A New Method to Generate Artificial Frames Using the Empirical Mode Decomposition for an EEG-Based Motor Imagery BCI

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-Computer Interfaces (BCIs) are crucial for neurorehabilitation in stroke and Disorder of Consciousness (DoC) patients.
• BCIs require user-specific training data for classifier development, which can be time-consuming.
• Extended training periods pose challenges for patient compliance and system usability.

Purpose of the Study:

• To investigate the efficacy of using artificial EEG frames to reduce BCI training time.
• To compare classifier performance trained with a mix of real and artificial data versus real data alone.
• To assess the potential of artificial data in mitigating the impact of artifacts in EEG signals.

Main Methods:

• Generating artificial EEG frames by applying Empirical Mode Decomposition (EMD) and mixing Intrinsic Mode Functions (IMFs).
• Training BCI classifiers using datasets with varying percentages of artificial EEG frames.
• Evaluating classifier performance and training time reduction across different subjects.
• Testing the method on a public dataset for validation.

Main Results:

• Up to 50% of real EEG frames could be replaced with artificial data in some subjects, halving training time (720s to 360s).
• In other subjects, replacing 12.5% of real frames reduced training time by 90s.
• The method effectively reduced the impact of artifacts by replacing corrupted frames.
• High performance was achieved using 87.5% artificial frames on a public dataset.

Conclusions:

• Artificial EEG frames generated via EMD offer a viable strategy to significantly reduce BCI training duration.
• This approach can enhance classifier robustness by mitigating artifacts and potentially improve patient rehabilitation outcomes.
• Further research is needed to validate these findings in patient populations and explore alternative IMF mixing strategies and BCI paradigms.