Chapter eight - Electrical impedance tomography image reconstruction based on autoencoders and extreme learning machines

Abstract

Electrical impedance tomography (EIT) is an image technique based on the application of an alternating electrical current on electrodes placed on the surface of the domain, which is also responsible for measuring the resulting electrical potentials. The main advantages of EIT are portability, low cost, and the nonuse of ionizing radiation. EIT image reconstruction depends on the resolution of direct and inverse problems, which are nonlinear and ill-posed. Several reconstruction methods have been used to solve EIT inverse problems, from Newton-based methods to bio- and social-inspired metaheuristics. In this chapter we propose a new approach: the use of autoencoders, a deep neural network with unsupervised training, to work like an intelligent filter, denoising the electrical potential data. After that, the use of random-weighted neural networks, specifically extreme learning machines (ELMs), to approximate sinograms from electrical potential data and, therefore, apply the classical backprojection algorithm for image reconstruction. We generated a database of 4000 synthetic 128 × 128 images. We trained 16 neural networks corresponding to a 16-electrode EIT system placed on a circular domain. Results were evaluated according to peak-to-noise ratio (PSNR) and structural similarity index (SSIM), as well as visual inspection. Images obtained with the proposed method were compared with images reconstructed without the autoencoders step.