High performance edge-preserving regularization in 3D SPECT imaging

We describe a parallel software for three-dimensional single photon emission computed tomography (SPECT) imaging. The software has been developed within a multi-disciplinary research activity aimed at improving both the mathematical models and the processing time in nuclear medicine. The numerical algorithm implements an edge-preserving regularization approach based on the total variation seminorm for solving the inverse and ill-posed problem describing the SPECT image reconstruction problem. The parallel software has been tested on a low-cost implementation of a parallel architecture made up of a Linux cluster (Beowulf). The proposed approach is compatible with diagnostic times in terms of availability of the results in a suitable turnaround time and of the overall cost of the system infrastructure.