Forward problem solution acceleration with graphical processing units (GPU) for electrical tomography imaging

Bartosz Matusiak, Andrzej Romanowski, Dominik Sankowski, Krzysztof Grudzieł

Institute of Applied Computer Science, Lodz University of Technology, Stefanowskiego 18/22, 90-924 Lodz, Poland

Abstract

The computational effort is associated in electrical tomography mainly with forward problem solution. As the recent developments in general purpose graphical processing units (GPGPU) shown possibilities of using parallel computing, the authors of this paper propose a novel algorithm dedicated to forward solving that accelerates the computations vs. classical implementations of imaging algorithms for central processors (CPU). The main contribution here is a method of matrix decomposition coupled with calculation groupings for efficient use of global and local memory usage with minimization of transfers between GPGPU and CPU memory. Tests were conducted for 8 and 16 electrode measurement image reconstructions with use of CUDA based implementation of algorithms and comparison tests for 32 and 64 bit platforms. Proposed algorithm designed for GPGPU enabled several times shorter computation time compared to the algorithm performed by the CPU, which can be directly translated into spatial and/or temporal resolution enhancement of reconstructed images.

Keywords: ECT forward problem, electrical tomography;GPGPU; GPU; parallel computing; numerical metods for tomography