Automated threshold-independent cortex segmentation by 3D-texture analysis of HR-pQCT scans

The quantitative assessment of metabolic bone diseases relies on tissue properties such as bone mineral density (BMD) and bone microarchitecture. In spite of an increasing number of publications using high-resolution peripheral quantitative computed-tomography (HR-pQCT), the accurate and reproducible separation of cortical and trabecular bone remains challenging. In this paper, we present a novel, fully automated, threshold-independent technique for the segmentation of cortical and trabecular bone in HR-pQCT scans. This novel post-processing method is based on modeling appearance characteristics from manually annotated cases. In our experiments the algorithm automatically selected texture features with high differentiating power and trained a classifier to separate cortical and trabecular bone. From this mask, cortical thickness and tissue volume could be calculated with high accuracy. The overlap between the proposed threshold-independent segmentation tool (TIST) and manual contouring was 0.904±0.045 (Dice coefficient). In our experiments, TIST obtained higher overall accuracy in our measurements than other techniques.