2D and 3D Alignment for Electron Microscopy via Graphics Processing Units

In this paper we address the GPU-based acceleration of the alignment of 2D and 3D images in the context of Electron Tomography. The alignment of images (2D) is part of the early stages in the 3D reconstruction of tomograms, since the images obtained by the electron microscope must be transformed to correct the misalignments inherent to Electron Microscopy. The alignment of volumes (3D) is applied in the so-called subtomogram averaging, where several subtomograms are aligned and averaged to improve the quality of the 3D reconstruction. The results show that 2D alignment can be executed x400 faster in comparison to a standard CPU. On the other hand, 3D alignment can be x200 faster.