Estimating Dynamic Functional Brain Connectivity With a Sparse Hidden Markov Model.

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Dynamic functional network connectivity (dFNC) from fMRI reveals brain organization and development.
• Resting-state fMRI (rs-fMRI) captures spontaneous brain activity.
• Existing dFNC methods like sliding windows have limitations, such as fixed window sizes.

Purpose of the Study:

• To propose a novel sparse Hidden Markov Model (HMM) for estimating dFNC.
• To represent time-varying neural processes as discrete brain states with interpretable connectivity.
• To validate the model on simulated and real rs-fMRI data and compare dFNC patterns across age groups.

Main Methods:

• Developed a sparse HMM integrating Gaussian HMM and Gaussian Graphical Model (GGM).
• Utilized Expectation-Maximization (EM) and graphical least absolute shrinkage and selection operator (glasso) for optimization.
• Applied the model to simulated BOLD time series and rs-fMRI data from the Philadelphia Neurodevelopmental Cohort (PNC) study.

Main Results:

• The proposed sparse HMM effectively captures both stationary and abrupt brain activity fluctuations.
• Interpretable functional connectivity networks were derived by enforcing sparsity.
• Significant differences in spatial and temporal dFNC patterns were observed between children and young adults.

Conclusions:

• The sparse HMM provides a robust method for estimating dFNC and understanding brain dynamics.
• The findings offer insights into the developmental trajectory of brain connectivity across childhood.
• This approach can advance research in developmental neuroscience and mental illness studies.