Abstract
The importance of the solution of the boundary integral equation for the exponentially growing solutions to the Schrödinger equation arising from the 2-D inverse conductivity problems is demonstrated by a study of reconstructions of simple piecewise constant conductivities on a disk from two methods of approximating the scattering transform in the D-bar method and from Calderón's linearization method.
Introduction
Electrical Impedance Tomography (EIT) is a non-invasive imaging technique that reconstructs the internal conductivity of a domain by applying voltages to the boundary and measuring the resulting currents. This process, mathematically described as Calderón’s inverse conductivity problem, is particularly challenging when dealing with "piecewise constant" conductivities—distributions characterized by sharp discontinuities at boundaries, such as those between different human organs.
The D-bar method provides a rigorous mathematical framework for solving this inverse problem using exponentially growing solutions to the Schrödinger equation. A pivotal step in this method involves computing a scattering transform from boundary data. However, this step is severely ill-posed and requires stabilization via approximation techniques. This paper investigates how different approximation strategies for the scattering transform impact the quality of reconstructions, specifically for high-contrast, sharp-edged targets.
Methodological Comparison
The study compares three distinct reconstruction algorithms:
- D-bar Method with $\mathbf{t}^{\text{exp}}$: This approach simplifies the complex boundary traces of exponentially growing solutions by replacing them with a basic exponential function ($e^{ikx}$).
- D-bar Method with $\mathbf{t}^{\text{B}}$: An advanced approximation where the authors solve an approximate boundary integral equation. By substituting the standard Green's function for the Laplacian in place of the complex Faddeev Green's function, they obtain significantly more accurate traces for the scattering transform.
- Calderón’s Linearization Method: A classic approach that assumes internal conductivity is a small perturbation from a constant background.
To manage the exponential blow-up of scattering transforms at high frequencies, all methods employ a frequency cut-off radius ($R$) in the complex plane.
Key Insights: Linearization vs. Non-linearization
A significant theoretical insight of this work is that Calderón’s method can be viewed as a heavily linearized version of the D-bar method using the $\mathbf{t}^{\text{exp}}$ approximation. This explains why both methods produce similar visual results and why both fail to recover the full amplitude of high-contrast objects. Linearization inherently assumes the target is a "small perturbation," which fundamentally limits its ability to reconstruct large jumps in conductivity.
In contrast, the $\mathbf{t}^{\text{B}}$ method captures the non-linear physics of the problem more effectively. By investing computational effort into solving the boundary integral equation, the method yields boundary traces that better reflect the underlying conductivity distribution.
Results and Discussion
The algorithms were evaluated using simulated numerical data, including single-jump and two-jump radially symmetric conductivities.
The $\mathbf{t}^{\text{B}}$ method consistently outperformed the other techniques in two key areas:
- Amplitude Recovery: For both highly conductive and resistive targets, the $\mathbf{t}^{\text{B}}$ method successfully recovered extreme conductivity values that the other methods significantly underestimated.
- Spatial Resolution: The $\mathbf{t}^{\text{B}}$ method tracked sharp jump discontinuities with much higher precision, whereas the linearized methods tended to blur the boundaries.
While the $\mathbf{t}^{\text{B}}$ method is computationally more demanding—requiring the solution of a linear system via GMRES—the resulting improvements in image fidelity are substantial.
Conclusion
This study demonstrates that the choice of approximation for the scattering transform is critical in the D-bar method. The $\mathbf{t}^{\text{B}}$ approximation, by incorporating more accurate boundary traces, provides a powerful tool for reconstructing the sharp, high-contrast features typical of biological tissues. Future research should focus on applying this method to clinical EIT data, such as monitoring mechanical ventilation, where accurate organ boundary resolution is essential.
Summary updated