Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EEG_RL-Net: Enhancing EEG MI Classification through Reinforcement Learning-Optimised Graph Neural Networks

Htoo Wai Aung, Jiao Jiao Li, Yang An, Steven W. Su

TL;DR

This work tackles EEG MI classification for BCIs by extending a graph-based neural approach with reinforcement learning. It builds EEG_RL-Net by reusing a pre-trained EEG_GLT-Net graph-convolution block at density $13.39\%$ and integrating a Dueling Deep Q-Network (Dueling DQN) RL agent to decide, in an episode of length $H=20$, whether to classify or skip each time point. Across $20$ subjects on PhysioNet, EEG_RL-Net achieves a mean accuracy of $95.36\%$, with an optimal reward configuration reaching $96.40\%$ and rapid classifications, vastly outperforming the prior EEG_GLT-Net performance of $83.95\%$. The RL approach also identifies time points that are ambiguous and defers their classification, contributing to safer, faster real-time BCIs. This work thus offers a practical pathway to high-accuracy, low-latency EEG-MI decoding and points to further gains via tighter graph-feature optimization.

Abstract

Brain-Computer Interfaces (BCIs) rely on accurately decoding electroencephalography (EEG) motor imagery (MI) signals for effective device control. Graph Neural Networks (GNNs) outperform Convolutional Neural Networks (CNNs) in this regard, by leveraging the spatial relationships between EEG electrodes through adjacency matrices. The EEG_GLT-Net framework, featuring the state-of-the-art EEG_GLT adjacency matrix method, has notably enhanced EEG MI signal classification, evidenced by an average accuracy of 83.95% across 20 subjects on the PhysioNet dataset. This significantly exceeds the 76.10% accuracy rate achieved using the Pearson Correlation Coefficient (PCC) method within the same framework. In this research, we advance the field by applying a Reinforcement Learning (RL) approach to the classification of EEG MI signals. Our innovative method empowers the RL agent, enabling not only the classification of EEG MI data points with higher accuracy, but effective identification of EEG MI data points that are less distinct. We present the EEG_RL-Net, an enhancement of the EEG_GLT-Net framework, which incorporates the trained EEG GCN Block from EEG_GLT-Net at an adjacency matrix density of 13.39% alongside the RL-centric Dueling Deep Q Network (Dueling DQN) block. The EEG_RL-Net model showcases exceptional classification performance, achieving an unprecedented average accuracy of 96.40% across 20 subjects within 25 milliseconds. This model illustrates the transformative effect of the RL in EEG MI time point classification.

EEG_RL-Net: Enhancing EEG MI Classification through Reinforcement Learning-Optimised Graph Neural Networks

TL;DR

This work tackles EEG MI classification for BCIs by extending a graph-based neural approach with reinforcement learning. It builds EEG_RL-Net by reusing a pre-trained EEG_GLT-Net graph-convolution block at density and integrating a Dueling Deep Q-Network (Dueling DQN) RL agent to decide, in an episode of length , whether to classify or skip each time point. Across subjects on PhysioNet, EEG_RL-Net achieves a mean accuracy of , with an optimal reward configuration reaching and rapid classifications, vastly outperforming the prior EEG_GLT-Net performance of . The RL approach also identifies time points that are ambiguous and defers their classification, contributing to safer, faster real-time BCIs. This work thus offers a practical pathway to high-accuracy, low-latency EEG-MI decoding and points to further gains via tighter graph-feature optimization.

Abstract

Brain-Computer Interfaces (BCIs) rely on accurately decoding electroencephalography (EEG) motor imagery (MI) signals for effective device control. Graph Neural Networks (GNNs) outperform Convolutional Neural Networks (CNNs) in this regard, by leveraging the spatial relationships between EEG electrodes through adjacency matrices. The EEG_GLT-Net framework, featuring the state-of-the-art EEG_GLT adjacency matrix method, has notably enhanced EEG MI signal classification, evidenced by an average accuracy of 83.95% across 20 subjects on the PhysioNet dataset. This significantly exceeds the 76.10% accuracy rate achieved using the Pearson Correlation Coefficient (PCC) method within the same framework. In this research, we advance the field by applying a Reinforcement Learning (RL) approach to the classification of EEG MI signals. Our innovative method empowers the RL agent, enabling not only the classification of EEG MI data points with higher accuracy, but effective identification of EEG MI data points that are less distinct. We present the EEG_RL-Net, an enhancement of the EEG_GLT-Net framework, which incorporates the trained EEG GCN Block from EEG_GLT-Net at an adjacency matrix density of 13.39% alongside the RL-centric Dueling Deep Q Network (Dueling DQN) block. The EEG_RL-Net model showcases exceptional classification performance, achieving an unprecedented average accuracy of 96.40% across 20 subjects within 25 milliseconds. This model illustrates the transformative effect of the RL in EEG MI time point classification.
Paper Structure (20 sections, 19 equations, 5 figures, 9 tables, 4 algorithms)

This paper contains 20 sections, 19 equations, 5 figures, 9 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview
  4. Dataset Description and Pre-processing
  5. Graph Feature Extraction
  6. Graph Representation
  7. Spectral Graph Filtering
  8. Training EEG_GLT-Net
  9. EEG MI Time Points GNN Features
  10. Problem Redefinition
  11. Data Preprocessing and Data Splitting
  12. Dueling Deep Q-Learning
  13. EEG_RL Algorithm
  14. Model Setting and Evaluation Metrics
  15. Results and Discussion
  16. ...and 5 more sections

Figures (5)

  • Figure 1: EEG_GLT-Net model aung2024eeggltnet: (a) Overall architecture (classifying EEG MI of one time point $\frac{1}{160}s$ of signals from 64 EEG electrodes), (b) Components inside the spectral graph convolution block, (c) Chebyshev spectral graph convolution
  • Figure 2: Overview of the EEG_RL-Net model: Incorporation of the pre-trained EEG_GCN Block at a 13.39% $m_g$ density from the EEG_GLT-Net, coupled with an RL Block
  • Figure 3: Agent interaction with EEG_RL Environment
  • Figure 4: Conversion of EEG MI time points into states using the pre-trained EEG_GCN Block, grouped into episodes comprising 20 states each
  • Figure 5: EEG_RL-Net's RL Block: Featuring the Dueling Deep Q Network (DQN), this component predicts the q-values linked to various actions