Statistical feature selection and classification models for Alzheimer's disease progression assessment

Alzheimer's disease (AD) is a progressive course neurodegenerative disease and the most common cause of dementia. Prediction of its progression from mild-cognitive impairment (MCI) to AD is still a challenging problem. This paper shows a system using tissue-segmented T1 magnetic resonance imaging (MRI) and neuropsychological tests including Mini Mental State Examination (MMSE) and cognitive assessment subscale (ADAS-Cog) to predict whether an MCI patient will evolve to AD within six months. Three alternative MRI feature extraction methods are considered: i) computation of brain region statistics by means of a regional atlas defining 116 regions of interest (ROIs), ii) voxel-based principal component analysis (PCA), and iii) voxel-based partial least squares (PLS). A t-test feature selection method is used to identify the most discriminant features for MCI to AD prediction that serve as input to a support vector machine (SVM) classifier that groups MCI subjects remaining stable during six months from those who progress to AD.