Hyperfast O(20484) image reconstruction for synchrotron-based X-ray tomographic microscopy

Tomographic image reconstruction, such as the reconstruction of CT projection values, of tomosynthesis data, PET or SPECT events, is computational very demanding. The most time-consuming step is the backprojection which is often limited by the memory bandwidth. Recently, we proposed several algorithms to speed up backprojection on various architectures [1], [2], [3], [4], [5], [6]. We now optimized our algorithms for the reconstruction of synchrotron CT rawdata by optimizing the memory layout of the data to make optimal use of the cache levels available. The code uses floating point arithmetic. Although the data are acquired in parallel geometry and parallel beam filtered backprojection seems straightforward the sheer amount of projection data and the large volumes (up to 2048³) are challenging. Often, expensive cluster solutions are used to reconstruct the data in an acceptable amount of time. Our aim is to show that standard PCs or standard servers allow for a similar or even higher performance, if programmed adequately.