| 英文摘要 | Virtual endoscopy is an integration of medical imaging, image processing, computer graphics and virtual reality techniques. It is meaningful for medical diagnosis because compared to traditional fiberopic endoscopy, it is noninvasive, cost-effective, highly ac- curate, and free of risks, easily tolerated by the patient, and furthermore, it can make physicians examine the regions that traditional endoscopy can not reach such as blood vessel. Virtual endoscopy is a new tool for medical diagnosis, surgery planning and opera- tion, which uses 3D medical images as information source, medical image processing and analyzing techniques as basis, computer graphics and virtual reality techniques as implementation tools, and aiming at noninvasive navigation and examination inside hu- man organs such as trachea, blood vessel and colon. Virtual endoscopy is multi-disciplinary, touching on many techniques in medical im- aging, pattern recognition, and computer graphics. This dissertation mainly concentrates on some key techniques in virtual endoscopy, such as centerline extraction, fairing sur- face generation, real time rendering, automatic and guided navigation etc. The contribu- tion of this dissertation is as follows: 1. A generic system framework for virtual endoscopy is proposed, including data aquisi- tion, image segmentation, centerline extraction, surface generation, real time ren- dering, camera control and virtual navigation. We analyzed some key techniques and presented our scheme in the corresponding chapter. 2. A new centerline extraction algorithm based on Hessian matrix was proposed. First, the distance transformation is performed. Then the initial path is obtained by com- puting the eigenvalues and eigenvectors of Hessian matrix. After that, the visibility test with adaptive visibility sphere radius, which is determined by the eigenvalues of Hessian matrix, is performed to remove useless voxels in the centerline. Finally, the path, with all the points staying away from the surface, is generated by Dijkstra's shortest path algorithm. Our method has two advantages. The first is that the visibility sphere radius is adaptive, not user-defined. The second is that the ridge extraction avoids the test to the large amount of useless voxels in the visibility test procedure. Experiments proved that the time cost in visibility test is 2-5 times less than Reliable Path method. 3. A fast mesh simplification algorithm combining half-edge data structure and modified quadric error metric (QEM) was presented. When half-edge structure is used, the adjacency queries between components of the mesh, such as vertices, faces and edges, can be quickly achieved and thus the run time is reduced remarkably. Ex- periments show that the initialization speed is 2 times as fast as QEM, and the; sim- plification speed is 2-4 times as fast as Q |
修改评论