Functional connectivity estimation in fMRI data: Influence of preprocessing and time course selection

A number of techniques have been used to provide functional connectivity estimates for a given fMRI data set. In this study we compared two methods: a 'rest-like' method where the functional connectivity was estimated for the whitened residuals after regressing out the task-induced effects, and a within-condition method where the functional connectivity was estimated separately for each experimental condition. In both cases four pre-processing strategies were used: 1) time courses extracted from standard pre-processed data (standard); 2) adjusted time courses extracted using the volume of interest routines in SPM2 from standard pre-processed data (spm); 3) time courses extracted from ICA denoised data (standard denoised); and 4) adjusted time courses extracted from ICA denoised data (spm denoised). The temporal correlation between time series extracted from two cortical regions were statistically compared with the temporal correlation between a time series extracted from a cortical region and a time series extracted form a region placed in CSF. Since the later correlation is due to physiological noise and other artifacts, we used this comparison to investigate whether rest-like and task modulated connectivity could be estimated from the same data set. The pre-processing strategy had a significant effect on the connectivity estimates with the standard time courses providing larger connectivity values than the spm time courses for both estimation methods. The CSF comparison indicated that for our data set only rest-like connectivity could be estimated. The rest-like connectivity values were similar with connectivity estimated from resting state data.