Three-dimensional Reconstruction Of Medical Images

3D reconstruction of medical images is a hot point of research in medical images processing domain at present. It can provide 3D graphics with real senses to users by getting 3D structure informations from 2D images. So, it is widely used in diagnosing, surgery layout, simulated emulation, etc.3D reconstruction include two main methods: surface rendering and volume rendering. Volume rendering has a large scale of computing, and highly depends on hardware. Both contour rendering and isosurface rendering are menthods of surface rendering. Because the former has the character of random, complex computing and poor effect of results of reconstruction, we use the method of isosurface rendering to reconstruction tissues and organs here. In the area of isosurface rendering, MC (Marching Cubes) is the classical algorithm. The algorithm builds voxel between neighboring slices of volume data , then gets density of a certain substance by experience. After setting threshold for surface of the substance which we want to get, and computing triangle mould of every voxel cube by use of linearity interpolation, the algorithm represents the surface of the substance as the form of triangle mesh. MC is a method of accurately defining voxel and isosurface , and becomes a member of pop algorithms of 3D reconstruction now, so it is applied to most merchant software.But the standard MC has some shortcomings: firstly, the standard MC picks up isosurfaces by threshold, however, threshold segmentation is invalid for picking up tissues or organs from some medical images; secondly, the standard MC pocesses cubes one by one, that is to say, all the cubes will be checked, and the algorithm spents 30%-70% of time to check the null units, so we need a reasonable data structure to travel the space data and accelerate the checking or filting of null units; thirdly, the standard MC has a large scale of triangles, normally, the tissue or organ reconstructed includes hundreds of thousands so much as millions of triangles, this means it hardly to execute real-time rendering or interaction; lastly, the standard MC can not get the very smoothly surface mesh, and there will be some unexpected accidented cases, especially in the case of big errors in oringinal data.