Title: Leveraging spatial dependencies on the cortical surface to improve estimation of subject-level brain organization
Authors: Amanda Mejia - Indiana University (United States) [presenting]
Abstract: A primary objective in many functional magnetic resonance imaging (fMRI) studies is localization of functional areas, regions of the brain exhibiting synchronous activity. This is true of both task and resting-state studies, where a goal is to identify regions that coactivate in the absence of a specific task. fMRI data is composed of small voxels, whose contributions to a functional area form a spatial field. These fields exhibit strong spatial dependence, since neighboring voxels tend to exhibit similar behavior. However, models used to estimate these functional areas have often considered voxels to be independent, resulting in loss of efficiency and power. Spatial Bayesian models have been proposed as a way to account for spatial dependence, but the complex dependence structure of fMRI is difficult to accurately represent with models simple enough to be tractable in high dimensions. A promising alternative is to use cortical surface fMRI (cs-fMRI), which projects the cortical gray matter to a triangular mesh manifold surface, where the spatial dependence structure is simplified. We propose leveraging spatial dependencies along the cortical surface through a Bayesian modeling framework with stochastic partial differential equation (SPDE) spatial process priors, which are built on a triangular mesh. We demonstrate this approach through task and resting-state fMRI studies and quantify its benefits through reliability studies using data from the Human Connectome Project.