Statistical inference of dynamic resting-state functional connectivity using hierarchical observation modeling.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Systems Neuroscience

Background:

• Spontaneous fluctuations in blood-oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) signals reveal synchronous activity between functionally related brain regions, forming functional networks.
• Current analysis techniques often assume static functional connectivity, which may not accurately represent dynamic changes, particularly in neurological conditions.
• Dynamic connectivity analysis aims to capture moment-to-moment variations in functional connections during an imaging session.

Purpose of the Study:

• To propose and validate a novel hierarchical observation modeling framework for statistical inference of dynamic functional connectivity.
• To assess the sensitivity, specificity, and reproducibility of the proposed method compared to sliding-window correlation analysis.

Main Methods:

• Developed a two-level linear model utilizing overlapping sliding windows of fMRI signals, accounting for the non-independence of consecutive windows.
• Synthesized fMRI datasets with constant and externally modulated functional connectivity to rigorously test the proposed framework.
• Compared the novel method against traditional sliding-window correlation analysis.

Main Results:

• The proposed method successfully identified statistically significant functional connections modulated by external input.
• Demonstrated superior sensitivity and specificity in detecting variable functional connectivity compared to sliding-window correlation.
• Real data analysis showed improved reproducibility and more discriminative estimation of dynamic connectivity.

Conclusions:

• The novel hierarchical observation modeling approach provides a statistically robust framework for inferring dynamic functional connectivity from fMRI data.
• This method offers enhanced accuracy, sensitivity, and reproducibility for studying time-varying brain network interactions, crucial for understanding neurological disorders.