Longitudinal FDG-PET features for the classification of Alzheimer's Disease

The majority of existing computer-aided diagnosis (CAD) schemes for Alzheimer's disease (AD) rely on the analysis of biomarkers at a single time-point, ignoring the progressive nature of the disorder. Recently, a method was proposed by Gray et al. [1] for the multi-region analysis of longitudinal fluorodeoxyglucose positron emission tomography (FDG-PET) images which reported classification improvements by using regional signal intensities combined with regional change over a 12 month period. In this paper we extend the approach proposed in [1] to the analysis of the entire brain pattern. Compared to [1], our method uses voxel-wise differences and avoids segmentation of the images into regions of interest. For our study, FDG-PET scans at the baseline and at 12-month follow-up of cognitively normal (CN), mild cognitive impairment (MCI) and AD subjects were retrieved from the Alzheimer's disease neuroimaging initiative (ADNI) database. For both AD and MCI identification, the best classification results were achieved by combining cross-sectional and longitudinal information rather than using only the cross-sectional data. Furthermore, the longitudinal voxel-based analysis outperformed multi-region analysis.