Similarity-constrained task-related component analysis for enhancing SSVEP detection

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) utilize subject-specific training algorithms like task-related component analysis (TRCA).
• TRCA extracts task-related components (TRCs) from electroencephalogram (EEG) signals to maximize trial reproducibility, but may retain trial-reproducible noise.
• This inherent noise can limit the effectiveness of SSVEP detection.

Purpose of the Study:

• To introduce a novel similarity-constrained TRCA (scTRCA) algorithm for SSVEP detection.
• To enhance the extraction of TRCs that are maximally correlated with SSVEPs by reducing task-related noise.
• To improve the overall performance and robustness of SSVEP-based BCIs.

Main Methods:

• Developed the scTRCA algorithm by incorporating similarity constraints into the TRCA objective function.
• Similarity constraints were introduced using covariance matrices between EEG training data and an artificial SSVEP template.
• Compared scTRCA against TRCA, multi-stimulus TRCA, and sine-cosine reference signal methods using two public datasets.

Main Results:

• The scTRCA algorithm demonstrated improved performance in target identification for SSVEPs.
• scTRCA significantly outperformed the other evaluated methods.
• Performance improvements were particularly notable when using insufficient training data, indicating enhanced robustness.

Conclusions:

• The proposed scTRCA algorithm effectively reduces task-related noise in SSVEP detection.
• scTRCA offers superior performance and robustness compared to existing methods, especially in scenarios with limited calibration data.
• This algorithm shows significant promise for advancing SSVEP-based BCIs.