NiChart Data Harmonization
To estimate and remove scanner-related batch effects in imaging variables we apply a statistical harmonization method, ComBat. The ComBat method is a Bayesian statistical technique aimed at removing batch effects in high-dimensional datasets.
The method estimates both the mean (location) and the variance (scale) of the residuals across batches using Empirical Bayes estimation, after correcting for additional covariates, such as age, sex and ICV.
NiChart data harmonization will be powered by the Combat-family software package that provides an ensemble of harmonization tools. Variants like ComBat-GAM offer the possibility to model selected covariates using splines, providing flexible adjustments to non-linear covariate associations. Combat can be used through a train/test paradigm, applying it on a training set to estimate batch effect parameters, and using the existing model to harmonize new data from the same batches.
Figure 1. Combat family of tools
Combat visualization and quality control (QC) package provides tools for evaluating batch effects and estimated parameters before and/or after harmonization.
Figure 2. Combat visualization and QC tool functions