GPU-based 3D cone-beam CT image reconstruction: application to micro CT

In tomographic image reconstruction, one often needs to calculate the forward- and backward-projections. In general, such a calculation constitutes the most computationally demanding component of a image-reconstruction process. It is has been recognized that the graphic processing unit (GPU) can be exploited for improving computational efficiency. In this work, we investigate and develop the GPU-based computation of both forward- and backward-projections in 3D cone-beam CT. The GPU-based backward-projection was incorporated into the FDK algorithm for reconstructing 3D images of mice from data acquired with a micro cone-beam CT. Our results indicate that the GPU implementation can reduce the computation time up to a factor of 40 for circular cone-beam scan.