Improved Injury Detection Through Harmonizing Multi-Site Neuroimaging Data after Experimental TBI: A Translational
G Kislik1, R Fox1, A V Korotcov2,3
1UCLA Brain Injury Research Center, Department of Neurosurgery, Geffen Medical School.
Biorxiv : the Preprint Server for Biology
|October 1, 2025
Summary
NeuroCombat data harmonization improves statistical power in multi-site neuroimaging studies of traumatic brain injury. By reducing site-specific confounds, it enhances the detection of true biological differences in preclinical models.
Area of Science:
- Neuroscience
- Medical Imaging
- Biostatistics
Background:
- Multi-site neuroimaging studies are crucial for increasing sample sizes and reproducibility in research on central nervous system (CNS) diseases.
- Data harmonization techniques like NeuroCombat aim to mitigate site-specific technical variations while preserving biological group differences.
- The efficacy of NeuroCombat in enhancing statistical power in preclinical models of CNS disease remains under investigation.
Purpose of the Study:
- To evaluate the effect of NeuroCombat data harmonization on statistical power in a preclinical model of traumatic brain injury (TBI).
- To assess the performance of NeuroCombat in the presence of data outliers and skewness common in multi-site studies.
Main Methods:
- Analysis of fractional anisotropy (FA) data from diffusion-weighted imaging in 184 adult rats across four sites, 3 and 30 days post-controlled cortical impact injury.
- Application of NeuroCombat harmonization after identifying and removing an outlier site with high data skewness.
- Comparison of statistical power and effect sizes before and after harmonization, using univariate and voxel-level analyses.
Main Results:
- NeuroCombat failed to remove site effects in data with >5% outliers and skewness, introducing variation.
- Following outlier removal and harmonization with a global sham population, effect sizes increased significantly (p<0.01) in data with existing group effects.
- Harmonization led to more similar voxel measurement distributions (Kolmogorov-Smirnov p<<0.001 to >0.01) and increased statistical power in the ipsilateral cortex.
- Improved detection of injury-affected regions by reducing confounds.
Conclusions:
- NeuroCombat can enhance statistical power and the reliability of findings in multi-site neuroimaging studies when applied appropriately, particularly after addressing data quality issues like outliers.
- This harmonization method aids in distinguishing true biological effects from technical site variations, improving the reproducibility of preclinical TBI research.
- Findings support the utility of NeuroCombat for accurate revelation of biological differences in large-scale neuroimaging consortia.
