Optimal sampling geometries for TV-norm reconstruction of fMRI data

This study explores the ability to reconstruct functional magnetic resonance imaging (fMRI) brain slices from a limited number of K-space samples. We use compressed sensing methods to reconstruct brain imaging activity using different K-space sampling geometries. To determine the optimal sampling geometry, we compute the reconstruction error. Here, for each geometry, we also estimate the optimal weighting parameters for the total variation (TV) norm and L-2 norm penalty functions. Initial results show that the optimal sampling geometry varies significantly as a function of the required reduction in K-space sampling density (for 60% to 90% reduction). Furthermore, the reconstructed fMRI slices can be used to accurately detect regions of neural activity from a largely reduced number of K-space samples.