Using surface data

[jaysonjeg]

Currently, the code takes Niimg-like objects. PairwiseAlignment also takes as input a mask (also Niimg-like object). How can we straightforwardly use surface space data with fmralign?

One idea: Make the input images numpy arrays (ntimepoints/ncontrasts x nvertices). 'clustering' could also become an array (nvertices,) This puts the onus on the user to convert any data format to numpy array before using fmralign. Ther user would also have to generate 'clustering' and mask the data before using fmralign. It improves usability without restricting to particular inputs data formats. One disadvantage of this is that passing full arrays to fmralign may consume too much memory, especially when doing many subjects in parallel.

[emdupre]

With the release of nilearn 0.10.2, we now have increased surface support; see, for example, this short demo of the current features.

The current plan is to continue improving support towards a full SurfaceMasker over the next couple of months. Since PairwiseAlignment heavily builds on NiftiMasker, my preference would be to wait until we can pull this functionality directly from Nilearn.

[emdupre]

This is, though, absolutely a feature that we should add !