Reproduce “Adaptive Gated Graph Convolutional Network for Explainable Diagnosis of Alzheimer's Disease using EEG Data” on OpenResearch

About this paper

Dominik Klepl, Fei He, Min Wu, Daniel J. Blackburn, Ptolemaios G. Sarrigiannis

Researchers from Coventry University, A*STAR, and the Universities of Sheffield and Royal Devon developed an Adaptive Gated Graph Convolutional Network (AGGCN) to diagnose Alzheimer’s Disease from EEG data with explainability. The AGGCN achieved 89.1% accuracy in eyes-closed EEG and revealed specific AD-related brain network alterations, including increased long-distance connectivity and the significance of 6-10 Hz frequency bands.

Key insights

• Adaptive, sample-specific graph learning is important for accurately capturing underlying brain connectivity patterns relevant to AD diagnosis from EEG.
• Dynamically weighting information contributions from different spatial scales through a gating mechanism in GNNs improves performance for EEG-based neurological disorder classification.
• Adaptive structure-aware pooling significantly enhances graph embedding, reduces model parameters, and aids in localizing critical brain regions for diagnostic explainability.
• The model's explainability features reveal that AD patients exhibit increased long-distance connectivity, decreased local coupling, and a reliance on local information processing.
• The 6-10 Hz (theta and low alpha) frequency bands are identified as particularly critical for distinguishing AD from healthy controls.

Summary by alphaXiv.