LOCAL SPATIAL FLOWS AND PROPAGATIVE ATTRACTORS: A NOVEL “FLOWNECTOME” FRAMEWORK FOR ANALYZING BOLD FMRI DYNAMICS

Robyn L. Miller, Victor M Vergara, Erik B. Erhardt, Vince D. CalhounbioRxiv 2024 · DOI

TL;DR

This paper introduces a novel "flownectome" framework to analyze spatial signal propagation in fMRI data, moving beyond traditional temporal correlation analyses. This new approach reveals structured patterns of local directional flows and propagative attractors, offering a fresh perspective on brain dynamics and identifying significant differences in these patterns in individuals with schizophrenia.

What Problem Does This Paper Address?

Traditional fMRI analysis primarily focuses on temporal signal fluctuations and correlations between brain regions. While this has been successful, it largely overlooks the spatial dynamics of how signals propagate within and between brain areas. The authors argue that this neglect is a "blind spot" limiting our understanding of how local spatial flows support or impede brain function. They aim to address this by developing a method to empirically measure and analyze these local directional flows and their larger-scale propagative consequences, creating a more comprehensive view of brain activity.

Methods

The core of the paper introduces the "flownectome" framework, which involves several key steps:

  • Spatiotemporal Gradients (STGs): The authors compute directional derivatives (gradients) of the BOLD signal in x, y, and z dimensions at each voxel and timepoint. This captures the instantaneous direction and magnitude of signal change.
  • Gradient-Driven Propagative Paths: Using these gradients, they simulate "paths" that trace the direction of signal flow over short time windows (15 TRs). These paths are initialized at every voxel and follow the direction of the strongest gradient.
  • Propagative Path Density (PPD) Maps: By aggregating these paths across all voxels within a time window, they create PPD maps, which represent areas of concentrated spatial signal inflow or "propagative energy."
  • Group ICA on PPD Maps (pdGICA): They apply Independent Component Analysis (ICA) to these time-varying PPD maps to identify recurring patterns of spatial signal propagation across subjects.
  • Spatio-Functional Connectivity: They then correlate the timecourses of these PPD components with functional network activity (derived from standard fMRI analysis) to understand how spatial propagation relates to functional network dynamics.

Datasets

The study utilizes resting-state fMRI data from a cohort of 314 individuals, comprising approximately equal numbers of schizophrenia patients (SZs) and healthy controls (HCs). The data was preprocessed using a standard pipeline with additional spatial and temporal smoothing. No new benchmarks were introduced; the focus is on a novel analysis of existing fMRI data.

Key Insights and Ideas

  • Spatial Flows are Structured: Local directional flows and propagative path densities exhibit structured patterns across the brain, often with plausible symmetries. This suggests that signal propagation isn't random but follows organized pathways.
  • Schizophrenia Alters Spatial Dynamics: The study reveals significant differences in these spatial flow patterns between individuals with schizophrenia and healthy controls. This highlights that disruptions in spatial signal propagation might be a key characteristic of the disorder.
  • Propagative Inflow Correlates with Activation: There's a positive correlation between the influx of spatial signal (propagative path density) into a functional region and that region's contribution to overall signal reconstruction (net gain in activation). This supports the idea that spatial inflow fuels local activity.
  • "Flownectome" Captures Novel Brain Dynamics: The novel "flownectome" framework, by analyzing spatial propagation, reveals patterns that are distinct from traditional functional connectivity and offers a new lens through which to view brain organization and dysfunction.

Key Results

  • Directional Flows Vary by Region: Different brain regions show distinct patterns of local spatial flow strength. For example, occipital and temporal lobes tend to have stronger local flows.
  • Schizophrenia Impacts Specific Regions: Schizophrenia significantly alters these flow patterns, often elevating them in frontal lobes and diminishing them in the hippocampus for x-directional flows, and reducing them in the cerebellum for y-directional flows.
  • PPD Maps Reveal Concentration Areas: Propagative path density maps highlight areas where spatial signal tends to accumulate, particularly in posterior brain regions.
  • Schizophrenia Elevates PPD in Certain Areas: In schizophrenia patients, there's elevated propagative density in frontal areas and default mode networks, suggesting altered spatial signal attraction.
  • Spatio-Functional Connectivity Differences: The relationships between spatial propagation patterns and functional network activity are significantly disrupted in schizophrenia, particularly involving regions like the thalamus and anterior cingulate cortex.

