Self-Attention GAN for Enhanced Electrical Impedance Tomography in Fetal Monitoring with Conductive Fabrics

A. K. Khambampati¹, M. Jeon¹, K. Y. Kim¹,^2*

¹Electronic Engineering Department, Jeju National University, South Korea, 63243
²Kyoto College of Graduate Sudies for Informatics, Kyoto, Japan, 606-8225

^*Email: kyungyk@jejunu.ac.kr

ABSTRACT

Electrical resistance tomography (ERT) offers a non-invasive, low-cost solution for physiological monitoring, making it well-suited for wearable fetal monitoring using conductive fabrics. However, such systems often suffer from noise and low-resolution reconstructions using traditional reconstruction methods. We propose a Self-Attention Generative Adversarial Network (SAGAN) to enhance ERT image quality by learning to map noisy inputs to high-fidelity conductivity distributions. The self-attention mechanism enables the model to capture global context, improving structural detail. The proposed SAGAN model is benchmarked against U-Net and Pix2Pix GAN, showing superior performance in terms of SSIM and PSNR. These results highlight the effectiveness of attention-based generative models for enhancing EIT image quality in smart fabric-based fetal monitoring applications.

Keywords: Conductive fabric, Electrical resistance tomography, Generative adversarial networks, Attention mechanism Attention.

Industrial Application: Applications of process tomography in industry, medical and NDT&E