Property Estimation of Thin Cortical Bone from Clinical QCT: An Analysis by Synthesis Approach

This dissertation addresses the challenge of assessing the thickness and density of thin cortical bone. Due to the limited spatial resolution of Quantitative Computed Tomography (QCT) the Nyquist-Shannon sampling theorem is violated, making direct reconstruction impossible. To overcome this, a probabilistic Analysis by Synthesis (AbS) framework is proposed that exploits the geometry of the bone and the anisotropy of the CT point spread function. By modeling uncertainties through a bone and measurement model, the AbS optimization is framed as a maximum a posteriori problem, using a Monte Carlo expectation maximization strategy to infer the bone parameters. The framework is validated on high-resolution micro-CT scans of real vertebrae, demonstrating its ability to accurately estimate cortical bone thickness and density. While there are limitations in mapping the spatial distribution of these parameters, the results are promising and further research offers the potential to refine and improve the method.

In this thesis a probabilistic framework that uses advanced modeling techniques to accurately estimate bone thickness and density, even when direct reconstruction isn't possible, is proposed. Readers will gain valuable insights into cutting-edge methods for improving medical imaging accuracy and a deeper understanding of how these techniques can be further developed for future applications.