Research On Auxiliary Diagnosis Of Alzheimer’s Disease Based On The Spatial And Temporal Characteristics Of Resting-state FMRI

As a neurodegenerative disease,Alzheimer’s Disease（AD）is seriously harming the health and quality of life of the elderly.The use of computer to assist the discrimination of AD images is beneficial to reduce the burden of doctors reading films,help doctors quickly screen AD patients,so as to intervene as early as possible,delay the process.Resting-state functional Magnetic Resonance Image（rs-fMRI）is a period of information collection in which subjects lie on their back,their whole body is relaxed and their brain is awake.rs-fMRI not only have temporal characteristics,but also have spatial characteristics.It not only includes the changes of Blood Oxygenation Level Dependent（BOLD）signal values of brain voxels during the acquisition period,but also includes the interaction of different brain regions in the acquisition process from space.Based on the analysis of the temporal and spatial characteristics of rs-fMRI data,this paper studies auxiliary model that comprehensively utilizes the temporal and spatial characteristics of rs-fMRI.The main research work includes:（1）Analysis of spatial and temporal characteristics of rs-fMRI.Aiming at the time characteristics of rs-fMRI,the minimum,maximum,mean,standard deviation,root mean square and periodicity of the brain voxel BOLD signal time series contained in rs-fMRI were statistically analyzed in this paper.In view of the spatial characteristics of rs-fMRI,different functional brain regions and their connections in rs-fMRI were represented as brain networks,and the characteristic path length,global efficiency and clustering coefficient of the network were statistically analyzed,laying a foundation for subsequent auxiliary diagnosis research.（2）Auxiliary model for AD diagnosis based on rs-fMRI spatial and temporal characteristics.As a temporal recursive network,Long Short-Term Memory（LSTM）is widely used in time series data processing,but it cannot capture the spatial characteristics of rs-fMRI.In order to comprehensively utilize the spatial and temporal characteristics of rs-fMRI,a Feature Weighted LSTM（FW-LSTM）model was proposed in this paper.The model calculates the connection frequency of each brain region and integrates it into the LSTM as a spatial feature weight.Therefore,it can comprehensively simulate temporal and spatial changes in rs-fMRI brain regions.In the Alzheimer’s Disease Neuroimaging Initiative（ADNI）data set,the FW-LSTM model was used to classify AD with 77.80% accuracy,76.41% sensitivity and 78.81% specificity.It is superior to the random forest model using only the spatial features of the brain and the onedimensional convolutional neural network model and LSTM model using only the temporal features of the brain region.（3）Auxiliary diagnosis for AD based on dynamic graph theory features of rsfMRI.The low-order functional connections constructed by the sliding window method can only capture the shallow relationships between brain regions,but cannot get the higher-order information in the correlation between brain regions.In order to further study the significance of dynamic time-varying spatial characteristics of rs-fMRI for AD auxiliary diagnosis,this paper proposed an AD auxiliary diagnosis based on dynamic graph theory features of rs-fMRI.By constructing high-order dynamic functional connections and using corresponding dynamic graph theory features as classification features,the multi-centrality support vector machine was used for AD auxiliary diagnosis.The accuracy,sensitivity and specificity of the proposed classification method can reach 93.98%,95.56% and 92.00%.The experimental results can verify the effectiveness of the proposed classification method based on the spatial and temporal characteristics of rs-fMRI in AD assisted diagnosis.