Functional Brain Network Based On Time-frequency Analysis And Its Application In Alzheimer’s Disease

Alzheimer’s disease(AD)is an irreversible neurodegenerative brain disease that severely affects patients’ quality of life.With the extensive application of computer technology in the medical field,more and more researchers begin to use artificial intelligence technology to study AD and have made some progress.Functional brain networks derived from resting state functional magnetic resonance imaging(rs-fMRI),as a non-invasive means of detecting brain abnormalities,provide a valuable opportunity for early intervention and control of AD.In functional brain networks,time-frequency domain analysis provides multi-dimensional information about brain activity.It allows researchers to better understand and explore how the brain works.Previous studies mainly explored the abnormal functional connectivity and topology of patients with brain diseases based on fixed frequency bands or sliding windows,which often ignored the rich multi-band information and time-varying differences of brain networks at different time periods.Based on the above problems,this paper constructed the classification framework of Alzheimer’s disease based on functional brain network in the time-frequency domain and developed the AD auxiliary diagnosis system.Specifically,this paper includes the following three efforts:(1)In view of the difficulty of fully capturing diagnostic and prognostic information with a single frequency band,a functional brain network classification framework based on multi-frequency fusion was proposed to mine the nonlinear correlation between different frequency bands.First,the blood oxygenation level dependent(BOLD)signal of a single frequency band was decomposed into two frequency bands by ensemble empirical mode decomposition(EEMD)method.Then,the functional brain network of the two frequency bands was integrated into a multi-frequency fusion network by similar network fusion(SNF)method.Finally,feature extraction and classification of fusion network are carried out.The validity of the proposed framework is validated using a dataset from the Alzheimer’s Disease Neuroimaging Initiative(ADNI).The experimental results show that the scheme can extract rich multi-band network features and biomarker information and achieve good classification accuracy.(2)Aiming at the problem that traditional dynamic functional brain networks cannot capture the time-varying differences of functional brain networks at different time periods,this paper proposed a functional brain network classification framework based on graph convolutional neural network(GCN).Firstly,multiple sliding windows were used to construct the dynamic functional brain network and the graph convolution neural network was used to learn the topology structure of the functional brain network.Then,the temporal characteristics of the dynamic functional brain network were captured by the long short-term memory network(LSTM).Finally,the feature weights were adjusted by the attention mechanism module and used for disease prediction.Through multiple experiments,the proposed scheme can effectively capture the evolution information of the importance of functional brain networks at different time periods,and obtain good prediction results.(3)Combined with the above two frameworks,this paper developed an AD auxiliary diagnosis system based on web platform.Doctors can realize the management,statistics and analysis of the data of patients with Alzheimer’s disease through the auxiliary diagnostic system.This system has passed the function test.It can meet the basic needs of auxiliary diagnosis,has a certain practical value.In summary,the framework and system proposed in this paper have a broad application prospect in the early diagnosis of AD.