Evolutionary ELM wrapper feature selection for Alzheimer's disease CAD on anatomical brain MRI

This paper proposes an evolutionary wrapper feature selection using Extreme Learning Machines (ELM) as the base classifier training algorithm, comprising a Genetic Algorithm (GA) exploring the space of feature combinations. GA fitness function is the mean accuracy of a cross-validation evaluation of each individual feature selection. The marginal distribution of the classification accuracy corresponding to a feature is used to measure feature saliency. The raw features are extracted as a voxel selection from anatomical brain magnetic resonance imaging (MRI). Voxel selection is provided by Voxel Based Morphometry (VBM) which finds statistically significant clusters of voxels that have differences across MRI volumes on a paired dataset of Alzheimer's Disease (AD) and healthy controls.