Advanced MRI techniques estimate proton density fat fraction (PDFF) by acquiring multi-echo, gradient-echo images to correct for R2*, with low flip angle and long TR to minimize T1 weighting. Images are analyzed with a multi-peak fat spectral model to correct for multi-frequency interference effects. These techniques assume that R2* of fat and water are identical. While this assumption permits accurate PDFF estimation, it has not been directly proven. As demonstrated by MRS, T2 of fat and water differ¹, suggesting that R2* of fat and water also may differ. The purpose of this study was to compare the R2* of fat and water in phantoms, and in adults with fatty liver disease. We used MRS to measure R2* of water and fat, since MRS permits reliable measurement of these values, while MRI estimates may be unstable.

Figure 1 shows that fat peaks are not simple singlets, but have complex structure due to j-coupling. The fat spectral model used by MRI and MRS to estimate PDFF in human liver in vivo neglects this complexity and instead assumes each peak is a broad singlet. This assumption causes the effective R2* (R2*^eff) of fat to be greater than the true R2* of fat, since R2* is correlated to peak-width. In this work, we use MRS to measure fat R2*^eff, rather than true fat R2*, since fat R2*^eff is the relevant parameter for PDFF techniques. Water is a single peak; hence, for water, R2* and R2*^eff are the same.

¹H STEAM MR spectra were acquired at 3 Tesla (GE Signa EXCITE HDxt, GE Healthcare, Waukesha, WI).

For phantom studies, a fat-water emulsion was scanned using a quadrature extremity coil. A 15x15x15 mm voxel was selected in the center of the phantom and, after auto-shimming a spectrum was acquired (TR 5,000 ms TE 10 ms, TM 5 ms, nsa 4). Additional spectra were then acquired with identical parameters, except with increasingly worse shim to increase magnetic inhomogeneity and hence R2* of fat and water. This allowed the change in fat R2*^eff with water R2* to be examined graphically.

For the human study, STEAM spectra were acquired using an 8-channel torso array coil from 46 adults with fatty liver disease, with MRS-determined PDFF > 5%. Subjects PDFF < 5% were excluded as fat R2*^eff cannot be measured when PDFF is that low. A 20x20x20 mm voxel was selected within the liver that avoided liver edges and large biliary or vascular structures. Following a single pre-acquisition excitation, five spectra (TR 3,500 ms, TM 5 ms) were acquired with a single average at progressively longer TEs of 10, 15, 20, 25 and 30 ms in a single 21 s breath-hold.

Spectra from individual channels were combined using singular value decomposition². Spectra were analyzed with the AMARES algorithm³ included in the MRUI software package⁴ using prior knowledge based on that used in many MRI PDFF assessment techniques¹. The fat spectrum was modeled with nine Gaussians with locations fixed relative to each other (5.29, 5.19, 4.2, 2.75, 2.24, 2.02, 1.6, 1.3 and 0.9 ppm) and with identical peak-width (that is each peak was assumed to have the same R2*^eff). The water peak was modeled by a single unconstrained Gaussian. The areas of the fat peaks were left unconstrained as the fat peak areas may not match those in the reference spectrum¹ after allowing for T2 decay. In human subjects, water R2* and fat R2*^eff were calculated for each TE, and the average values recorded.

Figure 2 compares water R2* and fat R2*^eff in the fat-water emulsion phantom. Water R2* ranged from close to zero (below the lower limit of normal liver R2*) to values observed in mild-to-moderate iron overload. At R2* values below those encountered in human liver, water R2* and fat R2*^eff differed. At water R2* values typical of those seen in vivo, we found close agreement between water R2* and fat R2*^eff.

Figure 3 compares water R2* and fat R2*^eff in subjects with hepatic steatosis. Water R2* and fat R2*^eff were correlated (slope 0.72, intercept 21.9 s^-1, R² 0.49). Mean water R2* was 68.8 s^-1 and mean fat R2*^eff was 71.1 s^-1; no significant difference was observed between these mean values (p = 0.13).

We demonstrated using MRS that there is good agreement between fat R2*^eff and water R2* in phantoms (across values typical of those in vivo) and in human subjects. These findings support the assumption that fat R2*^eff and water R2* can be treated as identical for measuring PDFF using MRI