Transforming Motor Imagery Analysis: A Novel EEG Classification Framework Using AtSiftNet Method

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery classification using electroencephalography (EEG) signals is crucial for Brain-Computer Interface (BCI) development.
• Existing methods often face challenges in effectively extracting and selecting relevant features from complex EEG data.
• The need for robust and computationally efficient feature extraction and selection techniques is paramount for BCI advancement.

Purpose of the Study:

• To propose and evaluate the AtSiftNet method, integrating Self-Attention for feature extraction and multiple feature selection techniques.
• To enhance the classification performance of motor imagery tasks using EEG signals.
• To assess the efficacy of the proposed method across various machine learning classifiers and validate its robustness.

Main Methods:

• EEG signals were denoised using multiscale principal component analysis.
• Self-Attention mechanism was employed for feature extraction from EEG trials.
• Eight different feature selection techniques were applied to extract the top 1 or 15 features.
• Five classification models, including Support Vector Machine (SVM), were used to evaluate performance.

Main Results:

• The AtSiftNet method, particularly with ReliefF and Independent Component Analysis feature selection, achieved high classification accuracies (up to 99.946%) for motor imagery.
• Support Vector Machine (SVM) classifier demonstrated excellent performance with the selected features.
• Five-fold cross-validation confirmed the model's robustness, yielding an average accuracy of 99.89%.

Conclusions:

• The AtSiftNet framework offers a resilient biomarker for motor imagery classification with minimal computational complexity.
• The proposed approach significantly enhances classification performance, making it suitable for practical Brain-Computer Interface applications.
• This study highlights the potential of combining Self-Attention with advanced feature selection for improved EEG signal analysis in BCI.