Electron tomography (ET) has been developed rapidly and applied widely in the recent years, as it provides the prospects of reconstructing non-uniform cells or macromolecules in nano scale. To overcome the deficits of current ET reconstruction software packages, especially the large computations of ET iterative reconstruction algorithms, we have developed a Graphic Processor Unit (GPU) powered software, ATOM. Its main functions include: 2D alignment, reconstruction parameters estimation, 3D reconstruction, and 2D image visualization etc. In 2D alignment part, iterative fiducialess shift and rotation alignment methods are applied. In 3D reconstruction part, it supports Back Projection (BP), several iterative reconstruction methods, and their parallel versions on GPU platform. The speedups of parallel iterative reconstruction methods are appealing, e.g. parallel Simultaneous Iterative Reconstruction Techniques (SIRT) can achieve 47 times speedup in contrast to serial SIRT. ATOM is an open source software and can run on any platform that supports Qt and CUDA libraries. It has a friendly graphic user interface along with detailed manuals.
Fig. 1 ATOM 1.0 contains both the Chinese version and the English version (Here we only introduce the English version). The interface of ATOM consists of four parts: “Basic steps” (This figure is carrying on the 2D alignment step), “Parameter input and output”, “In-process output”, and “Display window”. Pressing the “View” button will activate the independent “Display window”, see Fig.4. “In-process output” would output the in-process information in which people can check the running step and potential errors.
(a)
(b)
Fig. 2a (a).Fast Initial Model GUI. According to the input image series, we display one of the images in the display window. The two blue dotted lines separate the image to 3 parts and the red solid lines appoint where the model slices will be reconstructed. Users should choose the locations of the model slices manually and then press Start to generate the reconstruction slices.
Fig. 2b (b). Geometry Determination GUI. This figure shows one of the model slices. Users need to move the green solid lines to constrain the specimen border as much as possible, which could yield 4 border coordinates for this slice. Combining 12 coordinates from three slices, we could compute the geometry parameters of the sample.
Fig. 3 Reconstruction GUI. ATOM 1.0 provides three reconstruction algorithms-- BP, ART ,SIRT and their parallel versions. Besides, ATOM 1.0 provides two ways to calculate projection matrix which are Float type and 0-1 type. ATOM 1.0 also provides several initial models including BP, Zero, Random, and Already Exist. Already Exist means that users can do reconstruction based on the previous reconstruction result by other programs.
Fig. 4 Reconstruction GUI. In 2D alignment and 3D reconstruction step, users can press the View button to see the input image series, the output aligned series or the reconstructed 3D result. In this GUI, users can adjust the contrast ratio; choose some image to view; view 2D series sequentially; and choose to view in x-y axis, x-zaxis or y-z axis.
(a)
(b)
Fig. 5 The comparison of the tomograms aligned and reconstructed by IMOD and ATOM 1.0. (a) the tilt series of images are aligned and reconstructed by IMOD; (b) the same tilt series of images are aligned and reconstructed by ATOM 1.0 The parallel SIRT reconstruction algorithm in ATOM 1.0 is performed on Tesla c1060 GPU.
