Texture Analysis in Quantitative Osteoporosis Assessment

High resolution peripheral quantitative quantitative computed tomography (HR-pQCT) permits in-vivo assessement of trabecular microstructure at an isotropic voxel size of 82μm. The new imaging modality has potential in the differentiation of certain forms of osteoporosis based on the associated microstructural patterns. In this paper we propose an approach that assesses bone microarchitecture based on texture features extracted from the trabecular bone. The method is based on a three-dimensional gray level co-occurence matrix descriptor, and a k-medians clustering in the feature space to indicate characteristic categories of texture. The distribution of the microarchitecture classes allows for a differentiation between osteoporotic and healthy subjects. We report initial results for the repeatability of the clustering method and the feasibility of the differentiation between healthy and osteoporotic bone for 6 subjects.