Semiparametric Bayesian local functional models for diffusion tensor tract statistics.

Area of Science:

• Neuroimaging
• Biostatistics
• Computational Neuroscience

Background:

• Analyzing diffusion properties along white matter tracts is crucial for understanding brain development.
• Existing models often struggle to account for individual variability in fiber bundle shapes and covariate effects.

Purpose of the Study:

• To propose a semiparametric Bayesian local functional model (BFM) for analyzing multiple diffusion properties in white matter.
• To investigate the association between covariates (e.g., age, gender) and diffusion properties along white matter fiber bundles.
• To develop methods for identifying specific fiber segments significantly influenced by covariates and for subject clustering.

Main Methods:

• Developed a Bayesian functional model (BFM) incorporating a nonparametric Bayesian LPP2 prior and an infinite factor model.
• Implemented local hypothesis testing and credible bands for identifying significant associations while controlling for multiple comparisons.
• Utilized Markov chain Monte Carlo (MCMC) for posterior computation and a simulation study to assess finite sample performance.

Main Results:

• The BFM effectively models heterogeneity in diffusion properties and allows covariate effects to vary across subjects.
• Identified specific fiber segments where diffusion properties are significantly associated with covariates of interest.
• Demonstrated the model's ability to naturally group subjects into homogeneous clusters.

Conclusions:

• The proposed BFM provides a flexible and robust framework for analyzing diffusion properties in white matter tracts.
• BFM facilitates the investigation of neurodevelopmental trajectories by linking diffusion changes to covariates like age and gender.
• The model's application to infant neurodevelopmental studies highlights its potential in clinical research.