Deep learning-based estimation of lung collapse in electrical impedance tomography: A simulation and phantom study

Abstract Inhomogeneity of air volume distribution in the lungs of mechanically ventilated patients can lead to lung collapse and overdistention, increasing the risk of ventilator-induced lung injury. While Electrical Impedance Tomography (EIT) enables bedside monitoring of regional ventilation, traditional Degree of Lung Collapse (DoLC) assessment using the Global Inhomogeneity (GI) index is limited by lung segmentation. We propose a deep learning framework that directly estimates the DoLC from EIT images without lung segmentation. The model, trained on synthetic datasets simulating various lung conditions, achieved high accuracy with 05% error in both numerical and phantom tests. This approach oers a more consistent and automated solution for real-time lung function assessment.