Dementia-Related Features in Longitudinal MRI : Tracking Keypoints over Time

Abstract

We aim at developing new dementia-related features based on longitudinal MRI in order to differentiate various stages of Alzheimer’s disease.Current methods for dementia classification rely heavily on the quality of MRI preprocessing, especially on prior registration. We propose to avoid a possibly unsuccessful and always time-consuming non-rigid registration by employing local invariant features which are independent of image scale and orientation, and can be tracked over time in longitudinal studies. We detect and track such keypoints based on scale-space theory in an automatized image processing workflow, and test it on a standardized MRI collection made available by the Alzheimer’s Disease Neuroimaging Initiative (ADNI).Our approach is very efficient for processing very large datasets collected from different sites and technical devices, and first results show that characteristic scale and movement of keypoints and their tracks differ significantly between controls and diseased subjects.