Optimization of region-of-interest sampling strategies for hepatic MRI proton density fat fraction quantification

Abstract

Background: Clinical trials utilizing proton density fat fraction (PDFF) as an imaging biomarker for hepatic steatosis have used a laborious region-of-interest (ROI) sampling strategy of placing an ROI in each hepatic segment.

Purpose: To identify a strategy with the fewest ROIs that consistently achieves close agreement with the nine-ROI strategy.

Study type: Retrospective secondary analysis of prospectively acquired clinical research data.

Population: A total of 391 adults (173 men, 218 women) with known or suspected NAFLD.

Field strength/sequence: Confounder-corrected chemical-shift-encoded 3T MRI using a 2D multiecho gradient-recalled echo technique.

Assessment: An ROI was placed in each hepatic segment. Mean nine-ROI PDFF and segmental PDFF standard deviation were computed. Segmental and lobar PDFF were compared. PDFF was estimated using every combinatorial subset of ROIs and compared to the nine-ROI average.

Statistical testing: Mean nine-ROI PDFF and segmental PDFF standard deviation were summarized descriptively. Segmental PDFF was compared using a one-way analysis of variance, and lobar PDFF was compared using a paired t-test and a Bland-Altman analysis. The PDFF estimated by every subset of ROIs was informally compared to the nine-ROI average using median intraclass correlation coefficients (ICCs) and Bland-Altman analyses.

Results: The study population's mean whole-liver PDFF was 10.1 ± 8.9% (range: 1.1-44.1%). Although there was no significant difference in average segmental (P = 0.452) or lobar (P = 0.154) PDFF, left and right lobe PDFF differed by at least 1.5 percentage points in 25.1% (98/391) of patients. Any strategy with ≥4 ROIs had ICC >0.995. 115 of 126 four-ROI strategies (91%) had limits of agreement (LOA) <1.5%, including four-ROI strategies with two ROIs from each lobe, which all had LOA <1.5%. 14/36 (39%) of two-ROI strategies and 74/84 (88%) of three-ROI strategies had ICC >0.995, and 2/36 (6%) of two-ROI strategies and 46/84 (55%) of three-ROI strategies had LOA <1.5%.

Data conclusion: Four-ROI sampling strategies with two ROIs in the left and right lobes achieve close agreement with nine-ROI PDFF.

Level of evidence: 3 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2018;47:988-994.

Keywords: PDFF; fat quantification; region-of-interest; sampling strategy.

FIGURE 1

Comprehensive nine-ROI sampling strategy, where an ROI (green circle) is placed in each of the nine hepatic segments on multiple slices through the liver on the PDFF map.

FIGURE 2

Histogram demonstrating distribution of PDFF values across the study cohort.

FIGURE 3

Bland–Altman plot between left and right lobe PDFF values. The x-axis is a plot with a logarithmic scale. Although the mean bias of –0.125 (solid line) is small, a quarter of patients had a lobar difference of greater than 1.5 percentage points (dashed lines).

FIGURE 4

Histogram demonstration distribution of standard deviations of segmental PDFF values across the study cohort. Although the standard deviation of the segmental PDFFs was less than 1.5 percentage points for 69% of patients (293/391), some patients exhibited more substantial segmental variability.

FIGURE 5

Plot demonstrating PDFF values (y-axis) vs. segment (x-axis) for a representative sample of patients. Each line represents an individual patient in the cohort. Patients were sorted according to rank-order of increasing standard deviation of segmental PDFF, and every 10^th patient is shown. Darker lines correspond to patients with greater standard deviation of segmental PDFF values.

FIGURE 6

ICCs (y-axis) shown for sampling strategies by the number of ROIs used. Each dot represents a particular sampling strategy (511 total). ICCs increased as the number of ROIs increased. Strategies where the number of ROIs in the left and right hepatic lobes differed by no more than one (ie, balanced) are color-coded with darker colors. These balanced strategies tended to achieve better agreement with the nine-ROI average.

FIGURE 7

LOA bounds (y-axis) shown for sampling strategies by the number of ROIs used. Each dot represents a particular sampling strategy (511 total). LOA bounds decreased (closer agreement) as the number of ROIs increased. Strategies where the number of ROIs in the left and right hepatic lobes differed by no more than one (ie, balanced) are color-coded with darker colors. These balanced strategies tended to achieve better agreement with the nine-ROI average. In this study, 91% of four-ROI strategies achieved close agreement by the LOA metric, but 100% of balanced four-ROI strategies achieved close agreement.