Classification Of MCI Brain Network Based On Granger Causality Analysis

Mild cognitive impairment(MCI)is a transitional state between healthy elderly people and Alzheimer's disease(AD),commonly characterized by a degree of memory impairment and cognitive decline in the clinical manifestations.It haven't meet the criterion of the dementia,but has a high risk of transforming AD.The brain lesions of AD patients are serious and irreversible,and there is no effective treatment in clinical.However,the early detection and intervention treatment of MCI may reduce the probability of MCI conversion to AD,delay or even reverse the development of the disease.Therefore,the early diagnosis and detection of MCI has important clinical and social significance.The rapid development of neuroimaging has enabled researchers to conduct comprehensive research on MCI diseases and achieved considerable results.However,there is still lack of unified and effective methods in the early diagnosis of MCI,and the pathological mechanism of MCI is still not fully understood.Current studies mainly focus on the abnormal pattern of the MCI structure brain network and the functional brain network.There are few studies on the directed brain network of MCI,and further exploration is needed.This paper mainly studied the MCI brain network from the perspective of the directed brain network.The main contributions include the following two aspects:(1)MCI directed brain network anomaly pattern mining based on brain regions This study is carried out around the application of granger causality analysis(GCA)in neurological research.Firstly,resting-state functional magnetic resonance imaging(f MRI)data of normal elderly(NC),early mild cognitive impairment(EMCI)and late mild cognitive impairment(LMCI)were preprocessed,and divided brain regions according to the empirical anatomy template as the nodes of brain network.Then the granger causality analysis method was applied to construct the resting state directed brain networks of three groups.Secondly,based on the basis of complex network graph theory,the topological properties of directed brain network were calculated,and the appropriate feature selection method was chosen to select the features for MCI assistant diagnosis.Eventually,classified sample data of three groups using support vector machine classifier.Analyzed the classification results and the selected optimal feature subsets were discussed to give a physiological explanation.Exploring the abnormal pattern of the MCI directed brain network from the brain region level,and mining the imaging markers of MCI early diagnosis to assist the diagnosis and intervention treatment of clinical MCI patients.(2)MCI directed brain network anomaly pattern mining based on voxel The sample size was extended to the NC,MCI,and AD groups,and the hippocampus of the brain area,which is very important for the cognitive memory function of AD,was selected as the seed point.The granger causal analysis method was applied to construct the whole brain directed connectivity map of three groups based on voxel.Then,statistical analysis was conducted on the directed connectivity maps of three groups to investigate the abnormality of the directional connection pattern between the bilateral hippocampus and other voxels in the MCI brain compared with the normal elderly.Analyzing the difference between MCI and AD abnormality in the brain network to further understand the pathological development of MCI and the possible clinical manifestations.Exploring the abnormal connection pattern of hippocampus and other voxels in MCI patients at the voxel level to further understand the evolution pattern of directed brain network from MCI to AD.Directed brain network classification model of MCI achieved a better classification effect and provided an assistant technique for the early clinical diagnosis and detection of MCI.The results of feature selection showed the temporal lobe,frontal lobe,occipital lobe and parietal lobe in the brain of MCI patients developed pathological changes in varying degrees,and mainly concentrated in the temporal lobe and frontal lobe.The analysis of directed brain network based on voxel found the connectivity of bilateral hippocampus and many brain regions such as middle temporal gyrus,inferior temporal gyrus,cingulate gyrus and superior frontal gyrus demonstrated abnormal.The connectivity pattern differences between MCI group and AD group was more obviously.These studies revealed the effectiveness of constructing directed brain network based on granger causality analysis in the study of MCI and AD diseases,and provide more evidence for understanding the pathophysiological mechanisms of MCI and AD.