Abstract
Normative modeling (NM) enables detection and accurate quantification of individual-level deviations from a population reference. Even so, the training sample sizes needed to create well-calibrated diffusion tensor imaging (DTI) normative models, with known error, is largely unknown. It is also likely that sample size requirements for a known level of calibration will differ across white matter (WM) tracts. We trained hierarchical Bayesian regression (HBR) normative models of fractional anisotropy (FA) using harmonized multisite diffusion MRI from 54,812 individuals (ages 4-91) across 20 datasets, with an 8,812-subject test set (stratified by site and sex) held out from training. Using training subsamples of 5,000-45,000 subjects, we fit Gaussian HBR models for 22 ENIGMA-DTI WM regions, evaluating calibration via the mean absolute centile error (MACE) at the outer (0.05, 0.95) centiles. As expected, calibration improved with the training sample size for 20 of 22 regions, but the rate and extent varied widely by tract: the fornix, external capsule, posterior thalamic radiation, and cingulum improved most (10-26% MACE reduction, N=5,000 to 45,000), while the whole-skeleton Average and two other tracts showed no detectable improvement. Comparing pooled models against sex-specific models (up to 20,000 subjects/sex) revealed a modest but significant calibration advantage for sex-specific models in about half the regions tested, driven by calibration in females improving faster with sample size rather than by better calibration at matched N, an advantage still growing at N=20,000. These results provide tract-resolved, empirically validated guidance on cohort sizes needed for calibrated DTI-FA normative models and identify regions that may benefit from sex-stratified modeling as cohorts continue to grow.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/cdmri_013.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=iuaMCPgmH8
BibTex
@InProceedings{VilJul_RegionSpecific_MICCAISAT2026,
author = { Villalón-Reina, Julio E. AND Nabulsi, Leila AND Thomopoulos, Sophia I. AND Feng, Yixue AND Liou, Kenny AND John, John P. AND Lawrence, Katherine E. AND Nir, Talia M. AND Jahanshad, Neda AND Marquand, Andre F. AND Kia, Seyed Mostafa AND Thompson, Paul M.},
title = { { Region-Specific Sample Size Requirements for Calibrated Normative Modeling of Diffusion Tensor Imaging Fractional Anisotropy Across the Lifespan } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}