GPU acceleration of absolute EIT image reconstruction using simulated annealing

Electrical impedance tomography (EIT) is a portable low-cost medical image technique with fast time response characteristics. EIT can be approached as an optimization problem whose objective is to minimize the difference between the simulated and measured distributions. A preconditioned conjugated gradient is employed to solve a linear system with a symmetric sparse positive definite matrix. In order to increase its efficiency, a matrix format, the colored padded jagged diagonals storage (pJDS) format, is proposed. Parallelization is applied to several steps of the algorithm and at each step performance is observed to be superior to fast serial implementation. However, API overhead degraded the performance of the forward problem. Kernel consolidation combined with the pJDS format obtained a significant performance improvement. The inverse problem is solved as an optimization problem using the simulated annealing with adaptive neighborhood. While several instances of the conjugated gradients run on the GPU, the remaining processes are executed in parallel in the CPU. The GPU saturates at a speedup of 5 times as compared to CPU processing. (C) 2017 Elsevier Ltd. All rights reserved.