Algebraic Connectivity Reveals Modulated High-Order Functional Networks in Alzheimer's Disease

Giorgio Dolci, Silvia Saglia, Lorenza Brusini, Vince D. Calhoun, Ilaria Boscolo Galazzo, Gloria MenegazarXiv 2025 · DOI · arXiv

TL;DR

Researchers have moved beyond simple "point-to-point" brain maps to model the complex, multi-region ensembles involved in Alzheimer’s Disease (AD). By applying algebraic connectivity to weight these high-order "hypergraphs," the authors identified subtle functional disruptions that traditional methods miss, successfully linking tau protein buildup to cognitive decline through specific neural network failures.

What Problem Does This Paper Address?

Most functional MRI (fMRI) studies rely on "pairwise" connectivity—essentially a map of how Region A correlates with Region B. However, the human brain doesn't just work in pairs; it operates in complex, overlapping communities.

While hypergraphs (mathematical structures where an "edge" can connect any number of nodes) can model these high-order relationships, a major challenge remains: How do you assign a meaningful "weight" to a group of regions? Standard approaches often just average the correlations within the group, which discards vital information about the group's internal topology and robustness. This paper seeks a more mathematically rigorous way to "score" these functional ensembles to better detect the fingerprints of Alzheimer's.

Methods

The authors propose a novel pipeline for hypergraph modeling:

  1. Structure Extraction: Using LASSO regression on fMRI data from healthy controls, they identified "neighborhoods" of regions that predict each other's activity, forming the hypergraph's backbone.
  2. The Innovation ($a(\mathcal{G})$): To weight each hyperedge, they calculated its algebraic connectivity (also known as the Fiedler value). In graph theory, this is the second-smallest eigenvalue of the Laplacian matrix.
  3. Why it matters: Unlike a simple average, $a(\mathcal{G})$ measures the "synchronization potential" and robustness of a sub-network. A high value means the regions are tightly integrated; a low value suggests the network is fragile and prone to disconnecting.
  4. Validation: They compared this method against three baselines (Gaussian kernels, Pearson correlations, and LASSO coefficients) using machine learning (Random Forests) and mediation analysis.

Datasets

The study utilized data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI):

  • ADNI-3: Used for the primary analysis (587 subjects: 310 Healthy Controls, 199 Mild Cognitive Impairment, and 78 AD patients).
  • ADNI-2: Used as an independent cohort for external validation to ensure the findings weren't specific to one dataset.
  • Tau-PET & Cognitive Scores: Used to correlate functional changes with biological pathology and clinical symptoms.

Key Insights and Ideas

  • Robustness as a Biomarker: The study found that in AD patients, specific sub-networks (like the somatomotor and default-mode networks) show a significant drop in algebraic connectivity. This suggests that AD doesn't just "dim" brain activity; it fundamentally destabilizes the integration of neural communities.
  • The "Missing Link" in Mediation: One of the most compelling findings was that these hyperedge weights act as a mediator. It’s not just that tau protein causes memory loss; rather, tau buildup degrades the algebraic connectivity of the salience and somatomotor networks, which then leads to cognitive decline.
  • Compensatory Hyperconnectivity: Interestingly, some networks (like the dorsal attention network) showed increased connectivity in AD patients. This likely represents the brain's attempt to compensate for damage in other areas, a nuance captured more clearly by high-order modeling than by pairwise graphs.

Key Results

  • Sensitivity: The $a(\mathcal{G})$ method identified 6 significant hyperedges differentiating the groups, whereas standard Pearson correlation and Gaussian methods only found 2.
  • Classification Accuracy: Using $a(\mathcal{G})$ weights as features, a Random Forest model achieved 80% accuracy in distinguishing Healthy Controls from AD patients, outperforming all baseline weighting methods (which ranged from 65% to 77%).
  • Biological Grounding: Mediation analysis confirmed that connectivity in the salience/ventral attention and somatomotor networks partially explained the relationship between tau burden and ADAS-13 cognitive scores.

Limitations and Future Work

  • Structure Dependency: The hypergraph's "backbone" was built using LASSO regression. If the initial LASSO step misses a relationship, the subsequent algebraic connectivity analysis cannot recover it.
  • Connectedness Constraint: Algebraic connectivity is only strictly defined for connected subgraphs. While the authors didn't encounter disconnected hyperedges in this study, a more fragmented brain network might result in "null" scores.
  • Resolution Sensitivity: While the method worked on both 100-region and 400-region atlases, higher resolution introduces more noise. Finding the "Goldilocks" resolution for hypergraph modeling remains an open question.
  • Future Directions: The authors suggest applying this framework to other neurodegenerative diseases (like Parkinson's) or using it to track the longitudinal progression of AD over several years.

Summary updated

Abstract

Functional MRI is a neuroimaging technique that analyzes the functional activity of the brain by measuring blood-oxygen-level-dependent signals throughout the brain. The derived functional features can be used for investigating brain alterations in neurological and psychiatric disorders. In this work, we employed a hypergraph to model high-order functional relations across brain regions, introducing algebraic connectivity (a(G)) for estimating the hyperedge weights. The hypergraph structure was derived from healthy controls to build a common topology across individuals. The considered cohort for subsequent analyses included subjects covering the Alzheimer's disease (AD) continuum, encompassing both mild cognitive impairment and AD patients. Statistical analysis and three classification tasks: HC vs AD, MCI vs AD, and HC vs MCI, were performed to assess differences across the three groups and the potential of the hyperedge weights as functional features. Furthermore, a mediation analysis was performed to evaluate the reliability of the a(G) values, representing functional information as the mediator between tau-PET levels, a key biomarker of AD, and cognitive scores. The proposed approach identified a larger number of hyperedges statistically different across groups compared to state-of-the-art methods. The a(G) hyperedge weights also demonstrated a higher discriminative power in all three binary classifications. Finally, two hyperedges belonging to salience/ventral attention and somatomotor networks showed a partial mediation effect between the tau biomarker and cognitive decline. These results suggested that a(G) can be an effective approach for extracting the hyperedge weights, including important functional information that resides in the brain areas forming the hyperedges.
BibTeX
@misc{dolci2025algebraic,
  author = {Giorgio Dolci and Silvia Saglia and Lorenza Brusini and Vince D. Calhoun and Ilaria Boscolo Galazzo and Gloria Menegaz},
  title = {{Algebraic Connectivity Reveals Modulated High-Order Functional Networks in Alzheimer's Disease}},
  year = {2025},
  eprint = {2508.01252},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.01252},
  howpublished = {arXiv},
}
Figure