Image Reconstruction in Electrical Impedance Tomography through 1D-Convolutional Neural Network

Nomvussi, Serge Ayme Kouakouo; Mikulka, Jan

Image Reconstruction in Electrical Impedance Tomography through 1D-Convolutional Neural Network

Files

316-Nomvussi.pdf(1.08 MB)

Date

2025

Authors

Nomvussi, Serge Ayme Kouakouo

Mikulka, Jan

Publisher

Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií

Abstract

This paper presents a comparative analysis of image reconstruction performance using a 1D-Convolutional Neural Network (1D-CNN) against the Total Variation and the Gauss- Newton methods. The evaluation, conducted across multiple tests conditions, demonstrates that the 1D-CNN consistently outperforms both conventional methods in terms of correlation coefficient and structural similarity index (SSIM). In noise-free scenarios, the 1D-CNN achieves significantly higher correlation and SSIM values, indicating superior reconstruction accuracy. Furthermore, in the presence of noise (30 dB and 60 dB), the performance of the Total Variation and Gauss-Newton methods deteriorates considerably, whereas the 1D-CNN maintains high correlation and SSIM values, demonstrating strong robustness to noise. These findings highlight the effectiveness of deep learningbased approaches for image reconstruction, making the 1D-CNN a promising alternative to traditional reconstruction techniques.

Keywords

1D- convolutional Neural Network, Total Variation, Newton-Gauss, EIT

Citation

Proceedings I of the 31st Conference STUDENT EEICT 2025: General papers. s. 316-320. ISBN 978-80-214-6321-9
https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2025_sbornik_1.pdf

Document type

Peer-reviewed

Document version

Published version

Language of document

en

Document licence