| 英文摘要 | Three-dimensional medical visualization technique has been widely used in the fields of medical diagnosis, research and teaching. It made the dream of visualizing noninvasively human internal organs in their true shape and form become true. Since the data of 3D imaging is of large size and the reconstruction is time consuming, 3D reconstruction and 3D rendering are the key technology and difficult points. My dissertation was mostly based on 3D Medical Image Processing and Analyzing System (3dmed), which our lab has succeeded in developing it firstly in China with our own property right. In 1999, 3dmed gained the top three awards of "Asian Major 10 CT Scientific and Technological Advances". During the three years of doctoral work, as the project manager of Medical Image Processing Group, I attend a lot of research, development and management work.' In this paper, the digital geometry processing tools were applied to the medical imaging visualization, which includes edge detection, 3D reconstruction, mesh simplification, level of details and mesh optimization etc, and all the experiments were implemented in the 3dmed system. I summarize my three years research work in this paper. The main work of my dissertation is as follows: 1.A new surface generation scheme was proposed, which integrates segmentation and Marching Cubes algorithm. An appropriate segmentation method was applied to the image sequence according to the feature of the original medical image, and then the segmentation result (binary image data set) was used as the input of MC to generate iso-surfaces. In addition, we develop a surface-tracking algorithm based on region growing, which improve the efficiency by avoiding detecting empty space. The experiments show that the reconstructing speed is as ten times fast as the speed of standard MC, and it can reconstruct the 3D model which the standard MC can't. Another reconstruction of huge datasets was proposed, which is based on single surface tracking. We set up the caching mechanism and use triangle strip to represent the 3D model. We use the Visible Man dataset to do the experiments, and the results show that our method need less memory (just 1/3 of VTK), and the reconstructing speed is as 5-8 times fast as VTK, and the rendering time is just 1/4 of VTK. 2.Construct the level of details model with view-dependent subdivision, and apply it to virtual endoscope. First we simplify the mesh in hierarchy to get the base mesh, and then parameterize the mesh to map the original mesh to the base mesh, and use the view-dependent subdivision to resample. In the implementation of the view point principles, we adopt octree to index the change of the view point, and the model can be reused in the dynamic change. The experiments show that our method is efficient and is easy to implemented, and the model can be rendered in real time. The speed of our method is 200 times faster than Hoppe's VDPM, which can meet the requi |
修改评论