Title:A reconstruction algorithm of electrical impedance tomography based on one-dimensional convolutional neural network

Abstract:Electrical impedance tomography (EIT) is a novel computational imaging technology. In order to improve the quality and spatial resolution of the reconstructed images, the G-CNN and HG-CNN algorithms are proposed based on a one-dimensional convolutional neural network (1D-CNN) in this paper. The input of the 1D-CNN is the reconstructed conductivity distribution obtained by the GVSPM algorithm or the H-GVSPM algorithm. The reconstructed images with higher resolution are obtained through the calculation of 1D-CNN. Finally, the Hadamard product is applied to calculate the output of the 1D-CNN. In the simulation results of the lung cross-section models, the correlation coefficients of the G-CNN algorithm and HG-CNN algorithm maximumly are 2.52 times and 2.20 times greater than the GVSPM algorithm and H-GVSPM algorithm, respectively. In the results of the simulation and experiment, the reconstructed images of the G-CNN and HG-CNN algorithms are distortion-free. In addition, the artifacts of the reconstructed images are diminished after calculations of the Hadamard product. This research provides a reference method for improving the quality of the reconstructed images so that EIT is better applied in medical detection.