Distributed Block-Separable Ordered Subsets for Helical X-ray CT Image Reconstruction

Statistical reconstruction for low-dose CT can provide desirable image quality, but the computational burden still remains a challenge, particularly for large 3D helical scans. Parallel computing can help reduce computation times, but simple parallelization methods for CT reconstruction can be hampered by relatively large data communication times between nodes that do not share memory. This paper describes a blockseparable surrogate approach to developing algorithms that facilitate parallelization. These methods reduce communication between nodes and allow multiple independent updates on each node, while attempting to maintain convergence rates of recent accelerated algorithms. As a preliminary study, we investigated one version of the proposed algorithm in Matlab using a simulated 3D helical CT scan.