EEG classification for motor imagery BCI using phase-only features extracted by independent component analysis

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Accurate classification of electroencephalography (EEG) signals is crucial for developing reliable motor imagery brain-computer interface (MI-BCI) systems.
• Existing methods may face challenges in computational efficiency and feature extraction robustness.

Purpose of the Study:

• To propose and evaluate a novel feature extraction method for motor imagery EEG data classification.
• To achieve high classification accuracy using Extreme Learning Machines (ELM) with extracted phase-based features.
• To assess the computational efficiency of the proposed method for MI-BCI applications.

Main Methods:

• Utilized the BCI Competition-IV 2008 dataset IIa for binary classification of motor imagery EEG data.
• Applied Independent Component Analysis (ICA) to time series EEG data, followed by transformation to the Fourier domain.
• Extracted phase information from the Fourier spectrum to compute a maximized cross-correlation connectivity matrix, vectorizing its upper diagonal as features for ELM.

Main Results:

• Achieved a nested cross-validated classification accuracy of 97.80% (p < 0.0022) using the proposed phase-only features.
• The feature extraction and classification process demonstrated relative computational inexpensiveness.
• The method proved effective for binary classification tasks in motor imagery EEG.

Conclusions:

• The proposed novel feature extraction method based on EEG signal phase information is highly effective for motor imagery BCI.
• The method achieves excellent classification accuracy and is computationally efficient, making it suitable for real-time MI-BCI applications.
• This approach offers a promising avenue for advancing the reliability and performance of BCI systems.