Globally Optimal Energy Minimization Techniques - Bezier and Splines in Image Processing and Machine Vision - page 225

is needed is a fast, cheap, and reliable method to extract the shape of the
hippocampi from brain MRI that could be used as a screening test for early
Alzheimer's disease. Such a test may allow health workers to intervene before
serious brain damage occurs.
10.5 Conclusions
These globally optimal energy minimization methods are fast, easy to apply,
and tend to yield robust solutions. When using conventional active contours
based on local energy minimization, a great deal of effort is expended in
developing techniques for choosing the initial position of the contour, escap-
ing local minima, and determining stopping criteria. It is certainly true that
some effort must be expended on determining the search space and the en-
ergy function when using global energy minimization techniques. Yet, in our
experience, these techniques are much simpler to apply in practice and yield
more robust and accurate results. Note further that by carefully positioning
the search space, global energy minimization techniques can always find lo-
cally minimal energy solutions. In particular, for the globally minimal surface
approach, multiple sources and sinks can be used to guide the solution, pro-
viding many of the purported advantages of the original snakes of Kass et al.
The converse, however, is not true—local energy minimization techniques are
never guaranteed to find global solutions.
Future work is focussed on integrating these techniques with statistical
shape models to develop an 3D Expectation Maximization algorithm incorpo-
rating prior shape knowledge for detection and segmentation of known shapes.