number of nodes is The number of sides is , Represents the set of propagation paths The maximum calculated value in Representing brain region nodes to brain region nodes The strength of the connection between brain region nodes to brain region nodes There is no effective transmission path, making =0; Identify brain region nodes to brain region nodes propagation efficiency Corresponding number of edges brain region nodes to brain region nodes Target path length , among which, if =0, confirm If the propagation efficiency If there are multiple propagation paths with the maximum computed value, take the number of edges of the shortest path. brain region nodes to brain region nodes Target path length .

5. The mental illness classification system based on brain mapping data analysis according to claim 4, characterized in that, The feature extraction unit is specifically used for: Brain region nodes are calculated using the following method. With brain region nodes Weights between : ， in, brain region nodes With brain region nodes The weights between them.

6. The mental illness classification system based on brain atlas data analysis according to claim 3, characterized in that, The feature extraction unit is specifically used for: The nodal centrality of brain region nodes is calculated using the following method: ， in, brain region nodes The node centrality, brain region nodes The set of neighboring nodes, For the set of neighbor nodes The number of neighboring nodes in the data. brain region nodes With brain region nodes The strength of the connection between them.

7. The mental illness classification system based on brain atlas data analysis according to claim 3, characterized in that, The feature extraction unit is specifically used for: The nodal curvature index of brain region nodes is calculated using the following method: ， in, brain region nodes With brain region nodes The curvature index between nodes, brain region nodes With brain region nodes The strength of the connection between them brain region nodes The weighted node degree, brain region nodes The weighted degree of nodes.

8. The mental illness classification system based on brain mapping data analysis according to claim 3, characterized in that, The feature extraction unit is specifically used for: For each brain region node: Brain nodes Corresponding connectivity features, weighted node degree, node centrality, node curvature index, and brain region nodes Brain region abnormality feature data were spliced ​​together to obtain The row vectors, as brain region nodes Corresponding fusion features; The fusion features corresponding to each brain region node are numbered according to the brain region, forming a fusion feature matrix of m brain regions. This matrix serves as the input feature for the target object, with a matrix dimension of [missing information]. .

9. The mental illness classification system based on brain atlas data analysis according to claim 1, characterized in that, The classification model for mental illnesses is obtained by using a support vector machine model and training it with a sample set.

10. A method for classifying mental illnesses based on brain atlas data analysis, characterized in that, A mental illness classification system based on brain mapping data analysis, applicable to any one of claims 1-8, comprising: Using the data acquisition unit to run step S10: acquire brain network data and brain region abnormal feature data of the target object. The brain network data includes m brain region nodes and the connection strength between brain region nodes. The connection strength between m brain region nodes is represented by an m×m functional connectivity matrix. The brain region abnormal feature data is the abnormal index of m brain regions. The feature extraction unit performs step S20: extracting the connection features, topological features, and network features of the target object based on brain network data; and determining the input features of the target object based on the connection features, topological features, network features, and abnormal brain region features of the target object. Step S30 is executed using the disease classification unit: The input features of the target object are input into the trained mental illness classification model to obtain and output the mental illness classification result of the target object. The mental illness is schizophrenia, which includes multiple subtypes: paranoid, disordered, undifferentiated, and residual. The mental illness classification result includes the probability of the target object in each subtype of schizophrenia.

Images