A new investigation published in NeuroImage details a hitherto unaccounted for artifact of subject motion in functional connectivity analysis of resting state MRI data that has important implications for studies of functional connectivity, including those being done as part of the Human Connectome Project.
Jonathan Power and colleagues at Washington University in St. Louis demonstrate that subject head movement during scans produces significant changes in the intensity of the BOLD timecourse signal across the entire brain, particularly in images collected during and surrounding the movement. The study finds that commonly used processing measures to remove motion-related signal from the data such as image realignment and motion estimate regression do not adequately remove, nor cause, these head movement-induced artifacts.
The authors propose a process they call data “scrubbing” to identify and remove motion-corrupted frames using two indices of data quality: framewise head displacement and framewise rate of change of BOLD signal across the entire brain. Motion scrubbing the data significantly altered seed correlation maps within the four cohorts tested, revealing a general increase in medium- to long-range correlations and a decrease in many short-range correlations.
Reanalyzing their own data from a previously published study, the authors demonstrate how motion corruption can obscure patterns of functional connectivity within single cohorts, create spurious differences between cohorts, and impact the conclusions reached by a study. This characterization strongly suggests the need for many resting state functional connectivity MRI (rs-fcMRI) results to be critically revisited and that future studies should account for this substantial artifact.
Although the motion artifact is greater in particular subject populations that display higher levels of movement, such as children, it is found in data from all subject cohorts examined, in data collected at different locations, and on various scanners. Importantly, the artifactual basis of the motion effects described is not specific to rs-fcMRI only, rather it is a general feature of functional MRI that should also be evident, and similarly correctable, in task fMRI and diffusion imaging.
The paper was authored by Jonathan Power, Kelly Barnes, Abraham Snyder, Bradley Schlaggar, and Steven Petersen of Washington University School of Medicine.
Read the full article in NeuroImage: Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion