Machine learning approach to EIT image reconstruction of the human forearm section for different hand signs

Electrical impedance tomography (EIT) is an imaging technique used to reconstruct the conductivity of a target object from boundary voltages. In this study, we investigate suitable image reconstruction algorithms for EIT to enable the reconstruction of the conductivity distribution in the forearm section inferring muscle contractions at different hand signs. As EIT image reconstruction is an ill-posed inverse problem, the Gauss-Newton algorithm needs many iterations for the determination of suitable values of the regularization parameter and corresponding calculations of the Jacobian matrix. To reduce computational effort, we propose to use machine learning algorithms to directly reconstruct the EIT image. We explore the Radial Basis Neural Network (RBNN) and a one-dimensional Convolutional Neural Network (1D-CNN), which has been trained based on the measured EIT data for eight subjects, ten hand signs with ten trials. Both methods reach a low deviation at 0.0017 for RBNN and 0.0109 for CNN