Early prediction of Alzheimer’s disease using longitudinal descriptors over MRI data

Abstract

Alzheimer’s Disease (AD) is a chronic neurodegenerative disease characterized by memory loss and decline in other cognitive abilities. The neuronal death produces atrophy, which can be observed using Magnetic Resonance Imaging (MRI). It is known that the prodromal stage of AD, called Mild Cognitive Impairment (MCI), can progress to AD with a certain rate or stabilize without developing the disease. Therefore, the prediction of future progression from MCI to AD has a great potential to provide MCI patients preventive treatment against the dementia symptoms. The aim of this project is to discriminate between MCI patients that will remain stable (s-MCI) and patients that will progress to develop AD (p-MCI) based on the observed atrophy patterns from magnetic resonance images. MRI scans of the ADNI1 dataset from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database will be used. The scans will be processed to extract similarity maps, that will be analyzed using Machine Learning techniques to classify the disease outcomes. This work will extend the research performed by Sanroma et al (2017). They selected the hippocampus as region of interest to extract the similarity vectors, included one follow-up image per patient and used logistic regression to predict AD. This work extends the former paper in several ways: 1) expanding the region of interest from the hippocampus to the whole brain 2) including more follow-up images per patient at different time-points and 3) exploring the prediction using further classification algorithms, with the objective to achieve state-of-the-art performance and improve the results of the previous research