Construction And Analysis Of Brain Function Hypernetwork Based On Overlapping Group Lasso And Fused Lasso

Brain network analysis has been widely used in research in the field of neuroimaging.Traditional functional connection networks are mostly based on pairwise correlations to construct second-order relationships between brain regions.In order to effectively construct higher-order interactions between brain regions,a brain function construction method based on hypernetwork is proposed.A hypernetwork is a complex network based on the concept of a hypergraph.The hyperedges in the hypernetwork can be used to represent the interaction between multiple brain regions.The hypernetwork is constructed by a sparse linear regression model based on the resting state functional magnetic resonance imaging time series.Most of the existing sparse linear regression models are solved by the lasso method.Although the lasso method is widely used,it also has several disadvantages.First,when the order information about the characteristic variables is known in advance,this method obviously ignores the order relationship between the variables;secondly,this method cannot effectively explain the grouping effect information.In order to effectively solve the influence of grouping effect,related studies have introduced the group lasso method to construct the hypernetwork.Although the group effect problem is effectively solved,it is only applicable to disjoint groups,and it is still not suitable for overlapping groups.Be applicable.In order to analyze whether the above two deficiencies of the existing methods will adversely affect the structure of the brain function hypernetwork,this article hopes to find a mathematical modeling method to solve the two deficiencies,and try to apply it to the construction of the brain function hypernetwork model.The new model is applied to the classification and recognition of patients with depression to explore whether there will be better classification performance under the new model,and under what circumstances there will be the best classification performance.In this paper,two new mathematical modeling methods are applied to the construction of the brain function hypernetwork model,and optimization experiments are carried out under each model.The specific innovation work in this article includes the following three points:First,based on the application of the lasso method,when the order information about the feature variables is known a priori,the order relationship between the variables is ignored.This article proposes to apply the fusion lasso method to the brain function hypernetwork.In the construction of the model,the method of fusion lasso realizes the automatic group effect by penalizing the difference of the regression coefficients,so that the solution has the characteristics of piecewise constant,and can simply solve the problem of the order information of the adjacent variables.Second,based on the methods that have been applied to the construction of brain function hypernetwork,the problem of overlap between brain regions is not considered.This article proposes to apply the method of overlapping groups lasso to the construction of brain function hypernetwork.The structure of the brain is complex.Not only does it involve the cooperation of multiple brain regions when completing a specific function,but also from another perspective,the same brain region can act on a variety of specific functions,and at the same time,The same brain area may be involved in the realization of a variety of specific functions,which creates the problem of overlap between groups in the realization of specific functions in the brain area.The overlapping group lasso method takes into account the possible overlapping area problems between groups from the mathematical modeling,and applying it to the construction of the brain function hypernetwork model can effectively solve the overlap problem between brain area groups.Third,after completing the construction of the brain function hypernetwork model,to apply it to the classification and recognition of related brain diseases,it is necessary to find the features that have significant differences in the model constructed between the patient and the normal person.This requires The process of feature extraction and feature selection.In the methods that have been applied to the construction of brain function super-networks,only the three most commonly used clustering coefficient attributes from different angles are extracted in the process of feature extraction,and the attributes are relatively single.Under the overlapping group lasso brain function hypernetwork construction method,in the feature extraction stage,this study also extracted three other attributes,namely the shortest path,the number of super edges(edge degree)of the node,and the medium centrality of the node,and Clustering coefficient attributes are used for multi-feature fusion analysis together.The three newly added attributes are also commonly used features in the field of hypernetwork model research.The fusion analysis with the original features will have a significant impact on the classification effect of the final classification model.The research carried out experimental design based on the above three innovations,and compared with the original commonly used lasso method.The results showed that the hypernetwork constructed under the fused lasso method and the overlapping group lasso method is basically similar in structure to the hypernetwork constructed under the lasso method.When comparing and analyzing the clustering coefficient attributes of the models constructed by the three methods under the same subjects,the results show that the overlapping group lasso is highly correlated with the results under the lasso method,and the results under the fusion lasso and lasso method have Certain relevance.When analyzing the difference of the clustering coefficient attributes under the three methods,the results show that the results of the fusion lasso method and the lasso method are significantly different,and the results of the overlapping group lasso method and the lasso method are slightly different but not significantly different.In the comparison of the accuracy of the models constructed under the three methods using only the clustering coefficient features for the classification and recognition of depression,the results show that compared with the lasso-based hypernetwork construction method,the hypernetwork based on the fusion lasso The classification effect under the construction method has decreased,and the classification effect under the hypernetwork construction method based on the overlapping group lasso has been improved.In the experiment of fusion analysis by adding the shortest path,node degree,and medium centrality attributes and clustering coefficient attributes for feature selection after constructing the brain function hypernetwork model using the overlapping group lasso method,the results show that the results show that the combination of the shortest path,node degree,and medium centrality attributes is combined with clustering coefficient attributes.The comparative analysis of the effect data under the class coefficient can be obtained.Under the overlapping group lasso method,the classification performance after multi-feature fusion is significantly improved compared with the classification performance under the cluster coefficient alone,and the accuracy rate reaches 87.87%.This article is supported by a number of national,provincial and ministerial funds and projects.The key of this article is to improve some of the shortcomings of the original method at the level of the brain function hypernetwork construction method and at the feature extraction stage,so as to study the more different characteristics of the brain network structure of depression patients than normal people.Hope to provide some help for the early diagnosis and treatment of depression.This subject has received much attention from scholars at home and abroad.