Research On Alzheimer’s Disease Based On Brain Region Physiological Features And Brain Network Fusion Techniques

Alzheimer’s disease(AD)is a neurodegenerative disease that causes memory impairment,cognitive impairment and other symptoms in patients,which have serious impacts on their daily life and mental health.Studying the pathogenesis of AD is of great significance for improving the quality of life of patients.However,traditional methods for studying AD only explore brain networks or brain regions physiological characteristics based on blood oxygen-level dependent(BOLD)signals separately,without performing information fusion analysis on both.Therefore,this study will propose a new method to fuse brain functional network-brain region physiological characteristics,and carry out the following works:1.First,this thesis extracts physiological features of different brain regions based on BOLD signal: ALFF,f ALFF,Re Ho,and constructs partial correlation brain networks to reflect the physiological activity and information interaction of different brain regions.Then,this thesis adopts graph wavelet transform to fuse the brain network-physiological feature information and construct brain physiological attribute network(BPAN)to characterize the correlation between different physiological activities of different brain regions.Finally,this thesis carries out the following research based on BPAN:(1)Analyze the importance of different types of physiological activities in different brain regions and perform statistical tests,revealing that there are significant differences in the importance of different types of physiological activities in different brain regions(p-value < 0.05 after FDR correction);(2)Extract topological parameters and perform statistical tests on the subnetworks of BPAN(such as different lobes,left and right hemispheres),revealing that there are significant differences in the correlation of different types of physiological activities in different BPAN subnetworks(p-value < 0.05 after FDR correction).2.Based on the above work,this thesis extracts and statistically tests the BPAN parameters of AD patients and CN respectively.The results show that most of the BPAN topological parameters have significant differences between the two groups of subjects(p-value < 0.05 after FDR correction).In addition,this thesis also uses eight common classifiers to identify AD patients,and their average accuracy,recall,precision and AUC(Area under curve)values all reached over 99%.Finally,in order to further verify the effectiveness of the proposed new method,this paper also uses the following features to classify AD patients and healthy controls:(1)using different types of physiological features(ALFF,f ALFF and Re Ho)of each brain region separately;(2)using topological features(clustering coefficient,etc.)of brain networks separately;(3)vector combination of the above two types of features.Their average accuracy,recall,precision and AUC values were all lower than 90%.The results show that BPAN has the potential to be a new biomarker for identifying AD patients.In summary,this thesis proposes a novel method to fuse the brain network-brain region physiological features in a non-Euclidean way.The obtained BPAN network can not only reveal the potential associations between different physiological activities of each brain region,but also provide a new perspective for the exploration of brain mechanisms and brain diseases.