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, University of California at Los Angeles, Los Angeles, CA, United States.
Frontiers in Neurology
|September 5, 2025
Summary
Data harmonization using NeuroCombat can improve statistical power in multi-site neuroimaging studies of central nervous system (CNS) disease. Careful outlier removal and sham population pooling enhance its effectiveness in revealing true biological differences.
Area of Science:
- Neuroimaging
- Neuroscience
- Biostatistics
Background:
- Multi-site neuroimaging studies aim for larger, reproducible datasets to detect subtle effects.
- NeuroCombat is a data harmonization model that can remove site-specific technical variance.
- Its impact on statistical power in preclinical central nervous system (CNS) disease models remains unclear.
Purpose of the Study:
- To evaluate the effectiveness of NeuroCombat in harmonizing fractional anisotropy data from a multi-site preclinical traumatic brain injury study.
- To assess the impact of data harmonization on statistical power and the detection of injury-related pathology.
- To determine if NeuroCombat improves the reliability of multi-site neuroimaging data.
Main Methods:
- Analysis of fractional anisotropy data from 184 rats across four sites, 3 and 30 days post-controlled cortical impact injury.
- Application and evaluation of the NeuroCombat harmonization model, including outlier assessment.
- Statistical analysis of effect sizes, group differences, and voxel-wise pathology before and after harmonization.
Main Results:
- NeuroCombat failed to remove site effects in data with high outlier proportions (>5%) and skewness.
- After removing one outlier site and harmonizing using a pooled sham population, effect size and group effects increased (p < 0.01).
- Harmonization improved statistical power and the detection of injury-specific differences in the ipsilateral cortex.
Conclusions:
- NeuroCombat's effectiveness is contingent on data quality, particularly the proportion of outliers and skewness.
- Strategic outlier removal and sham population harmonization enhance NeuroCombat's utility.
- This optimized approach improves the reliability of multi-site neuroimaging studies by better distinguishing biological effects from technical confounds.
