An Automated System for 3D Segmentation of CT Angiograms - Visual Computing: Scientific Visualization and Imaging Systems - page 99

Fig. 5.3 Graphic demonstrating the effect for the shape filter. a Shows of the shape filter centred
at x and with the radius R. b The output of the shape filter [ 17 ]
Another issue in the segmentation of vascular structures in coronary CT images
is the so called 'kissing vessel artefacts', i.e. the non-coronary vessels in close
proximity to the arteries. This is a partial volume problem which is often
encountered in CT angiograms [ 23 ]. It may result to artificial vessel junctions and
thus distort the geometry of the vessels. To address this problem, Wong and Chung
[ 24 ] proposed the tracing of vessel centrelines and segmentation of their cross-
sections based on a probabilistic vessel-axis tracing framework. Their algorithm
allows user's interaction to produce the desired traces through the abnormal
regions, which contain the kissing vessels, lesion vessels (e.g. the stenosis) and
vessel junctions. The final segmentation in these regions is then derived from the
axis of the cross-sections.
5.3 Proposed Framework
This section presents the proposed two-step system for segmentation of coronary
arteries in CTA images by using both global and local intensity statistics. We deal
with the varying image brightness characteristics by computing regional statistics
locally, in the neighbourhood of the active contour. The global intensity constraint,
on the other hand, is utilised to evolve the contour to the desired boundaries
without being trapped in local minima. The proposed approach is implemented in a
Bayesian probabilistic framework to incorporate these two homogeneity con-
straints. Possible outliers, such as kissing vessel artefacts, are removed in the
subsequent stage by a novel slice-by-slice correction scheme, which uses multiple
regions competition.