The Brain Science Research Based On The Weighted Random SVM Cluster

Functional magnetic resonance imaging(fMRI)is an important imaging tool for exploring the structural and functional activities of the brain,which has been widely used in brain science research.It is of great significance for understanding the human brain by identifying the intrinsic brain activity patterns from the small-sample and high-dimensional fMRI data.In the past two decades,with the development of pattern recognition technology and machine learning algorithms,the classification research of fMRI has gradually become the interest of researchers.This paper proposed two interesting classification methods based on the fMRI data,and respectively used them for the identification analysis of mild cognitive impairment(MCI)and the progression analysis of Alzheimer's disease(AD).The main contents of this article are described as follows:(1)A new Weighted Random SVM Cluster(WRSVMC)algorithm was presented.The algorithm is mainly divided into two parts.One is to build multiple SVMs by randomly selecting samples and features for an ensemble classifier.Another is to give different weights to different SVMs in the voting process in ordor to optimize the performance of the model.This paper combined fMRI data and graph theory features,and used WRSVMC algorithm to classify MCI patients and normal controls.The results showed that the classification accuracy rate was as high as 87.67%,and the MCI-affected brain regions [such as gyrus rectus(REC.L),precentral gyrus(PreCG.R),olfactory cortex(OLF.L)and middle occipital gyrus(MOG.R)] were found,and provides a new way for the diagnosis of patients with mild cognitive impairment.(2)A new Weighted-evolutionary Random SVM Cluster(WERSVMC)algorithm was presented.The algorithm innovatively introduces the idea of evolution into the weighted random SVM cluster,which dynamically removes redundant features from all features to obtain optimal ones.Finally the abnormal brain regions can also be found.This paper combined fMRI data and functional connectivity features,using WERSVMC method to conduct experimental research on AD process.The results showed that the 90% and 88.89% accuracies were respectively obtained for the early MCI vs.late MCI and late MCI vs.AD classifications.In the process of exploring abnormal regions,several high-frequency brain regions(e.g.,superior temporal gyrus(STG.R),middle temporal gyrus(MTG.L)and insula(INS.L))are presented in the two groups of experiments at the same time,which suggested that these brain regions play crucial roles underlying the AD progression.But some brain areas only displayed high frequencies only in one group of experiment,which facilitated to understand differences in disease progression,and provides a new perspective for the pathological study of AD process.