Related Experiment Video
Updated: Jun 25, 2025
10:14
Author Spotlight: Enhancing Neurorehabilitation Through EEG, Motor Imagery, and Virtual Reality
Published on: May 10, 2024
941
An Intersubject Brain-Computer Interface Based on Domain-Adversarial Training of Convolutional Neural Network
IEEE Transactions on Bio-Medical Engineering
|May 23, 2024
Summary
This study introduces DA-TSnet, an end-to-end framework for attention decoding using electroencephalography (EEG). DA-TSnet effectively addresses interindividual variability, achieving high accuracy in both simulated and real online experiments for brain-computer interfaces.
Area of Science:
- Neuroscience and Biomedical Engineering
- Machine Learning for Signal Processing
- Brain-Computer Interface (BCI) Development
Background:
- Attention decoding using electroencephalography (EEG) is crucial for daily applications.
- High interindividual variability in EEG signals poses a significant challenge for universal model training.
- Existing methods often require extensive preprocessing and feature extraction.
Purpose of the Study:
- To propose an end-to-end brain-computer interface (BCI) framework, DA-TSnet, designed to overcome interindividual variability in EEG signals for attention decoding.
- To develop a model that integrates temporal and spatial feature extraction with domain-adversarial training.
- To validate the framework's performance through offline, simulated online, and real online experiments.
Main Methods:
- Developed DA-TSnet, an end-to-end framework combining temporal and spatial one-dimensional (1D) convolutional neural networks.
- Employed a domain-adversarial training strategy, jointly optimizing task and domain losses by maximizing domain loss and minimizing task loss.
- Conducted offline analysis, simulated online experiments on 85 subjects, and real online experiments on 22 subjects.
Main Results:
- Achieved a leave-one-subject-out (LOSO) cross-validation accuracy of 89.40% ± 9.96%, outperforming state-of-the-art attention EEG decoding methods.
- Demonstrated outstanding accuracy of 88.07% ± 11.22% in simulated online experiments.
- Attained an average accuracy surpassing 86% in real online experiments.
Conclusions:
- The end-to-end DA-TSnet framework eliminates the need for elaborate preprocessing, saving time and effort.
- Domain-adversarial training effectively addresses high interindividual variability in EEG signals, offering value for other EEG decoding tasks.
- The robust performance in both offline and online experiments highlights DA-TSnet's potential for reliable BCI applications.
