Medical 3D image processing applied to computed tomography and magnetic resonance imaging

Abstract

Existing microstructure parameters of computed tomography (CT) are able to compute architectural properties of the bone from ex-situ and ex-vivo scans while they are highly affected by the issues of noise and low resolution when applied to clinical in-vivo imaging. A set of improvements of the standard workflow for the quantitative computation of micro-structure from clinical in-vivo scans is proposed in this thesis. Robust methods are proposed (1) for the calibration of density values, (2) the binarization into bone and marrow phase, (3) fuzzy skeletonization and (4) the calibration of the CT volumes in particular for the computation of micro-structural parameters. Furthermore, novel algorithms for the computation of rod-volume fraction with 3D rose diagrams and fractal approaches are proposed and the application of local texture operators to diffusion tensor imaging is proposed. Finally an existing computer program for the application in radiology departments, Structural Insight, was improved and largely extended. In particular the methods of the microstructural calibration, the fractal and the texture operators showed significant improvements of accuracy and precision for the prediction of fracture risk and the quantitative assessment of the progress of Alzheimer's disease, in comparison to existing state-of-the art methods. The methods were tested on artificial and in-vitro data and as well on real-world computed tomography and magnetic resonance imaging (MRI) studies. The proposed novel methods improve the computation of bone characteristics from in-vivo CT and MRI in particular if the methods are combined with each other. In consequence, this allows to assess more information from existing data or to conduct studies with less ray exposure and regarding the MRI method in shorter time than nowadays required.