The Graphics Processor In The Cone-beam Ct Imaging Applied Research

Cone-beam CT has been a hot research topic in the CT field. Image reconstruction algorithms for cone-beam CT are the core of research on the software system of cone-beam CT. Image reconstruction algorithm is using projection data to obtain the image of the object. Forward-projection algorithm is used to simulate the process in which CT scanner acquires the projection data. Exact projection data collection is the key of iterative image reconstruction for cone-beam CT. Additionally, there are unavoidable noises in the process of collecting projection data, so it is deserved to research how to smooth and magnify the image effectively in order to observe the reconstructed image clearly. The amount of image data and projection data of cone-beam CT is larger than that of two-dimensional CT. The traditional algorithm for CT image reconstruction based on CPU is much slower than the algorithm based on GPU, with especial reference to iterative reconstruction algorithms. So it is another deserved research how to use the programmable pipeline for graphic processing unit (GPU) to accelerate the image reconstruction algorithms for cone-beam CT. In the paper, the problems mentioned above are solved as follows:(1) The forward projection algorithm is implemented based on the programmable pipeline for GPU. Most proposed algorithms simulate the projection data based on CPU. In this paper, the ray-casting method is applied to the forward projection algorithm for cone-beam CT based on GPU. And the experiments verify the algorithm in the Shepp-logan phantom. The results of experiment indicate that the algorithm achieves improved speed and the exact image.(2) Catmull_Rom interpolation is implemented based on the programmable pipeline for graphics processing unit (GPU). Then the algorithm is applied to magnify the CT images. Comparing with the nearest point interpolation and linear interpolation, Catmull_Rom spline function has the -continuous character. So the algorithm can avoid the mosaic appearance effectively. The result of experiment shows that the quality of image is excellent.(3) The algorithm for traversing voxels along a three-dimension line is implemented based on CPU. To verify the correctness of the algorithm, the scene that line traverses voxels is drawn based on GPU in the paper. The algorithm can be used to implement the forward projection of voxel-based object for the cone-beam CT, and can be also used to computer graphics in the collision detection.