A fully Bayesian inference framework for population studies of the brain microstructure.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biophysics

Background:

• Diffusion-weighted imaging (DWI) models are crucial for population studies of brain microstructure.
• Current methods relying on null hypothesis significance testing have limitations in reproducibility and detecting the absence of alterations.
• Existing scalar property extraction from DWI models lacks the ability to provide evidence for the absence of microstructural changes.

Purpose of the Study:

• To propose a novel Bayesian framework for population studies of brain microstructure using multi-fascicle models.
• To overcome the limitations of traditional significance testing in DWI-based population studies.
• To enable richer analyses of brain microstructure and improve the reproducibility of findings.

Main Methods:

• Development of a hierarchical Bayesian model for biophysical parameters of brain microstructure.
• Utilizing Hamiltonian Monte Carlo sampling for Bayesian inference.
• Generating joint posterior distributions over latent microstructure parameters for each group.

Main Results:

• The Bayesian framework demonstrated increased reproducibility of findings in population studies.
• The approach enables more nuanced analyses of brain microstructure compared to traditional methods.
• Successful application shown on both synthetic and in-vivo diffusion-weighted imaging data.

Conclusions:

• The proposed Bayesian framework offers a powerful alternative to traditional statistical methods for analyzing brain microstructure.
• This approach enhances the reliability and depth of insights gained from population-based neuroimaging studies.
• It opens new avenues for understanding brain microstructure and its alterations in health and disease.