Three-dimensional reconstruction of cerebral vasculature

This thesis examines the few-view reconstruction problem as it applies to imaging cerebral vasculature using digital subtraction angiography. A fully automated reconstruction algorithm is developed that circumvents the traditional 'correspondence problem' using only notions of consistency and connectivity. Here it is assumed that the object is a connected structure and that its centre-lines have been identified in 3 or more images. The first of three steps in the procedure involves generating a connected structure that represents the multiplicity of solutions that are consistent with any two (different) projections of the vascular tree. The second step assigns to each branch in this structure a measure of consistency based on its agreement with a third (or additional) view of the object. The problem then becomes one of propagating this information, via connectivity relationships and other consistency checks, throughout the above structure to distinguish between the branches comprising the imaged structure and the accompanying artifacts.This thesis also addresses the issue of how different imaging geometries affect the outcome of the reconstructed data. Here it is suggested that it may be possible to favourably affect the reconstruction by selecting certain projections based on what is known after two or more views.Simulations using mathematical models of varying complexity and experiments based on actual image data are used to validate the methods described here and demonstrate that it is possible to achieve adequate reconstructions with as few as three distinct views