We compared the predictive ability of the various neuroimaging tools and determined the most cost-effective, non-invasive Alzheimer's disease (AD) prediction model in mild cognitive impairment (MCI) individuals. Thirty-two MCI subjects were evaluated at baseline with [(18)F]-fluorodeoxyglucose positron emission tomography (FDG-PET), MRI, diffusion tensor imaging (DTI), and neuropsychological tests, and then followed up for 2 yr. After a follow up period, 12 MCI subjects converted to AD (MCIc) and 20 did not (MCInc). Of the voxel-based statistical comparisons of baseline neuroimaging data, the MCIc showed reduced cerebral glucose metabolism (CMgl) in the temporo-parietal, posterior cingulate, precuneus, and frontal regions, and gray matter (GM) density in multiple cortical areas including the frontal, temporal and parietal regions compared to the MCInc, whereas regional fractional anisotropy derived from DTI were not significantly different between the two groups. The MCIc also had lower Mini-Mental State Examination (MMSE) score than the MCInc. Through a series of model selection steps, the MMSE combined with CMgl model was selected as a final model (classification accuracy 93.8%). In conclusion, the combination of MMSE with regional CMgl measurement based on FDG-PET is probably the most efficient, non-invasive method to predict AD in MCI individuals after a two-year follow-up period.
Keywords: Alzheimer Disease; Diffusion Tensor Imaging; FDG-PET; MRI; Mini-Mental State Examination; Prediction.