Induced-Current Learning Method for Nonlinear Reconstructions in Electrical Impedance Tomography.

Area of Science:

• Biomedical Engineering
• Computational Imaging
• Electrical Engineering

Background:

• Electrical impedance tomography (EIT) reconstructs internal electrical properties from surface measurements.
• Nonlinear EIT problems pose significant reconstruction challenges.
• Current methods often struggle with speed, accuracy, and noise robustness.

Purpose of the Study:

• To propose and validate a novel Induced-Current Learning Method (ICLM) for solving nonlinear EIT problems.
• To develop a specialized Convolutional Neural Network (CNN) architecture for efficient EIT reconstruction.
• To enhance the robustness and speed of EIT imaging.

Main Methods:

• Developed a Cascaded End-to-End Convolutional Neural Network (CEE-CNN) implementing the ICLM.
• Designed a combined objective function and multiple labels to reduce EIT nonlinearities.
• Utilized induced contrast current (ICC) and updated electrical fields as input, with ICC as output.
• Incorporated skip connections to focus learning on unknown ICC components.

Main Results:

• ICLM, implemented via CEE-CNN, solves typical EIT problems in under 1 second.
• Achieved high-quality image reconstruction.
• Demonstrated significant robustness against measurement noise and modeling errors, including inaccurate boundary data.

Conclusions:

• The proposed ICLM with CEE-CNN offers a fast and accurate solution for nonlinear EIT.
• The novel input/output strategy and network architecture effectively address EIT challenges.
• ICLM shows strong potential for practical EIT applications requiring high fidelity and resilience.