Most water-fat separation methods based on chemical shift effect require multiple image acquisitions at different echo times, which prolong the total scanning time. Fast/turbo SE-based Dixon implementations are introduced to address long imaging time and multiple measurements¹, but these techniques are still inefficient for 3D volumetric imaging due to signal modulation along the echo train. Recently, to resolve aforementioned problems, variable-flip-angle (VFA) fast/turbo SE is developed². In addition, partial Fourier and/or parallel imaging techniques are incorporated with VFA fast/turbo SE imaging to speed up acquisition time but directly trade off with signal-to-noise ratio (SNR). To avoid multiple measurements and to tackle spatially variant noise amplification, we develop a novel water-fat separation method employing: 1) single-slab 3D VFA GRASE using phase-encoding blips for imaging time efficiency³, 2) phase-independent reconstruction exploiting spatially complementary information along the echo direction³, and 3) phase-corrected water-fat separation method using robust field distribution.

The proposed 3D VFA GRASE for rapid water-fat separation is shown in Fig.1, consisting of spatially non-selective excitation RF pulse followed by short, non-selective refocusing pulse trains with variable flip angles based on tissue-specific prescribed signal evolution, multiple readout gradients and phase-encoding blips like EPI fashion. Multiple echoes are shifted to -5π/6, π/2, and 11π/6 relative to the spin echo, which yield maximum SNR for all combination of water and fat within a voxel in fast/turbo SE imaging⁴. Each echo is grouped, and then phase-independent reconstruction is exploited to suppress spatially variant noise amplification and to preserve its phase information³. Prior to applying the water-fat separation algorithm, phase errors caused by switching the readout gradient polarities are estimated from projection data acquired in the phase-correction module (Fig.1), consisting of: 1) the polarities of the readout gradients in the phase-correction module are the same as those in the VFA GRASE imaging module, 2) the polarities of the readout gradients in the phase-correction module are flipped relative to the readout gradients in the VFA GRASE imaging module, and 3) turning off the phase-encoding gradient. For n^th echo, the estimated phase error, $$$\phi_{n}(x)=angle(pc_{1}g_{n}(x)/pc_{2}g_{n}(x))$$$, where x indicates the image domain, is fit to a polynomial, and then is subtracted from each echo image to align each k-space central position. In water-fat separation algorithm, robust field distribution is estimated using combined approach of region-growing and multi-scale method, wherein region-growing at the coarsest resolution and propagating the resulting estimates to the finer resolutions⁵. Thus, water and fat separated images can be obtained from multiple echo images demodulated field distribution.

Knee data was acquired in a healthy volunteer at 3T whole-body MR scanner (MAGNETOM Trio, Siemens Medical solutions) using an 8-channel transmit-receive knee coil. High resolution T₂-weighted knee images in the sagittal orientation with pseudo-linear reordering using conventional single-slab 3D VFA fast/turbo SE and the proposed 3D VFA GRASE to show the effectiveness of the latter in increasing time-efficiency over the former. The common imaging parameters were: TR/TE_eff = 2000ms/61ms, FOV = 153x153mm², in-plane matrix = 256x256, number of partitions = 144, readout bandwidth = 849Hz/Pix, echo spacing = 8.12ms, echo train length = 40, and shifting time with respect to the spin echo time = -0.4ms/1.2ms/2.8ms. Those specific to the proposed method were: EPI factor = 3, measurement = 1, number of self-calibrating signals for phase-independent reconstruction = 32x32, and imaging time = 8min 37sec. Those specific to the conventional method were: EPI factor = 1, measurement = 3, and imaging time = 75min. Those specific to the accelerated version of the conventional method were: EPI factor = 1, measurement = 3, subsampling factor in k_y-k_z space = 3, number of self-calibrating signals for conventional parallel imaging = 32x32, and imaging time = 25min 51sec. For PD-weighted knee images, center in-out reordering is exploited and imaging parameters are same as that for T₂-weighted knee imaging except for TR/TE = 1700ms/24ms. Water-fat separation without phase-correction to VFA GRASE leads to failed separation while complete water and fat separated images are achieved by applying phase-correction in readout direction (Fig.2). Fig.3 and Fig.4 represent T₂- and PD-weighted knee water-only images, respectively. Fig.3 and Fig.4 demonstrate the effectiveness of the proposed VFA GRASE over the conventional method in imaging efficiency without apparent artifacts and noise amplification. Imaging time of the proposed method is reduced 3-fold and 9-fold than that of version of the conventional method with and without acceleration, respectively.

The proposed 3D VFA GRASE effectively improves imaging efficiency without apparent loss of signal and image contrast to obtain water-fat separated imaging.