Segmentation of Vascular Networks from X-ray Angiograms
Occlusive vascular disease (e.g. arteriosclerosis) is a major and growing health problem worldwide. Traditionally, bypass surgery has been used to treat this condition but is ultimately unsatisfactory because it does not produce a cure. New drug treatments have attracted much research interest but a key obstacle here is the lack of any objective methodology for assessing the effectiveness of these treatments (before and after studies).
The clinical data sources are digital subtraction angiograms (DSAs) which are a routine assessment of patients presenting with symptoms of occlusive vascular disease. Novel image processing algorithms are required to analyze these DSA images which are of low quality though of high spatial resolution and likely to contain complex vascular networks.
Ultimately, we aim to model the haemodynamic capacity of the patient's vascular system using the geometric information extracted from DSAs. This can be regarded as objective, anatomically-based clinical evidence for assessing the curative effectiveness of any drug treatment, as well as for routine assessment of disease severity.
We have developed a tracking-based method of segmenting vascular networks which proceeds by fitting an imaging model to the normal intensity profile of the arterial segment, thus extracting vessel radius as well as automatically tracking the vessel from an easily identified start point. Our tracking method can also deal with vessel bifurcations.
Typical experimental results above show: (a) The original DSA image showing the vascular network, (b) The segmented vascular network indicated by the centre line (in red) and edges (green and blue lines, respectively), which were obtained by the proposed model-based tracking algorithm, and (c) The vessel segments traced using the well-known method of Sun. Red lines indicate individual scanlines. Our result (b) shows greatly improved tracking stability, more complete tracking and more reliable radius estimates.