A Novel Three-Phase Feature Selection Algorithm for Major Depressive Disorder Classification

摘要: Except for the “curse of dimensionality”, the challenge of inaccurate class labels is introduced in the high-dimensional feature selection (FS) problem in the realm of major depressive disorder (MDD) auxiliary diagnosis. This article proposed a novel three-phase FS algorithm, named aTPFS, for MDD classification using resting-state functional magnetic resonance imaging (R-fMRI) to address the above two challenges concurrently. aTPFS integrates relevance-guided clustering (Phase One and Phase Two) and a genetic algorithm (Phase Three). In the first two phases, features are sorted by their relevance and divided into feature clusters with low computational costs. After that, diverse representative feature subsets are generated for application across different samples, using an evolutionary algorithm in Phase Three. The efficacy of the proposed algorithm was verified on the multi-site REST-meta-MDD dataset in …