Algorithms on electrical impedance tomography, focusing on deep learning architectures and their implementations: A Review

Electrical impedance tomography (EIT) has garnered increasing attention in recent years, across different domains, as a promising alternative to traditional imaging techniques like X-rays due to its non-ionizing nature. Despite its potential, improvements in data collection methods and postprocessing techniques are essential to enhance image quality and generate useful metrics. This paper reviews several cutting-edge approaches that leverage deep learning to address these challenges, aiming to improve the resolution and accuracy of EIT images. Direct and indirect deep learning methods are systematically compared to establish a framework for method selection, guiding practitioners in choosing approaches based on specific application requirements. A key contribution of this work is the exploration of Bayesian learning as an effective method for standardizing and optimizing EIT systems, emphasizing its ability to incorporate model uncertainties and tolerances that are often neglected in conventional deep learning models. Additionally, the paper highlights emerging trends in EIT, including the use of generative adversarial networks (GANs), variational autoencoders (VAEs), and transformers, highlighting their strengths and weaknesses under different conditions. The trade-offs between computational speed and image quality are emphasized, underscoring the need to balance real-time processing demands with high-fidelity reconstructions in algorithm design. Furthermore, it presents solutions that are protected by patents, underscoring the industrial interest and commercial viability of EIT technology in various applications.