Cardiac-Related Signals Extraction in EIT using Generative Adversarial Networks

Abstract Cardiac-related signals (CRS) extracted from electrical impedance tomography (EIT) measurements are crucial for pulmonary perfusion monitoring and ventilation-perfusion ratio (V/Q) assessment. However, automatic extraction of CRS remains challenging. In this paper, we propose a deep neural network-based approach for CRS extraction that leverages adversarial training. A signal extraction network and a discriminator are trained jointly in a competitive manner, where the signal extraction network uses a 1D convolutional autoencoder architecture to extract CRS from EIT signals. The training objective combines a discriminative loss, which distinguishes between true and extracted signals, and a reconstruction loss for regularization. We conduct experiments on simulated datasets and compare our approach with two traditional baselines. Results show that our proposed method outperforms the baselines in terms of both quantitative and qualitative evaluations.