An Automated System for 3D Segmentation of CT Angiograms - Visual Computing: Scientific Visualization and Imaging Systems - page 97

Fig. 5.2 Illustration of the coronary arteries of human, including: the right coronary artery
(RCA), left anterior descending (LAD) and left circumflex (LCX)
5.2 State of the Arts in Vessel Segmentation
Despite numerous past and on-going research efforts in the past decade, seg-
mentation of vascular structures from CT images remains a challenging topic due
to the small size of vessels in medical images and the complex pathology. Since
their introduction as a means of front propagation-based segmentation method,
active contour models (also known as snakes) received a great amount of attention
by the medical image processing community [ 6 - 12 ]. Active contour models for
image segmentation iteratively deform a contour in order to minimise a user-
defined energy functional, which often depends on the shape of the contour and its
position in the image. Such methods are usually implemented using level sets [ 13 ],
where the boundaries of the object to be segmented are embedded as the zero level
of a higher dimensional level set function. Due to their ability to deal with
topological changes, such as region merging and splitting, level sets-based active
contour models are usually employed in segmentation of the vascular structures in
medical images. Active contour-based methods can be categorised in two folders,
when considering image-driven energy: edge-based and region-based models.
In the early edge-based models [ 11 , 14 , 15 ], the active contour deforms with a
speed F based on the derivatives of the image, which approaches zero at high
image gradients. These methods make use of the local edge information to stop