A computationally efficient 3D/2D registration method based on image gradient direction probability density function

详细信息    查看全文

• 作者：Soheil Ghafurian^aAuthor Vitae ; Ilker Hacihaliloglu^bAuthor Vitae ; Dimitris N. Metaxas^cAuthor Vitae ; Virak Tan^dAuthor Vitae ; Kang Li^a ; ^b ; ^c ; ^d ; ^e ; ^{kl419@rci.rutgers.edu}Author Vitae
• 关键词：Image-guided evaluation ; Histogram of gradient directions ; Feature-based registration ; 3D/2D registration
• 刊名：Neurocomputing
• 出版年：2017
• 出版时间：15 March 2017
• 年：2017
• 卷：229
• 期：Complete
• 页码：100-108
• 全文大小：1327 K
• 卷排序：229

文摘

Three-dimensional (3D) to two-dimensional (2D) registration is an essential problem in many medical applications. This problem aims at finding the rigid transformation parameters to match the projected image of a 3D model to the real one to estimate the 3D pose of the anatomical model. This class of image registration is computationally intensive due to the large number of solution assessments necessary to search the complex solution space. Moreover, the convergence of the solution process is contingent on a manual initialization of the solution close to the optimal solution. In this paper, we address both of these challenges by introducing a registration method which is significantly faster and less sensitive to initialization than existing methods. The method explores the properties of image gradient probability density function for registration and uses a weighted histogram of image gradient directions (WHGD) as the image feature. This simplifies the computation by searching the parameter space (rotations and translations) sequentially rather than simultaneously. Our experiments demonstrated that the proposed method was able to achieve sub-millimeter and sub-degree accuracy with 5% of the solution assessments needed by an established existing method. The accuracy was not sensitive to the initial solution as long as it was within 90° and 30 mm of the true registration, which is a substantial improvement over the existing methods.