Improving classification performance of motor imagery BCI through EEG data augmentation with conditional generative...

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Accurate electroencephalogram (EEG) classifiers are crucial for brain-computer interface (BCI) performance.
• Machine learning (ML) and deep learning (DL) models require large datasets, but EEG data collection is challenging due to variability and cost, leading to data scarcity and overfitting.
• This data scarcity hinders the development of reliable and generalizable EEG classifiers.

Purpose of the Study:

• To propose a novel EEG data augmentation (DA) framework using conditional generative adversarial networks (cGANs) to overcome the data scarcity problem.
• To enhance the performance and generalization capabilities of EEG classifiers for BCI applications.
• To validate the effectiveness of the proposed cGAN-based DA method against existing DA techniques and baseline models.

Main Methods:

• A novel EEG data augmentation (DA) framework utilizing conditional generative adversarial networks (cGANs) was developed.
• The framework was experimentally validated on two public motor imagery (MI) EEG datasets (BCI competition IV IIa and III IVa).
• The proposed DA method was evaluated in conjunction with eight different EEG classifiers, including traditional ML and state-of-the-art DL models, and compared against three existing DA methods.

Main Results:

• Most data augmentation (DA) methods, when applied with appropriate proportions, improved classification performance compared to no DA.
• The proposed cGAN-based DA method demonstrated superior performance improvement over other existing DA methods.
• The experimental results confirm the effectiveness of the proposed DA framework in enhancing EEG classifier performance for motor imagery-based BCIs.

Conclusions:

• The proposed conditional generative adversarial network (cGAN)-based data augmentation framework effectively addresses the challenge of EEG data scarcity in brain-computer interfaces (BCIs).
• This novel approach significantly enhances the classification performance of EEG classifiers, particularly for motor imagery tasks.
• The cGAN-based DA method shows promise as a valuable tool for improving the reliability and accuracy of BCIs.