Accelerating Simultaneous Algebraic Reconstruction Technique Based on CUDA-Enabled GPU

Iterative methods is a popular choice in image reconstruction fields due to its capability of recovering object information from incomplete acquisition data. However, it is computationally expensive due to frequent uses of forward and backward projections. In this paper, a method of accelerating SART based on CUDA-enabled GPU was proposed. We introduced a ray-driven method for forward projection and a voxel-driven approach for backward projection. The experimental results showed that the reconstructed images quality was not affected, but the reconstruction time was greatly decreased.