Dynamic effective connectivity

Area of Science:

• Computational neuroscience
• Neuroimaging analysis
• Dynamical systems theory

Background:

• Metastability and itinerant dynamics, characterized by spontaneous neuronal activity reorganization, are crucial in brain function.
• Dynamic functional connectivity (DFC) analyses using fMRI describe these patterns but lack mechanistic insight into neuronal underpinnings.
• Existing methods are descriptive and model-free, limiting the understanding of the neural basis of metastability.

Purpose of the Study:

• To propose effective connectivity (EC) analyses as a more suitable approach for investigating the neuronal basis of metastability.
• To introduce a probabilistic, generative model for hemodynamic fluctuations based on biologically-grounded models and dynamical systems theory.
• To establish the face validity of this novel model for analyzing non-stationary fMRI data and understanding brain dynamics.

Main Methods:

• Development of a generative model extending spectral Dynamic Causal Modelling (DCM) to simulate time-varying functional connectivity.
• Utilizing variational Bayes to recover key model parameters, including transition probabilities between connectivity states.
• Employing Bayesian model comparison for characterizing simulated data within and between subjects.

Main Results:

• The model successfully generates non-stationary fMRI time series that reflect underlying changes in effective connectivity states.
• Key parameters, such as transition probabilities and the nature of connectivity states, were accurately recovered from simulated data.
• The framework demonstrates face validity in capturing the complex, time-varying nature of brain functional architectures.

Conclusions:

• Effective connectivity modeling, particularly within a Bayesian framework, offers a powerful approach to uncover the neural mechanisms of brain metastability.
• The proposed generative model provides a method to analyze itinerant dynamics by characterizing transitions between distinct brain states.
• This framework is adaptable for studying metastability and itinerant dynamics in various non-stationary time series beyond fMRI data.