A Study of Various Feature Extraction Methods on a Motor Imagery Based Brain Computer Interface System

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Brain Computer Interface (BCI) systems leverage Movement Imagination (MI) for diverse applications.
• Current MI-BCI systems often use separate feature extraction and classification methods.
• Real-time applications necessitate efficient and accurate feature selection.

Purpose of the Study:

• To identify the most effective feature extraction methods for MI-based BCI.
• To evaluate feature performance using Linear Discriminant Analysis (LDA) classification.
• To select optimal features for real-time BCI navigation.

Main Methods:

• Applied a wide array of features to recorded 3-class MI data (left hand, right hand, foot).
• Utilized Linear Discriminant Analysis (LDA) for feature selection in offline mode.
• Recorded data specifically for left hand, right hand, and foot motor imagery tasks.

Main Results:

• Auto-Regressive (AR), Mean Absolute Value (MAV), and Band Power (BP) features demonstrated significantly higher accuracy.
• These selected features outperformed other tested feature sets.
• The Power Spectral Density (PSD) based alpha-Band Power (α-BP) feature yielded the highest average accuracy.

Conclusions:

• Selected features (AR, MAV, BP) were implemented for real-time BCI navigation.
• MI-based BCI systems exhibit subject-specific characteristics.
• PSD-based α-BP emerged as the most accurate feature for this MI-BCI system.