Abstract
PURPOSE: To improve the robustness of existing chemical shift encoding‐based water–fat separation methods by incorporating a priori information of the magnetic field distortions in complex‐based water–fat separation. METHODS: Four major field contributions are considered: inhomogeneities of the scanner magnet, the shim field, an object‐based field map estimate, and a residual field. The former two are completely determined by spherical harmonic expansion coefficients directly available from the magnetic resonance (MR) scanner. The object‐based field map is forward simulated from air–tissue interfaces inside the field of view (FOV). The missing residual field originates from the object outside the FOV and is investigated by magnetic field simulations on a numerical whole body phantom. In vivo the spatially linear first‐order component of the residual field is estimated by measuring echo misalignments after demodulation of other field contributions resulting in a linear residual field. Gradient echo datasets of the cervical and the ankle region without and with shimming were acquired, where all four contributions were incorporated in the water–fat separation with two algorithms from the ISMRM water–fat toolbox and compared to water–fat separation with less incorporated field contributions. RESULTS: Incorporating all four field contributions as demodulation steps resulted in reduced temporal and spatial phase wraps leading to almost swap‐free water–fat separation results in all datasets. CONCLUSION: Demodulating estimates of major field contributions reduces the phase evolution to be driven by only small differences in local tissue susceptibility, which supports the field smoothness assumption of existing water–fat separation techniques.