Neurophysiologically Meaningful Motor Imagery EEG Simulation With Applications to Data Augmentation

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery-based Brain-Computer Interfaces (MI-BCIs) show promise for neurorehabilitation and neuroprosthetics.
• Accurate MI pattern recognition in electroencephalography (EEG) is limited by data issues and challenges in acquiring large-scale, user-specific MI-EEG data for deep learning (DL).
• Existing EEG signal generation methods lack neurophysiological plausibility and user-specific integration for data augmentation.

Purpose of the Study:

• To introduce PySimMIBCI, a novel framework for generating realistic, neurophysiologically plausible MI-EEG signals.
• To address data limitations in MI-BCI research by providing simulated data for model validation and augmentation.
• To integrate user-specific neurophysiological information into the data generation process.

Main Methods:

• Developed PySimMIBCI, a framework integrating neurophysiological activity into biophysical forward models to simulate MI-EEG signals.
• Incorporated simulation of varying user capabilities and fatigue effects within the generated EEG data.
• Utilized simulated user-specific data for a novel data augmentation strategy.

Main Results:

• Simulated MI-EEG data demonstrated high resemblance to real EEG data.
• The data augmentation strategy using PySimMIBCI-generated data significantly outperformed state-of-the-art augmentation methods.
• Deep learning models trained with the proposed augmentation strategy showed performance enhancements of up to 15%.

Conclusions:

• PySimMIBCI provides a valuable tool for generating realistic MI-EEG data, overcoming limitations of real-world data acquisition.
• Simulated neurophysiologically plausible data effectively serves as a data augmentation resource for MI-BCI research.
• The framework facilitates improved training and performance of deep learning models for MI-BCI applications.