RSA region specificity


Dear all,

I am currently running a region of interest (ROI) RSA on the concatenated fMRI data and I am facing a weird pattern of activations. The general same structure of activations is present across all ROIs and weirdly enough is present also in cerebro-spinal fluid (CSF) and white matter. I haven’t still tested the differences between the regions and I am sure that there will be differences. However, is it common to have the same general pattern of activations even in the regions which theoretically should not show any specific patterns (the white matter and the CSF in my case)? Did anyone experience the same lack of region specificities?



Hi Mariam,

That sounds weird. Would you mind describing your problem in more detail? If I understand it correctly, you are finding very similar RDMs in lots of different brain regions? In general, this is not unusual when using data from within runs, although the pattern should obviously not be meaningful. Did you create your RDMs on data of the same runs or calculate (dis)similarities between runs? (the latter can help avoid problems due to autocorrelation within runs). How large are your RDMs, are they based on single trial betas? How did you calculate your betas? All of this can help figure out an answer to your problem. But in general, I would first have a look at the data before they enter the RDM stage (i.e. the data matrices that you use for calculating (dis)similarities). If you find large differences between conditions everywhere, this might also speak to confounds (e.g. head motion / respiration).



Agreed, this does not sound normal, but rather suggests a bug. You mention running a ROI-based RSA, but your question about seeing “activations” (to be picky, RSA results aren’t exactly “activations” but similarity patterns) in CSF and white matter makes me wonder if you did a searchlight analysis. Or is this a large “CSF” ROI?


I double Martin.
also it might be that you are picking up on consistent noise in your RDMs. That could only occur if your design is quite unbalanced. also if you smooth a lot, you might get some “structure” in WM/CSF due to smearing.