Diffusion-adapted spatial filtering of fMRI data for improved activation mapping in white matter

Abstract: Brain activation mapping using fMRI data has been mostly focused on finding detections in gray matter. Activations in white matter are harder to detect due to anatomical differences between both tissue types, which are rarely acknowledged in experimental design. However, recent publications have started to show evidence for the possibility of detecting meaningful activations in white matter. The shape of the activations arising from the BOLD signal is fundamentally different between white matter and gray matter, a fact which is not taken into account when applying isotropic Gaussian filtering in the preprocessing of fMRI data. We explore a graph-based description of the white matter developed from diffusion MRI data, which is capable of encoding the anisotropic domain. Based on this representation, two approaches to white matter filtering are tested, and their performance is evaluated on both semi-synthetic phantoms and real fMRI data. The first approach relies on heat kernel filtering in the graph spectral domain, and produced a clear increase in both sensitivity and specificity over isotropic Gaussian filtering. The second approach is based on spectral decomposition for the denosing of the signal, and showed increased specificity at the cost of a lower sensitivity.