Full Length ArticlesSignificant feed-forward connectivity revealed by high frequency components of BOLD fMRI signals
Introduction
Functional magnetic resonance imaging (fMRI) (Belliveau et al., 1991) using BOLD contrast (Kwong et al., 1992, Ogawa et al., 1992) has become an indispensable tool in non-invasive elucidation of brain areas subserving cognitive processes and behaviors. In addition to localizing individual functional areas, fMRI can also be used to reveal spatially distributed neuronal networks by studying either the temporal correlation (i.e., functional connectivity) or the causal modulation between brain areas (i.e., effective connectivity) activated by specific stimuli and tasks (for review, see (Friston, 2011b)).
The effective connectivity analysis of fMRI data has various implementations, including Structural Equation Modeling (SEM) (McArdle and McDonald, 1984), dynamic causal modeling (DCM) (Friston, 2011a, Friston et al., 2003, Penny et al., 2004), and Granger causality analysis (Granger, 1969). Different from SEM and DCM, where models of explicit directional influences among functional areas must be specified a priori, Granger causality analysis uses fMRI data to estimate the direction of information flow directly. The inference about the direction of information flow in Granger causality analysis is based on the effectiveness of the ‘prediction’ (Granger, 1969). Specifically, if one region is considered as the ‘source’, the prediction of the behavior at the ‘target’ region should be significantly improved when information about the ‘source’ is provided. In practice, most Granger causality analyses use fMRI time series and time-domain models in these prediction calculations. While the validity of Granger causality estimates using fMRI has been either supported (Abler et al., 2006, Deshpande et al., 2009, Eichler, 2005, Goebel et al., 2003, Kayser et al., 2009, Londei et al., 2006, Roebroeck et al., 2005, Sato et al., 2006) or questioned (David et al., 2008, Smith et al., 2012) because of regional difference in vasculature reactivity (Lee et al., 1995, Miezin et al., 2000), Granger causality analysis has been applied to many fMRI studies (for review, see (Stephan and Roebroeck, 2012)).
Note that Granger causality can be also formulated in the frequency domain (Brovelli et al., 2004, Geweke, 1982). Therefore the estimated causal modulations can be decomposed into different frequency ranges. When applied to fMRI, the sensitivity and specificity of such spectral Granger analysis is limited by the signal-to-noise ratio of fMRI BOLD signals at different frequencies. The spectral property of BOLD fMRI has been studied since the early days of fMRI (Weisskoff et al., 1993). Disturbances due to cardiac/respiratory fluctuations at characteristic frequencies (Beckmann et al., 2005, Birn et al., 2006) and low frequency drift (0–0.015 Hz) (Smith et al., 1999) have been reported. Interestingly, there is also empirical evidence suggesting that BOLD signal between 0.03 Hz and 0.06 Hz can be closely related to electrophysiological activity (Zuo et al., 2010) and gives greater small-world topology (Achard et al., 2006).
In this study, we use magnetic resonance inverse imaging (InI), a method of fast fMRI capable of sampling the whole-brain BOLD signal at 10 Hz with approximately 5 mm spatial resolution at cortex using a 32-channel head coil array at 3 T (Lin et al., 2006, Lin et al., 2008), to study the Granger causality spectrally up to 5 Hz. We specifically hypothesize that, at frequencies higher than 1 Hz, BOLD signals can still be used to provide estimates of directional information faithfully. Using a visuomotor two-choice reaction-time task, we first successfully identified clear feed-forward effective connectivity from visual to sensorimotor systems in our previous studies (Lin et al., 2013, Lin et al., 2014). Such feed-forward connectivity remains significant at frequencies up to 3 Hz. Our results suggest that the BOLD signal at frequencies higher than 2 Hz can still carry useful physiological information to disclose causal modulation.
Section snippets
Subjects and the task
Twenty-three subjects were recruited to this study with written informed consents approved by the Institute Review Board of National Taiwan University Hospital. Subjects were all right-handed. Part of this data set was used in previous analyses (Lin et al., 2013, Lin et al., 2014).
The experiment used a two-choice reaction-time task, where left or right visual hemifield reversing (8 Hz) checkerboard stimuli were presented to the subjects in a rapid event-related fMRI design. The hemifield
Results
The orders of AR model for all time series were listed in Table 1. On average, the time series were modeled with a 10- to 15-order AR model. Note that from the formulation (Eq. (3)), only one specific AR model was used for one set of the time series. In other words, the AR model order did not change across frequencies.
Fig. 1 shows the significant (p < 0.05) dominant directions of information flow estimated by Granger causality and isolated effective coherence at 0.1 Hz, 2.5 Hz, and 5.0 Hz. For
Discussion
This study is the first systematic study in revealing hemodynamic causal modulations among brain areas in a task-related network at frequencies up to 5 Hz. Importantly, taking the time domain analysis results as the reference, many feed-forward connections remained significant in the same direction between 1 Hz and 3 Hz. To minimize confounds of the physiological noise from cardiac and respiratory cycles, we used a Bayesian estimation approach (Sarkka et al., 2012) to track and suppress associated
Acknowledgment
The authors thank Dr. Pedro A. Valdés-Sosa for his insightful comments. This work was partially supported by NSC 101-2628-B-002-005-MY3, MOST 103-2628-B-002-002-MY3 (Ministry of Science and Technology, Taiwan), 100-EC-17-A-19-S1-175 (Ministry of Economic Affairs, Taiwan), MOHW103-TDU-PB-211-000026 (Ministry of Health and Welfare, Taiwan), NHRI-EX103-10247EI (National Health Research Institute, Taiwan), Finland Distinguished Professor (FiDiPro) program (TEKES), and the Academy of Finland.
