Harmonized Multisite MRI-Based Quantification of Human Liver Fat and Stiffness: A Pilot Study

Abstract

Background: Nonalcoholic fatty liver disease (NAFLD) is a leading cause of end-stage liver disease. NAFLD diagnosis and follow-up relies on a combination of clinical data, liver imaging, and/or liver biopsy. However, intersite imaging differences impede diagnostic consistency and reduce the repeatability of the multisite clinical trials necessary to develop effective treatments.

Purpose/hypothesis: The goal of this pilot study was to harmonize commercially available 3 T magnetic resonance imaging (MRI) measurements of liver fat and stiffness in human participants across academic sites and MRI vendors.

Study type: Cohort.

Subjects: Four community-dwelling adults with obesity.

Field strength/sequence: 1.5 and 3 T, multiecho 3D imaging, PRESS, and GRE.

Assessment: Harmonized proton density fat fraction (PDFF) and magnetic resonance spectroscopy (MRS) protocols were used to quantify the FF of synthetic phantoms and human participants with obesity using standard acquisition parameters at four sites that had four different 3 T MRI instruments. In addition, a harmonized magnetic resonance elastography (MRE) protocol was used to quantify liver stiffness among participants at two different sites at 1.5 and 3 T field strengths. Data were sent to a single data coordinating site for postprocessing.

Statistical tests: Linear regression in MATLAB, ICC analyses using SAS 9.4, one-sided 95% confidence intervals for the ICC.

Results: PDFF and MRS FF measurements were highly repeatable among sites in both humans and phantoms. MRE measurements of liver stiffness in three individuals at two sites using one 1.5 T and one 3 T instrument showed repeatability that was high although lower than that of MRS and PDFF.

Conclusions: We demonstrated harmonization of PDFF, MRS, and MRE-based quantification of liver fat and stiffness through synthetic phantoms, traveling participants, and standardization of postprocessing analysis. Multisite MRI harmonization could contribute to multisite clinical trials assessing the efficacy of interventions and therapy for NAFLD.

Level of evidence: 2 TECHNICAL EFFICACY STAGE: 2.

Keywords: elasticity imaging techniques; magnetic resonance imaging; magnetic resonance spectroscopy; nonalcoholic fatty liver disease.

Figure 1:

(A) Representative PDFF slice image of seven phantoms containing 4%, 10%, 15%, 25%, 40%, 45% and 55% fat fractions (FF) where coloration depicts the fat fraction as per the heatmap to the right; The relationship of the known/true phantom FF (x-axis) to the measured PDFF (B) or measured MRS (C) FF using four MRIs located at independent sites (blue dot = site 1, red dot = site 2, green triangle = site 3, dark blue triangle = site 4). Black line indicates a linear regression. D) The relationship of measured PDFF fat fraction (y-axis) and measured MRS fat fraction (y-axis) in phantoms, using the corrected MRS data from site 4. E) PDFF and MRS fat fractions as measured in four participants at four study sites. Top Left: Representative slice of PDFF maps from one participant across all four sites. Mean fat fraction (%) of selected ROIs are shown. Bottom Left: PDFF measured fat fraction (%) (y-axis) in four participants (x-axis) as measured using different scanners at each study site. Top Right: Representative spectra collected via MRS in one participant across all four sites. Resonances from water (4.7 ppm), methyl (CH3 at 0.9 ppm) and methylene (CH2 at 1.3 ppm) groups are also marked. Bottom Right: MRS measured fat fraction (%) (y-axis) in four participants (x-axis) as measured using different scanners at each study site (dark gray = site 1, light gray = site 2, medium gray = site 3, black = site 4). F) Relationship of PDFF and MRS fat fractions measured in four participants at four study sites. Each of the four study participants are represented as four data points where each data point represents a study site (blue = site 1, red = site 2, green = site 3 and dark blue = site 4) as plotted as a measure PDFF fat fraction (%) (y-axis) compared to MRS fat fraction (%) (x-axis). G) Bland-Altman plots showing the good degree of agreement among the measurements by different sites. Top Right: Plot for PDFF measurements, Bottom Left: Plot for MRS measurements and Bottom Right: Plot for MRE measurements. The limits of agreements with the mean were shown as the dashed red lines in plots for PDFF and MRS. Plot for MRE is a standard Bland Altman plot with 2 raters, the Red dashed line being at 2 standard deviations of the difference and the Green at the 3 standard deviations and the black dashed line is the mean of the stiffness measurements.