Topogical Correction of Subject-level Intrinsic Connectivity Networks

Noah Lewis, Armin Iraji, Robyn Miller, Vince Calhoun

Abstract: Over the last several decades, researchers have sought to capture the underlying functional activity of the human brain from functional magnetic resonance imaging (fMRI). One well-studied and promising avenue of research is independent component analysis (ICA) to capture the maximally independent set of elements known as functional networks. These functional networks are represented as spatial maps or 3D images in which each voxel has a value associated with the given network. The current state-of-the-art methods use group-level spatial maps as a reference to provide estimates of subject-level maps, which are vulnerable to the low contrast-to-noise (CNR) ratio of fMRI signals. However, such approaches do not account for all group-level spatial information. As such, their subject-level estimate is still quite noisy. This work presents a novel method that leverages the topological properties of the group maps to improve subject-level estimations. We show that adding topological similarity constraints also improves subject-specific information.