Purpose: The purpose of the study is to determine whether a novel semi-automated DIXON-based fat quantification algorithm can reliably quantify visceral fat using a CT-based reference standard.
Methods: This was an IRB-approved retrospective cohort study of 27 subjects who underwent abdominopelvic CT within 7 days of proton density fat fraction (PDFF) mapping on a 1.5T MRI. Cross-sectional visceral fat area per slice (cm2) was measured in blinded fashion in each modality at intervertebral disc levels from T12 to L4. CT estimates were obtained using a previously published semi-automated computational image processing system that sums pixels with attenuation - 205 to - 51 HU. MR estimates were obtained using two novel semi-automated DIXON-based fat quantification algorithms that measure visceral fat area by spatially regularizing non-uniform fat-only signal intensity or de-speckling PDFF 2D images and summing pixels with PDFF ≥ 50%. Pearson's correlations and Bland-Altman analyses were performed.
Results: Visceral fat area per slice ranged from 9.2 to 429.8 cm2 for MR and from 1.6 to 405.5 cm2 for CT. There was a strong correlation between CT and MR methods in measured visceral fat area across all studied vertebral body levels (r = 0.97; n = 101 observations); the least (r = 0.93) correlation was at T12. Bland-Altman analysis revealed a bias of 31.7 cm2 (95% CI [- 27.1]-90.4 cm2), indicating modestly higher visceral fat assessed by MR.
Conclusion: MR- and CT-based visceral fat quantification are highly correlated and have good cross-modality reliability, indicating that visceral fat quantification by either method can yield a stable and reliable biomarker.
Keywords: Biomarker; Morphometry; Proton density fat fraction (PDFF); Quantitative imaging; Visceral fat.