Biomedical Engineering Reference
In-Depth Information
4.9 Conclusion
We have presented an overview of the current state of deformable models in medical
image segmentation. They are a powerful tool for the given image conditions and
therefore play an important role. We have surveyed current snakes and level sets
methods that are examples for continues shape representation. These usually two-
dimensionally used approaches achieve good results with high computational costs.
However, they are highly dependent on a good initialization. Contrary to snakes,
level sets are able to adapt to different topology automatically. Discrete deformable
models are three-dimensional approaches, which can be represented as particle sys-
tems or through different mesh types. They can also be based on several physical
foundations. They provide great flexibility and haven proven useful in many cases.
Depending on the structure of interest, deformable models have to be carefully con-
figured and parameterized. An extension to deformable models is the incorporation
of prior knowledge to make them more robust in complex image conditions. These
knowledge-based deformable models exploit prior knowledge on shape and appear-
ance. They need a large training base and complex training but are very robust. We
have presented several forces that cover the tools needed for creating deformable
models. We have also extended the view by presenting other deformable model
approaches. Lastly we have presented the main approaches to the important initial-
ization problem.
A challenge for future research is the development of more general approaches.
Although very sophisticated and robust, deformable model applications are tuned
to specific organs and well-defined image conditions. Knowledge-based models for
example have to be trained with a large database of examples and image forces
like the intensity profiles have to be recreated if imaging sequences are modified.
Depending on the chosen approach, the initialization can be a critical task. Speed
and robustness still can be improved.
Acknowledgments
Thiswork has been funded by the EUFP7MarieCurie Initial TrainingNetwork
project MultiScaleHuman
(
http://multiscalehuman.miralab.ch
)
under Grant No. 289897.
References
1. Elnakib, A., Gimel, G., Suri, J. J., El-baz, A., & Gimel'farb, G. (2011). Medical image
segmentation: A brief survey. In A. S. El-Baz, U. R. Acharya, A. F. Laine, & J. S. Suri (Eds.),
Medical image segmentation
(pp. 1-39). New York: Springer.
2. Terzopoulos, D., Platt, J., Barr, A., Fleischer, K., Terzopoulost, D., & Fleischert, K. (1987).
Elastically deformable models.
SIGGRAPH Computer Graphics
,
21
(4), 205-214.
3. Nealen, A., Müller, M., Keiser, R., Boxerman, E., & Carlson, M. (2006). Physically based
deformable models in computer graphics.
Computer Graphics Forum
,
25
(4), 809-836.
4. Baraff, D., & Witkin, A. (1998). Large steps in cloth simulation. In
Proceedings of the 25th
Annual Conference on Computer Graphics and Interactive Techniques—SIGGRAPH '98
,
New York, USA, 1998 (pp. 43-54). New York: ACM Press.
