High-resolution conductivity reconstruction by electrical impedance tomography using structure-aware hybrid-fusion...

Area of Science:

• Medical Imaging
• Computational Electromagnetics
• Artificial Intelligence

Background:

• Electrical impedance tomography (EIT) is a non-invasive imaging technique for lung disease diagnosis.
• EIT faces challenges in reconstruction accuracy and robustness due to the ill-posed inverse problem.

Purpose of the Study:

• To develop a deep learning method for high-resolution conductivity reconstruction in EIT.
• To enhance EIT's robustness against measurement noise and modeling errors.

Main Methods:

• Proposed a novel structure-aware hybrid-fusion learning (SA-HFL) method.
• SA-HFL utilizes segmentation and conductivity reconstruction branches with a feature fusion module.
• Trained the network using numerical EIT datasets (regular and lung-shaped).

Main Results:

• SA-HFL outperformed five deep learning networks and the IGGS method in simulations and experiments.
• Achieved improved evaluation metrics: lower relative error (RE), higher mean structural similarity index (MSSIM) and peak signal-to-noise ratio (PSNR).
• Demonstrated efficient computational performance (FLOPs) and appropriate parameter execution.

Conclusions:

• Fusing multi-branch feature information significantly improves conductivity reconstruction accuracy in EIT.
• Multi-modal information fusion presents a promising future direction for EIT conductivity distribution reconstruction.