High-dimensional classification approaches have been widely used to investigate magnetic resonance imaging (MRI) data for automatic classification of Alzheimer's disease (AD). This paper describes the use of t-test based feature-ranking approach as part of a novel feature selection procedure, where the number of top features is determined using the Fisher Criterion. The proposed classification system involves five systematic levels. First, voxel-based morphometry technique is used to compare the global and local differences of gray matter in patients with AD versus healthy controls (HCs). The significant local differences in gray matter volume are then selected as volumes of interests (VOIs). Second, the voxel clusters are employed as VOIs, where each voxel is considered to be a feature. Third, all the features are ranked using t-test scores. In this regard, the Fisher Criterion between the AD and HC groups is calculated for a changing number of ranked features, where the vector size maximizing the Fisher Criterion is selected as the optimal number of top discriminative features. Fourth, the classification is performed using support vector machine. Finally, data fusion methods among atrophy clusters are used to improve the classification performance. The experimental results indicate that the performance of the proposed system could compete well with the state-of-the-art techniques reported in the literature.
Keywords: Alzheimer’s disease; Data fusion; Feature ranking; Fisher Criterion; Support vector machine; Voxel-based morphometry.
Copyright © 2015 Elsevier Inc. All rights reserved.