Cuda-based 3d Real-time Body Of Medical Image Rendering Technology

3D visualization technology is one of the hot research field in medical image science. It is interdisciplinary science that covers computer graphics, digital image processing, biological medicine and so on. The data in medical image is always large, and the data generated by the new generation of CT and MRI equipment have higher resolution. For such huge data, the existing 3D visualization algorithm based on CPU is hard to reach real-time interactive display effect. Although the early accelerating visualization algorithm based on the GPGPU(general computing based on GPU) can significantly increased the rendering speed, GPGPU has some shortages such as difficult programming, limit texture that unable to load a large-scale medical image data, and so on. So GPGPU have not solves the problem of real-time interactive perfectly. CUDA(Computer Unified Device Architecture) launched by NVIDIA is a product for general computing. It has easier programming, more powerful function and more suitable for GPU general calculation. The 3D visualization algorithm based on CUDA is a good solution to solving the shortages of the algorithm based on GPGPU, and reach high real-time interactive rendering speed and good quality.This thesis researched and summarized the architecture model CUDA, and launched in-depth research around CUDA used in 3D medical image volume rendering technology. On the other hand, when the slice spacing is too large, the aliasing phenomenon will occur, and the common interpolation methods such as linear interpolation, nearest neighbor interpolation can't improve the quality of image effectively. This thesis researched how to use radial basis function for interpolation among slices in order to improve the quality of image effectively.The primary contents of this thesis are as follows:1. Proposed an interpolation method between slices based on radial basis function. We use Multiquadric function as the radial basis function, and alone the direction perpendicular to the slice layer, take the four consecutive and adjacent original sample point, then construct interpolation function from which We get the new interpolation slice, at last a new 3D volume data can be gotten. The experiment results show that the radial basis function can improve the rendering quality significantly, and the aliasing phenomenon is controlled effectively.2. Research and implement the ray casting algorithm based on CUDA. This method let CPU and GPU do the collaborative work, and let GPU do the light sampling which is a highly parallel process. The volume data is bound in texture memory of CUDA, and the sampling location is calculated by specific algorithm. The tri-linear interpolation of sampling and the mapping from gray to RGBA are completed by texture memory. In CUDA architecture, the whole process become convenient and fast.3. Improved and implemented the shear warp algorithm base on CUDA. According to the architecture of CUDA, we improved the original shear warp algorithm: reduce the twice two-dimension sampling process to once, and make a detailed derivation of Coordinate transformation; considering the CUDA thread asynchronous problem, we abandon the lookup table method in original shear warp algorithm, and do the linear interpolation using the texture memory. The experiment results show that, in a multithreaded concurrent execution, the rendering speed is very fast, the real-time interactive become available, and the rendering quality is also well.