Wei Xia BEng, Abstract Co-Author: Nothing to Disclose

Xin Gao PhD, Presenter: Nothing to Disclose

Lei Wang PhD, Abstract Co-Author: Nothing to Disclose

Zhiyong Zhou, Abstract Co-Author: Nothing to Disclose

Ran Zhang BEng, Abstract Co-Author: Nothing to Disclose

Respiratory movement of the lung may bring trouble to radiation treatment planning and interventional treatment. A pulmonary respiration model needs to be established by 3D deformable registration of CT lung images. However, such registration process is a difficult and time-consuming task. This work’s aim is to reduce registration time and improve accuracy.

Registration of lung regions was performed on CT images from 4 patients, including 2 patients with nodules and 2 patients without lung disease. Pulmonary surfactant and vessels were first segmented from the lung images. The two parts was represented by point set. Each set of points was registered using coherent point drift (CPD) to obtain corresponding point displacement vectors. Then the squared Euclidean distance between point displacement vectors and the transformation vectors obtained by cubic B-spline was minimized as an objective function by the L-BFGS-B method. In addition, the component of transformation vectors was restricted to a certain size to avoid the folding of the image. Finally, the transformation obtained by point set registration was used as the initial parameters of a intensity information based registration which used mutual information, cubic B-spline and L-BFGS-B. The final transformation is the combination of the transformation derived by point set registration and intensity-information registration. A semi-automatic landmark annotating system was used to generate the corresponding landmarks for evaluating the accuracy by calculating distance error. 150~250 landmarks were detected in each image pair.

Compared with the method which was only based on mutual information, the experimental result showed that the proposed method reduced as much as 70% of the time consumption and had smaller (0.5%~13%) distance error.

The results demonstrated that the proposed method is able to accomplish deformable registration of 3D CT lung images with less time and better accuracy, which is useful to build pulmonary respiration model and localize the instrument in lung.

Xia, W, Gao, X, Wang, L, Zhou, Z, Zhang, R, 3D Deformable Registration of CT Lung Images Based on Point Set and Intensity Information .  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13014378.html