Virtual Reality (VR) Paradigm-Agnostic Motor Imagery Decoding Using Lightweight Network With Adaptive Attention...

Area of Science:

• Neuroscience
• Rehabilitation Engineering
• Brain-Computer Interfaces

Background:

• Motor imagery (MI) is crucial for brain-computer interfaces (BCIs) in neurorehabilitation.
• Existing 2D MI paradigms do not fully engage sensorimotor networks.
• There is a need for more immersive and effective MI-based BCI approaches.

Purpose of the Study:

• To design an immersive MI paradigm using virtual reality (VR).
• To develop a novel decoding algorithm integrating depthwise separable convolution and multi-head self-attention.
• To evaluate the algorithm's performance and generalizability in VR and 2D paradigms.

Main Methods:

• Developed a VR environment with 3D palm motion stimuli for immersive MI.
• Proposed a novel decoding algorithm combining depthwise separable convolution and multi-head self-attention.
• Evaluated the algorithm on BCI Competition IV-2a and PhysioNet MI datasets.

Main Results:

• The novel algorithm demonstrated superior classification accuracy compared to existing methods.
• Achieved an average 8% increase in kappa score over EEGNet for four-class MI tasks.
• Showed consistent performance across both VR and 2D paradigms, confirming robustness.

Conclusions:

• The immersive VR MI paradigm enhances user engagement in neurorehabilitation.
• The proposed decoding framework advances EEG-based intention recognition in VR.
• This approach offers a promising avenue for improving BCI applications in motor rehabilitation.