Early Diagnosis Of Alzheimer’s Disease Based On Spatiotemporal Characteristics Of Resting Functional Magnetic Resonance Imaging

Alzheimer’s disease(AD)is the most common cause of dementia,and its severity can result in death.The number of AD patients in China is expected to grow from 6 million to 28 million by 2050.The earlier stage of progression of AD disease is Mild Cognitive Impairment(MCI).Although there is currently no cure for AD,correctly identifying patients who are at the MCI stage can provide timely and effective drug treatment to delay the progression of AD disease.Neuroimaging has the advantages of safety,non-invasive and high accuracy,and has been widely used in the early diagnosis of AD.Resting state functional magnetic resonance imaging(rs-fMRI)can be used to map the changes in functional interaction between brain regions over time.It has temporal and spatial characteristics and has been used in the clinical diagnosis of AD.This thesis proposes an early diagnosis method of Alzheimer’s disease based on the combination of rs-fMRI spatiotemporal features,machine learning and deep neural network.The details are as follows:First of all,for the four groups of data of AD,early MCI(EMCI),late MCI(LMCI)and normal control(NC),the pretreatment of rs-fMRI was completed by using SPM12 and Gretna.According to the average time series obtained by preprocessing,a dynamic functional connection network is constructed based on Pearson correlation coefficient and time window.Then,because of the lack of spatial characteristics in traditional functional network,this thesis proposes a method of dynamic graph theory;Aiming at the problem that the characteristics of dynamic graph theory is too large to affect the classification performance of the model,this thesis designs a new feature selection algorithm,which is based on the filter selection algorithm and the wrapper selection algorithm to find the feature subset,and then constructs the AD/NC,EMCI/CN,EMCI/LMCI classification models based on SVM,thus realizing the early diagnosis of AD based on dynamic graph theory and machine learning.Finally,a number of comparative experiments are designed to verify the superiority and generality of the proposed method from three aspects:threshold selection,feature selection algorithm and classification model.Then,aiming at the machine learning model,which is generally divided into three steps:feature extraction,feature selection,and classification,and is trained as a sub-optimal model,this thesis proposes an end-to-end deep neural network model based on dynamic functional connection network.The r-fMRI spatiotemporal characteristics extraction model based on CNN and TCN,performs the decoupling and deep extraction of spatiotemporal characteristics.Thus,the early diagnosis of AD based on spatiotemporal features and deep learning is realized.Finally,several sets of comparative experiments are designed to verify the effectiveness of the proposed method from the selection of time window parameters,the selection of space convolution layer,the selection of time convolution layer and the comparison with the research status at home and abroad.To sum up,this thesis not only studies the early diagnosis of AD from the perspective of machine learning,but also uncouples and deeply mines the spatiotemporal features of rs-fMRI through the self-constructed spatiotemporal hybrid neural network model from the perspective of endto-end deep neural network,by extracting the features of dynamic graph theory and proposing a hybrid feature selection algorithm.By studying the early diagnosis method of Alzheimer’s disease based on the combination of rs-fMRI spatiotemporal characteristics,machine learning and deep neural network,the AD patients can be identified as soon as possible,and doctors can intervene in time to help delay the progress of AD disease and relieve the social nursing pressure and economic costs.