3D Segmentation of the Lung Based on the Neighbor Information and Curvature

A novel method for the automatic segmentation of the lung in X-ray computed tomography (CT) images is presented. In this paper, a maximum a posteriori (MAP) estimation framework, combining neighbor prior information and image gray level information, is used to extract the boundary of lung. The relationship of the left lung and the right lung is represented as a joint density function. We use the principal component analysis (PCA) to build the neighbor prior model in a set of training images. A double dimension reduction algorithm is developed to improve the efficiency. The model is formulated in terms of level set functions, and the surfaces evolve according to the associated Euler-Lagrange equations. Then we propose a new algorithm to refine the rough boundary generated by the MAP framework. This algorithm consists of two stages: 1. automatically detecting and rough fitting the region of lung hilum, 2. refining the fitting curve based on the curvature information.