Feature extraction of alzheimer's disease classification based on principal component and random subspace discriminant analysis

Abstract

Alzheimer's disease (AD) is one of the diseases which brings great influences on the lives of the people. AD classification can serve as a supportive tool to help the doctor to analyze the brain images. One of the important steps in AD classification is feature extraction. Among the feature extraction techniques, Principal Component Analysis (PCA) is a widely used machine learning approach. Nevertheless, it is hard to decide the number of dimensions to be extracted after the transformation. The accuracy of the classification can be greatly affected by the number of dimensions to be chosen. Therefore, this paper has developed a feature extraction method based on principal component and random subspace discriminant analysis (PCRSDA) to extract and select the features. The selection of the number of dimensions was determined by 10-fold cross validation where the features were selected randomly without replacement. The dataset in this paper was collected from Alzheimer's Disease Neuroimaging Initiative (ADNI) database across four time points. The classification results were 81%, 84%, 87% and 87% at time point of 24 months before stable diagnosis, 18 months before stable diagnosis, 12 months before stable diagnosis and at the stable diagnosis time point, respectively.