Comparing Steady-State Visually Evoked Potentials Frequency Estimation Methods in Brain-Computer Interface With the...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-Computer Interface (BCI) systems facilitate communication between users and external devices.
• Steady-State Visually Evoked Potentials (SSVEP) are a common BCI paradigm.
• Various feature extraction methods are employed to interpret SSVEP signals.

Purpose of the Study:

• To compare the performance of several feature extraction algorithms for SSVEP-based BCIs.
• To identify the most effective algorithm using minimal electroencephalography (EEG) channels.
• To evaluate computational efficiency alongside accuracy.

Main Methods:

• Comparison of Canonical Correlation Analysis (CCA), LASSO, L1-MCCA, MsetCCA, CFA, and MLR algorithms.
• Statistical analysis to determine performance differences.
• Evaluation across different numbers of EEG electrodes (1, 2, and 8 channels).

Main Results:

• MLR, MsetCCA, and CFA outperformed CCA, LASSO, and L1-MCCA with 8 EEG channels.
• CFA achieved the highest F-scores with only 1 or 2 EEG channels.
• CFA exhibited superior performance over MLR and MsetCCA on various electrode setups and offered the fastest computation time.

Conclusions:

• CFA is a highly promising algorithm for practical SSVEP-based BCIs, particularly under constraints of limited EEG channels and short time windows.
• The study highlights CFA's efficiency and accuracy, suggesting its potential for widespread adoption in BCI systems.
• CFA surpasses existing frequency recognition algorithms in specific real-world BCI scenarios.