Contrastive Pretraining for Ensemble Graph Neural Networks in Alzheimer’s Disease Classification

Alzheimer’s disease (AD) is a progressive neurode-generative disorder that necessitates accurate early diagnosis. We propose a novel training strategy using an ensemble graph neural network (GNN) with contrastive pretraining for AD classification, utilizing atlas-based graphs from the AAL atlas to ensure anatomically meaningful node semantics. This ensemble combines various GNN architectures to capture diverse connectivity patterns, while contrastive pretraining enhances robust latent representation before supervised training. Evaluated on the publicly available ADNI dataset, our approach outperforms state-of-the-art methods across multiple metrics. Additionally, integrating PGExplainer improves model transparency by identifying disease-relevant subgraphs and contributions from specific brain regions. Our framework enhances both classification accuracy and explainability in graph-based AD diagnosis.