default masked registration behavior in simple registration scripts

[stnava]

regarding this discussion: https://github.com/stnava/ANTs/commit/58cbd30bbfe8116d94847ec14622c45c3d268016

we might want to change default masked behavior in antsRegistrationSyN{Quick.sh}

two options that we are considering:

1. apply mask to all stages
2. only apply mask at the last stage ( e.g. SyN ) which would allow one to use global features for affine registration and then focus only the deformable mapping

obviously, we can enable both by adding a 2nd option ( applyMaskToAll ) but this would need to be implemented carefully .... for now, we use the above commits to document how to do this and as a placeholder until this issue is resolved.

[gdevenyi]

My experience thus far with masks is that masking at all stages isn't as robust as using unmasked followed by masks.

Following up on robustness, I found that this stages stack:

--transform Rigid[0.1]
--transform Similarity[0.1]
--transform Affine[0.1]

To be a bit better at handling large initial differences in brains (young ASD subjects vs MNI template for example)

[gdevenyi]

Followup,

I have also found that a masked affine stage (after unmasked affine) allows for better global correspondence of the region of interest (brains) before deformable registration.

[stnava]

thank you, both of these observations are helpful perhaps worth adding to antsRegistrationSyN{Quick}.sh as an option ... does seem logical to have this available. but i am beginning to wonder if we might want a separate antsMaskedRegistration.sh script.

[gdevenyi]

Well, I would envision antsMaskedRegistration.sh having the exact same options as antsRegistrationSyn{Quick}.sh so I don't see why the duplication of files (speaking of, I don't see why a --quick option wouldn't suffice in antsRegistrationSyN.sh)

I think the bash logic is simple and clear enough to insert extra stages based on the absence/presence of the mask option.

[stnava]

maybe --- if it is clear to you, do you want to make a pull request / some initial attempt at making such a script?

[gdevenyi]

Sure, I can take a look, although not till next week, I'm travelling to Boston tomorrow-Friday for meetings/teaching.

To be clear, what I will attempt to do is

• merge quick into script
• add usage of a fixed mask of affine (last scale) and syn (all scales)
• add the Similarity stage to improve robustness