Accumulation of fat in the pancreas is associated with development of metabolic syndrome, diabetes, and pancreatic cancer^{1, 2}. Advanced techniques, including 1H MRS and multi-echo MRI have been developed to estimate proton density fat fraction (PDFF) as a non-invasive, quantitative biomarker of pancreatic fat. Despite the clinical value of non-invasive pancreatic PDFF estimation, a clinical standard for the quantification of pancreatic fat has not been established. The purpose of this study is to assess intra- and inter-examination repeatability of 1H MRS and multi-echo MRI to estimate pancreatic PDFF.

In this IRB approved, single site, prospective study, 20 consecutive subjects were recruited from the UCSD NAFLD translational research unit, as these patients span a spectrum of body habitus, and scanned by experienced MR technologists at 3T (GE Sigma Excite HDxt). All sequences were obtained three times: twice without subject repositioning (to assess intra-exam repeatability) and then once more after taking subjects off and placing them back on the scanner table (to assess inter-exam repeatability).

Using anatomic landmarks, a 10 x 10 x 10 mm voxel was placed in the broadest portion of the pancreatic body, with care to include only pancreatic parenchyma and to exclude vasculature and surrounding adipose tissue. For each of the three repeats, the voxel location was replicated and, a 1H MRS STEAM sequence with the following parameters was obtained: four signal averages at TE 10 ms, to minimize T2 weighting, TR 3,500 ms, to minimize T1 weighting, and TM 5 ms to minimize J-coupling. Spectra were analyzed by a single, experienced analyst and were processed using the AMARES algorithm, as part of the jMRUI software package. Spectra were corrected for the fat included in the water signal using a pre-determined spectrum³.

For each repeat, a complex based three-dimensional multi-echo SPGR sequence centered over the pancreas was run in a single 21-s breath-hold with low flip angle (3°) at 7 ms TR. Six echoes were obtained per TR at TEs of 1.0, 1.8, 2.6, 3.4, 4.2, and 5.1 ms. A specialized reconstruction algorithm automatically generated T2*- and noise-bias corrected multi-fat-peak model PDFF parametric maps, which were then transferred off-line for further analysis. One region of interest (ROI) with area of 100 mm2 was placed in the body of the pancreas and co-localized to the center of the MRS voxel for the corresponding repeat, as shown in Figure 1.

For both MRS and MRI PDFF mean values were computed for each repeat. The intra-exam intra-class correlation coefficient (ICC) was computed between 1st and 2nd repeats, and inter-exam ICC was computed between 1st and 3rd repeats for both MRI and MRS pancreatic PDFF. 95% confidence intervals (CI) were computed for each ICC. Additionally, for both MRS and MRI a Bland-Altman plot was generated, and bias (mean of differences) and limits of agreement (LOA) were computed for repeats 1 and 2 (intra-exam), and for repeats 1 and 3 (inter-exam). Eighteen women and two men (mean age 49.8 yrs, range 25-76 yrs; mean BMI 28.3 kg/m2, range 19.3 and 43.9 kg/m2 ) were scanned. Mean MRS-measured PDFFs for repeats 1, 2, and 3 were 6.8, 7.5, and 10.7 %, respectively. Mean MRI PDFFs for repeats 1, 2, and 3 were 5.4, 5.4, and 5.1 %, respectively. As summarized in Table 1, the intra-examination ICC was 0.53 with 95%CI[0.11,0.79] for MRS and 0.70 with 95% CI[0.36,0.87] for MRI. The inter-examination ICC was 0.49 with 95% CI[0.07,0.77] for MRS, and 0.65 95%CI[0.27,0.85] for MRI. Intra-exam bias was -0.69 with 95% LOA [-10.3, 8.93] for MRS and -0.025 with 95% LOA [-4.8, 4.7] for MRI. Inter-exam bias was -3.75 with 95% LOA [-23.7, 13.3] for MRS and 0.41 with 95% LOA –[3.8, 3.8] for MRI. Figures 2-3 show Bland Altman plots. Our results suggest that MRI is more repeatable than MRS for measuring pancreatic PDFF, although formal comparisons were not made. The higher repeatability of MRI was particularly pronounced when subjects were repositioned between acquisitions, which more closely simulates the conditions in which these techniques might be applied clinically and in research. Importantly, the constraint that MRI ROIs be placed at the MRS voxel locations may have caused the true repeatability of MRI to be underestimated. Even higher repeatability may be possible by optimizing the MRI ROI placement strategy.