Comparing Techniques for Multi-Site Harmonization of Structural Connectivity

Tractography is a method to reconstruct white matter microstructure from DWI information and connectomics maps this reconstruction to a graph representation. We compute modularity, assortativity, global efficiency, and average betweenness centrality on this graph. We model changes in these measures with age and sex using DWI for healthy patients from two sites. Data from different sites requires harmonization to remove site-effects. We compare performances of ComBat and LinearRISH harmonization techniques at reducing CoV and removing confounding site-effects from associated linear models. We find that ComBat is effective at both and using LinearRISH in addition acts synergistically at harmonizing site differences.