In quantitative studies of fat infiltration in skeletal muscle, such as studies of e.g. diabetes¹, chemical shift-encoded imaging is commonly used. However, this technique requires short echo spacing to be reliable, especially at 3T, which is in contradiction to the high spatial resolution and high matrix size needed to resolve the detailed fat patterns in the muscle. The purpose of this study is to use a fat quantification technique based on T2-relaxation², and evaluate its accuracy and precision at both low and high spatial resolution compared to chemical shift-encoded imaging. Fat quantification based on the different T2 relaxation times of water and fat, rather than the difference in resonance frequency, may be a way to obtain very high resolution images of fat fraction (FF), in turn improving the differentiation between intermuscular and intramuscular fat.

A phantom was constructed using 6 vials filled with different ratios of water and Intralipid fat emulsion (20%, Fresenius Kabi) to achieve a range of FFs (0, 5, 10 and 20 %). In vivo data was acquired from a calf of a heathy volunteer. Two sequence types were used for phantom and in vivo data on a 3T scanner (Trio, Siemens Healthcare): multi spin echo (MSE) for the T2-based method and multi gradient echo (MGE) for the chemical shift-based method. Images were acquired at two matrix sizes (128x128 and 512x512). The following settings were used for the MSE-acquisitions: TE1/ΔTE/TR (phantom) = 11.7/11.7/2000 ms, TE1/ΔTE/TR (in vivo) = 9.20/9.20/2000 ms and BW₁₂₈/BW₅₁₂ = 425/391 Hz/px. The settings for the MGE-acquisitions were: TE1₁₂₈/ΔTE₁₂₈/TR₁₂₈ = 1.34/1.55/500 ms, TE1₅₁₂/ΔTE₅₁₂/TR₅₁₂ = 2.57/3.92/500 ms, BW₁₂₈/ BW₅₁₂ = 1776/651 Hz/px and flip angle = 12. Slice thickness and FOV were set to 5 mm and 280x280 mm² for all acquisitions.

FF was calculated from MSE-data by fitting the linear parameters W (water) and F (fat) in the signal equation $$$S = We^{-t/T2_{W}}+Fe^{-t/T2_{F}}$$$, using fixed values of T2_F and T2_W². The T2-values were obtained by drawing regions of interest (ROIs) in a T2-map, calculated from a monoexponential fit in each voxel. In the phantom measurement, the T2 values of water and Intralipid were used. Therefore, the FF map of the phantom represented the Intralipid fraction and were divided by five to get the actual FF values, knowing the true FF of undiluted Intralipid. In the in vivo case, ROIs were drawn in subcutaneous fat and in muscle tissue. An iterative least squares procedure, correcting for T2* and off resonance and including a multipeak fat model, was used to separate the fat and water signals from MGE data^3,4. To study the effect of deviations between assumed and actual T2-values on the estimation of FF, simulations were carried out.

The resulting FF maps of the phantom and calf using both methods are presented in Figure 1 and Figure 2, respectively.

In Figure 3, the mean estimated FFs are plotted against the known FFs in the phantom. Chemical shift-encoded imaging underestimates the FF at higher resolution while the T2-based method estimates the FF accurately. At the smaller matrix size and lower resolution, both methods perform well. The FF variation within the vials were markedly larger for chemical shift-encoded imaging than for the T2-method at the higher matrix size.

The estimated FFs in three muscle ROIs (tibialis anterior, soleus and gastrocnemius) are presented in Figure 4. Compared to the FFs obtained using chemical shift-encoded imaging at low resolution, the other approaches either under- or overestimates the FF.

The simulated errors in FF due to inaccurate T2_Fat and T2_Muscle are presented in Figure 5. These results suggest that it is more important to estimate T2_Muscle accurately compared to T2_Fat at the relevant FF (muscle FF ≈ 5%).

At higher resolution, chemical shift-encoded imaging underestimates the FF in phantom and in vivo whereas the T2-based method accurately estimates the FF in phantom but overestimates it in vivo. It is important to accurately estimate T2, especially of the muscles, to get a correct FF estimation. This, however, is not always easy to achieve with the presence of other not visible components within the ROI, e.g. fat. This may in part explain why the T2-method overestimates FF in vivo at high resolution, and why the T2-method at low resolution performs poorly in some muscle ROI. At low resolution, the FF calculation is also more sensitive to the effect of Gibbs artifacts which results in inferior performance.The phantom results show that the T2-based method can achieve an accurate estimation of FF at both high and low resolution, compared to chemical shift-encoded imaging.