A Novel Regularization Model for Demons Registration Algorithm on 3D Lung CT Images

The aim of this study is to propose a novel regularization model for demons registration algorithm on 3D lung cancer CT images. Accurate 3-dimensional registration between the initial and follow-up lung CT volumes remains a challenging task for the deformation at the image boundaries and the obvious anatomical structures. In order to improve the registration accuracy, the additional regularization constraints have to be introduced for describing the displacement field properties. Therefore, we propose a new method for registration, in which introduce a three-dimensional hybrid diffusion filter with continuous switch (HDCS) to substitute Gaussian filter. The proposed method exploits the benefits of edge-enhancing diffusion (EED) filtering and coherence-enhancing diffusion (CED) filtering, which not only takes into account local spatial smoothness, but also preserves edges and spherical or tubular structures in the lung. Moreover, the presented framework is capable of reducing the large deformations of lung shapes mismatch between the initial and follow-up lung CT volumes. Finally, to ensure the preservation of the objects biological topology, our proposed method is implemented in the space of diffeomorphisms, in which meaningful biological shapes can be found. Registration experiments between the initial and follow-up lung CT data sets show that our proposed method outperforms the state-of-the-art demons algorithm.