Artifacts due to B0 field off-resonance result in image distortions and blurring in non-Cartesian acquisitions. FieldMapNet is a learning-based method to map from each image in a spiral-in BOLD fMRI acquisition to a corresponding B0 fieldmap at that timepoint. We train FieldMapNet in a supervised fashion using custom data acquired at three echo times to generate ground truth dynamic fieldmaps. We then perform a tailored B0 correction at each fMRI timepoint using a model-based image reconstruction (MBIR). We show improved image space RMSE using FieldMapNet vs. a separately acquired GRE fieldmap, and a reduction in distortion artifacts and blurring.
This abstract and the presentation materials are available to members only; a login is required.