Sensitivity of two different reconstruction algorithms to body shape and electrode position errors in absolute EIT

Abstract

For image reconstruction in absolute EIT (a-EIT) on the human thorax, often a simple body geometry as a circular or elliptical cylinder is assumed. However, deviations from this assumption lead to distortions or artefacts in the resulting tomograms. Furthermore, inexact electrode positions affect the results. In order to investigate the influence of these factors on image reconstruction, synthetic data sets were calculated using 3D-FE modelling (COMSOL Multiphysics, COMSOL AB) with varying body shape and electrode positions. Additionally, measurements on a healthy volunteer were performed accompanied by exact determination of individual body shape and slight variation of electrode positions. Two different reconstruction algorithms were used: one based on an unstructured triangular parametrisation with FE forward modelling in combination with a Gauss-Newton inversion and another one based on a modified SIRT-algorithm with FD forward modelling using cylindric segments. The results show, that the ignorance of individual body shape and electrode positioning errors lead to distortions or artefacts with both algorithms. If the geometrical peculiarities are known, they easily can be included in the inversion process by the use of the unstructured parametrisation and so their influences are mainly diminished. If this information is not available exactly as it frequently will occur in practice, there are artefacts and distortions, which depend on the kind and the degree of geometry deviation and electrode position errors. The more simplified inversion using modified SIRT yields tomograms, which show at the outer positions stronger artefacts, but the reconstruction of resistivities in the lung regions is less disturbed. © Springer-Verlag 2007.