The Application Of Multilevel Clustering-evolutionary Random SVM Cluster On The FMRI

The brain is a complex biological organization composed of hundreds of billions of nerve cells.Deep understanding of the function and structure of the brain is one of the most challenging frontier scientific issues since the 21 st century.Functional magnetic resonance imaging(fMRI)as a non-invasive neuroimaging technique provides powerful data support for the brain-related research.Based on the fMRI data,this paper innovatively proposes a Multilevel Clustering-evolutionary Random Support Vector Machine Cluster(MCRSVMC)which is applied to the analysis in Asperger Syndrome(AS)and Mild Cognitive Impairment(MCI).The main content of the article is described as bellow:(1)This paper firstly proposed an interesting MCRSVMC algorithm.Based on the combination of multiple SVM base classifiers,the method of clustering evolution is introduced to construct the MCRSVMC to improve the final classification performance.This new model can be used to extract the optimal subset of features and further detect brain regions affected by the disease,which provides a new perspective for the study of brain diseases.(2)The MCRSVMC method was employed in the analysis of AS.Resting-state fMRI data of 63 AS patients and 72 Healthy Controls(HCs)was first acquired from the ABIDE database.Then,the graph theory metrics were constructed and used as input features of the MCRSVMC classifier.Finally,the MCRSVMC was used to classify AS patients from HCs.The results showed that the classification accuracy rate reached 95.24% based on the optimal features,and the brain regions affected by AS(such as the middle frontal gyrus,hippocampus,supplementary motor area)could be found out.The results of this study indicated that the new method can effectively help physicians to diagnose AS patients.(3)A series of experiments on the MCI process were performed by using the MCRSVMC.First,the resting-state fMRI data of 42 early MCI(EMCI)patients,38 late MCI(LMCI)patients and 36 HCs was obtained from the ADNI database.With considering the graph theory metrics,the MCRSVMC was used to perform two-class classifications on different cognitive states in the MCI process.The results showed that the accuracies of 89.47% and 90% for HC vs.EMCI and EMCI vs.LMCI were respectively achieved based on the optimal features.In the process of detecting the lesions,it was found that the brain regions(such as parahippocampal gyrus and posterior cingulate gyrus)were abnormal in two groups of experiments,indicating that these brain regions played key roles in the “HC EMCI LMCI” evolution.In a word,these experimental results demonstrated the enormous potential of the MCRSVMC approach in the study of MCI process.