fMRI image preprocessing crucially affects functional connectivity analysis
Grega Repovsš, Deanna M Barch
Department of Psychology, Washington University in Saint Louis, 63130 MO, USA
Despite an increase in the number of connectivity studies based on functional magnetic resonance imaging (fMRI), many methodological questions relating to preprocessing and analysis of imaging data to be used in connectivity studies still remain open and in need of empirical testing. To eliminate spurious correlations in the data, it has been proposed that variables representing nuisance covariates such as head motion parameters, deep white matter signal, ventricle signal and whole brain signal should be regressed from the imaging data before performing connectivity analysis. While such a strategy has been shown to improve resting state within subject functional connectivity results, the goal of this study was to assess the impact of using these methods in between subject task-related connectivity analysis. Thirty-eight individuals with schizophrenia and 38 controls were scanned using MRI while performing two-back working memory tasks with words and faces. Regions of interest and seed-correlation based functional connectivity analyses were performed on data after various stages of preprocessing. Between subject correlations were computed based on average task related brain activity as measured by fMRI BOLD response. Analysis of imaging data without additional preprocessing resulted in stable pattern of correlations that did not change significantly between task blocks and also showed robust group differences between patient and control groups. While spatial gaussian smoothing improved signal-to-noise ratio of the results with the raw data, elimination of nuisance covariates significantly changed the observed pattern of connectivity, which became highly unstable, differing substantially between task blocks. Analysis of correlation between nuisance regressors and task design showed substantial between and within subject variability. Significant differences in correlations between groups, tasks and scanning run number were observed. A plot of average within-run timeseries for selected regions showed pronounced deterioration of task related signal after removal of nuisance covariates. While removal of nuisance regressors might prove worthwhile in within-subject resting state studies, the results show that its use in task related between subject functional connectivity studies may not be appropriate and may reduce the signal of interest in such individual difference analyses.