Abstract
One of the major advantages of whole brain fMRI is the detection of large scale cortical networks. Dependent Components Analysis (DCA) is a novel approach designed to extract both cortical networks and their dependency structure. DCA is fundamentally different from prevalent data driven approaches, i.e. spatial ICA, in that instead of maximizing the independence of components it optimizes their dependency (in a tree graph structure, tDCA) depicting cortical areas as part of multiple cortical networks. Here tDCA was shown to reliably detect large scale functional networks in single subjects and in group analysis, by clustering non-noisy components on one branch of the tree structure. We used tDCA in three fMRI experiments in which identical auditory and visual stimuli were presented, but novelty information and task relevance were modified. tDCA components tended to include two anticorrelated networks, which were detected in two separate ICA components, or belonged in one component in seed functional connectivity. Although sensory components remained the same across experiments, other components changed as a function of the experimental conditions. These changes were either within component, where it encompassed other cortical areas, or between components, where the pattern of anticorrelated networks and their statistical dependency changed. Thus tDCA may prove to be a useful, robust tool that provides a rich description of the statistical structure underlying brain activity and its relationships to changes in experimental conditions. This tool may prove effective in detection and description of mental states, neural disorders and their dynamics.
