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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonary Medicine

Background:

• Mechanical ventilation can cause lung injury due to uneven air distribution.
• Electrical Impedance Tomography (EIT) monitors regional lung ventilation at the bedside.
• Current Degree of Lung Collapse (DoLC) assessment using Global Inhomogeneity (GI) is limited by lung segmentation.

Purpose of the Study:

• To develop a deep learning framework for automated DoLC estimation from EIT images.
• To overcome the limitations of lung segmentation in traditional DoLC assessment.
• To provide a more consistent and automated solution for real-time lung function monitoring.

Main Methods:

• A deep learning model was designed to directly estimate DoLC from EIT images.
• The model was trained on synthetic datasets simulating diverse lung conditions.
• Performance was evaluated using numerical simulations and phantom tests.

Main Results:

• The deep learning framework achieved high accuracy in DoLC estimation.
• The model demonstrated an error rate of 5% in both numerical and phantom tests.
• The proposed method eliminates the need for lung segmentation.

Conclusions:

• Deep learning offers a robust and automated method for DoLC assessment using EIT.
• This approach enhances the reliability and efficiency of real-time lung function monitoring.
• The technology has the potential to reduce ventilator-induced lung injury.