The female brain contains a larger proportion of gray matter tissue, while the male brain comprises more white matter. Findings like these have sparked increasing interest in studying dimorphism of the human brain: the general effect of gender on aspects of brain architecture. To date, the vast majority of imaging studies is based on unimodal MR images and typically limited to a small set of either gray or white matter regions-of-interest. The morphological content of magnetic resonance (MR) images, however, strongly depends on the underlying contrast mechanism. Consequently, in order to fully capture gender-specific morphological differences in distinct brain tissues, it might prove crucial to consider multiple imaging modalities simultaneously. This study introduces a novel approach to perform such multimodal classification incorporating the relative strengths of each modality-specific physical aperture to tissue properties. To illustrate our approach, we analyzed multimodal MR images (T(1)-, T(2)-, and diffusion-weighted) from 121 subjects (67 females) using a linear support vector machine with a mass-univariate feature selection procedure. We demonstrate that the combination of different imaging modalities yields a significantly higher balanced classification accuracy (96%) than any one modality by itself (83%-88%). Our results do not only confirm previous morphometric findings; crucially, they also shed new light on the most discriminative features in gray-matter volume and microstructure in cortical and subcortical areas. Specifically, we find that gender disparities are primarily distributed along brain networks thought to be involved in social cognition, reward-based learning, decision-making, and visual-spatial skills.
Copyright © 2013 Elsevier Inc. All rights reserved.