5ttgen neonatal algorithms #3193
Conversation
Co-authored-by: Manuel Blesa Cabez <mblesac@gmail.com> Co-authored-by: Paola Galdi <paola.galdi@gmail.com>
Co-authored-by: Manuel Blesa <mblesac@gmail.com> Co-authored-by: Paola Galdi <paola.galdi@gmail.com>
Co-authored-by: Manuel Blesa <mblesac@gmail.com> Co-authored-by: Paola Galdi <paola.galdi@gmail.com>
Co-authored-by: Manuel Blesa <mblesac@gmail.com> Co-authored-by: Paola Galdi <paola.galdi@gmail.com>
Co-authored-by: Manuel Blesa <mblesac@gmail.com> Co-authored-by: Paola Galdi <paola.galdi@gmail.com>
…put) Manuel Blesa <mblesac@gmail.com> Paola Galdi <paola.galdi@gmail.com>
…atal algorithms Co-authored-by: Manuel Blesa <mblesac@gmail.com> Co-authored-by: Paola Galdi <paola.galdi@gmail.com>
- Ensure that input files are unique and unambiguous. - Remove redundant logic across get_inputs() and execute() regarding use of hard segmentation. - Simplify logic of primary conversion loop. - Generate sub-cortical grey matter image from M-CRIB only once, rather than once per tissue; this should facilitate code execution without using the -force option.
Proposed code cleanup
Conflicts: docs/reference/commands/5ttgen.rst python/mrtrix3/commands/5ttgen/dhcp.py python/mrtrix3/commands/5ttgen/mcrib.py
|
clang-tidy review says "All clean, LGTM! 👍" |
|
Hi @Lestropie , several things have changed and I think some things in the algorithm should be edited to make it applicable to most datasets. Let me get back to you after ISMRM deadline, then I'll have time to work on this. One quick question, what are your thought about including docker/apptainer in the scripts? |
|
Hi @Lestropie , I found a way to access some neonatal data to test the algorithm, I'll send you an email with the detailed instructions |
|
Thanks @mblesac, will work my way through the data and produce something for CI.
Generally I would prefer to avoid: MRtrix3 could itself be invoked within a container (whether the official one or third party), and so itself attempting to invoke containers becomes a problem. It's not to say it can't be done if available, but IMO it's kicking the can down the road and could become a nightmare if utilised too much. I think the only current example of it is in Generally in such circumstances my preference would be to have the user invoke some tool and then have MRtrix3 take as input the derivatives of that tool, rather than MRtrix3 itself invoking the tool. It's not always possible, or might make the user interface a little clunkier as it's a smaller fraction of the requisite processing that can be encapsulated, but it provides a more clear distinction regarding what is MRtrix3 and what is not (which is an ongoing problem). |
- If it is possible to pass the M-CRIB template images through to ANTs without necessitating an explicit load of those images into the scratch directory, do so. - If the M-CRIB data have been downloaded and extracted by the user in a way that preserves the original sub-directory structure, navigate that structure to find the required images. - Continue processing even if only a subset of the M-CRIB images are found (this is exploited to speed up test execution). - Add test data and tests for 5ttgen mcrib.
|
clang-tidy review says "All clean, LGTM! 👍" |
|
|
- New alternatives for masking of image. - By default, map brain stem to fifth image volume, mirroring behaviour of 5ttgen hsvs. Co-authored-by: Manuel Blesa <mblesac@gmail.com>
- Merge VentralDC with brain stem component. - Refactor help page, including option groups and additional description. - Fix operation with piped input/output images. - Perform some limited intermediate cleanup. - Preserve relative ordering & sign of strides of input image in output images. - Avoid explicit tissue sum normalisation if using hard segmentation.
|
clang-tidy review says "All clean, LGTM! 👍" |
Supersedes #2393 (didn't want to suddenly start pushing to an external
masterbranch).This really needs test data. But it looks like access is pretty locked down.
@jdtournier @mblesac @paola-g: Is anyone able to provide an example dataset? Hopefully if I downsample a dataset prior to uploading to
script_test_data, it won't offend anybody...I also note that there's reference to BIDS layout in more recent data releases, to which the current code bears no resemblance. I'm guessing that different data releases have used different storage conventions?