Distributed image reconstruction of micro-CT data: Computation and communication issues

Three-dimensional iterative reconstruction of high-resolution cone-beam X-ray computed tomography (CT) data poses significant implementation challenges due to extensive computation and storage requirements. We first employ techniques to accelerate the reconstruction and reduce the problem size. We then distribute the remaining workload of the iterative image reconstruction to a cluster of PCs to make the computation more feasible. However there are performance trade-offs to consider between the number of nodes and parameters of the reconstruction with respect to the resulting communication costs. In this paper we analyze these trade-offs for our particular computing environment.