Iterative CT Reconstruction on the GPU

The computing power of modern GPUs makes them very suitable for Computed Tomography (CT) image reconstruction. Apart from accelerating the reconstruction, their extra computing performance compared to conventional CPUs can be used to increase image quality in several ways. In this paper we present our upgraded GPU based iterative reconstruction algorithm, including ML-TR (Maximum Likelihood algorithm for Transmission tomography), MAP estimation usingMRFGibbs priors with Huber function. We experimentally evaluated the impact of wide range of reconstruction parameters on the image noise level and the resolution, and we compared the performance of the reconstruction components in phantom studies with special focus on image noise, resolution and ring artefacts. Achieving more performance we extended our solution with multi-GPU support.