ComBat Harmonization With and Without Empirical Bayes Estimation for Resting-State Functional Connectivity in...

Area of Science:

• Neuroimaging
• Neuroscience
• Biostatistics

Background:

• Multi-site neuroimaging studies face challenges with batch effects.
• ComBat is a common tool for mitigating these effects.
• The role of empirical Bayes estimation in ComBat for resting-state functional connectivity (rsFC) is not fully understood.

Purpose of the Study:

• To evaluate the impact of empirical Bayes estimation in ComBat on rsFC measures.
• To assess site effect reduction and preservation of within-site variability in rsFC data from adolescents with mild TBI.

Main Methods:

• Resting-state fMRI data from 144 adolescents (CARE4Kids study) across six sites were analyzed.
• Functional connectivity was computed using Seitzman parcellation.
• ComBat, with and without empirical Bayes, was applied and assessed using ANOVA, ICC, age correlations, and principal component analysis.

Main Results:

• Significant site effects were observed across all networks before harmonization.
• Both ComBat approaches reduced site variability, with empirical Bayes yielding slightly higher residual site effects.
• Empirical Bayes consistently maintained excellent within-site consistency post-harmonization.
• Harmonization without empirical Bayes showed variable consistency across different site sample sizes.
• Both methods improved age associations and reduced principal component site effects.

Conclusions:

• ComBat effectively removes site variability in functional connectivity while preserving age-related associations.
• Empirical Bayes estimation in ComBat is superior in preserving original within-site variability in rsFC measures.
• This finding is crucial for accurate analysis of multi-site neuroimaging data, particularly in clinical populations like adolescents with mild TBI.