Implement GCNs-Net Model for EEG Motor Imagery Signal Decoding in EEGdecoder Repository

[Buddies-as-you-know]

Objective

The goal of this issue is to implement a Graph Convolutional Neural Network (GCNs-Net) model for decoding time-resolved EEG motor imagery signals, as outlined in the paper "GCNs-Net: A Graph Convolutional Neural Network Approach for Decoding Time-Resolved EEG Motor Imagery Signals". This model aims to enhance the precision of brain activity decoding in brain-computer interface (BCI) systems.

Background

Traditional EEG signal classification methods do not consider the topological relationship among electrodes. Recent neuroscience research highlights the importance of network patterns in brain dynamics, suggesting that the Euclidean structure of electrodes might not fully capture signal interactions. The GCNs-Net addresses this by using graph convolutional layers to learn generalized features from the graph Laplacian of EEG electrodes, based on the absolute Pearson’s matrix of overall signals.

Implementation Details

• Graph Convolutional Layers: Implement layers to process the graph Laplacian of EEG electrodes.
• Pooling Layers: Add layers to reduce the dimensionality of the data.
• Fully-Connected Softmax Layer: Develop this layer for final prediction.
• Accuracy Goals: Aim for the highest averaged accuracy, as achieved in the paper (e.g., 93.06% and 88.57% on the PhysioNet dataset).
• Dataset: Use the PhysioNet dataset and the high gamma dataset as mentioned in the paper.

Expected Outcomes

• The model should be robust and adaptable to individual variability.
• Ensure that the performance is reproducible across cross-validation experiments.
• Documentation on how to use the model with examples.

Additional Information

• The original paper and its methodologies should be closely followed.
• The source code for the existing DL library for EEG task classification can be found at EEG-DL, which may provide useful references.
• Collaboration with team members who have expertise in neural networks and EEG signal processing is encouraged.

[agramfort]

yes it would be great to grow the model zoo here !

@Buddies-as-you-know can you give us a hand ?