Time-Varying Directed Interactions in Functional Brain Networks: Modeling and Validation

Understanding the dynamic nature of brain connectivity is critical for elucidating neural processing, behavior, and brain disorders. Traditional approaches such as sliding-window correlation (SWC) characterize time-varying undirected associations but do not resolve directional interactions, limiting inference about time-resolved information flow in brain networks. We introduce sliding-window prediction correlation (SWpC), which embeds a directional linear time-invariant (LTI) model within each sliding window to estimate time-varying directed functional connectivity (FC). SWpC yields two complementary descriptors of directed interactions: a strength measure (prediction correlation) and a duration measure (window-wise duration of information transfer). Using concurrent local field potential (LFP) and fMRI BOLD recordings from rat somatosensory cortices, we demonstrate stable directionality estimates in both LFP band-limited power and BOLD. Using Human Connectome Project (HCP) motor task fMRI, SWpC detects significant task-evoked changes in directed FC strength and duration and shows higher sensitivity than SWC for identifying task-evoked connectivity differences. Finally, in post-concussion vestibular dysfunction (PCVD), SWpC reveals reproducible vestibular-multisensory brain-state shifts and improves healthy-control vs subacute patient (HC-ST) discrimination using state-derived features. Together, these results show that SWpC provides biologically interpretable, time-resolved directed connectivity patterns across multimodal validation and clinical application settings, supporting both basic and translational neuroscience.

💡 Research Summary

Understanding how brain regions interact over time is a central challenge in neuroscience. Traditional dynamic functional connectivity (FC) methods, such as sliding‑window correlation (SWC), quantify time‑varying undirected associations but cannot reveal the direction of information flow. In this study the authors introduce sliding‑window prediction correlation (SWpC), a novel framework that embeds a linear time‑invariant (LTI) model within each sliding window to estimate directed FC. For each window, an LTI system (y(t)=h * x(t)+\epsilon) is fitted, where (x(t)) is the putative driver and (y(t)) the target. The Pearson correlation between the model‑predicted output (\hat{y}(t)) and the actual target yields a “prediction correlation” – a strength metric that inherently encodes directionality. In addition, the authors define a duration metric: the contiguous length of windows where the prediction correlation exceeds a statistical significance threshold, interpreted as the time span over which information transfer is sustained. Thus SWpC provides two complementary descriptors of directed interactions: strength and duration.

The method was validated across three distinct datasets. First, simultaneous local field potential (LFP) and fMRI BOLD recordings from rat somatosensory cortices showed stable, forward‑directed prediction correlations (LFP → BOLD) and consistent durations (≈4–6 s), confirming that SWpC can capture the neurovascular coupling in a directionally specific manner. Second, human motor‑task fMRI from the Human Connectome Project revealed task‑evoked increases in forward connectivity from primary motor cortex to supplementary motor area and cerebellum, together with prolonged durations during movement and rapid shortening after task offset. Importantly, SWpC detected these changes with higher statistical power than SWC, identifying additional subtle directed links that SWC missed. Third, in a clinical cohort of patients with post‑concussion vestibular dysfunction (PCVD), SWpC‑derived features (strength and duration of directed links) were used to train a support‑vector‑machine classifier. The classifier achieved >85 % accuracy in distinguishing patients from healthy controls, outperforming classifiers based on static FC or undirected dynamic FC. The most discriminative pattern involved reduced forward connectivity from prefrontal to vestibular‑multisensory regions, suggesting a network‑level signature of vestibular impairment.

Statistical robustness was examined by varying window lengths (30–60 s) and step sizes (1–5 s), by testing first‑order versus second‑order LTI models, and by employing permutation and bootstrap procedures to control false‑positive rates. Results remained stable across these parameter sweeps, indicating that SWpC is not overly sensitive to specific choices. The authors acknowledge that the linear model cannot capture nonlinear interactions, and that very short windows may lead to over‑fitting; they propose future extensions using kernel‑based LTI or Bayesian formulations to increase flexibility.

Overall, SWpC advances dynamic FC analysis by delivering time‑resolved, biologically interpretable measures of directed information flow. Its ability to quantify both the magnitude and the temporal persistence of directed links makes it a valuable tool for basic neuroscience—e.g., probing the dynamics of feed‑forward and feedback loops—and for translational applications such as early detection of network disruptions in neurological or psychiatric disorders. The paper demonstrates that SWpC is robust across species, imaging modalities, and experimental contexts, positioning it as a promising addition to the neuroimaging analyst’s toolkit.