Research On Alzheimer Disease Neural Fingerprint Based On Resting-state Functional Magnetic Resonance Imaging

In recent years,the Alzheimer disease(AD)has become a major threat of human health with its incidence rate ascending year by year.Traditionally,AD diagnosis mainly depends on the doctor's diagnosis,MMSE and other unprecise methods,easily to misjudge thus delaying the timing of treatment.Early diagnosis of AD is very important for patients,because once happened,we would miss the opportunity to cure the disease,there will be no way to treat effectively.Early diagnosis and treatment of AD can help to improve the quality of life of the patients and delay the course of disease.The resting-state functional magnetic resonance imaging(rs-f MRI)which precisely reflects the brain changes on the resting state of individuals provides a quantitative approach to extract features,which has been introduced to distinguish AD patients from normal population,and achieve the result of early diagnosis.In this study,we carried out a comprehensive study of feature extraction based on rs-f MRI.In view of the fact that the current academic research only focuses on the unilateral features of MRI data,we constructed the feature model and screened the features based on the brain functional network,regional homogeneity(Re Ho)and amplitude of low-frequency(ALFF)fluctuation.In the experiment,we firstly extracted Re Ho and ALFF value of each voxel from each sample's MRI data,then we partitioned the brain according to the template.and the mean value and variance of those parameters were regarded as the feature components.In the next steps,we constructed the brain functional network and analyzed the network according to the graph theory,so we got all the features.The recognition performance of some pattern recognition algorithms were compared,and the best classifier were applied to the feature selecting experiment in the next step.The result shows that,those algorithms with simple principle didn't shows weak performance in the experiment,such as Support Vector Machine(SVM)with linear kernel function and Native Bayesian method,and SVM with linear kernel were selected as the best classifier to take part in the next experiment.We proposed a neural fingerprint building method based on Fisher score and linear-kernel SVM,and constructed the rs-f MRI neural fingerprint of AD,we can get a recognition of 100% using linear-kernel SVM on our feature sets.Firstly we calculated the Fisher score of each feature,and sorted the features according to the score.Then,under the participation of SVM,we finally got the best feature subsets.