Research On Optimized Implementation Of Volume Rendering Based 3D Visualization

Visualization of medical volume data is a process to reconstruct the discrete 3D data by interpolation, and then transform the data into image with 3D visual effect. It can be used to assist the doctor to plan the surgery, and simulate the operation using computer, which can greatly improve the accuracy and scientific of medical diagnosis. Volume rendering is a widely used technology for 3D visualization. But due to the large amount of computation, it often requires hardware support. For this reason, scholars have put forward a variety of methods to improve it. Most of them focused on the optimization at the design level.The thesis focuses on the optimization at the implementation level. After carefully comparing various wide-used volume rendering methods, the thesis devotes most of its effort to the ray-cast algorithm for volume rendering. Based on the volume rendering implementation in Visualization Toolkit (VTK), the thesis locates the bottleneck of the whole algorithm through systematic analysis and research, and then fully exploits the potential of the existing PC by carefully applying the fine-grained and coarse-grained optimization. The optimizations focus on the architecture of the processors. Through the analysis of the single instruction multiple data (SIMD) instructions and cache usage, the thesis applies the corresponding optimization strategy to reduce calculation and improve the efficiency of the algorithm.The thesis analyzes and optimizes the two bottlenecks of ray-cast algorithm: ray generation and re-sampling. To tackle the first bottleneck, Streaming SIMD Extension (SSE) is used to optimize the floating operation, and the penalty of branch mis-prediction is reduced. To tackle the second bottleneck, SSE2 is used to optimize the tri-linear interpolation, and performance penalty of cache miss is decreased by using data-blocking.In the thesis, various data sets with different characteristics are used to design the test cases. Test results show that, the optimized algorithm can obtain 30% to 50% efficiency gain without hardware upgrades, so ordinary PC can also obtain the better visualization results provided by the volume rendering algorithm.