3 D Medical Image Rendering Technology Research And Implementation Based On Cuda

Medical image registration and3D visualization technology is a hot research fieldof medical images. Medical image registration is a prerequisite to achieve theintegration of medical images，make the corresponding point on the spatial location andanatomical structure to achieve complete agreement through the transformation forspace. Medical image registration to provide a more comprehensive and accurate forthree-dimensional reconstruction image, thereby improving the accuracy andeffectiveness of diagnosis and treatment. Three-dimensional visualization of medicalimages that can be scientifically and accurate reconstruct the detected objects,toovercome the uncertainties caused by the limitations of medical imaging equipment, inorder to carry out qualitative and quantitative analysis. Visualization of medical imagesprovides physicians with a more realistic display, easy to multi-angle, multi-levelobservation and analysis. Medical images has been widely used in disease diagnosis andtreatment.3D medical reconstruction of various organs for doctors to have the diseaseanalysis, the use of computer simulation and planning of radiation therapy andorthopedic surgery. Medical image reconstruction due to the large amount of data, it isdifficult to achieve real-time interaction, using GPU powerful parallel computing powercan get high-quality rendering quality on the basis of real-time interaction. CUDA(Compute Unified Device) is NVIDIA’s GPU Computing products, the use of classC-language development, programming is simple, powerful, architecture system forGPU Computing.This thesis researched rigid registration of medical images and3D visualizationtechnology, Summarizes the classic algorithm of three-dimensional visualizationtechniques, proposed an accelerated medical image registration methods, and improvedrendering algorithm, so that whether it is to have some degree of improvement inrendering speed and rendering quality. On the other hand, for the rendering algorithmhuge computing and massively parallel computing features to achieve the draw based onthe CUDA accelerated to achieve a real-time interactive drawing requirements.The primary contents of this thesis are as follows: 1. Summarizes the classical algorithm for three-dimensional visualizationtechnology. Focuses on the MC surface rendering algorithms and ray tracing, and sheardeformation volume rendering algorithm and proposed optimization strategy based onsurface rendering, volume rendering.2. Researched medical image registration methods. Proposed method to Speed upthe rigid deformation medical image registration. This method use mutual informationas similarity measure which is more suitable for medical image registration. The mutualinformation do normalized to reduce the overlap region sensitivity and are not easily fallinto local minima. Improved Powell algorithm to optimize the alignment parametersmore accurately. Change to optimize the search for the initial value, reducing the time ofregistration.3. Improved two types of classical algorithm based point. The use of point octreesubdivision to achieve Splat algorithm of the adaptive level of detail. Proposed twomethods to solve the aliasing problem of the traditional sure splatting method. Shearperspective transformation to solve aliasing and improve the rendering efficiency,Filtering based on elliptical weighted and choose a different filter according to thedifferent distances also solve the aliasing problem.4. Achieved cuda acceleration point draw. Many calculations based on pointrendering algorithm in parallel to achieve,and the architecture of the GPU’s its verysuitable for parallel computing.Concurrent execution of multiple threads, capable ofreal-time interactive rendering image.