Spatial filter and feature selection optimization based on EA for multi-channel EEG.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) signals in Brain-Computer Interface (BCI) systems are typically band-limited.
• The band-limited multiple Fourier linear combiner (BMFLC) with a Kalman filter (KF) estimates EEG signal amplitudes in real-time within specific frequency bands.
• High-dimensional feature vectors from multi-channel EEG and BMFLC degrade classifier performance.

Purpose of the Study:

• To address the performance degradation in multi-channel EEG-based BCIs caused by high-dimensional feature vectors.
• To optimize spatial filtering and feature selection simultaneously using an evolutionary algorithm (EA).
• To propose and evaluate novel BMFLC-based BCI configurations.

Main Methods:

• Application of a real-valued evolutionary algorithm (EA) for optimizing spatial filters and feature selection.
• Encoding both spatial filter parameters and feature selection into the EA's solution.
• Optimizing the EA solution based on classification error.
• Proposing three distinct BMFLC-based BCI configurations.

Main Results:

• The evolutionary algorithm effectively tackled the high-dimensionality problem.
• The proposed BMFLC-KF configuration integrated with the covariance matrix adaptation evolution strategy (CMAES) demonstrated superior performance.
• Comparative analysis confirmed the effectiveness of the EA-driven optimization approach.

Conclusions:

• Evolutionary algorithms offer a robust method for optimizing feature extraction in EEG-based BCIs.
• The BMFLC-KF combined with CMAES presents a promising approach for enhancing BCI classification accuracy.
• This work provides a pathway for improving the efficiency and performance of BCI systems through advanced signal processing and optimization techniques.