ABSTRACT
This chapter exploits the intralayer self-closure structure of the optimal single surface detection problem, and formulates it as a minimum-cost closed set problem based on a nontrivial graph transformation scheme. It explores the interlayer self-closure structure of the pairwise interacting surfaces, which again enables us to model the optimal multiple surface detection (OMSD) problem as a minimum-cost closed set problem. The chapter utilizes the weight of both graph nodes and arcs to represent the desired segmentation properties for optimal single- and multiple-surface segmentation, which can incorporate a wide spectrum of constraints into the problem formulation. It focuses on convex smoothness penalty functions that are widely used in medical image processing and in Markov Random Fields. The chapter introduces the LOGISMOS method for multiobject, multisurface segmentation. It demonstrates its functionality on a knee bone/cartilage segmentation example.
