Classification of motor imagery BCI using multivariate empirical mode decomposition

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalogram (EEG) is crucial for brain-computer interfaces (BCI) but suffers from noise and non-stationarity.
• Extracting relevant frequency band information from noisy EEG signals is challenging for BCI applications.
• Existing frequency estimation algorithms struggle with low signal-to-noise ratios and closely spaced frequency bands.

Purpose of the Study:

• To evaluate the effectiveness of multivariate empirical mode decomposition (MEMD) for motor imagery BCI.
• To assess MEMD's capability in handling noisy and non-stationary EEG data.
• To compare MEMD, particularly its noise-assisted variant (NA-MEMD), against established methods for feature extraction in BCI.

Main Methods:

• Application of multivariate empirical mode decomposition (MEMD) for direct multichannel EEG processing.
• Utilizing the noise-assisted mode of MEMD (NA-MEMD) for enhanced time-frequency analysis.
• Comparative performance analysis using synthetic datasets and a standard motor imagery BCI dataset.

Main Results:

• Direct multichannel processing with MEMD significantly improves the localization of frequency information in EEG.
• NA-MEMD provides a highly localized time-frequency representation of EEG signals.
• MEMD-based methods demonstrate superior performance compared to other state-of-the-art techniques.

Conclusions:

• MEMD and NA-MEMD are effective tools for processing noisy EEG signals in motor imagery BCI.
• These advanced signal processing techniques enhance feature extraction accuracy for BCI applications.
• MEMD offers a promising approach for improving the robustness and performance of brain-computer interfaces.