Segmentation of trabecular bone for in vivo CT imaging using a novel approach of computing spatial variation in bone and marrow intensities

Abstract

Characterization of trabecular bone (TB) microarchitecture and computational modelling of bone strength are widely used in research and clinical studies related to osteoporosis, which is associated with elevated risk of fractures. Segmentation of TB network from the background marrow space is essential for quantitative assessment of the quality of TB microarchitecture and bone strength, which are key determinants of fracture risk. Clinical CT is rapidly emerging as a viable modality for in vivo TB microarchitectural imaging at peripheral sites. Here, we present a new method for TB segmentation using in vivo CT imaging of distal tibia. Our method is primarily based on computing the spatial variation in the background marrow intensity as well as the bone-marrow contrast. First, a new anisotropic diffusion algorithm is developed and applied to improve local separability of TB microstructures that uses Hessian matrix to locally guide the diffusion process. Subsequently, a new multi-scale morphological algorithm is developed and applied to determine spatial distribution of bone and marrow intensity values. The accuracy of the method was examined by comparing its performance with multi-user-selected regional thresholding for bone-marrow separation on in vivo CT images of ten subjects each containing twenty random regions of interest (ROIs). High sensitivity (0.93), specificity (0.93), and accuracy (0.93) of the new method were observed from experimental results. In addition, the impact of the new method on predicting bone strength was examined in a cadaveric study. Experimental results have shown that the new TB segmentation method significantly improves the ability (R2 = 0.82) of the computed TB thickness measure to predict actual bone strength determined by mechanical testing on TB cores.