Reducing a Localized Signal Fluctuation Artifact in fMRI using Spectral-Spatial Fat Saturation

INTRODUCTION The conventional fat saturation scheme for thin-slice fMRI is to use a 90° flip angle, spatially non-selective, 1D RF pulse at fat resonance frequency to rotate fat spins (or water spins under a B0 frequency offset that equals the fat-water frequency difference) in the whole 3D space to transverse plane. This scheme relies on a sufficiently strong following crusher gradient to remove the fat signals, which is an easy task when fat spins are at or close to the isocenter of the magnet. However, for locations further away from isocenter in the superior-inferior direction (e.g., neck or chest areas for brain fMRI), gradient linearity becomes increasingly worse and in some locations (e.g., >20 cm away from isocenter), gradients at some or all axes can become negligibly small, failing to crush out the fat spins in these areas. These fat spins produce artifactual signals that are aliased to contaminate the in-plane water signal. In the image domain, the artifactual signal affects one or multiple localized voxels because the imaging gradients that encode these out-of-plane signals have weak amplitude as well. The artifact can sometimes be seen as “clouds” in the background; more often, it overlaps with water signal and manifests itself as an additional temporal fluctuation in one or multiple voxels due to motion and/or system instability in the out-of-plane areas. For the latter case, the locations of voxels with the additional fluctuation are coherent through slices; therefore, one or multiple dark bands in the slice direction can appear in the sagittally or coronally reformatted signal-to-temporal-fluctuation-noise (SFNR) map. The additional temporal fluctuation can generate false result in the brain activation map and affect the accuracy of any subsequent data analysis. In this paper, we propose a spectral-spatial fat saturation method to remove the above-mentioned artifact.