Multi-site Normative Modeling of Diffusion Tensor Imaging Metrics Using Hierarchical Bayesian Regression.

Area of Science:

• Neuroimaging
• Biostatistics
• Computational Neuroscience

Background:

• Multi-site imaging studies enhance statistical power but require data harmonization.
• Hierarchical Bayesian Regression (HBR) presents an alternative to harmonization in normative modeling.
• Diffusion tensor imaging (DTI) provides insights into white matter microstructure.

Purpose of the Study:

• To evaluate the utility of HBR for multi-site normative modeling of DTI metrics across the lifespan.
• To compare linear, polynomial, and b-spline models within the HBR framework for age-dependency modeling.
• To demonstrate the application of HBR in detecting group differences, such as in carriers of genetic variants.

Main Methods:

• Applied HBR with linear, polynomial, and b-spline models to cross-sectional DTI data.
• Utilized data from over 1,300 healthy subjects aged 2-80 years across eight international sites.
• Modeled age-related changes in DTI metrics of white matter microstructure.

Main Results:

• Polynomial and b-spline models demonstrated superior performance in capturing DTI metric age dependencies compared to linear models.
• The HBR framework successfully modeled normative age trajectories for DTI metrics.
• The method was applied to identify microstructural differences in carriers of genetic copy number variants.

Conclusions:

• HBR provides a robust framework for multi-site normative modeling of DTI metrics, circumventing the need for data harmonization.
• Non-linear models (polynomial, b-spline) are more appropriate than linear models for capturing age-related changes in white matter microstructure.
• The HBR approach is valuable for detecting subtle group differences in brain microstructure, with model complexity influencing findings.