Grid processing techniques for the iterative reconstruction of large clinical positron emission tomography sinogram datasets

We have developed, implemented and validated a set of flexible and efficient iterative image reconstruction (IIR) and processing methods for clinical positron emission tomography (PET) static or dynamic studies. IIR techniques are shown to produce images of better contrast and signal-to-noise ratio than the conventional filtered back-projection (FBP). Their high computational cost has been a significant problem for the application of IIR in the clinical routine. We present a solution using grid distributed processing on a cluster of workstations. We show that IIR, even for multi-frame PET dynamic studies, does not require dedicated high-performance computing systems. Efficient reconstruction can be achieved with better management of the existing infrastructure in a PET Center. The grid-based, quasi-parallel implementation scheme allows the simultaneous reconstruction of several such studies on the available workstations operating on the network. This technology opens the way for the efficient clinical implementation and practical use of state-of-the-art IIR schemes.