R-fMRI reconstruction from k-t undersampled data using a subject-invariant dictionary model and VB-EM with nested...

Area of Science:

• Neuroimaging
• Biophysics
• Computational Neuroscience

Background:

• Resting-state functional magnetic resonance imaging (R-fMRI) provides crucial data on cerebral cortex functional networks.
• Achieving higher spatial resolution in R-fMRI typically involves trade-offs in scan speed using complex pulse sequences or k-space undersampling with signal priors.

Purpose of the Study:

• To develop a novel R-fMRI reconstruction framework to improve spatial resolution by undersampling in k-space and time.
• To introduce an advanced inference framework for R-fMRI reconstruction that provides uncertainty estimates.

Main Methods:

• Proposed a model-based R-fMRI reconstruction using a subject-invariant, spatially regularized dictionary prior.
• Developed a variational Bayesian expectation maximization with nested minorization (VB-EM-NM) inference framework.
• Evaluated the framework using simulated R-fMRI data and functional network estimation from brain R-fMRI reconstructions.

Main Results:

• The proposed framework significantly improves over the state-of-the-art in R-fMRI reconstruction.
• The method enables substantially higher spatial resolution compared to existing techniques.
• The VB-EM-NM inference framework provides uncertainty estimates for the reconstructions.

Conclusions:

• The novel R-fMRI reconstruction framework enhances spatial resolution and improves functional network estimation.
• The developed inference method offers uncertainty quantification, a key advantage over conventional approaches.
• This work paves the way for more detailed and reliable analysis of brain functional connectivity.