Reproducibility of hepatic fat fraction measurement by magnetic resonance imaging.

ABSTRACT

PURPOSE: To evaluate the reproducibility of magnetic resonance imaging (MRI)-determined hepatic fat fraction (%) across imaging sites with different magnet types and field strength. Reproducibility among MRI platforms is unclear, even though evaluating hepatic fat fractions (FFs) using MRI-based methods is accurate against MR spectroscopy. MATERIALS AND METHODS: Overweight subjects were recruited to undergo eight MRI examinations at five imaging centers with a range of magnet manufacturers and field strengths (1.5 and 3 T). FFs were estimated in liver and in fat-emulsion phantoms using three methods: 1) dual-echo images without correction (nominally out-of-phase [OP] and in-phase [IP]); 2) dual-dual-echo images (two sequences) with T2* correction (nominally OP/IP and IP/IP); and 3) six-echo images with spectral model and T2* correction, at sequential alternating OP and IP echo times (Methods 1, 2, and 3, respectively). RESULTS: Ten subjects were recruited. For Methods 1, 2, and 3, respectively, hepatic FF ranged from -2.5 to 27.0, 1.9 to 29.6, and 1.3 to 34.4%. Intraclass correlation coefficients were 0.85, 0.89, and 0.91 for each method, and within-subject coefficients of variation were 18.5, 9.9, and 10.3%, respectively. Mean phantom FFs derived by Methods 2 and 3 were comparable to the known FF for each phantom. Method 1 underestimated phantom FF. CONCLUSION: Methods 2 and 3 accurately assess FF. Strong reproducibility across magnet type and strength render them suitable for use in multicenter trials and longitudinal assessments.