Automatic Lung Segmentation of Volumetric Low-Dose CT Scans Using Graph Cuts

We propose a new technique for unsupervised segmentation of the lung region from low dose computed tomography (LDCT) images. We follow the most conventional approaches such that initial images and desired maps of regions are described by a joint Markov-Gibbs random field (MGRF) model of independent image signals and interdependent region labels. But our focus is on more accurate model identification for the MGRF model and the gray level distribution model. To better specify region borders between lung and chest, each empirical distribution of volume signals is precisely approximated by a linear combination of Gaussians (LCG) with positive and negative components. LCG models parameters are estimated by the modified EM algorithm. Initial segmentation (labeled volume) based on the LCG models is then iteratively refined by using the MGRF with analytically estimated potentials. In this framework the graph cuts is used as a global optimization algorithm to find the segmented data (labeled data) that minimize a certain energy function, which integrates the LCG model and the MGRF model. To validate the accuracy of our algorithm, a special 3D geometrical phantom motivated by statistical analysis of the LDCT data is designed. Experiments on both phantom and 3D LDCT data sets show that the proposed segmentation approach is more accurate than other known alternatives.