General additive models address statistical issues in diffusion MRI: An example with clinically anxious adolescents

Area of Science:

• Neuroimaging
• Statistical Modeling
• Developmental Neuroscience

Background:

• Standard statistical models struggle with complex diffusion-weighted MRI (dMRI) data, particularly white matter tract analysis.
• Existing methods often overlook data characteristics like distribution, type, interdependence, and numerous data points per tract.
• This limits the accurate detection of group differences in brain structure.

Purpose of the Study:

• To introduce Generalized Additive Models (GAMs) as a superior statistical approach for analyzing dMRI white matter tracts.
• To provide practical code and examples for implementing GAMs in dMRI research.
• To demonstrate GAMs' utility in identifying group differences and their association with cognitive functions.

Main Methods:

• Utilized dMRI data from 73 periadolescent participants (clinically anxious and healthy controls).
• Applied GAMs to test for group differences in white matter tracts, accounting for data properties and covariates.
• Investigated the association between identified tract differences and memory test performance.

Main Results:

• GAMs successfully identified significant group differences within specific white matter tracts.
• A positive association was found between the left uncinate fasciculus and memory overgeneralization for negative stimuli in higher anxiety groups.
• This association was not observed in the right uncinate fasciculus or anterior forceps.

Conclusions:

• GAMs are highly suitable for modeling dMRI data, effectively handling data complexity and covariates.
• GAMs offer a powerful tool for advancing diffusion-weighted imaging analyses.
• The findings highlight the potential of GAMs in uncovering neurobiological underpinnings of cognitive differences in clinical populations.