Premium
SU‐E‐J‐88: Deformable Registration Using Multi‐Resolution Demons Algorithm for 4DCT
Author(s) -
Li Dengwang,
Yin Yong
Publication year - 2012
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4734923
Subject(s) - image registration , computer science , algorithm , artificial intelligence , similarity (geometry) , process (computing) , computer vision , image (mathematics) , operating system
Purpose: In order to register 4DCT efficiently, we propose an improved deformable registration algorithm based on improved multi‐resolution demons strategy to improve the efficiency of the algorithm. Methods: 4DCT images of lung cancer patients are collected from a General Electric Discovery ST CT scanner from our cancer hospital. All of the images are sorted into groups and reconstructed according to their phases, and eachrespiratory cycle is divided into 10 phases with the time interval of 10%. Firstly, in our improved demons algorithm we use gradients of both reference and floating images as deformation forces and also redistribute the forces according to the proportion of the two forces. Furthermore, we introduce intermediate variable to cost function for decreasing the noise in registration process. At the same time, Gaussian multi‐resolution strategy and BFGS method for optimization are used to improve speed and accuracy of the registration. To validate the performance of the algorithm, we register the previous 10 phase‐images. We compared the difference of floating and reference images before and after registered where two landmarks are decided by experienced clinician. We registered 10 phase‐images of 4D‐CT which is lung cancer patient from cancer hospital and choose images in exhalationas the reference images, and all other images were registered into the reference images. Results: This method has a good accuracy demonstrated by a higher similarity measure for registration of 4D‐CT and it can register a large deformation precisely. Finally, we obtain the tumor target achieved by the deformation fields using proposed method, which is more accurately than the internal margin (IM) expanded by the Gross Tumor Volume (GTV). Furthermore, we achieve tumor and normal tissue tracking and dose accumulation using 4DCT data. Conclusions: An efficient deformable registration algorithm was proposed by using multi‐resolution demons algorithm for 4DCT.