Harmonization mitigates diffusion MRI scanner effects in infancy: insights from the HEALthy Brain and Childhood Development (HBCD) study
New ComBat-GAM method eliminates statistically significant scanner differences across six models in major brain study.
A large research team led by Elyssa M. McMaster from Vanderbilt University has published a critical methodological advance for the landmark HEALthy Brain and Childhood Development (HBCD) Study. The ongoing, large-scale longitudinal initiative aims to map population-level brain maturation from infancy but faces a common big-data challenge: variance introduced by using different MRI scanners across study sites can obscure genuine biological signals. The team systematically characterized these scanner model effects within the HBCD Data Release 1.1, which includes diffusion tensor imaging (DTI) metrics from a predetermined set of brain fiber bundles.
To solve this, the researchers applied a harmonization technique called ComBat-GAM (Generalized Additive Model) to the diffusion MRI data collected across six different scanner models. The results were striking. After harmonization, statistical analysis revealed zero significant differences between data distributions from any scanner model following False Discovery Rate (FDR) correction. Furthermore, they successfully reduced Cohen's f effect sizes—a measure of the magnitude of group differences—across all measured DTI metrics. This work, accepted for ISBI 2026, provides a cleaned, analysis-ready dataset and a proven protocol, enabling future investigations to focus purely on brain development rather than technical artifacts.
- The HBCD Study's ComBat-GAM harmonization eliminated all statistically significant scanner model differences after FDR correction.
- The method was applied to diffusion MRI data across six scanner models in the HBCD Data Release 1.1, reducing Cohen's f effect sizes for all metrics.
- This creates a reliable, multi-site foundation for analyzing infant brain development, removing a major confound in large-scale neuroimaging research.
Why It Matters
Enables reliable, large-scale analysis of infant brain development by removing technical noise, accelerating discoveries in early childhood neuroscience.