ComBat models for harmonization of resting-state EEG features in multisite studies.

Area of Science:

• Neuroscience
• Biostatistics
• Signal Processing

Background:

• Multisite resting-state electroencephalography (rsEEG) data pooling can introduce batch effects, compromising data integrity.
• ComBat models, originally for gene expression, are adapted for neuroimaging to mitigate these cross-site differences.
• Evaluating ComBat variants is crucial for harmonizing diverse rsEEG datasets.

Purpose of the Study:

• To assess the efficacy of four ComBat harmonization methods in a pooled rsEEG dataset from young and old adults.
• To determine which ComBat variants best control for batch effects while preserving biological variability.
• To investigate the impact of harmonization on age-related associations in rsEEG spectral features.

Main Methods:

• Four ComBat harmonization methods (neuroCombat, neuroHarmonize, OPNested-GMM, HarmonizR) were applied to preprocessed rsEEG signals (n=374).
• Oscillatory and aperiodic rsEEG features were extracted in sensor space.
• Age-related relationships were analyzed pre- and post-harmonization.

Main Results:

• Batch effects were confirmed in the rsEEG features.
• All ComBat methods successfully reduced batch effects and feature dispersion.
• HarmonizR and OPNested-GMM ComBat demonstrated superior performance.
• Harmonized spectral features, including Beta power and Alpha peak frequency, showed significant age-related associations.
• ComBat models preserved the direction of age effects and increased their magnitude.

Conclusions:

• ComBat models, specifically HarmonizR and OPNested-GMM ComBat, are effective for harmonizing sensor-space rsEEG spectral features.
• This harmonization workflow is valuable for multisite studies, ensuring reliable analysis of rsEEG data.
• The methods preserve biologically relevant associations, such as those related to aging.