基于图形处理器的X射线锥束成像模拟算法

摘要

To accelerate the simulation of X-ray cone-beam imaging, a GPU（ graphics processing unit） based algorithm is proposed in this paper. The algorithm generates X-ray image by accumulating the contribution of voxels along each X-ray. Intersection lengths of these voxels with X-ray are calculated by classifying the intersection types, which reduces the time-consuming dynamic branches compared to the famous incremental Siddon algorithm. To improve image quality, sampled values along X-ray are computed by GPU hardware supported linear interpolation instead of nearest interpolation used by the incremental Siddon algorithm. The experiment of the projection calculation of Shepp-logan phantom shows that the simulation speed is improved by 44% averagely as compared to the GPU-based incremental Siddon algorithm and a better image quality is achieved. Finally, the proposed algorithm is validated by the experiment with real measured data.%针对X射线锥束成像模拟计算量大、速度慢的问题，提出了一种基于图形处理器（GPU）的快速成像模拟算法。该算法沿着每条射线累加所经过体素对投影值的贡献量，实现了对X射线成像的模拟。在计算射线与体素的交线长时，采用分类处理交线的方法，减少了增量Sid—don算法的动态分支计算。为了提高投影图像质量，该算法还用GPU硬件线性插值采样取代Sid—don算法的邻近插值采样。对三维Shepp—logan模型的测试结果表明，该算法的速度比基于GPU的增量Siddon算法平均提高了44％，而且图像质量明显提高。最后，用实测数据进一步验证了算法的有效性。