Abnormal Pattern Mining Of AD Brain Network Based On MST

Alzheimer's disease(AD)is a major problem of public health.In recent years,numerous studies have used brain networks and graph theory to analyze abnormalities of brain function in AD and Mild Cognitive Impairment(MCI),and demonstrated that patients with MCI and AD have alteration in functional,abnormal topology in regional,and whole brain connectivity.The thresholds selection of the traditional complex network analysis method in the process of brain network construction will cause the network bottom layer to be incompletely connected,thus affecting the calculation and measurement of various network attributes,resulting in inconsistent research results of the brain network.In order to avoid the various influences brought by threshold selection,many studies have used minimum spanning tree(MST)to explore the topological changes of brain network.Although the MST itself can be used as a sparse representation of the whole network,it is somewhat implausible biologically,because the MST discards a large number of connections during the construction process and its edges do not form clusters or cliques.Therefore,it is significant important to seek a reasonable brain network analysis method to effectively explore the topological changes of AD brain and better assist the early diagnosis of AD.For the purpose of study the changes in brain network topology of MCI and AD,this paper analyzes the methodological problems in the traditional brain network research,uses the minimum spanning tree method to guarantee the full connectivity of the network,and proposes constructing a connected backbone network(CBN)by adding a certain connection to the minimum spanning tree according to a proportional threshold.This not only guarantees the full connectivity of the network but also preserves the small-world properties of the brain network,hoping to solve the problems in the traditional network analysis methods and valuable explore the methodological issues in brain network analysis.In addition,in order to explore the different manifestations of disease-related brain network topological changes in the overall network and the characteristics of nodes,this paper proposes hub disruption index to characterize the changes of different node attributes among the three groups of subjects.In the end,this paper sorted out the significant differences between the two groups in the minimum spanning tree and the connected backbone network,and used the SVM classification algorithm for classification studies.The contribution of the minimum spanning tree and connected backbone network characteristics to the research of brain network is illustrated by the classification effect.It is proved that the connected backbone network based on the minimum spanning tree can describe the real network structure better,capture the topological changes of the network,helps to unearth abnormal changes in the brain network of AD,and provides an important basis for early diagnosis of AD.The main accomplishments of this study are the following:(1)In response to the problem of the incomplete connectivity in the traditional network,this paper use the minimum spanning tree method to force all nodes of the network to be connected,and explore topological changes in the MCI and AD brain network frameworks.The MST global attributes and node attributes are calculated and statistically analyzed to extract the MST attributes with significant differences.(2)For the problem that the minimum spanning tree does not have clustering and clustering,this paper proposes a method that add a certain percentage of connections to the minimum spanning tree to form a connected backbone network,and analyze the abnormal pattern of AD brain network.Calculating the global attributes and local node attributes of the three groups of subjects unweighted network and statistically analyzing them,and extract network attributes with significant differences.(3)To better describe the overall performance of brain networks local topology changes,the concept of hub disruption index was proposed.Calculate the hub disruption index of the nodal attributes of each test group on the basis of healthy control,and perform statistical analysis on them to explore the topological changes between networks.(4)In order to better serve the network results to the early adjuvant diagnosis of AD.The significant MST attributes and connected backbone network attributes between groups were used to construct the feature space.The SVM classifier is used to train the classification model.The leave-one-out verification method is used to verify the classification model.The effect and contribution of different network characteristics on the classification effect of the three subjects is analyzed.The statistical analysis of the MST and connected backbone network attributes of the three groups of subjects and the statistical results of classification studies show that properties of MST based CBN greatly contributes to the improvement of the classification accuracy among the three groups of subjects,especially mild cognitive impairment and normal The role of classification between tests is very obvious,which provides a certain basis for the early diagnosis of AD.