Comparing intra- and inter-individual correlational brain connectivity from functional and structural neuroimaging data

Xin Di, Bharat B. Biswal, Alzheimer’s Disease Neuroimaging Initiative

TL;DR

This study uses rare longitudinal datasets—including one individual scanned over 16 years—to reveal that brain connectivity patterns derived from correlations between different people (inter-individual) are driven by distinct underlying factors. Structural correlations are primarily shaped by slow, long-term aging processes, while functional and metabolic correlations are driven by transient, "state-like" fluctuations in brain activity.

What Problem Does This Paper Address?

In neuroimaging, researchers often talk about "brain connectivity" by looking at how different regions correlate across a large group of people. For example, if people with a large hippocampus also tend to have a large prefrontal cortex, we might infer a "structural covariance" between those regions. Similarly, "molecular connectivity" is inferred from correlations in glucose metabolism (PET scans) across individuals.

The big question is: What actually causes these correlations? Are they driven by shared genetic blueprints, life experiences, or the simple fact that these regions work together? Because most studies are "cross-sectional" (one scan per person), it is difficult to separate stable individual traits from the effects of aging or the participant's mental state during the scan. This paper tackles this by looking at "intra-individual" correlations—how regions within the same person change together over decades.

Methods

The researchers compared two main types of correlation matrices:

  1. Intra-individual: Calculated by looking at how brain measures (like volume or metabolism) fluctuate across multiple sessions for a single person.
  2. Inter-individual: The standard approach, calculating correlations across a group of different people.

They focused on three primary measures:

  • Gray Matter Volume (GMV): A structural measure from MRI.
  • Regional Homogeneity (ReHo): A functional measure derived from resting-state fMRI.
  • FDG-PET: A measure of glucose metabolism (the brain's "fuel consumption").

To understand the drivers of these correlations, the authors used cubic age models to decompose the data into "age-related effects" (slow changes) and "residuals" (short-term fluctuations or noise). They then used Principal Component Analysis (PCA) to see how similar these different "connectivity" maps were to a gold-standard reference: resting-state functional connectivity (RSFC).

Datasets

The study leveraged two unique longitudinal sources:

  • The Simon Dataset: A remarkable "deep phenotyping" dataset of a single healthy male scanned 73 times over 16 years (ages 30 to 47). This provided a high-resolution look at how one brain changes over time.
  • The ADNI Dataset: Data from the Alzheimer’s Disease Neuroimaging Initiative, specifically focusing on healthy elderly participants who had at least five longitudinal FDG-PET and MRI scans.

Key Insights and Ideas

  • Aging is the Architect of Structure: In the single-subject (Simon) dataset, the correlations in gray matter volume were almost entirely explained by the aging process. Regions didn't just vary randomly; they grew or shrank in a synchronized fashion dictated by the person's age.
  • Function is a "State of Mind": Unlike structural measures, functional correlations (ReHo and PET) were driven by "state-like" variability. Even after removing the effects of aging, the residual fluctuations strongly mirrored known functional networks. This suggests these measures capture the brain's "vibe" or mental state during the specific hour of the scan.
  • The Hierarchy of Similarity: Functional measures (ReHo and PET) are much better proxies for actual neural connectivity than structural measures (GMV). In both datasets, metabolic and functional correlations showed significantly higher similarity to resting-state networks than structural covariance did.
  • Validation of Group Studies: The study found that inter-individual and intra-individual patterns are remarkably similar. This validates the common practice of using group-level correlations to infer brain organization, but with the caveat that these group patterns are a "mixed bag" of aging, genetics, and transient states.

Key Results

  • Simon Dataset: Intra-individual ReHo correlations matched resting-state connectivity ($\rho = 0.39$) better than GMV correlations did ($\rho = 0.29$).
  • ADNI Dataset: Replicated the findings in an older cohort. Intra-individual GMV correlations with resting-state connectivity were $\rho = 0.41$, while inter-individual were $\rho = 0.37$.
  • Metabolic Strength: FDG-PET showed the strongest results, with intra-individual metabolic correlations reaching a high similarity ($\rho = 0.71$) with functional connectivity.
  • PCA Clustering: The PCA analysis showed that all functional and metabolic measures (whether intra- or inter-individual) clustered together, while structural measures formed a distinct, separate group.

Limitations and Future Work

The authors acknowledge that their datasets cover specific age ranges (the Simon dataset is mid-life; ADNI is elderly). It remains to be seen if these "aging vs. state" dynamics hold true during the rapid brain development of childhood or adolescence.

Additionally, while the study compared structure and function, it did not include white matter tractography (DTI), which would provide a more direct "road map" of physical connections. Future work could integrate DTI to see if structural covariance aligns better with physical wiring or functional "traffic." Finally, the authors suggest that future inter-individual studies should more strictly control for age and sex to improve the interpretability of their "connectivity" findings.

Summary updated

Abstract

Inferring brain connectivity from inter-individual correlations has been applied across various neuroimaging modalities, including positron emission tomography (PET) and MRI. The variability underlying these inter-individual correlations is generally attributed to factors such as genetics, life experiences, and long-term influences like aging. This study leveraged two unique longitudinal datasets to examine intra-individual correlations of structural and functional brain measures across an extended time span. By focusing on intra-individual correlations, we aimed to minimize individual differences and investigate how aging and state-like effects contribute to brain connectivity patterns. Additionally, we compared intra-individual correlations with inter-individual correlations to better understand their relationship. In the first dataset, which included repeated scans from a single individual over 15 years, we found that intra-individual correlations in both regional homogeneity (ReHo) during resting-state and gray matter volumes (GMV) from structural MRI closely resembled resting-state functional connectivity. However, ReHo correlations were primarily driven by state-like variability, whereas GMV correlations were mainly influenced by aging. The second dataset, comprising multiple participants with longitudinal Fludeoxyglucose (18 F) FDG-PET and MRI scans, replicated these findings. Both intra- and inter-individual correlations were strongly associated with resting-state functional connectivity, with functional measures (i.e., ReHo and FDG-PET) exhibiting greater similarity to resting-state connectivity than structural measures. This study demonstrated that controlling for various factors can enhance the interpretability of brain correlation structures. While inter- and intra-individual correlation patterns showed similarities, accounting for additional variables may improve our understanding of inter-individual connectivity measures.
BibTeX
@inproceedings{di2025comparing,
  author = {Xin Di and Bharat B. Biswal and Alzheimer’s Disease Neuroimaging Initiative},
  title = {{Comparing intra- and inter-individual correlational brain connectivity from functional and structural neuroimaging data}},
  booktitle = {Brain},
  year = {2025},
  doi = {10.1007/s00429-025-02972-y},
}
Figure