NHSMM-MAR-sdNC: A novel data-driven computational framework for state-dependent effective connectivity analysis

Area of Science:

• Neuroscience
• Computational Neuroscience
• Data Science

Background:

• The brain exhibits intrinsic dynamics, including metastability, crucial for functional integration and segregation.
• Dynamic effective connectivity is vital for understanding brain metastability due to its causal inference capabilities.
• Existing methods for state-dependent effective connectivity are limited, necessitating new approaches.

Purpose of the Study:

• To propose a novel, data-driven computational framework for investigating state-dependent effective connectivity.
• To overcome limitations of existing models by allowing adaptive state number inference and explicit state duration estimation.
• To provide a tool for exploring brain metastability and itinerant dynamics.

Main Methods:

• Introduced NHSMM-MAR-sdNC: a nonparametric hidden semi-Markov model combined with a multivariate autoregressive model and state-dependent new causality.
• The framework is data-driven and free from biological assumptions.
• State number and duration distributions are directly inferred from observed data.

Main Results:

• Successfully identified the number of states adaptively and accurately mapped state-dependent causality networks using synthetic data.
• Revealed the dynamics of state-related causality networks in real-world resting-state fMRI data.
• Demonstrated the framework's capability to analyze complex brain dynamics.

Conclusions:

• The proposed NHSMM-MAR-sdNC framework offers a novel computational approach for identifying state-dependent effective connectivity.
• This method facilitates the identification and assessment of brain metastability and itinerant dynamics.
• The framework provides a powerful tool for advancing neuroscience research.