Biomedical Engineering Reference
In-Depth Information
Chapter 4
Deformable Models in Medical Image
Segmentation
Matthias Becker and Nadia Magnenat-Thalmann
4.1 Introduction
Medical imaging nowadays is part of the routine in hospitals. The acquired images
get better in quality and improvements are made to reduce the exposure of patients to
radiation. Modern techniques allow to capture small details and to differentiate be-
tween kinds of soft tissue. But with the increased number of images and their higher
resolution, interpretation has become a more complex task. Nonetheless, medical
image segmentation is required for applications like radiotherapy, preoperative plan-
ning and postoperative evaluation. Medical image segmentation, as described by
Elnakib et al. [
1
] in a survey, is the process of identifying regions of interest in
images. Approaches range from simple ones that only exploit intensity values or
region information to model-based ones that include a priori knowledge. The images
often suffer from noise, aliasing and anomalies or may contain gaps in boundaries,
providing challenges that are hard to handle with non model-based approaches.
Deformable models have been first proposed by Terzopoulos et al. in 1987 [
2
].
They provide a robust segmentation approach that uses bottom-up image-based con-
straints and top-down constraints from prior knowledge. Deformable models can be
curves or surfaces (or of higher dimension, e.g. for temporal segmentation). They
evolve under the influence of internal and external energies. The internal energy con-
trols the curves smoothness and the external energy aims to attract the model towards
boundaries in the image domain. Deformable models are an interesting approach as
they combine geometry (to describe the shape), physics (to simulate the behavior)
and approximation theory (for model fitting).
