Diagnosing Alzheimer’s disease: Automatic extraction and selection of coherent regions in FDG-PET images

Abstract

Alzheimer’s Disease is a progressive neurodegenerative disease leading to gradual deterioration in cognition, function and behavior, with unknown causes and no effective treatment up to date. Techniques for computer-aided diagnosis of Alzheimer’s Disease typically focus on the combined analysis of multiple expensive neuroimages, such as FDG-PET images and MRI, to obtain high classification accuracies. However, achieving similar results using only 3-D FDG-PET scans would lead to significant reduction in medical expenditure. This paper proposes a novel methodology for the diagnosis Alzheimer’s Disease using only 3-D FDG-PET scans. For this we propose an algorithm for automatic extraction and selection of a small set of coherent regions that are able to discriminate patients with Alzheimer’s Disease. Experimental results show that the proposed methodology outperforms the traditional approach where voxel intensities are directly used as classification features.