Abstract: Data-driven reconstruction techniques using deep neural network (DNN) architectures are applied more frequently in the field of electrical impedance tomography (EIT). The solution of the underlying ill-posed inverse problem may benefit from the possibilities of machine learning (ML). This contribution demonstrates, how knowledge on recurring sequences of EIT measurements (e.g. breathing cycles) may be used to improve the reconstruction. A combination of a Long Short-Term Memory (LSTM) and an Variational Autoencoder (VAE) is used.
