ABSTRACT
This chapter surveys the field of energy minimization as it applies to medical image segmentation (MIS). MIS remains a daunting task but one whose solution will allow for the automatic extraction of important structures, organs, and diagnostic features from medical images, with applications to computer-aided diagnosis, statistical shape analysis, and medical image visualization. Several classifications of segmentation techniques exist, including edge-, pixel-, and region-based techniques, in addition to clustering, and graph-theoretic approaches (Pham et al., 2000; Robb, 2000; Sonka and Fitzpatrick, 2000; Yoo, 2004). However, the unreliability of traditional, purely pixel-based methods in the face of shape variation and noise has caused recent trends (Pham et al., 2000) to focus on incorporating prior knowledge about the location, intensity, and shape of the target anatomy (Hamarneh et al., 2001). One type of approach that has been of particular 662interest to meeting these requirements is that of energy minimization methods due to their inherent ability to allow multiple competing goals to be considered.
