Tomographic segmentation and discrete tomography for quantitative analysis of transmission tomography data

• Main Author: van Aarle, Wim
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2012
• Subjects: Computer science; Medical imaging
• ISBN: 9781267357953; 1267357959

Computed Tomography (CT) is a non-destructive imaging technique for the visualization of internal structures. With CT, an object can be virtually reconstructed based on multiple X-ray projection images, taken under different angles. Many applications benefit from the use of CT: diagnostics in medicine, bio-medical research, materials science, biology, the diamond industry, etc. In case only a few projections are measured, or if these projections are inaccurate (for example due to dose restrictions), conventional reconstruction and segmentation techniques provide images that are insufficiently accurate for quantitative analysis. In this work, new techniques are proposed that attempt to create highly accurate tomographic images from as few projections as possible. Part 1: Tomographic segmentation: In conventional thresholding segmentation techniques, the choice of the threshold value is based solely on the reconstructed image and its histogram. With the proposed tomographic segmentation approach, also the available projection data is exploited. The optimal threshold is then found by minimizing the projection difference between the measured projection data and the forward projection of the segmented image. Very accurate segmentations can thus be computed. Part 2: Discrete tomography: In discrete tomography, reconstruction and segmentation are combined into a single algorithm. The Discrete Algebraic Reconstruction Technique (DART) is a popular, iterative reconstruction technique that, assuming that the number of grey levels in the reconstruction is limited and that these grey levels are known a priori, can generate highly accurate reconstruction from only a few projection images. An in-depth study was performed of this algorithm and solutions are proposed for some key issues that limit the applicability of DART in practical use. For the grey level estimation, which typically occurs by a manual trial-and-error process, various less user intensive methods are introduced: (1) automatic minimization of the projection difference between the measured data and a forward projection of the DART reconstruction, (2) semi-automatic estimation with the new DGLS algorithm, to be used prior to the reconstruction, and (3) using a tomographic segmentation method, as described in the first part of this work, in which the grey levels are adaptively estimated during the reconstruction process of DART. It was also noticed that DART is inherently unsuited for reconstruction of small objects (with respect to the size of the pixels in the reconstructed images) such as trabecular bone and metal foams. To solve this problem, it is proposed to combine DART reconstruction with a tomographic super-resolution approach.