DA-Net: A Dense Attention Reconstruction Network for Lung Electrical Impedance Tomography (EIT)

To achieve accurate reconstruction of lung contour and lesion structure based on electrical impedance tomography (EIT), this study proposed a novel method named dense attention network (DA-Net), a direct estimation model inspired by DenseNet. The dense connectivity is employed in DA-Net to enhance the information flow of the model, while the attention mechanism is also integrated to identify and focus on critical features, thereby boosting reconstruction accuracy. Simulation results demonstrate that the performance of DA-Net model exceeds 1D-CNN and ResNet50 in terms of accurate reconstruction of the lung contours, as well as effectively identifying the areas of lesions. Additionally, DA-Net exhibits good robustness. An experimental setup consisting of a circular flume is employed to evaluate the practicality of DA-Net. The imaging results indicate that the correlation coefficients (CCs) of the reconstructed images exceeded 0.83, and the relative errors (REs) remained below 0.27, demonstrating that experimental imaging accuracy can be effectively improved by DA-Net.