| The human brain is one of the most complex systems in nature.It realizes the exchange of information between different brain regions through neuron networks to complete cognitive activities.Studying the cognitive neural mechanism of the brain from the perspective of complex networks provides a new perspective for us to study the brain.Structural brain networks based on diffusion tensor imaging(DTI)can reveal various parameters of brain white matter fiber connections.For example,fractional anisotropy(FA)is more sensitive to the integrity of white matter structure,and the average Mean diffusivity(MD)MD measures the overall degree of diffusion of water molecules.Axial diffusivity(AD)and radial diffusivity(RD)correlate with axonal and myelin changes.A structural network with only one parameter cannot comprehensively reflect the brain white matter fiber connectivity information and cannot provide a comprehensive analysis.The brain has asymmetric left and right hemispheres in structure,and there is functional differentiation.The degree of lateralization is closely related to the cognitive function of the brain.However,previous studies have only focused on a single-parameter structural network,and the multi-parameter brain structural network of the brain and its laterality mechanism are still unclear.Calculating topological properties based on graph analysis methods can clearly understand the abnormal patterns of brain networks and explore the damage to brain structure caused by brain diseases.Therefore,this paper takes the disease of bipolar affective disorder as the research object,uses the embedded features extracted by the Node2 vec algorithm,splices the multi-parameter embedded features into a matrix,and then uses the cosine similarity algorithm to construct a multi-parameter fusion brain network.Lateralization analysis and statistical difference features are used to construct auxiliary diagnosis models to improve the accuracy of auxiliary diagnosis of bipolar affective disorder.The main innovative work and research results of this paper include:(1)This paper systematically analyzes four single-parameter structure brain networks based on DTI,FA,Lgi MD,Lgi AD and Lgi RD,and compares and analyzes their network topology properties.The results of the study found that compared with the control group,the FA network of patients with bipolar disorder(BD)was more sensitive to abnormalities in the frontal frame and limbic system of the brain and visual dysfunction,while Lgi MD,Lgi AD and Lgi RD 3 parameter networks.Increased sensitivity to brain emotional dysfunction.In addition,the integration of multiple single-parameter brain network analysis results found that the global integration and recovery functions of the BD brain were reduced,while the separation function was significantly increased,and the abnormal brain areas were mainly concentrated in the key areas of the frontal-cortical-striatal-thalamic circuit.The FA network was more sensitive to the abnormality of the frontal frame and limbic system and visual dysfunction,while the three parameter networks of Lgi MD,Lgi AD and Lgi RD were more sensitive to the emotional dysfunction of the brain.The results show that the brain network constructed based on multiple single parameters can more comprehensively detect changes in topological information,and provide complementary information for comprehensive analysis of changes in brain topology.(2)Based on the Node2 vec algorithm,this paper extracts the node embedding features of multiple single-parameter networks respectively,and then splices them into a node embedding feature matrix,reconstructs the fused brain network using cosine similarity,and calculates its topological properties and lateralization index.The results showed that compared with multiple single-parameter brain networks,the results of multi-parameter fusion brain network analysis also found that the global integration,restoration and dissociation dysfunction of the BD brain were more significant,and occurred in the parietal lobe,basal ganglia and temporal lobe.Significant changes,loss or even reversal of laterality in BD brain topology.In addition,network topology laterality also showed a significant correlation with clinical manic and depressive symptoms in BD.The research results show that compared with multiple single-parameter structural brain networks,the multi-parameter fusion network constructed in this paper can better reflect the more sensitive abnormal structural changes and laterality changes in the brain,and can be used as more effective biomarkers.(3)This paper uses a variety of machine learning algorithms to construct four classification models of SVM,random forest,logistic regression and decision tree for the topological properties and lateralization index features of multiple single-parameter networks and multi-parameter fusion networks,respectively.Classification effect.The results of the study found that the classification effect of using multi-parameter fusion network features was better than that of multiple single-parameter networks,and the accuracy of the SVM classifier was as high as 99.10%.In addition,after adding the lateralization feature to the model,the classification accuracy of the fusion network is significantly improved,reaching 99.23%.The research results show that fusion network features and lateralization features have certain clinical value in the application research of auxiliary diagnosis of BD. |
