search

nilearn / nilearn

Machine learning for NeuroImaging in Python
http://nilearn.github.io
Other
1.19k stars 602 forks source link

Adapt plot_haxby_searchlight_surface example to work with the new surface image API #4511

Closed Remi-Gau closed 1 month ago

Remi-Gau commented 2 months ago
Remi-Gau commented 2 months ago

FYI I think it may be necessary to create a "dataset fetcher" for the haxby dataset that returns surface object like was done for a couple of other fetchers for NKI and destrieux.

See for example:

https://github.com/nilearn/nilearn/blob/88be15f61bb9d794d643dbd3e98239239e3ee13d/nilearn/experimental/surface/_datasets.py#L91

bthirion commented 2 months ago

But the images of Haxby dataset are not correctly spatially normalized. So the projection to fsaverage will return misleading stuff.

Remi-Gau commented 2 months ago

Yet that's what we already do in this example. Should we add a warning in this example then?

https://nilearn.github.io/stable/auto_examples/02_decoding/plot_haxby_searchlight_surface.html#sphx-glr-auto-examples-02-decoding-plot-haxby-searchlight-surface-py

bthirion commented 2 months ago

I think we should.

Remi-Gau commented 1 month ago

Also would require running searchlight on the surface which I do not think we have at the moment: https://github.com/nilearn/nilearn/pull/4205#issuecomment-2386033946

bthirion commented 1 month ago

Maybe we should think of replacing Haxby dataset with another one on which we could run surface-based analysis ?

Remi-Gau commented 1 month ago

This would also require to have searchlight to run on surfaces. So I think I will close this for now and we may revisit it later.