I’ve downloaded Haxby’s visual object recognition dataset from OpenNeuro which is available in BIDS format. I also downloaded the preprocessed data that is generated using fmriprep, and it seems like it is transformed to an MNI space with (3.5 x 3.75 x 3.75) dimensions. So now I have difficulty generating a mask (ventrotemporal) using atlases in FSL because as I see, the MNI structures available there are not in (3.5 x 3.75 x 3.75) dimensions. Are there any atlases in FSL that provide structures in this particular space? If not, how can I generate a mask in this dimension? I’d appreciate your help. Thank you!
I think you might be confusing space with resolution and field of view. The MNI space can be expressed in any voxel size and field of view. This will mean that shape of the data in voxels will be different. Some tools take this into account and apply interpolation on the fly (fsleyes, Mango, nilearn), but some don’t and throw an error if the shape of the data in voxels is different (fslview). This does not mean that the file you are working with is not in MNI space - it only means that the data is not resliced in the exact shape as the particular atlas you are working with.
Practically speaking you need to reslice your atlas to the resolution of the files produced by FMRIPREP (for example using nilearn’s resample_to_img) or extract timeseries from the data using a tool that does interpolation on the fly (such as http://nilearn.github.io/connectivity/functional_connectomes.html).