Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EEG-GMACN: Interpretable EEG Graph Mutual Attention Convolutional Network

Haili Ye, Stephan Goerttler, Fei He

TL;DR

EEG-GMACN tackles interpretability and uncertainty calibration in EEG-based graph signal processing by introducing an Inverse Graph Weight Module and a Mutual Attention Mechanism within a graph convolutional framework. It encodes EEG signals into a graph using electrode topology, learns discriminative representations via GCNs with mutual attention, and derives interpretable electrode weights through Grad-CAM-like gradients blended with the adjacency. The method achieves improved classification performance and provides credible, electrode-level explanations and calibrated uncertainty estimates on the BCIII dataset, advancing transparent EEG analysis for clinical and neuroscience applications. This work lays groundwork for scalable, interpretable graph-based EEG models with practical relevance to brain–computer interfaces and neurological disease research.

Abstract

Electroencephalogram (EEG) is a valuable technique to record brain electrical activity through electrodes placed on the scalp. Analyzing EEG signals contributes to the understanding of neurological conditions and developing brain-computer interface. Graph Signal Processing (GSP) has emerged as a promising method for EEG spatial-temporal analysis, by further considering the topological relationships between electrodes. However, existing GSP studies lack interpretability of electrode importance and the credibility of prediction confidence. This work proposes an EEG Graph Mutual Attention Convolutional Network (EEG-GMACN), by introducing an 'Inverse Graph Weight Module' to output interpretable electrode graph weights, enhancing the clinical credibility and interpretability of EEG classification results. Additionally, we incorporate a mutual attention mechanism module into the model to improve its capability to distinguish critical electrodes and introduce credibility calibration to assess the uncertainty of prediction results. This study enhances the transparency and effectiveness of EEG analysis, paving the way for its widespread use in clinical and neuroscience research.

EEG-GMACN: Interpretable EEG Graph Mutual Attention Convolutional Network

TL;DR

EEG-GMACN tackles interpretability and uncertainty calibration in EEG-based graph signal processing by introducing an Inverse Graph Weight Module and a Mutual Attention Mechanism within a graph convolutional framework. It encodes EEG signals into a graph using electrode topology, learns discriminative representations via GCNs with mutual attention, and derives interpretable electrode weights through Grad-CAM-like gradients blended with the adjacency. The method achieves improved classification performance and provides credible, electrode-level explanations and calibrated uncertainty estimates on the BCIII dataset, advancing transparent EEG analysis for clinical and neuroscience applications. This work lays groundwork for scalable, interpretable graph-based EEG models with practical relevance to brain–computer interfaces and neurological disease research.

Abstract

Electroencephalogram (EEG) is a valuable technique to record brain electrical activity through electrodes placed on the scalp. Analyzing EEG signals contributes to the understanding of neurological conditions and developing brain-computer interface. Graph Signal Processing (GSP) has emerged as a promising method for EEG spatial-temporal analysis, by further considering the topological relationships between electrodes. However, existing GSP studies lack interpretability of electrode importance and the credibility of prediction confidence. This work proposes an EEG Graph Mutual Attention Convolutional Network (EEG-GMACN), by introducing an 'Inverse Graph Weight Module' to output interpretable electrode graph weights, enhancing the clinical credibility and interpretability of EEG classification results. Additionally, we incorporate a mutual attention mechanism module into the model to improve its capability to distinguish critical electrodes and introduce credibility calibration to assess the uncertainty of prediction results. This study enhances the transparency and effectiveness of EEG analysis, paving the way for its widespread use in clinical and neuroscience research.

Paper Structure

This paper contains 13 sections, 8 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Preprocessing and Feature Extraction
  4. Correlation Electrodes Search
  5. Mutual Attention Mechanism Module
  6. Inverse Graph Weight Module
  7. Experiments
  8. Datasets
  9. Baseline
  10. Evaluation Metrics
  11. Ablation Study
  12. Comparison Experiment
  13. Conclusion

Figures (2)

  • Figure 1: Overview of GSP Method with Interpretability Module for Revealing Electrode Weights in EEG Recognition Model.
  • Figure 2: Visual Comparison of Electrode Graph Weight Heatmaps Exported using the IEGW Module on the BCI III Test Set. (a) w/o mutual attention layers in the model, and (b) w/ mutual attention layers in the model.