Introduction

Fat suppression is particularly challenging at low field strength due to the relatively small frequency shift of fat (~80Hz at 0.55T compared to ~220 Hz at 1.5T). At low field strength, Dixon imaging with balanced steady-state free precession (bSSFP) sequence may be an appealing choice because it can combine the advantage of high SNR and scan speed of bSSFP with the reliable fat-water separation performance of Dixon¹. bSSFP signal presents a distinctive spectral response characterized by periodic “pass bands” of signal magnitude and phase offset between bands², whereas current implementations of Dixon methods¹ only assume uniform spectral response. This study proposed to adapt a two-point Dixon algorithm¹ to bSSFP signal model at 0.55T.

Methods

Modified two-point Dixon framework for bSSFP: In the original framework proposed by Berglund et al¹, the following signal functions were used:
$$S_1=(W+\ a_1\ F)b_0 \ \ \ \ $$ $$S_2=(W+\ a_2\ F)b_0\ b \ \ [1]$$
, where b₀ and b represent the water signal phase accumulated at TE1 and during ∆TE due to static field inhomogeneity (noted as ΔB₀ in the following), and a₁ and a₂ represent the phase accumulation of fat relative to water.
Assuming a uniform spectral response such as in the GRE or TSE signal model, a₁ and a₂ can be readily calculated using the chemical shift and relative amplitude of multiple fat peaks:
$$a_{1,2}=\sum_{p=1}^P \rho_p e^{i2\pi f_p\ TE_{1,2}} \ \ \ \ [2] $$
In the case of bSSFP, the multiple fat peaks lie in different “bands”, having different magnitude and (more importantly) phase response (Figure 1). The combined fat signal will accumulate phase differently from what was described in Equation [2]. To account for the bSSFP signal model, the following calculation of a₁ and a₂ are proposed:
$$a_{1,2}=e^{-i\varphi_{bSSFP}(f_m,\ TE_{1,2})}*\sum_{p=1}^P \rho_p e^{i\varphi_{bSSFP}(f_m+f_p,\ TE_{1,2})} \ \ \ \ [3] $$
, where $$$\varphi_{bSSFP}$$$ is the bSSFP signal phase response, which is a function of precession frequency. $$$f_m$$$ is the off resonance frequency of water caused by ΔB₀, and $$$f_p$$$ is the frequency shift of fat.

Numerical simulation: A numerical phantom was created, in which different water-fat mixtures were placed in a static background field that varies from -150 Hz to 100 Hz (Figure 2A). bSSFP steady state multi-echo signals were simulated using Bloch equations³ with the following sequence parameters: B0 = 0.55T, flip angle = 70°, TR/TE1/TE2 = 8.6/2.5/6.1 ms. The fat signal was modeled as six peaks⁴. Assuming the true ΔB₀ field was known, the derived water and fat images were compared when Equation [2] or [3] was applied.

Phantom experiment: Phantom experiments were performed on a 0.55T MRI scanner (MAGNATOM Free.Max, Siemens Healthcare, Erlangen, Germany). In one experiment, two bottles one filled with water and the other with oil were scanned using a standard bSSFP sequence with two-echo monopolar readouts (TR/TE1/TE2 = 8.6/2.5/6.1 ms, flip angle = 70). A shim gradient was applied during imaging to generate a significant field of ±250 Hz. Keeping the same shimming condition, the phantom was scanned again using a multi-echo GRE sequence to measure the ΔB₀ field. In another experiment, a meat phantom was scanned in a similar way with the same sequence parameters. Resulting water and fat images were compared when Equation [2] or [3] was applied. Fat suppression was evaluated using a “leakage index“ defined as the ratio of signals in the fat-suppressed and water-dominant regions in the water image.

Results

Numerical simulation (Figure 2) showed better fat suppression could be achieved when bSSFP signal model compared to GRE signal model was used in the two-point Dixon algorithm. The difference was most significant in the region of pure fat. Results of phantom experiments also showed cleaner fat suppression when bSSFP signal model was applied (Figure 3 and 4), although string-like artifact was observed at the locations where the fat-water relative phase map have discontinuities.

Discussion

This study has demonstrated that better fat suppression can be achieved in bSSFP-based two-point Dixon at 0.55T when bSSFP signal model was considered. This aligns with previous findings at 1.5T⁵. The study only investigated the effect of bSSFP signal response in one specific two-point Dixon algorithm¹. However, such effect may be different in other kinds of Dixon methods, such as classification-based algorithms. Another limitation of this study is that ΔB₀ map was obtained through a separate scan, which may be impractical for in-vivo scan. Future studies will aim to obtain accurate ΔB₀ distribution from the two-echo bSSFP data itself, potentially through iterative algorithms⁶.

Acknowledgements

No acknowledgement found.