Based on information about electrical potential in the perimeter of the cross section of an object (collected by sampling in simulated environment), an Electrical Impedance Tomography (TIE) system should be able to estimate an impedance distribution within this section, so non-invasive. Given an impedance distribution, the electric potential at the perimeter can be calculated by numerical methods using the finite element method (MEF), defining a function f such that phi = f (sigma) that has a sigma impedance map as input and returns the electrical potential of its perimeter. It is also known that there is a theoretical result ensuring that given a distribution of electric potential in the perimeter of an object, there is a single internal distribution of impedance (assuming isotropic field). The possibility to estimate a function f, inverse of f: sigma = g (phi), which is able to estimate the map of impedances from an electric potential distribution in the perimeter, is opened. It proposes an approach based on machine learning techniques (Autoencoders and Supervised Learning). The proposed methodology can be divided into 3 distinct phases: Generation of impedance maps; Calculation of electrical potentials in perimeters; and Construction of the inverse function g.
News published in Agência FAPESP Newsletter about the scholarship: