Quantification of Bone Marrow Types from High-Resolution MR Images in the Proximal Femur using Three Class Clustering

Jenny Folkesson¹, Julio Carballido-Gamio², Dimitrios C. Karampinos¹, Thomas Baum¹, Thomas M. Link¹, Sharmila Majumdar¹, Roland Krug¹

The purpose of this work was to evaluate the feasibility of marrow composition quantification in high resolution proximal femur bone MRI. Automatic volume of interest placement was performed using an atlas built from IDEAL images . Separation of trabecular bone, red and yellow marrow was performed using 3-class fuzzy clustering. In the test cohort (2 fracture, 8 non-fracture females), fracture subjects had lower red marrow fraction (p<0.05). These initial results suggest that marrow composition analysis is feasible using HR MRI, and could potentially help understand underlying factors related to aging and fracture history.