Scientific Session 09 — Gastrointestinal - Diffuse Liver DiseaseTuesday, May 2, 2017
2263. Optimization of ROI Sampling Strategies for Hepatic MRI Proton Density Fat Fraction
Hong C*, Wolfson T, Sy E, Schlein A, Sirlin C. University of California, San Diego, CA
Address correspondence to C. Hong (email@example.com)
Objective: Proton density fat fraction (PDFF) is the leading MRI-based biomarker for noninvasively quantifying hepatic steatosis. Clinical trials have relied on a comprehensive ROI sampling strategy in which an ROI is drawn in each of the nine hepatic segments, but this process is laborious and technically challenging. The purpose of this study was to optimize the sampling strategy for ROI placement by identifying the fewest number of ROIs required to achieve close agreement with the comprehensive strategy with intraclass correlation coefficients (ICCs) equal to 0.995 and limits of agreement (LOAs) less than ± 1.5%.
Materials and Methods: We performed a secondary analysis of adults with known or suspected nonalcoholic fatty liver disease (NAFLD) enrolled consecutively in prospective clinical studies from 2012 to 2016. Enrolled patients underwent confounder-corrected, chemical shift–encoded 3-T MRI to measure the PDFF. Demographics were recorded. For each examination, an ROI was placed in each of the nine anatomic hepatic segments. The PDFF values for the nine segments were averaged to provide a composite PDFF for each patient. Additionally, the PDFF was recalculated 511 times by averaging every combination of segments using one up to eight segments. The PDFF estimated by each combination was informally compared with that of the composite PDFF with nine ROIs using ICCs and Bland-Altman analyses. Additionally, the accuracy of each combination for diagnosing hepatic steatosis was assessed, with steatosis defined as PDFF greater than 5% by the composite with nine ROIs.
Results: A total of 398 patients were included (175 men, 223 women; mean age, 50 years; age range, 18–85 years). All strategies with one ROI had ICCs less than 0.995 and LOAs greater than 1.5%. Fourteen of 36 (39%) of strategies with two ROIs and 74 of 84 (88%) strategies with three ROIs had ICCs greater than 0.995. Only two of 36 (6%) strategies with two ROIs and 45 of 84 (54%) strategies with three ROIs had LOAs less than 1.5%. Any strategy with four or more ROIs had ICCs greater than 0.995. Of the 126 strategies with four ROIs, 115 (91%) had LOAs less than 1.5%. Among strategies with four ROIs, one strategy involved all four right-lobe segments, 20 involved three right- and one left-lobe segments, 60 involved two right- and two left-lobe segments, 40 involved one right- and three left-lobe segments, and five involved only left-lobe segments. LOAs in the strategies with four ROIs were less than 1.5% in 0/1 (0%), 16/20 (80%), 60/60 (100%), 38/40 (95%), and 1/5 (20%) strategies, respectively. The strategies with four ROIs and with two ROIs from each lobe achieved 98.2% aggregate accuracy (sensitivity, 98.8%, specificity, 97.4%) for the diagnosis of hepatic steatosis, defined as PDFF greater than 5% by the composite with nine ROIs.
Conclusion: These results suggest that sampling strategies with four ROIs can achieve excellent agreement and diagnostic accuracy compared with the composite PDFF with nine ROIs. With further validation, a simpler sampling strategy with four ROIs may become the new standard for measuring PDFF in clinical trials. Our observations indicate that two ROIs should be drawn in each hepatic lobe to ensure representative sampling.