PyMVPA Hyperalignment + fMRIprep Masking Template

I am currently going through the hyperalignment tutorial with fMRIprep outputs, and am wondering what masking should I be using for each dataset? To my understanding, I am looking at using the brain mask from the specific subject, but I believe that creates masked data that have different number of features across subjects. Would this be an issue?

Or instead, is there a template for fmri data? Is it a fair assumption that hyperalignment would want datasets to be masked the same way? Thanks!

Theoretically, hyperalignment doesn’t require the features to be the same for all subjects, and you can even hyperalign one brain region to another. At some point SearchlightHyperalignment required the feature set to be the same for all subjects, but the constraint should have been removed by now (if not, please let me know).

Specifically for fMRIprep outputs, I prefer to use the data (and masks) in the template space, such as MNI2009cAsym or fsaverage. These data have been aligned anatomically to the same template, and thus it’s completely fine to use an anatomical mask from the template.

1 Like

Thank you for the clarification! I will continue exploring Searchlight Hyperalignment using fMRIprep outputs (i.e. MNI2009cAsym or fsaverage) and bring up any issues if I encounter any here! Thanks again.