EEG Signal Prediction for Motor Imagery Classification in Brain-Computer Interfaces

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) typically require numerous electroencephalography (EEG) channels for accurate motor imagery (MI) classification.
• High-density EEG systems are costly, time-consuming to set up, and vulnerable to data loss from damaged electrodes, limiting practical applications.

Purpose of the Study:

• To develop a signal prediction-based method for high-accuracy MI classification using a reduced number of EEG channels.
• To assess the efficacy of elastic net regression for predicting full-channel EEG signals from a minimal set of central channels.

Main Methods:

• A signal prediction model was developed using elastic net regression.
• EEG signals from 8 central channels were used to estimate signals from 22 complete channels.
• Predicted EEG signals were utilized for feature extraction and MI classification.

Main Results:

• The proposed prediction method achieved an average MI classification accuracy of 78.16%.
• Performance varied among subjects, ranging from 62.30% to 95.24%.
• The method outperformed traditional few-channel and full-channel EEG approaches for MI classification.

Conclusions:

• The signal prediction method enables accurate MI classification using a reduced EEG channel set.
• This approach significantly mitigates the time and cost constraints associated with high-density EEG systems.
• The findings support the practical implementation of MI-based BCIs with fewer electrodes.