Check continuity of fast-track, esp. multi-session datasets and w/cohort normalization #224
Comments
|
Hello, Fast-track seems to break with my dataset on smriprep 0.7 / fmriprep 20.2.0, notably due to bids' build_path not considering potential session/runs within the name. I developed a dirty-hack which rely on BIDSLayout instead and works for me. commit: nuKs@bd72f83 I did not PR as I am not sure it's the best way to fix this, and if this works on other use cases. The included logging of missing files is convenient in any case. |
|
The patch also fixes anats within two sessions. The use case appears to be different as for single session anat, no template has to be created, while 2 sessions are averaged. In both case, "run-" is kept. Here are the outputs for anat derivatives (smriprep) of subsets of the day2day dataset, with single session: double session: |
|
@oesteban what are your thoughts on this? Do you have had any feedback about fast-track since the update? I could be missing something as well. edit: I have the same issue with the IBC dataset with smriprep 0.7.1 and fmriprep 20.2.1, some subjects have 2 anats so it is averaged and thus are working fine (these folders exist: /sub-xx/anat + 2x /sub-xx/ses-yy/anat) while some subjects have only 1 anat and fail (only this folder exists: /sub-xx/ses-yy/anat, /sub-xx/anat is missing). |
|
My patch seems to work, though not up to date with the fmriprep trunk (no config python file anymore etc). I did not compare my patch output with traditional fmriprep non-fast track output so i am not 100% sure everything is fine (might pick the wrong template etc). I can integrate it with the latest version if you think it's OK. edit: |
|
Hi @nuKs - thanks for your patience, we are seriously limited in time lately - What would you need from us to push this forward? Just check out your branch and test on our settings? Do you have examples that make the fast-track fail? |
|
Hi, no worries, thanks for sharing your work! What my patch does is to rely on pybids layout instead of bids build_path in order to retrieve the fast_track derivatives files. If you have worked on these bits, a simple quick look at my patch code, or even just the way I've described it, may allow you to see if there is any red flag regarding what I did. I made the assumption that fmriprep must have some additional step that picks up the right derivative within the list, and I didn't check this step's code so I am not 100% sure I patched the right way considering the derivative list returned by the patched function is now less restrictive than it used to be. I am perhaps being overly cautious, If you don't see any reason why the patch would cause issues, then everything should be fine,. So far, examples that make fast-track fails include:
In my case, IBC has been failing (without fix) for some subjects, modified/reduced day2day datasets too. I'll cleanup/update everything to the latest smriprep version so you can test and merge if you wish to. |
What my patch does is to rely on pybids layout instead of bids build_path in order to retrieve the fast_track derivatives files. Pybids layout does adapt to any bids-valid naming scheme, while build_path does not, but rather directly assumes the file name. So far, examples that make fast-track fails include: - subjects with a single T1 (as the path is different since no template is made) - subjects with T1 containing run-x segments within the name. EXAMPLE: In my case, IBC has been failing (without fix) for some subjects, modified/reduced day2day datasets too. IBC openneuro: https://openneuro.org/datasets/ds002685/versions/1.3.1 day2day upon request: https://bmcneurosci.biomedcentral.com/articles/10.1186/s12868-017-0383-y WARNING: This fix is less restrictive than previous version, thus might select too much files. This risk should properly be assessed. cf. nipreps#224
What my patch does is to rely on pybids layout instead of bids build_path in order to retrieve the fast_track derivatives files. Pybids layout does adapt to any bids-valid naming scheme, while build_path does not, but rather directly assumes the file name. So far, examples that make fast-track fails include: - subjects with a single T1 (as the path is different since no template is made) - subjects with T1 containing run-x segments within the name. EXAMPLE: In my case, IBC has been failing (without fix) for some subjects, modified/reduced day2day datasets too. IBC openneuro: https://openneuro.org/datasets/ds002685/versions/1.3.1 day2day upon request: https://bmcneurosci.biomedcentral.com/articles/10.1186/s12868-017-0383-y WARNING: This fix is less restrictive than previous version, thus might select too much files. This risk should properly be assessed. WARNING: freesurfer files aren't fixed. cf. nipreps#224
What my patch does is to rely on pybids layout instead of bids build_path in order to retrieve the fast_track derivatives files. Pybids layout does adapt to any bids-valid naming scheme, while build_path does not, but rather directly assumes the file name. So far, examples that make fast-track fails include: - subjects with a single T1 (as the path is different since no template is made) - subjects with T1 containing run-x segments within the name. EXAMPLE: In my case, IBC has been failing (without fix) for some subjects, modified/reduced day2day datasets too. IBC openneuro: https://openneuro.org/datasets/ds002685/versions/1.3.1 day2day upon request: https://bmcneurosci.biomedcentral.com/articles/10.1186/s12868-017-0383-y WARNING: This fix is less restrictive than previous version, thus might select too much files. This risk should properly be assessed. WARNING: freesurfer files aren't fixed. cf. nipreps#224
What my patch does is to rely on pybids layout instead of bids build_path in order to retrieve the fast_track derivatives files. Pybids layout does adapt to any bids-valid naming scheme, while build_path does not, but rather directly assumes the file name. So far, examples that make fast-track fails include: - subjects with a single T1 (as the path is different since no template is made) - subjects with T1 containing run-x segments within the name. EXAMPLE: In my case, IBC has been failing (without fix) for some subjects, modified/reduced day2day datasets too. IBC openneuro: https://openneuro.org/datasets/ds002685/versions/1.3.1 day2day upon request: https://bmcneurosci.biomedcentral.com/articles/10.1186/s12868-017-0383-y WARNING: This fix is less restrictive than previous version, thus might select too much files. This risk should properly be assessed. WARNING: freesurfer files aren't fixed. cf. nipreps#224
|
Jumping in that old convo: I noticed that the fast-track would fail if you process a single anat in a multi-session dataset (and results is output in a session folder) and then try to use that output as anat fast-track for fMRIPrep. Because |
|
Had made that PR if you want to check in, a bit old though. |
#219 and #223 will likely break some parts of the anatomical fast track. Let's make sure it works before cutting 0.7.
The text was updated successfully, but these errors were encountered: