Towards Predicting Future Impedance Distributions from Temporal Sequences of EIT Measurements Using a Recurrent...

The application of data-driven techniques in the field of electrical impedance tomography (EIT) has gained increasing attention in the past years. This is related to the possibilities of machine learning (ML) in addressing nonlinear, inverse, and even ill-posed problems, like the EIT image reconstruction. A currently used EIT application is monitoring intensive care patients. In clinical applications, the imaging reliability is crucial. This contribution demonstrates the potential of applying recurrent sequences of EIT measurements to a combination of a Long Short-Term Memory (LSTM) cells and a Variational Autoencoder (VAE) to predict the next time-instance impedance distribution. This could lead to a more accurate reconstruction or enable the detection of deviations from the established cycle before they actually appear.