Topography and behavioral relevance of the global signal in the human brain

Abstract

The global signal in resting-state functional MRI data is considered to be dominated by physiological noise and artifacts, yet a growing literature suggests that it also carries information about widespread neural activity. The biological relevance of the global signal remains poorly understood. Applying principal component analysis to a large neuroimaging dataset, we found that individual variation in global signal topography recapitulates well-established patterns of large-scale functional brain networks. Using canonical correlation analysis, we delineated relationships between individual differences in global signal topography and a battery of phenotypes. The first canonical variate of the global signal, resembling the frontoparietal control network, was significantly related to an axis of positive and negative life outcomes and psychological function. These results suggest that the global signal contains a rich source of information related to trait-level cognition and behavior. This work has significant implications for the contentious debate over artifact removal practices in neuroimaging.

Figure 1

Illustration of global signal beta map calculation. For each subject and each run, the global signal (X) was computed as the averaged timeseries across all cortical vertices. The global signal was then regressed from the timeseries of each vertex (Y), resulting in the GS beta map of a given run (β). Averaging across all runs within a given subject, we obtained the GS beta map of that subject. The GS beta maps for each subject were then used in subsequent PCA and CCA analyses (B).

Figure 2

Global signal topography. (A) Brain regions dominating the global signal, computed as the mean global signal beta map across all subjects per vertex. Brain regions with strong global signal include the visual cortex, posterior insula, central sulcus and cingulate sulcus. (B) Brain regions with high individual variation (standard deviation) of global signal topography include retrosplenial and visual cortex. (C) Global signal principal components computed across subjects. The patterns resemble the canonical brain networks regularly observed from decompositions of resting-state fMRI data.

Figure 3

Individual differences in behavior associated with the global signal. (A) CCA weights of each vertex on the first canonical variate pair. Strong positive weights are observed in the frontoparietal and salience networks, and strong negative weights are observed in the motor cortex. (B) Top 20% positive (blue) and negative (red) behavioral variable CCA weights displayed in a word cloud. The size of the text in the word cloud is proportional to the absolute value of that variable’s CCA weight.

Figure 4

Correlation of Global Signal Estimates with Head Motion. (A) Per-vertex correlation between global signal beta estimates and average head motion (FD and DVARS) across subjects. (B) Scatter plots between global signal (blue) and behavioral (red) CCA scores and average head motion (FD and DVARS). Estimated Pearson correlation coefficients (r) are displayed in the legend beside each label.