Brain Network Similarity Analysis Based On The New Graph Core

With the development of brain imaging and machine learning,brain network has gradually become an important way of brain research.After the modeling of brain network,machine learning technology is used to analyze brain network in various ways.The structure and function of the brain are complex and the signal processing is complex and diversified,which contains a lot of tag information and structured information.In the analysis of brain network using machine learning method,how to use this information to be recognized by the machine and used as feature vector in a reasonable way has become a focus of brain network research.In this paper,the similarity discrimination of brain network based on novel graph kernel is studied,and the following work is mainly completed.First,the application of existing graph nuclei in brain network classification is analyzed experimentally,and the deficiencies of these methods are proposed.Secondly,a weighted method is proposed to fuse the topological properties of brain networks and describe the structural information of brain networks in a more comprehensive way.This method in the real network of mild cognitive impairment in the brain discriminant classification accuracy of the data set is 76.00%,relative to the individual characteristics of the classification method has greatly improved,compared with the single extraction,characteristic path length,clustering coefficient of nodes as characteristic vector method,and the original data of 13.9%,15.7%,15.4%and 13.9%respectively.Second,a new kind of graph kernel is proposed.The main idea of this graph kernel is:take each node of brain network as the center,build the sub network of brain network together with several nodes that have the highest correlation with this node,and use discriminant function to analyze the similarity of each sub network,so as to judge the similarity of brain network The experimental results showed that the average accuracy was stable at 0.837,and the classification accuracy was higher than the previous method,which was 0.145 higher than the RBF core,0.172 higher than the Linear kernel,0.192 higher than the Poly kernel,and 0.225 higher than the Sigmoid kernel.