Related Experiment Video
Updated: Jul 29, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.9K
A GAN Guided Parallel CNN and Transformer Network for EEG Denoising.
IEEE Journal of Biomedical and Health Informatics
|May 23, 2023
Summary
This study introduces GCTNet, a novel deep learning model for removing artifacts from electroencephalography (EEG) signals. GCTNet effectively enhances EEG signal quality by capturing temporal dependencies and ensuring holistic consistency, outperforming existing methods.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Physiological artifacts contaminate electroencephalography (EEG) signals, compromising data quality and analysis.
- Deep learning methods offer advantages for EEG denoising but face limitations in capturing temporal artifact characteristics and ensuring holistic signal consistency.
Purpose of the Study:
- To develop an advanced deep learning network, GCTNet, for effective artifact removal in EEG signals.
- To address limitations in existing methods regarding temporal artifact characteristics and holistic consistency between denoised and clean EEG signals.
Main Methods:
- Proposed GCTNet, a Generative Adversarial Network (GAN) guided parallel Convolutional Neural Network (CNN) and Transformer network.
- The generator utilizes parallel CNN and transformer blocks to capture local and global temporal dependencies.
- A discriminator is employed to enforce holistic consistency between denoised and clean EEG signals.
Main Results:
- GCTNet demonstrated significant outperformance over state-of-the-art networks in various artifact removal tasks on both semi-simulated and real EEG data.
- Specifically, for electromyography artifact removal, GCTNet achieved an 11.15% reduction in Root Mean Square Error (RMSE) and a 9.81% improvement in Signal-to-Noise Ratio (SNR).
Conclusions:
- GCTNet offers a promising solution for practical EEG signal artifact removal.
- The proposed method effectively addresses limitations of existing deep learning techniques by considering temporal characteristics and holistic consistency.
