Exploratory factor analysis of brain networks reveals sub-networks related to cognitive performance

Properties of the brain´s structural networks can be analyzed by applying fiber-tracking techniques and network analysis to diffusion MRI. Here we applied exploratory factor analysis (EFA) to anatomical connectivity matrices, to identify brain networks whose properties predicted higher-order cognitive function. Using diffusion MRI scans from 104 healthy young adults, we computed connectivity matrices based on deterministic and probabilistic tractography (with the FACT and Hough transform methods). Both sets of matrices were submitted to factor analysis, to identify sub-networks relevant for predicting cognitive function. The Kaiser-Meyer-Olkin measure and Bartlett´s sphericity test were used to recover latent factors from the connectivity matrices, and only the Hough method yielded factorable outputs. Factor scores were related to fluid, crystallized, and spatial intelligence, and processing speed. Middle temporal and lateral prefrontal connectivity measures predicted all cognitive scores, except spatial intelligence. Cognitive performance was not predictable from global connectivity measures, which depended on the tractography method.