Electrical Impedance Tomography Image Reconstruction Based on Graph Laplacian and Sparse Learning

Electrical Impedance Tomography (EIT) is a non-invasive diagnostic technique capable of inferring the internal conductivity distribution of a target from the measured boundary voltage signal. However, estimating high-resolution conductivity images from undersampled voltage signals is a nontrivial task due to the highly ill-posed nature of the problem. Although supervised learning methods can greatly improve the accuracy of reconstructed conductivity images, such models rely on a large number of training samples, which are difficult to obtain in practice. To address this problem, sparse regression frameworks have been introduced in EIT within the unsupervised learning setting, based on the assumption of a smaller number of inclusions. In this work, we proposed an iterative learning approach, which simultaneously incorporates spatially sparse and graph priors. Experiments conducted on both synthetic and real-world data demonstrate the superiority of the proposed method.