A multi-domain graph convolutional network-based prediction model for personalized motor imagery action

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Motor imagery (MI)-based brain-computer interfaces (BCIs) decode imagined actions.
• Individual differences in MI are linked to cognitive EEG signals.
• Predicting personalized MI actions is crucial for BCI efficacy.

Purpose of the Study:

• To propose a Multi-domain Graph Convolutional Network (M-GCN) for personalized MI action prediction.
• To leverage cognitive EEG data for enhanced MI action decoding.
• To improve the accuracy and personalization of BCI systems.

Main Methods:

• Developed an M-GCN model integrating time, frequency, and spatial EEG features.
• Constructed multi-domain brain networks using various EEG quantization methods.
• Employed spectral Graph Convolutional Network (GCN) to analyze functional connectivity.
• Validated the model using a subject-independent, leave-one-subject-out cross-validation approach.

Main Results:

• The M-GCN achieved a prediction accuracy of 73.60% for personalized MI actions.
• Significantly outperformed baseline (15.87% improvement) and single-domain models (7.2% improvement).
• Demonstrated the effectiveness of multi-domain feature fusion and GCN in BCI.

Conclusions:

• The M-GCN accurately predicts personalized MI actions, enhancing BCI usability.
• Multi-domain feature fusion based on cognitive tasks and GCN is highly effective.
• This study offers a novel and efficient method for personalized BCI development.