BR-SFDA: A Source-Target Bidirectional Refined SFDA for Privacy Preserving EEG-based BCIs.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalography (EEG)-based brain-computer interfaces (BCIs) struggle with cross-subject decoding due to neural activity variability.
• Transfer learning is effective but raises privacy concerns regarding source subject EEG data.

Purpose of the Study:

• To propose a privacy-preserving framework for cross-subject EEG classification.
• To address challenges in EEG decoding, including inter-subject variability and data privacy.

Main Methods:

• A source-target bidirectional refined source-free domain adaptation (BR-SFDA) framework is proposed.
• BR-SFDA incorporates data augmentation, a multi-criteria fused metric for pre-training, and structured graph learning for self-supervised fine-tuning.
• The framework operates within a 'pretraining and fine-tuning' paradigm, refining both front-end and back-end processes.

Main Results:

• BR-SFDA demonstrated superior performance in cross-subject motor imagery decoding and emotion recognition tasks across four datasets.
• The framework effectively handles inter-subject variability and preserves data privacy.
• The study validated the effectiveness of data augmentation, filtering, structured graph learning, and domain adaptation components.

Conclusions:

• The proposed BR-SFDA framework offers an effective solution for privacy-preserving cross-subject EEG classification.
• This approach significantly improves BCI performance while safeguarding sensitive neural data.
• Future research can build upon this bidirectional adaptation strategy for enhanced BCI applications.