Fast distributed EM computation for PET image reconstruction using ordered subsets

An embedded system for fast two-dimensional image reconstruction of Positron Emission Tomography (PET) data on high performance loosely coupled processor networks is described. The distributed algorithm is based on Maximum Likelihood Expectation Maximisation (ML-EM) and includes message-passing performance optimisations derived from the Ordered Subset (OS-EM) algorithm. Results obtained from simulations on a workstation network confirms the algorithm performance and now allows the development of low cost and scaleable processor networks for the rapid image reconstruction of clinical PET data. An outline processor network architecture for PET image reconstruction is described.