Practical Approaches to Bone Marrow Fat Fraction Quantification Across Magnetic Resonance Imaging Platforms

Background Quantification of fat by proton density fat fraction (PDFF) measurements may be valuable for the quantification and follow-up of pathology in multicenter clinical trials and routine practice. However, many centers do not have access to specialist methods (such as chemical shift imaging) for PDFF measurement. This is a barrier to more widespread trial implementation. Purpose/Hypothesis To determine the agreement between fat fraction (FF) measurements derived from 1) basic vendor-supplied sequences, 2) basic sequences with offline correction, and 3) specialist vendor-supplied methods. Study Type Prospective. Population Two substudies with ten and five healthy volunteers. Field Strength/Sequence Site A: mDixon Quant (Philips 3T Ingenia); Site B: IDEAL and FLEX (GE 1.5T Optima MR450W); Site C: DIXON, with additional 5-echo gradient echo acquisition for offline correction (Siemens 3T Skyra); Site D: DIXON, with additional VIBE acquisitions for offline correction (Siemens 1.5T Avanto). The specialist method at site A was used as a standard to compare to the basic methods at sites B, C, and D. Assessment Regions of interest were placed on areas of subchondral bone on FF maps from the various methods in each volunteer. Statistical Tests Relationships between FF measurements from the various sites and Dixon methods were assessed using Bland-Altman analysis and linear regression. Results Basic methods consisting of IDEAL, LAVA FLEX, and DIXON produced FF values that were linearly related to reference FF values (P < 0.0001), but produced mean biases of up to 10%. Offline correction produced a significant reduction in bias in both substudies (P < 0.001). Data Conclusion FF measurements derived using basic vendor-supplied methods are strongly linearly related with those derived using specialist methods but produce a bias of up to 10%. A simple offline correction that is accessible even when the scanner has only basic sequence options can significantly reduce bias. Level of Evidence 2 Technical Efficacy Stage 1