Realizing the Application of EEG Modeling in BCI Classification: Based on a Conditional GAN Converter

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) modeling is crucial for brain-computer interface (BCI) development but faces challenges in practical application due to a lack of parameter guidelines and personalized data.
• Current EEG modeling in BCI primarily exists at a theoretical, qualitative level, creating a disconnect between foundational research and real-world usability.

Purpose of the Study:

• To establish a practical link between simulated EEG data and its application in BCI classification.
• To overcome limitations in acquiring personalized EEG data for BCI training by utilizing simulated data.

Main Methods:

• Developed a mathematical model for surface EEG simulation based on the physics of EEG generation, incorporating a parallel 3-population neural mass model, equivalent dipole, and forward computation.
• Designed a conditional generative adversarial network (GAN) converter to transform theoretical EEG simulations into a practical format, usable without individual bio-information.
• Integrated converted simulated EEG data into the training of BCI classifiers within a microexpression-assisted BCI paradigm.

Main Results:

• The inclusion of simulated EEG data in classifier training led to a significant improvement in overall BCI performance (P = 0.04).
• Test performance demonstrated an average increase of 2.17% ± 4.23, with a maximum improvement of 12.60% ± 1.81 when compared to training with insufficient real data.
• The study successfully established an initial connection between theoretical EEG simulation and BCI classification.

Conclusions:

• The proposed method, combining surface EEG simulation with a GAN-based converter, provides a viable solution for enhancing BCI applications.
• This approach effectively bridges the gap between theoretical EEG modeling and practical BCI classification, offering a novel way to leverage simulated data.
• The findings suggest a promising direction for improving BCI system robustness and performance, particularly in scenarios with limited real-world data.