A novel method for sparse dynamic functional connectivity analysis from resting-state fMRI

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Modeling

Background:

• Understanding dynamic functional connectivity (DFC) in brain networks is crucial.
• Estimating temporal dynamics from resting-state fMRI (rs-fMRI) is challenging with current methods.

Purpose of the Study:

• To develop a novel computational framework for sparse DFC analysis in high-dimensional rs-fMRI data.
• To overcome limitations of existing dynamic functional connectivity estimation methods.

Main Methods:

• Proposed HDP-HSMM-BPCA model: a temporal extension of probabilistic principal component analysis.
• Utilizes a hierarchical Dirichlet process (HDP) prior to overcome standard HMM limitations.
• Bayesian nonparametric hidden semi-Markov model (HSMM) automatically infers latent space dimensionality.

Main Results:

• HDP-HSMM-BPCA demonstrated superior estimation accuracy on synthetic data compared to existing models.
• Application to real rs-fMRI data revealed time-varying structures and sparse DFC patterns.
• The model effectively discovers underlying temporal structures in rs-fMRI data.

Conclusions:

• A new computational framework for sparse DFC analysis is presented.
• This framework facilitates the discovery of underlying temporal structures in rs-fMRI data.
• Enhances the study of brain functional connectivity dynamics.