Can Dynamic Causal Modelling (DCM) can tell us which parts of the brain talk to each other
Dynamic Causal Modelling (DCM) is a method of analysis. It is can be used on brain data obtained through functional magnetic resonance imaging (fMRI), based on the blood oxygenation, BOLD response, which is a measurable signal whenever a brain region is activated.
While typical fMRI analysis attempts to directly model the relationship between experimental manipulation and BOLD response, the DCM approach divides this into 2 steps and also covers the intermediary step by attempting to model directly how neural activity is modulated by the external input (external inputs being the experimental manipulation, ex. Instruction: “Look at this picture and name what you see”) as well as model how the neural activity then relates to the observed BOLD signal.
Unlike typical fMRI analysis, the DCM approach doesn’t just test one alternative hypothesis against the null hypothesis, which is inherently limited to questions like is there significant (beyond a certain threshold) activation in a certain brain region or not? Or does activation between two brain regions correlate or not? Instead the dynamic causal model compares a very large set of alternative models, explanations so to say, of how certain brain regions are connected. For instance, between the same brain regions, the connectivity can be very different in terms of directionality (which region is activated first, second, third,…?). This involves providing separate values for the parameter “connectivity”, one for the direction of flow from brain region A to brain region B and another (potentially smaller or larger) for the direction B to A. It is then possible to characterize the actual flow of activation. In addition to these multiple connectivity parameters, the DCM model also gives you values that estimate how much these connectivity values change as a result of the experimental manipulation. Now, this is unique to DCM because typical functional (rather than DCM) connectivity models only provide one value for the connectivity parameter, which is simply the correlation between two brain regions.
Let’s illustrate this with an example: We are interested in the connectivity between three different brain regions, so we study them in two different contexts (or experimental manipulations as discussed above). No need to get into specifics here, let’s just say one context involves the person in the brain scanner passively viewing someone fly through a maze while in the other context the same person has to actively navigate through it. Now when there is an increased connectivity strength from one region to another, the direction of this signal flow can be modelled because we have different connection strengths for each direction between these three regions. Of course, the network can be expanded to many more regions than just three, but this would involve increasingly complex and computationally demanding model comparisons. Comparing 6 instead of 3 brain regions would already result in over 30,000 possible connectivity models! For this reason, it is currently more common in neuroscience research to use DCM only when strong predictions about a limited number of brain regions that are likely to be connected or part of a particular network have already been formed a priori (beforehand). This can even be done based on the results of classical connectivity analysis, which are then taken a step further with DCM in order to uncover the directionality of signal flows. With the ever-increasing computational power of our brain imaging tools, DCM has incredible potential to be eventually applied to a vast number of potentially connected brain regions to discover entirely new networks in the brain.