Use this URL to cite or link to this record in EThOS:
Title: Non-rigid medical image registration using points, curves and parameterised surfaces
Author: Shah, Said Khalid
ISNI:       0000 0004 2717 9581
Awarding Body: University of East Anglia
Current Institution: University of East Anglia
Date of Award: 2011
Availability of Full Text:
Access from EThOS:
This thesis describes different point based non-rigid registration methods in general and the "fast" radial basis functions (RBFs) non-rigid registration method in particular. We propose and implement three techniques for identifying features like points, curves and surfaces in medical images which represent displacement fields among such images and are later used in developing three corresponding fast and accurate non-rigid registration techniques. Each registration technique involves four main steps: feature extraction, correspondence between the features, RBF fitting to corresponding features, and evaluating the fitted method to the 3 D image data. The main goal of developing these three techniques is to see the effect of the increasing number of salient features in the form of anatomical point landmarks, curves and parameterised surfaces on the overall performance (speed and accuracy) of the tested registration algorithms. The point and curve based Fast RBF registration methods use manually placed anatomical point landmarks and principal curves, respectively. They ensure sub-second registration of standard-sized X-ray C'I' and MR datasets without loss of accuracy as compared to competing methods. However, both methods have limited performance in recovering large deformation due to small numbers of point landmarks. The surface based Fast RBF registration uses surface parameterisation and reparameterisation techniques as pre-processing steps to increase the number of point landmarks on a surface, establishing initial correspondences between points of a surface-pair using a minimum distortion based global surface parameterisation algorithm. Experimental results demonstrate target registration errors less than 2mm on intra-subject registration of various size 256/\3 MR real datasets. in conclusion, it is shown that the performance of the Fast RBF algorithm is less sensitive to the increasing number of point landmarks as compared to competing algorithms without substantial loss of accuracy.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available