Error-Constraint Deep Learning Scheme for Electrical Impedance Tomography (EIT)

Abstract

As a new generation of noninvasiveness imaging modality, electrical impedance tomography (EIT) has broad application prospects to perform functional evaluation for human tissues and organs. Deep learning makes great breakthrough in recent years, and many different neural networks (NNs) have been applied to EIT image reconstruction for obtaining high resolution reconstruction images. However, the NN is hard to interpret, and the robustness and generalization ability of the network model cannot be guaranteed. In this study, a deep learning scheme namely error-constraint network (Ec-Net) is designed to remove the error of reconstruction images obtained by traditional reconstruction algorithm. By establishing the residual mapping between image and error, Ec-Net achieves more robust reconstruction performance and faster learning. Feature fusion module, dilated convolution block, and the structure of residual in residual are used to further improve the reconstruction accuracy for the inclusion boundary. The reconstruction results show that the proposed method can accurately reconstruct the irregular and sharp boundary of inclusions while showing good anti-noise ability for measurement noise higher than 30 dB. Even effective reconstruction can be obtained for new inclusions that have size/shape variations, which proves the generalization ability of Ec-Net model.