MPP-GNN: Adaptive Brain Module Detection for Alzheimer's Classification

Researchers have introduced a novel machine learning framework known as Meta Probabilistic Pooling GNN (MPP-GNN) aimed at classifying Alzheimer's disease (AD) through fMRI data. This model overcomes the shortcomings of current graph neural network (GNN) techniques that rely on a uniform number of functional brain modules for all individuals, failing to account for variability among subjects. MPP-GNN utilizes hierarchical adaptive graph partitioning to identify brain modules unique to each subject, which serve as an explicit prior for enhancing edge refinement and representation learning. This method is structured as a coupled, bilevel optimization challenge. Validation on two public fMRI datasets revealed enhanced classification accuracy. The findings are outlined in a paper on arXiv (arXiv:2607.28681).