Limitations of Bone Marrow Relative Fat Fraction Compared With Proton Density Fat Fraction

Objective: Changes in bone marrow fat content measured through relative fat fraction (rFF) obtained from dual-echo gradient-recalled echo (GRE) in- and opposed-phase (IOP) MRI have been proposed to evaluate treatment response for multiple myeloma. However, rFF suffers from several significant limitations that lead to inaccurate fat fraction measurements. In contrast, proton density fat fraction (PDFF) is the most objective and validated MRI metric of tissue fat content, and it is measured through confounder-corrected, multiecho, chemical-shift-encoded (CSE) MRI. The purpose of this study was to evaluate the linearity and bias of bone marrow rFF compared with PDFF. Methods: This single-center, retrospective study included 100 patients who underwent clinical MRI for liver fat/iron quantification at 1.5T and 3.0T (50 exams/patients for each field strength), which included dual-echo GRE IOP and commercial multiecho CSE MRI (IDEAL-IQ). One region of interest (ROI) was placed in each of the T12, L1, and L2 vertebral bodies. Per-ROI rFF was calculated using (S IP and S OP = signal intensities on IP and OP images, respectively). rFF was correlated with PDFF using linear regression and coefficient of determination ( R 2 ). Bland-Altman analysis evaluated rFF bias across the observed range for R2* and PDFF; mean bias and 95% limits of agreement (LOA) were reported. Results: Bone marrow rFF demonstrated no linearity against PDFF at 1.5T or at 3.0T ( R 2 = 0.032 and 0.057, respectively). Moreover, bone marrow rFF demonstrated significant bias with respect to PDFF at 1.5T and 3.0T, with significant bias that increases directly with bone marrow fat fraction. Conclusions: Bone marrow rFF is nonlinear and variably biased compared with PDFF and should not be used in research or clinical settings.