Machine Learning of Diffusion MRI U-Fiber FA Predicts Working Memory in MS

In a proof-of-concept study, tree-based ML models using fractional anisotropy (FA) of juxtacortical U-fibers predicted working memory Z-scores in MS patients with R²=0.49 (MSE 0.26), highlighting early dysconnectivity as a cognitive biomarker.

• 93 participants (35 healthy controls, 58 MS patients); diffusion MRI FA from 100 U-fiber regions analyzed via sequential forward selection and 15 regression models.
• Best model (CatBoost) achieved R²=0.488±0.128, MSE=0.256±0.068; most informative regions: left superior temporal and right inferior parietal.
• Single-center proof of concept with small sample; no external validation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.