Differentiation of Vertebral Metastases From Focal Hematopoietic Marrow Depositions on MRI: Added Value of Proton Density Fat Fraction

OBJECTIVE. The purpose of this study was to evaluate the added value of proton density fat fraction (PDFF) in differentiating vertebral metastases from focal hematopoietic marrow depositions. MATERIALS AND METHODS. The study included 44 patients with 30 vertebral metastases and 14 focal hematopoietic marrow depositions who underwent spinal MRI. The final diagnoses were based on histologic confirmation, follow-up MRI, or PET/CT. Two musculoskeletal radiologists with 1 and 15 years of experience independently interpreted both image sets (i.e., images from conventional MRI alone versus images from conventional MRI and PDFF combined). Using a 5-point scale, the readers scored their confidence in the malignancy of the vertebral lesions. The diagnostic performance (AUC) of the two image sets was assessed via ROC curve analyses. Sensitivities, specificities, and accuracies (for both image sets) were compared using the McNemar test. Kappa coefficients were calculated to assess interobserver agreement. RESULTS. Both readers showed improved diagnostic performance after PDFF was added (AUC, 0.840-0.912 and 0.805-0.895 for readers 1 and 2, respectively). However, adding PDFF did not significantly improve the sensitivity and specificity of either reader (p >.05). Interobserver agreement significantly improved from moderate (kappa = 0.563) to excellent (kappa = 0.947) after PDFF was added. CONCLUSION. The addition of PDFF to a conventional MRI protocol improved the diagnostic performance for differentiating vertebral metastases from focal hematopoietic marrow depositions but without resulting in significant improvement in sensitivity and specificity.

Automated summary

The study evaluated the added value of PDFF in differentiating vertebral metastases from hematopoietic marrow depositions. Results showed improved diagnostic performance with the addition of PDFF but without significant improvement in sensitivity and specificity. Interobserver agreement significantly improved after adding PDFF.