TSS-ConvNet for electrical impedance tomography image reconstruction.

Abstract

In this paper, we present a novel data-driven approach for solving ill-posed inverse problems, such as Electrical Impedance Tomography (EIT). Our approach introduces a new layer architecture composed of three paths: spatial, spectral, and truncated spectral paths. The spatial path processes information locally, while the spectral and truncated spectral paths provide the network with a global receptive field. Such architecture helps eliminate the ill-posedness and nonlinearity of the inverse problem. The three paths are interconnected, allowing for information exchange on different receptive fields with different learning abilities. The network has a bottleneck architecture which enables it to recover signal information from noisy redundant measurements. We call our proposed model Truncated Spatial-Spectral Convolutional neural Network (TSSConvNet). The model overcomes the receptive field limitation of the most existing models which use only the local information in Euclidean space. We trained the network on a large dataset that covers various configurations with random parameters to ensure generalization over the training samples. Our model achieves superior accuracy with relatively high resolution on both simulation and experimental data.