Abstract
The influence of global BOLD fluctuations on resting state functional connectivity in fMRI data remains a topic of debate, with little consensus. In this study, we assessed the effects of global signal regression (GSR) on effective connectivity within and between resting state networks (RSNs) - as estimated with dynamic causal modelling (DCM) for resting state fMRI (rsfMRI). DCM incorporates a forward (generative) model that quantifies the contribution of different types of noise (including global measurement noise), effective connectivity, and (neuro)vascular processes to functional connectivity measurements. DCM analyses were applied to two different designs; namely, longitudinal and cross-sectional designs. In the modelling of longitudinal designs, we considered four extensive longitudinal resting state fMRI datasets with a total number of 20 subjects. In the analysis of cross-sectional designs, we used rsfMRI data from 361 subjects from the Human Connectome Project. We hypothesized that (1) GSR would have no discernible impact on effective connectivity estimated with DCM, and (2) GSR would be reflected in the parameters representing global measurement noise. Additionally, we performed comparative analyses of information gain with and without GSR. Our results showed negligible to small effects of GSR on effective connectivity within small (separately estimated) RSNs. However, although the effect sizes were small, there was substantial to conclusive evidence for an effect of GSR on connectivity parameters. For between-network connectivity, we found two important effects: the effect of GSR on between-network effective connectivity (averaged over all connections) was negligible to small, while the effect of GSR on individual connections was non-negligible. In the cross-sectional (but not in the longitudinal) data, some connections showed substantial to conclusive evidence for an effect of GSR. Contrary to our expectations, we found either no effect (in the longitudinal designs) or a non-specific (cross-sectional design) effect of GSR on parameters characterising (global) measurement noise. Data without GSR were found to be more informative than data with GSR; however, in small resting state networks the precision of posterior estimates was greater after GSR. In conclusion, GSR is a minor concern in DCM studies; however, quantitative interpretation of between-network connections (as opposed to average between-network connectivity) and noise parameters should be treated with some caution. The Kullback-Leibler divergence of the posterior from the prior (i.e., information gain) - together with the precision of posterior estimates - might offer a useful measure to assess the appropriateness of GSR in resting state fMRI.
Highlights
The fMRI signal is corrupted by noise from several sources; for example, motion-induced noise, background noise, and nonneural physiological noise that arises from cardiac and respiratory processes (Liu, 2016)
The average root mean squared difference (RMSD) between connectivity with and without global signal regression (GSR) was 0.10, 0.06, and 0.05Hz for somatomotor network (SMR), salience network (SAL), and default mode network (DMN) networks, respectively, which is smaller than or equal to the heuristic threshold of 0.1Hz usually used in dynamic causal modelling (DCM) studies
In this study we investigated the effect of GSR on effective connectivity and parameters representing global observation noise, estimated with spectral DCM
Summary
The fMRI signal is corrupted by noise from several sources; for example, motion-induced noise, background (thermal) noise, and nonneural physiological noise that arises from cardiac and respiratory processes (Liu, 2016). This is a particular problem in resting state fMRI research, which usually aims to quantify low-frequency fluctuations in the absence of explicit perturbations. Extensive research has focused on developing and applying methods to de-noise the (resting-state) fMRI signal (e.g., Erdogan et al, 2016; Kasper et al, 2017; Rummel et al, 2013). Different types of corrections for GS fluctuations have been developed and applied, including GS regression (GSR), GS normalization, and GS subtraction (Liu et al, 2017)
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
