UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

3-D data handling and registration of multiple modality medical images

Oghabian, Mohammad Ali; (1994) 3-D data handling and registration of multiple modality medical images. Doctoral thesis (Ph.D.), University College London (United Kingdom). Green open access

[thumbnail of 3-D_data_handling_and_registra.pdf] Text

Download (15MB)


The many different clinical imaging modalities used in diagnosis and therapy deliver two different types of information: morphological and functional. Clinical interpretation can be assisted and enhanced by combining such information (e.g. superimposition or fusion). The handling of such data needs to be performed in 3-D. Various methods for registration developed by other authors are reviewed and compared. Many of these are based on registering external reference markers, and are cumbersome and present significant problems to both patients and operators. Internal markers have also been used, but these may be very difficult to identify. Alternatively, methods based on the external surface of an object have been developed which eliminate some of the problems associated with the other methods. Thus the methods which have been extended, developed, and described here, are based primarily on the fitting of surfaces, as determined from images obtained from the different modalities to be registered. Annex problems to that of the surface fitting are those of surface detection and display. Some segmentation and surface reconstruction algorithms have been developed to identify the surface to be registered. Surface and volume rendering algorithms have also been implemented to facilitate the display of clinical results. An iterative surface fitting algorithm has been developed based on the minimization of a least squares distance (LSD) function, using the Powell method and alternative minimization algorithms. These algorithms and the qualities of fit so obtained were intercompared. Some modifications were developed to enhance the speed of convergence, to improve the accuracy, and to enhance the display of results during the process of fitting. A common problem with all such methods was found to be the choice of the starting point (the initial transformation parameters) and the avoidance of local minima which often require manual operator intervention. The algorithm was modified to apply a global minimization by using a cumulative distance error in a sequentially terminated process in order to speed up the time of evaluating of each search location. An extension of the algorithm into multi-resolution (scale) space was also implemented. An initial global search is performed at coarse resolution for the 3-D surfaces of both modalities where an appropriate threshold is defined to reject likely mismatch transformations by testing of only a limited subset of surface points. This process is used to define the set of points in the transformation space to be used for the next level of resolution, again with appropriately chosen threshold levels, and continued down to the finest resolution level. All these processes were evaluated using sets of well defined image models. The assessment of this algorithm for 3-D surface registration of data from (3-D) MRI with MRI, MRI with PET, MRI with SPECT, and MRI with CT data is presented, and clinical examples are illustrated and assessed. In the current work, the data from multi-modality imaging of two different types phantom (e.g. Hoffman brain phantom, Jaszczak phantom), thirty routinely imaged patients and volunteer subjects, and ten patients with setting external markers on their head were used to assess and verify 3-D registration. The accuracy of the sequential multi-resolution method obtained by the distance values of 4-10 selected reference points on each data set gave an accuracy of 1.44±0.42 mm for MR-MR, 1.82±0.65 for MR-CT, 2.38±0.88 for MR-PET, and 3.17±1.12 for MR-SPECT registration. The cost of this process was determined to be of the order of 200 seconds (on a Micro-VAX II), although this is highly dependent on some adjustable parameters of the process (e.g. threshold and the size of the geometrical transformation space) by which the accuracy is aimed.

Type: Thesis (Doctoral)
Qualification: Ph.D.
Title: 3-D data handling and registration of multiple modality medical images
Open access status: An open access version is available from UCL Discovery
Language: English
Additional information: Thesis digitised by ProQuest.
Keywords: (UMI)AAI10046100; Biological sciences; Medical images
URI: https://discovery.ucl.ac.uk/id/eprint/10098025
Downloads since deposit
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item