Rescaling output template from buildtemplateparallel.sh

[cjasal]

Hi ANTS team! Great work, and thank you for sharing the great functionalities of this package.

I was running buildtemplateparallel.sh to create a template from T1-weighted Images. As an overview, the steps I did were:

1. Denoising using DenoiseImage
2. Skull Stripping
3. N4BiasFieldCorrection
4. Registering (rigid) all T1 to ICBM or MNI152
5. Then, using buildtemplateparallel.sh with 1 random subject as the initial target

However, I observed that the output template is in a scale of 0 to 7, and its histogram does not follow a typical T1-weighted image. I was planning to feed the template to FreeSurfer for parcellating the cortical and subcortical grey matter, but recon-all fails since it cannot detect WM, mainly due to the intensities. Are there any post-processing rescaling I can do? Or are there preprocessing steps I might have done wrong?

Thanks!

[cookpa]

BTP is an older script and doesn't have an option for this, but if you use antsMultivariateTemplateConstruction.sh, you can choose how to combine images with the -a option, and can avoid normalization (you should normalize intensities offline in this case). You could run it with one iteration using your existing template as the initial template (-z).

FYI there is also antsMultivariateTemplateConstruction2.sh which uses newer registration, but antsMultivariateTemplateConstruction.sh is closer to BTP in terms of the underlying registration, so if you're happy with the template it may be the better option.

[cookpa]

An even faster option is to pre-scale and then re-average the last set of Warped images from the BTP run. You would just normlalize them somehow and then call AverageImages. This gives you something close to the template from iteration (n-1).

These days, I typically use ANTsPyNet / ANTsTorch to do a fast tissue segmentation, then run N4, then scale the images so that the white matter has some constant mean (say, 1000). Then I use the template scripts with a plain average, no additional normalization.

[cjasal]

Thank you so much @cookpa!

I'm now running antsMultivariateTemplateConstruction2.sh and antsMultivariateTemplateConstruction.sh with the -a 0 function. I'll update here on the outcome. This is after denoising, correcting for 4NBias, and normalising the input image and having a constant mean of 1000.

I'll also try to do what you suggested about re-averaging the last set of Warped images from the BTP run and see if it pushes through recon-all for FreeSurfer. Thanks!