Emerging MRI-based proton-density fat-fraction (PDFF) and R2* (1/T2*) measurements in the liver have shown great promise as quantitative imaging biomarkers for the non-invasive detection, quantitative staging, and treatment monitoring of hepatic steatosis¹ and liver iron overload,² respectively. Typically, these measurements are performed by drawing regions-of-interest (ROIs) on PDFF and R2* maps to obtain quantitative estimates of the liver fat^3–5 and iron⁶ content, respectively. Despite the growing clinical and research interest in these techniques, a standardized approach for ROI-based measurements has not been established. Recent studies conducted at different sites have used a wide range of ROI sizes, placement, and number for measurements of liver PDFF^3–5,7 and R2*,⁶ which complicates the widespread dissemination of these techniques as reproducible quantitative imaging biomarkers. The purpose of this study was to evaluate the reproducibility of different combinations of ROI size, placement, and number for measurements of liver PDFF and R2* based on their inter- and intra-reviewer agreement. A secondary purpose was to establish practical and standardized guidelines for future acquisitions of ROI-based PDFF and R2* measurements.

53 patient liver MRI datasets were retrospectively analyzed for PDFF and R2* using ROI sampling methods that have been previously reported.^3–5,7 All datasets were collected as part of IRB-approved protocols and are HIPAA-compliant. The patients (mean [range] age: 51.9 [23–84] years; 26M/27F) were clinic patients undergoing abdominal MRI for a variety of clinical indications and were not suspected of having hepatic steatosis or liver iron overload.

All imaging was performed at 1.5T (Signa HDxt or Optima MR 450w, GE Healthcare, Waukesha, WI) using an 8- or 12-channel phased array cardiac or torso coil. Imaging was performed using an investigational version of a quantitative multi-echo spoiled gradient echo chemical shift-based water-fat separation method. Imaging parameters included: TR=13.5–3.7ms, TE₁=1.2–1.3ms, ΔTE=1.98–2.0ms, echoes=6, FOV=35x35–44x44cm, slice thickness=8–10mm, slices=24–32, receiver bandwidth=±83–125kHz.

Three reviewers analyzed the PDFF and R2* maps using nine circular ROI sampling paradigms that each used a different combination of ROI size, placement, and number. The ROI sizes were: 1) 1 cm², 2) 4 cm², and 3) The largest area that fit inside each placement designation, while avoiding large vessels, bile ducts, and obvious image artifact. The ROI placement designations were: 1) The left and right liver lobes, 2) The anterior, posterior, medial, and lateral segments of the liver, and 3) The nine Couinaud segments of the liver. The number of ROIs were: 1) Two ROIs (one per left and right liver lobe), 2) Four ROIs (one per anterior, posterior, medial, and lateral segment), and 3) Nine ROIs (one per Couinaud segment). Figure 1 summarizes these paradigms.

To evaluate the reproducibility of each paradigm, inter-reviewer agreement between all three reviewers was assessed with Intraclass Correlation Coefficients. Intra-reviewer agreement for two reviewers was assessed with Bland-Altman analysis.PDFF measurements had a mean ± SD [range] of 5.9 ± 8.9% [-0.01–41.7%]. R2* measurements had a mean ± SD [range] of 32.4 ± 10.3 s^-1 [12.2–82.1 s^-1]. Figure 2 shows inter-reviewer agreement assessed by Intraclass Correlation Coefficients for PDFF and R2*. Figure 3 shows Bland-Altman 95% Confidence Intervals for the intra-reviewer agreement assessments of PDFF and R2*. These results demonstrate a trend that the repeatability of PDFF and R2* measurements in the liver, as assessed by their inter- and intra-reviewer agreement, increases as the size and number of ROIs increases. Figure 4 and Figure 5 illustrate this trend.This study evaluated the reproducibility of different ROI sampling methods for liver PDFF and R2* measurements. Our results indicate that the repeatability of liver PDFF and R2* measurements improves as the size and number of ROIs increase, which increases the area of the liver being sampled. Although placing one largest-fit ROI in the nine Couinaud segments (Paradigm 9) resulted in better inter- and intra-reviewer agreement generally, it is not clear that this is the ideal paradigm because it is more complex, time-consuming, and it performed only slightly better in some assessments. Depending on the specific application and requirements of a study, Paradigms 6 and 8 may provide a good compromise between paradigm complexity and measurement repeatability. To conclude, this study highlights the improved repeatability that results when the sampling area of the liver is increased by using multiple, large ROIs to measure PDFF and R2*. Therefore, clinicians and researchers performing ROI-based measurements of liver PDFF and R2* should strive to sample as much area of the liver as possible by using ROIs that are large in both size and number.