ICA-based connectivity on brain networks using fMRI

This study introduces a novel data-driven approach for constructing large-scale functional brain networks. These networks are constructed by converting raw functional magnetic resonance imaging data into graphs using independent components analysis (ICA). Empirical evaluations were performed using data collected from three sites, which are part of a pediatric epilepsy consortium. The test data contained 30 control subjects and 29 pediatric epilepsy patients all of which were performing an auditory decision descriptive task, a language task paradigm. This approach is augmented by a unique graph thresholding technique based on the graph density function. The constructed networks were then analyzed using graph theoretical measures. The proposed network construction approach is weighed in merit to the traditional correlation approach and a modified version of it. The obtained results show that the ICA-based approaches improve considerably the delineation process of the patients' population from the controls' population, whereas the traditional methods show considerable overlap between the two populations. Furthermore, an investigation on the topology of the networks constructed show that all methods lead to a small-world topology conforming to previous brain functional studies.