MPP-GNN Improves Alzheimer's Classification with Adaptive fMRI Analysis

Yang Zhang, Xiao Zhou, Jonathan Warrell, Avram Holmes, Xuan Zhang, Mark Gerstein· August 3, 2026 View original

Key takeaways

  • MPP-GNN improves Alzheimer's disease classification using fMRI data.
  • It adaptively discovers subject-specific brain functional modules.
  • These modules explicitly guide connectivity pattern refinement and representation learning.
  • The model achieves superior AUC and aligns with known brain network organization.

Who benefits

HealthcarePharmaceuticalsMedical DevicesScientific Research

Summary

This research introduces Meta Probabilistic Pooling GNN (MPP-GNN), a novel graph neural network that performs subject-adaptive community detection for fMRI-based Alzheimer's disease classification. It addresses limitations of existing methods by discovering subject-specific brain modules and using them to guide representation learning, achieving superior classification accuracy.

Functional magnetic resonance imaging (fMRI) is a crucial tool for studying the brain, and Graph Neural Networks (GNNs) have shown promise in classifying brain disorders like Alzheimer's disease (AD) by analyzing functional connectivity. However, current GNN methods often assume a fixed number of functional brain modules across all individuals, overlooking significant inter-subject variability. To overcome this, researchers propose the Meta Probabilistic Pooling GNN (MPP-GNN). This model employs a coupled, bilevel optimization approach to adaptively partition brain graphs, identifying unique functional modules for each subject. These discovered subject-specific modules then serve as an explicit prior, guiding the refinement of connectivity patterns and the overall representation learning process within the GNN. MPP-GNN was validated on two public datasets for AD classification, where it achieved the highest Area Under the Curve (AUC) compared to established baselines. Furthermore, analysis revealed that MPP-GNN's discovered modules align well with canonical functional-network organizations, and it effectively identifies a network-level dedifferentiation pattern associated with Alzheimer's disease.

Why it matters

For healthcare professionals and researchers in neuroimaging, MPP-GNN offers a more accurate and personalized approach to diagnosing Alzheimer's disease using fMRI data. Its ability to account for individual brain variability could lead to earlier and more precise detection, facilitating timely intervention.

How to implement this in your domain

  1. 1Evaluate MPP-GNN's performance on internal fMRI datasets for Alzheimer's or other neurological disorder classification.
  2. 2Collaborate with AI researchers to adapt and deploy MPP-GNN in clinical research settings for improved diagnostic accuracy.
  3. 3Investigate the subject-specific brain modules discovered by MPP-GNN to gain deeper insights into individual disease progression.
  4. 4Integrate MPP-GNN into a multimodal diagnostic pipeline combining fMRI with other biomarkers for comprehensive patient assessment.

Original post by Yang Zhang, Xiao Zhou, Jonathan Warrell, Avram Holmes, Xuan Zhang, Mark Gerstein

"arXiv:2607.28681v1 Announce Type: new Abstract: Functional magnetic resonance imaging (fMRI) is a widely used technique for studying the brain. Recent methods that utilize graph neural networks (GNNs) for analysis of brain functional connectivity have shown great potential for th…"

View on X

Originally posted by Yang Zhang, Xiao Zhou, Jonathan Warrell, Avram Holmes, Xuan Zhang, Mark Gerstein on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses