Abstract
Background/Objectives: To comparatively evaluate the clinical reliability of automated single-location MRI-proton density fat fraction (PDFF), single-voxel magnetic resonance spectroscopy (MRS), and volumetric MRI-PDFF measurements for quantitative assessment of hepatic steatosis, using manual multi-regions of interest (ROI) MRI-PDFF as the reference standard. Methods: In this retrospective single-center study, adult patients who underwent liver MRI with both MRI-PDFF and MRS between December 2024 and January 2026 were included. Manual multi-ROI MRI-PDFF measurements were obtained from eight Couinaud segments and served as the reference standard. Automated single-location MRI-PDFF, single-voxel MRS, and volumetric MRI-PDFF measurements were extracted from system-generated reports. Fat fraction values and steatosis grades were compared using non-parametric tests, Spearman correlation coefficients, exact agreement rates, weighted Cohen's kappa statistics, and Bland-Altman analysis. Intra- and inter-observer reliability of manual multi-ROI measurements was assessed using intraclass correlation coefficients (ICC). Results: A total of 490 patients were included. Single-location MRI-PDFF and single-voxel MRS demonstrated high exact agreement with manual multi-ROI MRI-PDFF in steatosis grading (86.9% and 87.4%, respectively), with near-perfect agreement (weighted κ = 0.82-0.83). Volumetric MRI-PDFF showed lower exact agreement (76.1%) and a systematic tendency toward overestimation, with more frequent upward shifts in steatosis grade. All techniques showed strong correlations with manual measurements (ρ = 0.72-0.75). Manual multi-ROI MRI-PDFF demonstrated excellent inter- and intra-observer reliability (ICC > 0.97). Conclusions: Automated single-ROI MRI-PDFF and single-voxel MRS provide clinically reliable and time-efficient alternatives for hepatic steatosis assessment, while automated volumetric MRI-PDFF may introduce systematic bias toward overestimation.