Optimal Channel Selection of Multiclass Motor Imagery Classification Based on Fusion Convolutional Neural Network...

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) commonly use motor imagery (MI) for human-machine communication.
• EEG signals from MI present classification challenges due to noise and redundant channels, increasing costs.
• Optimal channel selection is crucial for efficient and accurate BCI performance.

Purpose of the Study:

• To propose an optimal channel selection method for multiclass MI classification using a Fusion Convolutional Neural Network with Attention (FCNNA).
• To enhance EEG signal feature extraction and reduce computational costs in BCI systems.
• To improve the accuracy and efficiency of MI-based BCIs.

Main Methods:

• Developed a CNN model with spatial and temporal filters to capture EEG signal features.
• Integrated a Convolutional Block Attention Module (CBAM) for enhanced feature extraction.
• Employed a genetic algorithm for optimal channel selection, offering fixed and variable channel options.

Main Results:

• The FCNNA model demonstrated a 6.41% improvement in multiclass classification over baseline models.
• Achieved 93.09% accuracy for binary classification (left-hand/right-hand movements).
• Cross-subject multiclass classification reached 68.87% accuracy, improving to 84.53% after channel selection.

Conclusions:

• The proposed FCNNA model with optimal channel selection significantly enhances EEG-based MI classification accuracy.
• The method is efficient in both channel selection and classification, providing superior results with reduced channel sets.
• This approach offers a cost-effective and high-performance solution for BCI applications.