Functional Network Topology Properties Analysis And Classification Research Under Multi-node Scale

The use of brain network templates to construct brain networks is a research focus in the field of brain network diseases.This method provides a more effective technique for the diagnosis of cranial nerve diseases.The study found that the difference in network node scale under different template definitions has a great effect on the structure of the network and its topological properties.However,the traditional single template construction network ignores the scale of the nodes and lacks the influence of different node scales on the trustworthiness of the topological properties.At the same time,how the difference in network node scale in the machine learning method affects the network structure and classification accuracy is still in studying.In order to solve these problems,based on the predecessors,this paper conducts an in-depth study on the influence of node scale on network topology properties,classification feature representation,selection strategy and classification accuracy.The following is the main work of this article:First,defining a multi-node template and construct a network.In this paper,the data set of the normal subjects published by New York University and the dataset of depression of Shanxi Medical University were used to divide the brain regions respectively,and five brain networks with different number of nodes were obtained.Secondly,exploring the effect of the number of different nodes on the topological properties.The local properties of brain networks of different node scales constructed by the data sets of the normal subjects exposed by New York University were extracted,and then the differences of topological properties were compared.Finally,the functional connections of the brain network and the reliability of the network topology properties were analyzed.Thirdly,exploring the impact of node scale on the selection and performance of classification features.Firstly,the local properties of brain networks with different node scales constructed by the depression data set are extracted,and then the statistical analysis is performed between the local properties by KS test.After judging discriminative features(significant difference properties)are selected as classification features.SVM is used to classify depression and normal people.Fourthly,the maximum correlation minimum redundancy method is used to analyze the redundancy between features,and the influence of the distance of the brain interval of different node scales on the redundancy is compared with the traditional statistically significant feature selection method at different node scales.Finally,the high-dimensional features in the experiment were tested for over-fitting problems.Comprehensive considerations show that with the increase of the number of spatial nodes,the reliability of the network structure is gradually improving,and the classification effect is also improving.The results show that the features obtained by different node scales are equivalent,that is,the template with a large number of nodes can't provide more effective features,but it can provide more effective features,which will lead to the improvement of classification accuracy.At the same time,the template with a large number of nodes is closer due to the distance between brain regions,and the degree of redundancy between features is also enhanced.It is also found that the traditional statistically significant feature selection method can be performed at different node scales,but the analysis results show that the traditional 0.05 threshold setting is too strict.The research was supported by the National Natural Science Foundation of China(61672374,61741212,61876124,61873178),Natural Science Foundation of Shanxi Province(201601D021073,201101D121135),Scientific and Technological Innovation Programs of Higher Education Institutions in Shanxi(2016139)and Key Research and Development(R&D)Projects of Shanxi Province(201803D31043).The research focus of this paper is on multi-node template definition and network construction.On this basis,the influence of network node scale on network topology properties and classification feature selection and performance is further discussed.In this paper,when applying the brain network topology properties to the machine learning method in the future,it provides a certain reference value for constructing the appropriate template for the network.