Robust Electrical Impedance Tomography for Respiratory Monitoring

Thoracic electrical impedance tomography (EIT) is a crucial bedside monitoring tool, particularly valuable in pulmonary and critical care medicine. However, its routine clinical application is restricted in the presence of poorly contacting electrodes, patient movements, medical interventions, and nursing procedures. This is because these practical factors induce impulsive noise in EIT boundary voltages, severely degrading EIT imaging performance. In this article, to handle this issue, we devise a robust EIT (REIT) imaging method, which is able to promote the clinical application of thoracic EIT. We first reformulate the EIT boundary voltage change model, where the additive noise is separated into dense Gaussian and sparse outlier components. We then exploit $\ell _{2}$ -norm and $\ell _{0}$ -norm to formulate the optimization problem, in which the former and the latter are employed to resist the Gaussian noise and the impulsive noise, respectively. Subsequently, we adopt proximal alternating minimization (AM) and projected gradient descent (PGD) to tackle the resultant optimization task. Despite that the suggested method introduces an auxiliary parameter, we propose an adaptive strategy for its determination. Furthermore, we prove that the objective value sequence is convergent and the variable sequence converges to a critical point with at least a sublinear rate. Numerical simulation and phantom experiment exhibit that the devised approach reconstructs higher quality EIT images than the state-of-the-art (SOTA) algorithms when EIT boundary voltages comprise strong interference. Finally, we test the suggested algorithm on two patients data measured in real clinical environments to show its practical robustness.