Comparing user-dependent and user-independent training of CNN for SSVEP BCI

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Steady-state visually evoked potentials (SSVEP) are crucial for brain-computer interfaces (BCIs).
• Convolutional neural networks (CNNs) offer potential for SSVEP detection but require efficient training strategies.
• Comparing user-dependent (UD) and user-independent (UI) training is vital for practical BCI applications.

Purpose of the Study:

• To comparatively analyze CNN training methodologies for SSVEP detection.
• To evaluate the efficacy of user-dependent (UD) versus user-independent (UI) training scenarios.
• To investigate the impact of different feature types (magnitude vs. complex spectrum) on CNN performance for SSVEP classification.

Main Methods:

• CNNs were trained using both UD and UI scenarios.
• Two feature types were explored: magnitude spectrum (M-CNN) and complex spectrum (C-CNN).
• Performance was benchmarked against established methods like canonical correlation analysis (CCA), TRCA, and FBCCA using two distinct SSVEP datasets.

Main Results:

• UD training generally outperformed UI training, as expected.
• The UI-C-CNN approach demonstrated performance comparable to UD-M-CNN.
• C-CNN methods consistently outperformed M-CNN, with UI-C-CNN offering a favorable balance between performance and training data requirements.

Conclusions:

• The proposed complex spectrum CNN (C-CNN) is a promising method for SSVEP detection.
• C-CNN achieves improved performance in both UD and UI training paradigms.
• The UI-C-CNN approach presents a practical and effective solution for SSVEP-based BCIs.