Multi-row detector computed tomography (MDCT) based generalizable tools for bone microstructure and strength analysis and their applications

Osteoporosis is a bone disease associated with enhanced bone loss, microstructural degeneration, and fracture-risk. Bone mineral density (BMD) explains only 60-70% variability in bone strength and the remaining is characterized by trabecular bone (Tb) strength and its microstructural basis. Multi-row detector CT (MDCT) has recently drawn interest as a viable modality for in vivo bone imaging due to its wide availability, fast scan speed, larger field-of-view, and ability to provide quantitative BMD. Although MDCT scanners allow visualization of Tb microstructure, its low spatial resolution causes smaller trabeculae to be artifactually broken following segmentation which makes computation of bone microstructural and strength measures from MDCT scans erroneous. Furthermore, differences in spatial resolution among different MDCT scanners create inconsistency in Tb microstructural measures derived from different scanners posing major challenge in multi-site study design.

In this thesis, new algorithms have been developed for accurate and consistent assessment of Tb microstructural quality and strength. A deep learning (DL) method is developed and validated for harmonizing low- and high-resolution MDCT scans. The DL network not only improves the structural quality of the low-resolution Tb images but also reduces the difference between the Tb microstructural measures derived from low- and high-resolution scans. Additionally, a finite element analysis (FEA) method is developed for MDCT-based computation of Tb strength without requiring segmentation of Tb micro-network from marrow space. Tb strength estimated using the proposed FEA method strongly correlates with micro-CT and experimentally derived reference measures and also shows very high repeat scan as well as inter-scanner reproducibility.