MRS provides a well-validated method of measuring liver fat fraction (FF%) in vivo [1] and more recently gradient echo (GE) MRI methods have also been used by fitting to various signal models [2]. For simplicity, the earliest fitting algorithms assumed a single fat peak [2], but a multiple fat peak model has been shown to provide improved fitting [3]. A common R2* is generally assumed for fat and water since dual-R2* models [4] suffer from poor noise performance, increased computational requirements and instability at 1.5 T [5]. As such, the consensus has been to use a multi-peak, single R2* model. This is surprising as MRS studies have shown that accurate liver FF% estimation requires T2 correction [1].To use GE-MRI and MRS liver data acquired before and after a hyper-energetic diet to assess the effect of T2 correction on GE MRI methods of estimating FF%.

20 overweight (BMI>25 kg/m2) healthy males were scanned before and after a two-week hyper-energetic (25% excess) diet, using a 3T Philips Achieva system with an XL torso coil.

MRS. Water suppressed and water unsuppressed MRS data were acquired from a 20x20x20mm³ voxel using STEAM at TE=20,30,40 and 60ms. Spectra were phase corrected and water and fat peak areas were measured. The fat-to-water ratio (R) at TE = 20ms was used to calculate FF%_MRSTE20. T2 values were determined from the multiple echo data and a correction factor was applied to provide a T2-corrected FF%_MRS.

MRI. 3D multi gradient echo images (FOV=400x400x200mm³, voxel=1.3mm) were acquired at 4 echo-times (TE₁=1.2ms,ΔTE=0.9ms), flip angle of 10⁰ and TR= 5.3ms [6]. The complex signal was fitted to 3 signal models shown in Figure 1 – a single peak model (SP), a multi peak model (MP) and newly proposed T2-corrected multi peak model (T2MP). Data was fitted using iterative least squares in Matlab [7] with noise bias resolved using magnitude discrimination [6]. The resulting values of FF% (Table 1) were averaged over the MRS ROI. The T2MP model used individual differences in 1/T2 between fat and water from MRS to estimate ΔR2* assuming that T2', which depends on field inhomogeneities, is equal for fat and water:

Equation 1: ΔR2* = 1/T2*_F -1/T2*_W = 1/T2_F + 1/T2' - (1/T2_W + 1/T2') = 1/T2_F - 1/T2_W

The inter-subject mean±SD measured from MRS was 38 ± 18 ms for fat and 24 ± 3 ms for water, and ΔR2* = 1.1 ± 2.0 x 10^-2 ms^-1. Figure 2 a and b show that FF%_SP and FF%_MP both correlated well with MRS, with FF%_MP giving less systematic difference (gradient ~1, intercept ~0, Fig 2b). When the MRS data was T2 corrected the correlation between FF%_MP and FF%_MRS reduced and the gradient of the correlation increased by approximately 33% (Fig 2c). The correlation between FF%_T2MP and T2-corrected MRS was 0.66 (Fig 2d) with less systematic difference (gradient ~ 1, y-intercept ~ 0).

Figure 3 shows that the change in average hepatic fat fraction between visit 1 and visit 2 increased significantly as measured using T2-corrected MRS (P ≤ 0.001). FF%_MP showed a smaller and non-significant increase (P = 0.7), whilst FF%_T2MP agreed between with FF%_MRS , but the change between visits was still non-significant (P = 0.1).

A previous 1.5T study found a reduced systematic error between MRI and MRS (TE = 25ms) for MP fitting compared with SP fitting [3]. However previous studies have shown that accurate measures of FF% using MRS require T2 correction of the spectra [1], suggesting that the good correlation and low systematic error between MRI and MRS results without T2 correction (Fig 2 a and b) are likely to result from similar systematic errors in both sets of data. Once the MRS data was T2-corrected the MRI data systematically overestimated (Fig 2c) in comparison. Correcting the MRI data for the change in between fat and water based on the spectroscopic T2 values (Eq. 1) reduced the systematic difference, but increased scatter. Notably, whereas the MRS data was able to detect a change in response to the two week diet, none of the MRI fitting methods currently being used were able to detect a significant change. Whilst the consensus has been to use single R2* fitting with a recent paper finding good accuracy at 1.5 T [5], the present study shows that dual R2* fitting is necessary at 3T.Despite advances in MRI water-fat separation techniques, MRS provides the most accurate method for measuring statistical changes in metabolic studies at 3T. Further developments in dual R2* MRI water-fat separation are needed.