Dear Jesús,
In which regions are you interested in? It might be easier to search for an atlas that includes probabilistic labels for the GM or WM structures of interest instead of trying to mask voxels containing CSF. But in any case, an atlas that also comes with files in native space (like the one suggested by Alexander) might especially be useful, as you could preprocess those images the same way as your data (e.g. if you plan to use Dartel or Shoot for a more accurate between-subject registration).
Please keep in mind that during preprocessing, signal from CSF voxels might already spread into non-CSF voxels to some extent due to interpolations (e.g. during realignment when working with functional data), especially when it comes to smoothing. Thus, it might be wise to mask the files before the smoothing, possibly even better still in native space (before normalisation). For that purpose, you would have to warp the atlas files and labels into native space, which is easy to achieve with atlases like the Hammersmith one.
Best
Helmut
|