AbstractIn many magnetic resonance imaging (MRI) applications, it is necessary to compare regions of interest (ROIs) on different images of the same patient. This comparison is often made difficult when the scanned tissue volume is not in exactly the same three dimensional location each time. Registration, the accurate alignment of the images through the determination of a transformation from one image space to another, is necessary so that ROIs may be compared correctly.
This thesis presents an implemented software system for the image processing and registration of MRI finger images. The particular application of this system is for patients suffering from rheumatoid arthritis. Firstly, features are derived from the images that will aid the registration process. The finger bones are considered to be the most reliable structures within MRI finger images and therefore, various image processing algorithms are applied to the images to create boundaries that are characteristic of the finger bones. In addition, a novel algorithm is presented which combines boundaries from many slices into a single image. Secondly, the rotational and translational offset between two images of the same finger is calculated. The Hough Transform is used to fit ellipses to the joint side of the two bones in the combined slice image. The displacement between the best-fit ellipse on images of the same finger provides the rotation and translation required to register the images. Finally, the calculated rotational and translational offset is applied to one of the images to register it to the other image. The system is applied to various data sets supplied by the University Hospital of Wales and is tested through fully worked examples. An analysis of the results is given.
|Date of Award||Nov 2000|
|Supervisor||Paul Roach (Supervisor)|