Evaluating Free Water Elimination for Tractometry in Neonatal Brains

Abstract

Diffusion MRI (dMRI) provides a unique non-invasive means of characterizing white matter (WM) organization and structural connectivity during early brain development. However, conventional dMRI pipelines were developed primarily for adult populations and may not adequately account for the distinct tissue composition of the neonatal brain. Incomplete myelination, immature microstructural organization, and reduced tissue packing contribute to elevated free-water content, which can decrease tissue specificity and compromise the reliability of downstream fiber-orientation modeling and tractometry. Free-water elimination (FWE) may address these limitations by separating isotropic free-water signal from anisotropic tissue signal. This descriptive proof-of-concept study evaluated the effect of FWE on neonatal tractometry using babyFWE, an FWE-based pipeline adapted for neonatal dMRI. BabyFWE was benchmarked against an otherwise matched conventional pipeline in 3 neonates from the Developing Human Connectome Project. Split-half reliability was evaluated for fiber-orientation distribution functions (fODFs), tract delineation, and tract profiles. Voxelwise fODF reliability was quantified using Pearson’s r² between corresponding split-half fODFs, tract-delineation reliability was assessed using weighted Dice coefficients between tract-density maps, and tract-profile reliability was measured using ICC(2,1) for DKI-FA, DKI-MD, DKI-MK, and DKI-AWF. Tract yield was evaluated separately using whole-brain, non-split tractography. BabyFWE generally improved reliability relative to conventional processing, although the effect differed among outcome measures. Voxelwise fODF reliability showed modest, spatially heterogeneous improvements, with participant-level mean and median changes averaging 8.0% and 7.3%, respectively. Weighted Dice coefficients increased for all 14 tracts, with an average increase of 71.1% (SD = 58.7%), indicating more consistent tract delineation. Tract yield increased for 7 of the 14 tracts, with no tract showing a negative mean difference. Tract-profile reliability also exhibited modest overall improvements, with variability among tracts and tissue properties. These findings provide preliminary evidence that FWE can improve multiple dimensions of neonatal tractometry reliability, with particularly consistent benefits for tract delineation and tract yield. Further validation in larger, independent cohorts is needed to establish generalizability and determine whether these reliability improvements also increase the biological and clinical validity of neonatal dMRI measurements.