A New Method for Medical Image Registration Based on Deformable Models: Application for Thorax CT Images

In 4D CT images of the thorax, deformation is caused by both the tissue movement and breathing. Tissue movement can be estimated at various time intervals by image registrstion. In this paper, a new deformable model is proposedfor non-regid medical image resgistration. The energy funational of the suggested algorithm consists of two internal (regulator) and external terms. By minimization of the energy functional, the optimal vector field which indicates the translation vectors of image pixels is obtained. The external energy is developed based on the well-known least-square-errors cost function (with the scaling coefficient of gray-levels). However, for local implementation of the translation, scaling, and rotation, we proposed minimization of the amplitude of the second-derivative of the vector filed as the internal energy term. For solution quality assessment, a number of 3D thorax CT images (available at the database http://www.dir-lab.com) at the end-inhale and end-exhale phases were used. In all benchmark images, the precise positions of a number of anatomical index points had been indicated by an expert. The performance of the proposed method can be evaluated by computing the distance between the corresponding index points in two images after registration. Experimantal results demonstrated that the average registration error of the proposed algorithm was at least 8% better than those of 11 counterpart algorithms.