Boosting template-based SSVEP decoding by cross-domain transfer learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) offer a promising communication pathway.
• Current SSVEP BCIs face challenges with data variability across sessions, subjects, and equipment.
• Improving the robustness and reducing calibration time are crucial for practical BCI applications.

Purpose of the Study:

• To develop a generalized transfer-learning framework for SSVEP BCIs.
• To enhance SSVEP decoding performance by leveraging cross-domain data transfer.
• To investigate the efficacy of least-squares transformation (LST) for cross-domain adaptation in SSVEP BCIs.

Main Methods:

• Enhanced template-based SSVEP decoding by integrating least-squares transformation (LST).
• Applied LST to transfer calibration data across different domains (sessions, subjects, EEG montages).
• Compared the LST-based method against standard task-related component analysis (TRCA) and naive transfer learning.

Main Results:

• Verified LST's effectiveness in mitigating SSVEP variability during cross-domain data transfer.
• Achieved significantly higher SSVEP decoding accuracy with the LST-based method compared to TRCA and naive transfer learning.
• Demonstrated superior performance of LST, particularly when calibration data is limited.

Conclusions:

• The LST-based transfer learning framework effectively leverages existing data across subjects and devices for SSVEP BCIs.
• The proposed framework significantly improves decoding accuracy over standard methods, especially under data scarcity.
• This approach facilitates plug-and-play functionality and broadens the practical applicability of SSVEP BCIs.