Generative Diffusion-Based Task Incremental Learning Method for Decoding Motor Imagery EEG.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Motor neurorehabilitation benefits from learning diverse motor imagery (MI) tasks.
• EEG-based brain-computer interfaces (BCIs) are effective for MI tasks.
• Current MI decoding methods struggle to balance plasticity for new tasks and stability for previously learned tasks.

Purpose of the Study:

• Propose GD-TIL, a generative diffusion-based incremental learning method for MI decoding.
• Address the challenge of balancing plasticity and stability in BCIs.
• Enhance neurorehabilitation through improved MI task learning.

Main Methods:

• Utilized data augmentation by segmenting and recombining EEG signals.
• Developed a multi-scale temporal-spatial feature extractor (MTSFE) integrating convolutions and attention mechanisms.
• Implemented a self-supervised task generalization (SSTG) mechanism and a prototype-guided generative replay (PGGR) module for incremental learning.

Main Results:

• Achieved continuous decoding accuracies of 80.20% and 81.32%.
• Demonstrated excellent plasticity and stability of the GD-TIL method.
• Outperformed state-of-the-art incremental learning methods in MI decoding.

Conclusions:

• GD-TIL shows significant potential for continuous neurorehabilitation.
• Highlights the synergy between MI-based BCIs and generative AI.
• Offers a promising approach for adaptive and robust brain-computer interfaces.