EEG artifact suppression using a Wasserstein generative adversarial autoencoder
Engineering Research Express · 01.07.2026 · 8 (13) · 135332
DOI: 10.1088/2631-8695/ae7ff2 ↗Kurz gefasst
Fachartikel in der Zeitschrift Engineering Research Express, online erschienen am 02.07.2026. Verfasst von Kumar Avinash Chandra und 2 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Gefördert von: Indian Council of Medical Research.
Abstract (Originalsprache)
EEG recordings are highly susceptible to artifacts such as ocular (EOG), muscular (EMG), and environmental noise, which degrade signal quality and hinder downstream analysis. In this work, we propose a Wasserstein Generative Adversarial Autoencoder (WGAN-AE) that jointly optimizes signal reconstruction and distribution alignment to achieve artifact suppression while preserving physiological fidelity. Unlike conventional AEs or GAN-based models, the proposed framework incorporates a signal-domain Wasserstein critic and latent-space regularization, enabling stable training and improved generalization across diverse artifact types. The model is trained in a self-supervised manner using paired clean–noisy EEG data without manual intervention and evaluated on multiple artifact types (EMG, EOG, and Gaussian noise). Experimental results demonstrate consistent improvements over ICA, wavelet-based methods, and conventional AEs, achieving lower RMSE and higher signal to noise-ratio (SNR) across EOG, EMG, and Gaussian noise conditions. Cross-dataset validation further confirms the robustness and generalization capability of the proposed approach, highlighting its suitability for real-time brain computer interface and clinical applications. For EOG artifact suppression, WGAN-AE achieved the lowest RMSE ( 7.03 ± 0.5 ) and MAE ( 5.79 ± 0.4 ), alongside the highest SNR ( 20.6 ± 0.8 dB) and correlation ( 0.91 ± 0.02 ). Comparable improvements were observed for EMG and Gaussian noise, with RMSE reductions of 15%–20% and SNR gains of 2–3 dB relative to conventional methods. All improvements were statistically significant ( p < 0.01 , paired Wilcoxon test). These findings demonstrate that WGAN-AE provides robust, data-driven artifact suppression while maintaining neural signal fidelity, supporting its potential integration into clinical neurophysiology and real-time brain–computer interface applications.
Förderung
- Indian Council of Medical Research
Zitieren
APA
Chandra, K. A., Upadhyay, P. K., & Mishra, S. K. (2026). EEG artifact suppression using a Wasserstein generative adversarial autoencoder. Engineering Research Express, 8(13), 135332. https://doi.org/10.1088/2631-8695/ae7ff2
Vancouver
Chandra KA, Upadhyay PK, Mishra SK. EEG artifact suppression using a Wasserstein generative adversarial autoencoder. Engineering Research Express. 2026;8(13):135332. doi:10.1088/2631-8695/ae7ff2
BibTeX
@article{Chandra2026,
author = {Chandra, Kumar Avinash and Upadhyay, Prabhat Kumar and Mishra, Sudhansu Kumar},
title = {EEG artifact suppression using a Wasserstein generative adversarial autoencoder},
journal = {Engineering Research Express},
year = {2026},
volume = {8},
number = {13},
pages = {135332},
doi = {10.1088/2631-8695/ae7ff2}
}
EUVIMED