Classification Research Based On Local Difference Minimum Spanning Tree Brain Network

The analysis and research of complex brain networks has been a hot topic in the field of neuropsychiatry in recent years.As a specific application of complex network theory in neurocognitive science,complex brain network plays an important role in understanding the pathogenesis of neuropsychiatric diseases.Resting-state functional magnetic resonance imaging(RS-fMRI)is one of the most important methods for studying brain diseases.Minimum spanning tree(MST)is one of the most widely used graph theory algorithms.As a new and effective research method,MST is active in the research of neuropsychiatric diseases.The method of minimum spanning tree can simplify the edges of the network and get the tree with the least total weight in all spanning trees under the condition of ensuring network connectivity.The unbiased method greatly simplifies the network structure and retains the core framework of the network.The calculation results will not be biased because of the influence of network size,average degree or density.It avoids the influence of network sparsity and other parameters on the network structure.At the same time,it ensures the interpretability of the network in neurology.It has been widely used in the field of neuroimaging research.Though many surprising achievements have been completed in this domain,there are still some problems need to be solved urgently.Previous studies have found that the traditional feature extraction method of minimum spanning tree uses local quantifiable indexes to classify brain diseases,ignoring the important role of low-weight connections and clusters in processing information in the brain network,resulting in loss of useful information in the network.Compared with other network features,its classification accuracy is significantly lower,and the validity of features and classification accuracy will be reduced.On this basis,this paper hopes to find a comprehensive method which can not only maximize the representation of inter-group differences,but also provide more effective classification features for service classification research.In order to solve these problems,a new method of feature extraction based on local difference network is proposed in this paper.The main innovative work of this paper is as follows:Firstly,network-based statistic(NBS)is used to identify connections with distinct functional connectivity between depression group and control group to connect the brain regions involved as the first step to construct a local difference network.NBS is an effective method to deal with non-parametric statistics of multiple comparison problems on graphs for statistical analysis of large networks.Many studies have used this method to identify connections related to experimental effects or inter-group differences and networks containing human connectors.Secondly,the minimum spanning tree functional connection network is constructed for each local difference subnet.Minimum spanning tree(MST)network maintains as high a connection strength as possible while guaranteeing network connection.This unbiased method can ensure that all nodes are connected to the network,the edges of the network are maximized,the network structure is greatly simplified,and the core framework of the network can still be retained.In this paper,we construct a local difference network with each brain region and its differentially connected brain regions as nodes.On this basis,we construct a minimum spanning tree functional connection network for each subnet,and make further analysis and research.Finally,the classification is studied based on the local difference minimum spanning tree brain network.As a complex network,the brain needs to be quantified in many ways.By calculating global and local indices on each local difference minimum spanning tree network,more effective features for classification can be obtained and the accuracy of classification can be improved.The results show that compared with the traditional classification method using minimum spanning tree to construct brain network,the proposed method can provide more effective features,which will significantly improve the classification accuracy.This paper provides an important reference for network construction and feature extraction in the future analysis of brain network topological attributes and application in machine learning.It also provides some help for medical assistant diagnosis and brain science,especially for the study of brain diseases.This paper has been supported by the National Natural Science Foundation Project(61876124),and as the main part of the "Analysis and Modeling of Highorder Complex Spatiotemporal Effects of Resting Functional Brain Networks".The research work has also been supported by the Shanxi Provincial Science and Technology Department Applied Basic Research Project(201801D121135),Scientific and Technological Innovation Programs of Higher Education Institutions in Shanxi(2016139),and Key Research and Development(R&D)Projects of Shanxi Province(201803D31043),and the Ministry of Education's Sell Network Next Generation Internet Technology Innovation Project(NGII20170712).The key point of this paper is to do some research on the local minimum difference spanning tree network and to discover the changes of brain network in patients with brain diseases,hoping to obtain markers that can help early diagnosis of brain diseases.This topic is very popular and important both at home and abroad.