Defining region-specific masks for reliable depth-dependent analysis of fMRI data

Abstract

In high-field fMRI research, anatomical reference information (e.g., gray matter (GM) segmentation, cortical depth delineation) is often defined in volumes acquired with pulse sequences subject to different distortions than those in functional volumes. In these cases, reliable interpretation of ultra-high resolution fMRI data depends on excellent cross-modal registration of functional volumes to reference anatomical volumes. In this paper, we describe a two-step approach to automating assessments of cross-modal registration quality for the purpose of guiding depth-dependent analysis. First, each functional/anatomical registration was scored by the ratio of the number of GM voxels in the functional data overlapping anatomical GM to the number of GM voxels in the functional data overlapping anatomical white matter (WM). This GM:WM overlap ratio provided an objective metric for determining whether an alignment algorithm had converged on a solution that would pass visual inspection. Second, surface-based maps indicating the consistency of overlap between functional and anatomical GM throughout the GM depth were generated for the entire region of cortex covered by the experiment. These maps served as a mask for the purpose of excluding regions where registration between functional and anatomical data was inadequate and thus unable to support depth-dependent analyses. We found, for both real and simulated data, that functional response profiles that were less biased toward superficial responses in regions where these metrics indicated satisfactory registration.