Limitations and Future Work

The authors acknowledge that their work is in its early stages and list several limitations and future directions:

  • Anatomical Artifacts: Better accounting for anatomical artifacts without hindering propagative flow analysis is needed.
  • Pipeline Optimization: Optimizing preprocessing and post-processing pipelines (e.g., window length for PPD maps) for sensitivity and robustness is crucial.
  • Spatial Resolution: Modulating the discretization of brain space (currently 3mm³) could allow for more flexible step sizes in propagation.
  • Directed Propagation: Investigating directed propagation between specific functional regions and time-varying "flow connectivity" through channels of propagative concentration is a key future step.
  • Clinical Biomarkers: The authors believe this framework offers a promising avenue for identifying more sensitive and condition-specific biomarkers for neuropsychiatric disorders.

Summary updated

Abstract

Although the analysis of temporal signal fluctuations and co-fluctuations has long been a fixture of blood oxygenation-level dependent (BOLD) functional magnetic resonance imaging (fMRI) research, the role of local directional flows in both signal propagation and healthy functional integration remains almost entirely neglected. We are introducing an extensible framework, based on localized directional signal flows, to capture and analyze spatial signal propagation and propagative attractor patterns in BOLD fMRI. Novel features derived from this approach are validated in a large resting-state fMRI schizophrenia study where they reveal significant relationships between spatially directional flows, propagative attractor patterns and subject diagnostic status. Plausibly, we find that spatial signal inflow to functional regions tends to positively correlate with net gain/loss in the region′s temporal contribution BOLD signal reconstruction. We also find that group ICA (pdGICA) performed on time-varying propagative density maps, which are whole-brain maps of spatial signal inflow to each voxel on successive 30 second windows (ie. propagative attractor maps), produce components that tend to concentrate predominantly in at most six or seven functional regions, in some cases focusing on as few as two. The relationship between the propagative attractor group ICA component maps and functional regions is not sharp, but is focused enough to render the pdGICA component maps functionally tractable. Temporal correlations between pdGICA component timeseries and net gain/loss functional region timeseries on corresponding windows echoes those traditionally observed between functional region timeseries when aligning each pdGICA component with the functional region with which it has greatest spatial overlap. Schizophrenia strongly disrupts the average correlative relationship between pdGICA components and certain functional regions, some of which tend to be implicated in schizophrenia e.g. the thalamus and the anterior cingulate cortex. Schizophrenia also strongly and pervasively linked to the importance of specific pdGICA components in reconstructing subjects′ observed time-varying propagative density maps. Over half of the 35 pdGICA component make significantly different average contributions to patient propagative density maps than to those of controls, with the functional footprints of impacted pdGICA components spreading over diverse functional domains. Finally, the magnitudes of local directional flows that carry propagation have spatially structured averages and structured, pervasive schizophrenia effects. The framework introduced here follows a new and fundamentally different data-driven approach to the BOLD fMRI signal. We believe that the empirical measurement of local directional flows and wider spatial signal propagation opens a plethora of new avenues through which to investigate healthy and disordered brain function using BOLD fMRI.
BibTeX
@misc{miller2024local,
  author = {Robyn L. Miller and Victor M Vergara and Erik B. Erhardt and Vince D. Calhoun},
  title = {{LOCAL SPATIAL FLOWS AND PROPAGATIVE ATTRACTORS: A NOVEL “FLOWNECTOME” FRAMEWORK FOR ANALYZING BOLD FMRI DYNAMICS}},
  year = {2024},
  doi = {10.1101/2024.01.26.576775},
  howpublished = {bioRxiv},
}
Figure