Machine Learning-Based Image Reconstruction in Wearable CC-EIT of the Thorax: Robustness to Electrode Displacement

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Electrophysiology

Background:

• Capacitively coupled electrical impedance tomography (CC-EIT) is a non-invasive imaging technique.
• Accurate electrode placement is crucial for reliable CC-EIT image reconstruction.
• Variations in electrode positioning can degrade image quality and diagnostic utility.

Purpose of the Study:

• To investigate the impact of variable electrode positions on thorax CC-EIT image reconstruction.
• To develop and evaluate a robust adversarial neural network for CC-EIT image reconstruction.
• To enhance the accuracy and reliability of CC-EIT imaging in the presence of electrode displacement.

Main Methods:

• Utilized a numerical thorax phantom with heart, lungs, aorta, and spine.
• Generated synthetic datasets with a wearable 32-electrode band and random electrode position variations.
• Trained adversarial neural networks on datasets with and without electrode displacement.
• Assessed image reconstruction quality using pixel-to-pixel metrics (RMSE, SSIM, CC, PSNR).

Main Results:

• Electrode displacement significantly degraded reconstruction quality without specific training data.
• Training neural networks with electrode displacement data substantially improved image reconstruction quality.
• Adversarial neural network demonstrated increased robustness to electrode placement inaccuracies.

Conclusions:

• Electrode placement variability is a critical factor affecting CC-EIT image reconstruction accuracy.
• Training CC-EIT neural networks with diverse electrode positions enhances robustness.
• This approach improves the clinical applicability of CC-EIT for thorax imaging.