Fat suppression techniques are widely used in MRI to suppress undesired signal from adipose tissue. Fat suppression could be achieved by fat saturation, inversion-recovery imaging or combination of these two using specially designed RF pulses. However, these methods are very sensitive to B₀ or B₁ inhomogeneities and may fail to fully suppress the fat signal [1], especially when large eddy currents induced by large diffusion gradient occurs. The fat signal in water image could potentially lead to quantification errors in statistical image analysis based on the water signal, such as in fMRI and DW-EPI. In EPI, fat signal shifted along phase encoding direction and can be treated as another simultaneously excited slice with controlled aliasing together with the water image, similar with the case in simultaneous multi-slice imaging (SMS) [2]. Then this fat image can be separated using tradition parallel imaging reconstruction methods [3]. In this abstract, we propose to use parallel imaging method to separate water and fat signal in EPI, then we further apply this method to remove fat residue in fat-suppressed EPI and DW-EPI.

All experiments were carried out on a Philips Achieva 3T MRI scanner equipped with an 8-channel SENSE head coil and a 16-channel abdomen coil. In the first experiment, single-shot spin echo EPI brain images were acquired with and without fat suppression to demonstrate the feasibility of this method. In the second experiment, fat-suppressed EPI and DW-EPI images were acquired in both brain and abdomen (with breath-hold). Scan parameters for brain were: TE/TR = 60/1000 ms, FOV = 230 × 230 mm², matrix size = 92× 92, echo spacing = 0.266 ms, slice thickness = 4 mm, and two b values (0, 200s/mm²). Scan parameters for abdomen imaging were: TE/TR = 13.7/407ms, FOV = 368 × 300mm², half-scan factor = 0.6, matrix size = 92×52, echo spacing = 0.266 ms, slice thickness = 8 mm and two b values (0, 200s/mm²).

At 3T, fat signal shifts N (= 440Hz × echo spacing × number of pixels in phase encoding direction) pixels relative to water image in EPI. For signal intensity S₁ in coil C₁ containing water and fat signal can be formulated as: S₁=C₁₁W₁+C₁₂F₁, where W₁ and F₁ are water and fat signal, C₁₁ is coil sensitivity at this location and C₁₂ represents the shifted fat coil sensitivity. These equations can be solved via linear inversion by SENSE [4]. For simplicity, coil sensitivity maps were generated from fat-suppressed EPI images by dividing image in each channel by their root-sum-of-square combination. Note that fat-suppressed EPI is not a prerequisite for coil sensitivity map generation.

Figure 1 shows multi-slice brain EPI images without fat suppuration and the water and fat images resolved by SENSE. The water and fat signal were separated without visually detected image quality degradation. Figure 2 shows multi-slice brain EPI and DW-EPI with fat suppression by SPIR. Fat residue was observed in subcutaneous region and effectively extracted with little water signal leaked into fat image. Figure 3 shows water and fat residue resolved by SENSE in multi-slice abdomen fat-suppressed EPI and DW-EPI, which also shows clean fat residue removal from the water image. This study has demonstrated that based on intrinsic chemical shift of water and fat, they can be separated using parallel imaging methods in EPI. This is particularly suitable for removing fat residue in fat suppressed EPI images. This method treats the EPI images with fat residue as simultaneous excited water and fat images, which are subsequently separated by SENSE method. The failure of fat suppression usually occurs when the main field is highly inhomogeneous or there exists eddy currents induced by large diffusion gradient, both of which can also induce geometric distortion in EPI type images. So the coil sensitivity maps need to have the same geometry in order to accurately separate fat and water image. In theory the SNR of separated images will decrease by parallel imaging reconstruction, but could be largely preserved by introducing regularization terms based on knowledge about water and fat distribution in the image. Also the decrease of SNR will be smaller if the fat shift is larger, as in SMS. This method is potentially more powerful in high-filed MRI, where the field inhomogeneities are a problem but the fat image has a larger shift to be effectively separated. In conclusion, we have presented a simple method to separate water and fat in EPI images and successfully applied it to remove fat residue in the brain and abdomen fat-suppressed EPI and DW-EPI images.