一种实用的基于GPU的Katsevich算法的实现

Katseivch算法是精确地解决锥束螺旋CT“长物体”问题的滤波反投影（FBP）形式的CT重建算法，不过它依然是非常耗时的，实际应用中必须考虑如何加速的问题。本文给出了一个使用通用显卡（GPU）对该算法进行并行加速的方法。同已有的方法相比较，几乎所有的计算都在GPU上完成，并且重建时所使用的积分范围是由PI线决定的，并且给出了算法所需要的精确地显存大小，10次数也被最大可能地降低。本文使用了标准数值模型对方法的速度、准确性和稳定性进行了验证。

A practical GPU-based implementation of the Katsevich algorithm

The Katsevich algorithm is an exact reconstruction method with the type of filtered back projection （FBP）. However, it is very time-consuming and parallel implementation has to be considered for practical use. Here we use graphics processing unit （GPU） to accelerate the algorithm. Compared with existing methods, we carry out all the calculations in GPU and uses PI line to determine the integration range. We also proposed an exact expression for the memory to be allocated in GPU. Furthermore, （IO） times are reduced to as low as possible. Simulations are used to show our method＇ s speed, precision and stability.