Person:

Tong, Yunjie

Resting state functional connectivity analysis is a widely used method for mapping intrinsic functional organization of the brain. Global signal regression (GSR) is commonly employed for removing systemic global variance from resting state BOLD-fMRI data; however, recent studies have demonstrated that GSR may introduce spurious negative correlations within and between functional networks, calling into question the meaning of anticorrelations reported between some networks. In the present study, we propose that global signal from resting state fMRI is composed primarily of systemic low frequency oscillations (sLFOs) that propagate with cerebral blood circulation throughout the brain. We introduce a novel systemic noise removal strategy for resting state fMRI data, “dynamic global signal regression” (dGSR), which applies a voxel-specific optimal time delay to the global signal prior to regression from voxel-wise time series. We test our hypothesis on two functional systems that are suggested to be intrinsically organized into anticorrelated networks: the default mode network (DMN) and task positive network (TPN). We evaluate the efficacy of dGSR and compare its performance with the conventional “static” global regression (sGSR) method in terms of (i) explaining systemic variance in the data and (ii) enhancing specificity and sensitivity of functional connectivity measures. dGSR increases the amount of BOLD signal variance being modeled and removed relative to sGSR while reducing spurious negative correlations introduced in reference regions by sGSR, and attenuating inflated positive connectivity measures. We conclude that incorporating time delay information for sLFOs into global noise removal strategies is of crucial importance for optimal noise removal from resting state functional connectivity maps.

It is widely accepted that the fluctuations in resting state blood oxygenation level dependent (BOLD) functional MRI (fMRI) reflect baseline neuronal activation through neurovascular coupling; this data is used to infer functional connectivity in the human brain during rest. Consistent activation patterns, i.e., resting state networks (RSN) are seen across groups, conditions, and even species. In this study, we show that some of these patterns can also be generated from the dynamic, systemic, non-neuronal physiological low frequency oscillations (sLFOs) in the BOLD signal alone. We have previously used multimodal imaging to demonstrate the wide presence of the same sLFOs in the brain (BOLD) and periphery with different time delays. This study shows that these sLFOs from BOLD signals alone can give rise to stable spatial patterns, which can be detected during resting state analyses. We generated synthetic resting state data for 11 subjects based only on subject-specific, dynamic sLFO information obtained from resting state data using concurrent peripheral optical imaging or a novel recursive procedure. We compared the results obtained by performing a group independent component analysis (ICA) on this synthetic data (i.e., the result from simulation) to the results obtained from analysis of the real data. ICA detected most of the eight well-known RSNs, including visual, motor, and default mode networks (DMNs), in both the real and the synthetic data sets. These findings suggest that RSNs may reflect, to some extent, vascular anatomy associated with systemic fluctuations, rather than neuronal connectivity.