Improving the Cross-Subject Performance of the ERP-Based Brain-Computer Interface Using Rapid Serial Visual...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Brain-computer interfaces (BCIs) offer communication channels but face challenges in cross-subject recognition due to individual, environmental, and system variability.
• Developing reliable and stable BCIs often requires extensive user-specific calibration, limiting their direct applicability.

Purpose of the Study:

• To develop a brain-computer interface (BCI) system with reduced calibration needs for new users.
• To enhance cross-subject classification and stability in electroencephalography (EEG)-based BCIs.

Main Methods:

• Employed Rapid Serial Visual Presentation (RSVP) as an evoked paradigm to minimize individual differences in event-related potential (ERP) patterns.
• Proposed and utilized the Correlation Analysis Rank (CAR) algorithm for improved cross-individual classification with minimal training data.
• Conducted experiments with 58 subjects, analyzing ERP similarities and calculating Information Transfer Rates (ITR) and Receiver Operating Characteristic (ROC) curves.

Main Results:

• RSVP evoked more consistent ERP patterns across subjects compared to the matrix paradigm, indicated by a higher average matching number (20 vs. 6).
• The RSVP paradigm achieved a 13% higher average Information Transfer Rate (ITR) of 43.18 bits/min compared to the matrix paradigm.
• The proposed CAR algorithm demonstrated superior performance over traditional random selection, achieving an AUC of 0.8 versus 0.65.

Conclusions:

• The RSVP paradigm effectively reduces individual differences in ERPs, making it favorable for cross-subject BCI classification.
• The CAR algorithm significantly enhances classification accuracy and efficiency in BCI systems.
• This combined approach offers a feasible method for achieving stable, reliable, and less calibration-dependent ERP-based BCIs.