Dimensionality Reduction Methods for Alzheimer's Disease Classification

Abstract

Alzheimer’s Disease (AD) is the most common type of dementia in the world. There is no cure for the disease. The early detection of the disease is the goal for better AD treatment. Computer-aided diagnosis serves as a supportive tool in AD diagnosis, which classifies the stages of AD from the three-dimensional (3D) brain images. In image processing, a 3D image will result in millions features. Therefore, apart from extracting significant features for AD classification, the feature extraction step also involves reducing the dimensions of the data. This paper aims to investigate the suitability of the current dimensionality reduction methods on AD classification. In addition, this paper also examines the impact of various intrinsic dimension estimation techniques on the dimensionality reduction techniques. The contribution of this paper is to conduct the comparative study with same dataset, which allows a comparison of the strengths of existing methods on AD classification. A total of 200 subjects with T1-weighted images were obtained at different time points from Alzheimer's Disease Neuroimaging Initiative (ADNI) database. The performance measurement of this paper is classification accuracy. The best approach among the methods discussed in this paper is the combination of discrete wavelet transform and principal component analysis, and it achieved 87% accuracy in average for the dataset collected at different time points. It reveals that the current techniques have the strength in extracting significant features for AD classification. However, the classification performance is influenced by the intrinsic dimension on dimensionality reduction. Therefore, the rationale and suggestions for improvement of the methods are discussed.