|
Abstract:
|
Neuroimaging data from large epidemiologic cohort studies often come from multiple scanners. The variations of MRI measurements due to differences in magnetic field strength, image acquisition protocols, and scanner vendors can cause bias and thus influence the interpretation of aggregated data. Therefore, various methodological efforts have been made to harmonize MRI data acquired from different scanners or protocols. ComBat, a method originally developed for harmonizing batch effect in genetics, has been employed in MRI research on this account. While ComBat has attained popularity in MRI studies, a few concerns have been noted. One of which is how ComBat treats unmeasured biological signals. To answer the question, we performed extensive simulations where biological signals were parameterized in terms of varying effect size, spatial extent, and covariance structure. Thereafter, ComBat was performed with simulated image data, where we manipulated amount of information given to ComBat. Results using full versus partial information were compared in various scenarios. As results, we found that hidden signals were mostly preserved after ComBat in most of the simulation scenarios. In conclusion, ComBat can be safely applied even in a situation that the full biological information is not available.
|
