Brain network dynamics fingerprints are resilient to data heterogeneity

Abstract

Large multi-site neuroimaging datasets have significantly advanced our quest to understand brain-behaviour relationships and to develop biomarkers of psychiatric and neurodegenerative disorders. Yet, such data collections come at a cost, as the inevitable differences across samples may lead to biased or erroneous conclusions. Previous work has investigated this critical issue in resting-state functional magnetic resonance imaging (rs-fMRI) data in terms of effects on static measures, such as functional connectivity and brain parcellations. Here, we depart from prior approaches and utilize dynamical models to examine how diverse scanning factors in multi-site fMRI recordings affect our ability to infer the brains spatiotemporal wandering between large-scale networks of activity. Building upon this premise, we first confirm the emergence of robust subject-specific dynamical patterns of brain activity. Next, we exploit these individual fingerprints to show that scanning sessions belonging to different sites and days tend to induce high variability, while other factors, such as the scanner manufacturer or the number of coils, affect the same metrics to a lesser extent. These results concurrently indicate that we can recover the unique trajectories of brain activity changes in each individual, but also that our ability to infer such patterns is affected by how, where and when we try to do so. Author summaryWe investigate the important issue of data heterogeneity in large multi-site data collections of brain activity recordings. At a time in which appraising the source of variability in large datasets is gaining increasing attention, this study provides a novel point of view based on data-driven dynamical models. By employing subject-specific signatures of brain network dynamics, we find that certain scanning factors significantly affect the quality of resting-state fMRI data. More specifically, we first validate the existence of subject-specific brain dynamics fingerprints. As a proof of concept, we show that dynamical states can be estimated reliably, even across different datasets. Finally, we assess which scanning factors, and to what extent, influence the variability of such fingerprints.