Individual Metabolic Network Construction And MCI Progression Prediction Based On Longitudinal FDG-PET

Mild cognitive impairment(MCI)is an intermediate stage between normal aging and Alzheimer’s disease(AD).The conversion of MCI to AD is generally a dynamic process,and longitudinal data tend to capture the development of disease over time.18 F fluoro-deoxy-glucose positron emission tomography(FDG-PET)has been proven to be a powerful tool for measuring cerebral glucose metabolism.Accurate prediction of MCI conversion by FDG-PET data is of great significance to the prevention of AD.In this study,we selected 79 subjects both baseline and multiple follow-up FDG-PET scans as well as cognitive scores.The subjects were divided into 33 progressive MCI(pMCI)patients and 46 stable MCI(sMCI)patients from the Alzheimer’s Disease Neuroimaging Initiative(ADNI).According to the method proposed in this paper,which combines the rough distance between brain regions and the exact distance between voxels in various brain regions,a method of building a single metabolic brain network is proposed,and the degree and clustering coefficient of network nodes are calculated,and the difference analysis is carried out.The metabolic brain network of two groups shown small-world properties.And the results showed that the abnormality mainly occurred in DMN,and the clustering coefficient decreased significantly in the left anterior wedge at the four time points,and with the development of the disease,the region with the significant decrease in clustering coefficient spread to the right cingulate gyrus and para-hippocampal gyrus.Subsequently,a classification model was constructed to predict MCI conversion.The network features are defined as the relevant connection coefficients of the monomeric metabolic brain network and the clustering coefficients of each node.The static features are the average metabolic intensity of each brain region,and the dynamic features metabolic intensity difference D and the metabolic intensity change rate R are defined based on the static characteristics.Using Lasso and F-score to select effective features,the SVM classifier based on RBF was trained and the classification result with the highest classification accuracy rate of 89.9% was obtained.In addition,we draw the following conclusions: the combination of static features at multiple time points may promote the prediction of MCI transformation,and dynamic features are more sensitive to MCI transformation than static features.Vertical data can provide supplementary information for horizontal data,and vertical data can reflect the changing trend of individual characteristics and improve the accuracy of classification.The introduction of longitudinal data and the construction of individual metabolic brain network may provide a new way to predict the development of MCI.