High-Resolution Lung Imaging with an Improved Physics-Informed Neural Network via Electrical Impedance Tomography

High-resolution lung imaging is critical for the early diagnosis and treatment of diseases such as pneumonia and acute respiratory distress syndrome (ARDS), particularly in scenarios that require real-time and continuous monitoring. Electrical impedance tomography (EIT) has emerged as a promising lung imaging technique due to its noninvasiveness, portability, and capability for real-time imaging. However, existing EIT reconstruction methods still face significant challenges in achieving high imaging resolution and maintaining physical consistency. To address these limitations, this paper proposes a high-resolution lung imaging method based on an enhanced physics-informed neural network (PINN) integrated with EIT. The proposed framework utilizes ResNet34 as its backbone and incorporates the EIT partial differential equation (PDE) as a physical constraint, forming a dual-network architecture consisting of a voltage distribution prediction network (V-Net) and a conductivity distribution prediction network (σ-Net). Furthermore, a realistic three-dimensional thorax-lung model is constructed from clinical CT data to generate diverse simulated datasets, thereby supporting both data-abundant and data-limited scenarios. Extensive simulation and experimental results demonstrate that the proposed method achieves superior imaging performance, with structural similarity index (SSIM) values consistently exceeding 0.86 and relative error (RE) below 0.27, even in challenging pathological conditions. The results highlight the model’s robustness and accuracy in accurately reconstructing both normal and injured lung tissues, indicating significant potential for clinical pulmonary monitoring and early disease detection.