References (64)
- et al.
Investigating directed influences between activated brain areas in a motor-response task using fMRI
Magn. Reson. Imaging
(2006) - et al.
Separating respiratory-variation-related fluctuations from neuronal-activity-related fluctuations in fMRI
Neuroimage
(2006) - et al.
The functional role of cross-frequency coupling
Trends Cogn. Sci.
(2010) - et al.
Time–frequency dynamics of resting-state brain connectivity measured with fMRI
Neuroimage
(2010) - et al.
Whole-head rapid fMRI acquisition using echo-shifted magnetic resonance inverse imaging
Neuroimage
(2013) - et al.
BOLD fractional contribution to resting-state functional connectivity above 0.1 Hz
Neuroimage
(2015) - et al.
Cortical surface-based analysis. I. Segmentation and surface reconstruction
Neuroimage
(1999) - et al.
Analyzing information flow in brain networks with nonparametric Granger causality
Neuroimage
(2008) - et al.
Identifying the default mode network structure using dynamic causal modeling on resting-state functional magnetic resonance imaging
Neuroimage
(2014) - et al.
Cortical surface-based analysis. II: inflation, flattening, and a surface-based coordinate system
Neuroimage
(1999)
Dynamic causal modeling and Granger causality Comments on: The identification of interacting networks in the brain using fMRI: model selection, causality and deconvolution
Neuroimage
Dynamic causal modelling
Neuroimage
Investigating directed cortical interactions in time-resolved fMRI data using vector autoregressive modeling and Granger causality mapping
Magn. Reson. Imaging
MR-encephalography: fast multi-channel monitoring of brain physiology with magnetic resonance
Neuroimage
Tracking dynamic resting-state networks at higher frequencies using MR-encephalography
Neuroimage
Event-related single-shot volumetric functional magnetic resonance inverse imaging of visual processing
Neuroimage
fMRI hemodynamics accurately reflects neuronal timing in the human brain measured by MEG
Neuroimage
BOLD-contrast functional MRI signal changes related to intermittent rhythmic delta activity in EEG during voluntary hyperventilation-simultaneous EEG and fMRI study
Neuroimage
Characterizing the hemodynamic response: effects of presentation rate, sampling procedure, and the possibility of ordering brain activity based on relative timing
Neuroimage
Comparing dynamic causal models
Neuroimage
Mapping directed influence over the brain using Granger causality and fMRI
Neuroimage
A simple view of the brain through a frequency-specific functional connectivity measure
Neuroimage
Dynamic retrospective filtering of physiological noise in BOLD fMRI: DRIFTER
Neuroimage
A method to produce evolving functional connectivity maps during the course of an fMRI experiment using wavelet-based time-varying Granger causality
Neuroimage
Investigation of low frequency drift in fMRI signal
Neuroimage
The danger of systematic bias in group-level FMRI-lag-based causality estimation
Neuroimage
A short history of causal modeling of fMRI data
Neuroimage
Testing for nonlinearity in time-series—the method of surrogate data
Phys. D
The oscillating brain: complex and reliable
Neuroimage
A resilient, low-frequency, small-world human brain functional network with highly connected association cortical hubs
J. Neurosci.
Laminar pattern of termination of the ipsilateral cortical projection from SII to SI in cats
J. Comp. Neurol.
Anatomical and functional assemblies of brain BOLD oscillations
J. Neurosci.
Cited by (30)
Imaging faster neural dynamics with fast fMRI: A need for updated models of the hemodynamic response
2021, Progress in NeurobiologyModel-based physiological noise removal in fast fMRI
2020, NeuroImageMapping Internet gaming disorder using effective connectivity: A spectral dynamic causal modeling study
2019, Addictive BehaviorsCitation Excerpt :A high-pass filter was also used to remove possible ultraslow fluctuations (<0.0078 Hz). We did not remove frequencies above 0.1 Hz, as they were confirmed to contain meaningful information in resting-state research (Lin et al., 2015). Then, we selected the MNI coordinates of nodes that have been previously reported in the DMN: the mPFC [3, 54, −2], PCC [0, −52, 26], left IPL [−50, −63, 32] and right IPL [48, −69, 35] (Di & Biswal, 2014; Razi et al., 2015; Tang et al., 2016), as the (8 mm radius) centre of the sphere to compute the principal eigenvariate and correct for confounds.
Advantages of short repetition time resting-state functional MRI enabled by simultaneous multi-slice imaging
2019, Journal of Neuroscience Methods
- 1
Both contributed equally.