Functional Connectivity in Default Mode Network During Resting State: An Evaluation of the Effects of Data Pre-processing

TL;DR

This study evaluates how different data pre-processing strategies, including ROI selection and physiological noise correction, influence resting state functional connectivity estimates in the default mode network.

Contribution

It systematically compares three ROI selection methods and assesses the impact of physiological noise correction on connectivity measures in resting state fMRI.

Findings

Physiological noise correction reduces artifactual correlations.

Individual ICA-based ROI selection increases connectivity estimates.

Both factors significantly affect the reliability of functional connectivity analysis.

Abstract

Resting state functional connectivity estimates from MRI measures has become a promising tool to characterize human brain networks. There are, however, limitations in the method since several sources of errors have been seen to significantly affect the final estimates. This has lead to a great interest in the field to do systematic investigations that help determine the most reliable and robust strategies to perform functional connectivity analysis. In the present study, we examine the influence of two aspects of data pre- processing in resting state functional connectivity analysis: the effect of criteria used to select nodes in the default mode network (DMN) for the computation of connectivity, and the effect of using or not physiological noise correction. Three different strategies of region of interest (ROI) selection were compared to define DMN node coordinates: (1) ROIs centered on atlas-based coordinates, (2) ROIs based on the result of group independent component analysis, and (3) ROIs based on the estimated DMN of each individual. The study was done using data of 15 healthy volunteers, which had resting state data available from a separate project. We found that both effects, ROI selection criteria for DMN nodes and physiological noise correction, have significant effects on the functional connectivity estimates. In particular, our results show that physiological noise correction introduces small but significant reductions in functional connectivity, consistent with a reduction of artifactual non-neural correlations introduced by physiological effects. Further, selecting DMN nodes based on the single subject ICA results introduced small but significant increases in functional connectivity, consistent with higher subject specificity of the node selection.