Understanding how brain activity translates into behavior is a grand challenge in neuroscientific research. Simultaneous computational modeling of both measures offers to address this question. The extension of the dynamic causal modeling (DCM) framework for BOLD responses to behavior (bDCM) constitutes such a modeling approach. However, only very few studies have employed and evaluated bDCM, and its application has been restricted to binary behavioral responses, limiting more general statements about its validity. This study used bDCM to model reaction times in a spatial attention task, which involved two separate runs with either horizontal or vertical stimulus configurations. We recorded fMRI data and reaction times (n=29) and compared bDCM to classical DCM and a behavioral Rescorla-Wagner model using goodness of fit-statistics and machine learning methods. Data showed that bDCM performed equally well as classical DCM when modeling BOLD responses and better than the Rescorla Wagner model when modeling reaction times. Notably, only using bDCM’s parameters enabled classification of the horizontal and vertical runs suggesting that bDCM seems to be more sensitive than the other models. Although our data also revealed practical limitations of the current bDCM approach that warrant further investigation, we conclude that bDCM constitutes a promising method for investigating the link between brain activity and behavior.
bioRxiv Subject Collection: Neuroscience