Abstract
X-ray fluoroscopy is an important tool in modern medicine and is used ever more frequently, particularly in the field of interventional neuroradiology. The health risks associated with X-rays are well known, and there is an impetus to reduce the levels of X-ray exposure to both patients and staff. This dose reduction effort must take place on many levels, from improvements in machine hardware to improved clinical techniques. The digital nature of modern X-ray fluoroscopy allows software methods to take a vital role in reducing X-ray exposure. The reduction of image noise and the enhancement of the appearance of vessels can aid the analysis of the data by eye and act as a prerequisite for further processing. Much of the work in the literature has focused on the reduction of image noise through the use of filters, be they spatial, frequency, temporal or a combination of the three. This thesis examines the effect of two different approaches to digital subtraction angiogram enhancement. The first method described is a nonlinear data fusion technique whereby individual frames in an angiogram sequence are regarded as separate 'sensors'. Each sensor votes for a certain intensity value at each pixel location. The fusion system takes account not only of the votes polled but also the neighbourhood surrounding each pixel to best estimate the true level of X-ray absorption. The second enhancement method models pixels as relaxation oscillators, with period determined by pixel intensity. A coupled network is constructed, first in two dimensions incorporating spatial information within the image, then in three dimensions, incorporating temporal information too. As the network evolves, neighbouring pixels with similar values become synchronous with one another, grouping together into homogeneous regions. This synchrony is manifest in the output as reduced noise and a more coherent structure to image regions.