search

ANTsX / ANTsPyNet

Pre-trained models and utilities for deep learning on medical images in Python
https://antspynet.readthedocs.io
Apache License 2.0
201 stars 29 forks source link

ANTsXNet Longitudinal Thickness - Atropos #144

Open dbrennan44 opened 3 weeks ago

dbrennan44 commented 3 weeks ago

Hi team, I'm working with TBI patients at an early time point - there's a lot of structural change such as swelling which makes the affine registration to the SST, even if SST is built with ANTsSyN, very poor.

In classic ANTsX SST priors could be nonlinear warped to the individual timepoint - is this appropriate for ANTsXNet with some tweaking? I realize "SST" cortical thickness is superior but the affine warp to SST is not usable in this use case.

Thank you!

ntustison commented 3 weeks ago

The ANTsXNet longitudinal thickness pipeline doesn't work that way. The closest thing to "tweaking" that you could do would be do refine the deep atropos weights for your specific data cohort.

cookpa commented 3 weeks ago

If I understand correctly, the antsxnet longitudinal script registers images to the SST space with an affine transform, but does not perform the shape normalization implemented in antsMultivariateTemplateConstruction2.sh.

In antsnetct, there's a different pipeline that calls antsMultivariateTemplateConstruction2.sh directly. This can support a nonlinear registration to the SST space. The deep_atropos posteriors on the SST are then transformed to session native space and classical Atropos is called in that space.

dbrennan44 commented 3 weeks ago

thanks @cookpa this may be the solution I need - I wasn't aware of antsnetct until your reply to my last post yesterday. I'll run it and post the results. Thank you!

cookpa commented 3 weeks ago

Great, please feel free to open issues over there if you have questions