Breaking the performance barrier in deep learning-based SSVEP-BCIs: a joint frequency-phase training strategy.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Deep learning shows promise for electroencephalography (EEG)-based brain-computer interfaces (BCIs) using steady-state visual evoked potential (SSVEP) classification.
• SSVEP signals possess both frequency and phase characteristics crucial for accurate classification.
• Current deep learning methods often neglect the combined utilization of frequency and phase information, limiting classification performance.

Purpose of the Study:

• To address the limitations of existing deep learning strategies in SSVEP classification.
• To propose and validate a novel Joint Frequency-Phase Training Strategy (JFPTS) for enhanced SSVEP classification.
• To fully exploit the dual frequency and phase properties inherent in SSVEP signals.

Main Methods:

• The proposed Joint Frequency-Phase Training Strategy (JFPTS) employs two distinct stages with specialized time-window sampling.
• The first stage utilizes a frequency prior-driven sampling scheme to optimize frequency component utilization.
• The second stage implements a phase-locked sampling scheme to improve intra-category phase consistency.

Main Results:

• Experiments conducted on two public datasets confirmed the efficacy of JFPTS.
• The JFPTS-enhanced deep learning model demonstrated superior performance compared to existing state-of-the-art methods.
• Performance significantly surpassed the established benchmark of task discriminative component analysis (TDCA).

Conclusions:

• JFPTS represents a novel training paradigm for deep learning in SSVEP classification.
• This strategy effectively leverages both frequency and phase characteristics of SSVEP signals.
• The proposed method is expected to advance deep learning applications in SSVEP-BCIs and encourage wider adoption.