Clinically Interpretable Transformer Model Reveals Neurodevelopmental Delay in Preterm and NICU Infants Using Diffusion MRI

Brain age estimation from infant MRI provides a quantitative marker of neurodevelopmental maturity that supports early risk stratification, prognosis, and therapy monitoring in preterm and NICU-treated infants. However, existing MRI-based models, ranging from feature-based regressors to CNNs and connectome approaches, remain limited by domain shift, modest accuracy, and unstable explainability. A recent transformer combining T2-weighted MRI and DTI achieved high accuracy, yet its clinical significance in vulnerable populations remains uncertain. We address these gaps using a multi-view transformer applied to DTI, integrating attention-based interpretability and a query-based, fast-processing approach.

This work has been accepted for presentation at ISMRM 2026; final proceedings text may differ.