Any advice for skull stripping subjects with severe atrophy (e.g. Alzheimer's disease)


I’m running several subjects from the ADNI dataset through fmriprep, and I’ve found a number of subjects where just a little too much brain is stripped off in the skull stripping (usually in the temporal an frontal regions, where the atrophy is particularly bad). This is a common problem in subjects with severe atrophy - I originally switched from using BET to fmriprep because it’s much better at it. However, it’s not perfect, and I was wondering if anybody had any better success using alternative templates.

I notice that fmriprep uses to do the skull stripping, and that it defaults to the OASIS brain template. Looking here it seems like this template is based on 15 subjects ranging in age from 19 to 34. I’m thinking that an older/wider range of subject ages might work better on AD brains. The webpage above lists 4 templates that come with antsBrainExtractor - IXI, Oasis, NKI-1 and Kirby/MMRR. fmriprep allows you to use Oasis and NKI. Is there a way to use the others? IXI in particular seems to go up to age 90, so it might be better for my purpose. Would this just involve getting IXI into templateflow, or is templateflow only for registration targets? I’d be happy to try stuffing the IXI images into templateflow, if I could then use it as a brain template.

Any other tips and tricks for stripping brains with lots of atrophy would also be welcome…


Hey Blaise. I believe that we can now use any TemplateFlow template for skull-stripping. If it’s saying that you can only use OASIS or NKI, then you’re probably on an older version.

To add support for IXI, the first thing to do would be to submit a new template to TemplateFlow. I’m not sure if we’ve gotten an outside contribution before, so if you would be willing to try, that would be a big favor to us. Once it’s in there, it should be usable by recent versions of fMRIPrep, such as the LTS.

I’ll give it a go. I’ll also upload the macaque that I actually did a few months ago but never managed to submit.

1 Like