Research On Uncertain Graph Mining And Feature Selection For Brain Network

At present,human beings have had extensive and profound knowledge of the natural world in which they live.In the process of human’s understanding of the world,subjects of various fields,such as physics,chemistry,and biology have been gradually formed.In these subjects,people have achieved remarkable achievements.And these achievements depend on a very complex system: the brain.Because of the complexity and particularity of it,people still know little about the mysteries of the brain.In recent years,with the development of science and technology and the improvement of research level,more and more researchers have begun to invest in the research of the brain.This field is also considered to have broad research prospects.Brain research is usually focused on the brain diseases,such as Major Depressive Disorder(MDD).In the past,the brain network method was used and constructed a certain graph of it,and frequent subgraph mining and discriminative feature selection were performed on the graph.However,because of the inherent uncertainty of the brain network connection structure.If the threshold method is used to convert it into a certain graph directly,whether it is a single threshold method or a multiple threshold method,information loss will be caused.If the brain network is constructed as an uncertain graph,the existing frequent subgraph mining algorithm for uncertain graphs is mainly designed for general graphs,ignoring the characteristics of the brain network,such as the uniqueness of the brain network nodes,which may affect the performance of brain network classification.In addition,the discriminative feature selection which is made on the certain graph’s frequent pattern is not applicable on the uncertain graph.There is no very effective method for the moment.In view of the above issues,this study builds an uncertain graph of brain network based on MRI image data,proposes a brain-network-based frequent subgraph mining algorithm for uncertain graphs,and proposes a new discriminative feature selection method.The main works of this article are as follows:First,the construction of the brain network.In order to fully compare the effects of the two network construction methods on the experimental results,constructs certain graph and uncertain graph of the brain network respectively.The certain graph network adopts the traditional binary method and uses sparsity to ensure that each network has the same number of edges.Uncertain graph network omits the binary process,but retains all non-negative edges in the fully connected network.Edge weights are calculated using Pearson correlation coefficients.Second,a frequent subgraph mining algorithm for brain networks is proposed.This algorithm is based on the uniqueness of node in the same brain network and the one-to-one correspondence between nodes in different brain networks,and draws on the idea of frequent subgraph mining in certain graphs and uncertain graphs.Experimental results show that this method effectively reduces the number and the support of frequent subgraph patterns.Third,a new brain-network-based discriminative feature selection method is proposed.This method uses the sum,mean,variance,skewness,kurtosis and other indicators of the probability in each positive and negative sample of each subgraph pattern.And generalizes the discriminative score function that can only be used in the certain graph to the uncertain graph.In order to evaluate the performance of the various methods,experiments were performed on each method under the experimental data sets,and the results were compared and analyzed.