Reconstruction of electrical impedance tomography images using chaotic ring-topology particle swarm optimization and non-blind search

Non-invasive imaging and e-health have been increasing in the last decades, as a result of the efforts to generate diagnostic technology based on non-ionizing radiation. Electrical Impedance Tomography (EIT) is a low-cost, non-invasive, portable, and safe of handling imaging technique based on measuring the electric potentials generated by the application of currents in pairs of surface electrodes. Nevertheless, EIT image reconstruction is still an open problem, due to its nature as an ill-posed problem governed by the Equation of Poison. Several numerical methods are used in order to solve this equation without generating anatomically inconsistent results. Particle swarm algorithms can be used as alternatives to Gauss-Newton and Backprojection well-known approaches, which frequently generate low-resolution blurred images. Furthermore, Gauss-Newton convergence to anatomically consistent images is not guaranteed, needing to set constraints like the number of anatomical structures present on the image domain. Herein this work we present EIT reconstruction methods based on the optimization of the relative error of reconstruction using chaotic particle swarm algorithms with non-blind initial search. We studied two forms of initialization: totally random and including an imperfect but anatomically consistent noisy solution based on Gauss-Newton reconstruction method, according to Saha and Bandyopadhyay's criterion for non-blind initial search in optimization algorithms, in order to guide the iterative process to avoid anatomically inconsistent solutions and avoid convergence to local minima. Results were quantitatively evaluated with ground-truth images using the relative mean squared error, showing that our results reached low error magnitudes. Qualitative evaluation also indicated that our results were morphologically consistent, mainly for classical PSO and ring-topology PSO with non-blind initial search.