Domain-aware domain-class adaptation network for motor execution to motor imagery EEG classification

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor imagery (MI) is crucial for electroencephalogram (EEG)-based brain-computer interfaces (BCIs).
• Transfer learning enhances MI-EEG decoding, but cross-task transfer from motor execution (ME) to MI remains underexplored.
• ME and MI tasks share analogous cortical activation patterns, suggesting potential for cross-task knowledge transfer.

Purpose of the Study:

• To address the challenge of cross-task transfer learning from ME to MI for improved EEG-based BCIs.
• To develop a novel deep learning model that facilitates knowledge transfer between ME and MI tasks.
• To enhance the performance and accessibility of MI-EEG decoding.

Main Methods:

• Proposed a domain-aware domain-class adaptation network (DDCA Net) with a domain-shared feature extractor and domain-specific re-weighting blocks.
• Employed domain-level alignment using Maximum Mean Discrepancy (MMD) to minimize feature distribution differences.
• Utilized a bi-classifier adversarial learning framework for implicit class-level alignment by ensuring decision boundary consistency across domains.

Main Results:

• DDCA Net significantly outperformed the within-task baseline, achieving a 7.71% increase in classification accuracy when 80% of target-domain samples were used for training.
• Successfully converted approximately 80% of previously BCI-illiterate subjects into BCI-literate users.
• Demonstrated the effectiveness of cross-task transfer learning from ME to MI using domain adaptation techniques.

Conclusions:

• This study is the first to demonstrate the feasibility of domain adaptation for cross-task transfer learning in MI-EEG classification.
• The findings highlight the potential of integrating ME and MI tasks for advanced BCI development.
• The proposed DDCA Net offers a promising direction for improving BCI performance and user accessibility.