Research On The Identification Of Mild Cognitive Impairment Based On Dynamic Functional Connectivity

Mild cognitive impairment(MCI)is an intermediate state between normal aging and Alzheimer’s disease.Dynamic functional connectivity(d FC)analysis using resting state functional magnetic resonance imaging(rs-fMRI)is an advanced technology to capture the dynamic changes of neural activity for brain disease identification.Due to the functional connectivity based on rs-fMRI data is extremely weak,there is still no objective,unified and effective biomarker for this disease.Therefore,an adaptive d FC modeling method is proposed to accurately track the time-varying characteristics of d FC.Then,in order to extract the most discriminative spatiotemporal features from d FC,three different MCI identification methods are designed to verify and reveal the abnormal connections related to MCI pathology.The specific research contents of this thesis are as follows:(1)An adaptive d FC extraction method based on Kalman filter is proposed in this thesis This method combines group lasso algorithm with Kalman filter algorithm to construct adaptive DFC.The topology of DFC can be determined by lasso group,and the dynamic change of function connection can be captured by Kalman filter algorithm point by point.The proposed method not only solves the problem of traditional sliding window parameter selection,but also provides more time-varying modes.The p H bilstm network is designed to verify the superiority and effectiveness of the proposed DFC extraction method.(2)An MCI identification method based on spatio-temporal attention network is proposed in this thesis.Similarly,in order to distinguish the most discriminative time points and functional connections,time attention module and spatial attention module which are more suitable for extracting DFC are proposed.In addition,the time-space regularization term is added to optimize the network training based on the cross entropy loss function in the designed sta bigru network.In this thesis,the temporal and spatial attention results of d FC are visualized,and the most discriminative functional connections and brain regions are explored.(3)An MCI identification method based on adaptive dynamic spatio-temporal graph convolution is proposed in this thesis.In order to learn the shared embedding and dynamic characteristics of functional connection and topology features,this thesis designs an adaptive graph learning method,which combines it with time,space and channel attention,and achieves the goal of end-to-end fusion learning space-time embedding.For the three classification tasks,the identification accuracy of MCI is improved compared with other comparison methods.It verifies the importance of extracting the function connection and the dynamic of topology.