Editorial: Advancing the measurement, interpretation, and validation of dynamic functional connectivity

David O’Connor, Catie Chang, Georgios D. MitsisFrontiers in Human Neuroscience 2023 · DOI · OpenAlex

Abstract

Dynamic functional connectivity (dFC) has emerged as a powerful framework for understanding the brain's temporal reconfigurations. This editorial reviews a collection of studies that advance the field by validating dFC through clinical interventions, assessing its reliability across imaging parameters, and improving computational workflows. Key findings include the utility of dFC in tracking stroke recovery following TMS, the identification of circuit-level dynamics via invasive recordings, and the development of high-efficiency pipelines for large-scale data analysis. Together, these contributions bridge the gap between fleeting neural signals and stable behavioral phenotypes.

Introduction

The human brain is a highly adaptive system, characterized by functional networks that reconfigure over seconds to support diverse cognitive demands. While traditional functional connectivity (FC) measures the average synchronization between regions over several minutes, dynamic functional connectivity (dFC) seeks to capture these transient states. Despite its potential, dFC has faced skepticism regarding whether observed fluctuations represent true neural signals or statistical artifacts. This editorial synthesizes recent research aimed at validating dFC metrics, enhancing their reliability, and optimizing the computational methods required to process high-dimensional connectivity data.

Clinical Validation and Causal Interventions

A primary challenge in dFC research is establishing a causal link between connectivity changes and clinical outcomes. Qin et al. (2021) addressed this by using transcranial magnetic stimulation (TMS) as a therapeutic intervention for stroke patients. By combining fMRI with Independent Component Analysis (ICA) and sliding-window dFC, they demonstrated that TMS-based rehabilitation partially restored connectivity between sensorimotor and cognitive control domains. Similarly, Lee et al. (2022) showed that intensive cognitive control training induced network integration changes that mirror natural adolescent development. These studies provide crucial evidence that dFC can serve as a biologically grounded marker for neuroplasticity and recovery.

Methodological Reliability and Acquisition Parameters

The transition of dFC from research to clinical settings requires standardized, reliable protocols. Cahart et al. (2022) evaluated the test-retest reliability of Leading Eigenvector Dynamics Analysis (LEiDA) in an aging population. Their investigation into multiband fMRI acquisition—which allows for sub-second sampling—highlighted the trade-offs between temporal resolution and signal-to-noise ratio. Such methodological studies are essential for determining how acquisition parameters influence the stability and interpretability of dynamic brain states.

Multi-scale Dynamics: From Circuits to Whole-Brain

Validation of dFC also benefits from cross-scale investigations. Keogh et al. (2022) utilized deep brain stimulation (DBS) in patients with neuropathic pain to observe short-term activation and coherence patterns within the dorsal anterior cingulate cortex (dACC). Their results showed that these local dynamics are modulated during learning tasks. In parallel, animal models provide a "ground truth" for these observations; Wang et al. (2022) recorded local field potentials in mice, demonstrating circuit-level connectivity shifts in the prefrontal cortex and thalamus during cognitive tasks. These findings suggest that the macroscopic fluctuations captured by fMRI have a solid foundation in local neuronal population dynamics.

Computational Efficiency and Interpretation

The high dimensionality of dFC data poses significant computational challenges. Sendi et al. (2022) introduced a two-stage pipeline for deriving connectivity states that is 25 times faster than traditional methods, enabling the analysis of much larger datasets without a loss in accuracy. To improve interpretability, Gonzalez-Castillo et al. explored low-dimensional embedding techniques, which compress complex connectivity patterns into more manageable forms. Additionally, Brandman et al. (2022) emphasized the importance of using naturalistic stimuli to bridge the gap between lab-based findings and real-world brain states.

Conclusion

The research highlighted in this editorial demonstrates significant progress in the measurement and validation of dynamic functional connectivity. By integrating multimodal imaging, clinical interventions, and advanced computational techniques, the field is moving toward a more robust understanding of how the brain's temporal dynamics underpin human cognition and behavior. Future research must continue to refine these methods to ensure their translational utility in clinical and real-world settings.

