Performance Improvement with Reduced Number of Channels in Motor Imagery BCI System

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Classifying motor imagery (MI) Electroencephalogram (EEG) signals is crucial for Brain-Computer Interface (BCI) systems.
• Reducing channel count in BCI systems enhances flexibility, portability, and efficiency, particularly for multi-class scenarios.
• Current methods often require numerous EEG channels for accurate MI classification.

Purpose of the Study:

• To investigate the efficacy of combining Electrooculogram (EOG) channels with a reduced set of EEG channels for MI classification.
• To challenge the notion that EOG channels solely contribute noise, demonstrating their utility in MI signal classification.
• To develop and validate an optimized deep learning model for MI signal classification with channel reduction.

Main Methods:

• Utilized advanced deep learning architectures, including 1D convolution blocks and depthwise-separable convolutions.
• Implemented a channel reduction strategy by integrating a minimal number of EOG channels with a reduced EEG channel set.
• Evaluated the proposed method on two distinct datasets: BCI Competition IV Dataset IIa (4-class MI) and the Weibo dataset (7-class MI).

Main Results:

• Achieved 83% accuracy on dataset 1 (4-class MI) using only 6 channels (3 EEG, 3 EOG).
• Obtained 61% accuracy on dataset 2 (7-class MI) with 5 channels (3 EEG, 2 EOG).
• Demonstrated superior performance compared to relying solely on a larger number of EEG channels.

Conclusions:

• Combining EOG and reduced EEG channels is an effective strategy for MI classification in BCI.
• The proposed deep learning model successfully leverages EOG data for improved classification accuracy.
• This channel reduction method enhances the practicality and efficiency of BCI systems for real-world applications.