Higher-order graph cuts and medical image segmentation

Energy minimization is regularly used for medical image segmentation. Higher-order energies are perhaps not as common, but are nevertheless being used increasingly often. Whereas the common first- order (pairwise) potential can directly model only the relationship between pairs of pixels, the higher-order potential can model more complex and useful relationships between more than two variables. For instance, sets of pixels, chosen according to the shape to be segmented, can be encouraged to be entirely in one segment or the other by higher-order terms. In this talk, I will describe methods for minimizing higher-order energies using graph cuts as well as some real-world examples of their applications in medical image segmentation that have been deployed in commercial medical imaging software.