Dual-channel TRCA-net based on cross-subject positive transfer for SSVEP-BCI

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Steady-state visual evoked potential-based brain-computer interfaces (SSVEP-BCI) show promise but suffer from low decoding accuracy and high inter-subject variability.
• Existing methods struggle to generalize across different users, limiting practical SSVEP-BCI applications.

Purpose of the Study:

• To enhance SSVEP-BCI decoding accuracy and information transfer rate.
• To reduce inter-subject variability for broader SSVEP-BCI adoption.
• To introduce a novel cross-subject transfer learning approach for improved SSVEP-BCI performance.

Main Methods:

• A dual-channel TRCA-net (DC-TRCA-net) method was developed, incorporating cross-subject positive transfer.
• An innovative Transfer-Accuracy-based Subject Selection (T-ASS) strategy was implemented for effective source subject selection.
• A deep learning network integrated with the SSVEP Domain Adaptation Network (SSVEP-DAN) was utilized, alongside cross-subject data augmentation.

Main Results:

• DC-TRCA-net demonstrated superior performance compared to existing networks on two large-scale public benchmark datasets.
• The T-ASS strategy effectively selected source subjects, mitigating negative transfer risks.
• Significant performance gains were observed, especially in complex experimental conditions, highlighting improved model generalization.

Conclusions:

• The proposed DC-TRCA-net method significantly enhances SSVEP-BCI decoding accuracy and information transfer rate.
• The T-ASS strategy and SSVEP-DAN integration contribute to robust cross-subject transfer learning.
• This approach offers a promising solution for developing more reliable and widely applicable SSVEP-BCI systems.