dc.description.abstract | In the field of image registration there is a pressing need for techniques that can assess the quality of a registration. Currently there is no widespread method to determine error in an image registration, a weakness of registration that adds a level of uncertainty to its results. While there do exist analytical solutions for error in fiducial-based registration, they do not allow for the estimation of error in individual cases. This dissertation presents the algorithm Assessing Quality Using Image Registration Circuits (AQUIRC), which estimates the quality of intensity-based and fiducial-based registration.
Image registration is used extensively in the medical imaging community for a variety of applications including, but not limited to, atlas-based segmentation, large scale probabilistic imaging studies, rigid registration of same-subject images acquired at different times, rigid registration of same-subject images acquired with different modalities, non-rigid registration of intra-subject images acquired at different times to track anatomical changes, and inter-subject non-rigid registration used to provide anatomical and functional information acquired from previous subjects. Fiducial-based registration is often used in the context of image-guided surgery (IGS) and is a critical component in determining the accuracy of an IGS procedure.
There are two main contributions to the field of image registration presented in this dissertation. First, it defines an algorithm that is the first to use the idea of redundancy in sets of registration circuits with multiple registrations which is then used to estimate the accuracy of a single registration. Second, the work described in this dissertation has inspired new work and new methods. The new work that has been inspired combined with the work described in this dissertation may expand the knowledge of registration accuracy.
In this dissertation a validation of the AQUIRC algorithm method is also presented. AQUIRC has been tested across multiple applications: fiducial-based registration (Chapter 4), affine and rigid image registration (Chapter 5), global non-rigid registration (Chapter 6), global atlas selection (Chapter 6), local non-rigid registration (Chapter 7, 8) and local atlas selection (Chapter 9). The results are analyzed using simulated and manually defined ground truth, and AQUIRC's applicability to each situation is discussed. | |