Study On The Analysis Method Of Unbiased Brain Network And Its Application In Alzheimer's Disease

Human brain is one of the most complex systems in nature.It is a complex and huge network composed of hundreds of millions of neurons.It is the physiological basis of information processing and cognitive expression.In recent years,the theory of complex network has been introduced into the research of brain science.The researchers have expressed it as a graph,analyzed the brain network from different angles by using the knowledge of complex network and graph theory,and made great progress.It provides new methods and means for the research of human brain.However,there are still some key problems to be solved in the construction and analysis of brain network.Firstly,the current threshold-based brain network construction methods are affected by the threshold selection,which makes the comparability of the network decrease,and at the same time makes the analysis results inconsistent.Secondly,how to mine the optimal attributes and structural features to identify the brain network,and how to accurately calculate the similarity of the brain network for clustering between the brain networks is also an urgent problem to be solved.Based on the complex network theory and graph theory,this paper discusses the evaluation method of the node centrality of the brain network,and the construction method of unbiased brain network is discussed aiming at the deviation of analysis result caused by threshold network.On this basis,the complex network indexes were compared among groups in order to find out the changing regularity of each index of the brain network under the condition of disease,and the classification model is constructed.In addition,from the angle of community structure,the differences between groups are analyzed,and the similarity between brain networks is measured,and the clustering model is constructed by using unsupervised learning method.Finally,the above methods were applied to Alzheimer's disease to analyze the differences of brain network between patients with Alzheimer's disease and normal controls,and to find out the imaging indexes of Alzheimer's disease diagnosis in order to assist clinical diagnosis.The main work of this dissertation includes:(1)A node centrality measurement method based on degree and k-core is proposed and applied to the classification of threshold brain networks.In the theory of complex networks,many methods for measuring the centrality of nodes are provided,but these methods measure the centrality of nodes from different and single angles.It is found that a single measurement index can not accurately measure the centrality of nodes.In order to solve this problem,this paper proposes a combination of the degree and k-core centrality to measure the centrality of the node of the brain network,then calculates the centrality of each node in the threshold brain network,and classifies the Alzheimer's Disease patients and the normal controls in threshold brain network.(2)The difference analysis method of topology structure of unbiased brain network is proposed,and the classification model is constructed.In order to avoid the influence of threshold selection on the network structure in the traditional threshold network,this paper introduces the minimum spanning tree deviation correction method to construct the unbiased brain network,and then analyzes the differences in the topology structure of the network.Kernel Principal Component Analysis method is used to capture differences and Alzheimer's Disease classification model is constructed.(3)An unbiased brain network classification model based on the combination of local attributes and topological features is proposed.Brain networks show significant differences in attributes and structures.Different features may provide different information.Considering that multiple features may reflect multiple complementary information,this paper extracts the features of local attributes and topological structures based on unbiased brain networks,fuses the two different features using synthetic kernels,and constructs Alzheimer's Disease classification model.(4)A similarity measurement method of brain network is proposed and a clustering model is constructed.when the data is assigned the correct label and the optimal feature is selected,the classification performance is usually very good.But it is time-consuming to extract and select the best features from a large number of data.In additional,due to the need for labeled data,it is highly dependent on the ground truth provided by the diagnosis of clinicians.Thus clinicians will spend a lot of time and effort to tag data.In order to reduce the workload of marking data,this paper proposes a method to measure the similarity of brain network by using cosine similarity and sub-network kernel.And the brain network clustering model is constructed with spectral clustering to identify patients with Alzheimer's disease.