EEG-based communication: evaluation of alternative signal prediction methods

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Brain-computer interfaces (BCIs) enable individuals to control external devices using neural activity.
• Electroencephalography (EEG) signals from the sensorimotor cortex can be translated into cursor movements on a screen.
• Optimizing the intercept in the EEG-to-cursor translation equation is crucial for equal accessibility of targets.

Purpose of the Study:

• To compare five different methods for selecting the intercept in EEG-based cursor control.
• To evaluate these methods based on performance balance and consistency across trials.
• To identify the most effective and practical intercept selection method for BCI applications.

Main Methods:

• Offline analysis of EEG data from trained subjects performing cursor control tasks.
• Implementation of five intercept selection methods: moving average, weighted sum, blocked moving average, blocked weighted sum, and blocked recursive sum.
• Evaluation of methods based on the balance of upward/downward cursor movements and trial-to-trial performance consistency.

Main Results:

• All five methods showed similar overall performance when data from all subjects were combined.
• The moving average, blocked moving average, and blocked recursive sum methods demonstrated greater consistency across individual subjects.
• The weighted sum and blocked weighted sum methods exhibited less consistent performance across subjects.

Conclusions:

• The moving average method, utilizing the five most recent top and bottom trial pairs, is recommended for its consistent performance and computational simplicity.
• This finding has implications for improving the usability and efficiency of EEG-based BCIs.
• Further research could explore adaptive intercept adjustments in real-time BCI operation.