In emission tomography the spatial distribution of radioactive tracer is reconstructed from projection data. The whole image volume is simultaneously calculated from the full set of projection values during the iteration steps of fully 3D image reconstruction. In this article we report on the development and implementation of an improved ML-EM algorithm, which implicitly corrects attenuation and scatter and minimizes reconstruction noise. To demonstrate the clinical usefulness results of the algorithm were compared to reconstructions with filtered backprojection. The improved image reconstruction algorithm has been parallelized for efficient execution on SMP clusters by adopting a hybrid parallelization strategy based on a combination of the MPI and OpenMP application programming interfaces. In order to employ the parallel image reconstruction software over the Internet, a Java-based framework is being developed allowing transparent access to remote high-performance computing platforms from clinical environments.

Top