An ISO-surfaces based local deformation handling framework of lung tissues

Abstract

Handling the deformations of the lung tissues in successive chest computed tomography (CT) scans of a patient is a vital step in any computer-aided diagnostic (CAD) system for lung cancer diagnosis. In this paper, we propose a new nonrigid registration methodology for the segmented lung region from CT data that involves two steps. The first step globally aligns the target-to-reference CT scans using an affine transformation based on ascent maximization of the estimated mutual information of the calculated distance map using the fast marching level sets method inside the segmented lung for both the target and reference objects. The second step is the local alignment of the target lung object in order to correct for intricate relative deformations due to breathing and heart beats. The local deformations are handled based on displacing each voxel of the target over evolving closed equi-spaced surfaces (iso-surfaces) to closely match the reference object. In order to displace the voxel on the iso-surfaces of the target lung object, the initial voxel-to-voxel match between target and reference lung objects is estimated by solving the 3D Laplace equation between each two corresponding iso-surfaces on the reference and target objects. Finally, the estimation of voxel-to-voxel match is refined through iterative energy minimization using a generalized Gauss-Markov random field (GGMRF) model. Qualitative and quantitative results demonstrate the promise of the proposed nonrigid registration framework.