I want to run some effective connectivity analyses on some fMRI data I have - to compare directional connectivity between group A and group B, and would love some advice on the best approach to use. The task is perception of famous faces.
Originally I intended to use DCM however, I’m not sure if I meet the assumption of there being some experimental manipulation. I am not comparing perception to baseline as there is not an appropriate one to compare against. Would it be more appropriate to use some spectral DCM technique? I saw some used in the context of resting state fMRI, which my data is not, however as there is no ‘pertubation’ - seems like this could be more appropriate.
thank you very much for your interesting question and welcome to neurostars, it’s great to have you here.
I’m not sure if we have a lot of DCM folks/experience here, however, could you maybe provide a bit more detail on your paradigm/task and intended analyses/hypotheses?
Thank you for the kind welcome and your reply about DCM!
I have two participant groups who vary in their subjective experience of perception and imagination. I am investigating neural correlate differences that reflect this. All participants do exactly the same task, they look at then subsequently imagine pictures of famous places and faces. The research questions is essentially whether the groups show difference in top-down and bottom-up connectivity in their visual perception-imagination network. i.e. for example does group 1 show increased information flow from occipital to frontal regions.
Upon further reading, it seems spectral DCM is inappropriate, as this assumes connectivity is roughly consistent throughout the paradigm, whereas in this case, there is a specific task (perceive an image when it appears).
ah ok, coolio, thx so much for providing more information!
From my limited experience/knowledge of DCM you would think that you have two options: “classic” statistics (e.g. t-tests, anova, etc.) and Parametric Empirical Bayes.
In the first, you would basically submit the obtained connectivity parameters to a two-sample t-test or e.g. 2x2 anova, testing for each connection if it is meaningfully different between the two groups. Afterward, you should correct your results for multiple comparisons, i.e. number of connections you tested. However, the problem with this approach is that the estimated distribution of the connectivity parameters and their uncertainty is neglected.
In the second, you would actually have the option to evaluate group and participant-specific effects more precisely. IIRC, this is the approach recommended by DCM folks/developers. You can find a bit more information on Parametric Empirical Bayeshere.