Neuroimaging techniques, such as PET and fMRI, are used to test hypotheses regarding local changes in neural activity in response to experimental manipulations. These changes are indirectly measured using raw counts or blood flow in the case of PET (Arndt et al., 1996) or susceptibility in the case of BOLD fMRI (Ogawa et al., 1993). Typically, data are acquired from the entire brain volume and analyses undertaken to identify subcomponents of this volume—voxels or regions of interest—in which significant signal changes have occurred (Friston et al., 1995c). Because neuroimaging experiments often test hypotheses regarding local changes in neuronal activity, variations in signal that are common to the entire brain volume (i.e., global blood flow in PET or global signal in fMRI) have been considered nuisance effects to be eliminated (Ramsay et al., 1993). Therefore, the bulk of discussion to date concerning global signals has regarded the appropriate method of their removal (Fox et al., 1988; Friston et al., 1990; Arndt et al., 1996). Specifically, the question of whether regional ‘‘activations’’ are proportional or additive to global signals has been debated at length, with the outcome of importance for scaling versus covariate approaches (Friston et al., 1990). Using one or another of these techniques, numerous PET and fMRI studies in which the effects of the global signal have been removed have been reported (e.g., Cabeza et al., 1997; Courtney et al., 1996; Jonides et al., 1993; Poline et al., 1996; Schacter et al., 1995; Vandenberghe et al., 1996). Relatively unaddressed, however, is the underlying validity of adjusting for the effects of global signal changes in the first place. Specifically, it must be asked if global signals behave as confounds or simple nuisance variables in additive models (Friston et al., 1995c). Formally, confounding exists if ‘‘meaningfully different interpretations of the relationship of interest result when an extraneous variable is ignored or included in the data analysis’’ (Kleinbaum et al., 1988). Confounding may be contrasted with a simple nuisance variable in that inclusion or exclusion of a confound will affect the expected relationship between the data and an independent variable of interest. In contrast, inclusion or exclusion of a simple nuisance variable will only affect the error variance of the model. In other words, covarying for confounds could change interpretation of statistical results both qualitatively (by changing the sign of relationships) and quantitatively (by either increasing or decreasing the significance of relationships), while covarying for simple nuisance variables could only quantitatively impact interpretation by increasing the significance of relationships (while preserving the sign). This difference in behavior between confounds and nuisance variables is due to correlation of the former with the independent variable(s) of interest. As global signals are spatial averages of the local signals of interest, which are themselves hypothesized to correlate with experimental treatments, it is reasonable to hypothesize that global changes would correlate with behavioral manipulations. If true, global signals will act as confounds, and hence interpretation of statistical results would be contingent upon whether a global signal covariate is either included or excluded. PET count data reported by Friston and colleagues (1990) demonstrated a relationship of global flow with experimental condition that was large with respect to the magnitude of the adjusted local effect (see Fig. 6 from that report). Similarly, Aguirre and colleagues (1997) reported that there was a significant correlation between observed global fMRI signals and an experimental paradigm. These two results suggest that global neuroimaging signals can be correlated with the experimental manipulation and are thus not necessarily simple nuisance variables. The implication is that covarying for global signal in PET and fMRI analyses is not simply, or necessarily, increasing power, but meaningfully changing the results and hence interpretation of these studies. To further address the issue of whether global neuroimaging signals could act as confounds, we analyzed an NEUROIMAGE 8, 302–306 (1998) ARTICLE NO. NI980367
[1]
D. Kleinbaum,et al.
Applied Regression Analysis and Other Multivariate Methods
,
1978
.
[2]
M. Mintun,et al.
Enhanced Detection of Focal Brain Responses Using Intersubject Averaging and Change-Distribution Analysis of Subtracted PET Images
,
1988,
Journal of cerebral blood flow and metabolism : official journal of the International Society of Cerebral Blood Flow and Metabolism.
[3]
Karl J. Friston,et al.
The Relationship between Global and Local Changes in PET Scans
,
1990,
Journal of cerebral blood flow and metabolism : official journal of the International Society of Cerebral Blood Flow and Metabolism.
[4]
Ravi S. Menon,et al.
Functional brain mapping by blood oxygenation level-dependent contrast magnetic resonance imaging. A comparison of signal characteristics with a biophysical model.
,
1993,
Biophysical journal.
[5]
Karl J. Friston,et al.
Changes in global cerebral blood flow in humans: effect on regional cerebral blood flow during a neural activation task.
,
1993,
The Journal of physiology.
[6]
Edward E. Smith,et al.
Spatial working memory in humans as revealed by PET
,
1993,
Nature.
[7]
K. Worsley,et al.
Local Maxima and the Expected Euler Characteristic of Excursion Sets of χ 2, F and t Fields
,
1994,
Advances in Applied Probability.
[8]
Karl J. Friston,et al.
Analysis of functional MRI time‐series
,
1994,
Human Brain Mapping.
[9]
Karl J. Friston,et al.
Statistical parametric maps in functional imaging: A general linear approach
,
1994
.
[10]
Karl J. Friston,et al.
Analysis of fMRI Time-Series Revisited
,
1995,
NeuroImage.
[11]
Daniel L. Schacter,et al.
Brain regions associated with retrieval of structurally coherent visual information
,
1995,
Nature.
[12]
Karl J. Friston,et al.
Analysis of fMRI Time-Series Revisited—Again
,
1995,
NeuroImage.
[13]
R. Turner,et al.
Characterizing Evoked Hemodynamics with fMRI
,
1995,
NeuroImage.
[14]
Richard S. J. Frackowiak,et al.
Functional anatomy of a common semantic system for words and pictures
,
1996,
Nature.
[15]
Leslie G. Ungerleider,et al.
Object and spatial visual working memory activate separate neural systems in human cortex.
,
1996,
Cerebral cortex.
[16]
Stephan Arndt,et al.
Normalizing Counts and Cerebral Blood Flow Intensity in Functional Imaging Studies of the Human Brain
,
1996,
NeuroImage.
[17]
Karl J. Friston,et al.
Reproducibility of PET Activation Studies: Lessons from a Multi-Center European Experiment EU Concerted Action on Functional Imaging
,
1996,
NeuroImage.
[18]
E Zarahn,et al.
Empirical analyses of BOLD fMRI statistics. II. Spatially smoothed data collected under null-hypothesis and experimental conditions.
,
1997,
NeuroImage.
[19]
Anthony R. McIntosh,et al.
Age-Related Differences in Neural Activity during Memory Encoding and Retrieval: A Positron Emission Tomography Study
,
1997,
The Journal of Neuroscience.
[20]
M. D’Esposito,et al.
Environmental Knowledge Is Subserved by Separable Dorsal/Ventral Neural Areas
,
1997,
The Journal of Neuroscience.
[21]
M. D’Esposito,et al.
Empirical analyses of BOLD fMRI statistics. I. Spatially unsmoothed data collected under null-hypothesis conditions.
,
1997,
NeuroImage.
[22]
M. D’Esposito,et al.
The Variability of Human, BOLD Hemodynamic Responses
,
1998,
NeuroImage.