When you define a contrast, it may reveal the same psychological processing as the behavior response does. Then it's dangerous to correlate brain activation resulted by the contrast and participants' behavior response. How can we justify it?
Hi,
This si a very interesting questione concerning fMRI data analysis.However, I advise to re ad the papers of Kriegeskorte a out circularity of analysis in fMRI.BYE
it would be considered double dipping if you ran say, a regression analysis on behavioral scores, identified a region that showed a significant effect of behavioral score on BOLD activity, and then created an ROI to extract parameter estimates to perform behavioral correlations with. The problem is that the analysis is circular in this case. However, if you identify a region based on BOLD magnitude differences between two conditions, create an roi, and extract parameter estimates for behavioral correlations, then this is not circular, but does introduce bias into the analysis. The least problematic way would be to define your roi based on a fully independent analysis (I.e. Functional localized) or based on a priori expectations from the literature.
I concur with the use of an a priori VOI (and not a VOI based on contrast activation in that same sample) as a way of avoiding this. You could see if someone has done an ALE study of that task and use the maxima from that external review. With handy tools like neurosynth.org and its dynamic meta-analytic platform, you can actually localize a VOI that has been recruited by dozens if not hundreds of studies of a similar task or cognitive construct, by entering relevant search terms.
I think extracting BOLD response magnitudes from a VOI derived from contrast activation is ok if you're simply trying to understand what drove the A-B contrast- greater BOLD response to stimulus A or a de-activation by stimulus B, for example.
Joseph, about identifying a region based on BOLD magnitude differences between two conditions, creating a ROI and extracting parameter estimates for behavioral correlations, what kinf of bias does this procedure introduce into the analysis ?
Thank you.
Authors should take care in how they present such analyses. I concur that an independent ROI works well. It is also perfectly acceptable to use the data-defined ROI and extracted data to correlate with the behavioral response to illustrate the fundamental relationship observed in such a regression model. This would be properly described as a posthoc, illustrative figure/analysis, but NOT as an independent analysis. I see authors and reviewers confusing this all the time. If the data are merely presented and interpreted as one effect, illustrated in brain and scatterplot, that is OK. Presenting as somehow independent or corroborative would be circular.