Research On Brain Network Classification Method Based On Graph Core And Feature Selection Algorithm

Alzheimer’s disease(AD)is a common progressive neurodegenerative disease of the elderly.Currently,drug therapy can only alleviate the symptoms but cannot cure them.This disease seriously affects the quality of life of patients and caregivers.Mild cognitive impairment is a condition between healthy elderly people and alzheimer’s disease.The diagnosis and treatment of mild cognitive impairment can effectively reduce the conversion rate of alzheimer’s disease.Currently,based on functional brain imaging data,researchers can construct and analyze functional brain networks for the diagnosis of mild cognitive impairment.With the development of machine learning,machine learning classification algorithm is applied to the classification of brain network,which has become a new way to diagnose mild cognitive impairment.In this paper,the classification of brain networks is studied based on graph kernel and feature selection algorithm.The main work is as follows: first,aiming at the situation that most graph kernels can not distinguish graph data with the same structure well,a graph kernel for brain networks is proposed,which judges the similarity between two networks by extracting the shortest path with node label in the network,It can better reflect the difference of topological structure between brain networks;secondly,based on the proposed graph kernel and principal component analysis,this paper implements the graph kernel dimensionality reduction algorithm,which can map the brain network data into vector data through graph kernel dimensionality reduction algorithm,so that the classification algorithm in machine learning can be used for brain network classification;thirdly,based on mutual information,this paper implements a feature selection method,which measures the redundancy between feature attributes by mutual information,can effectively reduce the redundancy between feature attributes.In order to verify the effectiveness of the above methods,the resting f MRI brain image data of patients with mild cognitive impairment and healthy elderly were downloaded from Alzheimer’s Disease Neuroimaging Initiative.Based on the preprocessing of brain image,the brain network was constructed,and the graph kernel verification experiment and classification experiment were carried out.In the graph kernel verification experiment,we use the improved graph kernel proposed in this paper to compare with the subtree kernel and the shortest path kernel.The experimental results show that the improved kernel can distinguish the differencesbetween the brain network data of healthy elderly people and patients with mild cognitive impairment more effectively than other graph kernel.In the classification experiment,the brain network is mapped to vector data by the graph kernel dimensionality reduction algorithm proposed in this paper;then,the feature selection algorithm proposed in this paper is used to select the feature of the brain network and realize the dimensionality reduction of the data;finally,the data is classified by using the support vector machine model,and the accuracy and recall rate are compared and analyzed.The experimental results show that: compared with the method of extracting single feature attribute as feature vector,the classification accuracy of the proposed method is increased by about 17% on average,the recall rate is increased by about18% on average,and the classification accuracy of the proposed method is improved by about 2% and the recall rate is increased by about 3.5% compared with the feature fusion method,thus,the validity of the proposed classification method is verified.