Correcting brain-wide correlation differences in resting-state FMRI.

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Data Analysis

Background:

• Resting-state functional magnetic resonance imaging (fMRI) is widely used to study brain function.
• Group comparisons in fMRI are susceptible to confounding factors like global signal fluctuations.
• Current methods such as global signal regression (GSReg) can introduce biases.

Purpose of the Study:

• To propose and evaluate average brain-wide correlations (GCOR) as a measure to identify and correct for global fluctuations in resting-state fMRI.
• To compare the performance of GCOR against GSReg in mitigating confounds and preventing false group differences.

Main Methods:

• Simulations were used to compare GCOR as a covariate and GSReg under various scenarios.
• The bias introduced by GSReg was analyzed as a function of the data's covariance matrix.
• GCOR's applicability was assessed on empirical data with motion-based subject grouping.

Main Results:

• GCOR is a conservative method that reduces global variations without generating false significant differences, unlike GSReg.
• GSReg introduces differential bias across brain regions, potentially leading to erroneous group findings.
• GCOR can reduce correlation differences between subject groups (e.g., high vs. low motion), but motion may not be the sole cause.

Conclusions:

• GCOR is a valuable complementary measure for resting-state fMRI, particularly for legacy data, to address global signal confounds.
• While effective, GCOR is not a substitute for rigorous level-I denoising.
• Further investigation is needed to understand the drivers of group differences observed in fMRI studies.