Hardware Accelerated Non-rigid Registration for the Evaluation of Cardiac Function

ResearchSpace/Manakin Repository

Show simple item record

dc.contributor.advisor Young, A en
dc.contributor.advisor Cowan, B en
dc.contributor.author Li, Bo en
dc.date.accessioned 2011-01-11T01:49:48Z en
dc.date.issued 2011 en
dc.identifier.uri http://hdl.handle.net/2292/6125 en
dc.description.abstract Cardiac magnetic resonance (CMR) imaging allows non-invasive access to the morphology and function of the heart with high precision and accuracy. Since a single cardiac examination results in a large amount of CMR images, the image processing for quantitative analysis of their underlying tissue deformation is complex and time-consuming. Recent advances in the Graphics Processing Units (GPU) available on standard low-price personal computers provide additional power for general-purpose computation. This leads to an increasing need for efficient image processing algorithms implemented in the faster hardware environment. In this thesis, three new non-rigid image registration algorithms for capturing the cardiac motion have been developed using the advanced graphics hardware. Firstly, a 2D non-rigid image registration algorithm for calculating the displacement between two consecutive images has been developed in the parallel architecture in order to be plugged into the graphics pipeline and accelerated by the advanced graphics hardware. This algorithm processes through all images in the forward and backward temporal directions to generate cyclic cardiac motion, and gains at least a fivefold speed improvement compared with a corresponding Central Processing Unit (CPU) implementation. Secondly, the 2D registration algorithm has been further extended into 2D+time directions, thus allowing tracking of the motion from all images to the reference image with the periodicity and temporal smoothness implicitly ensured by a spatio-temporal mathematics model. Thirdly, a novel registration algorithm based on a user-identified heart model has been developed in both 2D and 2D+time directions. To my best knowledge, this is the first attempt to registering image pixels within an anisotropic region. It enables an anisotropic geometric distribution of user-defined model parameters, which can be customized to the application, thereby achieving fast estimations with fewer degrees of freedom for a given level of accuracy than standard isotropic methods. Finally, one phantom dataset with analytical solution and two clinical CMR and tagged CMR datasets containing 36 and 30 patients have been used to quantitatively validate above algorithms in different clinical applications, i.e., epi- and endo-cardial boundaries tracking, mitral valves tracking, and tag tracking. No more than a 2 mm error on average in boundaries tracking is found in the middle short-axis slices. Furthermore, the 2D algorithm has also been implemented in the image re-constructor computer to enable in-line tracking of features in SSFP cine images during scanning. Compared with the standard manual approach, this led to a 30% saving in the executive time in the ventricular function analysis. en
dc.publisher ResearchSpace@Auckland en
dc.relation.ispartof PhD Thesis - University of Auckland en
dc.relation.isreferencedby UoA99207917814002091 en
dc.rights Items in ResearchSpace are protected by copyright, with all rights reserved, unless otherwise indicated. en
dc.rights.uri https://researchspace.auckland.ac.nz/docs/uoa-docs/rights.htm en
dc.rights.uri http://creativecommons.org/licenses/by-nc-sa/3.0/nz/ en
dc.title Hardware Accelerated Non-rigid Registration for the Evaluation of Cardiac Function en
dc.type Thesis en
thesis.degree.discipline Bioengineering en
thesis.degree.grantor The University of Auckland en
thesis.degree.level Doctoral en
thesis.degree.name PhD en
dc.date.updated 2011-01-11T01:38:25Z en
dc.rights.holder Copyright: The author en
pubs.elements-id 199129 en
dc.relation.isnodouble 11287 *
pubs.record-created-at-source-date 2011-01-11 en

Full text options

This item appears in the following Collection(s)

Show simple item record

http://creativecommons.org/licenses/by-nc-sa/3.0/nz/ Except where otherwise noted, this item's license is described as http://creativecommons.org/licenses/by-nc-sa/3.0/nz/


Search ResearchSpace

Advanced Search