Multi-Scale Assessment of Harmonization Efficacy on Resting-State Functional Connectivity

Raising trends of neuroimaging data sharing among different research centers, including resting-state functional magnetic resonance imaging (rs-fMRI) measurements, have driven to accessible large-scale sample and improvement of reliability and consistency of downstream analyses. However, in this context several concerns arise for non-biological confounding factors mainly related to differences in magnetic resonance scanners and imaging parameters among sites. Until now, there is limited knowledge of the impact of site-to-site variations in rsfMRI functional connectivity (FC) measures and the most suitable harmonization approach for mitigating such impact. In this study, we aimed to quantitatively evaluate the site-to-site variations in rs-fMRI FC patterns and how the widely used ComBat harmonization performs in removing them. A multi-scale analytical approach was adopted, from single pairs of regions to resting-state networks (RSNs) and to the entire brain. Our findings show that ComBat removes unwanted site effects from rs-fMRI FC measures while improving signal-to-noise ratio (SNR) in the data and RSNs identifiability. Further, we identify and visualized specific FC links highly affected by site, highlighting differences in such effects among RSNs. Overall, our findings demonstrate that ComBat is effective in harmonizing rs-fMRI FC measures, emphasizing also the overall RSNs identifiability and the enhancement of the majority of single RSNs in the entire brain connectome.