https://github.com/rishannp/motor-imagery---graph-attention-network

This is works in attempt to develop novel, state-of-the-art models for decoding EEG MI data from patient datasets. Specifically using GAT, highlighting their potential advantages.
https://github.com/rishannp/motor-imagery---graph-attention-network

amyotrophic-lateral-sclerosis bci-competition-iv-2a brain-computer-interface deep-learning eeg graph-attention-networks keras machine-learning motor-imagery-classification tensorflow

Last synced: 26 days ago
JSON representation

This is works in attempt to develop novel, state-of-the-art models for decoding EEG MI data from patient datasets. Specifically using GAT, highlighting their potential advantages.

• Host: GitHub
• URL: https://github.com/rishannp/motor-imagery---graph-attention-network
• Owner: rishannp
• License: gpl-3.0
• Created: 2024-09-09T13:13:27.000Z (about 2 months ago)
• Default Branch: main
• Last Pushed: 2024-09-14T21:30:34.000Z (about 2 months ago)
• Last Synced: 2024-10-13T00:41:33.394Z (26 days ago)
• Topics: amyotrophic-lateral-sclerosis, bci-competition-iv-2a, brain-computer-interface, deep-learning, eeg, graph-attention-networks, keras, machine-learning, motor-imagery-classification, tensorflow
• Language: Python
• Homepage:
• Size: 146 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Graph Attention Networks against Deep Learning Benchmark pipelines for 2 Class Motor Imagery Decoding in ALS and BCI Comp IV 2a. 

## Overview

Accurately classifying EEG signals, especially for individuals with neurodegenerative conditions like Amyotrophic Lateral Sclerosis (ALS), poses a significant challenge due to high inter-subject and inter-session changes in signal. This study introduces a novel three-layer Graph Attention Network (GAT) model for motor imagery (MI) classification, utilizing Phase Locking Value (PLV) as the graph input. The improvement underscores the effectiveness of graph-based representations in enhancing classification performance for neurodegenerative conditions. Evaluated across two datasets—the BCI Competition IV 2a dataset and a longitudinal ALS dataset with approximately 320 trials collected over 1-2 months—our GAT model not only achieves competitive accuracy but also shows robust performance across sessions and subjects. This stability is highlighted by statistically significant reductions in variance compared to state-of-the-art methods. These results support the hypothesis that phase-locking value-based graph representations can enhance neural representations in BCIs, offering promising avenues for more personalized approaches in MI classification. Our work emphasizes the potential for further optimizing GAT architectures and feature sets, pointing to future research directions that could improve performance and efficiency in MI classification tasks.

## Models Implemented

- EEGNet (Implementation from Scratch)

- EEGNet (Originally from V. Lawhern https://github.com/vlawhern/arl-eegmodels)

- DeepConvNet (Originally from V. Lawhern https://github.com/vlawhern/arl-eegmodels)

- ShallowConvNet (Originally from V. Lawhern https://github.com/vlawhern/arl-eegmodels)

- GCN (Personally developed, custom architecture)

- GCN (Personally developed, custom architecture)

- GAT (Personally developed, custom architecture)

## Results

## To Be Done:

- Prepare and Include the EfficientNet Implementations

- Publish ALS Dataset for open access and reproducibility

-