Model-based Segmentation with Active Contour Models
Another interactive segmentation tool is based on the T-snakes approach [1,2]. Snakes can be compared to balloon blown up inside a mould. In image processing the mould corresponds to the edges and the elastic properties of the balloon to contour/surface properties of an expanding closed contour/surface. A segmentation is realized when image forces and surface forces are balanced.
Segmentation with snakes is interpreted as an optimization of an energy functional including terms for image features and contour properties. The Segmentation starts with an initial two or three dimensional contour template which allows for locally defined segmentation parameters. Segmentation is an iterative optimization using gradient descent. The T-snakes approach was extended by calculating the image features separately for the x- and y-direction leading to a better point distribution on the contour. This results in more accurate segmentation results and better preservation of the topology of the points during segmentation. A reparameterization of the contour during segmentation ensures the topological adaption of the contour to the image data and a uniform distribution of the points on the surface. The final segmentation result can be influenced by user-defined weighting of smoothness of the contour and its alignment to edges in the image data.
The method was tested on CT scans of the prostate region with special emphasis on the segmentation of the bladder. The segmentation tool was evaluated with respect to its accuracy and efficiency. The results show that the algorithm is capable of separating the bladder from the prostate and that it can cope with moderate image artifacts. The efficiency of the algorithm can be described by the time it takes to segment an organ at risk. With the implemented approch a bladder can be segmented within three seconds on a standard PC (3 GHz) and a 3D cylindric template as initial contour.
Initial Contours for the 3D segmentation (left). Intermediate segmentation step. Final segmentation result (right)
© dkfz.de
References
[1] Kass M, Witkin A, Terzopoulos D: Snakes: Active Contour Models. Int J Comput Vis 1(4):321–331, 1988.
[2] McInerney T, Terzopoulos D: T-snakes: Topology adaptive snakes. Med Image Anal 4(2):73–91, 2000.