CUDA-based Real-time Reconstruction And Stereoscopic Display Of MIP

Medical visualization is a powerful tool that offers an insight into body organs and tissues. However, its application is often restricted by its lacking of real-time interaction because of long computing time. In this paper, we achieved real-time Maximum Intensity Projection (MIP) computing by optimizing original algorithm. Furthermore, we realized interactive stereoscopic display of the MIP images with the latest 3D vision technique.Firstly, OpenMP is used to exploiting the computation capacity of modern multi-core CPUS to accelerate MIP. An acceleration factor near the number of cores of the CPU is achieved though. Then we used CUDA technology developed by NVIDIA to exploit GPU's computing performance. Two algorithms were presented here, based on texture and shared-memory respectively. Among them, the texture-based algorithm achieves real-time computing target. Then we implemented a sub-volume reconstruction algorithm which estimates the static tissue signal intensity, and adjusts MIP image contrast accordingly.Finally, we used NVIDIA 3DVISION technique to display image. MRA data is reconstructed with slice-based texture ray casting algorithm. With a 120Hz display and a pair of specific glasses, true 3D MIP images can be viewed and manipulated interactively.