OASIS3 BOLD and Tau PET conversion

[DAVID AQUILUÉ LLORENS]

Dear Clinica team,

I am interested in converting both fMRI BOLD and Tau PET from the OASIS3 dataset to BIDS using your converter. Unforunately, I see in the documentation (and after running the converter) that these two neuroimaging modalities are not implemented yet.

I understand that maintaining your repo will surely take a lot of time and probably you will not be able to extend this converter in a timeline that fits my needs. I would be interested in undergoing the task by myself, if that seems reasonable for you, I would be happy to give something back for the value you've provided in my current project.

I would gladly appreciate any comment on why fMRI / tau PET conversion was not implemented, possible errors/issues that I may encounter and some tips on what parts of the current repo should I focus on reading to understand the inner workings of the dataset converters.

Thank you very much for all the work you do,

All the best,

David

[Alice Joubert]

Hello David,

Glad to hear clinica helped you in your project ! I am unsure as to why these modalities conversions were not implemented. It is either it was no present in OASIS3 at the time and/or nobody needed these. We would be glad to help and review your code if you decided to work on it since we were not planning on adding these ourselves. For parts of the repository you need to read you can concentrate on clinica/converters/oasis3_to_bids as not all our converters work the same way and it is quite specific. Sadly it is not very well described (missing docstrings...) or tested so it might not be very clear to understand without running it with breakpoints so don't hesitate if you have questions.

For potential issues you might face :

- fmri BOLD : the images are well defined in the clinical data we already download for the converter. All you need to do I think is modifying the mapping defined at line 149 in `oasis3_to_bids/_utils.py`. Any modality that is not defined in the mapping will not be converted.

- Tau PET : is a bit more tricky. It is not defined in the csv that lists all the PET scans which we currently use for all other PET modalities. That means you would need to search for where these clinical data are because we cannot convert without them. By investigating quickly I saw it was in a project "OASIS3 - AV1451" but I don't where to retrieve its clinical data. If you find them that would mean we need to modify the converter to be able to load it, curate it and merge it with other csv properly. 

Otherwise, if you decide to continue with this, I suggest we do the following :

- I set up a specific branch different from `dev` in the repository (for example named oasis3_adding_modalities)

- You can make you own fork (see the contributing guide here in the documentation) after that. 

- When you want to make a modification you set-up a branch originating from oasis3_adding_modalities and do a PR pointing that branch, not dev

- If you know how to do it don't hesitate to unit test functions you add

- We review, make comments, validate, merge

- We handle stuff on our side like the continuous integration data and tests, coding styles, etc on this branch

- We merge the oasis3_adding_modalities branch into dev when we are all happy with it

Additional requirements : 

- We use a linter for coding style which I think is suggested when you commit for the first time and don't have it. We ask you to use it to match the rest of the codebase

- We follow BIDS specifications as much as possible but I don't think it's going to come into play here

- It's good to test the converter on real data from OASIS3 and we suggest you do it when coding what you need. When you are finished you can send the list of subjects you used so we can enrich the testing database for OASIS3.

What do you think of this ? Do not hesitate to ask if any of what I said was unclear.

Best,

Alice

[AQUILUÉ LLORENS, DAVID]

Hello Alice,

I have started working on the implementation. Indeed, the Tau PET part of the OASIS3 was previously downloaded as a different dataset and it does seem a bit more tricky to implement. I have now some questions to better understand how to proceed with the development of the additions to the converter:

1- For questions and comments about the code, should I just go ahead with my considerations (using the linter and your contributing guide) and wait for the first PR for your feedback?

2- The OASIS3 Tau PET data comes in raw format (4D volumes, with different temporal acquisitions), which is quite different from the ADNI PET data. The ADNI-to-BIDS converter takes the already averaged and co-registered images (step 2 from the ADNI PET processed data) for the transformation. 

At this point I have implemented a pipeline that takes the 4D volumes, averages and co-registers them, which is what I currently require and I believe will be necessary for most use-cases. Additionally, this would significantly simplify the choice of PET file to be transformed to the BIDS format as for each session there may be several runs, with only one of them actually useful for PET computations (depending on some recording protocol). Do you have any recommendations on how this should be managed? I was thinking of running it if some checks are not passed before actually executing the converter...

3- I am considering running some LLM-assisted documentation (I am testing Claude Code) to document the missing docstrings in the OASIS3 converter, would that be ok for you? Should that be implemented in the same branch?

Sorry for the long message, it's my first time collaborating on an open-source project from an external lab and want to make the most out of the available time in my hands.

All the best,

David

Missatge de Alice Joubert <alice00...@gmail.com> del dia dv., 6 de març 2026 a les 12:00:

--
You received this message because you are subscribed to a topic in the Google Groups "Clinica" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/clinica-user/javfI6eDWQk/unsubscribe.
To unsubscribe from this group and all its topics, send an email to clinica-user...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/clinica-user/7fef013d-6099-4f0e-81e9-d5444250c7dbn%40googlegroups.com.

--

David Aquilué Llorens

PhD Student

Computational Neuroscience Group
https://www.upf.edu/web/cns

Center for Brain and Cognition 

Department of Information and Communication Technologies (DTIC-UPF)

Ramon Trias Fargas, 25-27 (Mercè Rodoreda building - Ciutadella campus)

08005 Barcelona                    

Office 24 335

[Alice Joubert]

Hello David,

1. You can go ahead if you desire to. I created a new branch "oasis3_new_modalities" from which you can create your own branches and in which we will merge your future pull requests. This allows us to do our own modifications on your work before merging into "dev" if need be. If you feel like you have more questions needing answers before spending time on code writing you can ask them here or on GitHub. We can also set-up time aside for a teams/zoom meeting if you have a list of points that need to be discussed and that is easier orally.

2. Unless I misunderstood, this point tackles two issues : 

- 4D volumes : we try to follow as much as possible the BIDS specifications where it is indicated 4D PET volumes should be saved as 4D --> it is not really mandatory but more of an indication. Would you like to share the pipeline so we can see how it works before taking a decision ? I assume this could be an option in the converter and not the default setting to stay close to BIDS specs.

- Having several runs per sessions : do you have a subject/session example where there were several runs per session (outside of 4D) ? Can you detail what checks you would perform ?

3. I am fine with this as long as you re-read what Claude wrote before submitting for review :-). You can start with this in one PR if you want to. Ideally 1 PR = 1 functionality as much as possible.

Best,

Alice

[AQUILUÉ LLORENS, DAVID]

Hi Alice,

Okay, I will start working on the branch you have created. If you don't mind, I have created an issue on the repo and maybe we can move the conversation there. I will try to add the pipeline tonight/tomorrow to one of the branches and provide some documentation/answers to your questions and will document it there.  

I will carry on with the FTP conversion first, then ideally the fMRI conversion, and finally, if I still have time will aim to improve documentation of the entire oasis3 converter.

Keep in touch,

David

 

Missatge de Alice Joubert <alice00...@gmail.com> del dia dt., 17 de març 2026 a les 15:26:

You received this message because you are subscribed to the Google Groups "Clinica" group.
To unsubscribe from this group and stop receiving emails from it, send an email to clinica-user...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/clinica-user/5f8e40c8-dd0e-42d9-865c-8ca77c890df5n%40googlegroups.com.