Electrical impedance tomography methods and algorithms processed with a GPU

The present paper discusses the process of parallelizing an algorithm for the reconstruction of an image acquired via electrical impedance tomography (EIT). The introductory section comprises a general description of EIT, the inverse problem, and regularization; in this context, the potential of the method for biomedicine, defectoscopy, and geophysical mapping is outlined. The following chapter then analyzes the objective function of the EIT inverse problem together with Tikhonov regularization. Besides setting up the objective function with a regularizing member, the authors also specify the differentiation equation for the iterative solution of the inverse problem via the Gauss-Newton method. Further, the time consumption of computing the Jacobian via a CPU compared to using a newly assembled program that exploits GPU-based parallel processing is investigated in detail. The program, utilizing the NVIDIA CUDA platform, employs parallelized computation of the individual columns of the Jacobi matrix, and this approach proved to be twenty times faster than the CPU-based sequential processing.