We propose a statistical learning approach to enhance contrast in EIT imaging of the brain. The conductivity is modelled as a combination of two unknowns, spatially distributed background and the skull layer. The discrepancy between the forward models with and without the skull layer is treated as a modelling error which is handled by the Bayesian approximation error approach. The approach is evaluated with simulations and phantom data.

This presentation is part of Minisymposium “MS20 - Advances in Reconstruction Methods for Electrical Impedance Tomography (3 parts)”
organized by: Melody Alsaker (Gonzaga University) , Samuli Siltanen (University of Helsinki) .

Authors:

Ville Kolehmainen (University of Eastern Finland)

Marko Vauhkonen (University of Eastern Finland)

Antti Nissinen (University of Eastern Finland)

Jari Kaipio (The University of Auckland)

Aku Seppänen (University of Eastern Finland)

Keywords:

bayesian methods, image reconstruction, inverse problems, machine learning, statistical inverse estimation methods