Performance of Parallel 3D Iterative Reconstruction Algorithms

3D Volume reconstruction from projections is a very interesting field of researching as it can be used within a wide range of civil and scientific applications. In this work we focus on high resolution structure determination of biological macromolecules by electron microscopy which is used on structural biology to help to understand their biological functions. Thousands of electronic microscope projections are processed to obtain high resolution reconstructions by applying certain regularized iterative algorithms. Those algorithms are very expensive in terms of computing time which means hours, days or even weeks on a single computer. In fact, for very complex specimens where the volume to reconstruct requires very high resolutions much more projections are needed which translates into an impossibility to run the reconstrucion on a single computer. Therefore, parallelization of the reconstruction algorithms is a natural way to solve these needs of computing power. In this article, we describe, analyze and compare five different parallel reconstruction algorithms. We will see which algorithms are faster, which scale better their weaknesses and their strong points. Key-Words: 3D Electron Microscopy, Tomography, Parallel Algorithms, 3D Reconstruction.