Cross-Subject Transfer Method Based on Domain Generalization for Facilitating Calibration of SSVEP-Based BCIs

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) commonly use spatial filtering for signal enhancement.
• Existing methods rely on individual calibration, leading to time-consuming sessions and increased user fatigue.
• This limits the practical usability and accessibility of SSVEP-BCI systems.

Purpose of the Study:

• To propose a novel cross-subject transfer method for SSVEP-BCI.
• To eliminate the need for subject-specific calibration data.
• To enhance SSVEP detection performance and improve BCI usability.

Main Methods:

• Employs domain generalization to transfer spatial filters and templates across subjects.
• Learns domain-invariant features by maximizing intra- and inter-subject correlations.
• Utilizes SSVEP data from target and neighboring stimuli for filter and template optimization.
• Constructs feature vectors using four types of correlation coefficients for SSVEP detection.

Main Results:

• The proposed cross-subject transfer method significantly improves SSVEP detection performance.
• Performance enhancement was validated across three independent SSVEP datasets.
• Outperforms existing state-of-the-art transfer learning methods.

Conclusions:

• The developed method offers an effective transfer learning strategy for SSVEP-BCIs.
• Eliminates the need for extensive data collection from new users.
• Holds significant potential for promoting practical, user-friendly BCI applications.