Motor Imagery Classification Using Effective Channel Selection of Multichannel EEG

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalography (EEG) is crucial for brain-computer interface (BCI) development, but practical implementation requires efficient data processing.
• Reducing computational complexity in BCI systems is essential for real-world applications.
• Motor imagery (MI) classification relies on identifying distinct cognitive patterns from EEG signals.

Purpose of the Study:

• To present an entropy-based approach for selecting effective EEG channels for motor imagery (MI) classification in BCI systems.
• To reduce computational complexity and improve classification accuracy by identifying and utilizing channels with higher information content.
• To validate the proposed channel selection method using established BCI datasets.

Main Methods:

• An entropy-based method was developed to calculate the information content of individual EEG channels.
• Channels with higher mean entropy across trials were selected as effective channels for MI classification.
• Common Spatial Pattern (CSP) was applied to sub-band signals of selected channels for feature extraction.
• Support Vector Machine (SVM) was used for classifying right-hand and right-foot MI tasks.

Main Results:

• The proposed entropy-based channel selection method effectively identifies informative EEG channels.
• The approach led to reduced computational complexity compared to using all available channels.
• Experimental results on public BCI datasets demonstrated superior performance over existing state-of-the-art techniques.
• Improved classification accuracy for motor imagery tasks was achieved.

Conclusions:

• The entropy-based channel selection strategy is effective for enhancing BCI performance.
• This method offers a computationally efficient way to improve motor imagery classification accuracy.
• The findings suggest a promising direction for developing more practical and accurate BCI systems.