Iterative Helical CT Reconstruction in the Cloud for Ten Dollars in Five Minutes

Iterative statistical X-ray CT reconstruction algorithms can improve image quality for low dose scans. Unfortunately, their clinical utility has been hampered by their enormous computational requirements; typical low-dose reconstructions require about an hour on commercial systems. Most existing parallel implementations use a shared memory programming model, limiting available parallelism. We investigate using a large compute cluster for a penalized weighted least-squares algorithm using ordered subsets (PWLS-OS), scaled to hundreds of cores to accelerate a single helical CT reconstruction problem. Using Amazon’s Elastic Cloud Compute (EC2) service, our experimental results show that a typical helical chest scan can be reconstructed in under five minutes at a cost under $10.