Multi-Modal EIT Image Reconstruction Using Deep Similarity Prior

Electrical Impedance Tomography (EIT) is a promising imaging technique with diverse applications from biomedical engineering to industrial process monitoring. Despite its potential, the practical utility of EIT is often limited by low spatial resolution due to the ill-posed nature of its inverse problem. This paper presents a novel multi-modal EIT image reconstruction algorithm that leverages neural network regularization to address this limitation. We first design a self-supervised Siamese network consisting of a dual Structure Encoding Network (abbreviated as SEN) with shared weights. Then, we construct the multi-modal EIT reconstruction with the network regularization, utilizing the learned similarity prior of the trained SEN. We compare our approach with conventional single-modal and multi-modal EIT image reconstruction algorithms to evaluate its effectiveness. Numerical simulations and real-world experiments demonstrate the superior performance of our approach over other algorithms by effectively integrating structural information from the auxiliary image to enhance the quality of the EIT image.