In this paper we present four different parallel implementations of the popular LM OSEM medical image reconstruction algorithm. While two of them use libraries such as MPI, OpenMP, or Threading Building Blocks (TBB) directly, the other two implementations use algorithmic skeletons of the Münster Skeleton Library Muesli to hide the parallelism. We compare the implementations w.r.t. runtime, efficiency, and programming style and show the resulting benchmarks which have been conducted on a multi-processor, multi-core cluster computer.