Abstract
The brain is a complex, multiscale dynamical system composed of many interacting regions. Knowledge of the spatiotemporal organization of these interactions is critical for establishing a solid understanding of the brain's functional architecture and the relationship between neural dynamics and cognition in health and disease. The possibility of studying these dynamics through neuroimaging data has catalyzed interest in time-resolved fluctuations in functional connectivity, often referred to as "dynamic" or time-varying functional connectivity (TVFC). However, debates regarding statistical validity, physiological origins, and cognitive relevance complicate interpretation. This article reviews the current literature, introduces core concepts, summarizes controversies, and presents a forward-looking perspective on how resting TVFC analyses can be rigorously and productively applied to investigate questions in cognitive and systems neuroscience.
Introduction: Beyond Static Connectivity
For decades, functional MRI (fMRI) research primarily focused on "static" connectivity, which involves averaging brain interactions over the duration of a 5-to-10-minute scan. However, the brain is inherently a multiscale dynamical system that does not remain in a single state. Time-Varying Functional Connectivity (TVFC) aims to capture how these neural "conversations" shift over shorter timescales, typically measured in seconds.
The transition from static to time-varying analysis has introduced significant challenges. fMRI data is notoriously noisy, leading critics to argue that observed "dynamics" may simply be sampling variability or artifacts caused by head motion, respiration, or changes in arousal (e.g., drowsiness). This review addresses the lack of consensus on terminology and the biological origins of these fluctuations.
The Rapid Expansion of TVFC Research
The field has seen an explosion of interest in TVFC methods over the last decade. This growth necessitates a critical evaluation of the tools used to characterize brain dynamics and the establishment of rigorous standards to separate true neural signals from statistical noise.
Methodological Frameworks and the Role of Null Models
Current TVFC methods generally fall into two categories:
- Data-Driven Approaches: These focus on observed BOLD signals using techniques like sliding windows (calculating correlation over moving time segments), state-based models (e.g., Hidden Markov Models to identify discrete brain states), and time-frequency analysis (using wavelets).
- Model-Based Approaches: These utilize biophysical models to simulate neural populations, which are then "inverted" to see if simulated dynamics match empirical fMRI data.
A critical methodological pillar is the use of Null Models. To claim a brain is "dynamic," researchers must demonstrate that the data differs significantly from a "stationary" surrogate—a synthetic dataset with the same mean and variance but no true time-varying interactions.
Key Insights: Distinguishing Signal from Noise
A primary concern in TVFC research is the "Map vs. Territory" problem: the risk of confusing a statistical method's output with underlying neural reality. Key findings from the literature include:
- Arousal as a Signal and Confound: Drowsiness and "microsleeps" are major drivers of TVFC. While this is a nuisance for cognitive studies, it is a vital signal for research into vigilance and sleep.
- Biological Validation: Cross-modal studies (e.g., fMRI combined with EEG or calcium imaging in animals) provide strong evidence that TVFC tracks actual electrical firing in neurons.
- Clinical and Behavioral Relevance: TVFC has proven to be a sensitive marker for conditions like schizophrenia, autism, and Alzheimer’s disease, and can predict behavioral performance or "mind-wandering" states.
- Reliability: Despite noise, TVFC patterns are remarkably stable within individuals over time, potentially serving as a "neural fingerprint."
Limitations and Future Directions
The field must address several hurdles to mature:
- Terminology: Moving away from the ambiguous term "dynamic" toward "Time-Varying Functional Connectivity" (TVFC) to avoid mathematical confusion.
- Preprocessing Sensitivity: TVFC is highly sensitive to motion. Aggressive cleaning, such as Global Signal Regression (GSR) and "scrubbing" (deleting high-motion frames), is essential.
- Causal Gaps: Most research remains correlational. Future studies should employ causal manipulations, such as Transcranial Magnetic Stimulation (TMS), to link connectivity changes directly to behavior.
- Standardization: Establishing a consensus pipeline is necessary to ensure results are comparable across different research groups.
Summary updated