Tracking conductivity variations in the absence of accurate stat evolution models in electrical impedance tomography

We present results on both linear and non-linear approaches in tracking conductivity variations in electrical impedance tomography. Throughout this study, we use both synthetic and measured data. The true system dynamics is considered as unknown and modelled as a random walk. In the linear reconstructions, the time evolution model is augmented with a Gaussian smoothness prior and results are shown using two different models for the covariance matrix of the process noise. Furthermore, we compare the reconstructions of the one step Gauss-Newton method to the Kalman filter on measured data from an adult human subject. In the non-linear study, we compare the performance of the extended Kalman filter against the particle filter on a simple test case. It is observed that the particle filter shows superior performance in tracking nonlinear/non-Gaussian conductivity variations.