Energy functionals for medical image segmentation: choices and consequences

Medical imaging continues to permeate the practice of medicine, but automated yet accurate segmentation and labeling of anatomical structures continues to be a major obstacle to computerized medical image analysis. Though there exists numerous approaches for medical image segmentation, one in particular has gained increasing popularity: energy minimization-based techniques, and the large set of methods encompassed therein. With these techniques an energy function must be chosen, segmentations must be initialized, weights for competing terms of the energy functional must be tuned, and the resulting functional minimized. There are a lot of choices involved, and their consequences are not always clear. In this thesis we explore the different consequences of these choices, and provide novel methods to overcome two of the more significant problems encountered: local minima, and parameter settings.