Replizierbarkeit von Dynamic Causa Modeling anhand fMRT Daten eines motorischen Paradigmas

Dynamic causal modeling (DCM) is a novel technique, which enables the investigation of effective connectivity between brain areas using fMRI data. While its validity has been repeatedly subject to investigation, there is little knowledge of the reliability and replicability of this method. The goal of the present study was to replicate the findings of Grefkes et al. (2008). In their study, participants performed a simple motor task while fMRI data was acquired. This data was analyzed using cDCM. The experimental conditions in the present study were closely matched to those of Grefkes et al., but the number of participants was increased to N=35. Regarding neural activity, the results presented by Grefkes et al. were highly replicable. Using the same framework conditions as Grefkes et al. (cDCM and BIC/AIC) and also using a more recent version of DCM (DCM10 and F) the results could almost exactly be replicated regarding coupling parameters and model selection. Using F as model selection criterion in cDCM the model selection differed whereas the key findings concerning coupling parameters could be replicated. Pool et al. (2013 & 2014) conducted two similar studies utilizing the same motor task and presented very similar findings. The present study underlines the importance of investigating seemingly minute changes in the DCM algorithms and model selection as these may have great influence on the results. Still this study finds a good replicability of cDCM results across different scanners and subjects - one critical precondition for using DCM to investigate brain function and especially with regard to future medical application.