Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.784947
Title: Motion correction of lung CT images with an application to oncology
Author: Thompson, Tony
ISNI:       0000 0004 7970 4915
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
The aim of image registration is to align sets of similar images which have been captured at different time points, from different perspectives or obtained using different imaging modalities (e.g. CT, MRI, X-ray). In oncology, image registration is a very powerful tool which can be used in an array of different applications such as anatomic image segmentation, 4D dose accumulation maps and lung ventilation maps. In order to align a given set of images, we first assign one image in the set to be the 'fixed' reference image to which we align the remaining 'moving' template images. We are then tasked with finding suitable transformations which deform the template images to match the reference image. In this thesis, we model the image registration problem mathematically through the minimisation of an energy functional. We begin by proposing an improved non-linear multigrid method, based upon the method proposed by Chumchob and Chen via a more accurate analysis of the scheme and new solver to improve convergence, accuracy and CPU time. Next we extend our improved Chumchob-Chen model to incorporate an additional constraint to prevent folding in the transformation, thus leading to physically accurate diffeomorphic registrations. After this we further extend our proposed constrained model to improve robustness with regard to accuracy in cases of severe folding, in addition to parameter choice. We then demonstrate these improvements using a combination of real lung CT images and a synthetic hand X-ray image set. Next we consider a different approach to addressing the problem of folding in the transformation, by formulating an inverse consistent image registration model based upon the model first proposed by Christensen and Johnson. Our proposed idea is to linearise the inverse consistency constraint in the Christensen-Johnson model, which is extremely expensive to compute with regard to computational cost due to its non-linear nature, in addition to implementing a fast non-linear multigrid scheme to further help reduce the computational cost of the model. We then perform some numerical tests on a mix of real CT images and a synthetic example, to highlight the advantages of our proposed inverse consistent model. Finally, we present three 3D image registration models based upon the models discussed throughout this thesis, in addition to 3D extensions of the associated non-linear multi-grid schemes. We then show some preliminary results, using eight examples taken from the Hugo image database along with clinically drawn contours for nine different objects within the CT images, comparing our proposed models with a state of the art commercial software used in hospitals.
Supervisor: Chen, Ke Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.784947  DOI:
Share: