Impact of autocorrelation on functional connectivity

Area of Science:

• Neuroimaging
• Brain Connectivity
• Statistical Modeling

Background:

• Autocorrelation in functional Magnetic Resonance Imaging (fMRI) time-series is well-studied for general linear models.
• Its impact on functional connectivity (FC) estimation has been recently questioned, with concerns about 'spurious' correlations.
• Previous research has largely overlooked the specific effects of autocorrelation on FC measures.

Purpose of the Study:

• To comprehensively investigate the impact of autocorrelation on functional connectivity estimates in fMRI data.
• To assess the effect of autocorrelation on Pearson correlation coefficients using theoretical approximation, simulation, and real fMRI data.
• To evaluate the necessity of autocorrelation correction for hypothesis testing in FC studies.

Main Methods:

• Theoretical approximation and simulation studies to analyze the effect of autocorrelation on Pearson correlation coefficients.
• Application and analysis of these effects on real fMRI datasets.
• Investigation of model order selection for autoregressive processes and frequency filtering effects.

Main Results:

• Autocorrelation was found to alter functional connectivity (FC) values in fMRI data.
• Despite alterations in FC values, hypothesis testing results remained largely consistent before and after autocorrelation correction.
• This consistency was observed for main effects and group difference testing (healthy controls vs. schizophrenia patients) in real data.

Conclusions:

• While autocorrelation influences raw FC values, its impact on the outcomes of statistical inference in FC studies is minimal.
• The findings suggest that conventional hypothesis testing on FC may be robust to autocorrelation.
• A preprocessing pipeline for connectivity studies, considering autocorrelation, model order, and filtering, is proposed.