An Improved Super-Resolution Image Reconstruction for Electrical Impedance Tomography

Electrical impedance tomography (EIT) is an innovative imaging technique used to recreate the distribution of conductivity in a region by utilizing measured current and voltage readings. In this paper, we introduce a novel method to improve the resolution of EIT images, based on the Efficient Sub-Pixel Convolutional Neural Network (ESPCN). The research involves fine-tuning the parameter weights of the loss function and training the model with datasets of COVID-19 Lung CT Scans Images. The method entails extracting features directly from Low Resolution (LR) images and using convolution operations in sub-pixel convolution layers to accurately capture and reconstruct the morphology and structure of lung tissue. This optimization is designed to solve the two-dimensional forward problem in lung EIT. We propose an Improved Efficient Sub-Pixel Convolutional Neural Network (IESPCN) model to tackle image inverse problems with irregular boundaries. Training the model with unstructured datasets enhances robustness and improves resolution in imaging effects, resulting in high-quality reconstructed images. Experimental results demonstrate that the lung EIT super-resolution reconstruction imaging model based on IESPCN effectively delineates the boundaries of lung organs. This method enhances both the structural fidelity and imaging quality of the output images, while also ensuring stable imaging effects across varying lung boundaries. Furthermore, it improves image resolution and significantly reduces artifacts in the reconstructed images.