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.
The human brain is a highly dynamic system, where functional connectivity (FC) evolves over time (Allen et al., 2012;Hutchison et al., 2013) and can exhibit asymmetric, directed interactions across brain regions (Friston, 2011;Mitra et al., 2015;Xu et al., 2021). Capturing these temporal and directional dynamics is critical for understanding neural processing, behavior, and the mechanisms underlying brain disorders. Traditional FC methods, such as sliding window correlation (SWC), have been invaluable for investigating temporal variability in functional magnetic resonance imaging (fMRI) data (Allen et al., 2012;Shakil et al., 2016). By segmenting time series data into overlapping windows, SWC enables researchers to explore how connectivity fluctuates over time, providing critical insights into brain network dynamics associated with cognition and disease. However, SWC is inherently limited to correlational analyses and lacks the ability to infer directionality in information flow. These limitations constrain its utility in characterizing hierarchical neural interactions, which are essential for understanding the mechanistic basis of both normal brain function and clinical conditions, such as post-concussion vestibular dysfunction (PCVD) (Smith et al., 2021).
In neuroimaging literature, methods for studying causality and correlation have long been viewed as distinct and divided, reflecting the differing goals and assumptions underlying each approach (Aedo-Jury et al., 2020;Friston, 2011;Friston et al., 2003;Keilholz et al., 2013;Liang et al., 2015;Park et al., 2018;van den Heuvel and Hulshoff Pol, 2010). Correlational approaches, such as SWC, focus on statistical dependencies without inferring directional influence, whereas causal models aim to uncover mechanistic interactions driving these dependencies. This conceptual separation arises from different goals: correlational methods measure symmetric statistical dependence without directionality, whereas causal models focus on asymmetric influence by distinguishing cause from effect (Peters et al., 2017). Recent work has advanced time-varying effectiveconnectivity modeling for EEG/MEG, often demonstrated in relatively low-to moderatedimensional channel/source settings, e.g., (Medrano et al., n.d.). Complementary developments are still needed to enable scalable, whole-brain, time-resolved modeling for resting-state fMRI, where hemodynamic filtering and high dimensionality impose distinct constraints. This practical gap has constrained efforts to integrate correlational and directed perspectives within a unified, time-resolved framework.
To bridge this gap, we propose sliding window prediction correlation (SWpC), a computational framework that embeds a validated causal linear time-invariant (LTI) model (Xu et al., 2017) within each sliding window. For each window and each direction 𝑥 → 𝑦 , SWpC predicts 𝑦 from 𝑥 via an LTI impulse response, then quantifies directed strength using prediction correlation: the correlation between predicted and observed 𝑦 in that window. In parallel, SWpC estimates a duration of information transfer as the window-specific impulse-response support (chosen by model selection), providing a time-resolved measure of how temporally extended the influence of 𝑥 on 𝑦 is. Importantly, strength and duration need not covary: interactions can be brief but strong, or weaker yet temporally sustained. Together, these complementary measures allow SWpC to characterize time-varying directional interactions while retaining the practical slidingwindow workflow used in dynamic FC studies.
Validation of directionality estimates from BOLD fMRI is particularly critical due to the variability in hemodynamic responses across brain regions. While prior studies have validated direction estimation using simulated data (Smith et al., 2011;Xu et al., 2017), the inherent limitations of simulations, including oversimplified assumptions, underscore the need for validation with biologically relevant data. To address this challenge, we leverage a diverse range of multimodal neuroimaging datasets, including concurrent LFP-BOLD recordings in rodents, human task-based fMRI, and patient fMRI data, which serve as robust benchmarks for testing the reliability and sensitivity of the proposed model.
Using high-content LFP-BOLD datasets, SWpC demonstrated its ability to reliably estimate directed interactions at both neuronal and BOLD levels, extending previous SWC results by adding directionality in strength and duration. Specifically, SWpC strength and duration were consistently estimated in both BLP and BOLD signals from S1L and S1R, exhibiting symmetries between the two regions, as their directional asymmetry consistently remained below scan variability.
Additionally, SWpC strength reflects LFP-BOLD correlations akin to those uncovered by SWC, with high BLP-BOLD correlates observed within the θ, low β, and upper bands (Garth John Thompson et al., 2013a), while SWpC dura
This content is AI-processed based on open access ArXiv data.