ABSTRACT
Timecourses that exhibit identical behaviour at distinct measurement occasions are reliable. Voodoo connectivity occurs when connectivity among brain regions exceeds within subject timecourse reliability. Thus, timecourse reliability limits the true detectable connectivity. We reproduced a working memory related connectome consisting of 561 paths obtained from 67 individuals. We tested >100000 fc-MRI pipelines and show that Savitzky Golay (SG) filters maximize true connectivity while conserving cognitively relevant changes of signals. This is noteworthy for approaches that focus on rapidly changing aspects of connectomes. Furthermore, SG filters detect zombie activity. These “resting state oscillations” are not under human control and contaminate working state signals. SPM pipelines exhibit more voodoo connectivity than SG pipelines. With the SPM pipeline, we observed a connectivity of r=0.44 and a poor true connectivity of r=0.23, but with the SG pipeline we observed a connectivity of r=0.59 and a fair true connectivity of r=0.43. The number of paths detected with fair true connectivity (r >0.4) was 4 for the SPM pipeline but 352 for the SG based pipeline. However, superior statistical properties of SG pipelines may not reflect neural reality. Hence, causal external validation of fc-MRI pipelines is crucial. Without such studies, different pipelines produce at best “alternative maps”.
