On the Implicit Representation of the Electrical Impedance Tomography Inverse Problem

Area of Science:

• Medical Imaging
• Computational Electromagnetics
• Machine Learning

Background:

• Electrical Impedance Tomography (EIT) is a noninvasive imaging technique.
• EIT faces challenges due to its ill-posed inverse problem, often requiring regularization.
• Neural networks present a novel approach to enhance EIT reconstruction accuracy.

Purpose of the Study:

• To investigate the efficacy of Deep Implicit Layers for EIT reconstruction.
• To combine fixed-point iteration and Newton's method within an implicit neural network framework.
• To improve the robustness of EIT models by incorporating random contact conductivity.

Main Methods:

• Utilized Deep Implicit Layers with fixed-point iteration and Newton's method.
• Solved the forward problem using the finite element method (FEM).
• Employed numerical phantoms for training and incorporated random contact conductivity for realism.

Main Results:

• The implicit neural network demonstrated effectiveness in distinguishing closely positioned objects.
• Model robustness was enhanced by modeling contact conductivity as a random variable.
• Contrast limitations were observed at the boundaries of the numerical phantoms.

Conclusions:

• Deep Implicit Layers show significant potential for advancing EIT reconstruction.
• Further research should explore advanced optimization techniques for improved performance.
• Expansion to 3D thorax reconstruction is a promising future direction for EIT.