Some ANTs interfaces will be broken

[armaneshaghi]

I'm documenting changes I need to make for ANTs interfaces, as a new built has broken some Nipype interfaces. Unfortunately, ANTs versioning and its backward compatibility are not the best.

• N4BiasFieldCorrection:

   --image-dimension needs to be changed to --image-dimensionality 
  

I think there are some changes we need to make for the registration interface, please add below as you find any broken interface.

[armaneshaghi]

• antsRegistration

collapse-linear-transforms-to-fixed-image-header

and

transformRigid

are no longer available.

[lsqshr]

An open pull request in Nov.14 has fixed the problem of collapse-linear-transforms-to-fixed-image-header #999. However it has not been merged?

[JDWarner]

I'll update #999 with the N4 bias correction change, and look into transformRigid.

[armaneshaghi]

Thanks. @satra do you suggest anything for backward compatibility? Shall we add a new input, such as version, so that the final output command line will change. For example an input for ants version may work like this:

antsRegistration.inputs.version = (a float)

But I do not like this solution, because it will complicate the life of new users. The the other (smarter way) would be to use Python subprocess to quickly scan the help of ants binary each time and set the flags.

[JDWarner]

👍 for the subprocess solution. Changing the flags is easy, adding robust logic to select the right one a little tricky. Unit testing that requires multiple ANTs versions ends up being hard, though, mainly thanks to Travis timeouts. Unfortunately I'm away right now and my laptop doesn't have ANTs built, so it's difficult for me to play with the parsing necessary.

I didn't bother with backwards compatibility for the collapse-linear-transforms-to-fixed-image-header change, as it was not doing anything previously for over 2 years.

[satra]

@armaneshaghi - there are a few things to consider.

• ants rarely comes out with versions and we often end up running a git version of ants. if we can convince the ants folks to at least create versions every month, then versioning can be used.
• there is a PR ENH: Adding new ants command line feature. #1004 by @hjmjohnson which reflects many of the changes, but merging that PR will break anyone without the latest version of ants.

so without having versions our only options might be to move some of these older interfaces to ants.legacy, but it will break some people's scripts.

@armaneshaghi - would you be willing to start a conversation on the ants thread about versioning? that would actually be most helpful, as nipype has support for versions through the min_ver and max_ver metadata flags. but this would also mean that we can resolve the version on an installation of ants. as far as i know, no version information is stored with the binaries.

[armaneshaghi]

I close this as the issue will be resolved by regular ANTs releases.

[blakedewey]

I know this is closed, but I just ran into this with ANTs and wanted to suggest an idea. The short CLI tag -d is used for both inputs even though the long tag has changed. At least for N4BiasFieldCorrection this might be a good reverse compatibility option until versioning can be worked out.

[oesteban]

👍 @blakedewey

[blakedewey]

Great! I'll put in a PR