The Design And Implementation Of Subtraction CTA Algorithm Based On CUDA

In recent years, CTA, as a non-invasive vascular examination method, has become the indispensable tool of angiography of head. However, CT angiography does not allow a straightforward understanding of vascular structure, since some of interesting parts can be located near bone and skull region. Therefore, the CTA images needs to be silhouetted to remove bone region to improve the display effect of vessels. The way to get boneless CTA images is to increase a CT scan before CTA scan, and to get the bone mole from CT images, and then to use the bone mold to silhouette CTA images, so as to achieve the purpose of subtraction CTA. Generally, the computational of substraction CTA algorithm is so heavy that the calculation time is too long to be tolerated on clinical applications.GPU(Graphic Processor Unit) is a parallel-computing device. Its powerful parallel computing capability makes it have good application in industrial design, game development, medical image processing and many other fields. And CUDA(Compute Unified Device Architecture) makes GPGPU (GPU general computing) becomes a reality, and makes GPU programming method more simple and convenient. CUDA programs can be completely implemented using C language. In this paper, we use CUDA to accelerate the processing of substraction CTA algorithm. The paper includes the following sections:First, we design and implement an automatic substraction CTA algorithm. It can automatically extract bones from CTA images without manual intervention; Second, we improve the substraction CTA algorithm with CUDA in a paralell way so that it can maximize the use of parallel computing capabilities the GPU; Finally, experimental results show that the proposed algorithm works well in substraction of CTA images, and the substraction CTA algorithm based on CUDA received about 2.5 times computing speedup with the same accuracy.