Single-point Dixon methods attempt to separate water and fat signal from a single image only. They are fundamentally limited in their ability to resolve mixed water and fat signal without additional a priori information, for instance on the present main field inhomogeneity. Nevertheless, single-point Dixon methods seem attractive, since they promise a substantial reduction in scan time compared to multi-point Dixon methods. This holds in particular for applications that are on the one hand time-critical and on the other hand undemanding as to the accuracy of the separation. Among these is notably contrast-enhanced MRA, in which a reasonable fat suppression is often sufficient. The purpose of this work was to improve single-point Dixon imaging in terms of SNR and CNR primarily in this application.The measured signal S is modeled in image space as
$$S=(W+cF)p,$$
where W and F denote the water and fat signal, c a factor describing the evolution of a pure fat signal in the absence of main field inhomogeneity, and p a phase error. p is derived either from S, assuming that either water or fat is present in individual voxels, or from other sources, such as an additional reference scan^1,2. W and F are then obtained by
$$F=Im\{S/p\}/Im\{c\},$$
$$W=Re\{S/p\}-F\cdot Re\{c\}.$$
Setting $$$c=e^{2\pi f_F \text{TE}}$$$, where f_F denotes the resonance frequency offset of fat relative to water and TE the echo time, the SNR in the water image is reduced by a factor of $$$sin(2\pi f_F \text{TE})$$$ relative to the SNR in the measured image.
To avoid loss of water signal, in particular for $$$Re\{c\}<0$$$, it is proposed to impose non-negativity constraints on W and F, for instance by
$$F=max\{Im\{S/p\}/Im\{c\},0\},$$
$$W=max\{Re\{S/p\}-F\cdot Re\{c\},0\}.$$
Moreover, to limit loss of SNR, especially for less favorable TEs, it is suggested to replace W or F by S if the estimated W and F indicate a strong dominance of one of the signal, for example by
$$F'=abs\{S\}, W'=0,$$
if $$$F>t W$$$, and
$$F'=0, W'=abs\{S\},$$
if $$$W>t F$$$, with a suitable threshold $$$t \geq 1$$$.The described modifications were evaluated in subtractionless first-pass peripheral MRA³. Data were acquired on patients on a 1.5 T Ingenia scanner (Philips Healthcare, Best, Netherlands). After injection of 0.1 mmol/kg Gadobutrol (Bayer Healthcare, Berlin, Germany), contrast-enhanced images were collected with a 3D T₁-weighted spoiled dual-gradient-echo sequence (TE₁/TE₂ = 1.8 ms/3.0-3.2 ms) at three stations. For the purpose of this work, only the images from the first gradient echo were used.A selected slice of one of the contrast-enhanced images is shown in Fig. 1. p was derived from the image itself in this case. The division by p removed the phase variations induced by main field inhomogeneity reasonably well, except in the heels. The application of the original single-point Dixon method to the corrected image yielded the water and fat images in Fig. 2. Due to the TE of 1.8 ms, at which the dephasing of water and fat signal amounts to about 2.5 rad or 140°, the SNR is noticeably reduced, especially in the water image. Corresponding water and fat images obtained with the modified single-point Dixon method are provided in Fig. 3. The SNR is evidently improved, as is the vessel-to-background contrast or CNR. The latter is particularly striking in the MIP and the difference images in Fig. 4. Due to the presence of contrast agent and of flow, the phase in the vasculature may differ from the phase in surrounding non-fatty tissue. This may lead to negative fat signal and reduced water signal in the vasculature using the original single-point Dixon method, whereas the proposed non-negativity constraints effectively prevent this. The choice of the TE in single-echo Dixon imaging is typically more critical to attain an adequate SNR than the choice of the TEs in multi-echo Dixon imaging, which compromises the expected reduction in scan time. The suggested selective, direct propagation of the measured signal to the water or fat signal may avoid the SNR penalty associated with less favorable TEs to a large extent and thus provide more flexibility in optimizing the employed sequence.