A Robust Convolutional Neural Network for Electrical Impedance Tomography Image Reconstruction

Electrical Impedance Tomography (EIT) is a widely used non-invasive imaging method in a variety of application fields. Most EIT applications require an accurate and high-quality reconstructed inner conductivity distribution image. Several solutions have been developed for reliable EIT image reconstruction, but they still suffer from limitations such as low spatial resolution and inconsistency in real measurement scenarios. In this paper, we propose using physically based, extensive simulation data to train a Convolutional Neural Network (CNN) model with 5-fold cross-validation. We generate a set of 10.000 data samples, which builds a comprehensive basis for its training and ensures reliability and generalization capability. The quantitative evaluation demonstrates good reconstruction performance, achieving a Mean Squared Error (MSE) of 0.0151 and an Image Correlation Coefficient (ICC) of 0.97 in the best-performing fold for the simulated data. The method's effectiveness is particularly validated through experimental measurements, where it significantly outperforms traditional Gauss-Newton reconstruction in terms of image sharpness and boundary definition. The algorithm is robust and easily applicable for EIT imaging.