In osteoporosis the assessment of the bone health status is important to identify fracture risk before are fragility fracture occurs. Invasive bone biopsies might be replaced by in-vivo virtual biopsies using high resolution peripheral quantitative computed tomography (HR-pQCT), which provide insights of the bone micrarchitecture at trabecular scale. Histomorphometry is usually applied to theses tomographic images by using stereology methods, however only a few post-processing methods capture image features beyond these global-based measurements.
The high isotropic spatial resolution of HR-pQCT scans has the potential to provide more insights of the bone microarchitecture, nevertheless the high information density needs to be processed to extract relevant features describing the the different bone patterns. The objective of the thesis is to develop a novel post processing technique of in-vivo images by classifying the bone microarchitecture with 3D-texture analysis in a supervised and unsupervised learn- ing manner. Since we are only interested in the trabecular compartment, it is necessary to have an accurate and reproducible segmentation algorithm. For this reason, we introduce a reproducible and novel segmentation approach for HR-pQCT data, which uses a fully automated threshold-independent image analysis algorithm based on local texture features. Texture analysis is used for feature extraction describing the bone microarchitecture by a statistical approach that captures the distribution and relationship of the intensities in an image. Clustering is applied to these extracted texture features which result in distinctive trabecular microarchitecture classes (TMACs). These classes represent trabecular bone regions with common texture characteristics representing different patterns of the bone.
Experimental results demonstrate the feasibility of 3D-texture analysis and trabecular bone clustering on HR-pQCT images of the ultradistal radius. These experiments include a preliminary application of the technique to a small set of HR-pQCT scans of postmenopausal women with and without fragility fractures. In addition, the clinical applicability of our method using a routine clinical multi detector computed tomography (MDCT) with optimized scan protocols shows promising results in advanced osteoporosis imaging and assessment.