Feature ranking for mild cognitive impairment and Alzheimer's disease diagnosis

This paper describes the experiment of analyzing MRI images of the brain of patients with Alzheimer's disease (AD), mild cognitive impairment (MCI) and normal control (NC). The aim of this research is to use new methods of extracting features from MRI images and selecting the most relevant features. The values of cortical thickness in regions of interest are used as a features for MRI images. The Support Vector Machine - Recursive Feature Elimination, Infinite Feature Selection, Relief-F, Feature Selection Via Concave Minimization and Support Vector Machine, Minimum Redundancy Maximum Relevance, Fisher, Unsupervised Discriminative Feature Selection, L0-norm Based Feature Selection, and Feature Selection via Eigenvector Centrality techniques for feature selection and ranking were used. In the experiment, the MRI images from ADNI database was used, which contains data from 539 subjects. In the result, the following maximum values of area under ROC curves for three datasets are obtained: NC-AD - 0.90, NC-MCI - 0.71, MCI-AD - 0.69. Classification accuracy with linear Support Vector Machine for the corresponding datasets: NC-AD - 80%, NC-MCI - 64%, MCI-AD - 65%. The results of this paper can be used for building a classifier in space of selected features.