Estimating dynamic brain functional networks using multi-subject fMRI data.

Area of Science:

• Neuroscience
• Data Science
• Computational Biology

Background:

• Brain functional connectivity is often assumed stationary, but dynamic variations are increasingly recognized.
• Analyzing these dynamic changes is crucial for understanding brain function.
• Existing dynamic connectivity approaches face challenges in detecting frequent, unknown change points.

Purpose of the Study:

• To develop a fully automated, two-stage approach for detecting rapid changes in brain functional connectivity.
• To pool information across subjects for robust change point estimation.
• To estimate brain networks within distinct dynamic states.

Main Methods:

• A fused lasso approach models time-dependent connectivity to estimate unknown change points in stage one.
• Sparse inverse covariance matrices are used to infer brain networks in each state phase in stage two.
• Extensive simulation studies and application to fMRI data were used for validation.

Main Results:

• The proposed method demonstrates improved power to detect rapid connectivity changes compared to existing approaches.
• The automated approach effectively identifies the number and location of change points.
• Successful application to saccade block task fMRI data.

Conclusions:

• The developed two-stage method provides a powerful tool for analyzing dynamic brain functional connectivity.
• This approach advances the understanding of time-varying brain organization.
• It offers a robust framework for identifying distinct network states in neuroimaging data.