Non-separable spatiotemporal deconvolutions improve decoding of neural activity from fMRI signals

The goal of many functional Magnetic Resonance Imaging (fMRI) studies is to infer neural activity from hemodynamic signals. Classical fMRI analysis approaches assume that the hemodynamic response function (HRF) is identical in every voxel, i.e. it is separable in voxel-space and time. This study demonstrates to our knowledge for the first time directly that although the nonseparable part is small, it significantly improves the decoding performance of intracortical neural signals from multivariate fMRI time series. Our results confirm previous findings using non-canonical HRFs and demonstrate that there is more neural information in fMRI signals than detected by classical analysis methods.