Dixon methods commonly rely on the assumption of smooth spatial variations of the main field within the FOV to resolve the inherent ambiguity in the separation of water and fat signal from the sampled composite signal. However, this assumption is typically violated near large susceptibility gradients and at larger distance from the isocenter of the magnet, where existing Dixon methods therefore tend to fail. To address the former areas, previous work suggested a simulation of susceptibility-induced main field inhomogeneity based on a patient model¹. In this work, an incorporation of prior knowledge of magnet imperfections in Dixon methods is proposed to improve the robustness of the separation in the latter areas.In theory, the magnet of an MR scanner is supposed to generate a static, homogeneous main field of a certain strength. In practice, however, it only achieves this approximately within a sphere or ellipsoid around its isocenter. Prior knowledge of spatial variations commonly exists in form of a model of the main field a particular type of magnet is designed to produce. Such a model characterizes the main field inhomogeneity caused by the magnet only. It does not describe further distortions of the main field introduced by susceptibility, i.e. the patient. Similar models of the gradient fields are routinely used in image reconstruction for distortion correction. In this work, a model of the main field is employed in image reconstruction to predict the main field inhomogeneity ΔB₀, to translate it to an offset of the resonance frequency Δf = γΔB₀/2π and an associated phase error 2πΔfTE at an echo time TE, and to correct the single-echo images for this phase error before applying existing Dixon methods for the separation of water and fat signal to them. Known, rapid spatial variations of the main field are thus eliminated, and the assumption of slow, smooth spatial variations is rendered more valid again.
This approach was evaluated in subtractionless first-pass peripheral angiography based on dual-echo Dixon imaging², which demands a high reliability of the separation over large FOVs. Experiments were performed on a 1.5 T Ingenia scanner (Philips Healthcare, Best, Netherlands). Patients were imaged after injection of 0.1 mmol/kg Gadobutrol (Bayer Healthcare, Berlin, Germany) with a 3D T₁-weighted spoiled dual-gradient-echo sequence (TE₁/TE₂ = 1.8 ms/3.0-3.2 ms) at three stations. Water-only images were reconstructed using mDIXON³.The offset of the resonance frequency Δf predicted for an off-center coronal slice with a FOV of 430 x 352 mm² using a model of the main field of the employed magnet is shown in Fig. 1. The main field varies only slowly within the homogeneity ellipsoid of the magnet, but rapidly beyond. The significant offsets in the corners of the FOV obviously correlate with the substantial spatial variations of the phase seen in the gradient-echo images in Fig. 2. This suggests that magnet imperfections are the primary cause of main field inhomogeneity in these areas. The improvements achieved in the separation with the proposed phase correction are illustrated in Figs. 3 and 4. Swapping artifacts occurring in the corners of the FOV are almost entirely eliminated. In this way, fat signal interfering with the visualization of the vasculature in the MIP is suppressed.The proposed incorporation of prior knowledge of main field inhomogeneity in Dixon methods permits using FOVs that exceed the homogeneity sphere or ellipsoid of the employed magnet. This is essential in subtractionless first-pass peripheral angiography, since a large virtual FOV has to be covered with as few stations as possible and hence relevant vasculature has to be imaged far from the isocenter of the magnet. Applications like off-center and whole-body imaging are expected to benefit similarly. The described approach obviously lends itself to a combination with other sources of prior knowledge of main field inhomogeneity, such as a separate B0 mapping or a simulation of susceptibility-induced distortions of the main field.