Classification of SPECT images of normal subjects versus images of Alzheimer’s disease patients

Abstract

This work aims at providing a tool to assist the interpretationof SPECT images for the diagnosis of Alzheimer’s Disease (AD). Our approach is to test classifiers, which uses the intensity values of the images, without any prior information. Such a classifier is built upon a training set, containing images with two different labels (AD patients and normal subjects). It will then provide a classification for any new unknown image. The main problem to be handled is the small number of available images compared to the large number of features (here the image’s voxels): the so-called small sample size problem. We evaluate here the ability of two linear classifiers to correctly label a set of 79 images. Our experiments show promising results. They also show that image classification based on intensity values only is possible and might be used for other applications as well.