fMRI and MEG Fingerprints Diverge at the Individual Level

1. Problem Statement

While functional connectivity (FC) profiles derived from functional Magnetic Resonance Imaging (fMRI) and Magnetoencephalography (MEG) are known to offer complementary views of large-scale brain organization, their relationship at the individual level remains poorly understood.

• The Gap: Previous research has established reasonable correspondence between fMRI and MEG at the population-average level. However, it is unclear how individual variability (the unique "neural fingerprint" of a specific subject) maps across these two modalities.
• The Core Question: Do the unique patterns of connectivity that distinguish one individual from another in fMRI data correspond to the unique patterns found in MEG data? Furthermore, how much of this individual variability is driven by underlying structural anatomy versus unique functional dynamics?

2. Methodology

The study utilized the Cam-CAN dataset, a large-scale neuroimaging resource, to analyze 543 participants who underwent both resting-state fMRI and MEG scans.

• Neural Fingerprint Derivation: The authors extracted subject-level FC profiles from both modalities. These profiles served as "neural fingerprints" intended to capture stable, individual-specific traits.
• Validation of Stability: They first verified that these fingerprints were robust by testing for within-subject, cross-session consistency and by demonstrating their ability to predict age and cognitive traits.
• Quantifying Cross-Modal Correspondence: To measure the relationship between fMRI and MEG fingerprints, the study employed two primary analytical frameworks:
  1. Variance Partitioning: To decompose the shared variance between modalities and determine how much is explained by structural connectivity (SC) and age.
  2. Representational Similarity Analysis (RSA): Specifically using Linear Centred Kernel Alignment (CKA) to quantify the preservation of pairwise inter-subject similarity patterns. This metric assesses whether if Subject A is similar to Subject B in fMRI, they are also similar in MEG.

3. Key Contributions

• Large-Scale Cross-Modal Analysis: This is one of the first studies to systematically compare individual-level functional fingerprints across fMRI and MEG in a large cohort (N=543) using the same participants.
• Decoupling Structure and Function: The study explicitly isolates the contribution of structural connectivity (SC) to functional variability, distinguishing between variance driven by anatomy versus unique functional dynamics.
• Methodological Integration: The application of CKA and variance partitioning to neuroimaging fingerprints provides a rigorous framework for assessing the alignment of different neuroimaging modalities at the individual level.

4. Key Results

The analysis yielded two primary findings that challenge the assumption of high cross-modal similarity:

• Low Individual-Level Correspondence: Despite strong similarity in population-average FC maps, the neural fingerprints at the subject level showed low correspondence between fMRI and MEG.
  • This was evidenced by low cross-modal shared variance.
  • Crucially, the CKA scores were low, indicating that the pattern of how subjects differ from one another in fMRI does not predict how they differ in MEG.
• The Dominance of Structural Variability:
  • Age-Related Variance: Structural fingerprints accounted for the majority of age-related variance in functional fingerprints for both modalities. Structural connectivity almost entirely explained the age-related variance shared between fMRI and MEG.
  • Unique MEG Information: While MEG functional fingerprints contained unique information regarding cognitive traits that was not explained by structure, this unique information was not shared with fMRI. In other words, the "extra" functional insight provided by MEG regarding cognition is orthogonal to what fMRI captures.

5. Significance and Implications

• Redefining Cross-Modal Integration: The findings suggest that fMRI and MEG, while complementary at the group level, capture distinct dimensions of individual variability. Researchers cannot assume that an individual's "fingerprint" in one modality will translate directly to the other.
• The Role of Anatomy: The study highlights that much of the individual variation in functional connectivity (especially related to aging) is actually a reflection of underlying structural variability. Once structural differences are accounted for, the remaining functional variance in fMRI and MEG diverges significantly.
• Clinical and Cognitive Applications: For studies aiming to use neural fingerprints for biomarkers (e.g., predicting cognitive decline or disease progression), relying on a single modality may miss critical, modality-specific information. Specifically, MEG offers unique functional insights into cognition that are independent of both structure and fMRI signals.
• Interpretation of "Surprising Lack of Similarity": The paper concludes that the lack of similarity in how subjects vary between electrophysiology (MEG) and hemodynamics (fMRI) is a fundamental property of brain organization, rather than a methodological artifact. This necessitates a more nuanced approach when integrating multi-modal neuroimaging data for precision medicine.