Hello,

I used the default pipeline for preprocessing to MNI. In the Denoising step I checked the histograms (QA plot) to see whether I had any outliers. I did have 2 Subjects with very flat and still shifted histograms after denoising. I tried to run the analysis again and now the histograms of the two subjects were fine but another subject was an outlier. A third run gave again another result. (I always made sure to use the right files so that I do not e.g. smooth my data two times..)

Reading "Conn: A Functional Connectivity Toolbox for Correlated and Anticorrelated Brain Networks" (Whitfield-Gabrieli and Nieto Castanon, 2012) I found that for creating the histogram a "random subset of 256 voxels" is used. Is this still the case? This would explain the different results but then another question arises to me: how can I detect outlier/bad subjects if the histograms changes so drastically if I do the same analysis twice? 


Thanks and best regards,
Eli