Robust Correlation for Link Definition in Resting-State fMRI Brain Networks Can Reduce Motion-Related Artifacts

It is well known that even small head movements introduce artifacts in resting-state functional magnetic resonance imaging data, and over the years, numerous methods were introduced to correct for this issue. The field of robust statistics, however, has not yet received much attention in this regard. In this article, we tested a recently developed statistical method called wrapping and compared it with two already established methods: data scrubbing and an independent component analysis-based approach for the automatic removal of motion artifacts (ICA-AROMA). A group of N = 120 healthy adult subjects were divided into high and low movement cohorts. The functional connectomes following wrapping, data scrubbing, and ICA-AROMA of the high movement cohort were compared with the mean functional connectome of the low movement cohort. Our results showed that wrapping could significantly decrease the Euclidean distance between connectomes of the two cohorts. Furthermore, wrapping was able to compensate the systematic effect of increased short distance correlations and reduced long distance correlations in functional connectomes, which often result from high subject motion. Our findings suggest that wrapping constitutes a valuable approach to correct for movement-related artifacts when estimating functional connectivity in the brain. Impact statement The influence of subject motion on functional magnetic resonance imaging (fMRI) data is still an actively discussed topic. However, to handle this problem, the field of robust statistics has not been given much attention yet. We want to fill this void by introducing and validating a recently developed method for calculating robust correlations. Our study shows that estimating robust correlations can improve fMRI preprocessing, and documents for a wider readership that fMRI analyses can benefit from new methods in the field of robust statistics.