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

• Published in: NeuroImage clinical 2022-01, Vol.33, p.102937, Article 102937
• Main Authors: Muncy, Nathan M., Kimbler, Adam, Hedges-Muncy, Ariana M., McMakin, Dana L., Mattfeld, Aaron T.
• Format: Article
• Language: English
• Subjects: Adolescence; Adolescent; Anisotropy; Anxiety; Anxiety - diagnostic imaging; Corpus Callosum; Diffusion Magnetic Resonance Imaging; Diffusion Tensor Imaging; DWI; GAM; Humans; MRI; Regular; Uncinate; White Matter - diagnostic imaging
• ISSN: 2213-1582; 2213-1582
• DOI: 10.1016/j.nicl.2022.102937

•Generalized Additive Models fit the nature of diffusion-weighted imaging data.•High anxiety participants had greater left uncinate FA values than low anxiety.•Left uncinate FA differences were positively related with memory overgeneralization. Statistical models employed to test for group differences in quantized diffusion-weighted MRI white matter tracts often fail to account for the large number of data points per tract in addition to the distribution, type, and interdependence of the data. To address these issues, we propose the use of Generalized Additive Models (GAMs) and supply code and examples to aid in their implementation. Specifically, using diffusion data from 73 periadolescent clinically anxious and no-psychiatric-diagnosis control participants, we tested for group tract differences and show that a GAM allows for the identification of differences within a tract while accounting for the nature of the data as well as covariates and group factors. Further, we then used these tract differences to investigate their association with performance on a memory test. When comparing our high versus low anxiety groups, we observed a positive association between the left uncinate fasciculus and memory overgeneralization for negatively valenced stimuli. This same association was not evident in the right uncinate or anterior forceps. These findings illustrate that GAMs are well-suited for modeling diffusion data while accounting for various aspects of the data, and suggest that the adoption of GAMs will be a powerful investigatory tool for diffusion-weighted analyses.