Impact of sample size and global confounds removals on estimates of effective connectivity

The interactions within brain networks are inherently directional and can be detected by using thespectral Dynamic Causal Modelling (DCM) for the resting-state functional magnetic resonance imaging (fMRI). The sample size and unavoidable presence of nuisance signals during fMRI measurementare the two important factors influencing stability of the group estimates of connectivity parameters. However, most of the recent studies exploring effective connectivity were conducted for rathersmall and minimally preprocessed datasets. Here, we explore an impact of these two factors by analyzing the cleaned resting-state fMRI data for the group of 330 unrelated subjects from the HumanConnectome Project database. We demonstrate that stability of the model selection procedure andinference of connectivity parameters are both dependent on the sample size. The minimal samplesize required for the stable Dynamic Causal modelling has to be about 50. Our results show thatglobal confounds removals have weak or moderate effect on DCM stability for the datasets properlycleaned from the artifacts