Multiclass Classification of Visual Electroencephalogram Based on Channel Selection, Minimum Norm Estimation...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Classifying multiclass visual electroencephalogram (EEG) signals is crucial for advanced brain-computer interface (BCI) applications.
• The nonlinearity and nonstationarity of EEG signals present significant challenges for accurate multiclass classification in BCI systems.
• Existing BCI systems face limitations in supervising multiple BCI tasks due to classification constraints.

Purpose of the Study:

• To develop and evaluate deep learning models for high-accuracy multiclass classification of visual EEG signals into 40 distinct classes.
• To address the challenges posed by EEG signal nonlinearity and nonstationarity for BCI applications.
• To enable multitask BCI applications by improving the classification of EEG data.

Main Methods:

• Implemented mutual information-based discriminant channel selection and minimum-norm estimate algorithms for EEG data enhancement and channel selection.
• Utilized deep EEGNet and convolutional recurrent neural networks (CRNNs) for classifying 40-class visual EEG data.
• Employed k-fold cross-validation to assess the performance of the proposed deep learning models.

Main Results:

• Achieved average classification accuracies of 94.8% with EEGNet and 89.8% with CRNNs.
• Demonstrated the effectiveness of the proposed methods in handling nonlinear and nonstationary EEG signals.
• Validated the approach using k-fold cross-validation, ensuring robust performance estimation.

Conclusions:

• The proposed deep learning approach offers a promising solution for multiclass visual EEG signal classification in BCI.
• The method enables multitask BCI applications with reduced channel usage (less than 50%) and network parameters (less than 110K).
• Satisfactory classification accuracies pave the way for more sophisticated and efficient embedded BCI systems.