20 May 2016 Parallel computing for simultaneous iterative tomographic imaging by graphics processing units
Author Affiliations +
In this paper, we address the problem of accelerating inversion algorithms for nonlinear acoustic tomographic imaging by parallel computing on graphics processing units (GPUs). Nonlinear inversion algorithms for tomographic imaging often rely on iterative algorithms for solving an inverse problem, thus computationally intensive. We study the simultaneous iterative reconstruction technique (SIRT) for the multiple-input-multiple-output (MIMO) tomography algorithm which enables parallel computations of the grid points as well as the parallel execution of multiple source excitation. Using graphics processing units (GPUs) and the Compute Unified Device Architecture (CUDA) programming model an overall improvement of 26.33x was achieved when combining both approaches compared with sequential algorithms. Furthermore we propose an adaptive iterative relaxation factor and the use of non-uniform weights to improve the overall convergence of the algorithm. Using these techniques, fast computations can be performed in parallel without the loss of image quality during the reconstruction process.
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Pedro D. Bello-Maldonado, Pedro D. Bello-Maldonado, Ricardo López, Ricardo López, Colleen Rogers, Colleen Rogers, Yuanwei Jin, Yuanwei Jin, Enyue Lu, Enyue Lu, "Parallel computing for simultaneous iterative tomographic imaging by graphics processing units", Proc. SPIE 9870, Computational Imaging, 987009 (20 May 2016); doi: 10.1117/12.2223466; https://doi.org/10.1117/12.2223466

Back to Top