An evaluation of the efficacy, reliability, and sensitivity of motion correction strategies for resting-state functional MRI

Estimates of functional connectivity derived from resting-state functional magnetic resonance imaging (rs-fMRI) are highly sensitive to artefacts caused by in-scanner head motion. This susceptibility has motivated the development of numerous denoising methods designed to mitigate motion-related artefacts. Here, we compare 8 popular retrospective rs-fMRI denoising methods, including methods such as regression of head motion parameters (with or without expansion terms), aCompCor, volume censoring (e.g., scrubbing and spike regression), global signal regression and ICA-AROMA, combined into 16 different pipelines. These pipelines were evaluated across five different quality control benchmarks in three independent datasets that were characterized by both high and low levels of motion. Pipelines were benchmarked by examining the residual relationship between in-scanner movement and functional connectivity after denoising; the effect of distance on this residual relationship; whole-brain differences in functional connectivity between high- and low-motion healthy controls (HC); the temporal degrees of freedom lost during denoising; and the test-retest reliability of functional connectivity estimates. We also compared the sensitivity of each pipeline to clinical differences in functional connectivity in comparisons between people with schizophrenia (SCZ; n = 50) and HCs (n = 121) and people with obsessive-compulsive disorder (OCD; n = 34) and HCs (n = 39). Our results indicate that (1) simple linear regression of regional fMRI time series against head motion parameters (with or without expansion terms) is not sufficient to remove head motion artefacts; (2) aCompCor pipelines can exacerbate motion artefacts in low-motion data; (3) the primary benefit of volume censoring comes from the exclusion of high-motion individuals rather than censoring of data in remaining participants; and (4) that ICA-AROMA consistently performed well across all benchmarks and datasets, particularly when applied after the exclusion of high-motion individuals. ICA-AROMA was also the most sensitive to clinical differences in case-control analyses, suggesting that its denoising efficacy is associated with enhanced power for detecting pathophysiological effects. Crucially, the comparison between HC and SCZ revealed that the specific choice of noise correction pipeline had a major effect on the findings, affecting both the location and direction of group differences. Putative increases in functional connectivity in patients only emerged in pipelines incorporating either global signal regression or aCompCor. Thus, group comparisons in functional connectivity are highly dependent on preprocessing strategy. We offer some recommendations for best practice and outline some simple analyses to facilitate transparent reporting of the degree to which a given set of findings may be affected by motion-related artefact.