Study On The Techniques For 3D Reconstruction And Visualization Of Medical Images

3D reconstruction from medical images is a multi-disciplinary subject. It is an important application of computer graphics and image processing in biomedicine engineering. It relates to the subjects of digital image processing, computer graphics and some related knowledge of medical. 3D reconstruction and visualization of medical images are widely used in diagnostic, surgery planning and simulating, plastic and artificial limb surgery, radiotherapy planning, and teaching in anatomy. Also, the reconstruction method can be used in reverse engineering to build CAD models, such as to build solid models based on industry computed tomography (ICT) images. Study on 3D reconstruction from medical images has important significance on science and worthiness in practical application.The main research contents of 3D surface reconstruction from medical images include image pre-processing, such as interpolating and filtering, segmenting and extracting tissues or organs of body, constructing 3D surface models, simplifying of meshes of the models, and cutting or stereo clipping the model. In this dissertation, key techniques for 3D reconstructing from medical images (such as CT, MRI images) are studied, and a 3D reconstruction system used for radiotherapy is developed.Segmenting and extracting tissues or organs from medical images are premises of 3D reconstruction models accurately. In this dissertation, an interactive segmentation method of 3D medical images based on gray-level information is presented. The segmentation process consists of following steps: filtering the images, creating the threshold values after analyzing the gray-level histograms of the whole 3D image and some slice images, binarizing the images, processing the images using properly mathematical morphology operation according to the feature of tissues or regions to extract, and filling the regions using seed fill algorithm.Marching cubes (MC) algorithm is a classical algorithm to extract iso-surface from regular volume data. Marching tetrahedra (MT) algorithm is improved based on MC algorithm. This dissertation has implemented these two algorithm to extract iso-surface from the segmented regions. 3D surface models are built up. MT algorithm has disadvantages of low reconstructing speed and data redundancy. To improve this algorithm performance, methods to deal with the relativities among tetrahedra in one voxel and relativities among voxels are developed. With the methods, repeated interpolating calculation is avoided, and surface reconstructing is speed up.3D surface model reconstructed with MC or MT algorithm contains huge number of triangles. It is quite difficult to render them in real time. A modifiedimesh simplification algorithm based on edges collapse algorithm presented by Hoppe has been designed. Surface models are simplified by using the algorithm. Running examples show that even 90 percent triangles had been reduced, the model still maintains the feature of the original.In order to analyze section of the reconstructed tissues and observe the size and structure of inner tissues in multi-tissue reconstruction and visualization, methods of cutting and stereo clipping of the reconstructed 3D models are presented. The 3D models are cut by a plane or a polyhedron. Lists of edges and vertexes in every cut plane are established. From these lists, the boundary contours are created and their relationship of embrace is ascertained. The region enclosed by the contours is triangulated using Delaunay triangulation algorithm. The visual model still maintains its correct topology structure. With these operations, inner tissues can be observed easily and it can aid doctors to diagnose.A method for building boundary representation solid model from surface model is presented in the dissertation. The steps of building surface geometry model from contours, data structure of solid geometry model and fundamental operations of modeling are discussed. Solid model including full geometry and topology information is built from surface model. This ap...