Feature Extraction And Classification Research Of Brain Network Based On Resting State FMRI For Alzheimer’s Disease

Alzheimer’s disease(AD),commonly known as senile dementia,is an age-related chronic neurodegenerative disease and the most common type of dementia.Mild cognitive impairment(MCI)is a transitional state between normal aging and dementia,10%-15% of MCI patients turn into AD every year.However,there is still a lack of effective biomarkers with high sensitivity in clinical practice.Functional magnetic resonance imaging(fMRI)has the advantages of high spatial resolution and strong computability.Researchers can analyze brain function and morphology from local and network aspects based on fMRI,and then explore the development process and early diagnosis methods of MCI and AD from multiple perspectives and modes.In order to better understand the effect of disease on brain function network and its potential as a biomarker,this study includes the following contents:(1)Most of the previous studies based on fMRI only focused on the blood oxygen level dependent(BOLD)signal of gray matter,and constructed the brain function connection(FC)network based on the region of interest(ROI)of gray matter,while the latest series of studies showed that the BOLD signal of white matter also contains related neural activity information.In order to investigate the brain network abnormalities caused by cognitive impairment more comprehensively and the possibility of brain network characteristics as biomarkers of the disease,this study extracts gray matter and white matter BOLD signals and constructs the FC network based on Pearson correlation coefficient,partial correlation coefficient as well as maximum information coefficient respectively.(2)The sliding window strategy is used to construct the dynamic brain network features.In addition to the root mean square(RMS)which is commonly used to extract the features of dynamic FC,this paper introduces sample entropy and permutation entropy to extract the features of dynamic FC,and the support vector machine(SVM)is used for classification,which aims to verify the effectiveness of various brain network features in identifying diseases and to discover abnormalpatterns of FC in the course of disease progression.(3)Previous studies have shown that cognitive impairments behave differently in different frequency bands.This study divides the BOLD signal into four frequency bands: full-band(0.01-0.08Hz),slow-3(0.073-0.198Hz),slow-4(0.027-0.073Hz),slow-5(0.01-0.027Hz),to further explore the classification performance of brain network features in different frequency bands,providing reference for the early screening and diagnosis of MCI and AD.The results show that the brain network constructed based on Pearson correlation coefficient and maximum information coefficient reflects similar pathological patterns,while which is constructed based on partial correlation coefficient reflects slightly different pathological patterns.The classification of brain network features based on partial correlation coefficients has the best performance with an accuracy rate of100%,and which based on maximum information coefficients has the second highest accuracy rate of 88.61%,both of them are better than those based on Pearson correlation coefficients.The analysis results of dynamic brain network characteristics show that in the progress of MCI,the complexity of patients’ brain areas has increased,and the regularity of FC has deteriorated.During the deterioration of LMCI to AD,the connection patterns between brain areas have changed,and the complexity of activity decreased.The overall performance of the features extracted based on sample entropy and permutation entropy is better than that based on root mean square extraction,the former has a classification accuracy rate of 100%.In addition,based on the classification performance of various characteristics,the results of this study indicate that the addition of white matter signals is helpful to the disease identification to a certain extent.