First and Second Order Gradients for Alzheimer's Disease Diagnosis

Abstract

Fluorodeoxyglucose Positron Emission Tomography (FDG-PET) is an effective modality in Alzheimer's disease (AD) diagnosis since it can capture the metabolism changes in the brain, even in the early stage of AD, which is known as Mild Cognitive Impairment (MCI). The widely used features for characterizing FDG-PET images are either voxel-wise or region-wise. In this paper, we attempt to characterize FDG-PET images from another point of view—gradients. For this purpose, the first and second order gradients are proposed to tackle the problem of AD diagnosis. Then the effectiveness of combined gradients is also investigated. The experiment results show that the first order gradients can give the best performance with an accuracy of 94.78% in AD diagnosis, which outperforms the state-of-the-art methods, while for classifying progressive MCI (pMCI) from stable MCI (sMCI), the combined gradients are suggested.