A Canonical Correlation Analysis-Based Transfer Learning Framework for Enhancing the Performance of SSVEP-Based BCIs.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Steady-state visual evoked potential (SSVEP) based brain-computer interfaces (BCIs) face a trade-off between high accuracy with extensive training data and low accuracy with minimal training.
• Existing methods struggle to balance performance and practicality, hindering real-world BCI applications.

Purpose of the Study:

• To develop a novel transfer learning framework using canonical correlation analysis (CCA) to enhance SSVEP BCI performance.
• To reduce the calibration effort required for new users by minimizing training data needs.

Main Methods:

• Proposed a CCA-based transfer learning framework (ASS-IISCCA) integrating intra- and inter-subject EEG data.
• Optimized spatial filters using CCA and developed an accuracy-based subject selection (ASS) algorithm to mitigate individual differences.
• Extracted features using optimized coefficients and recognized SSVEP frequencies via template matching.

Main Results:

• The ASS-IISCCA framework demonstrated significant improvements in SSVEP BCI performance.
• The method effectively reduced the number of training trials required for new users.
• Compared to the state-of-the-art Task-Related Component Analysis (TRCA), ASS-IISCCA showed superior results on a benchmark dataset with 35 subjects.

Conclusions:

• The proposed ASS-IISCCA framework offers a highly effective solution to the performance-practicality dilemma in SSVEP BCIs.
• This approach facilitates the real-world application of SSVEP BCIs by reducing calibration effort and improving accuracy for new users.