Paper ID: 2407.03335

Dual-Domain Deep D-bar Method for Solving Electrical Impedance Tomography

Xiang Cao, Qiaoqiao Ding, Xiaoqun Zhang

The regularized D-bar method is one of the most prominent methods for solving Electrical Impedance Tomography (EIT) problems due to its efficiency and simplicity. It provides a direct approach by applying low-pass filtering to the scattering data in the non-linear Fourier domain, thereby yielding a smoothed conductivity approximation. However, D-bar images often present low contrast and low resolution due to the absence of accurate high-frequency information and ill-posedness of the problem. In this paper, we proposed a dual-domain neural network architecture to retrieve high-contrast D-bar image sequences from low-contrast D-bar images. To further accentuate the spatial features of the conductivity distribution, the widely adopted U-net has been tailored for conductivity image calibration from the predicted D-bar image sequences. We call such a hybrid approach by Dual-Domain Deep D-bar method due to the consideration of both scattering data and image information. Compared to the single-scale structure, our proposed multi-scale structure exhibits superior capabilities in reducing artifacts and refining conductivity approximation. Additionally, solving discrete D-bar systems using the GMRES algorithm entails significant computational complexity, which is extremely time-consuming on CPU-based devices. To remedy this, we designed a surrogate GPU-based Richardson iterative method to accelerate the data enhancement process by D-bar. Numerical results are presented for simulated EIT data from the KIT4 and ACT4 systems to demonstrate notable improvements in absolute EIT imaging quality when compared to existing methodologies.

Submitted: May 12, 2024