Bayesian Neural Network to Solve the Inverse Problem in Electrical Impedance Tomography

In this contribution we present a method for solving the inverse problem in electric impedance tomography with Bayesian MLP neural network. The problem of reconstructing the conductivity distribution inside an object from potential measurements from the surface is known to be ill-posed, requiring efficient regularization techniques. We decompose the reconstruction problem to lower dimensional problem by principal component projection, and use very efficient Bayesian neural network to solve the reduced problem. This approach contains double regularization effect, first due to solving the inverse problem in the eigenspace, and second due to using neural networks that learn the distribution of feasible solutions from the training data. We show by simulations, that the proposed approach leads to rather accurate reconstruction results and facilitates estimation of other target values, such as the void fraction (the fraction of gas in liquid), directly without actual image reconstruction. We also demonstrate that the solutions are very robust against noise in inputs.