Discontinuity preserving image registration for breathing induced sliding organ motion
Year: 2013
Type: dissertation
Abstract: Image registration is a powerful tool in medical image analysis and facilitates \nthe clinical routine in several aspects. It became an indispensable device for \nmany medical applications including image-guided therapy systems. The \nbasic goal of image registration is to spatially align two images that show a \nsimilar region of interest. More speci�cally, a displacement �eld respectively \na transformation is estimated, that relates the positions of the pixels or \nfeature points in one image to the corresponding positions in the other one. \nThe so gained alignment of the images assists the doctor in comparing and \ndiagnosing them. There exist di�erent kinds of image registration methods, \nthose which are capable to estimate a rigid transformation or more generally \nan a�ne transformation between the images and those which are able to \ncapture a more complex motion by estimating a non-rigid transformation. \nThere are many well established non-rigid registration methods, but those \nwhich are able to preserve discontinuities in the displacement �eld are rather \nrare. These discontinuities appear in particular at organ boundaries during \nthe breathing induced organ motion. \nIn this thesis, we make use of the idea to combine motion segmentation \nwith registration to tackle the problem of preserving the discontinuities in \nthe resulting displacement �eld. We introduce a binary function to represent \nthe motion segmentation and the proposed discontinuity preserving \nnon-rigid registration method is then formulated in a variational framework. \nThus, an energy functional is de�ned and its minimisation with respect to \nthe displacement �eld and the motion segmentation will lead to the desired \nresult. In theory, one can prove that for the motion segmentation a global \nminimiser of the energy functional can be found, if the displacement �eld \nis given. The overall minimisation problem, however, is non-convex and a \nsuitable optimisation strategy has to be considered. Furthermore, depending \non whether we use the pure L1-norm or an approximation of it in the formulation \nof the energy functional, we use di�erent numerical methods to solve \nthe minimisation problem. More speci�cally, when using an approximation \nof the L1-norm, the minimisation of the energy functional with respect to the displacement �eld is performed through Brox et al.'s �xed point iteration \nscheme, and the minimisation with respect to the motion segmentation \nwith the dual algorithm of Chambolle. On the other hand, when we make \nuse of the pure L1-norm in the energy functional, the primal-dual algorithm \nof Chambolle and Pock is used for both, the minimisation with respect to \nthe displacement �eld and the motion segmentation. This approach is clearly \nfaster compared to the one using the approximation of the L1-norm and also \ntheoretically more appealing. Finally, to support the registration method \nduring the minimisation process, we incorporate additionally in a later approach \nthe information of certain landmark positions into the formulation of \nthe energy functional, that makes use of the pure L1-norm. Similarly as before, \nthe primal-dual algorithm of Chambolle and Pock is then used for both, \nthe minimisation with respect to the displacement �eld and the motion segmentation. \nAll the proposed non-rigid discontinuity preserving registration \nmethods delivered promising results for experiments with synthetic images \nand real MR images of breathing induced liver motion. \n
Source: edoc (University of Basel)
Author: Kiriyanthan, Silja
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W1770844853