# Graph Cut-Based Fine Segmentation of the Lung

The energy function to be minimized has a unary and pairwise terms. The unary term consists of three sub-terms: the likelihood term, the probabilistic atlas term, and the neighbor constraint term. In the postmortem lung segmentation, it is important to balance the probabilistic atlas term with the likelihood one, the latter of which is calculated using the rough segmentation result. So the proposed adaptive weights are defined by the Jaccard index between rough segmentation results and its most similar shape generated by a lung SSM. Note that the lower JI means more irregular shape of rough segmentation, most of which were caused by the failure in segmentation owing to severe pathologies and/or postmortem changes. Specifically, the JI is used as a weight for the probabilistic atlas term and (1-JI) is for the likelihood term. The neighbor constraint term is introduced to force the bottom surface of lung closer to top surface the liver extracted by the algorithm in the previous section.