Using nilearn.decoding.Searchlight, I successfully performed a searchlight analysis on each subject. For each subject I now have their accuracy maps. Ultimately I am hoping to show some sort of group-level analysis, in particular a group cluster-corrected stat map (e.g., t values) produced from each subject’s (smoothed) accuracy map in order to locate decoding hot spots at the group level.
What is the best way to go about this procedure after obtaining the maps from Searchlight? I notice that using AFNI's 3dClustSim requires a smoothness estimate based on residual images, which I don’t have from Searchlight. Conversely, I’m totally lost in FSL documentation and it’s unclear to me how to run a basic cluster correction with it (using what I’m guessing involves randomise and cluster?). Are there any other alternatives that are easy to use with accuracy maps (or a group-level t/z image)?
Are there any other alternatives that are easy to use with accuracy maps (or a group-level t/z image)?
Last I checked (a couple years ago), cluster correction of searchlights does not have very user friendly tools, and you’re probably going to be stuck implementing some of it yourself. The canonical paper I would recommend is Stelzer et al (2012) (summarized well and linked here). It’s probably worth searching for more recent papers, though.
there is also an easier to use implementation as part of the pymvpaw toolbox created by Ryan Stolier. He created some wrapper function for the pymvpa functions to more easily organise the data and implement the Stelzer et al 2012 algorithm . Perhaps it hepls:
Refreshing the topic.
I also performed Searchlight, but with RSA method, and wonder what is the best way to perform the group-level analysis to get cluster corrected stat map? Is there any reliable guide on how to do it (in python)? There’re a lot of ideas on how to do it, starting from simple one-sample t-test producing z image (which I did). However, I wonder if that’s enough?