PhD ceremony Ms. M. Abdoli: Development and evaluation of metal artifact reduction and image segmentation techniques in PET/CT
|When:||We 02-10-2013 at 16:15|
PhD ceremony: Ms. M. Abdoli, 16.15 uur, Academiegebouw, Broerstraat 5, Groningen
Dissertation: Development and evaluation of metal artifact reduction and image segmentation techniques in PET/CT
Promotor(s): prof. H. Zaidi, prof. R.A. Dierckx, prof. J. Pruim
Faculty: Medical Sciences
The production of artifact-free and quantitative PET images requires metal artifact reduction (MAR) of the CT images prior to application of the CT-based attenuation correction procedure. The various MAR methods are discussed in this thesis. The general conclusion is that sinogram-based MAR approaches are more accurate in obtaining reliable attenuation coefficients. Chapter 3 demonstrates that different implantable cardiac pacing devices can give rise to varying magnitudes of metal artifacts on CT images. The misalignment between PET images and attenuation maps produced more severe artifacts.
We describe various MAR methods, i.e. sinogram-based methods applying different interpolation techniques, ranging from a 1D spline interpolation method to a 2D Clough-Tocher interpolation technique. Each proposed technique improved the performance of the preceding one.
Delineation of target volumes on PET images is known to be a challenging task due to the inherent noisy and low-resolution characteristic of PET images. We assessed the performance of different categories of 18F-FDG PET image segmentation techniques. The deformable active contour model proposed in this thesis handles the intrinsic high noise level of PET images and the heterogeneity of the tumor uptake appropriately. Application of ADF makes the handling of the high level of noise in PET images feasible without influencing PET quantification. Moreover, integrating the contourlet transform of the image to the energy functional formulation makes the fitting terms more potent in attracting the evolving curve towards the object boundaries due to the enhancement of the TBR.