Research Article

Effective Preprocessing Procedures Virtually Eliminate Distance-Dependent Motion Artifacts in Resting State FMRI

Figure 4

The column (a) shows that censoring high-motion frames from RS-FMRI data decreases short-distance correlations and augments long-distance correlations. The Pearson and Spearman correlation differences are plotted as a function of the Euclidean 3D distance between brain locations in the upper and lower rows, respectively. The results for each seed pair averaged over 22 subjects are plotted as red dots. Blue circles are the grand mean of averaged correlation differences for equal numbers of brain location pairs in twelve segments (2,882 pairs per circle), to highlight the trend. In the preprocessing steps, 6 motion estimates with their first difference terms (MO) and tissue-based regressors with their first difference terms (GS; global, eroded white matter, and lateral ventricle signals) were regressed out. Columns (b) and (c) present the distance-dependent correlation biases of nuisance regression models GS and MO, respectively. Column (d) shows results when a localized and eroded WM signal is added in the regression model of (c). Column (e) shows the model of Column (d) with the addition of despiking. The censored time points of FMRI images were determined at > 0.25 mm in (a), and the same time points were used in the censoring process of all models.
(a) GS + MO 
(b) GS
(c) MO
(e) Despike + MO +WMeLOCAL