In my study, I utilized task-based fMRI and conducted activation analysis to define ROIs. Then I constructed an effective network using these ROIs in the same set of participants.
However, I am wondering the potential presence of double-dipping or circularity in this process. Is it necessary to rely on existing literature to select ROIs in order to avoid this issue.
I think we would need to know more about your planned analyses and hypotheses to judge this best. If you are worried, one thing you can do is use half of the time series to form the ROIs and other half in your connectivity analyses. That also gives you the opportunity to look at the robustness of your results by comparing results with using half 1 vs half 2 in defining your ROIs.