Summary updated

Abstract

The work by Qin et al 2021 highlights the utility of dFC methods in clinical research and demonstrates the causal effects of a clinical intervention on dFC [3]. In this study, multimodal experimentation methods, in the form of transcranial magnetic stimulation (TMS) and functional magnetic resonance imaging (fMRI), were used as a therapeutic intervention and to assess recovery from stroke. A combination of ICA for brain region definition, sliding windows for dFC calculation, and K Means clustering for state definition, were used to generate metrics for assessing stroke recovery. It was shown that partial recovery of connectivity between sensorimotor and cognitive control domains, as assessed by resting-state fMRI (rsfMRI) before and after the intervention (1-and 3-months post-stroke), was achieved by TMS based rehabilitation. Overall, the use of a causal intervention step suggests that disease recovery, in this case stroke, can potentially be assessed using dFC methods. Lee et al. 2022 present another intervention study showing changes in rsfMRI networks induced by cognitive control training [4]. Here, a six-week period of intensive cognitive training resulted in increased integration between functional brain networks, mirroring natural developmental changes in network organization across adolescence. This study further supports the utility of resting-state connectivity measures for assessing functional reconfiguration in brain networks.An important methodological question is the extent to which dFC measures are reliable across experimental procedures and data acquisition parameters. Cahart et al. 2022 investigated the test-retest reliability of single and multiband rsfMRI in healthy older adults using a dynamic analysis technique termed LEiDA [5]. Some prior work has shown that LEiDA is sensitive to cognitive ability changes in an aging cohort [6]. The work of Cahart et al. sought to assess the test-retest reliability of this analytical method across different fMRI acquisitions with varying levels of multiband acceleration. Multiband acquisition is highly relevant to dFC studies, as it can reduce the repetition time of fMRI from ~2 seconds down to below 1 second, potentially enabling dynamic cognitive processes to be more clearly resolved (albeit with complex tradeoffs in terms of signalto-noise ratio). Thus, studies that investigate the reproducibility of dynamic analysis methods across acquisition parameters can provide valuable information for applying and interpreting the results of dFC.The investigation of dFC need not be performed at the whole brain level. For instance, a study by Keogh et al. 2022 placed deep brain stimulation electrodes in 3 individuals with medically and surgically refractory neuropathic pain, centralized on the dorsal anterior cingulate cortex (dACC) and investigated short time activation and coherence patterns of the neuronal populations within the dACC [7]. In addition to revealing the characteristics of information flow within the ACC during a decision-making task, their results showed that these network dynamics were altered during learning. This study, along with Wang et al. 2022, which recorded local field potentials in mice from mPFC and mediodorsal thalamus [8], show that dynamic connectivity is altered at the circuit level during cognitive processes. These studies provide a concrete demonstration that specific short time activation and connectivity profiles manifest in response to certain behaviors. How these changes are reflected in more widely applicable macroscopic measurement modalities, however, remains to be determined.In addition to validation of brain imaging methods, attention must be paid to non-imaging phenotypes being estimated, and how different brain states can impact these measurements. In this context, the work of Brandman et al. 2022 is relevant; this study does not use brain imaging, but rather retrospective behavioral sampling to show how brain states can be manipulated over a short time period using naturalistic stimuli [9]. More grounded task paradigms are crucial for shifting findings obtained in a lab setting into the real world in a truly translational manner.A key challenge in analyzing whole brain dFC is the size of the derived data. This can yield challenges both in terms of computational demands and result interpretation. In work addressing the former issue, Sendi et al. 2022 constructed a two-stage dFC pipeline for deriving clusters (representing connectivity states) from large datasets [10]. Crucially, this pipeline was found to be 25 times faster than the conventional procedure of generating connectivity states, while producing states that are highly similar to those of the conventional approach.Finally, the study of Gonzalez-Castillo et al. addressed dFC interpretability issues, by investigating the use of low-dimensional embedding techniques, which can capture the variance observed in high dimensional methods but render it in a form that is more computationally feasible and more easily interpreted [not yet published].Collectively, the above studies highlight complementary approaches for relating dynamic measurements of brain connectivity to cognition and behavior, and across spatial and temporal scales. Further, they probe key methodological questions surrounding dFC analyses, such as computational efficiency and reliability across acquisition parameters. Overall, their findings shed new light on factors shaping dynamic brain connectivity and provide directions for future research.
BibTeX
@inproceedings{oconnor2023editorial,
  author = {David O’Connor and Catie Chang and Georgios D. Mitsis},
  title = {{Editorial: Advancing the measurement, interpretation, and validation of dynamic functional connectivity}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2023},
  doi = {10.3389/fnhum.2023.1206098},
}
Figure