Large multi-site studies that pool magnetic resonance imaging (MRI) data across research sites or studies, or that utilize shared data from imaging repositories, present exceptional opportunities to advance neuroscience and enhance reproducibility of neuroimaging research. However, both scanner and site variability are confounds that hinder pooling data collected across different sites or across different operating systems on the same scanner, even when all acquisition protocols are harmonized. These confounds degrade statistical analyses and can lead to spurious findings. Unfortunately, methods to address this problem are scant. In this study, we propose a novel denoising approach for multi-site multimodal MRI data that implements a data-driven linked independent component analysis (LICA) to efficiently identify scanner/site-related effects for removal. Removing these effects results in denoised data that can then be combined across sites/studies to improve modality-specific statistical processing. We use data from six different studies collected on the same scanner across major hardware (gradient and head coil) and software upgrades to demonstrate our LICA-based denoising approach. The proposed method is superior compared to the existing methods we tested and has great potential for large-scale multi-site studies to produce combined data free from study/site confounds.

Combining Multi-Site/Multi-Study MRI Data: Linked-ICA Denoising for Removing Scanner and Site Variability from Multimodal MRI Data