FITBIR database harmonization reveals 15 brain regions altered in TBI

A team led by Adam M. Saunders (Vanderbilt University) has tackled a major bottleneck in traumatic brain injury (TBI) research: heterogeneous, multi-site MRI data. They standardized and harmonized the entire Federal Interagency Traumatic Brain Injury Research (FITBIR) database, which contains over 45,529 MRI scans from 6,211 subjects across 26 studies. By converting data into the Brain Imaging Data Structure (BIDS) format and applying rigorous quality assurance, they created a clean cohort of 4,868 subjects with structural MRI and 2,666 with diffusion MRI.

Using UNesT deep learning segmentations, they extracted whole-brain metrics—mean fractional anisotropy, mean diffusivity, total intracranial volume—and volumes of 132 regions of interest. Generalized additive models for location, scale, and shape (GAMLSS) revealed statistically significant volume differences in 15 brain regions when comparing TBI patients to controls (q < 0.05, FDR-corrected). This work not only produces a ready-to-use harmonized dataset but also provides normative aging curves for these metrics, enabling future studies to detect TBI-related deviations more accurately.