GPU Accelerated 2.5D Solver for Fast and Accurate Modeling of EIT Forward Problem in Tube Geometries

P. Gebhardt, M. Gevers, T. Musch

Ruhr-University Bochum, Institute of Electronic Circuits, Universitaetsstr. 150, 44780 Bochum, Germany

Abstract

Solving the Electrical-Impedance-Tomography forward problem is an important task in simulating EIT measurement systems and reconstructing conductivity/impedance distributions. GPUs are a suitable hardware base for handling the large linear system the Finite-Element-Method generates to solve the forward problem. In this work, a hybrid CPU-GPU 2.5D FEM software has been developed to reduce the computing time needed to solve that problem, as well as a novel method for generating the FEM formulation in an analytical way. Furthermore, it is demonstrated how the computing performance scales with the mesh properties and how the order of the basis functions changes the overall behavior. Also numerical results are compared to experimental measurement data to proof the accuracy of the method.

Keywords: image processing; Electrical Impedance Tomography (EIT); 2.5D; FEM; GPU; CUDA